Elizabeth Couper-Kuhlen, Bernd Kortmann-Cause, Condition, Concession, Contrast - Cognitive and Discourse Perspectives (Topics in English Linguistics, No 33) (2000) PDF

Cause - Condition - Concession - Contrast
W
DE
G
Topics in English Linguistics
33
Editors
Bernd Kortmann
Elizabeth Closs Traugott
Mouton de Gruyter
Berlin · New York
Cause - Condition - Concession - Contrast
Cognitive and Discourse Perspectives
Edited by
Elizabeth Couper-Kuhlen
Bernd Kortmann
W
DE Mouton de Gruyter
G Berlin' New York 2000
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.
@ Printed on acid-free paper which falls within the guidelines of the

ANSI to ensure permanence and durability.
Library of Congress Cataloging-in-Publication-Data
Cause, condition, concession, contrast : cognitive and dis-

course perspectives I edited by Elizabeth Couper-Kuhlen,
Bernd Kortmann.
p. cm. - (Topics in English linguistics; 33)
Includes bibliographical references and index.
ISBN 3-11-016690-9 (alk. paper)
1. Grammar, Comparative and general-Clauses. 2. Gram-
mar, Comparative and general-Syntax. 3. Cognitive gram-
mar. 4. Discourse analysis. I. Couper-Kuhlen, Elizabeth.
II. Kortmann, Bernd, 1960- III. Series.
P297.C382000
415-dc21
00-033563
Die Deutsche Bibliothek - Cataloging-in-Publication-Data
Cause - condition - concession - contrast : cognitive and

discourse perspectives I ed. by Elizabeth Couper-Kuhlen ;
Bernd Kortmann. - Berlin ; New York : Mouton de Gruyter,
2000
(Topics in English linguistics; 33)
ISBN 3-11-016690-9
© Copyright 2000 by Walter de Gruyter GmbH & Co. KG, 0-10785 Berlin
All rights reserv~d, including those of translation into foreign languages. No part of this
book may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopy, recording, or any information storage and retrieval
system, without permission in writing from the publisher.
Printing: Werner Hildebrand, Berlin - Binding: Liideritz &. Bauer GmbH, Berlin.
Cover design: Christopher Schneider, Berlin.
Printed in Germany.
Contents
Introduction 1
Elizabeth Couper-Kuhlen and Bernd Kortmann
I. Cause
The relevance of causality 9

Paul Georg Meyer
On the processing of causal relations 35

Leo G. M Noordman and Femke de Blijzer
Domains of use or subjectivity? The distribution of three 57

Dutch causal connectives explained
Henk Pander Maat and Ted Sanders
Causal relations in spoken "discourse: Asyndetic constructions 83

as a means for giving reasons
Christine Gohl
II. Condition
Constructions with if, since, and because: Causality, III

epistemic stance, and clause order
Barbara Dancygier and Eve Sweetser
On affirmative and negative complex conditional connectives 143

Estrella Montolio
Pre- and post-positioning of wenn-clauses In spoken and 173

written German
Peter Auer
Counterfactual reasoning and desirability 205

Noriko McCawley Akatsuka and Susan Strauss
vi Contents
III. Contrast
Adversative connectors on distinct levels of discourse: A 235

re-examination of Eve Sweetser's three-level approach
Ewald Lang
Viewpoints and polysemy: Linking adversative and causal 257

meanings of discourse markers
l~Cott A. Schwenter
The treatment of contrasts in interaction 283

Cecilia E. Ford
IV. Concession
Concessives on different semantic levels: A typological 313

perspective
Mily Crevels
Causal and concessive clauses: Formal and semantic relations 341

Ekkehard Konig and Peter Siemund
Concession implies causality, though in some other space 361

Arie Verhagen
Concessive patterns in conversation 381

Elizabeth Couper-Kuhlen and Sandra A. Thompson
"that's true, although not really, but still": Expressing 411

concession in spoken English
Dagmar Barth
From concessive connector to discourse marker: The use of 439

obwohl in everyday German interaction
Susanne Giinthner
Index 469
Authors' addresses 473

Introduction
The theme of this book-endearingly referred to by its editors and

contributors as the CCCC or four-Cs volume-is a set of relations,
conceptual in nature but instantiated linguistically, which can be said to
hold typically between clauses or sequences of clauses in discourse.
Most, if not all, of our contributors will undoubtedly agree that each of
these relations can be realized or marked by different linguistic means,
e.g. by adverbials, particles, coordinating and subordinating conjunctions,
word order (see, for instance, the chapters by Barth, Dancygier/Sweetser,
Gohl, Montolio and Pander Maat/Sanders, which compare and contrast
different lexical connectives in the fields of causality, conditionality and
concessivity). Some of our contributors will undoubtedly maintain that
the C-relations can even hold in the absence of specific lexical or
syntactic markers (see, for instance, the chapters by Crevels, Gohl and
Meyer). Yet many of our contributors may disagree on whether these
relations are fundamentally semantic (see, e.g., the contribution by
Konig/Siemund) or fundamentally rhetorical/interactional in nature (see,
e.g., the chapters by Barth, Couper-KuhlenfThompson, Gohl and Meyer).
Those who think of the C-relations as semantic will be concerned to
represent their meaning (or the meanings of their markers) in a context-
independent, perhaps even a formal fashion, whereas those who see them
as basically rhetorical or interactive will address their interpretation (or
the interpretation of their markers) in specific-although perhaps
generalizable-contexts. Yet in this divergence-or rather, diversity-of
opinion we see one of the strong points of our endeavor. Indeed, the
novelty of the present volume lies not only in the cutting-edge research
which it presents but also in the fact that it embodies work at the frontier
of two very different approaches to language-cognitive linguistics and
discourse or interactional linguistics. By bringing these two traditions
together in one volume, we hope to initiate a dialogue in which the
respective bodies of work can be evaluated for their relevance to one
another.
The contributions collected here have been grouped roughly into
sections according to C-relation in the order: cause, condition, contrast,
concession. However, since some chapters explicitly address the
relationship between more than one relation, the section boundaries are by
no means rigid. In fact, this permeability is a reflection of deep-lying
2 Elizabeth Couper-Kuhlen and Bernd Kortmann
affinities and wahlverwandtschaften between the C-relations, which make

themselves apparent in both cognition and discourse.
Cause, condition and concession have long been known to have a
special relationship with one another as adverbial relations of
circumstance (cf e.g. Halliday 1985, Thompson/Longacre 1985,
Kortmann 1997). At times they have even been conceptualized in terms
of one another, a conditional relation being seen e.g. as a hypothetical
variant of a causal relation, a concessive relation as an inoperant cause
(Harris 1988, Konig 1986, 1988). The conceptual similarity between
cause, condition and concession is also reflected in the fact that languages
may encode them in the same way (Kortmann 1997). All three relations
lend themselves to expression via syntactic subordination; moreover,
single subordinators may be polysemous between cause, condition or
concession (e.g., if can express both condition and cause, for both cause
and concession; see the chapter by Konig/Siemund in this volume).
Dancygier/Sweetser (this volume) explore how this functional overlap
can be accounted for with respect to the connectives if, since and
because.
The relation of contrast, on the other hand, is typically expressed by
syntactic coordination and not unanimously thought of as an adverbial
circumstantial relation (Halliday 1985, Thompson/Longacre 1985). Yet
especially when it is considered from a pragmatic and/or discourse
perspective, contrast begins to have more in common with concession
(see e.g. Rudolph 1996) and even with cause and certain kinds of
condition. Sweetser (1990), for instance, applies her three-domain model
not only to causal, conditional and concessive sentences but also to
adversative sentences with the coordinator but (see Lang, this volume).
Moreover, as several of the discourse-based papers in this collection
argue, contrast is centrally implicated, e.g., in counterfactual
conditionality (Akatsuka/Strauss) and in concession (Barth, Couper-
Kuhlen/ Thompson). Contrast, specifically adversativity, is furthermore
involved in the discourse-marker use of Spanish si, a canonical
conditional conjunction (Schwenter) and in the discourse-marker use of
German obwohl, a canonical concessive conjunction (Giinthner). Finally,
contrast enters into wahlverwandtschaften with circumstantial C-rela-
tions, in particular with causality, in that justifications recurrently follow
contrasts in discourse, as Ford (this volume) shows. These findings
suggest then that the affinity between contrast on the one hand, and cause,
condition and concession on the other, has been underestimated in the
past. For this reason contrast has been included as a fourth C-relation
here.
Introduction 3
Despite the division of our collected papers into sections, there are-
as might be expected-recurrent themes which cut across the C-relations.
One of the motifs in the cognitively oriented papers, for instance, is
Sweetser's (1990) seminal work on domains or levels of interpretation.
Noordmanlde Blijzer (this volume) adopt her distinction between content
and epistemic level and show how it will help account for differing
degrees of cognitive complexity in causal sentences. Other papers are
more critical of Sweetser's model: Lang (this volume) takes issue with
her claim that content, epistemic and speech-act levels of interpretation
can be determined independently of syntactic structure and, like Crevels
(this volume), fmds it necessary to supplement Sweetser's three domains
with a fourth, textual level. Given a fourth level, Crevels fmds the model
useful for investigating the formal means which typologically different
languages deploy in each domain. Pander Maat/Sanders (this volume), on
the other hand, reject Sweetser's model as a means of accounting for the
distribution and interpretation of specific causal markers in Dutch and
propose in its stead a notion of subjectivity.
Another recurrent motif in several of the chapters collected here is
Fauconnier's theory of Mental Spaces (1985, 1997). Dancygier/Sweetser
(this volume) show that important distinctions can be drawn based on the
ways in which causal and conditional conjunctions participate in the
configuration of mental spaces. Verhagen (this volume) uses a mental-
space model to account, e.g., for the relationship between concession and
(negated) causality, with which Konig/Siemund (this volume) are also
concerned. Yet AkatsukaiStrauss (this volume) are critical of a mental-
space account, because it neglects what they see as an inherent dimension
of counterfactual conditionality, namely that speakers are expressing a
stance towards the events in question as desirable or undesirable.
AkatsukaiStrauss advocate a subjective dimension in the analysis of
counterfactual conditionality and in this sense are in agreement with
Pander Maat/Sanders, who see subjectivity as responsible for the
behavior of specific causal connectives in Dutch discourse. It is perhaps
worth noting that both AkatsukaiStrauss and Pander Maat/Sanders take a
corpus-based approach to C-constructions and for this reason belong
more in the discourse than the cognitive tradition. Significantly, it is in
relying on real data rather than on introspection that these two studies
independently establish a need-above and beyond mental-space or
domain considerations-for taking the speaker's (or subject of conscious-
ness's) stance into account in the analysis of C-constructions, especially
for counterfactual conditionality and causality. In this sense they are
initiating the kind of dialogue between cognitive and discourse
approaches to the C-relations which the editors have envisaged (cf. also
the chapters by Meyer and Konig/Siemund in this respect).
A second notion to be found in both cognitive and discourse-oriented
papers collected here is polyphony (Verhagen), or multiple viewpoints
(Schwenter). For Verhagen, the interpretation of concessive and negated
causal sentences as well as of all sentences involving epistemic C-
relations calls for the construction of two nearly similar mental spaces in
Fauconnier's theory-or, equivalently, for the assumption of polyphony in
Ducrot's sense of the term (1984, 1996). Schwenter also mentions
Ducrot's theory of polyphony as one way of conceptualizing viewpoints
and linguistic structure. He opts, however, for Roulet's (1984) distinction
between one viewpoint (monological) and two viewpoints (dialogical),
both of which can be found with either one speaker or two. Schwenter
finds that dialogicality is not only conceptually present in adversative
constructions but also (to a lesser degree) in 'exhortative' and epistemic
causals (i.e., ones which accompany exhortative speech acts and
inferential conclusions). The notion of polyphony or multiple viewpoints
then appears to be relevant to both cognitive and discourse-based
understandings of the C-relations. And it is clearly implicated in the
interactional model of concession which Couper-KuhlenfThompson (this
volume) propose, where multiple viewpoints often go hand in hand with
multiple speakers-although, as Barth (this volume) shows, they need not
do so.
Related to the dimension of multiple speakers is of course the question
of whether a piece of discourse-specifically, a C-construction-is
spoken or written. This is a recurrent motif in several of the discourse-
oriented papers collected here. Aside from questions of overall frequency
of use (see, for instance, Altenberg 1984, 1986 or Ford 1993), medial
considerations are relevant for C-constructions in at least two other ways.
One of these is with respect to the order of clauses, which Auer's chapter
in this volume investigates with respect to wenn-clauses in German (cf.
English 'when'/'if -clauses). A second issue relating to C-relations in
spoken and written discourse is the question of whether mediality has an
influence on the choice of C-marker. Montolio (this volume) argues that,
as in English, complex conditional constructions in Spanish such as a
condici6n de que ('provided that') and a menos que ('unless') express a
more restricted conditional relation than si 'if and are reserved for formal
written registers. And Barth (this volume) shows that the set of
concessive markers as well as the relative frequency of parataxis as
opposed to hypotaxis is significantly different when spoken English is
compared to written English.
Introduction 5
A further recurrent motif in the papers dealing with spoken discourse

is the use of C-connectives as discourse markers. This is an aspect of C-
relations which has been neglected in cognitively oriented studies. In fact,
the development of discourse markers from adverbial subordinators of
cause, condition, contrast and concession has only recently come to the
attention of discourse and interactional linguists (see e.g. Gohl/Giinthner
1999 for discussion of a causal discourse marker in German, and Lenk
1997 for discussion of contrastive discourse markers in English). In the
present volume it is the chapters by Schwenter on Spanish si and
Giinthner on German obwohl which deal with C-relations as discourse
markers. Whereas the obwohl phenomenon is comparable to paratactic
uses of although in English, the adversative and causal use of si has no
real equivalent with English if. But the evidence in both cases casts doubt
on developmental claims made in the literature to the effect that
conditional markers develop into concessive conditionals and from there
into concessives, while concessives do not develop any further (Konig
1986, 1988, Kortmann 1997). A line of development from conditionality
to adversativity and (paratactic) causality as with si, or from concessivity
to adversativity as with obwohl has gone unnoticed until now. Once
again, it is not insignificant that such insights have come from corpus-
based studies of spoken language. In fact, as Auer (this volume) points
out, a full understanding of spoken-language constructions can only be
achieved through corpus-based investigation. And this is another
recurrent theme in virtually all the discourse-oriented papers collected
here.
The interactional linguistic papers in this volume (Auer, Barth, Ford,
Gohl, Giinthner, Couper-Kuhlenffhompson) go one step further with
respect to corpus-based language study. They take it as axiomatic that
spoken language is first and foremost a tool for social action and that in
order to be understood fully it must be examined in its original habitat,
i.e. in everyday interaction. Viewed from this perspective, the C-relations
can be thought of as ways of carrying out social actions-causality being
instrumental in providing justifications or accounts (Gohl), contrast and
concession in negotiating agreement/disagreement (Barth, Couper-
Kuhlenffhompson, Ford). C-constructions in tum can be conceptualized
as linguistic resources or practices for carrying out the actions in
question, e.g. for conceding a point or correcting a prior claim
(Giinthner). In an interactional perspective, C-constructions are seen as
especially appropriate for or adapted to situated interactional needs and
tasks: a German wenn-clause in post-position-as Auer shows-is not
merely a positional variant of a pre-posed 'one (and thereby freely
interchangeable with it) but a tool 'designed' for or tailored to specific

contexts of usage.
A fmal motif-relevant to both cognitive and discourse-oriented
studies-is the relationship between asyndetic, paratactic and hypotactic
means of expressing C-relations. Although none of our chapters looks at
C-connectives diachronically, the implicational hierarchy for concession
that Crevels (this volume) offers in her typological study, namely Content
> Epistemic > lliocutionary > Textual, in conjunction with her marking
hypotheses has clear implications for language development. She
hypothesizes that the higher the semantic level, the more likely a
concessive relation is to be realized asyndetically. Moreover, if a
concessive relation is realized syndetically, the higher the semantic level
at which it is realized, the more likely it is that q rather than p will be
marked-where q-marking corresponds to parataxis and p-marking to
hypotaxis. The implicational hierarchy above corresponds, of course, to
what are thought to be increasing levels of cognitive complexity (see also
Noordmanlde Blijzer, this volume), whereas the cline asyndesis-
parataxis-hypotaxis corresponds to the unidirectionality hypothesis of
grammaticalization theory (HopperfIraugott 1993). Yet evidence from
spoken language does not fully conform to these hypotheses: asyndesis is
widespread, even for content-domain causality (Gohl, this volume) and
parataxis is more frequent than hypotaxis for the expression of
(epistemic) concessive relations (Barth, this volume). Furthermore, as
Schwenter's and Giinthner's chapters show, discourse markers which
originate from adverbial subordinators create parataxis out of hypotaxis
and thus reverse the unidirectional cline. In order to resolve such
contradictions we anticipate that more dialogue between cognitive and
discourse-oriented approaches to the C-relations will be necessary in the
future.
Meanwhile the reader is invited to appreciate the state-of-the-art
research documented here for cause, condition, contrast and concession.
In line with the title of the series, the present collection focuses on
English; yet it does not eschew the world beyond. Indeed, the
contributions by Noordmanlde Blijzer and Pander Maat/Sanders (Dutch),
by Auer, Gohl, Giinthner and Lang (German), by AkatsukaiStrauss
(Korean, Japanese) and by Montolio and Schwenter (Spanish) serve, each
in their own way, as a valuable corrective on what would otherwise be a
perhaps excessively Anglocentric perspective. In this sense the volume
presents a picture of the C-relations which is enriched by cross-linguistic
research.
Introduction 7
The editors would like to thank Verena Haser, Lieselotte Anderwald

and Manfred Krug for careful comments on earlier versions of the
contributions, Susanne Wagner for compiling the index, and Sabine
Conrad for checking all final versions for conformity to the publisher's
stylesheet. A word of heartfelt thanks also to Melitta Cocan for the
excellent job she has done in preparing the camera-ready copy.
Elizabeth Couper-Kuhlen, Konstanz Bernd Kortmann, Freiburg
References
Altenberg, Bengt
1984 Causal linking in spoken and written English. Studia Linguistica
38: 20-69.
Altenberg, Bengt
1986 Contrastive linking in spoken and written English. In: G. Tottie and
I. Baecklund (eds.), English in Speech and Writing. A Symposium,
13-40. Stockholm: Almqvist & Wiksell.
Ducrot, Oswald
1984 Le dire et Ie dit. Paris: Minuit.
Ducrot, Oswald
1996 Slovenian lectures/Conferences slovenes: Argumentative seman-
tics/Semantique argumentative. Ljubljana, ISH.
Fauconnier, Gilles
1985 Mental Spaces: Roles and strategies. Cambridge, MA: MIT Press.
Fauconnier, Gilles
1997 Cognitive Mappings for Language and Thought. Cambridge: Cam-
bridge University Press.
Ford, Cecilia E.
1993 Grammar in Interaction. Cambridge: Cambridge University Press.
Gohl, Christine and Susanne Gtinthner
1999 Grammatikalisierung von 'wei!' als Diskursmarker in der gespro-
chenen Sprache. Zeitschriftfur Sprachwissenschaft 18: 39-75.
Halliday, M. A. K.
1985 An Introduction to Functional Grammar. London: Edward Arnold.
Harris, Martin
1988 Concessive clauses in English and Romance. In: 1. Haiman and S.
A. Thompson (eds.), Clause Combining in Grammar and Dis-
course, 71-99. Amsterdam: Benjamins.
Hopper, Paul and Elizabeth C. Traugott
1993 Grammaticalization. Cambridge: Cambridge University Press.
Konig, Ekkehard
1986 Conditionals, concessive conditionals and concessives: Areas of
contrast, overlap and neutralization. In: E. C. Traugott, A. ter
Meulen, 1. S. Reilly and C. Ferguson (eds.), On Conditionals, 229-
246. Cambridge: Cambridge University Press.
Konig, Ekkehard
1988 Concessive connectives and concessive sentences: Crosslinguistic
regularities and pragmatic principles. In: 1. A. Hawkins (ed.),
Explaining Language Universals, 145-166. New York: Basil
Blackwell.
Kortmann, Bernd
1997 Adverbial Subordination. A typology and history of adverbial sub-
ordinators based on European languages. Berlin: Mouton de
Gruyter.
Lenk, Uta
1997 Marking Discourse Coherence. Functions of discourse markers in
spoken English. Tiibingen: Narr.
Roulet, Eddy
1984 Speech acts, discourse structure, and pragmatic connectives.
Journal ofPragmatics 8: 31-47.
Rudolph, Elizabeth
1996 Contrast: Adversative and concessive relations and their
expressions in English, German, Spanish, Portuguese on sentence
and text level. Berlin: de Gruyter.
Sweetser, Eve
1989 From Etymology to Pragmatics. Metaphorical and cultural aspects
ofsemantic structure. Cambridge: Cambridge University Press.
Thompson, Sandra A. and Robert E. Longacre
1985 Adverbial clauses. In: T. Shopen (ed.), Language Typology and
Syntactic Description, Vol. 2 Complex constructions, 171-234.
Cambridge: Cambridge University Press.
The relevance of causality-
Paul Georg Meyer
"In der Natur gibt es keine Ursache und

keine Wirkung". (Mach 41901: 513)
"The law of causality, I believe, like
much that passes muster among philoso-
phers, is a relic of a bygone age, surviv-
ing, like the monarchy, only because it is
erroneously supposed to do no harm."
(Russell [1912] 1921: 180)
"The most fundamental fact of the
world" (KamIah 1991, title)
"The cement of the universe" (Mackie
1974, title)
This article discusses interrelations between relevance and causality from

different angles: first, it is shown that the very notion of causality is subject
to a relevance constraint. In communicating about causes people tend to
seek out one or at most a limited number of possible causes from a possibly
infinite number of logically admissible conditions. The criterion for select-
ing this limited number is obviously relevance. This interrelation between
causality and relevance is analysed by way of a critical application of rele-
vance theory, with the result that the conditions of relevance of causal
statements are more similar to felicity conditions of speech acts than to a
cognitive relevance principle.
Second, causality is described as a major device for creating relevance in
discourse. The relevance of causality in discourse cannot easily be over-
stated. Nevertheless, causality is only one among several principles of text
organisation, though obviously the most relevant one.
Causality is an "elusive" concept (Ziv 1988: 543). We all seem to know

a causal relation when we see one, but as soon as we try to explain why
it is a causal relation, or to communicate what it is precisely that we
have seen, we run into problems. Like time and space, causality is taken
for granted by everybody and fully understood, so it seems, by nobody.
And, what is most disconcerting to scientifically-minded people,
"knowledge of causes is possible without a satisfactory grasp of what is
involved in causation" (Anscombe [1971] 1975: 67). Scientists have
10 Paul Georg Meyer
consequently sought to eschew the notion in their considerations alto-

gether (see mottoes). Philosophers, on the other hand, partly drawing on
scientific results (Reichenbach 1991), have stressed its crucial impor-
tance for the functioning of the world (see mottoes).
Almost a hundred years after Mach's and Russell's verdicts, philoso-
phers, psychologists, social and cognitive anthropologists, and linguists
are still wondering about causality in their different ways. The reason
why the "law of causality", or at least the notion of causality, is still
going strong, is independent of its status in nature or science. Causality
is simply an indispensable principle in the organisation of people's eve-
ryday lives, both on the individual and the micro-group level, and, a
fortiori, also on the level of societies, states, and international relations.
When I clean the blades of my electric lawn-mower with my fingers,
I rely on the law of causality, which tells me that once the plug is pulled
from the socket, the engine cannot start and make the blades rotate
again. When I am late for a meeting, I will try to find a cause l for my
being late, and include it in the account that will presumably be ex-
pected from me; and my account will presumably be much more accept-
able if I am able to give a satisfactory explanation for my being late.
When people claim money from an insurance company, their claim pre-
supposes the law of causality, which means that there was a cause for
the damage and depending on what the cause was, the insurance com-
pany will or will not pay. When the UN sets up a tribunal to indict and
convict war criminals, the legitimacy of the procedures hinges on the
law of causality, on the belief that certain people are responsible for
certain crimes, that is, that their actions were causally involved in these
crimes in a crucial way.
The law of causality has not been allowed to survive because people
believed it could "do no harm" (Russell [1912] 41921: 180, cf motto). It
has been allowed to survive because without a notion of causality, most
of our material-technical, micro- and macro-social, economic, legal, and
international life and institutions would not make sense and would break
down. 2
It will be shown presently that these remarks have more meaning in
the context of this article than the usual conventional introductory asser-
tions about the outstanding importance of an article's subject matter
(Swales 1981). The first reason for this is that the very topic of this arti-
cle is the relevance of causality. Second, the examples adduced above to
show the relevance of causality in an intuitive sense also show that, oc-
casional scientific necrologues on the concept notwithstanding, causal-
The relevance ofcausality 11
ity seems to have its locus in the social sciences and humanities. Every-
day action, group cohesion on all levels, economic transactions of all
kinds, notions of obligation and responsibility, law as the formalisation
of the latter notions, and many other topics of social sciences and hu-
manities crucially depend on causality. We may thus take the above
examples as a first approximation to a concept of causality that is lo-
cated in a social rather than in a physical world. In other words, the uni-
verse of which causality is the cement, the world in which causality is
the most fundamental fact, is primarily a socially constructed universe, a
socially constituted world.
This article does not undertake to solve the problem of causality. Its
objective is much more modest. Based on observations made in the phi-
losophical, scientific, psychological, social scientific, and linguistic lit-
erature, I will defend a set of linguistically inspired theses concerning
the relationship between causality and relevance (cf Meyer 1983: 125-
126), putting it into the context of the present discussion and of an
emerging research programme.
I will begin with a (very) brief history of the concept of causality
(Section 1). I will then show how the history of the concept, the way it
has been discussed in the literature, and the problems it has raised, seem
to indicate a certain constraint on the notion of cause that is best de-
scribed by a notion of relevance (Section 2). Next I will ask whether this
notion of relevance is explicable in terms of the presently most influen-
tial pragmatic theory of relevance (Sperber and Wilson 1995 2) (Section
3), and discuss problems arising from such an attempt. Finally, I will
argue for a basically discursive explication of the notion of causality in
the framework of a coherence relations approach, establishing causality
as one of several primitive universal principles that govern the construc-
tion of coherence in discourse (Section 4).
1. A brief history of causality
Talking of causality under a unitary label misleadingly suggests that the

label covers a unitary, homogeneous phenomenon. Breul (1997: 81-112)
has recently summarised the develoJ?ment of the notion of causality
from Aristotle via Hume into the 20 century. Among Aristotle's four
different kinds of causality (Metaphysics 5, 2, 1013a24ff), there is only
one'· (known as causa efficiens from the Latin translation) that is clearly
identifiable with a modern common-sense notion of causality. In the
12 Paul Georg Meyer
scientific and philosophical tradition of post-Renaissance Europe (e.g.

Hume), another notion of causality is brought forward, emphasising the
necessity and lawlikeness of the causal relation. Finally, the notion of
causality in that tradition dissolved into the mathematical notion of
function, as exemplified in Mach (1901) and Russell ([1912] 1921).
This notion is certainly far removed from everyday intuitive concepts of
causality. Strawson (1985) argued against a law-based notion of causal-
ity to cover everyday cases of causation:
A man, say, falls down a flight of stone steps as he begins the descent. The
steps are slippery and the man's mind is elsewhere. This is a sufficient ex-
planation of his fall. But of course not every preoccupied man falls down
every flight of slippery steps he descends. There is absolutely no question of
our formulating or envisaging exceptionless laws, ... , to cover all such
cases. (Strawson 1985: 131)
That is, all rational and enlightened people will assume that the
man's falling down the stairs is fully in accordance with some presumed
laws of nature and, in principle, fully explainable in such terms. And
yet, in ordinary language we do not expect such full "scientific" ac-
counts to be given, and indeed it is highly doubtful whether such full
causal accounts of singular events are useful, desirable or, for that mat-
ter, feasible.
For all practical purposes, we are content with much less than a list-
ing of all the conditions, regularities, probabilities, and laws of nature
that account for the event in question. A satisfactory explanation, in
fact, is not one that lists all this. A satisfactory explanation is concise
and concentrates on one or just a few causes. The problem that scientists
and philosophers have had with the notion of causality emerging from
this observation is that they cannot fmd a unique logical or mechanical
or otherwise physical characterisation of a state of affairs that happens
to be intuitively singled out as the "cause" of a certain event. Whatever
is named as "the cause" of something, does not seem to differ in any
significant way from other factors present in the situation in question.
There simply seems to be no logical or scientific basis for calling some-
thing a cause.
Logical analyses of the notion of cause have operated with necessary
and sufficient conditions. The two intuitively most plausible and (pre-
sumably for that reason) most influential defmitions have been Mackie's
(1965) "INUS condition" and the "counterfactual conditional", origi-
nally suggested by Hume and reintroduced into the modern debate by
David Lewis (1973). As we shall see in explicating what an INUS con-

dition is, the two are actually quite close to each other.
According to Mackie (1965) causes are best defined as "INUS" con-
ditions. An INUS condition is "an insufficient but necessary part of a
condition which itself is unnecessary but sufficient for the result. "
(Mackie 1965: 16).
Thus, to take Strawson's example quoted above, both the slippery
state of the stairs and the man's absentmindedness are causes of the
man's fall because they are
• insufficient: Slippery stairs or absentmindedness alone do not cause a
man to fall down a staircase, and would not have done so in that par-
ticular case.
• necessary: Had the man been more careful or had the stairs not been
slippery, the man would not have fallen.
• part: The two causes mentioned are only part of the set of conditions
causally involved in the man's fall. Thus, e.g., the man's trying to
descend the stairs in the first place is another condition on the man's
fall (that could also be called a cause on that account because it is an
insufficient but necessary part of the set of conditions leading to the
accident).
The set of such conditions taken as a whole is
• unnecessary for the result because the same effect might also have
been achieved by other causes, in an altogether different set of condi-
tions: Thus, e.g., the man might have fallen down the stairs because
somebody maliciously pushed him, or because he was drunk and the
steps were worn.
• sufficient (in that particular case) for the result because, after all, the
result was brought about.
According to Hume and David Lewis (1973) c is the cause of an
event e if and only if the counterfactual statement: "If c had not been
true, e would not have happened" is true. In other words, a cause is a
'conditio sine qua non'. The attentive reader will have noticed that in
explicating the notion of INUS condition above, we repeatedly had to
resort to counterfactual formulations: if the man had been more careful
or if the steps had not been slippery, he would not have fallen. Absent-
mindedness and the slippery state of the steps alone would not have
brought about the accident. The same accident could have been brought
about by altogether different causes, etc. It is thus clear that the INUS
definition of causality is only a more formalised and elaborate variant of
the counterfactual formulation of that definition.
14 Paul Georg Meyer
The problems with both these defmitions are manifold (Sosa 1975: 3-
4), but it is not my purpose to discuss them here. There is just one point
that I want to make: Mackie refers to a set of conditions that is unneces-
sary, but sufficient for the effect. The factor singled out as "the" cause is
not the whole set of conditions, but only an insufficient but necessary
part of this set. The problem that this definition raises is that it is not
very useful for an application of the concept of causality in the analysis
of everyday communication. The definition covers a number of condi-
tions that we would not call causes in a given situation. That is, Mackie
describes a necessary but not sufficient condition for calling an event a
cause. He describes a semantic precondition for a proposition to serve as
a causal explanation. Real events always have a large number of neces-
sary and sufficient conditions. Arguing that one circumstance was the
cause of a given event means foregrounding one insufficient and neces-
sary condition and stating it, and allocating all the other conditions that
jointly were unnecessary but sufficient to bring about the effect to the
background. Thus, the very problem of Mackie's definition is an indica-
tion of the basically pragmatic character of any reasonable notion of
causality that would be applicable in the analysis of discourse.
It is a situation that logical positivists and context-free semanticists
cannot easily cope with, but it is not unfamiliar to pragmatically ori-
ented linguists: reality is always understated and underdetermined by
the propositional content of utterances. The meaning of utterances al-
ways goes beyond what was said. There is always a set of meanings that
were meant but left unsaid. And there is always a background of knowl-
edge that is taken for granted, that is not problematised, that is simply
not at issue. If we believe Searle (1983: 141), this background contains
infinitely many propositions. And indeed, it is precisely this characteris-
tic of the set of sufficient conditions for a certain event that has led early
modem philosophers (such as Hume) to confine the notion of causality
to "lawful" instances and has driven 19th and 20th century scientists and
philosophers to despair.
2. What makes a cause a cause?
Let me try to illustrate this with Strawson's example (in fact, any exam-
ple of a causal relation would do). What is the set of conditions that is
jointly sufficient (though unnecessary) to bring about "the man's" fall
(for simplicity's sake let's call him Jones)? The two conditions men-
tioned by Strawson are in the foreground, of course: the steps were slip-
pery, and Jones was absent-minded. Apart from these, I have already
mentioned one further condition: the fact that Jones did try to descend
the stairs at the very time when the first two conditions were already
given. It seems that these three conditions, each of them independent of
the two others, were jointly sufficient for Jones' fall. But, as Strawson
(1985: 131) pointed out, "not every preoccupied man falls down every
flight of slippery steps he descends." So there must have been further
conditions that were necessary to make the set of conditions sufficient
for Jones' fall to occur: maybe if Jones had been wearing different shoes
(say, with non-slip rubber soles), he would not have fallen. Or, if there
had been a railing for him to hold on to, he would not have fallen. Or, if
his ankles had not been so stiff that day, he would not have fallen. Or,
perhaps, if he hadn't been drinking, in addition to his absentmindedness.
Or ...
On the other hand, we might also begin to question the most obvious
and plausible causes that were given: granted that the steps were slip-
pery, were they slippery enough for a man to slip? Granted that Jones
was absent-minded that day, was he absent-minded enough not to notice
the slippery state of the steps? That is, even the most readily accepted
explanations for an event are not undebatable. There is always a possi-
bility that the most plausible explanation is erroneous. Suppose that
Jones is absent-minded most of the time and the steps are always slip-
pery and he walks down these steps several times a day without slip-
ping. It is obvious that in this case we would still accept the explana-
tions given by Strawson to a certain extent, although we would have
evidence for the fact that the explanation is not sufficient. Even though
we would be willing to grant that there must have been other "causes"
for Jones' fall, the explanations "slippery steps" and "Jones' absent-
mindedness" are just too good to be discarded.
What this discussion is supposed to show is: as scientists and phi-
losophers found out long ago, there is no such thing as "the" cause or
"set of causes" of an event. Causality emerges as a discursive phenome-
non. Note that all the "causes" mentioned above could be adduced in
some kind of dispute over the "causes" of Jones' fall. But it depends
precisely on the kind of dispute as to which of the causes will be ad-
duced by whom.
Suppose Jones' wife had always been of the opinion that that stair-
case is dangerous and had always been warning Jones to use it (suppos-
ing there was an alternative available). In this case, the cause of Jones'
16 Paul Georg Meyer
fall that would be accepted by his wife would be the fact that Jones de-
scended the stairs in the first place. ("Did you have to use that danger-
ous flight of stairs? I told you one day you'd break your neck on it".) Or
that Jones descended the stairs while they were slippery ("Why couldn't
you wait until the steps had dried?") (Supposing, e.g., that they were
slippery because Jones' wife had just scrubbed them).3 For his wife, in
that case, the slippery state of the stairs would be taken for granted, just
like her husband's absentmindedness, which she has learnt to put up
with in long years of marriage.
Now take the case of an insurance company. Depending, of course,
on what kind of risk had been the object of the insurance, all kinds of
questions might be raised. Who was responsible for the slippery state of
the steps? Why did the flight of stairs have no railing? Was Mr. Jones'
absent-mindedness due to alcohol or drugs? What was the purpose of
his descent down the stairs? Was it on his way to work, for example?
Was Mr. Jones' stiflhess in the ankles due to an accident or some ill-
ness? Note that the insurance company's questions open up a much
wider background by considering what might provisionally be called
second- or third-order causes. The reason for this is that an insurance
company will tend to follow up causal chains until somebody is found
who can be blamed for an accident. Thus, for an insurance company or a
lawyer, the cause of Mr. Jones' fall might be the fact that the owner of
the house failed to replace a broken railing or to provide some kind of
slip protection measures for the dangerous stone steps, or the fact that
Mr. Jones' doctor prescribed some drug for him that affected his capac-
ity to keep his balance to a sufficient extent to cause, in connection with
the slippery state of the steps and Jones' absentmindedness, his fall
down the stairs.
That is, when considering the set of conditions jointly sufficient to
bring about an event, it is not sufficient to look at "first-order" causes
only. In any case, it is a question of focus and perspective what is re-
garded as "first-order cause". Take Mr. Jones' absentmindedness as an
example. Strawson said it "caused" Mr. Jones' fall. But looking more
closely at the actual event, we may be certain that innumerable other
"causes" intervened between Mr. Jones' absentmindedness and his fall.
Absentmindedness cannot immediately "cause" a fall. Before the fall
there must have been a slip. And it is difficult to ascertain at which point
in the causal chain leading to Mr. Jones' fall absentmindedness is sup-
posed to come in. Did absentmindedness cause him to slip, and the slip
cause him to fall? Or did he slip first, and did absentmindedness prevent
him from keeping his balance after the slip? Or did he slip and lose his
balance inevitably, and did absentmindedness prevent him from holding
on to something, which might have prevented his fall? Or was absent-
mindedness a limiting factor of all of Mr. Jones' movements that day,
and did it have disastrous effects all through the causal chain leading to
the accident?
Even if this was a real case, I doubt that all this detail could be ascer-
tained. It should be clear, however, that a distinction between first-order
and second- or third-order causes is impossible and has no systematic
status. What we regard as a cause is seldom the "immediate" cause (in
the sense that no intervening event between cause and effect can be
found) because our observations are seldom so fine-grained that "imme-
diate" causes would be noticed. Let us suppose that theoretically we
come to the conclusion that some basic muscular movement of Mr.
Jones immediately prior to his fall must have been the "immediate"
cause. Presumably neither Mr. Jones nor a by-standing observer would
have been able to observe and identify this "immediate" cause. Pre-
sumably this "immediate" cause was the inevitable result of some prior
events, conditions, and maybe of Mr. Jones' decisions, and these will
most likely be singled out as "the cause". We tend to look at events and
causal chains not from a mechanistic perspective, but from a perspective
that allows us to see and represent points of potential intervention (cf
Von Wright 1973). We look for conditions that could have been differ-
ent. And usually conditions can be different if people behave differently
or change conditions deliberately. Mr. Jones might not have used that
dangerous staircase at all that day. Mr. Jones could have been more
careful. The steps might not have been slippery (somebody could have
done something about that). So the insurance company, by opening up
an infinite space of conditions jointly sufficient for the accident, at the
same time limits the search to those states of affairs that are accessible
to human perception and intervention.
Let us now look at Mr. Jones himself How would he explain his fall
(supposing he survived it)? Again it would depend on a number of con-
tingent factors what would be a satisfactory explanation of the accident
for Mr. Jones himself So Mr. Jones might (most likely) blame the slip-
pery stairs, but it should be clear that he can only do so under certain
conditions. Suppose that Mr. Jones has been using these stairs for years,
they have always been slippery, he has never had an accident and has
never complained about the state of the steps. In that case it would be
somewhat difficult for Mr. Jones to blame the slippery steps. If he chose
18 Paul Georg Meyer
to do so, it would mean a "change of policy", so to speak. After years of

putting up with the slippery stairs, he would suddenly blame them for an
accident. Under a benevolent interpretation, this would mean his fall has
taught him a lesson, and he probably will have to do something about
the slipperiness of the stairs in the near future. Mr. Jones could also
blame his own absentmindedness, if he is willing to make such a face-
damaging admission. What is more likely (assuming Mr. Jones to be
vain like most people) is that he will blame some other condition that
does not figure in Strawson's account at all. He could resort to some
other explanation that would be more face-saving. "I was a bit stiff in
the ankles that moming" would do perfectly.
What this lengthy exposition of potential explanations for a fictitious
accident is meant to show is that explanations are not objective, based
on facts existing independently. In fact they are guided and informed by
people's interests. What explanations we are willing to give and willing
to accept as a "cause" is not so much a matter of fact, but in the best
case open to negotiation. Establishing a cause of an event is not so much
a fact-finding procedure, but rather a process of social interaction in
which diverging interests have to accommodated, from which different
consequences (even obligations) may arise (cf Hilton and Erb 1996:
303). It will thus be appropriate to look at the notion of causal explana-
tion in order to clarify the notion of cause. This will be done in the next
section.
It should, however, be emphasised that the above account does not
imply that "the truth" is at the mercy of subjective interests. All the ex-
planations mentioned above that could be adduced for Mr. Jones' fall
might be, strictly speaking, true. Truth is a semantic concept, and se-
mantically, all the conditions mentioned above as possible causes of Mr.
Jones' accident are INUS conditions: they are insufficient but necessary
parts of the whole set of conditions leading to the accident. They differ
not in terms of truthfulness but in terms of relevance in certain discur-
sive contexts and for certain people.
3. The relevance of causality
What does it mean for an explanation to be relevant in a given context?

The notion of relevance plays a crucial role in several theories of prag-
matics (Grice 1975; Hom 1984; Sperber and Wilson 1986, 1995). But
before I relate the previous discussion to these theories, I would like to
explicate in my own terms what I mean by "relevance" here. What

makes an explanation relevant? A good explanation raises an issue that
was either
- outside the "normal course of events", unpredictable, and out of con-
trol. This explains the occasional meteorite examples in the literature
(Stegmuller [1969] 1983: 911). Or it was
- within control of people, that is, something that could have been
otherwise, a condition that was amenable to change, but still to a
certain extent unexpected and outside the normal course of events.
So when the slippery steps are adduced as an explanation of Jones'
fall, it is implicated that perhaps they should not have been slippery, that
something should be done to prevent people from slipping there. If
someone takes the slippery stairs for granted as a fact of life, then Jones'
absentmindedness or his using the steps in the first place will be fore-
grounded.
Conditions that are absolutely unquestionable will never be men-
tioned as causes. Thus, the fact that Mr. Jones' body has weight, or, in
particular, that its specific weight is higher than the specific weight of
the atmosphere above the stairs and accordingly obeys the law of grav-
ity, would certainly not be accepted as an explanation for Jones' fall
except in circumstances (e.g., science fiction) where weight and gravity
are not to be taken for granted. Neither would the fact that Mr. Jones
was born be accepted as an explanation for his fall, although his birth is
an INUS condition on his fall, and if he hadn't been born, he couldn't
have fallen.
The closest equivalent to this informal notion of relevance for causal
explanations that I could fmd does not come from Gricean and neo-
Gricean pragmatics, but from speech-act theory. This should not be too
surprising if we accept that explanation or explaining is a speech act in
its own right. Like all illocutionary acts, it could have among its felicity
conditions a formulation reminiscent of Searle's classical formulation
(Searle 1969: 59)4
(1) Afelicitous explanation will adduce as a cause for a given event E

a condition C that is an INUS condition ofE in Mackie's (1965)
sense, where C is a proposition that would not have been true in
the normal course ofevents, or the truth of C is not obvious to the
hearer H.
20 Paul Georg Meyer
This formulation, as in all cases where Searle uses formulations of this

kind, has its problems, however. In particular, it is difficult to determine
what a "normal course of events" would be in each case, and what pre-
cisely is "obvious" to the hearer.
Our discussion above has shown something else. There is, strictly
speaking, no such thing as a "normal course of events"; and it is not
obvious what could be called "obvious to speaker or hearer"; but there
is, for each context of situation, a background (Searle 1983: 141).
Propositions which belong to the background are regarded as unques-
tionable, presupposed, imputed and taken for granted. The background
is never thematised except when communication breaks down or illocu-
tionary acts fail. What is part of the background in this sense cannot
felicitously be invoked as an explanation (like in most everyday con-
texts the fact that bodies have weight). We may thus stipulate:
(2) A felicitous explanation X will adduce as a cause for a given event

E a condition C that is an [NUS condition in Mackie's (1965)
sense and that is not part of the background (in the sense of
Searle 1983) for the description ofE orfor X
This formulation, I think, is quite analogous to what should replace the

corresponding felicity conditions in Searlean descriptions of illocution-
ary acts. A background fact cannot be felicitously made the content of
an assertion; it would be a typical case of a trivial assertion. Since ex-
planations are assertions, the above formulation provides only a special
case of such a felicity condition. Nor can background facts be made the
content of a directive: when you order a table in a restaurant, you don't
say you want chairs around it as well (unless it is a type of restaurant
where people usually sit or lie on cushions on the floor). And making
part of the background the content of a promise ("A happily married
man who promises his wife he will not desert her in the next week") "is
likely to provide more anxiety than comfort", as Searle (1969: 59) re-
marked.
Now this classical Searlean felicity condition, invoking obviousness
and the normal course of events, on most types of illocutionary acts, has
usually been equated with Grice's maxim of Relation ("Be relevant",
1975: 46-47), Horn's R-principle (1984) and Sperber and Wilson's
principle of RelevanceS (1986, 1995). It is doubtful to me whether this is
really the case. True, the Searlean felicity condition is approximately
consistent with colloquial meanings of the term relevance. But Grice's
maxim is a maxim of Relation, not of relevance (although its summary

uses the term). Grice's maxim of Relation has to do with the require-
ment that utterances should be related to the previous discourse and the
context of situation. In this way, it guarantees coherence of discourse.
But this is not what Searle had in mind when he referred to "the normal
course of events".
Hom's R-principle (1984), however, comes closer to the Searlean
notion because it incorporates half of the Gricean Quantity maxim as
well. In Gricean terms, utterly trivial statements or explanations violate
the Quantity maxim in that they give too little information. In Hom's
conception, this submaxim is combined with Grice's Relation maxim to
yield a complex R-principle, constituting a lower bound on how much
information should be given in an utterance.
Sperber and Wilson's (1986) conception of Relevance, although it
seems no more than a radicalisation of Hom's R versus Q duality, con-
stitutes yet another concept of 'relevance' because it includes Q- and R-
based considerations from the Hornian model in one formula. Relevance
is no longer one of several principles, but the only principle guiding
ostensive-inferential communication (Sperber and Wilson 1986: 155-
158). The R versus Q duality is converted into a quotient of "cognitive
effects" divided by processing effort. This means (counter to everyday
intuitions about relevance) that high processing costs of a message
(measurable in length, complexity, difficulty, amount of noise, etc.)
reduce the Relevance of that message.
Applying this conception to explanations, it is true that lengthy,
complicated, difficult explanations given in noisy surroundings or
through a fragile channel are certainly less acceptable to hearers. But
they may still be good, felicitous, relevant explanations in terms of the
above thesis (2).
This is not meant as an objection to Sperber and Wilson's theory-it
just points out a difference in conception. But there is another problem
that arises in connection with Sperber and Wilson's Relevance theory
when equating relevance as discussed here with their Relevance: The
Relevance of an event, Sperber and Wilson say (1986: 156-157), corre-
lates positively with the effect of that event on a given context of situa-
tion. What, now, is an effect? Doesn't this notion presuppose a notion of
causality? The event whose Relevance is to be measured is seen as a
cause having some effect on a situation. An event's Relevance is gauged
in terms of the relative effect of that event, seen as a cause. And it is a
cause that could be adduced in a felicitous causal explanation as defined
22 Paul Georg Meyer
in (2) above. It is a prerequisite for successful communication that an

utterance is not just an INUS condition for its contextual effect. It is the
very issue discussed by Sperber and Wilson that utterances should be
relevant causes of their contextual effects. If other INUS conditions of
that same effect (such as the Hearer's knowledge of the language used,
absence of noise drowning out the utterance, etc.) were more relevant in
a given situation, then the utterance itself would lose its relevance. The
problem is: combining Relevance theory and the relevance-based notion
of causality put forward here leads us into a vicious circle: we explicate
causality in terms of relevance only to find that relevance is explicated
in terms of causality!
I can presently see two ways out of this dilemma, without being sure
which of the two is to be preferred, and without being sure whether they
are actually mutually exclusive:
The first solution is less clear to me in all its detailed theoretical
implications, but there is some evidence in the literature in favour of it.
This solution would be to declare causality a semantic and cognitive
primitive (cf Ziv 1993: 21). This idea is actually an old one in linguis-
tics, first put forward formally, as far as I know, by postulating an
"atomic" semantic predicate CAUSE in generative semantics (McCaw-
ley 1968). The "linker" because is also contained in Wierzbicka's
(1991: 8, 1996: 70) lists of semantic primitives; though not in the origi-
nal one (1972: 16).
Lakoff's (1982: 163) idealised cognitive model (ICM) of causation
adds an interesting aspect to the discussion. Lakoff rejects the notion of
primitive in the traditional sense (1987: 279-280),6 replacing it by his
own notions of basic-level concept and image schema as the nearest
equivalents. Causation is obviously a basic-level concept in Lakoff's
terms. The interesting point is that basic level concepts have some char-
acteristics of traditional primitives in that they are "not put together by
fully productive principles of semantic composition" (ibd.). But they do
or may have internal structure. In this they differ from traditional primi-
tives. This internal structure is spelled out by Lakoff in idealised cogni-
tive models. The elements of Lakoff's model of causation do not men-
tion lawfulness or necessary and sufficient conditions at all. Similar to
Von Wright (1973), Lakoff regards agency and the transfer of energy as
essential features of prototypical causation. It should be noted that it is a
model of causation, not of causality, causation being understood as the
nominalisation of an agentive verb. It is plausible that prototypical cau-
sation is agentive; and it is quite conceivable that the abstract principle
of causality as used in discursive reasoning should be derived from

agentive causation. In this case, the cognitive experience of causality as
an abstract principle would be regarded as a notion that is gained by
abstraction from prototypical agentive causation.
There is also some justification to be found for the primitive analysis,
or rather non-analysis, of causality in the philosophical literature:
Anscombe ([1971] 1975) seems to advocate a view, summarised by
Sosa (1975: 4), that "causation is what it is and nothing else-and that
there is no analysis of causation that essentially involves conditionality
or lawfulness". And Sosa continues: "So far as I know no one has pub-
lished a successful analysis of causation by reference to conditionality
or lawfulness" (1975: 5). Note that Sosa, too, speaks of "causation"
rather than of"causality".
If it is true that causality is a primitive concept or very close to a
primitive concept of causation, we encounter a familiar phenomenon
that is often observable when we try to analyse a primitive concept.
Take the example of see. In accordance with Wierzbicka (1996: 78-79) I
would argue that it is a primitive concept. An obvious classical defini-
tion of see along the lines of genus proximum and differentia specifica
would of course be possible in terms of "visual perception" (Searle
1983: 61; Alm-Arvius 1993: 17). I argued in Meyer (1997: 121-122)
that visual, and presumably perception, too, cannot be defmed without
reference to the primitive concept see. It might just be that "analysing"
causality in terms of necessary and sufficient conditions and in terms of
relevance or lawfulness is a similar fallacy to "analysing" see in terms
of "visual perception".
It is quite conceivable that necessary and sufficient conditions are
conceptually secondary to causality; material implication as defmed by
means of truth tables in logic is an abstract notion far removed from the
conceptual lives of ordinary people. It is at least as far removed from
everyday concepts of if-then as an INUS condition is from everyday
causality. And, as we have seen above, relevance is possibly a notion
that cannot do without causality because it has to do with causal impact.
As this solution is compatible with Sperber and Wilson's account of
relevance and communication, and as Sperber and Wilson's account is
grounded in cognitive categories, it would mean that presumably causal-
ity is the cement of a cognitively appropriated world, on which both
perception and communication are based.
The other solution would be to develop a "causality-free" conception
of relevance. Maybe the formulation in (2) above is a step in that direc-
24 Paul Georg Meyer
tion. It would be a step "back" towards Grice, Hom, and Searle, away
from Sperber and Wilson. It would mean that relevance, and, in conse-
quence, causality, are categories of communication and interaction, not
of cognition (Edwards and Potter 1993: 24).
For the time being, I am inclined to pursue the second line of reason-
ing, interpreting causality as a discursive-interactional rather than a
cognitive phenomenon. There are two reasons for this: first, this line is
more in accordance with my own previous research on discourse
(Meyer 1975, 1983, 1996) and the role of causality in it. Second, it
seems to be the tum that psychological attribution research is taking at
the moment (Edwards and Potter 1993), so that further linguistic re-
search in that direction opens up promising channels for an interdisci-
plinary exchange of ideas.
As a textlinguist who has been workin~ in a coherence relations
framework (though under a different name), I have always been con-
vinced that causality has a major role to play in the explication of text
constitution and text coherence. But until recently I had tended to be-
lieve that causality was a principle "out there", in the world, being re-
ferred to and being made use of in discourse.
To see research in science, logic, psychology and social sciences
converge on the idea that causality is primarily located in discourse does
not come as a complete surprise, though, given the discursive-
constructionist tum in social sciences in general (Woolgar 1988; Latour
1987; Bazerman 1988) which has left none of the sacred traditional no-
tions of epistemology untouched. Expressions such as fact, truth, nature
or world cannot be used naIvely anymore after taking notice of that dis-
cussion (in which they invariably tum up in "scare quotes", and are ridi-
culed more often than seriously discussed).
Nevertheless, this "discursive turn" of the causality discussion sets
new tasks for discourse analysts as well. We cannot shift responsibility
for the notion of causality to scientists, psychologists, or philosophers
any more, but find ourselves as the branch of knowledge most immedi-
ately competent for a characterisation of that notion. While I remain
sceptical concerning social constructionism in general (Meyer 1997: 67-
74), I have allowed myself to be convinced by the overwhelming evi-
dence in the case of causality. It remains an open question, though,
which is the more appropriate approach, a cognitive or a discursive one.
My decision for a discursive approach is preliminary and inspired by
research-practical considerations.
4. Causality in discourse
So far I have considered the notion of causality as a concept external to

linguistics and found it subjected to relevance constraints. This led me
to adopt the perspective of causality as a discourse phenomenon in a
much more radical way than I had hitherto done.
Let me now start from the other end and see what can be said from
inside linguistics about interrelations between causality and relevance,
in particular from a textlinguistic perspective. Seen from this perspec-
tive, causality is not only subjected to relevance constraints, but also
creates relevance.
This can best be seen from the numerous examples, discussed con-
troversially in the literature, which show that asyndetic clause connec-
tions are often interpreted as causal connections, be they juxtaposing as
in (3) and (4), coordinating as in (5), or subordinating as the participle
construction in (6) (the examples are taken from a popularised social
science text):8
(3) People badly want a demonstration, a dramatisation of justice.

They need to have defined and reinforced for them what is right
and what is wrong.
(4) Penal methods by themselves will not put an end to crime. Even at
their fiercest, as in nineteenth-century England, ( ..) they did not
succeed.
(5) The participants remain anonymous and there is no one upon

whom the authorities can pin the offence.
(6) The <accusatorial> system is more adaptable than the inquisito-

rial one, allowing for practical changes in treatment methods to
be introduced easily during the sentence.
It seems that causality is such a basic principle that very often there
is no need to draw special attention to it, and the causal relation between
the two clauses is a matter of implicature. Even the direction of causal-
ity is left to the reader to infer in such cases. While in (3) to (6), the sec-
ond clause gives an explanation of some kind relating to the first clause,
the causal relationship is reversed in (7). The second clause states a con-
sequence of the fact reported in the first clause:
26 Paul Georg Meyer
(7) In all the countries concerned the population is growing quickly.

All will be faced sooner or later with the spectre of overpopula-
tion.
The precise character of the explanations in (3) to (6) has also been the
subject of much debate, but is of little concern for the present contribu-
tion (see Dancygier and Sweetser; Pander Maat and Sanders [this vol-
ume]).
Kortmann (1991: 118-121) has suggested a scale of informativeness
for adverbial relations. Causally based adverbial relations (concession,
condition, instrument, and cause) are to be found above a dividing line
in the middle of Kortmann's scale which separates more informative
from less informative adverbial relations, contrast being the only non-
causal relation figuring in the top group. Kortmann uses this scale in
explaining the semantic interpretation of free adjuncts and absolutes,
i.e., participle constructions in English. It could also be used for a rank-
ing of coherence relations. It is obvious, e.g., that a causal relation is
more informative than a temporal one because a causal relation implies
a temporal relation and only adds further information to this.
A standard analysis of examples (3) to (7) within the coherence rela-
tions approach would be to find specific coherence relations such as
"explanation" or "justification" to describe the relation between the two
clauses. Against such analyses much scepticism has been voiced, e.g.
recently by Dahl (1995: 259):
Causality in discourse is a rather fluid phenomenon whose relations with

what we would want to call rhetorical structure are of an indirect character.
The view I have argued against-that causality in discourse may be reduced
to a small set of rhetorical relations9-has to my knowledge never been
stated explicitly, but many treatments of discourse structure certainly give
the impression that such a reduction is possible.
I am not quite sure what is meant by "reduction" in the above quotation.

I agree with Dahl that the functions of causality in discourse are mani-
fold (see examples (3) to (7) above). But this is no reason why they
should not be capable of being reduced to one principle. There is one
thing that can be gained by an introduction of a principle of causality
into a coherence relations approach: the complexity brought about by
the "fluid" character of a fairly large number of different causally-based
coherence relations which are not easily delimited from each other may
be "reduced" by "reducing" them to a principle of causality. As was also
observed by Ziv (1993: 181), causality could be regarded as one of a

small number of basic universal reasoning principles that readers and
hearers use in interpreting clause sequences in texts.
I think that such principles of text organisation should have the fol-
lowing properties:
a. They stem from basic needs arising in human communication.
b. They therefore constitute reasons why we should need more than one
clause to say what we want to say, that is, motives to utter a text
rather than a single clause.
c. They represent ways in which clauses, sentences, or larger text
chunks can be relevant to each other.
d. They are not derivable from a further common principle.
Causality fulfils all the above criteria to a high degree:
a. It is a basic need in human discourse to explain, to justify, to reason
about causes, conditions and consequences. The close affinity be-
tween the primitive concept of causation and human action is no co-
incidence. People want to know about causes, reasons and conse-
quences because they need to act.
b. The most natural discourse strategy of explanation or stating conse-
quences is to add another clause.
c. The common use of asyndetic clause connections to explicate causal
relations in discourse shows that causality creates relevance on its
own.
To apply criterion d. we need to know what the other principles are. In
my own work (Meyer 1983, 1996) I have identified five principles so
far that fulfil the above conditions: Topic, Time, Clarification, Causal-
ity, and Persuasion. 10 Among these, Causality seems to be the most
"relevant" one, both in terms of text frequency, and in number and di-
versity of coherence relations derivable from them. But the other princi-
ples can also be shown to be important organising principles in texts
that are not derivable from the principle of Causality.
When we speak or write about a certain topic, then usually there is
more to be said about this topic than can be accommodated in one
clause. Often several topics have to be related to each other. This is how
the different forms of topic development arise which in tum give rise to
a number of different coherence relations such as "constant topic",
"topic progression", "split topic", etc. (cf. Danes 1970).
When we narrate a story, there is more than one event to narrate, and
these events are primarily and minimally related temporally. This is how
time comes in as an organising principle of texts.
28 Paul Georg Meyer
Furthermore, when we describe something, there is often the need to

clarify what we are referring to, that is, to spell out in more detail what
we wish to say about our topic, to compare, to give examples, to para-
phrase. All this cannot be done in one clause and thus, specific coher-
ence relations such as elaboration and contrast arise.
And it is clear that very often persuasion, that is, the need to con-
vince the hearers rather than just make them comprehend, can be the
motivation for adding another clause, too, giving rise to coherence rela-
tions such as "qualification" or "evidence".
If what we said above is true, our five principles also represent five
ways in which clauses or sentences can be relevant to each other in the
Gricean sense: they represent five different ways to fulfil the maxim of
Relation in coherent discourse. But the five principles are not on an
equal footing here. 11 Kortmann's scale can be applied to them, yielding
different degrees of relevance, depending on the way in which a clause
relates to the rest of the discourse.
The lowest degree of relevance that two sentences can have to each
other is brought about by their being related through their topic or top-
ics. Very often this does not suffice for text coherence to be brought
about. When a topic is split up into subtopics, or when the text proceeds
from one topic to the next, this is only admissible insofar as the subtop-
ics or the newly introduced topics are relevant in terms of some global
purpose of the text, or, to put it in Grice's terms: contributions to a co-
herent discourse must be "appropriate to the immediate needs at each
stage of the transaction" (Grice 1975: 47).
It is obvious that clarification and persuasion devices are highly rele-
vant, even in Sperber and Wilson's sense, to the purposes of the com-
munication of which they are part. Clarification strategies aim at lower-
ing the processing costs of hearers or readers: they make the text more
easily comprehensible. But also in Grice's sense, it may be said that
clarification may be the way in which one clause is relevant to another
in discourse, serving the "immediate needs", at this "stage of the trans-
action".
Persuasion devices are meant to heighten the probability of a contex-
tual effect of the message on the reader / hearer. They sometimes explic-
itly invoke relevance as an argument to impress the reader or hearer. In
Gricean terms, persuasion devices are applied at a certain stage of a
communicative transaction where scepticism is apparent on the part of
the hearer, or anticipated by a writer on the part of the reader.
When we narrate events that are in some temporal relation to each

other, these events must be also relevant to each other. This is why nar-
rated event sequences are so often causally connected as well. Two
events that are temporally related to each other do not necessarily make
a good story. But a causal chain of events is much more likely to be
accepted as suiting the "immediate needs" at this stage of the narration:
its narration may explain why something happened or point out conse-
quences of a certain event. More often than not, the question "What
happened next?" aims at learning about the consequences of a certain
event rather than about some causally unrelated event that happened to
happen next. And informing about the causes of a certain event fulfils a
basic need in human communication (see above). Thus, causality seems
to be the most relevant of the five principles mentioned. Explicating a
causal relationship in discourse is rarely questioned or regarded as
pointless. This is no real surprise given that, as we saw in the last sec-
tion, causality is constrained by relevance considerations. Causality is
thus much more intimately connected with relevance than mere tempo-
ral sequence (cf Kortmann's scale, 1991: 121).
5. Summary
In this paper I have discussed interrelations between causality and rele-

vance. I have tried to show from three different perspectives that there
are good reasons to regard causality as a primarily discursive phenome-
non.
1. The relevance of a causal statement is not ascertainable objectively,
but only in a discourse in which human interests and social obliga-
tions play a role.
2. The relevance of a causal statement is best described in a speech-act
framework. Its conditions of relevance are more similar to felicity
conditions of speech acts than to a cognitive relevance principle.
3. The relevance of causality in discourse cannot easily be overstated.
Nevertheless, causality is only one among several principles of text
organisation, though obviously the most relevant one.
If this is reasonable (and this article, I hope, contains arguments to
show that it is), then social-psychological causal attribution research
(Edwards and Potter 1993) becomes more important for a characterisa-
tion of the concept of causality than logical analyses. This is all the
more compelling for linguists, like myself, working in a functionally
30 Paul Georg Meyer
oriented typological framework. Thirty years of experience with this

framework have shown that what is likely to emerge as an interesting
typological parameter for such a framework, or as a non-trivial seman-
tic-pragmatic universal, is more likely to be found in everyday linguistic
usage, everyday reasoning, everyday argument, than in logically puri-
fied abstractions of these.
Social scientists working on causal attribution have long begun to
look to linguistics for answers to some of their questions (Edwards and
Potter 1993; Hilton, Jaspars and Clarke 1990; Hilton and Erb 1996). It is
time that linguists began to understand these questions and provide
more satisfactory answers.
Notes
* This paper is dedicated to Ekkehard Konig on the occasion of his 60 th

birthday.
I wish to thank the editors, Elizabeth Couper-Kuhlen and Bernd Kort-
mann, and, in particular, Verena Haser and Manfred Kmg (University of
Freiburg, Germany) for extensive (and partly devastating) comments on an
earlier draft of this paper. Although I failed to be convinced by some of their
arguments, they certainly were an invaluable help in reformulating and sup-
plementing my thoughts and in many cases provided me with further insights
into the problems concerned.
1. Some people might say that this would be a reason rather than a cause. I do
not wish to enter into the causes-reasons debate here, but if there is a distinc-
tion to be made, it would have to be between causes for events and reasons
for human actions (Beckermann 1977). Being late is not an action, so what-
ever causes my being late would be a cause rather than a reason.
2. It is perhaps interesting to note, in this connection, that the German word for
'reality', Wirklichkeit, is etymologically related to the verb wirken 'take ef-
fect' which in tum is etymologically related to English work). Literally, Wirk-
lichkeit could be translated as 'a coherent whole of things that take effect'.
3. I apologise for the somewhat stereotypic character of this whole example. I
could try to give a causal explanation for this, but I seriously doubt the poten-
tial relevance of such an explanation.
4. The formulation for promises (Searle 1969: 59), e.g., is: "It is not obvious to
bothS and H that S will do A in the normal course of events".
5. To distinguish Sperber and Wilson's concept of Relevance from others, in
particular the colloquial notion of relevance, I will capitalise the word when-
ever referring to Sperber and Wilson's principle.
6. I wish to thank Verena Haser and Manfred Kmg (University of Freiburg) for
pointing this out to me.
7. Other approaches with a similar set of basic assumptions, are Grimes (1975);
Mann and Thompson (1988); Hobbs (1983); Graustein and Thiele (1979)
and many others.
8. Klare, Hugh 1., Stress violence and crime, in: Mayne, Richard (ed.): Europe
tomorrow, London 1972,48-63.
9. Dahl uses Mann and Thompson's (1988) terminology, where "rhetorical
relation" corresponds to what is now called "coherence relation" in most of
the literature.
10. I use capital letters to indicate that I am talking about principles within a
framework that are not necessarily identical with their colloquial counterparts.
11. Grice himself (1975: 46) speaks about "different kinds and focuses of rele-
vance" whose exact nature is not made very clear. The maxim of relevance is
the least explicit one in Grice's account. Maybe the five principles discussed
here could contribute to a clarification.
References
Alm-Arvius, Christina
1993 The English verb see: a study in multiple meaning. (Gothenburg
studies in English 64.) Goteborg: University.
Anscombe, G.E.M.
1971 Causality and determination. Cambridge: Cambridge University :
Press.
1975 Causality and determination. In: Ernest Sosa (ed.), Causation and
conditionals, 63-81. Oxford: Oxford University Press.
Bazerman, Charles
1988 Shaping written knowledge: the genre and activity ofthe experimen-
tal article in science. Madison, WI: University of Wisconsin Press.
Beckermann, Ansgar
1977 Grande und Ursachen: zum vermeintlichen grundsiitzlichen Unter-
schied zwischen mentalen Handlungserkliirungen und wissenschaft-
lich-kausalen Erkliirungen. Kronberg: Scriptor.
Breul, Carsten
1997 Grammatik und Bedeutung der kausalen Satzverbiinde. (Linguisti-
sche Arbeiten 368.) Tubingen: Niemeyer.
Dahl, Osten
1995 Causality in discourse. In: Gert Rickheit and Christopher Habel
(eds.), Focus and coherence in discourse processing, 251-260. (Re-
search in text theory 22.) Berlin: de Gruyter.
Danes, Frantisek
1970 Zur linguistischen Analyse der Textstruktur. Folia linguistica 4: 72-
79.
Edwards, Derek and Jonathan Potter
1993 Language and causation: a discursive action model of description
and attribution. Psychological review 100: 23-41.
32 Paul Georg Meyer
Graustein, Gottfried and Wolfgang Thiele

1979 Towards an analysis of English texts. Zeitschrift fur Anglistik und
Amerikanistik 27: 62-74.
Grice, Herbert Paul
1975 Logic and conversation. Syntax and semantics 3: 41-58.
Grimes, Joseph
1975 The thread 0.[ discourse. (Janua linguarum. Series minor 207.) The
Hague: Mouton.
Hilton, Denis 1., Joseph M.F Jaspars and David D.Clarke
1990 Pragmatic conditional reasoning: context and content effects on the
interpretation of causal assertions. Journal of pragmatics 14:
791-812.
Hilton, Denis 1. and Hans-Peter Erb
1996 Mental models and causal explanation: judgements ofprobable cause
and explanatory relevance. Thinking and reasoning 2: 273-308.
Hobbs, Jerry R.
1983 Why is discourse coherent? In: Fritz Neubauer (ed.), Coherence in
natural-language texts, 29-70. Hamburg: Buske.
Horn, Laurence H.
1984 Toward a new taxonomy for pragmatic inference: Q-based and
R-based implicature. In: Deborah Schiffrin (ed.), Meaning, form and
use in context: linguistic applications, 11-42. Washington: George-
town University Press.
KamIah, Andreas
1991 The causal relation as the most fundamental fact of the world. Com-
ments on Hans Reichenbach's paper: the space problem in the new
quantum mechanics. Erkenntnis 35: 49-60.
Kortmann, Bernd
1991 Free adjuncts and absolutes in English: problems of control and
interpretation. London: Routledge.
Lakoff, George
1982 Categories: an essay in cognitive linguistics. In: Linguistic Society of
Korea (ed.), Linguistics in the morning calm: selected papers from
SICOL-1981, 139-193. Seoul: Hanshin.
1987 Women, fire and dangerous things: what categories reveal about the
mind. Chicago: Chicago University Press.
Latour, Bruno
1987 Science in action: how to follow scientists and engineers through
society. Milton Keynes: Open University Press.
Lewis, David
1973 Causation. Journal ofphilosophy 70: 556-567.
Mach, Ernst
1901 Die Mechanik in ihrer Entwicklung: historisch-kritisch dargestellt,
4th ed. Leipzig: Brockhaus.
Mackie, John Leslie
1965 Causes and conditions. American philosophical quarterly 2:
245-264.
1974 The cement ofthe universe: A study ofcausation. Oxford: Clarendon.
1975 Causes and conditions. In: Ernest Sosa (ed.), Causation and condi-
tionals, 15-38.
Mann, William C. and Sandra Thompson
1988 Rhetorical structure theory: toward a functional theory of text or-
ganization. Text 8: 243-281.
McCawley, James D.
1968 Lexical insertion in a transformational grammar without deep struc-
ture. Papers from the Regional Meeting of the Chicago Linguistic
Society 4: 71-80.
Meyer, Paul Georg
1975 Satzverknupfungsrelationen. Ein Interpretationsmode II fur situation-
sunabhtingige Texte. (Tubinger Beitrage zur Linguistik. 61.) Tubin-
gen: Narr.
1983 Sprachliches Handeln ohne Sprechsituation: Studien zur theoreti-
schen und empirischen Konstitution von illokutiven Funktionen in
'situationslosen' Texten. (Linguistische Arbeiten. 135.) Tubingen:
Niemeyer.
1996 Textfunktion-Textstruktur-Textanalyse: Zur Linguistik des Sach-
textes. In: Wolfgang Bomer and Klaus Vogel (eds.), Texte im
Fremdsprachenerwerb: Verstehen und Produzieren, 21-44. (Tubin-
ger Beitrage zur Linguistik 418.) Tubingen: Narr.
1997 Coming to know: studies in the lexical semantics and pragmatics of
academic English. (Forum fur Fachsprachenforschung 35.) Tubin-
gen: Narr.
Reichenbach, Hans
1991 The space problem in the new quantum mechanics. Erkenntnis 35:
29-47.
Russell, Bertrand
1912 On the notion of cause. Proceedings of the Aristotelian Society 13:
1-26.
1921 On the notion of cause. In: Mysticism and logic and other essays,
180-208. London: Longman.
Searle, John Rogers
1969 Speech acts: an essay in the philosophy of language. Cambridge:
Cambridge University Press.
1983 Intentionality: an essay in the philosophy ofmind. Cambridge: Cam-
Sosa, Ernest
1975 Introduction. In: Ernest Sosa (ed.), Causation and conditionals, 1-14.
Oxford: Oxford University Press.
Sosa, Ernest (ed.)
1975 Causation and conditionals. Oxford: Oxford University Press.
Sperber, Dan and Deirdre Wilson
1986 Relevance: communication and cognition. Oxford: Blackwell.
1995 Relevance: communication and cognition. 2. ed. Oxford: Blackwell.
34 Paul Georg Meyer
Stegmtiller, Wolfgang
1983 Probleme und Resultate der Wissenschaftstheorie und Analytischen
Philosophie. Bd. 1: Erkltirung Begrundung Kausalittit. 2., verb. und
erw. Aufl. Berlin: Springer.
Strawson, P.F.
1985 Causation and explanation. In: Bruce Vermazen and Merril B. Hin-
tikka (eds.), Essays on Davidson: actions and events, 115-135. Ox-
ford: Clarendon.
Swales, John
1981 Aspects of article introductions. (Aston ESP research reports 1.)
Birmingham: University of Aston in Birmingham / Language Studies
Unit.
Von Wright, Georg H.
1973 On the logic and epistemology of the causal relation. In: P. Suppes et
al. (eds.), Logic, methodology and philosophy of science, 293-312.
Amsterdam: North Holland.
1975 On the logic and epistemology of the causal relation. In: Ernest Sosa
(ed.), Causation and conditionals, 95-113. Oxford: Oxford Univer-
sity Press.
Wierzbicka, Anna
1972 Semantic primitives. (Linguistische Forschungen 22.) Frankfurt:
Athenaum.
1991 Cross-cultural pragmatics: the semantics of human interaction.
(Trends in linguistics: studies and monographs 53.) Berlin: Mouton
de Gruyter.
1996 Semantics: primes and universals. Oxford: Oxford University Press.
Woolgar, Steve
1988 Science: the very idea. ChichesterlNew York: Tavistock/Methuen.
Ziv, Yael
1988 On the rationality of 'relevance' and the relevance of 'rationality'.
Journal ofpragmatics 12: 535-545.
1993 Causality in language and in discourse. Journal ofliterary semantics
22: 177-185.
1993 Conjunction and causality: pragmatics and the lexicon. Pragmatics
and language learning monograph series 4: 17-27.
On the processing of causal relations
Leo G. M Noordman and Femke de Blijzer
This study deals with how different kinds of causal sentences are under-
stood. The central hypothesis is that sentences that more directly reflect a
causal relation in the world and in the cognitive representation of this
situation are easier to understand than other causal sentences. This hy-
pothesis is tested in reading-time experiments: longer reading times reflect a
greater complexity in the processing of the underlying infonnation. Three
factors are discussed that detennine how directly a causal sentence matches the
cognitive representation of causality: content vs. epistemic relations, linear
order of cause and effect in the sentence, and conceptual order of cause and
effect. It is demonstrated that each of these factors makes an independent
contribution to the complexity of understanding causal sentences. In addition,
it is demonstrated that these factors have an effect independently of a fourth
factor, namely the causal constraint between cause and effect.
1. Causality in cognition
Causality is an important ordering principle of human perception and hu-

man experience, and thus a central category in human cognition. It is fun-
damental both to the representation of human knowledge and to other cog-
nitive processes like predicting, explaining, and comprehending.
Our claim is that sentences that more directly reflect causality are easier
to understand than other sentences. We should therefore be explicit about
what we mean by understanding, and about how we conceive the cognitive
representation of causality. From a psycholinguistic point of view, we con-
sider the process of understanding as one in which the reader constructs a
representation of the information in the discourse. An important aspect of
that representation is that it is coherent, since the consecutive sentences in
a discourse are related to each other. Accordingly, the representation
contains relations between the sentences, or rather between the
information units that are expressed by the sentences. But that representa-
tion is also a representation of something else, i. e., of the world. Therefore,
the representation has relations with the world. The representation may be
related to the world in terms of a number of concepts, such as truth~n
which semantic theories have in general concentrated-possibility, and
36 Leo Noordman and Femke de Blijzer
plausibility. The distinction between relations in the representation and

relations of the representation to (a model of) the world is related to the
distinction made by Guenthner (1989) between D-relations (for discourse)
and T-relations (for truth). A similar distinction is made in theories about
psychological representations (Johnson-Laird 1983). In this conception,
understanding implies that the reader not only constructs a coherent repre-
sentation but also that the reader implicitly evaluates whether the informa-
tion in the discourse corresponds to the world, for example, whether it is
true or false, plausible or possible. Accordingly, understanding causal sen-
tences requires matching the information in the clauses with a cognitive
representation of the world, in this case with a cognitive representation of
the causality in the world.
This leads to the question of how causality is cognitively represented.
We will not develop an elaborate theory of the representation of causality.
We only assume that causal relations are represented as ordered cause-
effect pairs where the cause temporally precedes the effect. The represen-
tation of causal relations originates in our experience in the world. We
observe co-occurrences of events in the world. On the basis of these co-
occurrences we deduce causal relationships. We have a strong tendency to
interpret sequences of events in terms of causal relations, even when there
is no real causality involved (Michotte 1954). A child learns that when she
drops a glass, the effect is that it breaks into pieces. A child also learns to
use causal schemes in reasoning processes: if the glass is broken, it proba-
bly has fallen. The experience of causal relations between events in the
world is fundamental to the conceptual representation of causality. Given
the fact that causal relations derive from our experience of the world,
where causes temporally precede effects, we assume that causal relations
in our representation of the world reflect this experience and that they are
represented as cause-effect pairs with the cause preceding the effect.
2. Relevant factors in the processing of causal sentences
2.1. Content vs. epistemic relations
Causal sentences can be characterized with respect to a number of factors,

which presumably affect their processing. One factor is the distinction be-
tween content relations and epistemic relations (Sweetser 1990), some-
times also identified as semantic vs. pragmatic relations (Sanders,
Spooren, and Noordman 1992). A sentence that expresses a causal content
On the processing ofcausal relations 37
relation describes a real-world causality between two events or states in

the world. An example is (1):
(1) Because John worked hard, he passed the exam.
An epistemic causal relation expresses a conclusion by the writer/speaker

that is based on a causal relation in the world. An example is (2):
(2) Because John passed the exam, he must have worked hard
This sentence expresses that the speaker's knowledge of John's passing

causes the conclusion that he worked hard. A relation is a content relation
if the two clauses are related because of their propositional content. A
relation is an epistemic relation if the clauses are related by the speaker's
reasomng.
What can we expect with respect to the difference in processing be-
tween content and epistemic causals? Earlier, it was said that understand-
ing requires matching the information in the sentence with a model of the
world. Since a content causal sentence directly reflects a state of affairs in
the world, it may be easier to process than an epistemic causal sentence,
which does not directly express real-world causality. One can argue that an
epistemic causal expresses a real-world causality in a derived way. In fact,
an epistemic relation reflects a line of reasoning that is allowed by the co-
occurrence of events or situations in the world. The justification of that
reasoning is the contingency of events in the world. So, in this sense, an
epistemic relation is based on an underlying content relation. In under-
standing an epistemic relation, the reader has to check the possibility of the
underlying content relation in the world. Understanding an epistemic rela-
tion implies understanding the underlying content relation. We will illus-
trate this with two examples.
In sentence (2), the epistemic conclusion is justified by the underlying
content relation. The reason that one may infer from "John passed the
exam" to "he worked hard" is that ifyou work hard, you pass the exam.
In (3), there is the same underlying content relation: if you work hard,
you pass the exam.
(3) John worked hard, so he must have passed the exam.
The content relation is used in an act of concluding, which is expressed by

the connective so.
According to this analysis, content relations reflect experience in the

real world and epistemic relations are derived from these content relations.
Epistemic relations express real-world causality in a less direct way. One
might predict that processing an epistemic relation requires more time than
processing a content relation.
There is some support in the literature for this prediction. Noordman
(1979) asked readers to read conditional sentences that expressed content
and epistemic causal relations. An example of the first type is: "If John is
ill, he is not going to work". An example of the second type is: "If John is
not going to work, he is ill". The readers had to press a response button as
soon as they understood the sentence. The time was measured from the
moment the sentence was presented until the participant pressed a button.
Sentences expressing a content causal relation were processed 286 msec
faster than sentences expressing an epistemic causal relation. So, sentences
that express the causal relation more directly are processed faster than
sentences that express the causal relation in a less direct way. Similar re-
sults have been obtained by Traxler et ale (1997) for the understanding of
causal sentences and diagnostic sentences. In their experiment, these two
types of sentences corresponded to content and epistemic causals. The
interpretation by Traxler et ale is that diagnostic statements need a more
complex mental representation than causal statements; diagnostic state-
ments require the representation of an embedding phrase like: "someone
believes that. ..".
These data show that content and epistemic causal sentences are proc-
essed in a different way. A possible interpretation is that understanding
causal relations that describe events in the real world is easier than under-
standing causal relations that express a reasoning process. But we may not
yet derive this conclusion. When we scrutinize the sentences that are used
in the experiment by Noordman, it is clear that the difference between
content and epistemic relations is not the only difference between the sen-
tences. There are other factors that might play a role in the processing of
these causal sentences. One factor is what we call the conceptual order of
the clauses. Another factor is the linear order of the clauses. We will first
discuss these factors: conceptual order in Section 2.2. and linear order in
Section 2.3. Finally, we will discuss a factor that will be called causal con-
straint (Section 2.4.). In Section 3, we will present a reading-time experi-
ment in which these factors are investigated.
2.2. Conceptual order
Consider sentences (1) and (4):
(1) Because John worked hard, he passed the exam.
(4) John passed the exam, because he worked hard
Both sentences express a cause-effect relation. Conceptually, the direction

in which the causality is expressed is from cause to effect. In the concep-
tualisation underlying sentences (1) and (4), the effect is derived from the
cause. We will refer to this order as the conceptual cause-effect order.
There is a correspondence between the antecedent (John worked hard) and
the consequent (he passed the exam) in these sentences on the one hand,
and the cause and effect in the real world on the other hand. What con-
ceptually is the cause (working hard) is expressed as the antecedent in the
sentence; what conceptually is the effect (passing the exam) is expressed
as the consequent in the sentence. In these causal sentences, we define the
antecedent as the starting point in the causal deduction that is expressed by
the sentence and the consequent as the end point. In a content sentence, the
antecedent is the cause and the consequent is the effect, as in (1). In an
epistemic sentence, the antecedent is the argument and the consequent is
the claim, as in (2). Sentences that have the conceptual cause-effect order
exhibit what we call conceptual correspondence: they indicate that on the
basis of the cause the effect is derived. This order corresponds to the order
ofthe events in the world.
Sentence (2) expresses a effect-cause relation.
(2) Because John passed the exam, he must have worked hard
Conceptually, the cause is derived from the effect. The effect is considered
as a sign for the cause. (Therefore it was identified as an epistemic relation
in the first place). Sentence (2) has the conceptual effect-cause order.
There is an incongruence between what is cause and effect in the real
world and what is expressed in the sentence as antecedent and consequent.
It should be noted that the distinction between conceptual cause-effect
order and conceptual effect-cause order is not the same as the distinction
between content and epistemic relations. In fact, sentence (3) expresses an
epistemic relation, but the effect is derived from the cause.
40 Leo Noordman and Femke de Bli.Jzer
(3) John worked hard, so he must have passed the exam.
Therefore, sentence (3) has the conceptual cause-effect order. On the other
hand, the two distinctions are not completely independent of each other, as
is shown in Table 1. Content relations only occur in the conceptual cause-
effect order, because content relations directly reflect real-world causality.
Epistemic relations can have either conceptual cause-effect order or effect-
cause order.
Table 1. Examples of content and epistemic relations, in different conceptual and

linear orders.
conceptual order
cause-effect effect-cause
linear cause-effect effect-cause cause-effect effec~cause
order
content Because John John passed the -
relation worked hard, he exam, because
passed the he worked hard.
exam.
epistemic John worked John must have John must have John passed the
relation hard, so he must passed the worked hard, exam, so he
have passed the exam, since he since he passed must have
exam. worked hard. the exam. worked hard.
Since sentences with the conceptual cause-effect order reflect causality in

the real world more directly than sentences with the conceptual effect-
cause order, and since we assume that the more directly a sentence
matches a situation in the world, the easier it is to process, we predict that
sentences with cause-effect order (such as (1)) are easier to process than
sentences with effect-cause order (such as (2)). We prefer deriving effects
from causes to deriving causes from effects.
The basis for this assumption is our sensorimotor experience in the
world. We learn the notion of causality and we learn causal relations by
acting in the world, by observing that causes precede effects. If the cause-
effect order is conceptually more fundamental than the effect-cause order,
it seems likely that we prefer to reason from cause to effect instead of from
effect to cause.
Support for the assumption that conceptual cause-effect order is more
fundamental than effect-cause order is obtained by Tversky and Kahne-
man (1982). They found that reasoning from cause to effect is easier than
from effect to cause, even in situations in which the likelihood of the cause
given the effect is the same as the likelihood of the effect given the cause.
Tversky and Kahneman conclude from their studies that people "infer
effects from causes with greater confidence than causes from effects"
(page 118). People make use of cause-effect schemas, and causal infer-
ences that follow the normal cause-effect sequence are easier to make than
diagnostic inferences that reverse this sequence.
2.3. Linear order
The conceptual order of the clauses should be distinguished from the lin-
ear order. Both sentences (1) and (4), repeated below, express the concep-
tual order of cause-effect.
(1) Because John worked hard, he passed the exam
(4) John passed the exam, because he worked hard
The sentences differ in the linear order of the clauses. In (1), the first
clause expresses the cause and the second clause the effect. In (4), the first
clause is the effect and the second clause is the cause. Sentence (1) thus
has an order which is iconic with the world; sentence (4) is non-iconic.
What can we predict with respect to the processing of causal sentences?
An obvious prediction is that an iconic order facilitates processing. The
reason is similar to the one given in the section on conceptual order. If the
order of the clauses corresponds to the order of causality in the world (or
more specifically, to a model of causality), the matching process between
language and knowledge is easier (and thus the understanding of the sen-
tence).
However, on the basis of the literature one might make a different pre-
diction. Magliano et al. (1993) argue that inferences are made only about
causes and not about effects. They argue that readers make inferences
about causes because these explain the current situation (and the sentence
that expresses this situation). The inference forms a backward link. For-
ward inferences tend not to be made, because, at any point in a text, there
are numerous inferences one can make about the possible effects of the
situation expressed in the current sentence. Lexical decision data did in-
deed support the conclusion that inferences about causal antecedents are
made but inferences about effects are not. However, in this experiment the
causal constraints were rather weak (.30 and .26). We predict that in a
situation in which the cause strongly determines the effect and the effect
determines the cause, iconicity will facilitate processing of a causal sen-

tence. This prediction rests on the assumption that the cause-effect order
reflects the structure of our causal knowledge. These predictions were
tested by Noordman, Vonk and Meyer Viol (Noordman and Vonk 1998;
Meyer Viol 1984). In the experimental texts, a causal relation was ex-
pressed in two different orders. The following fragment is an example:
"In order to earn some money, John was cutting out weeds in his mother's
garden. It was a tough job, because the stinging-nettles were a metre high.
After two hours, he took a short break. The job was half finished. The sun
stood high in the sky and it was sweltering hot. He wiped away the sweat from
his forehead with his hand. He had touched the stinging-nettles with his hand.
His hand itched terribly."
In this fragment the cause sentence ("He had touched the stinging-nettles")
preceded the' effect sentence ("His hand itched terribly"). In the other ver-
sion of the same fragment, the cause sentence and the effect sentence were
reversed ("His hand itched terribly. He had touched the stinging-nettles").
It should be noted that the causal relations in both conditions have con-
ceptual cause-effect order. If we make the causality of the relation explicit,
we obtain the sequences: "because he had touched the stinging-nettles, his
hand itched terribly", and "his hand itched terribly, because he had
touched the stinging-nettles".
The materials in this experiment were constructed in such a way that,
first, the effects could very well be predicted on the basis of the causes
and, second, that the causes could very well be predicted on the basis of
the effects. This was achieved by two separate pilot studies to test and
improve the materials. The experiment showed that a "cause sentence
speeds up the processing of a subsequent consequence sentence, but a
consequence sentence does not speed up the processing of a subsequent
cause sentence... What this experiment demonstrates is a kind of iconicity
between cognitive structure and language" (Noordman and Vonk 1998:
205). If the linear order of the clauses is iconic with our model of causality
in the world, processing the sentence is speeded up. This has been demon-
strated in texts in which the effect is highly predictable given the cause,
and the cause is highly predictable given the effect. In the experiment of
Section 3, reported more completely in de Blijzer (1999), predictability
will no longer be held constant, but will be an independent variable.
Summarizing, we have made three distinctions: content vs. epistemic
relations, conceptual order, and linear order. In principle, these three dis-
tinctions yield a 2x2x2 matrix of eight different kinds of causal sentences
(see Table 1). As has already been noted, two cells are empty. A content
relation always has cause-effect order. If the conceptual order is effect-
cause, the relation is always epistemic. In Section 2.1, we discussed an
experiment that compared two kinds of sentences that differ on all three
factors. What we wish to do in the remainder of this paper is to tIy and
find evidence for the factors separately. Also, we would like to further
investigate a fourth factor, the predictability of causes and effects.
2.4. Causal constraint: the predictability ofcauses and effects
A fourth factor that may affect the processing of causal sentences is the
predictability of causes and effects in causal relations. How likely is the
effect given the cause and how likely is the cause given the effect? In par-
ticular, if a speaker draws a conclusion on the basis of an epistemic rela-
tion, it is crucial that there is a solid ground for drawing that conclusion
and that there are few possible alternatives for the conclusion. More spe-
cifically, when the derivation is from cause to effect, it is crucial that there
are few conditions that prevent the cause from having an effect. Following
Cummins et al. (1991), we will call such conditions "disabling condi-
tions". When the derivation is from effect to cause, it is crucial that there
are few other causes. We will call these causes "alternative causes". In the
previous experiment these factors were held constant.
Sentences (5) and (6) illustrate that disabling conditions affect the
plausibility of a sentence that expresses a derivation from cause to effect
(i.e., where the effect is derived from the cause).
(5) The woman must have got a sunstroke, since she was in the sun for
a long time.
(6) The girl must have got wet, since she jumped into the river.
In (5) it is quite likely that there are disabling conditions which could pre-
vent the effect (e.g., wearing a hat, sitting in front of a ventilator, throwing
water over your head once in a while or only being in the sun when it is
not too hot). So when a person has been in the sun for a long time, it is not
at all certain that the effect occurs and that this person does indeed get a
sunstroke. This line of reasoning, expressed by sentence (5), is rather im-
plausible. In (6), however, it is rather unlikely that·there are disabling con-
ditions that could prevent the effect. If someone jumps into the river, it is
highly likely that this person gets wet. Therefore, when we know the cause
we can easily draw the conclusion that the effect must have taken place.
When we derive the cause from the effect, the plausibility of a sentence
is affected by alternative causes, as is illustrated by sentences (7) and (8).
(7) The woman got a sunstroke, so she must have been in the sun for a
long time.
(8) The girl got wet, so she must have jumped into the river.
In sentence (7) it is unlikely that there are alternative causes for a sun-
stroke. Therefore, if someone has a sunstroke, we can easily draw a con-
clusion about the cause: this person has been in the sun for a long time.
This line of reasoning is very plausible. In sentence (8), however, it is
likely that there are alternative causes for being wet (e.g., rain, snow, tak-
ing a shower or a bath, a bucket of water). So, when a person is wet, it is
not at all certain that this person has jumped into the river. The line of
reasoning is not very plausible. It should be noted that these observations
do not regard the acceptability of the sentences as linguistic entities (they
all are acceptable), but the plausibility of the sentences as descriptions of
the world.
2.4.1. Causal constraint in logical deduction
Cummins (1995) and Cummins et al. (1991) have shown that disabling
conditions and alternative causes affect logical reasoning. For example,
people have difficulty with the following deduction:
(9) If you liein the sun for a long time, you get a sunstroke. Mary has
been lying in the sun for a long time. So Mary will get a sunstroke.
Relatively many subjects judged such a conclusion incorrect, although it is

a logically valid deduction. The interpretation is that there are many dis-
abling conditions. On the other hand, if there are few alternative causes,
relatively many subjects made the logically incorrect deduction from ef-
fect to cause:
(10) If you lie in the sun for a long time, you get a sunstroke. Mary got a
sunstroke. So Mary must have been lying in the sun for a long time.
Many subjects judged such a conclusion correct, because there are few
alternative causes. Cummins et al. showed that disabling conditions and
alternative causes affect judgments on correct and incorrect conclusions.
But in this study we are not interested in true/false judgments concerning
logical conclusions. We are interested in the understanding of causal sen-
tences. The question then is whether disabling conditions and alternative
causes also affect the understanding of causal sentences. Earlier we argued
that language understanding implies an implicit evaluation of the sentence
on the basis of a model of the world. Therefore, it is quite likely that the
understanding of causal sentences is sensitive to factors that affect the
validity of conclusions, and therefore that the understanding of causal
sentences is sensitive to disabling conditions and alternative causes. For
this reason, we incorporated causal constraint, operationalized in terms of
disabling conditions and alternative causes, as a variable in the experiment
to be reported in Section 3.
2.4.2. Causal scenarios
Sentences (5) to (8) illustrate two causal scenarios. The scenario underly-
ing (5) and (7)-if one lies in the sun for a long time, one gets a
sunstroke-has many disabling conditions (DC) and few alternative
causes (AC); the scenario underlying (6) and (8)-if one jumps into the
river, one gets wet-has few disabling conditions and many alternative
causes. These two kinds of scenarios will be used to study the role of
causal constraint. We collected a set of 71 causal scenarios. In order to
find appropriate sentences for the experiment, we presented these
scenarios to twenty participants in two judgment tasks. In one task,
participants had to assume that the causal event occurred and estimate on a
four-point scale how certain the effect was. For example: Suppose that the
girl jumps into the river. How certain is it that she will get wet? The higher
the score, the higher the causal constraint, the smaller the impact of
disabling conditions. The second task was just the reverse. Participants had
to assume that the effect event had occurred and to estimate on a four-
point scale how certain the cause event was. For example, suppose the girl
has got wet, how certain is it that she jumped into the river? The higher the
score, the higher the causal constraint, the smaller the impact of alternative
causes. For each scenario, the mean DC (disabling conditions) and AC
(alternative causes) scores over the twenty participants were computed as
well as the mean difference between the AC scores and the DC scores
(AC-DC). The 24 scenarios with the highest negative AC-DC difference

scores and the 24 scenarios with the highest positive AC-DC difference
scores were selected for the experiment. These groups of scenarios
constituted the two categories of the experimental materials: high on
disabling conditions and low on alternative causes on the one hand, and
low on disabling conditions and high on alternative causes on the other
hand.
3. Experimental study: the processing of causal sentences
The aim of this experiment was to find out whether the processing of con-
tent relations differs from that of epistemic relations and whether the proc-
essing of relations that have cause-effect conceptual order differs from
relations that have effect-cause conceptual order. Linear order was kept
constant: in all cases the first clause mentioned the effect and the second
clause the cause.
The experimental conditions in this experiment are illustrated in Table
2. The experiment was conducted in Dutch, but the table presents English
translations as well. There is only one type of content sentence: a sentence
type that expresses a cause-effect conceptual order. There were three types
of epistemic sentences. The fact that a sentence expresses an epistemic
relation was marked by the modal must have (in Dutch zal we/), or by the
conjunctions so or since (in Dutch: dus and want). In sentence type 3, the
epistemic character was indicated only by the conjunction so. The four
sentence types were applied to the two categories of scenarios.
Table 2. The experimental conditions according to type of relation (content vs.

epistemic), conceptual order (cause-effect vs. effect-cause) and causal
constraint (scenarios low on disabling conditions/high on alternative
causes vs. high on disabling conditions/low on alternative causes).
Causal constraint: low disabling conditions / high alternative causes

Relation Conceptual
type order
1 content cause-effect Het meisje is nat geworden, omdat ze in de rivier is
gesprongen.
The girl got wet, because she jumped into the river.
2 epistemic cause-effect Het meisje zal weI nat zijn geworden, want ze is in de
rivier gesprongen.
The girl must have got wet, since she jumped into the
nver.
3 epistemic effect-cause Het meisje is nat geworden, dus ze is in de rivier
gesprongen.
The girl got wet, so she jumped into the river.
4 epistemic effect-cause Het meisje is nat geworden, dus ze zal weI in de rivier
zijn gesprongen.
The girl got wet, so she must have jumped into the
nver.
Causal constraint: high disabling conditions / low alternative causes

Relation Conceptual
type order
1 content cause-effect De vrouw heeft een zonnesteek, omdat ze lang in de zon
heeft gezeten.
The woman got a sunstroke, because she was in the sun
for a long time.
2 epistemic cause-effect De vrouw zal weI een zonnesteek hebben, want ze heeft
lang in de zon gezeten.
The woman must have got a sunstroke, since she was in
the sun for a long time.
3 epistemic effect-cause De vrouw heeft een zonnesteek, dus ze heeft lang in de
zon gezeten.
The woman got a sunstroke, so she was in the sun for a
long time.
4 epistemic effect-cause De vrouw heeft een zonnesteek, dus ze zal wei lang in
de zon hebben gezeten.
The woman got a sunstroke, so she must have been in
the sun for a long time.
The causal relations were expressed by sentences without context. That is

a reasonable first step to find out the extent to which different factors play
a role in understanding, e.g., the difference between content and epistemic
relations and conceptual order. Further research will be required, for ex-
ample, to find out whether the context can overrule the effects of these
factors.
What kind of reading task is appropriate in this situation? We are inter-
ested in understanding causal relations. Understanding requires that the
sentence is related to a model of the world. Accordingly, when readers
understand a sentence, they should be able to evaluate whether the causal
relation expressed by the sentence is a possible causal relation in the
world. In the experiment, that was the task the readers were instructed to
perform: Is a causal relation possible between the two clauses? Since this
task requires that there are also impossible causal relations, so-called filler
items are incorporated in the experiment that expressed impossible rela-
tions (see Section 3.3.)
3.1. Procedure
The sentences were presented using a moving window paradigm. In this

paradigm, the sentences are presented in three parts: the first clause, the
conjunction and the second clause. Only one part is visible at a time; in the
other parts, the letters are replaced by hyphens. In the first presentation,
only the letters of the first part are visible; the letters of the other parts are
hyphens. As soon as the participant has pressed a button, the reading time
for the first part is registered and the hyphens of the second part are re-
placed by letters. In this way, the three parts were read consecutively and it
was possible for the reading times for the three parts to be measured sepa-
rately. The reading times for the third part are especially interesting, be-
cause at that moment and only then can the causal relation be understood
and does it become clear what kind of relation was expressed by the sen-
tence. Consequently, differences in processing time are expected to occur
in the third part. In this experiment, the third part is always the causal
clause.
3.2. Predictions
The four types of sentences (see Table 2) define the four conditions in this
experiment. The predictions for the reading times for the identical second
clauses (excluding the conjunction) in conditions 1, 2, and 3 are as fol-
lows. On the basis of conceptual order, the reading times for conditions 1
and 2 are expected to be shorter than for condition 3. The reading times for
the content condition 1 should be shorter than for the epistemic condition
2. This leads to a predicted increase in reading times for the conditions 1 to
2 to 3. The second clause in condition 4 is two words longer than in the
other conditions, so these reading times cannot be compared to the reading
times for the other second clauses, but the total sentence in condition 4 has
the same length (in terms of words and letters) as in condition 2. Based on
conceptual order, the reading time for sentences in condition 2 is predicted
to be shorter than in condition 4.
It is likely that these predictions are qualified by the causal constraint:
disabling conditions may make the cause-effect order more difficult, and
alternative causes may make the effect-cause order more difficult. More
specifically, for the scenarios that have low DC scores and high AC
scores, the low DC scores will make the cause-effect items easier, and the
high AC scores will make the effect-cause order more difficult. For the
scenarios that have high DC scores and low AC scores, the high DC scores
will make the cause-effect items more difficult and the low AC scores will
make the effect-cause items easier. So the difference between conditions 1
and 2 on the one hand and 3 on the other hand will be greater for the sce-
narios with low DC scores and high AC scores than for the scenarios with
high DC scores and low AC scores. In other words, an interaction is pre-
dicted between conceptual order and the two categories of causal con-
straint. No effect of causal constraint is expected on the comparison be-
tween content and epistemic sentences (condition 1 versus 2), because
both sentence types have the conceptual cause-effect order.
The effect of causal constraint can be investigated in an additional way.
The hypothesis is that the cause-effect order becomes more difficult when
the DC scores are higher. Consequently, there should be a positive corre-
lation between the DC scores and the reading times for the causal clauses
of conditions 1 and 2. The effect-cause order is supposed to become more
difficult when the AC scores are higher. Consequently, there should be a
positive correlation between the AC scores and the reading times for the
causal clauses of conditions 3 and 4.
3.3. Materials
As was explained in Section 2.4.2., the materials in the experiment con-

sisted of two groups of 24 scenarios each: the group of scenarios with high
DC scores and low AC scores, identified as high DC/lowAC and the group
of scenarios with low DC scores and high AC scores, identified as low
DC/high AC. For each scenario, four sentences were formulated corre-
sponding to the four experimental conditions. Four lists of materials were
composed, by pairing each of the 48 scenarios with a different experi-
mental condition in each list, so as to balance the lists with respect to the
scenarios and experimental conditions. Although sentence length does not
affect the way the hypotheses are tested, it is advisable to keep the sen-
tence length as constant as possible in order to reduce experimental noise.
The mean lengths of the causal clauses for the low DC/high AC and the
high DC/low AC scenarios were 8.25 and 8.33 syllables respectively. The
mean lengths of the complete sentences were also rather similar: 15.21 and
16.21 syllables respectively.
Given the fact that the task was to judge whether the causal relation
was possible, and all the experimental sentences required an affirmative
answer, filler sentences requiring a negative answer were needed. The
filler sentences expressed two clauses between which no causal relation
was possible. The clauses themselves made perfect sense and were quite
possible, so that the negative answer could only be based on the impossi-
bility of a causal relation, as it should be. An example is: "The letters have
become blurred because someone held the book upside down", and "Her
nose was bleeding because she had a hair cut". There were 32 filler sce-
narios. The filler scenarios were incorporated in the four lists and were
balanced with the experimental conditions in the same way as the experi-
mental scenarios.
Each of the four lists was presented to 8 participants. The participants
were students ofTilburg University, between 18 and 27 years of age.
3.4. Results and discussion
We are mainly interested in reading times. Since the data of participants

with many errors are not reliable, the reading time data of participants
whose error rate was higher than 5 % were not analysed, except in one
group in which two participants had an error rate of 7.5 %. Each reading
time of a particular reader for a particular item that wa~ extremely high or
extremely low in comparison to the average of the participant and the av-
erage of the item was replaced by an estimate based on the average of the
reader and the average of the item. In a similar way, reading times for
incorrect answers were replaced, since they are not informative. The per-
centage of substitutions was 3.1 %.
The reading times for the causal clauses are presented in Table 3.
Table 3. Reading times (msec) for the causal clauses.
1 2 3 4
content epistemic epistemic epistemic
cause-effect cause-effect effect-cause effect cause + modal
low disabling conditions / high alternative causes
1567 1730 2029 1985
High disabling conditions / low alternative causes

1380 1489 1620 1672
Mean
1473 1610 1824 1829
The reading times for the clauses in condition 1 are a significant 137 msec
shorter than for clauses in condition 2: F 1(1,28)=8.44, p<.Ol~
F2(1,40)=8.82, p<.Ol, supporting the hypothesis that content relations are
easier to process than epistemic relations. The reading times for clauses in
condition 2 are 214 msec shorter than for clauses in condition type 3:
F 1(1,28)=27.72, p<.OOl~ F2(1,40)=13.37, p<.OOI, supporting the hypothe-
sis that conceptual order plays a role: cause-effect relations are easier to
process than effect-cause relations.
There was a significant interaction between the experimental conditions
and the causal constraint, just as predicted. The difference between cause-
effect and effect-cause conceptual orders is greater for low DC/high AC
scenarios than for high DC /low AC scenarios. Specifically, the difference
between clauses in conditions 1 and 3 is greater for the low DC/high AC
than for the high DC/low AC scenarios: F 1(1,28)=7.53, p=.OI~
F 2(1,40)=4.66, p<.05. A similar pattern is observed for the difference be-
tween clauses in conditions 2 and 3, although the F 1 shows only a trend
and the F2 is not significant: F 1(1,28)=3.69, p=.065~ F2(1,40)=2.03, p=.16.
These interactions support the hypothesis that the cause-effect order is
more difficult when there are more disabling conditions and that the effect-
cause order is more difficult when there are more alternative causes.
As predicted, the difference between the clauses in conditions 1 and 2
is constant for the two categories of causal constraint. That should be the
case, since sentences in both conditions express cause-effect relations. One
prediction was not confirmed. The reading time for the complete sentence
2 was not shorter than for the complete sentence 4, as emerges from the
data in Table 4.
Table 4. Reading times (msec) for the complete sentences in conditions 2 and 4
2 4
epistemic epistemic
cause-effect effect-cause
low disabling conditions / high alternative causes
3770 3709
high disabling conditions / low alternative causes
3621 3441
mean
3695 3575
The basis for this prediction was that the conceptual order in sentence 2
but not in sentence 4 corresponds to the order of cause-effect in the world.
The prediction was justified by the fact that sentences 2 and 4 had the
same length. In sentence 2, the modal must have occurred in the first
clause; in sentence 4 it occurred in the second clause. The assumption was
that the effect of must have is constant. But this assumption appears to be
wrong. In fact, the reading times for the second clauses in condition 3 and
4 did not differ. This is surprising, since the Dutch clause in 4 (ze zal weI
lang in de zon hebben gezeten; "she must have been in the sun for a long
time") is two words longer than clause 3 (ze heeft lang in de zon gezeten;
"she was in the sun for a long time"). Interestingly, the same difference in
length occurred (in the original Dutch examples) in the first clauses of
conditions 2 (De vrouw zal weI een zonnesteek hebben; "The woman must
have got a sunstroke") and 4 (De vrouw heeft een zonnesteek; "The
woman got a sunstroke"). Here the longer clause did indeed require a
longer reading time; the reading times for the first clause in the conditions
2 and 4 were 1464 msec and 1130 msec respectively.
The interpretation is that must have in the second clause (condition 4) is
an indication that the clause expresses a conclusion on the part of the
writer. So must have facilitates the epistemic interpretation. If must have is
absent (in condition 3), the reader has to infer on the basis of so and hislher
knowledge of the world that the sentence expresses an epistemic conclu-
sion. A similar effect has been observed by Cozijn (in preparation) for the
conjunction because. The causal relation between two clauses can, but
need not be expressed by the conjunction because. If there is no conjunc-
tion, the sentence is shorter. But Cozijn found that the presence of because
speeds up reading, counteracting the increase in length. The facilitating
effect of must have is not expected for the first clause (conditions 2 and 4).
The first clause of condition 4 is initially interpreted as a content relation,
and during the first clause there is not yet any evidence that the sentence
expresses an epistemic relation, which otherwise would slow down the

processing ofthis first clause in condition 4.
Now we return to the predicted difference in reading time for the
complete sentences in conditions 2 and 4. The modal must have in the first
clause increases the reading time for sentence 2 and this cancels the
advantage of the conceptual order, leading to a reading time as long as that
for sentence 4. Since the effect of the conceptual order is smaller for the
high DC/low AC scenarios than for the low DC/high AC scenarios, the
increase due to the extra must have for the high DC/low AC scenarios is
greater than the advantage of the conceptual order. For the high DC/low
AC scenarios, this results in a significantly longer reading time for
sentence 2 than for sentence 4 (Fl(1,28)=4.97, p<.05; F 2(1,20)=4.86,
p<.05).
Finally, it was predicted that disabling conditions make the cause-effect
order more difficult, and that alternative causes make the effect-cause or-
der more difficult. This implies that the cause-effect sentences (conditions
1 and 2) should be easier for the low DC/high AC scenarios than for the
high DC/low AC scenarios, and that effect-cause sentences (conditions 3
and 4) should be easier for the high DC/low AC scenarios than for the low
DC/high AC scenarios. However, all the reading times of the high DC/low
AC scenarios were shorter than those of the low DC/high AC scenarios: F 1
(1,28)=39.89, p<.OOl; F2(1,40)=13.60, p<.OOl. This can be explained in
terms of the linear order of the clauses. It should be remembered that the
effect clause was always presented as the first clause and the causal clause
as the second clause. If the reader has read the effect clause, and tries to
anticipate the subsequent causal clause, the reading time for the causal
clause will be shorter if there are only few causes than if there are many
causes. This is exactly what the results show. So there is an effect of the
order of presentation of the sentences. Although we kept the order of pres-
entation constant, these results do show a linear effect in the processing of
the causal relations. The interesting result is that the effect of linear order
apparently overrules the effect of causal constraint within each of the con-
ceptual orders.
An independent confirmation of this result is obtained from the correla-
tions. It was predicted that there would be a positive correlation between
the DC scores and the reading times for the cause-effect order (conditions
1 and 2), and a positive correlation between the AC scores and the reading
times for the effect-cause order (conditions 3 and 4). But, as Table 5 dem-
onstrates, both orders correlate positively with the number of alternative
causes.
Table 5. Correlations between clause reading times (msec) and scores for
alternative causes (AC) and disabling conditions (DC) respectively.
1 2 3 4
content epistemic epistemic epistemic
cause-effect cause-effect effect-cause effect-cause + modal
AC .41* .37* .51 * .50*
DC -.01 .00 -.08 -.01
* p<.Ol
This makes perfect sense, since the effect clause was presented first, lead-
ing the reader to anticipate the cause sentence. The greater the predictabil-
ity ofthe cause sentence, the shorter, in fact, the reading time.
4. Conclusion
In our investigation of the processing of causal relations, a first distinction

was drawn between content causal and epistemic causal relations. Experi-
mental evidence from the literature (Section 2.1.) showed that processing
epistemic relations is more complex than processing content relations. But
it was argued that several factors were involved in the processing of the
causal relations that were investigated. Besides the distinction between
content and epistemic relations, distinctions between callSal relations can
be made on the basis of the correspondence of the conceptual order of the
clauses and a model of the world, on the basis of the correspondence of the
linear order of the clauses and a model of the world, and on the basis of
causal constraint between the cause and the effect. In this study, empirical
evidence was obtained for the effect ofthese four factors separately.
First, epistemic relations require more processing time than content re-
lations, as was indicated by the results of the experiment described in Sec-
tion 3. Understanding a sentence that describes a relation between states of
affairs in the world is easier than understanding a sentence that expresses a
reasoning about that relation. The explanation is that an epistemic relation
implies a content relation. The basis for an epistemic causal relation (a
reasoning) is a cause-effect relation in the world.
Second, there is an effect of conceptual order, as was indicated by the
results of the same experiment. Processing a causal relation from cause to
effect is easier than processing from effect to cause.
Third, the experiment also showed that the causal constraint has an ef-
fect. Processing from effect to cause is easy for scenarios that score low on
alternative causes. Processing from cause to effect is easy for scenarios
that score low on disabling conditions. But in both cases, the effect of
content vs. epistemic and the effect of conceptual order was observed.
Fourth, there is an effect of linear order. Presenting the cause first fa-
cilitates processing, as was indicated by the results of the experiment re-
ported in Section 2.3. An effect of linear order in processing was also
demonstrated in the experiment described in Section 3. In that experiment,
the effect clause was always presented as the first clause. The reading time
for the causal clause was shorter as it was the more predictable. Appar-
ently, readers try to anticipate the clause coming next.
Acknowledgment
We gratefully acknowledge many valuable contributions by Wietske Vonk

during all phases of this research and many valuable comments on an ear-
lier version of this paper by Lieselotte Anderwald, Elizabeth Couper-
Kuhlen, Verena Haser, and Bernd Kortmann.
References
Blijzer, F. de
1999 Causale scenario's. Het effect van de conceptuele volgorde en de
sterkte van de oorzaak-gevolg koppeling op de verwerking van seman-
tische en epistemische relaties [Causal scenarios. The effect of concep-
tual order and causal constraint on the processing of semantic and
epistemic relations]. MA thesis, Tilburg University, The Netherlands.
Cozijn, R
in preparation Integration and inference processes in understanding causal
relations. Ph.D. dissertation, Tilburg University, The Netherlands.
Cummins, D.D.
1995 Naive theories and causal deduction. Memory and Cognition 23: 646-
658.
Cummins, D.D., T. Lubart, O. Alksnis and R Rist
1991 Conditional reasoning and causation. Memory and Cognition 19: 274-
282.
Guenthner, F.
1989 Discourse: understanding in context. In: H. Schnelle and N.O. Bernsen
(eels.), Logic and Linguistics. Research Directions in Cognitive Science,
127-142. (European Perspectives 2.) Hove: Erlbaum.
Johnson-Laird, P.N.
1983 Mental Models. Cambridge: Harvard University Press.
Magliano, J.P., W.B. Baggett, B.K. Johnson and A.C. Graesser
1993 The time course of generating causal antecedent and causal conse-
quence inferences. Discourse Processes 16: 35-53.
Meyer Viol, W.P.M.
1984 Foregrounding and causality in inference making. M.A. thesis, Max
Planck Institute for Psycholinguistics, Nijmegen, The Netherlands.
Michotte, A.
1954 La Perception de la Causalite. Louvain: Publications Universitaires de
Louvain.
Noordman, L.G.M.
1979 lnferringfrom Language. Berlin: Springer.
Noordman, L.G.M. and W. Vonk
1998 Memory-based processing in understanding causal infonnation. Dis-
course Processes 26: 191-212.
Sanders, T.J.M., W.P.M. Spooren and L.G.M. Noordman
1992 Toward a taxonomy of coherence relations. Discourse Processes 15: 1-
35.
Sweetser, E.
1990 From Etymology to Pragmatics. Metaphorical and Cultural Aspects of
Semantic Structure. Cambridge: Cambridge University Press.
Traxler, MJ., A.J. Sanford, J.P. Aked and L.M. Moxey
1997 Processing causal and diagnostic statements in discourse. Journal of
Experimental Psychology: Learning, Memory, and Cognition 23: 88-
101.
Tversky, A. and D. Kahneman
1982 Causal schemas in judgments under uncertainty. In: D. Kahneman, P.
Slovic and A. Tversky (008.), Judgment under Uncertainty: Heuristics
and Biases, 117-128. Cambridge: Cambridge University Press.
Domains of use or subjectivity? The distribution
of three Dutch causal connectives explained*
Henk Pander Maat and Ted Sanders
This paper proposes a new account of the differences in meaning between

the three most frequent causal connectives in Dutch: daardoor, daarom
and dus. In an earlier corpus-analytic study, it was found that a distinction
between domains, such as the one proposed by Sweetser (1990), cannot
account for the distribution of these connectives.
Alternatively, the difference between the connectives is conceived of
here in terms of subjectivity or, more precisely, in terms of the distance
between the speaker and the Subject of Consciousness (SOC) responsible
for the causal relation. In such a subjectivity account, the three connectives
can be characterized as follows. In causal relations expressed by daardoor
'as a result' there is no SOC present because the causality is located outside
of any SOC. In relations expressed by daarom 'that's why' an SOC is
involved, but it is located at a certain distance from the Speaker~ in the case
of dus 'so', this distance is small or even absent. Hypotheses derived from
the subjectivity approach are supported by a new corpus-analytic study.
1. Causal connectives
The expression of causal relations in language is crucial for

communication. The conceptualization of causality is a fundamental
aspect of human cognition, and without much risk of overstatement it can
be claimed that every language has specific linguistic means to express
causal relations. When language users want to relate two discourse
segments, say clauses, in a causal way, as in (1) below, they can use
causal connectives like English because, so and since" or lexical cue
phrases like English as a result, that's why and on the grounds that.
Although all these linguistic means may express causality in one way or
another, it is clear that they cannot be used interchangeably. In English, it
is perfectly fme to use As a result / That's why / So to connect the
segments of (1), but it is impossible to use As a result to connect the
segments of (2). (This impossibility is expressed by the symbol #, which
signals an uninterpretable sequence; this symbol should not be confused
with *, which signals an ungrammatical sequence.) Also, it is at least
58 Henk Pander Maat and Ted Sanders
doubtful whether That's why could be used in (2), as indicated by the ?,

whereas ~~o would fit neatly.
(1) The neighbours suddenly left for Paris last Friday. As a result /
That's why / So they are not at home.
(2) The lights in the neighbours' living room are out. # As a result / ?
That's why / So they are not at home.
In other words, there are restrictions on the use of a cue phrase and
connectives. Ever since Halliday and Hasan's (1976) seminal work, it has
repeatedly been argued that the lexicon of connectives and cue phrases is
ordered according to the type of relation they can express (e.g. Martin
1992; Knott and Dale, 1994). For instance, temporal connectives can be
distinguished from causal connectives. Yet, as the examples (1) and (2)
show, the restrictions on the use of connectives and cue phrases are also
more subtle, i. e. they also hold within the class of causal relations as such.
Not all causal markers express the same type of causal relation. How then
can these restrictions be described?
In this chapter we report a study on the meaning and use of three
Dutch causal connectives: daardoor, daarom and dus. All three
connectives express causality in a 'forward' direction, that is, cause
precedes consequence. These connectives can best be translated in
English by means of the phrases we used in examples (1) and (2):
daardoor is similar to as a result / as a consequence, daarom can best be
translated as that's why and dus is quite similar to so. In (3) and (4) we
have translated examples (1) and (2) in Dutch.
(3) Ze zijn afgelopen vrijdag plotseling naar Parijs vertrokken.

Daardoor / Daarom / Dus zijn de buren niet thuis.
'They suddenly left for Paris last Friday. As a result / That's why/
So the neighbours are not at home. '
(4) Bet licht in de woonkamer is uit. # Daardoor /? Daarom / Dus zijn

de buren niet thuis.
'The lights in their living room are out. # As a result / ? That's
why/ So the neighbours are not at home. '
All three connectives seem to fit in (3), although they do express different
relations. Daardoor expresses a simple cause-consequence relation in
which the second segment (S2) might even be the unintended result of the
Domains o.fuse and subjectivity 59
first segment (S 1) ("They had to go to Paris"); with daarom the sequence

can be interpreted as a so-called volitional relation in which the first
segment contains the reason for an intentional action in S2 ("Going to
Paris is their reason for leaving home") and with dus S2 is a conclusion
based on S1 ("I conclude they are not there on the grounds that (I know)
they had to leave"). In (4), which can only be interpreted as a conclusion,
dus fits in very well, whereas daardoor leads to an unacceptable
sequence and daarom is at least odd (indicated by a'?').
These three connectives were selected because they are the most
frequently used ones expressing forward causality in Dutch (Uit Den
Boogaart 1975). Second, the differences between them seem to overlap
with distinctions put forward in recently developed theoretical proposals.
For instance, several linguists studying adverbial clauses distinguish
between content and epistemic layers of meaning (Kortmann 1997 for an
overview). Sweetser (1990) introduces a three-level approach to account
for differences in meaning and use of connectives. She distinguishes
between the content, epistemic and speech act domains. This type of
functionally and cognitively oriented approach seems attractive and
promising for the three Dutch causal connectives as well.
In this chapter we will first put such an approach to the test. We
conclude- that a domain theory does not account for our data. Still, the
testing of this hypothesis does give rise to an alternative account in terms
of subjectivity, which we will develop in more detail.
2. The distribution of three Dutch causal connectives:

Domains of use?
2.1. Content, epistemic and speech act domains
Sweetser (1990) has formulated an influential proposal on the

classification of connectives, which we will call the three-domain theory.
She demonstrates ambiguity in the use of conjunctions and claims that her
theory explains these ambiguities. Examples (5)-(7), for instance,
illustrate that English because has three readings: a content reading, an
epistemic and a speech act reading.
(5) John came back because he loved her.
(6) John loved her, because he came back.

(7) What are you doing tonight, because there's a good movie on.
In (5) John's love was the real-world cause of his coming back. (6)
means that my knowledge of John's return leads to the conclusion that he
loved her. An adequate paraphrase of (7) would be "I am asking what
you are doing tonight because there is good movie on". Distinctions like
these have been used by many linguists to account for differences in the
use of connectives and cue phrases (e.g. Halliday and Hasan 1976;
Martin 1992; Knott and Dale 1994), even though there are clear
differences in terminology and definitions (Bateman and Rondhuis, 1997;
Moore and Pollack 1992; Oversteegen 1997; Sanders and Spooren
1999).
How do such distinctions apply to the three Dutch causal connectives
introduced earlier? If we reconsider the two examples, (3) and (4), it was
already noted that in (3) all three connectives seem to fit in, although they
do express different relations. Daardoor expresses a content (Cause-
Consequence) relation, just like daarom (although this connective gives
rise to a volitional, a Reason-Action relation), and with dus the sequence
is of an epistemic nature: S2 is a conclusion based on S1. In (4), which
can only be interpreted epistemically, dus fits in very well, whereas
daardoor leads to an unacceptable sequence and daarom is at least odd.
Hence, at first sight, the domain theory seems relevant for the three
connectives. A first conclusion might be that daardoor can be described
as a connective used exclusively for content relations.
Similar conclusions of domain-specificity are reported for causal
connectives in other languages. For instance, it has been suggested that
German denn can only be used to express epistemic relations (among
others, Giinthner 1993; Keller 1995), just as French donc (Rossari and
Jayez 1996), Hungarian hat (Nemeth 1995) and English since are
supposedly impossible to use in the content domain (Sweetser 1990).
The attraction of the domain theory becomes even more apparent if we
consider the similarity between the domain distinctions and the categories
of coherence relations that have been proposed in cognitively inspired
work on relation categorization. In Sanders, Spooren and Noordman
(1992), for instance, it is shown how the set of coherence relations that
can hold between discourse segments can be decomposed in more basic
categories by making use of only four parameters. One of these
parameters, the Source of Coherence, which distinguishes between
semantic and pragmatic relations, shows great resemblance to Sweetser's
content-epistemic-speech act distinction (Sanders 1997; Knott 2000).
Given the 'basicness' of these conceptual distinctions it is attractive to
Domains ofuse and subjectivity 61
relate them to restrictions on the use of connectives. In such a view, the

lexicon of connectives would reflect the basic distinctions between
coherence relations in a systematic way.
No matter how attractive such a view, the categorization of these
linguistic devices is as yet not based on distributional data from actual
language use. In fact, none of the above ideas on 'domain-specificity' of
connectives and cue phrases are based on authentic occurrences of the
phrases in text corpora. By contrast, recent corpus studies do reveal that
existing categorization proposals, including that by Sweetser (1990),
cannot account for the data of connective distribution in a straightforward
way (Degand 1996 and Pander Maat 1998 and 1999). In an exploratory
cross-linguistic study, Knott and Sanders (1998) did find specific phrases
expressing the semantic-pragmatic distinction, but these tended to be
lexical signals or idiom chunks of the type on the grounds that
(exclusively semantic) and it follows that (exclusively pragmatic). The
most frequently used connectives did not seem to be very specific in
meaning and use.
Therefore, current suggestions on the 'domain-specificity' of
connectives are to be considered as hypotheses to be tested further in
empirical research of language in use. This was exactly the purpose of a
corpus study we conducted (Pander Maat and Sanders 1995): to
investigate to what extent the actual distribution of Dutch daardoor,
daarom and dus can be described in terms of the domains mentioned.
2.2. An empirical test ofthe domain theory
The use of dus, daarom and daardoor was studied in a corpus of 150
fragments, 50 for each connective. The corpus consisted of different text
types, argumentative / persuasive as well as descriptive / informative. It
contained annual reports and publicity brochures of Dutch companies,
public information brochures, articles from the readers' opinion page of a
Dutch quality paper (De Volkskrant 1994) and news articles from this
same paper.
Our methodology involved three steps. We first determined the
possible relational interpretations of these fragments without connectives.
Then we counted how often each connective expressed a certain relation.
In the final step, we checked whether the original connective could be
replaced by another. This substitution method is a way of testing
semantic intuitions (Knott and Dale 1994; Knott and Sanders 1998). The
questions are: Does substitution lead to a sequence which is still
acceptable? And, if acceptable, does the relational interpretation change

as a result of this substitution?
Here is a summary of our findings, formulated in terms of the relations
the three connectives can and actually do express.
• Daardoor can only express relations of the content non-volitional

type.
• Dus can express content volitional, epistemic, summary and
paraphrase relations, but not content non-volitional relations. It most
often expresses epistemic relations.
• Daarom can express content and epistemic relations. It most often
expresses content volitional relations.
Below are prototypical examples from the corpus.
(8) Een campagne in de VS duurt in de eerste plaats zeer lang. Een

kandidaat dient eerst de voorverkiezingen te winnen voordat hij
officieel wordt gekandideerd Een presidentiele campagne duurt
daardoor al gauw anderhalfjaar.
'In the first place, a campaign in the US takes a very long time. A
candidate has to win the pre-elections before he is officially
nominated. Daardoor 'as a result' a presidential campaign easily
takes one and a half years. '
(9) Vaste klanten zijn voor de Bijenkorf van vitaal belang: zij
besteden per jaar twee maal zoveel in de Bijenkorfwinkels als
andere klanten. Daarom heeft de Bijenkorf aan de Vaste
Klantkaart een aantal voordelen verbonden
'Regular customers are of vital interest to the Bijenkorf: in a year
they spend twice as much as other customers do. Daarom 'that's
why' the Bijenkorf has added a number of advantages to the
Regular Customer-Card. '
(10) Er blijkt weer een toename van het aantal besmettingen met het
HIV-virus te zijn. Ais AIDS-verpleegkundige zie ik de gevolgen
ervan. AIDS-preventie, daar komt het dus op aan.
'There appears to be an increase of infections with the HIV-virus.
As an AIDS-nurse, I see the consequences. Dus 'so' prevention of
AIDS, that's what it's all about. '
These results imply that only the use and meaning of daardoor can be
described in terms of relational domains, but even there we need the
additional criterion of volitionality, that is we need to determine whether
the relation concerns a reason for an intentional action or not. Daardoor is
restricted to the content domain, more specifically to relations of the non-
volitional content type. The difference between daarom and dus is
difficult to describe in terms of domains: both connectives regularly
express both kinds of relations. Hence we are faced with the situation that
the purported distinction between content and epistemic causality remains
unmarked in the field of most common Dutch connectives, while the
distinction between volitional and non-volitional relations is marked
linguistically by commonly used causal connectives. Does this imply that
(non-)volitionality has to be considered a more fundamental criterion than
the distinction between content and epistemic meaning?
In proposals for the categorization of coherence relations, volitionality
is indeed considered relevant but it is treated as a second-order distinction
within the content domain relations. For instance, Sanders et al. first
distinguish between semantic (equivalent to content) and pragmatic
(roughly equivalent to epistemic and speech act) relations, and then
separate volitional from non-volitional relations and epistemic relations,
as do Mann and Thompson (1988). Similarly, in the linguistic literature
on connectives and adverbial subordination, the first relevant distinction is
between content and epistemic.
It seems we might have to allow for the idea that, as far as the
typology of causality is concerned, a distinction of the type ["volitional]
may be as fundamental as the one between content and epistemic
domains (Stukker, Sanders and Verhagen 1999). Adding to this the
observation that volitional and epistemic relations are often lexicalized by
the same connectives, it makes sense to ask what the common
denominator in these relations might be.
3. Subjectivity in coherence relations and connectives
So far, the only clear contrast we have seen in the data from actual
language use is the one between daardoor on the one hand, and dus and
daarom on the other: daardoor only expresses non-volitional relations,
while daarom and dus may express both volitional and epistemic
relations. This contrast constitutes a problem for the 'domain theory',
since it cuts right through the content domain. Hence our task is to
develop a conceptualization that adequately captures the nature of the

contrast between non-volitional and other causal relations.
3.1. Subjectivity
The question now is: what is it that 'epistemic' causality has in common
with 'volitional' causality, so that speakers of Dutch easily use the same
vocabulary to express these two relation types, but not to express
relations of non-volitional causality? We want to suggest that what
epistemicity and volitionality have in common is that both crucially
involve an animate subject, a person, whose intentionality is
conceptualized as the ultimate source of the causal event, be it an act of
reasoning or some 'real-world' activity. That is, it seems that to us
humans a very fundamental distinction is the one between events
ultimately originating from some mind, and events that originate from
non-intentional causes; between causes that are crucially located in a
subject of consciousness, and those that are located in the inanimate
world. This distinction is so fundamental that it shows up in similar ways
at different linguistic levels and is often the only one marked explicitly by
means of some linguistic form (Verhagen 1995 and other contributions to
Stein and Wright 1995).
We will use the notion of subjectivity to specify this idea further. In a
sense, every linguistic utterance is subjective, that is, connected to the
point of view of some 'subject'. However, this use of the concept is not
very informative. A more interesting question regarding this notion is
whether this 'subject' is included in the semantic characterization of the
utterance. Langacker (1990), who applies the notion of subjectivity to
several linguistic phenomena, distinguishes between three situations (Pit
1997). First, the ground-the term he uses to refer to the speech event,
its participants and its immediate circumstances-may be entirely
external to the semantics of the utterance. This situation is exemplified by
(11) below. Second, the ground may be included in the scope of
predication as an offstage, unprofiled reference point. This is the case
when deictics like yesterday, tomorrow, etc. are used. Another example
is (12) below, in which the modal adverb probably invokes the present
speaker as the source for the probability judgement. Third, the ground
may be put onstage, as in (13). In this case, the ground is more or less
,objectified', that is, it is made part of the situation referred to in the
utterance.
(11) Peter is in Paris.
(12) Peter is probably in Paris.
(13) I think Peter is in Paris.
In an overview of the different accounts of subjectification and

perspective, J. Sanders and Spooren (1997) refer to sentences (12) and
(13) as instances of subjectification, by which they mean that the present
speaker accepts responsibility for the propositional content of the
utterance, and possibly also for its form. However, a piece of information
may also be connected to another subject ofconsciousness (from now on
SOC) than the present speaker. In this case, Sanders and Spooren speak
ofperspectivization. In (14), Peter is the SOC.
(14) Peter wants to go to Paris.
Normally, perspectivization requires indicators like verbs of cognition,

perception and evaluation, such as want in (14), as has been shown in
recent cognitive linguistic work on perspective and mental spaces
(Fauconnier 1994; Fauconnier and Sweetser 1996; J. Sanders and
Redeker 1996). In the following, we will analyze the role of a subject of
consciousness in relational interpretations. To determine to what degree a
certain coherence relation may be called subjectified or perspectivized,
we need to inspect both the related discourse segments and the relation
itself
3.2. The similarity ofvolitional and epistemic relations
Our task now is to identify the common characteristics of volitional and

epistemic relations. We start, however, by characterizing non-volitional
relations. These relations are presented as strictly objective. Not only does
the relation concern two real world situations, the relation itself is also
construed as an objective fact. Of course the speaker is committed to her
statements (i.e. the speaker cannot subsequently deny them), but the
crucial point is that the speaker is not involved in the events being
described, nor in constructing a causal relation between them. See
fragment (15), which corresponds to the first situation distinguished by
Langacker, in which the ground is external to the semantic representation
of the utterance.
(15) The sun rose. As a result, the temperature went up.
By contrast, consider the following epistemic relation:
(16) It was a small case, completely rusty. So it was made out ofiron.
The causal relation in (16) does not relate real-world situations, but two
beliefs of an SOC (Knott 2000). In this case the SOC is a concluder, that
is somebody making inferences. In (16), the concluder is identical to the
speaker. And the relation between these two beliefs is again situated
within the causal domain of this SOC. Hence, the contrast between non-
volitional and epistemic relations seems rather clearcut in that the
representation of epistemic relations assigns a very prominent role to an
SOC, while such a participant is entirely absent in our interpretation of
non-volitional relations.
As was indicated above, the distribution of causal connectives in
Dutch suggests a similarity of volitional and epistemic relations: dus and
daarom both express volitional and epistemic relations, whereas
daardoor encodes non-volitional relations. How should we conceptualize
this similarity? In our View, the representation of a volitional relation also
involves an SOC, namely the actor of the action in the second segment.
Consider fragment (17).
(17) It was eight 0 'clock. Claire put on the television.
Let us assume a volitional interpretation of (17), in which the time is the

reason for Claire to put on the television. The first thing to note about this
reading is that, although the speaker of (17) is taken to be committed to
the truth of the first segment, the cause for Claire's action is not the
objective fact of the time being eight o'clock, but Claire's mental
representation of this information. Likewise, although the speaker posits
the existence of the causal relation in the outside world, the causal
relation between reason and action is at the same time situated within
Claire's mental domain, since it is constituted by her decision-making
process, a process that she will generally be conscious of
The same goes for the action description in the second segment: this
description is valid both in the real world as conceived of by the speaker
and in the domain of the actor's beliefs. That is, normally describing an
action conveys not only the action taking place in reality as viewed by the
speaker, but also the action which the actor thinks she is performing.
Usually, this duality remains unnoticed, since in both domains the same
action is represented. The following fragment, inspired by Fauconnier

(1994,150-151) illustrates both the possibility of such conflicts and their
counterexpectational nature.
(18a) Calvin is drawing an elephant, but in fact it looks more like an

ant-eater.
(18b) Calvin thinks he is drawing an elephant, but in fact it looks more

like an ant-eater.
(18b) sounds more acceptable than (18a) because a divergence between

speaker perspective and actor perspective is more appropriate when
marked by a perspective indicator (thinks) in the action description itself
To sum up, the representation of volitional (content) relations always
involves both the mental domain of the speaker and the mental domain of
the actor. In general, the speaker adopts the actor perspective and in this
case, the actor is the SOC. In other words, the actor is the default SOC
for volitional relations. This holds for the relation as such, but also for the
segment containing the action.
We have already seen the clear difference in subjectivity between non-
volitional andepistemic relations, which was constituted by the presence
of an SOC in epistemic relations and its absence in non-volitional
relations. In the unmarked case the SOC, the concluding subject, is
identical to the present speaker. However, it is certainly possible to
represent the reasoning of a subject other than the speaker in epistemic
relations, as (19a, b) show.
(19a) The lights in the neighbours' living room were out. So they were
not at home.
(19b) Harry saw that the lights in his neighbours' living room were out.
He concluded they were not at home.
Subsequently we have argued that the presentation of volitional relations

involves an SOC as well, viz. the actor choosing a certain course of action
for a reason referred to in the first segment. This SOC, who is an actor in
volitional relations, is typically present in the second segment. Generally,
there is a certain distance between this SOC and the present speaker.
However, this distance decreases when the actor is referred to by I. The
distance also decreases when the difference between the point in time at
which the action is performed and the moment of speaking is reduced.
For these reasons the distance between the SOC and the present speaker
is larger in (20b) than in (20a).
(20a) / am tired, so / am leaving.
(20b) He was tired, so he left.
In conclusion, volitional and epistemic relations are similar in that their

representations involve an SOC. In both relations this SOC may be more
or less distant to the present speaker. However, epistemic relations seem
to have a built-in preference for speaker SOCs, a preference lacking in
volitional relations.
3.3. 'Dus ' and 'daarom ' differ in terms of subjectivity
In section 2 it was found that, although daarom and dus may both
express volitional and epistemic relations, dus is much more frequent in
epistemic relations than it is in volitional relations, while daarom shows a
slight preference for volitional relations. The hypothesis we want to
explore below is that this differential distribution is to be explained by a
difference between the two connectives with regard to the distance they
express between the SOC involved and the present speaker. More
specifically, our claim will be that daarom encodes a larger distance than
dus does. We will now derive some specific predictions from this general
claim, so that they can be put to the test in a corpus study.
Our first prediction presents itself in a rather straightforward way.
When dus really encodes a smaller distance between the SOC and the
speaker, the subject responsible for the action or the conclusion in dus-
segments should more often be identical to, or close to the speaker than
the actors and concluders in daarom-segments are. In order to test this
claim, we distinguish between four types of sacs, ranging from minimal
(1) to maximal (4) distance between the SOC and the speaker.
(1) Implicit speaker SOC included in the semantic representation of the

utterance.
Example: Baking cakes is fun.
(2) Explicit speaker SOC, referred to by /; this includes cited speakers.

Example: / bakedfour cakes or "/ bakedfour cakes", Calvin said.
(3) Explicit pronominal or nominal third-person SOC.

Example: He / Calvin bakedfour cakes.
(4) Contextually recoverable third-person participants invoked by a

passive construction.
Examples: Calvin loved baking cakes. So every birthday at least
four were baked (recoverable actor in the second segment).
The SOC-speaker distance is smallest for implicit speaker sacs.

Following Langacker (1985, 1990), we view explicit speaker references
as expressing a certain distance between the present speaker and the first-
person participant referred to in the utterance. Clearly, third-person sacs
are further removed from the speaker than first-person ones. Finally,
recent work on the passive shows that this construction indicates non-
identification with the actor, when compared with the corresponding
active construction (Comelis 1997). Therefore, implied agents in passive
constructions are placed at the bottom of our list: there is maximal
distance between the SOC and the speaker.
Our second prediction needs some additional introduction. As stated
earlier, we claim that dus prefers speaker sacs. If this is correct, dus-
fragments with third-person sacs would need additional marking,
indicating a non-default soc. Consider the following example.
(21) Het sloeg vijfuur. Dus pakte Frits zijn aktentas.

'The clock struck five. Dus 'so' Fritz took his briefcase.'
In this example, there are two possible interpretations. In the first one,
Fritz decides to go home on the basis of his knowledge of the time and his
knowledge of the need to go home at five. Under this interpretation, Fritz
is the SOC. In the second interpretation, the speaker presents Fritz'
behavior as an habitual action: always when the clock strikes five, Fritz
forgets all about his work and gets ready to leave. Under this
interpretation, the speaker is the SOC, because Fritz does not necessarily
consciously decide to leave work, he just does what he does every day.
If the segments in example (21) are connected by dus, the second
interpretation is preferred. If we want listeners to construct the first
interpretation, and we still use dus, we would need to mark the first
segment as being presented from the perspective of the non-default SOC:
the actor, not the speaker. Examples containing such non-default SOCs
are (21a) and (21b). In (21a) the SOC is explicit in the first segment; in
(21 b) it is contextually available on the basis of a segment m the

preceding discourse.
(21 a) Frits hoorde de klok vijfuur slaan. Dus pakte hi} zi}n aktentas.
'Fritz heard the clock striking five. Dus 'so' he took his briefcase.'
(21b) Frits keek op zi}n horloge. Het was al laat. Dus pakte hi} zi}n
aktentas.
'Fritz looked at his watch. It was late already. Dus 'so' he took his
briefcase. '
If this line of reasoning is correct, our prediction should be that in

comparison with volitional daarom-fragments, volitional dus-fragments
more often have a first segment in a third-person actor perspective or
contextually imply such an actor. However, this prediction only applies to
fragments with third-person actors: when the actor is the speaker, the first
segment is in the speaker-actor perspective by default, even when this is
not made explicit. The prediction of perspective continuity in third-person
volitional dus-fragments should be valid for third-person epistemic
fragments too. However, since third-person epistemic fragments are rare,
it remains to be seen whether the continuity hypothesis can be tested for
epistemic relations.
These are then the two hypotheses to be tested in corpora.
1. The SOC-speaker distance hypothesis: In dus-fragments, the distance
between SOC and speaker is smaller than in daarom-fragments.
2. The continuity hypothesis: In third-person dus-fragments, the first
segment will appear more often in the perspective of the actor (in
volitional relations) or concluder (in epistemic relations) than in third-
person daarom-fragments.
4. Corpus analysis
4.1. Corpus
25 dus- and 25 daarom-fragments were selected at random from a

corpus of a Dutch quality newspaper, the Volkskrant (1995). The corpus
was relationally stratified in the sense that the number of volitional and
1
epistemic relations for each of the connectives was equal. Three types of
fragments were not included in the corpus.
- Fragments with another connective immediately following dus or

daarom (e.g. dus omdat ... )
- Fragments in which the connective did not connect full sentences (e.g.
"Een mooie dus dure boot"/ 'A beautiful therefore expensive boat')
- Daardoor-fragments in which the connective did not refer to the
preceding segment as a whole, but rather to parts of it.
4.2. Results: volitional relations
For reasons of exposition, we start with the results regarding the second
hypothesis, the continuity hypothesis. In the analysis, we determined the
perspective of the first segment in the third-person volitional relations for
both connectives (daarom: n=2I; dus: n=I4). The results (see Table 1)
show that the proportion of fragments characterized by an explicitly or
implicitly continuous perspective for dus is significantly larger than that
2
for daarom (Chi =15.26, p < .001, df=2).
Table 1. The perspective of the first segment in third-person volitional daarom-

and dus- fragments: frequencies and column percentages.
81 has: daarom dus

Explicit actor perspective 10 (48%) 9 (640/0)
Contextually implied actor perspective 5 (36%)
Non-actor perspective (discontinuous) 11(52%)
Total 21(1000/0) 14 (1000/0)
Fragment (22) is a typical continuous dus-fragment, where he is both the

actor in segment (2a) (where segments are rougWy equivalent to clauses)
and the participant responsible for the evaluation in (Ib) and (Ic).
(22) (Anna Enquist debuteerde in 1991 met Soldatenliederen. Daar

had Theo Sontrop voor gezorgd. )
(Ia) Ze had eens een paar gedichten naar Maatstaf gestuurd, (Ib) en
daarover was hi} laaiend enthousiast geweest-(lc) vond haar
poezie meteen af (2a) Dus hi} schreefhaar meteen ofze nog meer
had, (2b) en dat werd toen die bundel.
'(Context: "She" refers to Anna Enquist, a Dutch writer. "He"
refers to Theo Sontrop, the Dutch editor of a literary journal called
Maatstaf[Criterion].)
72 Henk Pander Maa! and Ted Sanders
(la) She had sent some poems to Maatstaj, (lb) and he had loved
them-(lc) found her poetry had an immediate perfection. (2a)
Dus 'so' he wrote to ask her if she had any more (2b) and this
became her first collection of poems. '
The discontinuities in volitional daarom-relations predominantly occur

when the first segment of the relation contains the judgement of the
author. In (23), for instance, the author's judgement in (3) is marked by
"what is certain is that", whereas China is the actor in (4).
(23) (Is Noord-Korea bezig een kemwapen te ontwikkelen?)

(la) Voor aile duidelijkheid: (lb) het bewijs dat Noord-Korea met
dit soort snode plannen bezig is, is niet geleverd. (2a) De
Amerikaanse inlichtingendiensten geloven van wei, (2b) de
ministeries van Buitenlandse Zaken en Defensie in Washington
houden zich op de vlakte. (3) Vast staat dat een in een het nauw
gedreven No 0 rd-Korea met atoomwapens in Azie een
kemwapenwedloop zal veroorzaken. (4) Met name de regionale
kernmogendheid China spant zich, op Amerikaans aandringen,
daarom in de oude bondgenoot in te tomen.
'(Is North Korea working on the development of a nuclear
weapon?)
(1a) For the sake of clarity: (1b) the evidence that North Korea is
working on these infamous plans is not available. (2a) The
American intelligence agencies believe they are doing so; (2b) the
Department of Foreign Affairs and the Defense Ministry in
Washington remain silent. (3) What is certain is that a hard-pressed
North Korea can cause a nuclear arms race in Asia. (4) Daarom
'that's why' the regional nuclear power China, following American
suggestions, is making great efforts to restrain the former ally'.
As Table 1 shows, dus-fragments regularly have contextually recover-

able third-person participants, i.e. the participant is inferred on the basis
of the discourse environment. An example is fragment (24).
(24) (1) Maar al te vaak wordt het beeld van het buurland bepaald
door de media. (2a) Berichten over positieve zaken zijn geen
nieuws, (2b) dus concentreert men zich op het negatieve.
'(1) All too often the image of a neighbouring country is
determined by the media. (2a) Information about positive things is
not news, (2b) dus 'so' one concentrates on what is negative. '
In the volitional interpretation of (24), representatives of the media are

taken to be responsible for the evaluation ('that something is not regarded
as news') in (2a). That is, the participant of (2a) is to be recovered from
the preceding sentence. The author may very well disagree with (2a) and
disapprove of the action in (2b), which is motivated by it. Given the larger
context (a letter to the editor by the German ambassador in the
Netherlands pleading for mutual understanding between neighbouring
countries), this is the only plausible interpretation. This concludes our
discussion of the continuity hypothesis for now.
The results for the SOC-speaker distance hypothesis show that dus-
actors are indeed closer to the author than daarom-actors, see Table 2, in
which the possible participants are ordered from minimal to maximal
distance to the speaker (see section 3.3 for this identification hierarchy).
In the majority of cases, the speaker is the dus-participant, where the
speaker can be the author of the fragment as well as the cited speaker,
and usually appears in the text as I (occasionally we). By contrast,
daarom-participants are predominantly realized as a (nominally
indicated) third person. The difference in Table 2 is significant (Chi2 =
10.99, P < .05 one-tailed, df = 2), thereby supporting the second
hypothesis.
Table 2. The identity and realization of actor-participants in the volitional

daarom- versus dus-fragments: frequencies and column percentages.
Identity and realization daarom dus

Speaker explicit 4 (160/0) 11 (480/0)
Third person 13 (560/0) 9 (360/0)
Pronominal 2 6
Nominal 11 3
Contextually recoverable third-person agent 8 (32%) 5 (20%)
Total 25(100%) 25(100%)
We will now illustrate the data regarding the SOC-speaker distance

hypothesis. In daarom-fragment (23), the actor in (4) is a third-person
nominal (China). By contrast, dus-fragments more often have first-person
actors, as in (25). And in (24), there is a third-person participant
recoverable from the context.
(25) "In '68 is de cultuur daar verbannen en gedegradeerd tot 1,3

miljard spugende Chinezen die geen enkel gezag meer kunnen
opbrengen voor schoonheid. Zakken afval en stront gaan zo het
raam uit. Dat gaat mij te ver, dus neem ik de tweede klas ... "
'''In ' 68 culture was banned from there and degraded to 1,3 billion
spitting Chinese who cannot bring themselves to any respect for
beauty. Bags full of trash and shit simply go out of the window.
That is too much for me, dus 'so' I take the second class. '"
As shown in Table 2, dus and daarom show a clear difference with

respect to pronominal or nominal references to third-person actors:
daarom prefers nominal actors, while dus prefers pronominal actors (e.g.
fragment PZ», the difference being significant (Chi2 = 6.05, P < .05
two-tailed , df = 1). This indicates that dus-actors tend to have a higher
level of accessibility (Ariel 1988) than daarom-actors. This matches with
the fact that dus-actors are more often responsible for the first segment
than daarom-actors.
4.3. Results: epistemic relations
The SOC-speaker distance hypothesis was also investigated for epistemic

relations, with a method identical to the volitional study. The results are
presented in Table 3.
Table 3. The identity and realization of concluder-participants in the epistemic

daarom- versus dus-fragments: frequencies and column percentages.
Identity and realization daarom dus

Speaker implicit 14 (56%) 25 (100%)
Speaker explicit 2(12%)
Third-person nominal 9(32%)
All dus-participants are implicit speakers, while in daarom-connections

the second segment also contains explicit speakers and (nominally
indicated) third-person participants. There is clearly more distance
between participants (concluders) in the second segment and the author
for the daarom-fragments than for dus-fragments (Chi2 = 10.98, df= 1, p
< .001, two-tailed), thereby supporting the second hypothesis.
Since there are no third-person epistemic relations containing dus, we
cannot test the continuity hypothesis for epistemic fragments. Instead, we
carried out an exploratory study of the subjectivity relations between both
segments for the entire corpus of epistemic relations. For the purposes of
this analysis, all elements indicating deontic (have to) or epistemic
modality (must, I think, probably) and other lexical elements with
evaluative meanings (e.g. important, surprising) were considered as

indicators of subjectivity (Lemke 1998).
The results of this analysis are summarized in Table 4; illustrative
fragments follow below. Daarom and dus show different perspective
configurations (Chi2 = 11.60. df= 3, P < .01)
Table 4. Subjectivity configurations in epistemic daarom versus dus-fragments.
Configuration daarom dus

SI subj - S2 subj (continuous) 15 (600/0) 15 (60%)
SI subj-l - S2 subj-2 (discontinuous) 3 (12%)
SI obj - S2 subj 7 (280/0) 3 (12%)
SI obj - S2 obj 7 (28%)
Total 25 (100%) 25 (100%)
S1, S2 refer to the first and second segments connected by a connective.

Subj means Subjective and obj means Objective.
In the first and most frequent configuration, the same partIcIpant is

responsible for both segments. This is illustrated by fragment (26) (first-
person SOC) below.
(26) (1a) Drugs verwoesten mensenlevens, (lb) dus moeten drugs

strafrechtelijk bestreden worden.
'(1a) Drugs ruin human lives, (1b) dus 'so' drugs must be
combatted by means of criminal law. '
The second configuration is discontinuous and occurs only in daarom-

fragments. In fragment (27) the first sentence clearly adopts the
perspective of the Russian citizen, while the author takes full
responsibility for the second segment.
(27) (1a) Gewone burgers voelen zich kwetsbaar, (1b) overgeleverd

aan gangsters, gewone misdadigers, corrupte bureaucraten en de
mafia-ric) terwijl hooghartige bureaucraten hun zakken vullen.
(2) Daarom is het niet tegenstrijdig dat de kiezers zich uitspraken
voor de nieuwe grondwet en tegelijkertijd voor Zjirinovski
stemden.
'(1a) Ordinary citizens feel vulnerable, (1b) being at the mercy of
gangsters, common criminals, corrupt bureaucrats and the mob-
(Ic) while arrogant bureaucrats fill their pockets. (2) Daarom
'that's why' it is not contradictory that the voters voted for the new
constitution and at the same time for Zjirinovski. '
In the third configuration the concluder draws a conclusion on the basis of

an objective, factual statement: see (28). However, the most pronounced
difference between the epistemic daarom- and dus-fragments is found in
second segments which follow objective first segments. Daarom-
fragments with a factual S 1 always have an evaluation as their second
segment, as in (29). By contrast, dus-fragments more often contain
objectively formulated second segments, as in (30).
(28) (1) De VPRO-quiz was echter een open quiz. (2) Het stond
iedereen vrij om allerlei feiten op te zoeken in encyclopedie of
bibliotheek. (3) Of om overleg te plegen met familie, vrienden of
collega's. (4) Het hebben van een hoge score impliceert dus niet
noodzakelijkerwijs dat men ook veel verstand van wetenschap
heeft.
'(1) However, the VPRO-quiz was an open quiz. (2) Everybody
was allowed to look things up in an encyclopedia or a library. (3)
Or to consult family, friends or colleagues. (4) Dus 'so' having a
high
.
score
,
does not necessarily mean that one knows a lot about
SCIence.
(29) (la) Hi} had ervaring op die circuits, (lb) ik niet. (2) Hi} maakte
daarom theoretisch meer kans om punten te halen.
'(la) He was experienced on those circuits, (lb) I was not. (2)
Daarom 'that's why' in theory he had a better chance to score
points. '
(30) (1) Het punt is echter dat de verdachten die schuldjuist met klem
ontkennen en daarom bi} het hof hoger beroep hebben
aangetekend tegen hun veroordeling bi} de rechtbank. (2) Die
strafis dus nog niet definitief
'(1) The issue, however, is that the suspects vehemently deny their
guilt and have lodged an appeal against their conviction by the
court. (2) Dus 'so' this sentence is not final yet.'
Apparently, daarom is incapable of conveying an epistemic interpretation

in an entirely factual discourse passage. By itself: it fails to convey
subjectivity in the interpretation of the second segment and of the relation
itself.
This result fits very well with our account, which claims that dus and
daarom encode different speaker-SOC distances. In some contexts, no
SOC is present, either explicitly or implicitly (in the form of, for instance,
subjective elements). In these situations the use of dus is unproblematic.
The reason is that the processing instruction constituted by dus, i.e. 'find
an SOC close or identical to the speaker', can always be carried out
successfully because every utterance presupposes the presence of a
speaker. However, the instruction encoded by daarom, i.e. 'fmd an SOC
at a certain distance from the speaker' presents problems in situations
where the utterance itself does not present an SOC at all. After all,
selecting the speaker is impossible. In order to support selection of the
speaker, daarom needs linguistic elements indicating his presence,
though not referring to him as such.
5. Conclusion
In this paper we have adopted an empirical, corpus-analytic approach to

investigate the meaning and use of three frequent Dutch causal
connectives. It can be concluded that the lexicon of connectives cannot be
described in terms of distinctions well-known from accounts like
'domains' (Sweetser 1990) or layers of meaning, such as content,
epistemic and speech act.
Alternatively, we have proposed considering the difference between
the connectives in terms of subjectivity or, more precisely, in terms of the
distance between the speaker :md the SOC responsible for the causal
relation. In such a subjectivity account, the three connectives can be
characterized as follows. In causal relations expressed by daardoor 'as a
result' there is no SOC present because the causality is located outside of
this SOC. In relations expressed by daarom 'that's why' there is a certain
distance between speaker and SOC, whereas in the cases of dus 'so', this
distance is small or even absent. We have shown that this approach
accounts for the actual distribution of the connectives in newspaper
corpora.
The subjectivity approach may also have.further advantages. For one
thing, it points to the commonality between connectives on the one hand
and other linguistic phenomena which have been treated in terms of
concepts similar to subjectivity. For instance, it has long been recognized
that deictic reference is essentially relational in nature, in that it does not
primarily characterize the referent but the relation between the referent
and the contextual framework of the present utterance (Hanks 1992). The
relational dimensions at issue include the distance between the referent

and the speaker, both spatially (e.g. this vs. that; here vs. there) and
temporally (now vs. then), and the degree of active involvement in the
utterance event (1 vs. you vs. he/she). If the SOC of the causal relation
can be considered a kind of referent, the difference between dus and
daarom is conceptually parallel to the one between, say, here and there.
Alternatively, dus and daarom may be taken to encode different degrees
of speaker involvement in the causal relation (Degand and Pander Maat
1999).
Another phenomenon which has been treated in terms of subjectivity is
modality (Lyons 1977: 797-809; Coates 1983: 32-37; Nuyts 1992). In a
corpus study comparing Dutch modal adverbs (English gloss: probably)
to constructions involving modal adjectives (probable), Nuyts (1994: 45-
52) found that adverbs can only be used to express modal qualifications
"performatively", that is as expressing an epistemic commitment of the
present speaker, while adjectives may be used both performatively and
"descriptively", that is as reporting objective probability assessments or
epistemic evaluations by other people.
In other words, a distinction in terms of subjectivity is not only
lexicalized in the field of connectives but also in other semantic domains
like deixis and modality. In all three fields, speakers have to determine
their position vis-a-vis, or their involvement with, the entities and
propositions presented in their discourse. Language, then, seems to show
a systematic sensitivity to this expressive need.
Notes
* We wish to thank Bernd Kortmann, Elizabeth Couper-Kuhlen and Alistair

Knott for comments on an earlier version of this paper, and Alistair Knott
for correcting our English. Needless to say, all remaining errors are ours.
1. Our claim is that the difference between dus and daarom does not lie in the
intrinsically volitional or epistemic meaning of the connectives. So we
should be able to predict differences between the two connectives occurring
within the classes of volitional and epistemic relations respectively.
2. The difference between pronominal and nominal references to third-person
actors for dus and daarom was tested by a two-tailed test, since there was no
hypothesis concerning this difference.
References
Ariel, Mira
1988 Referring and accessibility. Journal ofLinguistics 24: 65-87.
Bateman, John A. and Rondhuis, Klaas-Jan
1997 Coherence relations: towards a general specification. Discourse
Processes 24: 3-49.
Coates, Jennifer
1983 The semantics ofmodal auxiliaries. Beckenham: Croom Helm.
Cornelis, Louise H.
1997 Passive and perspective. Amsterdam!Atlanta: Rodopi.
Degand, Liesbeth
1998 Het ideationele gebruik van want en omdat. [The Ideational Use of
want and omdat]. Nederlandse Taalkunde 4: 309-326.
Degand, Liesbeth and Henk Pander Maat
1999 Scaling causal relations in terms of speaker involvement. In: Levels
of Representation in Discourse, 45-53. Working notes of the
International Workshop on Text Representation. University of
Edinburgh.
Fauconnier, Gilles
1994 Mental spaces. Cambridge: Cambridge University Press.
Fauconnier, Gilles and Eve Sweetser (eds.).
1996 Spaces, worlds and grammars. Chicago: University of Chicago
Press.
Gtinthner, Susanne
1993 "...weil-man kann es ja wissenschaftlich untersuchen"-
Diskurspragmatische Aspekte der WortstellUng in WEIL-Satzen.
Linguistische Berichte 143: 37-59.
Halliday, M.A.K. and Ruqaiya Hasan
1976 Cohesion in English. London: Longman.
Hanks, William F.
1992 The indexical ground of deictic reference. In: Alessandro Duranti
and Charles Goodwin (eds.) Rethinking context. Language as an
interactive phenomenon, 43-76. Cambridge: Cambridge University
Press.
Keller, Rudi
1995 The epistemic weil. In: Dieter Stein and Susan Wright (eds.)
Subjectivity and subjectivisation. Linguistic perspectives, 16-30.
Knott, Alistair and Robert Dale
1994 Using linguistic phenomena to motivate·a set of coherence relations.
Discourse Processes 18: 35-62.
Knott, Alistair and Ted Sanders
1998 The categorization of coherence relations and their linguistic
markers: an exploration of two languages. Journal of Pragmatics
30: 135-175.
Knott, Alistair
2000 Semantic and pragmatic relations and their intended effects. In:
Ted Sanders, Joost Schilperoord and Wilbert Spooren (eds.),
Cognitive approaches to coherence. Amsterdam: Benjamins.
Kortmann, Bernd
1997 Adverbial subordination. A typology and history of adverbial
subordinators based on European languages. BerlinlNew York:
Mouton de Gruyter.
Langacker, Ronald W.
1985 Observations and speculations on subjectivity. In: John Haiman
(ed.), Iconicity in syntax, 109-150. Amsterdam/Philadelphia:
Benjamins.
1990 Subjectification. Cognitive Linguistics 1: 5-38.
Lemke, Jay L.
1998 Resources for attitudinal meaning. In: Functions ofLanguage 5: 33-
56. Amsterdam: Benjamins.
Lyons, John
1977 Semantics 2. Cambridge: Cambridge University Press.
Mann, William C. and Sandra A. Thompson
1988 Rhetorical structure theory: Toward a functional theory of text
organization. Text 8: 243-281.
Martin, J. R.
1992 English text. System and structure. Philadelphia: Benjamins.
Moore, Johanna D. and Martha E. Pollack
1992 A problem for RST: The need for multi-level discourse analysis.
Computational Linguistics 18: 537-544.
Nemeth, Enik6
1995 On the role of pragmatic connectives in Hungarian spoken
discourses. In: B. Warvik, S-K. Tanskanen and R. Hiltunen (eds.),
Organization in discourse. Proceedings from the Turku conference,
393-402. Anglicana Turkuensia 14.
Nuyts, Jan
1994 Epistemic modal qualifications: on their linguistic and conceptual
structure. Antwerp Papers in Linguistics 81. Antwerpen:
Universiteit Antwerpen.
1992 Subjective vs. objective modality: what's the difference? In:
Fortescue, Michael D., Peter Harder and Lars Kristoffersen (eds.),
Layered structure and reference in a functional perspective, 73-99.
Amsterdam: Benjamins.
Oversteegen, Leonoor E.
1997 On the pragmatic nature of causal and contrastive connectives.
Pander Maat, Henk
1998 Classifying negative coherence relations on the basis of linguistic
evidence. Journal ofPragmatics 30: 177-204.
1999 The differential linguistic realization of comparative and additive
coherence relations. Cognitive Linguistics 10: 147-184.
Pander Maat, Henk and Ted Sanders

1995 Nederlandse causale connectieven en het onderscheid tussen
inhoudelijke en epistemische coherentie-relaties. [Dutch causal
connectives and the distinction between content and epistemic
coherence relations.] Leuvense Bijdragen 84: 349-374.
Pit, Mirna
1997 Subjective and objective construal in coherence relations. Jan Don
and Ted Sanders (eds.), Utrecht Institute of Linguistics OTS
Yearbook 1997, 45-71. Utrecht: UiL OTS.
Rossari, Corinne and Jaques Jayez
1996 Donc et les consecutifs. Des systemes de contraintes differentiels.
Lingvisticte Investigationes XX: 117-143. Amsterdam: Benjamins.
Sanders, Jose and Gisela Redeker
1996 Perspective and the representation of speech and thought in
narrative discourse. In: Gilles Fauconnier and Eve Sweetser (eds.),
Spaces, worlds and grammars, 290-317. Chicago: University of
Chicago Press.
Sanders, Jose and Wilbert Spooren
1997 Perspective, subjectivity and modality from a cognitive linguistic
point of view. In: Wolf-Andreas Liebert, G. Redeker and L. Waugh
(eds.), Discourse and perspective in cognitive linguistics, 85-112.
Sanders, Ted
1997 Semantic and pragmatic sources of coherence. On the categorization
of coherence relations in context. Discourse Processes 24: 119-147.
Sanders, Ted lM., Wilbert P.M.Spooren, and Leo G.M. Noordman
1992 Toward a taxonomy of coherence relations. Discourse Processes 15:
1-35.
Sanders, Ted and Wilbert Spooren
1999 Communicative intentions and coherence relations. In: W. Bublitz,
U. Lenk and E. Ventola (eds.), Coherence in spoken and written
discourse, 235-250. Amsterdam/Philadelphia: Benjamins.
Stein, Dieter and Susan Wright (eds.)
1995 Subjectivity and subjectivisation. Linguistic perspectives.
Stukker, Ninke, Ted Sanders and Arie Verhagen
1999 Waar een wil is, is geen wet. De categorisering van causale relaties
binnen en tussen zinnen. [When there is a will, there is no law. The
categorization of causal relations within and between sentences.]
Gramma / TIT 7: 66-86.
Sweetser, Eve E.
1990 From etymology to pragmatics. Cambridge: Cambridge University
Press.
Uit den Boogaart, Pieter C.
1975 Woordfrequenties in geschreven en gesproken Nederlands. [Word
frequencies in written and spoken Dutch. ] Utrecht: Oosthoek,
Scheltema and Hoekema.
Verhagen, Arie
1995 Subjectification, syntax and communication. In: Dieter Stein and
Susan Wright (eds.), Subjectivity and subjectivisation, 103-128.
Causal relations in spoken discourse: Asyndetic
constructions as a means for giving reasons*
Christine Gohl
When speakers produce utterances offering reasons for a previous

utterance, they may do so either explicitly by linking these utterances, for
example, with a conjunctive element, or implicitly, i.e. by means of
asyndetic linkage. The analysis of asyndetically linked constructions poses
several problems regarding their interpretation as causally related
utterances. The lack of a conventionalized element signaling the causal
relationship raises questions concerning what counts as a causal relation
from the participants' perspective and how analysts can identify and
describe these unmarked constructions. After a discussion of general issues
with respect to asyndetic causal constructions, different environments in
which these occur are investigated. It will be argued that these
environments, constituted by different actions which call for accounting,
constrain the interpretation of an asyndetically linked construction.
1. Introduction
Earlier research on causal relations has dealt mainly with explicit lexical
markers, above all with subordinating conjunctions. More recent studies
on causal relations in spoken English draw attention to the paratactic use
of because (e.g. Ford 1994; Ford and Mori 1994; Couper-Kuhlen 1996;
Schleppegrell 1991), while studies on causal conjunctions in spoken
German focus, for example, on weil ('because') and the use of main
clause constructions in weil-clauses (e.g. Giinthner 1996; Pasch 1997;
Scheutz 1998; Uhmann 1998). This focus on the use of conjunctions
reflects the fact that they are indeed an important means for signaling
causal relations. That is not to say, however, that they are the most
frequent means of signaling causal relations in spoken interaction.
Especially in informal talk, a frequent device is the use of asyndetic
linkage. In these cases, speakers produce a reason for a previous action
without explicitly linking (e.g. via a causal connector) the utterance that
carries out the action to the utterance that is used to provide the account.
Although one can find various remarks in the literature on such asyndetic
causal structures (e.g. Blakemore 1987: Ch. 4.2.), they have not been
84 Christine Gohl
investigated empirically so far. In particular, there is a lack of research on

the use of such constructions in spoken interaction.
In this paper, I shall analyze asyndetic causal constructions in spoken
German discourse. This involves dealing with the following questions:
(1) What is a construction, i.e. in which cases can two (or more)
utterances be considered a structure in the sense that they form a
cohesive relation? (section 2)
(2) How is this cohesiveness signaled? How can we distinguish
adjacent utterances with no specific coherence relation from
adjacent utterances with a specific, for example, a causal relation?
(section 2)
(3) What counts as a causal relation in spoken interaction? How does
the term 'causal relation' relate to the notion of 'account' as it is
used in conversation analysis (e.g. Heritage 1984: 250-251, 266-
273)?1 (section 3)
(4) If there are no conventionalized elements present to signal the
causal relationship-how is it still possible to identify asyndetic
constructions as structures involving a causal relation, and how
can we arrive at a plausible interpretation of these constructions?
(sections 4.1-4.6)
With regard to (4), the analysis of the data will show that accounts
typically occur after particular actions. These actions, it will be argued,
2
establish constraints on the interpretation of an asyndetic construction.
2. Signalling connectedness in asyndetic constructions
Cohesiveness between utterances is typically signaled in some way.

Utterances that form a cohesive unit-also termed 'construction' in the
following-are connected either by explicit lexical marking or by more
implicit means, such as prosody. Dealing with asyndetic constructions,
i. e. with structures that are not connected by a grammaticalized element,
for example, a conjunction, an adverb, or a preposition, raises the
question: how is connectedness signaled in these cases?
A first indication as to the connectivity of asyndetically linked
utterances is their sequential proximity.3 By way of illustration, consider
excerpt (1). The participants are arguing about smoking: Erik and
somebody else want to smoke. Some other family members, however, are
still busy eating.
Causal relations in spoken discourse 85
(1) SMOKING (Schwab 4; 49:52)

lMaria: < <h> uwele ma: gsch no () Iguck da hat==s no
knoch[ele zum abnage. >
2Erik: [nix kriegt er me: mir musset rauche. ()
3?Erik: zum wohl. ((Anstoj3en von Gliisern))
4?: [(4 Silben)]
5Anna: fda kommt me tota:l untererniihrt kommt mer hoim,
6 bloj3 weil die rauche wollet.
, 1Maria: uwe dear do you want anything else (.) look there' re
bo[nes to pick. 4
2Erik: [he'll get nothing more we have to smoke. (.)
3 ?Erik: cheers. «clinking of glasses))
4?: [(4 syllables)]
5Anna: [we'll come home totally undernourished,
6 just because these guys want to smoke. '
The tum produced by Erik in line 2, nix kriegt er me: mir musset rauche
'he'll get nothing more we have to smoke', consists of two successive
tum-constructional units which together form a cohesive unit or
construction. The cohesiveness between these two units is of course
supported by the fact that a causal relation can be construed between
them. Yet this causal relation would not be sufficient to create a
construction, were the two units not in sequential proximity.
However, it is not only units produced by one speaker, as in the case
above, which may form a construction, but also units produced by two or
even more than two speakers. The latter case is rare in my data, but it
does exist (cf (14) CAT II below). The interactive production of
constructions by two speakers occurs more often (cf (10) CAR below).
Yet the majority of causal constructions in my data are produced by a
single speaker.
Another means of signaling connectedness is via the prosodic design
of two utterances, in particular the pitch level at the end of the first
utterance. There are two possibilities for signaling connectivity: (i) the
two utterances are produced in one intonation unit, (ii) the two utterances
are produced as two intonation units with the cadence of the first unit
signaling continuity. 5 The latter is found most often in my data in cases
where a causal construction is produced by a single speaker. In these
instances, the fmal pitch movement of the first unit does not reach the
bottom of a speaker's pitch range, i.e. does not end in 'period intonation'
86 Christine Gohl
(Chafe 1988: 2) (transcription symbol: .). It either just slightly falls

(transcription symbol: ;) or rises (transcription symbol: ,) or stays level
(transcription symbol: -), i.e. ends with 'comma intonation' (Chafe 1988:
2). Although in most cases the causal constructions in my data comprise
two intonation units, there are also cases in which a single speaker
produces both parts within one intonation unit. Excerpt (1) SMOKING
illustrates this. The speaker uses a single intonation phrase for the
production of the two units making up the asyndetic construction: nix
kriegt er me: mir musset rauche 'he'll get nothing more we have to
smoke'. The second part, mir musset rauche., justifies the action encoded
in the first part nix kriegt er me:, i.e. Erik's rejection of Maria's offer to
Uwe.
In excerpt the speaker uses two intonation units for the production of
one construction. The participants in this segment are the same ones as in
(1). They are talking about going to church after having had their
Christmas dinner. It is quite late in the evening.
(2) CHURCH (Schwab 3; 1:31:39)

Erik: gott sei dank muft 11 () < <all> Jheut nimme in(d) larch. >
i ddt schnarche.
(1.0)
Uwe: des isch ebe des was [(5 Silben)
Kai: [da: kommsch eh net
zum schnarche -
mir werdet eh schtande. ()
Maria: (nei)nei mir ganget so friih daft me net
schtandet.
'Erik: thank god I don't have to go to church today anymore.

I would snore.
(1.0)
Uwe: that's what [(5 syllables)
Kai: [you won't get round to snoring
there anyway -
we'll have to stand anyway.
Maria: (no)no we'll go so early that we won't have to stand.'
Kai's utterance contains a construction consisting of two intonation units.

The first part of the construction (da: kommsch eh net zum schnarche-
'you won't get round to snoring there anyway-') is accounted for in the
next intonation unit (mir werret eh schtande 'we'll have to stand

anyway').
A further hint as to the connectedness of two (or more) utterances is
the use of particular modal particles. In my data, particles that occur in
the second utterance are, above all, doch, ja, halt and eben. The literature
on German modal particles is abundant, and there is as much agreement
as disagreement on the meanings and functions of particular particles.
Generally speaking, there are two main functional components of modal
particles according to the literature (Hartmann 1986): (1) they signal the
speaker's attitude towards the proposition of the utterance; (2) they
6
connect the present utterance with the preceding utterance. The latter
point is much less controversial than the former and is the one which
concerns us here: the modal particles that occur in my data help to signal
the causal relationship, not in the sense that they convey a causal
meaning, but that they underline the connection between the units in
7
question (cf example (14) CAT II below).
8
The collection of asyndetic causal constructions in my corpus includes
constructions which exhibit at least one of the features discussed in this
section, i.e. sequential proximity, a single intonation contour or continuing
intonation at the end of a first contour, or the use of a modal particle.
Usually at least two of these characteristics occur together. Sequential
proximity, for example, usually co-occurs with the use of continuing
intonation or the use of a particle or both.
I will now specify my use of the terms 'causal relation' and 'account'
and come back to excerpts (1) and (2).
3. What constitutes a causal relation and an account?
In the case of asyndetic linkage, it is almost impossible to specify the

coherence relation that exists between two adjacent utterances without
looking at the broader sequential and contextual environment of the
9
utterances in question. This means that the specific relation between the
utterances is achieved via the interaction of their meaning and contextual
aspects (linguistic and non-linguistic context). Semantics alone is not
enough.
This can be nicely illustrated by looking at excerpts (1) and (2) again.
In both cases, the causal relation between the two units can be
reconstructed easily by taking into account the situational context and
encyclopedic knowledge: in (2), the speaker presents the fact that people
will probably have to stand in church as a reason for his claim that
88 Christine Gohl
snoring, i.e. falling asleep, will not be possible; it is not possible because
standing is not a good position for falling asleep. In (1), the knowledge
that is required in order to discover and make sense of the causal relation
springs from cultural conventions rather than from knowledge of, for
example, particular biological constraints upon human beings. The
speaker uses the argument mir musset rauche 'we have to smoke' as an
account for the production of the sentence nix kriegt er me: 'he'll get
nothing more:'. For reasons of politeness one should not smoke while
someone else is eating. Therefore, mir musset rauche 'we have to smoke'
justifies the first part of Erik's utterance (nix kriegt er me: 'he'll get
nothing more') by giving a reason for why he says that Uwe should not
have anything else to eat.
The analysis of these examples brings us back to question (3): what
counts as a causal relation in spoken interaction? When is a construction
a causal construction? When is it a construction with a different, e.g.
temporal relation? Which constructions are used by participants in
communication as a means for producing a causal relationship? How do
we know that participants understand a relation between two or more
adjacent utterances as causal? And how can we, from an analytic point of
view, give a plausible interpretation of this relation that goes beyond
intuitive judgment?
One possibility is to look at the way speakers treat and recipients
respond to the utterance in which the construction in question occurs. In
these cases, we can make use of the 'next-tum-proof-procedure', i.e. use
the participants' own interpretations as strong confirmation of the analytic
interpretation (cf Bergmann 1991: 218; Sacks, Schegloff, and Jefferson
1974: 728-729). There are several instances in my data in which a next
speaker refers to a prior asyndetic causal relation with an utterance that
makes it clear that he or she has understood the preceding construction as
containing a causal relation. In these cases, next speakers (or same
speakers in their next tum) use a causal conjunction, a causal adverb, or a
causal phrase like das ist der Grund warum 'that is the reason why' with
respect to the same relation that has been constructed asyndetically
before. These lexical elements or phrases give the analyst a clear
indication that the causality is indeed present for the participants.
Consider excerpt (1) SMOKING again: Erik's utterance consists of a
10
dispreferred second action (nix kriegt er me: 'he'll get nothing more')
and a following account for this dispreferred action (mir musset rauche
'we have to smoke' ): Erik, although not the addressee of Marie's offer,
rejects her offer to Uwe by implying that if Uwe continues eating, he
(Erik) and the other smokers will not be able to smoke. Anna, in her tum,
takes up both components of Erik's utterance, eating and smoking, and

connects these with the causal conjunction weil 'because': da kommt me
tota:l untererniihrt kommt mer hoim, blofJ weil die rauche wollet 'we'll
come home totally undernourished, just because these guys want to
smoke'. She thereby displays her understanding of the two components
as entities that are causally related in this particular context and displays
her understanding of Erik's previous utterance as a construction involving
a causal relation, although Erik did not use a causal connective.
Unfortunately, cases like these are rather infrequent in my data, which
makes it necessary to look for further regularities in the sequential
environment of asyndetic causal constructions. These regularities, as I
will show in the following analysis, are to be found in the kinds of actions
after which accounts occur. Second parts of causal constructions have in
common that they are accounts: they explain or justify a previous
conversational move, i. e. the action carried out by the first utterance.
Occasionally, the account relates to some non-linguistic action or to a
premise which is reconstructable only in connection with the context. The
vast majority of cases in my data, however, are those in which the causal
unit refers to a linguistic utterance or to implications that can be derived
directly from this utterance.
Part of the definition of accounts adopted here is covered by the notion
of 'account' as it has been employed in conversation analysis (e.g.
Atkinson and Drew 1979: 58-60; Heritage 1984: 250-251; 266-273;
Heritage 1988; Levinson 1983: 306-307, 334;), i.e. actions justifying
"disaffiliative (dispreferred) second actions to invitations, requests and
the like" (Heritage 1984: 272). Viewing causal relations in such a way
means including various types of utterances which do not seem to
establish causal relations at first glance. Consider the following excerpts
from Heritage (1984: 250). In excerpt (3), J asks a question about a train
that is supposed to go on a boat. In (4), the participants are talking about
a child: M wants to know whether the child is alright. J, however, does
not know anything about the child's welfare because the child has not
returned, i. e. J has not met with or talked to her yet.
(3) (Heritage 1984: 250)

J: But the trai: n goes. Does th 'train go 0: n th 'boa: t?
M: .h .h Doh I've no idea:. She ha: sn 't sai: d.
(4) (Heritage 1984: 250) «Concerning a child's welfare»

M: '8 alri:ght?,
J: Well 'e hasn ' c 'm ba-ack yet.
90 Christine Gohl
While it is easy to describe the causal relation produced by M in (3)

because the account-portion She ha: sn 't sai: d. refers to the preceding
action encoded in the utterance .h .h Doh I've no idea:., the situation in
(4) is a different one. Here, the account refers to an action that is not
encoded linguistically, but is constituted by the act of accounting itself,
namely J's inability to provide M with the information she has asked for.
Still, a causal relation is present, since Well 'e hasn' c 'm ba-ack yet.
provides a reason for another action, namely for the production of a
dispreferred second, in this case, an answer lacking the information
ll
requeste d.
This understanding of accounts has to be extended, though. As will
become clear in the course of my analysis, utterances explaining or
justifying a previous action occur not only after dispreferred second
actions, but also after particular first actions.
4. The production of accounts after particular actions
4.1. Analyzing causal talk
The identification of accounts produced via asyndetic causal construction

requires attentive data analysis, starting with the inductive search for
regularly occurring patterns in the original spoken data together with
transcript work. The features described in sections 2 and 3 prove helpful
in recognizing these action structures. In addition, I have made use of
Ford's fmdings with respect to the environments in which explanations
and justifications regularly emerge, either marked by because (1994 and
this volume) or unmarked (this volume). However, her finding that
accounts are frequently produced in contexts involving some kind of
contrast is only applicable to part of the material analyzed here. The
occurrence of accounts, it turns out, is not directly linked to the linguistic
structure which a previous utterance displays, but to the kind of action
accomplished by a previous utterance.
The next step in the analysis is to look for deviant cases, i. e. for cases
in which participants treat the non-production of an account as
problematic or in which they explicitly ask for the provision of an account
at a point when an account is due. These cases are typically less common
because they deviate from normatively required behavior (Heritage 1988:
133). Consider example (5), which is taken from a radio phone-in. The
context is as follows: the caller has problems with his mother, who
interferes in his relationship with his girlfriend. The psychologist is trying
to convince him that his problem is his own fear of coming to a decision:
she advises him to stand by his girlfriend and speak openly to his mother.
The caller, however, rejects this by saying that his mother does not want
him to see his girlfriend at all (line 9):
(5) DECISION (18/4; 15:30)

1Psychologin: mhm .h [abe darumJ gehts,
2Anrufer: [aberwenn aber (ich)]
3Psychologin: ob da: ne entscheidung (-) wirklich
moglich ist;
4 (2.5)
5Anrufer: ja mh
6 0.0)
7Anrufer: ich weifi es seIber aber==aber ich: :,
8 (0.5)
((10 intonation units left out, in which the caller raises objections))
9Anrufer: .hh[meine] mutter will==aber nicht
da:fi==ich zu meinerfreundinfahr.
10Psychologin: [del ]
11 (1.0)
12Psychologin: aha:;
13 (0.5)
14Psychologin: und:
15 (0.5)
16Psychologin: warum will die mutter das nicht?
17 (-)
18Anrufer: ja sie sagt h.
19 meine freundin hat drei kinder,
'lPsychologist: mhm.h [but that's] the crucial point,

2Caller: [but if but (I)]
3Psychologist: if a decison (-) is really possible;
4 (2.5)
SCalIer: yeah mh
6 (1.0)
7Caller: I don't know but I,
8 (0.5)
((10 intonation units left out, in which the caller raises objections))
9CalIer: 'hh[my] mother doesn't like me to see my girlfriend.
10Psychologist: [del ]
11 (1.0)
92 Christine Gohl
12Psychologist: aha:~
13 (0.5)
14Psychologist: and
15 (0.5)
16Psychologist: why doesn't she like it?
17 (-)
18Caller: well she says h
19 my girlfriend has three children,'
Here the account is only produced after the psychologist has explicitly
asked for it (line 16). Between the caller's utterance (line 9)-an
objection to the psychologist's insisting that a decision is necessary-and
the psychologist's question in line 16, there are three long pauses and two
short utterances which do not pursue any further topical talk. That is, the
psychologist can be heard as pursuing an account from her caller by
providing him multiple opportunities at which he can provide such an
account.
The analysis of such deviant cases leads to the conclusion that
"account giving is not merely an empirically common feature that is
associated with unexpected or unlooked for actions, but is a normatively
required feature of such actions. Since failures to provide accounts attract
either overt pursuits of them or sanctions, we can conclude that the giving
of accounts in such contexts is itself a morally accountable matter"
(Heritage 1988: 135).
In sections 4.2.-4.6., different environments will be described in which
accounts, realized by asyndetic linkage, regularly occur in my data. The
environments will be presented in the order of their frequency of
occurence: accounts after dispreferred seconds are most frequent, then
accounts after assessments, requests, and finally after complaints and
reproaches.
4.2. Accounts after dispreferred second actions
Various conversation analytic studies (e. g. Atkinson and Drew 1979, Ch.
2~ Heritage 1984: 265-269~ Pomerantz 1984) have shown that particular
actions, often organized as first parts of adjacency pairs, make a second
action relevant, i. e. these actions call for a second action by virtue of their
sequential implications. These second actions have alternative, but
nonequivalent, realizations: one realization is preferred, the other
dispreferred. After requests, suggestions, offers and invitations, an
acceptance is the preferred and unmarked second action. A refusal, or no

reaction at all, represents a dispreferred second. After assessments, an
agreement is the preferred response, disagreement the dispreferred one.
Dispreferred seconds' exhibit several common features, including delays,
prefaces and-as is relevant in the following-accounts. 12
Example (6) illustrates the provision of an account after a dispreferred
I3
second following a suggestion. Four family members are talking about
how to handle the meat they are having for dinner. Anne, the daughter,
suggests to her father (lines 1 and 2) that he use the scissors that are on
the table to cut the meat.
(6) SCISSORS (Schwab 8; 14:19)

lAnne: nimm halt die () schere;
2 fiir was haschn die;
3 (1.5)
4Fritz: da komm i mit der schere nicht durch;
5Erik: des sind doch knoche;
6Fritz: jetztet;
7 .hh
8Anne: ja grad deshalb hab i dacht,
9Erik: des kannsch beim e pock-pock;
10 da kannsch dknoche no durchschnipple;
,1Anne: take the (.) scissors;

2 what do you have them for;
3 (1.5)
4Fritz: I can't get through with these scissors;
5Erik: there are bones;
6Fritz: now;
7 .hh
8Anne: that's exactly why I thought,
98Erik: you can do that with a pock-pock 14 ;
10 there you can snip the bones;'
With his utterance in line 4, da komm i mit der schere nicht durch 'I
can't get through with these scissors', Fritz rej ects Anne's suggestion
nimm halt die () schere; 'take the (.) scissors;'. The preferred, and thus
structurally unmarked, second part following a suggestion would be its
acceptance. In such a case no further talk would be necessary. In (6),
however, Fritz offers an account for refusing to follow Anne's
suggestion, i. e. for why he does not put it into action. In providing a
94 Christine Gohl
reason for not taking up his daughter's suggestion, Fritz need not make
his refusal explicit: the action is evident although non-verbal. In this case,
asyndetic linkage is the only possibility for the production of the account
because there is no previous utterance or implication to which a
conjunctional clause could be linked.
Accounts occur after various kinds of such dispreferred second
actions. In excerpt (7), from a doctor-patient interaction, an account
follows a request for information that cannot be fully met by the second
speaker.
The doctor is a specialist in the field of reproduction technological
medicine and the patient, a woman with a hormonal disorder who is
seeking advice about problems that might occur if she wants to get
pregnant.
(7) FINDINGS (2; 5-16)

1Doctor: frau bade sie sind vom frauenarzt zu uns
geschickt worden wegen hormoneller sto: rung steht hier;
2 konnen sie mir da noch was niiheres dazu schon sagen,
3Patient: eigentlich vom hausarzt;
4Doctor: mhm,
5 (-)also ich hab angefangen (---) mit etwa zwanzig,
6Patient: fing das bei mir an mit der pille,
7Doctor: mhm,
8 ich hatte diane fiinfunddreifiich bekommen,
9 aber da hab ich von meinem ersten
frauenarzt die befunde noch nich;
10 [derwar] einfach noch im [urlaub;]
llDoctor: [mhm,] [mhm,]
12Patient: die schick ich dann noch nacho
'lDoctor: mrs bade you were sent to us by your

gynecologist because of a hormonal disorder is written here;
2 could you tell me a little bit more about this now,
3Patient: by my family doctor, actually;
4Doctor: mhm
5 (-) well I started to (---) at the age of twenty,
6Patient: I started to take the pill,
?Doctor mhm
8 I was given diane fiinfunddreillig, (name of the product)
9 but I don't have the fmdings from my first gynecologist yet;
10 [he was] still on [holiday;]
11Doctor: [mhm,] [mhm,]

12Patient: I will send them later. '
In lines 1 and 2, the doctor is asking the patient for some information
about her problem. After correcting one of the doctor's assumptions (line
3), the patient starts reporting on earlier treatment, thus producing the
answer that has been made relevant, i. e. sequentially necessary by the
doctor's request for information. She cannot completely comply with the
doctor's request, however (line 9). Immediately following this partly
dispreferred second action, she produces an account, explaining why she
does not have the findings from her first gynecologist yet: der war
einfach noch in urlaub; 'he was still on holiday;'. i.e. there was no way
of collecting the findings or having them sent.
The excerpt in (8) presents an interactive construction of a sequence
with the dispreferred-second-account structure. This construction shows
that recipients are sensitive to dispreferred actions even if a request for
information and a following dispreferred action occur within the context
of a narration and are embedded in a reported speech situation. Five
family members are talking about a television program on the fight
against crime. In the passage transcribed, the topic is undercover agents.
Fritz is telling the story of a man he knows personally because he used to
be a customer in Fritz's shop.
(8) UNDERCOVER (Schwab 3; 1:21:52)

1Fritz: (( ))
2 der war bei der polizei.
3 ie g,J ludwigsburg.
4 (1.5)
5 und is dann in=en g/ untergrund gegangen.
6 (1.5)
7 unlangst waren ersch e: die eltern da, (---)
8 un na habe gfragt ja: eh:,
9 wie gehts ihrem sohn und so e::,
10 < sagt er () wir wissen nichts, >
11Erik: hajaja
12 die krieget.h [die krieget () en vollstandig neue lebenslauf,
13Fritz: [< <f> wir wissen nichts, () >
14Erik: die krieget en dreifufJzig s e:, ()
15 un=e stattliches bankkonto,
96 Christine Gohl
16 (1.5)
17 na diirfe se 10s lege.
, 1Fritz: « »
2 he used to work as a policeman.
3 ie gJ, ludwigsburg. «name of policestation and town»
4 (1.5)
5 and then he went underground.
6 (1.5)
7 only recently his parents came in, (---)
8 and then I asked them well: eh:,
9 how is your son these days e::,
10 «p> he said (.) we don't know anything,>
11Erik: ohyeah
12 they get .h [they get (.) a completely new personal record,
13Fritz: [<<f> we don't know anything, (.»
14Erik: they get a three fifty s e, ((name of a car» (. )
15 and a large bank account,
16 (1.5)
17 and then they can get started. '
While Fritz is reporting on the conversation he had with the parents of the
former policeman (lines 8-13), Erik comes in (lines 11-12) and produces
an account for the parents' inability to give Fritz the information he had
asked for. Erik's utterance explains the production of a dispreferred
second by the characters in Fritz' story: the parents are not able to answer
Fritz's question wie gehts ihrem sohn und so e::, 'how is your son these
days e::,' because their son, after having gone undercover, was given a
completely new identity (die krieget () en vollstandig neue lebenslauf,)
and various other things that cut him off from his former life (lines 14-
17).
The accounts offered after dispreferred seconds in examples (6), (7),
and (8)-whether they follow requests for information (excerpt (7) and
(8» or suggestions (excerpt (6»-all refer to some kind of inability to
"carry out the proposed or required conversational action" (Heritage
1988: 136). In (6), the speaker accounts for why he is not able to use the
scissors for cutting the meat as Anne has suggested; in (7) the patient
accounts for her inability to completely comply with the doctor's request
for information, and in (8) a third speaker accounts for the inability of two
other-distant-speakers to carl)' out the action that has been called for
by the question asked, namely to provide an answer.
4.3. Accounts after assessments
Accounts are not only offered after dispreferred seconds; they also
emerge regularly after assessments. 15 These assessments 0 f en t ·
contaIn
lexical markers or phrases that make them 'strong assessments' in the
sense of what Ford has described as 'strong evaluation' (Ford 1994:
548). In German, some of these lexical markers are: besonders
'especially', wirklich 'really', sehr 'very', ganz 'totally', as well as
particular adjectives like abscheulich 'dreadful', wunderbar 'wonderful',
genial 'brilliant' or adjectives in their superlative form, e.g. am
schlimmsten 'worst' and nominal expressions like da war die holle los
'all hell broke loose' .
Accounts after strong evaluative assessments are more frequent in my
data than accounts after assessments that do not express a strong
evaluative stance. This observation leads to the conclusion that it is not
particular lexical items that are responsible for the production of an
explanatory unit, but the action carried out by the evaluative statement,
the assessment.
Let us look at transcript (9). The participants are talking about the
family's cat. The starting point for this discussion is Marie's utterance.
She thinks that the cat is sad because it did not get anything for
Christmas.
(9) CAT I (Schwab 3; 0:07:45)

lMarie: i mochtja nix (sage aber) die katz isch ganz traurig-
2 die hat nix kriegt.
3Fritz: <<f> grad eben ha () (ha n) neido>, ()
4 < des hat er net emol aguckt>.
5Erik: des isch doch dere katz ihr erschtes
weihnachte -
6 die weifi gar net dafi mer do [was kriegt== -
7?Marie: [was los isch
8Erik: etz niichsch jahr () da muj3 me re
vielleicht was gebe;
'lMarie: I don't want to (say) anything (but) the cat is really sad -
2 she didn't get anything.
3Fritz: «f> just now I gave her something>, (.)
4 «p> she didn't even look at it>.
5Erik: it's the cat's first Christmas, you see -
6 she doesn't know that [you get something at Christmas= -
98 Christine Gohl
7?Marie: [what's going on

8Erik: next year you'll probably have to give her something;'
With her utterance die hat nix kriegt 'she didn't get anything' Marie
gives an account for her assessment die katz isch ganz traurig- 'the cat is
really sad-' .
The following excerpt illustrates the interactive production of an
assessment-account sequence. The participants, mother and son, are
talking about the car the son has used to drive to his parents' home. This
car is obviously not his own, but belongs to the company he works for.
(10) AUTO (Schwab 3; 0:27:29)

lMarie: < <cresc> hasch du heut mit dem schone lauto
Jahre> J diirfe;
2Erik: ja:: i mu.f3 ja was darschtelle oder?
'IMarie: «cresc> were you allowed to drive> that

beautiful car today;
2Erik: yeah:: I have to keep up appearances, haven't I?'
With her utterance, Marie is doing two things that call for a second action
by the recipient. She is asking a question and is also making an
assessment of the car the son has used to visit his parents: mit dem
schone auto 'that beautiful car'. The content of this question is marked on
the prosodic level as something extraordinary: first, during its production
the utterance increases in volume; second, there is remarkable pitch
movement on auto Jahre, with the pitch falling from a very high level.
Marie's question - which constitutes the first part of an adjacency pair
- calls for an answer. This is produced by Erik immediately after Marie
has finished her tum (ja:: 'yeah::'). The assessment, however, requires a
separate acknowledgment. This is provided when Erik gives a reason for
what Marie has said (line 2). Note that Erik not only reacts to the content
of Marie's utterance, but also to its prosodic design, which conveys a
somewhat surprised and approving attitude towards the fact that Erik was
allowed to drive this beautiful car. This attitude, however, has a slight
ironic undertone, which is taken up by Erik in his utterance: both Marie's
question and Erik's answer are thus produced in a teasing, slightly ironic
style-judging from content and, especially in Marie's utterance, from
the prosody.
4.4. Accounts after requests
Asking a question is a particular form of requesting something from the

recipient, namely the provision of information (cf example (7)
FINDINGS, in which the doctor asks for detailed information about the
patient's medical history).
Consider now example (11). The participants are discussing the
appropriate time for smoking and the quality of the cigars a family
member has brought along. Erik was offered one of these cigars on an
earlier occasion. He now reports on the bad effects it had on him.
(11) CIGARS (Schwab 3; 0:0:37)

lMarie: do hocksch na un rauchsch st[att daj3 de unsre esse/
2Erik: [ja sind die
zigarre scho was,
3 also mir isch ganz ubel denoch gwase.
4Marie: er==hot==doch==gsagt die beschtn wo==s gibt-
5 (1.5)
6Fritz: was
7Erik: die zigarre,
'IMarie: then you will sit down and smoke

in[stead of our eating our foodl
2Erik: [are these cigars okay,
3 I felt absolutely sick after smoking one.
4Marie: he said the best ones you can get. -
5 (1.5.)
6Fritz: what
?Erik: the cigars,'
In line 3, Erik gives a reason for asking the question ja sind die zigarre
scho was? 'are these cigars okay?'. He asks if the cigars are okay,
because he felt sick after smoking one of them, implying with his
question that the cigars might not be okay. By offering an account, Erik is
giving the recipients some insight into the motive he has for asking the
question.
In the speaker makes a request in the form of an imperative. The
participants are talking about the food, the meat they are having for dinner
in particular.
100 Christine Gohl
(12) MEAT (Schwab 8; 15:18)

1Erik: tausche mer < deins isch kleiner.>
2Kai: gar net -
3 [fwe: noi nemm du i will blo) e schtii[ck; ==
4Kai: [nein (2 Silben) esse.
5Uwe: i will doch net solche brocken.
'IErik: let's swap «p> yours is smaller.>

2Kai: not true -
3Uwe: no you take it Ijust want a pie[ce;=
4Kai: [no (2 syl.) eat.
5Uwe: 1 don't want such big pieces. '
Erik asks Uwe if he wants to swap pieces of meat. Immediately after this
request, he offers a reason for requesting: deins isch kleiner 'yours is
smaller'. In his next tum (line 3), Uwe refuses to accept Erik's request
(noi nemm du 'no you take it') and accounts for the production of this
dispreferred second action in lines 3 and 5: i will blo) e schtiick; 'I just
want a piece', and i will doch net solche brocken 'I don't want such big
pieces' .
Excerpts (11) and (12) show that requests-in these cases formulated
as a question and as an imperative-are followed by accounts. Requests
are face-threatening activities because they put some obligation on the
recipient he or she has to do something for the speaker, either in the form
of a verbal action (provide the information requested) or in the form of a
non-verbal action (do the thing requested). This format makes requests
conversationally sensitive actions-and these are likely to be followed by
accounts.
4.5. Accounts after complaints and reproaches
A common feature of activities like complaining and reproaching is their

moral bias. "In these activities interlocutors focus on an infringement of
expectations concerning situatively appropriate behaviour and in doing
so, demonstrate their own orientation to moral rules and values"
(Giinthner 1994: 2). Reproaches make second actions by the person the
reproach is directed at relevant, i.e. they call for justifications or apologies
(Giinthner 1993: 3-6). In the data analyzed here, however, reproaches are
not only followed by such second actions (compare (13) below, where
Fritz denies the action he has been reproached for). First speakers, too,
i. e. the persons who are doing the reproaching, offer accounts for
reproaching someone or complaining about something.
In (13) the topic is once again the family cat. Anna is blaming Fritz for
having given his cold to the cat.
(13) COLD (Schwab 3; 1:49:38)

JAnna: mensch jetz hasch die katz scho angschteckt mit deiner
erkiiltung-==
2 der nie:st scho de ganze d[a:g;
3Fritz: [des hat er heut morge scho gmacht.
'lAnna: good heavens you've given the cat your cold=

2 he's been sneezing all d[ay
3Fritz: [he did it this morning already.'
From the external, visible fact der nie:st scho de ganze dag 'he's been
sneezing all day', Anna draws the conclusion that Fritz, who has a cold,
must have given it to the cat. She formulates this conclusion as a
reproach: both the content of this utterance and the way it is expressed
(e.g. the term mensch, which is very often used at the beginning or at the
end of an accusatory utterance) give it its reproachful character. Right
after this unit, Anna offers an explanation and a justification for the
conclusion and the reproach: the cat's been sneezing all day. 16
Reproaches are face-threatening activities, i.e. activities which are
likely to threaten both the face of the person the reproach is directed at
and the social relationship between the parties. Offering an account after
such an action makes a reproach more understandable because it reveals
the reasons behind it and gives the person blamed the chance to react to
the reproach in a way which is more customized than a mere refusal
would be: he or she can exculpate him- or herself by referring to the
reason for the reproach. This happens in line 3 in (13): Fritz makes it
clear that he refuses to take responsibility for the cat's cold because the
cat has been sneezing all morning. He implies that it must have caught
the cold elsewhere.
It will not come as a surprise that accounts after reproaches emerge
not only after a speaker has blamed some other participant for something,
but also after complaints about a third party. Consider the following
transcript. The content is the same as in CAT I: three family members are
talking about the meal they have just had and are considering the 'poor
cat', who seems to be sad because it did not get anything.
102 Christine Gohl
(14) CAT II (Schwab 3; 0:07:45)

lMarie: i mochtja nix (sage aber) die katz isch ganz traurig-
2 die hat nix kriegt.
3Fritz: <<f> grad eben ha () (ha n) neido>, ()
4 < des hat er net emol aguckt>.
5Erik: des isch doch dere katz ihr erschtes weihnachte -
6 die weifJ gar net daft mer do [was kriegt== -
7?Marie: [was los isch;
8Erik: etz niichsch jahr () da muj3 me re vielleicht was gebe;
'lMarie: I don't want to (say) anything (but) the cat is really sad-
2 she didn't get anything.
3Fritz: «f> just now I gave her something>, (.)
4 «p> she didn't even look at it>.
5Erik: it's the cat's first christmas you see -
6 she doesn't know that [you get something at christmas= -
7?Marie: [what's going on;
8Erik: next year you'll probably have to give her something;'
With his utterance des hat er net emol aguckt 'she didn't even look at it',
Fritz complains about the cat's behavior with respect to his action: he has
given the cat some food, but the cat did not appreciate it. Erik and Marie
account for this behavior with the utterances that follow. Thus, once
again, the causal relation is constructed interactively. Fritz produces an
utterance which is treated by a second speaker, Erik, as a consequence:
the cat does not appreciate the food because it is her first Christmas.
Therefore, you cannot expect her to react in a particular way. The second
part of Erik's utterance (die weifJ gar net dajJ mer do was kriegt- 'she
doesn't know that you get something at christmas-') provides another
potential reason for the cat's behavior. At the same time, another speaker
(probably Marie) comes in and completes another explanatory unit (die
weifJ gar net was los isch; 'she doesn't know what's going on;') for what
Fritz has said earlier (des hat er net emol aguckt 'she didn't even look at
it' ). Both explanatory units, Erik's and Marie's, contain reasons for the
cat's behavior, as well as consequences with respect to the first part of
Erik's utterance (des isch doch dere katz ihr erschtes weihnachte- 'it's
the cat's first christmas-'): hence, because it is the cat's first Christmas,
she does not know that you receive something on this occasion and she
does not know what is going on. The interactive nature of this production
of accounts shows that both speakers, as recipients, consider the action of
blaming the cat for not appreciating the food as an action that calls for
accounting.
4.6. Activity types as constraints on the interpretation of asyndetic

causa/constructions
In my data, asyndetically linked accounts follow dispreferred second

actions and various other types of action which are organized as
adjacency pairs in that they make a second-verbal or non-verbal-action
sequentially necessary: first assessments, for example, call for second
assessments (Pomerantz 1984), requests call for responses in the form of
an acceptance or refusal, reproaches for denial or admission (Heritage
1984: 269; Levinson 1983: 308,336). The causal units in the latter cases
account for the production of the first actions. However, not all first parts
of actions that are organized as adjacency pairs require accounts. Rather,
accounting occurs, in particular, after activities which are sensitive to the
social relationship between the parties. Requests and reproaches, are, like
dispreferred seconds, face-threatening actions-requests, because they
impose some obligation on others, reproaches, because they imply
criticism of others. Although assessments are not face-threatening in the
same way, they are still sensitive to possible disagreement-especially as
'strong assessments'-in that they express a speaker's highly subjective
evaluative stance towards a particular fact or event. Accounting after first
assessments makes a subjective position more understandable and thus
more likely to lead to a preferred second assessment.
The analysis of the data shows that participants are sensitive to the
production of accounts in these particular environments. Therefore, I
would argue that these activity types constrain the interpretations that
arise when a causal unit is not explicitly marked as such. An utterance is
much more likely to be interpreted as an account, i. e. as an utterance
explaining and/or justifying a previous conversational move, if the
preceding utterance, by virtue of its sequential and social implications,
calls for accounting.
Concerning accounts after dispreferred seconds, Heritage (1988: 136)
argues that they help to avoid threatening the social relationship between
the participants. Accounting for the inability to produce an answer, for
example, leaves the responsibility for the dispreferred second with the
speaker. Not accounting for a dispreferred second or denying the right of
the first speaker to carry out the first action would constitute a threatening
act, endangering the participants' relationship. The same holds for the
104 Christine Gohl
production of accounts after reproaches: reproaches are, by their very

nature, disaffiliative acts threatening the recipient's face. Giving a reason
for such a disaffiliative move makes the act of reproaching more
understandable because it reveals the motives behind the reproach.
With respect to requests, which, of course, are much less face-
threatening than reproaches, it can be argued that a request always puts
some obligation on the recipient(s): he or she is asked to do something for
the speaker, for example, provide him/her with the information requested.
Again, providing a reason for requesting makes the act more
understandable because it gives the recipient some insight into the
speaker's motives.
In sum, the interpretation of an asyndetic construction depends on the
features described in sections 2 and 3 and on the sequential environment
of the construction in question, i. e. on the activity it can be construed as
relating to.
5. Concluding remarks
This study shows that speakers may give reasons without usmg a
conventionalized causal connective element.
Different sequential environments were investigated in which speakers
produce such asyndetically linked causal relations in order to give
accounts. These environments are constituted by different actions which,
for various reasons, call for accounting. One group of these activities-
dispreferred seconds, requests, and reproaches-is particularly sensitive
to the social relation between participants, either because these activities
are face-threatening or because they put some obligation on the recipient.
The second group, including first (strong) assessments, is also socially
risky. Accounts following strong assessments make the production of a
preferred second more likely because they support what is often a highly
subjective attitude on the part of the speaker and thus make it easier for
recipients and possible next speakers to understand the speaker's stance
and react to it in a preferred format.
Ford has shown in her studies (1994 and this volume) that accounts or
explanations and justifications often emerge in contexts involving some
form of contrast, which mayor may not be marked, for example, by
contrastive markers or negation particles. The analysis of the asyndetic
causal constructions in my data, however, shows that only about half of
the accounts occur in contexts involving contrasts. This means that
accounts have a much broader context of occurrence. In fact, much
suggests that the production of accounts should be seen as motivated by

making reference to the kinds of actions which call for them. This is not
to deny that there may be a correlation between linguistic structure and
the kind of action that is carried out by an utterance-dispreferred
seconds, for example, often contain negation markers-but the
correlation is by no means obligatory. Consequently, analysis must go
beyond the linguistic realization of an utterance and ask what a speaker is
doing with a particular utterance.
Related to the issue of structure and action is the question of linguistic
practices. Which ones do speakers have at their disposal for the
production of reasons? What circumstances influence the choice of
linguistic means, especially the choice between explicit conjunctional
linkage and asyndetic linkage, when speakers produce accounts for
previous actions? First tentative analyses show that one of many possible
factors is whether speakers are giving reasons for their own actions or for
actions produced by other speakers: asyndetic linkage seems to be
preferred to conjunctional linkage in cases of interactively produced
constructions. Moreover, particular environments apparently allow only
asyndetic linkage. For example, when an account refers to a non-verbal
action, there is no previous utterance or implication to which a
conjunctional clause could be linked (cf example (6) SCISSORS).
These questions, which still need investigation, touch upon the general
issue of the linkage between (linguistic) practices and actions, discussed,
for example, in ScheglofI (1996, 1997). Further studies along the lines
suggested here could bring about a fuller understanding of the relation
between various linguistic practices and the action of 'giving reasons' in
spoken interaction.
Appendix: Transcription conventions
[] overlap
[]
(.) micro-pause
(-), (--), (---) short, middle or long pauses of approx. 0.25-0.75 seconds, up to ca.
1 second
(2.0) estimated pause of more than 1 second
and=uh slurring within units
lengthening, according to its duration
uh,ah, etc. hesitation signals, so-called "filled pauses"
? intonation phrase-final: high rise
intonation phrase-final: mid-rise
intonation phrase-final: level pitch
106 Christine Gohl
intonation phrase-final: mid-fall

intonation phrase-final: low fall
/ break-off
« » interpretive comments over a stretch of speech
( ) unintelligible passage
(such) uncertain transcription
~ » onllssion of text
pitch step up
j, pitch step down
«t» forte, loud
«p» piano, soft
«all» allegro, fast
«cresc» crescendo, becoming louder
Notes
* I am grateful to Dagmar Barth and Susanne Gtinthner as well as to the

editors of this volume for their valuable comments on earlier versions.
1. For the connection between the term 'account' and the more generic
phenomenon of interactional accountability discussed in ethnomethodology
and conversation analysis, see Heritage (1988: 128, 138-141).
2. I view speaking as a form of social action (Heritage 1984: 135) and therefore
linguistic utterances as a means for carrying out particular actions. From this
premise it follows that coherence relations, e.g. causal relations, that hold
between utterances also hold between actions, i.e. an utterance referring to
some other utterance does so because of the action(s) carried out by that
particular utterance. This applies to utterances to the extent that they consist
of one or more turn-constructional units. For the notion of turn-
constructional unit (TCD), see Sacks, Schegloffand Jefferson (1974).
3. A first analysis of explicitly linked causal structures compared with asyndetic
causal structures shows that explicit linkage allows for more conversational
space inbetween accountable action and account. Exceptions are, moreover,
deviant cases (Heritage 1988: 133-135), in which an account is produced
with a delay, usually when a first speaker does not offer the account
immediately but only after a second speaker has elicited its production (see
excerpt (5) DECISION in section 4.1.).
4. The translations are meant to convey more the content of the utterances than
the stylistic subtleties of the Swabian dialect.
5. Chafe (1988: 1) defines the prototypical intonation unit as "single coherent
intonation contour characterized by one or more intonation peaks and a
cadence that is recognizable as either clause-final or sentence-final."
6. For this connective function see also Franck (1980: 253) and Lindner (1991:
174). Besides emphasizing their connective function, Ickier (1994: 377) also
stresses the interactive role of modal particles: they "anchor particular
utterances in larger contexts, which are usually characterized by their
dialogic nature." For the interactive function of the particles doch, eben, and
ja in spoken discourse see also Liitten (1979).
7. Occasional remarks on the possible causal meaning of modal particles can be
found for doch (Feyrer 1998: 187-192), auch (Paul 1966), halt (Abraham
1983, Hinrichs 1983) eben andja (Abraham 1983).
8. This investigation is based on the analysis of approximately 7 hours of
spoken German comprising about 4 1/2 hours of informal conversational
data, i.e. family conversations, 1 1/2 hours of interactions in a medical
setting, and 3/4 of an hour of radio phone-ins. Most of the data was recorded
in the Swabian area of Southern Germany. I wish to thank Ute Lacher-
Laukenikat and project C4 of the SFB 471 for making their data available to
me.
9. For the interpretation of asyndetic conditional structures, i.e. paratactic
conditionals, see Thumm (1999).
10. For the notion of 'dispreferred second' and a more detailed discussion of
accounts after dispreferred second actions, see section 4.2.
11. In fact, in excerpt (3), the utterance'.h .h Ooh I've no idea:.' also constitutes
an account: M accounts for her failure to provide J with the information
requested (Heritage 1984: 250). Compare also excerpt (6) in section 4.2.
12. For more details, including the marked-unmarked distinction, and a survey
of the most important issues concerning preference organization, see
Levinson (1983: 332-345). On the use of causal markers in English and
Japanese conversations in dispreferred environments, see Ford and Mori
(1994).
13. For another example illustrating the production of a dispreferred second-
account-sequence after a suggestion, see excerpt (1) SMOKING. While in
excerpt (6) SCISSORS the declining component is not verbalized, in excerpt
(1) SMOKING the speaker formulates the declining of another speaker's
suggestion in a very explicit and highly marked way.
14. 'Pock-pock' is presumably the name for a kind of a fowl.
15. For a discussion of assessments and their sequential implications in spoken
German interaction, see Auer and Uhmann (1982). Some of their transcripts
show that second speakers may explicitly ask for the reason of an assessment
given by a first speaker if the latter does not account for the assessment him-
/herself (Auer and Uhmann 1982: 8-9).
16. The formulation der nie:st scho de ganze da:g ('he's been sneezing all day')
falls into the category of Extreme Case Formulations (Pomerantz 1986).
Such formulations are used by speakers when they justify complaining about
something or somebody to "legitimize a complaint and portray the
complainable situation as worthy of the complaint" (Pomerantz 1986: 228).
References
Abraham, Werner
1983 Starke Satzverkniipfer-Sinnsog und Partikelenquete. In: Harald
Weydt (ed.), Partikeln und lnteraktion, 25-39. Tiibingen: Niemeyer.
108 Christine Gohl
Atkinson, J. Maxwell and Paul Drew

1979 Order in Court. London: Macmillan
Auer, Peter and Susanne Uhmann
1982 Aspekte der konversationellen Organisation von Bewertungen.
Deutsche Sprache 10: 1-32.
Bergmann, Jorg R.
1991 Konversationsanalyse. In: U. Flick, E. von Kardoff, H. Keupp, L.
von Rosenstiel and S. Wolff (eds.), Handbuch Qualitative
Sozialforschung, 213-218. Mfinchen: Psychologie Verlags Union.
Blakemore, Diane
1987 Semantic Constraints on Relevance. Oxford: Blackwell.
Couper-Kuhlen, Elizabeth
1996 Intonation and clause combining in discourse: The case of because.
Pragmatics 6/3: 389-426.
Chafe, Wallace
1988 Linking intonation units in spoken English. In: John Haiman and
Sandra A. Thompson (eds.), Clause Combining in Grammar and
Discourse, 1-27. AmsterdamIPhiladelphia: Benjamins.
Feyrer, Cornelia
1998 Modalitiit im Kontrast: Ein Beitrag zur ubersetzungsorientierten
Modalpartikelforschung anhand des Deutschen und des
Franzosischen. Frankfurt am Main: Peter Lang.
Ford, Cecilia E.
1994 Dialogic aspects of talk and writing: because on the interactive
edited continuum. Text 14/4: 531-554.
Ford, Cecilia E. and Junko Mori
1994 Causal markers in Japanese and English conversations: A cross-
linguistic study of interactional grammar. Pragmatics 4/1: 31-61.
Franck, Dorothea
1980 Grammatik und Konversation. Monographien Linguistik und
Kommunikationswissenschaft 46. Konigstein, Ts.: Scriptor.
Gfinthner, Susanne
1993 "Kannst du auch fiber ANDERE Leute LAsTERN." Vorwiirfe als
Formen moralischer Kommunikation. Arbeitspapier Nr. 9.
MORAL-Projekt. Universitat Konstanz.
1994 Moral voices: Prosodic features of reproaches in "why"-formats.
Arbeitspapier Nr 11. MORAL-Projekt. Universitat Konstanz.
1996 From subordination to coordination? Verb-second position in
German causal and concessive constructions. Pragmatics 6/3: 323-
370.
Heritage, John
1984 Garfinkel and Ethnomethodology. Cambridge: Polity Press.
Hartmann, Dietrich
1986 Semantik von Modalpartikeln im Deutschen. Deutsche Sprache 14:
140-155.
Hinrichs, U.
1983 Konnen Abtonungspartikel metakommunikativ funktionieren? In:
Harald Weydt (ed.), Partikeln und lnteraktion, 274-290. Tubingen:
Niemeyer.
IckIer, Theodor
1994 Zur Bedeutung der sogenannten 'Modalpartikeln'. Sprach-
wissenscho:ft 19: 374-404.
Levinson, Stephen
1983 Pragmatics. Cambridge: Cambridge University Press.
Lindner, Katrin
1991 'Wir sindja doch alte Bekannte' The use of Germanja und doch as
modal particles. In: Werner Abraham (ed.), Discourse Particles,
163-201. Amsterdam: Benjamins.
Lutten, Jutta
1979 Die Rolle der Partikeln doch, eben, und ja als Konsensus-
Konstitutiva in gesprochener Sprache. In: Harald Weydt (ed.), Die
Partikeln der deutschen Sprache. 30-38. Berlin: de Gruyter.
Pasch, Renate
1997 WElL mit Hauptsatz-Kuckucksei im DENN-Nest. Deutsche
Sprache 3: 252-271.
Paul, Hermann
[1966] Deutsches Worterbuch, 6.A. Tubingen: Niemeyer.
Pomerantz, Anita
1984 Agreeing and disagreeing with assessments: some features of
preferredldispreferred turn shapes. In: 1. Maxwell Atkinson and
John Heritage (eds.), Structures of Social Action, 57-101.
Pomerantz, Anita
1986 Extreme Case Formulations: A way of legitimizing claims. Human
Studies 9: 219-229.
Sacks, Harvey, Schegloff, Emanuel A., and Jefferson, Gail
1974 A simplest systematics for the organization of turn-taking for
conversation. Language 50/4: 696-735
Scheutz, Hannes
1998 Weil in spoken German. Presentation given at the conference on
'Cognitive and Discourse Perspectives on Cause, Condition,
Concession and Contrast'. Oktober 1998. UniversiUit Konstanz.
Schegloff, Emanuel A.
1996 Confirming allusions. American Journal of Sociology 102/1: 161-
216.
1997 Practices and actions: Boundary cases of other-initiated repair.
Schleppegrell, Mary J.
1991 Paratactic BECAUSE. Journal ofPragmatics 16: 323-337.
110 Christine Gohl
Thumm, Markus
1999 How to do some things without 'if. Eine empirische Untersuchung
zu parataktischen Konditionalkonstruktionen im gesprochenen
Englisch. MA-thesis, Department of Linguistics, University of
Konstanz.
Uhmann, Susanne
1998 Verbstellungsvariation in Weil-Satzen: Lexikalische
DifIerenzierung mit grammatischen Folgen. Zeitschrift .fur
Sprachwissenschaft 17/1: 92-139.
Constructions with if, since, and because:
Causality, epistemic stance, and clause order*
Barbara Dancygier and Eve Sweetser
English causal and conditional conjunctions show significant overlap both

in functions and in the ranges of grammatical constructions in which they
occur. These phenomena are related: one reason why different conjunctions
can express similar meanings is that the lexical semantics of one
conjunction may explicitly involve meaning which is not present in the
other conjunction's semantics but can be expressed by accompanying
grammatical constructions. Using Mental Spaces theory (Fauconnier
1985/1994, 1997), we analyze the semantics of English if, since, and
because, examining the interaction between their lexical semantics and the
semantics of their grammatical surroundings. Among the factors involved
are predictive function and causal relations (Dancygier 1993, 1998;
Dancygier and Sweetser 1996, 1997), epistemic stance (Fillmore 1990a,b),
and the relation of the mental space to the speech setting (Sweetser 1990,
Dancygier and Sweetser 1996, Dancygier 1998).
1. Introduction
In this paper we will analyze the uses of causal and conditional

constructions using the conjunctions if, since, and because. These
conjunctions specialize in distinct areas of meaning, but their areas of
use overlap significantly: a comprehensive analysis, therefore, should
explain the overlap as well as the distinctions between them. It has been
noted, for instance, that since could substitute for if in (1) without
significant change in interpretation, and that a causal relation underlies
the contingency described in (2):
(1) If (as you say) he is in town now, maybe we should phone him?
(2) If I take my pills regularly, the symptoms go away.
In (1), imagine that the hearer has said earlier in the conversation that
the relevant person is in town now. In the speaker's reference to this
previous statement, she could use since instead of if, with quite similar
112 Barbara Dancygier and Eve Sweetser
effect. (2) can naturally be interpreted to mean that taking pills regularly
makes the symptoms disappear-a causal interpretation related to
readings of because and since. There is thus a significant degree of
overlap among the uses of the conjunctions involved. But the potential
for such interpretations will often depend on factors other than the
conjunction itself; notice that when is a plausible substitute for if in (2)
but not in (1), suggesting that the lexical and constructional contexts are
important in determinining possible construals of the relations between
conjoined clauses.
In positing functional overlap between these conjunctions, we are by
no means suggesting that they are synonymous--or even that they have
specific senses which are synonymous. On the contrary, we will try to
specify the source of the similarities as well as to clearly delineate the
differences. Both similarities and differences are best accounted for not
by focusing only on the conjunctions themselves, but by describing the
range of constructions they participate in. The main goal of our analysis
is thus to distinguish aspects of constructional meaning which are
involved in various uses of each of the conjunctions, and to show how
particular formal parts of constructions combine with each other to
contribute to varying interpretations of the whole. 1
We will bring up three relevant parameters in trying to compare and
contrast the four kinds of constructions. First, we will consider the role
of causal models in interpretation. Causality is involved in the
constructions in a variety of ways and on various levels of
interpretation. Constructional aspects of meaning, and the cognitive
domains in which these constructions are interpreted, can be as
important as the choice of conjunction, in bringing causality into the
interpretation. The conjunctions will also be compared in terms of the
type of mental space set-up they engage in; this is often also specifically
signalled by the use of verb forms marking negative epistemic stance.
Finally, we will review the constructions with respect to their roles in
backgrounding or foregrounding elements of interpretation in different
clause order configurations.
Our analysis of meaning will be framed in terms of Mental Spaces
theol)' (Fauconnier 1985, 1997), an extremely general formal theory
which provides mechanisms for talking about the ways in which we
connect cognitive structures with each other. To some extent, traditional
treatments of conditional semantics in terms of Possible Worlds have
aimed at capturing similar insights. Somewhat in the way that an if-
clause can be said to set up a Possible World within which a then-clause
Constructions with if, since and because 113
is construed, we would argue that an ~clause sets up a Mental Space

which is the background for the construal of the then-clause. For
example, in If his computer gets repaired, then he'll finish the paper by
Friday, the ~clause sets up a space wherein the computer gets repaired,
and within that space, the speaker predicts that the paper will get
finished by Friday.
However, mental spaces are a more general mechanism than possible
worlds, referring not only to very partial cognitive "world" or
"situation" constructions as well as to more complete ones, but also to a
variety of non-world-like structures which can be connected and
mapped onto other cognitive structures. For example, the two mental
spaces consisting of our mental list of restaurant customers and our
mental list of their orders are mapped onto each other to allow
utterances such as The ham sandwich wants his check now, wherein the
name of the food item ordered can be used to refer to the customer who
placed that order (Fauconnier 1985[1994]). World-like structures are not
the only consistent, complex mental models which humans develop, and
which humans can interconnect with each other.
We have exemplified the use of if as a mental space builder.
Numerous linguistic devices can play such a role: in 1984, in the
picture, possibly, etc. All such expressions require that the content of
the clause or sentence be considered in a mental space other than the
base space-the reality space of the hearer and the speaker. In what
follows we will show how the conjunctions if, since, and because
participate in the construction of mental spaces and how the kind of
space being set up affects the way the conjunctions are used.
2. Mental spaces, causality, and levels of interpretation
We will start this section by outlining types of mental spaces which are
particularly relevant to our interpretation of the use of conjunctions. We
will then show how these types of mental space set-ups correspond to
levels of interpretation of constructions and how causal relations are
established at each of these levels.
2.1. Mental spaces and cognitive domains
Causal and conditional constructions, and particularly if-conditionals,

are known to vary widely in function. It would be economical and
possibly even elegant to be able to attribute some of this functional
diversity to a few specific parameters of interpretation. In fact, mental
spaces theory opens up the possibility for such an elegant treatment, by
allowing us to talk about different kinds or classes of mental spaces. For
example, in a more or less "standard" conditional such as (3),the
speaker sets up a space of mental content (cf Sweetser 1990,1996;
Dancygier and Sweetser 1996)-a space which is about a possible state
of affairs in the world, namely the computer getting repaired.
(3) Ifhis computer gets repaired, he 'llfinish the paper by Friday.
Within this content-domain space (further filled out by the speaker's

and hearer's general knowledge about the individuals referred to), the
speaker predicts an added aspect of the content of this mental space-
the paper getting finished by Friday. As argued in Dancygier (1993,
1998), prediction in the content domain represents the central use of
conditional constructions.
"now (not Friday yet)"
Future Space = P Counter-Future Space =

(neutral stance) ",p
"his computer gets repaired" "his computer doesn't get repaired"
Figure 1. If his computer gets repaired, he'll finish the paper by Friday.
As seen in Figure 1, a predictive conditional such as (4)

automatically also sets up an alternative space, where P does not hold
(no computer repair) and Q also does not hold (no finished paper by
Friday). Although this is not a logical property of conditionals, it is a
necessary property of the predictive use of conditionals, as we have
argued elsewhere (Dancygier and Sweetser 1996, 1997); (3) would not
be relevant as a prediction if the paper were just as likely to be finished
regardless of the computer repair.
By contrast, in an example like (4), the speaker does not predict
something about a good Thanksgiving on the basis of information about
seeing the addressee before Thursday.
(4) If! don't see you before Thursday, have a good Thanksgiving!2
Rather, the speaker sets up a discourse context, a speech-act space

wherein Thursday has arrived without her seeing the hearer, and then
utters a speech act which is to be taken as effective within that space.
This can't quite literally be true, since of course the well-wishing is
uttered in the current context and can't exactly be retracted; but the
conditional certainly modifies the context of interpretation from the
previously accepted one. In particular, the hearer cannot now reply with
but isn't it a bit early for that? As we might expect from the non-
predictive nature of (4), no alternative space is set up; that is, although
the speaker knows she may see the addressee before Thursday, she is
not providing for that separate contingency. Her good wishes are offered
regardless.
P: S doesn't see H between

Monday and Thursday
Figure 2. If I don't see you before Thursday, have a good Thanksgiving.

A conditional can thus set up, imagine, and negotiate possibilities in

either the world of linguistically described content, or in the world of
current speech-act context and performance, where the events and
participants are the Speaker, the Hearer, and the speech act with its
interpretive context. Speech act conditionals have been noticed as a
special and interesting case for some time (Vander Auwera 1986,
Sweetser 1990). Mental spaces theory provides a simple mechanism for
the description and analysis of the difference between them and content
conditionals: conditional constructions may build speech-act spaces as
well as content spaces.
What other sorts of spaces might speakers construct with
conditionals? Sweetser (1990) has argued for a distinction between
content conditionals and epistemic conditionals, which follow the
speaker's reasoning process in a conditional manner, as in (5):
(5) If he finished the paper by Friday, his computer must have gotten
repaired.
In the described world of content, the paper's being fInished is not a

condition for the computer's being repaired, rather the reverse. But in
the epistemic world, where knowledge of effects limits possible
conclusions about causes, knowing when the paper was finished is a
condition which may be sufficient to warrant an inference about
computer repairs. In (5), an epistemic space is set up, wherein the
speaker believes or knows that the paper was finished, and it is in that
space that she draws the conclusion about the repairs. (Possible
paraphrase: 'If I know (or we know) that the paper was finished, then I
(we) must conclude that the computer was repaired. ')
A further point here is that the speaker of (5) is very possibly not
doing initial setup of a space wherein the paper is known to be finished.
We might say that she is evoking, rather than actively setting up, that
space, if she is borrowing structure set up by an interlocutor who has
just informed her that the paper was fInished by Friday.
Dancygier (1993, 1998) has set out a further variety of conditional
uses, as in (6), wherein metalinguistic negotiation is the subject of the
conditional relationship:
(6) Chris wants you to meet her fiance, if that's the right word for
him.
the speaker perhaps thinks that Chris is using fiance as a euphemism for
'live-in partner.' Crucially, the speaker is not here construed as
negotiating either the facts of the case (the content), or conclusions to be
drawn from them in the epistemic domain; nor is she conditionally
performing the speech act of stating that Chris wants the hearer to meet
a particular individual. She is apparently metalinguistically commenting
on, or presenting as merely conditional, her use of the label fiance to
refer to that individual.
In this case, we need a somewhat more complex mental space
structure, which we have referred to (Dancygier and Sweetser
1996,1997) as a metalinguistic space. A metalinguistic space
(generically diagrammed in Figure 3 below) consists of a speech-
context space, within which there is a space of content being discussed
and a space of linguistic forms available for referring to that content.
Metalinguistic conditionality concerns mappings between the content
space and the form space within a particular context. Here, the speaker
constructs a particular form-content mapping (uses fiance to refer to
Chris' partner), and then sets that in the context of a particular
metalinguistic space by calling the mapping into question explicitly (if
that's the right word). By implication, such a conditional usage at least
suggests that other metalinguistic spaces could be imagined, with
different form-content mappings; however, in (6) there is no explicit
attempt to evoke other specific mappings (e.g., partner).
Figure 3. Metalinguistic space

Every speech act, of course, involves interlocutors and a context and

the interlocutors' ideas of each other's understandings of the content and
of the forms used to express it. So all of this mental space structure is
present and accessible in any speech situation. However, not all of this
structure is necessarily actively at stake in a given conditional use. (6)
demands that we bring all of these elements into our diagram overtly,
because in a metalinguistic conditional the content-label mappings
themselves are what are being conditionally represented and compared.
In (4), we needed to overtly represent speech-act context spaces
involving S and H to analyze a speech-act conditional, but the content-
label mappings were not part of the conditional relationship. Rather,
they were taken for granted as background to it. A diagram of (5) would
need to represent possible epistemic spaces of S, but would take the S/H
context and label spaces for granted as background.
Our claim, then, is that in all these cases, a conditional construction
involves setting up a mental space (in the case of if conditionals, this is
the job of the ~clause), and requesting construal of something (in if
conditionals, the then-clause or main clause) within that space. Much of
the diversity of interpretation can be attributed to the fact that the spaces
themselves can be quite diverse sorts of objects, related to the linguistic
form in a variety of ways. Crucial to this analysis is the idea that
domains related to the current speech interaction are generally
privileged with respect to mental space construction (cf Sweetser
1990): they are automatically "accessible" for interpretation, even
without explicit mention, in ways that other spaces are not. 3 This fact
has wide-ranging effects on interpretation, far beyond conditional
constructions. As noted in Sweetser (1990), broad classes of linguistic
forms show the same kind of possibility for multiple interpretations
depending on the level or domain accessed. Parallel to conditionals are
causal and adversative conjunctions (as well as many coordinate
conjunction usages, which are beyond the scope of the present paper):
(7) .Joe turned down the stereo because Sam was studying.
(Causal relation is between state of affairs described in P and
event described in Q, i.e. between contents of the clauses.)
(8) Sam is (must be) studying, because Joe turned down the stereo.
(Causal relation is between speaker's knowledge about content
ofP and speaker's conclusion about content ofQ.)
(9) Could you turn down the stereo, because I'm trying to study.
(Causal relation is between the contextual state expressed in P
and the speech-act performance of the request in Q.)
(10) OK, Chris introduced me to her partner, since we're being

politically correct.
(Causal relation is between contextual situation P and a
particular form-content mapping used in Q.)
(11) Since you're a linguist, what's the Russian wordfor "blender"?

(Speech-act level: the addressee's professional status causes or
enables the current speech act of questioning. )
Examples such as these can be multiplied with ease in the domain of

conjunction. Some conjunctions seem to prefer certain domains: since in
English has preferentially a temporal sense in the content domain4
(Chris has been studying since he came home), but has readily
accessible causaVenablement senses in the other three domains (a
speech-act domain example is [11]). Ranging farther afield, Hom (1985)
has laid out in detail the differences between content and metalinguistic
uses of negation. And the well-known contrast between deontic and
epistemic modal uses is interpretable in terms of the content/epistemic
contrast here invoked. Thus Joe must be home by ten, because I say so
imposes a modality of compulsion in the content domain of Joe's
getting home. But Joe must be home by ten, because I always see his
light go on then involves instead a metaphoric 'compulsion' of the
speaker's reasoning processes: 'I am forced to conclude that he gets
home by ten, because 1 know that 1 always see his light go on then' (cf
Sweetser 1982, 1990, Talmy 1988).
What all these examples share is an implicit contribution of the
speech setting to the interpretation: some part of the linguistic form (for
example, because) need not be interpreted directly with respect to the
content expressed by surrounding forms, but can be interpreted as
meaning something about the speaker's mental processes or the speech
interaction, even though those mental processes or that speech
interaction have not been explicitly mentioned. So conditionals are not
alone in allowing speakers to implicitly access speech-act, epistemic and
metalinguistic structure as well as content, in building an interpretation.
As we shall see, however, each of the conjunctions participates in
mental space set-up in a somewhat different way.
2.2. Conjunctions versus constructions: sources of causality in the

content domain
As we claimed above, causal relations in content mental spaces are

understood in terms of standard causal models, including an active
causation model wherein one event or state of affairs actively brings
about another. Content-level causality also includes enablement (cf
Sweetser [1990]). Causal because occurs in examples like He is tired
because he's been up all night, where staying up late is presented as the
cause of ensuing tiredness, and also in sentences such as He will come
and spend Christmas with us because I'm paying for his flight, where a
free ticket enables the person to spend Christmas with his family, rather
than causing him to do so. Such content-level causal models constitute a
major part of the actual interpretation of constructions containing
because and if.
2.2.1. Two ways of depicting causality: because and if
Causal conjunctions like because introduce causality or enablement into

the interpretation as part of their lexical semantics, although other
constructional features and the context help the hearer locate the causal
link in the appropriate cognitive domain. (As we will try to show in
section 4, intonation patterns and clause order in constructions with
because signal content and non-content interpretation of causality in a
fairly regular way.) In the cases of non-causal conjunctions, causal
interpretations may arise from other aspects of the context and
surrounding grammatical constructions: She insulted me and I got angry
is easily interpreted as meaning that the insult caused the anger, on the
basis of iconic interpretation of clause order, plus real-world knowledge
about likely results.
Although because lexically expresses causality, the direction of the
causal chain is marked by the conjunction's placement in a larger
construction: the because-clause presents the cause, the main clause
presents the result. Consequently, even though the conjunction seems to
be solely responsible for the causal meaning, the way in which causality
participates in the interpretation of the sentence is also dependent on the
subordination construction. It may initially seem trivial to focus on the
obvious fact that the because-clause is marked as representing the cause,
as opposed to the caused event, in the causal relation. But as Lightbown
and Spada (1993) observe, children learning their first language may in
fact hypothesize that principles of iconicity override meanings
introduced at the lexical level and produce sentences like You took the
towels away because I can't dry my hands. In such cases, the strategy
used is to present events in their sequential iconic order (the towels
disappeared from the bathroom first, and consequently the child had
nothing to dl)' her hands with later on). The child needs to learn not only
the causal meaning of because, but also the way to use it
constructionally to mark causal connections correctly. 5
In many content-domain causal conjunction examples, only the
causal relation itself is asserted, while the two events or situations
between which the causal relation holds are presupposed background. In
the classic I'll do it because I want to, and not because you told me to,
the speaker does not assert either that she will do as she is told, or that
she wants to do so, or that she was told to do so: but she is necessarily
asserting that her actions are caused by her wishes and not by the
addressee's commands. .[fclauses generally set up a new mental space
distinct from the Base space of the speaker's assumed construal of
"reality", but because-clauses do not set up such a space. Main clauses
of content-domain because constructions decribe situations which may
already be known to hold in the interlocutors' current base space. 6 The
because-clause then offers a causal explanation of how that situation
came about, also not necessarily on the basis of brand-new information.
The causal relationship is at the core of the message.
Since is similar to because in being semantically causal, and in that
both are often said to presuppose the truth of their complements. These
differences interact with other parameters, such as information structure
and typical clause order. Since will be discussed in more detail in
sections 3 and 4, where we focus on epistemic stance and information
structure.
The conjunction if presents an interesting case. Unlike because, it is
not lexically causal. In the content domain, however, if-sentences
typically express causal or enablement relations among events. (12), for
example, would typically convey the message that the conditioner will
cause the hair to get softer.
(12) Ifyou use this conditioner, your hairwillfeel much softer.

Similarly, in example (3) above, it-is likely to be assumed that the repair
will enable the owner to start using the computer again.
Dancygier (1993, 1998) has argued that causal meanings enter the
interpretation of content conditionals via conditionals' primary function,
prediction. 7 Prediction in conditionals is a type of reasoning which
consists in setting up a hypothetical (typically future) mental space and
attempting to predict its consequences based on knowledge of typical
cause-effect chains and general world-knowledge. In (12) above, the
speaker sets up a space in which the hearer applies the recommended
conditioner, and predicts in that space that the hearer's hair will become
softer as a result. Such causal interpretations tend to arise in conditionals
which set up content mental spaces, where causation-based predictions
can be made on the basis of assumptions of sequential correlations
among events or states of affairs and general knowledge. But the
interpretation of speech-act or epistemic conditionals also involves
setting up causal relations: the speaker of Since you're a linguist, what
do you think of Chomsky? constructs relationships of causality and
enablement as part of the background to the act of asking a question.
The most notable formal feature distinguishing English predictive
conditionals is the combination of verb forms: for future-reference
conditionals, a present-tense verb form is used in the If-clause, and the
modal will in the main clause (cf Fillmore 1990a,b). This verb form
pattern characterizes the whole sentence (not the two clauses in isolation
from each other) as representing a certain kind of reasoning.
Predictiveness (and, consequently, causality) is part of the interpretation
of certain If-constructions as wholes, rather than of the conjunction if or
any other lexical element of the sentence. Let us also note that such
constructions do not assert any of the clauses. The If-clause cannot
express an assertion because the conjunction sets up a hypothetical
mental space, different from the base space where assertions could be
made. The main clause describes the predicted result in that same
hypothetical space, so it is not asserted either. What a predictive
conditional asserts is the causal dependency and correlation between the
events or states of affairs described by its clauses.
To sum up, content-domain conditionals are interpreted causally as a
result of predictive uses which are explicitly marked by formal markers
such as verb forms. This is one of the basic differences between
sentences with because, and those with if. In the former, the conjunction
itself introduces causality; in the latter, constructional meaning is the
source of the causal interpretation. Though the constructions with
because and if are different in a number of respects (verb forms, clause
order, mental space set-up, to mention just a few), they can both convey
information about the causal relations between the content of the

subordinate clause and that of the main clause. These differences and
similarities are related to other aspects of the constructions, such as their
epistemic stance and clause order, which we will consider in the
sections to follow.
2.3. Causality in other domains
As seen above, clauses marked by causal and conditional conjunctions

are often used to set up or evoke epistemic, speech act, or metalinguistic
mental spaces, in which relations are established among beliefs or
speech acts, not events or states of affairs. Does causality play any role
in such spaces?
Causal models are certainly present in the interpretation of non-
content domain constructions, but we will have to distinguish more than
one level of causal links to show how they contribute to the overall
interpretation. In the content spaces, causal links can be established
between events or states of affairs described. In the other domains,
similar links can be established between beliefs or between beliefs and
speech acts. In example (8) above (Sam is (must be) studying, because
Joe turned down the stereo), the causal relation is between the speaker's
knowledge about the content of P and the speaker's conclusion about the
content of Q. Similarly, in (11) (Since you're a linguist, what's the
Russian word for "blender"?, the belief expressed in the since clause
causes or enables the act of asking about Russian. We can interpret
metalinguistic uses similarly. For instance, in example (6) above (Chris
wants you to meet her fiance, if that's the right word for him) the
speaker uses the word fiance while also pointing out that the (possibly
mistaken) belief in fiance being the right word is all that enables her to
do so.
It is known that causal/enablement relations in non-content domains
often go in a direction opposite to the causal links in the content domain
(Sweetser 1990, Dancygier 1993, 1998). This is especially common in
the epistemic domain, where the speaker is often reasoning from effect
to probable cause. In (8), for example, we assume that Sam's need to
study normally causes Joe to tum his music down, so the fact of the
music being turned down enables one to infer that Sam is studying. In
fact, it is possible for epistemic sentences to be ambiguous, depending
on the underlying assumption about a content causal chain. For
example, a sentence like If Tim went to the embassy, he has his visa
documents can have two epistemic interpretations. If the speaker
believes that Tim was going to obtain his visa documents at the
embassy, she may use the information about Tim's visit there to infer
that he now has his documents. If, on the other hand, the speaker
believes that acquiring the documents enabled Tim to go to the embassy,
she concludes that he had them prior to the visit. In either case,
knowledge of cause-effect links in the content domain is used in a
reasoning process carried out in the epistemic domain.
Causal models can thus participate in the construction of meaning on
various levels. This is due not only to our understanding of causality,
which allows us to locate it in various types of mental spaces, but to our
understanding of the relations among those spaces, and, more generally,
to our understanding of speech interaction. This understanding is
dependent on what we may see as a nested complex of mental spaces.
At the center of this complex is some representation of the content of
what is said: events or states of affairs which are referred to by the
speaker. The content space need not make any reference to the speaker's
reasoning processes or to the speech exchange itself (although these
may also be addressed as explicit COlltent: I am asking you why you need
an extension on your paper). However, reasoning processes relevant to
the speech interaction almost inevitably make reference to the content
space: the speaker may, for example, be reasoning specifically about the
structure of the content space. In He must be home; I see his coat, the
speaker's use of must refers to her understanding that her reasoning
processes are under compulsion, not that the described subject's actions
are compelled; but the reasoning processes about the act of returning are
necessarily based on the speaker's understanding of the content world,
where both the return and the coat's presence reside.
The structure of the speech exchange will necessarily involve both
the idea that some content is expressed, and the understanding that the
speaker has reasoning processes related to that content, plus the added
understanding that there are a speaker and a hearer interacting in the
current speech setting, with all the social interactional issues attendant
upon conversational exchange. As we have seen, each of these levels of
mental space structure may involve causal models appropriate to the
level. But the nesting effect explains how different causal models can
enter the interpretation at the same time. This includes the possibility of
exploiting a cause-effect chain wherein P causes Q, as the necessary
background for an inferential chain wherein a belief that Q leads to a

conclusion that P.
3. Epistemic stance
As noted above, predictive conditionals with future reference typically

display a characteristic verb form pattern in English, with a present
tense verb in the protasis and the predictive modal verb will in the
apodosis. In fact, the same pattern is found in constructions with
temporal conjunctions like when: If/when we find the paper, we'll xerox
it right away. If and when have similar functions in this example, in that
both mark backgrounds to main-clause predictions.
The central difference between if and when is that when commits the
speaker to the reality of the space described in the when clause, even if
that reality will only take place in the future. If, on the other hand,
makes no such commitment. This contrast exemplifies the broader
phenomenon of epistemic stance (Fillmore 1990a,b), a parameter which
is by no means restricted to temporal and conditional clauses. When
identifies the speaker's beliefs with P as a description of the real state of
affairs (expresses positive epistemic stance); if, on the other hand,
expresses the speaker's lack of full positive stance with respect to the
content. The non-positive stance of if need not commit the speaker to a
negative or skeptical stance, but does indicate that she thereby distances
herself from full commitment to the contents of the ~clause. Other
aspects of a conditional construction may go further, and explicitly mark
the speaker's leaning towards non-belief in the reality of the described
situation. Sentences such as If we found the paper, we would xerox it
right away, If we had found the paper, then we would have xeroxed it
right away, and Even if we found/had found the paper, we wouldn't
xerox/wouldn't have xeroxed it right away express negative epistemic
stance towards the content of the conditional clause-and hence also
towards the content of the main clause, since that is situated in the
mental space defined by the conditional clause. The expression of
negative epistemic stance is formally marked by the use of verb forms
which in earlier work we have dubbed distanced, which pick up the
Present + will predictive pattern and backshift it, that is, contribute
added layers of past morphology. 8
We have seen examples of positive epistemic stance (with when) as
well as neutral and negative epistemic stance (with ij). We noted that
these constructions use a special predictive pattern of verb forms, in

both non-distanced and distanced format, and that the distanced forms
correlate with negative epistemic stance.
Since and because, our two remaining conjunctions, express positive
epistemic stance~ the clauses in their scope are seen as representing
"factual" or presupposed information. This aspect of their use seems to
result from the mental space set-up they engage in; in contrast with
conditionals, they do not build new hypothetical spaces, but instead
evoke aspects of whatever space is currently the base space. They also
do not require any special pattern of verb forms. As one might expect,
they do not allow the use of distanced verb forms, which are appropriate
only in the cases of hypothetical mental spaces marked with negative
epistemic stance.
However, not all conditionals are predictive. We have shown above
that causality-asserting predictive reasoning (and, consequently, the use
of predictive and distanced verb forms) is necessarily characteristic only
of conditionals setting up their mental spaces in the content domain. It
has been noted (cf Dancygier 1993, 1998; Dancygier and Sweetser
1996, Sweetser 1996) that the use of if in other domains need not
involve any restrictions on the use of verb forms, and that such
conditionals need not necessarily bring up another alternative space9 .
Recall our earlier discussion of speech-act uses of since such as Since
you're a linguist, what do you think of Chomsky? A parallel speech-act
conditional could serve a similar function: imagine a context where the
interlocutor has stated that he is a linguist, and the speaker says If you're
a linguist, what do you think of Chomsky? Here the speaker does not
intend to compare the space where the addressee is a linguist with other
possible spaces, nor to predict the speech act, but only to situate the
actually performed speech act in a well-defined space. In such non-
predictive cases, ~clauses may rely on context in ways which are
reminiscent of the more positive epistemic stance of since-clauses.
Are we now calling into question our previous argument that ifhas a
non-positive (neutral) epistemic stance, while since has an overtly
positive stance? One might initially feel that (13) expresses a positive
stance regardless of whether since or if is used.
(13) If/Since he's (so) hungry (as you say he is), he'll want a second
helping.
Sentences like (13) certainly suggest that the truth of the proposition
('he's hungry') has been asserted in earlier discourse. Rather than
setting up a novel space, they evoke a space already contextually
accessible. However, referring to a space available in previous discourse
does not necessarily mean that the speaker always aligns herself with its
content to the same degree. As argued by Dancygier (1998), ~f indicates
non-commitment to the clause it marks: but non-commitment covers a
wide range of possible attitudes, from strong disbelief to near-
commitment. Even in an example such as (13), where ~fand since seem
interchangeable, since sounds more completely positively aligned than
if, although in the presence of other strong indications of positive
attitude, the degree of non-commitment conveyed by if may be
negligible.
So there are important differences between if and since. If maintains
a neutral stance, while since takes a positive one. Also, since, unlike ~f,
has explicitly temporal senses. However, in appropriate contexts, there
is predictable functional overlap between the two conjunctions-
predictable, in the sense that it falls out naturally from a mental space
analysis of the constructions and contexts in question. But in defining
the possible contexts of overlap, we need a clear understanding of the
factors which make some particular assertion or claim accessible to, or
assertable by, some particular participant in the speech interaction. What
licenses a speaker to treat a particular content as already accessible to
the hearer, or as not being subject to questioning by the hearer? We shall
make use of the standard constructs of speech act theory to define
contexts where if and since are naturally isofunctional. As will become
evident, the relevant factors have a general application to the analysis of
the relationship between space-builders and informational structure:
space-building background information has to be brought up in a way
appropriate to its status in the speech context.
When fills a genuinely temporal space-building function, as does
since when used to mean 'from some past time to the present.' Non-
temporal since, however, builds up causal relations; it states a causal
relationship between the space which it evokes (rather than constructs)
and the main clause. An ~clause may, as in (13), merely re-enact the
building of an already accessible space--or it may build a truly novel
space, as in many predictive constructions. A causal since-clause,
however, always evokes a space which is presupposed by the speaker,
as accessible already to the audience.
Discourse accessibility of a mental space can arise in a variety of

ways. One of these ways is the speaker's right to access her
interlocutor's prior conversational record, quoting or summarizing or
making inferences from it.
(14) Since he's so inefficient, maybe you should make a formal

complaint.
(The speaker accesses her interlocutor's prior claims about
inefficiency, as background to her suggestion.)
Since, with its positive epistemic stance, marks (14) as explicitly

accepting the claim about the inefficiency-although of course this
acceptance may be insincere or ironic. We could expect the speaker to
have used if here, if she were not fully convinced or simply wanted to
avoid taking sides. Another source of evocation is the immediate
physical context of the exchange. In (15), since is appropriate in two
situations: one where the speaker is expressing her own knowledge
about the rain, and one in which the hearer has reported the rain to the
speaker who has not observed it firsthand:
(15) Since it's raining, we can't go outside after all.
As (15) shows, contextually evident information should normally be

marked with since rather than if, unless for some reason doubt is being
thrown on its validity. For example, a speaker who is physically
examining the hearer's clothing to decide whether it is warm enough or
weatherproof enough could say:
(16) Since you're wearing those shoes, we'd better not go to the
beach.
But for a speaker considering the possibility of the hearer's changing

into sneakers before departure, ifwould be a valid choice.
Typically, then, the context and the speaker's evaluation of the
situation (and in particular, her epistemic stance towards it) dictate the
choice between if and since. Certain categories of information are by
their nature presupposed to be known to the speaker, or to the hearer,
while others are not. In particular, speaker-based preconditions for a
speech act are presumed to be accessible to the speaker, while hearer-
based ones are not. Hearer-based preconditions, on the other hand, are
presumed to be uniquely accessible to the hearer; the speaker cannot

treat them as presupposed unless the hearer has already put them on the
conversational record.
Predictably, the non-presuppositional nature of if contrasts with the
presuppositional semantics of since in marking clauses expressing
speech-act preconditions. A speaker may hypothesize, but not
presuppose, the hearer's willingness to answer a question or comply
with a request or tolerate a potentially inappropriate remark:
(17) If/#Since it's not rude to ask, what made you get interested in
linguistics?
(since would be acceptable only if the hearer had already made it
plain that the question was acceptable.)
(18) If/#Since you don't mind, could you hold this stack ofbooks for a
moment?
(since is acceptable only if the hearer has indicated he doesn't
mind.)
(# is here used to indicate that the sentence so marked is grammatical,

but would be unacceptable without special context; one would have to
imagine that the addressees of (17)-(18) had just said things like, No,
it's not rude to ask about my personal intellectual development, or qr
course I don't mind holding somethingfor you.)
On the other hand, speaker-based preconditions demand a positive
epistemic stance; it is normal for a speaker to presuppose her authority
on these subjects. She does not need to hypothesize, normally, about
whether or not she is speaking technically:
(19) Since/#IfI'm speaking technically, I'll call it a misdemeanor and

not a crime.
(The if form would be acceptable as a general condition, meaning
'Whenever I'm speaking technically, ... ' It cannot, unlike the since
form, refer to the current speech act.)
And if the speaker is making a request on the preconditional basis of her

own desires, she is presumed to know (and not hypothesize) her stated
reasons for those wishes:
(20) Since/#If I'm stranded without a car, could you drop me at the
BART station?
If for some reason the speaker were not the most knowledgeable person
about her own access to a car, then if would seem more acceptable in
(20). (Imagine a situation where H, a service station employee, has just
told her that her husband picked up the car an hour ago, while she was
expecting to pick it up and drive home in it.)
Speaker and hearer are each presumed to be authorities on their own
states of mind and body, and not on each other's. Hence the speaker
should normally use since when evoking her own mental and physical
states, and ifwhen referring to the hearer's:
(21) If/#Since you're really tired, maybe we should stop work now.
(since is acceptable if the hearer has himself indicated his
tiredness. )
(22) Since/#IfI'm totally exhausted, maybe we should stop work now.
By this analysis, since should only be employed when the hearer

must concede the speaker's right to take a positive epistemic stance to
the content of the space-evoking clause. There remain a variety of
reasons to use if rather than since. First, of course, the material may
really be one towards which the speaker genuinely does not take a
positive stance (truly neutral or negative-stance hypothesizing), or about
which the speaker must defer to the hearer (as in hypothesizing about
the hearer's tiredness). But inheritance from the conversational record is
especially complex with regard to epistemic stance. On the one hand, if
the hearer has just said that it is raining, and there is no reason to doubt
his word, the simplest scenario may be to adopt it as given and mark it
with since. On the other hand, if (cf Akatsuka 1986; Schwenter 1997,
1998) marks a non-positive stance, which may mark genuinely reserved
judgement (If she was at· the party, I didn't see her), or may just
acknowledge that this claim is not the speaker's own. Quotative if can
thus be more polite than since, or less so, depending on whether the
issue is acceptance of the contents or taking credit for the claim.
How does since interact with the various parameters of interest to us?
Obviously, being positive in epistemic stance like when, it does not
cooccur with verb forms marking negative stance. Like if, it allows
interpretation of the relationship between the two clauses as existing in
different domains. There are content, epistemic, and speech-act uses of

since; and it is easy to find metalinguistic ones as well (cf Sweetser
[1990; chap. 4]):
(23) Since my plane was delayed, I'm phoningfrom London.

(Content the plane's delay-a situation treated as already
familiar to the hearer-is what necessitates the phone call.)
(24) Since he's phoning from London, his plane must have been
delayed.
(Epistemic: the presupposed knowledge of the phoning causes the
conclusion about the plane's delay.)
(25) Since you're so smart, when was George Washington born?

(Speech act the on-the-floor claim that the hearer is smart is what
provokes the query.)
(26) Since you're so picky, I'll just put another "trout" on the grill.
(Metalinguistic: The addressee has just corrected the speaker's
use of the term salmon to refer to the wrong species of fish. )
Sweetser comments that since may be preferentially used in the non-

content domains, precisely because of its presuppositional status; there
is a relationship between presupposed or given background-clause
material and non-content usage (cf. Dancygier 1998), as Nikiforidou
(1990) has also pointed out for Greek conditional forms. It is in the non-
content domains that it is often most useful to contextualize a main
clause, to make clear on what grounds or against what background
assumptions the speaker is performing the present speech act, drawing
the present conclusion, or using the present linguistic label.
Unlike when, which at the content level can refer to cotemporality at
any time relative to the speaker's present, since demands present or past
(realis) tenses in its content-level uses. It is not used to make content-
level predictions, and does not participate in the predictive use of tense;
backshifting does not occur in since-clauses, and in general predictive
uses of will in since-constructions can be interpreted as expressing
prediction at some other level than that of content. This may follow
from the fact that a conditional's business is to set up a (possibly
entirely unreal) space, as background to some further inferences, while
that of a causal conjunction is to link situations, claims, and conclusions

causally to a given background.
4. Clause order, space set-up, and causal/conditional relations
Given the flexibility of clause order in conditional and causal

constructions, how do different orderings of the main and subordinate
clause interact with space building? This is a very general issue indeed:
not only if=-conditionals, but most of the broader class of conditionally
used constructions, involve conventional use of the order of the
expressions of antecedent and consequent. It seems intuitively natural
that the space-builder clause should precede the clause which elaborates
the space; and this is of course consistent with the observed tendency for
if=-clauses to precede their main clauses. 10
However, the other order is certainly also possible, although (as we
shall see) it is not possible for all conditionals-the clause containing
the apparently space-building (or more neutrally, space-delimiting)
material can follow the main clause. Doesn't this violate some basic
understanding of how space-structuring proceeds? The answer is that
perhaps it might, if the cognitive processing of mental spaces were done
in a vacuum; but in fact, context may already have done a great deal of a
speaker's building for her, in which case an explicit space-building
clause may only be needed to confirm or clarify an already-constructed
mental space.
Looking first at if=-conditionals, we may note that all three of the
following organizations ofP and Q are possible:
(27) ifP, Q If the home computer breaks down, I'll work at my

office.
(28) Q, ifP I'll work at my office, if the home computer breaks

down.
(29) Q ifP I'll work at my office if the home computer breaks down.
(27), diagrammed in figure 4, follows our intuitively basic strategy of

first building a space defined by P, then adding the assertion of Q to the
structure of that space. Note that the relevant space could still be in
some sense pre-built by the discourse; (27) could respond to What 'II you
do ifyour home computer breaks down?
Set up conditional space defined by P, then assert Q within

that space. Space setup first, then contents.
There may be a background assumption that P is
already a cognitively accessible possibility, but Q is
not accessible in this way.
(Implicitly set up """'P, ~ space as well.)
Figure 4. If the home computer breaks down, I'll work at my office.
(28) instead asserts Q, without first specifying in what space Q applies.

Here it seems almost necessary that some background space be already
established, to make sure that the hearer will be adding Q to the relevant
space. The speaker can make sure that has been achieved by adding a
disambiguating if P of her own, following Q, but this would be an odd
way to introduce a genuinely new space. Note that in both (27) and (28)
we have standard predictive conditional structure, with the implicit set-
up of a """'P, ......,Q space alongside the P, Q space; the difference lies not in
the relations expressed but in the order of their construction and its
relation to the informational context of the discourse.
The contrast between (28) and (29) exemplifies a more general
contrast between conjuncts juxtaposed with and without comma
intonation. (29) would be an answer to Would you ever work at your
office? or When will you work at your office?, rather than to What will
you do if the home computer breaks down? Its major assertion is the
identity of the conditional relation under which the working in the office
would happen; the main clause content is already "on the floor." In
mental space terms, (29) begins by expressing a prediction Q, but in an
intonation pattern which makes it clear that Q is not being
independently asserted, and is therefore unlikely to belong to the base
space. The focal stress (marking new information), and the final
intonation fall (marking the end of an assertion), happen only once we
are identifying the space P within which Q is to be placed. Q is "old
information," while its space-setting is not.
So (27)-(29) correspond to different sequences of instructions for
building what are rather similar final mental space constructs. (27) is the
"canonical" sequence of setting up a space and then making a prediction
within it. (28) might be said to involve at least a potential pre-built P-
space, to which the utterance adds Q, subsequently confirming with an
~clause that yes, P was the intended space for elaboration by Q. (29)
could be seen as involving a presupposed Q, and asserting the
alternative space structure which links that Q with P rather than with ,-..;P.
These differing sequences of instructions are useful in different
contexts, just the way physical instructions to construct the same object
might usefully differ depending on which components of the object
were already pre-assembled or ready to the addressee's hand.
The three space set-up options presented in (27)-(29) differ in
foregrounding or backgrounding causal relations. (27), being a typical
predictive conditional, asserts the causal relation (rather than any of the
clauses) in the way discussed above. (28) is different in that it initially
asserts Q, and thus pushes the nature of the relation between P and Q
into the background. In fact, as Dancygier (1998) claims, this clause
order in conditionals is perhaps most common among speech act and
metalinguistic conditionals, which often perform a speech act in the
current space, and use the ~clause to negotiate the background structure
of that space. In (29), the stress on identifying the space in which the
"old" prediction is valid gives additional focus to the causal relation on
which the prediction is based. Dancygier (1998) argues that this clause
order and intonation pattern is available only to predictive, causality-
asserting conditionals, while all the other types of space-settings are
excluded. (29) thus represents a clause pattern which constructionally
requires causality to be asserted. 11
Similar clause order options are available for because-constructions,
but the preferences are distinctly different. While ~clauses typically
precede their main clauses, clauses with because typically follow theirs.
This regularity (observed by Ford 1993) is influenced by the fact that
because-constructions do not set up new spaces, but establish causal
relations in the base space. It is thus common for these constructions to
first mention the fact requiring explanation, and then establish the cause
of the state of affairs in question. Consequently, the main contrast
among because-sentences is between two intonation contours available
in the Q because P pattern.
Sweetser (1990, 82ff) discusses parallel examples involving causal
conjunction, as examples of Chafe's (1984) generalization about
"bound" (commaless) and "unbound" (comma) intonation: comma
intonation demands an interpretation involving assertion of both the
main clause and the causal connection between the clauses, while
commaless intonation is readily given an interpretation which
presupposes the assertion (in this case, prediction) embodied in the main
clause, and newly asserts only the causal connection between the
clauses. Thus (30) is likely to be understood as asserting that the reason
for Anna's love is Victor's resemblance to her first love, but
presupposing that Anna loves Victor; (31), on the other hand, asserts
that Anna loves Victor, as well as the reason why.
(30) Anna loves Victor because he reminds her o..f herfirst love.
(31) Anna loves Victor, because he reminds her ofherfirst love.
Sweetser comments that the bound interpretation is unlikely to occur

in conjunction with a speech-act or epistemic reading of the causal
relationship between the clauses. Thus, (31) but not (30) is readily
susceptible to non-content causal readings-for example, to conveying
the message that the inference about Anna's love for Victor is caused by
knowledge about a resemblance to her earlier sweetheart, rather than the
content-level message that the love is caused by the resemblance.
Sweetser's explanation for this contrast is that it would be exceptional to
presuppose the main clause of a speech-act or epistemic conditional,
since this would involve presupposing the very speech act being
performed, or the new conclusion being drawn.
The mental-space structures of (30) and (31) above are similar to the
conditional examples (28) and (29). The main difference between the
causals and conditionals is the preference for PQ clause order in
conditionals and QP clause order in causals, which appears to be a
consequence of the differing space-building functions of the two

conjunctions. Space-building background has a natural reason to
precede the content of a newly-built space, while a base-space state of
affairs in need of explanation may well reasonably be described first,
before giving the explanation for why or how it came to be that way.
In contrast with because and when, since seems awkward in the
commaless Q conjunction P construction: #1'1/ marry you since you're
so sweet, or #He invited me since I live next door are strange, though it
seems normal to say Since I live next door, he invited me or I'll marry
you, since you're so sweet. Since does not seem to allow foregrounding
or focus on the assertion of the causal relation, perhaps does not even
assert it 12 . In this, since contrasts with a wide range of temporal
conjunctions (before and after behave like when) as well as with
because and if. Compare the following examples of conjunctions in
constructions which employ stress or syntactic focalizing structures to
mark the focus of contrast:
(32) I love you because you're sweet, not because you're rich.
(33) We'll eat when Sue arrives, not now.
(34) They had coffee after lunch, not before dinner.
(35) #He invited me since I live next door, not since I'm his boss.
(36) It's because you're sweet that I love you, not because you're rich.
(37) It's when Sue arrives that we'll eat, not now (/not when Joe
arrives).
(38) It's after lunch that they had coffee, not before dinner.
(39) #It's since I live next door that he invited me, not since I'm his
boss.
Various causal and adversative conjunctions fall on one side or the other
of this contrast: though and while behave like since, and phrases like in
spite of the fact that or despite the fact that behave more like because.
Interestingly, since in its temporal sense does allow assertion and focus
of the temporal relation:
(40) It's since the pollen came out that I've been sniffling, not since I
moved to the new apartment.
(41) I've been sniffling since the pollen came out, not since I moved to
the new apartment.
Of the two causal conjunctions we have looked at only one, because,

readily allows focus on the causal relationship itself Interestingly,
another conjunction which can foreground the causal link in this way is
if, in its predictive use.
A look at clause order and intonation options thus reveals further
similarities and distinctions between the constructions under scrutiny.
Both if and because constructions allow the use of sentence-final
adverbial clause placement and commaless intonation to unambiguously
assert the causal relation between the contents of the clauses, while the
"comma" intonation usually renders the sentences ambiguous between
domains. On the other hand, since does not seem to allow foregrounding
or focus on the assertion of the causal relation. These facts are added
evidence of the varying degrees to which asserted, presupposed, and
construed causality are integral parts of building these constructional
meanmgs.
5. Conclusion
The functions of causal and conditional constructions can be compared

along a variety of parameters. We considered both the contribution of
conjunctions as lexical items, and the constructions in which these
conjunctions appear. Particular conjunctions participate in mental space
configurations in differing ways: some conjunctions set up their own
spaces, some evoke spaces set up in preceding discourse, some merely
elaborate the base space of the ongoing interaction. Further, important
differences arise with respect to the cognitive domains in which the
spaces are set up, but significant aspects of the overall interpretation
result from the spaces being nested one within each other.
Such parameters of interpretation correlate significantly with formal
features of constructions, such as the choice of verb forms, clause order,
or intonation. Some of these formal features may be claimed to invite
certain types of mental space set-up or contextual use, regardless of the
choice of conjunction. One good example of such a feature is clause
order and accompanying intonation, since we observe recurring patterns

of interpretation in constructions which use the same clause order and
intonation patterns.
It is often difficult to isolate the study of a small group of linguistic
forms. Temporal and adversative conjunctions, which were not
originally meant to appear in our data, had to be mentioned along with if
or since; otherwise, important facts about the behavior of our primarily
chosen conjunctions would have to be missed. As we have also claimed
elsewhere (Sweetser 1990; Dancygier 1993, 1998) coordinate
conjunctions and even conjunctionless and paratactic constructions 13
also share some constructional characteristics and aspects of
interpretation with if, because, and since.
What is needed is an analysis which uses parameters of
constructional meaning (verb forms, clause order, intonation, use of
mental space builders) to outline the range of constructions which
participate in the construal of related meanings (causality, sequentiality,
conditionality), and explores the similarities and differences between the
constructions with respect to these parameters. We hope to have shown
that such an analysis can reveal generalizations and aspects of
constructional meaning which accounts of individual conjunctions can
describe only partially.
Notes
* For Ekkehard Konig, on his sixtieth birthday.

1. It is impossible to cite all the works which have helped us to develop an
understanding of the semantics of grammatical constructions. Among salient
ones are Fillmore, Kay and O'Connor (1988), Fillmore and Kay (1999),
Langacker (1987, 1991a, 1991b), and Goldberg (1995). Sweetser (1999) deals
generally with the relation between lexical meaning and constructional
composition.
2. We are grateful to the late Suzanne Fleischman for this example as well as for
characteristically insightful comments on it.
3. By "automatically accessible," we mean that speakers do not need any special
shared experience, knowledge or context besides the fact of the speech
exchange, in order to have implicit access to these spaces.
4. Though see section 3 for a full discussion of uses of since in different
domains.
5. Traugott (1982) gives an illuminating discussion of the differences between
earlier stages of English, wherein both clauses in a causal relation were
marked, and the modern English situation.
6. Or, of course, in whatever space is currently being added to; the crucial fact is
that no new space need be set up when because is used.
7. For discussion of the relationship of then to predictive function, see
Dancygier and Sweetser (1997), Schiffrin (1992), and Iatridou (1991, 1994).
8. Conditionals with distanced form have been discussed extensively as
"counterfactual" or "unreal". The discussion of this issue exceeds the limits of
the present paper. The reader is referred to Sweetser 1990, Dancygier 1993,
1998, Dancygier and Sweetser (1996, 1997). For reasons why distanced
conditionals should not be treated as counterfactual, see Comrie (1986); for
discussion of the crosslinguistic connection between past tense forms and
distanced forms, see James (1982) and Fleischman (1989).
9. Cf. Dancygier and Sweetser (1996, 1997), for discussion of predictive and
non-predictive epistemic and metalinguistic conditionals, and for analysis of
reasons why content conditionals are necessarily predictive and speech-act
conditionals necessarily non-predictive (see also Sweetser (1990), Dancygier
(1998).
10. Cf Haiman 1980, 1986; also Ford 1993 on because vs. if.
11. All the patterns just discussed are also found among even if constructions,
with the same priorities and restrictions involved.
12. As Elizabeth Couper-Kuhlen has pointed out to us, this incompatibility with
non-comma intonation brings us back to Sweetser's (1990) original concern
about the preference of since for non-content uses, and raises doubts as to
whether examples like (23) should really be treated alongside other content
examples.
13. For discussion of parallels with paratactic conjunction, see Lakoff (1971),
Grice (1978), and Sweetser (1990).
References
Akatsuka, Noriko
1986 Conditionals are context-bound. In: Traugott et al. (eds.), 333-52.
Chafe, Wallace L.
1984 How people use adverbial clauses. BLS 10: 437-449.
Comrie, Bernard
1986 Conditionals: a typology. In: Traugott et al. (eds.), 77-99.
Dancygier, Barbara
1998 Conditionals and prediction: time, knowledge, and causation in
conditional constructions. Cambridge: Cambridge University Press.
1992 Two metatextual operators: negation and conditionality in English
and Polish. BLS 18: 61-75.
1993 Interpreting conditionals: time, knowledge and causation. Journal of
Pragmatics 19: 403-434.
Dancygier, Barbara and Eve Sweetser
1996 Conditionals, distancing and alternative spaces. In: Adele Goldberg
(ed.). Conceptual structure, discourse and language, 83-98. Stanford
CA: CSLI Publications.
1997 Then in conditionals. Cognitive Linguistics 8/2: 1-28.
Davis, W.
1983 Weak and strong conditionals. Pacific Philosophical Quarterly 64:
57-71.
Fauconnier, Gilles
1985 Mental spaces. Cambridge MA: MIT Press [Reprinted 1994,
Cambridge: Cambridge University Press].
1996 Analogical counterfactuals. In: Fauconnier and Sweetser (eds.), 57-
90.
1997 Cognitive mappings for language and thought. Cambridge:
Fauconnier, Gilles and Eve Sweetser (eds.)
1996 Spaces, worlds and grammars. Chicago: University of Chicago
Press.
Fillmore, Charles J.
1990a The contribution of linguistics to language understanding. In: Aura
Bocaz (ed.). Proceedings of the first symposium on cognition,
language, and culture, 109-128. Universidad de Chile.
1990b Epistemic stance and grammatical form in English conditional
sentences. CLS 26: 137-162.
Fillmore, Charles J., Paul Kay, and Mary Catherine O'Connor
1988 Regularity and idiomaticity in grammatical constructions: the case of
let alone. Language 63/3: 501-538.
Fillmore, Charles and Paul Kay
1999 Grammatical constructions and linguistic generalizations: the what's
X doing Y? construction. Language 75/1: 1-33.
Fleischman, Suzanne
1989 Temporal distance: a basic linguistic metaphor. Studies in Language
13/1: 1-50.
Ford, Cecilia E.
1993 Grammar in interaction: Adverbial clauses in American English
conversations. Cambridge: Cambridge University Press.
Goldberg, Adele
1995 Constructions. A Construction Grammar approach to argument
structure. Chicago: University of Chicago Press.
Grice, H. P.
1978 Further notes on logic and conversation. In: Peter Cole (ed.), Syntax
and Semantics, Vol. 9: Pragmatics, 113-127. NY: Academic Press.
Haiman, John
1980 The iconicity of grammar: isomorphism and motivation. Language
56: 515-40.
1986 Constraints on form and meaning of the protasis. In: Traugott et al.
(eds.), 215-227.
Hom, Larry
1984 Toward a new taxonomy for pragmatic inference: Q-based and R-
based implicature. In: Deborah SchifIrin (ed.), Meaning, form and
use in context: linguistic applications (GURT 1984), 11-42.
Washington, D.C.: Georgetown University Press.
1985 Metalinguistic negation and pragmatic ambiguity. Language 61:

121-74.
Iatridou, Sabine
1991 If then, then what? NELS 22: 211-225.
1994 On the contribution of conditional then. Natural Language
Semantics 2: 171-199.
James, Deborah
1982 Past tense and the hypothetical: a crosslinguistic study. Studies in
Language 6: 375-403.
Konig, Ekkehard
1986 Conditionals, concessive conditionals, and concessives: Areas of
contrast, overlap and neutralization. In: Traugott et ale (eds.), 229-
46.
Lakoff, Robin
1971 Its, and's, and but's about conjunction. In: Charles 1. Fillmore and
D. Terence Langendoen (eds.), Studies in Linguistic Semantics, 114-
149. New York: Holt, Rinehart and Winston.
Langacker, Ronald W.
1987 Foundations ofcognitive grammar, vol. 1: Theoretical prerequisites.
Stanford CA: Stanford University Press.
1991a Foundations ofcognitive grammar, vol. 2: Descriptive applications.
Stanford CA: Stanford University Press.
1991b Concept, image, symbol: the cognitive basis of grammar. Berlin:
Mouton de Gruyter.
Lightbown, P.M. and N. Spada
1993 How languages are learned. Oxford: Oxford University Press.
Nikiforidou, Kiki
1990 Conditional and concessive clauses in Modern Greek: A syntactic
and semantic description. Ph.D. dissertation, University of California
at Berkeley.
Schiffrin, Deborah
1992 Anaphoric then: aspectual, textual, and epistemic meaning.
Linguistics 30: 753-792.
Schwenter, Scott
1997 The pragmatics of conditional marking: Implicature, scalarity, and
exclusivity. Ph.D. dissertation, Stanford University.
1998 From hypothetical to factual and beyond: refutational si- clauses in
Spanish conversation. In: Jean-Pierre Koenig (ed.), Discourse and
cognition: Bridging the gap, 423-436. Stanford CA: CSLI
Publications.
Sweetser, Eve
1982 Root and epistemic modals: causality in two worlds. BLS 8: 484-507.
Press.
1996 Conditionals and mental spaces. In: Fauconnier and Sweetser (eds.),
318-333.
1999 Compositionality and blending: semantic composition in a

cognitively realistic framework. In: Theo Janssen and Gisela
Redeker (eds.), Cognitive Linguistics: Foundations, scope and
methodology, 129-162. Berlin: Mouton de Gmyter.
Talmy, Leonard
1988 Force dynamics in language and cognition. Cognitive Science 2: 49-
100.
Traugott, Elizabeth Closs
1982 From propositional to textual and expressive meanings. In: Winfred
Lehmann and Yakov Malkiel (eds.), Perspectives on Historical
Linguistics, 245-271. Amsterdam: Benjamins.
Traugott, Elizabeth Closs, Alice ter Meulen, Judy Snitzer Reilly and Charles A.
Ferguson (eds)
1986 On conditionals. Cambridge: Cambridge University Press.
Van der Auwera, Johan
1986 Conditionals and speech acts. In: Traugott et al. (eds.), 197-214.
On affirmative and negative complex conditional
connectives*
Estrella Montolio
This paper describes the characteristic features of clauses introduced by

complex conditional connectives (CCCs). Specifically, I analyze the
grammatical (§2), semantic (§3, §5, and §6) and pragmatic (§7) properties
of these clauses which distinguish them from ij=-conditionals. CCCs are
organized into two subsets that stand in complementary distribution: (i)
affirmative CCCs and (ii) negative CCCs. These subsets show essentially
the same grammatical behavior but differ in terms of their semantics (§ 1
and §4) and their pragmatics (§7). The data are drawn principally from
Spanish, but the results of the analysis are, in general, equally applicable to
English and French.
1. Preliminaries
Though it is true that the literature on conditionals is quite extensive,

conditional structures other than !fclauses have traditionally attracted
little attention from linguists. This is most definitely the case with what
might be called complex conditional connectives (Dostie 1987; Leard
1987; Montolio 1991b, 1999; Visconti 1994), which have received
anything but extensive treatment. Perhaps this has to do with the general
impression that complex conditional connectives (henceforth, CCCs)
are not only equivalent to one another but are also completely parallel to
!fclauses. 1 The goal of this paper is to refute such an idea and
demonstrate how mistaken it actually is. Contrary to the view of
equivalence, I shall show that CCCs cannot appear in the majority of the
contexts in which it is possible to use !fclauses. As we shall see, this is
because CCCs express a much more specific and restricted relationship
between the protasis and the apodosis than does if. I concentrate mostly
on Spanish data, but I believe that, for the most part, my observations
are equally valid for both Romance and Germanic Languages. 2
The connectives that we shall include in the category of CCCs
correspond to two complementary subsets (cf Montolio 1990, 1991b,
144 Estrella Montolio
1999). The first group, which we shall call the affirmative group (or
ACCCs), is made up of constructions such as:
(1) a. a condici6n de que (= French: acondition que)

'provided that', 'on the condition that', 'so long as'
b. con tal de que (= French: pourvu que)
'provided that', 'so long as'
c. siempre que 3 (= French: pourvu que)
'provided that', 'so long as'
Expressions such as the following are included in the second, negative

group (or NCCCs).
(2) a. a menos que (= French: amoins que)

'unless'
b. a no ser que (= French: amoins que)
'unless'
c. salvo que (= French: saufsi)
'except if
d. excepto que (= French: saufsi)
'except if
Spanish conditional connectives excepto si ('except if) and salvo si

('except if) are semantically analogous to the other connectives
included in this group (2), but, as if-compounds, they can combine with
either the subjunctive or the indicative. Given the fact that the rest of the
complex connectives to be analyzed here all trigger the subjunctive
mood, 1 have preferred not to pursue their description any further here.
Because these two subsets stand in complementary distribution, any
member from the ACCC group can appear in the same contexts as all
the others of this group but in none of the contexts required by NCCCs.
The complementary distribution of these subsets is due to a difference in
semantic content, which explains the pragmatic inadequacy of these
connectives when an ACCC is replaced by an NCCC (or vice versa)
within the same context. This can be seen in examples (3b) and (4b)
below. (I use the #sign to indicate that the utterance under consideration
is grammatically correct but pragmatically inadequate).
Complex conditional connectives 145
(3) a. Ire con vosotros al cine a condici6n de que/con tal de

que/siempre que me acompaneis luego a casa en coche.
'I'll go with you guys to the movies provided that/so long as
you take me home afterwards by car. '
b. # Ire con vosotros al cine a no ser que/a menos que me
acompaneis luego a casa.
'# I'll go with you guys to the movies unless you take me
home afterwards by car. '
(4) a. No ire con vosotros al cine a no ser que/a menos que me

acompaneis luego a casa en coche.
'I won't go with you guys to the movies unless you take me
home afterwards by car. '
b. # No ire con vosotros al cine a condici6n de que/con tal de
que/siempre que me acompaneis luego a casa en coche.
'# I won't go with you guys to the movies provided that/so
long as you take me home afterwards by car. '
The utterances in question become acceptable again if their polarity

is changed (see example (4a». Hence, we can say that the
complementary distribution of these two subsets (ACCCs and NCCCs)
stems from their complementary polarities. Notice that it is not a matter
of whether utterances with NCCCs should necessarily be negative (as
shown by a simple example such as Manana ire a hacer jogging a no
ser que llueva 'Tomorrow I'll go jogging unless it rains'). Rather, the
point is that its potential commutation by an ACCC always entails a
change in polarity (Manana ire a hacer jogging a condicion de que no
llueva 'Tomorrow I'll go jogging provided that it doesn't rain').
As we shall see below, the reason for the strangeness of the (b)
sentences is not, strictly speaking, a matter of ungrammaticality, i.e., a
problem in the syntax of the constructions. Instead, it probably has more
to do with a pragmatic inadequacy between the informational content of
the two clauses and the processing instructions conveyed by the
connectives in question. Despite these distributional constraints, it is
easily shown that si ('if) and its corresponding negative counterpart si
no ('if ... not') can appear in the ACCC and NCCC contexts,
respectively.
(5) a. Ire con vosotros al cine si me acompafiais luego a casa en

coche.
'I'll go with you to the movies if you take me home afterwards
by car.'
b. No ire con vosotros al cine si no me acompafiais luego a casa
en coche.
'I won't go with you to the movies if you won't take me home
afterwards by car.'
The distribution of the data leads me to believe that si 'if is the

prototypical and superordinate member of the category of conditional
structures. The connective si is semantically unmarked as compared to
CCCs (as well as other possible conditional constructions), which
express a more specific and restrictive type of condition. The import of
this observation is that CCCs are less abstract and more specific in
meaning than if and its negative counterpart if ... not. That is to say, if
and if ... not constitute the superordinate terms of the relationship, in
which CCCs are the subordinate terms.
2. The order and the value of p in CCCs
Although research on clause order in conditional structures has not

specifically examined CCC constructions, the data they provide seem to
show that clauses introduced by CCCs, unlike if-clauses, usually occupy
the second position in conditional constructions (for English, see
Traugott, 1997: 160 and Dancygier, 1998: 173; for French, cfr. Dostie,
1987: 177-178; see Montolio 1991b: 134-135, for Spanish data).
Therefore, in a high percentage of the cases, CCCs invert the normal or
unmarked order of conditional expressions, which of course tends
prototypically to be [if p, q]. CCC-ciause order, then, has to be related to
the schema [q ifp].4
Ramsey (1985: 405) observes that while preposed protases tend to
have discourse functions, postposed protases "look more like a
parenthetical comment for the main clause". The corpus-based studies
by Ford and Thompson (Ford and Thompson 1986, Ford 1993, 1997)
have corroborated this observation, namely that when the protasis is
preposed it either conveys information that is in some sense already
known or creates the framework for the following discourse. However,
when !fclauses are postposed, they function more like afterthoughts,

evaluations, justifications.
Prosodic features also play a role. When the postposed clause is
preceded by a pause [q, ifp], on occasion at least, the !fclause seems to
convey the sense of less probability. The expectation that the contents in
p will not be fulfilled is often taken advantage of in advertising slogans:
(6) a. Si puede mirela. 0 cambie, si puede. (Catchphrase for the

Spanish TV series Serie Rosa 'Pink Series') (EI Peri6dico de
Catalunya, March 21, 1991).
'If you can, watch it. Or change [the channel], if you can.'
b. Resistete, si puedes. (Advertising slogan for the Renault 19)
(EI Pais, September 15, 1992).
'Resist [the temptation], if you can.'
In each case the effect of postposition in conjunction with the

preceding pause clearly conveys an inferential interpretation that the
fulfillment of the modal puede(s) 'can' is improbable. In fact, in the
television commercial corresponding to the same Renault advertising
campaign as that in (6b) even a priest succumbs to the temptation. The
interpretation that 'p will not be easily fulfilled' is used as sort of a
challenge, a strategy to incite the consumer's interest.
I would like to suggest that the CCC-schema conveys an analogous
sense of improbability or exceptionality. The characteristic semantics of
postposed protases headed by CCCs represents an a-posteriori
specification that rectifies 5 the content of the main clause, by stipulating
the conditions under which q will or will not be carried out. In these
constructions the hearer is instructed that p entails a certain type of
correction of what is affirmed in q. As pointed out by Dancygier (1985,
1998) for unless constructions, we can propose that the relation between
q and p in CCC structures is not one of cause (as is prototypically the
case of conditionals), but rather one of circumstance. That is, p
constitutes a specification of the circumstances under which q can be
maintained. This can be seen in the following examples:
(7) a. A mi me parece que las especulaciones son muy poco

pertinentes, a no ser que seamos capaces de imaginarnos a
nosotros mismos convertidos en un anciano de setenta y siete
afios [ ..}. (El Europeo, October 1988, pp. 59-60).
'It seems to me that the speculations are rather inappropriate,

unless we are able to imagine ourselves turned into a 77-year-
old elderly person [... ].
b. La respuesta a tan dificil pregunta es sf, es posible broncearse,
siempre que la exposicion solar se realice de manera
inteligente. (Marie-Claire, February, 1993).
'The answer to such a difficult question is yes, it is possible to
get a tan, provided that exposure to the sun is done in an
intelligent manner.'
The specification implied by p is so strong that, orally, it is often

introduced by a long pause or, graphically, ~y a comma or even a
period. In many cases, CCC-ciauses constitute phonologically and
graphically independent utterances. The schema [ q. CCC p] is quite
frequent in Spanish. It can be observed in the following example:
(8) Sf, sf, vamos. Estoy convencido de que en mi vida futura la

filosofia sera mi hobby. Ahora, no mi profesion, desde luego. A
no ser que cambie mucho, 0 me vea en una situacion... economica
muy ahogada y me tenga que dedicar a dar clases. (Esgueva and
Cantarero (1981: 24).
'Yes, yes, well.... I'm convinced that in my life to come
philosophy will be my hobby. But not my job, that's for sure.
Unless my life changes a lot, or I find myself in a very tight
financial. .. situation and I have to earn a living teaching. '
As opposed to prototypical If-clauses, which are preposed, the order

[q, CCC p] is not iconic with the sequence of events in the 'real world'.
Instead, the order appears to be iconic with the mental processing of the
instructions to be given to the hearer. That is, CCC-ciauses tend not be
iconic with event structure, but rather with argumentational patterns, in
that their cognitive processing seems to mirror the sequence used in
assertion-restriction or the specification thereof
3. The biconditional value of CCCs
The hypothesis put forth here is the following: whereas the construction
if p designates a very broad and generalized condition, roughly
paraphrasable as 'in the event that p,' CCCs express highly specific
conditions. As noted above, CCCs establish a much more precise and

restricted relationship between p and q than if does. It is true that, as it
has been seen, !fclauses can convey a similar "restricted" relation
betwen p and q in certain grammatical circumstances (for instance, in
cases in which the subordinate clause is postposed and preceded by a
pause). The difference lies in the fact that if only suggests such a strong
contingency in the appropiate context, whereas CCCs make the
"restricted" relation explicit, which is consistent with the greater
transparency of CCCs.
The relationship between p and q in a CCC-structure is so strong
that, as we shall see, it cannot be canceled. In fact, the exceptionality of
the condition expressed by CCCs is such that it cannot be later
neutralized without leading to a semantically and pragmatically
unacceptable utterance (cf. Dostie 1987: 190).
Ducrot (1973: 267) and Fauconnier (1984: 147) have both pointed
out that there exist mechanisms that nullify the biconditional reading of
!fclauses, which is the normal interpretation in ordinary language
(known as conditional perfection). Fauconnier proposes that de toute
far;on ('anyway') can be employed to nullify the 'invited' inference of a
conditional reading (ifp, q , if not p, not q) (Geis and Zwicky 1971).
The neutralizing character of de toute far;on is confirmed by
"l'apparition simultanee possible de mais ou cependant" (Fauconnier
1984: 147).
(9) a. Si tu viens,} 'irai, mais si tu ne viens pas,} 'irai de toute fa~on.

b. Si til vas, ire, pero si no vas, yo ire de todos modos. 6
c. Ifyou go, I'll go, but if you don't, I'll go anyway.
However, as can be seen in (10), such expressions cannot be used

with CCC-ciauses.
(10) a. *J'irai a condition que tu viennes, mais si tu ne viennes pas,

} 'irai de toute fa~on.
b. *Ire a condici6n de que vayas, pero si til no vas, ire de todos
modos.
c. *1'11 go provided that you go, but if you don't, I'll go anyway.
These examples corroborate the hypothesis that the condition

established by CCC-ciauses is in no way defeasible. This behavior is
similar to that of the connective if and only if. A condicion de que
'provided that' and a menos que/a no ser que 'unless' convey unique or
exceptional circumstances that will allow (or block) the event expressed
in the main clause to occur. This explains why such CCC-constructions
cannot be iterated in coordinate structures:
(11) a. *,Sa/dre contigo a cenar a condicion de que no hab/emos de

trabajo y a condici6n de que me invites.
'I'll go out to dinner with you so long as we don't speak about
work and so long as you pick up the tab. '
b. *Me ire de esta empresa a no ser que me asciendan y a no ser
que me aumenten e/ sue/do.
'I'll leave this company unless they promote me and unless
they give me a raise. '
Notice that both constructions are perfectly viable with si:
(12) a. Sa/dre contigo a cenar si no hablamos de trabajo y si me

invitas.
'I'll go out to dinner with you if we don't speak about work
and if you pick up the tab. '
b. Me ire de esta empresa si no me ascienden y si no me
aumentan e/ sue/do.
'I'll leave this company if they don't promote me and if they
don't give me a raise. '
The reason for the unacceptability of the iterated CCCs lies in the fact
that it is simply not possible to present the only circumstance under
which something will or will not happen and then add another
exceptional condition. In short, CCCs disallow the presence of other
exceptional conditions.
4. ACCCs and NCCCs
Both groups (ACCC and NCCC) express exceptional conditions. The

meaning of a condici6n de que 'provided that' and other ACCCs in
general is restrictive: they establish the only condition under which q
can be realized. That is, ACCCs have approximately the same value as
[q only ifp]. ACCCs are therefore formulations of a favorable exception
for something to happen. NCCCs, on the other hand, indicate the only
situation by which q will not be realized. 7 That is, NCCCs convey the
instruction [q only if not pl. Hence, NCCCs convey something to the
effect of "in any situation other than p." In other words, NCCCs express
the exceptional circumstance by which the proposition expressed by the
main clause will not be fulfilled. Both types of CCCs express tff
relations. Both types of CCCs present a modal and argumentative value,
since they instruct the hearer whether a given circumstance is favorable
or disfavorable to the proposition in the main clause.
The perspective that the speaker adopts in the face of the possibility
of q 's realization is therefore quite different: while ACCCs signal that q
can only be realized if p, NCCCs express that q will not be realized only
in the case of p. This is what determines the affirmative character of the
first group and the negative character of the second. ACCCs convey the
necessary basis for something to occur; NCCCs express the
circumstance under which something will not be realized. Dancygier
(1985: 68) proposes quite insightfully that what unless really denies is q,
not p. Her proposal seems to be clearer than the one advanced by Geis
(1973): except if p, q. To observe a case in which the main clause
displays negative polarity (for instance, example (4)), we can
extrapolate from Dancygier's observation as follows in (13b):
(13) a. q unlessp q; [(not q) ifp] (1985: 65)

(b. not q unlessp notq; [ (q) ifp])
The negative character of NCCCs is the reason why a no ser que

'unless' and like connectives appear to be generally the equivalent of si
no 'if ... not' -clauses. In fact, however, there are quite a few differences
between the· two types of constructions (cf Geis 1973; Dancygier 1985;
von Fintel 1991; Dancygier 1998: 169-171). Consider, for instance, the
odd combination of NCCCs such as unless with negative-polarity
clauses which are perfectly acceptable with if (Cf *Unless .John cares a
bit for Mary, he shouldn't marry her vs. If John doesn't care a bit for
Mary, he shouldn't marry her (Geis 1973:283)).
In my opinion, regardless of whether they are equivalent to if or if ...
not clauses, the fundamental value of NCCCs is to specify in p the
circumstances under which q will not be fulfilled. The point is that the
circumstances present in p are construed as improbable or contrary to
what one might expect. The contrast between the two following
examples adds credibility to the hypothesis advanced here (cf Smith
1983: 21 for similar examples):
(14) a. Los cocodrilos no podrian sobrevivir si no tuvieran dientes

afiladisimos.
'Crocodiles couldn't survive if they didn't have extremely
sharp teeth. '
b. # Los cocodrilos no podrian sobrevivir a menos que tuvieran
dientes afiladisimos.
'Crocodiles couldn't survive unless they had extremely sharp
teeth. '
(14a) sounds perfectly normal but (14b) is surprising from a pragmatic

point of view. In (14a), the negation of the counterfactual structure
presupposes that crocodiles actually do have extremely sharp teeth. On
the other hand, (14b) is decidedly odd because the clause introduced by
the NCCC presupposes that the condition in p is exceptional. That is,
the condition is contrary to what would normally be expected. Since
NCCCs convey an exceptional obstacle for q, in an eternal-truth context
such as (14b) they are obviously inappropriate: the clause suggests that
crocodiles do not have extremely sharp teeth, and this conflicts whith
our encyclopedic knowledge.
This nuance of less expectation conveyed by unless-constructions
explains why they are often inappropriate for counterfactuals. As is well
known, counterfactual clauses are presupposed to be false:
(15) Yo Ie habria dado ese medicamento a la nina, si no 10 hubiera

desaconsejado el medico explicitamente.
'I would have given the medicine to the girl, if the doctor hadn't
advised against it explicitly. '
The utterance in (15) conveys a factual interpretation in which the

content of p is negated. The doctor is understood as indeed having
advised against giving the girl the medicine. Clauses headed by a menos
que 'unless' and a no ser que 'unless' convey information which is
necessarily understood as hypothetical, precisely because they carry a
sense of high improbability-and improbability is necessarily
hypothetical. Thus, a version of (15) with a menos que or a no ser que
indicates that the hypotheticality (i.e., nonfactuality) of p is real, since in
reality we do not know if the doctor did or did not authorize the
medicine:
(16) Yo Ie habria dado ese medicamento a la nina, a menos que/a no

ser que 10 hubiera desaconsejado el medico explicitamente.
'I would have given the medicine to the girl, unless the doctor had
advised against it explicitly. '
Precisely because they can only receive a hypothetical interpretation

and are primarily future-oriented (see Traugott 1997: 158 for the case of
unless), a menos que/a no ser que-clauses are incompatible with
counterfactuals that presuppose the creation of an 'impossible world.'
This is also shown by the behavior of the following examples.
(17) a. Si Colon no hubiera descubierto America, los indios habrian

sido mas felices.
'If Columbus hadn't discovered America, the Indians would
have been happier. '
b. *A no ser que Colon hubiera descubierto America, los indios
habrian sido mas felices.
*' Unless Columbus had discovered America, the Indians
would have been happier. '
As we have seen, a no ser que presents q as well as the possibility of not

q. Nevertheless, counterfactuals only allow for one interpretation: q, if
the clause is in negative form, or not q, if there are no overt markers of
negation. Hence, (17b) is odd because we know that Columbus did
actually discover America.
5. CCCs and contextually given conditions
As we have just seen, the greater specificity of CCCs-in contrast to the

neutral, unmarked character of si 'if-makes them inappropriate with
counterfactual structures. This same reason, I contend, stands behind
their inability to appear in constructions in which the content of p is
contextually asserted. Take for example sentence (18):
(18) A: Ya he acabado el trabajo.

'I've just finished the job.'
a. B: Pues si ya has aeabado e/ trabajo, sa/gamos a eenar por

ahi.
'Well, if you've just finished the job, let's go out for
dinner. '
b. 'B: *Pues a condici6n de que ya hayas aeabado e/ trabajo,
sa/gamos a eenar por ahi.
*'Provided that/so long as you've finished the job, let's go
out to dinner. '
In other cases, the assertion hidden in the protasis is less explicit, but it
is information that is present in the context of the speech event (the term
context should be understood in the broad sense of Sperber and Wilson
19962) . The thematic or given nature of such rrotases is demonstrated
by the fact that the protasis cannot be postposed:
(19) a. Si hay una pasion franeesa que no //eva camino de

desapareeer, esa es /a pasion por /a po/itiea. [E/ Pais, June 21,
1992]
'If there is one French passion that is not on its way to
disappearing, it is their passion for politics. '
b. *Es /a pasion por /a po/itiea, si hay una pasion franeesa que
no //eva camino de desapareeer.
'It's is a passion for politics, if there is one French passion that
is not on its way to disappearing. '
c. *A condici6n de que haya una pasion franeesa que no //eva
camino de desapareeer, esa es /a pasion por /a po/itiea.
'Provided that there is one French passion that is not on its
way to disappearing, it is their passion for politics. '
(20) a. Si no te /0 ha dicho es porque te quiere.

'Ifhe hasn't told you it's because he loves you.'
b. *Es porque te quiere, si no te /0 ha dicho.
'It's because he loves you, ifhe hasn't told you.'
c. *A no ser que te /0 haya dicho es porque te quiere.
*' Unless he's told you, it's because he loves you.'
Elsewhere (Montolio 1999) I have called such conditional structures
identificatives (19) and exp/ieatives (20), respectively. Unlike
prototypical (i.e., hypothetical) conditionals, the If-clauses here are far
from left-open (in the sense of Lehman 1974: 234; that is, the
proposition in the antecedent of a conditional sentence is not asserted).

On the contrary, these uses of si-clauses seem to be what I propose to
call left-given. This is shown by the fact that these clauses cannot be
postposed, as is demonstrated by the (b) examples, nor can they be used
with CCC-structures, as in the (c) examples.
In Spanish, as well as in French and English, the apodoses of if-
constructions can be-indeed frequently are-yes/no-questions «21a))
as well as wh-questions «22a) and (23a)).
(21) a. Sifuese obispo, ~condenaria el aborto?

'If you were a bishop, would you condemn abortion?'
b. *A condici6n de quefuese obispo, ~condenaria el aborto?
*'Provided that you were a bishop, would you condemn
abortion?'
(22) a. Si a Gonzalez Ie gusta tanto participar en la campana de un

referendum, ,",por que no organiza uno en su pais? dice
Philippe Seguin. [El Pais, June 21, 1992]
'If Gonzalez likes participating in the campaign for a
referendum so much, why doesn't he organize one in his
country? says Philippe Seguin. '
b. *A condici6n de que a Gonzalez Ie guste tanto participar en la
campana de un referendum, ,",por que no organiza uno en su
pais? dice Philippe Seguin.
*' Provided that Gonzalez likes participating in the campaign
for a referendum so much, why doesn't he organize one in his
country? says Philippe Seguin. '
(23) a. Si no estas preparada para tener un hijo, ,",por que demonios

has querido quedarte embarazada.
'If you're not ready to have a child, why the hell did you want
to get pregnant?'
b. *A menos que estes preparada para tener un hijo, ,",por que
demonios has querido quedarte embarazada.
*' Unless you're ready to have a child, why the hell did you
want to get pregnant?'
In view of examples such as these, it seems feasible to state that ACCCs

are difficult to combine with yes/no-questions, as is the case with
example (21b). Such a combination gives rise to utterances that are
clearly very strange. Furthermore, CCCs are, in general, difficult to

interpret with wh-interrogatives (e.g., (22b) and (23b». Actually, this
difference in behavior with regard to the two different basic types of
interrogatives does not come as much of a surprise. This is so because in
wh-interrogatives the tt:protasis functions as a framework for a latent
assertion (e.g.,(22a) and (23», whose superficial hypothetical form lays
the groundwork for and justifies the question that follows it. That is, the
use of a hypothetical form for contextually asserted information
constitutes a mechanism that attempts to safeguard the relevance of the
subsequent question (cf Haegeman 1984).
CCCs are also incompatible with so-called indicative
counterfactuals, since they also presuppose some kind of previous
assertion:
(24) a. Mira, si esa chica es guapa, yo soy Claudia Schiffer.

'Look, if that girl is pretty, I'm Claudia Schiffer.'
b. *A condici6n de que esa chica sea guapa, yo soy Claudia
Schiffer.
*'Provided that that girl is pretty, I'm Claudia Schiffer.'
(25) a. Si tit no eres rico, yo entonces soy un mendigo.
'Ifyou're not rich, then I'm a beggar. '
b. *A no ser que tit seas rico, yo soy un mendigo.
*' Unless you're rich, I'm a beggar.'
All of these kinds of conditionals fall under the heading of

nonpredictive conditionals (according to Sweetser 1990; Dancygier
1993, 1999). Note that, to a greater or lesser extent, they all constitute
"echoes" from some previous discourse. Furthermore, they can be
characterized by their lack of strict iconicity. The sequence of clauses is
not iconic with the sequence of events, though in fact there does seem to
be an iconic correspondence between the linear order of the clauses and
the sequence of steps taken in argumentation: Given x: X allows me to
assert ... (cft. Sweetser 1990: 128). That is, speakers use p to show the
hearer how they arrive at the assumptions in q. As we can see, CCC-
structures cannot appear in any context of epistemic conditionals, where
the tt:clause introduces some type of assumption present in the context
of the speech situation (Sweetser 1990: 128). That is, CCC-ciauses
cannot appear in any context whose content is given, assumed, or
clearly presupposed. The information that CCC-ciauses convey is
therefore necessarily rhematic. This is consistent with post-main-clause

order, because focus tends cross-linguistically to occur in sentence-final
position (Comrie 1986: 86).
The insertion of a CCC, especially NCCC, seems slightly more
acceptable--especially when they appear preposed-with speech-act
and metalinguistic conditionals than with any of the former types.
Nevertheless, such uses do not, by any means, constitute prototypical
contexts for CCCs. Let us consider Austin's celebrated example:
(26) a. There are biscuits on the sideboard ifyou want them.

b. #There are biscuits on the sideboard provided that/so long as
you want them.
Clearly there is something peculiar about (26b). Consider the examples

in (27):
(27) a. En espanol americano a eso Ie llaman 'durazno', SI no

confundo la palabra.
'In Latin-American Spanish that's what they call a 'durazno'
(peach), if I'm not confusing the word.'
b. ?En espanol americana a eso Ie Ilaman 'durazno', a no ser que
confunda la palabra. 9
'In Latin-American Spanish that's what they call a 'durazno'
(peach), unless I'm confusing the word.'
The ability of NNNC clauses to be inserted in these nonpredictive,

nonprototypical schemas can be explained by the parenthetical nature
that both !fclauses and CCC-ciauses acquire in these types of
conditionals: the independence of the propositional content of CCC-
clauses with regard to the main clause also corresponds to ·the
nonembedded character of the subordinate clauses in speech-act and
metalinguistic conditionals.
6. CCCs and the discourse deictic then
Another feature that characterizes CCC-constructions as noncentral

members of the family of conditional structures is their inability to
combine with the phoric then: 10
(28) a. Si hace sol, entonces iremos a la playa.

'/fit's sunny, then we'll go to the beach.'
b. *A condici6n de que/con tal de que/ siempre que haga sol,
entonces iremos a la playa.
'Provided that/so long as it's sunny, then we'll go to the
beach.'
(29) a. Si no llueve, entonces iremos a la playa.

'/fit doesn't rain, then we'll go to the beach.'
*A no ser que/ a menos que/ salvo que/ excepto que llueva,
entonces iremos a la playa.
*' Unless/except ifit rains, then we'll go to the beach.'
The presence of entonces ('then') in conditional schemas is a matter that
remains to be thoroughly studied in Spanish. Nonetheless, as pointed
out by Haiman (1978: 576), in contradistinction to the traditionally
studied schema if p, then q, the phoric then does not constitute by any
means an obligatory element in ordinary-language conditionals. In fact,
certain types, such as speech-act conditionals or metalinguistic
conditionals, reject it.
(30) a. ?? Si no me equivoco, entonces ese es tu primo.

??'/fI' m not mistaken, then that guy is your cousin. '
b. ?? Si se Ie puede llamar asi, entonces es una estructura
condicional.
??' /f you can call it that, then it's a conditional structure. '
Of course, speech-act and metalinguistic conditionals, by their very

nature, do not express a grammatical relationship where p entails q. By
contrast, entonces 'then' explicitly expresses that q follows from p.
Therefore, the semantics of these constructions clashes with that of then
and thus produces an unacceptable utterance.
In those constructions that do accept its presence, the insertion of
then is neither pragmatically, nor semantically identical to its omission:
it is one thing to imply that q follows from p, as in prototypical
conditionals using the schema [if p, q], and it is another to mark this
explicitly, as in [ifp, then q]. In any event, it should be noted that those
if-constructions that do allow the presence of then do not allow it when
the normal order of the clauses is altered, which happens to be the
schema normally used by CCCs:
(31) *Entonces ire a la playa, si llueve.

'Then I'll go to the beach, ifit rains.'
Nevertheless, the examples in (28) and (29) show that CCCs are
incompatible with entonces 'then' even when they appear in a preposed
position. Such a fact invites the question of just what the function of
then in a conditional is. There seems to be agreement that one of its
main functions is to act as an anaphoric emphatic correlate to the
protasis, i.e., something like a resumptive pronoun (Dancygier 1998;
Dancygier and Sweetser 1997; latridou 1992; Montolio 1991b, 1999).
One might say that the use of then constitutes an attempt on the part of
the speaker to insure that the interlocutor retains the hypothetical
framework on which the following clause or utterance is based. This is
clearly the case of conditionals in which various protases are linked to
one apodosis:
(32) Todos creemos tener el mismo derecho: si uno se puede hacer

millonario en dos dias, si no es el resultado de una vida de
esfuerzo, si se consigue con solo jugar a la Bolsa, con disponer de
informacion privilegiada, entonces todo el mundo cree que
tiene derecho a ganar mas de 10 que gana. [EI Pais, May 17,
1998]
'We all believe that we have the same right: if one can become a
millionaire in two days, if it's not the result of a life of hard work,
if it can be obtained by playing the stock market, by having
privileged information, then everybody believes that they have the
right to earn more than they do.'
As opposed to prototypical ~clauses, CCC-ciauses do not function

as a hypothetical framework for the assertion of q. Rather, as we have
seen, they restrict or specify what is expressed in q. They are not,
however, the basis for asserting the content of the main clause. These
facts account for why CCCs cannot collocate with then, since the
cohesive function of this element is to serve as an elliptical framework
for the apodosis. However, when the CCC-structure is included in a
larger framework, entonces 'then' is indeed acceptable.
(33) A: Papa, iftle Ilevaras al cine?

'Dad, will you take me to the movies?'
B: J.~i,
a condici6n de que acabes los deberes. Entonces te llevare
al cine.
'Provided that/so long as you finish your homework. Then I'll
take you to the movies. '
As can be seen, in (33) the insertion of entonces 'then' sounds perfectly

normal because its presence is not linked solely to the information in the
CCC-ciause. If this were the case, it would have given rise to an
unacceptable utterance:
(34) *A condici6n de que acabes los deberes, entonces te llevare al

cine.
*'Provided that/so long as you fmish your homework, then I'll
take you to the movies. '
Instead, in (33) B's answer, entonces ('then') is linked to a larger

discourse context, to the whole previous sentence ("yes, I will take you
to the movies, provided that you fmish your homework. Then --'if this
is the case' --, I'll take you").
7. The pragmatics of CCCs
In various papers, Fillenbaum has analyzed the relationship that exists

between a series of conditional constructions and certain speech acts
(see, specially, Fillenbaum 1986). He points out that the semantic
content of unless makes it appropriate to appear in deterrents, while at
the same time making it unnatural-sounding in inducements. This
semantico-pragmatic relationship between unless-constructions and the
illocutionary value of threats!! and prohibitions has been documented
diachronically for English (Traugott 1997) as well as for Spanish. Bartol
Hernandez (1992: 89) briefly comments that, historically, in the
majority of the cases, they are prohibitions (or negative orders,
expressed with can, should, or the subjunctive) whose compliance
carries an exception, a restriction, presented as a necessary condition.
(35) is an example given by this author:
(35) E si por aventura el Rey fuere de tan gran piedat que 10 quiere
dexar vivir, non 10 pueda facer a menos que no Ie saque los ojos.
(Fuero Real, 350)
'And if by chance the King were so kindhearted as to want to let

him live, he cannot do so unless his eyes are tom out. (Regional
Code, 350)'
Fillenbaum has shown that in a high percentage of cases provided that

appears in promises or inducements (e.g., Fillenbaum 1976: 148). The
fact is that it is very difficult to invert these roles; that is to say, the
formulation of a promise with unless or that of a threat with provided
that is highly infrequent (being generally inadequate). This is true even
if we change the polarity of the clauses, in accordance with the
complementary distribution of the CCCs that we have analyzed. As can
be seen, (36b) is no longer a deterrent (as (36a» but a promise not to do
something undesirable. In contrast, (36d) seems a threat more than an
inducement (in contradistinction to (36c»:
(36) a. Te dejare todo el verano sin vacaciones, a no ser que mejores

estas notas.
'I'll take away your vacation, unless you improve these
grades. '
b. # No te dejare todo el verano sin vacaciones, a condici6n de
que mejores estas notas.
#'1 won't take away your vacation, provided that you improve
these grades. '
c. Iremos de viaje a Disneylandia, a condici6n de que mejores
estas notas.
'We'll go to Disneyland, provided that you improve these
grades.'
d. # No iremos de viaje a Disneylandia, a no ser que mejores
estas notas.
#'We won't go to Disneyland, unless you improve these
grades.'
The favoring character of positive stimuli presented by provided that

is corroborated by the fact that it can paraphrase-with the right
modifications in the clauses-any paratactic conditional construction
with the conjunction and, but only in the event that it expresses an
inducement stimulus, such as:
(37) a. Dame un besito y te dare un caramelo.

'Give me a little kiss and I'll give you a piece of candy.'
b. Te dare un earamelo, a condici6n de que me des un besito.

'I'll give you a piece of candy, provided that you give me a
kiss. '
However, when these coordinate structures are the equivalent of a threat

the paraphrase is impossible. Compare the following sentences:
(38) a. Toea a mi hermano y te parto la eara.

'Touch my brother and I'll bash your face in. '
b. #Te parto la eara, a condici6n de que toques a mi hermano.
#'1'11 bash in your face, provided that you touch my brother. '
The illocutionary specialization of these connectives is derived from

their semantics. The lessening of expectation or high improbability
conveyed by a menos que ('unless') conveniently explains its
inadequacy as a stimulus: it does not seem pragmatically profitable to
try to act upon one's interlocutor under the pretext that something
probably WILL NOT happen.
This same reason explains the adequacy of unless for prohibitions
and threats. The coercive power of a threat lies in the fact that the
negative event for the hearer is presented as a general assertion, which
only has a distant possibility of being revised: X, unless Y. The hearer
must therefore either comply with the proposition in the unless-clause or
suffer the consequences. 12
To exemplify this, we can recreate a cliche-ridden dialogue from the
film-noir genre between a gangster type and "the other guy's girl." In
the improbable event that the gangster were to use a CCC-construction
instead of the more predictable paratactic structure, he could choose
between formulating his threat with provided that or unless.
Nevertheless, one of them would seem much more persuasive than the
other:
(39) a. No matare a tu novio, a condici6n de que te fugues eonmigo.

'I won't kill your boyfriend, provided that you run away with
me.'
b. Matare a tu novio, a no ser que te fugues eonmigo.
'I'll kill your boyfriend, unless you run away with me.'
The greater force of (39b) lies precisely in that the threatening action
(i.e., kill the boyfriend) is expressed in an affirmative factual manner.
This seems to be more effective than posing the matter in the converse
terms, as in (39a), where the threatening action is negated and the
restriction on this negated threat appears later.
The same pragmatic explanation can be applied to resolve the greater
or lesser inadequacy of utterances (36b) and (36d). As can be seen,
exchanging connectives causes the speech acts to be exchanged too.
Thus, (3 6b) seems to be a much weaker version of a threat than that
formulated in (36a), though it also could be interpreted as an
unconvincing stimulus. Insofar as the clear stimulating promise in (36c)
is concerned, it is converted directly into a threat as soon as its
connective is changed (36d).
Besides the illocutionary value generally equated with CCC-
structures, we can also pose the question of what their pragmatic
behavior is as compared to that of if-clauses. Also worth considering is
the question of whether there exist differences between the two subsets
of CCCs from such a point of view. An if-clause can appear in any
context that a CCC can, and in fact, speakers prefer to use the unmarked
connector if even when the insertion of a CCC is possible. The very
subtle differences that these constructions convey with regard to if (and
if ... not) are so faint that most speakers probably do not consider them
to be very operative. Furthermore, in the Romance languages, CCC-
clauses trigger the subjunctive mood, which according to the literature
on the matter is in the decline. This seems to be why CCC-ciauses are
more characteristic of formal, written registers than of spontaneous, oral
ones.
CCC-structures are, in effect, noticeably less frequent than if-clauses
in oral and spontaneous registers. In this sense, we can see the
correspondence between ontogenesis and linguistic universals, because
such constructions are acquired much later than if-structures. Although
most authorities on the matter of the acquisition of conditionals only
deal with if-clauses, Wing and Kofsky (1981) point out that just as
common sense would lead us to believe, CCC-structures are among the
latest to be acquired. Given their formal complexity and the specificity
of the condition they introduce, CCC-structures surface in child
language at a later stage than if-clauses and other types of synthetic
conditional structures, such as paratactic structures (McCabe et al. 1983,
Wing and Kofsky 1981). In the case of NCCCs, this late appearance
seems to be even more justified. NCCCs imply a greater cognitive-
processing complexity in that to the uncertainty of the truth value they
add a negative relationship between the two clauses.
My claim is that no matter what the speech act may be, ACCCs and
NCCCs are pragmatically very different from one another. When
speakers choose to use an unless-clause, they are often trying to avoid
committing themselves completely: q is said first, and afterwards, as a
kind of afterthought, the only situation in which their utterance will not
be true is signalled (P). This claim is consistent with the historical
corpus-based analysis of Traugott (1997), who concludes that in general
unless-clauses express 'possible events that will block promises, threats,
expectations, and even probable recurrence. '
(40) Creo que [el condicionamiento social y economico] impiden que

en la practica se pueda decir: yo me caso cuando quiera. A no ser
que la ... no se, tambien la situacion economica familiar del novio
y de la novia, su plan, claro (Esgueva and Cantarero 1981: 34).
'I think that [social and economic conditioning factors] keep one
from saying in practice: I'm going to get married whenever I
want. Unless the ... I don't know, also the boy's and the girl's
family financial situation, their plans, of course. '
Thus, in most cases, the use of unless can be interpreted as aface-saving

repair (cf Goffman 1967). In such cases the speaker tries to take
precautions against an interpretation of his utterance in terms of overall
truth, which could jeopardize the speakers "face" in the future. In this
sense, the use of unless is much like metadiscursive if=-clauses of the
type if I'm not mistaken, if that's the right word, if I remember
correctly, etc., and has to do with the strategy of guarding oneself from
possible criticisms from one's interlocutor.
On the other hand, ACCCs are more problematic from an interactive
perspective: once established, provided that cannot be mitigated. Such a
specific meaning explains-at least in part-its low frequency: speakers
prefer to propose a condition with an if=-clause, rather than impose a
condition with provided that. In fact, according to the data I have
worked with, speakers often use an ACCC-ciause when they are talking
about a third party. In effect, ACCCs can be too imposing, and therefore
are used typically when there exists a certain asymmetry ("power") in
the social status of the interlocutors. Thus, a tenured professor can easily
indicate to his graduate student:
(41) De acuerdo, Ie dirigire la tesis, a condici6n de que trabaje sobre

el subjuntivo.
'Agreed, I'll supervise your thesis, provided that you work on the
subjunctive. '
On the other hand, this utterance would not be very likely the other
way around; that is, it would be precarious if the graduate student
declared something like the following to his superior:
(42) Muy bien, dare sus clases a condici6n de que usted me ayude a
preparar la bibliografia.
'Okay, I'll teach your classes provided that you help me with the
bibliography. '
Such an utterance, without a doubt, would seem less risky formulated as

an ~clause:
(43) Muy bien, dare sus clases si usted me ayuda a preparar la

bibliografia.
'Okay, I'll teach your classes if you help me with the
bibliography. '
The rule of linguistic politeness which indicates that indirectness is

more courteous than direct expressions also explains why the ACCC-
structures can seem not very polite: with these structures, there is no
interpretation other than the imposition of a necessary condition. Such
constructions are learned; furthermore, they can also seem impolite if
they are used in direct address.
10. Summary
To summarize, I shall try to list the principal traits of CCC-structures

and place them in relation to other structures belonging to the family of
conditional structures:
(i) Non-predictive conditionals are non-prototypical.

a. CCCs cannot appear in non-predictive contexts (i.e., where, in
most cases, [ifp] equals [givenp]);
b. CCCs are related to predictive and future-oriented contexts.
(ii) Hypotheticals and counterfactuals are non-prototypical

(Dancygier 1993: 428).
a. CCCs rarely appear in counterfactual contexts;
b. CCCs are basically related to central, predictive conditionals
(i. e., prototypical conditionals).
(iii) In prototypical, predictive conditionals, CCCs are non-central

because:
a. whereas, in prototypical If-conditionals, clauses are related
through causal relationships, in CCC structures, clauses are
related through circumstantial relationships;
b. whereas the prototypical order of central If-conditionals if p, q
makes them iconic with the sequence of events, the
prototypical order of CCC-conditionals q, CCC P makes them
iconic with the order involved in processing assertion-
restrictions;
c. whereas the function of If-clauses is, prototypically, to
establish a state of affairs to be taken as the basis for
considering the consequent, the function of CCC-ciauses is to
restrict or specify the contents of the apodosis (hence, their
incompatibility with then).
Notes
* I would like to thank Maria Rosa Vila and Joseph Hilferty for the many hours
that they spent with me discussing many of the points expanded upon here, as
well as the editors of this volume, Bernd Kortmann and Elizabeth Couper-
Kuhlen, for their exhaustive observations on this chapter. I would also like to
thank lbon Sarasola for providing me with the Basque example in endnote 4. I
would further like to mention the help received from Joseph Hilferty in
making an intelligible English version. It goes without saying, however, that
all errors are of my own responsibility.
1. The lack of quality grammatical descriptions regarding CCCs is reflected in
the fact that most Spanish textbooks and reference grammars for foreign
learners give a rather jumbled treatment of these constructions. Furthermore,
they tend to present them as if they were mutually commutable with one
another as well as with ij=-clauses in general. This has the unfortunate side
effect, as I have seen in my own classes, of prompting foreign students to
produce unacceptable utterances.
2. It may be possible that the conclusions reached here are also largely
applicable to languages such as Italian and German. Given my unfamiliarity
with these languages, however, I leave it to others to confirm or disprove this
possibility. It should also be noted, however, that in Spanish--as in other

Romance languages--clauses containing CCCs are generally constructed
with the subjunctive mood. The inflectional contrast between the present and
the imperfect subjunctives allows speakers of the Spanish language to
differentiate the values that conditional constructions with if express through
the indicative/subjunctive opposition.
3. The Spanish ACCC siempre que (literally: 'always that') maintains a
temporal meaning in many of the utterances in which it appears. An
interesting question is which formal elements must be involved so as to
obtain a conditional reading; in this sense, it seems that the presence or
absence of a pause between the relevant clauses is decisive. Conditional
readings tend to be marked by a pause, whereas a lack of a pause seems to
signal a temporal interpretation. Compare the following examples:
(i) a. Nos traera un regalo siempre que vue/va de viaje. (temporal)
'She'll bring us a present every time she returns from a trip.'
b. Nos traera un regalo, siempre que vuelva de viaje. (conditional).
'She'll bring us a present, provided that she returns from a trip.'
4. The relation between CCC-ciauses and postposed ~clauses appears to be
corroborated by the fact that languages such as Basque, which do not possess
specific unitary connectives to convey such conditional nuances, often
express them by postposing the ~clause (as in the Basque ba):
(i) Hondartzarajoango ara eguzkia egiten
Beach ABL go FUT AUX(lPL/PRES/ITR) sun-ART
badu bederen
make-IMP COND-AUX(CONJ/TR/3SG) at least
'We'll go to the beach if it's at least sunny out.'
5. Something similar is suggested by Leard (1987: 164) upon isolating a
particular group of conditional expressions which he calls restrictives. Dostie
(1987: 191-192) also associates this rectifying mechanism with CCCs, though
she only considers those structures that I have included in the ACCC group.
6. Of course, such utterances would sound more natural if an optative
concessive conditional were used:
(i) Vayas 0 no vayas,
go-PRES-SUBJUNCT-you or NEG go-PRES-SUBJUNCT-you
yo ire.
I gO-FUT
'I'm going whether you go or not.'
7. NCCCs are, in a manner of speaking, related to concessive-conditional
connectives. This applies, specifically, to incluso si ('even if), since
concessive-conditionals express the idea of 'contrary to expectation'. On the
other hand, a no ser que ('unless') and incluso si ('even if) have, in a certain
sense, complementary meanings: a no ser que conveys the only circumstance
that will be an impediment for the fulfillment of q; in contrast, incluso si
entails that p will never be an obstacle for q to be realized.
8. In Spanish there is another type of conditional structure labeled a contrastive
conditional, which is characterized by the thematic nature of the protasis. It
does not appear to have a clear counterpart in English:
(i) Si Chatwin eligi6 vlaJar caminando, Paul Theroux redescubri6 las

glorias y zozobras del ferrocarril (El Europeo, May 1993).
*lfChatwin chose to travel by foot, Paul Theroux rediscovered the glory
and angst of the train.
However, just as in the case of other conditionals whose protasis expresses a
contextually given condition, this conditional cannot appear with a CCC, as is
demonstrated by the grammatical behavior of the sentences in (ii).
(ii) *A condici6n de que Chatwin eligiera viajar caminando, Paul Theroux
redescubrio las glorias y zozobras del ferro carri I.
*Provided that Chatwin chose to travel by foot, Paul Theroux
rediscovered the glory and angst of the train.
For more information on Spanish contrastive conditionals, see Montolio
(1999).
9. In fact, speech-act conditionals are a habitual context for a natural unless-
clause anteposition.Thus, for instance, the example at hand would seem more
plausible with a preposed CCC: A no ser que confunda la palabra, en espanol
americana a eso Ie llaman udurazno" 'Unless I'm confusing the word, Latin-
American Spanish they call this [i.e., a peach] "durazno'"
10. In Spanish there exist two particles that correspond to this value of then:
entonces (which, like then, coincidentally also has a temporal meaning) and
pues (which historically meant 'after'). In fact, the connective pues
'thus'seems actually to be much more frequent than entonces 'then' in spoken
Spanish conditionals and seems to correspond to a marker of sequentiality
(Dancygier, 1998: 178 et passim), that is, a particle that introduces
conclusions. This matter still needs to be examined further, but a first
approximation can be found in Montolio (1991a, 1999).
11. It is worthwhile pointing out that in a conversational context, especially
where direct reference is being made to the interlocutor, conditional threats in
Spanish frequently take the form of a special construction: como + VPRES-
SUBJUNCT. This construction seems to be quite similar to the Japanese -tewa
(Akatsuka, 1997) and the disjunctive paratactic conditional, as in: Cena 0 no
veras la television ('Eat your dinner or you can't watch TV'). For more on
such threatening conditionals in Spanish, see Montolio (1999).
12. To this effect, Brown and Levinson distinguish concrete types of acts that
threaten negative face and that are formulated in the following way: Threats,
warnings, dares (S[peaker] indicates that he (or someone or something) will
instigate sanctions against H[earer] unless he does A) (Brown and Levinson
1987: 65-66; the italics are mine).
References
Akatsuka, Noriko
1997 Negative conditionality, subjectification, and conditional reasoning.
In: Angeliki Athanasiadou and Rene Dirven (eds.), On Conditionals
Again, 323-354. Amsterdam: Benjamins.
Austin, John L.
1956 1ft and cans. Proceedings ofthe British Academy: 109-132.
Bartol Hernandez, Jose Antonio
1992 Algunas conjunciones y locuciones condicionales en la Edad Media.
In: Jose Antonio Bartol Hernandez, Juan Felipe Garcia and Javier de
Santiago (eds.), Estudios jilol6gicos en homenaje a Bustos Tovar,
vol. 1,85-110. Salamanca: Ediciones Universidad de Salamanca.
Brown, Penelope and Stephen C. Levinson
1987 Politeness. Cambridge: Cambridge University Press.
Comrie, Bernard
1986 Conditionals: A typology. In: Elizabeth Traugott, Alice Ter Meulen,
Judy Snitzer and Charles Ferguson (eds.), On conditionals, 77-99.
Contreras, Lidia
1963 Las oraciones condicionales. Boletin de Filologia de la Universidad
de Chile XII: 273-290.
Dancygier, Barbara
1985 If, unless, and their Polish equivalents. Papers in Contrastive
Linguistics XX: 65-72.
1990 Conditionals: Sequence of events and sequence of clauses. In: Jaeck
Fisiak (ed.), Further insights into contrastive analysis, 357-373.
1993 Interpreting conditionals: Time, knowledge, and causation. Journal
ofPragmatics 19: 403-434.
1998 Conditionals and prediction. Time, knowledge and causation in
conditionals structures. Cambridge: Cambridge University Press.
1997 Then in conditional constructions. Cognitive Linguistics 8(2): 1-28.
Dostie, Gaetane
1987 Etude semantique de quatre connecteurs conditionnels: a condition
que, pourvu que, en autant que et d'abord que. Le jran9ais moderne
LV/3-4: 174-203.
Ducrot, Oswald
1973 La preuve et Ie dire. Paris: Marne.
Esgueva, Manuel and Margarita Cantarero (eds.)
1981 El habla culta de la ciudad de Madrid. Madrid: Consejo Superior de
Investigaciones Cientificas.
Fauconnier, Gilles
1984 Espaces mentaux. Paris: Minuit.
Fillenbaum, Samuel
1986 The use of conditionals in inducements and deterrents. In: Elizabeth
Traugott, Alice Ter Meulen, Judy Snitzer and Charles Ferguson
(eds.), On conditionals, 179-195. Cambridge: Cambridge University
Press.
von Fintel, Kai
1991 Exceptive conditionals: the meaning of unless. North Eastern
Linguistics Society 22: 135-148.
Ford, Cecilia E. and Sandra A. Thompson

1986 Conditionals in discourse: A text-based study from English. In:
Elizabeth Traugott, Alice Ter Meulen, Judy Snitzer and Charles
Ferguson (eds.), On conditionals, 353- 372. Cambridge: Cambridge
University Press.
Ford, Cecilia E
1993 Grammar in interaction. Adverbial clauses in American English
Geis, Michael and Arnold M. Zwicky
1971 On invited inferences. Linguistic Inquiry 2: 61-66.
Geis, Michael
1973 If and unless. In: Braj B. Kachru, Robert B. Lees, Yakov Malkiel,
Angelina Pientrangeli and Sol Saporta (eds.), Issues in Linguistics.
Papers in honor of Henry and Renee Kahane, 231-253. Urbana:
University of Illinois Press.
Goffman, Erving
1967 Interaction ritual: Essays on face to face behavior. New York:
Garden City.
Haegeman, Liliane
1984 Pragmatic conditionals in English. Folia Linguistica XVIII/3-4: 485-
502.
Iatridou, Sabine
1991 If Then, then what? North Eastern Linguistics Society 22: 211-225.
Leard, Jean-Marcel
1987 La syntaxe et la classification des conditionnelles et des concessives.
Le Fran~ais Moderne LV (3/4): 158-173.
Lehmann, Christian
1974 Prinzipien flir "Universal 14". In: Hansjakob Seiler (ed.), Linguistic
Workshop 11, 69-97. Munich: Wilhelm Fink Verlag.
McCabe, Anne E., Susan Evely, Rona Abmmovitch, Carl M. Corter and Debra 1.
Pepier
1983 Conditional statements in young children's spontaneous speech.
Joumal ofChild Language 10: 253- 258.
Montolio, Estrella
1990 Expresi6n de la condicionalidad en espano!. Ph.D. dissertation.
Barcelona: Universitat de Barcelona.
1991a Asi, pues entonces, 10 mejor sera que pienses bien 10 de casarte.
Acerca de los procondicionantes en espana!. Foro Hispanico 2: 43-
53.
1991b On conditional structures with complex connectors. In: Johan van
der Auwera (ed.), Eurotyp working papers: Adverbialia, volume 1:
33-52.
1999 Capitulo 57: Las construcciones condicionales. In: Ignacio Bosque
and Violeta Demonte (eds.), Gramatica descriptiva de la lengua
espanola, Volume III: 3643-3737. Madrid: Espasa Calpe.
Ramsey, Violeta
1987 The functional distribution of preposed and postposed if- and when-
clauses in written discourse. In: Russell S. Tomlin (ed.), Coherence
and Grounding in Discourse, 383-408. Amsterdam: Benjamins.
Smith, Neil V.
1983 On interpreting conditionals. Australian Journal ofLinguistics 3: 1-
24.
Sperber, Dan and Deidre Wilson
19962 Relevance. Communication and cognition. Oxford: Blackwell.
Sweetser, Eve E.
1990 From etymology to pragmatics: Metaphorical and cultural aspects
1997 Unless and but conditionals: a historical perspective. In: Rene
Dirven and Angeliki Athanasiadou (eds.), On conditionals again,
145-167. Cambridge. Cambridge University Press.
Visconti, Jacqueline
1994 Entre hypotheses et conditions. Sur la semantique des connecteurs
conditionnels complexes. Cahiers de linguistique fran~aise 15: 77-
108.
Wing, C.S. and Ellin Kofsky
1981 Children's comprehension of pragmatic concepts expressed in
because, although, if, and unless. Journal ofChild Language 8: 347-
365.
Pre- and post-positioning of wenn-clauses
in spoken and written German *
Peter Auer
This paper is concerned with the distinction between pre- and postposi-
tioned (initial and final) wenn-clauses in German, and with the distinction
between written and spoken language. A simple cross-tabulation of the two
features [spoken/written] and [pre-/postpositioned] shows that initial wenn-
clauses are preferred in spoken German, but final wenn-clauses are pre-
ferred in written German. An attempt will be made to explain these find-
ings.
1. Introduction
The fmdings and discussions in this paper are corpus-based. They are
partly quantitative, partly qualitative. With respect to both dimensions,
the claim is that a full understanding of the syntax of (particularly) spo-
ken language eludes the possibilities of a purely introspective method-
ology. Of course, no (quantitative or qualitative) corpus-based investi-
gation can do without a strong reliance on the analyst's knowledge ('in-
tuition') about the language being researched; in fact, fmding valid gen-
eralizations always involves Gedankenexperimente, playing with struc-
tural changes in and recontextualizations of the 'examples' found to be
used by the informants. On the other hand, not even the empirical start-
ing point of the present investigation (i.e., the (differing) preferences of
spoken and written language for post- and prepositioning) is available to
a purely introspective approach, since it is of a quantitative kind. In or-
der to reach an explanation of these findings, this quantitative analysis
has to be complemented by an in-depth analysis of individual cases of
usage. Such an analysis will pay attention (a) to the in-time emergence
of syntactic patterns, including the details of their delivery such as hesi-
tations, reformulations, break-offs, etc., and (b) to the interactional as-
pects of this emergence, including hearer feedback (or lack of it) and
sequential placement. In this respect, spoken language research can
profit in important ways from conversation analysis.
174 Peter Auer
2. German wenn-clauses and English ~clauses: a brief syn-

tactic and semantic overview
This, of course, is not the place for a full contrastive analysis of the two
constructions. In order to facilitate the reader's access to the examples
to be discussed below, and in order to link up the present investigation
with previous ones on tf-clauses in spoken (and written) Engiish, a short
sketch of some important similarities and differences may, however, be
in order here.
2.1. Syntax
The syntax of English suggests a (misleading) parallel between pre- and

postpositioned adverbial clauses (including conditionals) since they can
usually be exchanged without structural changes in either the main or
the dependent clause. German syntax, on the other hand, treats the two
positions quite differently. While post-positioned adverbial clauses al-
ways occupy the so-called post-field (Nachfeld), which is not obliga-
tory, and are thus tagged on to an already complete syntactic pattern,
pre-positioned subordinated clauses may be (and in written, normative
language usually are) more tightly integrated into the syntactic structure
of the following main clause: they occupy the so-called front field (Vor-
feld) of the sentence, i.e., the uniquely available and obligatory position
before the finite verb. Moving adverbial clauses from one to the other
position therefore involves structural changes in the main clause:
(1) a. wenn sie==n JOB haben wollen, () mUssen sie==n bisschen da

aufn PUNKT kommen. (original utterance) 1
'if you want to have ajob, you need to get down to the point'
b. sie miissen n bisschen auf==n PUNKT kommen wenn sie==n jOb
haben wollen. (fabricated)
'you need to get down to the point if you want to have a job'
Positioning the wenn-clause in the post-field (lb) instead of the front-

field (la) implies that another constituent will fill this position (in the
present case, it is the subject pronoun Sie). The dominant syntactic pat-
tern in which pre-positioned adverbial clauses occur in written German
may therefore be called "integrative", while the English treatment is
"non-integrative" (cf Konig and van der Auwera 1988: 103-109 for this
Positioning ofwenn-clauses in German 175
terminology and some further remarks). In spoken German, however,

the fully integrated placement of the pre-positioned adverbial clause in
the front-field is only one possibility. Alternatively, pre-positioned
wenn-clauses may be followed by a resumptive particle (a local-
temporal adverbial such as dann or one of its regional equivalents, e.g.
na, denn, no, etc.; cf (2b»;2 or they may even be used in a non-
integrative way, rather like in English (cf (2c»:
(2) a. wenn sie==n JOB haben wollen, () mUssen sie==n bisschen da

aufn PUNKT kommen. (original utterance)
'if you want to have ajob, you need to get down to the point'
b. wenn sie==n JOB haben wollen, () dann mOssen sie==n bisschen
da aufn PUNKT kommen (fabricated)
(same meaning)
c. wenn sie==n WIRKlich n JOB hiitten haben wollen; () sie hiitten
dann SCHON n==bisschen aufn PUNKT kommen miissen
(fabricated)
'if you had really wanted a job, (then) you would have needed to
get down to the point'
For (2c), special conditions of use hold, and it may therefore be consid-
ered to be "marked" (cf Konig and van der Auwera 1988, Kopcke and
Panther 1985, Giinthner 1999, and below).
2.2. Semantics
The semantics of German wenn-clauses3 is not strictly equivalent to

English conditional tf-clauses either. The cognate of English if, German
ob (> Germanic *eba), has lost its original (OHG/MHG) function of in-
troducing conditional clauses, although remnants of this usage may still
be found in (etymologically) composite concessive conjunctions such as
Modem German ob+wohl and ob+gleich and in concessive conditionals
of the type ob X oder nicht...('whether X or not'). Filling the gap, the
temporal conjunction wenn (or rather, its predecessors, MHG
swenne/swanne), a cognate of English when, has taken over most of its
functions. As a consequence, the semantics of wenn-introduced clauses
oscillates between a temporal and conditional reading in the indicative
mood. (For this reason, German wenn-clauses, other than English tf-
clauses, cannot be called conditional clauses. 4 ) Other conjunctions are
176 Peter Auer
available for a nonambiguous encoding of a conditional or a temporal

relationship.
Disambiguation of wenn is possible on the basis of contextual infor-
mation (i.e., the semantics of the remainder of the clause and/or its con-
versational context) in some, but certainly not all cases. Even in the fol-
lowing examples (which are among the clearest in my data) the para-
phrases are not entirely beyond dispute. However, they do represent the
prevalent readings ofwenn-clauses:
(3) (temporal: consecutive)

dann MELD ich mich morgen bei ihnen? (-) wenn «=SOBALD,
SOWIE)) ich bei AUror angerufen hab,
'so I'll be in touch with you tomorrow as soon as I have given
"Auror" a ring'
(4) (temporal: simultaneous)

TRIFFST du den () wenn «=WAHREND, sOLANGE)) du in PEking
bist?
'will you see him while you are in Peking?'
(5) (temporal: iterative)

ich sprEch UNdeutlich, (1) LISpel auch n=b' etwas, (-) und (-) ich
sprEch dann after zu SCHNELL. (-) wenn «=JEDESMAL WENN)) ich
beGEIStert bin, oder (-) eh im element bin; (-) dann sprech ich zu
SCHNELL,
'I speak inarticulately, I also lisp a little, and then I often talk too
fast. whenever I am enthusiastic about something, or ehm get
carried away; (-) then I talk too fast,'
(6) (conditional: hypothetical)

er will sie jetzt wieder HElraten, und die haben so ne FRIST ehm in
den islamischen liindern dass innerhalb von nem halben JAHR
oder so, muss die frau dann wieder zum MANN zuriick wenn
«=FDRDEN FALL DASS, FALLS)) er sie DOCHwiederwil1 h.
'he now wants to marry her again and they have kind of a deadline
ehm in the Islamic countries that within half a year or so the wife
has to return to her husband in case he wants her back again'
(7) (conditional: factual)

«radio phone-in, psychotherapeutic consulting; the caller has
complained about having no-one to tum to with his marriage
problems; the therapist recapitulates and formulates her advice»
denn LETZTlich .h wenn ((=DA)) sie in ihrer verwAndtsch~ft
niemand HAbn mit dems REden kOnnen, .h iih is=danns BESte, (-)
sie wiirden zu am Eheberater gehn?
'for in the end, since you have nobody among your relatives who
you could talk to, ehm it's best then to tum to a marriage counselor'
Clearly, the temporal readings of wenn are not covered by English if,
but by when instead.
Some verbal and prosodic features of wenn-clauses may facilitate or
even enforce one or the other reading: (a) the temporal, non-iterative
reading is not available in sentences referring to past events; here, the
temporal conjunction als takes over (while English allows when); (b)
focussing adverbials such as stressed nur ('only') in the main clause
strongly suggest a conditional reading of the (following) wenn-clause;
(c) the particle schon (no English equivalent) in the wenn-clause sug-
gests a factual-conditional reading; (d) immer wenn ('always when')
instead of a simple wenn as a conjunction enforces a habitual-temporal/
contingent reading; (e) selbst wenn (' even if) and wenn ... iiberhaupt
('if ... at all') instead of a simple wenn enforce a conditional reading; (f)
subjunctive (Konjunktiv II) in the wenn-clause enforces a hypothetical-
conditional reading; (g) stressed wenn-conjunctions suggest a condi-
tional instead of a temporal reading.
Two special uses of wenn-clauses need to be mentioned. The first is
the expression of concessivity through the combination of wenn and
auch (wenn + auch or auch + wenn), roughly similar to Engl. even
if/even though: 5
(8) a. auch wenn sie KEInen job haben wollen, () miissen sie=n
bisschen da aufn PUNKT kommen. (fabricated)
'even if you don't want ajob, you need to get down to the point'
b. wenn sie auch keinen JOB haben wollen, () sie miissen n
'even though you don't want a job, you need to get down to the
point'
c. obWOIa sie KEInen job haben wollen, () miissen sie=n
178 Peter Auer
'although you don't want to have ajob, you need to get down to
the point'
The auch wenn-construction (8a) differs from obwohl-concessives

(Eng!. although, (8c» in that the truth of the proposition it expresses can
but need not be taken for granted ("neutral epistemic stance"; cf Fill-
more 1990, Couper-Kuhlen, 1999): whereas the proposition 'you don't
want a job' is not asserted in (8a)/auch wenn, it is in (8c )/obwohl. Auch-
wenn-clauses therefore differ from if=-conditionals and resemble true
(obwohl-) concessives in that the presupposed generic statement is nega-
tive (for the above example: 'someone who does not want a job does not
have to get down to the point'). At the same time, they differ from true
concessives and are similar to true conditionals in that the truth of the
antecendent may but need not be asserted. Note that, differently from
auch wenn pre-positioned wenn auch-clauses (8b) co-occur with non-
integrative word order in the consequent, and always receive a factual
interpretation.
Finally, it should be noted that German wenn-clauses occur some-
times as obligatory constituents of the verb. 6 (English often uses non-
finite forms such as participle or infinitive clauses for this purpose, al-
though if=-clauses are also possible.)
(9) das EINfachste, da ham sie RECHT, das war fiir uns, wenn sie
mal==n MOnat (-) im teleFONmarketing ARbeiten wiirden.
'the simplest solution for us, and here you are right, would be if
you could work in our direct marketing sector for a month'
Syntactically speaking, wenn-clauses of this kind can be replaced by

dass-(complement) clauses (das Einfachste ware, dass sie mal im Tele-
fonmarketing arbeiten). Semantically speaking, various differences re-
sult from the choice between dass- and wenn-complements; most of
them pertain to the presumed status of the information in the comple-
ment clause (cf Eisenberg 31994: 365f for some further discussion).
3. German wenn-clauses from a quantitative perspective
The observations in this section are based on a collection of 500 wenn-

7
clauses taken from a corpus of spontaneous, direct conversations. All
instances of wenn were considered for analysis, apart from obvious syn-
tactic break-offs in the wenn-clause, some non-reconstructable utter-

ances, and the comparative uses of als/wie wenn (see note 5). In Figure
1, the total of n=500 tokens is broken down according to the position of
the adverbial clause relative to the main clause: pre-positioning, post-
positioning, parenthetical positioning within the clause, 8 independent
use of the wenn-clause as a turn-constructional unit of its own, and a
residual category of ambiguous cases (e.g. apo-koinu constructions, see
below). There can be no doubt that the front position is preferred in
spoken German.
60,00
50,00
40,00
30,00
20,00
10,00
0,00
pre- paren- post-
single others
positioned thetical positioned
IJ percentage 56,00 4,00 32,60 5,40 2,00
Figure 1. Position of German wenn-clauses relative to the main clause~ n=500.
The results agree with Ford and Thompson's fmdings on If-clauses in

English conversations, according to which initials outnumber finals by a
ratio of 4:1 (n=316, initial=81%, fmal=19%; Ford and Thompson 1986:
362), with Ford's findings based on a smaller collection (n=52, 50% of
which where preposed, 35% postpositioned, and 15% single; cf Ford
1993: 24), and with more general claims about a universal preference
for pre-positioning of antecedents in conditional constructions (Green-
berg 1963). Note, however, that the preference found in the English data
for pre-positioning of conditional clauses does not extend to temporal
(e.g., when-) clauses; rather, Ford (1993: 24) found these to follow their
main clauses by a ratio of 2:1. Given the ambiguity of German wenn-
clauses (in the indicative mood) between a conditional and a temporal
reading, it may be asked if the preference for initial placement of wenn-
clauses holds for both. To answer this question, those instances ofwenn-
clauses have been singled out (n=203) which have either a clear tempo-
180 Peter Auer
ral or a clear conditional reading (based on the substitution tests and cri-
teria discussed in section 2.3).9
Of the n=203 disambiguated wenn-clauses, 24% have temporal, the
remainder conditional meaning. Exactly half of the wenn-clauses with
temporal meaning are prepositioned and postpositioned respectively.
(Of the 76% unambiguously conditional examples, about two thirds are
pre-positioned (55% of the total), one third is postpositioned (21 %).)
There is, then, a clear difference between conditional and temporal uses:
only for the former does the preference for pre-positioning hold. Since
the majority of German wenn-clauses are semantically ambiguous be-
tween a temporal and a conditional reading, this fmding also suggests
that, taken as a whole, they behave syntactically like (English) condi-
tional rather than temporal clauses.
Figure 2 shows the percentage of integrative, resumptive and non-
integrative constructions among the pre-positioned wenn-clauses in the
sample (n=280).
60,00
50,00
40,00
30,00
20,00
10,00
0,00
integrative resumptive non- others
integrative
25,72 47,83 18,48 9,42
Figure 2. Integration (%) of pre-positioned wenn-clauses into the subsequent main

clause (n=280)
The relatively large residual category ("others") covers wenn-clauses

plus subsequent main clauses within larger hypotactical constructions
(see below example (26)-(28)). Again, the results are very clear: re-
sumptive constructions are preferred to fully integrated and totally non-
integrated constructions. The canonical, integrative construction of
standard written German only plays a secondary role in spoken German.
Some comments on non-integrative wenn-clauses in German are

necessary at this point. Pre-positioned wenn-clauses occurring in the
pre-front field of a sentence are basically of two types (cf Auer 1996).
We find instances which cannot be positioned in the front field (i.e., in-
tegrated into the main clause); in other words, the only available pattern
for them is non-integrative syntax. This is sometimes for syntactic rea-
sons; in particular, yes/no-questions and imperatives, which are verb-
initial syntagms in German, do not have a front field, and in w-
questions, the w-question word is usually said to occupy the front-
field. 10 In these contexts, adverbial clauses either need to be post-
positioned (despite the general preference for pre-positioning), or to be
non-integrative. Of the 45 questions/imperatives in the sample, 16 have
pre-positioned wenn-clauses, i.e., non-integrative word order (cf (10»,
while 29 have post-positioning; this means that the normal preference is
reversed in this syntactic environment.
(10) ich mein ich muss ihnen (-) ganz SCHNELL und GANZ: vehement
sagen wenns IRgendwie gEht (-) fahrns InN
'I mean I have to tell you without hesitating and very vehemently:
if you can make it at all, go there!'
However, there are also semantic reasons why certain wenn-clauses

have to occur in the pre-front instead of the front field. This is the case
for "speech-act related" 11 wenn-clauses which do not conjoin two
propositions on the content level; often, they are used in order to miti-
gate subsequent face-threatening acts (such as, in the following exam-
ple, an interruption). The apodosis is asserted independently of the pro-
tasis, and this semantic independence corresponds with obligatory syn-
tactic non-integration:
(11) (Gob interview»

wenn ich (-) grad WElter ausfiihren darf; (0.5) Sie wissenja in de:
in der A Utoinduschdrie .h herrschen SEHR grofJe k'
konkurRENZ, markt
'if I may continue elaborating on that; (0.5) you know that in
the car industry there is a lot of competition «etc.»'
In such cases, the marked position of the wenn-clause in the pre-front

field helps to contextualize a marked (non-referential) semantic inter-
pretation.
182 Peter Auer
But there are also contexts in which non-integrative syntax is fre-

quent although not obligatory. For instance, there is a tendency for non-
integrative clause-combining to occur in concessive wenn nicht-
constructions:
(12) wenn auch die theoRIE; (-) eh (-) so IRgendwo mal gehOrt wurde
im KOPF? (-) eh das UMsetzen das ist ja das entSCHEldende,
'even though the theory (-) ehm (-) may have been heard
somewhere in one's head (-) the decisive thing is putting it into
practice'
Another frequent function of non-integrated wenn-clauses is topicaliza-

tion; in this case, the wenn-clause is typically followed by an anaphoric
pronoun back-referencing the proposition expressed in the wenn-clause
as a whole, or an element contained in it. In the following example, the
wenn-clause introduces a new discourse referent or topic; it is in many
ways equivalent to other topicalization constructions (such as a cleft
construction: was Ihre Fragen angeht, die konnen Sie jetzt stellen), with
the additional implication that the speaker is not certain about the rele-
vance of the new discourse referent for the co-participant.
(13) also wenn sie FRAgen ham zwischendufch, eh DIE konnen Sie
ruhig STELlen?
'well if you have any questions in between, ehm you can ask
THEM of course. '
A similar topicalization (not of a single referent, but of a whole proposi-

tion) is involved in the following example:
(14) un wenn ich mein Eltem anrufh wUrde, ==DS wiirde AUCH nix
bringn.
'and if I called my parents, that wouldn't be any use either. '
Here, the wenn-clause could even be entirely replaced by an infinitival

construction (meine Eltern anzurujen), since potentiality is already ex-
pressed by the conditional verb form wiirde... bringen and redundantly
coded by wenn. Finally, non-integrated wenn-clauses often express em-
phasis and lend an emotional meaning to the utterance: 12
(15) wenn WIRKlich==n ganzen tag das telefon klingelt, und acht
STUN' (-) man IS hinterher' < <acc>man WEISS was man>
getan hat. () geb ich ehrlich ZU
'if the phone really rings all day, and eight hOD' (-) afterwards
you are' you know what you have done. i have to admit that. '
In (15), the speaker describes her working-day in a call-centre and

wants to underline that dealing with callers is a tiring job~ one of the
strategies used to convey this meaning is the non-integration of the pro-
tasis into the apodosis.
4. Some reasons for pre- and post-positioning
What are the advantages of pre-positioning wenn-clauses? This question

seems less difficult to answer than the opposite one of why a certain
number of these clauses - roughly a third in our data - are post-
positioned. We will deal with each question in turn.
4.1. The advantages ofpre-positioning
To start with, it should be noted that the preference for pre-positioned

wenn-clauses is not just a quantitative fmding but is reflected in speak-
ers' changes in the design of an emerging syntactic pattern 'in mid-
stream'. Particularly striking are cases such as Ex. (5), in which a post-
positioned wenn-clause is retrospectively turned into a pre-positioned
one via what might be called an apo-koinu construction:
The K01VOV here, of course, is wenn ich beGEIStert bin, oder (-) eh
im element bin. It seems that the speaker, having completed the three-
part list of his verbal handicaps, wants to qualify the last item retrospec-
tively. He could have done this by simply adding the wenn-clause in the
post-field but recycles this last component instead, with the wenn-clause
inserted before it. The wenn-clause here is both final and initial. in-
stances in which a clause is broken off and a wenn-clause is inserted be-
fore it is re-started (as in (16)) are also evidence for the interactional
relevance of pre- vs. postpositioning.
(16) ichfahr (-) wenn (-) wenns iiberHAUPT geht

dennfahr ich NA:CH(er) erscht in Urlaub,
184 Peter Auer
'I'll have if if it works out at all

then I'll only have my holidays afterwards'
So why this additional effort? There seems to be some kind of cognitive

'naturalness' in the way in which conditionals create the ground - or, in
more recent but equally metaphorical parlance, set up a "mental space"
(Fauconnier 1985) - in which some hypothetical or factual proposition
is located. 13 For cognitive reasons, it is the grounding which (iconically)
precedes the focal proposition, and not the other way round. Ford, for
instance, suggests that "the prevalence of initially placed if-clauses may
reflect the general tendency to signal «... » that the interpretation of the
coming clause will be, in some general way, limited by the contents of
the if-clause" (1993: 15). Further evidence for the 'naturalness' of this
position can be derived from the affinity of conditional clauses and
topic-introducing devices (topics precede comments), for which some
evidence has been given in the preceding section (see Haiman 1978,
Ford and Thompson 1986 for an in-depth treatment of this line of argu-
mentation), and from the affinity of conditional and causal clauses
(where causes iconically precede their effects). The advantages of this
discourse function seem to outweigh the cognitive costs linked to the
deployment of a syntactic pattern which projects considerably into time.
It may not have been sufficiently taken into account in previous re-
search on clause positioning, however, that this projection in time has an
interactional side as well: 14 speakers who open up far-reaching syntactic
gestalts claim the tum for at least the time which is necessary to bring
them to a well-formed conclusion. In other words, producing a wenn-
clause gives the speaker the right and obligation to go on talking; it
functions as a tum-holding device until the formulation of the conse-
quent is completed. There are numerous cases in the data in which
highly complex turns emerge in this way, since the speaker uses the
space between a gestalt-opening wenn-clause and a terminating main
clause for detailing the "mental space" opened up by the first compo-
nent. Two elaborate examples (as they seem to be typical for institu-
tional talk) are (17) and (18):
(17) (Gob interview; applicant B is talking about his previous

employment in a West German consultant company which,
however, withdrew from East Germany, despite the fact that it
had highly experienced consultants»
Positioning ofwenn-c/auses in German 185
B: zum beispiel einen herren, () KELler?

(-) eh der () is: () FONFundzwanzig jahre
unterNEHmensberater? () der hat=n STAM!vfklientel in uh es ah
KAnada?
I: mhm,
B: und DER war naturlich, (-) ein FACHmann. (-) aber er KOMMT,
()
in die neuen BUNdesliinder? () «acc>er war ja nu> (-) eh
hatte es ja gar nicht mehr NDtig gehabt; =da () so [VIEL] (-) zu
REIsen,
I: [mhm,]
B: aber (-) er IS in die neuen BUNDdesliinder gekommen, (-)
um auch etwas zu be WEgen. (-) aber wenn er dann nur auf der
STRASse () sitzt, (-) und DANN (-) den () kliENten () mit nach
schweRIN nehmen muss um=n FDRderantrag zu stellen; (-) dann
wieder zur BANK, (-) und die BANK sagt () wir brauchen
erst=ne ZUstimmung von dem FDRderinstitut,
I: =mhm,
B: vor[her () ko]nnen wir nicht die geSAMTfinanzierung,
I: [(h) ]
 wie mit KDpenick. ja
B: und () fund er da]nn NUR aufder STRAsse ist;
I: [(h)]
B: (-) dann SAGT er das LOHNT sichfiir mich nicht. (-)
dann bleib ich LIEber () in nordrhein westFAlen.
'B: for instance a Mr Keller

ehm who has been a consultant for 25 years
he has his regular clients in the U.S. and Canada
I: mhm,
B: and he was a specialist of course. (-) but he is coming
to the New States (.) he certainly had (-) ehm he had no need to do
that any more; to travel so [much there
I: [mhm,
B: but (-) he did come to the New States, (-)
in order to get something moving. (-) but if he is on the road all
the time (-) and then (-) he has to take his client with him to
Schwerin in order to hand in the proposal for the subsidies; (-)
and then back to the bank, (-) and the bank says (.) first we need
the subsidizing body's consent
186 Peter Auer
I: mhm,
B: before [that we cannot (do) the total financing
I: [(h)
like with Kopenick 15
B: and (.) and he is just on the road then he says this isn't worth it
for me.
then I'd rather stay in North Rhine-Westphalia «a West German
state»'
This passage is embedded into a larger report the applicant gives of his
participation in a West German consultant agency in the New States,
which however closed down its East German office, making him redun-
dant. The interviewer does not seem to know the company and ques-
tions its importance on the market. The applicant counters by stating
that although small, the company had very professional consultants. At
the same time, he has to deal with the interviewer's innuendo that the
company withdrew from the East German market because it was not
working successfully. In this context, the case of "Mr. Keller" is men-
tioned, an experienced consultant who was disappointed by the
kafkaesque way in which state and bank authorities made it hard for
new enterprises to get subsidies, and returned to the Old States.
After he has been portrayed as a successful consultant who came to
East Germany mainly for idealistic reasons, "Mr. Keller's" dissatisfac-
tion with the situation is described in a complex tum construction which
starts out with a wenn-clause (wenn er nun aufder Straj3e sitzt... ). In the
given context, the interpretation is not hypothetical but refers to a (fac-
tual) state of affairs ('since he was always on the road... '), which is es-
tablished as the ground from which some conclusion can be drawn. Be-
fore this conclusion is reached, however, the speaker elaborates at con-
siderable length on the unfortunate situation in which "Mr. Keller" and
his clients found themselves; in four clauses each introduced by (und)
dann, the various fruitless journeys between the financing bank and the
state authorities in Schwerin are described. Towards the end of this
elaboration (securely produced by the speaker within the realm of his
own tum, since a syntactic projection - that of the when-clause - still
remains to be taken care of), the interviewer produces some recipiency
tokens which, although not claiming the tum (cf their reduced loudness,
indicating non-competitiveness), nevertheless acknowledge the
speaker's point: two laughter particles and one comment (wie mit
Kopenick) display understanding. Only after this feedback does the
speaker close the syntactic gestalt with two resumptive dann-clauses.

Their content is higWy predictable, given the fact that it has been men-
tioned before that the company closed down its East German branch. It
seems, then, that what the speaker wanted to convey by this complex
turn is not so much this consequent but rather the details of the situation
which led to it. The relevant information of this complex construction is
what is produced BETWEEN the initial wenn-clause and the final dann-
clauses. The speaker employs the projecting force of the first in order to
claim conversational space for himself: and makes use of this space as
long as he needs it to 'convince' the recipient of his point (as evidenced
by the recipient's responses). The 'orderly' conclusion of the turn is
produced as soon as this purpose is reached.
The following extract similarly shows how pre-positioned wenn-
clauses can be used to claim conversational space:
(18) (bulimia therapy)

M: aso ich hab ma mit einer zuSA!vfMgewohnt, ==
==und .h die hab ich EH nich so leidn kOnn un sie mich AUCH
nich,
und dann hab ich IMmer so.h (0.5)
und (-) DIE: is schon wesentlich DICker als ich; ==
==und dann hab ich ECHT immer gedacht (0.5)
ich hab so alles des (-) AUF se projeziert
und wenn se viel geGESsn hat, ==
==die hat sich .h SAHne n ganzn becher SAHne mit
Apfelschnittchen drin gegessn. ==
==und das warfiir mich ECHT der ABscheu. ==
<<fast>n hab ich gedacht> .h des is ja wohl (1.0) des is
FURCHTbar (1.0)
wie KAMmer denn sowas ESsn un auch noch mit gUtm ge WISsn.
'M: you see I once lived with a girl

and I couldn't really stand her and neither could she me
and then I always
and she really was a lot bigger than I was
and believe me I always thought
I projected everything on her
and when/if she ate a lot,
she put cream a whole cup of cream she ate with slices of apple in
it
188 Peter Auer
and to me that was really disgusting.

then I thought .h isn't that (1.0) that is really appalling
how can you eat anything like that and without even feeling
guilty. '
Once more, a speaker is involved in telling a story which in this case is

supposed to show how she projected her own feelings of guilt for eating
too much onto her flatmate. And once more, a wenn-clause is the first
component of a syntactically cohesive turn construction which spans six
intonation units. The speaker does not go into gestalt closure (apodosis)
after the wenn-clause, but rather parenthetically includes information
detailing the claim that the roommate 'ate a lot', and how she herself
reacted to that emotionally. Only then does a (dan)n-clause follow
which ties back to the initial part of the turn, where a story concerning
'projection' (ich hab so alles des aufse projeziert) was announced.
There is only one legitimate way for a recipient to share (or rather,
intrude into) the conversational space which a wenn-projection creates
for the current speaker: by becoming a co-speaker herself, i.e, by col-
laboratively producing the gestalt-closing apodosis matching the already
produced protasis (cf. Lerner 1991). Both inserted material between pro-
tasis and apodosis and collaborative constructions pivoting around this
transition suggest that there is some interactional work going on, and
that, at least in a substantial subgroup of examples, the construction is
not planned and executed as one whole, but rather develops in (at least)
two steps.
4.2. Why post-positioning at all?
If pre-positioned wenn-clauses are both cognitively more 'natural' and

interactionally more advantageous than post-positioned ones, why do
the latter occur at all? Two reasons have already been mentioned in sec-
tion 3: wenn-clauses may be used for expressing the temporal circum-
stances of an event, and since temporal adverbial clauses do not follow
the preference for pre-positioning, wenn-c1auses of this semantic type
need not do so either. I Secondly, it was shown that the absence of a
front-field in questions and other verb-initial syntagms makes their post-
positioning more likely. 17 There are, however, other important reasons.
First of all, it may be asked if there are any further syntactic envi-
ronments in which post-positioning is preferred or even necessary.
There is indeed another construction in which the front-field is not

available: that in which the wenn-clause plus subsequent clause are
themselves embedded into a larger construction. The various types of
embedding show different patterns with respect to the possibility of pre-
positioning. As in Ford and Thompson's English data (1986: 359), final
positioning is preferred "when a conditional clause occurs within a
nominalization, an infinitive, or a relative clause". Take, for instance,
the following case of a relative clause:
(19) (therapy session)

TM: s==ESsn isch wie? ein Teddybiir. ==
TW: ==ja,
TM: den' den sie: (-) .h mit sich RUMtragn. (2.0)
damit SIE < nich allEin sein miissn. >
un dem==mer (-) sich RANzieht, ja? (-) (-) wenns HARTwird;
(3.0) an dem=mer sich FESCHThiilt, (2.0) wem==mer EINsam
isch, (1.0) nd der ii: berall MIT muss.
'TM: eating is like a teddy bear.

TW: yeah
TM: who who you carry around with you.
so that you don't have to be alone.
and whom one holds close, right? (-) (-) when life becomes hard;
whom one clings to, (2.0) when one is lonely,
and who has to come along all the time. '
Both wenn-clauses in this extract are part of a relative clause introduced

by an oblique relative pronoun, i.e., their matrix clause is itself subordi-
nated, and therefore has verb-final syntax (cf the placement of the finite
verbs ranzieht and festhiilt). Here, the wenn-clause cannot be placed in
front of the relative clause (*und wenn's hart wird den man sich ran-
zieht); pre-positioning would require a superordinate main clause in-
stead of the relative clause (und wenn's hart wird, zieht man sich den
ran).18 The same applies to dependent clauses introduced by wie 'as',
obwohl 'although', weil 'because', etc. which likewise do not allow ini-
tial wenn-clauses.
However, subordination by the most frequent complementizer dass
('that') shows a different pattern. Here, we frequently encounter initial
placement of the pre-positioned wenn-clause be..fore the complementizer
dass:
190 Peter Auer
(20) MEIN interesse is natiirlich

WENN ich da: .h schon als POSTdoc auf==m ZEITvertrag bin;
dass ich wahrend de dieser ZEIT dann; (-) auch==n paar
ergebnisse MITnehme
'my interest of course is
if I am there as a post-doc on a temporary contract
that I can take at least some results with me during this time'
(21) und DESwegen war es natiirlich; (-) fir uns WUNschenswert; ()

<<scanning>WENN wir uns einigen KONNten, >
dass sie so friih wie MOGlich <dim>natiirlich anfangen.
'and therefore of course it would be desirable for us
if we could come to an agreement
that you start as soon as possible'
The additional stress on wenn in these examples may give us a clue to

the origin of this construction; arguably, it underlines the semantic link
between antecendent and consequent. Fronting the wenn-clause to a po-
sition before the dass-complementizer may be another way of focussing
on the semantic link established by wenn. 19 Note in passing that the
fronting of the wenn-clause renders its scope ambiguous both in (20)
and (21): it mayor may not include the initial phrases mein Interesse ist
natiirlichl ...ware es natiirlichfir uns wiinschenswert (i.e.: 'of course, if
I am only there as a post-doc on a temporary contract, then my interest
is to take at least some results with me' and 'if we could come to an
agreement it would of course be desirable for us that you start as soon as
possible' respectively).20
In addition to these syntactic constraints, there are semantic-syntactic
reasons for post-positioning wenn-clauses. In particular, wenn-clauses in
complement function are usually post-positioned (cf (9) above). As a
rule, the main clause contains an evaluative two-place predicate, with
the wenn-clause expressing the proposition which is evaluated.21 The
opposite serialization is not unacceptable; nevertheless, it is very rare.
The dominant pattern obviously parallels that of dass-introduced com-
plement clauses which can, but rarely do, precede the main clause as
well. Complements make up ca. 25% of all the post-positioned wenn-
clauses in the spoken materials investigated.
Finally, and most importantly, post-positioning of wenn-clauses is
linked to the pragmatic status of the proposition they express, and to the
interactional possibilities this position opens up both for the speaker and
the hearer. As outlined in section 1, final subordinated clauses in Ger-

man are added onto an already complete syntactic structure. They are
therefore a straightforward means for expanding a syntactic gestalt, and
thereby the tum-at-talk. This is particularly obvious in cases where syn-
tactically complete syntagms preceding the wenn-clause are marked as
terminal by intonation, e.g. by a pitch fall to the speaker's base line (full
stop in the transcription); the wenn-clause then appears as an after-
thought, or epexegesis (cf Auer 1991):
(22) «hypothetical talk about a situation in which two people are in

conflict over where to put the cup for the coffee; B is asked to
mediate»
B: ich wiird [einfach] die () die tasse kaffee NEHmen,
II: [h:m,]
B: und eh (-) WEGste//en. (-) .hja? ()
eh==s da stunden/ang streiteREIen gibt,
wiird ich sagen, a/[so: ] jetzt is:-
II: [hm,}
(0.5)
II: da hiitt==ich iirger [mit IHnen.
B: [sch/uss aus ENde?
II: [wenn SIE mir dann auch noch den KAFfee (wegschlieften.)]
B: [he he he he he he
'B: I would [simply take the (.) the cup of coffee

II: [mhm
B: and ehm (-) put it away. (-) you see?
before they start quarreling for hours,
I would say right [now it is
II: [hm,]
(0.5)
II: then I would have trouble [with you.
B: [over and out
II: [if you (shut away) my coffee
B: [he he he he he he'
At a point where B has already suggested to simply 'taking away' the

disputed coffee cup, but is in the middle of a syntactic construction
elaborating on this proposal (eh's da stundenlang Streiterei gibt wiirde
ich also sagen: Schluss, aus, Ende) II intervenes during an intra-tum
192 Peter Auer
hesitation pause to refute this solution: 'if you did that, there would be
trouble between the two of us' (i.e. between the mediator, B, and one of
the two people quarreling, i.e. himself). The utterance is linked to B's
proposal by the initial anaphoric da; it is semantically and syntactically
complete, and being marked by a final fall, it certainly is a candidate for
a complete tum. However, B does not pick up this refutation, but con-
tinues with the production of the unfinished syntagm in another piece of
simultaneous talk. Sequential structure and temporal development are
now out of phase: a response has been produced to an utterance which is
still in need of being completed, and is only completed after the re-
sponse. In this context, II's following wenn-clause, syntactically ex-
panding an already complete turn/syntagm, can be seen as a skillful way
of re-aligning sequentiality and timing: it re-instantiates II's refutation
of B' s proposal without repeating it, by retrospectively transforming a
simple construction into a hypotactical one with a post-positioned ad-
verbial clause. Semantically, this expansion adds nothing new: it just
restates what B herself has said before.
The possibility of such an expansion is not only available to the
speaker but also to the recipient, of course, who may become a co-
speaker and co-producer of the emerging syntactic pattern by adding a
wenn-clause himself/herself:
(23)
L: .h dann: eh () wir' der Hundwird auchjetz zunehmend ruhiger;
S: mHM (-) des GUT so; (-)
L: JAja des==also wird langsam (a)==richtiger HUND;
S: aHA
L: hm, (-)
S: wenn==er (nicht mehr) abhaut, (-)
hat (name) des ANgebot jetz fiir den zaun? ==
'L: then ehm beco the dog is becoming more and more calm now;
S: this is how it should be;
L: yeah slowly he's tuming into a real dog.
S: I see
L: hm,
S: if he doesn't escape (any more),
did NN get the offer for the fence in the meantime?'
So it is not only the transition between a wenn-clause and its subse-

quent main clause which is sensitive to turn-taking, but also the inverse
transition between a (main) clause and its subsequent wenn-clause. But
obviously, there is an important difference: while in the first case an
open syntactic projection is in play, in the second case the first speaker
has already come to an orderly completion of the sentence/turn.
Post-positioned wenn-clauses thus offer the possibility not only of
expanding a turn, but also of expanding a sentence by adding a post-
field constituent. At least example (22)22 also points to an important
pragmatic feature of such expansions: its low information value. Indeed,
this applies to a very large number of post-positioned wenn-clauses. Of-
ten it is the whole previous text which functions to build up the 'mental
space' that is necessary to come to the conclusion expressed in the main
clause, while the post-positioned wenn-clause only summarizes this pre-
ceding text, sometimes slightly changing the focus. In (24), the
introductory adverbial insofern explicitly establishes this resultative link
between pre-text and conclusion, while the post-positioned wenn-clause
just repeats what is known from the previous conversation anyway (the
wenn-clause is factual here):
(24) «after a long discussion of the applicant's career aspirations in the

bank, and an equally long description of the branch of the bank in
Stralsund and its sophisticated private customer service, which
seems to match these wishes»
(( ..)) das HAM wir alles in stralsUnd, also inSOfern, (-) eh ware
das==ne ideAle (-) STELle, (-) wenn sie (-) praktiZIEren wollen im
verTRlEBSbereich. im KUNDdennahen bereich.
'we've got all that in Stralsund, so in that regard, this would be
an ideal position if you want to be a trainee in the sales
department. in client-oriented business.'
As in other, similar cases of low-relevance wenn-clauses in final posi-

tion, the front-field is used here for a connecting (anaphorical) adver-
bial, which is preferentially placed in sentence-initial position, where its
indexical meaning is most easily processed. Since only one constituent
may be placed in the front-field, this position is not available for the
wenn-clause any longer.
Since post-positioned wenn-clauses are often of low pragmatic rele-
vance, upgrading their informational value requires special means; a
standard technique for doing so is the use of focussing particles such as
194 Peter Auer
(stressed) auch, nur or dann, or a combination of these. In this case, it is

the focussing particle which projects syntactically: it requires a constitu-
ent to follow which is in its scope. Therefore, wenn-clauses such as the
following cannot be treated in the same way as post-field wenn-clauses
in general: they do not expand an already complete syntactic pattern but
rather close a gestalt projected by the particle.
(25) «about wearing glasses))

ich zieh==se nur DENN ouf wEnn==i==se wIrklich (-) Effektiv
brOuch
'I only put them on then when I really and positively need them'
Ex negativo, the necessity of using such focussing particles in order to

upgrade the following wenn-clause to rhematic status is evidence for the
(sub)thematic status which wenn-clauses usually have in the post-field.
5. Pre- and post-positioning of wenn-clauses in written Ger-

man
In the last sections, it has been shown that wenn-clauses are preferen-
tially pre-positioned with respect to their main clauses in spoken Ger-
man, and that this serialisation has a number of cognitive and interac-
tional advantages. It has also been shown that the more marked struc-
ture, i.e. post-positioned wenn-clauses, which does occur in about a
third of all instances, has its own specific contexts of usage. These are
partly due to (a) syntactic constraints on pre-positioning in superordi-
nate clauses without a pre-field or in which two complementizers occur
in adjacent position; partly to (b) tum-taking (afterthought position);
partly to (c) semantic-syntactic reasons (wenn-clauses in complement
function are postpositioned); and partly to (d) pragmatic reasons (post-
positioned dependent clauses are thematic or subthematic, unless focus-
sing particles indicate the contrary).
In written German, wenn-clauses are generally less frequent than in
spoken discourse, a finding which contradicts the frequent claim that
spoken language avoids syntactically complex constructions: the fre-
quencies of wenn-clauses per 100 words in the corpus of spoken lan-
guage used above is 0.54, but in a corpus of written language, taken
from the newspapers DIE ZEIT (politics section) and Frankfurter All-
gemeine Zeitung (culture section), it is 0.33. 23 To put it differently,
every 186th word is wenn in our spoken corpus on an average, but only
every 300th word in our written texts. Once more, this fmding is in line
with comparative work on written and spoken English (Ford and
Thompson 1986, 354: 0.72 vs. 0.46; similarly: Beaman 1984 and Biber
1986), but also with previous work on German (Leska 1965: 450).
There are of course also qualitative differences between the wenn-
clauses used in the two corpora; in particular, certain rather idiomatic
patterns (constructions) seem to be more or less exclusively used either
in spoken or written language. For instance, the reduced wenn-(dann)-
constructions of spoken German (cf section 2.2. above) do not occur in
the newspaper corpus, while, on the other hand, the topicalizing causa-
tive construction wenn p dann (deswegen), weil q ('if p, then that is be-
cause of q'), as in (26), seems to be used exclusively in writing.
(26) (DIE ZEIT Nr. 8 17.02.1995)

Wenn wir das Leugnen von Auschwitz, anders als das Leugnen
der kopernikanischen Wende, unter Strafe stellen, dann deswegen,
weil es uns nicht nur hypothetisch angst macht.
'If we punish the denial of Auschwitz, and not the denial of the
Copernican revolution, then that is because it does not make us
feel afraid only hypothetically. '
Also, and contradicting received wisdom according to which written

language is more logical and more explicit, we fmd instances of wenn as
a conjunction in the newspaper texts with semantics which are exceed-
ingly vague, as for instance in (27):
(27) (F.A.Z 19.06.1993, S. 27 / Nr. 139)

Aber wenn zum stets und instiindig angestrebten "Weltniveau" der
DDR die Stellvertreterschaft des iiberragenden, fortschrittlichen,
darin sagar selbst "biirgerlichen" Kulturerbes gehorte, dann
iiufJert sich nun das AufschliefJen zur Weltoffenheit seltsam
kleinmiitig.
'But if the representation of an outstanding and progressive
cultural heritage, one which includes even the "bourgeois",
belonged to the ever and urgently sought after "international
standard" of the GDR, then growing into cosmopolitan open-
mindedness expresses itself rather timidly nowadays. '
196 Peter Auer
Here, the wenn-dann construction seems to vaguely express something

between adversativity and concessivity.
However, these differences only affect a relatively small number of
examples and are not directly linked to the positioning of the wenn-
clause. The important question for the present discussion is rather
whether the preference for pre-positioning of wenn-clauses is also to be
found in written German (as it is in written English, cf Ford and
Thompson 198624). Fig. (3) shows that this is not the case:
60,00
50,00
40,00
30,00
20,00
10,00
0,00
a percentage 51,60 1,60 1,00
Figure 3. Percentage of post- and pre-positioned wenn-clauses in written Ger-

man (n=626)
In the written materials, post-positioned wenn-clauses are almost one

and a half times more frequent than pre-positioned ones, while the op-
posite ratio is found in the spoken material. 25 As would be expected, the
number of integrative pre-positioned wenn-clauses is higher than in
conversational language (65.56%), and both the number of non-
integrative (9.28%) and resumptive (26.16%)26 structures is markedly
reduced. The preference for post-positioning is only slightly less pro-
nounced in unambiguously conditional wenn-clauses than in unambigu-
ously temporal ones (65 % vs. 71%) (n=301).
What could be the reason for this reversal of preferences for post-
and pre-positioning in written German compared to spoken language?
Three factors seem to be primarily responsible for it. First, the number
of (almost exclusively final) wenn-clauses with a preceding focus parti-
cle in the main clause is about eight times as high in the written as in the
spoken material (40 vs. 5 occurrences). An example is:
(28) (DIE ZEIT Nr. 7 10.02.1995)

Danach durfen Frauen ungewollte Schwangerschaften in den
ersten drei Monaten nur beenden, wenn sie sich vorher haben
beraten lassen: in einer Beratungsstelle und vom abtreibenden
Arzt.
'Accordingly, women may only terminate an unwanted pregnancy
during the first three months if they have undergone counselling:
in an advice centre and also by the physician who does the
abortion. '
Secondly, although embedded wenn-clause plus main clause construc-

tions are not more frequent in the written than in the spoken material, all
52 wenn-clauses of this type are post-positioned, while a majority of
them (34 of 54, all of which are embedded into dass-constructions) are
pre-positioned in the spoken materials. In other words, fronting of
wenn-clauses before the embedded matrix clause such as in (20) or (21)
does not occur in the newspaper texts.
Finally, one of the important reasons outlined above for frequent pre-
positioning in interactional language use is simply not applicable to
writing: this is the need for the speaker to claim conversational space for
the production of a larger tum, through projecting syntactically beyond
the current clause. Instead, another factor becomes relevant: in writing,
the wenn-clause may become so complex that processing it would be-
come difficult even in reading if it was pre-positioned with respect to its
main clause; cf for instance:
(29) (DIE ZEIT Nr. 03 13.01.1995)

Und man denkt an Talleyrands Feststellung: "Hochverrat ist eine
Frage des Datums", wenn man sich daran erinnert, daft Hans
Modrow in einer Phase als Reformer und Hoffnungstriiger galt,
aber in der niichsten fiir schuldig erachtet wurde, weil er
mitverantwortlich warfiir das DDR-System.
'And one thinks of Talleyrand's statement: "High treason is a
question of the date" when one recalls that Hans Modrow was
regarded as a reformer and as a source of hope in one phase, but
that he was found guilty in another, because he shared
responsibility in the GDR political system. '
198 Peter Auer
Neither of these reasons for post-positioning in written German can ex-

plain the difference between the English and the German results, of
course. Why should these same reasons not lead to a preference for
post-positioning in written English as well? If one was looking for a
structural explanation, one would probably try to find an answer based
on the most prominent difference between English if- and German
wenn-clauses, i.e. the semantic ambiguity of the latter. For instance, it
might be argued that since wenn can often be interpreted either condi-
tionally or temporally, German newspaper journalists try to disambigu-
ate their sentences by using other, strictly conditional conjunctions in-
stead, such as falls, im FaIle dass, fiir den Fall dass, sofern or soweit.
However, this hypothesis receives little empirical support in my data:
not only are these conjunctions very rare in the newspaper texts (a total
of 31 tokens!), they also fail to show a positional distribution different
from that of wenn (6 initial vs. 17 fmal tokens, with 8 parentheticals).27
Another possibility to express conditionality in German which is not
available in present-day English (apart from peripheral cases) is inver-
sion (as in: kommst du zu spat, bestraft dich das Leben := wenn du zu
spat kommst, dann bestraft dich das Leben 'if you are late, you will be
punished by life'). This possibility is almost never used in spoken Ger-
man because of its bookish and high-register connotations but its occur-
rence cannot be excluded in rather conservative newspapers such as DIE
ZEIT and F.A.Z.; and since the distribution of pre- and post-positioned
conditional clauses with inversion is unknown we cannot exclude a bal-
ancing effect, for instance due to a preference for pre-positioning in this
case. Since this syntactic pattern can only be quantified in syntactically
labelled corpora, there is no possibility to test this hypothesis in a
straightforward way in our materials. However, preliminary analyses of
some texts suggests that inversion hardly occurs in newspapers. Alterna-
tively, one might look for a non-structural explanation which would lo-
cate the reason for diverging English and German patterns on the textual
level, possibly in the stylistic preferences of English and German text
composition. The matter clearly awaits further investigation.
6. Conclusion
In this paper, I have looked at the placement of German wenn-clauses in

spoken and written texts. Various explanations for the general quantita-
tive results - i. e., that spoken German prefers pre-positioning, written
German post-positioning - have been presented and discussed on the

basis of individual conversational contexts in which wenn-clauses occur.
The general conclusion of this study is that the supposed parallel be-
tween 'left' and 'right' in syntax (suggested by parlances such as 'left
extraposition' vs. 'right extraposition', or 'left-adjoined' vs. 'right-
adjoined') is fundamentally mistaken when applied to spoken syntax; in
speaking, to be sure, there is no 'left' and 'right', but only 'earlier' and
'later'. At least for an approach to syntax which takes the in-time ('on-
line') emergence of (particularly) oral language units seriously, what is
dealt with first and what is taken care of later cannot be seen as a deci-
sion between two logical equivalents (as between 'right' and 'left').
Rather, it involves one of the most basic and far-reaching decisions a
speaker can make, with all kinds of cognitive, interactional and struc-
tural repercussions. 28
Appendix: Regularisation ofTranscriptions ofWord-Count(example)

original transcription: regularised transcription:
M .hh ich will UMgehend den (-) nachs- Ich will umgehend den nachsten An-
ten (-) ANrufer wieder einen Horer rufer, wieder einen Horer, begrii.6en:
begrii.6en guten Abend? Guten Abend.
A guten Abend, Guten Abend.
B guten Abend? Guten Abend.
(0.5)
A .hhh ja; ALso hh das proBLEM ah Ja, also das Problem aller steht auch
ALler steht ah auch ah so fib vor MIR so vor mir jetzt irgendwo. Wie und
jetzt irgendwo-wie wie und wo AN- wo anfangen am besten? Es ist eine
fangen am besten? etwas auBergewohnliche Problema-
B rnhm, tik, die vielleicht nicht so ganz haufig
A = ahrn: (--) es is eine etwas AUSser- in ihrer Sendung erscheint, obgleich
gewohnliche Proble ProbleMAtik, die das eigentlich eine total menschliche
vielleicht nicht nicht so ganz ah: : Angelegenheit ist. Ich lebe seit, urn
HAUfig in ihrer sendung erscheint, gleich mal auf Fakten zu kornrnen,
.hh obgleich des eigentlich ah::: ne ich lebe seit circa fiinfzehn Jahren
total MENSCHliche angelegenheit is- mit einern Mann zusarnrnen.
ich lebe: seit =um gleich mal auf
FAKten mal zu KOMM: ich lebe seit
hh fib circa FONFzehn JAHRN mit
einern MANM zusArnrnen? [.hh ]
B [ja] Ja?
A und fib auch ah recht GU:T eigentlich Und auch recht gut, eigentlich. Na-
=natiirlich mit den iiblichen ALLtags- tiirlich mit den iiblichen Alltags-
schwierigkeiten die iiberall existieren schwierigkeiten, die iiberall existie-
200 Peter Auer
original transcription: regularised transcription:
auch in HEterobeziehungen, ren, auch in Heterobeziehungen.

B ja, la.
A u:nd fih das=(dies) ist nicht unsere Und das ist nicht unsere Problematik,
probleMAtik eigentlich, wir kommen eigentlich. Wir kommen mit unserer
mit diesem mit unserer verANlagung Veranlagung sehr gut zurecht, also
sehr gut zuRECHT? also=[das]= ist das ist alles o.k. soweit.
alles o.k. soweit,
B [ja:] Ja.
Notes
* The corpus research on which this paper is based was supported in many and
substantial ways by Benjamin Stoltenburg. Thanks to Jessica Wallace for cor-
recting the worst blunders in my English, and to Susanne Giinthner as well as
to the editors of this volume for their comments on a previous version.
1. Transcription of the spoken extracts follows GAT-conventions (cf. Selting et
al. 1998); capital letters indicate stress positions. English translations are sim-
plified, particularly with respect to prosody and hesitation phenomena. In case
of conflict, less idiomatic versions have been chosen in order to give a better
impression of German syntactic structure.
2. The wenn-clause itself should be seen as adjoined to the resumptive particle,
i.e., as a co-constituent of the front-field; cf. Eisenberg (31994: 364f).
3. For a detailed discussion, see Metschkowa-Atanassowa 1983 and Zifonun et
al. 1997: 2280-2293.
4. A certain kind of ambiguity between a temporal and a conditional reading can
also be observed in the German question word wann (usually asking for tem-
poral information, 'at which time') which, when followed by a verb in the
subjunctive mood, often takes on a conditional meaning (wann wiirdest du
kommen = 'under which conditions would you come' or 'at which time would
you come'). Thanks to Bernd Kortmann for drawing my attention to this par-
allel.
5. For an analysis of these "concessive conditionals", cf. Konig 1985.
6. Among these wenn-clauses in the role of obligatory constituents, we may also
count comparisons using wie wenn and als wenn, as in: du kOmmsch dir
vielleicht vor wie wenn dir deine wErte verLORN gangn sin. ('maybe it seems
to you as ifyour values had been lost '). This usage of wenn will not be taken
into account in the following discussion, nor has it been included in the quan-
titative analysis.
A note in passing: some grammarians believe that wenn-clauses in comple-
ment function are obligatorily marked by a resumptive es (e.g. Eisenberg
31994: 365); this is not supported by my data, however.
7. Some 400/0 of the corpus are job interviews, mainly collected among north and
east German speakers, some 30% are therapeutic conversations, both in face-
to-face and in radio phone-in contexts, and the remaining 30% represent pri-
vate everyday conversations, partly on the telephone. In the latter two types of
data, southern German speakers prevail.
8. Embeddings of wenn-clauses into complex hypotactic constructions were not
counted as parenthetical.
9. The count excludes, in addition to all polyvalent cases, all factual conditionals
(Le. those expressing a positive epistemic stance), which are always non-
temporal, and all concessives, as well as wenn-clauses used as complements,
but includes counterfactual conditionals and reduced wenn-dann routines.
10. *Warum [wenn Du Kirschen magst] pfliickst Du Dir keine vom Baum? 'Why
[if you like cherries] don't you pick any from the tree?' therefore has to be
understood as parenthetical. Resumption by dann is also excluded here, Le.,
non-integration is the only option.
11. The term is used in a broader sense here than in Sweetser 1990. Details on this
construction may be found in Gfinthner 1999. Note that Sweetser's "epistemic
conditionals", although not "content conditionals", do not allow pre-front field
placement in German (* Wenn er sich jeden Tag voillaufen lasst, sie hat ihn
verlassen. 'If he gets drunk everyday, she has left him. ').
12. Cf. Konig and van der Auwera 1988: 128 ("assertive emphasis on a conse-
quent of a concessive allows non-integration"), Kopcke and Panther 1989:
700 ("high degree of ego involvement") and Gfinthner 1999 for details.
13. Cf., among others, Ford and Thompson 1986: 370; Ford 1993; Dancygier and
Sweetser 1996.
14. But see Ford 1993: 56.
15. Speaker I is most likely alluding to Zuckmayer's play (and a famous German
movie) Der Hauptmann von Kopenick, in which the Prussian state and army
authorities are caricatured.
16. This of course, leaves the question open why temporal adverbial clauses
should behave differently from conditional ones-a question which requires
an investigation of its own.
17. For a similar remark on English, cf. Ford and Thompson 1986: 369.
In some cases, however-though not in (26) with its oblique relative pro-
noun-the wenn-clause can follow the relative pronoun (das Essen ist wie ein
Teddybtir, der, wenn es hart ist, immer bei Ihnen ist, und der, wenn man ein-
sam ist, zum Festhalten da ist). But here we are dealing with parenthetical
placement in the middle field of the sentence; this is exceedingly rare in spo-
ken German.
19. Of course, wenn is not always stressed in fronted wenn-clauses. Cf. the fol-
lowing example:
ich (.) hab (-) FUNF jahre lang an der schule franzOsisch geHABTh, mir
jEhlts eigentlich an (.) PRAxis, .h aber: (-) ich bin aberZEUGT davon, =wenn
ich: eh eh OFters mal die geLEgenheit hatte zum beispiel in FRANKreich, eh
mich aufzuhaltn, .hh dass des: (-) eh SICherlich Ausbauftihig is.
'I had French at school for five years, actually I'm lacking practice, but I'm
convinced if I on occasion had the chance to spend some time for instance in
France, that I could work on it. '
20. The tendency to place the wenn-clause early in dependent constructions is also
evidenced by the fact that parenthetical placement immediately after dass is
202 Peter Auer
frequent (cf. Note 18). Often, a second, resumptive dass is added at the begin-
ning of the consequent:
kAnnst du ihm vielleicht (-) A USrichten dass ich ANgerufn habe?
und dass wenn er mit dem trelitz geSPROChen hat uber meine priifungk,
dass er sich dann irgenwie=mal=GANZ kurz bei mir mElden soli?
'could you perhaps tell him that I called?
and that if he has talked to Trelitz about my exams,
that he should give me a quick ring some time?'
21. The same was found in English conversations by Ford and Thompson 1986:
368.
22. In other-speaker produced post-positioned wenn-clauses this does not always
hold, since second speakers may choose this way of intimately linking their
speech to a preceding syntactic pattern but nonetheless produce unexpected
and even contradictory information under this 'disguise'.
23. More exactly, the corpus included the F.A.Z. Feuilleton-Glossen from Jan 8,
1993 (Ausgabe Nr. 6) to Dec 31, 1993 (Nr. 304) and DIE ZEIT politics sec-
tion of Dec 30, 1994 (No.1) to Feb 17, 1995 (Nr. 8). Transcriptions of con-
versational speech were regularized in order to make a comparative computer-
based word-count possible; see the appendix for an example.
24. The written corpus used by Ford and Thompson consisted of philosophical
essays, a professional text for automobile mechanics, and a personal narrative
account (1986: 355). The preference for pre-positioning held for all these
sources.
25. A separate count for the two newspapers shows that the quantitative results
are identical.
26. The resumptive particle is dann in 46 cases and so in 16 cases. Dann and so
are not freely interchangeable; in particular, so can be used with concessives
(introduced by auch/selbst wenn), while dann cannot. Cf.: Auch wenn das Ab-
geordnetenhaus erst im Herbst, voraussichtlich am 22. Oktober, gewtihlt
wird, so hat mit der Urabstimmung schon der Wahlkampf begonnen. ('Even
though parliament will not be elected until autumn, probably on October 22,
SO the electorial campaign has already begun with the strike ballot. ')
27. Of the 15 tokens in the spoken material, the three positions were about equally
distributed.
28. For a similar argument, cf. Thompson 1985.
References
Auer, Peter
1991 Das Ende deutscher Satze-Rechtsexpansionen im deutschen Ein-
fachsatz. Zeitschriftfur germanistische Linguistik 1991: 139-157.
1996 The pre-front field in spoken German and its relevance as a gram-
maticalization position. Pragmatics (Special issue, eds. Johannes
Wagner and Cecilia Ford), Vol. 6, No.3: 295-322.
Biber, Douglas
1986 Spoken and written textual dimensions in English: Resolving the
contradictory findings. Language 62: 384-416.
1999 Varieties of conditionals and their emergence in discourse. In: Aditi
Lahiri, Alexander Patschosvky and Christoph Schwarze (eds.), Is-
sues in interdisciplinary research on the lexicon = Working Paper
No. 99, University of Konstanz, Fachgruppe Sprachwissenschaft, 89-
130.
1996 Conditionals, distancing, and alternative spaces. In: Ann E. Goldberg
(ed.), Conceptual structure - Discourse and language, 83-98. Stan-
ford: CSLI Publications
Eisenberg, Peter
1994 GrundrifJ der deutschen Grammatik. Stuttgart: Metzler.
Fauconnier, Gilles
1985 Mental spaces. Cambridge: Cambridge University Press.
Fillmore, Charles J.
1990 Episternic stance and grammatical form in English conditional sen-
tences. In: Chicago Linguistic Society 26, University of Chicago,
137-162.
Ford, Cecilia E.
1993 Grammar in interaction. Adverbial clauses in American English
1997 Speaking conditionally: Some contexts for if-clauses in conversation.
In: Angeliki Athanasiadou (ed.), On conditionals again, 387-413.
Amsterdam: Benjamins,
Ford, Cecilia E. and Sandra A.Thompson
1986 Conditions in discourse: a text-based study from English. In: Eliza-
beth Traugott et al. (eds.), On conditionals, 353-372. Cambridge:
Greenberg, Joseph H.
1963 Some universals of grammar with particular reference to the order of
meaningful elements. In: Joseph H. Greenberg (ed.), Universals of
language, 73 -113. Cambridge, Mass. : MIT-Press.
Gtinthner, Susanne
1999 Wenn-Satze im Vor-Vorfeld: ihre Formen und Funktionen in der
gesprochenen Sprache. Deutsche Sprache 27(3): 209-235.
Haiman, John
1978 Conditionals are topics. Language 54: 564-89.
Konig, Ekkehard
1985 Where do concessives come from? On the development of conces-
sive connectives. In: Jacek Fisiak (ed.), Historical semantics, 263-
282. Berlin: de Gruyter.
204 Peter Auer
Konig, Ekkehard and Johan van der Auwera

1988 Clause integration in German and Dutch conditionals, concessive
conditionals, and concessives. In: John Haiman and Sandra A.
Thompson (eds.), Clause combining in grammar and discourse, 101-
133. Amsterdam: Benjamins.
Kopcke, Klaus-Michael and Klaus-Uwe Panther
1989 On correlations between word order and pragmatic function of con-
ditional sentences in German. Journal ofPragmatics 13(5): 685-711.
Lerner, Gene
1991 On the syntax of sentences-in-progress. Language in Society 20(3):
441-458.
Leska, Christel
1965 Vergleichende Untersuchungen zur Syntax gesprochener und ge-
schriebener deutscher Gegenwartssprache. Beitrage zur Geschichte
der Deutschen Sprache und Literatur 87: 427-464.
Metschkowa-Atanassowa, Sdrawa
1983 Temporale and konditionale "wenn "-Satze. Dusseldorf: Schwann.
Selting, Margret, Peter Auer, Birgit Barden, Jorg Bergmann, Elizabeth Couper-
Kuhlen, Susanne Gunthner, Christoph Meier, Uta Quasthoff, Peter
Schlobinski, Susanne Uhmann
1998 Gesprachsanalytisches Transkriptionssystem. Linguistische Berichte
34(173): 91-122.
Sweetser, Eve E.
1990 From etymology to pragmatics. Metaphorical and cultural aspects of
semantic structure. Cambridge: Cambridge University Press.
Thompson, Sandra
1985 Grammar and written discourse: initial vs. final purpose clauses in
English. Text 5: 55-84.
Zifonun, Gisela, Ludger Hoffmann and Bruno Strecker
1997 Grammatik der deutschen Sprache. Vol. III. Berlin: de Gruyter.
Counterfactual reasoning and desirability*
Noriko McCawley Akatsuka and Susan Strauss
This paper calls into question the popular view of counterfactual thinking
under the influence of formal logic, which considers the utterance of a
counterfactual sentence as an instance of complex and intricate reasoning
skills. Using naturally occurring data from English, Japanese, and Korean,
we will demonstrate that there exists a pattern of prototypical
counterfactual reasoning which appears to be natural and spontaneous to
every human being. Speakers/writers express a particular stance of
desirability versus undesirability toward a particular event, based on their
subjective evaluation of reality. Counterfactual conditionals are invoked as
a necessary step in this line of thinking.
1. Introduction l
This chapter represents a continuation of our inquiry into the most typical
usage of counterfactual conditionals in everyday life across language and
culture (Akatsuka 1997, 1999). We use the word "typical" here in the
sense of "natural and spontaneous," and thus underscore our claim that
human beings appeal to counterfactual reasoning in dealing with the
many aspects of everyday life that we find ourselves facing.
In the tradition of mathematics, philosophy and formal linguistics, the
conditional sentence has been regarded as the epitome of Man's rational
capacity, the height of Man's ability to reason logically. In the same vein,
it has long been argued that underlying the counterfactual conditional is a
similar, and perhaps even more logically complex type of reasoning,
engendering such analytical sentences as in (1), from Fauconnier (1985:
118).
(1)
A: If Napoleon had been the son ofAlexander, he would have won the
battle of Waterloo.
B: But he would have died long before that.
A: Well, suppose he lived a very long life, without ever ageing, or that
Alexander was resurrected in Corsica in the eighteenth century.
206 Noriko McCawley Akatsuka and Susan Strauss
Obviously, this string of utterances for Fauconnier represents a typical

instance of counterfactual conditionality. The utterance is clearly an
invented dialogue in which both participants realize the gross
counterfactuality (and absurdity) of the proposition 'if p'. And while the
type of reasoning displayed here could indeed be considered "complex,"
the view of counterfactual conditionality that we adopt actually has little
to do with this type of exercise in logic, truth values, and absurdity.
Based on the discourse data that we collected in American English,
Korean, and Japanese just following the 1994 Los Angeles earthquake
(hereafter, the earthquake data), we will show how and to what degree a
qualitative, cross-linguistic analysis can illuminate the role of the
speaker's evaluative stance of desirability. More specifically, we will
demonstrate the importance of the dichotomy of desirable/undesirable in
understanding the inherent ·relationship between counterfactual thinking
and everyday human existence.
The outline of the paper is as follows. In section 2, we will introduce
Fauconnier's (1985) "mental spaces" hypothesis as one current view on
counterfactuals. We will show first that while Fauconnier considers his
framework as "cognitive," it is actually inherently non-cognitive. We will
systematically demonstrate why this is so and why the framework fails to
account for usages of counterfactuals in everyday situations, by
introducing and then explicating a number of his invented examples in the
light of Akatsuka's desirability framework. In section 3, we will examine
spontaneous instances of counterfactual conditionals in the earthquake
data in the three languages (English, Korean, and Japanese) and will
contrast these with Fauconnier's constructed examples. We will also
present a discussion of the theoretical implications of this study. In
section 4, we will demonstrate that the present study sheds critical light
on the treatment of counterfactuals in Ford and Thompson (1986) and
Labov (1972). Section 5 is the conclusion.
2. Fauconnier (1985) and counterfactuality
2.1. Fauconnier (1985) and the tradition offormallogic
A close and careful examination of Fauconnier's (1985) analysis of

counterfactuality reveals that his approach is actually inseparable from the
tradition of formal logic. That is, the major concern of previous scholars
in philosophy and logic has centered on the logical problem of truth
conditions. Within that discipline, the primary aim was to specify under
what strict conditions a particular counterfactual statement can be viewed

as a case of 'true' or 'valid' reasoning (cf Jackson 1991; Fauconnier
1985; McCawley 1981). Example (2) excerpted from Fauconnier (1985:
109) encapsulates his approach to the phenomenon of counterfactuality.
As is clearly stated, rather than grappling with issues of logic and truth
conditions, the focus of analysis shifts to the process by which we
establish "counterfactual spaces"-an issue which Fauconnier labels a
"cognitive semantic question."
(2) Linguistically, we do not directly tackle the logical problem of truth

conditions for counterfactuals, but rather the cognitive semantic question
of how counterfactual spaces are set up and structured...Counterfactuality
is a case of forced incompatibility between spaces (emphasis added); a
space Ml is incompatible with another space M2 if some relation
explicitly specified in Ml is not satisfied for the corresponding elements
in M2 (1985:109).
Thus, he solves the truth conditional problem of logic by simply

proposing a hypothetical world. What is crucial here is that he
characterizes counterfactuality as a case of "forced incompatibility
between spaces." He does not, however, specify any other conditions
relating to the spaces M1 and M2, so that it is irrelevant which of the two
2
is the real world, R, and which is a hypothetical world, H. Further, in
spite of his labeling the problem as a "cognitive semantic" one, we
actually find no evidence of a true theoretical attempt to illuminate the
notion of counterfactuality from a cognitive semantic point of view. We
discuss this in detail in the next section.
2.2. "Space-builders"
In order to understand Fauconnier's view of counterfactuality, it is

perhaps best to first examine his theoretical construct "space-builders."
According to this framework, "space-builders" include various linguistic
elements such as conditionals (if ), verbs like wish, and negatives
such as not and prevent. The particular semantic properties of these
elements, according to Fauconnier, are supposed to be capable of setting
up a counterfactual mental space which functions in such a way that
'some relation,' which does not hold in the parent space, may be satisfied
in the newly created counterfactual space. In this light, consider the
utterances listed in (3a)-(3d), which appear in the original as examples
(3), (4), (5) and (6), respectively (1985: 109-110).
(3) a. If Lucky had won, I would be rich. I would have moved to Tahiti.
b. I wish Lucky had won. I would be rich.
c. Fortunately, the fire did not cross the highway. My house would
have been destroyed.
d. Luckily, the fire was prevented from crossing the highway. My
house would have been destroyed.
According to this framework, it is simply the linguistic elements not in

italics which are assumed to create the so-called counterfactual space-
that is, the conditional element if in (3 a), the verb wish in (3b), the
negation not in (3c), and the verb prevent in (3d).
However, there is one crucial observation missing here: what these
examples have in common is the fact that they are all subjective
statements. Utterances such as (3a) through (3d) can by no means be
construed as objective descriptions, whether invented by the researcher or
culled from spontaneous discourse data. In other words, the speaker of
(3a) and (3b) sounds unhappy or disappointed because what s/he was
hoping to happen did not actually happen. Conversely, the speaker of (3c)
and (3d) sounds happy and relieved because his/her home was not
destroyed by a particular fire which did not cross a particular highway.
In this light, Fauconnier's framework of mental spaces takes no
account whatsoever of an individual speaker's mental attitude, and thus
pays no special attention to the function of the adverbs fortunately and
luckily, which figure overtly in his own constructed examples (3 c) and
(3d). The function of these adverbials also accounts for analytic gaps in
examples (3a) and (3b). We will explicate this function in the next section
from the perspective of the concept of desirability.
2.3. Desirability and conditionals
We will begin this discussion by examining Fauconnier's analysis of (3c)

and (3d). According to Fauconnier, the lexical items not and prevent
represent the strongest type of lexical space-builder by virtue of the fact
that they deal in the domain of negation, and hence counterfactuality,
since semantically they are capable of reversing and/or negating some
expression of reality (i.e., factuality). Accordingly, lexical items such as
these are much stronger than the if and wish types, since the latter do not
always build counterfactual space. Fauconnier states:
(4) Clearly, counterfactuality may be lexically imposed, as it is by strong

negatives such as not and prevent in (5) [i.e., our (3c)] and (6) [i.e.,
our (3d)] (1985: Ill).
What is interesting here is the focused concentration on negation

markers as lexical agents for the construction of this so-called
counterfactual space. However, close examination of these sample
sentences reveals the force of other semantic factors beyond these
negators which playa significant role, if not a more significant role, in the
expression of counterfactuality.
To illustrate, let us begin by omitting the attitudinal adverbs
fortunately and luckily from Fauconnier's examples in (3c) and (3d). The
sentences so altered appear as (5a) and (5b) below:
(5) a. ??Thefire did not cross the highway. My house would have been
destroyed.
b. ?? The fire was prevented from crossing the highway. My house
would have been destroyed.
While the change consists simply of a seemingly minor lexical deletion,

the occurrence of the counterfactual in the second sentence is rendered
unnatural in both examples. The negation markers nonetheless are still
intact-a fact which now casts serious doubt on the claim that
counterfactuality is 'lexically imposed' by such elements as negatives.
We propose instead that the phenomenon of counterfactuality is
inherently related to the critical role of the speaker's stance of desirability.
And while we do not deny the importance of negation in this
phenomenon, we do not recognize negation as the single factor involved.
That is, negation comes into play in conjunction with, and only in
conjunction with, an expression of the speaker's stance of desirable vs.
undesirable.
This becomes much clearer when we compare Fauconnier's examples
with their Japanese counterparts. The English expression,
'Luckily/Fortunately + p (proposition) happened' corresponds most
closely to the Japanese sentence pattern, p-te yokatta. What this grammar
pattern expresses is precisely the speaker's attitude, desirable, towards
the past event/state of affairs described by the proposition, p. Observe
(6):
(6) p-te, yokatta.

[desirable} was lucky
'(I, it, etc.) was lucky that p.'
Now, let's compare Fauconnier's original example from (3c), repeated

below as (7), with the Japanese translation of the same utterance in (8).
(7) Fortunately, the fire didn't cross the highway. My house would
(8) Kaji ga dooro no mukoo ni moeutsura-nakute, yokatta.

Fire SM. highway over burn-cross-not fortunate
(Moshi) moeutsutte-itara, uchi ga yakete itadaroo.
IRREALIS burn-cross-if my house SM had burnt down-MDL
'I was lucky that the fire did not cross the highway. If it had, my
house would have been destroyed. '
The essential difference between the two examples is that the conditional
antecedent appears in the Japanese example, (8). That is, unlike English,
Japanese syntax does not normally tolerate monoclausal conditionals. And
once again, we see the overt representation of the speaker's attitude,
desirable, towards the past event/state of affairs described by the
proposition, p. In this case, p is synonymous with 'the fire did not cross
the highway' (=fact).
To illustrate in English, we repeat examples (Sa) and (5b) below.
(5) a. ??The fire did not cross the highway. My house would have
been destroyed
b. ??Thefire was preventedfrom crossing the highway. My house
would have been destroyed.
Note that when we supply an overt conditional antecedent in both

examples, the awkwardness disappears. Observe (Sa') and (5b'):
(5) a' The fire did not cross the highway. If it had, my house would
(5) b' The fire was prevented from crossing the highway. If it hadn't
been, my house would have been destroyed.
Now, (Sa') and (Sb') both contain the explicit conditional antecedent
'if not P,' and it is here that we can begin to understand what the speaker
is actually doing by invoking counterfactuality: she is presenting the
reason why she considers it desirable that p, i.e., 'the fire did not cross
the highway' (=fact), happened. The counterfactual situation, i.e., 'not p'
or 'the fire's crossing the highway,' would have led to the loss of her
house. We informally represent this line of reasoning, in (9):
(9) It was desirable that p (=fact) happened. If 'not p' (=counter to

fact) had happened, it would have led to undesirable consequences
(='not q' ).
Contrary to Fauconnier's claim, then, it is not the lexical items not and
prevent that build the so-called counterfactual space. Rather, we maintain
that it is the speaker who invokes counterfactual thinking as an integral
step in the line of reasoning, as delineated in (9). Thus, the use of
negation figures strongly in the expression of p vs. 'if not P,' as well as q
vs. 'not q,' or in other words, in the speaker's expression of stance of
desirable vs. undesirable, and not in the construction itself of the so-
called 'counterfactual space. '
Let us now tum to Fauconnier's analysis of (3a) and (3b), repeated
below as (lOa) and (lOb) for convenience.
(lOa) IfLucky had won, I would be rich. I would have moved to Tahiti.
(lOb) I wish Lucky had won. I would be rich.
According to Fauconnier's framework, it is the "space-builders" if and

wish that are responsible for the appearance of the counterfactuality in
these examples. Now, compare the same examples, with the first
sentence of each altered to express the fact p that Lucky (a racehorse) did
not win a particular competition, and the second sentence altered to
include an overt conditional antecedent expressing 'if not p,' as noted in
(lOa)' and (lOb)':
(lOa)' Unfortunately, Lucky did not win. If he had won, I would be rich.
IfI were rich, I would have moved to Tahiti.
(lOb)' Unfortunately, Lucky did not win. I wish he had won. If he had
won, I would be rich.
We claim that an adequate theory of counterfactuality must explain on

a principled basis that the workings of counterfactual reasoning in
Fauconnier's examples (lOa) and (lOb) are exactly the same as those in
(lOa)' and (lOb)'. According to our analysis, the line of reasoning
involved here will be illustrated informally as in (11).
(11) It was undesirable that p (=fact) happened. If 'not p' (=counter-to-

fact) had happened, it would have led to desirable consequences
('not q").
It will be obvious that (9) and (11) represent the identical pattern of
reasoning, with the sole difference being the speaker's desirability
assessment taking opposite values. The line of reasoning delineated in (9)
and (11) will be referred to henceforth in this study as "prototypical
counterfactual reasoning."
We have demonstrated that Fauconnier's framework is an i1mpersonal
one, not unlike those proposed by scholars from the tradition of formal
logic and mathematics; the view is necessarily limited to an incomplete
observation of the distribution of counterfactuals. Crucially, the
framework fails to account for all four examples (i.e., (3a)-(3d» as
squarely representative of prototypical counterfactual reasoning.
Unwittingly, too, even example (1), based on an invented dialogue
positing absurd conditions ('if Napoleon had been Alexander's son'),
expresses a strong stance of desirability ('he would have won the Battle
of Waterloo'). And what Frenchman would find this anything but
desirable?
We now tum our discussion away from constructed examples and
examine authentic discourse excerpted from the earthquake data in
English, Japanese, and Korean, where "prototypical" counterfactuals
emerge spontaneously.
3. A qualitative analysis of the earthquake data3
On January 17, 1994, at approximately 4:30 a.m., a strong earthquake hit

Los Angeles, significantly startling its residents. January 17 happened to
be a Monday, and that Monday happened also to be a national holiday
(Martin Luther King's birthday). Because of this, many people had
planned not to go to work that morning. Commuter traffic was
exceptionally light, and many people were still asleep when the
earthquake hit. Because of these and other factors, the casualties were
4
rather light, given the magnitude of the quake .
Approximately three weeks following the incident, researchers from
UCLA collected 'earthquake stories' from dyads of native speakers of
English, Japanese, and Korean. Subjects were predominantly UCLA
graduate students, though some undergraduates also participated. In all
cases, we attempted to ensure that the dyad participants did not know
each other previously, though here too some of the Japanese data reveals
that a handful of subjects did know their interlocutor to some extent. For
the most part, however, subject pairs were intended to be strangers.
The only instructions given to the subjects were to tell each other
about the experiences they had had during the earthquake, and that the
data recording session would last for approximately 20-25 minutes. Other
than these criteria, no explicit instructions were given. The interactions
were both audio and video taped and were transcribed using the
conventions of Conversation Analysis (Atkinson and Heritage 1984).
What is remarkable is that counterfactual conditionals emerged in all
three languages, and in a number of cases we find significant overlap in
terms of the types of issues captured with counterfactuals (e. g., 'if it had
happened during rush hour,' 'if it had happened in a place other than
LA,' 'if it had happened before the recent earthquake retrofitting of some
of the campus buildings,' etc.). None of the conversational dyads was
aware of what the other dyads had said in other recording sessions.
Further, because the study was designed solely to elicit spontaneous talk
in the respective languages, no mention of counterfactuals was ever made
at the inception. It was not until we transcribed and analyzed the data that
we discovered this parallelism.
In the discussion that follows, we limit ourselves to one representative
sample from each language. In all three examples, the subjects
systematically express their reactions to the relatively light damage and
small number of injuries and deaths, in the light of the strength of the
earthquake itself
Interestingly, the abstract line of reasoning of each case is quite similar
to what we have noticed in Fauconnier's examples (3c) and (3d), even
though Fauconnier's examples are monologic and constructed, while our
data derive from spontaneous, natural dialogue. Instead of the initial
occurrence of such attitudinal adverbs as fortunately and luckily as we
witnessed in the Fauconnier examples, we observe here the speakers
exchanging and building on the expression "we were lucky", in all three
languages. We will call this activity 'the opening.' Just following this
sequence emerges the counterfactual utterance, 'if not p' had happened,
'not q,' would have happened in all three languages. Henceforth, we will
adopt the formula 'if not p, then not q' to represent counterfactual
conditionality.
Let us now examine each excerpt, starting with English.
3.1. English: "Actually it's good that this happened now"
The speakers, T and H, were just talking about the fact that friends living
in the dorms said that the building seemed to sway at the time of the
earthquake; T then adds that some of the UCLA dormitories had recently
undergone seismic retrofitting: "actually, it's good that this happened now
because..." In other words, "we were lucky," hence the opening
sequence.
Observe, too, that after this opening T immediately appeals to
counterfactuals. What is of particular importance here is T's initial clause
"I jus' kept thinking," which precedes the counterfactual construction,
underscoring just how naturally and just how spontaneously we appeal to
counterfactual reasoning.
(12) English:
«participants: T=Female, H=Male))
(a) Opening
T1 Actually it's good that this happened now because (then)

these dorms were just respir- retrofitted () fer- () y'know ()
earthquakes () two to three years ago.
(b) 'if not p, not q'
==> T2: I jus' kept thinking if this earthquake had been in

==> the middle ofthe afternoon on a day that wasn't a
==> holiday how ba: :d this would have been, like the San
==> Francisco quake. 1- That was at five 0' clock in the ==>
afternoon on a Friday "hh or something like that It ==>
was just. y'know I think it-I think there would have ==>
been so many more lives lost and this would have been
==> a rea: : I tragedy.
In the stretch of talk in section (b) 'if not p, not q' we witness the
gradual emergence of a complex set of thoughts, all of which express
some counter-to-fact notion with respect to some aspect of the reality of
this earthquake. That is, first T proposes a time and a day different from
the actual time and actual day of the incident p and then indicates that the
consequences q would have been much worse (or undesirable), a line of
reasoning we encountered earlier in example (9), repeated below as (13):
(13) It was desirable that p (=fact) happened. If 'not p' (=counter-to-

(='not q').
Crucially, the speaker attitude underlying this utterance surfaces in the

linguistic manifestation of how T expresses the degree of potential
undesirability of the consequences ('not q'), had the situation played out
under the, counterfactual conditions she just established (i.e., 'different
time, different day'): her utterance is structured with an exclamatory
degree of intensity marker how, coupled with an ostensible sound stretch
in the negative adjective ba:d (I jus' kept thinking how ba::d this
would've been).
But she doesri't stop there. We actually witness her imagination
emerge linguistically, as she now upgrades the potential counterfactual
consequences by not only changing the time of the occurrence, but now
the place, i.e., San Francisco, where in 1989 a comparable earthquake
occurred during the height of rush hour traffic. And once again, the
pattern of reasoning is identical to that in (13).
The significance of this upgrade is twofold: first, it establishes a
recognizable frame of reference for this new level of negative
consequences by recalling an actual incident where those very same
conditions held (i.e., different time of day, different day of the week, and
now a different place); it is a well-known fact that many more people
suffered and died in the San Francisco quake. Second, it underscores all
the more the crucial relationship between desirability stance and
counterfactual thinking by again linguistically matching the upgraded
scenario with an equally upgraded utterance: there would have been so
many more lives lost and this would have been a rea:: I tragedy, clearly
more semantically intense than the initial 'how bad this would've been,
and notice too the audible sound stretch on the intensifier in a rea:: I
tragedy.
3.2. Co-construction in Korean and Japanese
We have just observed an example from the English database where a

single speaker imagines a scenario different in specific respects from the
actual one, in order to emphasize her current stance of desirability toward
the actual incident.
In the Korean and Japanese examples that follow, we note a similarity
in this type of complex thinking; only here, the imagination belongs to
both parties as they collaboratively participate in picking out specific
aspects of the actual incident which could have been different, and in
jointly constructing the potential undesirable circumstances.
3.2.1. Korean co-construction
First we present the Korean extracts to illustrate the co-construction of

counterfactual reasoning. Example (14) below is framed much in the
same way as its English counterpart in (12). That is, it begins with an
assessment in (a) concerning the fortunate fact that the earthquake
happened very early in the morning, followed by the utterance of a
counterfactual scenario in (b) where the participants underscore just how
lucky they were. There, we will observe that the counterfactual
conditional 'if not p, not q' is collaboratively constructed by the two
speakers.
H opens this sequence with the Korean expression 'be lucky' tahayng
ita, and also notably ends the sentence with the same expression. He
continues his turn with mannyak-ey, 'if foreshadowing that some
conditional utterance will most likely follow. And T provides an
agreement token, kurehcho 'that's right. '
(14) Korean:
«participants: H=Male, T=Male))
(a) Opening
==> HI: tahayng-i-n kes-un cihca saypyek-ey cicin

lucky-COP-ATTR thing-TM dawn at earthquake
tn-ssunikka tahayng-i-cyo
occur-CONN lucky-COP-TAG
'The lucky thing is that the earthquake happened at
dawn, that was lucky, huh. '
H2: mannyak-ey..
if (lit. in a one in ten thousand chance)
'if
==> TI: kureh-cho, ku..

be like-TAG FLLR
'That's true.'
Directly following this assessment sequence, T then builds upon H's

foreshadowed conditional, by introducing a counterfactual element
concerning the time that the earthquake occurred: 'traffic' manha-ssu-I -
ttaykena, 'if it had been at a time when there was a lot of traffic,' in other
words 'if not p.'
H produces the consequent clause 'not q' koyngcanghi manhun
salam-tul-i ama, 'my, a lot of people would probably have died,' which
itself is an emotionally charged statement, underscored all the more by
the exclamatory marker ama '(oh) my. '
(b) 'if not p, not q'
==> T2: ce 'traffic' manha-ssu-I -ttaykena xxx ... freeway'

FLLR traffic be a lot-PST-ATTR time freeway
'ifit had been at a time when there was a lot of traffic

or...the freeway."
==> H3: koyngcanghi manhun salam-tul-i ama

rather many-ATTR people-PL EXCL
cwuke-ss- keyss-cho
die-PST-MDL-MDL
'my, a lot of people would probably have died.'
The Korean example represents a very neat case where one speaker
produces the conditional antecedent, 'if not p' clause and the second
speaker the consequent, 'not q.' And the interactional dynamic which
engenders such a collaborative construction is an interesting one.
Even more revealing, however, is the Japanese example, where we
find the most elaborate instance of collaboratively produced
counterfactual reasoning of the three excerpts under investigation.
3.2.2. Japanese co-construction
In the stretch of talk that follows, we observe an opening interactional

pattern much like that in example (14), where one participant, in this
case, H, expresses the assessment that it was "lucky" the earthquake
happened so early in the morning. This is responded to in the very next
line by A with an expression indicating a stance of agreement, "I thought
about that, too."
What is interesting here is that in both the Korean and the Japanese
examples, the participant who utters the assessment does so in such a
way that co-participation is invited from the outset through the use of
interactional particles. In the case of the Korean utterance, the speaker
ends the assessment with -ci, functioning at times like a TAG-like
particle. Similarly, in the Japanese example below, we find the use of nee,
which among other functions, also serves as a TAG-like marker. And, in
each case, their interlocutors do provide both an appropriate and relevant
response, i. e., utterances providing agreement.
(15) Japanese:
«participants: H=Male, A=Female))
(a) Opening
==> HI : demo nee asa de yokatta desu yo nee

but PRT morning in lucky-PST COP PRT PRT
gozen yo)i han de?
a.m. four 0' clock halfpast at
'But, see, we were lucky it happened in the morning.

at four thirty, right?'
==> AI: atashi mo omotta-n desu yo.

1 too think-PST-PLN-NOML COP PRT
'I thought about that, too.'
In the lengthy collaboratively built sequence that follows, H expands

his own initial utterance and begins to weave a "complex" mesh of
counterfactual thought with his interlocutor, first by altering the time of
the earthquake, from the early morning hours (4:00 a.m.) to the current
time, i.e., the approximate time of recording (4:00 p.m.), and then by
altering the place in conjunction with the changed time ('if it had
happened in the library at a time when students were in it'). And finally,
the relevance of an altered time emerges in the talk, referring to the
freeway overpass which collapsed as a result of the quake, beginning at
line A3, and resumed again from H7 through the end of the excerpt.
(b) 'ifnotp,notq'
==> H2: moshi nee kore ga nee imagoro=

ifPRT this 8M PRT now around
'If (it had happened) around now [i.e., about 4 p.m.],'
==> A2: =kore ga nee, yuugata no:

this 8M PRT evening PRT
'(if it had happened) in the evening,'
H3: ima toka ne

now or something PRT
'like now or something. '
==> A3: rasshuji no go-ji toka ne ha ha ha ha

rush hour GEN five o'clock or something PRT
'at rush hour like five 0' clock ha ha ha ha ha'
H4: soo SOD SOD yuugata no
right right right evening GEN
rasshuji no goji toka
rush hour GEN five 0' clock or something
'Right, right, right at rush hour, at like five o'clock

in the evening. '
A4: ha haha haha ha

'ha haha haha ha'
==> H5: anD toshokan nanka de mada gakusei nanka

FLLR library or something LOC still students something
ir<[irujikan dattara:
be at time COP-COND
'Well, like at the library, if it had happened at a time when

students were still there. '
A5: [iru san yoji toka:

be at three four 0' clock or something
'like three or four o'clock or,'
H6: moo atama kara saa hon ga ochitekite:

EXCL head from PRT book SM fall-come-GER
'well, the books would have fallen on their heads and,'
A6: aa zaa - sugokatta desu nee.

EXCL ONM terrible-PST COP PRT
'Oh, like 'CRASH'-it would've been terrible.'
H7: haiuei furiiuei nanka mo zenbu nee

highway freeway something too all PRT
'the highways, freeways, all of them,'
A7: zenbu
all
'all of them'
==> H8: sundanshichattara moo

be cut- COMPLTV-COND EXCL
'If they had (all) been cut'
==> A8: kanbotsushichatte:

collapse-COMPLTV-GER
'they would have collapsed and'
==> H9: shisha nanka...juubai gurai

dead people or something ten times about
n n natta deshoo nee?
LOC become-PST MDL TAG
'The death toll would've been ten times higher, right?'
(c) Closing
==> A9: nn atashi mo omotta-n su asa de yokatta-n daroo naa tte

I too think-PST-PRT moming-INST good-PST-PRT-MDL-PRT
'So I thought too that we were lucky it happened in the

morning.'
HIO: nnn
'yeah'
AIO: minna neteru toki d~tta kara

everyone sleep-te PROG time COP-PST-PLN because
'Everybody was sleeping at that time, so'
HII: honto hukoo chuu no saiwai desu yo.

really disaster in GEN blessing COP PRT
'Right, it was the only blessing within this disaster. '
All: nn=
'yeah.'
HI2: =nn nn
'yeah.'
In the first line of this sequence, H2, foreshadows his conditional

thought with the word moshi 'if,' and then sets the counterfactual hour to
be 'around now' imagoro. In response, A builds on H's utterance by
repeating it almost verbatim, with the exception of one change: rather
than recycling imagoro, a temporal deictic term whose meaning can only
be recovered from the situated context, she substitutes the time reference
with yuugata, thus slightly upgrading the level of specificity of the
counterfactual time in question. And she substitutes and upgrades the
time reference yet once more in A3, by specifying even more detail in this
counterfactual switch: rashuji no go-ji toka '(if it were) at rush hour, like
5:00 p.m. '-a piece of information which will become supremely
relevant later in the talk.
From H2 through A4, the speakers have created a counterfactual
scenario in which the earthquake could have occurred at a time other than
at the actual time that it did. At H5, they carry this one step further by
now zooming in on a specific location which could have suffered serious
effects, had that quake indeed happened at the altered hour. What is
crucial here, as well, is the fact that this specific location is one which
both purportedly know very well-the library, ('well [suppose this
happened] like at the library, if it had happened at a time when students
were still there'). Note too, that H is the one who proffers this new
location (H5), and in mid-utterance, as he struggles through the verb iru

('to be-at a location'), A actually begins her tum, by supplying the verb
iru as well as the temporal specification of 3 or 4 0' clock.
And their collaborative imagining continues, as they jointly create a
scene, complete with vivid visuals, supplied by H (moo atama kara saa
hon ga ochitekite 'books would have fallen on their heads'), and
onomatopoetic expressions (aa zaa 'crash!'), supplied by A.
Observe that this very description represents the first instance of a
consequent clause in this entire sequence, and that this consequent clause,
just like its antecedent, is collaboratively produced by both participants,
ending with the assessment of sugokatta desu ne 'it would have been
terrible,' at A6.
At H7, H continues to build on the counterfactual scenario to include
all freeways and highways, thereby creating a much more intense and
dangerous situation than the actual one, zenbunee sundanshichattaramoo
'if all freeways and highways had been cut (like the one that actually was
cut'). A in tum concurs with the first half of the undesirable consequent
kanbotsushichatte: 'they would have collapsed, and', which H completes
by providing the actual upshot of this entire conversational sequence and
line of reasoning, shisha nanka juubai gurai n natta deshoo nee, 'the
death toll would have been ten times higher, right?'
Crucially, as soon as the above conclusion is jointly reached, the
speakers go back to where they left the conversation in the opening. That
is, they again congratulate themselves by saying asa de yokatta 'we were
lucky it was in the morning. '
3.3. Summary
To sum up, in the excerpts from all three languages, we find the
consistent parallel whereby speakers invoke counterfactual thinking to
express the pattern of reasoning originally introduced as (9) and (13), and
repeated below as (16):
(16) It was desirable that p (=fact) happened. If 'not p' (=counter-to-

(='not q').
Every instance of counterfactual conditional is invoked to highlight the

speaker's appreciative attitude towards the actual incident. That is, in
each language, we find that speakers have created a domain of
imagination in which they alter one or more specific aspects of a given

reality (either on their own, as in the English example, or in collaboration
with their interlocutor, as in the Korean and Japanese examples). In all
three cases, we have witnessed how speakers shifted coordinates of time,
coordinates of place, or both, and then viewed the situations from a
counterfactual perspective which enabled them all to appreciate just how
lucky they were that the damages caused by such a large earthquake were
so mmor.
3.4. Discussion
Why are there such significant overlaps in the contents of counterfactual

conditionals uttered by the American, Korean, and Japanese speakers
when they recall the occurrence of the earthquake from the point of view
of the common experience of the community? In the discussion that
follows, we suggest that this seemingly simple question is the key which
will shed light on the cognitively dynamic aspects of counterfactual
reasomng.
3.4.1. The predictability of 'not p'
In response to the question above, we propose that the counterfactual

utterances by the American, Korean, and Japanese speakers overlap in
the way they do for the reason that the contents of 'not p' are highly
predictable. The speakers in all three languages, in producing
counterfactual utterances, reflect their shared common sense about the
earthquake. In other words, irrespective of language and culture,
residents of metropolitan cities like Los Angeles share the view that if an
earthquake were to occur early in the morning while most residents are
still asleep, casualties would be relatively small, even if freeways and
large public buildings were to suffer heavy structural damage, precisely
because few people would have been present in those areas.
Thinking this way, we realize that the processes of imagination we
have observed in the cross-linguistic data are not really "complex" mental
activities. This is the reason why the two Japanese speakers can enjoy the
lengthy collaborative construction of counterfactual reasoning, as
illustrated by the laughter tokens ha ha ha ha ha (lines A3 and A4) and
their enthusiastic endorsement of the other party's comments soo SOD SOD
'right right right' (H4). Note that they are not co-reasoning to create new
ideas. There is absolutely no exchange of new ideas going on here.

Instead, in these extended stretches of talk the speakers are jointly
repeating what they have been repeating to themselves and to others and
what they have heard repeated on so many occasions since the earthquake
first hit five weeks earlier. And by repeating what they know and what
basically everybody else already knows, they are congratulating
themselves by conveying the message, "We are really lucky. It could've
been so much worse."
3.4.2. The speaker's belief: "if 'not p' actually happens, then terrible
consequences will result"
In sharp contrast to Fauconnier's 'absurd' conditions in example (1) ('if

Napoleon had been the son of Alexander'), all instances in the earthquake
data of 'not p' of "if 'not p' had happened" actually represent a variety of
situations which could have indeed happened and could have resulted in
much greater casualties. Why is this the case? We suggest that when the
speaker utters "if 'not p' had happened," this utterance reflects the
speaker's belief that if 'not p' does indeed happen, then terrible
consequences will result. In other words, each instance of counterfactual
reasoning in the earthquake data is endorsed by the same speaker's belief
that "if 'not p' actually happens, then 'not q' (=terrible consequences)
will result." What we observe here is the co-existence of conditional
judgment and counterfactual judgment in the consciousness of the
speaker-which explains why the earthquake data contains so many
instances of expressions of great relief
3.4.3. The 'causal link' between p and q5
In order to explain the co-existence of conditional reasoning and

counterfactual reasoning in the consciousness of the speaker that is
observable in the earthquake data, we will need a dynamic theory of
conditionals which postulates that the connection between p and q is an
integral part of the grammatical meaning of natural language conditionals.
Logicians have long noted that normally there is some kind of connection
between p and q. It was partly due to the difficulty in pinpointing the
exact nature of this connection in the framework of formal logic that they
generally concluded, regardless of their personal stand on the analysis of
~f, that this connection should be treated as a problem of pragmatics
rather than grammar (e.g., Quine 1950, Grice 1975, Stalnaker 1975). The
position of these logicians has been more or less inherited by many
linguists (e.g., Geis and Zwicky 1972, Haiman 1978, Karttunen and
peters 1979, McCawley 1981). It is important to note that Fauconnier
(1985: 113-115) explicitly rejects the idea that there is an inherent
connection between p and q of both indicative conditionals and
counterfactual conditionals. Fauconnier even takes the position that the
counterfactuality of q in counterfactual conditionals is implicature and
therefore cancelable. Basically, then, Fauconnier (1985) treats natural
language conditionals very much like mathematical conditionals. Note
that in the case of mathematical conditionals, there is no connection
between p and q of 'p ::> q'; they are independent propositions. Further,
there is no sequential relationship between the two, either. However,
McCawley (1981) observed that in all English conditionals, p is
temporally and/or causally or epistemologically prior to q. We now come
to the conclusion that it is far from being an accident that Fauconnier's
mental space theory has failed to account for the prototypical
counterfactual reasoning phenomenon in our discussion in Section 2.
4. More on aspects ofprototypical counterfactual reasoning
4.1. Ford and Thompson (1986): 'contrast'
It will be illuminating here to examine Ford and Thompson's (1986)

analysis of conditionals in the light of the present discussion. Ford and
Thompson (1986) is a pioneering study of discourse functions of the if-
clause as a link between the preceding context and the following
discourse. Based on their quantitative analyses, they have identified four
basic discourse functions shared by their written and spoken data, one of
which is to present a contrast with the preceding context. This view is
shown by the formula in (17) below. What is relevant to our study is that
(18) is provided as a clear case illustrating that such a contrast can be
indicated by means of a counterfactual. (18) is an excerpt from a personal
narrative about the behavior of a chimpanzee by the name of Nim
Chimpsky, trained to use American Sign Language. (17) and (18) appear
in the original as examples (6) and (8), respectively (1986: 359).
(17) X. (But) if not X, then Y.

(18) Nim's aggression increased mainly because of the necessity of

introducing more and more teachers into his life... 1f it had been
possible for him to have grown up with a small and stable group of
caretakers, he would have experienced fewer separations from his
trusted caretakers and had far fewer opportunities to test his dominance
through aggression. (Terrace 1979: 145).
According to our analysis, what underlies (18) is the same line of

reasoning as delineated in (19) below. This reasoning is identical to that
expressed earlier in (16), with the exception that the speaker's
desirability stance has taken the opposite value. (19) was previously
introduced as (11), and together with (16) these expressions represent
instances of 'prototypical counterfactual reasoning.'
(19) It was undesirable that p (=fact) happened. If 'not p' (=counter-to-

fact) had happened, it would have led to desirable consequences
('not q").
The writer of (18) has a warm and sympathetic understanding of the

poor chimp's aggressive behaviors resulting from his unfortunate
upbringing. This has been made unmistakably clear by his appeal to
counterfactual reasoning. We agree with Ford and Thompson that
'contrast' is indeed a useful notion in understanding the discourse
function of the counterfactual conditional; however, we propose that what
is contrasted in (18), is not 'if not X' and its preceding context X, but
rather the undesirable reality 'caused by X' and the desirable situation Y
which could have taken place.
4.1.1. Clancy, Akatsuka, Strauss (1997)
Strong empirical support of our re-analysis of Ford and Thompson's

formula comes from our recent cross-linguistic research project (Clancy,
Akatsuka, and Strauss 1997). Based on a total of 84 hours of spontaneous
discourse data, we examined how American, Japanese, and Korean
parents use conditionals when talking to children less than three years of
age. Since the subjects were so young, these interactions were
particularly rich in such adult-produced speech acts as commands,
prohibitions, and granting permission. We discovered that
overwhelmingly the most frequent type of conditional uttered by the
caregivers was what we called "predictive" conditionals, which predict
that if one does an undesirable thing, then undesirable consequences will

result (American English 63%, Japanese 74%, Korean 80%).
Observe the following example where an American mother is talking
to her two-year-old child who wants Santa Claus to bring her two dolls
for Christmas:
(20) Mother: You have to be a good girl. If you are not a good girl,
Santa Claus won't bring you anything.
Throughout the data, the caregivers often gave warnings, criticisms, and
advice to their children, constituting a particular type of persuasive
speech act. Typically, this pattern of persuasion takes the following form:
An initial utterance expressing "do X" or "don't do X" is followed by the
provision of a reason, couched in a conditional statement. In other words,
the message takes the general shape of "if children don't listen to what
adults say, undesirable consequences will result." What is being
contrasted here are desirable consequences versus undesirable
consequences. We can informally represent this line of reasoning involved
here as (21):
(21) It is desirable that p will happen. If 'not p' happens, it will lead to
undesirable consequences.
There is an obvious parallelism between prototypical counterfactual

reasoning and the pattern of reasoning delineated in (21). From this, it
will be clear to the reader that the speaker's stance of desirability plays a
critical role in understanding the many usages of conditional utterances in
our daily lives, not just the role it plays in prototypical counterfactual
reasoning alone.
4.2. The discourse domain: The case ofa book review
Because the prototypical counterfactual serves to underscore the

speaker's stance of desirability or undesirability toward a particular
consequence, the discourse domain in which this kind of reasoning
manifests itself can be quite broad. In order to illustrate this point, let us
look at a one-page book review which appeared in the "Dividing Line"
column of a recent issue of Time magazine (October 4, 1999: 90). The
book is entitled Gone Boy and was written by a father whose 15 year old
son was killed by a fellow student in a college campus shooting incident.
The reviewer, Jack E. White, is a long-time friend of the author~ the

review is directly related to White's stance on gun control.
The column heading reads "Elegy for a Gone Boy," in large, bolded
black letters, accompanied by the subtitle, "A father's searches for
meaning in his son's murder," which appears in much smaller font. It is
clear that the function of the heading and its subtitle is not simply to
provide an efficiently abridged abstract of the contents of the book-it is
an unmistakable indicator of the reviewer's mental attitude toward an
undesirable reality.
The review begins as a personal narrative, which appears in (22)
below:
(22) GREGORY BIGSON AND I MET 36 YEARS AGO AS FRESHMEN

AT Swathmore College. Greg was 15-the same age his elder son
Galen had reached in 1992 when he was slaughtered in an act of
senseless violence. Galen was in his second year at Simon's Rock
College in Great Barrington, Mass., when a fellow student named
Wayne Lo went berserk and shot up the campus with a cheap imported
rifle, killing Galen and a teacher and wounding four others.
White devotes a sizable amount of space to the body of the article,

which provides an account of how Galen's grieving father spent seven
long years in the investigation into the cause of his son's murder. He goes
on to report that the result is "a tapestry of shared pain and guilt in which
everyone, including the killer is in some sense a victim." The review ends
with the concluding remarks reproduced in (23) below. Notice that this
utterance is a textbook case of prototypical counterfactual reasoning, "if
'not p' had happened, it would have led to desirable consequences."
(23) None of this would have happened if Wayne Lo, at the age of 18,
had not been able to walk into a gun store, flash his driver's license
and $129 and walk out with a deadly weapon. Or if he had not been
able to have 200 bullets sent to him at Simon's Rock College by a
mail-order arms company. To my friend, Greg, there is a
straightforward conclusion to be drawn from the mystery of Galen's
death. "We've just got too many guns in this country. We've got to get
rid of them". Anyone who reads Gone Boy will find it hard to disagree.
By appealing to counterfactual reasoning, the reviewer is providing the

reader with detailed and specific new information about just how easily
Wayne Lo was able to buy a gun, have the ammunition sent to him at
school, and actually carry out this fatal shooting. The bottom line here is
that there are indeed too many guns in America.
Thus, counterfactuals do not just describe unrealized possibilities; they

can and do eloquently report what actually happens in the real world.
4.3. Labov (1972): 'comparators'
This section will serve as a preview of our next project, which involves a
re-examination of Labov's (1972) seminal study of narratives in the light
of our prototypical counterfactual reasoning hypothesis. Specifically, we
are interested in Labov's treatment of counterfactuals in relation to his
key notions of "evaluation" and "comparators". Simply stated, Labov's
"evaluation" is a syntactic unit with which the narrator shows why a
particular event is a reportable experience; "comparators" are
grammatical devices that the narrator utilizes to evaluate the narrative.
According to the original framework, neither notion is related to the
speaker's evaluative stance of desirability. Consider the following excerpt
(1972: 387) (emphasis added), where Labov cites one of the most
dramatic danger-of-death stories to illustrate the notion of what
constitutes a comparator:
(24) One of the most dramatic danger-of-death stories was told by a retired
postman on the Lower East Side: his brother had stabbed him in the
head with a knife. He concludes:
And the doctor just says, "Just about this much more", he says,
"and you'd've been dead".
Comparators then include negatives, futures, modals (emphasis added),
quasimodals, questions, imperatives, or-clauses, superlatives, and
comparatives, more or less in increasing order of syntactic complexity.
As stated here, Labov (1972) treats counterfactual utterances as

evaluating clauses because they utilize "modals".
We are impressed by Labov's perceptive observations. For example,
he continues the above discussion by saying that "A comparator moves
away from the line of narrative events to consider unrealized possibilities
and compare them with the events that did occur." Unfortunately,
however, the question of "what precisely is it that the speaker is
evaluating with those devices called comparators?" remains unaddressed.
Further, as we have just pointed out in the previous section,
counterfactuals do not just consider unrealized possibilities. Skillful
speakers/writers such as the narrator of this story know that they can
effectively report what actually happened by appealing to counterfactual
reasoning, and in so doing, they convey the message "I was really lucky."
Throughout this study, we have attempted to demonstrate that there exists

a pattern of "prototypical counterfactual reasoning," which appears to be
natural and spontaneous for every human being. We represent this pattern
informally as in (25) and (26).
(25) It is desirable that p happened. If 'not p' had happened, it would

have led to undesirable consequences (='not q').
(26) It is undesirable that p happened. If 'not p' had happened, it would

have led to desirable consequences (='not q').
These two represent the identical pattern of logic.

We have argued that prototypical counterfactual reasoning in the
earthquake data calls for a cognitively dynamic theory of natural language
conditionals which can account for the fact that there is a necessary,
simultaneous co-existence of conditional and counterfactual conditional
reasoning in the consciousness of the speaker. That is, each occurrence of
"if 'not p' had happened" in the cross-linguistic data entails the same
speaker's belief that if 'not p' actually happens, then 'not q' (=terrible
consequences) will result. This analysis explains why the earthquake data
is permeated by expressions of such a great sense of relief In this line of
thinking, we realize that the seemingly "complex" imagining processes
that we observed in the cross-linguistic data are not really "complex"
mental activities because 'not p' is highly predictable; it is a reflection of
the shared common sense of the residents of Los Angeles about the
earthquake and its damages, irrespective of language and culture.
We have further argued that prototypical counterfactual reasoning
calls for a theory of natural language which postulates that the antecedent
and consequent of both indicative and counterfactual conditionals are not
only sequentially ordered but they are 'causally' related to each other.
What emerges from the study of prototypical counterfactual reasoning,
then, is the understanding of the fundamental differences between the
grammar of the natural language conditional and that of the mathematical
conditional. Recall that in the case of the mathematical conditional, the
antecedent and consequent are independent propositions. Moreover, there
is no ordering relationship between them. Our study suggests strongly
that the popular view of counterfactual thinking as "complex" is a by-
product of theoretical frameworks which consider that the natural
Counter/actual reasoning and desirability 231
language conditional has a similar structure to that of the mathematical

conditional.
In conclusion, we hope to have demonstrated instead that
counterfactual reasoning as it occurs typically in everyday situations is
inextricably related to an expression of the speaker's stance toward the
desirability or undesirability of a particular outcome.
Abbreviations
ATTR ATTRIBUTIVE
COMPLTV COMPLETIVE
COND CONDITIONAL
CONN CONNECTIVE
COP COPULA
EXCL EXCLAMATION
FLLR FILLER
GEN GENITIVE
GER GERUNDIVE
:MDL MODAL
NML NOMINALIZER
ONM ONOMATOPOETIC
PLN PLAIN FORM
PRT PARTICLE
PST PAST
SM SUBJECT MARKER
TAG TAG MARKER
TM TOPIC MARKER
Notes
* We would like to dedicate this study to the memory of James D. McCawley.

If Jim had not written McCawley (1981) we would never have learned the
joy of discovering the logic of natural languages.
The authors have contributed equally to the production of this chapter and
are both equally responsible for its contents.
1. The authors are grateful to the two volume editors, Bernd Kortmann and
Elizabeth Couper-Kuhlen, and to Sandy Thompson for their insightful
comments and constructive criticisms, which greatly enhanced the quality of
this paper.
2. See Akatsuka (1985) for a full discussion of the epistemic scale where this
relationship is discussed in cognitive terms.
3. The Japanese data were collected and transcribed by Shoichi Iwasaki and his
research assistants, Eri Yoshida and Yumiko Kawanishi. Data collection and
transcription for the Japanese data appearing here were partially supported
by the University of California's Pacific Rim Study Grant (1992-94). The
English and Korean used in this study data were collected and transcribed
independently from the Iwasaki data.
4. The magnitude of the Northridge earthquake was 6.6. In all, 51 people lost
their lives and over 5,500 people were injured. For the sake of comparison,
the earthquake in Kobe, Japan, which occurred exactly one year later at
approximately the same time of the morning, measured 7.2 on the Richter
scale~ the number of fatalities in the Kobe earthquake exceeded 5,000.
5. See Akatsuka (1986) for an earlier discussion of this point based on
introspective data
References
Akatsuka, Noriko
1985 Conditionals and epistemic scale. Language 61: 625-639.
1986 Conditionals are discourse-bound. In: Elizabeth Closs Traugott,
Alice ter Meulen, Judy Snitzer Reilly, Charles A. Ferguson (eds.),
On conditionals, 333-352. Cambridge: Cambridge University Press.
1991 Dracula conditional and discourse. In: C. Georgopoulos and R.
Ishihara (eds.), Interdisciplinary approaches to language: Essays in
honor of S.-Y. Kuroda, 25-37. Dordrecht/BostonILondon: Kluwer
Academic Publishers.
1997 On the co-construction of counterfactual reasoning. Journal of
1999 Towards a theory of desirability in conditional reasoning. In: Akio
Kamio and Ken-ichi Takami (eds.), Function and Structure, 195-
213. AmsterdamlPhiladelphia: Benjamins.
Atkinson, J. Maxwell and John Heritage (eds.)
1984 Structures of social action: Studies in conversation analysis.
Clancy, Patricia M., Noriko Akatsuka and Susan Strauss
1997 Deontic modality and conditionality in adult-child discourse: a
cross-linguistic study. In: A. Kamio (ed.), Directions in functional
linguistics, 19-57. Amsterdam: Benjarnins.
Fauconnier, Giles
1985 Mental spaces. Cambridge, MA: MIT Press.
Ford, Cecilia E. and Sandra A. Thompson

1986 Conditionals in discourse: A text-based study from English. In:
Elizabeth Closs Traugott, Alice ter Meulen, Judy Snitzer Reilly,
Charles A. Ferguson (eds.), On conditionals, 353-372. Cambridge:
Cambridge University Press
Geis, Michael L. and Arnold M. Zwicky
1971 On invited inferences. Linguistic Inquiry 2: 561-566.
Grice, Paul H.
1975 Logic and conversation. In: Peter Cole and Jerrold L. Morgan
(eds.), Syntax and semantics 3: Speech acts, 41-58. New York:
Academic Press.
Haiman, John
Jackson, Frank (ed.)
1991 Conditionals: Oxford readings in philosophy. New York: Oxford
University Press.
Karttunen, Lauri and Stanley Peters
1979 Conversational implicatures. In: C.-K. Oh and D.A. Dinneen (eds.),
Syntax and semantics 11: Presupposition, 1-56. New York:
Academic Press.
Labov, William
1972 The transformation of experience in narrative syntax. In: William
Labov (ed.), Language in The Inner City: studies in the Black
English vernacular, 354-396. Philadelphia: University of
Pennsylvania Press.
McCawley, James D.
1981 Everything that linguists have always wanted to know about logic
**(but were ashamed to ask). Chicago: University of Chicago Press.
Quine, Willard Van Orman
1950 Methods oflogic. New York: Henry Holt and Co.
Sacks, Harvey, Emanuel Schegloff and Gail Jefferson
1974 A simplest systematics for the organization of turn-taking for
conversation. Language 50: 696-735.
Stalnaker, Robert
1968 A theory of conditionals. Reprinted in: William L. Harper, Robert
Stalnaker and Glenn Pearce (eds.) (1981), 1ft: conditionals, belief,
decision, chance, and time, 41-56. Dordrecht/Boston: Reidel/
Kluwer.
Terrace, Herbert
1979 Nim: A chimpanzee who learned sign language. New York:
Washington Square Press.
White, Jack E.
1999 "Elegy for a Gone Boy" Book review in Time magazine October 4,
1999.
TEXTS:
1994 Los Angeles earthquake data:
English data: Susan Strauss, 1994
Japanese data: Shoichi Iwasaki, 1994
Korean data: Susan Strauss and Yumiko Kawanishi
Adversative connectors on distinct levels of
discourse: A re-examination of Eve Sweetser's
three-level approach *
Ewald Lang
While basically acknowledging the discemability of three levels on which a

sentence is interpreted (content, epistemic and speech act level), the paper
tackles Sweetser's claim that the correct interpretation of interclausal
connections depends not on form, but on a pragmatically motivated choice.
I argue for a division of labour between grammar and pragmatics proper by
(i) showing that there are more level-specific structural differences at issue;
(ii) proving the notion "epistemic" to be feasible only if statements,
assumptions, and inferences are kept distinct; (iii) providing an account for
the unavailability of content-level interpretation for adversative
conjunctions.
1. Introduction
In Chapter 4 of her exciting book From Etymology to Pragmatics (1990),

Eve Sweetser offers a fresh view of the notorious diversity of
interpretations we observe with causal and adversative connectors. She
rejects the view that connectors like because, although or and and but are
lexically polysemous in the classical sense and suggests instead that the
extremely varied uses of these connectors should be explained in terms of
pragmatic ambiguity. Drawing on Sweetser (1990:76-78) the main tenets
of her approach can be summarized as in (I)-(ill):
(I) One meaning, several uses
In polysemy, a morpheme has several related semantic values; in

pragmatic ambiguity <... >, a single semantics is pragmatically
applied in different ways according to pragmatic context.
(1990:76)
236 Ewald Lang
(II) Three levels of interclausal connection ("conjunction"l)
... conjunction may be interpreted as applying in one of (at least)

three domains [where] the choice of a "correct" interpretation
depends not on form, but on a pragmatically motivated choice
between viewing the conjoined clauses as representing content
units, logical entities, or speech acts. (1990:78 ) [emphasis mine
-EL]
The domains of interpretation are illustrated by clear-cut examples

such as (la-c) below, all of which have in common that the second clause
asserts a proposition in the form of a statement.
(1) a. John came back because he loved her. [content or fact

level]
b. John loved her, because he came back. [epistemic level]
c. What are you doing tonight,
because there's a good movie on. [speech act level]
The levels are distinguished depending on what the clauses are taken
to represent and hence on what sort of entities the causal relation is taken
to obtain. Thus, in (la) it is two content units between which a factual
causal relation is asserted to obtain. More exactly: the content units are
two propositions whose couching in declarative sentences enables them
to render assertable statements. The linking pattern for a causal relation
on the content level thus amounts to STATEI\ffiNT because STATEI\ffiNT.
In (1 b), the first clause is not asserted as a statement but only as an
assumption which is inferred from what is stated as fact in the second
clause. The causal connector marks the second clause as sufficient
evidence justifying the assumption rendered by the first clause. The
linking pattern on the epistemic level is ASSUMPTION, because
STATEMENT (EVIDENCE). The hypothetical status of what is asserted by
the first conjunct can be made explicit by embedding it under modals -
cf (1 b-i, ii) or by rephrasing it as an inference - cf (1 b-iii):
(1) b-i. John must have loved her, because he came back.
b-ii. John undoubtedly loved her, because he came back.
b-iii. From the fact that he came back,
I conclude that John loved her.
Adversative connectors on distinct levels ofdiscourse 237
In (Ic), the first clause serves the purpose of a speech act of asking
whose performance is being justified by the statement rendered by the
second clause. The linking pattern thus is SPEECH ACT, because
STATE:MENT (JUSTIFICATION). The because-clause does not form part
of the question but is linked instead to an understood assertion which, if
explicitly inserted, may read like this:
(1) C-l. What are you doing tonight? I am asking that

because there's a good movie on.
In the course of presenting a variety of cases that are less clear-cut

than the ones in (la-c) above, Sweetser does of course mention
disambiguating features that may (but need not) be present in the
grammatical structure of the conjunction.
For example, she hints at the role of intonation: commaless intonation
is linked with the content reading in (Ia) but unavailable to the other
readings in (lb) or (lc); the latter require a so-called "comma intonation"
such that the clause-final intonation drop marks the first clause as an
independent assertion. Sweetser also discusses the role of iconic ordering
of clauses in different domains (see her Ch. 4.2; more on this in § 3
below).
The overall tendency of her approach, however, is based on the
following claim :
(ill) Pervasiveness of pragmatic ambiguity
Given sufficient context, we can almost always force either a

content-conjunction reading or an epistemic-conjunction reading
on any pair of clauses conjoined by because; it is just harder to
find reasonable contexts for some readings than for others. <...>
[in addition to because] therefore, since, so, although, and
despite < ... > all show such multiple usage. (1990:77-8)
2. Objections and amendments
While 1 fundamentally agree with claim (I) One meaning, several uses (I
consider this to be the only promising way of dealing with connectors!),
and while 1 am basically sympathetic to the notion of "pragmatic
ambiguity", 1 see some serious difficulties in accepting the claims in (II)
and (ill).
238 Ewald Lang
The notion of "pragmatic ambiguity" is in need of refmement and

elaboration. To achieve this aim Sweetser's approach will be relativized
by
(N) a. challenging the ambiguity claim advocated in (IT) and (ill);

b. inspecting more closely the features based on which the
supposed levels are distinguished;
c. checking the consistency and exhaustiveness of the three-level
distinction.
Pragmatics is not considered to be the sort of stop-gap device which

Sweetser purports it to be; rather, what Sweetser calls "a pragmatically
motivated choice of interpretation" (see (IT) above) will be shown to draw
heavily on structural information available from the conjoined clauses as
well as the connectors occurring in the conjunctions at issue. Specifically,
I will defend (and provide evidence for) the following counter-claims,
each of which has its bearing on the three points of criticism in (N). For
ease of reference, these claims will be prefixed by "C-". The rest of this
section will concentrate on (C-l), the others will be discussed in § 3.
(C-l) There are more level-specific structural differences in the way

in which clauses are conjoined than Sweetser's claims (I) - (ill)
would admit.
(C-2) The lexical items, i.e. the connectors that are used to connect
clauses on the three levels of interpretation, display more level-
specific distributional restrictions than the sample of cases
discussed by Sweetser would suggest.
(C-3) The three-level distinction proposed by Sweetser must be

supplemented by another level-provisionally called "textual
progression" or "discourse perspective"-which is not located
within the Sweetserian hierarchy of levels but cuts across them.
(C-l) emphasizes the role of structural information in determining the

interpretation of a given conjunction. Its import can easily be
demonstrated by taking up Sweetser's introductory set of examples. A
closer look at (la-c) immediately reveals that there is a whole range of
structural dimensions with respect to which the clauses that are linked
together by connectors differ. Spelling out (C-l) in detail means checking
the clauses of what Sweetser calls a "conjunction" for a possible display
of some crucial structural features, e.g. those listed in (CI-I) to (CI-5)

below.
(CI-I) Are the clauses of the same or of different clause type? (in
morphosyntactic terms of being declarative, interrogative, or
imperative sentences)
It is obvious that a content-level interpretation in the sense of

Sweetser's paragon example (la) is confined to clauses which share
declarative mood and thus may render STATEMENTS. Taking this as a
necessary condition for content-level interpretation implies that any
conjunction whose clauses differ in type is excluded from being
interpreted on that level and is thus confined to being interpreted on
epistemic and/or speech act level. 2
Sharing declarative mood is a necessary, though not a sufficient,
condition for a conjunction of clauses to be interpretable on the content
level. Consider again (lb) and a version of (1 c) in which the SPEECH ACT
of ASKING is not grammatically overt but concealed in a declarative:
(1) b. John loved her, because he came back.

c'. I am asking you what you are doing tonight,
because there's a good movie on.
This is the point where further criterial features enter the picture. One of
them, of course, is comma intonation. Due to its interaction with other
features the indicative role of comma intonation is more far-reaching than
Sweetser assumes. Below the level of sharing declarative mood there are
important indicators of clause-internal differences which interact with
comma intonation. Thus, in isolation, the first clause of (1 b) may render a
STATEMENT as does the first clause in (la). Both behave differently,
though, if we add an adjunct like presumably or a parenthetical like I
guess, which mark the hypotheticality or assumption status of the
assertion being made:
(1) a'. Presumably John came back because he loved her.

John, I guess, came back because he loved her.
b'. Presumably John loved her, because he came back.
John, I guess, loved her, because he came back.
While in (la') hypotheticality markers like presumably take scope over

both clauses, their scope in (1 b') is confined to the first clause. Thus the
240 Ewald Lang
function of comma intonation to mark the first clause as an independent

assertion is confirmed by its effect of marking scope boundaries. Similar
observations can be made for negation and other operators. Despite
looking alike on the surface, the first clauses in (la, Ia') and (Ib, Ib')
differ in scope determination as well as in intonation, which undoubtedly
both form part of core grammar, and hence provide distinct cues for the
interpretation to be assigned to the conjunction as a whole.
Thus despite the fact that the first clauses in (1a) and (1 b) are alike in
sharing declarative mood, they also differ. The first clause in (Ia) asserts
a proposition in the form of a factual statement thus allowing for a
content-level interpretation of the conjunction. The first clause in (1 b),
however, asserts a proposition in the form of an ASSUMPTION as
evidenced by comma intonation and characteristic scope properties. The
assumption status of the first clause is crucial to epistemic-Ievel
interpretation of the conjunction.
The next dimension in support of level-specific structure formation as
claimed by (C-l) is mentioned by Sweetser in discussing "iconic
ordering" in coordinate structures. However, the point at issue can be
extended to causal connectors and exploited in a more systematic way.
(C 1-2) Restrictions on the order of clauses with subordinators
(Ia) on the content level allows both orders-John came back

because he loved her and Because he loved her John came back, though
there is a difference that distinguishes backgrounded vs. focussed
material on the level of discourse-related information structure. In any
case, preposing the subordinate clause does not prevent a content-level
reading as shown by (2a). Coherence is maintained independently of the
order of clauses. However, as shown in (2b-c'), Sweetser's examples
(Ib) and (Ic) as well as their German equivalents do not allow of
preposed because/weil-clauses if they are to keep their respective
readings on the epistemic or speech act level (marked * and ??,
respectively):
(2) a. Weil er sie liebte, kam John zuriick. [content]

'Because he loved her John came back. '
b. Weil er zuriickkam, liebte John sie. *[epistemic]
'Because he came back, John loved her. '
b'. Weil er zuriickkam,

LIEBTE John sie vermutlich / denke ich. [epistemic]
'Because he came back,
John presumably / I guess LOVed her. '
c. Because there's a good movie on,
what are you doing tonight? ??[speech act]
c'. Because there's a good movie on, I'd like to ask you:
what are you doing tonight? [speech act]
Thus, unlike (lb), a conjunction like (2b) cannot be interpreted as

justifying the conclusion "he loved her" from the fact that "he came
back". What is lacking here is the sort of coherence that is rendered by
the original ordering. To ensure coherence in such cases requires
additional lexical or prosodic focus indicators, as is illustrated in (2b').
Similarly, while clause type difference and intonation break may serve as
compensatory cues, (2c) is still incoherent. An amended version of it
should read like (2c').
Sweetser recognizes the role of conjunct ordering in discussing
conjunctions with coordinators like and, or and but by stating that
pragmatically "the primary, independent conjunct precedes the secondary,
dependent conjunct" (1990:98). Yet, she simply ignores the ordering
effects in conjunctions with subordinators. The facts in (2b, 2c) above
show that (Cl-2) has to be taken as a condition on level-specific structure
formation, which in tum has serious repercussions on Sweetser's
ambiguity claim in (II)-(ill) above.
First, it is a grammatical category feature of subordinators like
because, since, etc. to allow for clause preposing while keeping their
interpretation constant-cf (2a). If this built-in property fails, as with
(2b) for example, this has to be acknowledged as a disambiguating
grammatical fact, too. It is on this basis that pragmatic interpretation
starts to work, contrary to the ambiguity claim in (II).
Second, unlike subordinators, coordinators like and, or, but (though
notfor or German denn) are grammatically unrestricted as to the order of
conjuncts and hence do not provide category-based grammatical
information that is taken up by pragmatic interpretation. Among other
things, it is the sequential ordering of the conjuncts that is submitted to
pragmatic interpretation (along the lines discussed by Sweetser).
Third, in view of this, we should expect coordinate conjunctions to be
less fertile, hence more ambiguous or vague, in providing grammatically
determined information on which, following Sweetser, the "pragmatic
choice of the' correct' interpretation" may draw.
242 Ewald Lang
In what follows it will be shown that this assumption is by no means

justified. Due to space restrictions the further pursuit of (C-I) to (C-3)
will be confined to coordinate conjunctions whose clauses are linked by
adversative connectors. Moreover, we will employ German data3 and
present arguments which draw on results from current work being carried
out in the connector project at the Institut fur deutsche Sprache
Mannheim (for details see HdK [=Pasch et al. (forthcoming)]).
3. Adversative connectors
3.1. How ambiguous are aberlbut-conjunctions?
From the set of about 30 lexical items that may be subsumed under the
label "adversative connector" in German we will select only those needed
to illustrate (C-I) to (C-3).
Let us first turn to coordination involving the connector aber, which
(like but in English) is commonly considered as the standard adversative
connector. Compared with connectors such as trotzdem ('nevertheless'),
hingegen ('whereas'), or dennoch ('nonetheless', 'all the same'), aber is
much less restricted regarding distribution and hence to a much higher
degree prone to "pragmatic ambiguity". Nevertheless, the supposed
ambiguity of aber-conjunctions is drastically reduced by cues which
emerge from structural features of the coordinate clauses in the domain of
sentence structure. In addition to CI-I and CI-2, the relevant features
include:
(CI-3) Matching or independently varying syntactic clause structure
(CI-4) Parallel or independently varying information structure
(C 1-5) (Un)availability of clause-internal deletions like "gapping",

"right node raising", etc. (Lang 1991, see also th. 4 below).
Let us now look at some data that are representative of the range of cases
to be taken into account.
(3) a. Dein Vortrag war sehr lang, aber das soli keine Kritik sein.
'Your talk was rather lengthy, but this is not a criticism. '
b. Wir sind mit dem Manuskript in Verzug,

aber welcher Autor halt schon Termine ein?
'We are late with the manuscript,
but what author keeps the dead-lines?'
c. Willst du heute etwa ins Kino-aber ich komme nicht mit.
'Do you want to go to the movies today-but I'm not coming
along'
All examples in (3) are easily recognizable as cases to be interpreted on

the speech act level. Just recall the relevant indicators mentioned in (C I-
I), namely "clause type difference" - cf (3b, 3c), "intonational break" -
cf (3a), and what was said about "independent clause-internal structures"
in CI-3 to CI-4 (i.e. lack of parallelism as to constituency and topic-focus
articulation). Just as in (lc) above, it is the conjunct-internal differences
which prevent the examples in (3) from being interpretable on the content
level and qualify them for speech act level interpretation. Generally
speaking, the reason is this: aber/but marks what is rendered by the
second clause as being in contrast with inferences that might be drawn
from the speech act that is performed by uttering the first clause. In short,
what counts are the conjunct-internal differences in terms of C1-1 to C1-
4. If such differences are missing, more effort is needed to determine the
respective levels of interpretation. Consider the following examples,
which Sweetser tends to locate at the epistemic level.
(4) a. Hans hat Unmengen von Zigarren gehortet,

aber er raucht gar nicht.
'Hans has hoarded an obscene number of cigars,
but he doesn't even smoke'
b. John keeps six boxes ofpancake mix on hand,
but he never eats pancakes. (cf Sweetser (1990: 100»
Sweetser comments: "But presents two conjuncts which clash with each
other in some way ... At the epistemic level, the available premises may
clash with an apparently necessary conclusion" (1990: 100). The first part
of the quote is unproblematic; but what exactly is "epistemic" about the
entities connected by aber/but? (4a, b) instantiate the well-known type of
"denial of expectation" interpretation of but-conjunctions (Lakoff
1971:133). In both cases, the clauses are mutually independent, hence
semantically compatible STATEMENTS. Given this, (4a, b) qualify as
coordinations of statements at the content level with the additional
requirement that the interpretation involves inferring an ASSUMPTION
244 Ewald Lang
which is in contrast with the two statements being presented side by side.
This analysis reduces the "epistemic" aspect of (4a, b) to the inferences to
be drawn and thereby does justice to the structurally determined level-
distinction. The same point was made in Foolen (1992:83).
3.2. Problems with interpreting but/aber at the content level
The awkwardness of Sweetser's use of the notion "epistemic" is

furthermore revealed by the stand she takes on cases of what since Lakoff
(1971:133) is commonly known as "semantic opposition" but. Sweetser
(1990:103) claims: "The but in (56) [= (5a) below] does indeed indicate
contrast: an epistemic contrast between two semantically opposed
propositions" (italics mine - EL). From this we may conclude that (5b)
lacks such an epistemic contrast, other things being equal. Again, (5b)
with and is a clear-cut case of content-level conjunction.
(5) a. John is rich but Bill is poor. [= Sweetser's (56), 1990:103 ]

b. John is rich and Bill is poor.
c. John is rich, but Bill is rich, too.
d. *John is rich but Bill is rich.
What we are left with is the problem of where to locate (5a): it does
indeed indicate a different contrast from (5b). I doubt, however, whether
this contrast can be accounted for in terms of the "epistemic level"
without rendering this notion completely void. I would contend that the
alleged "epistemic" difference between (5a) and (5b) is that the
interpretation of (5a) requires some additional inferencing which the
interpretation of (5b) does not. Thus, (5a) and (5b), like (5c) when
compared with (5d), clearly differ as to the conditions they impose on the
context in order to get interpreted properly.
According to an observation first made by Bellert (1972), the
interpretation of but/aber-conjunctions involves at least two semantic
contrasts to be read off or inferred from the conjuncts. (5d) contains the
overt contrast John-Bill but fails to provide cues to reconstruct the
second contrast needed, i. e. an INFERRED ASSUMPTION which is in
opposition to the STATEMENT Bill is rich. This pragmatic insufficiency
has its grammatical counterpart: (5d) can be considered deviant because
it does not allow a contrastive stress pattern - just try and you·will see'
Moreover, contrary to most work done in the field, we claim that the
"semantic opposition" (more exactly, the occurrence of antonymous
predicates in logically compatible STATEMENTS) in (Sa) is not a crucial

feature of this type of conjunction (as shown by (5c) which lacks
opposition); and that this opposition is by no means the carrier of the
contrast indicated by aber/but. Instead, the contrasting entity has to be
inferred from outside the coordination. However, the opposition in (5a) as
well as its absence in (5c) serve as structural clues to reconstructing it.
We will return to this in 3.3 below.
Now, if (5a) cannot be located on the epistemic level, why not
consider it a content-level conjunction - very much like (5b)? Sweetser
(1990:103) rejects this idea by stating: "I have not been able to unearth
any indubitable content-conjunction examples with but" and by offering
"as a plausible explanation for the use of but in only two domains"
[epistemic level and speech act level - EL] the following consideration:
"what does it mean to say that A and B "clash" or "contrast" in the real
world? How can discordance or contrast exist outside of the speaker's
mental concept of harmony or non-contrast?"
Sure enough, the contrast induced by aber/but is not something to be
looked for in the real world but something to be established by the
interlocutors in assessing what, by way of coordinate conjoining, is
presented as co-existing in the world. But this does not prevent aber/but-
conjunctions from being interpretable at the content level, that is, as
conjoining two STATEl\1ENTS which lay the ground for additional
inferencing. Sweetser's problem with interpreting but at the content level
is caused by confusing two notions which, though not unrelated, have to
be kept distinct: inference (whether a proposition is asserted or inferred)
and epistemic status (whether a proposition forms the content of a factual
STATEMENT or an ASSUMPTION).
The alternative approach suggested here tries to avoid this confusion.
It draws on the fact that the interpretation of an adversative conjunction
involves relating the conjoined clauses to an ASSUMPTION which-if not
available in explicit terms (see (V) below)-has to be inferred from
outside the conjunction. Based on these considerations it is posited that
(V) (a) adversative (and probably also concessive) connectors inher-

ently contain pointers to previous information available from
the context, and due to this
(b) adversative conjunctions necessarily involve some back-
tracking that may well go beyond the domain of sentence
structure and operates on (what may be called) the level of
"textual progression" or "discourse perspective".
246 Ewald Lang
Following (C-3) it is this level which provides the appropriate means to

interpret adversative conjunctions, and therefore we consider it a
necessaIy supplement to Sweetser's approach. Note that the level of
"discourse perspective" must not be understood as an additional level
within, or on top of, the Sweetserian hierarchy, but rather as embracing
the three levels discussed so far as grammatically determined patterns of
discourse formation. Recall that C1-1 to Cl-5 apply to sentence structure
and thus clearly contribute to discourse formation and comprehension.
3.3. Adversative conjunctions and the division of labour between

grammar and pragmatics
This section will provide an outline of the way in which sentence and
discourse level interact in determining the interpretation of adversative
conjunctions. To begin with, note that the semantic contribution of
aber/but in combining two clauses into a conjunction is twofold. In a
nutshell, it may be conceived of as:
(VI) (a) conjoining semantically compatible and non-inclusive

propositions that-depending on the number of structural
parallels between the conjuncts-may be bundled up or
enumerated as instances of a "common integrator" (Lang
1984, 1991, in print; see also fn. 4 below);
(b) indicating that the assertion rendered by the second clause
is in contrast to an ASSUMPTION that either may be read
off, or must be inferred from, previous information
(Braul3e 1998).
(VIa) states what aber/but share with und/and; therefore (VIa) is the
basis for (Sa) and (Sb) having identical truth conditions. This is a feature
of aber/but that probably nobody will deny. (VIb) is what separates
aber/but from undland, and hence (VIb) is what causes (Sa) and (Sb) to
have distinct conditions of use as regards semantic coherence. As such,
(VIb) is perhaps not as uncontroversial as (VIa), so it may be worth
commenting on.
What has been stated in (VIa + b) should not be understood as
amounting to "but = and + pragmatics". This would defmitely miss the
target. The conditions of use of aber/but and other adversative connectors
are not simply a matter of "pragmatically motivated interpretation" (as
Sweetser suggests); rather they are rooted in the lexical semantics as well
as in the syntactic category features of these connectors. On the other

hand, I do not deny the role which contextually induced pragmatic
interpretation inevitably has to play in specifying the "correct" reading
which a conjunction of clauses eventually gets assigned. I differ from
Sweetser, however, by claiming that in order to account for (what she
calls) "pragmatic ambiguity", the division of labour between grammar
and pragmatics has to be redefined. Regarding aber/but as characterized
in (VIb), this division of labour may be determined as follows:
That part of (VIb) which reads "indicating that the assertion rendered
by the second clause is in contrast to an ASSUMPTION" has to be spelled
out primarily in terms of grammatical structure formation (including C1-1
to CI-5), while the other part, which reads "an ASSUMPTION that either
may be read off, or must be inferred from, previous information", is
where pragmatics proper enters the picture.
It seems reasonable to consider the interpretation of an adversative
conjunction as involving a search device which is determined by a
specific source-target relation. Triggered by the lexical indication that a
contrast must be established, the device takes the assertion rendered by
the second clause as the source from which to look for an appropriate
ASSUMPTION that meets the condition of contrast, i.e. the target. Though
the search device is to some extent guided by grammatical structure, as
claimed by (Cl-l) to (CI-5), it is inevitably accompanied and
supplemented by pragmatically based reasoning. So the part to be played
by pragmatics proper is to provide supplementary means to find out what
the target, i.e. the INFERRED ASSUMPTION, is.
Note that by dividing up (VIb) between grammar and pragmatics as
suggested afore, it is not an accident that the notion "ASSUMPTION"
figures in both, and rightly so. The INFERRED ASSUMPTION forms the
bridge between grammar and pragmatics in interpreting adversative
conjunctions. Almost all studies of adversative connectors have attempted
to account for it in one way or another (for a survey see Rudolph 1996).
What is new is the way to tackle it as proposed in (VIb).
Source and target are both propositional but differ as to their epistemic
status. That the target (i.e. the proposition which the second clause is put
into contrast with) has to be an ASSUMPTION, not a factual STATEMENT,
can be seen if, for example, (4a) is extended by a clause that renders the
target in explicit terms, cf the bold-faced additions in (6):
248 Ewald Lang
(6) a. Hans hat Unmengen von Zigarren gehortet,

aber er raucht gar nicht. [= 4a]
'Hans has hoarded an obscene number of cigars,
but he doesn't (even) smoke'
b. Hans hat Unmengen von Zigarren gehortet,
*folglich raucht er, aber er raucht nicht.
' ... hence he smokes, but he doesn't smoke'
c. Hans hat Unmengen von Zigarren gehortet,
also konnte man vermuten, daB er raucht,
aber er raucht nicht.
, ... so one could assume that he smokes,
but he doesn't smoke'
If the supposed target of contrast is expressed in the form of a

FACTUAL STATEMENT, the result is a contradiction-cf (6b), but if it
appears with overt markers of ASSUMPTION status as in (6c), the inter-
pretation of the entire construction is consistent. This is what follows
from the grammar part of (VI), which requires the target of contrast to be
an ASSUMPTION.
Let us now turn to the pragmatic aspects of (VI). If the clauses
conjoined by aber/but both represent STATEMENTS (as is the case with
the majority of adversative conjunctions), the ASSUMPTION forming the
target of contrast has to be inferred. Here is where pragmatics comes
into play by providing the means to infer the target assumption from
previous information. Reconsidering the conjunctions in (4) and (5) above
will then amount to the following analysis.
In (4a/6a), the inference is obviously drawn by resorting to world
knowledge about smokers such that the STATEMENT rendered by the
first clause (Hans has hoarded an obscene number ofcigars) is taken as
a premise which by abductive reasoning makes it plausible to assume
"Hans smokes" and thus to yield a suitable target assumption. By the
way, the fact that this inference is not based on deductive but on
abductive reasoning provides additional evidence of the ASSUMPTION
status of the target.
In (Sa-d), repeated as (7a-d) below, the inferential steps leading to the
target are less clearly set out. (This is what caused the trouble for
Sweetser's approach, cf § 3.2 above.) How come?
The clauses conjoined by aber/but in (5/7) are clear-cut STATE-
MENTS, so the target of contrast has to be inferred from outside the
conjunction. Moreover, the clauses conjoined are absolutely parallel as to
their internal structure. So they cannot but be taken as instances of a
"common integrator".4 The common integrator, which has to be

identified within the conjunction (see fu. 4 below) or else has to be
inferred from outside the conjunction, then serves as the location frame
within which the contrast is to be placed. Thus, in comparison to (4a16a),
the conjunctions in (5/7) require additional cues from the discourse level
in order to be assigned an appropriate interpretation.
The simplest way to illustrate these context conditions consists in
showing what sort of questions the respective conjunctions in (7) might
serve to answer. Suitable questions (added in 11...11 ) thus reveal the
distinct effects that are caused by the structural differences of the
conjoined clauses.
(7) a. John is rich but Bill is poor.

liAs for John and Bill, are they both rich? II
b. John is rich and Bill is poor.
II As for John and Bill, what about their income? II
c. John is rich, but Bill is rich, too.
II As for John and Bill, they aren't both rich, are they? II
d. *John is rich but Bill is rich.
II??? II
The structural differences are clearly brought out by the fact that none
of (7a-d) is replaceable with any of the others as an appropriate answer to
the respective question-context given in II. ..11. Assessing (517) once more,
we may say: in isolation and in written form these conjunctions are not
really ambiguous but are underspecified regarding the cues they need to
get interpreted.
Finally, we will look at those grammatical features of adversative
connectors that form the basis of the division of labour between grammar
and pragmatics just outlined. The most relevant point to note is this:
indicating a contrast in the way suggested in (VIb) is a semantically built-
in feature of aber/but which correlates with likewise built-in syntactic
properties of these connectors. In contrast to and, or, etc., adversative
coordinators like aber/but display the following features:
First, aberlbut allows for binary connection only. This is the basis for
marking the assertion following aberlbut as the source from which to
look for an appropriate CONTRASTING ASSUMPTION, that is, the target.
Second, aber Ibut is intrinsically asymmetric in marking what has
been called a "change of perspective" (Spooren 1989): for example, (4a)
or (7a) with the reversed order of conjuncts would not render the same
250 Ewald Lang
information on the level of discourse perspective, because source and

target are reversed.
Third, unlike und/and, which allows for scope ambiguities, aber/but is
a marker of scope boundaries: negation or other scope-taking operators
cannot take scope over aber/but, cf:
(8) Hans ist nicht dumm undfaul [ambiguous as to scope of negation]

'Hans is not stupid and lazy'
(9) Hans ist nicht dumm, aberfaul [scope of negation unambiguous]

'Hans is not stupid but he is lazy'S
With these features in mind let us now examine a couple of aber/but

conjunctions which display subtle but telling differences in the
information structure of the conjuncts.
To begin with, note that in written form (10) below is underspecified
as to its information structure and hence to the way it is linked with
previous discourse. However, when (10) occurs as a continuation of (11)
or (12), which (along the lines of (5/7) above) form distinct discourse
contexts which (10) has to match, we realize that (10) assumes
correspondingly distinct patterns of information structure shown in (11)
and (12), respectively. (In the following examples, SMALL CAPS indicate
main stress, I and \ contrastive stress, labeled square brackets topics and
foci, respectively.)
(10) Hans ist krank, aber seine Frau geht arbeiten.

'Hans is off sick but his wife is at work'
(11) II Wie geht's den beiden? II 'How are the two doing?'
a. Tl[/Hans] F[ist KRANK], aber T2[seine Frau\] F[geht ARBEITEN]

b. T[[Hans] und [seine Frau]] sind F[[IKRANK] bzw. [auf ARBEIT\]]

'Hans and his wife are off sick and at work, respectively'
(12) II Was regt dich so auf?11 'What's upsetting you?'
a. F[[Hans] ist KRANK], aber F[[seine Frau] geht ARBEITEN]

b. F[[Hans] ist KRANK], dennoch F[[seine Frau] geht ARBEITEN]

'Hans is off sick, nevertheless, his wife is at work'
c. F[Obwohl [Hans] KRANK ist], F[geht [seine Frau] ARBEITEN]

'Although Hans is off sick, his wife is at work'
Let us take a closer look at the way in which information structure at

sentence level and at discourse level interact. The contexts adduced in
(11) and (12) each form a discourse topic whose internal structure
determines the way in which it is elaborated by the sentences following it.
The question in (11) marks the NP den beiden 'the two' as a (con-
tinued) topic and specifies the VP as question focus, thus determining
focussed VPs as answers. In view of this, (I1a) shows a topic-focus
articulation which elaborates that of the discourse topic in a strictly
parallel way: the topic den beiden is resumed by the topical subjects
Hans and seine Frau, respectively; the VP question focus is instantiated
by the VPs ist krank and geht arbeiten, respectively. This mode of
parallel elaboration is evidenced also by two symmetry features: (i) the
clauses in (lIa) could be reordered without changing their links to the
discourse topic; (ii) (IIa) can be regrouped into a respectively
construction like (lIb). Hence, (11a,b) instantiate what might be called
"strictly parallel elaboration of a discourse topic".
In (12), where was ('What') refers to a whole proposition, the
question is such that it requires all-focus sentences as answers. Given
this, (10) is assigned the information structure in (I2a), which determines
the conjuncts as all-focus sentences at the discourse level. The
informational status of Hans and seine Frau is open: syntactically they
are sentence topics, but it depends on further discourse factors whether or
not they resume a discourse topic as they do in (I1a,b). Now, (12a) is a
predetermined elaboration of the discourse topic Was regt dich so auf?
('What's upsetting you?'), but there is no need for the elaboration to
spread over a conjunction of clauses-Hans ist krank ('Hans is off sick')
would suffice as an answer to (12) alone. In case the elaboration involves
a conjunction of clauses, these are not bound to the sort of parallelism we
observe with (11a,b). The relation between the all-focus clauses in (12a)
is not dependent on the discourse topic. Hence, (I2a) is what might be
called a "non-symmetric elaboration of a discourse topic. " This is
evidenced by (12b,c), which answer the question by couching the same
asserted propositions in concessive constructions. Concessive
constructions6 , no matter whether coordinate like (I2b) or subordinate
252 Ewald Lang
like (12c), are inherently asymmetric regarding the relation between the
clauses they link.
To summarize: due to their distinct information structure, (12a-c) are
not replaceable with (1Ia,b) while keeping the respective discourse
contexts constant, and vice versa.
Furthermore, the examples in (11) and (12) reveal an important
difference in the way interclausal aber/but relate to the discourse. In the
non-symmetric cases, the target from which the ASSUMPTION that
contrasts with the STATE1vffiNT in the second conjunct ('His wife is at
work') has to be derived is supplied by the first conjunct. So, in (12a-c) it
is the assertion of John's being off sick that triggers the relevant
inference: sick people need care, and spouses make good candidates for
this job.
In (11a,b), which form symmetric continuations of the given discourse
topic, the target that licenses the use of aber/but is not as easy to
reconstruct. Structurally, the question Wie geht's den beiden? ('How are
the two doing?') does not provide an appropriate target per se; all it does
is to license a list of and-conjoined items, cf (11 b). This is where
pragmatics enters the picture. In interpreting (1Ia), we face the following
situation: if searching the wider context for a suitable target fails, we are
left with no alternative but to take the first conjunct Hans ist krank
('Hans is off sick') and generate from it the implicature that the news
about his wife will be of similar quality.
Given that, as claimed in (VIb), aber/but has the built-in feature of
triggering an inference, the notion of a but of "semantic opposition"
(introduced by Lakoff (1971» seems to be misconceived: semantic
opposition per se never suffices to license the use of aber/but.
Sweetser's three-level approach has proved to be valuable as a heuristic

means in sorting the various interpretations that conjunctions can be
assigned. However, in order to serve as a means of explanation, it needs
to be modified in some respects.
Regarding the level of grammatically determined sentence structure,
Sweetser's level-distinction can be more precisely reconstructed in terms
of criteria like Cl-l to CI-5, among which Cl-l (sameness or difference
of clause type) is the most crucial one. Given this, we might reconstruct
what should be kept of Sweetser's three-level distinction of conjunction
interpretation in the following way:
(VII) a. Content-level interpretation is reserved for conjunctions of

declarative clauses that render STATEMENTS (i.e. entities
to be related to the world via truth-conditions); see (la, 3a,
4a,5b,6a).
b. Epistemic-Ievel interpretation is confined to conjunctions of
declarative clauses, of which at least one is marked as
rendering an ASSUMPTION; see (lb',lb-i, 1b-ii, 1b-iii,
2b').
c. Speech act level-interpretation is available for conjunctions
of non-declaratives (with reservations about or-con-
junctions like (2» as well as for conjunctions of clauses
differing as to type but with one clause being marked as
rendering a SPEECH ACT; see (lc, 1c', 1c-i, 2c, 2c', 3a-c).
In addition, Sweetser's approach has to be modified in a way that

would enable it to account for contextual requirements imposed on the
interpretation of adversative conjunctions. The level that is assumed in
(C-3) to do this is the all-embracing level of "discourse perspective". This
is the level where (a) the information structure of sentences or clauses
interacts with that of the discourse containing them-as exemplified by
the cases in (7) and (10)-(12); (b) where the peculiar syntax and
semantics of adversative connectors can be accounted for in a promising
way; (c) where the division of labour between grammar and pragmatics
can be defined - cf (VI a+b).
Pragmatics comes in when grammar does not provide sufficient cues
to get an adversative conjunction properly interpreted.
Notes
* This paper is dedicated to Ekkehard Konig on the occasion of his 60 th

birthday.
I wish to thank the editors for useful comments on an earlier draft of this
paper and Chris Wilder for improving my English.
1. Note that Sweetser uses the term "conjunction" for the syntactic operation of
clause-linking without distinguishing coordination and subordination. In
order to avoid confusion with the word class "conjunctions" I will call the
latter "connectors" (subdivided into "coordinators" and "subordinators") and
the constituents they link "conjuncts"~ otherwise I will follow Sweetser's
terminology.
2. It may be noted in passing that sameness of clause type with non-declaratives
raises some problems concerning the conjoining of clauses on the speech act
254 Ewald Lang
level. I will take for granted that pragmatically (i) is asymmetric, while (ii) is
symmetric regarding the alternatives presented:
(i) Give me liberty or give me death! [= Sweetser's (46), p. 98]
(ii) Give me a salami sandwich or a hotdog!
What needs to be clarified, however, is this: is it justified to consider (ii) as
an or-{;onjunction of two speech acts? Speech acts normally resist being
presented as alternatives.
Another open question is whether we are to take (ii) as a conjunction of two
speech acts or rather as a single speech act of request which comprises an
alternative of how the request might be complied with?
Again, these questions show that a closer examination of the internal
structure of the conjuncts seems to be unavoidable.
3. The advantage of illustrating the issue at hand with German data consists in
the fact that aber is an unambiguous adversative connector. It does not share
the range of additional uses and readings that are covered by but, as in
German these are rendered by separate lexical items like sondern, auf3er etc.
Analysing but-constructions in comparison with aber-constructions thus will
help us to pick out only those having adversative readings.
4. Without going into details (see Lang 1984, 1991, in print), let me just point
out that the derivation of the Common Integrator (CI) of a given conjunction
(a) is sensitive to surface structure parallels shared by the conjoined clauses,
and (b) can best be illustrated by means of a (multiple) question to which the
given conjunction would serve as an appropriate answer.
So while (i)-(iv) share the propositional formula "someone transfers
something to someone" as constituent part of their respective Cis, the
specific make-up of the latter rests on the information structure of the
clauses, specifically on the (non-)coreference of the items instantiating the
argument places. (Non-)coreference is indicated by (non-) coindexing.
Coreference requires destressing: if e.g. the beast is coindexed with the dog
the former serves as an anaphoric epithet for the latter.
(i) Grandpa will bequeath the dogk to his grandchildreni and
Grandma (will)(bequeath) the beastv*k to us j/*i
CI: WHO will bequeath WHAT TO WHOM?
(ii) Grandpa will bequeath the dogk to his grandchildreni and/but
Grandma will sell the beastk/*l to us i/*j
CI: WHO will transfer the dogk to USi IN WHAT WAY?
(iii) Grandpa and Grandma will bequeath the dogkand the beastv*k
to their grandchildreni and to us j/*i, respectively
CI: WHO will bequeath WHAT TO WHOM?
(iv) Grandpa will bequeath, and Grandma will sell, the dog to us
CI: WHO will transfer the dogkto USi IN WHAT WAY?
Note that the internal structure of the conjuncts in (i) allows for regrouping
the parallel constituents into a respectively construction like (iii) which leads
to the same CI as (i) but is, for instance, unavailable for (ii). On the other
hand, (ii) can also be rendered by a right node raising construction like (iv)
which leads to the same CI as (ii) but is unavailable for (i). So (iii) and (iv)
may serve as overt syntactic diagnostics for information structure and CI
differences that seem to be concealed when conjunctions like (i) or (ii) are
presented in written form only.
5. (8) and (9) point to an important difference between German aber and
English but regarding scope determination under negation. Since German
adversative coordinators for contrast (aber) vs. correction (sondern) are
lexically distinct, their scopal behaviour is built-in and need not be indicated
by other means-ef. (i, ii). English but-constructions, however, have to
render the distinction of contrast vs. correction under negation by distinct
conditions on deleting non-focussed material in the second clause. Thus, but-
constructions like (iii) expressing contrast disallow, whereas but-
constructions like (iv) expressing correction require, the deletion of repeated
material in the second conjunct (for details see Lang (1991).
(i) Hans ist nicht dumm, aber (er ist) faul [contrast]
(ii) Hans ist nicht dumm, sondem (er ist) faul [correction]
(iii) Hans is not stupid but *(he is) lazy [contrast]
(iv) Hans is not stupid but (*he is) lazy [correction]
The same point can be made to explain the difference in the following cases:
(v) John ist reich, aber Bill (ist) arm.
(vii) John is rich but Bill *(is) poor.
(vi) John ist reich, aber Bill nicht lauch.
(viii) John is rich but Bill *(is) not Itoo.
Despite the differences shown, aber and adversative but both mark scope
boundaries in preventing propositional operators from taking scope over
them.
6. The relationship between aber-conjunctions and concessive constructions
cannot be tackled here, but see the contributions of Barth and of Crevels (in
this volume).
Another point that deserves mentioning is the fact that German aber (unlike
English but) is not confined to interclausal positions, but can also float in the
second clause (this is why in HdK [=Pasch et a1. (forthcoming)] aber has
been classified as a "connective particle"). At first glance, this distributional
flexibility seems to support the view that aber is most prone to "pragmatic
ambiguity". It can be shown, however, that being allowed to float within the
second clause is what enables aber to syntactically mark the information
structure relation of the second clause with respect to the discourse topic as a
non-symmetric one, similar to (12a-c) above:
(12) d. F[[Hans] ist KRANK], F[[seine Frau] geht aber ARBEITEN]
'Hans is off sick, his wife is, nevertheless, at work'
The prosodic details of contrastive vs. concessive clause linking are being
investigated in a current DFG project of mine which forms the source of the
data discussed here.
256 Ewald Lang
R~ferences
Bellert, Irena
1972 On certain properties of the English Connectives and and but. In:
Senta Plotz (ed.), Transformationelle Analyse, 327-356. Frankfurt
am Main: Athenaum.
BrauBe, Ursula
1998 Was ist Adversativitat? aber oder una? Deutsche Sprache 26:138-
159.
Foolen, Ad
1992 Review of Sweetser 1990. Lingua 88:76-86.
HdK [= Pasch, Renate, Ursula Brausse, and Eva Breindl]
(to appear) Handbuch der deutschen Konnektoren. 2 vols. Ms. Mannheim:
Institut flir deutsche Sprache.
Kortmann, Bernd
1997 Adverbial subordination. A typology and history of adverbial sub-
ordinators based on European languages. (EALT 18.) Berlin/New
York: Mouton de Gruyter.
Lakoff, Robin
1971 If's, and's and but's about conjunction. In: Charles J. Fillmore and
D. T. Langendoen (eds.), Studies in linguistic semantics, 114-149.
New York: Holt, Rinehart & Winston.
Lang, Ewald
1984 The semantics of coordination. (SLCS 9.) ArnsterdamlPhila-
delphia: Benjarnins.
1991 Koordinierende Konjunktionen. In: Semantik/Semantics, Art. 26,
597-623. (HSK 6.) BerlinlNew York: Mouton de Gruyter.
(in print) Die Wortart "Konjunktion". In: Lexicology/Lexikologie, Art. 88,
(HSK 17.) Berlin/New York: Mouton de Gruyter.
Raible, Wolfgang
1992 Junktion. Eine Dimension der Sprache und ihre
Realisierungsformen zwischen Aggregation und Integration.
Heidelberg: Winter.
Rudolph, Elisabeth
1996 Contrast: adversative and concessive relations and their
expressions in English, German, Spanish, Portuguese on sentence
and text level. (Research in text theory 23.) BerlinlNew York:
Mouton de Gruyter.
Spooren, Willibrordus
1989 Some Aspects of the form and interpretation ofglobal constrastive
coherence relations. PhD Dissertation, University of Nijmegen.
Sweetser, EveE.
1990 From etymology to pragmatics. Metaphorical and cultural aspects
Viewpoints and polysemy: Linking adversative
and causal meanings of discourse markers
Scott A. Schwenter
This article examines the use of a connective discourse marker in Spanish

(si), which can be used to convey both adversative and causal meaning, de-
pending on the discourse context. The semantic/pragmatic link between the
two uses of the connective, which are considered distinct, but overlapping,
polysemies, is mediated by viewpoint. The refutational meaning of the ad-
versative polysemy is paralleled by the high-scalar meaning of the causal
polysemy. This parallelism results from the transfer of dialogical features
from the adversative use, typically found in dialogues, to the causal use,
typically found in monologues. The study provides a general framework for
analyzing the role of viewpoint in cases of polysemy.
1. Introduction
The representation of more than one viewpoint in a single lexeme or con-

struction can be illustrated by a multitude of forms in English or any other
language. A prototypical example of how multiple viewpoints or "voices"
get encoded into linguistic structure can be seen clearly by looking at one
of the "e" relations forming the theme of this book: concession. In an
English concessive sentence like the following,
(1) Even though it rained, they played the baseball game.
what is strictly encompassed by the speaker's point of view is clear: the

rain did not affect whether the baseball game was played or not. But be-
yond this, another point of view is evoked which cannot necessarily be
ascribed to any particular individual, namely the expectation (based on
knowledge of "how the world works") that the rain would have been suf-
ficient cause for the baseball game not to be played. Though not explicitly
stated in (1), this expectation, and the point of view from which it arises,
is taken into account by the speaker, and grammatically indexed by the
concessive ('even though') construction.
In this paper, I intend to show that taking multiple viewpoints into ac-
count is a useful practice for establishing connections between distinct
258 Scott A. Schwenter
polysemies of individuallexemes. I will focus in particular on adversative

(contrastive) and causal polysemies of the Spanish discourse marker si,
which derives from the conditional marker si 'if (and is unrelated to to
the sf of affirmation meaning 'yes'; indeed, as will be shown below, its
meaning is often quite close to 'no'). Although I will be interested pri-
marily in the synchronic connections exhibited by these polysemies, such
an analysis is not necessarily restricted to synchrony. In fact, this kind of
analysis has been advanced for some time now with regard to the rise and
structure of conditional constructions, which are claimed to present in
condensed form (i.e. in a complex sentence) a discursive sequence con-
sisting of a (yes-no) question (speaker A), an implied affirmative answer
(speaker B), and a stated consequence following from B's answer
(speaker A) (Haiman 1978; Jespersen 1940).
Although the title of this paper makes explicit the connections to two
of the four "C" relations that make up the theme of this volume-contrast
(of which I take adversativity to be a subtype; cf Schwenter 1999a: Ch. 4
and below) and causality-the full story of what I am about to present
actually brings another of the "C" relations into the picture, namely condi-
tionality. I will focus most of my attention on discourse marker (DM)
uses of Spanish si, which, in its most canonical use, is a conditional
marker meaning 'if, marking "non-assertion" of the protasis, as in the
constructed example (2a). With other functions, however, it is found in
assertions with declarative sentence form as an adversative DM (2b), akin
to certain uses of 'but', or as a "justificatory" DM or connective (2c-d),
akin to certain uses of 'because' (the latter two often, but not always, with
exclamatory intonation):
(2a) Si traen el pastel, me 10 como.

'If they bring the cake, I'll eat it.'
(2b) A: jComete un poco de pastel!

'Eat some cake!'
~ B: jSi ya he comido!
'But I've already eaten!'
(2c) Date prisa. jSi se va el autobus!

'Hurry up. (Because) the bus is leaving!'
(2d) Juan no viene a la fiesta. Si su madre no 10 deja.

'Juan won't come to the party. (Because) his mother won't let
him.'
Viewpoints and polysemy 259
In the constructed examples (2b-d) si does not contribute to the pro-

positional content of the utterance-indeed, it could be removed without
affecting this content in the least-but rather signals how the utterance in
which it appears relates to the preceding utterance (cf Fraser 1988). In
this function, then, si is like many other DMs, which primarily express
procedural as opposed to conceptual meaning (Blakemore 1996) as well
as a "metatextual" comment on the text being constructed (Traugott
1999). In (neo-)Gricean terms, si in each of these examples conveys a
conventional implicature which is non-truth-conditional but also uncan-
celable; the specific semantic content of this conventional implicature, as
well as its pragmatic effects, are what I intend to flesh out in the discus-
sion below. Throughout the paper, I will take a neo-Gricean perspective
on implicature, and in particular on scalar implicature (cf Hom 1989,
among others).
The remainder of this paper is structured as follows. In section 2 I
provide an overview of the analytical framework used throughout. Sec-
tion 3 analyzes uses of si as an adversative DM. Section 4 analyzes uses
of si as a causal DM in two different contexts, exhortative and epistemic,
and summarizes the polysemies of DM si. Section 5 offers concluding
remarks, including ways of extending such an analysis to some English
forms. The data to be analyzed in the paper come principally from a 40-
hour corpus of colloquial conversational Spanish collected by the author
in Alicante, Spain, and indicated by [ALC]. The sources of other natu-
rally-occurring examples will be given in the text, and constructed exam-
ples will also be used when necessary.
2. Analytical framework
Let me begin by stating explicitly my position regarding the concepts of

contrast and adversativity. Contrast is not a strictly linguistic notion, but
rather a general cognitive ability linked to structures of expectation. I de-
fine contrast as "[t]he perception of difference, at some level and by any
means possible, between two entities comparable on some other dimen-
sion" (Schwenter 1999a: 126). Adversativity, on the other hand, is a
strictly linguistic notion. It involves "[c]ontrast between different points of
view as these are constructed in language use (Schwenter 1999a: 127,
emphasis in original). Contrast can be manifested indirectly in language,
for example in a sentence like John is short and he's a good basketball
player. Here, the conjunction does not explicitly signal contrast, but hear-
ers familiar with the link between height and basketball-playing ability
will perceive it nevertheless. Adversativity differs in that it typically pro-

vides explicit marking of clashes between different viewpoints, as in John
is short but he's a good basketball player, where the conjunction but
forces an interpretation in which the speaker asserts (supposed) incom-
patibility between shortness and basketball-playing ability.
Now, although there are a number of ways to conceptualize the inter-
play between viewpoints in language use and linguistic structure (for
some examples, see Ducrot 1984, 1996), in this paper I will utilize a
rather basic schema for analysis. Following Roulet (1984), I distinguish
between monologues (one physical speaker) and dialogues (two physical
speakers), as well as between monological (one "viewpoint") and dialogi-
cal (two "viewpoints") discourses. These two sets of distinctions can be
merged to yield four different combinations of speaker/viewpoint struc-
ture, as displayed in Table 1:
Table 1. Structure of discourse by speakers and "viewpoints"
No. of "Viewpoints"
No. of Speakers One Two
(Monolo~ical ) (Dialo~ical)
One (Monologue) Monologue/ Monologue/
Monological Dialo£ical
Two (Dialogue) Dialogue/ Dialogue/
Monological Dialogical
The distinction between monological and dialogical viewpoints in this

paper is closely tied to the argumentative function of utterances in their
discourse context, and in particular to whether the viewpoints are con-
strued as argumentatively parallel, i.e. oriented towards the same conclu-
sion, or as argumentatively opposing, i.e. oriented towards different con-
clusions. To illustrate what is meant by this, consider again the English
concessive sentence in (1) above. Here it is possible to isolate both the
stated viewpoint that the game was played despite the rain, and the under-
lying expectation that rain would cause the game to be cancelled. Thus, in
this case, a single speaker uttering (1) is simultaneously advancing two
viewpoints which are in some sense incompatible. Therefore, in accor-
dance with the schema in Table 1, a concessive sentence is a monologue,
yet expresses a dialogical viewpoint.
As the discussion below will make clear, the discreteness of the four
combinations of speakers and viewpoints illustrated in Table 1 is some-
what illusory. Instead, a gradient perspective on viewpoints appears nec-
essary, since the vestiges of indisputably dialogical interactions remain
palpable in meanings/functions that otherwise appear monological in

viewpoint. It is possible to determine, for instance, different degrees of
"addressee orientation" or, from another perspective, the degree to which
the meaning/function of a lexeme is based strictly in the perspective of the
speaker. This is not meant to imply, however, that the schema provided in
Table 1 is misguided. Indeed, it would have been much more difficult to
recognize the gradient nature of viewpoints without this starting point for
the analysis.
3. Si as an adversative DM
The adversative polysemy of si is typically found in dialogues, and is

always dialogical-it signals a confrontation of "incompatible" view-
points. The utterance introduced by si presents a "retort" (Montolio 1999)
to another speaker/viewpoint, and has a "refutational" (Schwenter 1998)
pragmatic value, indicating the exclusive relevance of the speaker view-
point. As a retort, the speaker's viewpoint is presented in the context of
another, competing viewpoint (often the immediately preceding utterance
of an interlocutor), and via inference excludes the competing viewpoint
from further consideration in the discourse.
To illustrate this process in action, consider the exchange in (3), taken
from the Valencia corpus of colloquial conversational Spanish (Briz [ed.]
1995). Here, a couple is engaged in a rather heated argument, and appear
to be on the verge of "breaking up" (§ = "latching" between two speak-
ers; / = .5 second pause):
(3) A: § mira yo te quiero// y cre- y creo que /0 SABES/// pero

NO/no puedo DEMOSTRARTELOJ 0 sea no no puedo dedi-
carte todo /0 que tit necesitas
'Look, I love youl/ and I thi- and I think that you KNOW it/II
but NO/ I can't DEMONSTRATE IT TO YOU, I mean I can't
give you everything that you need. '
~ B: pero si yo no te pido que me /0 demuestres
'but SI I'm not asking you to demonstrate it to me.'
[Briz (ed.): 1995: 83]
B's response specifically picks up on A's claim no puedo demostrarte/o

and contravenes an assumption inferrable from A's utterance: (A thinks
that) B wants A to demonstrate his love for B. In other words, what B is
refuting here is not any part of A's utterance per se, but rather his (prag-
matic) presupposition about what B wants him to do (viz. demonstrate his

love for B).
The adversative nature of these si-marked utterances is often rein-
forced by the adversative DM pero, as in (3). However, such reinforce-
ment is not necessary for the adversative meaning to be conveyed, and si
is often used on its own to signal refutation of a competing viewpoint. 1 In
(4) si is used by M to introduce a refutation of (part of) the propositional
content derivable from A's utterance:
(4) [A and M want advice for their next vacation]

A: S~ .ha viajado por toda Europa.
'S has traveled all over Europe. '
~ M: iSi no ha salido de Espana!
'SI she hasn't been out of Spain!'
(ALC)
Here, A's position that S has traveled extensively is directly denied by M

via negation. Since A's utterance also conversationally implicates that S
would be a good person to ask for advice, M's retort functions, via impli-
cature, to refute the validity of this implicit line of reasoning.
Semantic contradiction as in (4) is not a necessary feature of si-
marked utterances. The refutation involved may be based in considera-
tions of relevance (in the Gricean sense), as in the next example, where
the rhetorical argumentation introduced by si is presented as the only one
relevant to the discourse situation.
(5) [R is watching A drink orange juice]

R: A, no te bebas el zumo tan rapido, que te va a doler la
barriga.
'A, don't drink the juice so quickly, because your stomach
is going to hurt. '
~ A: Si tengo sed.
'SI I'm thirsty.'
R: Bueno, pues 10 que te de la gana.
'Fine, then (do) whatever you want.'
(ALC 7b.OI15.338)
The argumentative sequence of the first two turns in (5) can be summa-
rized as in (5'), where the parenthesized segments are conclusions left
implicit, but easily inferrable from the exchange.
(5') R: beber rapido ~ dolor de barriga (:. no beber rapido)

'drink quickly' ~ 'stomach ache' ( :. 'don't drink quickly')
A: tener sed ~ (heber rapido)
'be thirsty' ~ ('drink quickly')
It is important to notice that the utterer (=A) of the si-marked turn does
not dispute the "truth" of the proposition ("your stomach is going to
hurt"), but she does dispute its relevance and the relevance of the associ-
ated argumentative sequence within the discourse situation. Her response
to R signals the exclusive relevance of the "be thirsty" argument and, as a
consequence, justifies the conclusion "drink quickly".
Refutation with si may also be directed towards conversational impli-
catures that go beyond the literal meaning of the proposition, like the one
generated by A's utterance in (6):
(6) [B (age 6) looks strangely at a glass of a purple beverage]

B: i,Que es eso?
'What is that?'
A: Es el que te gusta a ti.
'It's the one (flavor) that you like.'
~ M: Si a ella Ie gustan todos los sabores.
'SI she likes all of the flavors. '
(ALC)
The cleft construction in A's utterance leads to an upper-bounded scalar

implicature from "it's the flavor that you like" to "it's the only flavor that
you like" (cf Atlas and Levinson 1981). M's si-marked utterance contra-
venes the upper bound implicated by A's utterance.
Finally, si can also head an utterance employed to object to more
strictly metalinguistic aspects of utterances, such as the pronunciation of a
word:
(7) A: Manana tengo que dar clase, sobre la GENEOLOGiA.

'Tomorrow I have to teach a class, about geneology.'
~ R: Si es GENEALOGiA.
'SI it's genealogy.'
(ALC)
R's objection in (7) is to the appropriateness of A's pronunciation of the

word genealogia 'geneology', not to any propositional content derivable
from A's utterance. The correction of A's pronunciation by R implicates

that A's rendering of the word was inappropriate.
As the preceding examples illustrate, si-marked utterances can intro-
duce a wide range of opposition types between two (or more) viewpoints.
This should not be surprising, since refutation as a discourse-pragmatic
strategy is not limited to denying and correcting the "truth" of prior
propositions. Like negation, it is also free to target "oppositions which are
created in the pragmatic aspects of utterance interpretation" (Ladusaw
1996: 323; cf Hom 1989), e.g. rhetorical argumentation, conversational
implicatures, and judgments of metalinguistic appropriateness.
3.1. "Hybrid" uses
The bulk of the examples of si presented in 3.1 above had an overwhelm-

ingly adversative flavor, since in each case two speakers were in "opposi-
tion" to each other. There are examples of other interactions which on the
surface appear to express a similar kind of adversativity, but on further
inspection reveal themselves to be subtly-yet crucially-different from
most of those seen above. The one exception among the preceding exam-
ples is (6), which is a "metapragmatic" (Schwenter 1999a) example ob-
jecting to a scalar implicature, but not to the propositional content or to
the argumentative orientation of the prior utterance. Examples like (6) are
therefore distinct because their interpretation is not so obviously constru-
able in strictly adversative terms, i. e. strictly as a case of opposition be-
tween (two) competing viewpoints. Another example of this type is (8):
(8) M: Creo que te han tintado MAL el pelo.

'I think they dyed your hair POORLY. '
~ B: iSi me 10 han tintado FATAL!
'SI they dyed it HORRIBLY!'
(ALe)
In (8), M makes a hedged assertion about the dye job done on B's hair,
and characterizes it as done mal 'poorly'. B' s response disputes this
characterization of the dye job; but instead of characterizing the dye job
positively, B strengthens the negative orientation already made explicit by
the adverb in M's assertion, stating that the dye job was donefatal 'hor-
ribly, lit. fatally,.2 Such an example expresses some degree of adversativ-
ity, since there is a clear correction of one speaker's evaluation by another
speaker's evaluation. At the same time, because the two speakers actually
AGREE on the overall implicated point-the dye job was not a good
one-the rhetorical purpose of each utterance in the exchange is parallel.
Here there is no ambiguity between a dialogical, adversative interpreta-
tion and a monological interpretation with parallel arguments. Rather,
both components of meaning combine to produce the overall interpreta-
tion of the example.
The adversativity expressed by a particular si-marked utterance can be
diminished even more when a "correction" of (some aspect of) an inter-
locutor's utterance cannot be identified:
(9) [Looking at a bicycle for a 4 year-old boy]

S: Mira, se puede bajar el asiento tambien.
'Look, you can lower the seat also.'
~ A: Si incluso se puede bajar mas.
,SI you can even lower it more. '
(ALe)
In the situational context of this example, S and A are inspecting the bi-
cycle, and S' s utterance is accompanied by the physical act of lowering
the bicycle seat a certain amount. A's utterance is in response to the (non-
locutionary) implication that the seat cannot be lowered beyond the point
to which S has lowered it. This is where the adversativity of the example
lies. Nevertheless, the contributions of both speakers are argumentatively
oriented in the same direction, each constituting positive assessments of
the bicycle in question, and implicating a conclusion like "we should buy
this bicycle" (which, in fact, they did). The co-oriented nature of the ut-
terances is made clear by the appearance of incluso 'even' in A's tum,
which marks the expressed proposition as a parallel, but stronger, argu-
ment for the aforementioned conclusion (which A took to be implicated
by what S said and did) (Kay 1990; Konig 1991).
The two preceding examples, along with (6) above, illustrate that,
even in dialogues, the degree of opposition marked by si is not fixed, but
sensitive to contextual factors, lexical relationships, etc. The "hybrid" na-
ture of each example is due primarily to the parallel argumentative orien-
tation of the asserted propositions: in (8) both the mal proposition and the
fatal proposition are directed towards a negative evaluation of B's hair,
while in (9) the propositions derivable from both utterances are being
used to evaluate the bicycle positively. The greater adversativity ex-
pressed in (8) as opposed to (9) results from the explicit correction of a
weaker predicate with a stronger one.
4. Si as a "causal" DM
4.1. "Exhortative" contexts
DM uses of si where it functions to signal "connection" between two ut-

terances that form part of a complex discursive act in monologue, such as
(2c) and (2d) above, can be divided into two classes. The first is com-
prised of a paratactic sequence of two utterances, normally separated by
period intonation. Specifically, this class consists of an utterance express-
ing a command or exhortation which is "motivated" by a following utter-
ance prefaced by si. The second class consists of a similar two-utterance
sequence linking a speaker's subjective assessment with an argument
serving to justify that assessment. The distinction between the two classes
corresponds roughly to the now well-known division between "speech-
act" and "epistemic" causality (Sweetser 1990). Hence from a pragmatic
perspective they are distinct and, following Sweetser, I consider them to
be pragmatically ambiguous. Both differ from adversative DM si in that
they cannot be prefaced by pero 'but'. However, they are also similar in
th~t both are typically expressed in paratactic structures. This is the norm
for "speech-act" and "epistemic" causals, which assert the content of
their clauses. Both types are moreover opposed to "content" ("real-
world" or "direct") causals, which presuppose the content of their clauses
and tend to display features of subordination (Couper-Kuhlen 1996, inter
alia). In this section I examine the first class of uses, where "exhortation"
is the central pragmatic act being performed.
To the extent that any uses of si like those in (2c) and (2d) above have
been considered in the literature, the explanation for them has been as-
similated to dialogue-based, adversative uses like those in §3 above. Con-
sider for instance the following naturally-occurring example, from a re-
cent article analyzing the functions of "independent" si:
(10) [To the disgusted husband whose favorite soccer team has just lost
three goals to none]
No te preocupes, jhombrel, si eljUtbol no es importante.
'Don't worry about it, man! SI football [=soccer] isn't important.'
(Porroche 1998: 236)
In her analysis of this example, Porroche points out that the target of the
husband-directed utterance is not an actual utterance produced by the
husband, but rather "a belief that ... forms part of the mental universe of
the speaker, but which, at times, can also be attributed to the interlocutor"
(Porroche 1998: 236~ my translation). The second part of this explanation

is, of course, implicitly recognizing the multiple viewpoints discernible in
these examples: the wife believes that her husband thinks soccer is (very)
important, and attempts to provide a correction to his viewpoint. How-
ever, at the same time, the segment introduced by si can also be under-
stood in a causal sense, since it presents a piece of motivation for the
negative imperative No te preocupes 'Don't worry', and, moreover, goes
on to provide a reason not to worry.
The causality involved in examples like (10) above or (12) and (13)
below thus concerns the relationship between a command/exhortation and
motivating material which provides "just cause" for what the command
requests. In this "exhortative" use, there exists a clear overlap between si
and discourse connective uses of que (lit. 'that') which perform a similar
function (cf Garcia 1996: 49-52)-indeed, que could have been used in
place of si in (10) above. Consider the constructed (near) minimal pair in
(11 ):3
(IIa) Come, que no te queda casi nada.

'Eat, QUE there's hardly anything left.'
(11 b) Come. Si no te queda casi nada.

'Eat. SI there's hardly anything left.'
Either of these utterances could be uttered by a parent to a child who

is still sitting at the table with food in front of her, long after the rest of the
family has finished eating. Both markers preface an asserted proposition
that functions pragmatically as an argument in favor of carrying out the
action demanded by the imperative come 'eat', and both could be sup-
pressed, i. e. the causal relationship between the imperative come and the
declarative sentence that follows is inferrable without any intervening
connective element. Pragmatically, the difference between the two utter-
ances is that the second one with si is more "emphatic"~ it would be in-
terpreted as a stronger (more persuasive) attempt on the parent's part to
get the child to finish the food than the first utterance with que, which
presents the "motivation for eating" in a more neutral way, pointing out to
the child the fact that she has little food remaining, and possibly softening
the command to eat (cf Pons 1998). Put somewhat differently, the sec-
ond utterance seems to assume a priori stronger opposition from the child
towards eating-it strives to take into account the child's perspective and
presents a corrective measure to that perspective. In short, it displays dia-
logical features. Thus, though the propositional content of the two exam-
pIes in (11) is equivalent, the version with si offers stronger motivation

for eating than the version with que.
Unlike most of the dialogue-based examples shown above, example
(11 b) would most likely not be interpreted as a reply to an explicit utter-
ance by an interlocutor. As a result, further reinforcement of si by the ad-
versative DMpero, a possibility in all of the adversative examples in §3,
would be odd in (11 b), unless the child had previously stated or implied
that there was an overabundance of food left on her plate (i. e. as justifica-
tion for not eating more). The (im)possibility of prefacing si-marked ut-
terances with pero thus constitutes a handy diagnostic for distinguishing
adversative uses (usually dialogues) from those with a more causal sense,
like the monologue example in (11 b).
A naturally-occurring example where eating was precisely at issue can
be seen in (12), where A (the mother) produces two utterances which
relate back to the command to eat in the first utterance:
(12) [Mother to her 18-month-old son, who is sitting at the dinner ta-
ble]
A: jVenga hombre! jCome!
'Come on man! Eat!'
A: te quedan dos bocaditos.
'there are two bites left. '
~ A: Si ya no te queda casi nada.
'SI there's hardly anything left. '
[ACA,3/21/99]
Here, both of A's final two utterances can be seen as trying to per-
suade the child to eat, or more accurately, to finish eating. Both of these
utterances can thus be considered "pragmatically subordinate" to the ini-
tial tum where the command is uttered; the si-marked utterance being the
"stronger" of the two. Indeed, it is interesting to note that if the proposi-
tional contents of A's latter two utterances are switched, the sequence
will become pragmatically anomalous, as shown in (12'):
(12') A: iVenga hombre! iCome!

A: Ya no te queda casi nada.
~ A: ?? Si te quedan dos bocaditos.
The reason for this is that the sequence introduced by si must be inter-
pretable not only as motivation for carrying out the action demanded by
the imperative, but also as the "strongest" piece of motivation among
those available. Here, the negative sentence explicitly minimizes the

quantity of food left, especially through the use of casi nada 'almost noth-
ing', thereby providing the strongest argument among those put forward
in favor of eating. By contrast, the sentence te quedan dos bocaditos
'there are two bites left' presents the quantity in positive terms, leaving
strictly to conversational inference the pragmatic connection between it
and the imperative come. What this type of example illustrates, then, is
that si in this use encodes a component of scalar meaning, in effect
imposing a ranking on propositions that are being used as motivation for
the action requested.
The dialogical nature of the preceding examples, though not linguisti-
cally explicit, does appear to be easily inferrable from the discourse con-
text. In each case, one can assume a reluctance or "opposition" on the
part of the addressee towards carrying out an action (e.g. eating) or, as in
(10), an emotional opposition to "not worrying" about the losing soccer
team. Moreover, the si-marked utterances are plainly addressee-directed,
since they attempt to motivate or persuade the addressee to carry out the
action specified in the command. But one might ask whether it is possible
to generalize this analysis to all "exhortative" examples. An example
which, at first glance, might not seem analyzable in this way is (13):
(13) B: jcono! aqui /e da e/ so/// jmecagiien /a put-!§

'Damn! the sun is shining here// son of a bi-!'
D: § i,ahora te enteras?/
((estam-)) una mierda ahi
'you just figured that out? «we-)) a piece of shit there'
~ C: mira! a/ Toni/ si /e sobra un pedazo dee- dee- de
hierba por ahi/ co/gando
'Look at Toni/51 he's got a piece of plant there/ hanging'
D: [dos dos]
'two two'
A: [i,a quien?J
'who?'
C: en e/ bocadi//o//parece que //evas unn§
'in the sandwich// it looks like you've got a'
A: § jcono! es- son ajos tiernos
'damn! it's- they're green garlic'
[Briz (ed.) 1995: 68]
Here there is no expressed opposition to the command 'look at Toni' ut-

tered by C. However, this utterance represents an obvious and abrupt
change in topic, as is evidenced by the exchange between B and D at the

beginning of the example. C's goal is to capture the other participants'
attention and redirect it towards Toni. It appears, then, that C's si-marked
utterance-presenting an argument in favor of looking at Toni-can also
be interpreted as attempting to overcome opposition, specifically as a way
of persuading the others to heed his command by refocusing their atten-
tion on Toni. Given the discussion above of si and que (which could re-
place si in [13]), C's choice of si to provide linkage between the com-
mand and the assertive motivation that follows is not arbitrary: the scalar
strength of si lends the propositional content a sense of "noteworthiness"
that que would not necessarily convey, and thereby makes a stronger re-
quest for the addressees' attention.
Because the illocutionary force of the two utterances linked by si in the
above examples differs (a command and an assertion related to that
command), it is tempting to assimilate them fully to the "speech-act" ex-
amples of Sweetser (1990). However there exists an important differ-
ence. Sweetser's speech-act examples consist of sentences in which it is
the act itself that is motivated or "enabled" by the causal clause, as in
What are you doing tonight? Because there's a good movie at the thea-
ter. In this sentence, the causal clause provides the motivation for per-
forming the speech act of asking the question-in other words, it provides
a "context of relevance" for the question. The "exhortative" examples
with si differ: they do not provide motivation for the speech act itself, but
rather for the action specified in the command (e.g. looking, eating). In-
stead of being based in considerations of speech-act relevance, their pur-
pose is fundamentally persuasive in nature.
4.2. Epistemic contexts
Like the exhortative examples seen in 4.1, in this section we will once
again consider discursive sequences consisting of two utterances with the
structure [VI + si V2]. Once again, there is a paratactic relationship be-
tween the two utterances, and in this relationship the si-marked utterance
V2 remains "pragmatically subordinate" to VI. Specifically, VI repre-
sents an inferential conclusion that the speaker presents as "warranted" or
"justified" by the content of an assertion in V2 (cf Schiffrin 1987: 202).
Because they refer to speaker-internal cognitive positions, these examples
are labeled "epistemic"; indeed, they appear to be good examples of cau-
sality in the epistemic domain, to use Sweetser's (1990) terminology.
Two representative examples of si used as a DM in epistemic contexts

can be seen in (14) and (15):
(14) [talking about a friend who has cellulitis]

Q: Se que todas las mujeres la tenemos.
'I know that all women have it. '
Q: Pero ella, es que es una cosa por de mas.
'But her, it's a thing that is too much. '
~ Q: Si encima Ie han dicho que caminara un poco para
reducirla.
'SI they even told her to walk a little to reduce it. '
A: Pero no.
'But she doesn't. '
[ALC]
(15) [A has been telling C how she likes the actor Mel Gibson]
A: Y a L tambien Ie gusta mucho.
,And L likes him a lot too. '
~ A: Si incluso Ie grabe yo una pelicula de el.
'SI I even recorded one of his movies for her.'
A: i,C6mo se Ilamaba?
'What was it called?'
C: i,EI hombre sin rostro?
'The Man without a Face?'
A: Si, esa es.
'Yes, that's it.'
[ALC]
In (14) Q evaluates her friend's cellulitis negatively as "a thing that is

too much". This evaluation is then justified by the si-marked utterance
that follows. Presumably the underlying reasoning involved is this: given
that doctors have told the friend to walk to reduce her cellulitis, it is rea-
sonable to conclude that the cellulitis is "too much". The force of this ar-
gument gains strength in the context of Q's initial assertion that "all
women have it", but despite that fact not all women require the interven-
tion of doctors. In (15) A's statement that her friend L likes Mel Gibson a
lot is justified by A's following assertion that she recorded one of Gib-
son's movies for L. Both examples show si in combination with scalar
additive particles (encima and incluso) which, in these contexts, can both
be translated as 'even'. The pragmatic force of such particles is to mark a
proposition as a strong-if not the strongest-argument for a contextual
conclusion (Kay 1990; Konig 1991; Schwenter 1999c), and to imply that
other, weaker, arguments for that same conclusion also exist. The combi-
nation of si + scalar particle thus signals very strong speaker commitment
towards an inferential conclusion, which, as in (14), is often a controver-
sial and/or critical assessment. Notice that, unlike the exhortative exam-
ples seen above, epistemic examples like (14) and (15) do not contain
explicitly addressee-directed utterances. To the extent that an opposing
viewpoint can be identified, or even inferred, it is construed in opposition
to the conclusion, not to any aspect of the justificatory utterance intro-
duced by si. As a result, it is in the speaker's best conversational interest
to support the conclusion to the greatest degree possible.
It is interesting to note that most of the examples of si introducing an
assertion that justifies (some aspect of) speaker viewpoint, as in (14) and
(15) above, are functionally quite similar to the epistemic use of the
causal connective porque 'because'. Indeed, porque could replace si in
(14) and (15) without affecting the propositional content in the least, with
the proviso that there would typically be a shorter intonation break
(comma intonation) between the connected segments. The most salient
semantic/pragmatic difference between the two markers is that speakers
intuit a tighter "fit" between the content of the two segments with si so
that, for instance, (14) would be more likely used in a situation in which
the speaker felt that there already was or possibly would be opposition to
her opinion ("her cellulitis is too much") stated in the first segment. The
use of si instead of porque thus tends to reflect emphatic epistemic com-
mitment to the truth of the conclusion. While it is possible that porque
could be used in such a context, the relevant point regarding the epistemic
porque construction is that it has a wider domain of application than, and
is pragmatically unmarked with respect to (cf Lambrecht 1994: 16-17),
the epistemic si construction.
Further evidence for this particular analysis comes from the two con-
nectives' distinct co-occurrence possibilities with epistemic expressions
like me parece 'It seems to me' or creo 'I think.' These epistemic ex-
pressions function to attenuate speaker commitment to the truth of the
propositional content of an assertion. While such expressions seem com-
pletely natural with porque, they are by contrast pragmatically bizarre
with si, as the following examples illustrate (one might imagine these ut-
tered by a speaker A in a context in which another interlocutor B assumes
that Juan is coming to their party that same night):
(16a) (Me parece que/Creo que) Juan esta en(ermo, porque 10

he visto en el medico.
'(It seems to me thatlI think that) Juan is sick, because I
saw him at the doctor's office. '
(16b) (*Me parece que/*Creo que) Juan esta enfermo. Si 10 he

visto en el medico.
'(It seems to me thatII think that) Juan is sick. SI I saw
him at the doctor's office. '
Likewise, the well-known epistemic use of Spanish synthetic future

morphology to convey probability or inferred certainty (like English
That'll be John calling upon hearing the phone ring) is wholly compatible
withporque, but not with si:
(17a) Juan estara enfermo, porque 10 he visto en el medico.

'Juan must be (lit. will be) sick, because I saw him at the doctor's
office. '
(17b) *Juan estara enfermo. Si 10 he visto en el medico.

'Juan must be (lit. will be) sick. SI I saw him at the doctor's of-
fice. '
The principal explanation behind these patterns of pragmatic accept-

ability appears to reside in the strong degree of epistemic commitment
that si expresses, not only to the asserted segment that it introduces
(whose content is usually not under dispute), but also, and principally, to
the justificatory relationship between that segment and another segment
that expresses a potentially more controversial speaker position. In this
use, si is "modally harmonic" (cf Bybee, Perkins, and Pagliuca 1994),5
i. e. in epistemic concord, with non-attenuated forms like the indicative
mood, but is clearly not harmonic with markers of lessened epistemic
commitment like those seen in (16) and (17). By contrast, porque carries
fewer restrictions with respect to strength of speaker epistemic commit-
ment it can appear felicitously in both epistemically attenuated and non-
6
attenuated environments.
As a causal DM in epistemic contexts, then, si expresses justification
of an emphatic sort, as shown by its scalar contrast with porque. The co-
occurrence of si with scalar additive particles as in (14) and (15) above is
therefore not surprising, as these too are informationally emphatic forms.
Note however that it would be misleading to ascribe an additive meaning

to si, as is made clear by (18):
(18) R: "Que vas a hacer con la cortina?

'What are you going to do with the curtain?'
A: Pues arreglarla.
'Well fix it.'
A: ;,No ves que esta rota?
'Don't you see that it's ripped?'
[A takes the curtain and shows the ripped part to R]
A: Sera muyfacil.
'It'll be very easy. '
~ A: Si total, nada mas es cuestion de coserle esta parte de aqui.
'SI after all, it's only a question of sewing up this part here. '
[ALe]
Here A's inferential conclusion that fixing the curtain "will be very
easy" is warranted by the si-marked assertion that follows. The DM total,
which is most plausibly translated in (18) as 'after all', indicates that the
following proposition represents the only argument relevant to the conclu-
sion. And nada mas 'only, lit. nothing more' further minimizes the de-
scription of the effort involved in fixing the curtain. A's argument for how
easy fixing the curtain will be, then, constitutes the only argument for
coming to this conclusion, i.e. si appears in a non-additive context. All of
these markers thus work in tandem to justify the conclusion, which "fol-
lows" from the assertion in A's fmal tum.
The exhortative examples in the preceding section included a strategy
of lexical minimization (e.g. through negation) in a si-marked utterance as
motivation for carrying out the action specified in an imperative-mood
command. But as the preceding example shows, such minimization is not
limited to exhortative examples-epistemic examples display it also. An-
other epistemic example that uses minimization in support of a previously
stated conclusion is (19). Here, Y is attempting to justify her opinion,
stated at the outset, that the man she's talking about doesn't pay attention
to anyone:
(19) [Y describing a man that she knows]

Y: Es que el, es una persona que no Ie hace caso a nadie.
'it's that him, he's a person who doesn't pay attention to
anyone.'
~ Y: Si una vez fuimos a su casa y se qued6 ahi sentado sin decir

nada ... ini se movi6!
'SI once we went to his house and he just sat there without
saying anything ... he didn't even move"
[ALe]
The negative words in Y's second utterance mark an absolute endpoint:

nada 'nothing' and ni 'not even'. They are used in service ofY's conclu-
sion that the man doesn't pay attention to anyone, and run counter to ex-
pectations that Y holds regarding another person's behavior when in the
presence of visitors in that person's home. In addition, the clause ini se
movi6! 'he didn't even move" licenses an inference about other things
the man refuses to do when in the company of others, further sanctioning
the conclusion "he doesn't pay attention to anyone".
The connection between all the examples in this section is that in each
case it is strongly implicated that (i) the si-marked justification offered for
the conclusion is the strongest available, and (ii) the conclusion simply
follows, as a matter of course, from this justification.
4.3. Summary
Implicit in the organization of this paper has been the position that the
adversative DM si and the causal DM si are distinct polysemies of the
7
same lexeme. This is supported on discourse-structural grounds: the ad-
versative DM presents, within a dialogue, a dialogical "retort" directed at
an interlocutor's utterance, while the causal DM introduces an assertion
which provides motivation/justification for a preceding utterance by the
same speaker. By contrast, the exhortative and epistemic uses of the
causal DM are not distinct polysemies but rather offer a case of prag-
matic ambiguity (Hom 1989; Sweetser 1990): what si is expressing in
each use is the same semantic content, the interpretation of which differs
due to the pragmatic domain of application. The two polysemies also dif-
fer with respect to collocational properties: perhaps most importantly,
only the adversative DM can be prefaced, perhaps in every case, by the
adversative DM pero 'but', which reinforces the contrast expressed by si
(Montolio 1999). Utterance-initial pero is not found with either sub-type
8
of causal DM si.
Despite these distinctions, however, it should also be obvious from the
foregoing discussion that the boundaries between each use of si as a DM
are not clear-cut, and that the strong dialogical nature of the adversative
examples endures, albeit to a lesser extent, in the causal examples, where

it is now partially interpretable as an aspect of scalar meaning. Indeed,
one can arrange the uses presented above in relative fashion with respect
to the degree to which they are dialogical or monological. This is shown
in the table below:
Table 2. Uses ofDM si, by viewpoint
Dialogical Monological
Adversative +++
Exhortative ++ +
Epistemic + ++
This display helps illustrate what has been stressed above: exhortative
and epistemic uses of si do not lack dialogical features completely. The
importance of these features is somewhat diminished in these uses, but
paradigmatic contrasts with other forms like porque still indicate a dia-
logical residue: a more salient feature now is high scalar commitment to a
conclusion, as the examples in 4.2 illustrate. The same cannot be said for
adversative uses of si, which always appear to be exclusively dialogical
(even in monologue examples, see Montolio 1999). The greater dialogical
nature of exhortative uses of DM si, as compared with the epistemic uses
in Table 2, results from the addressee-directed characteristics of the for-
mer-most importantly, the assumption of strong opposition on the part
of the addressee-which are exhibited to a lesser extent by the epistemic
uses.
5. Conclusion
An obvious implication of the analysis presented in this paper is that there

is an underlying affinity between (some types of) adversativity and (some
types of) causality. This affinity seems commonsensical, especially if one
considers a pragmatic function/strategy like "justification" to be one that
speakers employ when they sense that their positions may potentially be
in dispute. But it also shows that certain pragmatic relationships, for in-
stance, refutation, which hold between (strictly) dialogical viewpoints,
may have as their (relatively more) monological counterparts other kinds
of pragmatic relationship, like justification. Thus, while it is not the case
that all types of adversative expression have a counterpart in the domain
of causality, the meaning conveyed by Spanish si in the first conceptual

domain does have an analogue in the second.
The analysis also shows that the distinction between dialogical and
monological viewpoints is not absolute, but rather gradient. The meaning
associated with adversative DM si, especially its function as a marker of
often vehement opposition between viewpoints, is carried over into the
strong scalar meaning of causal uses of si, which, I have argued, conserve
aspects of the dialogical features characteristic of adversative uses. A
strong type of adversativity, namely refutation, is mirrored by the high-
scalar argumentative strength expressed by causal si, which, indeed,
strives to present "irrefutable" justification for speaker-based epistemic
conclusions. Put more generally, then, dialogical "footprints" can be
found in lexemes and constructions which otherwise appear similar to
their more strictly monological counterparts. The gradience revealed here
between types of viewpoint parallels the gradience among the conceptual
domains which are under investigation in this volume (cf Konig 1986;
Kortmann 1997).
Though still under-utilized as an analytical device, the framework em-
ployed in this paper can and has been extended to the study of other lex-
emes. For instance, Traugott (1997: 6) argues th~t the historically distinct
uses of English after all are linked by "dialogic use, whether in dialog or
monolog". In addition, adversative and additive uses of English infact, as
in (20) and (21), respectively, are also amenable to this type of analysis:
(20) They think he's crazy, (but) in fact he's quite sane.
(21) That movie was bad, in fact it was terrible.
These uses, which arise at different times diachronically (Schwenter and

Traugott 2000), can be distinguished synchronically by the dialogical
viewpoint in the adversative example and the monological viewpoint in
the additive example (Schwenter 1999a). Many other differences be-
tween the two types of in fact can also be related to this difference in
viewpoint: the greater bleaching of the nounfact in (21), the sense of "re-
formulation" in the same example, and the epistemic scale in (20) versus
the "rhetorical" or "argumentative" scale in (21).
Finally, even lexemes whose different uses do not appear to constitute
separate polysemies convey nuances of meaning which are sensitive to
the number of viewpoints involved. This is the case, for instance, of epis-
temic sentence adverbs like certainly in dialogues like these:
(22a) A: John isn't smart.

B: He certainly is!
(22b) A: John sure is smart.

B: He certainly is!
In both examples the main semantic contribution made by certainly is to

signal and reinforce speaker B' s full epistemic commitment to the propo-
sition "John is smart". Because of this semantic parallelism, it does not
seem plausible to consider the two uses to be different polysemies. None-
theless, as a result of the distinct argumentative orientation of B' s utter-
ance with respect to A's in each case-anti-oriented in (22a) vs. co-
oriented in (22b)-it is feasible to identify certainly as a dialogical prag-
matic marker of disagreement in (22a) but as a monological marker of
agreement in (22b).
Notes
1. See Schwenter (1999b) and Montolio (1999) for discussion of the pero si
combination.
2. The scalar relationship between the two adverbs in (8) is easily demon-
strated. Consider the results when they are inserted into a syntactic frame of
the type "X but not Y", where X represents the "weaker" adverb, y the
"stronger" one:
(ia) Te 10 han tintado mal, pero no fatal.
'They dyed it badly, but not horribly.'
(ib) *Te 10 han tintado fatal, pero no mal.
'They dyed it horribly, but not badly.'
The infelicity of (ib) results from the fact that any speaker of Spanish would
order these adverbs as <fatal, mal> along a semantic ("Horn") scale, such
thatfatal semantically entails mal, but not vice-versa.
3. The intonational break between the imperative and the utterance segment
justifying it tends to be longer when si introduces the latter. This is reflected
by the comma intonation in (11a) as opposed to the period in (11b).
4. Note too that the translation equivalent of because in Spanish, porque, could
not replace si in the examples seen above. In fact, upon translating examples
(10) through (13) into English I am unable to replace si with any English
connective form.
5. My use of "modally harmonic" with respect to discourse markers extends the
term beyond that of the scholars cited, who limit it to cases involving a mo-
dal verb and other devices like adverbs.
6. The differences between this connective use of si and epistemic porque are
very similar to those separating discourse marker uses of English after all
and epistemic because (Traugott 1997). Nevertheless, after all and si in

these uses are not always possible translations of each other.
7. In Schwenter (1999b), I have pointed out the extreme similarity between si
and the Japanese connective datte, which can be translated as either 'but' or
'because', depending on its context of use (Mori 1994), and which "by itself
indicates disagreement" (Ford and Mori 1994: 56). Unfortunately, these au-
thors do not tackle the issue of whether the separate uses of datte constitute
distinct polysemies of a single lexeme.
8. Another potential difference not mentioned in this paper is the intonation
pattern of each use. According to Montolio (1999), utterances introduced by
adversative DM si are characterized by a rise-fall pattern with strong stress
on the final tonic syllable. In my data from Alicante, this pattern is found
mainly with adversative examples in which si is reinforced by pero, and
therefore could be interpreted as an iconic reflection of the speaker's
strengthened adversative stance. But whatever the case may be for the adver-
sative DM, it appears that causal OM si does not occur with this unique in-
tonation pattern, though at times increased loudness (as in an exclamatory
utterance) can be found.
References
Atlas, Jay David, and Stephen C. Levinson

1981 It-clefts, informativeness, and logical form: radical pragmatics (re-
vised standard version). In: Peter Cole (ed.), Radical pragmatics, 1-
61. New York: Academic Press.
Blakemore, Diane
1996 Are apposition markers discourse markers? Journal of Linguistics
32: 325-47.
Briz, Antonio (ed.)
1995 La conversacion coloquial (Materiales para su estudio). (Cuadernos
de Filologia 16.) Valencia: Universitat de Valencia.
Bybee, Joan, Revere Perkins, and William Pagliuca
1994 The evolution ofgrammar. Chicago: University of Chicago Press.
1996 Intonation and clause combining in discourse: the case of because.
Ducrot, Oswald
1984 Le dire et Ie dit. Paris: Minuit.
1996 Slovenian lectures/Conferences Slovenes: Argumentative seman-
tics/Semantique argumentative. Ljubljana: ISH.
Ford, Cecilia, and Junko Mori
1994 Causal markers in Japanese and English conversations: A cross-
linguistic study of interactional grammar. Pragmatics 4: 31-61.
Fraser, Bruce
1988 Types of English discourse markers. Acta Linguistica Hungarica
38: 19-33.
Garcia, Serafina
1996 Las expresiones causales y finales. Madrid: ArcolLibros.
Haiman, John
Horn, Laurence R.
1989 A natural history ofnegation. Chicago: University of Chicago Press.
Jespersen, Otto
1940 A modern English grammar on historical principles, Volume 5:
Syntax. London: George Allen and Unwin.
Kay, Paul
1990 Even. Linguistics and Philosophy 13: 59-111.
Konig, Ekkehard
1986 Conditionals, concessive conditionals and concessives: Areas of
contrast, overlap and neutralization. In: Elizabeth Closs Traugott,
Alice ter Meulen, Judy Snitzer Reilly, and Charles A. Ferguson
(eds.), On conditionals, 229-46. Cambridge: Cambridge University
Press.
1991 The meaning offocus particles: A comparative perspective. Lon-
don: Routledge.
Kortmann, Bernd
1997 Adverbial subordination: A typology and history of adverbial sub-
ordinators based on European languages. Berlin: Mouton de
Gruyter.
Ladusaw, William A.
1996 Negation and polarity items. In: Shalom Lappin (ed.), The hand-
book ofcontemporary semantic theory, 321-41. Oxford: Blackwell.
Lambrecht, Knud
1994 Information structure and sentence form. Cambridge: Cambridge
University Press.
Montolio, Estrella
1999 iSi nunca he dicho que estuviera enamorada de ell Sobre construc-
ciones independientes introducidas por si con valor replicativo. Ora-
lia 2: 37-69.
Mori, Junko
1994 Functions of the connective datte in Japanese conversation. In: No-
riko Akatsuka (ed.), Japanese/Korean Linguistics, Vol. 4, 147-63.
Stanford: CSLI Publications.
Pons Borderia, Salvador
1998 Conexi6n y conectores: estudio de su relaci6n en el registro infor-
mal de la lengua. (Cuadernos de Filologia 27.) Valencia: Universi-
tat de Valencia.
Porroche Ballesteros, Margarita
1998 Sobre algunos usos de que, si yes que como marcadores discursivos.
In: Ma Antonia Martin Zorraquino and Estrella Montolio Duran
(eds.), Los marcadores del discurso: Teoria y analisis, 229-42. Ma-
drid: ArcolLibros.
Roulet, Eddy
1984 Speech acts, discourse structure, and pragmatic connectives. Jour-
nal ofPragmatics 8: 31-47.
Schiffrin, Deborah
1987 Discourse markers. Cambridge: Cambridge University Press.
Schwenter, Scott A.
1998 From hypothetical to factual and beyond: refutational si-clauses in
Spanish conversation. In: Jean-Pierre Koenig (ed.), Discourse and
cognition: Bridging the gap, 423-35. Stanford: CSLI Publications.
1999a The pragmatics of conditional marking: Implicature, scalarity, and
exclusivity. (Outstanding Dissertations in Linguistics.) New York:
Garland.
1999b Discourse markers and the PAISN distinction. Ms., The Ohio State
University.
1999c Two types of scalar particles: evidence from Spanish. Advances in
Hispanic linguistics. In: Javier Gutierrez-Rexach and Fernando
Martinez-Gil (eds.), 546-61. Somerville, MA: Cascadilla Press.
Schwenter, Scott A. and Elizabeth Closs Traugott
2000 Invoking scalarity: the development of in fact. Journal ofHistorical
Pragmatics 1: 7-25.
Sweetser, Eve E.
Press.
1997 The discourse connective after all: A historical pragmatic account.
Paper presented at the 16th International Congress of Linguists,
Paris, July.
1999 The rhetoric of counter-expectation in semantic change: A study in
subjectification. In: Andreas Blank and Peter Koch (eds.), Histori-
cal semantics and cognition, 177-196. Berlin: Mouton de Gruyter.
The treatment of contrasts in interaction *
Cecilia E. Ford
Relying on data from natural conversations and emphasizing the on-going

analyses afforded by speakers as they respond to one another, this study ex-
plores the functions of contrasts in interaction. Considering contrast as
broadly including neutral contrast, concession, and antithesis, and recogniz-
ing that contrasts can be manifested as disagreements across interlocutors, a
recurrent rhetorical pattern is found wherein contrasts are followed by ex-
planations or solutions. Furthermore, cases of contrast deployed without ex-
planation or solution are found to be followed by talk in which interlocutors
treat the contrast as requiring further elaboration. Unelaborated contrasts are
also found to be used in a variety of ways related to activity and speaker
role. The methodology is an adaptation of conversation analysis, with the
talk of interactants themselves-specifically talk following contrasts-used
as a primary resource for interpreting the functions of the rhetorical patterns
under scrutiny.
1. Introduction
Documenting the recurrent resources and practices of language use is a

primary goal for researchers in both discourse-functional linguistics (for
recent reviews see Cumming and Ono 1997; Tomlin et al. 1997; Hopper
1998, among others) and conversation analysis (see Atkinson and Heri-
tage 1984; ten Have and Psathas 1990; Whalen 1992, among others).
Understanding linguistic structures and how such structures are com-
bined in use, is fundamental to the study of language development, lan-
guage change, and language as social action. As part of a functional re-
search program focusing on language in use, the present study examines
the realization of a rhetorical relation which is recurrently deployed in
spoken interaction: the relation of contrast. I explore how speakers-
participants in interaction-treat contrasts through their talk directly
following the textual relation, with attention to how that subsequent talk
both builds on and interprets the functions of the contrasts themselves.
My attention will thus be on the functions of contrasts as they are elabo-
rated on and responded to by participants in face-to-face interaction.
This method of inquiry is an adaptation of the conversation analytic
284 Cecilia E. Ford
practice of relying on the talk of interactants as a primary source for the

analysis of functions of talk-in-interaction (see Petrakyla 1997).
For the purposes of this study, a contrast consists of two text spans
which present an opposition. This type of textual relation subsumes the
definitions of contrast, antithesis, and concession in Rhetorical Structure
Theory (RST) (Mann and Thompson 1988). Included in my collection
are cases where two parts of a text, generally (but not exclusively)
clauses, are presented as entities, actions, or propositions which could
be:
(a) comprehended as the same in many respects

(b) comprehended as differing in a few respects
(c) compared with respect to one or more of these differences
(Mann and Thompson 1988)1
A feature of RST which makes it especially useful for the broad study of
rhetorical relations is that it allows for relations to be either explicitly
marked or inferable based on the content of text spans in question; thus
no explicit linguistic marking is required for the interpretation of two
spans as being in an oppositional relation. Furthermore, RST places no
fixed limit on the size of text spans to be related, though in practice the
lowest unit has often been the clause.
Example (1) offers an initial sense of the type of pattern I will be ex-
amining. The quote is from a magazine interview, and the speaker is a
student who organized a protest at her high school. She is describing the
effort it took:
(1) (From Ford 1994)
<" lIt sounds eas~,1

CONTRAST ~
!but i t wasn' ~
EXPLANATION
,..--------------------------,
'cause pep assembly's a real big deal,
especially at homecomdng; the whole school
gathers
The first part of the contrast ("It sounds easy") explicitly refers to an
assessment that the interviewer (or the magazine readers) could poten-
The treatment ofcontrasts in interaction 285
tially make;2 the second part of the contrast ("but it wasn't") stands in
opposition to the first. What follows the contrast is an explanation of the
opposition: the speaker explains that the setting for the protest was a
particularly important and consequential one. Here then, through her
subsequent talk, the speaker shows her understanding of the contrast as
presenting a problem needing resolution, and the resolution she provides
is an explanation supporting one pole of the contrast. At one level of
rhetorical analysis, we see that the incompatibility between the parts of
the contrast is addressed and resolved through the explanation.
At another level, when we take these rhetorical moves in their con-
text we see a contrast plus explanation doing particular persuasive work.
A contrast, here, functions as a format for presenting two perspectives,
of which the speaker identifies strongly with only one. Through this rhe-
torical combination, the speaker efficiently acknowledges a potential
incompatibility but builds support for her own perspective as she adds
an explanation for one pole of the contrast.
In this paper, I examine the functions of contrasts in interaction, with
a specific focus on the relationship between a contrast and the talk
which follows. While the rhetorical relation of contrast is a format for
the expression of diverse actions, a pattern is evident: across contexts,
contrasts are treated as needing resolution in subsequent talk; that is,
they are regular building blocks for larger actions involving explana-
tions or solutions for problematic states of affairs. In the data I have ex-
amined for this study, there is indeed a pattern of explanation and solu-
tion relations following contrasts; however, as we will see, it is also evi-
dent that the larger context of activity and the role of a speaker in an
activity are mediating factors affecting how and whether further resolv-
ing talk is treated as relevant after a contrast.
The remainder of the paper will be organized as follows: In section 2,
I describe my data sources; section 3 offers some examples of contrasts
in spoken English. Section 4 sketches the lexical and grammatical en-
coding devices associated with contrasts in the present data. In 5 and 6,
bearing in mind that the social and sequential context of any piece of
talk is central to the formulation and interpretation of what that talk is
doing, I discuss the rhetorical combinations found in the data:
CONTRAST + EXPLANATION and CONTRAST + SOLUTION. In
section 7, I examine cases where contrasts are not followed by explana-
tions or solutions. I discuss relationships between activity type, speaker
role, and the location of the contrasting action in an interactional se-
quence in relation to the absence of further talk in these cases. Section 8
286 Cecilia E. Ford
concludes the paper with a general review and discussion of the find-
Ings.
2. Data sources
Two different videotaped conversations provide the primary data for

this study; both involve spontaneous talk-in-interaction. One is in an
institutional setting and includes a science teacher and a class of high
school students. The other source is an evening gathering of five female
friends in a home setting.
The science class, in an urban U.S. public high school, was video-
taped twice a week during the course of a semester (Spring 1997) as part
of a broader longitudinal study of language socialization among diverse
students. 3 I have concentrated on the interaction in one 50-minute pe-
riod. The class hour involves pre-class interactions before the official
bell rings, a teacher-fronted segment (including many student contribu-
tions), and a laboratory activity during which one group of four female
students was videotaped. 4 The recording comes from the second semes-
ter of a two-semester course, so the students and teacher are quite famil-
iar with one another.
The second source is a 20-minute segment from a three-hour video-
taped gathering, involving five women between the ages of 30 and 46
(taped in December 1995). The women are familiar with each other to
varying degrees: Terry (T) and Pam (P) are domestic partners, Abby (A)
is a friend of theirs, and Rachel (R) is more of an acquaintance of T and
P. R was invited by a friend ofT and P's (Stacy), who is out at the store
during this segment of talk.
3. Contrasts in conversation
In spoken language, contrasts have been previously documented in a

variety of uses. A contrast can be produced as part of a more planned
public address, as part of a longer turn in informal interaction, and
across speakers in the expression of disagreement.
Contrastive combinations are treated as significant units of talk in
public speaking events. Audiences orient to contrasts by responding to
such pairings with applause or other forms of group action (Atkinson
1984):
(2) simp~i£ied from Atkinson (1984) Xs app~ause; =

the contrasting units are bracketed on the ~e£t5
Heath: ~owo the la:bour:: (0.4) prime mdnister and

his collea:ques are boasting in this election
campaign (0.7) that they have brought
infla:tion down from the disah:strous level
CONTRAST=l-
r twenty six percent (1.4)
But we are: entitled to inquire
(0 • 4)
Who put i t up to twenty six percent=
Audience: hehh[eh
Audience: [x-xxXXXXXXXXXXXXXX ...
Contrasts are produced by single speakers either as explicitly oppos-

ing propositions, or as negations of presupposed but not explicitly stated
information, as in (3), from the data:
(3) A teacher is handing out papers £or an assignment
T: This thing that you're getting now, (.) the maze?

?: ( )
T: Okay?
( .)
=>T: You don't have to turn this in.
( .)
T: Listening?
( .)
T: I'm gonna give you this 'cause you're gonna write
on it,
To understand this example, one needs to know that it is the custom in

this class that written assignments which the teacher distributes be
turned in once they are finished. In this instance, the teacher explicitly
tells the students that this sheet with a maze on it is an exception. At the
arrow, the teacher produces a clause which references, through nega-
tion, a corresponding affirmative expectation; that is, in stating that the
students "don't have to turn this in", the teacher indexes a shared under-
288 Cecilia E. Ford
standing in this class that assignment sheets in general are to be returned

to the instructor.
Contrast may also occur within the matrix of an interactional se-
quence, as in example (4). During a laboratory activity involving can-
dles, the lights in the classroom have been turned off, but at this point
most of the students have fmished the task:
(4) ("I'monna" = 'I am going to')
T: PEOPLE. I-I THINK MOST OF YOU ARE DONE WITH THE

CANDLES,= I'MONNA PUT THE LIGHTS ON.
(0.3)
::) K: WE PREFER THEM OFF.
- -
::) T: You prefer them aw- I prefer them on, I can see:.
These examples illustrate the occurrence of contrasts in spoken lan-

guage. In the case of audience response, in particular, prior research
supports an interpretation of these complex units as real and relevant to
the participants themselves.
4. Forms of contrast
For the current study, I collected 50 instances of what could broadly be

interpreted as utterances and utterance sequences expressing contrasts,
with reference to the Rhetorical Structure Theory definitions of contrast,
antithesis, and concession, as discussed in section 1. In addition to in-
stances where both poles of a contrast were expressed, I also collected
cases where a negative clause implicitly indexed a corresponding af-
firmative, as in example (3) above. Cases of negation can be understood
as responding to or constructing a context in which the corresponding
affirmative is relevant or assumed (Giv6n 1978; Tannen 1979; Fair-
clough 1989).
The collection also includes disagreements across speakers as a form
of contrast, as in (4):
( 4) (repea ted)
T: PEOPLE. I-I THINK MOST OF YOU ARE DONE WITH THE

CANDLES,=_I'MONNA PUT THE LIGHTS ~.
?: ( (O.~) ) 7-
K: [WE PREFER THEM IOF~~
T: You prefer them aw- I prefer them~, I can see:.
The second explanation is embedded in the first: as can be seen in (4),

the linguistic markers of contrast in this collection are in line with what
has been described in the work of Halliday and Hasan (1976), Martin
(1993), and Rudolph (1996), and, for discourse in science classes,
Lemke (1993). These markers include a variety of lexical and gram-
matical strategies and devices for expressing dissimilarity, including
antonymy (in adjectives, nouns, and prepositions), negation, compara-
tives, and conjunctions. Adverbs associated with contrasts were actu-
ally, really, and only. Temporal oppositions such as then versus still
were also represented. Finally, prosodic highlighting (contrastive stress,
pitch and rhythmic parallelism) was a frequent feature in contrasts.
The defining criterion for contrasts in this study was that incompati-
bility be expressed, whether in content, through linguistic marking, or
by both. The collection contains cases of contrasting clauses, phrases,
and words; and contrasts may be produced by single speakers or across
speakers, with negation and disagreement included.
In sections 5 and 6 I provide analyses of the two types of rhetorical
combination that form the majority of cases in my data: CONTRAST +
EXPLANATION and CONTRAST + SOLUTION. The cases are di-
verse with respect to the functions of the contrasts, i. e., the presentation
of the incompatibility or opposition itself, serves different interactional
ends. However, in common across cases, we fmd participants treating
contrasts as presenting problems in need of explanation or solution.
5. Contrast + Explanation
In my data base, a pattern was evident in which contrasts were followed

by explanations. The textual relation I am calling explanation corre-
sponds to what, in an early form of Rhetorical Structure Theory, was
categorized as REASON relation. As defined by Fox (1987), this textual
290 Cecilia E. Ford
relation involves one span "which makes a statement about something"

and another span "which provides the reason for that statement (either
the reason for making it or the reason for it being so)" (1987:82). In my
data the statement for which a reason is provided involves a contrast.
The rhetorical cluster of CONTRAST + EXPLANATION is produced
in a variety of different interactional sequences, within a single
speaker's talk and across speakers.
An initial example comes from a teacher-dominated portion of the
science class. In (5) the science teacher is involved in a lecture, which
includes prompts for student responses. In this excerpt, the teacher pro-
duces two contrasts and two explanations.
(5)
T: As I move away, (.) does my height get
different?=
CONTRAST (1) ~ T: =No.
?: 0No. 0
EXPLANA- (1):::> T: I look smaller,because I'm farther away,
TION' I I I
CONTRAST (2) EXPLANATION (2)
First, the teacher answers his own question, "does my height get differ-
ent?" with the contrasting "No." Next, he explains this negation by con-
ceding that while there is an illusion of getting small, it is only because
of the added distance. And within that explanation, he introduces yet
another contrast, the contrast between his original denial of difference
and his subsequent concession that there is an appearance of difference.
This contrast is highlighted by the stress on look (as opposed to actually
being different). As with the first contrast, the second is followed by the
explanation, "because I'm farther away."
The contrasts in (5) are vehicles through which the science teacher
highlights expectations and assumptions, thus leading the students
through a discovery process. The first contrast draws attention to the
fact that there is an incompatibility between the image and the object
being reflected. This highlighted contrast then leads into the explana-
tion. Within that explanation is the more subtle contrast between ap-
pearance and reality, indexed through the contrastive stress on look.
And again, the opposition highlighted through the contrast leads into an
explanation component.
The conversation recorded in a home setting also contains instances

of the CONTRAST + EXPLANATION cluster. The exchange in (6)
takes place as A, R and T have been playing in the board game Piction-
ary. R remarks on the striking difference between where T's team was
located just a short while earlier and where they are now-far ahead on
the gameboard relative to R's team. As an explanation, T suggests that it
was the participation of Stacy on T's team which allowed them to ad-
vance so far ahead:
(6) R: 0: :kay.
R: W'll yeah. Look at that.
(0.5)
R: You were back here'n
CONTRAST:::) [we were way over there.
A: [Yau: :p?
A: Yup?
EXPLANATION :::) T: CAN YOU SAY ( . ) STACY?
T: [huh huh huh huh huh
A: [eh [ heh heh heh
R: [uh hah hah hah hah hah hah hah hah=
A: =I say we keep her
To understand T's tum as an explanation requires a bit of background in

North American popular culture. T is using a currently common joking
formula borrowed from the children's television show, "Mr. Rogers."
Mr. Rogers has an excessively gentle style of teaching, which contextu-
alizes the addressee as very young and just learning new concepts. Mr.
Rogers introduces something new and then prompts the children in the
viewing audience to try saying the new word: "Can you say (.) __?"
Imitating the same stylized intonation, including the pause before the
new word, comedians as well as regular folks in North America have
taken to using this question format as a way of stating something obvi-
ous, thus treating the situation as one in which the answer or "new
word" should have already been clear. The question is now a joking
formula for stating the obvious; the formulation imitates a nursery
school teacher's request for choral repetition of a new vocabulary item.
T's tum, "Can you say (.) Stacy?", is presented in an exaggerated,
mock-didactic intonation and at a raised amplitude. After Stacy joined T
and A's team, they moved ahead on the gameboard; thus, "Stacy" is the
explanation for the discrepancy between where the team was and where
292 Cecilia E. Ford
they are now. That A and R understand the joke is evidenced by their
laughter. A's "I say we just keep her" also demonstrates that the refer-
ence to Stacy's being on the team is interpreted as an explanation for the
team's success.
In (6), the contrast is presented by one speaker and the explanation
by another. By providing an explanation, T treats R's statement of the
contrast as unfinished in itself and calling for elaboration. Thus, in terms
of its treatment by the conversation participants, we can analyze this
contrast as expressing an opposition that is remarkable and puzzling, an
observation that seems to draw further talk providing a resolution, even
if it is done in a joking format.
Example (4), repeated once again, is from the setting of the science
class, but the activity is quite different from the one in (5), in which the
teacher was in more firm control of the direction of the talk. Here the
students are working in small groups on laboratory tasks. There is much
more student talk in this portion of the class; and even as the teacher
announces an action he's about to take, at least one student feels free to
object:
(4) (repea ted)

T: PEOPLE. I-I THINK MOST OF YOU ARE DONE
--
WITH THE CANDLES, =_ I 'MONNA PUT THE
LIGHTS ON.
(0.3)
?: [
CONTRAST~ K: [WE PREFER THEM OFF.
CONTRAST~ T: You prefer them aw- I prefer them on,
EXPLANATION~ I can see:.
The student's turn presents an unelaborated contrast; in stating her pref-

erence, she offers no explanation or justification. However, when the
teacher partially repeats K' s contrasting turn, we see another case of the
CONTRAST + EXPLANATION cluster: the teacher adds his own con-
trast, "I prefer them on", and he follows this with the explanation, "I can
see." In this case, the contrast is formed across speakers as a disagree-
ment, and thus the explanation comes as the account portion of what
conversation analysts understand as a dispreferred response.
As reviewed by Levinson (1983:333-335), conversation analytic re-
search has shown that turns which are in some sense disagreements (in
CA terms "dispreferred responses") regularly contain components ex-
plaining or otherwise accounting for the disagreement. Thus, from a

conversation analytic perspective, the contrast in (4) would be a case of
a dispreferred response with a characteristic account portion:
(4) (Teacher's tuxn)
T: You prefer them aw- I prefer them on, I can see:.
11 11
DISAFFILIATION ACCOUNT
Accounts, as Levinson (1983) defines them, are "carefully formulated

explanations for why the (dispreferred) act is being done" (334). The
rhetorical relation of explanation is associated with dispreferred tum
formats in cross-tum contrasts.
In a study of adverbial clauses in interaction (Ford 1993), I noted that
because regularly introduced account portions of disaffiliative re-
sponses. This can be seen in the following example taken from a phone
conversation. The topic is R's new job:
(7) simp~i£ied £rom Ford (1993:115)
A: Did you get your (.) first pay check from it?
( .)
A: [At least?
=>R: [NO: I won't get that for a couple of weeks yet.=
A: =Oh,
( .)
A: [W'l
=>R: ['Cause i t takes a long time.
A's extension, "at least?", displays the pursuit of agreement from R. At

the first arrow, R instead produces a negative and disaffiliative re-
sponse, which is followed, at the second arrow, by a because-prefaced
account for this disagreement.
In a subsequent study (Ford 1994), following up on the observation
that because introduced accounts for disaffiliative responses, I looked at
the use of because in different types of talk and writing. I found that if
disaffiliation was understood as a form of contrast, then the pattern of
294 Cecilia E. Ford
because-prefaced units introducing accounts could be generalized across

text types.
While my findings in that study supported the observation that be-
cause regularly functioned to introduce accounts for contrasts, I did not
look for a broader pattern beyond units connected with that specific
causal conjunction. In other words, the 1994 study involved a collection
of cases based on linguistic form, the connector because: in a data base
of instances of because, I found that a pattern in their immediately pre-
ceding textual environments regularly involved contrasts. 6 In the present
study, however, I fmd evidence of a more general pattern whereby the
contrasts and explanations come in sequenced combinations regardless
of the use of any particular formal connector.
This pattern of explanation following contrast was common in the
more freely interactional portions of the science class. In (8), just before
the class officially begins, student J has asked for change for his fifty
dollar bill. His intent is to buy some candy from S. S now offers him
change in the form of a check she has received from someone else in
payment for the candy she is selling (a second-party check):
(8)
s: I have a check.
S: Eight fifty.
CONT~T=> J: Nah. I won't take- I don't take second-
party checks.
S: eh huh huh huh huh
EXPLANATION=> J: I don't got no way of telechecking 'em,
J's refusal (contrast) is followed by an explanation in which he cites

his inability to verify whether the check is good ("telechecking" is a
process used by stores to get clearance for cashing checks). The tum
format includes normative features of disagreeing responses: the declin-
ing "Nah" and the account portion (what 1 am calling an explanation), "I
don't take second party checks... 1 don't got no way of telechecking
'em." Note also that there are nested contrasts and explanations here, as
in (5), above. J's "Nah" is followed by the explanation that he does not
take second party checks, and that negation is further followed by the
explanation that he has no way of electronically verifying the validity of
the checks.
As a final example of contrast plus explanation in my data, (9), be-
low, is an instance which supports the claim that explanations after con-
trasts are relevant not only from a discourse analyst's perspective, but
also, and in a significant manner, to the interactants themselves. In (9) T
presents an assertion which contrasts with a prior assumption (line 8),
but she does not immediately provide an explanation that might resolve
the asserted incompatibility. Note that both A and R treat the absence of
an explanation as problematic and deserving of further attention.
A and R have been remarking on the fact that there are many pictures
of people and sheep on the walls of T and P's dining room, where they
are seated. At lines 4-7, R begins to refer to the people in the pictures as
relatives, "grandma" and "grandpa":
(9)
1 R: I see there's a lo:t of pictures of sheep.
2 A: Oh yeah. there's more in the other [r(h)oo:m
3 t(h)oo.
4 R: [There
5 there's sheep there, Right by grandma.
6 A: ha ha ha
7 R: Gramma, (.) an' (.) an' (.) and grandpa,and the
8 the sheep and [baby sheep.
9 =>T: [We don't know these people.
10 (. )
11 T: t>or< these people.
12 R: y~ dotn't?
13 ( .)
14 T: No.
15 ( .)
16 A: Do you know the one [just below that one
though?
17 ?: [just
18 (0.5)
19 A: the one in the mid[dle?
20 T: [No.
21 T: The little cryin' girl?
22 A: No?
23=>R: [You just like the >pictures.<
24=>T: [We just like- silver: frames and
25 pictures,
26 R: And what about the sheep.=Do you know the
27 sheep?
296 Cecilia E. Ford
28 A: Eh(h) [heh heh heh

29 R: [Ub heh heh heh
30 T: These have been in our family [for
31 gener(h) at(h) ions
32 R: [Ha ha ha ha ha
At line 9, T produces a tum which contrasts with R's representation (4-

7) of the people as relatives. Instead of agreeing with R's statements, T
asserts that she and her partner P "don't know these people." It is after
this contrasting assertion that T fails to provide an explanation. What
happens next supports the existence of an expectation, on the parts of
these participants, for an explanation to be offered as a resolution to the
incompatibility expressed through the contrast. They treat the contrast
here as a remarkable announcement needing elaboration.
At 12, with marked pitch height, R expresses surprise, with "You
do l'n' t?", a form of repetition which regularly functions as a prompt for
further elaboration on the part of the prior speaker. But at 14, T, rather
than providing an explanation now that her contribution has been treated
as noteworthy, simply reasserts her negation. From 16 through 22, A
tries to salvage the assumption of T and P's familiarity with the subjects
of the pictures: she draws attention to persons in other pictures, asking
whether T and P are at least familiar with them. However, T continues
to respond with unelaborated, unexplained denial (line 20).
It is not until lines 23 and 24 that the sought after explanations are
offered, and at that point they are produced by R and T, in overlap. R
offers her guess at a possible account, "You just like the pictures," and
simultaneously, T gives her explanation, "We just like- silver: frames
and pictures." Only after these explanations does the talk move beyond
the problematic contrast. What follows is a joking question from R
about the sheep; and T joins in the joking by pretending that the sheep
have been in her family "for generations."
In the clear orientations of the interactants in (9) to something pro-
blematic and unresolved after T's contrasting assertion, we find further
evidence for a shared expectation that explanation should follow at least
certain types of contrasts.
In addition to the cases in my data, I have found other instances of
CONTRAST + EXPLANATION in published studies of conversation.
Notably, cases reported for English in the volume Analysing Causal
Conversation (Eggins and Slade 1997) and in a study of tum-
constructional units in German by Margret Selting (1998). While space
limitations preclude the inclusion of those cases here, and while the au-
thors do not draw specific attention to a CONTRAST +
EXPLANATION rhetorical cluster in their data, I take its occurrence in
their examples as suggestive of a more general pattern, one that is likely
not exclusive to my data nor to English conversations.
In sum, in the present data, we find a rhetorical pattern involving
contrasts, in a variety of functions, regularly followed by explanations.
As shown here, contrasts may be used pedagogically to highlight puz-
zles or problems; they may be the vehicles for observing remarkable
states of affairs; and, through interactional sequences, they may involve
disagreements between speakers. Significantly, explanation is a treated
by interactants as normatively relevant subsequent to contrasts in a vari-
ety of functions. Whether produced by single speakers or across speak-
ers, and whether prefaced by because or not, speakers in these data
regularly treat explanations as relevant after contrasts.
In addition to the pattern of contrasts followed by explanations, an-
other set of cases from my data involve contrasts followed by solutions.
Examining these cases provides further perspective on sequential con-
texts in which elaboration is added after a contrast. These cases also
support the observation that contrasts are used to present incompatibili-
ties or problems which participants treat as needing resolution.
6. Contrasts + Solutions
A further pattern was evident in the data whereby contrasts were fol-
lowed either by solutions or by displays that solutions, if not actually
stated, were understood as relevant or expected. In Rhetorical Structure
Theory, a relation of SOLUTIONHOOD requires that one span present
a problem and that the "situation presented in [the other text span be] a
(partial) solution to the problem" (Mann, Matthiessen, and Thompson,
1992:72). While Mann et al. do not point to the rhetorical relation of
CONTRAST as a pattern used in expressing the problem in the solu-
tionhood schema, they do note that the "scope of a problem includes...
conditions that carry negative values, either expressly or culturally, in-
cluding...frustrations" (1992:72). This resonates with Longacre's (1983)
notion of frustration in relations between predications. Frustration in a
text relation involves an expectancy which is countered. Contrasts ap-
pear to be common textual expressions of situations that go against ex-
298 Cecilia E. Ford
pectations. The expression of solutions after contrasts evidences partici-

pants' interpretation of contrasts in just this way.
Solutions also have commonalities with explanations in such rhetori-
cal contexts. Both types of additions treat contrasts as problematic and
needing further talk in the service of resolution. In the case of explana-
tions, resolution is in the form of support or justification. With solu-
tions, the contrast is resolved through the expression· of a potential rem-
edy.
In (10), R offers a solution to the problem A has just expressed
through a contrast. A and P have been Christmas shopping at a store that
carries bird feeders. P is now on the phone in the another room, and A is
telling R about their shopping dilemmas:
(10) "she" =P
A: She's trying to decide if sh[e wants to SPEND=
R: [She has a lot of
sheep.
A: =thirty five dollars on her dad. for Christmas.
R: Oh.
A: And she really doesn't.
=> (0.4)
A: But- (.) the gift that she thought of was thirty
five dollars, and she can't find i t anyplace
else.
R: Mm-mrmn.
A: so,
=> R: Has she tried Chickadee Depot.
There are actually two contrasts in this example: (1) spending $35 or
not, (2) not wanting to spend $35 (for which finding the gift more
cheaply would be a solution) but being unable to find the item anywhere
else for less. Note the pause after the first contrast, an indication that R
is treating A's tum as not yet complete. Given the pattern I am observ-
ing, this would be a relevant place for an expression offering resolution,
either an explanation or a solution. A normative pattern whereby con-
trasts are elaborated in such ways could explain R's lack of uptake fol-
lowing this first contrast and during the pause at the first arrow; she may
be awaiting an explanation, for example.
The second arrow points to R's response to the second contrast. This
contrast is between P's not wanting to spend $35 but not being able to
find the birdfeeder elsewhere for less. By asking whether P has tried
"Chickadee Depot," a local bird store, R is proposing a potential solu-
tion. In suggesting this solution (in the form of a question), R displays
her interpretation of the contrast as expressing a problem, for which a
search for solution is relevant.
In the science class, not surprisingly, there was a great deal of talk
focusing on problems and solutions. In the following case, the teacher
treats a solution as relevant in the context of a contrast, but he suggests
that the students solve the problem on their own. This strategy seems
particularly appropriate to one of the pedagogical goals in this discourse
context: to teach the students to formulate and solve physics problems.
(11)
T: If I stand thi-Oand this is the mirror
th- the screeno here, (0.3) I'm at a
«pointing to wall in front of him»
certain height, right? Look in.
~ If I go back farther, am I smaller?
CONTRASTS I'm still the same height, right?
( .)
J: Yeah,=
ORIENTATION T: =Try (.) If you care to. (.) try this

TO SOLUTION uh (.) problem, ( .) but doing i t wi th
uh scale drawing.
The teacher's contrasts here are formulated as questions, and although

he does not provide a solution, he does explicitly name what he has pre-
sented a "problem," and he suggests that the students may want to solve
it.
So in addition to being treated as presenting incompatible states of
affairs needing explanation, as described in section 5, contrasts can also
be treated as presenting problems, after which solutions are relevant.
Both solutions and explanations are elaborations which contextualize
preceding talk as unfinished, problematic, and needing resolution. The
present data provide us, then, with a basis for positing two kinds of rhe-
torical clusters and for using these clusters as a basis for interpreting the
work that contrasts do in talk-in-interaction.
300 Cecilia E. Ford
7. Contrasts without explanations or solutions
In a major portion of the cases I have analyzed, speakers either produce

or treat as problematically absent the patterns described in sections 5
and 6; that is, contrasts are either followed by explanations or solutions,
or speakers treat the lack of such elaboration as notable, through delays
or through further talk in pursuit of resolution. However, the collection
also includes instances of contrasts after which explanations and solu-
tions are not produced nor are they explicitly treated as relevant. In a
small subset of these instances, the contrasts appear to function perfectly
well alone, that is, they are in no way treated as unresolved in their con-
texts. The functions of contrasts which are treated in this way is worthy
of further investigation. In the majority of these non-explanation, non-
solution cases, however, there are good sequential and functional rea-
sons for a lack of orientation to resolution.
One type of contrast which was left unelaborated involves connec-
tives used as markers of discourse organization, without any literal con-
trast. Contrastive marking works as a format for text cohesion without a
concomitant expression of a problem or other content in need of resolu-
tion. Thus, while most of the contrasts in the collection involved some
contrastive marking and some real incompatibility at the content level,
several cases involve conjunctions serving text organizational functions
but not encoding any content level incompatibility. In such cases,
contrastive marking structures chunks of discourse, rather than focusing
on contrasts in meaning. In (12), the science teacher is involved in the
activity of discussing itemized tasks with which the students have
already been working. He refers briefly to one item and then moves to
another, using contrastive markers as a way of shifting the addressees'
attention, without drawing any problematic incompatibility into focus.
In introducing the next item in his list, he uses contrastive expressions
and marked stress: "Now this one, (.) but this one." In addition to the
use of but, we can see a use of now which Schiffrin (1987:234-40)
describes as expressing "a speaker's progression through the discourse
time of a comparison."
(12) T's first and second uses of this refer to

different tasks on which the students are working
T: After today this one will be easy. Now this
one, (.) but this one >I don't want to take
an hour on it.<
After using contrastive devices to introduce the shift "in discourse time"
to a next item, the teacher does not go on to produce any kind of expla-
nation or solution. The contrast functions only at a discourse organiza-
tional level, and subsequent talk does not attend to any content-level
opposition.
More interesting than the discourse structuring uses of contrastive
marking are cases in which sequential context, and the related factor of
speaker role in a local interactional activity or setting, are associated
with unelaborated contrast. Contrasts which come at the close of story-
telling sequences, for example, appear to function more as concluding
assessment than as focused expressions of problems which might call
for further elaboration and resolution. (13) is such a case. T has been
telling a story about her partner P's nephew; R and A respond with turns
which deliver displays of appreciation, action types which are typically
produced in the sequential contexts of story endings. In this case, the
displays are done through negative statements:
(13)
T: He was cute,>you know, sending the thank you
card before Ch(h)r(h)istm(h)as(h) ,<
hh[uh huh huh huh
R: [Yeah (h) huh huh hah
R: Stra[tegic.
T: [( )
T: We saw right through him.
=> R: [He is no: t a dumb kid.
A: rYe [h
T: [eh heh huh huh huh huh huh.
(): .hhh
(0.2)
=> R: [No sense in sending i t in August.
=> A: [Well at least he didn't send his list to
Santa.=
T: =That's right
A: °Yeah. °
While there are contrasts implicit in these negations, they are not treated
as invoking any need for explanation or solution. This seems related to
their coming as contributions to an assessment sequence at the close of a
story telling. As such, rather than presenting focused topics of talk on
~02 Cecilia E. Ford
their own, they are specifically produced as positive and agreeing re-
sponses to the upshot of T's story, a story about a clever and somewhat
manipulative child (note T's "We saw right through him"). In the sense
that they are affiliative with the teller's displayed stance and with the
story's gist, contrasts in such contexts are probably less likely to be
treated as needing resolution or even any special attention or elaboration
in themselves. That is, they may well express incompatibility at some
level (incompatibility with the expectation that children are guileless,
and that thank you notes should be sent soon after the action for which
they express gratitude); however, these turns are fully aligned interac-
tionally with the stance displayed in the immediately preceding talk. 7
In another kind of unelaborated contrast, the lack of orientation to
resolution seems to serve as an enactment or indexing of authority. In
the classroom context, there are explicit rules of conduct and there is a
clear authority figure. There are instances in which the teacher makes a
negative command or when a student makes a command which is based
on institutional rules, but no explanation or solution follows. While
there may be incompatibility expressed, the call for elaboration is art-
fully ignored in a manner which claims or underscores institutionally
supported rules and authority.
In one case, a student says to her classmates, "No talking." As this is
a rule during specific classroom activities (though one that is often vio-
lated), it is not surprising that the student offers no further explana-
tion-the explanation being treated as self-evident from shared knowl-
edge. And when the teacher at one point says, "Don't complain", in fail-
ing to produce a subsequent explanation, he may well be enacting his
role as authority. These cases suggest there may be institutional contexts
and specific activities (issuing directives based in or with reference to
institutional authority) in which a lack of explanation or solution is in
fact a resource for aligning with or reinforcing authority. Thus, explana-
tions may be relevantly absent in certain types of turns produced by per-
sons who are either referring to institutional authority or enacting it. It
must be noted, however, that this pattern is by no means absolute; recall
example (4), in which the teacher offers an explanation for his prefer-
ence for having the lights on, "I can see."
Related to the enactment of authority are cases of correction, them-
selves expressions of special access to relevant knowledge. (14) is a
case of third tum repair, involving a speaker repairing a misunderstand-
ing of her tum after the recipient has produced a responsive tum (Sche-
gloff 1992):
(14)
R: Ohm (.) who drew the last time. me or her.
T: I did.
( .)
T: very poorly.
=> R: No.=Me or her.
In the game they are playing (Pictionary) the drawing of pictures alter-
nates between teams and between teammates. Interpreting R's question
is complicated by the fact that both teammates and teams alternate in
this way. T understands R's first tum to be asking who among all the
players drew the last time. It turns out that R does not want to know
which individual among all the players was the last to take a tum; R's
question pertains to her team only: Was it she or her teammate who
drew last? The overlapped "me or her" in the first line specifies this ver-
sion of the question, but T is already responding at that point.
In the contrasting tum, R offers a repeat of her own previous tum,
and in so doing, she speaks from authority-having exclusive access to
what she actually meant in the first place. An explicit explanation in this
context would likely seem irrelevant and redundant (e.g., "No. Me or
her, because that's what I meant").
Finally, as a corollary to the treatment of contrasts as problems, there
are cases where contrasts express complaints or troubles-tellings (Jeffer-
son and Lee 1992). It can be a delicate matter whether or not the recipi-
ent of a complaint should offer a solution ("advice giving") or should
take the role of a sympathetic listener ("troubles recipient"). In (15), A
is describing an ongoing dispute in her extended family. The contrast
involves the families' living remarkably close to each other but still ar-
guing about where to spend their time:
(15)
A: ... she an'- her husband, (.) her parents
and her husband's parents, live within
~ six blocks of each other here in town?
CONTRAST'iR: Oh (.) hm hmmm
A: and ih- (.) where they spend most of their
time is like this major bone of
>contention about whether they spend time
at ~ parents' house< or his parents'
house?
304 Cecilia E. Ford
( .)
A: Six blo[cks-
R: [Six blocks away?
A: °Yeah. °
(1.1)
A: °It's insane.o
The incompatibility lies in the contrast between the proximity of the

dwellings and the existence of tension about where time is spent. In this
instance, the contrast is indeed treated as problematic, but this treatment
is not made evident though any following explanation or a solution.
Note that after presenting the contrast, A repeats "Six blocks," a key
element of the contrast. With the repeat, A prompts R for a response
which would align with the remarkableness of the proximity of the
families relative to the intensity of their issue regarding where to spend
time. R produces a surprised uptake, "Six blocks away?", in overlap
with A's prompt. After a pause, A goes on to express her stance even
more directly by assessing the situation negatively with "It's insane."
A's prompt, R's surprised uptake, and A's negative assessment all attest
to the fact that the contrast is interpreted as presenting a problem. Yet
neither participant produces a move toward a resolution.
This case suggests that contrasts interpreted as being in the service of
troubles telling may be less likely to be followed explanations or solu-
tions than are contrasts in other interactional activities. It is precisely the
lack of explanation or solution that makes the state of affairs count as
"trouble." Relevant to this analysis is the work of Jefferson and Lee
(1992), who illustrate similarities between advice-giving and troubles-
telling sequences. In (15), by taking the stance of a surprised recipient,
R embodies the role of "troubles recipient" rather than "advice giver."
Similarly, A's subsequent negative assessment treats the contrast as a
problem, but in this case as a problem in a "troubles-telling sequence"
rather than as an occasion for searching out remedies; A does not con-
textualize herself as one in search of a solution. While R is not strongly
affiliating as a sympathetic recipient here, at the same time, she does not
offer any solution or explanation.
It may be the case, then, that when contrasts are used in the service
of expressing troubles or complaints, the absence of an orientation to the
relevance of a solution or explanation reinforces the complaining force
of the contrast. The problematic incompatibility leads to further nega-
tive assessment rather than moves toward resolution.
The examples and discussion in this section underscore the fact that
contrasts are by no means always followed by explanations or solutions,
or even by clear orientations to the need for resolution. The contexts and
functions documented here provide a wider view of the uses of contrasts
in interaction, an area worthy of further study. Contrasts can be used in
text organizational functions, exploiting the contrastive format to
achieve a shift in focus without placing the contrast itself in focus. Such
subordination of the role of the contrast relative to a broader context of
activity is also seen in story closings. In these contexts, interactants may
use negative statements as presentation formats for otherwise agreeing
or affiliating turns without implicating any need to focus on the con-
trasts themselves as topics for further development.
Contrasts can also express commands and perform repairs. The
speakers of such turns act in the role of authority, act with reference to
shared institutional rules, or display their special access to knowledge
precisely through not explaining their contrastive moves. That is, the
authoritative stance can be seen as both associated with and enacted
through the absence of explanation. Finally, contrasts may be used in
the service of troubles telling. The lack of a move toward resolution is
one of the interactional features that distinguishes troubles-telling from
advice-giving sequences. Instead of proposing explanations or solutions,
interactants produce assessments which underscore treatment of the
contrast as indeed expressing trouble not easily remedied.
8. Conclusion
In this study, I have examined some of the functions of contrasts in in-

teraction, using the talk that follows contrasts as a special source for in-
terpretation. I have found significant orderliness associated with uses of
contrast, with explanations and solutions either coming immediately
after contrasts or oriented to as being expected and relevant in such con-
texts. In the present data, I have also observed that contrasts without
elaborations are artfully used for a variety of functional ends. In some of
these instances-commands from authority, other correction, and trou-
bles telling-particular actions seem to achieve their functional force at
least in part precisely through the absence of moves toward explanation
or solution. Thus, contrasts may still be associated with explanations
and solutions, but a speaker can index the authority from which s/he is
speaking by strategically producing an unelaborated contrast, or she
306 Cecilia E. Ford
may display an interpretation of a problem statement as a complaint by

offering a show of sympathy rather than a move toward remediation.
Finding evidence that CONTRAST + EXPLANATION and
CONTRAST + SOLUTION constitute schemas-recurrent and expect-
able rhetorical clusters-builds a foundation for further research into the
contexts for and encoding of such rhetorical constellations. Logical next
steps in this research agenda will be several: A first will be the study of
linguistic coding options for these paired actions, and the circumstances
in which different coding options are used. When and to what ends are
contrasts and explanations explicitly marked with contrastive or causal
connectives? Looking at languages other than English, how does this
action sequence fare cross-linguistically? Thinking diachronically, one
could explore whether these recurrent rhetorical combinations leave a
mark on linguistic practices over time (Hopper and Traugott 1993). At a
more sociolinguistic level, in reference to the cases of troubles telling
and speaking with institutional authority, it would be revealing to know
how interactional roles and relationships are invoked and performed
through contrasts as well as through other linguistic and pragmatic
means. Perhaps demanding explanations or solutions serves as a vehicle
for challenging the roles invoked in these contexts (e.g., troubles recipi-
ent, commanded/controlled subject).
In responding to an earlier version of this paper, Susanne Giinthner,
Harrie Mazeland, and Elizabeth Couper-Kuhlen each noted that con-
trasts are not actions in the same sense that explanations or proposals for
remedies (solutions) are. In rethinking this research, I have taken this
point most seriously; it suggests that while contrasts, solutions, and ex-
planations are all relations which language users can be shown to em-
ploy in constructing turns at talk, these categories of relation are in-
vented by analysts, and there is good reason to question the degree to
which these categories reflect a comparable type of rhetorical connec-
tion (see also Ford 1986 for a related critique of RST). In the current
version of this paper, I have taken pains to consider what is made evi-
dent through the talk of the participants themselves. What I hope to have
demonstrated is that, in treating contrasts as needing solutions and ex-
planations, the participants in these interactions themselves offer us a
basis for the functional analysis of these contrasts.
Notes
* I thank Elizabeth Couper-KuWen, Donna Dallos, Christine GoW, Susanne

Giinthner, Bernd Kortmann, Harrie Mazeland, and Sandra Thompson for help
in my thinking for this research and for feedback on various drafts of the pa-
per. I take full responsibility for my use of their feedback.
1. Mann and Thompson's definition of "concession" involves an acknowledge-
ment of "potential or apparent incompatibility" between two units of text be-
ing compared (1988). Given that this relation has the expression of contrast
within it, I included such cases in my collection (see the concession, "I look
smaller" in (5), below).
2. In another investigation, I am looking at the initial parts of contrasts. In this
instance, the use of the perception verb, sound, expresses an evidential dis-
tance between what is apparent and what is real, projecting that what is real
may be expressed next. The second part of the contrast makes good on this
projection.
3. These data come from a longitl;ldinal, ethnographic study headed by Jane
Zuengler at the University of Wisconsin-Madison.
4. The laboratory interaction is analyzed in Ford (1999).
5. Transcription symbols
Symbol Interpretation
(.) A short, untimed pause
(0.3) A timed pause
hhh Audible breath
thi- Hyphen indicates a sound cut off
Latching, rush into next turn or segment
[ Brackets indicate the onset of overlap
she Prominent stress
SHE Higher volume than surrounding talk
°sheo Lower volume than surrounding talk
she: Sound stretch
Low falling intonation
? High rising intonation
Intermediate intonation contours:
level, slight rise, slight fall
>talk< At a faster tempo than surrounding talk
6. While contrasts formed the most prevalent textual environment for subse-
quent use of because, that study also pointed to "strong evaluations" as regu-
larly followed by because-prefaced continuations (Ford 1994:548).
7. I know of no studies looking specifically at the use of negative statements in
affiliative assessments such as these.
308 Cecilia E. Ford
Re..ferences
Atkinson, J. Maxwell
1984 Public speaking and audience responses: some techniques for invit-
ing applause. In: 1. Maxwell Atkinson and 1. Heritage (eds.), Struc-
tures of Social Action, 370-410. Cambridge: Cambridge University
Press.
Atkinson,1. Maxwell and John Heritage (eds.)
1984 Structures ofSocia/Action. Cambridge: Cambridge University Press.
Cumming, Susanna and Tsuyoshi Ono
1997 Discourse and grammar. In: T. A. van Dijk (ed.), Discourse as Struc-
ture and Process, 112-137. London: Sage.
Eggins, Suzanne and Diana Slade
1997 Analysing Casual Conversation. London: Cassell.
Fairclough, Norman
1989 Language and Power. New York: Longman.
Ford, Cecilia E.
1986. Overlapping relations in text structure. In: Scott DeLancey and Rus-
sell S. Tomlin (eds.), Proceedings ofthe Second Annual Meeting of
the Pacific Linguistics Conference, 107-123. University of Oregon:
Department of Linguistics.
1993 Grammar in Interaction: Adverbia/ Clauses in American English
Conversations. Cambridge: Cambridge University Press.
1994 Dialogic aspects of talk and writing: because on the interactive-
edited continuum. TEXT 14(4): 531-554.
1999 Collaborative construction of task activity: Coordinating multiple
resources in a high school physics lab. Research on Language in So-
cia/Interaction 32 (4): 369-408.
Ford, Cecilia E., Barbara A. Fox, and Sandra A. Thompson
1996 Practices in the construction of turns: the "TCU" revisited. Pragmat-
ics 6(3): 427-454.
Fox, Barbara A.
1987 Discourse Structure and Anaphora: Written and Conversational
English. Cambridge: Cambridge University Press.
Given, Talmy
1978 Negation in language: Pragmatics, function, ontology. In: Peter Cole
(ed.), Syntax and Semantics, Vol. 9,69-112. New York: Academic
Press.
Goodwin, Charles
1979 The interactive construction of a sentence in natural conversation. In:
George Psathas (ed.), Everyday Language: Studies in Ethnomethod-
ology, 97-121. New York: Irvington.
1981 Conversational Organization: Interaction between Speakers and
Hearers. New York: Academic Press.
1995 Co-constructing meaning in conversations with an aphasic man. Re-
search on Language and Socia/Interaction 28(3): 233-260.
Goodwin, Marjorie Harness

1990 He-Said-She-Said: Talk as Social Organization among Black Chil-
dren. Bloomington, Indiana: Indiana University Press.
Halliday, M. A. K., and Ruqaiya Hasan.
Have, Paul ten, and George Psathas (eds.)
1990 Studies in the Social Organization of Talk and Embodied Activities.
Washington, D.C.: University Press of America.
Hopper, Paul J.
1998 Emergent grammar. In: Michael Tomasello (ed.), The New Psychol-
ogy of Language: Cognitive and Functional Approaches to Lan-
guage Structure, 155-176. Mahwah, New Jersey: Erlbaum.
Hopper, Paul J., and Elizabeth C. Traugott
Jefferson, Gail
1973 A case of precision timing in ordinary conversation: overlapped tag-
positioned address terms in closing sequences. Semiotica 9(1): 47-
96.
1978 Sequential aspects of story telling in conversation. In: J. Schenkein
(ed.), Studies in the Organization of Conversational Interaction,
219-248. New York: Academic Press.
Jefferson, Gail and John R. E. Lee
1992 The rejection of advice: managing the problematic convergence of a
"troubles-telling" and a "service encounter" In: Paul Drew and John
Heritage (eds.), Talk at Work: Interaction in Institutional Settings,
521-548. Cambridge: Cambridge University Press.
Lemke, Jay L.
1990 Taking Science: Language, Learning, and Values. Norwood, NJ:
Ablex.
Lerner, Gene H.
1991 On the syntax of sentences-in-progress. Language in society 20(3):
441-58.
1996 On the "semi-permeable" character of grammatical units in conver-
sation: conditional entry into the tum space of another speaker. In:
Elinor Ochs, Emanuel A. Schegloff, and Sandra A. Thompson (eds.),
Interaction and grammar, 238-77. Cambridge: Cambridge Univer-
sity Press.
Lerner, Gene H., and Tomoyo Takagi
1999 On the place of linguistic resources in the organization of talk-in-
interaction: A co-investigation of English and Japanese grammatical
practices. Journal ofPragmatics 31(1): 49-75.
Longacre, Robert E.
1983 The Grammar ofDiscourse. New York: Plenum.
Levinson, Stephen C.
1983 Pragmatics. Cambridge, UK: Cambridge University Press.
310 Cecilia E. Ford
Mann, William C., and Sandra A. Thompson

1988 Rhetorical structure theory: toward a functional theory of text or-
ganization. TEXT 8(3): 243-81.
Mann, William C., Christian M.I.M. Matthiessen, and Sandra A. Thompson
1992 Rhetorical structure theory and text analysis. In: W.C. Mann and S.
A. Thompson (eds.) Discourse Description: Diverse Linguistic
Analyses ofa Fund-raising Text, 39-78. Amsterdam: Benjamins.
Ochs, Elinor, Emanuel A. Schegloff, and Sandra A. Thompson (eds.)
1996 Interaction and Grammar. Cambridge: Cambridge University Press.
Orestrom, Bengt
1983 Turn-taking in English Conversation. (Vol. 66). Lund: CWK
Gleerup.
PerakyHi, Anssi
1997 Reliability and validity in research based on tapes and transcripts. In:
David Silverman (ed.), Qualitative Research: Theory, Method and
Practice, 201-220. London: Sage.
Rudolph, Elizabeth
1996 Contrast: Adversative and Concessive Relations and their Expres-
sions in English, German, Spanish, Portuguese on Sentence and Text
Level. New York: Gruyter.
Sacks, Harvey
1987 On the preferences for agreement and contiguity in sequences in
conversation. In: Graham Button and John R. E. Lee (eds.), Talk and
Social Organization, 54-69. Clevedon, England: Multilingual Mat-
ters.
Sacks, Harvey, Emanuel A. Schegloff and Gail Jefferson
1974 A simplest systematics for the organization of turn-taking for con-
versation. Language 50(4): 696-735.
Schegloff, Emanuel A.
1980 Preliminaries to preliminaries: can I ask you a question? Sociological
Inquiry 50: 104-52.
1988 Discourse as an interactional achievement II: an exercise in conver-
sation analysis. In: Deborah Tannen (ed.), Linguistics in Context:
Connecting Observation and Understanding, 135-158. Norwood,
NJ: Ablex.
1992 Repair after next turn: the last structurally provided defense of inter-
subjectivity in conversation. American Journal of Sociology 97(5):
1295-1345.
1996 Turn organization: one intersection of grammar and interaction. In:
Elinor Ochs, Emanuel A. Schegloff, and Sandra A. Thompson (eds.),
Interaction and Grammar, 52-134. Cambridge: Cambridge Univer-
sity Press.
Schiffrin, Deborah
1987 Discourse Markers. Cambridge: Cambridge University Press.
Selting, Margret
1998 TCUs and TRPs: The Construction of Units in Conversational Talk.
InLiSt NO.4. (Interaction and Linguistic Structures.) Sekretariat LS
Angl. Sprachwissenschaft. Konstanz, Germany: Universitat Kon-
stanz.
Tannen, Deborah
1979 What's in a frame? Surface evidence for underlying expectations. In:
Roy A. Freedle (ed.), New Directions in Discourse Processing, 137-
181. Norwood, New Jersey: Ablex.
Tomlin, Russell S., Linda Forrest, Ming Ming Put, and Myung Hee Kim
1997 Discourse semantics. In: T. A. van Dijk (ed.), Discourse as Structure
and Process, 63-111. London: Sage.
Whalen, Jack
1992 Conversation analysis. In: E. F. Borgatta and M. L. Borgatta (eds.),
Encyclopedia ofSociology, 303-310. New York: Macmillan.
Concessives on different semantic levels:
A typological perspective *
Mily Crevels
It will be argued (a) that concessive clauses are expressed on four different
semantic levels: the content level, the epistemic level, the illocutionary or
speech-act level and the textual level, and (b) that there exists a systematic
cross-linguistic correlation between the semantic level at which adverbial
clauses apply, on the one hand, and the way they are expressed, on the
other. An illustration of this correlation is the fact that languages may use
different conjunctions to express adverbial relations if these apply at
different levels. The discussion of three hypotheses stands at the centre of
this paper (section 3). These hypotheses have been formulated to investigate
whether, depending on the semantic level, concessive clauses show
distinctive formal properties and have been tested against data from a wide
range of languages.
1. Concessive relations
I will start by discussing the major properties of conceSSIve

constructions, drawing heavily on Konig (1986, 1988 and 1994).
Someone who produces an utterance of the general format under (1), is
committed to the truth of both propositions (cf Konig 1988: 146).
(1) Although p, q
An example is given in (2):
(2) Although it is raining, I am going outfor a walk.
Moreover, a connection is implied between the propositions of the two

related clauses in question: the speaker asserts these two propositions
against the background assumption that the two types of situations
which p and q describe are generally incompatible. This implication has
the status of a presupposition rather than of an entailment (cf Konig
1986: 233). For an example such as (3a) the concessive presupposition
may be expressed as in (3b):
314 Mily Crevels
(3) a. Even though he had not eaten for days, he looked strong and
healthy.
b. {f one does not eat for days, one normally does not look strong
and healthy.
Konig (1988: 147) formulates the relevant presupposition as follows:
(4) If p', then normally --q'
Concessive clauses occur in all positions where adverbial clauses are

permitted in a language. In many languages they may either precede or
follow the main clause. However, as Konig (1994: 679) has pointed out,
concessive clauses differ from other types of adverbial clauses in a
number of ways: (i) in contrast to most other types of adverbial clauses,
there do not seem to be many languages with a concessive interrogative
pronoun analogous to English when (Time), why (Reason), how
(Manner), etc., (ii) concessives cannot be the focus of a focusing adjunct
(focus particle) like only, even, just, especially (in contrast to a causal
construction as in (5a», (iii) concessives cannot occur as focus in a cleft
sentence (5b) and (iv) concessives cannot be the focus of a negation or a
polar interrogative (5c).
(5) a. i. Only because it was raining ...

ii *Only although it was raining ...
b. i. It was because it was raining that ...
ii *It was although it was raining that ...
c. i. Was he harassed because he was a boxer?
ii ?Was he harassed although he was a boxer?
As Konig claims, all of these divergent properties seem to be

manifestations of a single syntactic/semantic l constraint on the use of
concessive clauses: they cannot be focused. This constraint is generally
considered to indicate that the relevant clauses are less tightly integrated
into a main clause than other types of adverbial clauses.
Concessives on different semantic levels 315
2. Distinct levels of linking
2.1. Sweetser's domain theory and layers in Functional Grammar
Sweetser (1990) clearly demonstrates the contrast between root,

epistemic and 'speech-act' uses of modal verbs and multiple uses of
conjunctions, such as conditional if. Linking between three different
types of entities, i.e. (i) real or hypothetical situations, (ii) aspects of
knowledge or (iii) speech acts, therefore takes place at the content, the
epistemic or the illocutionary level. In Crevels (1994, 1998) I have
shown that these distinctions may also be made for concessive
constructions in Spanish, and that Sweetser's multiple semantic
domains correspond, respectively, to the predicational, the propositional
and the illocutionary layers within the layered structure of the clause in
Functional Grammar (cf Dik et al. 1990). The idea of this structure is
that layers of lower complexity are fully contained within layers of
higher complexity. Thus, every speech act contains a proposition, and
every proposition contains a predication. In addition, a fourth layer, the
so-called text layer (Crevels 1994, 1998, 2000; Hengeveld 1997), is
required. Concessive clauses pertaining to this layer have to do with the
organization of the discourse and, therefore, apply to text units which
may contain more than one sentence.
In Functional Grammar, six layers are claimed to be relevant for the
analysis of clause structure in natural languages (Hengeveld 1989, 1992,
1997; Crevels 1994, 1998, 2000). As Hengeveld (1996: 119) points out,
a serious problem lies in providing evidence for the validity of each of
these layers in main clauses. An important way to solve this problem is
to study the properties of subordinate constructions. Since subordinate
constructions may be classified according to the highest layer they
contain, and since each of the layers present in the hierarchical clause
model may be turned into a subordinate construction, the study of
different types of subordinate constructions will lead to a better
understanding of the differences between the layers.
A strong argument in favour of the existence of the different
semantic levels is the fact that languages may use different conjunctions
to express the same kind of adverbial relation. French parce que
'because' is used specifically for content conjunction, while puisque is
the correct causal conjunction at the epistemic or illocutionary level (cf
Sweetser 1990: 82, 156).
316 Mily Crevels
(6) a. II va I 'epouser parce qu'ill 'adore.

'He's going to marry her because he adores her.'
b. (Mais si,) il va I 'epouser, puisqu'ill 'adore.

'(But of course,) he's going to marry her, since he adores her.'
Likewise, Latin quoniam, English since and Spanish ya que show a

strong tendency towards an epistemic or speech-act reading, rather than
toward a content-conjunction reading (Quirk et al. 1985; Bolkestein
1991; Crevels 1994, 1998).
2.2. Entity types
Extending the analysis proposed by Lyons (1977: 442-447), Hengeveld

(1989, 1992, 1998) classifies the semantic types of adverbial clauses on
the basis of the entity types they designate. Linguistic units may refer to
entities of six different types. In view of the text layer, Hengeveld's
typology of entities may, therefore, be extended by adding fifth order
entities, which consist of text units that can be evaluated in terms of
their thematic continuity (cf Table 1).
Table 1. Entity types
Entity type Description Evaluation

Zero order Property or Relation Applicability
First order Individual Existence
Second order State of affairs Reality
Third order Propositional content Truth
Fourth order Speech act Felicity
Fifth order Text unit Thematic Continuity
2.3. Integration
Zero and first order entities cannot be realized by concessive adverbial

clauses, but the other four types are indeed relevant. Examples are given
in (7):
(7) a. Although it's raining, we're goingfor a walk.

b. He's not at home, although his car is parked in front ~f the
house.
c. Even though I am calling a bit late, what are your plans for
this evening?
d. I speak and write Serbian, Albanian, Turkish and Dutch, but I
cannot express my true feelings in any other language than
~omani. .A lthouF now that I come to think of it, I have done
It many tImes ...
The difference between (7a) and (7b) corresponds to the distinction

Lyons (1977) makes between second and third order entities, i.e.
between states of affairs and propositional contents. The concessive
construction in (7a) is an example of what Sweetser calls a real world or
content relationship: although in the real world the rain indeed forms an
obstacle, it nevertheless cannot prevent our going for a walk. In (7b) the
concessive construction concerns an epistemic relationship: the
concessive connective combines two items of knowledge, a premise and
a conflicting conclusion. Thus, in this particular example, the speaker,
even though s/he knows that the car of the person in question is parked
in front of his house, reaches the conflictive conclusion that this person
is not at home. In (7c) we find an example of an illocutionary
relationship; in this case the concessive clause forms an obstacle to the
speech act expressed in the main clause, a possible paraphrase being: 'I
know that 1 should have phoned you sooner, so 1 normally wouldn't
phone you this late to ask you what you are doing this evening'.
Concessives at this speech-act level are often closely related to the
conversational maxims of Grice (1975); thus, the concessive clause in
(7c) appeals to Grice's principle of informativeness, which in some
cases allows the hearer to read into an utterance more information than
it actually contains. In (7d), fmally, an example of a textual relationship
is given. In this case the concessive clause Although, now that I come to
think of it, I have done it many times ... stretches over a whole series of
preceding utterances, signalling an unexpected turn in the discourse
context, a possible paraphrase being 'I come to the conclusion that 1
have expressed my true feelings many times in Serbian, Albanian,
Turkish and Dutch, so 1 normally wouldn't first say that I speak and
write Serbian, Albanian, Turkish and Dutch, but that I cannot express
my true feelings in any other language than Romani' .
318 Mily Crevels
2.4. A four-level approach to concessives
In the content domain a concessive connection indicates that the event

or the state of affairs described in the concessive clause forms an
obstacle for the event or the state of affairs described in the main clause,
but does not impede its realization.
(8) She's just given birth to a beautiful baby girl although she's fifty-
two.
In the epistemic domain concessive connection expresses the idea that

the speaker, in spite of being convinced of the content of the concessive
clause, still reaches the opposite conclusion contained in the main
clause. In other words, in the epistemic domain concessive conjunction
will mark the impediment of a belief or a conclusion. Example (9) does
not express any factual conflict, but a conflict between the conclusion
and the potential counterargument expressed in the concessive clause.
(9) He left his wife and children, although he loved them very much.
In the speech-act domain the content of the concessive clause does not
form an obstacle for the realization of the event or the state of affairs
described in the main clause, but raises obstacles for the realization of
the speech act expressed by the speaker in the main clause.
(10) Although it's none ofmy business, your behaviour is a disgrace.
The speaker's recognition of the fact that s/he has no business

commenting on the interlocutor's behaviour is in no way incompatible
with the speaker's belief that the interlocutor's behaviour actually is a
disgrace. A possible paraphrase of (10) might be 'I know that 1 have no
business commenting on your behaviour, so 1 normally wouldn't do so',
or 'I know that 1 have no business commenting on your behaviour, so 1
normally wouldn't perform the speech act of commenting on it'.
Let us consider, finally, concessive conjunction at the text level,
where the modification is based on an unforeseen turn in the discourse
context:
(11) My favourite poster is, I think, a French one for Nesquik, which
shows a sophisticated-looking small boy leaning nonchalantly
against something and saying that thanks to Nesquik he went hack

on to milk. He really looks a nice child. Though, there are some
Adchildren that one would feel quite ashamed to have around the
house ... (Greenbaum 1969: 68)3
In contrast to illocutionary concessives, textual concessives do not

modify the main clause of a concessive construction, but generally a
whole preceding text unit which may be composed of various sentences.
lliocutionary concessives are always part of a single construction, and,
therefore, specify or modify a single preceding or following speech act;
textual concessives, on the other hand, usually modify a whole series of
preceding propositions, rather than just single utterances and often seem
to be functioning as an afterthought.
Although in written language one is confronted with the bias of
punctuation conventions, textual concessive clauses do not always
manifest themselves in the form of a simple sentence. Moreover, they
typically exhibit main clause word order in those languages, like Dutch
and German, where main and subordinate clauses are distinguished on
the basis of word order (Konig and van der Auwera 1988, Giinthner
1993). In (12a) we have an example of a concessive with main clause
word order, whereas (12b) shows the typical V-2 order:
Dutch (Indo-European)
(12) a. Ze komt vast en zeker; hoewel, hi} haar kan je maar nooit
weten ...
'She will certainly come; though, one can never be sure with
her ... '
b. Ze komt vast en zeker, hoewel je hi} haar maar nooit weten kan
'She will certainly come, although one can never be sure with
her ... '
This difference in word order between main and subordinate clauses

may well be explained by so-called iconicity principles, or, in other
words, principles which govern natural form-function correlations (cf.
Givan 1984: 30-31).
Different subtypes of concessive clauses manifest different degrees
of subordination to and integration into a main clause. As mentioned
above, concessive clauses cannot be focused--even in their standard
use--and tend to take wide scope over any operator in the main clause.
320 Mily Crevels
Therefore, it is assumed that they are less tightly connected to a main

clause than conditional, temporal or causal clauses. The higher the
semantic level they pertain to, the looser they seem to be integrated into
a main clause and the more paratactic-like the construction becomes.
3. Hypotheses
On the basis of the discussion in section 2.4 it is now possible to

investigate whether the expression of concessive clauses at the four
semantic levels shows distinctive formal aspects. To do this I have
constructed a small questionnaire containing nine concessive
constructions in English at the four relevant levels with two content, two
epistemic, three speech-act and two textual concessives, respectively.
With the help of native speakers and specialized linguists I have
collected data on a sample of 30 extant languages. 4 The consultants
were requested to express the concessive constructions in their own
language and to use a concessive linker where possible. The data of the
sample languages will now be analyzed on the basis of three specific
hypotheses which may be subjected to further typological investigation.
The basis for each of these hypotheses is constituted by the following
hierarchy of semantic levels, which is based on the concessive relations
given in (7):
(13) Content level> Epistemic level> lliocutionary level > Text level
The content level, which is the semantically least complex level, is

followed by the semantically more complex epistemic level, which in
tum is followed by the illocutionary or speech-act level, and, finally, the
text level. In the concrete application of this hierarchy we will see that
in each case there is a correlation between the semantic level of the
concessive construction and its formal expression. Thus, by testing this
hierarchy it will be possible to establish the validity of each of the four
categories which belong to it. Of the three hypotheses that are to be
tested, the first two have a bearing on the way concessive constructions
are marked, while the third one has to do with the nature of concessive
linkers.
3.1. The marking ofconcessive constructions
3. 1. 1. Syndetic versus asyndetic constructions
The first hypothesis to be tested ties in with other observations and

hypotheses in the domain of clause-linking in functional typology (cf
van der Auwera (ed.) 1998; Devriendt et al. (eds) 1996; Haiman and
Thompson 1988). If we may assume that a concessive clause becomes
less and less integrated into its main clause and more and more
paratactic-like the higher its semantic level is, we may assume at the
same time that languages are more likely to use asyndetic strategies to
express a concessive construction if this construction belongs to a
higher semantic level. Thus the first hypothesis, represented in (14),
may be formulated as follows:
HYPOTHESIS 1
The probability that a concessive construction will be expressed by
asyndetic means increases as the construction pertains to a higher
semantic level.
(14) Content > Epistemic > Illocutionary > Text

Syndetic > > > Asyndetic
Table 2 provides a schematic representation of the clause combining

strategies used to express the various concessive constructions: '-'
stands for syndetic and '+' for asyndetic concessive linking. For
instance, in Kiwai-a complex Papuan language spoken in south-
western Papua New Guinea, Western Province, Fly River
Delta--syndetic and asyndetic constructions may occur at the speech-
act and text level. This implies that my consultants in some cases use
syndetic means to express an elicited illocutionary concessive
construction, and on other occasions resort to asyndetic means.
The results of the testing of Hypothesis 1 are interpreted as evidence
for three cut-off points within this hierarchy, since, as can be seen from
the bottom end of Table 2, in Lakxota (South Dakota, USA) asyndetic
linking may take place at the epistemic, the speech-act and the text
level, while Pima (Arizona, USA) and Kiwai show signs of asyndetic
marking at the speech-act and text level, and in Hungarian, Boboda
(Burkina Faso), Kannada, Burmese, Turkish, Uzbek and Mongolian
asyndetic concessive constructions only occur at the text level. The
322 Mily Crevels
almost extinct Australian Aboriginal language Duunggidjawu from

south-eastern coastal Queensland provides a good example of a
language which uses juxtaposition of clauses as a clause-combining
strategy for all concessive constructions. Examples (15) and (16) show
that although in Duunggidjawu the word order is very free and varying,
as is the case in many Australian languages, it is rigid for the expression
of concessive clauses, which are always preposed.
Duunggidjawu (Australian; Stephen Wurm p.c.)

(15) dyinnang jO:ri wakka galang yingi,
foot his NEG good be:INCPAST
guyumgu biye dyingi.

camp.to returning he. went.there
'He went back to the camp although he had broken his foot.,5
(16) gurangur jo: ri minna na barang jO: ri

spear his this and boomrang his
ganna yingi, jO: wakka guyum-gu.

there are he NEG camp-in
'He's not in the camp, although his spear and his boomerang are
there. '
Table 2. Schematic representation of asyndetic constructions
Language Content Epistemic Illocutionary Text

Amharic
Wolaitta
Bahasa Indonesia
Tahitian
Motu
Dargi
W-Greenlandic
Basque
Ket
Kwaza
Mestreechs
Romani (Erli)
San

Yoruba
Nahuatl
Lokono
Japanese
Cantonese
Finnish
Hungarian -/+
Boboda -/+
Kannada -/+
Burmese -/+
Turkish -/+
Uzbek -/+
Mongolian +
Pima -/+ -/+
Kiwai -/+ -/+
Lakxota -/+ -/+ -/+
Duungidjawu + + + +
In Lakxota, which shows a cut-off point between the content and the
epistemic level, syndetic means are used at the content level, but an
asyndetic strategy is possible at the higher levels. Consider examples
(17) and (18) with a Lakxota content concessive and an epistemic
concessive, respectively:
Lakxota (Amerind; Violet Catches p.c.)

(17) Ishtagxungxa eyash, ishtamanza un shni.
blind although glasses wear NEG
'He doesn't wear glasses although he sees very little.'
(18) Txawicu na cinca tewichaxila nus,

wife and off-spring love supposedly
awich-ayushtan.
them-left
'He left his wife and children, although he loved them very much.'
As mentioned above, Kiwai shows a cut-off point between the epistemic

and the speech-act level, since syndetic means are used at the content
and the epistemic level, but an asyndetic strategy is possible at the
324 Mily Crevels
speech-act and the text level. Consider the epistemic concessive in (19)
and the speech-act one in (20):
Kiwai (Indo-Pacific; Stephen Wurm p.c.)

(19) Nanie nou kara moto tatan ai-r-otoi,
although his car house beside FAcT-itsBJ-stand
nou pai moto-wa.

he NEG house-in
'He's not at home, although his car IS parked in front of the
house.'
(20) Mo pai ubi ai-r-erea ro k-emarogo-gido,

I NEG wish fact-sBJ-be yOU:SG vN-scolding-ENWH
ro uba ai-r-erea.
yOU:SG bad FACT-sBJ-be
'Although I should be minding my own business, your behaviour
is a disgrace. '
Another cut-off point is the one between the illocutionary and the text
level. In Turkish, for example, syndesis is the only option on the
illocutionary level (21), while at the text level asyndetic linking is also
possible (22).
Turkish (Altaic; Hiiseyin Demirel p.c.)

(21) Beni ilgilendir-me-me-si-ne ragmen
me concem-NEG-VN-3SG:POSS-DAT although
senin hareket-ler-in {fok kOtii.

your behaviour-PL-2sG:Poss very bad
'Although I should be minding my own business, your behaviour
is a disgrace. '
(22) Ben ingilizce konuruyor ve yazl-yor-um

I English speak-PROG and read-PROG-ISG
ama his-ler-im-i tiirkr;e dl~lnda

but feeling-PL-ISG:poss-ACC Turkish except
ba~ka bir dil-de anlat-am 1-yor-um

other one language-Loc express-IMPOT-PROO-ISO
As11nda dii~n-iince birf(ok deja

actually think-OER several time
yap-t 19 -1m -1 hat1r11-yor-um.

do- VN -ISO:POSS- ACC remember-PROO-Iso
'I speak English, and I write it, but I cannot express my true
feelings in any other language than Turkish. Although, now that I
come to think of it, I have done it many times ... '
At the text level it is sometimes difficult to establish whether a

construction contains a concessive linker or not. Thus I have classified
(22) as an instance of asyndetic linking because as11nda does not
convey any concessive connotation and, moreover, because the
concessive nature of the sentence does not change when asl1nda is left
out.
3.1.2. Clause marking
If it is true that concessive clauses get more and more paratactic-like as

they pertain to a higher semantic level, the following is to be expected:
in eliciting a subordinate construction of the general format 'although p,
q', the possibility that the informant will mark the q (apodosis) instead
of the p (protasis) increases as the construction pertains to a higher
semantic level. This allows us to formulate the following hypothesis
based on hierarchy (13):
HYPOTHESIS 2
The probability that in an elicited subordinate concessive construction
the q will be marked instead of the p increases as the construction
pertains to a higher semantic level. 6
(23) Content > Epistemic > lliocutionary7

p > > q
Table 3 shows a schematic representation of the clause-marking

strategies used to express the various concessive constructions. In
326 Mily Crevels
Mongolian, for instance, both p and q marking is possible at the speech-

act level (P/q). This means that my informants in some cases choose to
mark the p in order to express an elicited illocutionary concessive
construction, and in other cases the q (apodosis). Furthermore, pq stands
for correlative marking in which both the concessive subordinate clause
and its subsequent main clause are marked. IRR, finally, stands for the
absence of an overt concessive linker.
Again there is evidence for two cut-off points within the hierarchy
tested in Table 3: (i) between the content and the epistemic level, since
the Omotic language Wolaitta (Ethiopia) and Kwaza-a higWy
endangered language isolate spoken in Rondonia (Brazil)--tend to mark
the p at the content level but may mark the q at remaining levels; (ii)
between the epistemic and the speech-act level, since Hungarian,
Nahuatl (Mexico), Kannada (South India), the Erli dialect of Romani
(Bulgaria), Bahasa Indonesia, Uzbek, Basque, Mongolian, Mestreechs
(Maastricht, The Netherlands) and Pima tend to mark the p at the
content and epistemic levels but may mark the q at the speech-act level.
Consider examples (24) and (25), which exemplify a content and an
epistemic concessive in Kwaza, respectively:
Kwaza (Isolate; Kyjkaw Made, and Hein van der Voort p.c.)
(24) e 'ka) dwy-le'te tso'roi=e-ki afY- 'na
leg break-coNc run=go-DECL house-Loc
'He ran all the way home although he had broken his foot.'
(25) a'riii cutsikja- 'sinwa-hy-wara j a- 'he-tse

tapir stand-yard-NR-but he exist-NEG-DECL
'He is not at home, although his horse is in the yard. , 8
Table 3. P versus q marking
Language Content Epistemic Illocutionary

Duunggidjawu IRR IRR IRR
Amharic P P P
Tahitian P P P
Kiwai p P pllRR
Ket p P p
Burmese p P p
San p P p
Boboda p P p
Language Content Episternic Illocutionary

Yoruba p p p
Turkish p p p
Dargi p p p
Japanese p p p
Finnish p p p
Lakxota p p/IRR p/IRR
Cantonese pq pq pq
Motu pq pq pq
W-Greenlandic p p p/pq
Lokono p/pq p/pq p/pq
Hungarian p p p/pq/q
Nahuatl p p p/pq/q
Kannada p p p/q
Romani (Erli) p p p/q
Bahasa Indonesia p p p/q
Uzbek p p p/q
Basque p p p/q
Mongolian p p p/q
Mestreechs p p p/q
Pima p p q/IRR
Wolaitta p p/q p/q
Kwaza p p/q p/q
While (24) shows the same marking of the p as the example in the
questionnaire does, (25) shows a marking of the q. As observed above,
the marking of the q in this last Kwaza example goes hand in hand with
a more paratactic or coordinate-like type of construction.
The cut-off point between the epistemic and the speech-act level can
be observed in Bahasa Indonesia. Consider examples (26) and (27),
which exemplify an epistemic and an illocutionary concessive,
respectively:
Bahasa Indonesia (Austric; Melnie Tanudjaja p.c.)

(26) Ia tidak dirumah,
he NEG at.home
walaupun mobil-nya diparkir didepan.

although car-poss parked outside
'He's not at home, although his car is parked outside.'
328 Mily Crevels
(27) Saya dapat mengerti persoalan anda, namun

I EMPH understand problem your nevertheless
berusahalah supaya pekerjaan anda besok selesai.

try COMPLR work your tomorrow finished
,Although I understand your problems, get the work done
tomorrow!'
3.2. Concessive linkers
3.2. 1. Lexical contiguity
On the basis of the hypothesis in Greenberg (1978) that there is a

systematic relation between typological variation and diachronic
change, Hypothesis 3 may finally be formulated. Among other things
this hypothesis predicts the distribution of polyfunctional concessive
linkers.
HYPOTHESIS 3
A concessive linker can only be used to introduce constructions on
different semantic levels if these levels form contiguous categories in
hierarchy (13).
This hypothesis predicts the (non-)occurrence of the patterns in (28):
(28) Content > Epistemic > lliocutionary > Text

Linker 1 + + + +
Linker 2 + + +
Linker 3 + +
Linker 4 +
Linker 5 + + +
Linker 6 + +
Linker 7 +
Linker 8 + +
Linker 9 +
Linker 10 + +
Linker 11 +
Linker 12* + +
Linker 13* + +
This third hypothesis is also borne out in the languages investigated, i. e.

the linkers appear at contiguous semantic levels, as shown in Table 4. 9
Moreover, it provides proof of a cut-off point between the content and
epistemic level, since Japanese, Mestreechs and Kwaza have a
concessive linker that only applies to the content level. Consider the
Mestreechs examples with the content concessive in (29) and the
epistemic concessive in (30):
Mestreechs (Indo-European; Ineke Jongen and Elisabeth Jongen-

Kobben p.c.)
(29) Er heet geine bril
he has no glasses
ondaanks tot 'r slech zuut.

in.spite.of.the.fact that he bad sees
'He doesn't wear glasses although he sees very little.'
(30) 'R ging weg bei z 'n vrouw en kinder,

he went away PREP his wife and children
allewel er va61 vaan hun heelt.

although he much PREP them loved
Furthermore, a wide range of languages show a cut-off point between

the epistemic and the speech-act level. Consider the Kannada epistemic
and speech-act examples in (31) and (32), respectively.
330 Mily Crevels
Table 4. Lexical contiguity

Duunggidjawu IRR IRR IRR IRR
Tahitian A A A A
Motu A A A A
San A A A AB
Cantonese A A A B
Ket AB AB AB B
Finnish A A A B
Nahuatl A A AB AB
Kiwai A A AB/IRR B/IRR
Hungarian ABC ABC BCDE BCF/IRR
Kannada A A ABC C/IRR
Romani (Erli) A A AB BC
Boboda A A B C/IRR
Turkish A A AB C/IRR
Uzbek A A ABCD BE/IRR
Mongolian A A AB AIIRR
Dargi ABCD ABCD AC E
Basque AB AB ABCD DEFG
W-Greenlandic A A AB ACD
Bahasa Indonesia AB AB ABCD ABCD
Burmese A A B B/IRR
Lokono A ABC ABCD D
Amharic AB ABC ACD DE
Wolaitta ABCD ABCDE ABCE AE
Yoruba A AB ABC B
Japanese AB BCDE DE DE
Pima A AB B/IRR B/IRR
Mestreechs A BC BC BC
Lakxota A A/IRR A/IRR A/IRR
Kwaza A B B B
Kannada (Elamo-Dravidian; D.N.S. Bhat p.c.)

(31) avan-ge heNDati-makka-La me: Ie oLLe pri: ti
he-DAT wife-child-PL on much love
i-dd-ar-u: avO avara biTT-ikki

be-PREs-COND-EMPH he them left-coMPL
ho:-d-d
gO-PAST-3MSG
(32) a:nu he:Lu-d-ara ni:nu namb-add-i-pp-al-u: sa:ku,

I say-it-GEN you believe-NEG-be-sBJv-EMPH possible
a: dare a:nu tappitastd alIa

but I wrongdoer not
'I'm innocent, although you probably don't believe me.'
As can be seen in the Appendix, -ar-ude is the only linker used at the
content and epistemic level in Kannada; the speech-act level, however,
shows the use of other linkers as well.
Finally, Table 4 also shows us proof for a cut-off point between the
speech-act and the text level. Consider, for example, (33) and (34) with
a Dargi (Daghestan) illocutionary and textual concessive, respectively:
(33) Dargi (Caucasian; Helma van den Berg p.c.)

he-Ia Giyan-ti nu-ni
yOU:OBL-GEN difficulty-PL:ABS me-ERG
irg-uli-xalli, !nnei diafal taman+d-ar-a

understand-GER-although work tomorrow end+pl-do-IMP
,Although I understand your problems, get the work done
tomorrow!'
(34) A: se d-udi-ehe -w nusa-ni,

what pl-drink-FUTlp-INT We-ERG
farqJi-Ia, cagir-Ia Ahmad?

wodka-GEN wme-GEN Ahmed
B: farqJi-Ia, sen+ahen-nu nusa-ni d-a ?ari

wodka-GEN why+be.not-so We-ERG pi-enough
d-erdi-i-I-ra
pl-drink-AOR-GER-l
332 Mily Crevels
A: 'What shall we drink, wodka or wine, Ahmed?'

B: 'Wodka probably, although we already drank enough.'
In Dargi the concessive markers -Gi and -xalli are attached-with or

without the emphatic linker -ra--to a non-finite verb form at the
content, epistemic and speech-act level~ the text level, however, only
allows for the adverbial linker sen+ahen-n in combination with a finite
verb form.
4. Conclusion
In the foregoing I have tried to demonstrate that concessive

constructions may be used on four different semantic levels, and that
there exists a systematic correlation between the semantic type of these
constructions and the way in which they are expressed. The testing of
the three hypotheses against the data from 30 languages has furnished
clear evidence for the various cut-off points within these languages,
which again provides a clear proof of the linguistic reality of the various
semantic levels that have been claimed.
Abbreviations
1 first person IND indicative

2 second person INT interrogative
3 third person IRR irrelevant
ABS absolutive LOC locative
ACC accusative M masculine
AOR aorist MIT mitigation
COMPL completive NEG negative
COMPLR complementizer NR norninalizer
CONC concessive OBL oblique
COND conditional PL plural
DAT dative pI non-human 3rd person plural;
DECL declarative human 1st-2nd person plural
EMPH emphatic POSS possessive
ERG ergative PREP preposition
FACT factitive PROG progressive
FUT future SBJ subject
GEN genitive SBJV subjunctive
GER gerund SG singular
IMP imperative v verb

IMPOT impotential V-2 verb second
INCPAST incomplete past VN verbal noun
Appendix: Table ofconcessive linkers
Language Content Epistemic Speech-act Text

Amharic b(i)-V-i"m b(i)-V-i"m b(i)-V-i"m
[AFRO-ASIATIC]
liw-V -m liw-V -m
nagar gi"n nagar gln
gin gi"n gin
liiniigiiru
Bahasa walaupun walaupun walaupun walaupun
Indonesia meskipun meskipun meskipun meskipun
[AUSTRIC]
biarpun biarpun
namun namun
Basque arren arren arren
[ISOLATE]
nahiz eta nahiz eta nahiz eta
ba-Vere
baina baina
hala ere
dena dela
beno
Boboda kasiJr:; kasiJr:;
[NIGER-
hale
KORDOFANlAN]
hale ...ye
yEri
IRR
Burmese pemE? pemE?
[SINO-TIBETAN]
da be mE? da be mE?
IRR
Cantonese seuiyihn... seuiyihn... seuiyihn...
[SINO-TIBETAN]
daahnhaih daahnhaih daahnhaih
batgwo
Dargi -Gi -Gi -Gi
[CAUCASIAN]
-Gi-ra -Gi-ra
GER-ialli GER-xalli GER-ialli
GER-ialli-ra GER-ialli-ra
334 Mi/y Creve/s

sen+anen-nu
Duunggidjawu IRR IRR IRR IRR
[AUSTRALIAN]
Finnish vaikka vaikka vaikka

[URALIC-YUKAGHIR]
toisaa/ta
Hungarian pedig pedig
[URALIC-YUKAGHIR]
ambar ambar ambar ambar
ugyan ugyan
ugyan ... de
de
es megis
IRR
Japanese GER-mo
[KOREAN-lAPANESE]
-noni -noni
-nimokakawarazu
-ga -ga -ga
-keredomo -keredomo -keredomo
Kannada -ar-ude -ar-ude -ar-ude
[ELAMO-DRAVIDIAN]
-are nija
a: dare a: dare
IRR
Ket qaj qaj qaj
[ISOLATE]
qo °t qo °t qo °t qo °t
Kiwai nanie nanie nanie
[INDO-PACIFIC]
nanime nanime
IRR IRR
Kwaza -/e'te
[ISOLATE]
-wara -wara -wara
Lakxota eyash eyash eyash eyash
[AMERIND]
IRR IRR IRR
Lokono barhin barhin barhin
[AMERIND]
barhin ... barhin ...
kiaki kiaki
barhin ... ma barhin ... ma
ma ma
Mestreechs 6ndaanks tot
[INDO-EUROPEAN]
a/Jewel a/Jewel a/Jewel

meh meh meh
Mongolian bolovc bolovc bolovc
[ALTAIC]
IRR
Motu ena be ... to ena be ... to ena he ... to ena be ... to
[AUSTRIC]
Nahuatl maske maske maske maske

[AMERIND]
maske ... pero
Pima le:da le:da
[AMERIND]
haba haba haba
IRR IRR
Romani (Erli) nisto kaj nisto ka} nisto ka}
[INDO-EUROPEAN]
makar ka}
ama ama
San alaan€ alaan€ alaan€ alaan€
[NIGER-KORDOFANIAN]
sini
Tahitian noa atu a noa atu a noa atu a noa atu a
[AUSTRIC]
Turkish ragmen ragmen ragmen

[ALTAIC]
ama
zaten
IRR
Uzbek bolsa ham bolsa ham bolsa ham
[ALTAIC]
khaqiqatta khaqiqatta
togridir
khaqiqatta ...
ammo
ammo
IRR
W-Greenlandic -galuar- -galuar- -galuar- -galuar-
[ESKIMO-ALEUT]
-galuar- ...
kisianni (-Ii)
kisianni(-Ii)
-Ii
Wolaitta -kko-kka -kko-kka -kko-kka -kko-kka
[AFRO-ASIATIC]
-kko-nne -kko-nne -kko-nne
-sinka -sinka -sinka
-iidi-kka -iidi-kka
336 Mily Crevels

-sin -sin -sin
Yoruba bi 6 til~ j~ pe bi 6 til~ j~ pe bi 6 til~ j~ pe
[NIGER-KORDOFANIAN] am~
am~
§ugb~n
Notes
* Without the help and patience of numerous native speakers and linguists this
paper would not have had its actual form. I cannot thank them enough for their
generosity in providing me with valuable data. Furthermore, I am grateful to
Elizabeth Couper-Kuhlen, Verena Haser, Kees Hengeveld, Bernd Kortmann,
Manfred Krug, and Hein van der Voort for their thorough comments on an
earlier version of this paper.
1. The restriction on focusability as such may not be the manifestation of a
syntactic/semantic, but rather of a pragmatic constraint.
2. Giinthner (this volume) argues that concessives of this type with German
obwohl 'although' are not concessive in the sense of Konig, because the
presupposition 'Ifp, then normally --q' supposedly does not hold.
3. It should be noted that Though in (11) does not function as a subordinator, but
as an adverbial linker, or conjunctional adverb. For a detailed discussion on
subordinators, adverbial linkers and coordinators the reader is referred to
Crevels (2000).
4. The sample .languages have been selected according to the Diversity Value
(DV) method as proposed by Rijkhoff and Bakker (1998). Contrary to other
methods (e.g. Dryer 1987), in which typological and geographic independence
are a sample's most important criteria, the chance offinding maximal variation
is the greatest when the DV method is applied, precisely because this method
creates maximal genetic distance between the languages which are finally
included in a sample. For a detailed description of the sampling procedure the
reader is referred to Rijkhoff and Bakker (1998) and RijkhofI et ale (1993).
5. Note that the translations between single quotes correspond to the sentences in
the questionnaire.
6. Obviously, the marking of the apodosis (q) indicates a more paratactic or
coordinate-like type of construction.
7. It goes without saying that textual concessives are not relevant in this context,
since textual concessives, as we have seen before, do not modify the main clause
of a concessive construction, but a whole preceding text unit which may be
composed of various sentences.
8. Note that by marking the q with -wara 'but' the consultant has used a
coordinating strategy to express the questionnaire sentence. For a more detailed
account of the different clause combining strategies used by the consultants see
Crevels (2000).
9. The characters in Table 4 correspond to the concessive linkers listed for each
language in the Appendix. Thus, the A corresponds to the first listed linker,
the B to the second, etc.
References
Bolkestein, A. Machtelt
1991 Causally related predications and the choice between parataxis and
hypotaxis in Latin. In: R. Coleman (ed.), New Studies in Latin
linguistics, 427-451. AmsterdamlPhiladelphia: Benjamins.
Crevels, Mily
1994 La conexi6n concesiva en espanol: una aproximaci6n funcional. MA
thesis, Department of General Linguistics and Department of
Spanish, University of Amsterdam.
1998 Concession in Spanish. In: Mike Hannay and A. Machtelt Bolkestein
(eds.), Functional grammar and verbal interaction, 129-148.
Amsterdam/Philadelphia: Benjamins.
2000 Concession: a typological study. Dissertation, University of
Amsterdam.
Devriendt, Betty, Louis Goossens and Johan van der Auwera (eds.)
1996 Complex structures: A functionalist perspective. BerlinlNew York:
Mouton de Gruyter.
Dik, Simon C.
1989 The theory of Functional Grammar. Part 1: The structure of the
clause. Dordrecht: Foris.
Dik, Simon C., Kees Hengeveld, Elseline Vester and Co Vet
1990 The hierarchical structure of the clause and the typology of adverbial
satellites. In: Jan Nuyts, A. Machtelt Bolkestein and Co Vet (eds.),
25-70.
Dryer, Matthew S.
1987 Large linguistic areas and language sampling. Studies in Language
13: 257-292.
Giv6n, Talmy
1984 Syntax: a functional-typological introduction, 2 vols. Amsterdam!
Philadelphia: Benjamins.
Greenbaum, Sidney
1969 Studies in English Adverbial Usage. London: Longman.
Greenberg, Joseph
1978 Diachrony, synchrony and language universals. In: Joseph
Greenberg, Charles A. Ferguson and Edith A. Moravcsik (eds.),
Universals ofHuman Language, Vol. 1: Method and Theory, 61-92.
Stanford: Stanford University Press.
Grice, H.P.
1975 Logic and Conversation. In: Peter Cole and Jerry Morgan (eds),
Syntax and Semantics, Vol 3: Speech Acts, 41-58. New York:
Academic Press.
338 Mily Crevels
Gtinthner, Susanne.
1993 "...weil-man kann es ja wissenschaftlich untersuchen"-
Diskurspragmatische Aspekte der Wortstellung in WEIL-Satzen.
Linguistische Berichte 143/1993: 37-59.
Haiman, John and Sandra A. Thompson (eds)
1988 Clause combining in grammar and discourse.
AmsterdamlPhiladelphia: Benjamins.
Hengeveld, Kees
1989 Layers and operators in Functional Grammar. Journal ofLinguistics
25.1: 127-157.
1990 The hierarchical structure of utterances. In: Jan Nuyts, A. Machtelt
Bolkestein and Co Vet (eds.), 1-23.
1992 Non-verbal predication: theory, typology, diachrony. Berlin/New
York: Mouton de Gruyter.
1996 The internal structure of adverbial clauses. In: Betty Devriendt,
Louis Goossens and Johan van der Auwera (eds.), 119-147.
1997 Cohesion in Functional Grammar. In: John H. Connolly, Roel M.
Vismans, Christopher S. Butler and Richard A. Gatward (eds.),
Discourse and Pragmatics in Functional Grammar, 1-16.
Berlin/New York: Mouton de Gruyter.
1998 Semantic type and expression format: On the internal structure of
adverbial clauses. In: Johan van der Auwera (ed.), 335-419.
Konig, Ekkehard
1986 Conditionals, concessive conditionals and concessives: areas of
contrast, overlap and neutralization. In: Elizabeth Traugott, Alice Ter
Meulen, J." Snitzer Reilly and Charles A. Ferguson (eds.), On
conditionals, 229-246. Cambridge: Cambridge University Press.
1988 Concessive connectives and concessive sentences: Cross-linguistic
regularities and pragmatic principles. In: J. Hawkins (ed.),
Explaining Language Universals, 145-166. New York: Blackwell.
1994 Concessive clauses. In: R.E. Asher (ed.), The Encyclopedia of
Language and Linguistics, 679-681. Oxford: Pergamon Press.
Konig, Ekkehard and Johan van der Auwera
1988 Clause integration in German and Dutch: conditionals, concessive
conditionals and concessives. In: John Haiman and Sandra A.
Thompson (eds.), 100-133.
Lyons, John
1977 Semantics, 2 vols. Cambridge: Cambridge University Press.
Nuyts, Jan, A. Machtelt Bolkestein and Co Vet (eds)
1990 Layers and levels of representation in language theory.
AmsterdamIPhiladelphia: John Benjamins.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik
1985 A comprehensive grammar of the English language. London:
Longman.
Rijkhoff, Jan and Dik Bakker
1998 Language sampling. Linguistic Typology 2-3: 263-314.
Rijkhoff, Jan, Dik Bakker, Kees Hengeveld and Peter Kahrel
1993 A method of language sampling. Studies in Language 17.1: 169-203.
Sweetser, Eve E.
1990 From etymology to pragmatics. Metaphorical and cultural aspects of
semantic structure. Cambridge: Cambridge University Press.
van der Auwera, lohan (ed.)
1998 Adverbial constructions in the languages of Europe. Berlin/New
York: Mouton de Gruyter [with D6nall P. 6 Baoill].
Causal and concessive clauses:
Formal and semantic relations
Ekkehard Konig and Peter Siemund
It is an intuition of long-standing that concessive constructions are some-

how the negative or contradictory counterpart of causal constructions. This
intuition is expressed by terms like "incausal", "anticause" or "inoperant
cause" that are frequently used instead of "concessive". It is shown that this
intuition, which is supported by a wide variety of facts across languages,
can be explicated by analysing the meaning of causal and concessive con-
structions in such a way that the external negation of the former is equiva-
lent to the internal negation of the latter. A semantic analysis for concessive
constructions is proposed which meets this criterion of adequacy. Further-
more, it is shown that this opposition between concession and causality can
also be observed in "interactive patterns of conceding". What is negated in
such interactive schemas of concession is the assumption of an interlocutor
that some fact is a reason for a specific conclusion.
1. Introduction
It is a well-known fact that among the set of circumstantial ("adverbial")

relations usually distinguished in grammars for natural languages there
is a subset whose members are related to each other in interesting ways:
"conditional", "causal", "concessive conditional" and "concessive" are
the labels traditionally used for these relations as well as the connectives
(i.e. prepositions, conjunctions, adverbs) expressing these relations. The
following sentences are instances of these relations:
(1) a. Ifpeople smoke, they will damage their health. (conditional)

b. Even ifyou smoke, you may become very old. (concessive con-
ditional)
c. John's lungs are not in good shape because he smokes.
(causal)
d. Even though John smokes, he is still in good health. (conces-
sive)
342 Ekkehard Konig and Peter Siemund
If we arrange these four relations in a two-dimensional diagram we

can describe some of the properties shared and some of the oppositions
as follows:
Table 1. Sense relations
Hypothetical Factual
Harmony conditional causal
Dissonance concessive conditional concessive
A first, vertical division distinguishes relations whose arguments are

factual from those whose arguments are hypothetical. Both sentences
with causal and sentences with concessive adverbial clauses assert the
truth of the two clauses thus combined, in contrast to conditionals and
concessive conditionals. The horizontal division distinguishes between
relations whose arguments are in harmony with general tendencies, i. e.
they describe typical sequences of situations or a typical concomitance
of situations in the world, as opposed to relations between situations that
are not in harmony with such tendencies, i. e. express a dissonance with
general regularities of cooccurrence. Given these similarities and oppo-
sitions it is to be expected that we may find either similarities in the
formal expression between conditionals and causals, on the one hand,
and between concessive conditionals and concessives, on the other. Al-
ternatively, languages may primarily give expression to the vertical as-
pects of relatedness and code causals and concessives in some similar
fashion and also employ similar formal devices for conditionals and
concessive conditionals. 1 Finally, we should not expect any similarity in
the expression of causal constructions and concessive conditionals. All
of these expectations are clearly substantiated by a variety of languages.
A typology of adverbial relations could therefore be based on the crite-
rion of whether languages give priority to the horizontal or vertical rela-
tions.
In addition to the structure described in Table 1 there are some more
specific ways in which adjacent items are related to each other. Many
analyses of causal connectives assume that causality involves a presupposi-
tion of counterfactual conditionality (p because q == p & q ; -, p ~ -, q; cf.
Dowty 1979: 100ff.). Concessive conditionals are clearly further elabora-
tions of (simple) conditionals, brought about by the addition of the scalar
particle even or by formulating a conditional relationship for an extreme
antecedent, as in (2):
Causal and concessive clauses 343
(2) I would not marry you ifyou were THE LAST MAN ON EARTH.
Concessive conditionals typically develop into genuine concessives

or may at least be used as such whenever the truth of the antecedent is
contextually given (cf (3a». Moreover, the example from Shake-
speare's Hamlet shows that English though still had a concessive condi-
tional meaning ('even if) in Early Modem English:
(3) a. Even ifyou dislike museums - which I know you do - the Mu-
seum ofModem Art is worth a visit.
b. I'll speak to it though hell itselfshould gape and bid me hold
my peace. (Sh. Hamlet I.ii.244)
These specific types of relatedness can be summarised as follows (cf.

Kortmann 1997: 197ff.):
(4) a. causal constructions presuppose conditionals

b. conditionals can be elaborated, enriched to concessive condi-
tionals
c. concessive conditionals may change into concessives
This leaves open the question as to the precise relationship between

causality and concessivity and it is this question we intend to specifi-
cally address in this paper.
There is a clear intuition that concessive clauses and complex sen-
tences with such clauses are rarely, if ever, used for the speech act or
speech activity of conceding and that they should rather be analysed as
standing in opposition to causal constructions. As a consequence, it has
often been suggested that the term "concessive" is really a misnomer
and that labels like the following would be much more appropriate: "in-
causal", "inoperant cause", "anticause", "cause contraire", "condition
contrecarre", etc. On the other hand, studies of concessive patterns in
verbal interaction have often discussed formal properties that are hardly
ever subsumed under the label "concessive" in grammatical analyses
and grammars (cf Couper-Kuhlen and Thompson 1998). More often
than not, however, the assumption that there is some relation of opposi-
tion between causal and concessive constructions is simply posed and
defended as being based on clear intuitions without much concern for
the basis of this intuition. Our first goal therefore is to provide such a
basis: it will be shown that a wide variety of facts clearly support the
view that concessive constructions are in some sense the opposite of

causal constructions. These facts will later be confronted with various
analyses of concessive constructions and used as criteria of adequacy in
an evaluation of these proposals.
2. Causal and concessive connectives: phenomena in need of

explanation
The first fact is the equivalence of sentences like the following, given a
certain intonation pattern and information structure: 2
(5) a. / The house is no less comfortable because it dispenses with

AIR-conditioning. /
b. / The house is no LESS comfortable / although it dispenses with
AIR-conditioning. /
As is indicated by the capitalisation, (5a) is meant to be read as one

tone group, with one nuclear accent, whereas (5b) has two focal accents
and thus contains two tone groups. Since the equivalence in question
cannot be the result of a synonymy between because and although, it
must be due to a different interaction between negation and the adver-
bial adjuncts, as indicated by the intonation. What we have is a differ-
ence in the scope of the negation: wide scope over both clauses in (5a)
and narrow scope over the main clause only in (5b). What the two ex-
amples in (5) show, therefore, is that a negated causal construction with
wide scope of the negation can be paraphrased by its concessive coun-
terpart with narrow-scope negation. Moreover, we find the same para-
phrase relations between causal constructions with double (both external
and internal) negations and simple non-negative concessive construc-
tions:
(6) a. It is NOT the case that Brutus's love for Caesar was a reason
for not killing him.
b. Brutus killed Caesar, although he LOVED him.
What our examples also show is that the relevant relations of scope
go together with specific intonation patterns, which are ultimately indi-
cators of different information structures. Whether or not differential
scope is simply a consequence of information structure is a question that
will be taken up again later. Note furthermore that the wide-scope nega-
tion in (5a) and (6a) is not an instance of the focusing not ... but nega-
tion. The distinction between a focusing, or "replacive", not ... but ne-
gation with a focused adverbial clause and a sentential negation that
does not relate to the adverbial clause is illustrated by the following
minimal pairs. There is a fall-rise on money in (7a), whereas the intona-
tion of (7b) is falling:
(7) a. / John is not unhappy because he lost a lot ofMONEY / , but ...
b. / John is NOT unhappy because he has lost a lot ofmoney. /
An analogous contrast can be observed in (8 alb): the scope of interroga-

tion is wide in both cases, but the adverbial clause is focused in (8a) and
thematic in (8b).
(8) a. Are we racists because we do business with SERBIA? ( ... or is it

because ... ?)
b. Are we RACISTS because we do business with Serbia?
(Does our doing business with Serbia make us racists?)
What is rejected in (5a), (6a) and (7b) is the causal connection be-
tween the two situations described in the clauses and, as a result, also
the content of the main clause. The subordinate clause is not affected by
the negation and can therefore be assumed to be presupposed. The fol-
lowing paraphrases of (7b) show this even more clearly:
(9) a. Losing a lot ofmoney has not made John unhappy.

b. John has lost a lot ofmoney without therefore being unhappy.
Closely related to the first set of facts is the observation that the
paraphrase relations discussed above can also be found in the corre-
sponding paratactic constructions. In such examples the cause of a con-
sequence is identified in a separate sentence preceding the negated con-
sequence:
(10) a. John has lost a lot ofmoney. He has not become a poor man
because ofthat.
b. John has lost a lot ofmoney. He has not become a poor man in
spite ofthat.
Since the presuppositions of a word or construction can always be

spelt out in the preceding co-text, these examples also suggest very
forcefully that the adverbial clauses in both causal and concessive con-
structions may be presupposed. Note, moreover, that one possible Ger-
man translation of the examples in (10) contains both a causal and a
concessive adverbial (in either order), clearly the result of blending for-
mal properties of two equivalent but formally different constructions: 3
(11) Johann hat viel Geld verloren. Deshalb ist er trotzdem kein armer
Mann geworden. / Trotzdem ist er deshalb kein armer Mann ge-
worden.
A related phenomenon in English can be found in sentences with

comparatives and the causal connective for:
(12) John is not an Oxbridge man, though none the worse for that.
Our third observation relates to a frequent, but very surprising pat-

tern of semantic change. Causal connectives may develop into conces-
sive ones, even though the reverse direction has not been attested. In
Old English for was the most common causal connective and Scandina-
vian languages still use a cognate form of this connective. In Modem
English, for can still be used in a causal sense, but apart from the adverb
therefore, this use of for has become somewhat marginal. The conjunc-
tion for is very formal, even bookish in its stylistic quality and the
prepositionfor is restricted to a few contexts and fixed collocations:
(13) a. He was rewardedfor his bravery.

b. We could hardly see for the thickfog.
c. for fear of, for want of,for this reason, etc.
In addition to this causal use and to various others, for may also have
a concessive interpretation. In fact, the concessive interpretation is the
only option whenever for combines with all:
(14) a. For all his personal responsibility for the manipulation of

power, he has not done anything illegal.
b. He can be very selfish, but he is not unpleasant to work with
for all that.
That the concessive interpretation of for should require a supportive

marker all is not really surprising. Such a quantifier can be found as a
component in many concessive connectives (cf English withal, al-
though, albeit, all the same; German bei all, allerdings; French toute-
fois, etc.). What is remarkable, though, is that the same connective
should have both a causal and a concessive meaning and that the latter
should be derived from the former.
The clearest evidence for the assumption that there is a relationship
of contradictory oppositeness between causal and concessive construc-
tions would be provided by languages where a concessive is the result
of combining a causal connective with a negator. Unfortunately, such
cases seem to be extremely rare, but they are certainly attested. In
Lezgian kiligna is the aorist converb of kiligun 'look' and literally
means 'having looked, having considered'. This deverbal postposition is
primarily used in the sense of 'because, in accordance with, according
to'. The causal meaning is particularly prominent when kiligna is com-
bined with a masdar. 4 The negated form kilig tawuna (literally 'not
looking, disregarding') has the meaning ,despite' (Haspelmath 1993 :
224). A related fact is the use of negated expressions as concessive con-
nectives, such as the following: German ungeachtet, English regardless,
French n 'empeche que, Dutch ondanks, Finnish huolimatta,5 etc. In all
of these cases, the relevant expressions have some kind of negated
causal meaning (e.g. empecher 'not cause that', dank(zij) 'because of).
Moreover, negated expressions of proportional correlation such as Eng-
lish nevertheless, nonetheless, French neanmoins, etc. can also be re-
garded as negated counterparts of basically causal expressions: NEG (the
more p, the more q) = (p; nevertheless -.q).
3. Some recent proposals
More often than not, the intuition that concessive relations are semanti-
cally the opposite of causal relations is simply mentioned rather than
explicated (cf Hermodsson 1973, 1994; Azar 1997). In what follows,
we will discuss some recent attempts at providing such an explication as
a preparatory step towards presenting our own explication.
The most recent contribution to the topic comes from Di Meola
(1998). The basic assumption guiding his proposal is that concessive
relations do not involve only one underlying causal relation, as assumed
in many previous studies, but that there are two such relations operating
in the background. The following example can be used to illustrate this

idea:
(15) Although Carol is ill, she goes to work today.
According to Di Meola, this concessive construction is based on the

hypothetical causal relation given in (16A) as well as on the real relation
in (16B):
(16) A: Since Carol is ill, she cannot go to work.

B: Since there is an important business meeting today, Carol goes
to work.
The relations expressed in (16) amount to the schema given below

(cf Di Meola 1998: 338):
(17) AI: being ill ~ A2: not go to work

B1: important business meeting ~ B2: go to work
The consequents of these two causal relations are mutually exclusive

whereas the antecedents are ordered hierarchically. The important point
about concessive relations is that two causes compete with one another
and that the stronger or more important one wins out (B 1 in our case).
What Di Meola's idea essentially boils down to is that concessive con-
structions code antecedent and consequent of two different causal rela-
tions (AI - B2). A noteworthy property of this analysis is that the 'suc-
cessful' cause (B 1) is almost never linguistically coded. Di Meola there-
fore defines concessivity as "hidden causality".
One of the problems for this analysis identified by Di Meola himself
is that there appear to be concessive relations without "hidden causes".
For cases like (18) it is not plausible to assume double causality (in
(18a» e.g. it is impossible to reconstruct a conceivable cause why the
flat is located in a nice area), and various stipulations are necessary in
order to explain them within this framework. As a matter of fact, the
assumption of causality, hidden or not, is not particularly fortuitous in
these cases to begin with.
(18) a. Although the flat is too small, it is located in a nice area.

b. He was laughing, although he was desperate.
c. The ostrich is a bird, although it cannotfly.
However, Di Meola's theory also encounters problems with much

less complicated examples. Consider the following sentences:
(19) a. Although the flat is expensive, it is not well-equipped.

b. Although the pizza is expensive, it does not taste good.
c. Although the programme is bad, many people watch television.
d. Although he has a lot ofmoney, he does notfeel happy.
In all of the above examples, general assumptions of plausibility

about the interrelation of events are violated (expensive flats are nor-
mally well-equipped, etc.) and hence they are easily explained within
more traditional accounts of concessivity. To reconstruct a hidden
cause, however, appears quite difficult. Let us try to translate one of the
above examples into the schema proposed by Di Meola (e.g. (19b)).
(20) AI: Pizza is expensive ~ A2: Pizza tastes good

Bl: ? ~ B2: Pizza does not taste good
The problem obviously consists in the fact that it is not easy to re-
cover a cause for the bad taste of the pizza. Although such a cause might
in fact exist, say decayed ingredients, under the most plausible interpre-
tation this sentence simply expresses that the pizza is overpriced for its
taste. 6 Although the symmetrical nature of Di Meola's analysis is very
appealing, examples like (19) make one wonder if it is at all the case
that concessive constructions encode a "hidden cause".
To avoid all these problems, Di Meola's analysis should perhaps be
restricted to what Sweetser calls linking at the content level (Azar's
1997 "indirect-rejection concessivity"). Such a move might exclude (18)
as potential counterexamples. But even for this subset of concessive
constructions, there are fundamental problems: the second cause (B 1) is
never encoded, nor does it seem to play any role as background assump-
tion for conversational inference, i. e. a sentence like (19b) can never
give rise to a conversational inference for which propositions like 'the
pizza is cheap/old ... ' are used as a premise. Moreover, its basic seman-
tic outlines are never contextually given. Above all, Di Meola has noth-
ing to say on any of the phenomena discussed in the preceding section.
Finally, there is also the problematic view, which has not been men-
tioned so far, that concessive constructions are somehow non-
cooperative, involving as they do "hidden causality". If the Gricean
maxims of cooperative conversation are an essential basis for any com-
municative interaction, it is highly implausible that an uncooperative

pattern should become formally entrenched in most if not all languages.
Another very different attempt to explicate the relationship between
causality and concessivity can be found in Konig (1991), where this
relationship is seen as a specific instance of duality. As is well known, a
relationship of duality holds between two scope-bearing expressions if
the sentences containing them can be translated into one another by
means of internal vs. external negation. A duality square is usually used
for the sake of visualisation, with dual relations holding between diago-
nally opposed comers. Clear representatives manifesting this relation-
ship are the modal operators possible/may vs. necessary/must, permit vs.
force, the phase quantifiers already vs. still and universal vs. existential
quantifier, as is shown by the following equivalencies (cf Lobner
1990): t/x Px == -,3 -, Px; -,t/x Px == 3x -,Px.
Applied to the relationship between causality and concessivity, this
now predicts that (although p, q) should be equivalent to -,«because p)
-,q) and this prediction is indeed borne out by the data. As already indi-
cated in (6), the following two sentences are equivalent in meaning:
(21) a. It is NOT the case that Brutus's love for Caesar was a reason
for not killing him.
b. Brutus killed Caesar, although he LOVED him.
A relationship of duality should also manifest itself in the equiva-

lence of (22a) and (22b).
(22) a. The road is not getting wet although it is raining.

b. It is not the case that the road is getting wet because it is rain-
ing.
And again this prediction is correct. Unfortunately, there are prob-

lems with the duality view, as already indicated in Konig (1991). Con-
cessive sentences cannot be transformed into causal ones by applying
internal and external negation. (23a) shows the relevant negative form
of (22a), but is not equivalent in meaning to (23b), if it is meaningful at
all.
(23) a. It is not the case that the road is getting wet although it is rain-
ing.
b. The road is getting wet because it is raining.
Pasch (1992a, 1992b) rejects this analysis for several reasons. The
first reason is that the truth conditions for concessive constructions and
negated causal constructions do not come out identically, according to
her analysis. But this argument depends very much on the details of the
analysis. The perspective of Konig (1991) is different: our intuitions
about an opposition between causality and concessivity, which mani-
fests itself in the phenomena discussed in section 2, should be taken as a
basic problem for their semantic analysis. In other words, the ability to
explicate this relationship should be taken as a criterion of adequacy for
such analyses. Pasch's second reason is more to the point and impossi-
ble to refute. The relationship between 'because p, q' and 'although p,
q' is not a clear instance of duality, since one does not get the expected
paraphrases, except for the case where a wide-scope negation of a
causal construction clearly corresponds to a concessive construction
with the negative counterpart of the main clause. All the pairs discussed
in section 2 (e.g. (5), (9» are of this kind. But, there is simply no way
one can get from a negated concessive causal construction to a causal
one. The relevant paraphrase relations seem to hold only in one direc-
tion. Moreover, Pasch is right in pointing out that the internal negation
of a causal construction «because p) --, q) does not admit of a conces-
sive paraphrase «although -,p) q):
(24) a. Since it is so hot, there are not many people in the streets.
b. Although it is not hot, there are many people in the streets. =I:-
(24a)
Worse still, causal constructions that make sense lose this quality as
soon as the main clause is negated and therefore have no paraphrase in
terms of concessivity:
(25) a. The streets are wet, because it has rained.

b. The streets are not wet, because it has rained.
So, Pasch's conclusion that the relationship between causality and

concessivity is not a clear instance of duality is certainly right. Still,
there are the interesting cross-linguistic facts discussed in section 2. As
in most studies of concessivity, Pasch is not aware of such facts, nor is
she interested in explaining them. She does, however, have some brief
suggestions for an explication of the equivalence between sentence pairs
such as (5alb). Before we discuss these suggestions we have to take a

detailed look at her analysis of causal and concessive connectives.
4. Towards an explication
Pasch's analysis of causal and concessive connectives (Pasch 1992a) is

ultimately a precise formulation of an analysis proposed informally in
various other discussions of this topic (cf for instance Konig 1988,
1994). We will neglect these precise formal details and present a
slightly simplified summary of her analysis.
In terms of categorial grammar, adverbial conjunctions have, strictly
speaking, the semantic category <t, <t, t», since they combine with
clauses (category <t» to form adverbial adjuncts (category <t, t», i.e.
expressions that combine with another clause to form a sentence. If we
simplify matters slightly, however, we can simply regard adverbial con-
junctions as two-term relations and represent both causal and concessive
constructions as in (26):
(26) a. because p, q
b. although p, q
Given such simplified representations, Pasch's analysis can be sum-

marised as follows:
Table 2. Causality and concessivity
causal constructions concessive constructions

because p, q although p, q
p ~ q (presupposition) p ~ --, q (presupposition)
p & q (assertion) p & q (assertion)
As far as their assertion is concerned, both causal and concessive

constructions are simply factual, i.e. they entail both the content of the
subordinate and that of the main clause. Consider (lc/d) again for illus-
tration: (lc) entails 'John's lungs are not in good shape' as well as 'John
smokes' and (ld) has the entailments 'John smokes' and 'John is still in
good health' respectively. The relevant part of their meaning can there-
fore be represented as a logical conjunction of the two propositions. The
difference lies in the presupposition. Loosely speaking, the two situa-
tions linked by a causal connective are in keeping with general tenden-
cies, regularities, sequences of situations, whereas the sequence or con-

comitance of situations described in a concessive construction goes
against the tide: p & q is asserted given a general tendency 'if p then not
q', i.e. the very opposite of q normally goes together with p. The arrows
in Table 2 stand for material implication of propositional logic.
There are two points where this analysis is in need of modification.
The background assumption against which the two clauses of a conces-
sive construction are asserted seems to involve some kind of generalisa-
tion over the two specific situations asserted. Thus, a sentence like Al-
though it is raining, Fred is going for a walk seems to have a presuppo-
sition like 'if it is raining, Fred does not normally go out for a walk' or
'People do not go out for a walk if it is raining', or even 'The more it
rains,' the less people go out for walks'. All of the attempts at capturing
the relevant presupposition involve some kind of quantification and
generalisation of the specific propositions p and q. These arguments also
apply, mutatis mutandis, to causal constructions. We will therefore rep-
resent the presupposition of both causal and concessive sentences as 'P
~ Q' and 'P ~ -,Q', respectively. The second problem concerns the
status of the first conjunct 'p' in concessive sentences. A wide-spread
type of concessive connectives is (partly) composed of expressions
whose basic meaning can simply be described as 'emphatic assertion of
truth'. Connectives like German zwar, obzwar, obwohl, obschon, Eng-
lish true, although, albeit, etc. provide illustration. 7 Moreover, the con-
tent of the subordinate clause does not seem to be affected by putting
concessive sentences into interrogative form (Was he fired, although he
did not take part in the demonstration?). It is therefore quite plausible to
assume that concessive constructions with wide-scope negations pre-
suppose the content of the adverbial clause. These two modifications
result in the following revised version of the analysis given in Table 2: 8
Table 3. Causality and concessivity (revised)

because p, q although p, q
P --7 Q~ p (presupposition) P --7 --, Q~ p (presupposition)
p & g (assertion) p & g (assertion)
In view of the problems mentioned at the beginning of our paper, the

question now is, of course, whether this analysis can explain the phe-
nomena discussed in section 2. As already mentioned, Pasch's analysis
certainly does not stand alone in completely neglecting such cross-
linguistic observations. At the end of her paper, however, she does have
some suggestions for the analysis of equivalent pairs like (Sa) and (Sb),
which we repeat here for convenience:
(27) a. / The house is no less comfortable because it dispenses with

AIR-conditioning. /
b. / The house is no LESS comfortable / although it dispenses with
AIR-conditioning. /
One condition for the equivalence between a negated causal con-

struction and a concessive construction with a negated main clause is -
according to Pasch - the thematic character of the causal clause. As a
result of this thematic character the clause introduced by because be-
comes part of the presuppositions of the whole sentence and is thus not
affected by the negation (Pasch 1992a: 31). This corresponds more or
less to what is said above about such constructions: the negation has
wide scope (signalled inter alia by the fact that we have a single tone
group), the adverbial clause is not focused, but presupposed. What is
negated in such cases is the causal relation between the two clauses:
(28) NEG ([this house dispenses with air conditioning] CAUSE [it is less
comfortable])
If a certain fact p does not bring about a certain situation q, then q is

clearly not the case. Let us now spell out in detail the steps demonstrat-
ing the equivalence between (27a) and (27b):
Table 4. Negated causality

(i) --, (because p, q) although p, --, q
(ii) P -7 Q~ p (presuppositions) P -7 Q~ p (presupp.)
(iii) --, (p & q)
(iv) P & --, g (since p is a presupp.)
Since the negation of a causal construction does not affect its pre-
suppositions, these go through unchanged (cf line (ii)). The negation
relating to the whole causal construction in line (i) can therefore only
relate to the assertive part of its meaning, i.e. the conjunction p & q (cf
line (iii)). Since we also assume that the causal clause represents pre-
supposed material, the negation can only affect the main clause, as is
indicated in line (iv). The external negation of a causal construction is

therefore shown to be equivalent to the internal one of a concessive con-
struction.
Thus there is practically no disagreement between Pasch's analysis
and the one proposed in Konig (1991) with regard to such examples as
(27a/b). What must go over board, though, is the furthergoing claim of
Konig that what we have here is an instance of duality. The analysis in
Table 3 and Table 4 thus comes fairly close to explicating the phenom-
ena described in section 2 and meeting a central criterion of descriptive
adequacy relevant for any theory of concessivity.
5. Discourse approaches to the analysis of concessive relations
A discourse view of concessive relations and concession (cf Thompson

1987, Thompson and Mann 1987, Couper-Kuhlen and Thompson 1998,
Couper-Kuhlen and Thompson, this volume) may lead to a very differ-
ent picture from the one given above. First, it does not seem to be a
problem to use the term "concession" or "conceding" for the relevant
data, whereas grammar-oriented studies of the written language have
always had difficulties in accepting those terms as appropriate labels.
Secondly, the markers seem to be very different ones. The standard
connectives of written English even though, although, in spite oj, never-
theless, still, just the same, etc. do not seem to play an important role in
spoken discourse, whereas but seems to be the by far most frequent
marker of concessive relations. This also means that the data typically
studied in such discourse-based approaches are very different from the
phenomena studied above, the following being a typical example:
(29) A: (After all) he is your brother. (X)

B: True (X'), but I cannot/orgive him this time. (Y)
The complementarity in the data and the markers of the relations

studied also lead to very different characterisations of the basic func-
tions of concessive markers. Sentence-oriented studies of written dis-
course typically characterise the basic function of concessive sentences
as the assertion of a remarkable fact, the assertion of a sentence that is
true in unfavourable circumstances. In interactive, discourse approaches
the analysis of concessivity typically leads to a characterisation of the
"concessive relation" as involving (a) acknowledging another's point
and (b) going on to make one's own (contrasting) point (cf Couper-
Kuhlen and Thompson 1998: 29).
Let us now take a closer, critical, look at such a discourse-oriented
approach. Couper-Kuhlen and Thompson (1998: 30), for example, char-
acterise the distinctive features of the concessive relation in interaction
by formulating the following Cardinal Concessive Schema (where X
and Yare understood to be potentially incompatible):
Table 5. Cardinal Concessive Schema
Cardinal Concessive Schema example

X A: (that guy is healthy as an ox) except for his
liver
X' B: as much as he abused his liver,
Y he is still as healthy as an ox.
As far as we can see, this analysis leads to the following problems:

first of all, there is no precise characterisation of 'potential incompatibil-
ity'. Secondly, the claim that a concessive schema involves three rather
than four propositions or interactive moves seems equally problematic
to us. It may be the case that the fourth step is typically a covert one, but
it is also clear that X in the schema given above is put forward as an
argument for something: for the conclusion 'He is not in perfect health'
in Table 5 and 'You ought to forgive him' in (29). What the speaker B
does in the schema depicted in Table 5 is to react both to the argument
X uttered by A and to the inference based on this argument, which we
can call Z. It is, in our opinion, this conclusion Z that Y is in conflict
with, not its argumentative basis X. The reason why this is not obvious
in Table 5 is that here Y is more or less the negative counterpart of X
anyway. If examples like (29) are taken, on the other hand, it is clear the
X could be given as a reason for a wide variety of implicit recommenda-
tions Z: 'You ought to give him that loan', 'You need to put in a word
for him', 'You cannot neglect him', etc. It is precisely to this implica-
ture that B reacts with utterance Y. But this means that the schema in
Table 5 has to be expanded as follows:
(30) Cardinal Concessive Schema

X (therefore Z)
X'
Y (where Y is contradictory to Z)
Reformulated in this way, the Cardinal Concessive Schema is per-

fectly in line with what was said above. In such interactions the inter-
locutor rejects the argumentative schema 'x. Therefore Z', i.e. NEG [X.
Therefore Z], by acknowledging X and rejecting the conclusion Z.
We should also add here that a clear contradictory relationship be-
tween Z and Y is only a special instance of a more general relationship.
As is shown by examples such as the following, taken from Azar
(1997), the contradictions concern the consequence Z of X and the con-
sequence W of Y:
(31) A: Why are you late for work?

B: My car broke down. (X) (+> I could not get here earlier. (Z»
A: I see (X'), but there is always public transport. (Y) (+> You
could have gotten here earlier. (W»
6. Conclusion
Our discussion was meant to show that we can only explain some inter-
esting phenomena in the formal coding of concessive relations across
languages if we succeed in providing a clear explication of the contra-
dictory relationship intuitively felt to oppose causal to concessive con-
structions. A few attempts at explication were presented and critically
examined. The basic outlines of such an explication, which we claimed
to be an indispensable criterion of adequacy for any analysis of conces-
sivity, seem to be fairly clear. It was also shown that there is no need to
develop different analyses for concessive constructions (i.e. complex
sentences with concessive adverbials) and interactive concessive
schema or adversative sentences. The analysis developed above also
subsumes the relevant concessive schema and the traditional (non-
interactive) "adversative" or "contrastive" sentences. What we need to
distinguish are, of course, different concessive sentences such as "di-
rect-rejection concessive" vs. "indirect-rejection (argumentative) con-
cessive" (cf Azar 1997) or concessive vs. adversative sentences. Unfor-
tunately, there is no unambiguous, generally accepted terminology in
this domain. What the discourse-oriented studies have shown is that the
interactive/non-interactive distinction needs to be made for all argumen-
tative or "indirect-rejection" concessives. Perhaps the term "adversa-
tive" should be restricted to all argumentative concessives (p but q) that
lack the interactive component, i. e. to such cases where two arguments
supporting contradictory conclusions are brought up by the same

speaker.
Notes
1. Note that we are not assuming that there will be many cases where the same
connective is used both in a concessive and in a causal interpretation. As is
pointed out in a typological study by Kortmann (1997: 202) this is very rarely
the case. What we do expect on the basis of the analysis developed below is
that a concessive connective will often be a negated counterpart of a causal
one. Such cases do indeed exist in a wide variety of languages: in view of, for
instance, may be regarded as a causal preposition and both regardless and
irrespective of can plausibly be analysed as the negated counterparts. On the
other hand, the relatedness between causal connectives and concessive ones
may only manifest itself in the use of the indicative mood (as in German) as
opposed to the subjunctive mood, which is used in concessive conditionals
and concessive clauses in French. Note also that negated correlative markers,
which are closely related to causal expressions, are frequently used as conces-
sive connectives (cf. English nevertheless, nonetheless).
2. A major part of the following discussion is based on Konig (1991).
3. Such cooccurrences of causal and concessive connectives seem to be only
possible in cases where we have epistemic or argumentative linking in the
sense of Sweetser (1990), rather than linking at the content level. A combina-
tion of causal and concessive adverbs always amounts to rejecting a conse-
quence or conclusion (held by an interlocutor) based on the preceding sen-
tence as a premise. Examples like the following therefore have an interactive
character, i.e. if both sentences are uttered by the same speaker, they suggest
that the second is a rejection of an opinion held by the hearer:
i. Karl hat mir sehr weh getan. Deswegen ist er aber trotzdem kein
schlechter Mensch.
'Karl really hurt me. But that does not mean he is a bad guy. '
ii. Ich habe zwei Wochen lang fast nichts gegessen. Deswegen hahe ich
trotzdem nicht abgenommen.
'I have not eaten anything for almost two weeks. 1 have not lost any
weight for all that.'
4. 'Masdar' is the term traditionally used for nominalisations of verbs or verb
phrases in Caucasian languages. Like the gerund in English, masdars have the
external syntax of noun phrases, but the internal syntax of verb phrases.
5. -matta is a derivational affix with a privative meaning: V-matta roughly
means 'without Ving'.
6. Note that it would not be possible to assume Pizza is cheap as Bl because this
simply contradicts AI. What is necessary is a stronger cause.
7. German zwar, like English true, derive from expressions encoding the notion
of 'truth' and the second components of obwohl and obschon are particles that
may be used for focusing positive polarity in declaratives ('verum focus'):
Karl hat das WOHL/SCHON zu verantworten.
8. The difference between Table 2 and Table 3 is that the subordinated clause is
analysed as representing presupposed information and that the implicational
presupposition is formulated as a connection between general tendencies (P,
Q) rather than specific instantiations (p, q).
References
Azar, Moshe
1997 Concession relations as argumentation. Text 17: 3.
Bybee, Joan et al.
1994 The Evolution of Grammar. Chicago: The University of Chicago
Press.
Couper-Kuhlen, Elizabeth and Sandra A. Thompson
1998 On the concessive relation in conversational English. In: Fritz-
Wilhelm Neumann and Sabine Schtilting (eds.), Anglistentag 1998
Erfurt. Proceedings, 29-39. Trier: Wissenschaftlicher Verlag.
Di Meola, Claudio
1997 Der Ausdruck der Konzessivittit in der deutschen Gegenwartsspra-
che. Theorie und Beschreibung anhand eines Vergleichs mit dem
ltalienischen. (= LA 372.) Ttibingen: Niemeyer.
1998 Zur Definition·einer logisch-semantischen Kategorie: Konzessivitat
als 'versteckte Kausalitat'. Linguistische Berichte 175: 329-349.
Dowty, David R.
1979 Word Meaning and Montague Grammar. Dordrecht: Reidel.
Haspelmath, Martin
1993 A Grammar ofLezgian. Berlin: Mouton de Gruyter.
Hermodsson, Lars
1973 Inkonditionalsatze. Zur Semantik der sogenannten Konzessivsatze.
Studia Neophilologica 45: 298-305.
1994 Der Begriff "konzessiv". Terminologie und Analysen. Studia Ne-
ophilologica 66: 59-75.
Konig, Ekkehard
1988 Concessive connectives and concessive sentences: cross-linguistic
regularities and pragmatic principles. In: John Hawkins (ed.),
Explaining Language Universals, 145-166. Oxford: Blackwell.
1991 Concessive relations as the dual of causal relations. In: Dietmar
ZaefIerer (ed.), Semantic Universals and Universal Semantics, 190-
209. Dordrecht: Foris.
1994 Concessive clauses. In: R. E. Asher (ed.) The Encyclopaedia ofLan-
guage and Linguistics. Oxford: Pergamon Press.
Kortmann, Bernd
1997 Adverbial Subordination. A Typology and History ofAdverbial Sub-
ordinators Based on European Languages. (= Empirical Approaches
to Language Typology 18.) Berlin: Mouton de Gruyter.
Lobner, Sebastian
1990 Wahr neben Falsch. Duale Operatoren als die Quantoren natilrli-
cher Sprache. (= Linguistische Arbeiten 244.) Tubingen: Niemeyer.
Pasch, Renate
1992a Sind kausale und konzessive Konstruktionen Duale voneinander? (=
Arbeiten des Sonderforschungsbereichs 282 - Theorie des Lexikons,
Nr. 31.) Dusseldorf: Heinrich-Heine-Universitat.
1992b Kausale, konzessive und adversative Konnektive: Konnektive als
Mittel des Ausdrucks von Diskursprasuppositionen. Milnstersches
Logbuch zur Linguistik 1: 33 - 48.
Sweetser, Eve
1990 From Etymology to Pragmatics. Cambridge: Cambridge University
Press.
Thompson, Sandra A.
1987 'Concessive' as a discourse relation in expository written English.
In: Brian D. Joseph and Arnold M. Zwicky (eds.), A Festschrift for
lise Lehiste, 64-73. (Working Papers in Linguistics 35.) Columbus,
Ohio: Ohio State University.
Thompson, Sandra A. and William C. Mann
1987 A discourse view of concession in written English. In: Scott De-
Lancey and Russell Tomlin (eds.), Proceedings of the Second An-
nual Pacific Linguistics Conference, 435-447. Eugene, Oregon: Uni-
versity of Oregon.
Concession implies causality, though in some
other space
Arie Verhagen
This paper addresses the issue of the relationship between concessivity and
causality, in the sense that these concepts can and should be construed as
parallel, concessivity being in some sense a negative counterpart of
causality. It is argued that in order to avoid the risk of internal contradiction
and other conceptual problems, it is necessary to construe the relationship in
terms of partly but not completely similar 'points of view', or mental
spaces. While aiming for an explanation of obvious facts of interpretation
and usage, the discussion focuses on clarification of conceptual questions.
1. Introduction
It is well known, as noted by Konig and Siemund (this volume), that

causal and concessive relations and the linguistic means for marking
them are conceptually related in interesting ways. Theoretical analyses
of these relations should therefore provide explanations for this
connection. The kind of phenomena Konig and Siemund draw attention
to involve, inter alia, a parallelism in the possible interpretations of
sentences such as the following:
(1) The house is no less comfortable because it dispenses with air-

conditioning.
(2) The house is no less comfortable, although it dispenses with air-

conditioning.
Especially if realized with a particular intonation contour, (1) can be

read as a paraphrase of (2): with the negation taking wide scope,
including the causal clause, (1) can express the same relationship as (2)
with the negation taking narrow scope, excluding the concessive clause.
Roughly: denying "the house is less comfortable because it dispenses
with air-conditioning" can be equivalent to stating "it is no less
362 Arie Verhagen
comfortable" while conceding "it dispenses with air-conditioning".

Even if one doubts the equivalence of (1) and (2) in all circumstances,
the fact that for certain purposes or in certain respects they can be taken
as parallels suffices to establish an interesting connection between
concessivity, causality, and negation, one which poses a criterion of
adequacy for theories of concessivity.
Konig (1991) explores the idea that this connection might be
captured in a very direct way, viz. by considering it an instance of
duality. As already pointed out at the end of that paper, there appear to
be some fundamental problems for this approach as soon as one goes
beyond the very first examples that inspired it (cf also section 3 of
Konig and Siemund (this volume». One of the problems is the
asymmetry in the relationship. Negation of causality may lead to an
interpretation as concessive, but (wide scope) negation of concessivity
does not allow for an interpretation as causality: It is not the case that
John failed his exams although he worked hard cannot be interpreted as
"John passed his exams because he worked hard"-if it is interpretable
at all. Another issue is that in this way the relationship can at best be
stipulated, not explained. It is evident that only a wide scope
interpretation of a negation of causality can correspond to a concessive
reading, but why should this constraint hold in precisely this way? All in
all, it seems unavoidable to conclude that concession cannot be taken as
negated causality in any simple sense. However, this should not lead to
the conclusion that the original observation was misguided-the parallel
between causality and concessivity still requires an explanation.
In fact, there is yet another set of interesting connections between
causality and concessivity (or more generally: contrast), viz. the fact
that they can be applied in parallel ways in different conceptual domains
in the sense of Sweetser (1990). Interestingly, it is also true that this
theory at some point runs into the same problem that the parallel is not
complete, and that it is not obvious whether the difference can be
accounted for in a principled manner. It is this connection that I will use
as a starting point for an alternative analysis of the relationship between
causality and concessivity.
2. Causals and concessives in two domains
One of the main motives for Sweetser's domain distinction is that it

provides a mechanism for accounting for the fact that similar types of
Causality in some other space 363
polysemy show up in different expressions. Consider the ambiguity of

modal verbs in terms of deontic vs. epistemic interpretations. He must
come home is an example of deontic usage; it describes some obligation
in the world (viz. in the person referred to). He must be home (by now)
illustrates epistemic usage; it locates an obligation (viz. to conclude "He
is home" on the basis of the available evidence) in the conceptualizer
rather than in the world being conceptualized. It is no coincidence that
this type of polysemy resembles the difference between distinct types of
usage of the English conjunction because, as exemplified in the contrast
between (3) and (4):
(3) John passed his exams because he worked hard.
(4) John worked hard, because he passed his exams.
In (3), the causal relation holds between the facts that John passed his
exams and that he worked hard (i.e. in the "content domain"). In (4), on
the other hand, it holds in the conceptualizer's mind, i.e. the proposition
that John worked hard is epistemically caused by an argument based on
the knowledge that he passed his exams. Postulating a systematic
conceptual distinction between content and epistemic domains for the
application of the meaning of modal verbs as well as connectives
provides a unified account for such phenomena, and thus has great
explanatory power.
At first sight, Sweetser's approach is further corroborated by the fact
that it also applies to connectives with some negative aspect of meaning,
i.e. concessives. However, precisely this application also produces a
conceptual dilemma, as we shall see.
On the one hand, cases like (5) and (6) seem to be clear negative
parallels to (3) and (4):
(5) John did not pass his exams although he worked hard.
(6) John did not work hard, although he passed his exams.
Intuitively, one could say that in (5) the real-world causal connection
between passing exams and working hard is in some sense negated, so
that this would be a case of concession in the content domain.
Analogously, what is negated in (6) seems to be the validity (in this
case) of the inference from the knowledge of John's passing his exams
364 Arie Verhagen
to the conclusion of his working hard, so that this would be a case of

concession in the epistemic domain (cf the examples and paraphrases of
'adversative' conjunctions in Sweetser 1990:79). Thus what we have
here is another interesting connection between causality and
concessivity: if the latter is the negative counterpart of the former, then,
given the theory of domains, we should expect it to be applicable in a
parallel fashion in the distinct domains, and this is precisely what
appears to be the case.
On the other hand, Sweetser explicitly draws attention to the fact that
for the adversative conjunction but (arguably even more general and
semantically simpler than concessive although), it is hard to find clear
examples of content usages. The point is that a contrastive relation
always allows for a construal as a relation of reasoning, involving
relations between arguments and conclusions, and is never clearly
restricted to real-world relations. Consider the following parallel to the
concessive examples above:
(7) John worked hard, but he did not pass his exams.
The contrast signalled by but cannot be sufficiently explained in terms

of the real world facts of working hard and not passing; after all, these
situations are not at all incompatible in the real world, as is clear from
the fact that they co-occur quite regularly. In order to explain the
contrast it seems necessary to invoke, among other things, some
reasoning process on the basis of general and specific knowledge
licensing the expectation that John could pass his exams, so that the
contrast can be construed as holding between this expectation and the
actual fact of John's failing.
In a way, such a move even seems natural in some general sense, as
it is actually not really clear what the content of the concept of 'contrast'
should be if we were to try to locate it completely in the world, i. e.
without reference to some person perceiving or conceptualizing a
contrast (Sweetser 1990: 103-104). However, such a move does create a
dilemma. If we call the use of but inherently epistemic (acknowledging
that the concept 'contrast' requires reference to reasoning), then we
would seem obliged to call although inherently epistemic as well,
because this also involves contrast: (5) is as contrastive as (7), and it
does in fact seem to require reference to the same kind of (unfulfilled)
expectations. But calling all uses of although epistemic would deprive
us of a way to account for the difference between (5) and (6), as well as
for the parallel between this pair on the one hand, and (3) and (4) on the
other.
My purpose in the remainder of this paper is to argue that the effects
previously analyzed in terms of domains can and should be derived
from the construction of mental spaces, with (especially) negative
semantic elements taken as instructions for setting up alternative mental
spaces, and that precisely this reconstruction avoids the dilemma just
noted. More exactly, an approach using mental spaces, as dynamically
constructed domains for the interpretation of linguistic elements, allows
us to simultaneously (a) account for the parallel between (3)/(4) and
(5)/(6), (b) claim that contrastive-as opposed to causal-relations
necessarily involve reference to some reasoning process, and (c)
account for the connection between concessivity and (negated)
causality.
3. Elements of a solution
Let me begin by illustrating the idea with the relatively simple example
(5), repeated here for convenience:
(5) John did not pass his exams although he worked hard.
As is well established in the literature on concessives, the specific

nature of such a relation crucially involves the fact, as Konig (1991)
described it, that two propositions are asserted "against the background
of an assumption."l Analyses explicating this idea in terms of logical
operators have been undertaken by Pasch (1992a, 1992b, 1994:16-27)
and Konig (1991, 1994), among others. There are differences among
these analyses, but the general idea is as follows:
(8) "p although q" means:

a. Truth conditions: p & q
b. Presupposition: q -+ -'p
Taken at face value this leads to an internal contradiction (implying both

p and -'p, or both q and -'q), and is thus minimally in need of additional
mechanisms or assumptions. Pasch, acknowledging this, seems to
assume that a speaker of anything of the form (8) actually utters
something that is necessarily false. Others are clearly more hesitant to
366 Arie Verhagen
draw such a conclusion. In any case, what seems to be needed is to be

more clear about what exactly is meant by the idea of "background
assumption", "discourse presupposition", or whatever it is called (cf
Konig 1991). It is precisely at this crucial point that Mental Space
theory (Fauconnier 1994, 1998; Fauconnier & Sweetser (eds.) 1996) can
offer clarification, in a very natural way. Put informally, the way we
interpret (5) is that the speaker, working with some general rule
licensing certain inferences (in this case for example "Normally,
working hard increases your chances of passing" or "The harder you
work, the likelier it is that you will pass your exams", and their
corollaries), reports her knowledge of a particular situation that may be
judged to satisfy the criteria for application of the rule, but nevertheless
does not conform to it. In Mental Space terms: the speaker makes us set
up a 'point of view' in which a general rule, or "topos" (cf note 1),
holds of the form {P~Q} (the capital letters indicate that the
propositions involved are generalizations, not particulars). Furthermore,
in this mental space not only the general topos holds, but also some
particular proposition p, which counts as an instance of P.
Consequently, within this mental space, the topos combined with p
2
licenses a causal inference, viz. of another particular proposition q.
Represented graphically:
p: "John worked hard"

q: "He passed his exams"
Figure 1. Single Mental Space configuration
However, understanding the concessivity of John did not pass his exams
although he worked hard involves understanding that such a mental
space is to be set up as a second space projected from the speaker's
actual mental space. By uttering although p the speaker instructs the
hearer to set up two similar but distinct mental spaces in which p is true;
in one (Spacel, with the same structure as in Figure 1, but now projected
from the speaker's own mental space) the causal inference "p therefore
q" is valid, while in the speaker's mental space (Spaceo), the truth of q is
denied so that this causal inference is not valid. Represented
graphically:
Spaceo

Figure 2. Elementary Mental Space configuration for concessivity
This analysis provides an immediate account, in a natural way, for the

often noted close relationship between concessivity and causality
without the problems involved in 'mixing' them in one system of
representation. It seems to capture a crucial feature of the concept of
concession: someone acknowledges that in highly similar circumstances
a mind very similar to one's own draws a valid causal inference, while
this inference is actually not valid.
The representation in Figure 2 uses the standard apparatus of Mental
Space theory, especially connectors across spaces (the lines connecting
elements in one space to another). It includes the idea that information
may get mapped from one space to another without explicit specific
instruction (cf Fauconnier 1998). In particular, the topos {p.... Q} may
transfer from Space! (in which it is necessary as licensor of the causal
inference) to Spaceo 'for free': one would normally interpret the speaker
as still adhering to the idea that hard work ordinarily increases the
chances of passing (considering this particular case an exception to the
rule). But strictly speaking this interpretation is optional (hence the
parentheses in Spaceo in Figure 2); it may be cancelled, for instance,
when the speaker goes on to construe her observation as an argument
against the general assumption (as in: Although he worked hard, John
failed the exam; so you see that you are wrong in suggesting that
working hard leads to success). In such a case, she still evokes a mental
space in which the topos and the related causal inference hold, but then
368 Arie Verhagen
the topos is explicitly excluded from the speaker's own space. In any
case, the topos is neither necessarily entailed nor contradicted within
Spaceo, something that is very hard to capture without distinguishing
between the two spaces.
4. Excursion: mental spaces and multiple voices
I just characterized concession as envisaging another mind, similar to

one's own, drawing a causal inference that is valid in highly similar
circumstances but not in actuality. This formulation suggests a parallel
to another approach to the analysis of 'viewpoints' in a discourse, viz.
Ducrot's theory of "polyphony" (Ducrot 1984, 1996). The concepts of
mental spaces, as I have so far used the term, and Ducrot's concept of
multiple 'voices' being present in the interpretation of a single
utterance, may be seen as no more than different labels for what is
essentially the same fundamental insight. To an important extent, I do in
fact believe this to be true. However, there are some differences that one
should be aware of
First, the term "mental spaces" seems to put more emphasis on an
individual's mental capacity for entertaining different viewpoints,
whereas "multiple voices" more strongly evokes the inherent dialogic
nature of verbal communication. At this point I actually see no
incompatibility; the individual's capacity can be seen as something that
is put to use in actual linguistic communication (when the abstract
mental spaces are assigned to the minds of actual people, as it were), as
well as something that has emerged in the individual as a result of
interaction with actual other persons during childhood. In other words,
though much may still be unknown about the empirical issue of the
precise relationship between individual and social cognition and the
ways they develop, there does not seem to be an a priori conceptual
opposition between mental spaces and polyphony.
Second, Ducrot distinguishes between three specific types of voices
in an utterance, especially (1) physical producers of pieces of discourse,
(2) 'locutors' (those who may be held responsible for the content of the
utterance, and who do not necessarily coincide with the physical
producers; I will designate these as 'conceptualizers' below), and (3)
'enunciators' (those whose viewpoints are represented in the discourse).
I do not know of an explicit parallel to these distinctions in mental space
terms. Something that is conceivably parallel to the distinction between
the latter two notions will tum out to be relevant for the discussion
below, but I think it is an immediate consequence of other features of
the analysis and does not require independent stipulation.
Third, Ducrot's concept of polyphony is intimately linked to his
theory of argumentativity in natural language. At this point I think there
is both a real difference between the mental space and the polyphony
approaches, and a point where Mental Space theory can profitably be
augmented, if we take 'argumentation' not so much in the sense of
verbal conflict, but more in the sense of 'trying to influence a person's
point of view', 'orienting one's addressee(s) towards some conclusion'.
I think this notion of argumentative 'orientation' (in my view more
important for linguistic analysis than the notion of argumentation in
itself) is very useful, and can help us to see how mental spaces work in
certain areas (such as scalar semantics, or certain phenomena of
language change). Still, for the purposes of the analysis of discourse as
such, the mental space terminology seems to me to involve the
minimum of necessary assumptions concerning the nature of discourse
interpretation. Analysts will agree, I assume, that any approach to
discourse interpretation has to take into account that people can manage
multiple viewpoints simultaneously, whereas the question if these are
'better' viewed as individual psychological phenomena or as dialogic,
social ones, requires more specific assumptions and evidence. This is
the reason why I prefer to use the mental space terminology here. But
let us return to the main theme now.
5. Solving the paradox
5.1. Epistemic causality
Setting up two related but distinct mental spaces in the construction of a

concessive relationship is crucial in avoiding the attribution of
contradictory beliefs to the conceptualizer of a concessive utterance, as
seems unavoidable in an analysis that tries to analyze concessivity only
in terms of manipulations of propositions (cf the beginning of section
2).3 Another interesting feature of the mental space approach is that the
same apparatus holds some promises for the analysis of (concessive)
conditionals and counterfactuals (cf If John had worked hard, he would
have passed or Even if John had worked hard, he would not have
passed) and thus for an account of the similarities (as well as the
370 Arie Verhagen
differences) between these kinds of relations and concessives. 4 But in

the present context it is especially relevant that the idea of levels of
interpretation can be reconstructed, at least partly, in terms of mental
spaces: whenever the interpretation of a discourse relation (or a
connective) between two segments involves the construction of two
similar mental spaces such that certain inferences are valid in one but
not in the other, then this interpretation is necessarily epistemic. What
makes an example like (5) look like a case of argumentation, hence of
an epistemic relation, is the construction of more than one mental space
in which different inferences are valid. But what makes it look like
'concessivity in the content domain' is the fact that the background
causal inference in Space 1 relates to the world. The qualitative
difference between causal and concessive relations is that the former do
not necessarily lead to the construction of multiple spaces while the
latter do. It is the case of a causal relation within a single mental space
that provides the instances of pure 'content causality'. It is the parallel
case with two mental spaces representing distinct epistemic stances
towards the same causal relations that provides instances of apparent
'content concessivity' .
At the same time, the presence of two mental spaces makes even
such content concessives into cases of epistemic relations, and thus
parallel to epistemic causality. The point is that the actual use of an
epistemic causal relation, by Gricean principles, also involves the
construction of multiple spaces, in the following way. Recall that the
epistemic interpretation of a causal relation or connective as in (4),
repeated below, amounts to construing the second segment as an
argument for the conclusion in the first:
(4) John worked hard, because he passed his exams.
Content causality, e.g. stating John passed his exams and then going on
to say how this came about (. .. because he worked hard) allows for the
first clause to be taken as representing established knowledge, and the
second as the only news. But when the second segment is an argument
as in (4), then its being uttered would be beside the point, violating the
maxim of relevance, if the first clause was already taken to represent
undisputed knowledge at its being uttered: one does not provide support
for information that one knows will remain unchallenged.
Consequently, if a speaker provides an argument in the second clause
(i.e. produces a relation of epistemic causality), she projects a mental
space with a non-positive epistemic stance (unlike her own) towards the
proposition expressed in the first clause:
Space! Spaceo

Figure 3. Elementary Mental Space configuration for epistemic causality5
By the same reasoning, the content reading of the causal relation in (3)
does not involve the construction of another mental space representing
another epistemic stance:
(3) John passed his exams because he worked hard.
As has been observed frequently, the content reading provides only one
statement, e.g. an answer to a single question of the type "Why?" or
"How?", and thus in a sense presupposes the proposition that John
passed his exams; this is what I called the "undisputed information"
interpretation of the first clause above. Thus in a content reading of p
because q the speaker precisely does not project a mental space in
which the validity of p is not (completely) certain. 6 We can now
formulate what it is that makes epistemic causality parallel to contrast
(including, in particular, concessivity): the construction of two mental
spaces with distinct epistemic stances towards a proposition.
5.2. Epistemic concessivity
We have now provided an account, by using the mental space approach,

of the parallel between 'simple' concessivity and both content causality
(the inference in one mental space relates to real-world causality), and
epistemic causality (the interpretation involves two mental spaces with
372 Arie Verhagen
distinct epistemic stances). The remaining question is how the apparent

'epistemic concessive' (6) fits in:
(6) John did not work hard, although he passed his exams.
What appears to be going on here is recursive mental space building. An

abductive causal reasoning process as in Figure 3 is embedded in the
space representing the projected epistemic stance in Figure 2 (Spacel).
Since such a reasoning process itself involves two mental spaces, what
results is the constellation of mental spaces depicted in Figure 4: 7
Space. Spaceo
although

Figure 4. Mental Space configuration for 'epistemic concessivity'
That is, in uttering (6): John did not work hard ["not p" in Spaceo],
although he passed his exams ["q" in Spaceo], in particular through the
use of although, the conceptualizer projects another mental space
(Spacel), containing the abductive inference that John must have been
working hard ["p" in Spacel] because he passed his exams ["q" in
Spacel], and then invalidates this inference by denying its conclusion.
This analysis is corroborated by the fact that (6) allows for the following
explication: "Actually, the inference that John must have been working
hard considering that he passed his exams, is invalid: it is both true that
he passed [=second clause], and that he did not work hard [=first
clause]". The applicability of such a paraphrase is predicted on the basis
of the present analysis: since the mental space projected immediately
from the conceptualizer's space contains an instance of epistemic
causality, this space projects yet another mental space as its background,
given our analysis of epistemic causality in the previous section.
The parallel between epistemic causality and epistemic concessivity

is that both involve a mental space containing argumentation, which
therefore projects yet another mental space; the difference is that in the
latter case this constellation is the background of the conceptualizer's
space rather than that space itself The constellation of the two leftmost
spaces in Figure 4 is identical to the one for epistemic causality depicted
in Figure 3, the two rightmost spaces in Figure 4 correspond to the
elementary constellation of concessivity in Figure 2. The parallel
between concessives in general and epistemic causality is that both
kinds of relationships involve more than one space with a difference in
epistemic stance towards a proposition.
It is the possibility of embedding spaces that allows us to
simultaneously account for differences and commonalities between the
pairs of sentences that we started out with. The mental space approach
thus provides a way to account for multiple levels of interpretation
without multiplying levels in the theory: the same mechanism of
projecting a mental space from another one may sometimes be applied
recursively.
5.3. Concessivity and negated causality
I now want to show how the mental space approach also provides an
analysis of the connection between concessivity and negated causality,
i. e. the (partial) overlap between the wide-scope interpretation of (1)
and the narrow-scope interpretation of (2), repeated here for
convenIence:
(1) The house is no less comfortable because it dispenses with air-

conditioning.
(2) The house is no less comfortable, although it dispenses with air-

conditioning.
What the relevant interpretations have in common is that an alternative

space is set up in which a causal inference "the house is less
comfortable" (q) is licensed on the basis of a topos of the type "Air-
conditioning normally makes a house more comfortable", and that this
inference is invalidated in the conceptualizer's space, while it is true in
374 Arie Verhagen
both spaces that "the house dispenses with air-conditioning". The

mental space representation of this interpretation is given in Figure 5:
Spaceo
p: "The house dispenses with air-conditioning"

q: "It is less conformtable"
Figure 5. Mental Space representation of (1)/(2)
It is now evident that the concessive interpretation of (1) entails that the
element because marks a causal relationship in another mental space
than the conceptualizer's. The relations between the conceptual
structure of Figure 5 and linguistic material in (1) are as indicated by
means of italicization in Figure 6:
Spaceo

q: "It is less conformtable"
Figure 6. Mental Space representation of (1)
In the concessive interpretation, the causal conjunction must mark a

relation in another mental space than the conceptualizer's; the latter is
the one containing the denial of q, and the associated linguistic element
not. This explains why it is the so-called wide-scope interpretation of
negation that corresponds to the concessivity: in this interpretation, the
because-relation cannot hold in the same space as the negation, thus it is

itself negated. Note, furthermore, that Figure 6 actually gives a more
accurate picture of the interpretation than a one-dimensional formula for
wide scope negation of the type "not (q because p)", as the latter does
not explicitly indicate that "not q" is actually asserted in the
conceptualizer's space.
Note that the interpretation of the element because does not have to
be assigned to another mental space when it occurs in the context of
negation. But when it is interpreted as belonging to the same mental
space as not, then the same conceptualizer takes responsibility for the
negation and for the causal inference, and a so-called narrow scope
interpretation is the result.
By itself the causal connective because does not indicate the
relevance of multiple viewpoints, but in a context of two mental spaces
it may apparently be used to mark causality in either one of these, as
long as the overall interpretation of the utterance is not self-
contradictory. Another example of a causal connective demonstrating
this possibility is the following, from a Dutch newspaper text:
(9) De nieuwe tariefstructuur is marktconform, maar daarom niet per

definitie klantvriendelijk.
'The new price structure is in accordance with market rules, but
therefore not by definition customer-friendly.'
The causal connective daarom ('therefore') evokes an inference from

being in accordance with market rules to being customer-friendly, and
since the latter is denied by the conceptualizer while the former is
acknowledged, the causal inference must be assigned to another mental
space than the conceptualizer's.
The conceptual structure represented in Figure 5 may also be marked
in other ways. As we have argued, a concessive connective like
although is itself an indication for projecting a mental space containing
a causal inference that is invalidated in the conceptualizer's space. That
is, the conceptual structure that is evoked by marking causality in a
projected space-as in (1) and (9)-may also be evoked by a single
element that conventionally indicates this structure, i. e. a concessive
connective. As Figure 7 shows, the mental space configuration evoked
by although in (2) shares its structure with the constellation in Figure 6:
the same conceptual relationships (depicted in a general way in Figure
5) are realized in linguistically different ways.
376 Arie Verhagen
Space! Spaceo

q: "It is less comfortable"
Figure 7. Mental Space representation of (2)
Thus, the connections between causality and concesslVlty that we

started this discussion with appear to fallout from the mechanisms of
Mental Space theory automatically.
What is more, it also follows from this approach that the relationship
is not one of complete duality, for a reason that may now appear very
simple. What would it amount to, in this view, to try to negate a
concessive relation? This would have to involve the projection of a
mental space containing both mental spaces involved in the concessive
relation, and their connections-not simply another mental space
projected from the space functioning as the background to the
conceptualizer's one. In section 5.1, we claimed that the qualitative
difference between causal and concessive relations is that the former do
not necessarily lead to the construction of multiple spaces while the
latter do. And there is no way to get a single mental space constellation
as in Figure 1 from the multiplication of mental spaces. On the contrary,
that way one can only get more and more complicated constellations;
recall that a sentence such as (10) is virtually uninterpretable:
(10) ?11t is not the case that John failed his exams although he worked
hard.
Interestingly, examples that have the same grammatical structure as (10)

and that appear to fit the above characterization (a mental space that
contains both other spaces and their interconnections) are not at all hard
to find:
(11) Beatie finds that although she has lost a lover she has gained her
freedom.
In fact, even a negation in the matrix clause of such constructions seems

to be possible in principle:
(12) They could not believe that although John had worked hard, he
failed his exams.
What seems to make this example so much better than (10) is the
factivity of the matrix predicate in (12); when I say X could not believe
that p, I am still committed to the truth of p myself, unlike when I say It
is not the case that p. Thus, in both (11), a non-negative sentence, and
(12), negative but factive, the embedded propositions are not
contradicted in the primary conceptualizer's mental space: in Spaceo,
there simply is no denial of "p although q". The problem with (10) on
this view is that it contains two instructions (not the case and although)
to set up mental spaces as alternatives to the conceptualizer's, and that
these cannot be properly related to each other.
It is interesting to note that with a non-factive matrix predicate the
interpretation of the entire sentence becomes unclear:
(13) ?They did not conclude/think that although John had worked
hard, he failed his exams.
When I say X did not think that p, my utterance may well be taken as an
indication that I am at least uncertain about p myself, unlike the
situation with factive predicates. Consequently, if the complement
clause contains a concessive one, this would be in the scope of the
'uncertainty operator' and the same problem of distinct spaces with a
non-positive epistemic stance would arise as in (10). Whatever the
details of the analysis of these cases, they help to show clearly that the
problem with the interpretation of a negation having a concessive clause
in its scope is a problem of manipulating points of view: when there are
several of them, it may be hard or even impossible to relate them to each
other in a consistent manner.
378 Arie Verhagen
6. Conclusion
In retrospect, the crucial problem in previous approaches to the

connection between causality and concessivity seems to lie in the
implicit single-viewpoint conception of negation, as no more than the
reversal of truth value. This creates the illusion that two negations
cancel each other out, whereas in natural language and cognition
negation (also as a part of concessivity) involves the construction and
manipulation of different viewpoints (cf, again, Ducrot 1996, and
earlier work), or in terms of the present framework: mental spaces. This
framework, motivated independently of the issue that concerned us here,
turns out to offer a prospect for a truly explanatory account. Since only
a very limited range of phenomena has been taken into account in this
paper, an important task is, of course, to test the approach against a
larger body of data. Yet I believe that it has proven worthwhile to spend
some space and energy on a mainly conceptual issue: what we
understand precisely when we call a phenomenon a case of 'concession'
or 'negated causality' crucially involves the manipulation of multiple
distinct though interconnected viewpoints; therefore it differs
qualitatively and irreversibly from the conceptualization of causality per
se, which need not demand the construction of more than one mental
space.
Notes
1. There is quite a variety of labels for this concept, such as topos, (discourse)
presupposition, conventional implicature, defeasible implication. For some
analysts these are probably not the same, but the differences do not have to
concern us here. The term for this kind of "inference licensing rules" that I
will use myself later in this paper is "topos". It stems from Anscombre &
Ducrot's theory of argumentation in natural language, e.g. (1983, 1989). See
also Ducrot (1984, 1996).
2. Thus I am in full agreement here with Konig and Siemund's emendation of
Pasch's analysis (cf. their section 4). What they fail to note, however, is that it
is not only a matter of descriptive adequacy that the 'background assumption'
involves some kind of generalization, but actually one of conceptual
necessity. As observed with respect to (8) , this schema (as well as the right
hand column of Table 2 in Konig and Siemund (this volume» leads to a
straightforward contradiction, as it entails both q and -'q. In order to avoid
this, only the propositions p and q themselves should be allowed to represent
particulars ("singular statements" in terms of Popper (1972», thus allowing
for verification (cf. Popper 1972: 27, 33, and elsewhere); the background
assumption must be some general rule whose logical form does not allow for
verification. For similar reasons, the background assumption must be 'only'
general and not universal~ if it were truly universal, e.g. with q in the
presupposition taken as "Vx(j{x»", the use of a concessive, asserting p&q,
would again entail a contradiction (p and "'P). These points have been
elaborated in other contexts before (cf., for example, Lagerwerf (1998) and
some references cited there), and especially in work in the tradition of
Anscombre and Oucrot's theory of argumentation in language. As Oucrot
(1996: 144) suggests, it might in fact be much older: "When we say ['It's
warm, let's go for a walk'], we do admit that there might be exceptions but
that does not prevent the topos from being valid, which is the point this highly
famous formula attributed to Aristotle makes: 'exceptions make it possible to
uphold the rule in unforeseen cases' .".
3. Precise terminology may be important here. Notice that what is to be avoided
is the attribution of contradictory beliefs. It is, of course, no problem to
entertain contradictory propositions. In fact, Mental Spaces is an explication
of what is involved in doing just that.
4. It should perhaps be pointed out that this description is not a full semantic
analysis of a linguistic element such as although (though it does provide
essential conceptual structure that will have to enter into such an analysis).
For example, this description does not yet provide a basis for distinguishing
between although and but, whereas the functions of these words are not
identical.
5. Abductive reasoning is a sufficient but not a necessary condition for the
construction of multiple spaces: a deductive inference as in may also be
construed as an argument to take away doubt. In domain terms, and
informally: content causality may always be 'used as' epistemic causality, but
the reverse does not hold.
6. This is a point, by the way, where I disagree with a specific polyphonic
analysis. It is sometimes proposed that such a distribution of old-new-
information should be analyzed in terms of two voices, one asking the
question ("Why?" or "How?"), the other providing the answer. In my view,
such a move threatens to destroy the usefulness of the concept, since it would
actually make it quite hard again to represent the difference between content
and epistemic causality.
7. In these and the following mental space representations of concessives, I will
leave out the connectors between the topos-representations (one obligatory
and one optional) in the different spaces, in order to make the pictures
somewhat simpler to read.
References
Anscombre, Jean-Claude, Oswald Oucrot

1983 L'argumentation dans la langue. Bruxelles: Mardaga.
1989 Argumentativity and informativity. In: Michel Meyer (ed.), From
Metaphysics to Rhetoric, 71-87. Dordrecht, etc.: Kluwer Academic
Publishers.
380 Arie Verhagen
Ducrot, Oswald
1984 Le dire et Ie dit. Paris: Editions Minuit.
1996 Slovenian Lectures/Conferences Slovenes. Argumentative Semantics/
Semantique argumentative. EditortEditeur Igor Z. Zagar. Ljubljana:
ISH Institut za humanisticne studije Ljubljana.
Fauconnier, Gilles
1994 Mental Spaces. Aspects of Meaning Construction in Natural
Language. Cambridge: Cambridge University Press.
1998 Mental spaces, language modalities, and conceptual integration. In:
Michael Tomasello (ed.), The New Psychology of Language.
Cognitive and Functional Approaches to Language Structure, 251-
279. LondonIMahwah, NJ: Lawrence Erlbaum Ass..
Fauconnier, Gilles, and Eve Sweetser (eds.)
1996 Spaces, Worlds, and Grammar. Chicago/London: The University of
Chicago Press.
Konig, Ekkehard
1991 Concessive relations as the dual of causal relations. In: Dietmar
ZaefIerer (ed.), Semantic Universals and Universal Semantics, 190-
209. Dordrecht: Foris.
1994 Concessive clauses. In: R.E. Asher (ed.), The Encyclopedia of
Language and Linguistics, 679-681. Oxford: Pergamon Press.
Lagerwerf, Luuk
1998 Causal Connectives have Presuppositions. Effects on Discourse
Structure and Coherence. Den Haag: Holland Academic Graphics.
Pasch, Renate
1992a Sind kausale und konzessive Konstruktionen Duale voneinander?
[=Arbeiten des Sonderforschungsbereich 282-Theorie des Lexikons,
Nr. 31]. Dusseldorf: Heinrich-Heine-Universitiit.
1992b Kausale, konzessive und adversative Konnektive: Konnektive als
Mittel des Ausdrucks von Diskursprasuppositionen. Munstersches
Logbuch zur Linguistik 1, 33-48.
1994 Konzessivittit von wenn-Konstruktionen. Tubingen: Narr.
Popper, Karl R.
1972 The Logic ofScientific Discovery. London: Hutchinson & Co. [First
English edition 1959]
Sweetser, Eve E.
1990 From Etymology to Pragmatics. Metaphorical and Cultural Aspects
ofSemantic Structure. Cambridge: Cambridge University Press.
Concessive patterns in conversation *
Elizabeth Couper-Kuhlen and Sandra A. Thompson
This paper argues that Concession in conversation is in a very fundamental

sense dyadic, involving a three-part sequence in which a first speaker makes
some point (X) and a second speaker acknowledges, or concedes, the
validity of this point (X') but goes on to make a potentially contrasting
point (Y). This basic pattern, the Cardinal Concessive, has several
variations which involve, for instance, a different ordering of the parts or an
implied Y. Two mechanisms for projecting Yare examined, semantic
partitioning and implicative prosody. In addition, several functional and
social implications of the Cardinal Concessive and its variations are
considered. Perhaps the most interesting of these is that Concession may be
used for expressing alignment as well as disaligmnent by conversationalists.
1. Introduction
Students of discourse have long been fascinated by a recurrent set of

morpho-syntactic patterns in which a speaker or a writer appears to be
'conceding' that an addressee is, or might be, right about something that
the speaker or writer goes on to take issue with. Here is an example
from our conversational database (see section 2 below): 1
(1) Joanne and Lenore are discussing a mutual friend who has
engaged in substance abuse.
1 Joanne:but he's healthy as an OX,

2 that guy.
3 (---) (h) that guy is HEAL[THY as an o:x.
4 a Lenore: [his liver,
5 a ~ except for his liver.
6 Joanne: (-) yeah,
7 but I'm saying,
8 is like,
9 you know,
10 b ~ as much as he's ABU:SED his liver,
11 and all other (-) other things in his life,
12 he's still HEALTHY as an OX.
382 Elizabeth Couper-Kuhlen and Sandra A. Thompson
In this example, there are several points at which 'conceding' appears;

we focus here on the most explicit one. In lines 4 and 5, as indicated by
the 'a' arrows, Lenore points out that their mutual friend's liver is not
healthy. In line 10, at the 'b' arrow, Joanne acknowledges that there is
something right about what Lenore has said, namely that his liver is not
healthy, but in line 12 sticks by her claim that he is 'still healthy as an
ox'.
For this paper we are interested in four recurrent features of patterns
like the one in this example. First, much as the everyday use of the word
'concede' suggests, conceding is something that people do in talking
with each other. Interactionally, acknowledging Lenore's claim allows
Joanne to disagree in a mitigated form by first (weakly) agreeing with it.
As noted by Pomerantz (1984), conceding is one way speakers have for
carrying out a potentially disruptive disagreement. As this example
demonstrates, conceding involves a 3-part interactional sequence:
1st move A: States something or makes some point

2nd move B: Acknowledges the validity of this statement or
point (the conceding move)
3rd move B: Goes on to claim the validity of a potentially
contrasting statement or point
Secolld, as we are defining it here, conceding is in a very fundamental

sense dyadic: in its basic form it requires (at least) two parties, one who
concedes and one to whom (or for whom) the concession is made. 2
Third, the speaker doing the conceding is doing so in the context of
going on to make his/her own potentially contrasting point. Thus,
'conceding' can be understood as one way in which speakers can deal
with expressing 'contrast' (as discussed in Ford, this volume).
Fourth, the nature of the potential contrast in the third move is open,
since people's inferential capabilities are open. That is, the exact way in
which X and Y are understood by the participants as contrasting is not
definable in advance, but is interpreted and negotiated by the
participants in the situated context. In some contexts, for example, Y is
taken by the participants to directly contrast with X, while in others Y
may contrast with an inference from X, rather than X itself
How have students of language recognized conceding? The notion of
conceding has been of interest to linguists primarily in two arenas. First,
Concessive has been higWighted as a type of clause combination.
Within this tradition, scholars have investigated concessive clauses in
relation to their main clauses, and particularly with respect to concessive

connective morphemes (especially Konig 1985a, 1985b, 1986, 1988,
Altenberg 1986, Sweetser 1990, Karantzola 1995, Giinthner 1999, this
volume). A second area in which Concession has been studied is as a
text relation (see, e.g., Longacre 1983, Martin 1992, Mann and
Thompson 1987a, 1987c, 1988, 1992, Thompson 1987, Matthiessen
1995, Rudolph 1996, Azar 1997, and Reinhardt 1997).
Conceding has also been a concern of rhetoricians.
Perelman/Olbrechts-Tyteca (1969) cites the 'classical masters of
rhetoric', in particular Quintilian (1921-1933), as urging 'it is a good
thing never to refute [one's opponent's] arguments in such a way that he
seems a poor advocate' (454). One way to do this is concession, which
expresses the fact that one gives a favorable receipt to one's opponent's real
or presumed arguments. By restricting his claims, by giving up certain
theses or arguments, a speaker can strengthen his position and make it
easier to defend, while at the same time he exhibits his sense of fair play
and his objectivity (488).
We will see that the way English conversationalists use concession does
in fact exhibit these very characteristics. More recently,
Antaki/Wetherell (1999) identify what they call 'show concession', a
pattern not identical but related to the ones described here, in which
speakers address some vulnerability in their starting claim and, by
making a show of conceding it, achieve the rhetorical effect of
fortifying their own position.
Building on these studies, we became fascinated with the ways in
which mundane conceding is accomplished by conversationalists. We
incorporate the traditional view of concession as involving a conflict of
expectations with our idea that the third move in a conceding format
involves a potentially contrasting point of view. We incorporate the
rhetorical perspective in our view of conceding as being situated (often)
in a larger context of argumentation.
In this paper we would like to explore how and why a speaker
concedes another's point in everyday interactions. The questions we
address include:
a. How does Concession structure conversational exchanges?

b. When are speakers likely to use it and what interactional work
does it do?
c. What grammatical formats do speakers use to express
Concession?
Examination of conversational data suggests answers to these questions.

The data show that conceding is indeed something that speakers do
frequently. Certain actjvities (e.g. assessments, blamings, tellings) are
especially rich environments in which conceding can do important
interactional work.
2. Database
The data for this study were taken from a variety of British and
American sources, including conversations from the Corpus of Spoken
American English (DC Santa Barbara)3 as well as our own collection of
conversations and radio call-in show data. We identify the source of
each of our examples below.
In carrying out this piece of research, we studied our transcripts for
all cases in which we felt, as native speakers, that conceding was being
done. For us, as suggested above, 'conceding' involves, first,
acknowledging another's point (though not necessarily 'giving in' or
'yielding' 4), and second, going on to make one's own «potentially)
contrasting) point. We also included in our collection instances in which
one or the other of these parts was missing but strongly projected. We
collected a database of approximately 200 examples and examined them
for recurrent interactional and grammatical features. In this paper we
will report on the most robust patterns.
3. The Cardinal Concessive
As a first step towards understanding and describing the conceding data

we have found, we propose defining what we will call a 'Cardinal
Concessive' pattern. On analogy to a Cardinal vowel, a Cardinal
Concessive can be defined in abstract terms and used as a reference
schema for actual instantiations in conversation. The schema is
maximally explicit, in that its three parts-which we represent with X,
X' and Y-are realized verbally, and are maximally distinctive with
respect to other possible hypotactic clause combinations. The pattern
identified in example (1) just above is an example of a Cardinal
Concessive and could be represented provisionally as:
(1 ') Schematic representation of Cardinal Concessive In (1)

(provisional)
A: X That guy's liver is not healthy

B: X' As much as he's abused his liver (He's abused his
liver a lot)
Y He's still healthy as an ox
Or even more abstractly:
(1 ") Schematic representation of Cardinal Concessive
A: X
B: X'
y
(where X and Yare understood by participants as potentially
contrasting)
We use 'Concession' and 'Concessive' with a capital 'c' to refer to this

interactional schema. In this paper, we will further illustrate the
Cardinal Concessive, and then go on to discuss variations on the
Cardinal Concessive schema and their implications for interaction.
Consider now another instantiation of the Cardinal Concessive
schema in our database. This example will allow us to introduce another
important dimension of conversational conceding. Again, we have
labeled the moves we understand the speakers to be making with X, X'
andY.
(2) Charles and Steve, currently in NYC, are lamenting the advent of
Starbuck's coffeehouses, which Steve says will 'blowout' the
small privately owned places in cities like Minneapolis.
1 X Charles: but even the privately owned places here-

HERE,
2 X are so CHARACTERLESS I
3 (-) I mean by and large.

4 X' Steve: (--) OUTSIDE [of the Village they are.

5 Charles: [you know.
6 Y Steve: there's [some in the Village that are
nice.
7 Charles: [outside of the Village, yeah.
8 yeah,
9 that you certainly wouldn't wanna lose.
10 but everywhere ELSE,
11 you know they're all like,
12 CHRO:ME and BLACK [( ) CHAIRS,
13 Steve: [oh up here it's
horrible.
14 Charles: and ( ) pictures and so forth.
15 Steve: yeah, they OBVIOUS -
they're not places where they want you to
LINGER.
Charles claims in lines 1-2 that even. the privately owned coffeehouses
'here', i.e. in New York City, are so characterless. When Steve does not
immediately second this assessment, Charles downgrades it somewhat
by appending by and large (line 3). In the next tum, a conceding tum,
Steve acknowledges that Charles' assessment is partially true: he agrees
that privately owned places outside Greenwich Village are indeed
characterless (line 4). But he goes on to maintain that in another sense
Charles' claim is wrong: there are privately owned places in the Village
which are nice (line 6).
A prominent feature, then, of this example is the partial nature of the
conceding move, labeled X', at line 4. In other words, in his conceding
move, Steve sets up a contrast between two aspects of a single
proposition, which we will call 'partitioning'. Thus he acknowledges
that Charles is right about places being characterless, but only about the
places outside of the Village. But there's some in the Village that are
nice. In fact, one of our most striking findings was that speakers very
often accomplish conceding by acknowledging only part of what the
other speaker has said. In other words, one of the recurrent ways in
which a speaker concedes another's point is to acknowledge that part of
it is, or may be, true.
Again, we can characterize this interaction in terms of the Cardinal
Concessive schema as follows:
(2') Schematic representation of Cardinal Concessive in (2)

(provisional)
A: X The privately owned places here are so

characterless
B: X' Outside the Village they are
Y There's some in the Village that are nice
(where X and Y are understood by participants as potentially
contrasting)
Not only is the Cardinal Concessive a practical analytic device for

describing actual concessive patterns in discourse, but in fact our data
suggest that it is also a schema which interactants-whether consciously
or unconsciously-orient to in the production of talk (see also Ono and
Thompson 1995). Evidence for this can be found, for instance, in
collaborative completions (Lerner 1991, 1996, Ono and Thompson
1996). In the following extract, for example, Bee produces the X of the
Cardinal Concessive schema, and Ava concedes that Bee is partially
right (X'). Bee then fInishes the sequence by producing Ava's Y, as it
were:
(3) Bee and Ava (on the phone) have been joking about each other's
knowledge of where the other is.
1 Bee: I FINALLY said something right.

2 X ( 0 . 2) YOU ARE HOME. ( hhhh)
3 X' Ava: yeah- I BELIEVE so.
4 X' PHYSICALLY anyway.
5 Bee: yea:h.
6 Y not MENTALLY (h) though, (hh)
In this excerpt Bee produces the first part of a Concessive schema in

line 2. Ava then jokingly acknowledges that she is home, but only
partially: she's home physically. That is, she does this conceding by
partitioning Bee's X, just as we saw Steve doing with his in the Village
in example (2). In doing this partial acknowledgment, Ava so strongly
sets up a contrast between 'physically' and 'not physically' that Bee can
collaboratively provide the Y of the Cardinal Concessive schema, in this
case that she is not home mentally. We would argue that it is the shared
knowledge of, and familiarity with, the Cardinal Concessive schema
which enables Bee and Ava to collaborate in the joint production of this
pattern.
In this section, we have illustrated the Cardinal Concessive pattern

and one common way it is produced (with partitioning), as well as
provided evidence that it is a schema to which speakers can be shown to
systematically orient in ordinary conversation. In the next section, we
suggest two other benefits to approaching our concessive data in terms
of a cardinal schema.
4. The Cardinal Concessive as an analytic tool
Aside from its interactional reality, the Cardinal Concessive schema has
two further advantages as an analytic tool: (i) it provides for maximal
flexibility with respect to the lexico-syntactic representation of its parts,
and (ii) it underspecifies the linkage between X' and Y. We expand on
each of these features below.
4.1. Lexico-syntactic forms ofthe constituent parts
The Cardinal Concessive schema allows for its constituent parts X, X'
and Y to be realized by linguistic units of varying size: words, phrases,
clauses or whole chunks of discourse-all possibilities which are in fact
attested in our database. In examples (1)-(3), all three parts of the
Cardinal Concessive schema were phrases or clauses. But in the
following case, X' is realized by a single word:
(4) Marie has called in to this Berkeley radio phone-in program

during the 1991 Gulf War to talk about an anti-war demonstration
which took place the day before and in which she took part. The
demonstration erupted into violence.
1 Marie: I DO- I DO wanna say

2 that I really think a lot of people were
SINCERELY there,
3 just to make the opinion KNOWN that,
4 (0.7) it was STILL time to negotiate,
5 and we didn't need to start the BOMBING.
6 x and (0.6) i hate to SEE: uhm,
7 X (0.3) EVERYBODY get a bad NAME for that,
8 x and AS you say,
9 x tha t ' s wha t makes i t AMERI CA .
10 ->X'Leo: yeah.
11 Y but it IS a shame.
12 Y and it uh UNFORTUNATELY: uh::,

13 Y (.) the fact that there are more people
like YOU marie,
14 Y than the REST,
15 Y (0.6) than the BA:D apples uh,
16 Y (.) kind of gets LOST in all of this.
In this excerpt Marie expresses the point that at least some of the
participants in the demonstration simply wanted to make their opinion
known that the bombing had started too early, before efforts at
negotiation had been exhausted, and that they should not be disparaged
because of this. In line 8 she appends a reference to an earlier remark by
the moderator that freedom of expression, as long as it is peaceful, is a
fundamental American right. Leo, the moderator, initially agrees with
this (yeah) (line 10), but then proceeds to evaluate what happened as a
shame (line 11), an assessment which is potentially incompatible with
the positive light in which the basic right to freedom of expression has
been presented. Yet his use of the Cardinal Concessive schema allows
him to come across as claiming that both points hold.
In this example the X' move has minimal size, being realized by the
single word yeah. 5 However it can be more substantial, as we saw in (1),
where Joanne concedes by saying as much as he's abused his liver. In
the following case the conceding move is considerably more substantial:
(5) On this call to a British radio phone-in program, Mrs. Etchalls and
the moderator Dick Hatch are comparing the advantages and
disadvantages of a penal system in which people who are
convicted of felonies wear electronic bleepers rather than go to
prIson.
1 X Etchalls: everybody KNOWS,

2 X whoever you SEE don't they,
3 X they know you've done SOMETHING,
4 X whereas ( 0 . 7) YOU know,
5 X once you're- you're going (all) away,
6 X a lot of people could just think you've
X gone on a HOLIDAY or something li:ke
tha:t.
7 X' Hatch: (0.4) I suppose they COULD,
8 X' YES I suppose it would mean that
X' you'd have to spend quite a lot of time
INDOORS,
9 X' or else you'd have to wear a large
MUFFLER all the time.
10 X' or pretend you had a stiff NECK,

11 X' and wear one of those big PINK things
they give you.
12 Etchalls: (0.2) yeah,
13 X' Hatch: or something like THAT.
14 y (0.6) but I must say I THINK -
15 y it's the one sort of opinion I HAVE on
the matter,
16 y I think I would rather THAT than go to
prison,
17 because (0.2) PRISON must be VERY soul
destroying. don't you think,
In the first tum of this extract, Mrs. Etchalls is weighing in in favor of

prison: electronic bleepers are visible signs of one's conviction;
whoever you see ... they know you've done something (lines 2-3),
whereas, she maintains, with prison a lot of people could just think
you've gone on a holiday or something like that (lines 5-6). This
antithesis implies the claim that prison is better, an X whose formulation
extends over six intonation phrases. Dick Hatch's next tum is now
occupied with conceding both of Mrs. Etchalls' points: line 7, I suppose
they could acknowledges that with prison people could think one has
gone on holiday, and lines 8-11 acknowledge that bleepers would be
visible and make special countermeasures necessary in order to resolve
the problem. This X' is five intonation phrases long. Mrs. Etchalls
acknowledges his concession in line 12, whereupon Dick Hatch
proceeds to carry out a countering move (Y): but...! think! would rather
that than go to prison (lines 14-16). As indicated by the Xs, X's, and Ys
in the left margin, this example shows that the constituent parts of the
Cardinal Concessive schema can be quite lengthy.
The lexico-syntactic shape of the Concessive pattern is thus reducible
or expandable depending on local contingencies in the interaction.
Using Xs, X's and Ys in the Cardinal Concessive template allows us to
capture the wide divergence of lexico-syntactic forms we find speakers
using in the construction of interactive Concessive patterns.
4.2. Links between X' and Y
A second benefit of the Cardinal Concessive schema is that it does not

specify how the two parts, X' and Y, of speaker B's tum are joined to
one another. This allows for the variety of links which we find attested
in our corpus. For instance, the relation between X' and Y may not be
marked by any morphological markers at all, as was the case in example
(2) above. Here Steve merely juxtaposes the two claims outside of the
Village they are (so characterless) and there's some in the Village that
are nice. Yet X' and Y are linked by a parallelism between the two
prepositional phrases and contrastive stress on in in the second one.
As was illustrated in (4), speakers may link Y to X' with the
coordinating conjunction but. And, as the following example
demonstrates, speakers may also have recourse to subordinating
conjunctions in constructing Concessive patterns:
(6) Mr. Butler has called in to this Radio Manchester phone-in

program to take exception to the point which an earlier caller,
Mrs. Sack, made, namely that contaminated meats should not be
used in the production of pet food. He refers here to dogs.
1 Butler: and I just WONDER,

2 (0.2) WHY we SHOULD worry TOO much
about what we put in their FOOD,
3 (0.6) er if- if- when you-
4 as soon as you let them off the LEAD,
5 or as soon as we SEE people LET them
off a LEAD,
6 they immediately go SNIFFING ROUND,
7 (0.5) picking up all and sundry from the
GROUND,
8 X and I just wonder what's the point in
worrying
X about what goes in the TIN.
9 (1.2)
10 X'Hatch: well YES,
11->Y except that I mean you- you HAVE to
RECOGNISE,
12 Y as er (.) Mrs Sack said,
13 Y that there ARE a GREAT many PEOPLE,
14 Y who- whose LIVES are:
15 Y BUILT ROUND (.) to some extent,
16 Y (0.4) their- their PETS,
Mr. Butler's point is expressed in a form related to that of a rhetorical

question: I just wonder what's the point in worrying about what goes in
the tin (lines 8-9), which in essence claims 'there's no point in
worrying... '. Following a noticeable pause of 1.2 seconds, Dick Hatch
initially agrees with what his caller has just said: well yes (line 11). But
he goes on to formulate a sense in which he does not agree with this
point: you have to recognize ... that there are a great many people ..
whose lives are built round ... their pets (lines 13-16). The conclusion to
be drawn from this remark is presumably that owners are justified in
being concerned about the well-being of their pets. Notice that Dick
Hatch's counter-point Y is linked to his prior X' by the subordinator
except that (line 12).
Thus a variety of means of explicitly linking X' and Y-as well as
the absence of any explicit link-are attested in our corpus. 6 The
Cardinal Concessive schema does justice to this fact by underspecifying
how its constitutive parts are conjoined.
In this section we have tried to show that approaching our data in
terms of a Cardinal Concessive schema has several advantages,
allowing us to show what all of our examples have in common in terms
of what the interactants are doing and how they understand what they
are doing. We have seen that acknowledging a previous speaker's point
to have been (partially) right and then going on to make one's own
contrasting point is a recurrent format with which speakers counter each
others' claims. In the next section, we will examine the two most
frequent variations on the Cardinal Concessive schema.
5. Variations on the Cardinal Concessive
The Cardinal Concessive is characterized· by (i) the explicit presence of

all three steps or moves (X, X', Y) and (ii) their presence in the order X,
X', Y. Yet our data suggest that we want to also allow for variations on
the schema involving either the absence of one of the parts or a change
in the order of the parts, or both. We suggest that what the speakers are
doing in the excerpts that follow is best understood in terms of the
Cardinal Concessive pattern. We will also see that each of these
variations has interactional consequences.
5.1. Concessive variation with [X Y X'j order
Rather than first conceding the validity of an interlocutor's point and

then countering it, speaker B on occasion comes in immediately in the
next tum with a potentially incompatible point or claim and only
subsequently acknowledges that speaker A's prior point was also
(partially) valid. For example:
(7) Prior to this excerpt Charles has talked about a French friend
Didier, who he describes as 'a musician and kind of cabaret
artist' . The conversation then shifts to an occasion on which
Charles heard Didier play some of his own songs and to another
occasion on which Karen, a present co-participant, heard Didier's
manager Ray sing. The topic of talk now reverts to Didier.
1 X Steve: w'll i- dz-- this guy's a CABARET artist,

2 I mean -
3 does [he PERFORM in
4 y Charles: [well I- I don't think he does that
PROFESS I ONAL-
5 X' I mean he does -
6 x' he: -he's had like a COUPLE of
en [gagements,
7 Karen: [oh Ray's his MANAGER.
8 Charles: (--) HAH
9 Karen: I forgot,
10 right.
11 Charles: o:h MAN.
12 Karen: and they- he just had a gig at (--)
STARBUCKS
13 (-) on the EAST side.
In line 1 of this excerpt, Steve begins to ask Charles more about

Didier's professional status. With his i- dz-, he begins and cuts off two
questions; these might well have been questions starting with is, and
then does. What Steve says next picks up from an earlier comment by
Charles that Didier is a cabaret artist. Steve reasserts this earlier claim of
Charles' as background to his question in line 3 does he perform in ....
In line 4, Charles intervenes immediately with his counter-claim Y to
correct the misinterpretation: I don't think he does that professional-
(line 4). This remark by Charles suggests that Steve's assumption was
wrong if 'cabaret artist' is understood to refer to a professional
occupation. So far then we have a pattern which looks rather like a
simple contrast. Yet notice that Charles aborts his tum during the word
professionally, which is produced with a sound cut-off; moreover, he
proceeds to weaken his prior counter-statement by stating that Didier
does (line 5). Although this tum is also aborted, the contrastive stress on
does nevertheless indicates that there may be a sense in which the label
of 'cabaret artist' is justified. Charles then proceeds to add that Didier
has had a couple of engagements (line 7), which suggests that he is at
least semi-professional. Accordingly, we have labeled lines 5-7 as X'.
Thus the pattern we see here is first a denial or contradiction (Y) of a

prior statement and then an acknowledgement that this statement is in
one way or another (partially) right after all (X').
Notice now that one of the consequences of the reverse order of parts
in B's tum, namely Y X', is that the prominence relations between X'
and Yare also potentially reversed. In the Cardinal Concessive it is
typically the second move (Y) by speaker B which is the thrust of the
tum and has implications for what happens next (see also Pomerantz
1984:75). This sequential implicativity (ScheglofflSacks 1973) is rather
naturally encouraged by the end position of Y. Yet when the X' Y order
of the Cardinal Concessive is reversed, it is X' which is in end position
and thus potentially in a position to receive more attention in subsequent
talk. In fact, this is precisely what happens in (7). Karen's next turns
point to two further ways in which Didier is professional: he has a
manager (Ray's his manager, line 8) and he recently performed in
public (they-ejust had a gig at (1.5) Starbuck's, lines 12-13). Therefore,
instead of Charles' counter-move Y (= Didier is not professional), it is
rather his conceding move X' (= there are ways in which Didier is
professional) which is implicative for following talk.
At times in our data, however, speakers explicitly frame their talk in
such a way as to avoid having their conceding move be the prominent
part of their tum. There are at least two strategies that we have observed
speakers using to compensate for the natural prominence of X' in end
position. One of these is for speaker B to reiterate y after the production
of X'. Consider (8):
(8) Two friends Ann and Barb are talking here about a mutual friend
of theirs who has become seriously ill and given up on life. Barb
has just asked rhetorically "What about the people who care about
her and the people who are around?"
1 X Ann: she DOESN'T HAVE many people around her

anymore.
2 y Barb: but she DOES.
3 y she DOES still have a lot of people who
CARE,
4 X' (0.3) she doesn't have anybody she'll let·
get CLOSE to her,
5 -> Y but she's got a LOT of people who care.
6 (.hh) and a LO:T of people (.h)
7 you know the I:RMGARD Trassbachs,
8 (0.7) an- and (.h) CHRISTI:NE,
9 and CASS:,
and people ALSO,
11 who've never been CLOSE friends of hers,
12 (0.3) but who'd BEND over backwards (0.2)
for this woman,
When Ann claims that their friend doesn't have any people around her
anymore (line 1), Barb contradicts this with a Y in her next tum: but she
does (line 2). Next Barb reasserts her counter-claim with she does still
have a lot ofpeople who care (line 3). This formulation re-specifies the
predicate in question as having people around her who care. Yet Barb
goes on to acknowledge, with an X', that in one sense Ann's point
holds: she doesn't have anybody she 'II let get close to her (line 4). This
acknowledgement, like the one in example (7), serves to weaken the
force of the prior contradictory move by suggesting that in one way the
interlocutor's statement was right. Yet in contrast to (7), the conceded
part, X', does not remain in end position: Barb now reiterates her initial
counterclaim (Y): but she's got a lot ofpeople who care (line 5). This in
effect expands the sequence by turning X' into a conceding move
which-as in the Cardinal Concessive-is followed up by a counter-
claim Y. Moreover, Barb now proceeds to construct a list of people who
care, thereby underlining the validity of her counterclaim.
What Barb does here can be thought of as first disagreeing, and then
retracting somewhat by weakly (partially) agreeing. In fact, in this
variation on the Cardinal Concessive, where speaker B's conceding
move X' follows the contrasting claim Y, we often have the vernacular
sense that speaker B is 'backing down'.
A second way in which the sequential implicativeness of an X' in
end position can be avoided is to downplay it lexically and prosodically.
This is frequently done by using the unstressed particle though to
represent X'. For instance:
(9) In this excerpt Laura has been talking about her frequent
backaches. Ruth has reported her doctor's claim that everybody
gets backaches because the backbone is 'built for an animal to
walk on all fours' .
1 Mom: ACTUALLY,
2 X that HELPS,
3 X (-) when you have a backache.
4 Laura: really?
5 X Mom: (--) is to -
6 X (-) to get down on your hands and KNEES,
7 X and walk AROU-

8 X and crawl AROUND,
9 Laura: oh,
10 Y not crou -
11 Y,X' not ALL the way DOWN though,
12 Mom: unhunh,
The X in the pattern instantiated here is provided by Mom, who states

that backaches can be helped if you get down on your hands and knees
and... crawl around (lines 2-8). Laura counters this immediately with a
Y, oh, not crou- ... not all the way down (lines 9-11). That is, she is
heard to be asserting that Mom's claim is not true-at least not wholly
true, her formulation not all the way down suggesting the possibility of a
contrast between 'all the way down' and 'part way down'. Thereafter,
Laura appends the particle though, an abbreviated way of conceding the
(partial) validity of Mom's prior statement. (In making this
interpretation we follow Pomerantz, who says that as a final particle
though does the work of "claiming to agree with the prior while
marking, and accompanying, a shift in assessed parameters which
partially contrasts with the prior" (1984:63).7 )We have indicated this
with the notation [Y X'] in line 11. We might gloss Laura's turn as
saying 'It won't help a backache if you go down all the way, though
what you say is partially true' .
We note that Laura's conceding move is not only lexico-syntactically
minimal, it is also prosodically reduced: though does not form a
separate intonation phrase but instead is 'cliticized', as it were, to the
prior intonation phrase, which in this case comprises the entirety of Y,
where it appears in post-tonic position. As a consequence, the import of
X' is scarcely noticeable; Y remains prominent despite the fact that it is
not in end position. One indication of this in example (9) is the fact that
in subsequent talk the speakers orient not to how getting down on all
fours helps a backache (X), but to how going down all the way is not
helpful (Y): e. g., Mom produces a negative agreement token in line 12.
So a favorite variation on the Cardinal Concessive schema, then, is
the X Y X' pattern, when speakers present their counterclaim first, and
then concede that all or part of the previous speaker's claim has some
validity.
5.2. Concessive variation with [X X'} only
The second type of variation we find attested in our corpus-albeit

much less frequently-is the pattern [X, X']: in other words, Y is
missing. That this should be an option is in a way surprising, since we
have argued with respect to the Cardinal Concessive schema that the
third move is the thrust of speaker B' sturn. But under certain conditions
(to be specified shortly), the Y move can be so strongly projected by a
speaker's X' that its explicit expression is made redundant. Here is an
example in which we have found speakers actually refraining from
producing a projected Y by design, appealing to a tacit understanding of
it and thereby furthering collusive alignment:
(10) Harold, Miles, Pete and Jamie are gossiping about a neighbor who
has recently produced a number of 'brat' kids.
1 X Harold: does she even have a b- a MAN?
2 X I guess she MUST.
3 Miles: (0.5) does she have a WHAT?
4 Jamie: [a MA.N.
5 Harold: [a MA.N.
6 Pete: ( .hh)
7 x' Jamie: she has SOME [kind of a-
8 x' Miles: [at least TEMPORARILY,
9 Pete: [yeah,
10 Harold: [yeah.
11 Jamie: heh heh (.h) [at ONE time
12 x' Harold: [for about FIVE minutes,
13 x' PROBABLY.
14 Miles: heh
15 Pete: heh
At the beginning of this excerpt Harold asks a question: Does she even
have a man? (line 1), which he then proceeds to answer himself: I guess
she must (line 2). This establishes the point 'She must have a man',
which we are calling X. Following a short side sequence for repair (lines
3-5), Jamie and Miles now each produce, in partial overlap, a conceding
move with respect to this point. Jamie's she has some kind ofa- (line 7)
has strong stress on some, Miles' at least temporarily (line 8) has strong
stress on temporarily.
What we see here, then, is Jamie and Miles each partitioning
Harold's claim X in a different way. In Jamie's she has some kind of a-
(line 7), the stress on some sets up a contrast between 'some kind of
man' vs. 'the real kind of man'. In Miles' at least temporarily (line 8),
the stress on temporarily sets up a contrast between 'temporarily' vs.

'permanently'. In each case, these contrastive stresses on the first part of
the contrast make it possible to project the second part. 8 And in each
case the counter-move Y (= 'but she doesn't have the real thing' or
='but she doesn't have one permanently') is missing. The projectability
of these counter-moves, however, serves as common ground for the
merriment which Jamie, Miles and Pete share (lines 11 and 14-15). The
fact that Harold too understands the implications of the concessive
pattern is nicely demonstrated when he chimes in with another
instantiation of X': For about five minutes (line 12) using identical
prosody.
The second variation on the Cardinal Concessive that we have found
well represented in our data, then, is one in which the third move,
speaker B's counter-claim, what we are calling Y, is missing. In all such
cases of concession in our data, Y can only be missing if it is strongly
projected by the first two moves, X and X'. Our analysis of example
(10) reveals one of the strategies by which this projection is
accomplished, with prosody. The other common strategy has been seen
in a number of our examples, including (10), namely what we have been
calling partitioning.
We have omitted the Y in our labels in example (10), since, as we
have said, Y is in fact missing. In the next section, we will slightly
revise this labeling, and examine these two prominent strategies by
which speakers accomplish conceding without making their counter-
claim Y explicit.
6. Projectability of Y
As we have noted, there are two mechanisms which provide for the
projectability of Y and render its actual production in interaction
negotiable, both of which are in evidence in (10). One is semantic and
the other prosodic; we will briefly examine each in this section.
6.1. Semantic partitioning
The semantic mechanism by which the counter-claim Y can be

projected involves what we have been calling 'partitioning'. Partitioning
occurs when X' sets up a binary contrast with respect to some aspect of
X, for instance with respect to size or amount (little vs. a lot), type (fake
vs. real), time (now vs. then), location (inside vs. outside), frequency
(sometimes vs. often), chance of occurrence (likely vs. unlikely), out-
ward appearance (looking or seeming vs. really being), etc. In (10), for
example, as we have shown, two partitions are in evidence, one
concerning the type of man the neighbor has, the other concerning the
period of time during which she has had a man. The concessive move
involves confirming one aspect of the partition, e.g. one type of man or
length of time, but disconfirming the other. Since the contrast estab-
lished is binary, very often evoking one part naturally calls to mind the
other. Moreover, the affirmation of one part implies the denial of the
other (and vice versa). Thus the implicational relations in (10) could be
more realistically captured if the X' moves were represented as [X' +Y-
implied]:
(10') Schematic representation of Concessive variation in (10) (revised)
X A: She must have a man

X' + V-implied B: She has some kind of a-
X A: She must have a man

X' + V-implied B: At least temporarily
In these cases, then, one type of man the neighbor has or one time
period of her having a man is affirmed; the other type and time period
are by implication disconfirmed.
Our analysis of the way partitioning allows Y to be projected but not
made explicit also gives us a means to schematically enrich our analysis
of the conceding that Steve does in example (2), repeated in short form
here:
(2') Extract from (2) above
1 X Charles: but even the privately owned places here-

HERE,
2 X are so CHARACTERLESS,
3 (-) I mean by and large.
4 X' Steve: (--) OUTSIDE [of the Village they are.
5 Charles: [you know.
6 Y Steve: there's [some in the Village that are

nice.
7 Charles: [outside of the Village, yeah.
8 yeah,
As we suggested above, Steve is able to claim the simultaneous validity

of two potentially incompatible situations, namely that privately owned
coffeehouses are characterless and that privately owned coffeehouses
are nice, by partitioning the proposition with respect to its location: he
sets up the contrast 'outside of the Village' and 'in the Village' to do
this. Due to the fronting of 'outside of the Village' as well as to its
contrastive accenting, his X' thus strongly projects a counter-claim Y
with an opposite location. In this case, however, unlike (10), the
projected Y move is subsequently made explicit. The pattern could thus
be represented as:
(2") Schematic representation of Concessive variation in (2) (revised)
X A: The privately owned places here are so

characterless
X' + Y-implied B: Outside of the Village they are
Y-explicit There's some in the Village that are nice
What we see here, then, is that partitioning often projects what the
speaker's next move, the (potentially contrasting) counter-claim Y, will
be. Sometimes Y is then offered explicitly, as we see in Steve's
comparison between outside of the Village and in the Village in example
(2). But sometimes Y is left inexplicit, often with humorous or satirical
overtones, as we saw in Jamie's she has some kind of a- and Miles' at
least temporarily in example (10) about the neighbor's relationships.
Example (10) thus illustrates the way in which this partitioning strategy,
precisely because it is so frequently employed in doing conceding, can
be used to project a counter-claim Y which then does not need to be
made explicit.
6.2. Implicational prosody
The second mechanism which is available for projecting a counter-claim

Y is prosodic in nature. Two intonational devices will have this effect:
contrastive accent placement and a fall-rise tone (or its functional
equivalent).
Contrastive accent placement is in evidence when the word or
syllable which carries the 'main' or primaI)' accent in an intonation
phrase falls on a grammatical word or some lexical item other than the
last one in an intonation phrase (Couper-Kuhlen 1986). For instance, in
example (2), as we have noted, Steve says outside of the Village they
are with strong prominence on the preposition outside. This projects a
contrast with the lexical opposite of 'outside', namely 'inside'. And in
(10) Jamie stresses some in she has some kind of a-. (If this tum had
been completed, it presumably would have been 'she has some kind of a
man'). This accent placement also evokes a binary contrast, although it
does not make the other member of the contrast explicit. Contrastive
accent placement is particularly effective when it is used in an X' to
project an upcoming Y.
A second prosodic device we find speakers using to implicate an
upcoming contrast is a fall-rise tone or its variant, the truncated fall
(Couper-Kuhlen 1986). The fall-rise is found in the absence of
contrastive accent placement but also in conjunction with it. In the
following instantiation, speaker B uses both contrastive accent
placement and the fall-rise contour to project an upcoming Y:
(11) Talk in this family conversation has centered around making up

clues to the birthday presents which Marcia's daughter Kendra is
about to open. Kevin has pointed out in prior talk that his wife
Wendy has tried her hand for the first time at making a clue and
there has been some mild teasing about what kind of clue Wendy,
as an amateur and relative newcomer to the family, might have
made. Wendy has responded with Stop it. Don't make fun of me,
cued as playful in a child-like voice.
1 Wendy:
[but don't make FUN of me. ]
2 X Marci:
[you look pretty PROUD of your]self
Wendy,
3 X'+Y-impl Wendy: WE:LL,
4 X'+Y-impl I'm a LITTLE proud, =
5 Y-hypoth =except if you [think it's STUPID, heh
6 Kevin: [ha
7 Y-hypoth Wendy: [then [I'll feel BA:D.

8 Kevin: [ha [ha ha ha
In this excerpt Wendy repeats her playful plea don't make fun of me
(line 1), while Marci suggests in overlap that Wendy actually looks
rather proud of her clue (line 2). In response, Wendy acknowledges that
Marci's assessment is partly true-she does feel a little proud-but she
goes on to claim that should the others find her clue stupid, then
something potentially incompatible will hold, namely she will feel bad.
(As indicated, Y is expressed hypothetically here.) Partitioning is
signaled in this instance not only by contrastive accent placement, with
the main accent on little rather than proud, but also by the fall-rise tone,
which begins with a pitch step-up on lit- and is spread over the rest of
the tail of the intonation phrase. The falling movement on little is quite
marked, while the movement on proud is only minimally rising.
The projecting prosodic devices discussed in this section are not used
consistently enough in our materials to make any strong claims about
,concessive intonation'. We have many cases of concessive patterns
which lack this kind of marking. Yet when contrastive accent placement
and/or fall-rise tone are found,. they often cue partitioning and are thus
hints that Concession may be involved. Furthermore, when we find an
example in which Y is missing, we always find either or both projecting
prosodic devices and partitioning.
7. Functional considerations
What we hope to have shown is that Concession in English conversation

can be understood in terms of a cardinal schema to which speakers can
be demonstrated to orient, according to which they (partially)
acknowledge another's point (X'), and then go on to make their own
potentially incompatible point (Y). In this section we would like to
touch on what we think are the most interesting functional and social
implications of our analysis of Concession.
The patterns we have identified here apply first and foremost to
dyadic interaction. 9 Rather than being random or coincidental
formations, we would argue that they have grown out of tasks which
participants fmd themselves continually confronted with in interaction.
The Cardinal Concessive, for instance, is clearly well adapted to
prefacing disagreement with partial or weak agreement in the context of
an assessing activity (Pomerantz 1984). As Pomerantz and others have
argued, prefacing a disagreeing move with an initial agreement is a way

speakers have of negotiating alignment with one another in talk. We
believe that our example (2), about the coffeehouses inside and outside
the Village, is a particularly apt demonstration of the strategic use to
which the Cardinal Concessive together with partitioning can be put. It
is perhaps not insignificant that agreement is ultimately reached in this
episode.
Related to disagreement-prefacing is the use of the Cardinal
Concessive schema in an argumentative context, as we find often
happens in our radio phone-in materials. Here the conceding move X',
especially if it is enlarged upon, as in (5) about prison vs. electronic
bleepers, has a decidedly rhetorical effect: by first building up an
opponent's case to its maximal strength, one makes one's own counter-
argument all the more devastating when delivered. 10 In an
argumentative context speakers need not be aiming for alignment; in
fact, in the call from which example (5) is taken, interlocutors never
reach consensus on whether bleepers or prison is better for felons.
But Cardinal Concessives are found not only in oppositional
environments. They can be a very effective way of saying-not 'You
are partly right but mostly wrong' but-'We're both right, but let's look
at things my way'. The following example illustrates this:
(12) Connie, Vera and her boyfriend Keith are talking about Vera's
father, who is suffering from the aftermath of a knee operation. In
prior conversation it has emerged that Vera's mother has led her
to question whether the operation was necessary in the first place.
Just prior to this excerpt Connie has pointed out that Vera's father
was in considerable pain before the operation.
1 Kei: how come (0.2) you're letting this

CRA[ZY person change your mind,
2 Ver: [right.
3 Kei: you [know.
4 Ver: [well s- it's like she made me
believe that,
5 YEAH,
6 you know the doctors ARE (.) you know
horrible,
7 and yes this is ridiculous,
8 an .hh gu- I gue[ss -
9 Con: [why do you think your -
10 X mom is probably just -
11 X Ver: JEALOUS.
12 Kei: eh h[eh,
13 X'Con: [well I mean pr - maybe that too,
14 Y but she just can't HANDLE (0.2) [people
close to her being in PAIN.
15 Ver: [reality.
Connie's contribution in line 9 appears to be questioning Vera's

assessment of her mother's behavior: why do you think your-. Your
serves as a pivot for Connie's next tum, which initiates an alternative
explanation for Vera's mom's behavior: (your) mom is probably just-
(line 10). This partially complete tum is now collaboratively completed
by Vera with jealous (line 11). Thus, the claim which is on the floor at
this point is, from Vera's perspective, 'Mom is probably just jealous',
which we have accordingly labeled X. In her next tum Connie
acknowledges that this may be true well I mean...maybe that too (line
13), a conceding move with respect to Vera's position (X'). Yet she
goes on to present another way of viewing the matter: she just can't
handle (0.2) people close to her being in pain (line 14).11 Connie is thus
explicitly treating the two positions, Vera's and her own, as being
compatible (see too in line 13). In this case we argue··.that Connie's
counter-claim Y, while potentially incompatible with Vera"s claim X, is
being presented by Connie as not in fact incompatible. 1bis example
thus shows that Concession may be used as an alignment \strategy just
as readily as it may be used as a disalignment strategy.
Recall that in addition to the Cardinal Concessive pattern, we
presented in section 5 two variations on that pattern which recur in our
data. We have found that one of these, namely the variation in which the
conceding move, X', follows the counter-claim, has consistent
functional correlates. This is the variation which we labeled as [X Y X']
in section 5. 1. Here, recall, speaker B produces a contradiction or denial
immediately, only to modify or weaken it subsequently. When the X'
move remains in end position, thus becoming implicative for subsequent
talk, speaker B will often be heard as backing down from a claim which
is found to be too strong. The concessive pattern with [X Y X'] order
thus serves as a subtle means of self-correction.
[X Y X'] variations with X' realized as though, by contrast, are
adapted to quite different environments. Here because the conceding
move is so strongly backgrounded, it is the Y move, the contradiction or
denial, which goes on record. It can be mitigated in force if, as in
example (10), partitioning is used; recall that Laura's not all the way is a
milder way of disagreeing with the statement that helps when you have
a backache ... is to get down on your hands and knees and ... crawl
around than would be 'no it doesn't', because it opens up a way of
looking at the matter which leaves some room for the other to be
(partially) right.
8. Conclusion
In this paper we have shown how Concession is accomplished in

conversational English, how speakers reveal their knowledge of
Concessive patterns, and how they manipulate them for their own
interactional ends.We have suggested that a useful way to analyze these
interactions is to postulate a Cardinal Concessive schema, in terms of
which we can understand how speakers accomplish a variety of
interactional activities.
Functionally, we have seen that there is a range of evidence that
speakers of English orient to and manipulate an interactional Concessive
schema as they negotiate and jointly construct meanings. This schema
emerges from the frequent use speakers can make of it for such tasks as
prefacing disagreement and creating 'agreement' by acknowledging the
simultaneous existence of two potentially incompatible perspectives.
We have also seen that the variations on the Cardinal Concessive
schema that we have described are regularly used to modify, weaken, or
'back down' from a previously stated position or to highlight Y rather
than X'.
We hope to have made a contribution to the emerging field of
'interactional linguistics' by demonstrating that Concession flourishes in
the everyday speech of users of English, that grammatical, lexical, and
prosodic choices yield a range of ways in which Concession is
manifested in conversation, and that starting from a Cardinal schema
provides a rich way of accounting for the linguistic properties of this
interactional activity.
Numbered line One intonation phrase

CAPITALS Primary stress
'Final', i.e. low falling intonation
? 'Appeal', Le. high rising intonation
'Continuous', i.e. slightly falling or rising intonation
Truncated intonation phrase

Syllable lengthening
Syllable cut-off
(.) Micropause
- -- ), ( --- )
( ), ( Very short, short or not so short pause (estimated)
(1.0) Measured pause
[ Overlapping talk
[
(.h) Inbreath
(h) Outbreath
( Uncertain transcription
Notes
* We are grateful to Susanna Cumming and Gene Lerner for valuable
discussion of the issues brought up in this paper and especially to Dagmar
Barth, Cecilia Ford, Christine GoW and Susanne Gfinthner for reading and
commenting on an earlier version. Any remaining errors are our own
responsibility. Both authors contributed equally to the conceptualization and
writing of this paper.
1. Unless otherwise indicated, all of the excerpts given as examples in this paper
stem from face-to-face interaction. We have standardized transcriptions to
facilitate reading. For notational conventions, please see the appendix.
2. This notion was implicit in the 'Concessive' relation which Mann and
Thompson identified in written texts (see below).
3. We are grateful to John W. Du Bois, director of the CSAE, for his efforts in
creating the CSAE and making its contents available to us, and to Steve
Albert for sharing his data with us.
4. See, however, Kotthoff 1993 and Vuchinich 1984, 1990 for work in which
concession is understood this way.
5. A minimal token like yeah may of course be merely signalling recipiency
rather than conceding a point. To distinguish between the recipient-signal use
and the conceding use, we suggest that it may be helpful to consider the
prosodic realization of the minimal token. If, as in (4), the token forms a full-
fledged intonation phrase of its own and is not produced with reduced
prosody or in overlap, there are grounds for regarding it as a conversational
move in its own right, which in the context under discussion would have
conceding force. At the same time, as Susanne Gfinthner has suggested to us,
the yeah but format can be seen as a conventionalization of the Cardinal
Concessive pattern, which in many instances of its use has lost all or most of
its concessive force.
6. It is also possible for X' and Y to be expressed hypothetically and linked by
if... then. For instance in one of the Gulf War calls, the moderator claims one
bomb is a big difference from making a hundred or a thousand. and nobody
said he'd make more than one or two (the reference is to Saddam Hussein),
whereupon his interlocutor replies well even then the damage with just one
bomb is quite a bit. We propose that this sequence can be analyzed as:
A: X nobody said he'd make more than one or two

B: X' even then (=even if he made only one bomb)
y the damage with just one bomb is quite a bit
The Cardinal Concessive schema thus also allows for what have been called
'concessive conditionals' (Konig 1986).
7. Etymologically the particle though and the subordinator though are related,
the latter being the result of a process of grammaticalization via which an
adverb becomes a subordinating conjunction (Hopperrrraugott 1993).
According to Quirk et al. (1972) the conjunct though is often "an informal
equivalent to an abbreviated subordinate clause with the conjunction though
as subordinator" (1972:525).
8. See Sacks (1987) for discussion of a similar case.
9. There are interesting derivative forms which appear in monadic discourse but
for reasons of space we will not be able to go into them here (see, however,
Couper-KuhlenIThompson, 1999).
10. See also Quintilian, cited in Perelman/Olbrechts-Tyteca 1969.
11. Notice that Vera also proposes a candidate completion of Connie's as yet
unfinished 'she just can't handle:' with 'reality' (line 15), which Connie
however does not acknowledge.
References
Altenberg, Bengt
1986 Contrastive linking in spoken and written English. In: Gunnel Tottie
and Ingegerd Backlund (eds.), English in speech and writing, 13-40.
Uppsala: Academiae Upsaliensis.
Antaki, Charles, and Wetherell, Margaret
1999 Show concessions. Discourse Studies 1: 7-27.
Azar, Moshe
1997 Concession relations as argumentation. Text 17.3: 301-316.
1986 An introduction to English prosody. Ttibingen: Max Niemeyer and
London: Edward Arnold.
Wilhelm Neumann and Sabine Schuelting (eds.), Anglistentag 1998
Erfurt: Proceedings, 29-39. Trier: Wissenschaftlicher Verlag.
Gtintlrnner, Susanne
1999 Entwickelt sich der Konzessivkonnektor obwohl zum
Diskursmarker? Grammatikalisierungstendenzen im gesprochenen
Deutsch. Arbeitspapier Nr. 95, Fachgruppe Sprachwissenschaft,
Universitat Konstanz.
Hopper, Paul 1. and Elizabeth C. Traugott
Karantzola, Eleni
1995 Let's talk about concession: the case of the Modern Greek particle
as. Journal ofPragmatics 24.1/2: 55-75.
Konig, Ekkehard
1985a On the history of concessive connectives in English. Diachronic and
synchronic evidence. Lingua 66: 1-19.
1985b Where do concessives come from? In: 1. Fisiak (ed.), Historical
semantics, historical wordformation, 263-282. Berlin: Mouton.
1986 Conditionals, concessive conditionals and concessives: areas of
contrast, overlap and neutralization. In: Elizabeth Traugott, Alice ter
Meulen, Judy Reilly, and Charles Ferguson (eds.), On conditionals,
229-246. Cambridge: Cambridge University Press.
1988 Concessive connectives and concessive sentences: cross-linguistic
regularities and pragmatic principles. In: John A. Hawkins (ed.),
Explaining language universals, 145-166. Oxford: Blackwell.
Kotthoff, Helga
1993 Disagreement and concession in disputes: on the context sensitivity
of preference structures. Language in Society 22: 193-216.
Lerner, Gene H.
1991 On the syntax of sentences-in-progress. Language in Society 20:
441-458.
1996 On the 'semi-permeable' character of grammatical units in
conversation: conditional entry into the turn space of another
speaker. In: Elinor Ochs, Emanuel Schegloff, and Sandra A.
Thompson (eds.), Interaction and grammar, 238-271. Cambridge:
Longacre, Robert E.
1983 The grammar ofdiscourse. New York: Plenum Press.
1987a Rhetorical Structure Theory: a theory of Text Organization, prepared
for Livia Polanyi (ed.), Discourse Structure, Ablex. Also available as
Information Sciences Institute Research Report 87-190, 4676
Admiralty Way, Marina del Rey, CA 90292-6695.
1987b A discourse view of concession in written English. In: Scott
DeLancey and Russell Tomlin (eds.), Proceedings of the Second
Annual Meeting of the Pacific Linguistics Conference, 435-447.
Eugene: University of Oregon Linguistics Department.
1988 Rhetorical Structure Theory: towards a functional theory of text
organization. Text 8.3: 243-281.
1992 Relational discourse structure: a comparison of approaches to
structuring text by 'contrast'. In: Shin Ja Hwang and William R.
Merrifield (eds.), Language in context: Essays for Robert E.
Longacre, 19-45. Dallas: SIL.
Martin, J. R.
1992 English text: system and structure. Amsterdam: Benjamins.
Matthiessen, Christian
1995 Lexicogrammatical cartography: English systems. Tokyo:
International Language Sciences Publishers.
Ono, Tsuyoshi and Sandra A. Thompson

1995. "What Can Conversation Tell us about Syntax?", In: Philip W. Davis
(ed.), Alternative linguistics: Descriptive and theoretical modes,
213-271 Amsterdam: Benjamins.
Ono, Tsuyoshi and Sandra A. Thompson
1996 Interaction and syntax in the structure of conversational discourse.
In: Eduard Hovy and Donia Scott (eds.), Discourse processing: an
interdicsiplinary perspective, 67-96. Heidelberg: Springer-Verlag.
Perelman, C. and L. Olbrechts-Tyteca
1969 The new rhetoric: A treatise on argumentation. Notre Dame and
London: University of Notre Dame Press.
Pomerantz, Anita
preferred/dispreferred turn shapes. In: 1. Maxwell Atkinson and John
Heritage (ed.), Structure of social action, 57-101. Cambridge:
Quintilian
1921-1933 The institutio oratoria of Quintilian, translated by H. E. Butlet.
NewYork: G.P. Putnam's Sons.
Quirk, Randolph, Sydney Greenbaum, Geoffrey Leech, and Jan Svartvik
1972 A grammar of contemporary English. London and New York:
Longmans.
Reinhardt, Mechthild
1997 Some notes on concessive 'but'. In: Mechthild Reinhardt and
Wolfgang Thiele (eds.), Grammar and text in synchrony and
diachrony: in honor of Gottfried Graustein, 233-244.
Frankfurt/Main: Verlag Iberoamericana.
Rudolph, Elisabeth
1996 Contrast: adversative and concessive relations in English, German,
Spanish, and Portuguese. Berlin: W. de Gruyter.
Sacks, Harvey
1987 On the preference for agreement and contiguity in sequences in
conversation. In: G. Button and 1.R.E. Lee (eds.), Talk and social
organization, 54-69. Clevedon: Multilingual.
Schegloff, Emanuel A., and Sacks, Harvey
1973 Opening up closings. Semiotica 8: 289-327.
Sweetser, Eve
Press.
Thompson, Sandra A.
1987 'Concessive' as a discourse relation in expository written English.
In: Brian D. Joseph and Arnold M. Zwicky, (ed.), A Festschriftfor
lIse Lehiste, 64-73. (Working Papers in Linguistics 35.) Ohio: State
University.
Vuchinich, S.
1984 Sequencing and social structure in family conflict. Social Psychology
Quarterly 47: 217-234.
410 Elizabeth Couper-Kuhlen and SandraA. Thompson
Vuchinich, S.
1990 The sequential organization of closing in verbal family conflict. In:
A. D. Grimshaw (ed.), Conflict talk. Sociolinguistic investigations of
arguments in conversations, 118-138. Cambridge: Cambridge
University Press.
"that's true, although not really, but still":
Expressing concession in spoken English
Dagmar Barth
This contribution argues that despite what is claimed in the research litera-
ture concession in speech is not rare, though the means used to realize it in
this mode differ from those in writing. After an introduction to the struc-
tural realization of concessive relations in spoken English discourse, two of
the constructions in question will be examined in more detail: those that
occur most frequently, concessive but-constructions, and those that are tra-
ditionally associated with concession, a/though-constructions. The paper
will briefly list the reasons for the preference of paratactic concessive but-
constructions in spoken discourse and then concentrate on concessive a/-
though-constructions. It identifies several discourse functions and relates
them to differences in the realization of the concessive schema in compari-
son to concessive but-constructions.
1. Introduction
One of the claims in research literature on concessive relations is that they

are rare in spoken language (Di Meola 1997: 10, Hermodsson 1978: 65).1
At first sight, this opinion is supported by the results of quantitative stud-
ies: of the 197 adverbial constructions which Ford (1993) identifies in her
corpus, only 3 are concessive (for similar results Stojanova-Jovceva
1979; Brunet 1995), and this is often explained by the assumed complex-
ity of concessive constructions. Konig (1988) also suggests that "[a]n
elaborate inventory of concessive expressions ... presupposes, it seems, a
certain tradition of argumentative writing." (1988: 145, my emphasis -
D.B.) Analyzing a corpus of spoken English, however, soon reveals that
speakers, too, can concede their interlocutor's point, though their means
may differ from those of writers (Kaufmann 1974: 2).
These means have, however, not been investigated closely: most of the
studies so far have been restricted to ideal sentences (e. g., Konig 1994;
Lotscher 1988). If "real" examples are examined at all, they are usually
taken from written language (e.g., Di Meola 1997; Rudolph 1996; Ford
1993). Moreover, even empirical studies are mainly restricted to investi-
412 Dagmar Barth
gating "established" concessive connectors, such as although 2 , preposi-

tions and particles (among others Altenberg 1986; Grote, Lenke, and
Stede 1997; Ford 1993; but Reinhardt 1997). Asyndetic relations (e.g.,
Konig 1988; Di Meola 1997; Rudolph 1996) and the use of but as a con-
cessive connector (Halliday 1985; Konig 1985a) are only mentioned.
Finally, in most of the studies, and not only those on concession, even
with "real" examples there is often no context provided.
Thus concessive relations in speech have, for the most part, escaped
attention (cf Di Meola's claim) and if identified at all, have not been con-
sidered worthwhile looking at in their own right. This is what the present
contribution attempts to rectify. I will first show that speakers also engage
in conceding and that this action is in fact quite frequent. We will then
look at two concessive constructions more closely: those realized with but
and those with although. The frequency of but-constructions in express-
ing concession in spoken discourse will be explained by its advantages as
a paratactic construction. The paper will then focus on concessive al-
though-constructions. We will examine their discourse functions and
relate them to the construction's specific way of realizing the concessive
schema.
2. Concessivity in spoken language
In order to be able to show the frequency of concessive constructions in

spoken language they need to be identified. For this purpose the onoma-
siological approach developed by Couper-Kuhlen and Thompson (1999,
this volume) was adopted. Their model, drawing on Rhetorical Structure
Theory (e.g., Mann and Thompson 1986, 1992) and Conversation Analy-
sis (e.g., Schegloff, Jefferson, and Sacks 1977; Pomerantz 1984; Sacks
1987), points to the inherently interactive nature of concessivity: one
speaker acknowledges a claim made by another, which in turn allows her
to disagree in a mitigated form. The following study is based on this ap-
proach. I will first briefly summarize Couper-Kuhlen and Thompson's
model (2.1.) and some of its variations (2.2.).
2.1. The Cardinal Concessive schema
The Cardinal Concessive schema depicts an (idealized) situation in which

speaker A makes some statement or point (X), whereupon speaker B
acknowledges that this may indeed hold (X') but claims that something
Expressing concession in spoken English 413
which could be understood to be incompatible also holds (Y). More

schematically this looks as follows:
A: X (stating something or making some point)

B: X' (acknowledging the validity of this statement or point,
the conceding move)
Y (claiming the validity of a potentially contrasting
statement or point)
Example (1) illustrates the Cardinal Concessive pattern.
(1) Wally and friends: Exchange rates (28.01; 23:54)3

(British English dinner conversation between two elderly couples.
Mora and Wally are telling Betsy and Dave about their trip to Aus-
tralia. Betsy who is also planning to go there is wondering when to
exchange currencies. Wally has just suggested doing so at the
bank.)4
X 01 Mora: .hhh but you dOn't get a very good rAte ofexchAnge at
those () [those () banks.
X' Wally: [nO too trUE you don't; nol
Y 05 but it wi il it's bEtter than () [doing it at the bA: r;
Betsy: [but whEn do you change
YOUR money;
The example begins with Mora taking up Wally's advice to exchange

currencies at the bank by pointing out that it may not be so advantageous,
as the exchange rates at banks are bad (lines 1-3). This is taken up and
agreed with by Wally in line 4, who then counters it with the claim that
even though this may indeed be the case, there are other places where the
exchange rates are even worse (lines 5-6). Apparently not yet fully satis-
fied, Betsy restates her question in lines 7-8. The concessive schema
looks as follows:
X Mora: (but) you don't get a very good rate of exchange

(implied: you have to worry about something)
X' Wally: no, too true
Y but it's better than doing it at the bar
(implied: you don't have to worry that much)
414 Dagmar Barth
2.2. Variations on the Cardinal Concessive schema
The Cardinal Concessive schema is not the only way to accomplish con-
cession. There can be variation with respect to the sequence of the ele-
ments of the concessive schema, the number of participants involved in
producing it and the kind of marking. This paragraph illustrates these
variations and gives a rough indication as to their frequency. For this pur-
pose 8.5 hours of mainly spontaneous spoken, public and private British
and American English data have been searched thoroughly for all realiza-
tions of the concessive relation.
2.2.1. Variation in the sequence of elements
In one dyadic variation, the sequence of X' and Y is reversed. This is the
case in about 34% of the examples in our corpus. Example (2) illustrates
such a case:
(2) Presidential debate: Jim Baker (20.01~ 55:12)

(American English public TV discussion; second part of the
presidential debate between Bush, Clinton and Perot, in which
journalists ask the candidates questions. Susan Ross is pointing to
the small number of women and ethnic minorities in key positions
in Bush's government.)
X 01 Sus: .hh butwhE:nwe IOO:k;

at the cIrcle ofthe kEY: pEople clOsest to you;()
your Inner cIrcle ofadVIsors;
05 wE see whIte mEn Only. ()
whY: ?
and whEn will that chAnge.
[23 lines omitted in which Bush defends himselfby listing women who
hold influential positions in his government]
Y 30 Bush:you can lOOk all arOUnd,

and you'll see first clAss strOng WOmen;
X' uh: () jim bAker's a mA:n. ()
yeah i agrE=i plead gUIlty to that;
Y' but lOOk arOUnd who's/- ()
35 lOOk who's arOUnd wIth him there.

i mean thIs is a little defENsive on your part
Here, the counterclaim Y precedes the acknowledging move X': Bush

first defends himself against the accusation of discriminating against
women and ethnic minorities and only then, ironically, acknowledges that
part of the accusation is true. Thus, the concessive pattern looks roughly
as follows:
X Sus:
why do you discriminate against women and ethnic mi-
norities in key positions
Y Bush: I don't, look at the women I have in my staff
X' yes, there is a white man, Jim Baker
Y' but there are so many women that he hardly counts
Interestingly, after the acknowledging move, Bush returns to his earlier

defense. This is marked by but in line 34. 5 Thus X', rather than merely
following Y, is framed by two utterance parts. The two utterance parts
express one and the same idea and are therefore referred to as Y and y',
in analogy to X and X'.
2.2.2. Variation in the number of participants
Apart from the order of X' and Y, variation can also concern the number
of participants involved in the expression of the concessive relation.
Whereas the Cardinal Concessive schema is dialogic, in monologic vari-
ants one single speaker produces a concessive relation. Here pseudo-
dyadic and monadic variations can be distinguished. In the pseudo-dyadic
variation the speaker produces an entire Cardinal Concessive on her own,
enacting the roles of both A and B, as in (3):
(3) Presidential debate: Incumbent (20.01~ 1:23:32)

(American English public TV discussion; first part of the presi-
dential debate between Bush, Clinton and Perot, in which the
candidates react to questions posed by the moderator. Here Bush
is defending himself as the current incumbent against Governor
Clinton's attacks by accusing Clinton himself)
01 Bush: hE keeps criticizing Us.

criticizing mE.
416 Dagmar Barth
x I'm the incUmbent.

X' fine.
Y 05 but HE's an incumbent.
and we've got to look at All the facts;
In the example above, Bush refers to Clinton's criticim of him (line 2)

saying that he (Bush) is the incumbent (X, line 3). This remark was pre-
sumably made to suggest that Bush had failed as President despite his
position of responsibility. Bush acknowledges Clinton's claim that he
(Bush) is the incumbent with fine (X', line 4) and then counters it by
pointing to Clinton's own status: but he's an incumbent (Y, line 5) with
stress on he. This implies that the public should also look at Clinton and
his success or lack of it in office. Hence all three parts of the concessive
schema are realized, though in different roles, by one single speaker,
Bush. Although pseudo-dyadic examples are interesting from the point of
view of the Cardinal schema, they amount to only about 2% of all the
relevant examples in my corpus and are thus rather rare.
Monadic variations, in contrast, with 36%, are quite frequent in our
corpus. Here, the speaker assumes the existence of some prior X
which-though not explicit-can be acknowledged in X'. This is illus-
trated in (4):
(4) Don Vogel: Flashy (8.02; 55:14)

(American English radio show, Don Vogel and T.D. Mishky are
broadcasting from the Minnesota state fair. T.D. has asked Don,
who is blind, how he manages to dress well-matched. Don has
answered that he only buys matching clothes so that he cannot
make wrong choices.)
01
TD: there's a lEsson in thatfor All ofus.
Don: uh hm. ()
yEAh it's very EAsy.
X' now you cAn't dress rEAl FLASHy, ()
05 by dOing it that wA~·
Y bU:t () you know i'd rAther be () able to sort ofmA:tch;
than to be realflAshy.
[you knOw what i mean?
10 TD: [uh hm,
Don: so: - ()
After T.D. has praised Don's method of buying clothes as an example to

be emulated (lines 1), Don answers this compliment by downplaying his
own merit (lines 2-3) (Pomerantz 1984). He continues by acknowledging
a possible disadvantage to an otherwise extremely advisable practice: you
can't dress real flashy by doing it that way (X', lines 4-5), which could
imply that it is not such a good practice after all. However, he dismisses
this possible counterargument by assigning less importance to being able
to dress flashy: ]' d rather be able to sort ofmatch (lines 6-9).
2.2.3. Variation in the kind of marking
What may also be surprising in the light of earlier research is that in none
of the examples given so far is the concessive relation marked by al-
though or even though. Instead it is marked by but.
Table 1 presents a frequency list for the marking of concessive rela-
tions in our corpus of spoken data. For comparison, it is complemented
by one for concessive markers in written data, which is taken from Ru-
dolph (1996).6
Table 1. Frequency list of realizations of concessive relations in spoken and

written English
Spoken data (in 8.5 hours) Written data

(total: 317) (adapted from Rudolph 1996: 5)7
260 but-constructions though
36 asyndetic constructions although
6 and-constructions in spite of
(of all these: 19 with final though despite
3 with ... though ... ) even though
4 on the other hand-constructions even when
3 and yet-constructions ... though ...
2 although-constructions although - in fact
1 initial though-construction although - nevertheless
1 yet-construction though - all the same
1 nevertheless-construction even if
1 while-construction albeit
1 apart from that-construction ... as ...
lor-construction ... that ... etc.
8
The number of concessive relations found in our spoken language corpus
clearly counters claims that concession is rare in spoken language (cf. 1).
In fact, against the overall amount of data analyzed, it is an action which
418 Dagmar Barth
is accomplished quite frequently. Note also that markers such as on the

other hand, which are traditionally categorized as adversative (e.g., Ru-
dolph 1996: 4~ Quirk et al 1985: 635-636), had to be included, as in the
context of some of our examples they were clearly concessive.
Furthermore, although Table 1 supports claims of greater lexical
variation in written as opposed to spoken language (e.g., Altenberg 1986~
Kaufmann 1974), the items at the top of the two lists differ considerably:
whereas though and although lead the list for written concessive rela-
tions, they are rather infrequent in speaking, except for fmal though,
which Rudolph (1996: 400) explicitly excludes from the category of con-
cessive markers. The top item on the list of spoken concessive relations,
but-constructions, is not included by Rudolph at all. 9
In the following we will try to explain the frequency and distribution of
concessive but- and although-constructions. We will begin with conces-
sive but-constructions.
3. Explaining the predominance of but-constructions for the

expression of concessivity in spoken language
One does not have to look very far to account for the overwhelming pre-
dominance of but-constructions in expressing concessivity in speaking.
As paratactic constructions, they facilitate on-line production. But they
also provide an opportunity for face work by leaving the speaker room to
manoeuvre and by attending to the recipient's need for politeness.
3.1. On-line production
In line with earlier research, the observed distribution of concessive con-

structions in spoken as compared to written language can be explained by
the advantage of coordinating constructions for spoken language. Alten-
berg (1986) states that the preference for but in speaking, as against al-
though, for example, is a general preference for coordination rather than
for a particular connective.
Spoken language has been said to place special demands on both
speaker and hearer because a large amount of data needs to be produced
and processed within a short period of time. In addition, the hearer cannot
return to "difficult" stretches of text in order to process them again,
unless she interrupts the flow of discourse with explicit requests for re-
pair. Similarly, the speaker's return to "difficult" stretches in the produc-
tion of the text results in interruptions of the flow of discourse. Paratactic

relations are said to simplify this high speed on-line language production
and processing, as the speaker can circumvent cognitively more demand-
ing subordinate structures (e.g. Chafe 1988; Ochs 1979; but e.g. Beaman
1984).
3.2. Room to manoeuvre: saving the speaker'sface
Another advantage of paratactic constructions is that they balance the

weight of the propositions connected. Whereas hypotactic constructions
immediately subordinate the opposing argument (Grote, Lenke, and
Stede 1997), coordinate concessive constructions grant some weight to
this argument at first, and then contrast it with Y. The speaker thereby
preserves a certain amount of "room to manoeuvre", which can be neces-
sary when opinions about the state of the world still need to be negotiated,
i.e., argued about. This way the propositions still have an equal chance to
be valid, as each of the participants tries to make her point (cf (1)). Only
at the end of this negotiation does it become clear which of the arguments
is dominant, i. e. which one of them can be subordinated. Having used
coordinating structures the speaker can then still "change sides" without
lO
losing too much face by suddenly adopting an opposing point of view
(Mulkay 1985).
3.3. Politeness: saving the interlocutor's face
The strategy of lending equal weight to both propositions has a second

advantage for concessive but-constructions, namely the possibility of
saving not only the speaker's own face but also that of her interlocutor.
Whereas hypotactic constructions immediately subordinate the opposing
argument and thereby make the face-threatening suggestion that an inter-
locutor's argument is less valid, paratactic constructions allow the
speaker to include as much material in the conceding part as is necessary
to save the opponent's face. Thus, Pomerantz (1984) mentions conces-
sive but-constructions as formatting a dispreferred action, namely the
expression of disagreement, whereby various means such as explicit first
agreement (X') can be used to delay the dispreferred move. This delay
provides the first speaker with a chance to reformulate and/or correct the
point on which her conversational partner is projecting disagreement (also
Koerfer 1979).
420 Dagmar Barth
In sum, coordinating constructions prove to be very useful for the re-

alization of concessive relations in spoken language. This explains their
frequency and in tum the infrequency of although-constructions as one
representative of hypotactic relations in spoken language. The question
which this raises, however, is: why and for what purposes are the latter
still used in spoken language?
4. Explaining the occurrence of although-constructions in spo-

ken language: four major discourse functions
In order to answer this question a collection of spontaneously produced!!

concessive but- and although-constructions was compiled from the 8.5
hour corpus mentioned above. The number of concessive although-
relations was extended by examining a further 50 hours of spoken Eng-
lish data. Thus, the composition of the subcorpus was as follows:
Table 2. Examples included in the further investigation
Construction Number of examples

12
concessive although-constructions 26
concessive but-constructions 260
Looking at the concessive although-constructions in our subcorpus in

their broader context it becomes apparent that they can fulfill at least four
functions:
1. Restricting previous claims (4. 1.),

2. Introducing additional information (4.2.),
3. Forestalling possible objections (4.3.) and
4. Summarizing the previous exchange of arguments (4.4.).
Interestingly, with the exception of the third, these functions are of a dis-
course-organizational nature. This is paralleled by the fact that in some
although-constructions a topic-organizing function, such as topic shift or
topic insertion, can also be detected. Further evidence for these "functions
on a different level" is that they all occur in overlap, either with the con-
ceding function or, partly, with each other. Therefore, the percentages in
Table 3, the table which illustrates the frequency of these functions in the
examples investigated, exceed 100%.
Table 3. Frequency of discourse functions of concessive although-constructions
Function Number of examples

(total: 26)
Restricting previous claims 19 (730/0)
Introducing additional information 15 (580/0)
Forestalling a possible objection 13 (50%)
Summarizing a previous exchange of argu- 3 (12%)
ments
Insertion 14 (56%)
Topic shift 4 (16%)
4.1. Restricting previous claims
The most frequent function which concessive although-constructions in

spoken discourse fulfill is that of restricting previous claims. These can
have been stated explicitly (restricting previous utterances) or merely
implied (restricting conclusions) (also Rudolph 1996: 410-422; Konig
1985b; Giinthner, this volume, for German obwohl 'although'). Restrict-
ing constructions are characterized by a second proposition which limits
the validity of the previous claim, as is illustrated in (5):
(5) Colin Blakemore: Ulcer (24.04; 1:03:50)

(British English radio programme. Antony Clair is interviewing
Professor Colin Blakemore, a well-known neurologist. Clair has
been trying to pinpoint reasons for Blakemore's success, one of
which he assumes to be Blakemore's having suffered from a gas-
tric ulcer in his youth, which could have been fatal.)
01 Clair: .hh so there's a sense oflIving what on

borrowed tIme even;
that this is tIme you hadn't expEcted;
you pAck itfull;
05 you nEver know when the/ -
-> =although infact that partIcular problem was EAsed as uh
the Ulcer,
and the blEEding,
[and!
10 Blake: [i hAd an operAtion when i was i think twEnty;
uh gastrActomy,
422 Dagmar Barth
Following his statement in lines 1-5 that Blakemore is now living on bor-
rowed time, giving him extra energy to accomplish his job so success-
fully, Clair seems to have realized that his claim was slightlyoverdrama-
tized. In lines 6-9 he downplays it by referring to a fact which is familiar
to him as a well-informed journalist, namely that that particular problem
was eased. He thus restricts the validity of X, Blakemore's suffering be-
ing the reason for his success today.
Although this example may be interpreted as a correction at first sight,
it is important to note that it is still restrictive: Clair only points to an im-
provement in Blakemore's condition; he does not claim that there has
never been any problem. Thus, in contrast to a correction, where the va-
lidity of the previous claim is annulled and which can therefore not be
considered concessive (Giinthner, this volume), in this restrictive example
it is still valid: Blakemore's earlier suffering could still be one of a num-
ber of reasons for his success today, even though it may not be the only
one as Clair's initial statement suggests. In contrast to German, where the
corrective function of obwohl 'although' is already so established that
there are purely corrective examples (Giinthner, this volume), in English
there is still a certain note of concession-restriction present in almost all
potentially corrective examples. Nevertheless, assigning X and Y in these
cases is extremely difficult: although usually provides a hint as to which
part of the structure is X'. In this case, however, what follows although is
more important for the rest of the conversation, i.e. it is Y rather than X'.
This underlines the status of these restrictive examples as a link between
concession and correction (Giinthner, this volume).
Restricting the validity of the previous claim is a strategy often em-
ployed when the speaker expects an unfavourable response, such as dis-
agreement (Pomerantz 1984), to her utterance. Clair also uses it in order
to be agreed with by his interlocutor. That Blakemore was in fact likely to
disagree with Clair's initially overstated claim is indicated by his collabo-
ration in producing the restriction before Clair has even finished his tum
(lines 9-10).
4.2. Introducing additional information
Closely related to the restrictive function of although-clauses is that of

introducing additional information. This close relationship manifests itself
in the fact that it is often additional information which limits the validity
of the previous claim. In (5) above, for instance, Clair's information that
Blakemore's suffering ended at a particular point in time downplays the

drama of his earlier statement. Another example is given in (6):
(6) Larry Bensky: Common Cause (71.01~ 51:42)

(American English radio show, Larry Bensky, the host, is inter-
viewing Ann McBride, director of Common Cause, an American
grass-roots movement for making politics more efficient. They
are talking about support for their goals by politicians, such as the
Republican congressman Shace. He, when interviewed the week
before, admitted that with the present reform measure they only
achieved a compromise.)
01 Larry: he bAsically said;

... [2 lines reported speech omitted]
it's nOt even very gOOd;
05 but it's All we can get thrOUgh there;
X [ ==thAt's] ==not==really gOOd enough.
X' Ann: [well]
well It is nO:t good enough as it is right NOW;
althOUgh in the last wEEk,
uh our understAnding is,
that they have wOrked to strEngthen their IObby/ ()
lObbyist gIft ban, ()
y .hhh but hE is at lEAst talking about
15 pUtting these Issues on the TA: ble;
and i 'II tEll you whAt;
X Larry: the reform measure you got through is not really good
enough
X' Ann: well, it is not good enough as it is right now
X'a although we heard that they have worked on strengthen-
ing the lobbyist gift ban
y but he is at least putting these issues on the table
In this example, Ann first partly agrees with Larry's criticism that the
reform measure they have achieved is not good enough as it is right now
(lines 7-9). The stress on now indicates that at some other point in time it
may be better. She then points out that the situation is, in fact, not as bad
as her initial statement might suggest, since they have achieved partial
success in working on a lobbyist gift ban (X' a, lines 10-13). The informa-
424 Dagmar Barth
tion given in X' a is less directly related to the line of argumentation in that
it gives additional background information, whose relevance in the pre-
sent context is signalled by although. 13 At the same time it restricts the
validity of X' and thus in a way already foreshadows an upcoming coun-
tering Y. Ann then, in fact, returns to the original concessive but-
construction and completes it in lines 14-15.
4.3. Forestalling a possible objection
A third, quite frequent function of concessive although-constructions is

that of forestalling a possible objection (also Konig 1988: 148). If speak-
ers anticipate counterarguments ,to their claim or point, they may try to
invalidate them within their own tum by making them explicit in X' be-
fore their interlocutor has a chance to state them. This was already illus-
trated in (5) above, where Clair revises his earlier overly dramatic claim
just in time. Another example is given in (7):
(7) Bernie Hermann: The rising sun (44.01; 49:35)

(American English radio-phone-in show with Bernie Hermann on
the dropping of the Hiroshima bomb. Bernie is defending it by re-
ferring to the war situation at that time; he agrees with the previ-
ous caller, who has also pointed out that there was a danger of
Russia possibly invading Japan.)
01 Bern: although the russians dId declare war on japan very very
lAte,
a:nd they dId go into manchUria;
Y but/h we were concErned, ()
05 about their getting tOO strong a fOOthold in Asia.
and perhAps; ()
X' uh althOUgh uh mister trUman never sAId that,
Y' uh uh that might have been an (Underlying)/
10 anOther one ofthe rEAsons, ()
whY the bomb was drOpped.
Bernie is of the opinion that the bomb may also have been dropped be-
cause the United States was concerned about their [Russia's - D.B.]
getting too strong a foothold in Asia (Y, lines 4-7,9-12). Nevertheless,
he is aware of an objection which other people calling in on his show
could possibly make, namely that no one ever mentioned this as a possi-
ble reason for the dropping of the bomb. He orients towards this by con-
ceding that this is not a reason given officially: Mister Truman never said
that (X', line 8), and thus forestalls such possible objections. The conces-
sive schema then looks as follows:
Y Bernie: we were concerned about their getting a foothold in

Asia and perhaps
X' although mister truman never said that
Y that might have been another reason why the bomb
was dropped
As the objection is only a possible one, i.e. not made explicit on the sur-
face, there is no X in the concessive schema.
In a sense this function was also illustrated in (6) above in that the ad-
ditional information, which depicts the situation more accurately than in
Ann's initial reaction to Larry's claim, could also be interpreted as fore-
stalling a possible objection from her collaborators in Common Cause.
4.4. Summarizing a previous exchange ofarguments
A fourth kind of although-relation does not anticipate possible objections

but accomplishes a more global discourse task, in that it summarizes the
points discussed in the preceding co-text in order to dismiss them. This is
illustrated in (8):
(8) Any Questions: Bargain (47.01; 13:13)

(British English radio discussion in which the panel answers
questions from the audience. Here they are discussing a proposed
pay deal with the TGWU according to which its members would
be given private medical treatment. After another panel member
has criticized this proposal, Enoch Powell now takes a tum.)
x 01 Pow: i suppose i stArt with a prEjudice in favour ofthOse who

are bArgaining,
arrIving at a bargain at the End ofit,
which eh sUIts both pArties.
Y 05 but uh i dOn't like,
pAY:,
in the form of() frInge bEnefits or bEnefits in kInd,
or Anything Other than straight wAges;
426 Dagmar Barth
10 i think it Always ends/ lEAds in the end to trOUble,

... [4 lines elaborating on the trouble omitted]
y uh so,
X' althOUgh () i wouldn't wish to interfEre with such a bar-
gain,
i wouldn't wish gOvernment or lEgislation to
20 interfere with such a bargain being arrived at,
Y i i hOpe that sort ofbargain is not going to become tYpi-
cal.
because ... [5 lines on the reason omitted]
uh On the matter ofuh () bUYing () cAre-
After having pointed out how difficult it is to arrive at a bargain in remu-

neration negotiations (lines 1-4), Powell admits that payment in the form
of... anything other than straight wages ... always ... leads in the end to
trouble (lines 5-15). Up to this point Powell has contrasted two proposi-
tions: X, that bargains which suit both parties are to be valued, i.e. prom-
ising some trade union members advantages is not to be criticized, and Y,
remuneration other than in straight wages leads to trouble and is to be
criticized. After his lengthy explanation of Y, he summarizes his point,
redundantly marked by so (line 16), and, by packaging X and Y neatly in
an although-construction, at the same time makes clear which of the two
propositions has more weight. The concessive schema looks as follows:
Y Powell: so
X' although I wouldn't wish anyone to interfere with such a
bargain (once it has been arrived at)
(implied: such bargains are to be valued, i.e. promising
advantages is not to be criticized)
Y I hope such bargains are not going to become typical
(implied: such bargains are to be criticized)
These summarizing although-constructions are those which most closely

resemble concessive relations in writing, but, as Table 3 shows, they are
rare in speech. This can be explained by the fact that spoken/unedited
language is usually used to negotiate opinions about the state of the
world, i.e. the relationship between X and Y, whereas written/edited lan-
guage reports the result of this negotiation. Consequently, constructions
which report the result of these negotiations, as it were, are more likely to
occur in the medium which is most often used for this purpose, i.e. writ-
Ing.
We will now see that these discourse functions nicely match the way
in which although-constructions realize the concessive schema. For this
purpose they were compared with the most frequent realization of con-
cessive patterns in our corpus: concessive but-constructions.
5. Differences in the realization of the concessive schema: con-

trasting although- and but-constructions
Four types of differences were found between these two concessive con-
structions. These include the number of speakers involved in producing
the relation (5.1.), the realization of the elements of the concessive pat-
tern, in particular X and X' (5.2.), the sequence of the concessive ele-
ments X' and Y (5.3.) and the distribution according to text type (5.4.).
5.1. The number ofspeakers
Table 4 shows that, whereas concessive but-constructions are mainly

produced by more than one speaker, the majority of the concessive al-
though-constructions is realized monologically. This correlates with tasks
such as giving additional information (cf (6», forestalling a possible ob-
jection from one's interlocutor (cf (7» or summarizing the previous dis-
cussion (cf (8», which are usually accomplished by one of the interlocu-
tors.
Concessive but-constructions, in contrast, can be used for both tasks:
dealing with explicitly stated counterarguments by other speakers in the
form of the Cardinal Concessive pattern (cf (1» or its dyadic variation
(cf (2», and dealing with possible objections (cf (4». This is reflected in
the almost even distribution of the examples over one and more than one
speaker in Table 4. The slight predominance of dialogic patterns with but-
constructions can be explained by the discourse origin of concessive but-
constructions (Hopper 1987 on emergent grammar; Konig 1985a,
1985b).14
Table 4. Number of speakers in concessive but- and although-constructions 15
Number of speakers but (total: 260) although (total: 26)

one 122 (47%)*** 22 (85%)***
more than one 138 (53%)*** 4 (15%)***
428 Dagmar Barth
5.2. The realization ofthe elements ofthe concessive schema
In concessive but-constructions, the X-element, i.e. the argument of the

first speaker, is predominantly made explicit (cf Table 5 and example
(1)). In concessive although-constructions, in contrast, the X-element is
frequently not expressed. This correlates with tasks such as giving addi-
tional information (cf (6)) as well as forestalling a possible objection (cf
(7)), which, due to the nature of these tasks, imply that the information
and/or the objection have not yet been made explicit, i.e. that there is no
x.
Table 5. Realization of X in concessive but- and although-constructions
Realization of X but (total: 260) although (total: 26)

not realized 110 (42%)*** 20 (77%)***
realized 150 (58%)*** 6 (23%)***
Tasks such as those just mentioned also explain why concessive al-
though-constructions frequently introduce a 'new' X': X cannot be taken
up as, again, with additional information and possible objections it is not
(yet) realized on the surface. With the majority of the concessive but-
constructions speaker B takes up X in X' (cf Table 6 and example 2).
Table 6. Realization of X' in concessive but- and although-constructions
Realization of X' but (total: 260) although (total: 26)

new (no X) 110 (42%)*** 20 (77%)***
taking up X 150 (58%)*** 6 (23%)***
5.3. The sequence ofX'- and Y-elements
Also the most frequent function of although-constructions, restricting the

validity of a previous claim, correlates significantly with one of the sur-
face characteristics of this construction: only those claims which have
already been made can be restricted. Therefore, when the although-
element initiates a restriction, X' must occur in post-position (cf (5)).
This explains the predominance of postposed although-clauses in Table
7. The frequency of split position, i.e. X' occurring between Y and y',
can be explained by the fact that speakers frequently return to their initial
point-whether it has been downgraded, as with restrictions, or not, as

with additional information inserted.
In concessive but-constructions, usually a preceding explicit X is taken
up. Therefore X' occurs mainly in pre-position, i.e. before Y, the valid
argument (cf Table 7 and example (1». There are also a number of ex-
amples to be found where X' is inserted between Y and Y' either fore-
stalling a possible obj ection (cf (4» or admitting the truth of part of the
preceding X after it has already been countered (cf (2».
Table 7. Position of X' with respect to Y in concessive but- and although-

constructions
Position of X' but (total: 260) although (total: 26)

pre-position 178 (68%)*** 1 ( 4%)***
post-position 18 ( 7%)*** 14 (54%)***
split position 64 (25%)* 11 (420/0)*
5.4. The distribution according to text type
The concept of text type (Werlich 1976: 39-41) has been adopted for the
present study because it was assumed that the constructions are more
sensitive to their immediate context than to the nature of the speech event
as a whole (text sort), i.e. that the heat of a discussion in a press confer-
ence is more likely to be reflected in the way a politician tries to defend
herself, for instance, than in the otherwise emotional neutrality character-
istic of this genre.
Werlich distinguishes five text types: descriptive, narrative, exposi-
tory, instructive and argumentative. A text can and usually does consist
of more than one text type. For instance, whereas on the whole the tele-
vised presidential debate can be described as predominantly argumenta-
tive, we can undoubtedly also find clearly descriptive, narrative and ex-
pository chunks in it. 16
Table 8 shows that both constructions prefer descriptive-expository
environments (cf examples (1) and (8». However, there are also signifi-
cant differences with respect to their distribution over the text types.
Whereas concessive but-constructions also occur quite frequently in ar-
gumentative environments (Konig 1988) (cf Table 8 and example (2»,
this context is rather rare for concessive although-constructions. Instead,
the latter occur more frequently in narrative-descriptive parts of texts (cf
430 Dagmar Barth
(7)) and, in addition, they are slightly more frequent in descriptive-

expository contexts than but-constructions.
Table 8. Distribution of concessive but- and although-constructions according to

text type
Distribution according to but (total: 260) although (total: 26)

text type
descriptive-expository 154 (59%)* 20 (77%)*
instructive 18 ( 70/0) 1 ( 40/0)
narrative-descriptive 19 ( 70/0) 4 (15%)
argumentative 69 (27%)*** 1 ( 4%)***
This can be explained by the fact that-as mentioned briefly above-

although-constructions predominantly occur in contexts in which the
opinion about the state of the world has already been decided upon, for
instance, when the previous exchange is being summarized. Of the text
types mentioned, functions used for "reporting" on the state-of-the world,
i.e. the relationship between X and Y, are more likely to occur in report-
ing text types such as descriptive-expository and narrative-descriptive
text types than in argumentative ones.
Concessive but-constructions are preferred when it is not yet clear
which of the arguments will "succeed". By assigning equal weight to
both propositions, they leave the speaker room to manoeuvre and are
face-saving for the interlocutor (cf 3.2, 3.3).
To conclude, in most cases the surface characteristics of concessive
although-constructions neatly match the particular tasks these construc-
tions fulfill in spoken discourse, although it must be noted that from an
exclusively synchronic point of view it remains difficult to say whether
the form has promoted certain functions or vice versa. Concessive but-
constructions, in contrast, seem to be more suited to dialogic exchanges
on the state of the world, i.e. the relationship between X and Y. This issue
will have to be addressed in further research.
6. Summary
In this paper it becomes clear how the investigation of concessive rela-

tions in a corpus of spoken language, including the broader context of the
examples, can indeed deliver new insights.
Firstly, I demonstrated that concession is not rare in spoken language.
Spoken and written language differ, rather, in the frequency and distribu-
tion of particular kinds of concessive constructions. For reasons relating

to on-line production and processing, room to manoeuvre and politeness,
paratactic constructions such as concessive but-constructions are pre-
ferred to although-constructions, which are traditionally considered hypo-
tactic.
Secondly, although but-constructions constitute the overwhelming ma-
jority of concessive patterns in spoken English, although-constructions
are still used. A closer investigation has shown that, in addition to
conceding, they can fulfill various, mainly discourse organizational
functions, such as restricting previous claims, introducing additional
information, forestalling possible objections and summarizing the
previous exchange of arguments.
Thirdly, a quantitative analysis of the realization of the concessive
schema with although-constructions and their distribution has revealed a
number of surface differences apart from the paratactic-hypotactic dis-
tinction (cf Table 9).
Table 9. Concessive but- and although-constructions in comparison
Concessive but- Concessive although-

constructions constructions
Discourse functions: interactively dealing with discourse-organizational
directly preceding utterance tasks:
• restricting the valid-
ity of previous claims
• introducing addi-
tional information
• forestalling a possible
objection
• summarizing a pre-
vious exchange
Surface differences:
number of speakers more than one one
realization of concessive
elements: X mainly realized often not realized
X' taking up X 'new'
position of X' pre-position post-position
split position
text type descriptive-expository descriptive-expository
argumentative narrative-descriptive
Thus, most examples of concessive but-constructions exhibit a pattern

which is closer to that of the Cardinal Concessive schema. Concessive
although-constructions, in contrast, are more often monologically pro-
432 Dagmar Barth
duced. Seldom do they include X-elements, so that frequently X' intro-

duces a 'new' proposition instead of acknowledging an X that has already
been uttered. In addition, the sequence of X' and Y is often reversed and
although-constructions occur predominantly in descriptive-expository and
narrative-descriptive text types.
Finally, it has been shown that these differences nicely match with the
discourse functions of concessive although-constructions. The restricting
function, for example, requires positioning of the claim to be restricted
before the although-clause, thereby fostering post-position of X'. The
function of introducing additional information, in turn, can explain the
frequent use of split position for although because the information is usu-
ally-in accordance with its digressive nature-inserted between two
parts of the proposition Y (cf also weil 'because' in Gohl and Gfulthner
1999). The function of forestalling a possible objection implies that X has
not been made explicit before, whereas the function of summarizing X
and Y in the previous exchange makes concessive although-constructions
more likely in such reporting text types as the descriptive-expository and
narrative-descriptive. All in all, concessive although-constructions seem
to be preferred for single-speaker tasks of a more discourse-
organizational nature.
Differences in the surface characteristics of concessive although- and
but-constructions suggest that the latter, in tum, may be better suited for
dealing with frequently needed local interactional tasks.
ACcent primary accent latching

Accent secondary accent lengthening
.hh in-breath
final intonation falling to low (h) laugh particle
final intonation falling to mid (.) pause
? final intonation rising to high / break-off
final intonation rising to mid () suggested transcription
final level intonation « » meta-comment
)( claim [] overlap
)(' acknowledgement
Y potentially.incompatible claim one line =:: one intonation unit
-> line relevant for analysis
Notes
1. I would like to thank the editors and their collaborators for valuable com-
ments on earlier versions of this paper. Special thanks is due to Elizabeth
Couper-Kuhlen, Christine Gohl and Susanne Giinthner.
2. More recently, even though has also been taken into consideration as a proto-
typical marker for concessivity (cf. Konig 1994).
3. These numbers indicate the exact location of the example in the corpus: the
initial digits specify the tape, the following digits the time of occurrence.
4. For transcription conventions cf. appendix.
5. This polyfunctionality can be considered an indication of the relationship
between but as a clausal conjunction and a discourse marker (cf. Schiffrin
1987).
6. The approach still needs to be systematically applied to written data.
7. Rudolph claims that the order of the items roughly reflects their frequency in
writing; in particular she refers to though and although as "the main and
most frequent connectives..." (Rudolph 1996: 6), not giving actual numbers,
however. Apart from frequency she employs ordering criteria such as seman-
tic peculiarity and diachronic development. Concessive conditionals have
been neglected in Table 1, except for even ifwhich can be both concessive
and concessive conditional (cf. Kjellmer 1989).
8. This number includes all constructions which can be categorized as conces-
sive according to Couper-Kuhlen and Thompson's criteria.
9. In her survey of the literature, however, Rudolph mentions Kaufmann's
(1974) observation that in spoken language "...the concessive thought is ex-
pressed through a construction with the very frequent aber." (Rudolph 1996:
201) (cf. also Konig 1991: 632).
10. For the notion of face cf. Goffman (1967, 1971); for the notion of politeness
cf. Brown and Levinson (1987). A concise summary is given in Foley
(1997).
11. This excluded examples occurring in texts written to be spoken, such as
advertisements, news and radio features.
12. Although the actual analysis only included this admittedly small number of
although-constructions, my claims are supported by the examples which a
search in the spoken part of the Bank of English on-line sample corpus
yielded (British English, approximately 9 million words).
13. In addition, the prosodic realization of the although-part underlines its status
as an insertion, in that lines 10-13 are produced at a higher speech rate and
with a lower pitch.
14. The fact that but can also be used as a concessive marker in monologic con-
cessive relations once again proves that it is a legitimate grammatical means
for producing concessive constructions, on a par with although and even
though as in traditional accounts.
15. The examples have been archived, classified and statistically analyzed by
means of the Systemic Functional Coder developed by Michael O'Donnell at
Edinburgh University (UK). In the following tables, statistically significant
434 Dagmar Barth
differences have been indicated by * (100/0 chance of error), ** (50/0 chance

of error), *** (2% chance of error).
16. As Werlich's text-type categories descriptive, narrative and expository are
difficult to distinguish, I have merged them into descriptive-expository and
narrative-descriptive respectively for this study.
R~ferences
Altenberg, Bengt
1986 Contrastive linking in spoken and written English. In: Gunnel Tot-
tie and Ingegerd Backlund (eds.), English in speech and writing. A
Symposium, 13-40. (Acta Universitatis Upsaliensis, Studia Anglis-
tica Upsaliensis 60.) Uppsala, Stockholm: Almquist and Wiksell.
Beaman, Karen
1984 Coordination and subordination revisited: Syntactic complexity in
spoken and written narrative discourse. In: Deborah Tannen (ed.),
Coherence in spoken and written discourse, 45-80. Norwood, New
Jersey: Ablex.
Brown, Penelope and Stephen Levinson
1987 Politeness: Some universals in language usage. Cambridge: Cam-
Brunet, Jacqueline
1995 La subordination: chronique d'un declin annonce. In: Hanne L.
Andersen and Gunvar Skytte (eds.), La subordination dans les lan-
gues romanes, 57-68. (=Etudes Romanues 34.) Institut d'Etudes
Romanes, Universite de Copenhague.
Chafe, Wallace
1988 Linking intonation units in spoken English. In: John Haiman and
Sandra A. Thompson (eds.), Clause combining in grammar and dis-
course, 1-27. Amsterdam/Philadelphia: Benjamins.
Wilhelm Neumann and Sabine Schiilting (eds.), Anglistentag Erfurt
1998. Trier: Wissenschaftlicher Verlag.
Di Meola, Claudio
1997 Der Ausdruck der Konzessivitat in der deutschen Gegenwartsspra-
che: Theorie und Beschreibung anhand eines Vergleichs mit dem
ltalienischen. Tiibingen: Niemeyer.
Foley, William A.
1997 Anthropological linguistics. An introduction. Oxford: Blackwell.
Ford, Cecilia
1993 Grammar in interaction: Adverbial clauses in American English
GofIman, Erving
1967 Interaction ritual: Essays on face-to-face behavior. New York:
Doubleday.
1971 Relations in public: Microstudies of the public order. New York:
Harper and Row.
Gohl, Christine and Susanne Gunthner
1999 Grammatikalisierung von weil als Diskursmarker in der gesproche-
nen Sprache. Zeitschriftfur Sprachwissenschaft 18: 39-75.
Grote, Brigitte, Nils Lenke and Manfred Stede
1997 Ma(r)king concessions in English and German. Discourse Proc-
esses 24: 87-117.
Haiman, John and Sandra A. Thompson (eds.)
1988 Clause combining in grammar and discourse. Amster-
damIPhiladelphia: Benjamins.
Halliday, M.A.K.
1985 An introduction to functional grammar. London: Arnold.
Hermodsson, Lars
1978 Semantische Strukturen der Satzgefuge im kausalen und konditiona-
len Bereich. Uppsala: Almqvist and Wiksell.
Hopper, Paul
1987 Emergent grammar. In: General session and parasession on gram-
mar and cognition. Edited by Berkeley Linguistics Society, Febr. 1-
16, 1987, 139-157.
Kaufmann, Gerhard
1974 Zur "konzessiven" Beziehung. Zielsprache Deutsch 5: 1-21.
Kjellmer, Goran
1989 Even lfand even though. English Studies 3: 256-269.
Koerfer, Armin
1979 Zur konversationellen Funktion vonja aber. Am Beispiel uiliversi-
tarer Diskurse. In: Harald Weydt (ed.), Die Partikeln der deutschen
Sprache, 14-29. BerlinlNew York: Mouton de Gruyter.
Konig, Ekkehard
1985a On the history of concessive connectives in English. Diachronic and
synchronic evidence. Lingua 66: 1-19.
1985b Where do concessives come from? On the development of conces-
sive connectives. Jacek Fisiak (ed.), Historical semantics, 263-282.
BerlinlNew York: Mouton de Gruyter.
1988 Concessive connectives and concessive sentences: Cross-linguistic
regularities and pragmatic principles. In: John A. Hawkins and
SandraA. Thompson (eds.), Explaining language universals, 145-
166. New York: Blackwell.
436 Dagmar Barth
1991 Konzessive Konjunktionen. In: Arnim von Stechow and Dieter

Wunderlich (eds.), Semantik. Ein internationales Handbuch der
zeitgenossischen Forschung, 631-639. Berlin/New York: Mouton de
Gruyter.
1994 Concessive clauses. In: The encyclopedia of language and linguis-
tics, 679-681. Edited by R. E. Asher. Oxford: Pergamon, Vol. 2.
Lotscher, Andreas
1988 Textgrammatik adversativer, restriktiver und konzessiver Satzver-
kntipfungspartikeln. Kopenhagener Beitrage zur germanistischen
Linguistik 24: 65-97.
1986 Relational propositions in discourse. Discourse Processes 9: 57-90.
1992 Relational discourse structure: A comparison of approaches to struc-
turing text by 'contrast'. In: Shin Ja 1. Hwang and William R. Mer-
rifield (eds.), Language in context: Essays for Robert E. Longacre,
19-45. Summer Institute of Linguistics and The University of Texas
at Arlington Publications in Linguistics, Publ. 107.
Mulkay, Michael
1985 Agreement and disagreement in conversation and letters. Text 5:
201-227.
Ochs, Elinor
1979 Planned and unplanned discourse. In: Talmy Givon (ed.), Syntax
and semantics. Vol. 12, nb. 4, 51-80. San Francisco/London: Aca-
demic Press.
Pomerantz, Anita
1984 Agreeing and disagreeing with assessments: some features of pre-
ferred/dispreferred tum shapes. In: J. Maxwell Atkinson and John
Heritage (eds.), Structures of social action: studies in conversation
analysis, 57-101. Cambridge: Cambridge University Press.
1985 A comprehensive grammar ofthe English language. London: Long-
man.
Reinhardt, Mechthild
1997 Some notes on concessive but. In: Mechthild Reinhardt and Wolf-
gang Thiele (eds.), Grammar and text in synchrony and diachrony.
In honour of Gottfried Graustein, 233-244. FrankfurtJMain: Ibero-
americana.
Reinhardt, Mechthild and Wolfgang Thiele
1997 Grammar and text in synchrony and diachrony. In honour of
Gottfried Graustein. FrankfurtJMain: Iberoamericana.
Rudolph, Elisabeth
1996 Contrast: Adversative and concessive relations and their expres-
sions in English, German, Spanish and Portuguese on sentence
level and text level. BerlinlNew York: Mouton de Gruyter.
Sacks, Harvey
1987 On the preference for agreement and contiguity in sequences in
conversation. In: Graham Button and John R.E. Lee (eds.), Talk and
social organization, 54-69. Clevedon: Multilingual Matters.
Schegloff, Emanuel, Gail Jefferson and Harvey Sacks
1977 The preference for self-correction in the organization of repair in
conversation. Language, 53: 361-382.
Schiffrin, Deborah
1987 Discourse markers. Cambridge: Cambridge University Press.
Stojanova-Jovceva, Stanka
1979 Untersuchungen zur Stilistik der Nebensatze in der deutschen Ge-
genwartssprache. Beitrage zur Geschichte der deutschen Sprache
und Literatur 100: 40-178.
Werlich, Egon
1976 A text grammar ofEnglish. Heidelberg: QueUe and Meyer.
Weydt, Harald (ed.)
1979 Die Partikeln der deutschen Sprache. BerlinlNew York: Mouton de
Gruyter.
From concessive connector to discourse marker:
The use of obwohl in everyday German
• • *
Interaction
Susanne Gunthner
This paper-intended as a contribution to research in "interactional

linguistics" and particularly to the study of "grammar in discourse"-
analyses the uses and functions of obwohl 'although' constructions in
everyday German conversations. In present-day German, speakers use
obwohl with subordinate and with coordinate word order. I will argue that
participants in everyday conversations use different variants of obwohl -
constructions for signalling different discourse functions, and that these
variants cannot be treated either functionally or grammatically as
equivalent. Whereas obwohl-constructions with syntactic integration are
used to indicate concession, those with syntactic non-integration are used
to limit or correct the validity of the assertion made in the preceding
utterance. In this second type of construction, obwohl takes over the role of
a discourse marker by indicating an upcoming correction or disagreement.
1. Introduction
This paper deals with the relation between grammar and discourse, and
thus, with the use and functions of syntactic constructions in everyday
conversation (Schegloff/OchsfIhompson 1996; Schlobinski 1997). In
analyzing obwohl-constructions in everyday colloquial German inter-
action, this study aims at uncovering how grammatical resources are
exploited by interactants to fulfill communicative functions. 1
In spoken colloquial German, speakers frequently make use of
syntactic constructions treated as ungrammatical by traditional German
grammar: the use of obwohl 'although' and weil 'because' with main-
clause syntax (Eroms 1980; Gaumann 1983; Kuper 1991; Schlobinski
1992; Giinthner 1993, 1996; Keller 1993; Wegener 1993; Pasch 1997;
Uhmann 1998). The increasing use of main-clause syntax in seemingly
"subordinate" concessive and causal clauses has led to heavy criticism
among self-announced "caretakers" (Sprachpjleger) of the German
language, and has even resulted in the foundation of an "advocacy group"
2
(Aktionsgemeinschaft) to save German subordinate-clause order.
440 Susanne Gunthner
Moreover, the increasing tendency to use main-clause syntax in obwohl-

and weil-constructions has led to discussions about the future of German
verb order in subordinate clauses.
German, which has verb-second as its basic word order in simple and
main clauses, requires final position of the fmite verb in subordinate
3
clauses. Thus, adverbial clauses introduced by adverbial subordinators
such as obwohl 'although'-according to German grammar-should dis-
play verb-fmal order. German thus provides a clear signal for the
grammatical incorporation of one clause into another. Consider examples
(1) and (2):4
(1) BOOK PROJECT

lEva: der hat nochfiinfundzwanzigfehler gefunden,
2 obwohl schon korrektur gelesen war.
lEva: he has another twenty-five mistakes found,

2 although already proofread was.
'lEva: he found another twenty-five mistakes,

2 although it was already proofread. '
The obwohl-clause in (1) shows final positioning of the finite verb war
'was'. During the last ten to fifteen years, however, in spoken colloquial
German as well as in certain written genres that reproduce colloquial
language (e.g. in interviews, dialogues in advertising, and in hybrid
genres like e-mail communication and chat talk) speakers frequently use
obwohl-clauses with verb-second position. 5 The following example is
taken from a conversation between Lisa and Mara. Lisa has just
mentioned that she and her boy-friend have decided to go on vacation on
the Maldive Islands this year. In her reply Mara says she wishes she
could go too:
(2) MALDIVE ISLANDS

22Mara: ich ddt SO GERN mit. (echt)
23 (0.5)
24 obwohl ZEHN stundenjlug () des fand ich doch net so gut.
22Mara: I would so much (come) along (really)

23 (0.5)
24 although a ten hour flight (.) this would find I not so good.
From concessive connector to discourse marker 441
'22Mara: I would love to come along. (really)

23 (0.5)
24 although a ten hour flight (.) I wouldn't like that.'
Mara's obwohl-utterance displays main-clause order with the finite verb

6
({and) in verb-second position; this word order is considered un-
grammatical by many reference grammars.
On the basis of an analysis of conversational data, I maintain that
participants in everyday conversation use different obwohl-constructions
for different discourse functions: whereas obwohl-constructions with
syntactic integration (final positioning of the finite verb) are mainly used
to indicate a concessive relationship between two clauses, those with
syntactic non-integration (second position of the finite verb) are used to
restrict or even cancel the validity of a preceding utterance. I suggest that
in the latter construction, obwohl has lost those grammatical features that
identify it as a concessive subordinator. Instead obwohl assumes the role
of a discourse marker indicating an upcoming correction or
· agreement. 7
dIS
The following analysis is based on 67 everyday informal German con-
versations (30 to 180 minutes in length) among friends and family mem-
bers (dinner conversations, chats over coffee, telephone interactions) col-
lected from 1983 to 1998 in different parts of Germany.
2. Obwohl as a concessive subordinator
In everyday spoken interaction, when speakers make use of obwohl to

indicate a concessive relationship between two clauses, they generally use
verb-final position in the obwohl-clause, thus marking it as grammatically
incorporated into the main clause.
Example (3) is taken from a conversation over breakfast. Klara is
telling her friends how she came to join the Green (i.e. environmentalist)
party:
(3) PARTY MEMBER

34Klara: ich bin dann- bin dann doch mitglied geworden, (-)
35 obwohl ich eigentlich nie in ne partei (r)eintreten wollte.
36 das war dann () ne ne ziemlich «hi> active zeit>.
'34Klara: then I-then I became a member after all, (-)

35 although I actually never wanted to join a political party
36 this was a rather «hee> active time>.'
As Quirk et ale (1985: 1098) state, "Concessive clauses indicate that the
situation in the matrix clause is contrary to expectation in the light of what
is said in the concessive clause". Klara's utterance ich bin dann- bin
dann doch Mitglied geworden, 'then 1- then I became a member after all'
is presented as contrary to the expectation raised by the concessive clause
obwohl ich eigentlich nie in ne Partei (r)eintreten wollte 'although I
actually never wanted to join a political party' . The concessive
construction thus depends-as Konig/van der Auwera (1988: 107)
indicate-on a "relationship of 'normal incompatibility' or dissonance
between the two component propositions"; i.e. concessive constructions
license the inference that the two sentences asserted to be true "are
instances of situations that do not normally go together" (Konig 1994:
8
681). This can be stated as presupposing a conditional:
If p', then normally -, q'.
Accordingly, the discourse presupposition in the example (3) is: if
someone has never wanted to join a political party, then he/she normally
does not become a member.
Logically, concessive clauses entail both their component clauses, i.e.,
both "p" (she never wanted to join a political party) and "q" (she became
a member after all) are treated as valid (Konig 1991; 1994).9
MannfThompson (1992: 39), who analyze concessive relations within
the framework of Rhetorical Structure Theory, argue that speakers make
use of concessive relations
to promote a particular belief or action in the presence of apparent contrary

information. The belief or action being promoted is in the nucleus, and the
apparent contrary information in the satellite. The speaker acknowledges
the apparently contrary information, but then advances the nucleus
anyway, showing that s/he does not regard the two as genuinely
incompatible. This tends to remove the satellite as an obstacle toward
favoring the nucleus.
Applied to our example, Klara presents the information in the nucleus ich
bin dann- bin dann doch mitglied geworden, 'then 1- then I became a
member after all' as valid, despite the seemingly contradictory
information in the satellite: ich wollte eigentlich nie in ne partei
(r)eintreten 'although I actually never wanted to join a political party'. In
so doing, she removes the statement in the satellite as an obstacle, and
presents the information in the nucleus as the dominant or major

information. As line 36 indicates, the information in the nucleus (Klara's
joining the party) remains the topic for the following sequence.
Whereas in example (3) the obwohl-clause was in fmal position, in the
following example (4) it is in the initial position ("obwohl p, q,,):10
(4) MANUSCRIPT
22Anna: obwohl das noch iiberarbeitungsbediirftig isch
WOLLT sie 's mitnehm.
'22Anna: although it still needs to be revised

she wanted to take it along'
Also in the case of (4) the speaker (Anna) promotes the validity of the
information in the nucleus (sie wollt es mitnehm; 'she wanted to take it
along' in spite of apparent contrary information in the satellite.
Prosodically, the obwohl-clause and the following main clause form one
intonational contour and are thus realized as a single intonational
trajectory.
Concessive obwohl-clauses can also be inserted into the main clause.
The following example is part of an episode in which Urs talks about his
political activism as a student:
(5) GOOD FRIDAY

82Urs: und dann hatte ICH also,
83 <<all> obwohl ich ja wie gesagt bei der jungen union war, >
84 zufiilligerweise die MAO BIBEL dabei.
85 nur um in der schule zu zeigen
'82Urs: and so I carried,

83 «all> although I was as I said before a member of the junge
. 11
UnIon, >
84 the mao bible with me, by chance.
85 only to show in school'
With this complex sentence, Drs asserts two facts (he was a member of
the Junge Union and he carried the Mao Bible with him) against the
background of the assumption that these two do not normally go together:
if you are a member of a conservative party, you do not carry around a
Mao Bible. The satellite status of the inserted parenthetical obwohl-clause
is supported by an increase in tempo and thus by the contextualization of

a low degree of relevance. 12
As the previous examples have shown, obwohl-constructions with
subordinate verb order have the freedom of different types of ordering
relative to the main clause: "q, obwohl p", "obwohl p, q", and even
inserted obwohl-clauses. Furthermore, concessive obwohl-constructions
show additional features of subordinate clauses: they can be within the
scope of a question: (4) WOLLT sie's mitnehm, obwohl es noch uberar-
beitungsbedurftig isch? 'did she want to take it along although it still
needs to be revised?'; and they can be transformed into a prepositional
phrase with trotz 'in spite of: Trotz der Tatsache, daft es noch
iiberarbeitungsbediirftig ist, wollt sie's mitnehm 'in spite of the fact that
it still needs to be revised, she wanted to take it along'.
3. Corrective uses of obwohl
In this section we will consider a use of obwohl-constructions, frequently

found in informal spoken German, which differs formally and functionally
from the concessive use of obwohl discussed in the previous section.
3.1. The initiation ofself-correction with obwoW
Speakers in colloquial German frequently use obwohl to initiate self-

correction. 13 In (2) we observed a speaker using obwohl with main-
clause syntax to introduce a correction of her former statement:
(2) MALDNE ISLANDS

22Mara: ich diit SO GERN mit. (echt)
23 (0.5)
24 obwohl ZEHN stundenflug () desfiind ich doch
net so gut.
25Lisa: hm () des macht mirjetzt gar nix [aus.]
26Mara: [i-ich]
27Mara: find- () ich halt (so) lange Fluge echt kaum aus.
28Mara: (finds) gruselig.
'22Mara: I would love to come along. (really)

23 (0.5)
24 although a ten hour flight (.) I wouldn't like that.
25Lisa: hm (.) 1 don't mind [it.]

26Mara: [1- I]
27Mara: find- (.) 1 can hardly stand (such) long flights.
28Mara: (find it) horrible.'
In contrast to complex sentences with concessive obwohl-clauses,-

where speakers assert that p and q are both valid but normally do not go
together-the speaker (Mara) in (2) first asserts "q" (ich dat SO GERN
mit. (echt)); then, after a short pause, she utters obwohl "p" (obwohl
ZEHN stunden flug [J des land ich doch net so gut.), thereby correcting
the validity of "q".l This type of obwohl-clause contradicts one of the
central semantic characteristics of concessives: the entailment of "p" and
"q".
From the perspective of Rhetorical Structure Theory, too, this use of
obwohl can no longer be subsumed under the definition of concessive
relations. Contrary to a concessive relation, in which "the speaker
acknowledges the apparently contrary information, but then advances the
nucleus anyway, showing that s/he does not regard the two as genuinely
incompatible" (MannfThompson 1992: 39), the speaker treats the two
utterances as incompatible in corrective obwohl-constructions. By
uttering the obwohl-clause, the information in the "nucleus" is rectified.
The meaning of this corrective use of obwohl is somewhat like "as
against what 1 just said", and thus corresponds to what Halliday/Hasan
(1976: 252ff) call the "corrective type" among "adversative relations". 15
Syntactic non-integration of the obwohl-clause (in 2) is supported by
prosody: both utterances (ich dat SO GERN mit. (echt) and obwohl
ZEHN stunden flug () des fand ich doch net so gut.) have their own
intonation contours and are separated by a pause. Furthermore, as the
example shows, syntactic non-integration provides the option of adding
"main clause phenomena" (Green 1976), e.g. left-dislocation phenomena:
obwohl ZEHN stunden flug () des fand ich doch net so gut' although a
ten hour flight (.) 1 wouldn't like that'.
In the following example (6), Mia first suggests (lines 17 and 18) that
Bea should ring her doorbell, then she will come down and join Bea for
their planned bike tour. But then she changes her mind (lines 21 and 22),
using obwohl to introduce a correction of her original proposal:
(6) BIKE TOUR

17Mia: klingel einfach;
18 ich [komm] dann runter.
19Bea: [okay.]
20 (0.5)
21Mia: obwohl () ich komm doch besser bei dir vorbei.
22 des isch glaub =ich gschickter.
'17Mia: just ring the bell;

18 I'll [come] down.
19Bea: [okay.]
20 (0.5)
21Mia: although (.) I'd better come and get you at your place.
22 that's more convenient l=think. '
Also in this case, the two constructions-the obwohl-clause and the

preceding one-are produced in two steps. First, Mia suggests: klingel
einfach; ich komm dann runter. After a moment of reflection-signalled
by the pause-she revises her former proposal and produces a correction:
obwohl () ich komm doch besser bei dir vorbei. Again, as in (2),
syntactic non-integration is supported by prosody: the two clauses (lines
18 and 21) carry their own intonation contours, and the pause between
the two utterances indicates their discontinuity.
Corrective obwohl-utterances no longer represent a satellite or a minor
act; on the contrary, they achieve the status of a nucleus or major act,
rectifying the preceding statement. 16 The syntactic organization (main
clause syntax) as well as the prosodic design (prosodic non-integration)
iconically reflect the elevation of its status to a major act. Consequently,
certain syntactic operations typical of subordinate clauses are no longer
possible in the case of corrective uses of obwohl (i. e. they would change
the meaning of the construction): *"Obwohl ich doch besser bei dir
vorbei komm, komm ich dann runter?"; *"Trotz meines Vorbeikommens
komm ich runter". Iconicity is also at work in the linear ordering of
corrective obwohl-utterances: the linear organization "q, obwohl p"
corresponds to the fact that the obwohl-construction is added after the
speaker realizes that the preceding utterance needs to be rectified.
Syntactic and prosodic non-integration go hand in hand with the
pragmatic independence of the obwohl-utterance and the preceding one:
both utterances have their own illocutionary forces. 17
As the following two examples (7) and (8) show, the obwohl-
construction and the preceding utterance can even represent two different
sentence types. In (7) Eva has prepared green tea and offers some to
Hans:
(7) GREEN TEA

44Eva: willsch mal proBIERe?
45Hans: hm. ich MAG kein griinen tee.
46 (0.5)
47Hans: obwohl GEB mir doch mal ne (-) h ' HALBE tasse vol/.
'44Eva: you wanna try some?

45Hans: hm. I don't like green tea.
46 (0.5)
47Hans: although pour me half a cup.'
Hans first rejects the offer and states his dislike for green tea (line 45).
After a short pause, he changes his mind and asks for a half a cup (line
47). Here, the obwohl-clause corrects the validity of the preceding speech
act (Hans' refusal). Not only do both clauses have their own prosodic
contours, they also have their own illocutionary forces. In the next
example (8), the obwohl-utterance corrects a preceding question:
(8) PARTY
64Kati: kommt der eigentlich au?
65 (-)
66Kati: [0 obwohl) () s=isch mir eigentlich ega:/. 0
67Lars: [[glaub schon))
'64Kati: is he coming too?

65 (-)
66Kati: [0 although] (.) 1 don't care really.o
67Lars: [[I assume so)]'
Examples such as (7) and (8) clearly reveal that the obwohl-utterances
are no longer within the scope of the illocutionary force of their main
clauses (Lehmann 1991: 526).
The correction introduced by obwohl may involve various levels of the
preceding statement: whereas in the case of (2), (6) and (7) the correction
was related to an explicitly stated utterance, the following example (9)
shows that obwohl can also be used to initiate the correction of a possible
implication. Ulla is talking about a neighbor who has suddenly fallen
seriously ill, and comes to the conclusion that one should consider oneself
lucky to have enjoyed good health so far.
(9) ILLNESS
36Ulla: do: kann man bis=jetzt=no=eigentlich (-)
37 Tal. Tal. Tal () no ganz FROH sei. gell?
38 (0.5)
39 OBWOHL man weij3 jo gar net was in oim SCHLUMMERT
40 (1.5)
41 vielleicht sen mir au scho bald mol DO. GWESE.
43Rolf: des weij3 mer halt [nie:.]
44Ulla: [hajo]
45Rolf: do steckt mer halt net drinne.
'36Ulla: up to now one can actually (-)

37 knock on wood (.) be glad. right
38 (0.5)
39 although you never know what's slumbering inside of you
40 (1.5)
41 perhaps our time in this world will soon be up
43Rolf: one never knows [for sure.]
44Vlla: [of course]
45Rolf: it's not up to us. '
In lines 36-37 VIla states that up to now one can be glad, inferring that
she has been healthy so far. However, after a short pause, the obwohl-
clause introduces a sudden change in her perspective: obwohl man weij3
jo gar net was in oim SCHLUMMERT 'although you never know what's
slumbering inside of you' and thus corrects the suggested inference. The
correction does not affect the factual content of the preceding statement,
but a possible conclusion which may be drawn from that utterance. Thus,
corrective obwohl-constructions may also operate on the level of possible
conclusion, i.e., what Sweetser (1990) calls the "epistemic domain".
Syntactic non-integration (do: kann man bis=jetzt=no==eigentlich (-)
Tal. TOl. TOl () no ganz FROH sei. gell? (0.5) OBWOHL man weij3jo
gar net was in oim SCHLUMMERT.) is again echoed by prosodic
discontinuity: the two clauses have their own intonation contours and are
separated by a pause of 0.5 seconds. The obwohl is marked by a local
increase in volume and a rise in pitch.
Corrections using obwohl can also relate to a term used in a preceding
utterance. 18 In (10) Elli talks about a relative who plays various
instruments:
(10) VIOLA
46Elli: der spielt do- doch au (soviel=[i=weifi)] CELLO.
47Hans: [de karl]
48Hans: alles MDGLICHE [(spielt=der).}
49Elli: [obwohl} (-) des isch glaub=i
50E/li: a BRATSCHE (was er noch spielt.)
'46Elli: he ev- even plays (as=[far=as=i=know)] CELLO.

47Hans: [karl]
48Hans: all sorts of things [(he=plays).]
49Elli: [although](-)I=think it is
50Elli: it's a VIOLA (that he also plays).'
Again, a concessive interpretation is not possible. The obwohl-construc-

tion does not presuppose: *if I think it's a viola, then normally he does
not playa cello. Instead, Elli corrects herself
In the examples presented in this section so far, the scope of the
connecting function of obwohl is extended from the clause level
(interclausal linking) to the linking of two (relatively) independent
utterances. As the following example (11) illustrates, the scope .of obwohl
can even exceed the level of linking two utterances to that of linking an
utterance with a larger discursive sequence. 19 Anja and Lisa are talking
about John, a mutual acquaintance:
(11) NEW YORK

56Anja: jaja (so is er halt)
57Lisa: ein LANG[WEILER}.
58Anja: [ja ((hi)) ja}.
59Lisa: null pep. echt.
60Anja: ja des stimmt.
61Anja: obWOHL du:, ()
66 als wir (gero und ich) () damals in new york waren,
67 da da HAT der sich ENGA[GIERT},
68Lisa: [hm}
69Anja: uns die stadt-
70 is =ja=seine=heimat[stadt,}=
71Lisa: [ja]
72Anja: gezeigt,
73 und (uns) da () rumgefiihrt,
74 war to- () total voll engagement und [WIEF],
75Lisa: [echt?]
75Anja: ich hab den (hi) echt kaum wieder(hi)erkannt.
'56Anja: yes yes (that's the way he is)

57Lisa: a [BOREl
58Anja: [yeah «hee» yeahl
59Lisa: no pep really
60Anja: yeah that's right
61Anja: although you know:, (.)
66 when we (gero and I) (.) were in new york that time,
67 he he really was very [active],
68Lisa: [hm]
69Anja: showing us the city-
70 it=s=his=home[town,]=
71Lisa: [yes]
72Anja: and
73 he showed (us) around there,
74 was to- (.) totally full of enthusiasm and [involved],
75Lisa: [really?]
75Anja: I hardly (hee) recognized (hee) him. '
After Anja has agreed to Lisa's assessment that John is a

"LANGWEILER" (bore; line 57) and has "null pep. echt" (no pep
really; line 59), she introduces a correction sequence (61) initiated by
obwohl. Here, obwohl does not introduce a corrective utterance but a
short narrative sequence. This narrative reconstruction refutes the
preceding evaluation of John in the sense of "as against what we just
said" (Halliday/Hasan 1976: 254). Obwohl assumes the role of a framing
device which not only marks the discourse-pragmatic function of the
following clause but of the larger narrative sequence as well. Thus, the
scope of corrective obwohl can go far beyond the clause level, contrary to
adverbial subordinators. Their scope is rather "local" (i.e., it does not
exceed their immediate environment).
3.2. The initiation ofdisagreement with obwohl
Obwohl in spoken interaction is not only used by a speaker to correct his

or her own utterance. Second speakers frequently use obwohl to displ~
their disagreement with their co-participant's preceding assessments. 0
Consider (12): Lina and Ira are discussing Ira's plans to defend her
doctoral thesis once she will have started her teacher training program:
(12) TEACHER TRAINING

54Lina: des referendariat ischja auch ganz schon stressig.
55Ira: ach obwohl (-) d 'eva die sagt immer,
56 in baden-wiirttemberg ist das nichts im vergleich zu bayern.
'54Lina: teacher training is quite demanding.

55Ira: ah although (-) eva she always says,
56 the situation in baden-wiirttemberg is nothing compared to that
in bavaria. '
After Lina's statement about teacher training, Ira produces a disagree-

ment introduced by the interjection ach and an obwohl: ach obwohl (-)
d' eva die sagt immer, in baden-wiirttemberg ist das nichts im vergleich
zu bayem 'ah although (-) eva she always says the situation in baden-
wiirttemberg is nothing compared to that in bavaria'. Thus, obwohl-in
combination with ach - is used here as a "disagreement preface"
(Pomerantz 1984: 72). Ira then quotes Eva as an authority in order to
modify Lina's previous assessment. Due to its syntactic non-integration,
the obwohl-clause functions like a non-integrative obwohl-clause uttered
by a single speaker: it corrects the validity of the preceding turn.
Obwohl is also used in the following example to introduce an
upcoming disagreement. Klaus and Hans are discussing the quality of
non-alcoholic beer:
(13) SUMl\1ER HEAT

lKlaus: das is echt s 'BESTE BIER. (-)
2 ich mein von den alkoholfreien.
3 (-)
4Hans: hhm. obwohl es gibt schon BESSERE.
5 zum Beispiel BECKS is bei weitem TRINKBARER.
, 1Klaus: this· is really the best beer (-)

2 I mean among the alcohol-free ones
3 (-)
4Hans: hhm. although there are better ones
for example Becks is far more drinkable'
After Klaus' assessment concerning the high quality of the beer he is

drinking, a short pause occurs, which may already indicate upcoming
disagreement. Then, Hans uses a combination of hhm and obwohl to
introduce his disagreeing second assessment: obwohl es gibt schon
BESSERE. 'although there are better ones' .
As we have seen, corrective obwohl is not restricted to self-correction
but can also be used as a dissent marker by a second speaker to frame an
upcoming disagreement. The obwohl-construction refutes the preceding
utterance and provides a partially or totally contradictory second assess-
ment. It reads something like: "As against what you just said...".
What is it that makes obwohl such a suitable candidate for a
disagreement token? A plausible explanation seems to be that obwohl
does not articulate an explicit, "on record" dissent (such as 'no' or 'that is
not true'), but-due to its potential concessive character (i.e., its
rhetorical indication of a concession)-a weakened and more face-saving
form of indicating an upcoming disagreement. 21
In all the examples presented in sections 3.1. and 3.2. obwohl was
shown to no longer function as a concessive subordinator, but as a marker
indicating an upcoming correction or disagreement. The degree of the
correction can vary from a partial correction of the preceding utterance
(11) to a total correction which cancels the validity of the preceding
statement or speech act (6). It can correct the factual content of the
preceding utterance (2), as well as a (performed) communicative act (6),
(7) and (8), a possible conclusion or inference which the recipients may
draw from the ~receding statement (9), or a term used in the preceding
utterance (10).2 Besides self-correction it can also be used to introduce a
disagreeing second assessment and thus function as a dissent marker (12)
and (13).
The various levels on which these corrective uses of obwohl differ
from concessive obwohl-clauses are summarized below:
Concessive Corrective
obwohl obwohl
(q, obwohl p) (q, obwohl p)
Discourse function: in spite of seemingly change in perspective:

contrary information the validity of "q"
"q" is presented as valid is corrected by "p".
("q" is the major act) ("p" is the major act)
dialogical use:
disagreement
Formal characteristics:
validity of "q" "q" remains valid the validity of"q" is

cancel1ed~
dialogical use:
the validity of "q"
is questioned
syntactic properties syntactic integration syntactic non-integration

(final positioning (second positioning
of the finite verb; of the finite verb;
subordinate clause main clause pheno-
operations are possible) mena may appear)
position of the
obwohl syntagma initial and final final only
prosodic features prosodic integration only prosodic

and non-integration non-integration23
are possible
pragmatic features the obwohl-clause the obwohl-clause

is within the scope of and the preceding
the illocutionary one have their own
force of the main clause illocutionary forces
scope of connection connection of two connection of two

clauses utterances or larger
chunks of discourse
4. Restrictive obwohl as a link between concessive and

corrective obwohl
So far, the analysis of the use of obwohl in colloquial German suggests

that there are two variants of obwohl-constructions which represent
strictly separate entities. This is not the case, however. The semantics of
the concessive subordinator obwohl constrains the functional variation of
the corrective obwohl in various ways, and the two uses of obwohl still
share major features. (i) Both variants of obwohl function as cohesive
elements (Halliday/Hasan 1976). But whereas the concessive obwohl is
used to join two clauses (or smaller units), the corrective obwohl connects
utterances or even larger discursive chunks and represents a cohesive
element on the discourse level. (ii) Both variants of obwohl depend on a
sort of dissonance or incompatibility. However, whereas in the case of
concessive obwohl the speaker asserts "two facts against the background
of an assumption that situations of these types are incompatible and do
not normally go together" (Konig 1994; and Konig in this volume), in the
case of corrective obwoh/ the speaker asserts a dissonance or in-
compatibility between her/his upcoming utterance and her/his preceding
one. The element of dissonance is transferred to the discourse level. Thus,
the semantics of concessive obwoh/ is still partly inherent in the corrective
use of obwohl: it also expresses a dissonance or incompatibility. It loses
the conceding element, however. Thus, there is a loss in the semantic
complexity of the source item.
These two uses of obwoh/ (as a concessive subordinator and a
corrective marker) can be seen as two ends of a scale, which has
intermediate gradations. One such intermediate option is the restrictive
use of obwohl. This use does not share the typical characteristics of a
concessive relation (it does not presuppose "if p, normally"" q"), nor does
it cancel the validity of the preceding utterance (as the corrective obwohl
does). Instead, it restricts the preceding statement or a conclusion which
can be drawn from the preceding utterance. 24 Restrictive uses of obwohl
show verb-final as well as verb-second position.
The following transcript shows a restrictive obwoh/-construction with
final ordering of the finite verb. Elli and Fritz are talking about Greta and
her husband Hans, who "adores" her:
(14) FRIEDA'S BIRTHDAY

34Elli: der der hot doch vie:: Iflir die [gmacht.]
35Fritz: [hajo]
36E//i: obwoh/==s scho au grenze flir==en gebe hot.
'34Elli: he he's done a lot for [her.]

35Fritz: [right]
36Elli: although he has reached his limits at times. '
The obwohl-clause does not express a general dissonance or a factual

conflict between the two statements: der der hot doch vie/ flir die
[gmacht.] 'he he's done a lot for her.' and obwoh/==s scho au grenze
{fiir==en] gebe hot. 'although he has reached his limits at times'. Instead
it prevents a possible conclusion that could be drawn from the first
statement: 25 that Hans has done just everything for her, without any
limits; i.e., a possible inference is restricted. The traditional definition of a
concessive relation "if p, then normally"" q" cannot be applied here: *"if
someone has his limits, then normally he will not do a lot for someone
else". However, as with concessive relations, both "p" and "q" are
entailed and treated as valid; i.e., "p" (obwohl==s scho au grenze
{fiir==en] gebe hot., 'although he has reached his limits at times' as well
as "q" (der der hot doch viel fiir die [gmacht.], 'he he's done a lot for
her.' are logically true.
In our example (14), syntactic integration (via verb-final positioning in
the obwohl-construction) marks a close connection between the two
clauses. The prosodic realization, however, indicates that the two clauses
are produced as distinct units: the utterance in line 34 carries a fmal
intonation contour which marks completeness. Syntactic non-integration
(via verb-second positioning in the obwohl-construction) would be
possible without any change in meaning: der der hot doch vie:: 1fiir die
gmacht. obwohl s==hot scho au grenze fiir==en gebe. And indeed, in
colloquial German, speakers also use main clause syntax with this kind of
restrictive obwohl-construction.
In (15), Karl is talking about the composer Penderecki and his musical
oeuvre. In line 4 he produces an obwohl-construction in order to restrict
the validity of his preceding statement:
(15) PENDERECKI
lKarl: nein==nein ()
2 der is (sicher) nicht hi SCHRILL. (-)
3 was sicherlich der fall ist,
4 <<all> obwohl ich kenn nicht sein gesamtwerk, >
5 seine Musik ist zeitgenossisch aber nicht dogmatisch.
, 1Karl: no=no (.)

2 he (surely) isn't hee jarring. (-)
3 which surely is the case,
4 «all> although I am not familiar with his complete works,>
5 his music is contemporary but not dogmatic. '
With the obwohl-utterance Karl restricts the validity of his former

utterance was sicherlich der fall ist 'which surely is the case': he admits
that he is not familiar with Penderecki's complete works.
However, the obwohl-construction here again does not yield the
traditional presupposition for concessive relations; i.e., "if p', then
normally -, q'" cannot be applied: *"ifI am not familiar with his complete
works, then normally it is surely not the case".
The restriction of a preceding statement or a possible conclusion from
a preceding statement has the character of an afterthought. Consequently,
it is not surprising that the linear order is such that the obwohl-clause is
generally post-posed: having stated the "matrix", the speaker sees the
necessity of restricting the validity of this statement or of a possible
conclusion which may be drawn from it.
One should add, however, that the distinctions between the different
uses of obwohl are not clear-cut. Sometimes it remains unclear whether
an obwohl-utterance totally cancels or only restricts a preceding
utterance. Thus, we arrive at a continuum ranging from the concessive
subordinator obwohl via the restrictive use to obwohl as a correction
marker.
Table 1. Synchronic variants of obwohl in colloquial German
Concessive subordinator Restrictive obwohl Corrective obwohl

q obwohl p q, obwohl p q, obwohl p
p and q are valid q or a conclusion from q correction of q (or of a
which can be inferred is conclusion from q)~
restricted dialogical use: the
validity of q is questioned
verb final position verb final or verb verb second position
second position
prosodic integration and prosodic non- prosodic non-integration
non-inte~ration integration
initial and final final positioning of the final positioning of the
positioning of the obwohl-clause obwohl-clause
obwohl-clause
Due to its possible main clause syntax (and resulting main clause
phenomena), its prosodic independence and its loss of freedom in the
linear order, obwohl in its restrictive use can no longer be classified as a
subordinate conjunction. 26 Furthermore, due to the fact that restrictive
obwohl-constructions no longer presuppose "If p, then normally --, q",
they no longer represent concessive relations in the traditional sense. The
corrective use of obwohl is even further away from its original use as a
concessive subordinator: it does not show features of subordination nor
does it possess the central semantic characteristic of concessives; i.e. the
entailment of "p" and "q". Finally, its scope is extended to the level of
connecting utterances and units of discourse.
Given that corrective obwohl has lost syntactic, semantic, pragmatic
and discourse-functional properties which are characteristic for
concessive subordinators,27 the question arises: has the concessive
subordinator obwohl been reanalyzed as a discourse marker in colloquial

German?
5. Corrective obwohl as a discourse marker?
In his chapter on discourse deixis, Levinson (1983: 87£) refers to those

words which are used to indicate "the relationship between an utterance
and the prior discourse":
...there are many words and phrases in English, and no doubt most
languages, that indicate the relationship between an utterance and the prior
discourse. Examples are utterance-initial usages of but, therefore, in
conclusion, to the contrary, still, however, anyway, well, besides, actually,
all in all, so, after all, and so on. It is generally conceded that such words
have at least a component of meaning that resists truth-conditional
treatment. .. What they seem to do is indicate, often in very complex ways,
just how the utterance that contains them is a response to, or a continuation
of, some portion of the prior discourse.
Levinson's description comes close to what Fraser (1996: 186) defmes as

"discourse markers":
... an expression which signals the relationship of the basic message to the
foregoing discourse. [...] They (Le. "discourse markers"; S.G.) provide
instructions to the addressee on how the utterance to which the discourse
marker is attached is to be interpreted.
Although definitions vary a great deal, there are a number of

characteristics which most studies on discourse markers identify:28
(i) discourse markers are predominantly a feature of oral discourse;
(ii) they are drawn primarily from the syntactic classes of conjunctions,
adverbials, and prepositional phrases;
(iii) because of their frequency in oral language, discourse markers are
stylistically stigmatized and negatively evaluated (especially in
written or formal discourse);
(iv) they are usually "short items";
(v) they appear in utterance initial position;
(vi) they occur either outside the syntactic structure or loosely attached
to it and hence have no clear grammatical function;
(vii) they are optional;
(viii) they are marginal grammatical categories;
458 Susanne Ganthner
(ix) they. are multifunctional, operating simultaneously on the local

(semantic, syntactic etc.) and global (pragmatic) level.
Based on these characteristics, corrective obwohl appears to be a

straightforward instance of a discourse marker: it is a feature of oral
discourse and is stylistically stigmatized as "ungrammatical", it is a "short
item", which is only loosely attached to the syntactic structure and hence
has no clear grammatical functions, it is optional, it mainly functions as a
pragmatic and discourse grammatical category and is multifunctional at
the same time. One could even argue that it occurs initially, as the
corrective utterance is independent from the preceding utterance, and
obwohl cannot be moved from its utterance-initial position to a position
within or at the end of the utterance. 29 Since corrective obwohl indicates a
relationship of dissonance between an utterance (or a longer discursive
chunk) and the prior discourse, it can be assigned to the sub-group of
"corrective markers" (Holker 1990; GulichIKotschi 1995) or "contrastive
discourse markers" (Fraser 1998).
However, contrary to Fraser's (1990) claim that the pragmatic
meaning or function of discourse markers is clearly separate from the
semantics of their "homophonous forms" (1990: 389), I maintain that the
meaning and function of the discourse marker obwohl is still related to its
traditional meaning/function as a concessive subjunctor: corrective
obwohl still carries the meaning potential of dissonance and
incompatibility, even though it has lost its conceding element.
As for the different variants of obwohl in colloquial German, we can
now suggest the following cline, which from a synchronic perspective can
be thought of as a continuum, on which the forms available are
systematically arranged (Hopperffraugott 1993: 6£): subordinate
conjunction> (coordinate conjunction» discourse marker. 30
This cline is to be seen as a gradual process with overlaps: the
different variants co-exist and the traditional concessive meaning of
obwohl constrains the meaning/function of the later variants. At the same
time, this cline-which reveals features of semantic and syntactic
reclassification and an increase in pragmatic function-contradicts
traditional assumptions of grammaticalization, e.g. the unilinear view of
31
clause combining from loose paratactic to tight hypotactic constructions.
6. Uses of concessive conjunctions as corrective markers in

other languages
The different variants of obwohl, ranging from a concessive subordinator

to a discourse marker, raise the question of whether there are similar
reinterpretations of concessive subordinators as corrective discourse
markers in other languages. Is there a more general tendency in spoken
interaction, namely that of a development from subordinate conjunction
(via coordinate conjunction) to discourse marker? Can although, for
example, also be used as a discourse marker indicating a correction in
colloquial English? And do we fmd similar developments in non-
European languages?
In discussing the development of although-constructions from less to
more paratactic, Hopperffraugott (1993: 184) refer to a similar case in
English. They treat the paratactic use of although, which is apparently
frequently used by college students, as a "counterexample" to the
unidirectionality in clause combining and claim that this use of although
(in the sense of however) is a "hypercorrection, resulting from literacy
and learned punctuation, and therefore a temporary innovation". In his
analysis of "contrastive discourse markers", Fraser (1998) also refers to
although in the use of however. "Contrastive discourse markers'·' are used
to signal a "contrastive relationship between the S2 they introduce and a
foregoing S2, although in some cases they signal more than simple
contrast"; i.e., they "signal that the speaker intends the explicit message
conveyed by S2 to contrast with an explicit or indirect message conveyed
by SI" (Fraser 1998: 303ft). According to Fraser (1998: 313), although
in the sense of however "can have as its target the direct SI message (... )
and an indirect message conveyed by S 1".
As Barth's study (in this volume) shows, speakers of colloquial
English (not only college students) make use of although in order to
introduce a partial correction of previous utterances. However, in contrast
to the German corrective marker obwohl, the correction introduced by
although does not reach the extent of a total annulation of the validity of
the preceding statement.
In discussing the different uses of the Spanish concessive conjunction
aunque, Crevels (1998) adds a "textual level of linking" to Sweetser's
three domains (content, epistemic and speech act), and states that in this
textual use, aunque takes over the function of a "rectifying concessive"
(Crevels 1998: 140). It signals an "unexpected tum in the discourse
context, a possible paraphrase being 'if I conclude all of a sudden that
(... ), then normally I would not previously have said that (... )" (Crevels
1998: 134). The examples Crevels provides show that in its textual use,
Spanish aunque comes close to the restrictive obwohl: "the concessive
clause is not meant as a plain comment on the felicity of the preceding
assertion, but rather modifies the preceding text unit by excluding certain
interpretation options which have become available to the hearer in the
course of it. (... ) Rectifying concessive clauses are even more loosely
linked to a main clause: They can only follow the main clause and
typically exhibit main clause order in languages like German and Dutch"
(Crevels 1998: 141).
Onodera's (1995) analysis of the development of the connectors demo
and dakedo in Japanese illustrates a similar process from concessive
conjunctions to discourse markers. Demo and dakedo develop from
clause-final concessive subordinators, indicating referential contrast to
clause-initial discourse markers used to signal contrastive actions. Parallel
to this semantic, syntactic and pragmatic change "the scope of the
connecting function is enlarged" from the sentence to the discourse level,
and the pragmatic functions are expanded (Onodera 1995: 422ff).
These studies show that the use of a traditionally concessive sub-
ordinator as a contrastive discourse marker is not restricted to German,
but that various languages show similar developments. 32 Still, there are
differences concerning the extent of the contrast: in English, paratactic
although may be used to restrict a preceding utterance, but "q" is still
treated as valid. In German and Japanese, however, those contrastive
markers originating from concessive connectors (obwohl, demo and
dakedo) can be used to cancel a preceding utterance or communicative
action.
7. Conclusion
In this paper I have discussed obwohl-constructions in contemporary

spoken German. Speakers not only use obwohl as a subordinator
expressing concessive relations, they also use obwohl-constructions to
restrict and cancel the validity of prior discourse. Thus, speakers
reanalyze this traditional subordinator as a discourse marker indicating an
upcoming correction or disagreement. These shifts in discursive function
involve shifts in the grammatical features of obwohl-constructions.
However, the different uses of obwohl are not arbitrary or disparate; they
are related to a certain potential inherent in obwohl, namely the display of
an incompatibility or dissonance. This dissonance can be between the two
statements expressed, between a statement and a possible conclusion, or

between two actions.
Furthermore, the ongoing development of obwohl from a concessive
subordinator to a corrective marker, and thus from an element of sentence
grammar to an element of discourse grammar, is in accordance with
tendencies attested for various aspects of the grammar of spoken
discourse; i. e. functional shifts of linguistic elements from the sentential
to the textual and discursive level. 33
(The transcription conventions follow GAT "Gesprachsanalytisches

Transkriptionssystem", Selting et al. 1998):
[ja das] finde ich conversational overlap

[du ab]
( -) short pauses of less than 0.5 sec.
(0.5) pauses of 0.5 sec. and longer
( ) unintelligible
(gestern) uncertain transcription
latching
? intonation phrase-final: rising
intonation phrase-final: slightly rising
intonation phrase-final: falling
, intonation phrase-final: slightly falling
<<?>word word> global high pitch
<<?>word word> global low pitch
a: lengthening
°no o soft voice
NEIN loud voice
mo«hi»mentan laugh particles within the utterance
hahaha laughter
«hustet» nonlexical phenomena (e.g. coughing)
Notes
* This study is closely connected to the project "Adverbial clause relations in

spoken English and German" in the Sonderforschungsbereich 471 at the
University of Konstanz. I would like to thank Peter Auer, Elizabeth Couper-
Kuhlen, Dagmar Barth, Christine Gohl, Verena Haser, Bernd Kortmann and
Manfred Krug for their comments on an earlier version of this paper.
1. Cf. also Giinthner (1996~ 1999).
2. Cf. Eisenberg (1993).
3. Cf. Kortmann (1998) on characteristics of adverbial subordinators, and cf.

Hengeveld (1998) on adverbial clauses in various European languages.
4. In these two examples a word-by-word gloss will be given in addition to the
English translations.
5. Cf. see also Metrich (1980), Gaumann (1983), Kiiper (1991), Giinthner
(1996; 1999).
6. Due to its main clause order, left dislocation is possible in this obwohl-
sentence.
7 Cf. also Gtinthner (1999).
8. In general, definitions of concessive relations operate with terms such as
"dissonance", "incompatibility", "contrast", "unexpectedness", "surprise",
etc. Cf. Di Meola (1997) and Crevels (1998).
9. I will use "p" and "q" to refer to concessive relations: "p" refers to the
obwohl-clause, "q" to the matrix clause (i.e. "obwohl p, q" and "q, obwohl
p").
10. Initial obwohl-clauses are rarely used in my data. Among the 28 uses of
concessive obwohl in my data there are only two cases of obwohl in initial
position. .
11. The "Junge Union" (young union) is the youth organization of the
conservative party (the Christian Democrats) in Germany.
12. Cf. Gumperz (1982) on the use of prosody to contextualize meaning in
interaction; cf. Uhmann (1992) on speech rate and the contextualization of
relevance.
13. For the conversational organization of repair and correction sequences cf.
ScheglofIlJeffersonlSacks (1977) and Jefferson (1987).
14. Syntactic integration would suggest a different meaning (i.e. a concessive
reading): ich dat so gern mit, obwohl ich zehn Stunden Flug nicht so gut
fand. If we include the modal particle doch, which is used in the original
version, the concessive meaning is less obvious.
15. As corrective obwohl-constructions lack the acknowledgment that an
assertion holds, despite its potential incompatibility, they no longer represent
concessives but mark contrast relations (cf. Couper-KuhlenlThompson, this
volume).
16. Cf. Grote/Lenke/Stede (1997: 92) for "major" (nucleus) and "minor acts"
(satellite) in concessions.
17. Cf. Givan (1990: 826) for iconicity and syntactic, semantic and pragmatic
independence.
18. Cf. also Gaumann (1983).
19. Cf. Giinthner (1999).
20. Cf. Pomerantz (1984) for the interactive organization of disagreement. Cf.
Giinthner (1996; 1999) for obwohl as a disagreement marker.
21. Cf. Brown/Levinson (1978) for face-saving strategies.
22. Thus, as with the concessive use of obwohl, the corrective use of obwohl
operates on different levels of discourse: (i) it can link propositions ("content
domain"); (ii) it can link aspects of knowledge; i.e. a statement and a
possible conclusion ("epistemic domain"); and (iii) it can link speech acts
("speech act domain"); cf. Sweetser (1990) and Crevels (1998).
23. Corrective obwohl is sometimes used in those written genres which

reproduce dialogical, colloquial language (e.g. reconstructions of interviews,
advertisement dialogues, private letters, e-mail communication, chat groups,
etc.). In these cases, the discontinuity is marked by a comma or period before
the obwohl-clause. Furthermore, the obwohl is sometimes separated from the
following correction by a colon or dash. The following text segment is taken
from an interview in a weekly newspaper with a German model. The
interviewer asks her whether the prejudice that models have no brains hurts
her:
Q: So ein Urteil tut weh?
A: Ach, es start mich nicht, obwohl: Es argert mich schon ein biJ3chen.
'Q: Does such a judgement hurt?
A: Ah, it doesn't bother me, although: it actually does make me a bit
angry.'
(WOCHENPOST, 6 April 1995).
24. Lerch (1929: 334) refers to the special case of "restrictions" among
concessive relations, and has treated them as an intermediate stage between
adversativity and concessivity. On restrictive uses of concessive conjunctions
also cf. Rudolph (1996: 411), who states: "The state of affairs affirmed in the
main clause is restricted in its validity by the state of affairs affirmed in the
concessive clause. The procedure of restricting serves to make clearer what
was meant and offers the opportunity of avoiding misunderstandings." Cf.
also Konig (1994: 681) on "rectifying" concessive clauses.
25. This use of obwohl is related to Sweetser's (1990: 79) "epistemic domain".
Furthermore, it comes close to what Halliday/Hasan (1976: 252f.) describe as
the "internal use of adversative relations". The meaning is something like "as
against what the current state of the communication process would lead us to
expect, the fact of the matter is..." (Halliday/Hasan 1976: 254).
26. The fact that restrictive obwohl-constructions also show main clause syntax
and thus reveal typical characteristics of coordinate constructions contradicts
Rudolph's (1996: 415) assumption that "the restrictive type is of exclusively
semantic characteristics", and can thus be considered "as a subordinate
clause". However, as Rudolph's data is limited to written material, it is not
surprising that she does not find restrictive uses of obwohl with verb-second
position. Cf. also Konig (1994: 681) on "main clause order" in (what he
calls) "rectifying concessive clauses".
27. Cf. Kortmann (1998: 458-460) on characteristics of adverbial subordinators.
28. Cf. Schiffrin (1987: 34ff.)~ Brinton (1996: 29ff.), Fraser (1998: 301),
Jucker/Ziv (1998: 1), Giinthner (1999).
29. It occupies the topological position of the German Vor-Vorfeld (pre-front-
field)~ cf. Auer (1996).
30. Cf. Kortmann (1998) on characteristics of subordinators and coordinators.
Cf. also Traugott (1995a) and Tabor/Traugott (1998) on the
grammaticalization of discourse markers.
31. Cf. Giinthner (1999) and Gohl/Giinthner (1999) who discuss present uses of
obwohl and weil as discourse markers and thus "the exploitation of old forms
for novel functions" within the theory of grammaticalization.
464 Susanne Giinthner
32. Cf. also Konig's (1994: 681) discussion of rectifying concessive clauses. He
mentions the use of French encore que for this subtype of concessive
constructions. Even in Latin the concessive connectors quamquam and etsi
(although) were sometimes used to restrict or correct the preceding sentences.
Cf. Kuhner/Stegmann's (1962) Latin grammar.
33. Cf. Traugott (1995a~ b), Auer (1996), Gtinthner (1999) and Gohl/Gtinthner
(1999).
References
Auer, Peter
1996 The pre-front field position in spoken German and its relevance as a
grammaticalization position. Pragmatics 6 (3): 295-322.
Brinton, Laurel J.
1996 Pragmatic Markers in English. BerlinlNew York: de Gruyter.
Brown, Penelope and Stephen C. Levinson
1978 Universals in language usage: Politeness phenomena. In: Esther
Goody (ed.), Questions and Politeness, 56-311. Cambridge:
Crevels, Mily
1998 Concession in Spanish. In: Hannay, Mike and A. Machelt
Bolkestein (eds.), Functional Grammar and Verbal Interaction,
Di Meola, Claudio
1997 Der Ausdruck der Konzessivittit in der Deutschen Gegenwarts-
sprache. Tubingen: Niemeyer.
Eisenberg, Peter
1993 Der Kausalsatz ist Dicht zu retten. Praxis Deutsch 118: 10-11.
Eroms, Hans-Werner
1980 Funktionskonstanz und Systemstabilisierung bei den begrundenden
Konjunktionen im Deutschen. Sprachwissenschaft 5: 73 -115.
Fraser, Bruce
1990 An approach to discourse markers. Journal of Pragmatics 14: 383-
395.
1996 Pragmatic markers. Pragmatics 6 (2): 167-190.
1998 Contrastive· discourse markers in English. In: Andreas Jucker and
Yael Ziv (eds.), Discourse Markers. Descriptions and Theory, 301-
326. AmsterdamIPhiladelphia: Benjamins.
Gaumann, Ulrike
1983 'Weil die machen jetzt bald zu '. Angabe- und Junktivsatz in der
Deutschen Gegenwartssprache. Goppingen: Kummerle.
Giv6n, Talmy
1990 Syntax. A Functional-Typological Introduction. Amsterdam:
Benjamins.
Gohl, Christine and Susanne Giinthner

1999 Grammatikalisierung von weil als Diskursmarker in der
Gesprochenen Sprache. Zeitschrift fur Sprachwissenschaft 18: 39-
75.
Green, Georgia
1976 Main clause phenomena in subordinate clauses. Language 52: 382-
397.
Grote, Brigitte, Nils Lenke and Manfred Stede
1997 Ma(r)king concessions in English and German. Discourse
Processes 24: 87-117.
Giilich, Elisabeth and Thomas Kotschi
1995 Discourse production in oral communication. A study based on
French. In: Uta Quasthoff (ed.), Aspects of Oral Communication,
30-66. Berlin: de Gruyter.
Gumperz, John J.
1982 Discourse strategies. Cambridge: Cambridge University Press.
Giinthner, Susanne
1993 ' ...weil-man kann es ja wissenschaftlich untersuchen'-
Diskurspragmatische Aspekte der Wortstellung in WEIL-Satzen.
Linguistische Berichte 143: 37-59.
1996 From subordination to coordination? Verb-second position in
German causal and concessive constructions. Pragmatics 6 (3): 323-
370.
1999 Entwickelt sich der Konzessivkonnektor obwohl zum
Diskursmarker? Grammatikalisierungstendenzen im Gesprochenen
Deutsch. Linguistische Berichte 180: 409-446.
Halliday, Michael A. K. and Ruqaiya Hasan
Hengeveld, Kees
1998 Adverbial clauses in the languages of Europe. In: lohan van der
Auwera (ed.), Adverbial Constructions in the Languages ofEurope,
335-420. Berlin/New York: Mouton de Gruyter.
Holker, Klaus
1990 Franzosisch: Partikelforschung. Lexikon der Romanischen
Linguistik V (1): 77-88.
Hopper, Paul and Elizabeth C. Traugott
Jefferson, Gail
1987 On exposed and embedded correction in conversation. In: Graham
Button and 1. R. E. Lee (eds.), Talk and Social Organization, 86-
100. Clevedon: Multilingual.
Jucker, Andreas and Yael Ziv
1998 Discourse markers: Introduction. In: Andreas Jucker and Yael Ziv
(eds.), Discourse Markers. Descriptions and Theory, 1-12.
AmsterdamIPhiladelphia: Benjamins.
466 Susanne Gilnthner
Keller, Rudi
1993 Das epistemische WElL. Bedeutungswandel einer Konjunktion. H.
J. Heringer and G. Stotzel (eds.), Sprachgeschichte und
Sprachkritik, 219-247. BerlinINew York: de Gruyter.
Konig, Ekkehard
1991 Konzessive Konjunktionen. In: Arnim von Stechow and Dieter
Wunderlich (eds.), Semantik. Ein internationales Handbuch der
Zeitgenossischen Forschung, 631-639. Berlin: de Gruyter.
1994 Concessive Clauses. In: R. E. Asher (ed.), The Encyclopedia of
Language and Linguistics, 679-681. Oxford: Pergamon.
Konig, Ekkehard and Jan Van der Auwera
1988 Clause integration in German and Dutch conditionals, concessive
conditionals and concessives. In: John Haiman and Sandra A.
Thompson (eds.), Clause Combining in Grammar and Discourse,
Kortmann, Bernd
1998 Adverbial subordinators in the languages of Europe. In: Johan van
der Auwera (ed.), Adverbial Constructions in the Languages of
Europe, 457- 562. BerlinlNew York: Mouton de Gruyter.
Kuhner, Raphael and Stegmann Carl
1962 Ausfuhrliche Grammatik der Lateinischen Sprache. Darmstadt:
Wissenschaftliche Buchgesellschaft.
Kuper, Christoph
1991 Geht die Nebensatzstellung im Deutschen verloren? Zur prag-
matischen Funktion der Wortstellung in Haupt- und Nebensatzen.
Deutsche Sprache 19: 133-158.
Lehmann, Christian
1991 Grammaticalization and related changes in contemporary German.
In: Elizabeth C. Traugott and Bernd Heine (eds.), Approaches to
Grammaticalization, 493-535. Amsterdam: Benjamins.
Lerch, Eugen
1929 Historische Franzosische Syntax. Leipzig: O. R. Reisland.
Levinson, Stephen C.
1983 Pragmatics. Cambridge: Cambridge University Press.
1992 Relational discourse structure: A comparison of approaches to
structuring text by 'contrast'. In: S. Hwang and W. R. Merrifield
(eds.), Language in Context: Essaysfor Robert E. Longacre, 19-46.
Arlington: University of Texas Press.
Metrich, Rene
1980 Zur Syntax und Semantik von 'obwohl' und 'wenn auch'. Paris:
Centre Universitaire du Grand Palais.
Onodera, Noriko O.
1995 Diachronic analysis of Japanese discourse markers. In: Andreas
Jucker (ed.), Historical Pragmatics. Pragmatic Developments in the
History ofEnglish, 393-437. AmsterdamlPhiladelphia: Benjamins.
Pasch, Renate
1997 Weil mit Hauptsatz-Kuckucksei im Denn-Nest. Deutsche Sprache
3: 252-271.
Pomerantz, Anita
preferred/dispreferred tum shapes. In: John M. Atkinson and John
Heritage (eds.), Structures ofSocial Action: Studies in Conversation
Analysis, 57-101. Cambridge: Cambridge University Press.
1985 A Comprehensive Grammar of the English Language. London:
Longman.
Rudolph, Elisabeth
1996 Contrast. Adversative and Concessive Relations and their
Expressions in English, German, Spanish, Portugese on Sentence
and Text Level. Berlin/New York: de Gruyter.
Schegloff, Emanuel A., Gail Jefferson and Harvey Sacks
1977 The preference for self-correction in the organization of repair in
conversation. Language 53 (2): 361-382.
Schegloff, Emanuel A., Elinor Ochs and Sandra Thompson (eds.)
1996 Interaction and Grammar. Cambridge: Cambridge University Press.
Schiffrin, Deborah
1987 Discourse Markers. Cambridge: Cambridge University Press.
Schlobinski, Peter
1992 Funktionale Grammatik und Sprachbeschreibung. Eine Unter-
suchung zum Gesprochenen Deutsch sowie zum Chinesischen.
Opladen: Westdeutscher Verlag.
Schlobinski, Peter (ed.)
1997 Syntax des Gesprochenen Deutsch. Opladen: Westdeutscher Verlag.
Selting, Margret, Peter Auer, Birgit Barden, Elizabeth Couper-Kuhlen, Susanne
Giinthner, Uta Quasthoff and Peter Schlobinski
1998 Gesprachsanalytisches Transkriptionssystem (GAT). Linguistische
Berichte 173: 91-122.
Sweetser, Eve E.
1990 From Etymology to Pragmatics. Metaphorical and Cultural Aspects
ofSemantic Structure. Cambridge: Cambridge University Press.
Tabor, Whitney and Elizabeth C. Traugott
1998 Structural scope expansion and grammaticalization. In: Anna
Giacalone-Ramat and Paul Hopper (eds.), The Limits of
Grammaticalization, 229-272. Amsterdam: Benjamins.
Traugott, Elizabeth C.
1995a The Role of the Development of Discourse Markers in a Theory of
Grammaticalization. Paper presented at the 12th Inter~ational
Conference on Historical Linguistics. University of Manchester.
1995b Subjectification in grammaticalization. In: Dieter Stein and Susan
Wright (eds.), Language, Subjectivity and Subjectivisation, 31- 54.
Uhmann, Susanne
1992 Contextualizing relevance: On some forms and functions of speech
rate changes in everyday conversation. In: Peter Auer and Aldo di
Luzio (eds.), The Contextualization of Language, 297-336.
1998 Verbstellungsvarianten in weil-Satzen: Lexikalische Differenzie-
rung mit grammatischen Folgen. Zeitschrift fur Sprachwissenschaft
17.1: 92-139.
Wegener, Heide
1993 weil-das hat schon seinen Grund. Zur Verbstellung in
Kausalsatzen mit WElL im gegenwartigen Deutsch. Deutsche
Sprache 4: 289-305.
Index
account 87,89-105,292-294,296 134-138, 236, 347f., 354, 359,

adverbial clause 363, 370f.
post-positioned 174, 197, 181, conceptual order 35, 38f., 40-
428f. 43,46-49,51-55,58
pre-positioned 174, 179, 428f. linear order 35, 38, 40-42, 53-
adversative connector 118, 136, 55,156,456
138,235,242,249,253 causality 1-3, 5f., 9-14, 18, 21-27,
adversativity 2, 5, 196, 258f., 264f., 29f., 35- 42, 57-59, 63f., 77, 88,
276f. 112f., 120, 122-124, 126, 134,
argumentation/argumentative 61, 137f., 141, 258, 266f., 270,
148, 151, 156, 260, 262-265, 276f., 341-343, 348, 350f., 354,
277f., 356f., 369f., 373, 383, 361f., 364f., 367, 369-373,
404,411,424,429-431 375f.,378
assessment 78, 92f., 97f., 103f., 212, hidden causality 348f.
216-218, 222, 265f., 272, 284, causals 4, 37f, 138,266, 342~see
301, 304f., 384, 386, 389, 403, also causal connective
405, 450-452 content 35-38, 370f.
assumption 4, 40, 42, 52, 123, 156, epistemic 38, 370f.
236, 239-241, 243-246, 248, causals, concessives, equivalence of
252f., 261,276, 295f., 313, 341, 344, 350f., 354,362
348f., 353, 365- 367, 393, 443, causals, concessives, paraphrase
454 relations 344, 351, 361
asyndetic construction 6, 27, 83-87, causation (concept of) 12, 22f., 27,
92, 103-105, 321f., 417 120
clause marking 284, 320f., 325, 391
clause order 111f., 120-122, 132,
134f., 137f., 146, 156-158, 240,
background 14, 16, 20, 113, 118,
443f., 446, 460~ see also causal
121f, 125, 127f., 131, 133f,
relation~ linear order~ adverbial
136, 240, 291, 313, 348f., 353,
clause, pre-positioned, post-
365f., 370, 372f., 376, 393,424,
positioned~ conditional clause,
443,454
pre- vs. post-posed
clause-combining strategy 321f.
co-construction 188, 192, 216, 218-
causal connective 3, 57-60, 63, 66, 222
77,89,104,118,136,272,306, cognition 2, 24, 35f., 57, 65, 368,
342, 346f., 352, 375 378
causal constraint 35, 38, 41, 43-45, coherence relation 11, 24, 26-28, 56,
47,49,51,53-55 60f, 63, 65, 84, 87, 256
causal relation 2, 9, 12, 14, 25-27, common integrator 246, 248
35-38, 40, 42f., 47f., 50, 52-55, complaint 92, 100f., 303f., 306
57f., 64-66, 77f., 83-85, 87-90, conceding!concession/concessivity
102, 104, 111, 113, 120-123, 1-6, 26, 177, 196, 257, 283f.,
470 Index
288, 290, 341, 343, 348-353, 119-121, 123, 126, 131, 181,
355, 361-369, 371-374, 376, 236, 239f., 243-245, 300, 313,
378, 381-387, 389f., 393-400, 318, 320, 323, 326, 329, 349,
403-405, 411-414, 417-420, 363,370
422, 425, 430f., 439, 452, 454, contrast 2, 136, 225f., 244f., 247-
458 249, 258f., 283-285, 288-294,
Cardinal Concessive Schema 296-301, 303-306, 364, 382,
356f., 384-392, 394-398, 459f.
402-405,442-446,427-431 conversation 84, 89, 92, 96, 111,
concessives 5, 178, 251f., 313f., 173, 193, 222, 283, 286, 291-
318-320, 336, 342f., 357, 362f., 293, 296, 349, 381, 384, 388,
365,370,373,445,456 393, 401-403, 405f., 413, 422,
conditional clause 439-441
pre- vs. postposed 146, 148, conversation analysis 84, 89, 92,
154f., 157, 159, 173, 179f., 173,213,283, 292f.
181, 183-194, 196, 322 corpus analysis 5, 70, 77f., 146,
conditional perfection 149 164,173, 178f., 194f., 259, 417,
conditionals 5, 114-116, 118f., 122, 420
125f., 132, 134f., 139, 143, 147, correction 147, 264-267, 302, 305,
154, 156-159, 163, 165-168, 404, 422, 439, 441, 444-448,
174f., 178, 184, 205-208, 210, 450,452,456, 459f.
213,223-226,230, 342f., 369 corrective marker 454, 458f., 461
complex conditional connective
143
content 116, 139 desirability 205f., 208f., 212, 215f.,
counterfactual 2f., 12f., 139, 226f.,229,231
152f., 156, 166, 205-209, disagreement 5, 87, 93, 103, 278,
211-218, 221-231, 342 283, 286, 288f., 292f., 297, 355,
epistemic 116, 122, 135, 156 382, 403f., 405, 419, 422, 439,
speech-act 118, 126, 139, 157f., 441, 450-453, 460
168 discourse function 146, 184, 225f.,
conjunction 1-3, 6, 46, 48, 52, 59, 411f., 420f., 427, 431f., 439,
83f., 88f., 111-113, 118-123, 441,452
125-127, 132, 135-139, 147, discourse grammar 461
161, 175, 177, 195, 198, 209, discourse marker 5f., 8, 257f., 275,
218, 236-253, 259f., 289, 294, 439, 441, 457-460
300, 313, 315f., 318, 341, 346, division of labour between grammar
352, 354, 363f., 374, 391, 4Q2, and pragmatics 235, 246f., 249,
456- 460 253
connective If., 5f., 37, 57-63, 68, duality 21, 66, 350f., 355, 362, 376
7Of. , 75, 77f., 143-146, 149, Dutch 3, 6, 46, 52, 57-61, 63f., 66,
151, 162f., 167f., 257f., 267, 70f., 77f., 317, 319, 347, 375,
272, 300, 317, 341, 344, 346f., 460
352f., 355, 359, 363, 370, 375, dyadic 381f., 403, 414-416, 427
383,418
content domain/level/relation 36-38,
40, 43, 51f., 54, 59f., 63, 114,
Index 471
enablement 119-123 213, 226, 239, 260, 264, 283-

entity type 316 286, 293, 299, 305f., 343f., 350,
epistemic domain/level/relation 3, 356f., 368, 383, 385f., 390, 399,
35-40, 42f., 46f., 51, 53f., 59f., 403, 406, 432, 439, 441, 450,
62f., 65-68, 70, 74-76, 78, 116, 459
118, 126f., 131, 236, 240f., 243- interclausal connection 235f.
245, 270-274, 313, 315, 317f., interclausallinking 449
320, 323, 326, 329, 331, 363f., intonation 85-87, 91, 105f., 120,
370,448 133-135, 137-139, 188, 191,
epistemic stance IIIf., 121, 123, 237, 239-241, 258, 266, 272,
125f., 128-130, 178, 370-373, 291, 344f.,361, 390, 396,402f.,
377 405f., 432, 445f., 448,455, 461
exceptional condition 147, 150, 152
explanation 26f., 283, 285, 289-294,
296-306 Japanese 6, 168,205f., 209f.,212f.,
causal explanation 21 216-218, 223, 226f., 323, 327,
329f.,460
justification 2, 5, 26f., 90, 100,
focus/focussing 134, 136f., 177, 103f., 147, 237, 258, 268, 272-
193f., 243, 250f., 314, 344f., 276,292
254
French 60, 143, 155
Korean 6, 205f., 212f., 216-218,
223,226
German 2, 4-6, 8, 60, 73, 87, 97,
166, 173-175, 178-181, 184,
186f., 191, 195, 198, 240-242, levels of interpretation 3, 36-38, 59-
296, 319,336, 346f., 353, 421f., 63, 112f., 238, 243, 252f., 313,
439-441,444,453,455-461 315,332,370,373,459
spoken German 83f., 173, 175, lexical contiguity 328, 330
179f., 194f., 198,444,460
writtenGerman 173f., 180, 194,
196, 198f. mental space 3f., 65, 111-114, 117-
128, 132, 134, 137f., 184, 193,
206-208,225,361,365-378
hypotaxis/hypotactic 4, 6, 180, 192,
420,431,458
narrative 27, 29, 225, 228f., 429-
432,450
iconicity/iconic 41f., 120f., 140, negation/negative 104f., 119, 144-
148, 156, 166, 184, 237, 240, 146, 151-153, 208f., 211, 240,
319,446 249f., 262, 264, 274, 287-290,
imagination 215f., 223 294, 296, 301, 314, 341, 344f.,
informativeness (scale of) 26, 317 347, 350f., 353-355, 361f.,
interaction 5, 18, 24, 49, 51, 83f., 374f.,377f.
87f., 94, 105, 111, 118f., 124,
127, 137, 173, 183-188, 199,
472 Index
parataxis/paratactic 4, 6, 83, 138, semantic domain 78, 315

161-163, 266, 270, 320f., 325, semantic partitioning 381, 399
327, 345, 431, 458f. semantic primitives 22
point of view 35, 64, 88, 152, 163, Spanish 2, 4-6, 143f., 146-148, 155,
207, 223, 257, 366, 369, 383, 157f., 160, 166-168, 257-259,
416,419,430 261,273,277, 315f.,459f.
politeness 88, 165, 418f., 431 speech act domain/level/relation 3,
polyphony 4, 368f. 59f., 119, 126, 131, 157f., 181,
polysemy 235, 257-259, 261, 275, 235f.,237,239f.,243,245,253,
277f.,363 270, 313, 317f., 320f., 323f.,
pragmatic ambiguity 235, 237f., 326f., 329, 331f.
242, 246f.,275 speech setting 111, 119, 124
prediction 38, 41f., 48f., 51f., 68-70, spoken language 5f., 173, 194, 196,
114f., 122, 125, 131, 134f., 199, 286, 288, 411f., 417f., 420,
165f.,350 430
predictive function 111, 139, 226f. statement 235-237, 240, 243, 252f.
presupposition 262, 313f., 336, 342, subject of consciousness/SOC 3, 57,
346, 352-354, 365f., 442, 455 64-70, 73-78
projectability 398 subjectivity 3, 57, 59, 63f., 67f., 74-
prosody 84-87, 98, 147, 177, 289, 78
381, 398, 401, 445f. subordinator 2, 5f., 240f., 336, 392,
440f., 450, 452-453, 456f., 459-
461
quantifier 347, 350
topic-focus articulation 243, 251

reasoning process 36, 38, 116, 119, topos 366-368, 373
124, 364f., 372
refutation 192, 262-264, 276f.
Relation (maxim of) 20f. viewpoint 4, 64, 257, 260-262, 264,
Relevance (principle of) 20-22 267, 272, 276f., 368f., 375, 378
representation 35f., 38 volitionality/volitional 63-68, 70-74
request 89, 92, 96, 99f., 103f., 267,
418
restriction 58, 61, 126, 139, 148f., word order 1, 178, 181, 319, 322,
159f., 163, 166, 238, 240f., 273, 439-441
336, 422, 428f., 455
restrictive use 421f., 454, 456
rhetorical pattern 283, 297
Rhetorical Structure Theory/RST
284, 288f., 297, 306, 412, 442,
445
scope 190, 239f., 250, 319, 344f.,

254, 272f.
semantic change 346
Authors. addresses
Noriko McCawley Akatsuka Barbara Dancygier

Dept. of East Asian Languages & English Language and Culture Program
Cultures Simon Fraser University
University of California at Los Angeles 580-515 West Hastings Street
276-B Royce Hall Vancouver, B.C.
Los Angeles 90095 V6B 5K3
USA Canada
akatsuka@ucla.edu baska@netcom.ca
Peter Auer Femke de Blijzer

Deutsches Seminar I Faculty of Arts
Albert-Ludwigs-Universitat Freiburg Tilburg University
Postfach P.O. Box 90153
D-79085 Freiburg NL-5000 LE Tilburg
Germany The Netherlands
auerp@ruf.uni-freiburg.de f.deblijzer@kub.nl
Dagmar Barth Cecilia E. Ford

Germanistisches Institut Dept. of English
Martin-Luther-Universitat Halle- University of Wisconsin-Madison
Wittenberg 600 N. Park Street
Luisenstr. 2 Madison, WI 53706
0-06099 Halle/Sa. USA
Germany ceford@facstaff.wise.edu
barth@germanistik.uni-halle.de
Christine Gohl
Elizabeth Couper-Kuhlen Sprachwissenschaft
Sprachwissenschaft (Anglistik) Universitat Konstanz
Universitat Konstanz Postfach 5560
Postfach 5560 0-78434 Konstanz
0-78434 Konstanz Germany
Germany Christine. GoW@uni-konstanz.de
elizabeth.couper@uni-konstanz.de
Susanne Giinthner
Mily Crevels Sprachwissenschaft
Dept. of Comparative Linguistics Universitat Konstanz
Leiden University Postfach 5560
PO Box 9515 0-78434 Konstanz
NL-2300 RA LEIDEN Germany
The Netherlands Susanne. Guenthner@uni-konstanz.de
crevels@rullet.leidenuniv.nl
474 Authors' addresses
Ekkehard Konig Leo Noordman

Institut flir Englische Philologie Faculty of Arts
Freie UniversiUit Berlin Tilburg University
Gosslerstr. 2-4 P.O. Box 90153
0-14195 Berlin NL-5000 LE Tilburg
Germany The Netherlands
koenig@zedat.fu-berlin.de Noordman@kub.nl
Bernd Kortmann Henk Pander Maat

Englisches Seminar I Dept. of Dutch/Centre for Language and
Albert-Ludwigs-Universitat Freiburg Communication
Rempartstr. 15 Utrecht University
0-79098 Freiburg Trans 10
Germany NL-3512 JK Utrecht
kortmann@ruf.uni-freiburg.de The Netherlands
henk.l.pandermaat@let.uu. nl
Ewald Lang
Zentrum flir Allgemeine Ted Sanders
Sprachwissenschaft Dept. of Dutch/Centre for Language and
Jagerstr. 10/11 Communication
0-10117 Berlin Utrecht University
Germany Trans 10
Lang@zas.gwz-berlin.de NL-3512 JK Utrecht
The Netherlands
Ted. Sanders@let.uu.nl
Paul Georg Meyer
Institut flir Anglistik III
RWTHAachen Scott Schwenter
Karmanstr. 17-19 Dept. of Spanish and Portuguese
0-52062 Aachen The Ohio State University
Germany 266 Cunz Hall
paul.meyer@anglistik3.rwth-aachen.de 1841 Millikin Road
Columbus, OR 43210-1229
USA
Estrella Montolio schwenter.1@osu.edu
Facultat de Filologia
Dept. de Filologia Hispanica (Lengua
Espanola) Peter Siemund
Universitat de Barcelona Institut flir Englische Philologie
Gran Via de les Corts Catalanes, 585 Freie Universitat Berlin
08007 Barcelona Gosslerstrasse 2-4
Spain 0-14195 Berlin
montolio@fil.ub.es Germany
siemund@zedat.fu-berlin.de
Authors' addresses 475
Susan Gayle Strauss Sandra A. Thompson

Dept. of Speech Communication Dept. of Linguistics
The Pennsylvania State University University of California
234 Sparks Building Santa Barbara, CA 93106
State College, PA 16802 USA
USA sathomps@humanitas.ucsb.edu
sgs9@psu.edu
Arie Verhagen
Eve Sweetser Dutch Linguistics
Dept. of Linguistics Leiden University
University of California at Berkeley P.O. Box 9515
1203 Dwinelle Hall 2300 RA Leiden
Berkeley, CA 94720-2650 The Netherlands
USA Arie. Verhagen@let.LeidenUniv.nl
sweetser@cogsci.berkeley.edu
Topics in English Linguistics
Edited by Bernd Kortmann and
Elizabeth Closs Traugott
Mouton de Gruyter . Berlin· New York
1 Niels Davidsen-Nielsen, Tense and Mood in English. A Comparison with Dan-

ish. 1990.
2 Historical English Syntax. Edited by Dieter Kastovsky. 1991.
3 English Computer Corpora. Selected Papers and Research Guide. Edited by
Stig Johansson and Anna-Brita Stenstrom. 1991.
4 Donka Minkova, The History ofFinal Vowels in English. The Sound ofMuting.
1991.
5 Lia Korrel, Duration in English. A Basic Choice, Illustrated in Comparison
with Dutch. 1991.
6 Andreas H. Jucker, Social Stylistics. Syntactic Variation in British Newspa-
pers. 1992.
7 Ken-ichi Takami, Preposition Stranding. From Syntactic to Functional Analy-
ses. 1992.
8 Bas Aarts, Small Clauses in English. The Nonverbal Types. 1992.
9 New Directions in English Language Corpora. Methodology, Results, Soft-
ware Developments. Edited by Gerhard Leitner. 1992.
10 History of Englishes. New Methods and Interpretations in Historical Linguis-
tics. Edited by Matti Rissanen, Ossi Ihalainen, Terttu Nevalainen and Irma
Taavitsainen. 1992.
11 Early English in the Computer Age. Explorations through the Helsinki Corpus.
Edited by Matti Rissanen, Merja Kyto and Minna Palander-Collin. 1993.
12 Towards a Standard English: 1600-1800. Edited by Dieter Stein and Ingrid
Tieken-Boon van Ostade. 1993.
13 Studies in Early Modern English. Edited by Dieter Kastovsky. 1994.
14 Ronald Geluykens, The Pragmatics of Discourse Anaphora in English. Evi-
dencefrom Conversational Repair. 1994.
15 Traute Ewers, The Origin ofAmerican Black English. Be-Forms in the HOO-
DOO Texts. 1996.
16 lIse Depraetere, The Tense System in English Relative Clauses. A Corpus-
Based Analysis. 1996.
17 Michiko Ogura, Verbs in Medieval English. Differences in Verb Choice in
Verse and Prose. 1996.
18 Spanish Loanwords in the English Language. A Tendency towards Hegemony
Reversal. Edited by Felix Rodriguez Gonzales. 1996.
19 Laurel J. Brinton, Pragmatic Markers in English. Grammaticalization and
Discourse Functions. 1996.
20 Christiane Dalton-Puffer, The French Influence on Middle English Morphol-
ogy. A Corpus-Based Study on Derivation. 1996.
21 lohan Elsness, The Perfect and the Preterite in Contemporary and Earlier
English. 1997.
22 Carl Bache and Niels Davidsen-Nielsen, Mastering English. An Advanced
Grammar for Non-native and Native Speakers. 1997.
23 English in Transition. Corpus-based Studies in Linguistic Variation and Genre
Styles. Edited by Matti Rissanen, Merja Kyto and Kirsi Heikkonen. 1997.
24 Grammaticalization at Work. Studies of Long-term Developments in English.
Edited by Matti Rissanen, Merja Kyto and Kirsi Heikkonen. 1997.
25 Axel HUbler, The Expressivity of Grammar. Grammatical Devices Expressing
Emotion across Time. 1998.
26 Negation in the History ofEnglish. Edited by Ingrid Tieken-Boon van Ostade,
Gunnel Tottie and Wim van der Wurff. 1998.
27 Martina Hacker, Adverbial Clauses in Scots: A Semantic-Syntactic Study.
1998.
28 Ingo Plag, Morphological Productivity. Structural Constraints in English Der-
ivation. 1999.
29 Gustav Muthmann, Reverse English Dictionary. Based on Phonological and
Morphological Principles. 1999.
30 Metaphor and Metonymy at the Crossroads. A Cognitive Perspective. Edited
by Antonio Barcelona. 2000.
31 Generative Theory and Corpus Studies. A Dialogue from 10 ICEHL. Edited
by Ricardo Bermudez-Otero, David Denison, Richard M. Hogg and C. B. Mc-
Cully. 2000.
32 Manfred G. Krug, Emerging English Modals. A Corpus-Based Study ofGram-
maticalization. 2000.
33 Cause - Condition - Concession - Contrast. Cognitive and Discourse Per-
spectives. Edited by Elizabeth Couper-Kuhlen and Bernd Kortmann. 2000.

Elizabeth Couper-Kuhlen, Bernd Kortmann-Cause, Condition, Concession, Contrast - Cognitive and Discourse Perspectives (Topics in English Linguistics, No 33) (2000) PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Elizabeth Couper-Kuhlen, Bernd Kortmann-Cause, Condition, Concession, Contrast - Cognitive and Discourse Perspectives (Topics in English Linguistics, No 33) (2000) PDF

Загружено:

Авторское право:

Доступные форматы

Cause - Condition - Concession - Contrast

@ Printed on acid-free paper which falls within the guidelines of the

Library of Congress Cataloging-in-Publication-Data

Cause, condition, concession, contrast : cognitive and dis-

Die Deutsche Bibliothek - Cataloging-in-Publication-Data

Cause - condition - concession - contrast : cognitive and

The relevance of causality 9

On the processing of causal relations 35

Domains of use or subjectivity? The distribution of three 57

Causal relations in spoken "discourse: Asyndetic constructions 83

Constructions with if, since, and because: Causality, III

On affirmative and negative complex conditional connectives 143

Pre- and post-positioning of wenn-clauses In spoken and 173

Counterfactual reasoning and desirability 205

Adversative connectors on distinct levels of discourse: A 235

Viewpoints and polysemy: Linking adversative and causal 257

The treatment of contrasts in interaction 283

Concessives on different semantic levels: A typological 313

Causal and concessive clauses: Formal and semantic relations 341

Concession implies causality, though in some other space 361

Concessive patterns in conversation 381

"that's true, although not really, but still": Expressing 411

From concessive connector to discourse marker: The use of 439

Authors' addresses 473

The theme of this book-endearingly referred to by its editors and

affinities and wahlverwandtschaften between the C-relations, which make

A further recurrent motif in the papers dealing with spoken discourse

interchangeable with it) but a tool 'designed' for or tailored to specific

The editors would like to thank Verena Haser, Lieselotte Anderwald

Elizabeth Couper-Kuhlen, Konstanz Bernd Kortmann, Freiburg

Paul Georg Meyer

"In der Natur gibt es keine Ursache und

This article discusses interrelations between relevance and causality from

Causality is an "elusive" concept (Ziv 1988: 543). We all seem to know

consequently sought to eschew the notion in their considerations alto-

1. A brief history of causality

Talking of causality under a unitary label misleadingly suggests that the

scientific and philosophical tradition of post-Renaissance Europe (e.g.

David Lewis (1973). As we shall see in explicating what an INUS con-

2. What makes a cause a cause?

to do so, it would mean a "change of policy", so to speak. After years of

3. The relevance of causality

What does it mean for an explanation to be relevant in a given context?

explicate in my own terms what I mean by "relevance" here. What

(1) Afelicitous explanation will adduce as a cause for a given event E

This formulation, as in all cases where Searle uses formulations of this

(2) A felicitous explanation X will adduce as a cause for a given event

This formulation, I think, is quite analogous to what should replace the

maxim is a maxim of Relation, not of relevance (although its summary

in (2) above. It is a prerequisite for successful communication that an

of causality as used in discursive reasoning should be derived from

So far I have considered the notion of causality as a concept external to

(3) People badly want a demonstration, a dramatisation of justice.

(5) The participants remain anonymous and there is no one upon

(6) The <accusatorial> system is more adaptable than the inquisito-

(7) In all the countries concerned the population is growing quickly.

Causality in discourse is a rather fluid phenomenon whose relations with

I am not quite sure what is meant by "reduction" in the above quotation.

observed by Ziv (1993: 181), causality could be regarded as one of a

Furthermore, when we describe something, there is often the need to

When we narrate events that are in some temporal relation to each

In this paper I have discussed interrelations between causality and rele-

oriented typological framework. Thirty years of experience with this