A New Architecture For Functional Grammar PDF

A New Architecture for Functional Grammar
≥
Functional Grammar Series 24
Editors
Casper de Groot
J. Lachlan Mackenzie
Mouton de Gruyter
Berlin · New York
A New Architecture
for Functional Grammar
edited by
Marı́a de los Ángeles Gómez-González
Mouton de Gruyter
Berlin · New York
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.
앪
앝 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
A new architecture for functional grammar / edited by J. Lachlan

Mackenzie, Marı́a de los Ángeles, Gómez-González.
p. cm. ⫺ (Functional grammar series ; 24)
Includes bibliographical references and index.
ISBN 3-11-017356-5 (pbk. : alk. paper)
1. Functionalism (Linguistics) 2. Discourse analysis. I. Macken-
zie, J. Lachlan II. Gómez-Gonzáles, Ma. de los Ángeles (Marı́a
de los Ángeles) III. Series.
P147.N478 2004
4011.118⫺dc22
2004049946
ISBN 3-11-017356-5
Bibliographic information published by Die Deutsche Bibliothek
Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data is available in the Internet at ⬍http://dnb.ddb.de⬎.
” Copyright 2004 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin.
All rights reserved, 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.
Printed in Germany.
Table of contents
Preface ......................................................................................... vii

Abbreviations ................................................................................ xiii
The Architecture of a Functional Discourse

Grammar
Kees Hengeveld ....................................................................... 1
Functional Grammar from its inception
Matthew P. Anstey ................................................................... 23
Behind the scenes: Cognition and Functional
Discourse Grammar
Carlos Inchaurralde ................................................................ 73
The question of discourse representation in
Functional Discourse Grammar
John Connolly .......................................................................... 89
Focus of attention in discourse
Francis Cornish ....................................................................... 117
The complementarity of the process and pattern
interpretations of Functional Grammar
Michael Fortescue .................................................................... 151
Functional Discourse Grammar and language
production
J. Lachlan Mackenzie ............................................................... 179
Comment clauses, Functional Discourse Gram-
mar and the grammar-discourse interface
Peter Harder ............................................................................ 197
Functional Grammar and the dynamics of
Discourse
María de los Ángeles Gómez-González ....................................... 211
The problem of subjective modality in the
Functional Grammar model
Jean-Christophe Verstraete ..................................................... 243
vi Table of contents
Remarks on layering in a cognitive-functional

language production model
Jan Nuyts .................................................................................. 275
Discourse structure, the Generalized Parallelism
Hypothesis and the architecture of
Functional Grammar
Ahmed Moutaouakil ................................................................. 299
Towards a speaker model of
Functional Grammar
Dik Bakker and Anna Siewierska ............................................. 325
Epilogue
Kees Hengeveld ........................................................................ 365
Index of names .............................................................................. 379

Index of subjects ........................................................................... 384
Preface
This volume arises indirectly from the Ninth International Conference on

Functional Grammar (ICFG9), which was held at the Universidad National
de Educación a Distancia (UNED), Madrid, from 20 to 23 September 2000.
Some of the chapters derive from papers given there, others have been spe-
cially commissioned. This book addresses, and offers solutions to, a
problem that was keenly felt at that conference: Functional Grammar (FG),
the model of the organization of human languages proposed by the late
Simon Dik, was failing to achieve one of its central ambitions, to account
for regularities of form that are attributable to the use of language in dis-
course.
The first chapter of this volume, by Kees Hengeveld of the University
of Amsterdam (Netherlands), proposes that the solution to the problem lies
in turning the original model on its head. He rejects the bottom-up ap-
proach, which started with the individual predicate and worked upwards to
a fully specified underlying representation, in favour of a top-down ap-
proach, which takes the speaker’s intention to communicate as its starting
point. At the same time, he develops suggestions in Dik’s posthumous
work that a FG should have a modular architecture, proposing three levels
of analysis for each discourse unit to be analysed: the interpersonal, the
representational and the expression levels. Within each of these levels, the
nesting of layers which came to be the trademark of FG in the nineties has
been preserved. Hengeveld sketches various ways in which this architec-
ture, to be known as Functional Discourse Grammar (FDG), is more
adequate than the previous model, by showing how it can account for phe-
nomena involving interactions between levels.
All but the last of the remaining chapters offer reactions by other schol-
ars to Hengeveld’s bold proposals. Matthew Anstey, Sessional Lecturer at
the School of Theology, Charles Sturt University (Australia) and PhD can-
didate at the Vrije Universiteit, Amsterdam (Netherlands), offers a
thoughtful, critical and thoroughly documented history of FG from its ear-
liest beginnings to the present. He shows how FG, in its various guises, has
struggled with the same seven problems throughout its existence and, even
in its metamorphosis as FDG, continues to face many of the same difficul-
ties.
viii Preface
FDG brings grammar closer to cognitive science in introducing a

Cognition module. In his chapter, Carlos Inchaurralde of the University of
Zaragoza (Spain), assesses the cognitive adequacy of FDG. While stressing
the impenetrability of cognitive processes, he points out various ways, for
example in a language like Japanese with morphosyntactic marking of
speakers’ awareness of social roles, that cognition impacts linguistic form
in a manner that can be modelled within FDG.
The following chapter, by John H. Connolly of the University of
Loughborough (United Kingdom), turns to another module of FDG, that
for the Communicative Context. He details the multiplicity of contextual
factors that can influence the form of a discourse, and proposes a notation
for representing contextual information. He distinguishes carefully between
information which belongs to the contextual module and that which is
more properly assigned to the interpersonal level. The chapter contains a
detailed application of Connolly’s proposals to an invented mini-discourse.
Discourse analysis is also central to the chapter by Francis Cornish, of
the University of Toulouse (France), but his data are all taken from attested
language use. His purpose is to show how the new FDG architecture pro-
vides a sound environment in which to account for phenomena of topic and
focus in discourse. He achieves this aim by confronting FDG with work
from the Columbia School of Linguistics (CS), which he finds to lack the
tools needed to deal with the flow of discourse; on the other hand, incorpo-
ration of CS’s concern with attention and accessibility would further
strengthen FDG treatments of language use.
Michael Fortescue, of the University of Copenhagen (Denmark), ob-
serves a certain contradiction in the fact that FDG mimics language
production while remaining a grammar. His position is that we need to ad-
dress both Process and Pattern in a complementary fashion. He illustrates
this thesis with data from Nootka, a language whose structures are heavily
controlled by considerations of discourse pragmatics. Fortescue concen-
trates on Focus, showing that in languages like Nootka it belongs to both
Process and Pattern; in a language where Focus is not encoded grammati-
cally, it will be absent from the account of Pattern. For FG, the conclusion
might be to retain the traditional architecture for Pattern and adopt the new
for Process.
The tension between static grammar and dynamic language production
also underlies the contribution by J.Lachlan Mackenzie, of the Vrije Uni-
versiteit Amsterdam (Netherlands). He proposes a variant on Hengeveld’s
proposals, dubbed Incremental Functional Grammar, in which the interper-
sonal and expression levels account for real-time processes. The
Preface ix
interpersonal component analyses the Move as a sequence of Acts, each of

which is composed of temporally successive Subacts. The role of the repre-
sentational component, in this view, is that of a declarative grammar,
providing semantic frames which organize focal and topical units into pro-
positional form.
Peter Harder, of the University of Copenhagen (Denmark), welcomes
Hengeveld’s three-tier model, sharing Fortescue’s view that we need to
recognize both the ‘cooking’ and the ‘recipe’, as Harder has it, even at the
cost of a certain duplication of information. He concentrates on construc-
tions in which there is a mismatch between the two, with especial reference
to cases where the main clause serves to ‘hedge’ the content of the subor-
dinate clause, as in I'm afraid John is ill.
In the chapter by María de los Ángeles Gómez-González, of the Univer-
sity of Santiago de Compostela (Galicia, Spain), it is again FDG’s potential
for treating the dynamics of discourse and of clause construction that is
central. She confronts Hengeveld’s proposals with other recent develop-
ments in cognitively oriented linguistics, bringing them together in her
model of Incremental Discourse Cognitive Grammar. She shows how
clauses with complex beginnings result from the speaker’s incremental
manipulation of the addressee’s focus of attention.
Modality has long been a crucial issue in FG. Its treatment in FDG is
the subject of the contribution by Jean-Christophe Verstraete, of the Uni-
versity of Leuven (Belgium). The chapter concludes that the new
architecture offers an improvement in this respect, allowing – as Ver-
straete’s new data demand – a four-way distinction such that both
epistemic and deontic modality (analysed as tensed and tenseless at the
representational level) can be either subjective (located at the interpersonal
level) or objective (and located at the representational level).
The ambition of a Functional Discourse Grammar is comparable in
many ways to the scope of Functional Procedural Grammar as developed
by Jan Nuyts, of the University of Antwerp (Belgium). In a chapter which
shares Verstraete’s concern with modality, he provides a critical review of
FDG, arguing that layering is properly situated in cognition, and not in the
grammar itself. Nuyts asks grammarian to be content with dealing with
grammatically coded distinctions, and not to dabble in the study of concep-
tualization, which is non-linguistic, but hierarchical.
Directly opposed to such functional minimalism, as Fortescue calls his
analogous proposal, is the contribution by Ahmed Moutaouakil (University
of Rabat, Morocco), who argues for an expansion of the layering principle
upwards into the structure of discourse and downwards into the inner struc-
x Preface
ture of the word. Moutaouakil proposes a Generalized Parallelism Hy-

pothesis which entails that all products of linguistic activity, from text to
word, are to be seen as richly layered discourses, with the proviso that the
realization of the many layers is increasingly constrained as one descends
to ever smaller units.
The final reaction to Hengeveld’s proposal takes the form of a substan-
tial chapter by Dik Bakker, of the University of Amsterdam (Netherlands),
and Anna Siewierska, of the University of Lancaster (United Kingdom).
Unlike their colleagues, they concentrate on the expression level, imple-
menting proposals for an incremental, dynamic view of the morphosyntax
of utterances. They propose a formalism that works top-down, left-to-right
and depth-first and permits constrained inheritance and percolation. They
then return to a central issue of this book, the relation between grammar
and production, between pattern and process, arguing that FDG offers a
model of the speaker’s actual processes of production, and backing this up
with a detailed analysis of a few lines of transcribed speech.
This volume closes with an Epilogue by Kees Hengeveld, in which he
gives various responses to issues raised in the preceding chapters and then
refines the model originally proposed. This chapter can thus be read as rep-
resenting the current status quæstionis in FDG. Hengeveld begins by
identifying FDG as a pattern model: although the structures of languages
reflect the processes of communication, FDG does not model those proc-
esses. Yet he indicates his sympathy with dynamic implementations of the
model. He goes on to welcome the various contributions to clarification of
the cognitive and contextual components of the model and ends by offering
an expanded version of FDG.
The editors hope that this book will give a lively impression of the de-
bates currently being conducted about the architecture of FDG. As Anstey
says, now “FG could do with more engineers and builders”. The current
volume can be seen as a blueprint for the work of those labourers.
We hereby acknowledge the financial support provided by the Spanish
Ministry of Science and Technology (Ministerio de Ciencia y Tecnología,
MCYT), the European Regional Development Fund (Fondo Europeo de
Desarrollo Regional, FEDER), and the Xunta de Galicia (XUGA) under
project BFF2002-02441, PGIDIT03PXIC20403PN, entitled "Discourse
Analysis in English: Aspects of Cognition, Typology and Second Language
Acquisition/Análisis del discurso en lengua inglesa: Aspectos cognitivos,
contrastivos y de adquisición", and by the Netherlands Organization for
Scientific Research, for making it possible for Dr Gómez- González to visit
the Netherlands to work on the production of this book.
Preface xi
Our gratitude goes to Susanna Hop MA for her invaluable assistance

with word-processing. Last but not least, we wish to thank the authors of
the various chapters for their contributions, and for their patience as they
awaited the appearance of this book.
J. Lachlan Mackenzie María de los Ángeles Gómez-González

Amsterdam Santiago de Compostela
Abbreviations
A Discourse Act; addressee FPG Functional Procedural

ADS Archetypal Discourse Grammar
Structure FrmFtrs formal features
Ass Assertive sentence GFT General Functional The-
C communicated content; ory
clause GH Givenness Hierarchy
c content GivTop Given Topic
CB Cobuild Corpus GPH Generalized Parallelism
CDS current discourse space Hypothesis
Cl Clause GPSG Generalized Phrase
Config configuration Structure Grammar
ContrFoc Contrastive Focus H hearer
CS Columbia School of Lin- HDPSG Head Driven Phrase
guistics Structure Grammar
D discourse IDCG Incremental Discourse
DCH Discourse Categories Hi- Cognitive Grammar
erarchy IFG Incremental Functional
DECL declarative Grammar
DM Discourse Model IL Interpersonal Level
DRS Discourse Representation ILL Illocution (of the expres-
Structure sion)
DRT Discourse Representation INFRM Informing move
Theory IS Information Structure
D-Topic Discourse Topic ISO International Standardi-
E State of Affairs zation Organization
ECC extraclausal constituent IT Interpersonal Theme
EL Expression Level L actual language
EMT Extended Multiple Lab label
Theme Lcog general cognitive repre-
ER Expression rules sentation language
ET Experiential Theme Lex Lexeme
F Property or Relation LFG Lexical Functional
FDG Functional Discourse Grammar
Grammar LIBMSEC Lancaster IBM Spoken
FG Functional Grammar English Corpus
FncFtrs functional features
xiv Abbreviations
LIPOC Language Independent Ps speaker

Preferred Order of Con- PSp speaker
stituents R Referential Act; relation-
Lkr language of knowledge ship
representation ResTop Resumed Topic
Llog logical language for logi- RL Representational Level
cal reasoning. RP Referential Phrase
Lur language of underlying RST Rhetorical Structure
representations Theory
M move; manner; message S Sentence; speaker
MNLU Grammatical Module of SDRS Segmented Discourse
Natural Language User Representation Structure
NewTop New Topic SET Setting(s)
NLU Natural Language User SFG Systemic Functional
NVC non-verbal Grammar
communication SoA State of Affairs
OSI Open Systems Intercon- ST Situation Theory
nection Reference Model SubCat Subcategorization
p.c. personal communication SubTop Sub Topic
P Participant in Speech T Ascriptive Act; discourse
Event type
p propositional content TT Textual Theme
P1 Predication-internal u utterance
clause-initial position U utterance; discourse unit
P2 External clause-initial UC Underlying Clause
position UR Underlying Representa-
Pa addressee tions
PAddr addressee UtterFoc Utterance Focus
Para Paragraph WCF West Coast Functional-
POS positive ism
Pr Problem X Individual; proposition
PredFoc Predicate Focus x term variable
PrP Predicate Phrase
The Architecture of a Functional Discourse
Grammar
Kees Hengeveld
1. Introduction
Since the beginning of the nineties, a significant part of the research carried
out within the Functional Grammar framework has been directed at the ex-
pansion of Functional Grammar (FG) from a sentence grammar into a
discourse grammar.1 There are several reasons why FG should aim at such
a development. First of all, there are many linguistic phenomena that can
only be explained in terms of units larger than the individual sentence: dis-
course particles, anaphorical chains, narrative verb forms, and many other
aspects of grammar require an analysis which takes the wider linguistic
context into consideration. Secondly, there are many linguistic expressions
which are smaller than the individual sentence, yet function as complete
and independent utterances within the discourse. This requires a conception
of utterances as discourse acts rather than as sentences, as has been shown
in Mackenzie (1998).
Hannay and Bolkestein (1998) argue that the proposals2 which have
been developed aiming at the expansion of FG into a grammar of discourse
represent two different approaches. In the first, the discourse level is cov-
ered by additional hierarchically superordinate layers. This approach,
called the upward layering approach in Hannay and Bolkestein (1998), is
exemplified by Hengeveld (1997) and Moutaouakil (1998). In the second
approach, the discourse level is handled by a separate component, linked to
the grammatical component through an interface. Hannay and Bolkestein
(1998) call this the modular approach, examples of which are Van den
Berg (1998) and Vet (1998).
In this chapter I want to claim that an adequate model of the grammar of
discourse requires the integration of these two approaches, i.e. I will argue
that both the application of extended layering and the recognition of vari-
2 Kees Hengeveld
ous levels of analysis are necessary. The model of a Functional Discourse

Grammar (FDG) presented here is thus both hierarchical and modular. A
major feature of the model is that it works in a top-down fashion, that is,
decisions at higher levels and layers of analysis determine and restrict the
possibilities at lower levels and layers of analysis. This feature of FDG will
be treated first, in Section 2. Section 3 then presents a general outline of the
model, focusing on the various levels of analysis and the complex inter-
faces linking them to one another. The layers to be distinguished at the
various levels are presented in detail in Section 4. Section 5 looks at the
dynamic top-down construction of basic linguistic expressions within FDG
by analyzing a series of illustrative examples. More complex examples
which involve intricate interactions between the various levels of analysis
are discussed in Section 6. The paper is rounded off in Section 7.
2. Top down
In Levelt (1989) the speech production process is described as a top-down
process, running from intention to articulation. His analysis suggests that
the speaker first decides on a communicative purpose, selects the informa-
tion most suitable to achieve this purpose, then encodes this information
grammatically and phonologically and finally moves on to articulation.
Levelt shows that there is ample support in psycholinguistic research for
this conception of speech production.
The speech production model used in FG (Dik 1997a: 60) has a quite
different orientation. It starts out with the selection of predicate frames that
are gradually expanded into larger structures, which when complete are
expressed through expression rules. In view of Levelt’s (and many other
psycholinguists’) findings, this organization of the grammar runs counter to
the standard of psychological adequacy that FG should live up to (Dik
1997a: 13–14).
In the model defended here production is therefore described in terms of
a top-down rather than a bottom-up model. This step, apart from having a
higher degree of psychological adequacy, is crucial to the development of a
grammar of discourse: in a top-down model, the generation of underlying
structures, and in particular the interfaces between the various levels, can
be described in terms of the communicative decisions a speaker takes when
constructing an utterance, as will be illustrated in Sections 5 and 6.
The architecture of a FDG 3
3. Outline of the model

Figure 1 gives the basic outline of FDG. It shows that FDG distinguishes
three interacting levels: the interpersonal level, the representational level,
and the expression level, in that hierarchical order. The presence of these
three levels as separate modules within the model is the major difference
from earlier upward layering proposals.
Layering applies at each level separately, i.e. each level is organized hi-
erarchically, as will be shown in the next section. This hierarchical
organization of the model is the major difference from earlier modular pro-
posals.
Mapping rules link the interpersonal to the representational level, in
those cases in which semantic content is necessary for the transmission of a
certain communicative intention. Expression rules then link the interper-
sonal and representational levels to the expression level. In cases in which
only pragmatic content has to be transmitted, expression rules directly link
the interpersonal to the expression level. The various linking mechanisms
may be interpreted as interfaces which define the possible correspondences
between layers at different levels.
The three levels interact with a cognitive component and with a com-
municative component. The cognitive component represents the (long-
term) knowledge of the speaker, such as his communicative competence,
his knowledge of the world, and his linguistic competence. The speaker
draws on this component at each of the three levels.
The communicative component represents the (short-term) linguistic in-
formation derivable from the preceding discourse and the non-linguistic,
perceptual information derivable from the speech situation. As far as the
linguistic information is concerned, the communicative component is fed
by the interpersonal and expression levels, and feeds the representational
level in order to enable later reference to earlier acts and expressions, as
will be illustrated in Section 6. Of course, short-term information may be
selected for long-term storage and is in that case passed on to the cognitive
component.
4
INTERPERSONAL LEVEL
Kees Hengeveld
Mapping rules
REPRESENTATIONAL LEVEL
Expression rules
CONTEXT
COGNITION
COMMUNICATIVE
EXPRESSION LEVEL
Figure 1. Outline of FDG GRAMMAR

4. Levels and layers
4.1. Introduction
As was mentioned in the previous section, each level of analysis in Figure
1 is organized hierarchically. In this section I will first of all review each of
the levels separately. Then I will present the full model and compare it to
the earlier layered sentence model, as presented in Hengeveld (1989).
4.2. The interpersonal level

The hierarchical structure of the interpersonal level is presented in Figure
2.
(M1: [(A1: [ILL (P1)S (P2)A (C1: [...(T1) (R1)...] (C1))] (A1))] (M1))
Figure 2. The interpersonal level
Note that this representation is non-exhaustive, in the sense that there

are higher levels of discourse organization which are not captured here. In
4.6 I will return to this point.
At the interpersonal level a central unit of analysis is the move (M), de-
fined in Kroon (1995: 66), following Sinclair and Coulthard (1975), as “the
minimal free unit of discourse that is able to enter into an exchange struc-
ture”. As such, the move is the vehicle for the expression of a single
communicative intention of the speaker. Examples of such communicative
intentions are inviting, informing, questioning, threatening, warning, rec-
ommending, etc.
In order to achieve his communicative intention, the speaker executes
one or more discourse acts (A), defined in Kroon (1995: 65) as “the small-
est identifiable units of communicative behaviour”. A move consists of one
central act, which may be supported by one or more subsidiary acts. Every
act may be characterized in terms of its illocution (ILL), by which I here
mean the illocution as coded in the expression3 (Dik 1997b: ch. 11). Illocu-
tions are represented as abstract illocutionary frames,4 which take the
participants (PN) in the discourse act, i.e. the speaker (PS) and the addressee
(PA), and the communicated content (C), i.e. the information transmitted in
the discourse act, as their arguments. In order to build up the communi-
6 Kees Hengeveld
cated content the speaker may have to execute one or more ascriptive acts
(T) and one or more referential acts (R): it is the speaker who refers to enti-
ties by using referring expressions,5 and it is the speaker who ascribes
properties to entities by applying predicates to these referring expressions.6
4.3. The representational level

The hierarchical structure of the representational level is presented in Fig-
ure 3.
(p1: [(e1: [(f1) (x1)] (e1))] (p1))
Figure 3. The representational level
Note that, again, this representation is non-exhaustive. There are higher

levels of semantic organization which are not captured here (see Section
4.6).
In transmitting his communicative intention the speaker will in most
cases have to fill his utterance with basic semantic content, i.e. with de-
scriptions of entities as they occur in the non-linguistic world. These
entities are of different orders: third-order entities or propositional contents
(p); second-order entities or states of affairs (e); first-order entities or indi-
viduals (x); and zero-order entities or properties (f). Within the maximally
hierarchical representation given in Figure 3 the propositional content (p1)
contains the description of a state of affairs (ei), which contains the descrip-
tion of a property (fi) and the description of an individual (xi). Note,
however, that all entity types may also be expressed directly, i.e. non-
hierarchically, through lexical items.
4.4. The expression level

The hierarchical structure of the expression level is presented in Figure 4.
(Para1: [(S1: [(Cl1: [(PrP1: [(Lex1)] (PrP1)) (RP1: [(Lex2)] (RP1))] (Cl1))] (S1))] (Para1))
Figure 4. The expression level
This representation is just a simplified example of what the expression

level might look like. It is an example, since every language has its own
expression possibilities, which lead to different expression units in their

grammars. It is simplified, since the expression possibilities of a single
language are generally much more refined than in the example I have given
here. This simplified example will suffice, however, to illustrate the points
I want to make in this chapter. It is a representation of constituent structure
which starts at the level of the paragraph (Para), which may consist of one
or more sentences (S), each of which may contain one or more clauses
(Cl), which may contain one or more predicate phrases (PrP) and referen-
tial phrases (RP), each of which may contain one or more lexemes (Lex).
It is important to note that the expression level corresponds with what in
Levelt’s production is the product of grammatical and phonological en-
coding. Articulation, the final step, is not a level within the grammar, but
the actual output of the grammar.
4.5. Integration
The levels and layers discussed so far are given in Figure 5. The elements
in boldface at the interpersonal and representational levels in this figure
correspond to units that were present in the layered representation of clause
structure defended in Hengeveld (1989) and its upward-layering elabor-
ation in Hengeveld (1997). The correspondences may be listed as follows:
M Move M Move
E Speech Act A Discourse Act
ILL Illocution ILL Illocution
PN Speech Act Participant PN Discourse Act Participant
X Propositional Content p Propositional Content
e State of Affairs e State of Affairs
x Individual x Individual
f Property or Relation f Property or Relation
Figure 5. Hengeveld (1989/1997)
The major differences at the interpersonal and representational levels,

then, concern the presence of the variables C, T and R at the interpersonal
level. In other words, a major feature characterizing the current proposal is
downward layering at the interpersonal level. I will now briefly discuss
each of these variables in contrast with their representational neighbours.
Examples will follow in Section 5.
The introduction of the variable T at the interpersonal level makes it
possible to distinguish systematically between ascription as an act of the
8 Kees Hengeveld
speaker, and the entity type which is described within this act of ascription.
Often the speaker will use the description of a zero-order entity (f) to give
content to his ascriptive act, but he might also use, for instance, a first-
order entity (x) in a classifying or identifying construction. Similarly, the
variable R allows for a systematic distinction between the act of referring
on the one hand, and the entity type referred to on the other. Frequently the
speaker will use the description of a first-order entity to give content to his
referential act, but reference to other types of entity is equally possible.
The introduction of the variable C opens up a way to distinguish the in-
formation communicated in a discourse act from the nature of the entity
type the description of which is used to transmit that information. As a re-
sult, it is no longer necessary to assume that every discourse act contains a
propositional content, i.e. a third-order entity. In many circumstances it is
sufficient, for instance, to communicate information by simple reference to
a first-order entity.
A further difference between Hengeveld (1989, 1997) and the current
proposal concerns the presence of the Expression Level in the model. The
major motivating factor for the introduction of this level is the existence of
meta-linguistic expressions (Sweetser 1990) or reflexive language (Lucy
1993). This phenomenon will be illustrated in Section 6.
4.6. Upward layering

The previous sections have shown that each level or module in the proposed
model has its own layered structure. Section 4.5 has stressed the relevance of
further downward layering at the interpersonal level. Further upward layer-
ing is necessary too, but will not be dealt with here. A major point, however,
is that upward layering is not restricted to the interpersonal level, but is a fea-
ture of all levels of analysis. Thus, at the interpersonal level there may be
linguistic reasons to distinguish, for instance, between the layers of Turn and
Exchange in dialogues; at the representational level languages may give spe-
cial treatment to the layers of Episode and Story in narratives; and at the
expression level there may be reasons to distinguish layers, for example Sec-
tion and Chapter in written communication. In each case, the possible
mappings of interpersonal to representational to expression categories have
to be determined partly on a language-specific basis. Thus, a Move is
mapped onto an Episode onto a single sentence in narratives in many lan-
guages of Papua New Guinea, whereas it is commonly expressed through a
paragraph in most Western European languages.
(M1:[(A1:[ILL(P1)S(P2)A(C1:[...(T1)(R1)...](C1))](A1))](M1))
Mapping rules
(p1:[(e1:[(f1)(x1)](e1))](p1))
Expression rules
CONTEXT
COGNITION
COMMUNICATIVE
((Para1:[(S1:[(Cl1:[(PrP1:[(Lex1)](PrP1))(RP1:[(Lex2)](RP1))](Cl1))](S1))](Para1)))
The architecture of a FDG
GRAMMAR
9
Figure 6. Levels and layers

10 Kees Hengeveld
5. Making choices at the interpersonal level

Within the model proposed here, the construction of linguistic expressions
can be interpreted as a decision-taking process on the part of the speaker.
This process applies in a left-right and top-down fashion.
Left-right decisions apply at the interpersonal level only. For instance,
only after deciding on the communicative intention Warning for a certain
move may the speaker select, for instance, the appropriate discourse acts
Vocative, and Imperative. Only after deciding on the discourse act Voca-
tive does the necessity to execute a referential act become obvious, etc. The
result might then be something like (1):
(1) George, watch out for that tree!
Top-down decisions are of a more complex nature. These concern the

decisions the speaker makes with respect to (i) the semantic content neces-
sary to successfully execute an interpersonal act, and (ii) the expression
category necessary to successfully transmit his communicative intentions.
In what follows I will restrict myself to some examples of this decision-
making process. In order to facilitate the interpretation of the representa-
tions I use tables in which the various levels correspond with rows: the first
row contains the interpersonal units, the second row the representational
units, and the third row the expression units. Furthermore, I concentrate on
the representation of the relevant part of the example under consideration.
Within the example this part is printed in italics; within the representation
it is separated from the remaining part by means of vertical lines.
Let us first consider a lexical, i.e. ready-made C. If the speaker wants to
express his frustration about the way things are going he may select a lexi-
cal item which serves this purpose directly. An expression such as damn
has pragmatic, not semantic content. Therefore, the speaker may move di-
rectly from the interpersonal to the expression level:
(2) Damn!
(A1: [EXPR (P1)Sp (P2)Addr (C1) ] (A1))

-----
(Lex1)
The expressive illocution takes care of the prosodic contour of this one-
word expression.
The next example concerns a lexical R. If the speaker wants to draw the
attention of someone present in the speech situation he may simply call his
name. Here we have a referential act which makes use of a lexical item
(Lex) which does not have semantic content, but only referential content.
Therefore, the speaker may move directly from the interpersonal to the ex-
pression level again:
(3) John!
(A1: [VOC (P1)Sp (P2)Addr (C1: [ (R1) ] (C1))] (A1))

-----
(Lex1)
Now consider a case in which the speaker draws on the representational

level in order to transmit his communicative intention. The content com-
municated (C) here is the description of a third-order entity (p) expressed
in a clause (Cl):
(4) The Plaza Santa Ana is the best place to go.
(A1: [DECL (P1)Sp (P2)Addr (C1) ] (A1))

(p1)
(Cl1)
The same propositional content (p), expressed as a clause (Cl), may oc-
cur as the vehicle which the speaker uses to execute a referential act (R):
(5) I want to know whether the Plaza Santa Ana is the best place to go.
(A1: [DECL (P1)Sp (P2)Addr (C1: [ (R1) ] (C1))] (A1))

(p1)
(CL1)
This may be contrasted with a case in which a referential act (R) again
refers to a propositional content (p) but is expressed by means of a referen-
tial phrase:
(6) I want to know your opinion.

12 Kees Hengeveld
(A1: [DECL (P1)Sp (P2)Addr (C1: [ (R1) ] (C1))] (A1))

(p1)
(RP1)
Ascriptive acts (T) often make use of the description of a zero-order en-
tity (f) and are then expressed by means of a lexeme (Lex), as in the next
example:
(7) The Plaza Santa Ana is wonderful, don’t you think?
(A1: [DECL (P1)Sp (P2)Addr (C1: [ (T1) ] (C1))] (A1))

(f1)
(Lex1)
But the speaker may also decide on a first-order entity (x), expressed as
a referential phrase (RP) to transmit the same kind of information, as in the
following example:
(8) The Plaza Santa Ana is a wonderful place, don’t you think?
(A1: [DECL (P1)Sp (P2)Addr (C1: [ (T1) ] (C1))] (A1))

(x1)
(RP1)
The examples just given serve to illustrate how linguistic expressions

may be seen as the product of a top-down decision process on the part of
the speaker, with a certain independence, within limits, of the three levels
distinguished within the model.
6. Complex interactions between levels
6.1. Introduction
Let me now turn to more complex interactions between the various levels.
As Figure 1 already showed, the communicative context feeds into the rep-
resentational level. The preceding discourse is of course part of this
communicative context, and units within this discourse may be used for
later reference. This is achieved in the model presented here by having
these units reappear within the representational level. In this way we may
account for constructions like the following:
(9) a. If you behave well, I’ll let you read my poems.

b. Is that a threat or a promise?
(10) My brother-in-law, if that’s the right word for him, is a poet.
In (9b) the demonstrative that refers to the preceding move. In (10) it

refers to the preceding lexeme brother-in-law. These examples thus illus-
trate that elements from both the interpersonal and the expression levels are
available for later reference once they are produced, i.e. they may become
part of the representational level in ensuing communication. Reference to
elements from the interpersonal and expression levels will be studied sepa-
rately below. In order to account for these cases the model presented here
(Figure 7 below) allows for the copying of elements from the interpersonal
and the expression levels to the representational level via the communica-
tive context.
In order to show their different status, variables from both levels are
written with lower-case letters when they are used. In the next sections I
will give some examples of how these variables are used in analyzing a va-
riety of constructions which involve metacommunicative and metalin-
guistic expressions.
14
(M1: [IN (A1: [ILL (P1)S (P2)A (C1: [...(T1) (R1)...] (C1))] (A1))] (M1))
Mapping rules
Kees Hengeveld
(m1: [(a1: [(c1: [...(t1) (r1)... ] (c1))] (a1))] (m1))

(p1: [(e1: [(f1) (x1)] (e1))] (p1))
(para1: [(s1: [(cl1: [(prp1: [(lex1)] (prp1)) (rp1: [(lex2)] (rp1))]
(cl1))] (s1))] (para1))
CONTEXT
COGNITION
Expression rules
COMMUNICATIVE
((Para1:[(S1:[(Cl1:[(PrP1:[(Lex1)](PrP1))(RP1:[(Lex2)](RP1))](Cl1))](S1))](Para1)))
Figure 7. Levels and layers GRAMMAR

6.2. Referents from the interpersonal at the representational level

In this section I will discuss two construction types that can only be prop-
erly understood if we allow units from the interpersonal level to enter the
representational level: hedged performatives and identity statements.
Consider the following examples from Spanish:
(11) a. (Me) temo que Juan esté enfermo.

to.me I.am.afraid that Juan is.SUB ill
‘I'm afraid that Juan is ill.’
b. Me temo que Juan está enfermo.
to.me I.am.afraid that Juan is.IND ill
‘I'm afraid that Juan is ill.’
In (11a) the speaker simply expresses his state of mind. The embedded
clause, in which the subjunctive is used, represents what he fears might be
the case. In (11b) the indicative is used in the embedded clause. This sen-
tence, unlike (11a), is an example of a so-called hedged performative
(Fraser 1975), in which the embedded clause represents the actual informa-
tion the speaker wants to transmit, but which he ‘hedges’ since he thinks
the addressee might not like what he has to say. In cases like these the ac-
tual communicated content (c) is hidden in the embedded clause (Cl),
which itself is the expression of a referential act (R), so that this sentence
may be represented as follows:
(A1: [DECL (P1)Sp (P2)Addr (C1: [... (R1) ...] (C1))] (A1))
(c2)
(CL1)
A second case in which units from the interpersonal level figure at the
representational level concerns so-called identity statements (Declerck
1988, Hengeveld 1992, Keizer 1992) as illustrated in (12):
(12) The Morning Star ís the Evening Star.
Sentences like (12), with a prosodically prominent copula, serve the pur-
pose of stating that the act of referring to an object by using a certain name
is equivalent to the act of referring to that same object by another name;
hence they are statements about the validity of acts of reference. Therefore
the representation of e.g. the Morning Star may be as follows:
16 Kees Hengeveld
(r2)
(RP1)
This representation states that in (12) the speaker executes a referential act
by making reference to a referential act (r), which is expressed as a referen-
tial phrase (RP).
6.3. Referents from the expression level at the representational level

Units from the expression level may enter the representational level as well. I
will discuss two cases here: metalinguistic conditionals and direct speech.
An example of a metalinguistic conditional is given in (13):
(13) This concert, if you want to call it that, isn’t exactly what I was waiting for.
The word that refers to the preceding word concert, which is a case of ref-
erence to the code rather than to the message. Thus there is a referential act
(R) in which reference is made to a lexeme (lex) which is expressed as a
lexeme (Lex), as indicated in the following representation:
(lex1)
(Lex2)
A second case in which reference is to the code rather than to the mes-
sage concerns direct speech. Reporting direct speech may be interpreted as
a form of mimicry (Clark and Gerrig 1990), where direct speech can be
seen as imitated code. This is evident from the fact that direct speech re-
ports respect the original language and/or dialect, as in:
(14) He said: “¿Cómo estás?”.
and that direct speech reports may contain meaningless noise, as in:
(15) He said: “gagugagugagu”.
The latter example furthermore shows that the imitated code can be any
part of the expression level.
In (14) the speaker refers (R) in the second argument of the verb say to
a previous sentence (s) which in the actual expression is repeated through
imitation. This example may thus be represented as follows:
(s1)
(S1)
7. Conclusion
In this chapter I have presented a basic outline of FDG and illustrated its
appropriateness by analyzing a number of construction types that would
have been difficult to handle in earlier versions of FG. Many aspects of
FDG require further elaboration. These aspects can be grouped together
into five categories:
(a) What are the restrictions on left-right decisions within the production
process, i.e. what are the systematic restrictions on the internal con-
stitution of the interpersonal level?
(b) What are the restrictions on top-down decisions within the produc-
tion process, i.e. what do the interfaces between the three levels of
grammar look like?
(c) What is the internal structure of the cognitive component and how
does it interact with the three levels of grammar?
(d) What is the internal structure of the contextual component, particu-
larly with respect to the representation of the non-linguistic context,
and how does it interact with the three levels of grammar?
(e) None of these questions is new to FG. I hope that the model of FDG
presented here will provide the basis for an integrated approach to
these central issues in linguistic theory.
Notes
1. This chapter is the product of long and lively discussions with a great num-
ber of people. The Amsterdam FG-DISCO group has met at irregular
intervals over the last few years, and has been a very inspiring environment
for discussion of the topics dealt with in this chapter. I am indebted to the
members of this group, Machtelt Bolkestein, Mike Hannay, Caroline Kroon,
Lachlan Mackenzie, Rodie Risselada, and Co Vet, for the many open-
minded and inspiring discussions we have had. A special word of thanks
goes to Mike Hannay, for a revival of the group’s activities when the time
was there, and to Lachlan Mackenzie for joining this revival. Outside the
FG-DISCO group, and extending over the same period, I have had countless
18 Kees Hengeveld
discussions with Gerry Wanders about the topic of this chapter, in which
she has manifested herself as a critical and generous sparring partner. I am
grateful to her for her help. Given the extensive interaction with all of these
colleagues over a long period of time, it is hard to do justice to their indi-
vidual contributions to the contents of this chapter. As a result, I do not
want to claim originality for many of the ideas presented in this chapter,
only for the way these are put together.
2. See Van den Berg (1998), Connolly (1998), Connolly et al. (1997b),
Crevels (1998), Gómez Soliño (1996), Hengeveld (1997), Jadir (1998),
Kroon (1997), Liedtke (1998), Mackenzie (1998, 2000), Moutaouakil
(1998), Rijkhoff (1995), Steuten (1997, 1998), Vet (1998).
3. I take a broad view of coded illocution here, in that among the encoding
possibilities I include not only sentences types, but also prosodic encoding,
morphological encoding, and conventionalized lexicalization patterns.
4. This slot may alternatively be occupied by a performatively used speech-act
verb.
5. Cf. Lyons (1977: 177): “... the speaker ... invests the expression with refer-
ence by the act of referring”.
6. Cf. Lyons (1977: 161): “For example, in saying of a particular flower that it
is red, we ascribe to it the property of redness, but we predicate of it the
predicate ‘red’”.
References
Berg, Marinus E. van den
1998 An outline of a pragmatic functional grammar. In: Mike Hannay and
A. Machtelt Bolkestein (eds), 77–106.
Clark, Herbert H. and Richard J. Gerrig
1990 Quotations as demonstrations. Language 66: 764–805.
Connolly, John H.
1998 Information, Situation Semantics and Functional Grammar. In: Mike
Hannay and A. Machtelt Bolkestein (eds), 167–190.
Connolly, John H., Anthony A. Clarke, Steven W. Garner and Hilary K. Palmen
1997a A functionally oriented analysis of spoken dialogue between indi-
viduals linked up by a computer network. In: John H. Connolly, Roel
M. Vismans, Christopher S. Butler and Richard A. Gatward (eds),
33–58.
Connolly, John H., Roel M. Vismans, Christopher S. Butler and Richard A. Gat-
ward (eds)
1997b Discourse and Pragmatics in Functional Grammar. Berlin and New
York: Mouton de Gruyter.
Crevels, Mily
1998 Concession in Spanish. In: Mike Hannay and A. Machtelt Bolkestein
(eds), 129–148.
Declerck, Renaat
1988 Studies on Copular Sentences, Clefts and Pseudo-clefts. Leuven:
Leuven University Press and Dordrecht: Foris.
Dik, Simon C.
1997a The Theory of Functional Grammar. Part I: The Structure of the
Clause (K. Hengeveld, ed.) [2d rev.]. Berlin and New York: Mouton
de Gruyter.
1997b The Theory of Functional Grammar. Part II: Complex and Derived
Structures (K. Hengeveld, ed.). Berlin and New York: Mouton de
Gruyter.
Fraser, Bruce
1975 Hedged performatives. In: Peter Cole and Jerry L. Morgan (eds),
Speech Acts (Syntax and Semantics 3). New York: Academic Press.
Gómez Soliño, José S.
1996 La organización jerárquica de los textos desde una perspectiva fun-
cional. In: M. Teresa Caneda Cabrera and Javier Pérez Guerra (eds),
Os Estudios Ingleses no Contexto das Novas Tendencias. Vigo: Uni-
versidade de Vigo.
Hannay, Mike and Machtelt Bolkestein
1998a Introduction. In: Mike Hannay and A. Machtelt Bolkestein (eds),
vii–xii.
Hannay, Mike and Machtelt Bolkestein (eds)
1998b Functional Grammar and Verbal Interaction. Amsterdam and Phila-
delphia: Benjamins.
Hengeveld, Kees
1989 Layers and operators in Functional Grammar. Journal of Linguistics
25: 127–157.
1992 Non-verbal Predication: Theory, Typology, Diachrony. Berlin and
New York: Mouton de Gruyter.
1997 Cohesion in Functional Grammar. In: John H. Connolly, Roel M.
Vismans, Christopher S. Butler and Richard A. Gatward (eds), 1–16.
Jadir, Mohammed
1998 Textual cohesion and the notion of perception. In: Mike Hannay and
Keizer, M. Evelien
1992 Reference, Predication and (In)definiteness in Functional Grammar.
Dissertation, Free University Amsterdam.
20 Kees Hengeveld
Kroon, Caroline
1995 Discourse Particles in Latin: A Study of nam, enim, autem, vero and
at. Amsterdam: Gieben.
1997 Discourse markers, discourse structure and Functional Grammar. In:
John H. Connolly, Roel M. Vismans, Christopher S. Butler and
Richard A. Gatward (eds), 17–32.
Levelt, Willem J. M.
1989 Speaking: From Intention to Articulation. Cambridge, MA: MIT
Press.
Liedtke, Frank
1998 Illocution and grammar: a double level approach. In: Mike Hannay
and A. Machtelt Bolkestein (eds), 107–128.
Lucy, John A. (ed.)
1993 Reflexive Language: Reported Speech and Metapragmatics. Cam-
bridge: Cambridge University Press.
Lyons, John
1977 Semantics. 2 Vols. Cambridge: Cambridge University Press.
Mackenzie, J. Lachlan
1998 The basis of syntax in the holophrase. In: Mike Hannay and A.
Machtelt Bolkestein (eds), 267–296.
2000 First things first: towards an Incremental Functional Grammar. Acta
Linguistica Hafniensia 32: 23–44.
Moutaouakil, Ahmed
1998 Benveniste's récit and discours as discourse operators in Functional
Grammar. In: Mike Hannay and A. Machtelt Bolkestein (eds), 25–
42.
Rijkhoff, Jan
1995 Bystander and social deixis: some programmatic remarks on the
grammar/pragmatics interface. Working Papers in Functional
Grammar 58. Amsterdam: University of Amsterdam.
Sinclair, John M. and Richard M. Coulthard
1975 Towards an Analysis of Discourse: The English Used by Teachers
and Pupils. London: Oxford University Press.
Steuten, Ans A. G.
1997 Business conversations from a conversation analytical and a
Funcytional Grammar perspective. In: Connolly et al. (eds), 59–73.
1998 Structure and coherence in business conversations. In: Mike Hannay
Sweetser, Eve E.
1990 From etymology to pragmatics: metaphorical and cultural aspects of
semantic structure. Cambridge: Cambridge University Press.
Vet, Co
1998 The multilayered structure of the utterance: about illocution, modal-
ity and discourse moves. In: Mike Hannay and A. Machtelt
Bolkestein (eds), 1–24.
Functional Grammar from its inception
Matthew P. Anstey
1. Introduction
Hengeveld’s chapter (this volume) was written with two purposes in mind.1
On the one hand, it is the latest attempt within Functional Grammar to for-
mulate an explicit model for the detailing of natural language phenomena.
On the other, it is a manifesto for those working within the Functional
Grammar movement, a framework wherein scholars may locate their contri-
bution to the general theory. This new architecture, therefore, is not only a
design for a theoretical ‘space’. It is also, metaphorically speaking, a design
for a collective, academic space, a conceptual meeting-place for those work-
ing in the many divergent subdisciplines of the linguistic sciences who wish
to identify themselves to some degree with Functional Grammar.
Accordingly, this chapter is an evaluation of Hengeveld’s proposal in
light of these two purposes, as seen from the perspective of the evolution of
Functional Grammar. Three important questions will be addressed. As a
framework for Functional Grammar, what is its relationship to previous
models? Does it address the objections raised against previous models? As a
framework for Functional Grammarians, does it create sufficient space for
the Functional Grammar academic community to continue to share one roof?
2. Preliminaries
2.1. Chronological Overview

Clearly, such a historical overview needs to be highly selective. I have
concentrated on the core of Functional Grammar, organizing the material
24 Matthew P. Anstey
into five periods. Each FGn refers not only to the period but also to its ca-
nonical FG publication. FG4, however, refers to both the period and
Hengeveld’s Functional Discourse Grammar.
(a) FG0 covers the period prior to 1978, the central work being Dik’s
dissertation (1968) on coordination;
(b) FG1 covers 1978 to 1989, beginning with the publication of Func-
tional Grammar (Dik 1978a);
(c) FG2 covers 1989 to 1997, beginning with the publication of The The-
ory of Functional Grammar (Dik 1989b);
(d) FG3 covers 1997 to September 2000, beginning with the publication
of the two volumes of The Theory of Functional Grammar (Dik
1997a; 1997b);
(e) FG4 extends from the ninth International Conference on Functional
Grammar (Madrid, Spain) onwards, where Hengeveld’s (2000)
Functional Discourse Grammar was first presented.
Figures 1 and 2 provide an overview of these five periods.2
Total Publications
120
100
80
60
40
20
0
78
80
82
84
86
88
90
92
94
96
98
Figure 1. Functional Grammar publications from 1978 to 19983

FG from its inception 25
FG0 Dik 1968. Coordination

1968-1978 Dik 1977. Stepwise lexical decomposition
Dik 1978. Functional Grammar
Dik 1980. Studies in Functional Grammar
Hoekstra et al. 1980. Special edition on Functional Grammar. Glot 3(3–4)
Bolkestein et al. 1981. Predication and expression in Functional Grammar
Martín Mingorance 1981. Translation of Dik 1978 into Spanish
’82: Functional Grammar workshop during 13th Int. Congress of Linguists, Tokyo, Japan
Dik 1983. Advances in Functional Grammar
FG1 ’84: 1st Functional Grammar conference, Amsterdam, The Netherlands
1978-1989 ’85: Functional Grammar symposium on predicate operators, Amsterdam, The Netherlands
Bolkestein et al. 1985. Syntax and pragmatics in Functional Grammar
Bolkestein et al. 1985. Predicates and terms in Functional Grammar
’86: 2nd Functional Grammar conference, Antwerp, Belgium
’87: Functional Grammar symposium on the computer, Amsterdam, The Netherlands
Nuyts and de Schutter 1987. Getting one's words into line
van der Auwera and Goossens 1987. Ins and outs of the predication
’88: 3rd Functional Grammar conference, Amsterdam, The Netherlands
Dik 1989. The theory of Functional Grammar
Moutaouakil 1989. Pragmatic functions in a Functional Grammar of Arabic
’90: 4th Functional Grammar conference, Copenhagen, Denmark
Nuyts et al. 1990. Layers and levels of representation in language theory
Hannay and Vester 1990. Working with Functional Grammar: descriptive and computa-
tional applications
Siewierska 1991. Functional Grammar
’92: 5th Functional Grammar conference, Antwerp, Belgium
Keizer 1992. Reference, predication and (in)definiteness in Functional Grammar
Rijkhoff 1992. The noun phrase: a typological study of its form and structure
FG2
Fortescue et al. 1992. Layered structure and reference in a functional perspective
1989-1997
’94: 6th Functional Grammar conference, York, England
Engberg-Pedersen et al. 1994. Function and expression in Functional Grammar
’95: Simon Dik dies after three years of illness
Kroon 1995. Discourse particles in Latin
’96: 7th Functional Grammar conference, Córdoba, Spain
Devriendt et al. 1996. Complex structures. A functionalist perspective
Engberg-Pedersen et al. 1996. Content, expression and structure: Studies in Danish Func-
tional Grammar
Butler et al. 1997. A fund of ideas: recent developments in Functional Grammar
Connolly et al. 1997. Discourse and pragmatics in Functional Grammar
Dik 1997. The theory of FG. Part I: the structure of the clause
Dik 1997. The theory of FG. Part II: complex and derived structures
’98: 8th Functional Grammar conference, Amsterdam, The Netherlands
FG3
Olbertz et al. 1998. The structure of the lexicon in Functional Grammar
1997-2000
Hannay and Bolkestein 1998. Functional Grammar and verbal interaction
Faber and Usón 1999. Constructing a lexicon of English verbs
Martín Arista et al. 1999. Nuevas perspectivas en Gramática Funcional
’00: 9th Functional Grammar conference, Madrid, Spain
Mackenzie 2000. First things first. Towards an Incremental Functional Grammar
FG4
Pérez Quintero 2001. Special edition of Revista Canaria de Estudios Ingleses. Challenges
2000
and Developments in Functional Grammar.
onwards
’02: 10th Functional Grammar conference, Amsterdam, The Netherlands
Mackenzie and Gómez-González 2003. A new architecture for Functional Grammar
Figure 2. Historical Overview of Functional Grammar

2.2. Central Problems

In addition to the evolution of FG, we wish to pay attention to several re-
current problems that emerge. These provide various perspectives from
which to evaluate Hengeveld’s proposal. It is beneficial to articulate them
at this stage.4
(a) PR1. The problem of structure – what place, if any, does constituent
structure have in the grammar?
(b) PR2. The problem of underlying representations – what do the un-
derlying representations (URs) actually represent and how should
they be interpreted?
(c) PR3. The problem of verbal interaction – how does FG relate to
communication as process?
(d) PR4. The problem of functional primitives – how are the primitives
defined and applied? How many are needed?
(e) PR5. The problem of discourse – how does FG account for linguistic
phenomena beyond the sentence level?
(f) PR6. The problem of psychological adequacy – what does psycho-
logical adequacy mean and does FG fulfil it?
(g) PR7. The problem of formalization – how should the notation for-
mally and explicitly represent language structure?
3. FG0 - prior to 1978
3.1. Publications prior to FG0

Two of Simon Dik’s earliest writings provide helpful background to FG0.
Firstly, Dik (1966: 406; cf. 1967b) indicates that his first disagreement
with generative-transformational grammar concerned the strict division be-
tween grammar and semantics. For example, the words manner, many,
manifest, manly, manifold and so forth all contain the string ‘man’ but such
a purely formal observation “can never tell us which of these [recurrences]
are grammatically significant and which are not”. Therefore, some seman-
tic criterion is necessary to determine this. Dik concludes that “no
grammatically relevant element can be found in complete independence of
meaning”. Nevertheless, he maintains, “[w]ithout being committed to rec-
ognize the exclusive rights to the title of ‘grammar’ claimed for it, one
should not underrate the value of the contribution that especially the trans-
formational component of generative grammar can make to general

linguistic theory” (Dik 1966: 410).
Secondly, the only published evidence of Dik’s familiarity with Prague
School linguistics is in a book review wherein he first mentions “function-
alism” (Dik 1967a). Dik approves of Jakobson’s view of the ‘means-ends
model’ of language, writing “[t]his means that each fact of language is
evaluated not only with respect to the system as a whole, but also with re-
spect to the ultimate function it fulfils in the larger setting of extra-
linguistic reality” (Dik 1967a: 87). These two basic meanings of ‘func-
tional’ appear in all of Dik’s writings.
Of more interest is Dik’s criticism of the Prague School’s theme-rheme
analysis. He argues that their analysis is “arrived at impressionistically”
and lacks scientific rigour. “In the first place”, he writes, “it is rather mis-
leading to subsume it under syntax and treat it as a part of grammar.
Rather, it should be regarded as a phenomenon depending on the interpre-
tation of the utterance in context and situation and thus already
presupposing the grammatical (and the semantic) structure of the utterance.
In the second place, if the distinction is really to be workable, there should
be some principled basis for finding theme and rheme in any given sen-
tence” (Dik 1967a: 86). This is the view now widely advocated in
Functional Grammar (cf. Bolkestein 1998).
3.2. The theory

Dik presents (proto-)Functional Grammar (FG0) as a chapter in his 1968
dissertation on coordination. He was motivated to provide an explanation
of coordination that could overcome the problems in the approach advo-
cated by generative grammar. Moreover, Dik thinks that the problems are
not just with how generative grammar explains coordination but with the
theory itself. Although Dik provides nine points of divergence of FG0 from
“any version of ‘restricted’ constituent structure grammar” (1968: 199),
two are central and will come to characterize all FG models. These are the
“independent introduction of grammatical functions” and a “different no-
tion of ‘derivation’”.
Dik’s justification of grammatical functions as being basic to linguistic
description consistently starts from the observation that formally equivalent
constituents do not have equivalent linguistic properties. Therefore, con-
stituents must have additional functional properties. Dik does not claim
that this insight is original and is happy to assume Longacre’s (1965: 65)
definition: “By function is meant the particular office or role of one distin-
guishable part of a construction type in relation to other parts of the same

construction”.5
What precisely then is a ‘grammatical function’? Dik writes that “…
grammatical functions are irreducible aspects of grammatical structure,
which can be partly correlated to formal features …, but cannot possibly be
completely reduced to these” (1968: 154). His argument is essentially one
of explanatory power: if we assume that a grammatical system has ‘gram-
matical functions’, then such a system provides a superior account of
linguistic data. However, the axiom of functional primitives is itself not
verifiable.6
It is clear, moreover, that psychological adequacy is already a motif in
his thinking: “And we can explain the general features of human language
if we can show them to be conditioned by general features of the human
constitution, in particular, human mental and physiological properties. The
latter kind of explanation, however, is not as such within the competence of
the linguist, but should rather be explored by close collaboration between
linguists and psychologists” (Dik 1968: 11–12).
Dik’s “different notion of ‘derivation’” is the forerunner to the ‘avoid
transformations’ constraint found in FG1 onwards. “The basic (negative)
property of a functional grammar”, he writes, “… is that it does not include
transformational rules” (1968: 163). To achieve this Dik posits a
monostratal underlying representation of the functional, categorial, and
constitutional elements of each linguistic expression.
More specifically, FG0 has three basic components: constituents (that is,
lexical categories) such as verb, noun, adjective, article; categories such as
noun, pronoun, noun phrase; and functions such as subject, object, predica-
tor, modifier, indirect object, question. How then does one select the
necessary functions to describe a language? Dik provides the following
stipulation, anticipating in many respects the central philosophy of FG1-4:
“At least so many functions are set up for a language as there are gram-
matical (not semantic) differences between the linguistic expressions of
that language which cannot be correlated to differences in constituency
and/or in categorization” (1968: 176).
FG0 distinguishes four sets of rules that relate constituents, categories,
and functions together to form an ‘independent linguistic expression’. The
application of these rules results in a single UR that combines functional
(CAPITALS), categorial (normal) and constituent (italic) elements. For ex-
ample, the sentence (1) has the underlying structure (2) (Dik 1968: 198):7
(1) The man came.

(2) ile(s (sdecl (SUBJ (np (npsg (DET(art(the) )+HEAD (nsg (man) ) ) ) )+PRED
(fv (fvintr (fvintr past (fvintr past 3d ps sg(came) ) ) ) ) ) ) )
Dik considers the various possible relationships between semantics and

grammar and concludes, along the lines of Weinrich (1966), “[i]n my opin-
ion these various developments point to the possibility that [grammar is
dependent on semantics] might finally prove to the right one” (1968: 293).
This revealing quotation underlines the grammatical nature of the functions
Dik posits. “Perhaps we could even go further”, he continues, “and include
the full semantic description of a linguistic expression in its grammatical
specification”. These are, however, “speculations which have little more
than a programmatic value. They might, however, provide interesting pos-
sibilities for a further elaboration of the theory of functional grammar”
(1968: 293).
Thus, for Dik FG0 is an exclusively grammatical, non-semantic system,
yet it includes ‘functional primitives’. The semantics of a language is a dif-
ferent matter. This bracketing out of semantics from the system of
grammar will change in subsequent FG models, where the UR is inter-
preted as conveying grammar and meaning.
Matthews’s (1969: 350) review of Dik’s dissertation is noteworthy. He
views the dissertation positively but has grave doubts about FG0. Although
modelled after tagmemics,8 FG0, unlike tagmemics, has no primitive that
corresponds to syntagmeme (or to Halliday’s structure). Matthews takes
this oversight as problematic to the whole concept of FG0: “But until he
does so, his present formalization is literally not worth a moment’s consid-
eration” (1969: 358). In other words, this is the first sign of the problem of
structure (PR1). Matthews’s observation boils down to asserting that the
URs of FG0 contain insufficient information to generate (via expression
rules) the correct surface form. Bakker and Siewierska (2000), over thirty
years later, conclusively demonstrate that this problem remains; but see
Bakker and Siewierska (this volume).9
4. Dik’s writings between 1968 and 1978

The “interesting possibilities” that a semantically based grammar could of-
fer occupied Dik for the rest of his life. In particular, the decade after his
dissertation was spent on eclectically combining various components to as-
semble FG1. Three of his publications in this period are illustrative.
Dik (1973b: 838) first refers to the predicate frame and operators when
he writes, “In spite of differences in formulation and elaboration, many

semantic theorists agree in that the basic form of semantic representations
should look like the structure given.”
prop
pred arg 1 arg 2 arg 3
GIV E JOH N MAR Y BO OK
Figure 3. Dik’s first predicate frame
Dik also examines briefly the possible basis for universal semantic
categories. He rejects Chomsky’s ‘Innateness Hypothesis’ as a candidate
and prefers what he calls the ‘Means-End Hypothesis’ (presumably from
Jakobson 1963; see Dik 1967a) whereby language is an instrument of so-
cial interaction. The exploration of ‘communicative competence’10 across
various cultures could hopefully lead to the positing of universal semantic
categories.
Dik (1975) provides the theoretical background to FG1’s States of Af-
fairs typology. He adopts two criteria for distinguishing states of affairs:
dynamism and control. For the first time Dik introduces explicit semantic
functions into underlying structures, using the variable ‘M’ for ‘in the
manner of’. Thus he suggests that sentence (3) is “correctly expressed” as
(4):
(3) Annette dances beautifully.

(4) s1(dance(Annette))s1 and beautiful(Ms1)
In 1977 Dik delivered a paper on stepwise lexical decomposition (Dik

1977). The central idea is that meaning definitions of predicates are given
by a subset of simpler predicates from the lexicon of that same language
(cf. Faber and Mairal Usón 1999: 58–65).
5. FG1 - 1978 to 1989
5.1. The Theoretical Framework

The first monograph devoted entirely to Functional Grammar (Dik 1978a)
appeared ten years after Dik’s dissertation. Dik writes:
In this book I develop a theory of Functional Grammar with the following

main distinguishing properties: (i) it is conceived from a functional point of
view on the nature of language: that is, one in which a language is regarded
as an instrument of social interaction; (ii) it makes crucial use of functional
notions of three different levels: the semantic, the syntactic, and the prag-
matic levels; …
Consequently, the functional paradigm (1978a: 1) must deal with two

rule systems: those that govern verbal interaction, ‘pragmatic rules’, and
those that govern the formation of linguistic expressions, ‘semantic,
syntactic, and phonological rules’. Both sets are “social in nature” (1978a:
1) with the linguistic rules being “instrumental with respect to” the verbal
interaction rules (1978a: 2).
We therefore discern three notions of functionalism: language as a
whole is functional with respect to social interaction (the functional para-
digm); individual linguistic expressions within a language are functional
with respect to their uses in particular instances of verbal interaction; and
the individual components of linguistic expressions are functional with re-
spect to other components in the expression (cf. Nichols 1984).
Dik (1978a: 6–9) argues that in addition to descriptive adequacy as
found in generative grammar, FG1 adheres to three standards of explana-
tory adequacy: pragmatic, psychological, and typological (cf. Butler 1991).
He writes (1978a: 6) that “[w]e want a Functional Grammar to reveal those
properties of linguistic expressions which are relevant to the manner in
which they are used, and to do this in such a way that they can be related to
a description of the rules governing verbal interaction”. Unfortunately, to
date there has been little formalization of such pragmatic ‘rules’. Inadver-
tently, Dik created the problem of verbal interaction (PR3) by setting an
agenda for Functional Grammar that was never satisfied.
Dik (1978a: 15) introduces three basic categories – nouns, verbs, and
adjectives – which he describes as “basic predicates along the lines of Bach
(1968)”.11 These three types of lexical predicates are embedded in a predi-
cation containing a predicate frame and optional satellites. The central
concept of FG1 is essentially the combining of semantic, syntactic and

pragmatic functions with predicate arguments.
The semantic functions “are somewhat reminiscent of those distin-
guished in Fillmore’s (1968) Case Grammar” (Dik 1978a: 31). Fillmore
writes that “[t]he case notions comprise a set of universal, presumably in-
nate, concepts which identify certain types of judgements human beings
are capable of making about the events that are going on around them…”
(1968: 24). Dik (1978a: 39) is quite circumspect about semantic functions:
“… the above proposal must remain tentative. … it may turn out that the
distinctions made above are not, after all, as relevant as I think them to be.
Much more research will be needed for a satisfactory solution of this ques-
tion to be found.” The problem of primitives (PR4) is inherent in any
grammar that has primitives, particularly semantic ones.
The two syntactic functions, Subject and Object, are considered ‘primi-
tive[s]’ of ‘grammatical relations’, in accord (partially) with Relational
Grammar (Dik 1978a: 114; Fillmore 1968; Johnson 1977).
Dik provides barely any references in his chapter on pragmatic func-
tions, seeing the prevailing research as containing “much difference of
opinion and much terminological confusion” (1978a: 129). Gebruers
(1983; cf. Piťha 1985) helpfully observes that Bühler was the intellectual
precursor both to Mathesius, founder of the Prague Linguistic Circle, and
to Reichling, Dik’s doctoral supervisor. This connection possibly explains
FG1’s (pragmatic) similarity to Prague.12
5.2. The Formal Notation

The notation of FG1 has four components: the predicate frame, the term
structure, the predication structure, and the ‘Outline of Functional Gram-
mar’.
The structure of the predicate frame is not given abstractly, but it may
be clearly ascertained to be as follows (Dik 1978a: 29, 48; cf. Dik 1978c).13
(5) [predicatetype (x1: <s.r.>)SF … (xn: <s.r.>)SF]SoA where n ≤ 4
The notation of the term structure is (1978a: 16; influenced by Dahl

1971):
(6) (ω xi : φ1 (xi) : φ2 (xi): … : φn (xi) )

In this case, ω covers term operators, ωxi defines the universe of poten-
tial referents, restricted firstly by the subset of which φ1 (xi), the Head, is
true. This subset is then restricted by a subsequent subset of which the
Modifier φ2 (xi) is true, and so forth.
The predication is as follows:
{ [ φ (x1) (x2) ... (xn)] (y1) (y2) ... (yn) }
predicate arguments satellites
terms
nuclear predication
extended predication
Figure 4. The predication structure in FG1 (Dik 1978a: 26)
The overall model of FG is as follows:
F U N D
P R E D I C A T IO N
C O N S T R U C T IO N
P R E D I C A T IO N S
E X P R E S S IO N
R U L ES
L IN G U IS T IC
E X P R ES S IO N S
Figure 5. The overall layout of FG1 (from Dik 1980b: 2)
One could ask what the four arrows represent in the picture. Is this a pro-
ductive model that mirrors a Speaker’s process, in which case the arrows
represent the transition from intention to articulation? Alternatively, is this

a procedural model? If so, of what procedure – the steps a computer simu-
lation may go through in mimicking a language? That this same model
gives rise to two different interpretations of Functional Grammar signals
the beginning of the problem of psychological adequacy (PR6): how does
such a standard apply to the model? To illustrate these different interpreta-
tions compare Van der Auwera (1983: 437 fn.5) who comments: “Another,
rather implausible assumption is that the order of the FG formation would
reflect the mental processes a speaker has to go through while forming a
sentence.” Fortescue (1985: 113), on the other hand, writes that “FG seems
to me interpretable as a model of real-time sentence production (this is of
course not the only way it can be usefully envisaged)”.
Finally, the formal notation intrinsically poses the problem of formal-
ization (PR7), since it aims for explicitness and testability (see Bakker
1994; Hengeveld 1999). This problem became acute in the late mid-
nineties when a plethora of notational variants competed for superiority.
5.3. Reviews
Having outlined the salient points of FG1, we are now in a position to con-
sider briefly the reviews it received, most of which were very positive.
Comrie (1979: 275) is therefore representative when he writes that “[t]hese
critical comments should not detract from the very positive promises that
Functional Grammar holds out” (see also Comrie 1980; Dik 1980a). Most
reviewers, however, noted the incompleteness of the theory and were
“looking forward” to its explication. Almost all of the problems identified
so far are observed by various people.
Hymes (1979: 306), who clearly influenced Dik, notes the problem of ver-
bal interaction (PR3): “It might be fair to say that Dik understands FG to be
preferable as a way of analysing grammar as communicative means, but leaves
the analysis of communicative ends, and the linkage between means and ends,
to others, or for another time” (cf. Prideaux 1981; Hymes 1983).
Bauer (1980: 52) picks up on the problem of primitives (PR4): “Objec-
tions of this kind – especially the difficulty of adequately motivating
assignments [of semantic/syntactic functions] – were probably the most vi-
tal factor behind the flagging interest in case grammar, and it seems to me
that in this respect, FG is no further ahead. Its viability depends on solu-
tions to these problems”.
Piťha (1980: 266) concentrates on the problem of underlying represen-
tations (PR2): “More problematic is the author’s term ‘semantic,’ denoting
not only the semantics of nuclear predications, but also the semantics of
fully developed linguistic expressions in their surface realizations. It is not
clear whether the author … works with two notions: the semantics of
predicate frames and the semantics of surface structures.”14
Watters (1980: 166) observes the problem of structure (PR1): “Perhaps
the most glaring weakness is the lack of any extended discussion about the
place and role of constituent structure and categorial information. Much of
this is left implicit in the discussion of term formation and the ordering of
constituents”.
5.4. Developments of FG1

FG1 sparked an enormous amount of interest, propelled along by Simon
Dik’s charismatic leadership. In the next ten years, many publications were
eventually to become assimilated into FG2, usually with some modifica-
tions by Dik. I have grouped them thematically.
5.4.1. The problem of structure (PR1)
Dik never provides an explanation of how he constructed FG1 from the

various components he appropriated and modified. But to understand the
problem of structure (PR1) in contemporary FG work we need to consider
why constituent structure seems to be absent from FG1. I submit the
following ‘reconstruction’.15
The critical objective for Dik was to find a way for the URs of his
grammatical theory to be independent of autonomous syntax. His approach
was quite inventive. Firstly, he simply removed all references to constitu-
ent structure (such as noun phrase, head, and so forth) from the URs,
reducing them to lexical items arranged in a combination of functional
relationships – semantic, syntactic,16 and pragmatic relationships.
Secondly, he ‘hides’ this stranded structural information in three sepa-
rate locations, employing three strategies as follows:
(a) individuation – the predicate frames, which are stored in the lexicon,
assign to every verb in a language a specific number of argument
slots, each slot being specified for a semantic function, thereby obvi-
ating the need for an abstract (VP-type) structure in the UR;
(b) semanticization – the internal dependency relations of the compo-
nents of noun phrases are isomorphically instantiated in the term
structure as semantic relations, thereby eliminating the need for an

abstract NP structure;17
(c) deferment – sentential and clause structure, including linearization
information, is reified into word-order templates, which, although
‘existing’ per se, are prohibited from contributing their structural in-
formation until after the UR is fully specified with functional
information.
In other words, nouns and adjectives are assembled into ‘term frames’
(that is, term structures), which are assembled into verb-bearing ‘predicate
frames’, which are assembled (with other ‘term frames’ as satellites) into
‘predication frames’ (that is, URs), which are mapped onto ‘sentence
frames’ (that is, word-order templates). The genius of Functional Gram-
mar is that this approach creates the illusion, in my opinion, that there is
no autonomous constituent structure, when in fact the four frames have an
a priori status in the model.18 The presence of covert structure in the
model explains FG’s ability to describe certain (often distributionally
dominant) types of language data and its difficulties with other types of
language data.
For the sake of argument, however, let us assume that my ‘reconstruc-
tion’ is incorrect and that there is genuinely no constituent structure in
Functional Grammar. In this case, the success of Dik’s approach in de-
scribing linguistic data raises the obvious question: why does it work at all?
Why does Functional Grammar demonstrate that (a large majority of) syn-
tactic structure is apparently epiphenomenal, a mimicry of semantic
structure? And why can systematic syntactic deviations from this isomor-
phism so often be tied to pragmatic features of the communicative context?
Finding thorough answers to such questions is the quintessential task of the
functional linguist, as it would demonstrate that the fundamental functional
hierarchy of influence – pragmatics > semantics > syntax – is not just an a
priori belief but a persuasively substantiated linguistic theorem.19
5.4.2. The problem of the underlying representations (PR2)
According to Dik, the URs of FG1 supposedly only contain lexical items in
functional relationships with one another. Such a view facilitated Dik’s
shift to understanding the URs as bearers of meaning. No additional
semantic interpretative module is necessary, since the UR contains all the
information necessary to deduce the meaning of the sentence. As Dik
(1973b: 835) himself remarks, the “semantic content of a linguistic expres-

sion” is “the information which, given the role of context and of
extralinguistic knowledge, it is necessary and sufficient to assign to a lin-
guistic expression in order to explain its different final interpretations”.
This shift, I submit, lies at the origin of the problem of underlying repre-
sentations (PR2), because Dik wants the underlying representation to do
two different tasks: to structure the information that contributes to the in-
terpretation of a linguistic expression (URs as ‘semantics’) and to structure
the instructions (that is, the functional relations between the information
units) that will generate a linguistic expression (URs as ‘syntax’).20 Dik
(1986a: 11) clearly indicates this when he states that they “…are meant to
contain everything that is needed to retrieve the semantic content of the
predication on the one hand, and to specify the form of that expression on
the other”.
The problem is illustrated by Dik’s pragmatic explanations of linguistic
phenomenon such as definiteness. Does the definiteness operator in under-
lying representations represent the intrinsic definiteness of a term (a
semantic UR) and so is included for any intrinsically definite term,21 or
does it represent a trigger to generate the ‘definite article’ where need be?22
Neither account seems sufficient even for English, with its fairly restrictive
use of the definite article. Consider the following three sentences, anno-
tated with both types of UR:
(7) a. Mr Smith went home grumpy.

intrinsic: (dsx: ‘Mr Smith’: ‘grumpy’)
generative: (sx: ‘Mr Smith’: ‘grumpy’)
b. The grumpy Mr Smith went home.
generative: (dsx: ‘Mr Smith’: ‘grumpy’)
c. A grumpy Mr Smith went home.
generative: (isx: ‘Mr Smith’: ‘grumpy’)
The expression rules component would obviously prefer the ‘genera-

tive’ UR as input, but this is plainly at the expense of semantic accuracy.
A further complication arises when we read Dik’s explanation of defi-
niteness (1978a: 61; cf. 1973a): “The difference between ‘definite’ and
‘indefinite’ can in my opinion only be captured in pragmatic terms: by
means of a definite term the Speaker expresses the fact that he acts on the
presupposition that the Addressee can identify the particular intended refer-
ent(s) of the term in question.” This suggests that neither type of UR pre-
sented above represents the pragmatic significance of the use/non-use of the
article.23 Once we recognize from the above examples that ‘intrinsic’ and
‘generative’ can be understood as ‘semantic’ and ‘syntactic’ respectively, the
problem of the URs (PR2) can be understood as reconciling the tripartite
functional hierarchy of influence with a monolinear formal notation.24
5.4.3. Extreme Functionalism
Strangely enough, in spite of the absence of constituent structure in FG1, in

1983 Dik is still willing to admit its theoretical existence. He writes
(1983b: 75): “Does this mean that every structural property must some-
where find a functional explanation? No: as with all human instruments,
the functional requirements put on a language leave quite a bit of leeway
for alternative specifications of non-functional properties; historical devel-
opments unavoidably create rudimentary properties …; and conflicts
between different functional requirements may even create dysfunctional-
ity in given areas of linguistic organization”.
Hence one could imagine that either FG1 would make room for such
structure or Dik’s views would change. Three years later the latter occurs.
Dik (1986b: 11) writes: “Saying that a certain feature of linguistic design
or change cannot be functionally explained is tantamount to saying that we
have not yet been able to find a functional explanation for that feature”.
Autonomous, non-functional structure is completely excised.
5.4.4. The problem of psychological adequacy
Nuyts (1983; cf. also 1985, 1992) is the first to attempt to graft FG into a
broader cognitive model of grammar and to take seriously the all-
encompassing framework of pragmatics. Nuyts concludes, “It is obvious
that Dik’s theory is not a complete functional language theory: (a) it does
not deal with cognitive structures; (b) the pragmatic rules are well-nigh ab-
sent (apart from the pragmatic functions); (c) a discourse component is
missing. … Thus FG is intended to be a grammar in the restricted sense of
the word (a formal description of only the verbal aspects of language), and
not in its wider sense” (1983: 384).
From 1983 onwards it is clear that Dik is likewise increasingly con-
cerned with the psychological adequacy of FG1. We can chart the
problematic course that Dik sought to navigate in this regard. The problems
begin when Dik makes a crucial redefinition of psychological adequacy. In
FG1 psychological adequacy is defined weakly as “[a grammar] should not
be incompatible with strongly validated psychological hypotheses about
language processing” (Dik 1978a: 7 [emphasis mine]). But in 1983 he
writes, “… FG would like to relate as closely as possible to psychological
models of linguistic competence and linguistic behaviour. … [W]hy should
a grammar be neutral as between producing and comprehending linguistic
expressions, when producing and comprehending such expressions is just
what the grammar is there for?” (1983b: 76 [emphasis mine]).
Dik (1986a) attempts to integrate FG1 into a theory of verbal interac-
tion, or more precisely, “an integrated model of linguistic interpretation”.
Psychological adequacy reaches its most extreme (1986a: 6): “psychologi-
cal adequacy implies that a grammar developed according to FG
specifications should be a good candidate for incorporation into operational
models of natural language users” (cf. Dik 1989b: 13, 1997a: 13).
But clearly, for such a model to be accurate, it must represent humans
and their participation in verbal interaction. Dik adopts a simple but haz-
ardous strategy (1986a: 2): “the form of … knowledge representation
structures can be derived from the theory of Functional Grammar.” In
1989, this equivalence between URs and human cognition is taken to its
logical conclusion, where he proposes the following equation (Dik 1989a:
100):
(8) Lcog =Lur = Lkr = Llog

Lcog = general cognitive representation language
Lur = language of underlying representations
Lkr = language of knowledge representation
Llog = logical language for logical reasoning.
Not surprisingly, a thorough critique was forthcoming. Hesp (1990) ap-

praised Dik’s computational Functional Grammar model, raising serious
doubts about the possibility of Dik’s interpretation of psychological ade-
quacy. Due to the subsequent inattention to this issue, the role and
formalization of conceptual representations in Functional Grammar re-
mains unresolved.
Finally, a double irony in the history of Functional Grammar’s relation-
ship to psychology should be noted. One the one hand, since its inception it
has been a theory that explicitly seeks psychological adequacy, and yet it is
formal linguistic theories such as Chomsky’s that have been the most in-
fluential in psycholinguistics. On the other hand, it is arguably the case that

FG’s commitment to formalization (derived from the merging of predicate
logic with functional primitives) that has led to its appropriation in the
study of information sciences such as knowledge engineering (Weigand
1992), textual scenario analysis (Rolland and Ben Achour 1998), and or-
ganizational analysis (Steuten and Dietz 1998).
5.4.5. The layered ‘revolution’
Between FG1 and FG2 the most important development is the introduction
of the layered structure for URs. This development was a gradual expan-
sion of the predication structure.25
Dik (1979) presents tense and aspect (τ α) distinctions as ‘predicate op-
erators’; Dik (1981b) is the first to include them in the formal notation:27
(9) π1/2 predβ (ω x1)n
De Jong (1980) proposes indicating the illocutionary function of a sen-

tence as a predication operator. For example, the Ass(ertive) sentence (10)
could have the UR in (11):
(10) JOHN opened the door.

(11) Ass openV (xi: JohnN (xi))AgSubj (xj: doorN (xj))GoObj
In 1986, Kees Hengeveld, inspired by Searle (1969), Lyons (1977), and

Foley and Van Valin (1984),28 presents for the first time a (three-)layered
UR that contains an abstract illocutionary frame and operators at each
level. Predications are placed in the illocutionary frame, just as terms are
placed in the predicate frame (Hengeveld, 1986: 77; cf. Hengeveld 1987):29
(12) π4 ILL (π3 Xi: [π1 predβ (ω x1)n] (Xi))
Hengeveld (1988) uses the ‘E’ variable for speech acts and the ‘e’ vari-
able for predications (Vet 1986), thereby creating the fourth layer:30
(13) E1: [π4 ILL (S) (A) (π3 X1: [proposition31] (X1))] (E1)
(π2 e1: [π1 predβ (ω x1)n] (e1))

Not only is the clause layered horizontally, it is now also layered verti-
cally, the two layers signifying two different aspects of language: the
interpersonal and the representational. So sentence (14) is represented as
(15):32
(14) It seems that it is possible that he can cure blindness.

(15) E1: [DECL (S) (A) (XI: [seemV (XJ: [ ] (XJ))] (XI))] (E1)
(Pres e1: [possibleA (ej: [canV cureVinf (xj: p3 (xj))Ag (xk: blindnessN (xk)Go] (ej))Ø] (e1))
Hengeveld (1989) brings his layered structure to full fruition in his arti-
cle “Layers and Operators”. The structure is unchanged from 1988, but
Hengeveld provides a clear explanation of each component of the model
and many examples of various structures. More importantly, he lays out a
method of handling verbal complementation and clause combining.
5.4.6. Functional Grammar in Spain
In 1981 Leocadio Martín Mingorance translated FG1 into Spanish (Dik

1981b). This led to the establishment of the Spanish FG school, which has
concentrated on lexicology. Martín Mingorance (most recently, 1998) also
proposed a new synthesis, the Functional Lexematic Model, which has
been the basis of the important work of Faber and Mairal Usón (1999) and
others in recent years (see Mairal Usón and Pérez Quintero 2002).
5.5. Summary
We see that FG1 generated much work and faced many challenges. The
occasion arose to produce a new publication to pull these sometimes-
incompatible threads together.33 It is to this publication that I now turn.
6. FG2 and FG3 - 1989 to 2000
6.1. Introduction
If read in isolation, FG2 (Dik 1989b) appears to represent a major overhaul
of FG1, when actually it mainly incorporates developments of the model
suggested in the preceding years. And for reasons explained below, FG3
(Dik 1997a, 1997b) is not that different from FG2. So, assuming the reader
is familiar with FG2/3, we will focus solely on those developments that led
to Hengeveld’s FG4, except to notice that FG2/3 adopts Hengeveld’s layered
structure of the clause but collapses the two tiers (interpersonal and repre-
sentational) into a single tier.34 Thus, the structure is as follows.
(16) π4 Ei: [π3 Xi: [π2 ei: [π1 [predβ (ω x1)n] σ1] σ2] σ3] σ4
6.2. The impact of Simon Dik’s death

Simon Dik fell ill in 1992 and died three years later, doing little new re-
search in this period.35 To fulfil the goal of a second volume to FG2
(TFG2), it was decided to publish FG3 in such a way that “the text pre-
sented … is intended to reflect Simon Dik’s views as closely as possible”
(Hengeveld in Dik 1997b: viii). And since between 1989 and 1992 he was
preoccupied with computer implementations of FG2, FG3 is almost identi-
cal to FG2, except for the two chapters on term structure, which were
revised to take into account (aspects of) the work of Rijkhoff (1992). TFG2
is of course all new material, but theoretically is mostly an application of
TFG1. Xinzhang Yang (2000) and Verstraete (2000) provide reviews of
TFG2.
6.2.1. Hengeveld’s Layered Structure
Hengeveld continued after FG2 to develop his layered structure. In 1990, he

for the first time presents the notation complete with operators and satel-
lites,36 including a new σ5 ‘clause satellite’. The model is now (Hengeveld
1990a):37
(17) E1: [π4 ILL:σ4 (S) (A) (π3 X1: [predication] (X1): σ3 (X1))] (E1):σ5 (E1)
π2 e1: [π1 predβ:σ1 (ω x1: predN (x1))n] (e1): σ2 (e1)
Rijkhoff (1992) made a substantial contribution to the layered structure

of the clause with his proposals for noun phrase structure. A fascinating
observation he made was the parallel between NP structure and the struc-
ture of the extended predication, whereby nouns have operators (ω) and
satellites (θ) in three layers, as represented in his (simplified) NP structure:
(18) (ςi: [ ω3 [ ω2 [ ω1 predN (xi) θ1 ] θ2 ] θ3 ])
Hengeveld (1992) is an important article. Firstly, he incorporates Dik’s

(1989: 50) zero-level ‘f’ variable into the representation for all predicates
and an illocutionary ‘F’ variable. Thus the layered structure, with operators
added, is:
(19) E1: [π4 F1: ILL (F1) (S) (A) (π3 X1: [predication] (X1))] (E1))
π2 e1: [(π1 f1: predβ (f1)) (ω x1: (f2: predN (f2)) (x1))n] (e1)
Hengeveld justifies the ‘f’ and ‘F’ variables using his characteristic ar-
gument: because a predicate and illocution may serve as an antecedent for
anaphoric reference, a variable for the antecedent needs to be included in
the UR. Secondly, Hengeveld clarifies the place of adverbs in Functional
Grammar and provides functional definitions for parts of speech. Finally,
we see the first hint that the representational level is not always necessary
when Hengeveld briefly mentions that interjections such as Ouch! may
have a structure as follows:
(20) E1: (fi: ouchInt (fi)) (EI)
Hengeveld (1997) introduces a rhetorical layer above the interpersonal

layer, which consists of a discourse (D), a discourse type (T), and a move
(M). Hengeveld changes (S) and (A) in the illocutionary frame to (P1)S and
(P2)A and allows recursion at all levels to indicate the hierarchical structure
of discourses. This approach to discourse became known as the ‘upward
layering approach’. The layered structure is accordingly:
(21) D1: [(T1: TYP (T1)) (M1: [speech act] (M1))n] (D1)
(E1: [π4 F1: ILL (F1) (P1)S (P2)A (π3 X1: [predication] (X1))n] (E1))n
(π2 e1: [(π1 f1: predβ (f1)) (ω x1: (f2: predN (f2)) (x1))n] (e1))n
In FG4 the Move ‘M’ has moved down into the interpersonal level and
the rhetorical level is missing. However, it would seem that Hengeveld still
believes in such a level since he writes, “there are higher levels of dis-
course organization which are not captured here” (this volume: 5).
6.2.2. Some additional problems
Hengeveld’s 1997 model is still quite different from FG4. We need to re-
view briefly the factors that motivated the other changes in the FG4 model.
Bolkestein (1985a) is probably the first author to articulate clearly the
problem of discourse (PR7) in FG. She notes how discourse-level consid-
erations affect the code of language. Many other studies considered
discourse within the FG framework, but it was probably Kroon (1995) who
established this problem in the centre of contemporary FG debate. She pro-
vided the important definitions of ‘discourse move’ and ‘discourse act’
adopted in FG4.
Harder (1989) is to my knowledge the first to point out that the ‘E’ vari-
able cannot be interpreted as referring to a speech act in the way that the
other variables refer to their respective referents, because an utterance is a
speech act. Bolkestein (1992) also remarks that ‘E’ refers ambiguously
both to a clause, which is a product of a speech act, and to a speech act it-
self. Vet (1998) demonstrates that the use of ‘E’ in direct speech implies
that the speaker ‘says’ a speech act. Hengeveld (this volume) therefore
makes two important changes: he changes ‘E’ to ‘A’ (a discourse act) and
he introduces the expression layer to allow the UR to refer to the language
code, as he explains in his Section 6.2.
Van der Auwera (1992: 336) argues that Hengeveld’s use of variables is
at times “an ontological mix-up” between reference and denotation. This
criticism motivates Hengeveld’s (this volume) introduction of referential
and ascriptive acts in the interpersonal level, indicated by ‘R’ and ‘T’ re-
spectively.
Moutaouakil (1996) and Mackenzie (1998) extend Hengeveld’s (1992)
treatment of sentence fragments that omit certain layers in the layered
structure altogether but still remain discourse acts. This explains why in
FG4 the presence of a particular layer does not entail that all lower layers
within its scope have to be present.
Vet (1998) and Van den Berg (1998) suggest that discourse phenomena
are better handled by a modular approach. Vet’s motivation for the modu-
lar approach is that the phenomena of verbal interaction “obey different
rules” from those of the linguistic expression code. Vet also points out that
such a module is necessary regardless of discourse phenomena, to account
for pragmatic illocutionary conversion for instance. The introduction of the
communicative context in FG4 is Hengeveld’s modular proposal.
7. FG4: The New Architecture

We are now in a position to consider our three introductory questions in the
light of the evolution of Functional Grammar and the problems associated
with it, with most attention being devoted to the second question.
7.1. FG4 as a framework for Functional Grammar
7.1.1. What is FG4’s relationship to previous models?
A newcomer to FG4 who has only read FG2/3 would understandably strug-
gle to see the connection between the two, but the above survey
demonstrates the natural progression of Functional Grammar as envisioned
by Hengeveld and as shaped by many people over the last 15 years.
The different appearance of the notation should not be taken to mean
that the core commitments of Functional Grammar have changed. On the
contrary, FG4, for all its notational sophistication, is still basically a non-
transformational theory of grammar whereby linguistic expressions are
generated by expression rules operating on underlying representations sig-
nifying functional relationships between lexical items encased in various
frames, hierarchically nested within each other.
This continuity between Hengeveld’s model and FG0-3 should not over-
shadow the important discontinuities. As the discussion below will
illustrate, there are two defining characteristics of Hengeveld’s model that
are important progressions in the evolution of Functional Grammar: the
shift from predicate-centricity to pragmatic-centricity and the trifurcation
of the traditional Functional Grammar underlying representation into
interpersonal, representational, and expression layers.
Moreover, our survey amply documents the rapid and eclectic way in
which FG has continually adapted to the stream of linguistic data, criti-
cisms and suggestions in which all linguistic theories find themselves
existing. FG4, which in the light of the survey above presents itself as a di-
rect descendant of previous models, continues this tradition that began with
“speculations which have little more than a programmatic value” (Dik
1968: 293). My suggestion that a few non-functional elements may have
strayed into various FG models and that the theory has struggled with vari-
ous problems does not entail that its core commitments are indefensible,
only that their implementation has not been perfect.
7.1.2. Does it address objections raised against previous models?
Focusing on the problem of structure (PR1), we may concede that the ex-
pression level is an interesting addition to Hengeveld’s model, but his
examples are not clear as to the precise nature of this level. Hengeveld (this
volume: 7) characterizes it as “a representation of constituent structure”;
the expression rules fill this constituent structures in generating linguistic
expressions, since as Hengeveld observes, “the expression level … is the
product of grammatical and phonological encoding” (this volume: 7)
The problem of structure involves deciding if there is any non-
functional structure in language, and if so, deciding where it belongs in
FG4. From the perspective of speech production, the question is whether
the Speaker himself contributes constituent structures from a sort of ‘syn-
tacticon’ in the way he contributes lexemes from the lexicon (and
functional primitives from the ‘semanticon’). Hengeveld hints at such a
process when he writes, “Top-down decisions … concern the decisions the
speaker takes with respect to (i) the semantic content …, and (ii) the ex-
pression category necessary to successfully transmit his communicative
intentions” (this volume: 10) Yet consider his example (2), here renum-
bered as (22):
(22) Damn!
(A1: [EXPR (P1)Sp (P2)Addr (C1) ] (A1))

-----
(Lex1)
The choice of expression category, in this case EXPR, is not a matter of

constituent structure but an abstract illocutionary type. This presumably re-
sults in an isomorphic function-to-form correspondence between the type
and its (grammatical and prosodic) expression (cf. Van Buuren 1985; Lied-
tke 1998). Similarly, in footnote 2, Hengeveld (this volume: 18) lists
among the “encoding possibilities” of the illocution “prosodic encoding,
morphological encoding, and conventionalized lexicalization patterns”. So
although the expression level is clearly a representation of constituent
structure, it seems that, as in previous FG models, only the expression rules
manipulate such structure – the Speaker can only influence the structure
indirectly by choosing semantic and lexical primitives.
On this interpretation of FG4 we would have to say that it does not re-
solve the problem of structure,38 but the absence of detailed examples
illustrating the working of the expression level leaves open the possibility
of alternative interpretations of the model.
Let us now turn to PR2: the problem of the underlying representations.
It is difficult to evaluate FG4’s treatment of this problem since Hengeveld
does not discuss it. He only writes that “… the construction of linguistic
expressions can be interpreted as a decision-making process on the part of
the speaker” (this volume: 10).
Nevertheless, one could argue that Hengeveld addresses this problem
indirectly by introducing a third interpretation of underlying structures. Let
me explain. The interpersonal level clearly represents the intention of the
speaker, indicating discourse moves, discourse acts, ascriptive acts, and
referential acts. The variables for these do not have ‘meaning’ per se, as
they are outside C, “the information communicated in a discourse act”
(Hengeveld this volume: 8). Neither could they be triggers for the expres-
sion rules to generate the correct linguistic expression. In a word, the
interpersonal layer appears to be representing the pragmatics of communi-
cation, but pragmatics in a particular way. On the one hand, the
interpersonal layer represents pragmatics in a very broad sense, including
the illocutionary and intentional aspects of language use. On the other, it
represents pragmatics in a rather narrow sense, as Hengeveld restricts the
interpersonal layer to representing only grammatically coded reflections of
communication. So although a language may be used for dozens if not
hundreds of different types of discourse acts, illocutionary acts, intentions,
and so forth, it is only those with a unique grammatical expression that will
be represented.
It should be clear from my historical survey that the solution to PR2 is
not simply to clarify the status of URs, but in a principled manner to iden-
tify and untangle the components of URs and thus work out what belongs
where. And although Hengeveld does not address PR2 explicitly, his clear
separation of what can be understood as pragmatic, semantic, and syntactic
layers paves the way, in my opinion, for a resolution of this problem. In
other words, Hengeveld’s three layers correspond to the three possible in-
terpretations of the single-layered UR of FG1-3. To illustrate, let us return to
our three example sentences illustrating definiteness, this time with a pos-
sible FG4-like annotation:
(23) a. Mr Smith went home grumpy.

interpersonal: Decl (R: ‘Mr Smith’: ‘grumpy’)39
representational: (dsx: Mr Smith: grumpy)
expression: (NP: /mstə smθ/ … AdjP /grmpi /)
b. The grumpy Mr Smith went home.

interpersonal: Decl (RFoc: ‘Mr Smith’: ‘grumpy’)
expression: (NP: /ðə grmpi mstə smθ/)
c. A grumpy Mr Smith went home.
interpersonal: Decl (R: ‘Mr Smith’: ‘grumpy’)
expression: (NP: /ə grmpi mstə smθ/)
This representation has several notable features. Firstly, the expression

layer indicates the referential unity of the three utterances: in each the
Speaker refers to Mr Smith. Secondly, the interpersonal layer can represent
grammaticalized pragmatic differences if need be. Thus for sentence (b) we
could argue that the speaker intends both to refer and to draw attention to
what he is referring to, written formally as RFoc. Thirdly, the semantic, or
intrinsic definiteness of ‘Mr Smith’ is shown in all three utterances in the
representational layer, avoiding the dilemma that A grumpy Mr Smith…
poses for FG1-3. Fourthly, the variation in syntactic structure of the three
utterances is visible in the expression layer. So for the sake of argument,
suppose that linguists agreed that (c) was simply a variation of (a) contain-
ing no semantic or pragmatic differences. In this hypothetical case the
language provides the Speaker a choice between multiple syntactic realiza-
tions for the same pre-syntactic input.
The point of this brief analysis is obviously to demonstrate that FG4
provides a framework to tackle such linguistic data, a framework I would
suggest is eminently superior to FG1-3. The linguistic product emerging
from the complex interplay between pragmatics, semantics, and syntax can
now be more adequately described. More importantly, coupled with vari-
ous linguistic, philosophical, and psycholinguistic arguments, such data, it
is hoped, can be convincingly explained.
In other words, Hengeveld’s contribution in FG4 is to introduce a tripar-
tite formal notation corresponding to the tripartite functional hierarchy of
influence: pragmatics > semantics > syntax becomes instantiated as inter-
personal layer > representational layer > expression layer. It is my view
that this change alone constitutes the most important development for
Functional Grammar, for without it the theory can not unravel the Gordian
knot that has arisen from the well-intended but ultimately self-defeating
conflation of pragmatic, semantic, and syntactic information onto the
monolinear URs of FG1-3.
Our next concern is PR3, i.e. the problem of verbal interaction. FG4 at-
tempts in part to solve the problem of verbal interaction by creating a
pragmatic module that interacts with the grammar. While I would like to
see a pragmatic module developed, I suggest that it may not work the way
Hengeveld proposes, for two reasons. Firstly, let us consider Hengeveld’s
example (9), here numbered (24):

b. Is that a threat or a promise?
Let us break this down into a rough approximation of an FG4 account of

how this could work for our two participants, PA and PB, each with their
Pragmatic Module, PMA and PMB respectively.40
(25) a. PA forms: M1: A1: [Decl (C: [letV (IPro) (e: [readV (youPro) (my
poems)Go]) [(behaveV: wellAdv (youPro)]Cond]]]
b. The ERs produce S1: “If you behave well, I’ll let you read my poems.”
c. PA copies UR1 and its expression (as m1 and s1) into PMA
d. PB hears: “If you behave well, I’ll let you read my poems.”
e. PB copies s1 and m1 into PMB
f. PB forms: M2: [A2: [Int (C: [{threatN or promiseN} (Am1)Ø]]]
g. The ERs produce S2: “Is that a threat or a promise?”
The problem lies with the use of the ‘copies’ in step (c) and particularly
(e). Such an approach to pragmatic interaction is too reductionistic: what
feeds into the communicative context is not abstract underlying structures
but interpretations and inferences based on linguistic expressions. And since
each participant constructs their own communicative context, we must as-
sume some Gricean principle of mutual trust whereby each participant is
doing a faithful job in tracking participants, identifying discourse moves, es-
tablishing deictic reference points, and so forth. Much communication
involves checking that everyone is ‘doing his or her job’ properly. But like
interpretations and inferences, communicative principles and conventions
resist formalization. Implementing these into the model is no simple task.
Secondly, let us consider Hengeveld’s example (13), here number (26):
(26) This concert, if you want to call it that, isn’t exactly what I was waiting for.
Hengeveld writes that “[t]he word that refers to the preceding word
concert, which is a case of reference to the code rather than the message”
(this volume: 16). Again, let us break this down into stages to illustrate the
problem, using Hengeveld’s three-line notation.41
1. A1: [Decl (C1: [(R1)]), 6. (C2:[(T1T2R2R3R4)])Cond 11. (C3: [(T3 R5])

2. (prox sei: concert)Ø 7. (want (you)Ag [call 12. neg e: (exactly ..)
(Ø)Ag (it)Go ( that)Ref]Go42
3. Ph1: /ðs knst/ 8. Cl1: /f j wnt t k:l 13. Cl2: /znt gzæktl .. /
t ðæt/
4. [ðs knst] 9. [f j wnt t k:l t 14. [znt gzæktl ..]
ðæt]
5. cl1, etc. copy to CC 10. cl2, etc. copy to CC 15. cl3, etc. copy to CC
Figure 6. A representation of example (26)
The obvious problem is timing: step #5 must occur before step #7. Oth-
erwise, the speaker cannot formulate C2, as he cannot refer (by using that,
see note 42) to what R1 refers to (by using concert) until both he and the
listener have encoded and decoded concert instantiated as [knst] in the
communicative context.43
The objection to FG4 is not that this order of production is not a priori
impossible, but that the notation records no ‘chronology of composition’
and thus is static, whereas verbal interaction is manifestly dynamic. Fur-
thermore, it seems that such linking of underlying representations in a
chain of verbal interaction is fraught with cognitive and notational prob-
lems (not least of which would be the size of the notation). Dik’s (1989b:
52) warnings about the ‘quasi-productive’ interpretation of FG need to be
heeded.
Therefore I suggest that FG4’s account of verbal interaction is problem-
atic because on the one hand verbal interaction necessarily involves
interpretation, inference, and co-operation, none of which FG4 accounts
for, and on the other because it is dynamic while FG4 is not.
Now let us move on to PR4: the problem of primitives. FG4 has nothing
to say about (the problem of) functional primitives. This is unfortunate
considering the problems that have been raised concerning these. Having a
new architecture is of limited value if the building materials have a ten-
dency to show stress under pressure.44 What we can surmise is that the
division of the UR into three layers entails the movement (and perhaps re-
definition) of the traditional Functional Grammar semantic, syntactic, and
pragmatic primitives into new locations within the model.
PR5, i.e. the problem of discourse, is specifically addressed by FG4.

Hengeveld (this volume) seeks to deal with discourse phenomena by in-
cluding discourse-sized units in the interpersonal (and possibly a higher
rhetorical) layer and by feeding linguistic information into the modular
‘communicative context’. That is, he attempts to combine the upward-
layering approach with the modular approach (cf. Mackenzie 2000).
The most significant improvement in FG4 is the notational severing of
the model from FG’s sentence-centric shackles. I say ‘notational’ since it is
ironic that in FG0, Dik (1968: 199) advocates “no restriction to sentence as
highest unit” and in FG1 he writes (1978a: 15) that “FG is meant to cover
any type of linguistic expression, to the extent that the internal structure of
that expression is governed by grammatical rules. It is thus not restricted to
the internal structure of sentences … ”. FG4 merely fulfils what Dik always
envisioned for FG. The motivation for this change, however, arose through
FG’s encounter with linguistic phenomena, not with a desire to align the
notation with the vision. Further studies need to be done to see how benefi-
cial the model is for the description of discourse.
Our next concern is PR6, or the problem of psychological adequacy.
Hengeveld suggests that FG4 has “a higher degree of psychological ade-
quacy” (this volume: 2) than FG3. This is justified on the grounds that it
starts from the interpersonal layer and works down towards the expression,
while FG3 moves from the lexicon to predicates through to full expres-
sions.
It would seem that when Hengeveld talks of psychological adequacy he
is actually addressing the problem of the underlying representations: how
are we to interpret the model, what is it a model of? But this poses a prob-
lem: FG4 is strictly a model of grammar, verified and refined by linguistic
data, and applied to linguistic data; it proves its worth by how good a job it
does of explaining grammatical phenomena. It is not a model of speech
production.
Thus the model itself is agnostic with respect to psychological proc-
esses. Linguists may attach symbolic meanings to the model that are
analogous to cognitive-linguistic phenomena but they are just that, sym-
bolic meanings. Therefore I am not convinced that FG4 (or FG0-3) can be
psychologically adequate in any meaningful sense. This may trouble some
(and relieve others), so I will present a brief sketch of three ways of under-
standing the relationship between psychology and FG4.
Firstly, we would do well to study Dik’s original use of psychological
adequacy in FG1. Using LIPOC as an example (Dik 1978a: 192–212), we
see that LIPOC derives from linguistic data as a general word-ordering
principle and exists irrespective of any psychological theory. Dik notes fur-
thermore that LIPOC is explanatory with respect to linguistic data, but
what would explain LIPOC? He writes that “[w]e expect that such [ex-
planatory] principles can be found in psychological mechanisms bearing on
the perceptibility of constituents and on the human capacity for processing
complex information” (Dik 1978a: 212). Note that we do not start with a
psychological principle and then inject it into the FG model. By analogy,
the proof of the pudding with respect to the interpersonal layer in FG4
(where the strongest claims for psychological adequacy are made) is
whether it accounts for linguistic data. Do languages show clear evidence
that discourse moves ‘affect’ illocutions, that illocutions ‘affect’ proposi-
tions, and so forth? If so, then irrespective of psychology, one can justify
representing this ‘chain of command’, as it were, iconically in the linguistic
model. Psychologists may have something to say about such phenomena,
or they may not. In other words, I suggest that the way Dik used psychol-
ogy to explain an already observed fact such as LIPOC, P1, and so forth,
should be applied to layering.
Secondly, an inverse situation obtains when the linguist encounters in-
disputable psycholinguistic phenomena, such as Hengeveld’s mention of
the ‘intention to articulation’ processing pathway. Does this justify iconic
representation in the model? Not necessarily, because the model is imitat-
ing linguistic phenomena and not psychological. A case in point is the
‘chronology of composition’ problem presented in example (27) above.
The so-called ‘time course’ of speech production constitutes a sine qua non
for psycholinguistic grammatical models (Van Turennout 1997) but is to-
tally absent from almost every current linguistic theory. Similarly, with
respect to the three-layered structure, I do not think that Hengeveld can jus-
tify this division on evidence from psycholinguistic research, but only on
linguistic grounds.
So then, what do we do with such research? It provides a conversation
partner to engage with, one whose views may open up new approaches to
knotty problems. Sometimes this partner may be called upon ‘to give evi-
dence’ in a disputed linguistic case, but only alongside the witnesses of
language. It is collaborative, not authoritative (cf. Berg 1998; Boland
1999).
Thirdly, the very concept of psychological adequacy itself, as applied to
linguistics, involves a confusion of categories. It is like saying that a theory
of psychology should attain linguistic adequacy – one is simply talking
about two non-contiguous academic disciplines, which nevertheless, like
physics and biology, are in many ways deeply intertwined. The mistake is
to make one academic discipline have evaluative authority over another.

Even when psychology does bolster our case, we should beware of saying
that our model has ‘psychological adequacy’. Perhaps we could say instead
that the symbolic meaning we attribute to our model coheres well with
psycholinguistic evidence.
The last issue that needs to concern us here is PR7, that is, the problem
of formalization. We are now in a position to evaluate the core of FG4: its
formal structure. How does it line up with linguistic data? Irrespective of
the hypothetical pathway taken by the Speaker in deriving URs, the nota-
tion can only be justified if it provides superior linguistic explanations.
We should observe from the outset that the greater part of Hengeveld’s
article is a justification of an extension of the formal notation to capture
many subtle referential problems that “would have been difficult to handle
in earlier versions of FG” (this volume: 17). Hengeveld is therefore moti-
vated by problems of actual language data (hedged performatives, identity
statements, metaconditionals, and direct speech). Consequently, a crucial
question regarding FG4 is, in my opinion, whether the capacity of the
model to describe referentiality exceeds the capacity of grammar. Rather
than make a definitive judgement on this question, I wish to raise three ba-
sic issues that need to be considered.
Firstly, Hengeveld clearly sides with the view that the URs are bearers
of meaning: “the speaker refers to entities”, “the speaker … ascribes prop-
erties to entities” (this volume: 6), and since most of his problems are refer-
ential he has invested the model with referential precision. Whatever a pro-
form refers to, Hengeveld needs a variable for it. So let us take Hengeeld’s
example (9) again, now numbered (27), and pretend B replied differently:

b. Is that Keats or Donne?
In this case the demonstrative that would refer to ‘my poems’. This in-
dicates that the expression rules must select the same pro-form, that, to
represent both types of anaphora. Thus FG4, based on URs as meaning-
bearers, models a process of referential neutralization whereby a single
pro-form does many referential duties. An alternative model, based on a
pragmatic account of reference, would have the Speaker select that to trig-
ger an interpretative strategy whereby the Addressee determines the most
appropriate antecedent (as suggested by Harder 1996a: 237–243). The
point is clear: the solution to the problem of URs (PR2) affects how one
chooses between these alternatives.
Secondly, the FG4 account is not easily verifiable, as the UR could be

interpreted as containing more referential information than is encoded in
the language. As Harder writes: “If we view … referents in pragmatic
terms, however, they are merely aspects of the world of discourse after the
utterance, not semantic units which are introduced as entities referred to”
(1992: 320; Harder’s emphases). Not that FG is a stranger to such conun-
drums: witness the semantic neutralization across a single case-marked
noun in Latin or the enormous multifunctionality of genitive constructions
that in many languages reduces to a single morphosyntactic form. Such
phenomena are unfortunately often presented as dichotomous ‘division of
labour’ problems between grammar and pragmatics: is reference better ex-
plained pragmatically or syntactically? But surely the job of reference is
itself shared between syntax and pragmatics. The grammar allows particu-
lar pro-forms to refer to certain types of entities (which is why *Is such a
threat or a promise? is ungrammatical) and the Addressee’s interpretative
strategies choose the appropriate entity from the referential “inventory”
that the pro-form introduces. Lyons captures this balance well when he
writes (1977: 197) that “[t]he linguist can contribute to the study of refer-
ence by describing the grammatical structures and processes which
particular language-systems provide for referring to individuals and groups
of individuals. It does not follow, however, that the linguist must be con-
cerned with the actual reference of expressions in his analysis of the
grammatical structure of system-sentences”.
Thirdly, in defence of FG4, we must remember that models such as this
one are not language-specific. Rather, they are designed to represent the
superset of all possible linguistic phenomena. Thus crucial evidence
against this part of FG4 would be to show that no language has a particular
lexeme used for referring to previous discourse acts. But for a particular
language, we simply need to determine how the language grammaticalizes
referential subtleties. The issue here is to distinguish falsifying FG4 in gen-
eral from demonstrating that an aspect of FG4 is irrelevant to a particular
language.
FG4 is an ambitious formalization of the URs of linguistic expressions.
This formalization in FG4 has been aligned with the functional hierarchy of
influence, allowing descriptions of linguistic data that represent more
clearly than in FG1-3 the pragmatic, semantic and syntactic factors that in-
teract in language use. Further research is necessary to determine if such
superior descriptions lead to superior explanations.
7.2. FG4 as a framework for Functional Grammarians

It is always difficult to answer such questions as whether FG4 creates
sufficient space for the Functional Grammar academic community to con-
tinue to share the one roof. The biggest problem is that the model is part of
a larger whole that needs articulation. Theoretically, we have seen that FG4
touches upon many major areas that concern the FG community, but re-
mains silent on substantial issues such as the primitives and the lexicon.
As shown above, there has been no substantially new canonical FG
monograph since 1989. If anything, this survey underlines the need for
such a work, one that is perhaps more encompassing of diversity among
the FG community than previous volumes. From a practical perspective,
those teaching FG also deserve an up-to-date presentation of the theory,
and newcomers to the theory need a more user-friendly initiation if
Hengeveld’s increasingly elaborate notation becomes widely accepted.
Even the question what contemporary FG theory is, is hard to answer with-
out such a reference work.
Perhaps the word ‘architecture’, therefore, is particularly apt. By
analogy, I suggest FG could do with more engineers and builders. FG has
been in a design frenzy for many years now, as if having the right blueprint
will eventually manufacture the building all by itself. Needless to say, I
suggest that addressing perennial problems concerning the theory’s founda-
tions may offer more fruitful resources for development.
Notes
1. I would like to thank Lachlan Mackenzie, Kees Hengeveld, Dik Bakker,
Chris Butler, and the late Machtelt Bolkestein for their assistance.
2. The references in Figure 1 are also indicative of the important geographical
centres of current Functional Grammar research: The Netherlands, Belgium,
Spain, Denmark, and Morocco.
3. The official Functional Grammar bibliography (De Groot and Olbertz 1999)
lists 1,141 references for 1978 to 1998. If one includes post-1998 publica-
tions and Dik’s works from 1966–1977, the total is at least 1,300.
4. Many of these problems have been identified by those working in computa-
tional implementations of Functional Grammar (Bakker 1994; Kwee 1994).
There are other important problems, such as the problem of the lexicon –
what place do observed semantic and syntactic regularities in the lexicon
have in Functional Grammar? Are there language-independent lexical
primitives, and more basically, where do particles, interjections, adverbs
and prepositions belong? (Mackenzie 1992; Olbertz, Hengeveld and

Sánchez García 1998; Engberg-Pedersen et al. 1996; Mairal Usón and Van
Valin 2001); and the problem of sub-disciplines – how does Functional
Grammar integrate with the various linguistic sub-disciplines such as dia-
chrony, typology, language acquisition, sign language, and so forth (Bakker
1998; Boland 1999)?
5. Compare Dik (1997a: 26): “… functional statements specify the relations of
constituents to the constructions in which they occur”. Dik also notes that
this idea is found in the works of Tesnière, Strang, Piťha, and Buyssens.
6. One wonders why Dik does not refer to Whorf’s concept of ‘covert categories’
as summarized by Fillmore (1968: 3): “Many recent and not-so-recent studies
have convinced us of the relevance of grammatical properties lacking obvious
‘morphemic’ realizations but having a reality that can be observed on the basis
of selectional constraints and transformational possibilities”.
7. The abbreviations are as follows: ile: independent linguistic expression, s:
sentence; decl: declarative, np: noun phrase, fv: finite verb, intr: intransitive.
Note that the ‘ile’ is (potentially) larger than a sentence, as in question-
answer pairs. Dik suggests that language users probably do not know how
to produce such structures. Note the contrast with FG3, “…speakers are able
to construe underlying clause structures and map these onto linguistic ex-
pressions” (Dik 1997a: 56).
8. Dik writes, “… [FG0] can be viewed as a particular reconstruction of tag-
memic notions” (1968: 159).
9. They write (2000: 18) that “[i]t is then an empirical matter whether … we
will have to rehash the overall organization in the sense that we assign ele-
ments and features to the UR level that would formerly be assigned to ER
and the other way around”. Since the URs are already semantically com-
plete, in the context of their paper, I understand these “elements and
features” to refer to types of constituent structure.
10. Dik quotes this from Campbell and Wales (1970). In every subsequent pub-
lication he quotes instead from Hymes (1972). Hymes’s article is based on
his 1966 lectures, subsequently published in a monograph in 1971. Inciden-
tally, it should not be assumed that Dik borrows this phrase from Hymes
appropriately. Is there anything substantial in FG that would absolve it of
Hymes’s pointed critique of ‘Garden of Eden’ linguistics: “The controlling
image is of an abstract, isolated individual, almost an unmotivated cognitive
mechanism, not, except incidentally, a person in a social world” (1972:
272)?
11. Mackenzie (1987) is the only one to observe that the FG view of nouns as
monovalent instead of avalent results directly from Dik’s appropriation of
Bach’s logical-semantic formalism, which in turn derives from Carnap,
Reichenbach, and others.
12. A further clue to their origin was supplied by Lachlan Mackenzie (p.c.),
who relates this story: “I asked Simon why there were so few references to
other literature in the pragmatic functions chapter in Dik (1978). He said
that he had finished most of the rest of the book, but needed to add a vital
chapter on pragmatic functions. To do so, he needed to shut himself off
from the world for a few days, and therefore booked a hotel (in London, I
think it was). He took no or a few (I can’t remember) books with him, and
composed and completed the chapter in a few days. What references there
are were added later, back at his desk so to speak. This must all have been
around 1976/1977”.
13. In this structure ‘type’ is the lexical category (verbal, nominal, adjectival),
xn are terms, ‘s.r.’ are selection restrictions on the term, and SF is the se-
mantic function of the term in the state of affairs (SoA) designated by the
(nuclear) predication. It is hard to know where Dik derives the idea of selec-
tion restrictions from, since the only clue is an allusion to Weinrich (1966).
The notion was regularly used in generative writings in the 1960s (Katz and
Postal 1964: 15; Chomsky 1965: 95).
14. For recent critical evaluations of predicate frames in Functional Grammar
see Mairal Usón and Van Valin (2001) and García Velasco and Hengeveld
(2002).
15. He also dislikes Dik’s chapter on pragmatic functions: “It is deplorable that
Dik’s FG only sporadically and clumsily stumbles towards the findings of
the Prague school, rather than taking its starting point from there. … Regret-
fully, I must conclude that this chapter of Dik’s book lags behind the other
ones in scientific importance” (Piťha 1980: 268).
16. This reconstruction is purely hypothetical and other Functional Grammari-
ans undoubtedly view the situation differently. The point is that it is only by
understanding if and how Functional Grammar accounts for syntactic struc-
ture that the theory can be critically evaluated.
17. ‘Syntactic’ in FG is perspectival and not related to constituent structure.
18. Thus, the symbols ‘:’, ‘,’ and and – that is, restriction, apposition, and coor-
dination – represent micro-functional semantic relations inside the term
structure, which in turn enters into macro-functional relations with other
terms and predicates.
19. As evidence, consider Dik’s (1981a) article on ‘displacement’. It can also
be observed that between FG1 (1978) and Bakker and Siewierska (2000)
few publications discuss autonomous constituent structure.
20. The hierarchy is simply a succinct way of expressing the oft-quoted position
of Dik (1987b: 7): “… pragmatics is seen as the all-encompassing frame-
work within which semantics and syntax must be studied. Semantics is
regarded as instrumental with respect to pragmatics, and syntax as instru-
mental with respect to semantics”.
21. Siewierska (1991: 228 fn. 7) notes: “This unclarity in regard to the semantic
vs. semantico-syntactic nature of the underlying structures recognized in FG
creates some uncertainty in regard to the type of criteria to be used in the
determination of these underlying structures”.
22. “In order to be able to generalize over all definite terms, however, we as-
sume that even in such cases [as proper nouns and personal pronouns] the
definiteness operator is present in the underlying structure of the term” (Dik
1978a: 60).
23. Another such example is in Dik’s (1983a: 235) pragmatic interpretation of
reflexivity. He sees the Addressee as inheriting the following task from the
speech situation: “Given a two-place relation R2 and a single entity a, con-
struct an interpretation in which R2 is applied to a”.
24. The author who most clearly identifies PR2, but more from the viewpoint of
instructional semantics, is Harder (1990, 1992, 1996a, 1996b; see also
Mackenzie 1987 and Keizer 1992b: 148–157). He argues, for instance, that
“there is strictly speaking no semantics” (1992: 305) in the accepted inter-
pretation of FG, in the sense that “meaning” can only be attributed to an UR
once it is interpreted, and no such interpretation mechanism exists in FG.
25. In the structures given below, I have standardized the symbols to illustrate
the continuity in the model’s development. They are structurally unchanged
of course. The superscript ‘n’ indicates stacking. The subscript ‘β’ indicates
lexical category (N, V, A, and so forth).
26. It is in Dik (1989b: 137) that he first mentions Seuren’s work Operators and
nucleus (1969b). Seuren (who reviewed Dik’s dissertation, Seuren 1969a) notes
that propositional (i.e. illocutional) operators were proposed as early as 1946 by
Lewis and that modal operators were proposed as early as the Middle Ages.
Seuren’s complete list is as follows: the sentence qualifiers are ASSertion,
QUestion, IMPerative, and SUGGestion and the qualifiers are NEGation, Pres,
Fut, Perf, Past, U (for universal or gnomic tense), Poss, Nec(essity), and
Perm(itted). In the light of this work, it is perhaps surprising that FG took its
time to adopt a more sophisticated operator structure.
27. The predicate and predication operators are π1 and π2 respectively in FG2
onwards, but at this stage of FG development they are conflated so I have
used π1/2.
28. Hengeveld joined the FG discussion group in 1985. Here he first encoun-
tered Foley and Van Valin (1984) when it was reviewed by Machtelt
Bolkestein.
29. Hengeveld (1986, 1987) calls X a predication, but it later is correctly rela-
belled as a proposition. Hengeveld (1986: 120, n. 6) notes that Van Schaaik,
at the June 1985 FG symposium, was the first to distinguish between π3 and
π1 operators (cf. Van Schaaik 1983). Incidentally, Hengeveld (1986) is also
the first to mention in print the distinction between ILLs, ILLe, and ILLa, as
presented in Weijdema et al. (1982), but Dik introduced it first in his 1985
FG lecture notes.
30. Vet’s (1986) paper is very important to Hengeveld, who regularly quotes
from it, since he adopts Vet’s argument that predicational adverbs restrict
the predication, thus justifying representations such as the following:
Pierre arrived at eight o’clock

PAST ei: [arriveV (d1xi: PierreN (xi))Proc] (ei): [eight o’clockN] (ei)
The two (ei)’s in the structure mean that both Pierre arrives and eight
o’clock apply to the discourse event (ei). This somewhat technical point ex-
plains Hengeveld’s preference for each variable in the structure to appear
symmetrically on both sides of the layer the variable has scope over. See
Mackenzie (1987) for an important and neglected alternative to this argu-
ment.
31. Hengeveld (1990a: 4) subsequently corrects his terminology to predication
in this notation since the proposition is actually π3 X1: […] (X1). Note, how-
ever, that in the same volume, Hengeveld (1990: 107) again incorrectly uses
proposition.
32. The reason that two ‘X’ and two ‘e’ variables occur is that epistemic and
objective modality are expressed respectively by the lexical items (seemV
and possibleA). Note that Hengeveld’s use of such lexeme-based forms in
this example prohibits the canonical use of operators to represent grammati-
calized semantic categories. This practice possibly diminishes the
typological strength of the notation.
33. There are three other contributions worth mentioning. Vester (1983) adds
Change and Momentaneous to the SoA typology, based on the works of
Vendler and Dowty; Brown (1985) introduces a lot more sophistication re-
garding term operators; and Dik (1987) provides a typology of entities using
Bunt’s (1985) work.
34. See Keizer (1992a) for an excellent discussion of other more subtle differ-
ences, particularly Dik’s unexplained introduction of the ‘f’ variable for
zero-order entities.
35. Since I only encountered Functional Grammar in 1999, I cannot speak of
the broader (and clearly significant) impact of Simon’s death on the FG
community in general. Machtelt Bolkestein (p.c.) kindly provided this brief
comment: “After Simon’s death his PhD students were re-distributed over
the other FG-minded available professors, and naturally their own works
underwent both a slowing down at least for a while, and perhaps also some
changes. All in all the first year after his death there was a very depressed
atmosphere, which picked up after (and perhaps partly due to) the Córdoba
conference on FG [1996]”.
36. Hengeveld (1990) suggests that all five satellites are restrictive satellites,
which accounts for the ‘:’ preceding each one. However, in the same publi-
cation, Dik, Hengeveld, Vester and Vet (1990: 63) argue that σ1 satellites
are always restrictive, σ3-5 are always non-restrictive and only σ2 is either re-
strictive or non-restrictive.
37. Because Hengeveld wants the σ1 to bind a variable and because in FG3, σ1
modifies the variable-less “core predication” (Dik 1997a: 87), he must bind
it directly to the predicate. So the scope of σ1 is ambiguous between predi-
cate and core predication. Similarly, the σ4 binds to ILL as there is not yet
an ‘F’ variable in this model.
38. As early as 1990, Van Valin noted that for FG “the development of such a
theory [of non-relational syntactic structure] and an explicit means of repre-
senting syntactic structure is essential” (1990: 225). Mairal Usón and Van
Valin (2001) reiterate this sentiment. A ‘syntacticon’ could contribute to
solving this problem and provide a natural place to provide discourse parti-
cles and other clause combiners. Incidentally, the term ‘semanticon’
actually comes from Dik (1973b).
39. ‘Mr Smith’, Mr Smith, and /mωstχ smωθ/ etc. represent conceptual, lexical,
and (Australian English) phonological encoding respectively. I understand
the expression layer to signify that which is input for the articulatory com-
ponent of Functional Grammar.
40. Hengeveld writes that “… the demonstrative that refers to the preceding
move”, which is why ‘Am1’ is the subject of S2 in line (f). In FG, ‘A’ is
used for anaphora, and coreferential uses of pronouns are understood to be
anaphoric. It should be noted that Hengeveld’s statement is not unproblem-
atic. The pronoun that may be a case of what Lyons calls impure textual
deixis (quoted in Levinson 1983: 87 from Lyons 1977: 670). Such uses of
pronouns seem both to refer (in this case to the state of affairs referred to be
the first speaker’s utterance) and to be anaphoric (in this case to the state-
ment/move made by the first speaker’s utterance).
41. The numbers represent a possible order of production. ‘CC’ stands for
‘communicative context’, and ‘Ph1’ for ‘phrase one’. The fifth line repre-
sents the copying mechanism proposed in FG4 to account for verbal
interaction. I am assuming that the two ascriptive acts in step #6 are want
and call and that the three referential acts are you, it, and that, since anaph-
ora and discourse deixis are both referential.
42. The notation of steps ##6–7 is not at all straightforward. In FG1-3 one would
probably have something as follows for step #7: (f1: want (xj: you)Ag [f2: call
(Ø)Ag (Ae1)Go (Al1)Ref]Go, where Aei (‘it’) is coreferential with ei: concertlex1
and Al1 (‘that’) is discourse-deictic (see Levinson 1983: 86) with (just the
lexeme) concertlex1, the “l” standing for ‘lexeme’. This perfectly illustrates
the sort of pragmatic-semantic conflation FG4 seeks to avoid. I understand
FG4 as locating reference in the interpersonal layer; thus in R3 the speaker

refers to what R1 refers to and in R4 the speaker refers to the lexeme con-
cert. This allows direct entry in the representational layer of the pronominal
forms the speaker chooses to ‘represent’ the content of the referential acts
R3–4.
43. A similar problem regarding the use of Tails to clarify the main clause in
response to a non-verbal clue from the Addressee was noted by Piťha
(1980).
44. Some sources of pressure include the SoA typology (Rijksbaron 1989);
Subject and Object in patient-prominent languages such as Lak (Helmbrecht
1992); the problem of assigning semantic functions to terms with no rela-
tionship to a predicate (Mackenzie 1990); and the presence of linguistically
questionable primitives such as ‘sim’ and ‘post’ in the FG notation (Harder
1995: 232–235).
References
Auwera, Johan van der
1983 Review of Bolkestein et al. (eds), Predication and expression in
Functional Grammar 1981. Journal of Pragmatics 7: 433–442.
1992 Free relatives. In: Michael Fortescue, Peter Harder and Lars Kristof-
fersen (eds), 329–354.
1997 Cosubordination. Working Papers in Functional Grammar 63. Am-
sterdam: University of Amsterdam.
Auwera, Johan van der, and Louis Goossens (eds)
1987 Ins and Outs of the Predication. Dordrecht: Foris.
Bach, Emmon
1968 Nouns and noun phrases. In: Emmon Bach and Robert T. Harms
(eds), Universals in Linguistic Theory, 91–122. New York: Holt,
Rinehart and Winston.
Bakker, Dik
1994 Some formal and computational aspects of the Functional Grammar
machine model. In: Elisabeth Engberg-Pedersen, Lisbeth Falster Ja-
kobsen and Lone Schack Rasmussen (eds), 421–440.
1998 Language change: Some implications for the theory of Functional
Grammar. Working Papers in Functional Grammar 66. Amsterdam:
University of Amsterdam.
Bakker, Dik and Anna Siewierska
2002 Adpositions, the lexicon and expression rules. In: Ricardo Mairal
Usón and Maria Jesús Pérez Quintero (eds), 125–177.
Bauer, Winifred
1980 Review of Dik’s Functional Grammar 1978. Archivum Linguisticum
11: 50–57.
Berg, Marinus E. van den
Berg, Thomas
1998 Linguistic Structure and Change: An Explanation from Language
Processing. Oxford: Clarendon Press.
Boland, Annerieke
1999 Functional Grammar and first language acquisition: The acquisition
of adverbs in a functional approach. Unpublished Master’s thesis.
Department of General Linguistics, University of Amsterdam.
Bolkestein, A. Machtelt
1985 Cohesiveness and syntactic variation: Quantitative vs. qualitative
grammar. In: A. Machtelt Bolkestein, Casper de Groot and J. Lach-
lan Mackenzie (eds) (1985b), 1–14.
1992 Limits to layering: Locatability and other problems. In: Michael
Fortescue, Peter Harder and Lars Kristoffersen (eds), 387–407.
1998 What to do with Topic and Focus? Evaluating pragmatic informa-
tion. In: Mike Hannay and A. Machtelt Bolkestein (eds), 193–214.
Bolkestein, A. Machtelt, Henk A.Combé, Simon C. Dik, Casper de Groot, Jadran-
ka Gvozdanović, Albert Rijksbaron and Co Vet.
1981 Predication and Expression in Functional Grammar. New York:
Academic Press.
Bolkestein, A. Machtelt, Casper de Groot and J. Lachlan Mackenzie (eds)
1985a Predicates and Terms in Functional Grammar. Dordrecht: Foris.
1985b Syntax and Pragmatics in Functional Grammar. Dordrecht: Foris.
Brown, D. Richard
1985 Term operators. In: A. Machtelt Bolkestein, Casper de Groot and J.
Lachlan Mackenzie (eds) (1985b), 127–145.
Bunt, Harry
1985 Mass Terms and Model-Theoretic Semantics. Cambridge: Cam-
bridge University Press.
Butler, Christopher S.
1991 Standards of adequacy in a Functional Grammar. Review of Dik’s
The Theory of Functional Grammar, Part I: The Structure of the
Clause 1989. Journal of Linguistics 27: 499–515.
1999 Nuevas perspectivas de la Gramática Funcional: Los estándares de
adecuación de la teoría. In: Javier Martín Arista, Francisco J. Ruiz de
Mendoza Ibánez, Ricardo Mairal Usón and Christopher S. Butler
(eds), 219–256.
Butler, Christopher S., John H. Connolly, Richard A. Gatward and Roel M. Vis-
mans (eds)
1997 A Fund of Ideas: Recent Developments in Functional Grammar.
Amsterdam: IFOTT.
Campbell, R. and R. Wales
1970 The study of language acquisition. In: John Lyons (ed.), New Hori-
zons in Linguistics, 242–260. Harmondsworth: Penguin.
Chomsky, Noam
1965 Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Comrie, Bernard
1979 Review of Dik’s Functional Grammar 1978. Studies in Language 3:
267–276.
1980 Comments on Dik’s Functional Grammar. In: Michael Kac (ed.),
Current Syntactic Theories: Discussion Papers from the 1979 Mil-
waukee Syntax Conference, 29–33. Indiana University Linguistics
Club.
ward (eds)
1997 Discourse and Pragmatics in Functional Grammar. Berlin and New
Dahl, Östen
1971 Nouns as set constants. Gothenburg Papers in Theoretical Linguistics
3.
Devriendt, Betty, Louis Goossens, and Johan van der Auwera (eds)
1996 Complex Structures: A Functionalist Perspective. Berlin and New
Dik, Simon C.
1966 Review of Robins’s General Linguistics: An Introductory Survey.
1964. Lingua 16: 399–412.
1967a Review of Vachek’s The Linguistic School of Prague: An Introduc-
tion to its Theory and Practice 1966. Lingua 18: 80–89.
1967b Some critical remarks on the treatment of morphological structure in
transformational generative grammar. Lingua 18: 352–383.
1968 Coordination: Its Implications for the Theory of General Linguistics.
Amsterdam: North-Holland.
1973a Crossing co-reference again. Foundations of Language 9: 306–325.
1973b Some remarks on the notion ‘universal semantics’. In: Patrick Suppes,
Leon Henkin, Athanase Joja and Gr. C. Moisil (eds), Logic, Methodol-
ogy, and Philosophy of Science IV: Proceedings of the Fourth Inter-
national Congress for Logic, Methodology and Philosophy of Science,
Bucharest, 1971, vol. 4, 833–843. (Studies in Logic and the Foundations
of Mathematics 74.) Amsterdam: North-Holland.
1975 The semantic representation of manner adverbials. In: Albert Kraak

(ed.), Linguistics in the Netherlands 1972–1973, 96–121. Assen: Van
Gorcum.
1977 Stepwise lexical decomposition. Paper presented at the Nijmegen
conference on the Foundations of Semantics.
1978a Functional Grammar (North-Holland Linguistic Series 37). Amster-
dam: North-Holland.
1978b Stepwise Lexical Decomposition. Lisse: Peter de Ridder Press.
1978c Word order in a Functional Grammar. In: Wolfgang U. Dressler and
Wolfgang Meid (eds), Proceedings of the 12th International Con-
gress of Linguistics, 349–351. Innsbruck: Universität Innsbruck.
1979 Seventeen sentences: Basic principles and application of Functional
Grammar. Publikaties van het Instituut voor Algemene Taalweten-
schap, 22. University of Amsterdam.
1980a Reply to Comrie’s comments. In: Michael Kac (ed.), Current Syntac-
tic Theories: Discussion Papers from the 1979 Milwaukee Syntax
Conference, 35–40. Bloomington: Indiana University Linguistics
Club.
1980b Studies in Functional Grammar. London and New York: Academic
Press.
1981a Discrepancies between predication and expression in natural lan-
guages. In: A. Machtelt Bolkestein, Henk A. Combé, Simon C. Dik,
Casper de Groot, Jadranka Gvozdanović, Albert Rijksbaron and Co
Vet, 19–39.
1981b Gramática Funcional. Traducción, Glosario de Términos Técnicos e
Introducción a la Edición por Fernando Serrano Valverde. Trans.
Leocadio Martín Mingorance. Madrid: Sociedad General Española
de Librería.
1981c Predication and expression: Statement of the problem and outline of
the theoretical framework. In: A. Machtelt Bolkestein, Henk A.
Combé, Simon C. Dik, Casper de Groot, Jadranka Gvozdanović, Al-
bert Rijksbaron and Co Vet, 1–17.
1983a On the status of verbal reflexives. In: Liliane Tasmowski and Domi-
nique Willems (eds), Problems in Syntax, 321–355. (Studies in
Language 2.) Ghent: Communication and Cognition.
1983b Some basic principles of Functional Grammar. In: Shiro Hattori, Ka-
zuko Inoue, Tadao Shimomiya and Yoshio Nagashima (eds),
Proceedings of the XIIIth International Congress of Linguists, Au-
gust 29–September 4, 1982, Tokyo, 74–88. Tokyo: Tokyo Press.
1986a Linguistically motivated knowledge representation. Working Papers
in Functional Grammar 9. Amsterdam: University of Amsterdam.
1986b On the notion ‘functional explanation’. Working Papers in Func-

tional Grammar 11. Amsterdam: University of Amsterdam.
1987 A typology of entities. In: Johan van der Auwera and Louis Goos-
sens (eds), 1–20.
1989a Towards a unified cognitive language. In: Frieda J. Heyvaert and F.
Steurs (eds), Worlds behind Words; Essays in Honour of Prof.Dr.
F.G. Droste on the Occasion of his Sixtieth Birthday, 97–110. Leu-
ven: University of Leuven.
1989b The Theory of Functional Grammar. Part 1: The Structure of the
Clause. Dordrecht: Foris.
Clause. Berlin and New York: Mouton de Gruyter.
Structures. Berlin and New York: Mouton de Gruyter.
Dik, Simon C. (ed.)
1983 Advances in Functional Grammar (Publications in Language Sci-
ence 11). 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.
Engberg-Pedersen, Elizabeth, Lisbeth Falster Jakobsen and Lone Schack Rasmus-
sen (eds)
1994 Function and Expression in Functional Grammar. Berlin and New
Engberg-Pedersen, Elizabeth, Michael Fortescue, Peter Harder, Lars Heltoft and
Lisbeth Falster Jakobsen (eds)
1996 Content, Expression and Structure: Studies in Danish Functional
Grammar (Studies in Language Companion Series 29). Amsterdam
and Philadelphia: Benjamins.
Faber, Pamela B. and Ricardo Mairal Usón
1999 Constructing a Lexicon of English Verbs. Berlin and New York:
Mouton de Gruyter.
Fillmore, Charles J.
1968 The case for case. In: Emmon Bach and Robert T. Harms (eds), Uni-
versals in Linguistic Theory, 1–88. New York.
Foley, William A. and Robert D. Van Valin Jr.
1984 Functional Syntax and Universal Grammar. Cambridge: Cambridge
University Press.
Fortescue, Michael
1985 Anaphoric agreement in Aleut. In: A. Machtelt Bolkestein, Casper de
Groot and J. Lachlan Mackenzie (eds), (1985a), 105–126.
Fortescue, Michael, Peter Harder and Lars Kristoffersen (eds)

1992 Layered Structure and Reference in a Functional Perspective. Am-
sterdam and Philadelphia: Benjamins.
García Velasco, Daniel and Kees Hengeveld
2002 Do we need predicate frames? In: Ricardo Mairal Usón and María J.
Pérez Quintero (eds), 95–123.
Gebruers, Rudi
1983 S. C. Dik’s Functional Grammar: A pilgrimage to Prague? LAUT
Paper. 115: 1–20.
Groot, Casper de and Hella Olbertz (compilers)
1999 Functional Grammar Bibliography. www.functionalgrammar.com.
Hannay, Mike and A. Machtelt Bolkestein (eds)
1998 Functional Grammar and Verbal Interaction. Amsterdam and Phila-
delphia: Benjamins.
Hannay, Mike and Elseline Vester (eds)
1990 Working with Functional Grammar: Descriptive and Computational
Applications. Dordrecht: Foris.
Harder, Peter
1990 Tense, semantics and layered syntax. In: Jan Nuyts, A. Machtelt
Bolkestein and Co Vet (eds), 139–164.
1992 Semantic content and linguistic structure in Functional Grammar. On
the semantics of ‘nounhood’. In: Michael Fortescue, Peter Harder
and Lars Kristoffersen (eds), 303–327.
1996a Functional Semantics. A Theory of Meaning, Structure and Tense in
English. Berlin and New York: Mouton de Gruyter.
1996b Subordinators in a semantic clause structure. In: Betty Devriendt,
Louis Goossens and Johan van der Auwera (eds), 93–118.
Helmbrecht, J.
1992 An exercise in Functional Grammar - a formal representation of a
Lak folk-tale. Unpublished paper.
Hengeveld, Kees
1986 Aspects of a Functional Grammar of Spanish [Master’s thesis]. Uni-
versity of Amsterdam.
1987 Clause structure and modality in Functional Grammar. In: Johan van
der Auwera and Louis Goossens (eds), 53–66.
1988 Illocution, mood, and modality in a Functional Grammar of Spanish.
Journal of Semantics 6: 227–269.
25: 127–157.
1990a The hierarchical structure of utterances. In: Jan Nuyts, A. Machtelt
Bolkestein and Co Vet (eds), 1–24.
1990b Semantic relations in non-verbal predication. In: Jan Nuyts, A.

Machtelt Bolkestein and Co Vet (eds), 101–122.
1992 Parts of speech. In: Michael Fortescue, Peter Harder and Lars
Kristoffersen (eds), 29–55.
1997 Cohesion in Functional Grammar. In: John H. Connolly, Roel M. Vis-
mans, Christopher S. Butler and Richard A. Gatward (eds), 11–16.
1999 Formalizing functionally. In: Michael Darnell, Edith Moravcsik,
Fritz Newmeyer, Michael Noonan and Kathleen Wheatley (eds),
Functionalism and Formalism in Linguistics (Volume 2: Case Stud-
ies), 93–105. Amsterdam and Philadelphia: Benjamins.
2000 The architecture of a Functional Discourse Grammar. 9th Interna-
tional Functional Grammar Conference, 20–23 September, 2000.
Madrid. [Revised for this volume].
Hesp, Cees
1990 A critique of FG-CNLU. Working Papers in Functional Grammar
35. Amsterdam: University of Amsterdam.
Hoekstra, Teun, Harry van der Hulst and Michael Moortgat (eds)
1980 Special edition of Glot 3. 3/4. (Functional Grammar.)
Hymes, Dell
1972 On communicative competence. In: J. B. Pride and J. Holmes (eds),
Sociolinguistics, 269–293. Harmondsworth: Penguin.
1979 Short notice of Dik’s Functional Grammar 1978. Language in Soci-
ety 8: 305–306.
1983 Short notice of Dik’s Studies in Functional Grammar 1980. Lan-
guage in Society 12: 137.
Jakobson, Roman
1963 Efforts towards a means-ends model of language in interwar Conti-
nental linguistics. In: Christine Mohrmann, F. Norman and Alf
Sommerfeldt (eds), Trends in Modern Linguistics, 104–108.
Utrecht/Antwerp: Spectrum.
Johnson, David E.
1977 On relational constraints on grammar. In: Peter Cole and Jerrold M.
Sadock (eds), Grammatical Relations, 151–178. New York: Aca-
demic Press.
Jong, Jan R. de
1980 On the treatment of focus phenomena in Functional Grammar. Glot
3: 89–115.
Katz, Jerrold J. and Paul Postal
1964 An Integrated Theory of Linguistic Descriptions. Cambridge MA:
MIT Press.
Keizer, M. Evelien
1992a Predicates as referring expressions. In: Michael Fortescue, Peter
Harder and Lars Kristoffersen (eds), 1–27.
1992b Reference, predication and (in)definiteness in Functional Grammar:
A functional approach to English copular sentences. Doctoral disser-
tation, Free University of Amsterdam.
Kroon, Caroline
1995 Discourse Particles in Latin. A Study of nam, enim, autem, vero and
Kwee Tjoe Liong
1994 Programmer’s Reflections on Functional Grammar: More Exercises
in Computational Theoretical Linguistics. (Studies in Language and
Language Use) Amsterdam: IFOTT.
Levinson, Stephen C.
1983 Pragmatics. Cambridge: Cambridge University Press.
Liedtke, Frank
1998 Illocution and grammar: A double level approach. In: Mike Hannay
Longacre, Ronald E.
1965 Some fundamental insights of tagmemics. Language 41: 65–76.
Lyons, John
1977 Semantics. Cambridge: Cambridge University Press.
1987 The representation of nominal predicates in the fund. Working Pa-
pers in Functional Grammar 25. Amsterdam: University of
Amsterdam.
1990 Let’s get our heads together: A reply to van der Auwera. In: Mike
Hannay and E. Vester (eds), 133–144).
1992 Places and things. In: Michael Fortescue, Peter Harder and Lars
2000 First things first. Towards an Incremental Functional Grammar. Acta
Mackenzie, J. Lachlan and María A. Gómez-González (eds)
2003 A New Architecture for Functional Grammar. Berlin and New York:
Mouton de Gruyter.
Mairal Usón, Ricardo and Robert D. Van Valin

2001 What Role and Reference Grammar can do for Functional Grammar.
Revista Canaria de Estudios Ingleses 42. 137–166.
Mairal Usón, Ricardo and María Jesús Pérez Quintero (eds)
2002 New Perspectives on Argument Structure in Functional Grammar.
Berlin and New York: Mouton de Gruyter.
Martín Arista, Javier, Francisco J. Ruiz de Mendoza Ibánez, Ricardo Mairal Usón
and Christopher S. Butler (eds)
1999 Nuevas Perspectivas en Gramática Funcional. Barcelona: Ariel.
Martín Mingorance, Leocadio
1998 El Modelo Lexemático-Funcional. (Amalia Marín Rubiales, ed.).
Granada: University of Granada.
Matthews, Peter H.
1969 Review of Dik’s Coordination: Its Implications for the Theory of
General Linguistics 1968. Lingua 23: 349–371.
Moutaouakil, Ahmed
1989 Pragmatic Functions in a Functional Grammar of Arabic. Dor-
drecht: Foris.
Nichols, Joanna
1984 Functional theories of grammar. Annual Review of Anthropology 13:
97–117.
Nuyts, Jan
1983 On the methodology of a functional language theory. In: Simon C.
Dik (ed.), 369–386.
1985 Some considerations concerning the notion ‘psychological reality’ in
Functional Grammar. In: A. Machtelt Bolkestein, Casper de Groot
and J. Lachlan Mackenzie (eds) (1985b), 91–105.
1992 Aspects of a Cognitive-Pragmatic Theory of Language. On Cogni-
tion, Functionalism and Grammar. Amsterdam and Philadelphia:
Benjamins.
Nuyts, Jan, A. Machtelt Bolkestein and Co Vet (eds)
1990 Layers and Levels of Representation in Language Theory: A Func-
tional View. Amsterdam and Philadelphia: Benjamins.
Nuyts, Jan and Georges de Schutter (eds)
1987 Getting One’s Words into Line. On Word Order and Functional
Grammar. Dordrecht: Foris.
Olbertz, Hella, Kees Hengeveld and Jesús S. Sánchez García (eds)
1998 The Structure of the Lexicon in Functional Grammar. Amsterdam
Pérez Quintero, María Jesús (ed.)
2001 Revista Canaria de Estudios Ingleses 42. Sección Monográfica:
Challenges and Developments in Functional Grammar.
Pit’ha, Petr
1980 Review of Dik’s Functional Grammar 1978. Journal of Pragmatics
4: 265–270.
1985 Functional Grammar in Holland. The Prague Bulletin of Mathemati-
cal Linguistics 44: 63–77.
Prideaux, Gary D.
1981 Review of Dik’s Functional Grammar 1978. Language 57: 717–720.
Rijkhoff, Jan
1992 The noun phrase: A typological study of its form and structure. Doc-
toral dissertation, University of Amsterdam.
Rijksbaron, Albert
1989 Aristotle, Verb Meaning and Functional Grammar; Towards a New
Typology of States of Affairs. With an Appendix on Aristotle’s Dis-
tinction between Kinesis and Energeia. Amsterdam: Gieben.
Rolland, Colette and Camille Ben Achour
1998 Guiding the construction of textual use case specifications. Data and
Knowledge Engineering 25: 125–160.
Schaaik, Gerjan van
1983 A functional analysis of aspects of Turkish grammar. Unpublished
Master’s thesis. Department of General Linguistics, University of
Amsterdam.
Searle, John
1969 Speech Acts: An Essay in the Philosophy of Language. Cambridge:
Cambridge University Press.
Seuren, Pieter A. M.
1969a Echte en onechte taalkunde. On Dik’s Coordination: Its Implications
for the Theory of General Linguistics 1968. De Gids 132: 25–42.
1969b Operators and Nucleus: A Contribution to the Theory of Grammar.
(Cambridge Studies in Linguistics 2.) Cambridge: Cambridge Uni-
versity Press.
Siewierska, Anna
1991 Functional Grammar. London: Routledge.
Steuten, Ans A. G. and Jan L. G. Dietz
1998 Guidelines for bridging the gap between literal sentence meaning
and communicative intention. The Language/Action Perspective. 1–
14.
Turennout, Miranda I. van
1997 The electrophysiology of speaking: Investigations on the time course
of semantic, syntactic, and phonological processing. Doctoral
dissertation. MPI Series in Psycholinguistics 1. Katholieke Universi-
teit Nijmegen, The Netherlands.
Valin, Robert D. Jr. van

1990 Layered syntax in role and reference grammar. In: Jan Nuyts, A.
Machtelt Bolkestein and Co Vet (eds), 193–232.
Verstraete, Jean-Christophe
2000 Review of Dik’s The Theory of Functional Grammar, Part 1: The
Structure of the Clause (2nd edition), Part 2: Complex and Derived
Constructions. Language 76: 210–212.
Vester, Elseline
1983 Instrument and Manner Expressions in Latin. Assen: Van Gorcum.
Vet, Co
1986 A pragmatic approach to tense in Functional Grammar. Working Pa-
pers in Functional Grammar 16. Amsterdam: University of Amster-
dam.
1998 The multilayered structure of the utterance: About illocution, modal-
ity and discourse moves. In: Mike Hannay and A. Machtelt Bolke-
stein (eds), 1–24.
Watters, John R.
1980 Review of Dik’s Functional Grammar 1978. Lingua 50: 155–171.
Weigand, Hans
1992 Assessing Functional Grammar for knowledge representation. Data
and Knowledge Engineering 8: 191–203.
Weijdema, Willy, Simon Dik, Margo Oehlen, Clara Dubber and Akke de Blauw
1982 Strukturen in verbale interaktie. Muiderberg: Coutinho.
Weinrich, Uriel
1966 Explorations in semantic theory. In: Thomas A. Sebeok (ed.), Cur-
rent Trends in Linguistics, vol. 3, 395–477. The Hague: Mouton.
Xinzhang Yang
2000 Review of Dik’s The theory of Functional Grammar, Part 1: The
structure of the Clause (2nd edition), Part 2: Complex and derived
constructions. Word 51: 308–313.
Behind the scenes: Cognition and Functional
Discourse Grammar
Carlos Inchaurralde
1. Introduction
Psychological adequacy has always been a claim of Functional Grammar
(Dik 1989: 13), but it has always been problematical in its application (cf.
Butler 1991, 1999; Hesp 1990a, 1990b; Nuyts 1990, 1992: 223–236).
Hengeveld (this volume) faces some aspects of the problem and describes a
new architecture for a Functional Discourse Grammar (FDG), in which a
top-down model is adopted and communicative decisions precede represen-
tations. This model is coherent with the findings of Levelt (1989) and other
psycholinguists, according to whom the speech production process runs from
intention to articulation. This is fine as far as production is concerned, but we
should not forget that in linguistic interaction there is decoding as well as en-
coding. The production model describes the process from the communicative
intention to the message. However, there is also a reception process on the
hearer’s part, in which the message comes first, and from which both the
content and the communicative intention are obtained. In the pages below I
will try to show how both processes can be integrated into a single model
that preserves the architecture proposed by Hengeveld.
2. Modularity of the mind: different kinds of storage

Cognition plays an important role in the FDG model as a specific compo-
nent that interacts with the interpersonal, representational, and expression
levels, which have a hierarchical structure.1 The interpersonal level is
linked to the representational level, and the interpersonal and representa-
74 Carlos Inchaurralde
tional levels are linked to the expression level by means of expression

rules. Obviously, the influence of the cognitive component on any of the
levels should have an observable result on the structures produced, and the
information that can have such an influence may be of many different
kinds. If we assume, as Hengeveld claims, that it “represents the (long-
term) knowledge of the speaker, such as his communicative competence,
his knowledge of the world, and his linguistic competence” (this volume:
3), it is important to bear in mind that we are dealing with many different
types of information that may even have different psychological reali-
zations.
In modular approaches such as Fodor’s, our linguistic competence, that
is, our capacity to create and interpret language, has many layers that work
independently of others: our use of syntax, phonetics, morphology is auto-
matic, with very little influence from central systems. The most important
properties of modular systems, according to Fodor (1983), are information
encapsulation (they perform the operations that they have been assigned
independently of other systems) and domain-specificity (they use specific
information that is not shared with other modular systems or central sys-
tems).2 It is the autonomous, domain-specific nature of the different
‘modules’, which are specialized in different kinds of tasks, that distin-
guishes this conception from the holistic one, in which all mental processes
share mental resources.
In a manner parallel to this modularity of the mind, the architecture of
the FDG model assumes some degree of modularity for the different levels
in language production, but this structure is described in such a way that
there is also the possibility of having some influence from a general cogni-
tive component. Thus, this model allows some room for the claims of the
interactionist perspective, which assumes that there is interaction across all
mental processes, including the linguistic ones. This is thus an intermediate
point of view. But, in any case, the status of ‘linguistic competence’ (or
even that of ‘communicative competence’) will need to be very different
from that of ‘knowledge of the world’, which has a direct influence on the
choice and the interpretation of vocabulary and on the inferences that are
possible in discourse. For this kind of information, long-term knowledge
could easily be equated with long-term memory.
2.1. Long-term knowledge and encyclopaedic knowledge

There is considerable evidence that supports the well-known distinction be-
tween two types of memory: short-term and long-term (Loftus and Loftus
Cognition and FDG 75
1976); and the permanent store of our knowledge would correspond to the
second kind. A further distinction within long-term memory, i.e. that be-
tween episodic memory (in which we recall particular episodes or events)
and semantic memory (in which we integrate all our knowledge about the
world; Tulving 1972) does not add much, since both kinds may be useful
for interacting with the interpersonal, representational and expression lev-
els.
Long-term knowledge makes itself very evident in vocabulary interpre-
tation, when we resort to all the encyclopaedic information stored in
individual lexical items. Descriptions of how this encyclopaedic informa-
tion deep in our minds can be used and linked to single words appear in the
work of some cognitively oriented authors. Langacker (1987: 162), a cog-
nitive linguist, suggests that lexical units are merely points of access, and
that therefore “concepts are simply entrenched cognitive routines”. More
precisely:
The entity designated by a symbolic unit can (...) be thought of as a point of

access to a network. The semantic value of a symbolic unit is given by the
open-ended set of relations ⎯ simple and complex, direct and indirect ⎯ in
which this access node participates. Each of these relations is a cognitive
routine, and because they share at least one component the activation of one
routine facilitates (...) the activation of another. (Langacker 1987:163)
Langacker does not suggest this description out of the blue. It is based
on proposals by psychologists and computer scientists for models that in-
volve semantic networking. The main hurdle is how to determine, for every
access node, the extent of the corresponding conceptual network. Since it is
possible to think of a network in terms of an unending chain of associations
between nodes, there must be some mechanism, congruent with current
psychological evidence, that helps determine which nodes get a place in the
network, in other words which nodes are ‘relevant’. The keyword here is
activation. This is a term used in connectionism to designate the degree of
strength that the different nodes in a neural network may have.
Models of cognition that incorporate connectionist ideas use this term
profusely. Anderson (1983), for instance, in his ACT model, distinguishes
declarative memory from production memory and working memory. To re-
trieve information from declarative memory so that we can use it in our
working memory, a certain amount of energy is needed in the form of acti-
vation. The various modes of activation for concepts in a network explain
the link between words and encyclopaedic knowledge dynamically.
Also from psychology come the now well-established assumptions of

so-called ‘schema’ theories (Bartlett 1932; Rumelhart and Ortony 1977;
Rumelhart 1980), according to which everything we know about the world
is stored in large mental knowledge structures. Different labels have been
used to refer to these structures, depending on the type of information and
the way it is stored: ‘frames’, ‘scenarios’, ‘scripts’, ‘situations’, ‘idealised
cognitive models’, and so on (Minsky 1975; Schank and Abelson 1977;
Sanford and Garrod 1981; Fillmore 1982; Lakoff 1987, etc.) are all equally
useful in understanding how people use vocabulary in context. For in-
stance, Wierzbicka (1994, 1997) has pointed to the existence of different
‘scripts’ for interaction in different cultures and has drawn attention to how
words have uses and meanings that are highly dependent on culturally-
shared knowledge structures. All this seems to imply that it is necessary to
leave behind us the fixed-entry approach, provided by a dictionary perspec-
tive on the lexicon, and adopt some sort of distributed, encyclopaedic view.
However, at the same time, from an FG perspective, we do need fixed
entries in the Fund, where all predicate frames and terms are placed at our
disposal for the building of predications, and this is the place where the
cognitive component can act as input. The cognitive component is the in-
terface between our mental encyclopaedia and the lexicon that is available
to us. The cognitive component provides a ‘lexicopaedia’, in the form of a
fixed-entry list with points of access to a whole semantic network (cf. In-
chaurralde 2000). We cannot deny the evidence that shows us how
individual lexical units are a psycholinguistic reality,3 which gives psycho-
logical support to the existence of a lexicon in the Fund; but at the same
time it is important that we admit that there exists semantic information of
an encyclopaedic nature, and this is a natural and simple way of incorporat-
ing it in the FDG model.
2.2. Long-term knowledge and the communicative component

Knowledge of the relevant parameters that can affect the coding of the in-
terpersonal level should also be part of the long-term information held by
the cognitive component, especially if we bear in mind that this is the first
level in Hengeveld’s model. There are elements in the communicative con-
text which, once they become available to the cognitive component, are
relevant in further processing, and this fact is even more apparent in lan-
guages with a strong grammaticalization of pragmatic phenomena that deal
with the social relationship between the interlocutors in the communicative
situation. Japanese, for instance, attaches a lot of importance to the per-
ceived distance between the interlocutors, which relates to two different

factors that work together. One factor is politeness, which is in turn related
to a strong need to keep up one’s face, as well as the other person’s, in
Japanese culture. The second one is the strength of hierarchical relations in
a group-oriented culture. In Japanese, apart from the neutral ‘unmarked’
register, there is a special register, which is called keigo and has three dis-
tinct sub-registers, sonkei-go (respectful language), kenjoo-go (humble
language), and teinei-go (honorific language). Sonkei-go is used to show
respect towards a person whom we place ‘above’ us. This person may have
higher social status or may be somebody who deserves good treatment (a
client, or somebody we have just met for the first time). Kenjoo-go is used
by speakers who place themselves ‘below’ their interlocutors. Speakers
may also show respect by downgrading their own ‘face’ before others’. Fi-
nally, teinei-go is used when formal language is required. Since this
register can be used for any topic of conversation, its main function is sim-
ply to raise the level of formality involved.
These registers involve special choices in the vocabulary employed, but
they also require changes in the construal of the underlying situation,
which in turn influences its representation. Thus, in the FDG model, a
change of register would involve a different specification in the interper-
sonal level, especially concerning the speaker (Ps) and the addressee (Pa)
and their relationship, which cannot be properly codified in the hierarchical
structure as it stands. For Japanese, something like (Ps = Low) for humble
language, or something like (Pa = High) for respectful language could be
used, or a grading system which could show the relative social distance be-
tween both. For instance, (Pa = 10) and (Ps = 5) would show that the
addressee is socially higher than the speaker. It is important to notice that
theoretically it is possible to have (Ps = High) if the speaker is the Emperor,
or some kind of god in a tale, but the possibility is very unusual. In Eng-
lish, it could be compared to a character speaking with third person verbal
forms instead of using the first person. In Japanese, as there are no differ-
ent verbal forms according to person, it is the special register used that
helps us to know about whom something is being said.
The cognitive component, according to the specification present in the
interpersonal level, would then choose a suitable construal for the represen-
tation of content. Let us give an example. In sonkei-go, some verbs change.
Instead of taberu ‘to eat’, or nomu ‘to drink’, Japanese speakers use the
expression meshi-agaru, which literally means ‘to ascend food’. Instead of
miru ‘to see, to look’, we have goran ni naru ‘to become a viewing’. In
general, the expression o + verb root + ni naru can be used, so that we
have, for instance, the form o-kaki ni naru ‘to become a writing’ instead of
kaku ‘to write’. These forms are very lexicalized, but they still involve dif-
ferences in cognitive construal that require a different form at the
representational level. In these cases, even if we do not accept changes at
this level, it is clear that they do require different forms at the expression
level, which justifies the input from both the cognitive component and the
interpersonal level. There are many expressions in kei-go that differ only
lexically. We have the honorific prefixes o- and go- (e.g. o-tomodachi ‘his
friend’ instead of tomodachi ‘friend’), alternative forms (e.g. achira instead
of asoko ‘there’), and special words for the different kinds of kei-go (e.g.
iku ‘go’ and kuru ‘come’, which both change into irassharu in sonkei-go
and into mairu in kenjoo-go and teinei-go).
Of course, the information about the relative social distance of both in-
terlocutors is part of the communicative settings. General knowledge about
different kinds of communicative settings (as well as of different text ‘gen-
res’ or conventional text structures), with their implications for different
performative acts – cf. Austin’s (1962) mention of ‘felicity conditions’ – or
for the use of different communicative functions, is also part of the long-
term information stored in the cognitive component.
2.3. Linguistic competence and knowledge of the world

Hengeveld refers to linguistic competence as being part of the long-term
knowledge available to the cognitive component. If by linguistic compe-
tence we mean knowledge of the phonology, grammar, and vocabulary of
a language, then obviously we would have to include here the knowledge
of all vocabulary items and all rules of the grammar, as well as phonologi-
cal rules. However, although the form and certain combinatorial
characteristics of vocabulary items can be considered to be linguistic
knowledge, the above-mentioned ‘lexicopaedic’ information seems to go
beyond this kind of knowledge. It seems more suitable to consider lexical
linguistic information to be specifically the kind of information that is
available in the lexicon for predicates in terms of form, syntactic category,
quantitative valency, qualitative valency, selection restrictions, meaning
postulates, and meaning definitions (Dik 1989: 68ff.). This kind of infor-
mation, stored in long-term memory, is useful for a metalinguistic usage
of language, and for monitoring language production, but vocabulary
choice seems to have some degree of automatization. The same can be
said of syntactic and phonological rules. There seems to be some evidence
of independent modular capacities that deal with syntax and phonology in
an autonomous way (as can be seen, for instance, in the study of certain
aphasias, as in Saffran 1982, Saffran et al. 1980, or Schwartz 1987); but,
at the same time, there is also some evidence that interaction does take
place (cf. Olson 1973, Shatz 1977).
Another consideration is that we have a certain knowledge of the struc-
ture of external reality, which is based on bodily and sensory experience, as
cognitive linguistics has argued (cf. Radden 1992, Lakoff 1990). This
means that we have an internal representation of that structure in terms of
entities that interact in time and in space, with certain properties and cer-
tain relations, and we also have an idea of the causal structure of events in
the world. This long-term knowledge of an experiential nature can be
mapped onto knowledge about word classes and grammar categories
(nouns, verbs, categories of number, person, tense, aspect, mood, modality,
etc.) and this mapping can be assumed to take place in the cognitive com-
ponent, which is like a black box to the Functional Discourse Grammar
model, since the three levels in this model can receive its input but can nei-
ther see nor influence its contents.
3. The computer metaphor

The most interesting feature of the Functional Discourse Grammar model,
however, may not lie in its structure, but rather in its functioning. I there-
fore want here to put forward a conceptualization, congruent with the
encoding-decoding process that takes place in actual communicative inter-
action, of what this functioning might look like. In order to do this, I will
use a model from computer science.
Cognitive psychology has always been characterized, since its start in
the late fifties, by its repeated use of the computer metaphor in all the mod-
els it has presented. Broadbent (1968), for instance, used flow diagrams to
show the workings of our faculty of attention. Flow diagrams have also
been used to show the different kinds of memories (long-term memory,
short-term memory, and, later on, working memory) as storage places out
of which or into which a flow of information comes and goes (Atkinson
and Shiffrin 1968, Baddeley and Hitch 1974). Now that we have entered
the 21st century, psychologists do not feel so enthusiastic about using flow
diagrams, so that the situation has been reversed, with computer scientists
now looking to the brain itself for a model in the form of connectionism.
However, there is still room for the computer metaphor in the explana-
tion of the mind’s workings, and obviously there is room too for that
metaphor in the explanation of linguistic activity, which clearly is cogni-

tive in nature. Functional Grammar has always followed the computer
metaphor, or rather, the computer-language metaphor, with its Prolog-like
formalisms which make it so easy to implement in computer programs.
Nevertheless, Functional Grammar has always had the same operational
problem as Prolog: the logical relations are clearly expressed, and unifica-
tion and deduction are easy; but the actual procedure, the real machinery
that works behind the scenes, is hidden from us and has no specification –
not even a tentative one – of any kind.
What I suggest here is that we again resort to the computer metaphor
when looking at the processes that are covered in Hengeveld’s model. If
we take the two sides of communicative interaction, the speaker’s and the
addressee’s, as the two poles of a chain of actions, then the entire encod-
ing-decoding process would follow a process similar to that which is
currently used when two operating systems are connected by means of a
protocol. The layer model is customary in the architecture of present-day
operating systems. In the OSI Standard (Open Systems Interconnection
Reference Model),4 for instance, there are several different layers: the
physical layer, the data link layer, the network layer, the transport layer,
the session layer, the presentation layer, the application layer, and the user
program. The different layers may generate requests for those below them
(the lowest of all being the physical layer). For instance, a word processor
(the user program layer) may generate a request for a certain routine that
opens windows with some given colours and a given size (the presentation
layer). This means that the user program layer generates a request which is
passed on to the application layer and then goes on down to the presenta-
tion layer. This layer performs the routine and sends an acknowledgement
primitive that goes up to the user layer again to acknowledge receipt of the
request and the performance of the action requested (see Figure 1). In this
respect, the Functional Grammar model proposed assumes this kind of
communication moving up and down the different layers.
However, the interesting point in our use of the operating systems
metaphor is how we can employ the basic model for communication across
local area networks as a model for natural language communication be-
tween individuals in a Functional Discourse Grammar framework. In
modern information systems, the communication is arranged in fixed-
length chunks of data called packets. These packets, as well as handshak-
ing messages for regulating the communication, are transmitted between
two interconnected systems as ‘frames’ created by a data-link layer and
then transmitted by the physical layer. A request from the user program
triggers another request from the application layer, in the form of a ‘frame’
made up of some data and a header.
User program
REQUEST ACKNOWLEDGEMENT+ROUTINE
Application
layer
REQUEST ACKNOWLEDGEMENT+ROUTINE
Presentation
layer
Session
layer
… … …
Figure 1. An example using the OSI model
This request goes down to the next layer, where it is taken as the data of
a further packet with a new header; it then goes down to the next layer, and
so on. In the entire process down to the physical layer, where the flow of 1s
and 0s is transmitted along the communication lines, packets of a given
level are included as data in packets of the next level, creating a multilay-
ered structure that is very familiar to Functional Grammar researchers. This
is the encoding process. At the other end, the receiver of these packets
starts the process of unpacking the chunks of information received. At the
data-link layer a packet is obtained with a header and data. If the data has
some instruction that can be performed at this level, the process stops here;
but, if the data is a packet of a higher level, the unpacking process goes on
until it reaches the level at which the data contains instructions that can be
executed. In our analogy, the sender operating system would be the en-
coder of the linguistic message, and the receiver operating system the
decoder. Any request generated at a given level would need to be proc-
essed at the same level by the receiver of the message (see Figure 2).
Level of transaction
Actual communication (Physical layer)
Figure 2. The encoding-decoding process
Now, since this procedure actually works in local area networks and is a
fine model of phenomena that take place in real language, we can adopt it
as a procedural basis for Hengeveld’s model. Let us have a look at some
examples. Let us imagine a Japanese businessman who is waiting for a su-
perior. When he meets him, he says:
(1) Aa, Tanaka-san desu ka. Omachi shite orimashita

‘Oh, Mr. Tanaka. (I) have been waiting (for you)’
The process that has taken place is as follows. First, he has to encode
his communicative intention. The cognitive component comes into action,
retrieving long-term information about communicative settings and the dif-
ferent possibilities available and also drawing information from the
communicative context. The interactants have a different relative position
on the social scale, and this fact should be encoded at the interpersonal
level. The ‘request’ that is sent to the first layer is formalized as follows:
(2) [A1: [DECL (P1 = Low)Sp (P2 = High)Addr (C1: [...] (C1))] (A1)]
This level sends the ‘request’ to the next layer (= level in FDG), but appar-
ently it does not change the representation of the propositional content,
which stays as it would have been, without any indication of the difference
in relative social power between the interlocutors:
(3) [pi: [[PastProg ei: matsuv (x1)Ag (x2)Go] (ei)] (pi)]
This means that the request has to go down to the next layer (= level in
FDG), where it produces some changes at the expression level. More con-
cretely, the form o-machi shite is employed instead of matte ‘waiting’
(from matsu ‘to wait’); likewise, the form orimashita is used instead of
imashita (‘was’ or ‘had been’ in this example). It is at this level (the
equivalent to the physical layer in the computer communication model)
that the form is transmitted. Then, in decoding, the process is repeated, but
this time it goes upward from the most ‘physical’ level to the actual level at
which the request is processed. The ‘package’ is fetched and ‘unpacked’ at
the expression level. The data obtained from processing the expression
then goes to the representation level. At this level, the representation can be
correct from a propositional point of view, but there is still a package of
data to be processed which cannot be interpreted there. The package then
goes up to the interpersonal level, where it gets the interpretation it needs,
which is important, because the addressee needs to evaluate the adequacy
of the interpersonal settings that the speaker is using for the current com-
municative context. At all stages, of course, the receiver has needed to
consult the long-term information stored in the cognitive component in or-
der to be able to interpret the data packages.
This approach is also compatible with discourse processing, since both
macro-planning (concerned with the preparation of large stretches of
speech) and micro-planning (concerned with grammatical and lexical
choices; cf. Butterworth 1980: 159) can be encoded in some way in the
data packages involved. There are some marks at the expression level that
clearly encode information relevant to the interpersonal level (e.g. dis-
course markers, which show relations in discourse). This can also be
applied to pauses, which are significant as indicators of processing time.
As Butterworth (1980: 155) explains, “(...) the more the delays, the more
cognitive operations are required by the output”. As it can be assumed that
long pauses reflect more complicated cognitive processes than short
pauses, they can be signals of discourse planning and show relevant land-
marks of the cognitive processes involved (cf. Schilperoord 1998). As they
are involuntary, it is questionable whether we should think of pauses as
some kind of unit analysable at the expression level, but there is no doubt
that the addressee can use them to reconstruct discourse structure.
4. Conclusion
We have examined here, from a cognitive perspective, some questions
which are relevant to the role of cognition in the Functional Discourse
Grammar model proposed by Hengeveld (this volume). His model assigns
a very important role to the cognitive component, since cognition provides
input to all three levels proposed. However, cognition itself remains rather
mysterious to the model. The view adopted here is that this cannot be oth-
erwise, since cognition is situated outside the main core of the model.
Nevertheless, some assumptions can be made about the role of cognition,
particularly with regard to the various inputs from the cognitive compo-
nent. The cognitive component contains not only general knowledge about
the world, but also knowledge about relevant communicative parameters
and about linguistic competence itself. We have argued that these are dif-
ferent kinds of information. Some are linked to more central cognitive
systems in the mind, while some others are not if we accept the evidence
supporting the modular approach proposed by Fodor (1983). A computer
metaphor has also been put forward that helps conceptualize the workings
of the model in a more understandable way, and cognition has its place in
it. However, we must realize that while cognition, as a general capacity,
cannot be ignored within the Functional Discourse Grammar model, it is
wise to assume that giving a very detailed account of its internal workings
would go beyond the scope of this grammatical approach and would lead
us into matters of a fully psychological import. And this would take us
away from our main object of study, the linguistic message.
Notes
1. Note that I am referring to modularity of the mind and not to modularity
within the FG model.
2. Other properties of modular systems (cf. Fodor 1983) are that they have a
compulsory operation, which is automatic and independent of voluntary
processes; they work at a speed faster than other cognitive processes; their
output representations are incomplete and need to be completed by central
systems, which have more complete information; they are innate, since they
have a pattern of ontogenetic development; they are realized by means of a
fixed neural structure, that is, they tend to be clearly localized in certain
fixed areas of the brain; moreover, their alterations are highly specific and
selective, by contrast with what happens in central systems.
3. I give some examples in Inchaurralde (2000: 98–100).
4. The Open Systems Interconnection Reference Model is a set of standards

developed by the Open Interconnection Committee – sponsored by the In-
ternational Standardization Organization (ISO) – to facilitate intercon-
nectivity between computer systems.
References
Anderson, John R.
1983 The Architecture of Cognition. Cambridge: Harvard University Press.
Atkinson, Richard C. and Richard M. Shiffrin
1968 Human memory: a proposed system and its control processes. In
Spence, Kenneth W. and Janet T. Spence (eds), The Psychology of
Learning and Motivation (vol. II), 89–195. New York: Academic
Press.
Austin, John L.
1962 How to Do Things with Words. Oxford: Clarendon Press.
Baddeley, Alan D. and Graham Hitch
1974 Working memory. In Gordon H. Bower (ed.), The Psychology of
Learning and Motivation (vol. VIII), 47–89. New York: Academic
Press.
Bartlett, Frederick C.
1932 Remembering. A Study in Experimental and Social Psychology.
Cambridge: Cambridge University Press.
Broadbent, Donald E.
1968 Perception and Communication. London: Pergamon.
Butler, Christopher
1991 Standards of adequacy in Functional Grammar. Journal of Linguis-
tics 27: 499–515.
1999 Nuevas perspectivas de la Gramática Funcional: los estándares de
adecuación de la teoría. In Christopher Butler, Ricardo Mairal Usón,
Javier Martín Arista and Francisco José Ruiz de Mendoza (eds),
Nuevas perspectivas en Gramática Funcional, 219–256. Barcelona:
Ariel.
Butterworth, Brian
1980 Evidence from pauses in speech. In Butterworth, Brian (ed.), Lan-
guage production: Volume 1. Speech and talk, 155–176. London:
Academic Press.
Dik, Simon C.
1989 The Theory of Functional Grammar. Part I: The Structure of the

Clause. (Second, revised edition. Edited by Kees Hengeveld.) Berlin
and New York: Mouton de Gruyter.
Constructions. (Edited by Kees Hengeveld.) Berlin and New York:
Mouton de Gruyter.
1982 Frame semantics. In Linguistic Society of Korea (ed.), Linguistics in
the Morning Calm, 111–137. Seoul: Hanshin.
Fodor, Jerry A.
1983 The Modularity of Mind. Cambridge MA: MIT Press.
Hengeveld, Kees
This vol. The architecture of a Functional Discourse Grammar.
Hesp, Cees
1990a A Critique of FG-CNLU. Working Papers in Functional Grammar
1990b The Functional Grammar computational natural language user and
psychological adequacy. In: Jan Nuyts, A. Machtelt Bolkestein and
Co Vet (eds), 295–312.
Inchaurralde, Carlos
2000 Lexicopedia. In Peeters, Bert (ed.), The Lexicon-Encyclopedia Inter-
face. Amsterdam: Elsevier, 97–114.
Lakoff, George
1987 Cognitive models and prototype theory. In: Neisser, Ulric (ed.), Con-
cepts and Conceptual Development. Ecological and Intellectual Fac-
tors in Categorization, 63–100. Cambridge: Cambridge University
Press.
1990 The invariance hypothesis: Is abstract reason based on image-
schemas? Cognitive Linguistics 1: 39–74.
Langacker, Ronald W.
1987 Foundations of Cognitive Grammar. (Vol 1. Theoretical Prerequi-
sites.) Stanford: Stanford University Press.
1983 Monitoring and self-repairs in speech. Cognition 14: 41–104.
1989 Speaking: From Intention to Articulation. Cambridge MA.: MIT
Press.
Loftus, Geoffrey R. and Elizabeth F. Loftus
1976 Human Memory: The Processing of Information. Hillsdale NJ: Law-
rence-Erlbaum.
Minegishi Cook, Haruko
1996 The Japanese verbal suffixes as indicators of distance and proximity.
In Pütz, Martin and René Dirven (eds), The Construal of Space in
Language and Thought, 3–27. Berlin and New York: Mouton de

Gruyter.
Minsky, Marvin L.
1975 A framework for representing knowledge. In: Winston, Patrick H.
(ed.), The Psychology of Computer Vision, 211–280. New York:
McGraw-Hill.
Nuyts, Jan
1990 Linguistic representation and conceptual knowledge representation.
In: Jan A. Nuyts, A. Machtelt Bolkestein and Co Vet (eds), 263–293.
1992 Aspects of a Cognitive-Pragmatic Theory of Language. Amsterdam
Nuyts, Jan, A. Machtelt Bolkestein and Co Vet (eds)
1990 Layers and Levels of Representation in Language Theory. Amster-
dam and Philadelphia: Benjamins.
Olson, Gary M.
1973 Developmental changes in memory and the acquisition of language.
In: Timothy E. Moore (ed.), Cognitive Development and the Acquisi-
tion of Language, 145–157. New York: Academic Press.
Radden, Günter
1992 The cognitive approach to natural language. In: Martin Pütz (ed.),
Thirty Years of Linguistic Evolution, 513–541. Amsterdam and
Philadelphia: Benjamins.
Rumelhart, David E.
1980 Schemata. The building blocks of cognition. In: Rand J. Spiro, Ber-
tram C. Bruce and William F. Brewer (eds), Theoretical Issues in
Reading Comprehension. Perspectives from Cognitive Psychology,
Linguistics, Artificial Intelligence, and Education, 33–58. Hillsdale
NJ: Erlbaum.
Rumelhart, David E. and Andrew Ortony
1977 The representation of knowledge in memory. In: Richard C. Ander-
son, Rand J. Spiro and William E. Montague (eds), Schooling and
the Acquisition of Knowledge, 99–135. Hillsdale NJ: Erlbaum.
Saffran, Eleanor M.
1982 Neuropsychological approaches to the study of language. British
Journal of Psychology 73: 317–337.
Saffran, Eleanor M., Myrna F. Schwartz and Oscar S. M. Marin
1980 Evidence from aphasia: Isolating the components of a production
model. In: Brian Butterworth (ed.), Language Production. Volume 1:
Speech and Talk, 221–241. London: Academic Press.
Sanford, Anthony J. and Simon C. Garrod
1981 Understanding Written Language. New York: Wiley.
Schank, Roger C. and Robert P. Abelson

1977 Scripts, Plans, Goals and Understanding. Hillsdale NJ: Erlbaum.
Schilperoord, Joost
1996 It’s about Time. Temporal Aspects of Cognitive Processes in Text
Production. Amsterdam: Rodopi.
Schwartz, Myrna F.
1987 Patterns of speech production deficit within and across aphasia syn-
dromes: Application of a psycholinguistic model. In: Max Coltheart,
Giuseppe Sartori and Remo Job (eds), The Cognitive Neuropsychol-
ogy of Language, 163–199. Hillsdale, NJ: Erlbaum.
Shatz, Marilyn J.
1977 The relationship between cognitive processes and the development
of communication skills. In Charles B. Keasy (ed.), Nebraska Sym-
posium on Motivation 1977, 1–42. Lincoln: University of Nebraska
Press.
Tulving, Endel
1972 Episodic and semantic memory. In: Endel Tulving and Wayne
Donaldson (eds), Organization of Memory, 382–403. New York:
Academic Press.
Wierzbicka, Anna
1994 Cultural scripts. A new approach to the study of cross-cultural com-
munication. In Martin Pütz (ed.), Language Contact, Language
Conflict. Amsterdam and Philadelphia: Benjamins, 69–87.
1997 Understanding Cultures through their Key Words. English, Russian,
Polish, German, and Japanese. Oxford: Oxford University Press.
The Question of Discourse Representation in
Functional Discourse Grammar
John H. Connolly
1. Introduction
Ever since its inception, FG has been a pragmatically oriented theory, and a
good deal of pragmatically based research on the grammar of natural lan-
guages has been carried out within its framework. However, it was not
until the publication of Dik (1997a, 1997b) that the foundational text of FG
came to include a specific chapter devoted to discourse, focusing upon
units of language larger than the clause.
In the field of language studies outside of FG, there has, of course, been
a considerable amount of research into discourse from a variety of different
standpoints, stretching back for many years. Some, though by no means all,
of the principles established in this pre-existing work are incorporated into
the final chapter of Dik (1997b).
Now Hengeveld (this volume) has proposed a new architecture for FG,
which he terms Functional Discourse Grammar (FDG). Significantly, this
framework accommodates discourse as an integral part of the model. What
has not yet evolved, however, is a consensus on the issue of how to repre-
sent discourse. Hence, the present paper will be concerned with a
discussion of this question.
2. The Treatment of Discourse
2.1. Discourse in Dik’s FG Framework

First, though, we need to conduct a brief overview of previous research. An
appropriate place to begin is with a consideration of the scope of discourse.
90 John Connolly
In other words, what phenomena belong under the heading of discourse

studies and thus merit consideration in respect of the design of frameworks
for discourse representation?
We may begin by noting that discourse can be regarded either as a pro-
cess of communicative interaction or as the product of that process. Both
these perspectives need to be accommodated, and we shall seek to do so
here, using the term ‘discourse’ in whichever sense is appropriate in any
particular context.
The approach to discourse studies presented in Dik (1997b) may be
summarized as follows. Discourse as a product is viewed as a coherent
stretch of spoken or written language, which normally contains more than
one clause, and which may also contain one or more extraclausal constitu-
ents (ECCs). Discourse as a process can take the form of a monologue or
dialogue, and the interlocutors (or participants) are labelled S (for
speaker/writer) and A (for addressee) as appropriate.
Any discourse process will be embedded within a Discourse Event,
which is a type of social phenomenon. There are several important aspects
of discourse which apply to the event rather than the content of the lan-
guage employed. These include (i) the identity of the interlocutors and their
relationship to each other, (ii) the time, place and setting of the discourse,
and (iii) the prevailing social conventions which govern the acceptable use
of language. When embarking upon a discourse, several decisions need to
be made which tend to affect the construction of the whole piece of lan-
guage concerned. These include the choice of genre (such as a lecture or an
interview) and the selection of style (which may be formal or informal, po-
lite or impolite, and so forth).
As the discourse proceeds, the interlocutors develop a mental picture
which enables each new utterance to be interpreted in the light of what has
gone before, and also in the context of the interlocutors’ pragmatic infor-
mation concerning the discourse situation and, indeed, the wider universe.
This evolving mental representation, which has a degree of commonality
among all the interlocutors, is termed a Discourse Model (DM). Because it
is constantly changing, a DM is dynamic, and because it can never include
all the pragmatic information that is conceivably relevant, it constitutes
only a partial model of the discourse concerned.
As stated in Dik (1997a: 300–304), each individual clause has a basic il-
locutionary value which is to be viewed as a coded instruction whereby S
seeks to modify A’s pragmatic information. The possible illocutionary val-
ues are declarative, interrogative, imperative and exclamative. These types
are recognized on the basis that they tend to be coded differentially in the
The question of discourse representation 91
grammars of natural languages, unlike the speech act categories of, for ex-
ample, Searle (1976). However, in Dik (1997b: 419–421) it is proposed, in
addition, that whole discourses, or sections thereof, may have illocutionary
values, which are duly termed Discourse Illocutions, and which set the de-
fault values of the individual clauses within the stretch of language
concerned. Dealing with illocutions in terms of discourse also involves dis-
tinguishing between (i) the illocution as intended by S, written ILLS, (ii)
the illocution as coded in the expression, written ILLE, and (iii) the illocu-
tion as interpreted by A, written ILLA; see Dik (1997b: 230). This entails
countenancing the use of speech act categories additional to the four
grammatically encoded values; see Connolly et al. (1997a) for an example
of the use, within the FG tradition, of categories proposed by Searle (1976)
and Hancher (1979).
Certain other phenomena, too, tend to persist for whole sequences of
clauses or indeed for entire discourses. Such phenomena include tense and
discourse topic. Furthermore, the maintenance of topical continuity is con-
nected with the creation of chains of anaphoric reference, which frequently
extend across sentence boundaries.
Within the framework of Dik (1997b), a discourse event is structured in
two layers. The first is the Interpersonal Layer, which is concerned with the
human relations between the interlocutors and with their mentalities in
relation to the discourse and to each other. The interpersonal layer is
subdivided into the Interactional, which focuses on the interrelationship
and interplay between the participants, and the Attitudinal, which is con-
cerned with their states of mind, emotions and judgments, as reflected in
the discourse. The second layer is the Representational, which is related to
the actual language produced by the interlocutors. This layer is subdivided
into the Organisational, which deals with how the material that the dis-
course consists of is arranged and presented, and the Contentive, which is
concerned with the facts and States of Affairs communicated.
Discourse has (it is claimed) an essentially hierarchical structure,
though sometimes this is complicated by phenomena such as interruptions,
embeddings and recursions. The hierarchical structure can be discerned
within both the interpersonal and representational layers. In the interper-
sonal layer the lowest-ranking unit is the speech act (which will carry a
particular illocutionary value). Units of this type can combine into longer
sequences of speech acts. In a dialogue, sequences of one or more speech
acts are further organized into turns, and the whole sequence of turns
makes up the discourse. As a complication, patterns known as Adjacency
Sequences (such as question-answer or question-answer-reaction) can often
92 John Connolly
be identified as successions of speech acts over two or three turns. From

the point of view of the representational layer, on the other hand, the dis-
course is divided into episodes, sub-episodes, sub-sub-episodes and so on
(for example, chapters, sections and paragraphs in a book). Ultimately,
these units are composed of individual propositions, which constitute the
lowest-ranking type of discourse unit within the representational layer.
Another aspect of discourse appears in the form of rhetorical relations,
such as elaboration or contrast. A well-known approach to these phenom-
ena is found in the Rhetorical Structure Theory (RST) proposed by Mann
and Thompson (1987) and discussed in relation to FG by Gulla (1997). In
this approach, an individual rhetorical relation associates two stretches of
discourse, one of which is called the nucleus and the other the satellite. For
instance, an elaboration is considered to be the satellite of what it elabo-
rates, the latter being the nucleus. Dik (1997b) accepts that rhetorical
relations of the kind recognized in RST should be accommodated into the
FG-based approach to discourse.
2.2. The advent of a modular framework

If Dik’s (1997b) approach to discourse is accepted as being indicative of
the scope of discourse studies as far as FG is concerned, then we may
safely assume FDG to be in conformity with this. Nevertheless, there still
remains the question of how exactly to accommodate discourse within the
descriptive framework. Let us, therefore, now summarize how this issue
has been addressed within FG from 1997 onwards.
One approach to discourse has been to treat it as an additional level (or
set of levels) above the clause. This approach is described in Hengeveld
(1997), and has become known as upward layering. The additional level
was originally called the Rhetorical Layer, wherein speech acts are
grouped into moves, and moves are organized into a particular type of dis-
course (such as a conversation). A move is realized by a paragraph within a
monologue or by a turn within a dialogue. This view of discourse organisa-
tion is very reminiscent of the Sinclair and Coulthard approach to
discourse, where acts are grouped into moves, moves into exchanges, ex-
changes into transactions, and transactions into interactions; see Coulthard
(1977: 107). In fact, Steuten’s (1998a, 1998b) work on the analysis of
business conversations leads her to suggest replacing Hengeveld’s original
rhetorical layer with an Interactional Layer in which acts and exchanges
comprise the main organisational units within dialogue.
Some writers have urged a stronger separation between grammar and

discourse than is entailed by upward layering, contending that discourse
should be handled within a separate module. The latter is termed the Prag-
matic Module (as opposed to the Grammatical Module) by Vet (1998a,
1998b), Bolkestein (1998) and Liedtke (1998), the Discourse Module (as
against the Sentence Module) by Kroon (1997), and the Communicative
Unit (as distinct from the Grammatical Unit) by Steuten (1998a, 1998b).
Kroon argues that the lowest-ranking unit in the discourse hierarchy, which
she calls the Discourse Act, is not necessarily equivalent to the clause,
since ECCs, too, can function as discourse acts. For Vet and Steuten, the
difference in nature between the process of discourse and the product of
that process, in the form of clauses, motivates their being treated separately
in the descriptive framework. In fact, the idea of some kind of modular
framework seems also to be envisaged in Dik (1997b: 409).
In Vet's model the grammatical module and the pragmatic module are
linked by an interface. Van den Berg (1998) develops this interface into a
level in its own right, calling it the Message Module. In Van den Berg's
framework, the pragmatic module is the controlling unit of the discourse
process, and deals with both the construction of moves and with the man-
agement of the social context within which the discourse takes place. The
grammar module provides the lexical and grammatical material for the
composition of the message, but the latter is actually put together by the
message module in a manner determined by instructions from the prag-
matic module.
In Hengeveld’s FDG, three levels are recognized, namely the Interper-
sonal Level, the Representational Level and the Expression Level.
Moreover, each of these levels is conceived of as a separate module, and
each is internally layered. Rules serve to connect the modules, as appropri-
ate. In this way, Hengeveld seeks to combine the best points of previous
approaches. Another important advance is the recognition in FDG of a
communicative component and a cognitive component. Both of these com-
prise essential elements of the context with which the three language
modules interact.
2.3. Other areas of discourse studies

The FG approach to discourse embraces numerous phenomena. Neverthe-
less, it by no means encompasses everything that is addressed in the
broader field of contemporary discourse analysis. It will be appropriate,
94 John Connolly
therefore, to mention briefly a selection of other areas which at least some

linguists would also consider to lie within the scope of discourse studies.
Firstly, during the last twenty years or more, there has been a good deal
of research into the relationship between language, power and ideology.
This area of work is often termed Critical Language Study or Critical Dis-
course Analysis; see, for instance, Fairclough (1989) or Fairclough and
Wodak (1997). Its central premise is that power is unevenly distributed in
human society, and that existing inequalities of power are maintained
partly with the aid of sets of supposedly common-sense assumptions,
known as ideologies. For example, the prescriptive linguistic ideology that
there is a correct way of speaking and writing English is still accepted
without question by many people in the UK, and its maintenance, with the
help of expressions like ‘bad grammar’, is something that tends to suit the
purposes of the establishment. Authors in the tradition of Systemic Func-
tional Grammar have made moves towards incorporating ideology, along
with genre, into their theoretical framework; see, for instance, Martin
(1992: 496). However, this has not become common practice within other
approaches.
Secondly, the work of Bakhtin has attracted a fair amount of interest;
see, for instance, Clark and Holquist (1984). In particular, we should men-
tion here his theory of Voice. To take an example, suppose that speaker S1
makes some utterance U1, and that this event is subsequently recounted by
another speaker S2 in the form of utterance U2, such that U2 contains a
piece of reported speech, structured along the lines of ‘S1 said U1’. Utter-
ance U1, when thus embedded into utterance U2, has, in a sense, two
speakers: S1 (its originator) and S2 (its reporter), and their voices intermin-
gle into a polyphony. It is also possible to push this notion of voice further,
until it is no longer necessarily identifiable with a particular speaker, for
instance when someone is said to speak with the voice of a scientist. This
implies the recognition of a generic as well as an individual voice as being
a possible component within a polyphony.
Thirdly, Kress and Van Leeuwen’s semiotically based approach to dis-
course deserves to be taken into account here. Kress and Van Leeuwen
(1996: 39) remind us that text is in general multimodal. For example,
speech is typically accompanied by non-verbal communication (NVC),
such as gesture. Indeed, NVC can be used as a substitute for speech within
a dialogue; for instance, from the point of view of communication, a nod is
as effective as the word yes in response to a question (assuming a face-to-
face situation). Furthermore, written text has additional, non-linguistic but
nevertheless communicative dimensions, such as the layout of a document,
or the type and quality of the surface on which the text is found. In short,
discourse can reasonably be viewed as a multimedia process. Given such a
view, the advantage of a semiotic approach is that it has the potential to ac-
commodate the linguistic and non-linguistic aspects of discourse within a
single, coherent framework.
3. Discourse representation
Even now we have by no means covered everything that might be regarded
as part of discourse studies. Nevertheless, our survey certainly shows that
there are many matters to consider in designing an adequate representation
for discourse. Let us, therefore, now discuss the question of representation.
3.1. Notation in FG and FDG

Given that not all FG authors have adopted the same framework for the de-
scription of discourse, it is not surprising that there is no consensus, either,
on the notation to be employed for representing discourse or discourse
models. In the chapter on discourse in Dik (1997b) we find some hints as
to notation. Dik proposes that illocutions at levels higher than the clause
should be represented along the following lines:
(1) ILL(DISCOURSE-EPISODE)
He also adopts the convention of showing hierarchical structure by

means of indentation. A simple example devised in the spirit of this princi-
ple is the following:
(2) ENTER DISCOURSE

ENTER EPISODE 1
Sequence of speech acts
LEAVE EPISODE 1
ENTER EPISODE 2
Sequence of speech acts
LEAVE EPISODE 2
LEAVE DISCOURSE
The ENTER-LEAVE pairs act as brackets around the stretches of language

which they enclose. In the present chapter we shall call them Delimiters.
96 John Connolly
The hierarchical structure of discourse is indicated by means of horizon-

tal trees in Kroon (1997). In the outline example in Fig 1, A and B are
interlocutors in a dialogue:
A: Utterance act
move
Utterance act
exchange
B: Utterance act move
Figure 1. Example of Kroon’s style of discourse representation
Kroon also, where appropriate, labels acts and/or moves as either central or
subsidiary.
Hengeveld (1997) employs a means of representing discourse which re-
flects his idea of upward layering. However, this will not be considered in
detail here, as it has been overtaken by Hengeveld’s more recent work, to
which we shall shortly return. Suffice to say that the method of discourse
representation in Hengeveld (1997) takes the form of a vertical tree struc-
ture. Steuten's notation takes the form not of a tree but of a table; see the
fifth chapter of Steuten (1998a). A simple and slightly adapted example is
seen in Table 1, which is based on the assumption that each illocutionary
act in the discourse has previously been identified and numbered in se-
quence.
In positing a non-linguistic part of the transaction in addition to the lin-
guistic discourse, Steuten particularly has in mind business conversations,
where the transaction could include, for instance, the exchange of goods
and money.
Gulla (1997) proposes a means of representing a combination of FG and
RST with the help of a bracketed structure. The design of this representa-
tion is illustrated in the outline example in Figure 2, which is based on a
stretch of discourse where two propositions occur in sequence, and propo-
sition 2 serves to elaborate proposition 1.
Table 1. An illustration of Steuten's method of discourse representation
COMMUNICATIVE UNIT GRAMMATICAL

UNIT
Level 1: Level 2: Level 3: Level 4:
Transaction Exchange Interactional Act Illocutionary Act
Opening Greeting 1
Greeting 2
Information Question 3
Conversation Answer 4, 5
Reaction 6
Closing Valediction 7
Valediction 8
Non-linguistic
Part
RELATION elaboration
NUCLEUS proposition 1
SATELLITE proposition 2
Figure 2. Outline of an RST-style discourse representation
Vet (1998a), working within the framework of Modular Functional

Grammar, distinguishes between the speech act and the utterance which
constitutes the product of that act. The structure of an utterance is exempli-
fied in the following, where ‘u’ stands for ‘utterance’, ‘c’ for the ‘content’
of that utterance, ‘POS’ for the truth-value ‘positive’ and ‘p’ for ‘proposi-
tional content’:
(3) a. Jay likes Kay.

b. u1: [DECL (c1: [POS p1: [Pres e1: [likeV (d1x1: JayN (x1))ZeroSubj
(d1x2: KayN (x2))RefObj] (e1)] (p1)] (c1))] (u1)
Such a characterization belongs to the grammatical module. In the prag-

matic module, on the other hand, Vet accommodates the discourse activity,
together with the mental considerations which motivate it. For instance,
take the sequence:
98 John Connolly
(4) Jay likes Kay. He dates her.
The account of this in the pragmatic module would be something like that
shown in Figure 3:
WANT (S) (S inform A of the romantic association between Jay and Kay)
MOVE M1 : (S) WANT ((S) INFORM (A) OF (R))
E1 E2
ASSERT (S)(m1 )(A) ASSERT (S)(m2 )(A)
LIKE (JAY, KAY) DATE (JAY, KAY)
[L] [L]
m' 1 m' 2
Figure 3. Example of Vet’s style of discourse representation
This can be interpreted as follows. The desire which S has to pass on to

A the information about Jay and Kay’s relationship (R) leads S to make a
discourse move M1, which is split into two speech acts, E1 and E2, which
are intended to transmit messages m1 and m2 respectively to A. In the case
of m1 the conceptual content of the message is represented as LIKE (JAY,
KAY). This is expressed in the form of actual language by means of a
function denoted as L, which encapsulates the choices made by S in order
to put the content of the message into words. The output of L is denoted as
m'1, which represents the conceptual content of m1 together with the
choices just mentioned. Finally, m'1 gives rise to u1 (whose internal details
were set out earlier), which is the underlying linguistic form of the utter-
ance that expresses the content concerned. In this way, the function L
manages the interface between the pragmatic and grammatical modules.
This principle is encapsulated within the condensed formula given by Vet
(1998a: 12), namely L(m1) = u1.
Vet’s proposals are expanded and slightly modified by Connolly (1998),
who elaborates the representation of the discourse move to include the be-
liefs of the speaker that lie behind the move, and also the intended outcome
of the move, as illustrated in the following example, in which C stands for
‘content of ’:
(5) M1:
[[BELIEVE (S) (NOT (KNOW (A) C(m1, m2)))], Precondition
[WANT (S) (KNOW (A) (C(m1, m2)))], Precondition
[CONVEY (S) (m1, m2) (A)], Nucleus
[KNOW (A) (C(m1))] Postcondition

[BELIEVE (S) (KNOW (A) (C(m1, m2)))] Postcondition
[NOT (WANT (S) (KNOW (A) (C(m1))))]] (M1) Postcondition
The postconditions list the anticipated outcomes of the move. Of course,

these may or may not coincide with the actual outcomes!
The above are by no means the only proposals in the FG literature
which relate to the representation of discourse. However, they give a rea-
sonable feel for the variety of notations that have been put forward, prior to
the emergence of FDG.
In FDG the representation of discourse structure is distributed among
the three modules. Within the interpersonal module the units recognized
include the move, the act and the illocution. The formulation of the inter-
personal structure proceeds along the following lines, where M stands for
Move, A for Discourse Act, P for Participant, ILL for Illocution, C for
Communicated Content, T for Ascriptive Act and R for Referential Act:
(6) (M1: [(A1: [ILL (P1)S (P2)A (C1: [...(T1) (R1)...] (C1))] (A1))] (M1))
In dialogues the layers of Turn and Exchange are also recognized.

At the representational level the propositional content is addressed.
The formulation of structure follows the following pattern, where p stands
for Propositional Content, e for State-of-affairs, f for Property and x for In-
dividual:
(7) (p1: [(e1: [(f1) (x1)] (e1))] (p1))
Again, higher layers are recognized if appropriate, for instance Episode

and Story within the narrative genre.
100 John Connolly
The formulation of structure at the expression level is carried out along

the following lines, where Para stands for Paragraph, S for Sentence, Cl for
Clause, PrP for Predicate Phrase, RP for referential Phrase and Lex for
Lexeme:
(8) (Para1: [(S1: [(Cl1: [(PrP1: [(Lex1)] (PrP1)) (RP1: [(Lex2)] (RP1))] (Cl1))]
(S1))] (Para1))
Once more, allowance is also made for higher layers, for example Sec-
tion and Chapter in books.
3.2. Discourse Representation Structures

Outside of FG there have also been various attempts to develop methods of
representing discourse. Perhaps the best known of these have emerged
from Discourse Representation Theory (DRT); see Kamp and Reyle
(1993). Espousing from the very outset a communicative (or functional)
view of language, Kamp and Reyle develop a detailed theory of discourse
interpretation, based upon concepts drawn from model-theoretic semantics.
In their exposition they make extensive use of a type of semantic represen-
tation called a Discourse Representation Structure (DRS). A DRS, in its
most basic form, contains (i) a list of symbols, such as x or y, each of
which represents a discourse referent, (ii) a list of predicates, such as Jay(x)
or Kay(y), indicating the respective individuals for which the aforemen-
tioned symbols stand, and (iii) an indication of the relationship among the
individuals designated by the symbols (for instance, ‘x likes y’). The rela-
tionship concerned is derived from a syntactic analysis of the sentence
concerned.
x y
Jay(x)
Kay(y)
x likes y
Figure 4. DRS-style representation of the sentence Jay likes Kay
By way of notation, Kamp and Reyle combine the contents of a DRS

into a list enclosed inside a box, as shown in Figure 4. Discourses contain-
ing more than one sentence can, of course, be accommodated, as for
example in Figure 5. Kamp and Reyle go on to elaborate this notation in

various ways, which are, however, beyond the scope of the present brief
survey.
x y u v
Jay(x)
Kay(y)
X likes y
u=x
v=y
U dates v
Figure 5. DRS-style representation of the discourse Jay likes Kay. He dates her.
3.3. Segmented Discourse Representation Structures

Asher (1993) proposes a somewhat different version of DRT from Kamp
and Reyle. His version incorporates rhetorical relations, and is based
around a unit called a Segmented Discourse Representation Structure
(SDRS). A slightly adapted example is given in Figure 6.
k1 := x, s1, y k2 := u, s2, v
Jay(x) u=x
s1-like(x,y) s2-date(u,v)
Kay(y) v=y
Cause(k1,k2)
Figure 6. SDRS-style representation of the discourse Jay likes Kay. He dates her.
As will be apparent, it has been assumed here that the content of the
first sentence is the cause of that of the second.
3.4. Situation Theory

A brief mention should also be made here of Situation Theory (ST), and in
particular Devlin's (1991) version of ST, which is discussed in relation to
FG by Connolly (1998). Although not a fully-fledged theory of discourse,
Devlin's version of ST does include a means of describing the situation in
102 John Connolly
which an utterance is produced. According to Devlin, a description of this

nature will include the following information: (i) who is addressing whom,
when and where, (ii) what is being uttered and (iii) which individuals are
being referred to by means of which lexical items; see Devlin (1991: 237).
Full details of the relevant ST-style notation need not be considered here,
but a rough indication of what is involved in the description of an assertive
speech act is the following:
(9) << speaking-to, Speaker, Addressee, Location, Time, true >>

and
<< utters, Speaker, Proposition, Location, Time, true >>
and
<< refers-to, Speaker, word 1, Referent 1, word 2, Referent 2, ...,
Location, Time, true >>.
Clearly, this is a more elaborate characterisation of a speech act than is

conventional in FG, and thus merits our attention in the present context.
4. A framework for discourse representation

From what has been said so far, it is clear that the representation of dis-
course involves coping with a large amount of information. We are also
faced with two conflicting goals. On the one hand, we would wish our rep-
resentation to provide an adequate coverage of discourse phenomena. On
the other hand, we would want our notation to be easy to read, as well as
being straightforward and quick to produce. What is presented below is a
compromise between these desiderata.
The problem of representation is complicated by the fact that discourse
has more than two dimensions. We need to accommodate temporal (or else
spatial) sequencing, hierarchical structure, different levels of analysis and
also relational phenomena (in particular adjacency sequences and rhetorical
relations). All these need to be projected onto the two-dimensional page.
It is fairly natural to represent temporal sequencing as progressing verti-
cally down the page, and it is convenient to organize the levels of analysis
on the horizontal dimension across the page, with each level being con-
tained in a separate column, while maintaining a horizontal alignment
between entries which correspond to one another. In fact, it is proposed
that we employ a series of columns, to accommodate each of the following:
(10) a. The contextual level

b. The interpersonal level
c. The representational level
d. The expression level
As will be immediately apparent, we are here honouring the distinction

between the discourse itself and the context in which it is situated. This de-
cision is to some extent motivated by Dik's distinction between discourse
proper and discourse event. It will be recalled that the phenomena which
Dik includes in the analysis of the discourse event are: who the interlocu-
tors are, how they relate to one another, where, when and in what
circumstances they are communicating, and what social conventions are in
force. These considerations are all matters of context, and therefore if we
are to accommodate them, we need to recognize a contextual level of de-
scription. The contextual level is not actually part of language itself, but is
intended to be seen as a component of the description of discourse.
It may be noted that the idea of a contextual level has echoes in other
frameworks besides FDG. For example, Hasan (1999: 224), writing within
the framework of Systemic Functional Grammar, proposes that in describ-
ing language we should recognize four strata: context, semantics, lexico-
grammar and phonology. Moreover, Stamper (1997), in his discussion of
applied semiotics, puts forward a six-level model in which the social world
is recognized as a stratum which encompasses much of the context in
which human communication takes place; see further Connolly (2000).
In fact, the contextual level turns out to be very useful, as it provides a
natural home for a number of factors that need to be accommodated within
a full account of discourse. These include (where relevant):
(11) a. The identity of the participants, their attitudes, statuses and back-
grounds;
b. Any other sources of voices which blend into the discourse;
c. Bystanders;
d. The non-linguistic properties of the medium or media of communica-
tion employed;
e. The non-verbal communication;
f. The time and place;
g. The setting;
h. The prevailing cultural, social and institutional conventions governing
communication;
i. The referents of terms used in the discourse;
j. The pre- and postconditions and outcomes of particular discourse acts.
104 John Connolly
As may be apparent, the above list strongly echoes the work of Firth (1957:
203) and Hymes (1972).
Let us illustrate this with the help of the following brief excerpt from a
conversational dialogue:
(12) Hugh (making move M101): ( = P1)

Jay likes Kay. ( = Message m151)
He dates her. ( = Message m152)
Irene (making move M102): ( = P2)

Gosh! ( = Message m153)
Ella told me that he didn’t go for tall women. ( = Message m154)
(Ella = AV1, i.e. additional
voice 1)
A possible contextual description of this interactive discourse could be

the following, which contains information (that will here be assumed to be
true) along the lines suggested above:
(13) P1 = Hugh
P2 = Irene
AV1 = Ella
All have equal social status
All are from the USA
All have conventional American social attitudes towards romantic involve-
ments
No bystanders
Participants are using normal phonation
Participants are smiling
Time = 09.40 p.m., 4 February 2000
Place = Irene’s apartment in Kansas City
Setting is an informal dinner party
Prevailing communicational conventions are those of English-speaking
Americans
Referents of:
x251 (JayN)
x252 (KayN)
x253 (EllaN)
x254 (womanN)
Preconditions(M101) =
BELIEVE (P1 NOT (KNOW (P2 C(m151))))
WANT (P1 KNOW (P2 C(m151)))
Postconditions(M101) =
KNOW (P2 C(m151))
KNOW (P2 C(m152))
BELIEVE (P1 KNOW (P2 C(m151)))
NOT (WANT (P1 KNOW (P2 C(m151))))
Outcome(M101) = Postconditions(M101)
KNOW (P1 C(m153))
KNOW (P1 C(m154))
Clearly, any contextual description will have to be selective. There is no

possibility of including every fact about the situation. The basic principle is
to include what one feels to be relevant, insofar as one’s knowledge and the
evidence permits. Thus, our example is intended merely to provide an idea
of how a contextual description might look.
One particular point to note is how the multimedia aspects of discourse,
including non-verbal communication, have been allowed for in the contex-
tual level. Admittedly, this probably does not go far enough to satisfy the
point made by Kress and Van Leeuwen about the multimodal nature of
text, but it is, hopefully, a step in the right direction, as it explicitly incor-
porates multimedia phenomena into the FDG-related framework.
106 John Connolly
The description of the discourse proper begins at the interpersonal level.

Internally the interpersonal level has a hierarchical structure, which can be
represented with the help of indentation, somewhat after Dik (1997b: 423).
The units in the hierarchy will be taken to be those of Sinclair and
Coulthard, given that many of these have been adopted somewhere in the
FG literature, and some of them are already recognized in FDG.
By way of illustration, an interpersonal level description of the same
dialogue is the following:
(14) ENTER DISCOURSE INTERACTION D1

Genre = conversation
Style = informal, unremarkably polite, concise
Ideology = neutral
Discourse topic = miscellaneous idle chatter
Discourse tense = present
...
ENTER TRANSACTION Tr5
Topic of transaction = gossip about mutual acquaintances
ILL(Tr5) = representative
...
ENTER EXCHANGE Ex18
Topic of exchange = Jay and Kay’s romantic association
ENTER MOVE M101
PSp = P1, PAddr = P2
M101: [CONVEY (PSp) (m151, m152) (PAddr)] (M101)
ENTER ACT A151
A151: [ASSERT (PSp) (m151) (PAddr)] (A151)
ILLS(A151) = ILLA(A151) = representative
C(m151) =
REFER(x251, x252)
ASCRIBE(f201) AS PER p171
LEAVE ACT A151
ENTER ACT A152
C(m152) =
REFER(x251, x252)
A152 = Elaboration(A151)
LEAVE ACT A152
LEAVE MOVE M101
ENTER MOVE M102

M102: [CONVEY (PSp) (m153, m154) (PAddr)] (M102)
ENTER ACT A153
A153: [EXCLAIM (PSp) (m153) (PAddr)] (A153)
ILLS(A153) = ILLA(A153) = expressive
Attitude(A153) = surprise
A153 = Acknowledgement(A151, A152)
LEAVE ACT A153
ENTER ACT A154
CONTAIN(A154, BLEND(PAddr, AV1))
C(m154) =
REFER(x253, P2, x251, x254)
ASCRIBE (f203) AS PER p173
ASCRIBE (f204, f205) AS PER p174
LEAVE ACT A154
LEAVE MOVE M102
LEAVE EXCHANGE Ex18
LEAVE TRANSACTION Tr5

...
LEAVE DISCOURSE INTERACTION D1
The description is set out in a manner derived through combining the

practice of several different FG authors, but organized in terms of the Sin-
clair and Coulthard categories. Following the opening delimiter (‘ENTER
DISCOURSE INTERACTION D1’), we have a specification of various
overall attributes of the discourse, such as the genre, the style, the ideology
and the overall topic. For example, the style is informal (witness the inclu-
sion of the interjection gosh), unremarkably polite (in that it is courteous
but in no way obsequious) and concise (being composed of short sen-
tences), while it is ideologically neutral in that it does not contain any overt
marker of, for instance, feminism, capitalism, and so on. Next comes the
sequence of transactions, of which our example happens to show the fifth,
i.e. Tr5. The details of the latter are placed between its opening delimiter
(‘ENTER TRANSACTION Tr5’) and its closing delimiter (‘LEAVE
TRANSACTION Tr5’); we shall return to these particulars in a moment. At
the end of the sequence of transactions comes the final closing delimiter
108 John Connolly
(‘LEAVE DISCOURSE INTERACTION D1’), which terminates the dis-

course.
The opening delimiter of the transaction in our example is followed by a
statement of the topic of the transaction concerned. After that we have the
sequence of exchanges, of which our example shows the eighteenth, i.e.
Ex18, which is enclosed between the delimiters ‘ENTER EXCHANGE
Ex18’ and ‘LEAVE EXCHANGE Ex18’. The opening delimiter is followed
by a statement of the topic of the exchange in question, after which are set
out the series of moves of which the exchange consists. As the illustrative
discourse is a dialogue, it is necessary to begin the description of each
move with a specification of who is the speaker (PSp in Hengeveld’s nota-
tion) and who is the addressee (PAddr). For instance, in the case of move
M101, the speaker is shown to be participant P1 and the addressee the par-
ticipant P2. Next comes a statement of the essence of this move, which is
for the speaker to convey to the addressee a series of two messages, i.e.
m151 and m152. This is accomplished by means of acts A151 and A152. (An
act is abbreviated A in Hengeveld’s notation.)
The description of an act requires at least three lines. A simple example
is found in act A151. Firstly, we need a specification of the essence of the
act, which in this case is that the speaker asserts a particular message (m151)
to the addressee. Secondly, we need a statement of the illocutionary force
of the act, for which purpose in the present paper we employ Searle’s
(1976) categories. In the current example, the illocutionary force of the act
from the point of view of both the speaker and the addressee is ‘representa-
tive’. Thirdly, the discourse processes of reference and ascription are
acknowledged. The line ‘REFER(x251, x252)’ states that the individuals x251,
x252 (which are primarily constituents of the representational level, and
given lexical shape as JayN and KayN at the expression level) are referring
to certain phenomena in the context. These referents are enumerated in the
contextual description. Thus, the interactional level serves as a bridge be-
tween language and context, and hence fulfils the function of the pointers
mentioned in Connolly (1998: 184).
The line ‘ASCRIBE (f201) AS PER p171’ serves to assign the property
f201 (another constituent of the representational level, given lexical shape as
likeV at the expression level) to the individuals x251 and x252, in the manner
indicated in the representational level proposition p171. This proposition is
shown in outline form in (15a), with the corresponding expression-level
structure (up as far as the clausal layer, and in outline form) being given in
(15b), while the outline propositions and expression-level structures for the
second, third and fourth acts in the exchange are shown in (16–18). Formu-
lae such as ‘AS PER p171’ are employed in order to avoid having to dupli-
cate, at the interpersonal level, information that needs in any case to be
stated at the representational level.
(15) a. (p171: [(e171: [(f201) (x251) (x252)] (e171))] (p171))

b. (Cl171: [(PrP231: [(likeV)] (PrP231)) (RP281: [(JayN)] (RP281)) (RP282:
[(KayN)] (RP282))] (Cl171))
(16) a. (p172: [(e172: [(f202) (x251) (x252)] (e172))] (p172))

b. (Cl172: [(PrP232: [(dateV)] (PrP232)
(RP283: [(hePron)] (RP283)) (RP284: [(herPron)] (RP284))] (Cl172))
(17) a. NO PROPOSITIONAL CONTENT

b. Gosh
(18) a. (p173: [(e173: [(f203) (x253) (p174: [(e174: [NOT(f204) (x251) (x254: (f205)
(x254))] (e174))] (p174)) (P2)] (e173))] (p173))
b. (Cl173: [(PrP233: [(tellV)] (PrP233)) (RP285: [(EllaN)] (RP285))
(Cl174: [(PrP234: [(not goV for)] (PrP234))
(RP286: [(hePron)] (RP286)) (RP287: [(womanN: tallA)] (RP287))] (Cl174))
(RP288: [(mePron)] (RP288))] (Cl173))
Other information can also be included within the description of an act.

First of all, rhetorical relations can be specified, as for instance in act A152,
which is described as an elaboration of the preceding act, by means of the
line ‘A152 = Elaboration(A151)’. Relationships within adjacency sequences
can also be specified. For example, within the description of act A153, the
final line, ‘A153 = Acknowledgement(A151, A152)’, shows the act in question
to be an acknowledgement of the acts comprising the previous move. In
addition, the adoption of a particular attitude can be indicated. This, too, is
exemplified in act A153 (corresponding, as it does, to the interjection Gosh),
which is shown to be marked by surprise, through the inclusion of the line
‘Attitude(A153) = surprise’.
There are quite a number of points to be made in relation to the form of
discourse representation just presented. Let us begin with some matters of
notation. Firstly, it is assumed that every type of unit in the hierarchy is
numbered in sequence, starting from the opening of the discourse. Hence,
for instance, Move M1 would be the first move in the entire interaction, and
not just within the transaction of which it forms part. Secondly, no use has
been made of boxes, in contrast to the DRT notation. This is in order to
make the representation quicker to produce. Thirdly, the notation as exem-
110 John Connolly
plified above is not intended to be overly prescriptive. The discourse ana-

lyst is expected to include what is relevant to his or her purposes. The
notation as exemplified above is intended to be fairly explicit, and in some
circumstances it would doubtless be possible to omit some of it as being
obvious, although this is less likely if the notation is being used in the con-
text of Computational FG. In any case, though, it is quite acceptable to
omit information that is not relevant to the task in hand, or alternatively to
add further information if necessary. For instance, if it desired to label acts
as ‘central’ or ‘subsidiary’, after the fashion of Kroon (1997), then this can
easily be done (by means of lines like ‘ENTER MAJOR ACT301’ or ‘EN-
TER SUBSIDIARY ACT302’, or by some suitable alternative). Another
point of clarification concerns the units within the interaction. It is assumed
that each member of the hierarchy (apart from the act) consists of a se-
quence of one or more units of the rank below. Thus, for instance, a move
can be composed of just a single act.
Let us now consider the coverage of the interpersonal level description.
The latter is constructed around the discourse interaction, revealing the hi-
erarchical structure of the latter, but it also accommodates statements about
attitudes, for example ‘Attitude(A153) = surprise’. It incorporates broad-
scope parameters such as discourse topic, discourse tense and large-scale
illocutionary force, as well as recognising ILLA, ILLS and their values. In
addition, it accommodates both adjacency sequences (for instance ‘A152 =
Elaboration(A151)’) and rhetorical relations (for example ‘A152 = Elabora-
tion(A151)’).
It should be noted that the overall framework accommodates certain
discourse-related concepts that are not strongly associated with the FG tra-
dition. These include ideology (alluded to in the above example by the
statement ‘ideology = neutral’), as well as voice in Bakhtin’s sense of the
term, reflected in the statements ‘AV1 = Ella’ in the contextual description
and ‘CONTAIN(A154, BLEND(PSp, AV1))’ in the interpersonal level rep-
resentation. Following the example of authors in, for example, DRT and
Situation Theory, the referring elements are picked out (for instance by
means of the statement ‘REFER(x251, x252)’) in the interpersonal level de-
scription, which have corresponding entries at the contextual level (in this
case ‘Referents of: x251 (JayN), x252 (KayN)’). Note that the words in brack-
ets here are intended merely as glosses for predicates which would
normally be incorporated, instead, into the representational level and ex-
pression level description. All the other information in Devlin’s
characterisation of a speech act, as illustrated earlier, also finds a place in
one or other of the levels of description envisaged here.
How do the present proposals compare with those of Hengeveld (this

volume)? On the one hand, Hengeveld’s treatment of the interpersonal
level of discourse is more concise, and will no doubt suffice for many cir-
cumstances in which discourse description is not the primary topic of
interest. On the other hand, the present proposals are more appropriate
when a detailed treatment of discourse is required, particularly since they
include phenomena such as adjacency relations, which are omitted from
Hengeveld’s (admittedly non-exhaustive) presentation.
There is another significant difference between the two sets of propos-
als, which concerns the recognition or otherwise of structures such as
paragraphs above the sentence layer. The stance tentatively taken here is
that paragraphs can be seen as the expression-level counterpart of moves.
Similarly, chapters and sections can be seen as the expression-level foot-
prints of interpersonal-level units; thus a chapter expresses a transaction
and a section expresses what we may term a sub-transaction. In this way,
the hierarchical structure of discourse can be simply attributed to the inter-
personal level, in which case its expression at lower levels is purely
derivative, and therefore does not really need to figure in the lower levels
of the linguistic description, where it adds nothing fresh. This stance ac-
cords with Verschueren's (1999: 130) view that there are no purely
structural units above the sentence. It also simplifies the description, in
dispensing with what Dik calls the organisational level. In these respects,
then, it seems a preferable alternative to other approaches. However, to re-
peat, this proposal is tentative and will need to be examined more fully on
another occasion. If it should prove to be untenable, then there would be no
inherent difficulty in incorporating the higher-ranking structures into the
expression level description.
Finally, we need to consider how the four levels of description all fit to-
gether. Ideally, it would be satisfying to set the discourse description out in
four parallel columns. However, limitations of space may well prevent that,
in which case we may set out the contextual and interpersonal levels along-
side each other as in Table 2, and then deal with the other two levels as in
Table 3. The first column of Table 3 allows for a cross-reference to each
act, which is dealt with in more detail in Table 2. In the second column of
Table 3 we place the representational-level description of the propositional
content of the act concerned, and in the third column its expression-level
description.
112 John Connolly
Table 2. Illustrative excerpt from a multi-level discourse description: contextual

and interpersonal levels
Contextual Description Interpersonal Description

ENTER MOVE M101

M101: [CONVEY (PSp) (m151, m152)
(PAddr)] (M101)
ENTER ACT A151
A151: [ASSERT
(PSp) (m151) (PAddr)] (A151)
ILLS(A151) = ILLA(A151)
= representative
ILLS(A151) = ILLA(A151)
= representative
C(m151) =
REFER(x251, x252)
LEAVE ACT A151
ENTER ACT A152
A152: [ASSERT
(PSp) (m152) (PAddr)] (A152)
ILLS(A152) = ILLA(A152) =
representative
C(m152) =
REFER(x251, x252)
LEAVE ACT A152
LEAVE MOVE M101
KNOW (P2 C(m151))
KNOW (P2 C(m152))
BELIEVE( P1 KNOW (P2 C(m152)))
Table 3. Illustrative excerpt from a multi-level discourse description: representa-

tional and expression levels
Interpersonal Representational Expression

Act Description Description
A151 p171 Cl171
e171
f201 Prp231
likeV
x251 RP281
JayN
x252 RP282
KayN
The above is offered as a feasible way of representing discourse for the

purposes of FDG. As it stands, for the sake of readability it does not em-
ploy a fully formal notation, though if necessary, for instance if it were to
be incorporated into a computing system, a properly formal version could
be developed from it.
5. Conclusion
In this chapter a framework for the representation of discourse has been
proposed which is intended to accommodate the main phenomena of inter-
est to analysts and to be compatible with FDG. An important feature of the
representation is that it treats context as a separate level of description. Un-
doubtedly the framework stands in need of further development, but it does
hopefully serve to indicate how the various and numerous phenomena of
discourse can be seen to fit together into a coherent whole.
References
Asher, Nicholas
1993 Reference to Abstract Objects in Discourse. Dordrecht: Kluwer.
1998 What to do with topic and focus? Evaluating pragmatic information.
In: Mike Hannay and A. Machtelt Bolkestein (eds), 193–214.
Clark, Katerina and Michael Holquist
1984 Mikhail Bakhtin. Cambridge MA: Harvard University Press.
114 John Connolly
Connolly, John H.
1998 Information, situation semantics and Functional Grammar. In: Mike
Hannay and A. Machtelt Bolkestein (eds), 167–189.
2000 Accommodating natural language within the organisational semiotic
framework. Proceedings of the Third International Workshop in Or-
ganisational Semiotics, 1–13. Stafford: University of Staffordshire.
Connolly, John H., Anthony A. Clarke, Steven W. Garner and Hilary K. Palmén
1997a A functionally oriented analysis of spoken dialogue between indi-
viduals linked by a computer network. In: John H. Connolly, Roel
M. Vismans, Christopher S. Butler and Richard A. Gatward (eds),
33–58.
ward (eds)
1997b Discourse and Pragmatics in Functional Grammar. Berlin and New
Coulthard, Michael
1977 An Introduction to Discourse Analysis. London: Longman.
Devlin, Keith
1991 Logic and Information. Cambridge: Cambridge University Press.
Dik, Simon C.
Fairclough, Norman
1989 Language and Power. London: Longman.
Fairclough, Norman and Ruth Wodak
1997 Critical discourse analysis. In: Teun van Dijk (ed.), Discourse as So-
cial Interaction. London: Sage, 258–284.
Firth, John R.
1957 Papers in Linguistics 1934–1951. London: Oxford University Press.
Gulla, Jon Atle
1997 Combining Functional Grammar and Rhetorical Structure Theory for
discourse representation. In: Connolly et al. (1997b), 75–89.
Hancher, Michael
1979 The classification of co-operative illocutionary acts. Language in
Society 8: 1–14.
delphia: Benjamins.
Hasan, Ruqaiya
1999 Speaking with reference to context. In: Mohsen Ghadessy (ed.), Text
and Context in Functional Linguistics, 219–321. Amsterdam and
Hengeveld, Kees
1997 Cohesion in Functional Grammar. In: Connolly et al. (1997b), 1–16.
This vol. The architecture of a Functional Discourse Grammar, 1–21.
Hymes, Dell
1972 Models of the interaction of language and social life. In: John J.
Gumperz and Dell Hymes (eds), Directions in Sociolinguistics: The
Ethnography of Communication, 35–17. New York: Holt, Rinehart
and Winston.
Kamp, Hans and Uwe Reyle
1993 From Logic to Discourse: Introduction to Modeltheoretic Semantics
of Natural Language, Formal Logic and Discourse Representation
Theory. Dordrecht: Kluwer.
Kress, Gunther and Theo van Leeuwen
1996 Reading Images: The Grammar of Visual Design. London: Rout-
ledge.
Kroon, Caroline
Liedtke, Frank
Mann, William C. and Thompson, Sandra A.
1987 Rhetorical structure theory: A framework for the analysis of texts.
Papers in Pragmatics 1: 79–105.
Martin, Jim R.
1992 English Text: System and Structure. Amsterdam and Philadelphia:
Benjamins.
Searle, John
1976 The classification of illocutionary acts. Language in Society 5: 1–24.
Stamper, Ronald
1997 Organisational semiotics. In: John Mingers and Frank Stowell (eds),
Information Systems: An Emerging Discipline?, 267–283. London:
McGraw-Hill.
Steuten, Ans A. G.
1998a A Contribution to the Linguistic Analysis of Business Conversations
within the Language/Action Perspective. Delft: Steuten.
116 John Connolly
1998b Structure and coherence in business conversations. In: Mike Hannay

Van den Berg, Marinus
Verschueren, Jef
1999 Understanding Pragmatics. London: Arnold.
Vet, Co
1998a The multilayered structure of the utterance: about illocution, modal-
ity and discourse moves. In: Mike Hannay and A. Machtelt Bolke-
1998b Epistemic possibility in the layered structure of the utterance. In:
Mike Hannay and A. Machtelt Bolkestein (eds), 149–165.
Focus of attention in discourse
Francis Cornish
1. Introduction
My goal in this chapter is to compare two functionalist theories of indexi-
cal reference in discourse in terms of topichood, seeking to critically
evaluate each both in its own terms and in terms one of the other.1 Ulti-
mately, I aim to show where each theory may usefully illuminate the other,
resulting in a more adequate overall account of this complex and still not
fully understood area of language use.
I will begin by characterizing in Section 2 the North-American Colum-
bia School of Linguistics’ account of what it calls the Focus and Deixis
systems of attention concentration in discourse, and will then re-examine in
Section 3 the four Topic subtypes recognized in standard FG (Dik 1997a:
Ch. 13; also Hannay 1985a, 1985b, 1991; Bolkestein 1998, 2000;
Mackenzie and Keizer 1991; Siewierska 1991: Ch. 6; Gómez-González
2001) in the light of the earlier characterization of these systems. Finally,
Section 4 briefly discusses how the conception of topic (and focus) as-
signment which emerges from this comparison might be integrated into the
new Functional Discourse Grammar (FDG) model put forward by Kees
Hengeveld (this volume).
In terms of this new Functional Discourse Grammar framework, the as-
signments of Topic and Focus functions which I shall be discussing clearly
fall within the Interpersonal Level recognized within this format – the na-
ture of these pragmatic functions arguing strongly in favour of the top-
down relationship between the three descending levels of (respectively) In-
terpersonal, Representational and Expression, which structure the new
model.
A brief word, first of all, on the Columbia School (CS), which is rela-
tively little known as such in Europe (even though the work of several CS
118 Francis Cornish
linguists is individually quite well known: the work of Robert Kirsner on

the Dutch demonstratives, of Erica García on the Spanish clitic pronoun
system, or of Ricardo Otheguy on the Spanish articles, to take but three ex-
amples). The Columbia School, founded by the late William Diver, is a
functionalist sign-based theory which rests on the twin pillars of Saus-
surean structuralism and the primacy of the communicative function of
language. It is an avowedly inductive theory, suspicious of a priori catego-
ries, notions and constructs. All such theoretical devices must be
demonstrated to be necessary on the basis of evidence of patterning in ac-
tual texts (there are shades of FG here – though CS adheres much more
strictly to this principle). CS linguists seek out the existence of micro-
systems such as those of tense, aspect, mood, number, case, the demonstra-
tives, and so on, in terms of Saussurean constellations of values, or so-
called signal-meaning hypotheses: the existence of systematic oppositions
of forms correlates with systematic oppositions in meaning, or sense (signi-
fied in Saussurean terms) within the micro-system.
From the user’s point of view, an actual message is fleshed out on the
basis of a given set of values expressed via certain signals belonging to the
systems at issue, in conjunction with relevant features of the context
(cotext and context-of-utterance – including the speaker/writer’s hypothe-
sized intentions in producing the utterance in question). That is, as in
Sperber and Wilson’s Relevance Theory of understanding in context, as
also in standard FG and the new FDG model, the formal structure of an ut-
terance (CS does not recognize the traditional category sentence) always
underdetermines the message(s) which it can serve to convey, given par-
ticular contexts. The gap between (structural) meaning (or signified), based
on given networks of formal and semantic oppositions, and message, has to
be bridged via inferences on the part of the language user.
CS linguists operate very much at the frontier between clausal structure
and textual product, in the sense that they make heavy use of text counts of
given features and their combinations, leading to statistically significant
tendencies or skewings across texts from different genres. These skewings,
once apparent, are then used as empirical validations of the hypotheses set
up to describe and account for the behaviour of given forms in context. The
introduction by Ellen Contini-Morava to Contini-Morava and Goldberg
(1995) gives a very good overview of CS theory.
Focus of attention in discourse 119
2. The systems of Focus and Deixis in CS
2.1. Focus
The CS conception of Focus should not be confused with the sense of Fo-
cus in FG terms in the context of so-called Information Structure
(henceforth IS). See Gundel (1999) and Cornish (1999: §5.3.2, 2000a:
§2.2) for clarification of the relationship between these two senses of ‘Fo-
cus’ in the literature. The reader will need to be alert to this in what
follows, and to keep his or her eye firmly on the terminological ball. To
keep the two intended senses of ‘focus’ distinct, I will capitalize this term
throughout when referring to the CS conception, and will use only initial
capitalization for the FG one.
FOCUS in CS terminology does not mean ‘the most important informa-
tion2 conveyed by a clause, representing the difference between P(A)S (the
speaker’s representation of the addressee’s current state of pragmatic
knowledge) and (PS) (the speaker’s own current state of pragmatic knowl-
edge)’, as in the standard FG account, but rather ‘what the speaker is
assuming the hearer is already concentrating on, what s/he has at the fore-
front of his/her consciousness at the time of utterance’. This is therefore
more akin to the FG notion of Topic than of Focus. Only participants (ar-
guments) with respect to the Event represented in a clause fall within the
CS FOCUS systems. The CS value NOT IN-FOCUS is then (confusingly)
equivalent by default to the information-structural sense of Focus – al-
though there is no prediction within CS theory that constituents bearing
this value will in fact have Focus status (in the Information Structure
sense). Wallis Reid (pers. comm.) confirms my understanding that IN-
FOCUS status is in fact equivalent to that of Topic in the IS sense, and that
NOT IN-FOCUS equates with non-topical. He goes on to point out that the
value NOT IN-FOCUS only applies to participants that could have been
IN-FOCUS (i.e. clausal topics). The instruction conveyed to the under-
stander via a constituent bearing the value [IN-FOCUS] (the case of the
term Jane in (1a) below, under its topic-comment interpretation) is ‘pay
more attention to the referent at issue, as it will be important in the subse-
quent discourse’, and via one carrying the value [NOT IN-FOCUS] (a
tarantula in (1a), and, presumably, the entire sentence in (1b)), ‘pay
somewhat less attention to the referent involved’.
120 Francis Cornish
(1) a. Jane [IN-FOCUS] has just spotted a tarantula [NOT IN-FOCUS]

(categorical utterance: ‘predicate-focus’ in Lambrecht’s 1994 terminology).
b. A bomb exploded yesterday morning in Armington Valley high street.
(thetic utterance: ‘sentence focus’ pace Lambrecht 1994)
I am assuming here that in the immediate context of (1a), the discourse has
been dealing with Jane, that is, that this referent is topical in Dik's (1997a)
sense of the term. Thus, (1a) should be understood as a reply to the question
What did Jane do?, or What about Jane?, where the IS Focus would corres-
pond to the content of the VP has just spotted a tarantula. Alternatively,
(1a) could be a response to the question What has Jane just seen/spotted?,
where the IS Focus would correspond to the second argument a tarantula
alone (‘Jane has just seen/spotted something’ being presupposed, following
Erteschik-Shir 1997: 13). (1a) may be analysed as an instance of the classic
‘topic-comment’ (categorical) information structure, since a relation of
‘aboutness’ may easily be attributed to the referent of the subject expression
and to the denotation of the predicator-cum-second argument (given the
type of context indicated above), as expressed in (2a):
(2) a. As for/About Jane, she has just spotted a tarantula.

b. #As for/About a tarantula, it has just been spotted by Jane.
(2b), which is very unnatural as a paraphrase of the relevant part of (1a),

clearly shows that the second argument a tarantula does not bear the topic
(i.e. [IN-FOCUS]) relation with regard to the remainder of the utterance.3
Two well-known problems with assessing written, decontextualized
sentences of the type in (1a) are first, precisely the lack of a context which
would enable the assignment of a given information-structural analysis on
the basis of that context; and second, the lack of an explicit prosodic struc-
ture which would motivate that assignment. This latter point makes itself
felt particularly acutely in English.
(1b) as a thetic utterance does not presuppose any particular prior con-
text; as such, it could occur discourse-initially. CS would not assign IN-
FOCUS status to the subject term a bomb here, since the predication is in-
transitive. As such, it would be assigned the value ‘MORE FOCUS
(needed)’ (see below), in comparison to inverted constructions where the
subject term is postverbal. The fact that the value ‘MORE FOCUS
(needed)’ is assigned to this referent correctly predicts that it is not (yet)
topical: for here there is no ‘aboutness’ relation involved between its refer-
ent and the proposition; the entire utterance is thetic (presented as ‘all-new’
information to the addressee) – an event-reporting utterance in Lam-
brecht’s (1994) terms.4 That the argument expression a bomb, in subject
position, does not bear the topic, or an [IN-FOCUS] relation with respect to
the remainder of the utterance, is shown by (3):
(3) #As for/About a bomb, it exploded yesterday morning in Armington Valley

high street.
However, were this utterance to correspond to a transitive sentence (as in A

bomb destroyed the Armington Valley city centre yesterday morning), the
value IN-FOCUS would then be assigned to the referent of the subject
term, thereby incorrectly predicting it to be the topic of that utterance.
Now, there is an ambivalence in the conception of FOCUS within the
CS framework: on the one hand, an item signalled as IN-FOCUS repre-
sents information which the speaker is assuming the addressee is already
concentrating on; and on the other, the addressee is being tacitly instructed
to maintain the concentration differential between this IN-FOCUS item and
other NOT IN-FOCUS items, for the immediately ongoing discourse. The
following quotation from Huffman (forthc.: 31) reflects this ambivalence:
The system of Focus deals with the centering of attention on one of the par-
ticipants in an event. The opposition here is a binary one: a participant is
either IN FOCUS or NOT IN FOCUS. A participant IN FOCUS is thereby
stated to be the center of attention with respect to the particular event, and
when a participant is signalled to be NOT IN FOCUS, the speaker is
asserting that this one should not have the hearer’s attention focused on it.
(Huffman forthc.: 31 [emphasis mine])
Although for convenience, illustrations of the FOCUS system are often

given in the context of a single clause, CS linguists point out that it can only
really be detected within a complete text, since it is the relation of partici-
pants to the wider discourse which the participant FOCUS values
instantiate.5 The Columbia School system of FOCUS has two basic values,
then, restricted to encoded differential attention contrasts solely with respect
to the arguments of finite verbs:6 for canonical transitive predications (in FG
terminology, A1 subject - predicator - A2 object), IN-FOCUS for the A1 sub-
ject, and NOT IN-FOCUS for the A2 object (Reid 1991:178). The latter
value (NOT IN-FOCUS) would also be attributed to a third participant (i.e.
an argument which is nuclear, that is, not preceded by a preposition), as in
122 Francis Cornish
John bought Mary a book. The predicator is not involved in the CS FO-
CUS system, only argument expressions (i.e. ones denoting participants)
are. As suggested above, this distinction would appear to hold only in the
case of so-called categorical utterances, where the utterance as a whole may
be construed as conveying information about the entity denoted by the topic
(here subject) expression – although CS linguists do not appear to draw the
distinction between categorical and thetic utterances.7 All entities coded as
non-participants (i.e. non-nuclear arguments, as well as satellites, in FG
terminology) – e.g. via a prepositional phrase – are ipso facto presented as
not entering the FOCUS system at all. In languages possessing case-marked
clitic pronouns, nominative clitics signal HIGHER-FOCUS, while non-
nominative clitics signal LOWER-FOCUS.
As already mentioned, where there is only one participant, in an intran-
sitive clause, for example, the two values are said to be MORE FOCUS for
the participant in preverbal subject position, and LESS FOCUS when it is
postposed. The contrast here is then between two construction types, not
one between participants (arguments) within the same predication. That is,
these word order signals form an oppositional micro-system; see in particu-
lar Huffman (1993) and Reid (1991: 180–181, ex. (29)), for justification
and practical illustration based on attested texts. (4a,b) from Reid (1991)
illustrate:
(4) a. A fly (MORE FOCUS) was swimming in my soup. (Reid 1991: ex. (26))
b. There was a fly (LESS FOCUS) in my soup. (Reid 1991: ex. (27))
Note that both (4a) and (4b) may be analysed as thetic utterances, not as
categorical ones. Reid’s annotation here goes some way towards implying
this in (4a), since the assignment of ‘MORE FOCUS (needed)’ (as in the
case of the term the bomb in example (1b) above) entails that the referent at
issue is not already IN-FOCUS (i.e. topical). As such, neither occurrence
of the NP a fly in (4a) and (4b) would be construed as a topic expression,
both of them necessarily carrying a high level of pitch-accent.8 Now, while
the analysis ‘LESS FOCUS (needed)’ may be plausible as an interpretation
of a fly in (4b), that of ‘MORE FOCUS (needed)’ is not evidently so in the
case of the token functioning as subject in (4a), since the thetic interpreta-
tion of the containing utterance here treats the referent of this expression,
not as an entity having a separate existence, but as an integral part of the
situation conveyed as a whole (see for example Siewierska 1991: 161–162,
who says of her comparable example (18b) A tiger chased a tourist: “...in
using (18b) the primary purpose of the speaker is not to present the refer-
ents but to establish the event”; see also on this point Hannay 1991: 146).
In fact, (4b) involves not just inversion, but also so-called there-
insertion,9 a specifically ‘existential’ construction. In this connection, Bir-
ner and Ward (1998: 102–106) argue that there is a discourse-pragmatic
constraint affecting the postverbal NP in this construction, to the effect that
its referent must not be ‘hearer-old’ (i.e. assumed to be already famil-
iar/known to the hearer); no such constraint is claimed to regulate the
canonical word-order variant, as seen in (4a), however. This restriction
may be seen to stem from what Hannay (1985a: 101) argues is the raison
d'être of presentative constructions of this kind, namely to “assert the exis-
tence of an entity or state of affairs in the discourse world”. Siewierska
(1991: 161–162) distinguishes between the two construction types also
within an FG framework in suggesting that sentences of type (4a) serve to
report a bare event, while existential ones, as in (4b), function to introduce
or present and highlight a specific entity (or alternatively, a State of Affairs
as a whole). Huffman (1993) presents the three construction types as a
three-way decreasing scale of FOCUS marking: subject-verb order confer-
ring a high level of attention focus on the referent of the subject term, that
of verb-subject a lower or mid level, and the there + subject + extension (in
Hannay's 1985a analysis), the lowest level of the three.
(5b) is a naturally-occurring example, contrasted with (5a), which is
adapted from it. The immediate left-hand and right-hand co-text of (5b) is
given under (6a) and (6b), respectively. Moreover, the immediate macro-
topic of the segment in which (6a), (5b) and (6b) occur is ‘the city of Anti-
bes’.
(5) a. ...the Picasso museum in the old Grimaldi château is of greater interest.
b. ...of greater interest is the Picasso museum in the old Grimaldi châ-
teau.(Extract from The Holiday Which? Guide to France by A.Ruck
1982: 112)
(6) a. ...In the old town [of Antibes] the cathedral, like so many others, has
an ill-lit Bréa; (4b).
b. ...(6a) + (5b) The rest of Antibes is a modern and bustling resort...
In comparing (5a) and (5b) as occurring in the left- and right-hand contexts
specified in (6a) and (6b), respectively, it would seem that for the subject-
predicate ordering as in (5a) to be fully coherent and natural in this context,
the adversative connective but would be expected preceding it. The force
of this connective is in fact implicit in the original predicate-subject order-
124 Francis Cornish
ing as in (5b), the attested version, but the surface presence of this item is
not essential in such a context. I would suggest that the reason for the
choice of predicate-subject ordering in (5b) is entirely due to the need to
relate (5b) to (6b), its immediate context, via a relation of comparative con-
trast. The very fact of mentioning a feature of a town in a travel guide such
as this (here the fact that Antibes cathedral has an “ill-lit Bréa”) implies
that it is potentially ‘of interest’ to the prospective visitor. This is then the
topic of utterance (5b) (‘features of potential interest to the visitor to Anti-
bes’), and is the reason why the prepositional phrase is preposed in this
context, connecting directly in this way with its immediate discourse con-
text. In FG terms, it would be placed in P1 position – an intra-clausal
position reserved (among other features) for constituents bearing special
pragmatic values. This, then, is the common thematic feature holding be-
tween the two sentences, in terms of which the propositions they express
are connected. See Hannay (1991a: 145) for a presentation and analysis of
a very similar example (his (19a)). Hannay notes that the preposed adjecti-
val phrase Particularly interesting... in his example is “topical inferrable
information” (on the basis of the immediate context which he provides),
and that it is a “staging device” rather than a Topic per se. As with CS the-
ory, so with FG only nominal expressions (terms) may be assigned one of
the Topic functions.
Now, one may agree with the CS analysis here that the referent of the
term the Picasso museum in the old Grimaldi château in (5a) is the (new)
topic of this utterance (‘As for the Picasso museum, this/it is of greater in-
terest’), and therefore bears the value ‘[MORE FOCUS (needed)]’, since it
is the subject of an intransitive predication; and that the predicational
prepositional phrase of greater interest is non-topical, not participating in
the FOCUS system (since it does not correspond to a participant – i.e. ar-
gument, in FG terms). But surely this degree of [FOCUS] (i.e.
concentration of attention) on the preverbal term in (5a) is not greater than
that which the inversion construction illustrated in (5b) manifests? Of
course, the difficulty here is that we are in grave danger of confusing the
CS sense of Focus (and what it is applied to) and the information-structural
one. CS would say that more attention-focus is concentrated on the prever-
bal term in the non-inverted construction in (5a) than on the postverbal one
in (5b). At the same time, it is apparent that in their spoken realizations, the
comparative adjective greater in (5a) (being part of the information-
structure focus) would have a more prominent pitch-accent on it than it
would in (5b). The placing of the subject term the Picasso museum in the
old Grimaldi château in postverbal predicate position in (5b) seems indeed
to “direct the attention of the hearer(s) to its intension, by being uttered”

(Erteschik-Shir 1997: 11), which is how this linguist defines the IS sense of
Focus; but this is not the case in (5a), where it is in preverbal, canonical
subject position.10 Erteschik-Shir’s (1997: 14) ‘lie-test’ confirms this dif-
ference. The lie-test consists in contradicting various parts of an utterance,
and seeing whether the connection between the two utterances is coherent.
It is essentially a test of what is asserted in the utterance at issue
(Erteschik-Shir 1997: 15), and so reflects the potential IS Focus domain(s).
(7) a. A: ....of greater interest is the Picasso museum in the old Grimaldi
château.
b. B: That’s not true! The ill-lit Bréa is much more interesting...(adapted
from the context of (4b) in The Holiday Which? Guide to France, 1982:
112 )
Compare the naturalness of (7b) as a contradiction of the assertion in (7a),

with the more bizarre character of (8b) as a contradiction of the informa-
tion conveyed by the preverbal PP in (7a) (here (8a)):
(8) a. A:....of greater interest is the Picasso museum in the old Grimaldi châ-
teau.
b. B: #That’s not true! Of no interest at all is that museum.
Clearly then, by the ‘lie-test’ the most prominent information in (5b) is

that associated with the postverbal term, not the preverbal one. And the in-
version construction seems equally clearly to place even greater informa-
tional prominence on the postverbal term than is the case in the unmarked
topic-comment construction illustrated in (5a).
The CS claim, then, that inversion constructions such as (4b) and (5b)
involve the signalling of less attention on the postverbal referent than on
the preverbal one in non-inverted constructions seems to be valid only un-
der the epistemic interpretation of the CS FOCUS systems – where the
addressee’s attention is simply assumed not already to be concentrated on
the referent at issue. Otherwise there would appear to be a clear conflict in
discourse values: for CS theory, the postverbal subject acts as an instruc-
tion to the addressee to pay LESS attention to the referent of this term (than
if it had occurred in preverbal subject position); and for IS linguists
(Erteschik-Shir, Lambrecht, and others – including Dik and other FG lin-
guists), the referent of this term is focal information, to which the
addressee’s attention is specifically being directed.11
126 Francis Cornish
In this regard, Birner and Ward (1998: 156–194) argue that the function
of the preverbal constituent in inversion constructions is to link the utter-
ance to the prior discourse context (see the PP of greater interest in the
attested example (5b)), and that, for the inversion to be felicitous, the post-
verbal constituent must not represent information which is more
‘discourse-old’ than that of the preverbal one. This is clearly the case with
the term the Picasso museum in the old Grimaldi château in (5b). More-
over, the verb in inversion constructions is a ‘light’ one: intransitive,
copular, or denoting existence or emergence and so forth.
For Huffman (1993), verb-subject12 ordering (where it is possible) is a
means of unobtrusively introducing a new referent, with maximum conti-
nuity with respect to the preceding co-text. Subject-verb ordering, on the
other hand, always assumes that the subject term’s referent is important
and salient in relation to the immediate concerns of the current discourse;
thus, when a new referent is introduced in this position, there is an effect
of discontinuity which is not apparent in the case of verb-subject order-
ing.
In the text from which (5b) was taken, the subsequent co-text does not
go on to mention the referent concerned (the Picasso museum in the old
Grimaldi château) – but neither does it do so in the case of the cathedral
in (6b), whose referent is in fact introduced via subject-verb ordering in
this sentence. And unlike the term the Picasso museum in the old Gri-
maldi château in (5b), the cathedral in (6a) would receive the value IN-
FOCUS, since it is the preverbal subject of a transitive, and not intransi-
tive, predication. In the case of the two sentences (6a) and (6b), the subject
expressions the cathedral (IN-FOCUS), and the rest of Antibes (MORE
FOCUS (needed)) would both correspond to the FG Topic type known as
‘SubTopic’, since they each pick out an aspect of the town of Antibes, the
immediate macro-topic of the text unit in which they occur. It is this more
macro-discourse relation which, in my view, makes for the coherence of
the three sentences (in the order in which they actually occur in the text).
The writer’s assumption that the reader can easily infer the intended part-
whole relation between the referents of the two preverbal subject expres-
sions and the immediate macro-topic lessens the tension created by the
placing of (relatively) new material in this position.
It is possible that the feature which Huffman ascribes to texts generally
is in fact genre-specific, occurring no doubt in narrative texts – W. Gold-
ing's Lord of the Flies and W. Cather's O Pioneers!, used as a testing
ground by Huffman – but not in travel guides such as the one illustrated by
the extracts in (6a), (5b) and (6b).
One last general point about CS FOCUS systems: given that, as Hannay
(1991: 140), Reinhart (1981: 56), Lambrecht (1994), Erteschik-Shir (1997)
and others point out, a given clause may sustain several different topic and
focus assignments, depending on its context at the utterance level, postver-
bal argument terms (direct and indirect objects) may fulfil the topic
function. CS theory, as presently constituted, would seem to have no means
of predicting these possibilities.
2.2. Deixis
DEIXIS in CS terms is said to constitute a grammatical system whose sub-
stance (in Saussurean terms) is “degree of insistent pointing toward the
intended referent” (cf. Diver n.d., 1995: 34–37 on the Latin demonstrative
and anaphoric pronouns; Leonard 1995 on the Swahili demonstratives, and
Aoyama 1995 on the Japanese ones). The degree of DEIXIS (required con-
centration of the addressee’s attention) is claimed (Diver, n.d.: 12) to
reflect the following principle (known as Principle G):
The more obvious the referent, the lower the item [indexical expression-
FC] on the deictic scale used to refer to it.
The deictic scale which Diver (n.d.: 11) gives in this respect is formu-
lated in terms of the Latin demonstratives (see also Dik’s 1997b: 223
similar table (21) structured in terms of preferred choices of Latin indexical
forms for realizing NewTops, Focus, and GivTops, as well as Bolkestein
2000, for FG). Bolkestein examines the discourse conditions under which
each of these forms (with the exception of ipse) occur, including of course
the zero subject pronoun, which would be placed at the extreme low end of
the scale in (9) below, as well as the subject relative pronoun qui.
Ipse, the emphatic demonstrative, is at the highest end of Diver’s scale,
and the clitic reflexive se is at the lowest end. The referent of se is nor-
mally what is in focus with respect to the associated finite verb. This is by
virtue of the reflexive relation with the nominative expression, if present,
where any nominative term and the verb inflection already encode HIGH
FOCUS. According to Principle G, the less obvious the speaker assumes
the intended referent in context to be for the addressee, the more a form
from the upper part of the scale will tend to be used; and correspondingly,
the more obvious, the more one from the lower part will tend to be chosen.
128 Francis Cornish
(9)
HIGHEST
ipse
hic
Deixis iste
ille
is
se
LOWEST
Hence, to use my own English example, in (10), the distal demonstra-

tive determiner is used initially, since the intended referent is new to the
discourse, the addressee being assumed not already to have noticed it:
(10) [It is dusk, and John and Mary are returning from a shopping trip. As John
is parking the car, Mary exclaims:]
Good God! Look at that incredibly bright light! [Mary gestures towards a
point in the evening sky] What on earth do you think it could be? (Cornish
1999: 26, ex. (2.6a))
Here, the indexically ‘stronger’ determiner that, together with the lexical
component of the term and the associated gesture, are used initially, since
the addressee John is assumed not already to be attending to the bright
light in question; while the indexically ‘weaker’ ordinary pronoun it is used
in the second sentence to refer back to something assumed by this point to
be within his attention focus.
But it is not just the ‘degree of obviousness’ of a given referent in a dis-
course which is the criterial factor in the organization of the field of
DEIXIS, in the CS sense. It is also, and mainly, that referent’s degree of
importance, as judged by the speaker, within the immediate discourse at
the point of use, which determines which form will be chosen from within
the relevant scale. This is the so-called ‘direct’ strategy for referent
establishment, the strategy in terms of the assumed degree of obviousness
of the referent being the ‘indirect’ one.
DEIXIS, in the CS conception, is claimed to be encoded by (among
other expression-types) clitic pronouns of different varieties. Clitic reflex-
ive pronouns in languages such as Latin, French and Spanish (but not in
English, where reflexive pronouns are not clitic and so have different dis-
tributional properties) signal the lowest degree of DEIXIS (in this sense of
the term), namely that very little effort need be expended by the addressee
in order to retrieve the intended referent – why? Presumably because it is
IN-FOCUS;13 and non-nominative, non-reflexive clitics signal a high(er)
level of DEIXIS – since here, the intended referent is no doubt (assumed
to be) NOT IN-FOCUS. Let us take the ambiguous example (11) from
French as a basis for illustration:
(11) Elle [LOW DEIXIS] sait bien que la décision lui [HIGH/LOW DEIXIS] incombe.
‘Shei well knows that the decision is up to himk/heri/j.’
One point which (11) immediately illustrates is that, contrary to CS predic-

tions, under the coreferential interpretation of the subject pronoun elle and
the dative clitic pronoun -lui,14 both pronoun forms must bear identical
(LOW) DEIXIS status, the coreferential interpretation pulling the dative
clitic within the dominion of the antecedent subject one (see Van Hoek
1997 for this type of approach to pronominal anaphora, within Langacker’s
Cognitive Grammar framework). Under the non-coreferential interpreta-
tion, it is possible that the DEIXIS status of the two referents concerned
differs. (11) shows the difficulty encountered by accounts such as the CS
one (also Gundel et al.’s 1993 ‘Givenness Hierarchy’), which frame their
scales in terms of individual lexical items, rather than in terms of given
uses of forms.
2.3. The CS FOCUS and DEIXIS systems compared

However, this distinction between the subsystems of FOCUS and DEIXIS
is not entirely convincing, since it fails to capture what is common to both
(which is a good deal, in fact) in a revealing way. In addition, there is the
ever-present danger of redundancy, depending on which definition of the
FOCUS system is retained: that is, ‘attention is already concentrated on a
given referent’, or ‘attention needs to be so concentrated’. It would be
tempting to suggest (as would seem logical, after all) that (a) the very same
morphemes which encode the value LOW DEIXIS are signals of
HIGH/CENTRAL FOCUS, and that (b) those morphemes signalling HIGH
DEIXIS simultaneously encode the value LOW/PERIPHERAL FOCUS.
There is evidence for the (b) relationship in the shape of the French clitic
complement pronouns lui-, leur-, le-, la-, and les-, which are said to encode
the values PERIPHERAL (= LOW) FOCUS and HIGH DEIXIS (see
Huffman’s 1997: 211 Figure 5.3). In the case of (a), however, the French
reflexive se is claimed to encode the values LOW DEIXIS, but also PE-
130 Francis Cornish
RIPHERAL (LOW) FOCUS (Huffman 1997: 211). Yet it is at least plausi-

ble that an alternative analysis in terms of CENTRAL FOCUS might be
made of French se, owing to its obligatory coreference with the nominative
term (‘subject’) which acts as its controller (see also footnote 13 above).
For García (1977: 152), Spanish se is outside the FOCUS system alto-
gether.
Furthermore, in the case of Latin, as Wallis Reid (pers. comm.) points
out, the demonstratives ipse and hic are at the top of the Deictic Scale (see
(9) above), and so clearly encode the value HIGH DEIXIS; and yet since
both pronouns may be fully declined, each may take nominative case (giv-
ing them HIGH FOCUS status) or alternatively an oblique case (where
they would signal LOW or PERIPHERAL FOCUS: in the latter case, this
combination would be as predicted by (b) above). Similarly, the
demonstrative is, which occurs near the bottom of the Deictic Scale in (9),
may take an oblique case, and so encode the value LOW FOCUS.
Now, given that these combinations hold, how can we make sense of
the apparently contradictory instructions which the ‘non-conforming’ com-
binations signal? That is, HIGH FOCUS + HIGH DEIXIS: ‘the intended
referent is assumed already to be in the addressee’s attention focus, but will
need a high degree of effort in order to locate it’; and LOW FOCUS +
LOW DEIXIS: ‘the intended referent is not assumed already to be in the
addressee's attention focus, but at the same time it is not necessary to ex-
pend a great deal of effort in order to locate it’. In my view, the only
sensible interpretation of these apparently contradictory combinations of
instructions would be as follows: ‘the intended referent is/is not already in
the addressee's attention focus (FOCUS system), but the referent at issue
is/is not important for the current discourse purpose (DEIXIS system)’.
Thus, it is the direct, not the indirect strategy associated with the DEIXIS
system which necessarily comes to the fore in such cases.
More thought clearly needs to be given to these kinds of relationships
between the various values encoded by given forms from within the two
systems at issue here, in order to try to better integrate the two complemen-
tary systems of DEIXIS and FOCUS in CS theory.
3. The four Topic statuses recognized in standard FG

Functional Grammar uses both discourse-pragmatic and formal-coding cri-
teria to distinguish Topics in relation to Foci. Two basic properties of
discourse relevant to Topic and Focus status are those of topicality and fo-
cality, considered as functions of the speaker's assumptions regarding

his/her addressee's current pragmatic knowledge state at any given point in
the discourse. Topicality has to do with the set of referents “‘about’ which
information is to be provided or requested in the discourse” (Dik 1997a:
312). This notion is close to Brown and Yule’s (1983) concept topic
framework. Focality, on the other hand, “attaches to those pieces of infor-
mation which are the most important or salient with respect to the
modifications which S [the Speaker] wishes to effect in PA [the addressee's
current state of pragmatic knowledge], and with respect to the further de-
velopment of the discourse” (Dik 1997a: 312). Importantly, topicality and
focality are not mutually exclusive: areas where they overlap include con-
structions involving Contrastive or Parallel Focus. The ‘aboutness’ test
which we used in Section 2.1 is employed to determine the Topic relation
between an entity and a predication, as is the Question-Answer test (which
also determines the potential Focus domain(s) of the utterance concerned).
The latter test establishes the degree to which a given (declarative) utter-
ance may be construed as a natural or coherent response to a question
about a specified referent, providing the information required. This type of
test is criticized by Gómez-González (2001), though it would seem that no
very convincing evidence is marshalled against it.
Now, Lambrecht’s (1994: §4.1.2) important distinction between topic-
referent and topic-expression is reflected in the FG approach to topicality
and Topic assignment, in that the tests just mentioned serve to isolate the
referent bearing the Topic function in relation to a given Focus domain
within an utterance; however, the various formal criteria needing to be sat-
isfied in order for a constituent to be analysed as Topic obviously relate to
the topic expression, in Lambrecht’s terms. That is, in order for a discourse
entity to be recognized as a Topic of some kind, its linguistic exponent
must be singled out in a systematic way by the language system involved:
via morphological form, a particle, a specific type of construction (e.g. the
left-detached construction), a specific word-order pattern, or a particular
prosodic form. As we shall see, this form-oriented aspect of Topic recogni-
tion in FG has taken priority over the discourse-cognitive dimension15 (as
developed in Reinhart 1981, or Lambrecht 1994, for example), in that it
has motivated the recognition of a Topic type known as New Topic. How-
ever, according to the discourse-cognitive criteria, such referents are not
topics at all within the clause in which they are expressed: rather they are
clearly focal (see the references listed in footnote 23 on this issue). More-
over, it has led one pair of authors within FG (Mackenzie and Keizer 1991)
to conclude that no Topic function exists at all in English, since according
132 Francis Cornish
to them, there is no systematic coding in that language which would single

out expressions bearing the Topic function in relation to other, non-topical
expressions (but see Hannay 1991: 142, fn. 9 for an opposing view). I be-
lieve that this over-reliance on systematic, overt coding as the main
condition determining the existence of particular pragmatic functions in a
language is a reflection of the fact that the assignment of such functions is
standardly assumed to take place within the grammar, as the final specifi-
cation before the complete underlying clause structure representation is
transferred to the expression component for its formal realization. A num-
ber of linguists within FG have argued against this locating of pragmatic
function assignment16 within the grammar, for a variety of reasons (e.g.
Vet 1998, Bolkestein 1998, Van den Berg 1998). More recently the type of
model presented by Hengeveld (this volume) is conducive to representing
the looser kind of relationship between a pragmatic module and a gram-
matical one which these authors argue for. See also Hannay’s (1991) five
proposed message management modes (to be discussed in Section 4 below)
and their relevance for grammatical coding.
Now, as Siewierska (1991: 148) points out, the subdivision of the Topic
function within FG was designed to relate the specification of clause struc-
ture to the wider discourse setting in which the clause under analysis may
occur. The four Topic statuses recognized in FG (Given, Sub, Resumed
and New) would appear to subsume the distinction in CS between FOCUS
and DEIXIS. Given, Sub and Resumed Topics would realize CS DEIXIS,
and New Topics, FOCUS (i.e. they would correspond to the CS value NOT
IN-FOCUS). FG GivTops, SubTops and ResTops clearly form a scale of
deicticity, in the CS sense of the term, GivTops retrieving obvious refer-
ents, where no competition is involved between referents of the same type:
examples used so far in this chapter are it in the second sentence of exam-
ple (10), and elle in (11) (where this pronoun is unaccented).
SubTops17 require the drawing of a minimal inference from a contextu-
ally given set of entities to the existence of a particular member of that set,
or from a whole to one of its parts, and so on – thus, encoding a slightly
higher degree of DEIXIS than that associated with GivTops: examples
would be the preverbal subject terms the cathedral and the rest of Antibes
in examples (6a) and (6b) above. Both these definite terms manifest a part-
whole relation within the immediate macro-topic of their containing dis-
course segment, namely the city of Antibes.
Finally, ResTops encode the highest degree of deicticity of the three
subtypes, the referent being assumed to exist within the current discourse
model, but no longer active or present in the addressee’s current con-
sciousness. Thus the intended referent is assumed to be the least ‘obvious’

of the three at the point of use. An attested example is given under (12):
(12) ... Elliot Morley, the Countryside minister, said he thought it was possible,
with the new funds available, to aim at not just halting, but reversing the
declines [of UK farmland bird species] in the medium term. “I'm in no doubt
about the scale of this task, but I believe it is realistic to aim to do this by
2020”, he said. (Last paragraph of a 9 paragraph newspaper article ‘Ministers
pledge £1bn to save farmland birds’⎯The Independent, 7 August 2000).
The demonstrative term this task in the fourth line of this extract refers
back over the two preceding discourse units to the initial (macro-discourse)
topic, introduced in the first, namely, the UK government’s commitment to
halt and reverse the massive decline in farmland bird species in that coun-
try by the year 2020. Neither of Dik’s (1997a: Ch. 13) other two formal
criteria supposedly accompanying terms with the ResTop function appear
here, however (a connective indicating the start of a new discourse unit,
and a specific reference back to the discourse unit to which the ResTop
expression effects a return pop). It is the fact that the term itself occurs in a
unit which is discourse-final, marking the conclusion of the discourse as a
whole, I believe, which obviates the need for these two supposedly neces-
sary conditions here. I will deal with the fourth subcategory of Topics
recognized in FG (‘NewTops’) later in this section.18
Now, an important difference in emphasis arises between the two theo-
ries, regarding the way in which each views the discourse function
associated with the expression-types at issue. For FG (and also other cogni-
tively-oriented approaches like the ‘Givenness Hierarchy’ of Gundel,
Hedberg and Zacharski, 1993), the context-bound Topic functions are
predicated solely on the discourse-cognitive status which the speaker is as-
suming the referent to enjoy in the addressee’s discourse model at the point
of use. For CS, on the other hand, the signals of DEIXIS are said to serve
as instructions to the addressee to concentrate a given level of attention on
the referent involved: HIGH – the referent is important to the current dis-
course concerns and so must be foregrounded relative to other referents;
MID – the referent is of medium importance with regard both to HIGH
DEIXIS and LOW DEIXIS referents; and LOW – the referent is of only
background, ‘scene-setting’ (for example) importance. As Leonard (1995)
and Aoyama (1995), respectively, point out in relation to the attested Swa-
hili and Japanese examples which they present, discourse referents are not
simply designated ‘decrementally’ with ever-decreasing levels of DEIXIS
134 Francis Cornish
marking as they are introduced, established, and maintained anaphorically

throughout a discourse.19 Leonard’s (1995) DEIXIS scale as realized by the
Swahili demonstratives is given in (13):
(13) h HIGH DEIXIS

h-o MID DEIXIS
le LOW DEIXIS
The Swahili DEIXIS system (Leonard 1995: 273, item (1))
The LOW DEIXIS Swahili demonstrative -le may indeed function

anaphorically to maintain reference to a recently-introduced entity, by
virtue of its differential deictic meaning ‘no particular effort of
concentration is needed to retrieve this referent’, as illustrated in Leonard’s
(1995: 272) example (2a), here presented as (14); note that -le here would
be glossed in English by the definite article the:
(14) Alimwona kijana. Kijana yu-le alisimama mlangoni.

‘He saw a youth; le-youth (LOW-DEIXIS) was standing at the door’
Clearly also, as expected, where a new referent is introduced into the

discourse, and where that referent is also of importance to the speaker and
to the current discourse purpose, a HIGH DEIXIS marker is used:
(15) [An oathgiver explains an oathing procedure to a man. He says:]

“First I’ll speak, you’ll listen, then when I’m finished ⎯ nataka wewe
useme h-ivi “Mikale mikale...”
‘I want you to speak in h- (HIGH-DEIXIS) manner [i.e. say the following]:
“Mikale mikale...” (Leonard 1995: ex. (3))
The HIGH DEIXIS marker h- here is clearly equivalent to the English

proximal demonstrative this, which Strauss (1993) argues is the HIGH
DEIXIS member of the triplet this/that/it in that language, whereby the
speaker is conveying a high degree of personal, subjective involvement
with the referent, instructing the addressee to accord a high degree of atten-
tion-concentration on the referent, and to assume it is going to be important
for the ensuing discourse. The distal demonstrative that is said to realize
the value MID DEIXIS/FOCUS,20 and the ordinary inanimate pronoun it,
the value LOW DEIXIS/FOCUS. As with Leonard’s (1995) attested exam-
ples of Swahili texts, Strauss gives examples where English HIGH DEIXIS
this is used in subsequent (instead of, or even as well as, initial) reference
to a given entity:
(16)
L 30
Caller: You spoke a little while ago to a man.hhh with regard to uh
law,//an’ murder.
Crandall: Yes ma’am...
L 70
Caller: ...you were talking about, or was he anyway,//that man
Crandall: well, no:: Now you see this the- again we come back to this:
Murder is the act of killing someone, by, an- a specific
dee:duh, described in the law of the sovereign state in which
you live, as being unjustified.
Caller: Now/
Caller: Yuh know/
L 76
Caller: By the d- conversation that chu had with this man: my im-
pression was that there’s something drastic’ly wrong with us
as a nation. Because I- uh when you say, that the Federal
government cannot interfere,.hh in u-uh:: condoned murder.
Like you-
Crandall: [In a- in a- in a murder case⎯
Caller: [Like you mentioned in Mississippi
(Strauss 1993: ex. (12))
Strauss points out (1993: 410) that the use of the HIGH
DEIXIS/FOCUS expression this man in the first line of L 76, as fourth
mention of the referent involved, is not evoking ‘new information’, but is
rather motivated by the caller’s attempt to regain her turn, lost to Crandall
at line 71.
Another attested example (from my own corpus this time, here a news-
paper article) is given under (17):
(17) [Beginning of article “God sends a sign to the heathen in a Chevy”, The In-
dependent, 11 August 1999, p. 13]
You are driving along in the relaxed American way, just getting from A to
B and minding your own business, when your attention is grabbed by a gi-
gantic billboard, “That ‘Love Thy Neighbour’ Thing”, reads the neat, white
text, “I meant it – God”.
136 Francis Cornish
This particular hoarding, which thrust itself into my line of sight some-
where on Interstate 81 in southern Virginia at the weekend, is the latest in
my growing collection of “God” adverts...
From a purely ‘referent-tracking’ point of view, there is clearly no mo-

tivation for the use of a HIGH DEIXIS marker at the start of the second
paragraph of this extract. The particular billboard at issue has been intro-
duced by the indefinite nominal a gigantic billboard in the first paragraph,
and it is already in the reader’s focus of attention by the end of that para-
graph. If the adjective particular were omitted from the second-mention
demonstrative NP, then that hoarding or the hoarding, or even (if the ap-
positive relative clause following it were absent too), simply it, would have
been appropriate. The HIGH DEIXIS proximal demonstrative phrase is
clearly chosen here in order to underline the writer’s empathy or strong
subjective involvement with the referent at the point recounted in her car
journey when she first noticed it – and also to signal the fact that the refer-
ent is being envisaged in a slightly different context in this second
paragraph, as simply one of many such billboards proliferating along
American highways (the generic topic of that and the subsequent para-
graphs, in fact).
Nevertheless, the CS account of DEIXIS in terms of three degrees of ef-
fort needed by the addressee to concentrate his/her attention on the
intended referent is not general enough. I have argued (in Cornish 2001)
that these progressively greater degrees, as we move up the scale, of ‘urg-
ing’ or ‘insistent pointing’ on the speaker’s part that the addressee attend to
the referent may be predicted from what can be seen to be more basic cog-
nitive-interactional values encoded by the members of a given ‘deictic
scale’. If we take as a model the three English indexicals this, that and it,
for example,21 then the proximal (marked) variant has the effect of placing
the referent within the speaker’s discourse sphere, signalling his/her strong
subjective involvement with that referent and allowing the inference that
this referent does not constitute ‘negotiated’, mutually-validated informa-
tion. From this and the ‘egocentricity’ principle which clearly regulates
discourse interpretation generally, it can be deduced that the speaker con-
siders the referent at issue to be highly important for the ensuing discourse,
and that the addressee should pay special attention to it. Particular uses of
this in English such as colloquial ‘indefinite’ this and anticipatory or cata-
phoric this (both instances of discourse deixis in my view), as in (18a, b),
are natural extensions of this value:
(18) a. There was this terrible boy at the circus, who....

b. Now listen to this: a man went into a butcher’s shop and bought a
whole pig...
The distal (unmarked) variant, that, would have the discourse-interactional

value whereby its referent is placed outside the speaker’s subjective sphere
and into a shared, ‘negotiated’ space, in which the speaker’s and ad-
dressee’s stances are in co-alignment. As such, the referent is not
(necessarily) envisaged as representing new information for the addressee,
and thus the somewhat lower degree of insistent urging that s/he attend to it
is predicted (see also Cheshire 1996 and Glover 2000 for similar analyses
of English this and that). The ‘recognitional’ or ‘reminder’ use of that pro-
vides a good illustration (see also the demonstrative term that ‘Love Thy
Neighbour’ thing in example (17) above):
(19) Tony Blair was said to be on the phone last week to his German oppo,
Chancellor Schröder, trying to talk through an upbeat final instalment of
that dismal industrial soap opera called Longbridge.
(The Guardian, 3 May 2000, p. 12)
Finally, the non-demonstrative, inanimate (unaccented) pronoun it sig-

nals not only that its intended referent constitutes shared information, but
also that that information is currently in the addressee’s (and the speaker’s)
ongoing attention focus. Again, this value implies that the addressee need
expend no particular effort at all in locating and retrieving the intended ref-
erent, since it is (assumed to be) ‘right there’, in his/her current
consciousness.
Unlike both the CS account of DEIXIS and the Gundel et al. (1993)
Givenness Hierarchy, which are formulated in terms of particular lexical
items, the four FG Topic types are characterized in terms of purely dis-
course-functional roles: although there are claimed to be certain
prototypical indexical markers which realize such roles (zeros and unac-
cented 3rd person pronouns for GivTops, definite lexically-headed terms
for SubTops, expanded definite or demonstrative terms for ResTops, and
indefinite lexically-headed terms for NewTops), it is also clear that, in
given contexts, other expression types may also be used to realize each of
these functions (for example, the demonstratives, whether as determiners
or pronouns, may in fact realize all of them – except ResTops, where de-
monstrative NPs rather than pronouns are required: see Cornish 1998 for
some discussion, based on an English-language newspaper article).
138 Francis Cornish
This would seem to me to be desirable, since as we have seen, other,

strategic-discourse as well as expressive and viewpoint-signalling factors
come into play in addition to purely referent-tracking ones in the use of in-
dexical expressions. However, no allowance seems to be made within FG
for proactive, strategic uses, based on interactional or discourse-con-
structional motivations such as the ones which we briefly saw in examples
(16) – (17). But the FG account has the advantage over the CS and the
Givenness Hierarchy (GH) ones, in that specific lexical items are not ‘tied’
to particular positions on a hierarchy of forms in terms of relative indexical
‘strength’.22 In any case, we have seen that the CS scales in terms of de-
grees of insistent pointing towards the intended referent may be predicted
on the basis of more fundamental discourse-interactional values; and it is
likely that the GH may similarly be subsumable under a more precise dis-
course-interactional characterization of each of the form types represented
on the Hierarchy.
FG NewTops, on the other hand (the most problematic of the four sub-
types recognized in the theory), clearly fall within the FOCUS system
distinguished in CS theory. In fact, where such NewTops are introduced in
postverbal position (as is most often the case in Dik’s 1997a presentation:
see (20) below for an illustration), they would be viewed in CS theory, as
we saw earlier, as being NOT IN-FOCUS – that is, as not worthy of the
addressee’s attention at all! So here we have what at first sight is a diamet-
rically opposite analysis of a given form type by each theory. Actually, as
we saw in the case of the comparison between non-inverted constructions
such as (5a) and inverted ones such as (5b), the contrast is a paradox rather
than a true contradiction, since CS is saying that Dik’s NewTops are non-
topical (which is correct, as we shall see below), while in FG they realize
one type of Focus (in the Information Structure sense); and clearly, both
statuses can simultaneously characterize the same given constituent. In
fact, even within FG as they are defined in Dik (1997a: Ch.13), they can be
analysed as simply constituting in the unmarked case one type of Comple-
tive Focus, since we are dealing here with the introduction of a referent
which is anticipated to play a role in the subsequent discourse. Dik (1997a:
312) in fact specifically mentions NewTops in illustrating the possibility of
overlap between topical and focal elements in a discourse. He characterizes
NewTops as the introduction of new participants or entities about which
the speaker intends to say something later on, and by virtue of that intro-
duction, as constituting the main point of the containing utterance. His
example of this is (20):
(20) All of a sudden we saw A GIGANTIC SHARK. (Dik 1997a: 312, ex. (6))
However, it is debatable whether in fact the referent of the NP A GIGAN-

TIC SHARK in this context (in fact, a thetic utterance) is topical at all – that
is, whether the utterance of (20) may be construed as a categorical rather
than thetic one, and ‘about’ a gigantic shark: witness the incoherence of
#As for a gigantic shark, all of a sudden we saw one/it. Applying
Erteschik-Shir’s (1997) ‘lie-test’ clearly shows that the content of a gigan-
tic shark in (21) is part of the focal information conveyed via this
utterance:
(21) A: All of a sudden we saw A GIGANTIC SHARK.

B: That’s not true! It was a BABY WHALE.
The notion of NewTop is in fact an incoherent one itself, if it is con-

ceived (as I believe it must be) in discourse-contextual terms. After all,
discourse topics do not emerge fully-fledged just through being introduced
into a given discourse; they have to be established, and then maintained if
they are not to fade from salience or be superseded by subsequent topics
(see Cornish 1998 for some discussion). New topics may be established on
the basis of the introduction of a new referent into a discourse, as in the
case of a gigantic shark in (20); but I think that it is a mistake to believe
that the referents of such expressions ipso facto have the status of topics.23
Topic introduction and establishment is now widely conceived within the
literature as being a joint, cooperative undertaking, involving the coordina-
tion of both speaker and addressee, and is not the prerogative of the
speaker alone. In any case, all the examples of NewTops presented in Dik’s
(1997a: 315-318) subsection on NewTops (§ 13.3.1) in fact fall within
thetic, not categorical utterances, where Dik specifically makes the point
that such referents tend to be introduced via expressions occurring towards
the end rather than at the beginning of the clause.
4. Pragmatic functions within the new FDG model

Quite clearly, the assignment of pragmatic functions within the new Func-
tional Discourse Grammar model (Hengeveld, this volume) must be part of
the Interpersonal Level specified in the model, a level which subsumes and
‘controls’ the specifications made at the two lower levels in the system
(those of Representation and Expression). The marking of information-
140 Francis Cornish
structural relations at the level of the clause needs to be sensitive both to

features of the communicative context, and to the speaker’s evaluation of
the addressee’s current cognitive state. This can be done once the Cogni-
tion and Communicative Context components of the FDG model which
feed and are in turn fed by the Interpersonal Level (as indicated in
Hengeveld, this volume: Fig. 1) are adequately fleshed out.
Adopting Hannay’s (1991: 148) five modes of message management
(All-New information, Topic, Reaction, Neutral and Presentative), it is
clear that this initial planning stage must be situated at the Interpersonal
Level recognized in the FDG model, where communicative intentions are
formed, expressed minimally by the Move, which is realized by one or
more constituent Discourse Acts. See also for a more specific elaboration
of this component Van den Berg’s (1998) ‘pragmatic’ and ‘message’ mod-
ules within his ‘pragmatic functional grammar’; here, pragmatic functions
are handled by the ‘message module’ (1998: 100). I assume his ‘pragmatic’
module would correspond to the Interpersonal Level in Hengeveld’s FDG
model, and that his message module would correlate with the FDG Repre-
sentational Level. Examples illustrating Hannay’s message management
modes presented so far in this chapter are (1b) and (4a) for All-New, (1a)
for Topic, and (4b) and (5b) for Presentative. The Neutral mode is said not
to involve an initial Topic or Focus expression and comprises a ‘dummy’
subject (it or there). An example would be It was surprising that the
Greens won the election. As for the Reactive mode, it involves the prepos-
ing of immediate, expressive-subjective material bearing the Focus
function, as in Hannay’s (1991: 143) example (2b): (2) Q: Did you get
wet?… (b) Wet? Bloody soaking I was.
Now, clearly, each message management mode, with the exception of
the first (the All-New mode), presupposes a specific kind of context. This
must be represented in the Communicative Context component, whose task
is to provide a representation of the discourse developed prior to the utter-
ance about to be produced, as well as of the prevailing situational context.
A very important part of this discourse representation must be (a) an indi-
cation of the current (local and global) macro-topics and (b) the structure of
this discourse (i.e. the level or ‘unit’ of the discourse under development
which the discourse has reached). The Presentative, Neutral and All-New
modes may serve to initiate a new discourse unit, since their essential func-
tion is to introduce new referents or states of affairs. In the case of the
Presentative mode, as we saw in example (5b), there is often a preposed
thematic locative, temporal (or more generally, circumstantial) phrase in-
troducing the constituent manifesting it; in which case, the content of the
clause as a whole is presented as tightly connected with the preceding co-

text. Thus, such clauses may serve either as transitions between one dis-
course unit and the next, or as links between utterances within the same
unit.
Constructions realizing the All-New mode, however, introduce wholly
new situations or events, and re-set the space, time and thematic coordi-
nates within a given discourse. In the case of Reaction mode clause
realizations, these are mainly restricted to face-to-face spoken language of
a dialogic kind (as in the illustrative example given above); as its very
name suggests, the Reaction mode presupposes a ‘first pair-part’ to which
it constitutes a ‘reaction’. Thus, its clausal exponents are restricted to oc-
curring within minimal spoken discourse units (the adjacency pair in
conversational discourse). The Topic mode too – that is, where the Topic
type at issue is a GivTop – is by definition restricted to intra-unit occur-
rences. In terms of the intra-clause Topic functions, most clearly, ResTops
(manifesting the highest level of CS deicticity of the three discourse-bound
FG Topic functions) would refer to an entity evoked within an earlier, dis-
tinct discourse unit in relation to the one in which their exponents occur.
But (as Dik 1997a: 325 himself implies), they could not open a new dis-
course unit on their own: Dik claims that a specific connective signalling
the start of a new discourse unit is needed, presumably for this purpose.
SubTops, too, may signal the start of a new discourse unit, but one which is
clearly subordinate to the immediately preceding one – or which is at least
at the same level of subordination as this unit.
Whereas the Communicative Context component within the FDG model
must clearly keep track of the preceding current discourse, as we have
seen, and must feed this information into the Interpersonal Level in order
for it to be able to manage the organization of future messages, the Cogni-
tive component must contain long-term representations of both linguistic
and non-linguistic kinds: encyclopaedic information concerning real-world
properties and relationships, as well as personal information assumed by
the current speech participants to be mutually shared, and which is relevant
to and evoked via the current discourse. It must also contain information
regarding semantic relations between lexemes (such as hyponymy, mero-
nymy, synonymy and antonymy), thus motivating the choice at the
Expression Level of appropriate first restrictors within definite term struc-
tures manifesting GivTop or SubTop functions within the wider discourse
(see Hannay, 1985b: 55–56, exx. (26) – (35) for the range of semantic or
pragmatic relations in terms of which SubTops may be realized).
142 Francis Cornish
With this subsuming, controlling relationship between the higher-level

Interpersonal Level and the lower, intermediate-level Representational one,
which forms a message on the basis of the pragmatic structuring specified
at the Interpersonal Level, Focus assignment or the lack of such assignment
may be shown to have an effect on the semantic nature of the predicator
selected at the lower Representational level, something which was impos-
sible or very difficult to show under the standard model of FG (Dik 1997a:
Ch. 13).24 Bolkestein (1998: 204–206) likewise gives Latin examples
where the assignment of Focus function to given constituents in a clause
may block Subject or Object assignment to certain other constituents. Thus,
in her words (p. 205), “the location of main Focus must be part of the input
for the rule that takes care of syntactic function assignment”. All these phe-
nomena show clearly that pragmatic function assignment must take place
before, and not after, the stages at which semantic and syntactic properties
are established within the clause. In any case, as the various tests, such as
the Question-Answer test for Topichood and potential Focus domains,
clearly indicate, information from the wider discourse context in which a
given clause is set is crucial for the assigning of pragmatic functions to
relevant constituents within it.
The structure of the discourse corresponding to a particular Move, then,
is established at the Interpersonal Level in the new model, in terms of a co-
herence relation with respect to the last such Move, and to the wider
discourse and thematic structure as specified by the Communicative com-
ponent. The mode of message management adopted at this level will in part
determine at the lower Representational Level a particular underlying
clause structure, with Topic and Focus assignments marked (as appropri-
ate) as a function of the message structure indicated at the Interpersonal
Level. Finally, the Expression rule component will convert these specifica-
tions into an actual object-language expression (but now, there is no
requirement of a one-to-one relation between a type of systematic coding
in the language in question and a given Topic or Focus function: Topic ref-
erents, in particular, are established as such independently of a given
coding type, and indeed are already marked as such at a more language-
exclusive level of representation – the Interpersonal Level). So topics (ref-
erents) may be such without there needing to be dedicated language-
specific coding devices for expressing them, the Interpersonal and Repre-
sentation Levels now being distinct and containing non-isomorphic types
of units.
5. Conclusions
Both theories (CS and FG, but CS more so than FG) need to take account
more centrally of the thetic/categorical distinction in predicting possible
Topic-Focus relations in discourse – and in doing so, CS theory needs to
take on board the information-structural dimension of the flow of utter-
ances in discourse, in addition to its purely topic-based account. It could
usefully take cognizance of the fundamental FG distinction between topi-
cality and focality in discourse, as distinct from the assignment of
particular Topic or Non-Topic (i.e. IN-FOCUS or NOT IN-FOCUS) roles
to given constituents within the clause. As it is, its notion of Focus of At-
tention leaves the focality dimension totally out of account – even though
its presence is implicit in CS analyses.
Under the conception where FOCUS in CS is defined in terms of the
addressee’s being instructed to concentrate or otherwise his/her attention
on a given referent, it is basically no different from the system of DEIXIS
(at least under the indirect strategy interpretation), whose signals likewise
enjoin the addressee to accord different levels of attention concentration on
a particular referent, as a function of its assumed degree of obviousness for
the addressee or reader – but where three degrees of ‘insistent urging’
rather than two, as for FOCUS, are involved. Viewed in this light, there is
no essential difference between the two systems (apart, as we have seen,
from the partially complementary set of clitic pronoun signals involved in
coding each system). There would appear to be two interpretations where
the FOCUS and DEIXIS signalling systems have differing functions: (a)
where CS FOCUS has the cognitive/AI interpretation (where the referent
concerned is simply assumed already to be at the forefront of the ad-
dressee’s consciousness at the point of use): here, the FOCUS system
signals what is considered to be topical or laying claim to the addressee’s
conscious awareness, and the DEIXIS system (under its ‘indirect’ interpre-
tation) indicates how to retrieve various sorts of discourse referents, given
their already-existing FOCUS level; and (b) where the morpheme signal-
ling DEIXIS has the direct status of indicating the level of discourse
importance of a given referent, from the speaker/writer’s point of view.
In fact, with the systems of DEIXIS and FOCUS, CS theory only con-
cerns itself with topic/non-topic signalling in discourse – i.e. with what the
speaker is or is not centrally talking about in a given utterance; his or her
marking of what s/he wants the addressee to do with (or to) this referent
(via the various types of Focus function, in the IS sense) appears to be left
out of account.
144 Francis Cornish
FG, for its part, might usefully pay more heed to establishing criteria for
determining the various attention-focus or accessibility statuses of given
referents in the flow of discourse, thus motivating the successful prediction
of the use of the various subtypes of Topic expression in accessing them
(this point is also made by Gómez-González 2001: 165). Finally, one of the
four subtypes of FG Topics, NewTops, may, I think, be safely abandoned
(since its provision is already catered for, in the unmarked case, by Com-
pletive Focus), thereby simplifying the system as a whole.
Notes
1. This chapter is a revised and expanded version of a paper presented at the
9th International Conference on Functional Grammar, held at the Universi-
dad Nacional de Educación a Distancia, Madrid, from 20 to 23 September
2000. My thanks in particular to Wallis Reid, Kees Hengeveld and Knud
Lambrecht as well as the editors of the present volume for their very helpful
comments on earlier versions.
2. This term should be understood here in the sense which Lambrecht (1994)
adopts, namely where discourse information is that which is derived by an
addressee from the combination of a topic with a focus (within a topic-
comment utterance), or in more general terms, from that of items one with
another in the flow of text. In CS theory, FOCUS (‘IN-’, ‘NOT IN-’,
‘MORE’ or ‘LESS’) is assigned by the speaker-writer to individual argu-
ments (participants) both in relation to the particular predicator chosen
(what it calls ‘EVENT’) within the clause, and to the wider discourse. I am
indebted to Knud Lambrecht for this clarification.
3. As Knud Lambrecht points out (pers. comm.), this is partly due to the fact
that the term in question is indefinite (thereby marking its referent as uni-
dentifiable by the addressee), and not definite (and hence potentially
identifiable). It is also due, of course, to its position to the right of the verb,
and to the fact that, in its spoken realization, it would receive a nuclear
pitch-accent.
4. Lambrecht and Polinsky (1997) present a range of distinctive properties
characterizing thetic utterances cross-linguistically.
5. In this connection, see Reid’s (1991: 181–182) presentation and analysis of
a newspaper report of a marathon race, as well as Huffman’s (1993) appli-
cation of the system to two fictional narratives.
6. I am indebted to Wallis Reid for this characterization of the CS FOCUS
system.
7. For good, fairly recent discussions of the thetic vs. categorical distinction,
see Rosengren (1997), and Lambrecht and Polinsky (1997).
8. See Birner and Ward (1998: 159), as well as Gundel (1999: 299–300) on the
prosodic difference between ‘semantic focus’ (i.e. unmarked focus in a
topic-comment structure), and ‘contrastive focus’.
9. See in particular Hannay (1985a) for an FG account of this construction.
10. In English, of course, preverbal subjects may also be focal, since (given an
appropriate context) they may receive a high pitch-accent, and thereby con-
vey new information relative to that context (A: Who ate the fish? B: JANE
ate the fish/did).
11. See also the point made by Reinhart (1981: 57) that the description ‘focus
of speaker’s [and addressee’s] attention’ may characterize topics and foci
alike.
12. Strictly speaking, Huffman as a CS linguist would not use the traditional
terms ‘subject’ and ‘verb’ here, preferring the descriptive terms ‘(first) par-
ticipant’ and ‘event’, respectively. However, I use the more traditional
terms here for ease of comparison with the IS conception.
13. This is not always the case, however: for example, in Huffman's (1997:
209–210) analysis of French clitic pronouns, the 3rd person clitic reflexive
se is claimed to blend CENTRAL and PERIPHERAL (as he calls it) FO-
CUS, since on the one hand it automatically links up with the IN-FOCUS
‘subject’ term (nominative in form, if a clitic pronoun: il/ils, though not
elle/elles, since these forms may occur in other relations to the EVENT
(verb) than that of P1 (initial argument)); and on the other, it involves a ‘re-
mention’ of the same participant. Nonetheless, I think that a case could be
made for French clitic se to encode the value CENTRAL FOCUS, by virtue
of its automatic binding by the HIGH FOCUS nominative term bearing sub-
ject function with respect to the verb, and in relation to the latter’s
inflectional form. In FG, predications containing a reflexive pronoun, in re-
lation to ones with a full term phrase (i.e. transitive clauses), are considered
to have undergone ‘argument reduction’, the reflexive pronoun being the
surface marker of this relation. Such clauses are clearly intransitive (like in-
herently pronominal verbs in French and other such languages). Hence the
reflexive (clitic in French) pronoun does not correspond to an argument se-
mantically, and cannot therefore be assigned any value pertaining to
argument expressions. See García (1977) for an excellent analysis of Span-
ish se within the CS framework.
14. The hyphen preceding given pronoun forms indicates that the clitic variant
is what is intended.
15. As Siewierska (1991: 148) points out, “though a constituent may be topical
or focal on discourse grounds, unless its topical or focal status is coded in
the structure of the clause, there is no basis for recognizing a special clause-
bound level of pragmatic organization distinct from the semantic and syn-
tactic levels of clause structure”.
146 Francis Cornish
16. That is, the selection of given constituents as bearing Topic or Focus func-
tion in terms of the wider discourse context and the assignment of these
functions to expressions conforming to the formal coding requirements at
the same level.
17. See Hannay (1985b) on SubTopics.
18. Siewierska (1991: §§6.3.2–6.3.3) and Gómez-González (2001: §§5.2.1-
5.2.4) present critical surveys of the four types of Topic function within
standard FG.
19. That is, where a ‘HIGH DEIXIS’ marking might be used to introduce a ref-
erent, a ‘MID DEIXIS’ one to establish it as a topical discourse referent in
the current stretch of discourse, and subsequently a ‘LOW DEIXIS’ mark-
ing to maintain its current topicality (as long as it is not superseded by a
new referent bearing a higher claim to topic-worthiness).
20. Strauss (1993: 403–404), though recognizing that the three-member system
of indexical pronouns she is dealing with would normally fall within the CS
DEIXIS system, nevertheless terms it ‘FOCUS’. This terminological shift is
highly significant in the light of the discussion in Section 2.3 above.
21. See also the studies of the Finnish demonstratives by Östman (1995) and
Laury (1997), or of the Dutch demonstrative determiners by Maes (1996:
ch. 4).
22. Admittedly, the GH is an ‘implicational’ scale, so that particular forms may
move up or down the scale according to the drawing of certain implicatures
based on one or the other of Grice’s two Quantity maxims.
23. See also Mackenzie and Keizer (1991: 193–194), Hannay (1991: 138), and
Siewierska (1991: 162) for a similar view within standard FG. Outside FG,
see also Lambrecht (1994: 129–130, 168). Mackenzie and Keizer (1991:
194) suggest that Dik's (1997a) ‘NewTops’ should be analysed as a type of
‘Presentative Focus’ (a subtype of New or Completive Focus). Gómez-
González (2001: §5.2.4) provides a survey of published criticisms of the al-
leged topical status of so-called ‘NewTops’.
24. See Cornish (2002: §4) on this particular problem, presented in terms of
French examples where subject and (intransitive) verb are inverted follow-
ing a clause-initial locative satellite, and where it is the predicate corres-
ponding to the head noun of the subject term expression which is assigned
the Focus function, thereby acting as predicator (example: A l’horizon
couve un orage ‘On the horizon (there) brews/is brewing a storm’); or
where canonical subject–verb order obtains, the verb then receiving Focus
function and assuming its full lexical value as predicator for the clause (ex-
ample: A l’horizon, un orage couve ‘On the horizon, a storm is brewing’).
References
Aoyama, Takashi
1995 Deixis and value: a semantic analysis of the Japanese demonstra-
tives. In: Ellen Contini-Morava and Bonnie S. Goldberg (eds), 289–
320.
Birner, Betty J. and Gregory Ward
1998 Information Status and Noncanonical Word Order in English. Am-
2000 Discourse organisation and anaphora in Latin. In: Susan Herring,
Pieter van Reenen and Lene Schoesler (eds), Textual Parameters in
Older Languages, 107–137. Amsterdam and Philadelphia: Benja-
mins.
Brown, Gillian and George Yule
1983 Discourse Analysis. Cambridge: Cambridge University Press.
Cheshire, Jenny
1996 That jacksprat: an interactional perspective on English that. Journal
of Pragmatics 25: 369–393.
Contini-Morava, Ellen and Barbara S. Goldberg (eds)
1995 Meaning as Explanation: Advances in Linguistic Sign Theory. Berlin
Cornish, Francis
1998 Les ‘chaînes topicales’: leur rôle dans la gestion et la structuration du
discours. Cahiers de Grammaire 23: 19–40.
1999 Anaphora, Discourse, and Understanding. Evidence from English
and French. Oxford: Clarendon Press.
2000a L’accessibilité cognitive des référents, le centrage d’attention et la
structuration du discours: une vue d’ensemble. Verbum 22: 7–30.
2000b The Functional Grammar conception of discourse anaphora: A (con-
structive) critique. Working Paper in Functional Grammar 73.
Amsterdam: University of Amsterdam.
2001 “Modal” that as determiner and pronoun: The primacy of the cogni-
tive-interactional parameter. English Language and Linguistics 5:
297–315.
2002 'Downstream' effects on the predicate in Functional Grammar clause
derivations. Journal of Linguistics 38: 247–278.
Dik, Simon C.
1997a The Theory of Functional Grammar. Part 1: The structure of the
clause. Berlin and New York: Mouton de Gruyter.
148 Francis Cornish
Dik, Simon C.
1997b The Theory of Functional Grammar. Part 2: Complex and derived
constructions. Berlin and New York: Mouton de Gruyter.
Diver, William
(n.d.) Avoidance of the obvious: the pronoun as a minimax solution. Un-
published course text for Linguistics G 6801.
1995 Predictions from the hypotheses about the characteristics of the texts
(§3.4.4.1), and Measuring the strength of the correlation (§3.4.4.2),
extracts from “Theory”. In: Ellen Contini-Morava and Bonnie
S.Goldberg (eds), 34–35 and 35–37.
Erteschik-Shir, Nomi
1997 The Dynamics of Focus Structure. Cambridge: Cambridge University
Press.
García, Erica
1977 On the practical consequences of theoretical principles. Lingua 43:
129–170.
Glover, Karen
2000 Proximal and distal deixis in negotiation talk. Journal of Pragmatics
32: 915–926.
Gómez-González, María de los Ángeles
2001 The Theme-Topic Interface. Evidence from English. Amsterdam and
Gundel, Jeanette K.
1999 On different kinds of Focus. In: Peter Bosch and Rob van der Sandt
(eds), Focus. Linguistic, Cognitive, and Computational Perspectives,
293–305. Cambridge: Cambridge University Press.
Gundel, Jeanette K., Nancy Hedberg and Ron Zacharski
1993 Cognitive status and the form of referring expressions in discourse.
Language 69: 274–307.
Hannay, Mike
1985a English Existentials in Functional Grammar. Dordrecht: Foris.
1985b Inferrability, discourse boundness and subtopics. In: A. Machtelt
Bolkestein, Casper de Groot, and J. Lachlan Mackenzie (eds), Syntax
and Pragmatics in Functional Grammar, 49–63. Dordrecht: Foris.
1991 Pragmatic function assignment and word order variations in a Func-
tional Grammar of English. Journal of Pragmatics 16: 131–155.
delphia: Benjamins.
Hengeveld, Kees
Huffman, Alan
1993 Cognitive and semiotic modes of explanation in Functional Gram-
mar. Talk given at Rutgers University, 10 October 1993.
1997 Categories of Grammar: French lui and le, Amsterdam and Phila-
delphia: Benjamins.
forthc. The Control-Focus interlock. In: Modern English: A Columbia
School Grammar.
Lambrecht, Knud
1994 Information Structure and Sentence Form: Topic, Focus, and the
Mental Representations of Discourse Referents. Cambridge: Cam-
Lambrecht, Knud and Maria Polinsky
1997 Typological variation in sentence-focus constructions. In: Singer,
Kora, Randall Eggert and Gregory Anderson (eds), Papers from the
Panels, 33rd Regional meeting of the Chicago Linguistic Society.
189–206.
Laury, Ritva
1997 Demonstratives in Interaction. The emergence of a definite article in
Finnish. Amsterdam and Philadelphia: Benjamins.
Leonard, Robert A.
1995 Deixis in Swahili: attention meanings and pragmatic function. In:
Ellen Contini-Morava and Bonnie S. Goldberg (eds), 271–287.
Mackenzie, J. Lachlan and M. E. Keizer
1991 On assigning pragmatic functions in English. Pragmatics 1: 169–
215.
Maes, Alfons
1996 Nominal Anaphors, Markedness, and the Coherence of Discourse.
Leuven: Peeters.
Östman, Jan-Ola
1995 Recasting the deictic foundation, using Physics and Finnish. In: Ma-
sayoshi Shibatani and Sandra Thompson (eds), Essays in Semantics
and Pragmatics in Honor of Charles J. Fillmore, 247–278. Amster-
Reid, Wallis
1991 Verb and Noun Number in English: A functional explanation. Lon-
don and New York: Longman.
Reinhart, Tanya
1981 Pragmatics and linguistics: An analysis of sentence Topics. Philoso-
phica 27: 53–94.
150 Francis Cornish
Rosengren, Ingrid
1997 The thetic/categorical distinction revisited once more. Linguistics 35:
439–479.
Siewierska, Anna
1991 Functional Grammar. London and New York: Routledge.
Strauss, Susan
1993 Why this and that are not complete without it. In: Katie Beals et al.
(eds), Proceedings of the 29th Regional Meeting of the Chicago Lin-
guistic Society, 403–417. Chicago Linguistic Society.
Van Hoek, Karen
1997 Anaphora and Conceptual Structure. Chicago: University of Chi-
cago Press.
Vet, Co
1998 The multilayered structure of the utterance: about illocution, modal-
ity, and discourse moves. In: Mike Hannay and A. Machtelt Bolke-
The complementarity of the process and pattern
interpretations of Functional Grammar
Michael Fortescue
1. Introduction
From its inception, the Functional Grammar formalism has in principle
been interpretable in two ways, either as a static system of functional
choices, or as a dynamic process model (more specifically, a production
model). Interest in the second kind of interpretation has increased with the
rapprochement between cognitive and functional linguistics in recent years,
and not everyone within the FG sphere has been as sanguine as Dik himself
in considering the formalism directly implementable as a computer model
of cognition.1 What I want to explore in this chapter is the question of what
the consequences of interpreting FG as a process model really are.
If one wants the FG model to represent production and/or comprehen-
sion directly, one should, I would suggest, base it on underlying cognitive
processes which have an intuitive feel of plausibility about them (not, for
instance, on hypothetical neural circuits at some dehumanized ontological
level). As has been argued by Harder (1992: 313ff.), one aspect of the
model that simply does not fit a process/procedural interpretation is the
first-order predicate formalism of the lowest level of the model.2 Taken
straight from the abstract world of logicians like Quine (1960), representa-
tions of this kind, with their variables, quantifiers and open predicates, are
frankly not obvious candidates for psychological ‘reality’. Strawson, in de-
scribing the cognitive origin of ‘perspicacious grammar’, specifically
warns against introducing variables and quantifiers as opposed to individ-
ual particulars at the basis level of the logical representation of
propositions (Strawson 1974: 117). That is because it is individual particu-
lars (and their more abstract, higher-order entities derived by ‘sub-
152 Michael Fortescue
stantiation’) that enter into the act of reference, not variables. The basic
cognitive processes as I see them are rather Whiteheadian ‘prehensions’,
the grasping and assimilation of both perceptual and conceptual ‘data’,
specifically ‘propositional prehensions’, which link the prehension of a
predicate to the prehension of a set of logical subjects (the totality of rele-
vant ‘entities’ bound by the predicate – cf. Fortescue 2001: §1.2.5). At this
level of cognition we are not yet talking of linear chains of linguistic signs:
the processual ‘concrescences’ whereby multiple data-to-be-expressed are
integrated on the way towards a determinate linguistic output should be
taken as having their own internal complexity (unless purely automatic re-
sponse is involved).
Now Dik’s term variable ‘x’ in underlying predications simply stands
for a first-order entity (as opposed to a proposition ‘X’, for example). In a
nuclear predication variables and their first restrictors are replaced by
terms. ‘Firstness’ is all that determines which predicate restrictor is to be
treated as term head by binding with a preceding variable. This kind of no-
tation is innocuous and does indeed serve a useful purpose, namely in
relating underlying clause structure to the lexicon, where individual predi-
cates (including those defining terms) can be represented as a predicate
frame relating directly to a particular kind of State of Affairs. However,
this is a matter of pattern, not of process. There is no guarantee that the
straightforward linkage between predicates and SoAs has any direct rele-
vance for cognitive processes as such.
I would like now to suggest that we may need two versions of the FG
model, a Process and a Pattern version, rather than just one (Dik’s original
one, say) serving both purposes – or, if there is agreement on a monolithic
framework again, then it should at least be of such a nature as to be inter-
pretable in two clearly distinct ways and not remain indeterminately
‘hybrid’. This has been the subject of much debate in recent years within
FG circles, in the form of discussions of the upward limit of layering of the
theory. Dik himself proposed that his model should be envisaged as em-
bedded in – or side by side with – a distinct discourse grammar (Dik 1997:
409ff.), and this idea has been followed up in two principal ways (as re-
flected in the various articles in Bolkestein and Hannay 1998),3 either by
advocating the extension of the original model to include such a ‘discourse
grammar’ or by calling for their treatment as two separate, interacting
modules. I would suggest that neither of these options is really what we
need if the original functional model is to be seen in cognitive terms. For
you cannot simply embed pattern within process (or vice versa). The model
as it is (as pattern) needs a process interpretation – where by ‘process’ I re-
Process and pattern interpretations 153
fer to all procedurally interpreted levels of the model, from the choice of
basic predicate (and its relationship to the lexicon) up to the highest level
of illocutionary intention (and its relationship with still broader discourse
context).
My own position as regards the upper limit of the model as such is thus
twofold: from the pattern point of view (the actual ‘emic’ coding choices of
grammar) it is the illocution, just as Dik proposed. From the process view,
illocutions are simply an intermediate stage of organization whereby com-
plex intentions (perlocutions, etc.) find their way to expression via the
available linguistic code. The upward limit from this viewpoint would cor-
respond to the Whiteheadian ‘subjective aim’ defining the individual
communicative act. Such an intentional stance is, along with propositional
content, an integral part of the Whiteheadian judgment (which, unlike those
of standard propositional logic, is not limited to the assignment of positive
and negative truth-values). The way in which the intention to utter such a
judgment is worked out with the help of the ‘template’ of grammar is one
of the constraints on pure expression. A grammar as such is simply a set of
abstract ‘forms of definiteness’ (Whiteheadian ‘eternal objects’), and the
FG model (as pattern) is best regarded as a higher-level set of generaliza-
tions across the grammars of many languages. The obvious direction to
move in describing a process interpretation of FG is thus top-down (com-
pare also Hengeveld this volume).
Now a cornerstone of FG is the notion that pragmatics contains seman-
tics, which in turn contains morphosyntax (as in Dik 1989: 7). So if
pragmatics really embraces all levels of grammar it should be seen to ex-
tend down to the very lowest level of the process model it embraces,
namely the organization of predications. This will be my starting point in
what follows. Also for Hannay (1991: 146ff.) the ‘place to start’ is with
pragmatic/message organizing (choices to do with new versus given infor-
mation, etc.).4 Note the difference from my proposal, however: Hannay
sees his message management module or component as an interface be-
tween a product-oriented model and a process one (of verbal interaction).
In fact he indicates choice of ‘mode’ as taking place at the highest level of
the extended FG model, i.e. as part of the ‘pattern’ grammar, e.g. TOPIC
(mode) DECL E1: [X1: ...]. My own approach is not really modular at all:
both discourse and grammar have their pattern and their process aspects,
only the pattern aspect of discourse is far more heuristic, relying heavily on
inference and background in ways that are not necessarily coded ‘emically’
in any grammar, and its patterns are across situations of use and means of
achieving ends, not across coding categories. The consequence of taking
the overarching nature of pragmatics seriously must, it seems to me, result

in a process interpretation of language that does not set discourse and
grammar side by side on the same ontological level, but rather treats gram-
mar as abstract pattern or template at the service of communicative
purposes. Let us avoid formalizing the patterns of pragmatics as if they
were merely an extension of the patterning of grammar. This amounts to
having one’s cake and eating it as regards the choice between incorporating
a discourse ‘level’ into the existing FG model (in the manner of Hengeveld
1997) or interfacing it with a separate ‘discourse grammar’ (in the manner
of Hannay 1991). It is in fact compatible with Hengeveld’s more recent
proposal (this volume) for combining the two approaches to discourse
within the FG model.5
In this chapter I shall be extending the notion of letting the highest (dis-
course-pragmatic) level of a processual interpretation of the FG model ‘dip
down’ to the lowest, predicate level, as proposed for predicate formation
rules in Koyukon and West Greenlandic (Fortescue 1992: 131). Languages
vary on this dimension, some allowing whole propositions as input to cer-
tain derivational rules, others strictly limiting input to stem predicates (as
English, by and large, does).6 In the following section I shall specifically
look at relevant phenomena in Nootka. My representations will focus on
the level of the (extended) predication (= Whiteheadian proposition, as in-
gredient in the ‘judgment’, which corresponds more or less to Dik’s
‘proposition’), since that is where I claim the link to psychological reality
must be made. The ‘missing’ relationship between this level and the lexi-
con (with its reflection of SoAs) will be returned to in the final section.
2. Choice of predicate in Nootka

In Nootka, as in the other Wakashan and neighbouring Salishan languages
of western Canada, the distinction between lexical nouns and verbs is noto-
riously slippery. Moreover, the choice of constituent that is treated as
(main) sentence predicate often goes contrary to European expectations.
What corresponds to an adverbial or a quantifier, for instance, might be
chosen as predicate, followed by a bare (structurally indeterminate) verbal
or nominal expression whose relation to the predicate – e.g. as subject – is
very loose (and must be inferred). Thus the following sentences appear in
Mithun (1999: 188–189, citing Nakayama 1997):
(1) hu:ak-a-qu:s mamu:k

early-TEL7-CONDIT.1SG work
‘I would work early’
(2) a:n-u:a-qu:-č hauk umi:qs-ak
only-do.together.with-TEL-CONDIT-INFER eat mother-POSS
’He used to eat alone with his mother’
(3) mu:-a-qu: maayi hi:
four-TEL-CONDIT.3 family there.in.house
’There used to be four families living in this house’
(4) sačica-am-it-iš-a wai:q
unceasingly-CAUS-PAST-IND.3-PL homesick
‘They constantly got homesick’
Swadesh (1936: 84) gives a further example with a locative predicate

(from a dialect with a different indicative paradigm):
(5) hi-ap-’a-’at-ma waič

there-CAUS-TEL-SHIFT-INDIC sleep
’He is allowed to sleep there then’
It is important to realize that although the constituent that is chosen as

predicate in such sentences is somehow focused or emphasized, one should
not assume that we are dealing here with Focus assignment at the prag-
matic-function level of the traditional FG sort (although the etic functions
concerned do of course overlap). The ‘focused’ element is rather what the
speaker in a general way considers to be ‘newsworthy’ in the discourse
context (this reflects the strategy of ‘most newsworthy first’ found in many
North American languages with pragmatically controlled word order – cf.
Mithun 1987). Note that the language does have a separate contrastive fo-
cus construction when focus is on a term, using predicates uħ (for subject)
or u:kwi (for object), as in:8
(6) a. qaħsa:p uħ John muwac

kill do.it John deer
‘(It was) John (who) killed the deer’
b. qaħsa:p-’a kwatya:t u:kwi qwayac’i:k
kill-TEL Kwatyaat do.it.to wolf
‘Kwatyaat killed the wolf’
The phenomenon in (1) to (5) is part of a structural phenomenon which

Nakayama calls ‘serialization’, which combines multiple clauses into units
expressing a single State of Affairs (actually some of the example sen-
tences strictly speaking represent other parallel clausal relationships such
as ‘modification’, which do not allow word-order variation according to
pragmatic factors in the way serialization as Nakayama defines it does).
This is an event-encoding strategy for combining predicates into informa-
tionally dense predications – dense in terms of participant structure or
event specification – which cannot be expressed in a single word (Naka-
yama 1997: 119ff.).9 The shared inflectional suffixes (none of them
actually obligatory) are generally gathered on the first constituent, what-
ever it is. The non-initial element can take a more limited array of suffixes
(mainly for aspect, as in (7) below). Serialization proper serves a number
of discourse purposes besides that of the kind of construction illustrated
above. Sentences such as (2) do not really bear any special focus according
to Nakayama (pers. comm.), since there is no flexibility of ordering choice
with the particular predicate involved. Similarly, the order is obligatory
with negative and interrogative words such as the following, where there is
no special focus – or rather only ‘default’ focus on the initial constituents:
(7) wik’-a hauk-ši

not-TEL eat-MOM
‘He did not eat’
(8) aqish-w’it’as-a-it-ħ suw’a inm’i:-či
why-about.to-TEL-PAST-INTERR you snail-INCHO
‘Why would you become a snail?’
(9) wa:-yaq-pi-mit-ħ awa: town
which-most-PAST-3SG.INTER near town
‘Which was the closest town?’
In (9) note that even the modifying suffixes that belong semantically with
the adjectival predicate awa: has been positioned on the interrogative
constituent appearing as predicate of the whole sentence.
In the following the adverbial predicate glossed ‘very’ can be said to be
well-suited as focus owing to its inherent meaning:
(10) i:ħ kwi:s-ħi

very strange-DUR
‘It is very strange’
Note that it can still remain as main sentence predicate when converted
semantically to a transitive verbal expression by addition of the ‘inverse’
(‘shifter’) suffix -’at. This indicates action on an undergoer higher on the
animacy scale than the subject (here 3SG outranks 3PL).10 A rough gloss of
(11) would be ‘they did it very much to him, being strange’, with being
strange less central to the construction than doing very much.
(11) i:ħ-’at kwi:s-ħi

very-SHIFT strange-DUR
‘People treated him very strangely’
A parallel phenomenon is found within the nominal phrase, as in the

following, where the only inflection present (the definite article) is attached
to the modifier constituent (which must come first), even though it goes
semantically with the head:
(12) camaħta-i qwayac’i:k

real-DEF wolf
‘the real wolves’
There is clearly no marked focus on the modifying constituent here.

To return to the verbal sentence examples, the next question is: if the
initial element of serialized construction is the (main) predicate, how is the
logical subject of these sentences expressed? There are several possibili-
ties. It may be an overt argument later in the sentence, or some referent
inferred from context, or it may be a whole verbal expression understood
as ‘nominalized’ (as in the first sentence). As can be seen, there is no direct
marking of what the subject of the sentence is – with the exception of the
few examples of pronominal inflection. Compare also the following exam-
ples of the clausal possessor-incorporating construction, whereby the
possession relationship is marked by a special affix on the main sentence
predicate immediately preceding an incorporated pronominal marker of the
possessor:
(13) m’aw’a:-a-’at-ukw-ina hiyiqtup

deliver-TEL-SHIFT-POSS-1PL things
‘He brought our things’ (or ‘We had our things brought by someone’)
(14) y’ap’ic-uk-kw-is siy’aq
blue-DUR-POSS-1SG I (SUBJ)
‘Mine is blue’
Note that the independent pronoun in (14) is in its subject form, not its
possessor form: the first-person possessor affixed to the first word has
subject properties, as Nakayama puts it, and the possession marker itself
produces a verbal expression. In such constructions the possessor is topic,
i.e. salient in the discourse, and treated as a grammatical argument of the
predicate expression. Nakayama argues elsewhere that the category of
syntactic ‘subject’ (and ‘object’), as opposed to semantic ‘agent’ (and
‘undergoer’), is in general difficult to maintain for this language (Naka-
yama 1997: 99ff.).
3. Nootka phrasal affixes and zero-verbalization

Now it could be argued that the ‘suffixes’ attached obligatorily to the initial
constituent in the serialization type of construction are enclitic and that that
constituent is not ‘really’ the predicate of the sentence at all. But Naka-
yama argues that this is not the case, for whether the suffixes concerned are
inflectional or derivational they have properties that are un-clitic-like. Thus
in Nakayama (1994: 266) he compares the following sentences, all contain-
ing ‘phrasal affix’ –i:  ‘make’:
(15) ča:pac-i:
canoe-make
‘He made a canoe’
(16) u-i: čapac
nice-make canoe
‘He made a nice canoe’
(17) mu:kw-i: u čapac
four-make nice canoe
‘He made four nice canoes’
As he argues, the ‘verbalizing’ suffix here takes a whole phrase as its base
(though it must morphosyntactically stand on the first word of the phrase,
whatever that is). Its scope properties are directly comparable to those of
such West Greenlandic affixes as -liur- ‘make’, which can also be added to
an NP leaving the modifying element stranded (in the instrumental case).11
Whether one can jump from this to the claim that these suffixes are not en-
clitics at all is perhaps a matter of definition: they are at least not
prototypical enclitics, being lexically much weightier than clitic elements
usually are.
There is another kind of lexical suffix in Nootka than the illustrated

‘governing’ type, namely the ‘restrictive’ type as in (18), which has no par-
allel in West Greenlandic. There are a great many suffixes of this type for
expressing location and instrument, etc., in Nootka, as in most of the lan-
guages of the surrounding Northwest American linguistic area:
(18) quac-’aq
person-inside
‘There is a person inside’
Here the topic is person (tracked as such in following discourse), as op-

posed to canoe in (15) for example. The verbal/nominal/adverbial status of
the suffix is lexically quite indeterminate – it is only in the context of a
predication, such as this sentence, that it is clear that its function is to mod-
ify (for location) the left-most constituent understood as a zero-verbalized
nominal, i.e. in an existential context. It only differs from other predicates
in being an obligatorily bound form.
This leads us back to the question of the status of nominal versus verbal
expressions in Nootka in general, for as Nakayama puts it, although it is
not easy to differentiate between isolated verbs and nouns, it is generally
clear what element is functioning as a verbal or a nominal, either because a
verbal morpheme functioning as a nominal or a nominal one functioning as
a (sentence) predicate is marked as such by affixation or because the dis-
course context renders it apparent (when, for instance, serialization is
involved). In the following sentences (from Swadesh 1936, phonemicized
according to Nakayama) the verbal affix on the first word and the nominal
one on the second make this quite clear, whichever of them is chosen as
sentence predicate:12
(19) mamu:k-ma qu:as-i

work-INDIC man-DEF
‘The man is working’
(20) qu:as-ma mamu:k-i
man-INDIC work-DEF
‘The working one is a man’
The situation is somewhat obscured by the fact that sentential affixes

(inflections) are not obligatory, unlike, say, the inflections of West
Greenlandic. Nevertheless, a lexical item referring to an entity (a proto-
typical ‘nominal’) cannot function as a verb except in copular sentences of
identification and class-assignment or the like (as above), i.e. where one
alternatively could talk of a zero-marked derivational copula (something
actually quite widespread in the languages of the world; see Fortescue
1998: 62ff.). Likewise, a lexical item referring to an action or state (a pro-
totypical ‘verbal’ morpheme) generally has to be nominalized in order to
function as a nominal (e.g. by addition of the definite article, as above).
4. Word classes and the relationship between predication and

lexicon in FG
The question I want to turn to now is this: how are the facts about Nootka
sketched above to be fitted into the FG model? First, the choice of the
‘most important/newsworthy’ element that is positioned first in the typical
clause (as in sentences 1–5) could in theory be made at the level of prag-
matic function assignment (wherever that may be). On the other hand,
Nootka ‘phrasal’ suffixes (as in 15–17), like semantically similar West
Greenlandic ones, can be treated by individual predicate formation rules,
and its ‘zero copula’ construction (as in 20) by term-predicate formation
rules in the manner of Dik (1989: 172) or, preferably, by non-verbal predi-
cate frames in the manner of Hengeveld (1992: 32). The lack of evidence
for the assignment of syntactic Subject status to particular arguments (as
mentioned in connection with sentences 13–14) would seem to be ac-
counted for by aligning Nootka with those numerous languages where FG
recognizes that Subject assignment simply does not occur. Furthermore,
Dik’s fundamental principle of treating both nominals and verbals as un-
derlying predicates would seem to be amply justified for this language,
where all lexical content morphemes are indeed predicates in the sense that
they can function – given an appropriate morphosyntactic ‘slot’ – as
verbals or nominals or modifiers of either. However, with the standard ver-
sion of FG, designed principally for non-polysynthetic languages, there are
problems with all of these suggestions.13
Specifically as regards clause-initial ‘focused’ constituents, it appears
that the choice of first constituent is not simply a matter of Focus in Dik’s
original pragmatic function sense. The constituent chosen is either one dis-
playing general ‘newsworthiness’ (in the discourse context) or one required
by the lexical items and/or construction involved (as in 7–9). There has
been much discussion within FG in recent times as to whether Focus (and
other discourse-related functions) really belong in the model here at all or
should be part of a separate discourse grammar or module (as discussed
briefly in Section 1 above). Nootka is a language in which complex dis-

course-pragmatic factors affect the basic syntactic choices that modulate
the linear succession of constituents, and not all of this complexity can be
fitted into the simple binary choices of FG pragmatic function assignment.
In part this is a matter of the interface between lexicon and basic predica-
tion. ‘Newsworthiness’ is a complex matter of weighting various
contextual and lexical factors, not one of simple ‘digital’ choices. In other
words, it is hard to argue here for a discrete pragmatic function assignment
level of the classical FG kind.14
The problem with the lack of Subject assignment, on the other hand, is
that Nootka does have a morphologically marked inverse construction (ex-
tending also to impersonal subject constructions, as in 10), which is in part
a matter of syntactic function level ‘perspectivization’ but is in part also
determined (and quite automatically so) by semantic animacy – similar to
the situation with the inverse construction in Athabaskan languages.
The ‘emic’ (coded) grammatical categories of Nootka appear to be ex-
ceptionally context- (and semantics-) dependent. As regards parts of
speech, Nakayama distinguishes, as we have seen, only verbal and nominal
expressions (apart from uninflected particles), both of which are possible
as arguments in serialized/complement syntactic configurations or as modi-
fiers. Basic clause structure is, as Nakayama puts it, imposed top-down
(Nakayama 1997: 78ff.). It is the basic propositional articulation of the
speech chain that holds it all together structurally.
Here is a language with maximal pragmatic control of syntax – much
inference from context is needed to interpret its syntactic cohesion, though
its pragmatic articulation is transparent. It is a language in which ‘high-
level’ illocutionary or pragmatic factors may reach down all the way to af-
fect the choice between predicate and argument status. There are very few
structural clues to signal this: no inflectional categories are obligatory, and
if these are overtly expressed they are typically attached to the first words
of a phrase, whatever its function, with a multiplicity of functions and
predicate frames being associated with individual lexical morphemes. In
sentences in which constituents expressing time, negation, or other ‘pe-
ripheral’ adverbial elements are chosen as the main clause predicate, as in
(2), context-dependent emphasis of some sort is generally involved – al-
though this is not always strictly a matter of Focus. It may rather be a
matter of default, ‘unmarked’ focusing (e.g. with the negative ‘verb’,
which must come first). The first element chosen in the serialization con-
struction, where there is a choice, is simply the ‘most newsworthy’ in the
broader discourse context.
As regards ‘phrasal suffixes’, the situation is much the same as I have

described for West Greenlandic (Fortescue 1992: 131), namely that if this
is a matter of derivation in the Fund (a perfectly reasonable assumption),
then such a derivational process must be able to ‘dip down’ (or rather ‘up’,
depending on the kind of diagramming of the FG model one has in mind)
into layers of clause structure where not just extended predications but
even whole propositions have been elaborated as input for the derivational
process concerned. The bound derivational suffixes of Nootka and West
Greenlandic need in any case to be treated independently of the predi-
cate/term formation processes into which they might enter and belong in
the lexicon with their own predicate frames. As the examples from Japa-
nese and English in note 5 show, this cross-level recursivity of derivation is
not limited to polysynthetic languages.
Finally, as regards the ‘zero copula’ construction, if one chooses to fol-
low Dik’s approach to copular constructions, a conversion of a lexical
‘term’ to a predicate expression is assumed, but if all lexical items are am-
biguous as to term or verbal status in Nootka, there is no need for such a
conversion at all – unless one first produces a term by term formation and
then converts it back to a predicate by term-predicate formation, a seem-
ingly unmotivated extra complication for this type of language. If, on the
other hand, one follows Hengeveld’s approach mentioned above, one
avoids having to postulate ad hoc predicate formation rules, and can treat
non-verbal predicates in terms of predicate variables (f1), just like verbal
ones. However, there is very little evidence in Nootka for distinct predicate
frames (with predetermined argument sets) being associated with particular
non-verbal predicate construction types any more than with verbal predi-
cates, so it is hard to tell if this is an advantage or not. I shall return to this
matter below.
In sum, the nub of the problem is the relationship between the lexicon –
with its predicate frames corresponding to discrete SoAs – and the rest of
the model (in particular the choice of elements in the basic predication). A
given lexical morpheme (or “predicate”) in Nootka may enter into various
constructional “frames”, with the valency even of a specific verbal frame
being rather vague compared to the situation in more familiar language
types. In Nootka, lexical morphemes are not anchored in the lexicon in
quite the same way as in more familiar languages: it would appear to be a
typological fact one must live with (and FG claims typological adequacy)
that some languages have a much closer correspondence between States of
Affairs (SoAs) and lexical items than do others (compare Fortescue 1992
for another polysynthetic language – Koyukon – which does, unlike
Nootka, have a very close correspondence between the two). Perhaps the
mistake is to take the SoAs of the original model to be truly universal.
An attempt to incorporate a typology of languages with varying degrees
of rigidity of word-class distinctions within FG has been made by
Hengeveld (1992: 45ff.). At first sight Nootka would appear to go with
languages like Tongan, representing the highly ‘flexible’ type, in which
predicates can be used in practically any function. This contrasts with
highly ‘rigid’ languages like Cayuga, which may have only one part of
speech, the predicate, but where this must be morphologically derived to
function as other parts of speech than the verb. In fact Hengeveld’s ex-
tremes meet, since in ‘rigid’ languages any one morpheme may be
extremely flexible as regards morphosyntactic context, and the real contrast
is with ‘specialized’ languages like English with a full array of morphosyn-
tactically distinct parts of speech. The problem with fitting Nootka into this
typology is that the distinct morphological marking of predicate words in
the one or the other sentential or phrasal function is not obligatory, so that
it partakes of both the Tongan and the Cayugan type depending on whether
there is suffixation present or not (and, as we have seen, suffixes tend to be
shifted to the first word of constructions, whatever it is). Hengeveld (pers.
comm.) suggests that Nootka might represent a ‘rigid’ language in which
the relevant morphology has relatively recently lost its obligatory status.
It seems then that some languages, like Nootka, are simply more con-
trolled by discourse pragmatics than others which are more lexically
controlled and in which there is a closer link to SoAs.15 Nootka will, for
instance, split what in other languages correspond to unitary predications
into two clauses (of varying degrees of tightness) to emphasize the salience
of the undergoer argument. If this is so, we must look for another place in
the model (or an extension of it) where such matters as discourse-
determined choice of initial predicate can be accommodated. Short of put-
ting virtually all of the morphosyntax of Nootka within a ‘discourse
module’ (on the nature and categories of which there is no consensus
within FG today), the only obvious possibility would seem to be to intro-
duce discourse factors determining the choice of main clause predicate at
the lowest level of clause structure (e.g. by a feature ‘most newsworthy’),
obviously a severe, and undesirable, disruption of the model. This can be
overcome, I would suggest, by clearly distinguishing between a ‘process’
and a ‘pattern’ interpretation of the FG model. It is the former that must be
reassessed in the light of the kind of evidence I have presented. Let me
elaborate the point.
Nootka understood as ‘pattern’ (or grammatical ‘template’) presents

ample support for Dik’s treatment of both nominals and verbals as underly-
ing predicates.16 On the other hand, the question of which element to treat
as main clausal predicate and which as argument (and of how many clauses
to split up one proposition into) is hardly relevant to the pattern perspective
at all if, as I have shown, this is determined by higher discourse-pragmatic
factors, which must ‘reach down’ from the topmost level. It would only be
relevant if there were a more or less direct link between the State of Affairs
to be expressed and a particular predicate frame corresponding to it in the
lexicon (built up usually around a verbal core), but this link is precisely
what is extremely tenuous in Nootka, where the main predicate could just
as well be an adverbial or a quantifying expression, for example. It is diffi-
cult to see how the pattern perspective could accommodate ‘percolation’
down from the level of illocution (or beyond) in the way that Hannay
(1991) has suggested, since this would have to extend right down into the
lexicon itself – besides which, ‘percolation’ is surely a process and not part
of a pattern description at all.
The nature of the link between clause/predication structure and dis-
course articulation is also much less direct in Nootka than in English,
where the clause can accommodate several arguments and satellites of sin-
gle events. By contrast, a unitary predication may in Nootka be broken into
several clauses, e.g. with one salient referent in each. The Nootka clause is
structurally very simple and less flexible than in English as regards meet-
ing more elaborate discourse needs than just indicating ‘newsworthiness’.
It makes sense to describe the pattern grammar of Nootka as being without
syntactic – and indeed with very limited pragmatic – function assignment.
(This is something the FG model already allows for.) Thus Focus in the
general sense of ‘newsworthiness’ could in theory be left right out of the
grammar as an etic, not an emic category, although in the serialization con-
struction, where it triggers an actual ordering choice, it had better be
retained.17 The morphological ‘inverse’ can in turn be treated as a matter of
predicate formation in the Fund rather than of syntactic function assign-
ment, just like causative (and other morphological) valency-shifting
derivation processes. In other words, all that has to go from the traditional
pattern model is the direct link between lexical items and specific SoAs.
From the ‘process’ perspective, on the other hand, one can indeed en-
visage higher-level pragmatic choices (at the level of the illocution or
above) percolating all the way down to the basic predicate level and affect-
ing the choice of (main) predicate. Compare the following two
representations of sentence (1) above (I would work early, with early cho-
sen as predicate); (21a) is formulated within standard static FG grammar

(with a traditional Foc introduced at the pragmatic function assignment
level); (21b) is a representation of actual processes leading to the expres-
sion concerned on the basis of a unitary (though complex) Whiteheadian
prehension:
(21) a. DECL E1: [X1: [HABIT e1: mamu:k (d1x1 :1p(x1))Ag (hu:ak)TimeFoc (e1)]
(X1 )](E1 )
b. DECL ((hu:ak)Foc? (HABIT mamu:k (1SG)Act))
The representation in (21b) should be understood as being embedded in a

specific discourse-pragmatic background which articulates the embedded
proposition into a focal part (in the broadest sense) and a presupposed part
(i.e. as the source of the focal marking on hu:ak ‘early’). This corresponds
to an outer Whiteheadian prehension of the concept ‘early’ applied to (i.e.
qualifying) an inner proposition consisting in turn of an ‘indicative’ pre-
hension of the subject I and a ‘conceptual’ prehension of the predicate
work. The articulation of the predicative structure reflects the higher (or
rather outer) level choice: the two form an indivisible unitary actual occa-
sion of language use and cannot be separated in either time or space. The
sum, when coupled to an appropriate illocutionary intention (here simpli-
fied to ‘DECL’), corresponds to a Whiteheadian judgment.
Note that the features of the discourse-level trigger which cumulatively
assign Focus are not ‘emic’ elements of any finite system of contrasts but
simply a collection of relevant presuppositions and communicative inten-
tions, whose weighted effect is to incite the choice of early as main clausal
predicate (the question mark on the feature is meant to leave open the ques-
tion whether it is the same Foc as in 21a). I assume that also the ‘telic’
affix -a is introduced by percolation from the outer discourse level, func-
tioning as it does to indicate that the predication as expressed does indeed
represent one unitary clause (the speech chain is thus modulated to reflect
the propositional structure). This affix would be difficult to introduce at all
in the static FG model (as in 21a), since its packaging function is purely a
matter of discourse management.
In contrast to (21a), representation (21b) reflects the selection of early
as the basic predicate right from the start. In (21a) this choice would have
to be suspended until the operation of higher levels of the model closer to
the word-ordering expression rules – up to that point work would have to
be regarded as the basic predicate selected to correspond to the given SoA
(with the adverbial satellite, or predicate restrictor if one follows Henge-

veld’s approach, being focused later). This can hardly be called psycho-
logically plausible.
Recall what was suggested above as regards two-tiered thetic proposi-
tions: these too need to be indicated as such at the initial phase of a process
representation.18 Note also that term variables have been eradicated from
(21b). Moreover, the semantic role Actor has replaced Ag, since these roles
are, according to Nakayama (1997: 90ff.), restricted to Actor and Under-
goer in Nootka. The role Actor is required by the predicative construction
(an intransitive declarative clause) rather than being closely tied to the
lexical predicate itself. The Actor is no doubt the default argument associ-
ated with mamu:k (the default function of which is in turn verbal), but
other arguments, e.g. of place, could optionally have been indicated instead
of the Actor (nominal arguments are not usually marked morphologically
for semantic or syntactic role).
As to how the Nootka sentences we have looked at relate to Hannay’s
specific ‘message management’ types, one could say that (1–5) and (19–
20) at least would correspond to his REACTION mode (presupposing a
discourse context that requires Focus on the first non-term element of the
utterance). (6a) and (6b) would presumably also be in that mode, but with
contrastive Focus on a core term (one of the few constructions in Nootka
where Focus does not lie on the initial constituent). (7–9) could be seen as
reflecting the same mode, but perhaps NEUTRAL is more fitting, since the
initial position of the element concerned is lexically determined, there be-
ing no choice as to what is regarded as the sentential predicate; note that
Focus here has to be indicated in the lexicon as adhering to these words,
since they automatically trigger initial positioning. (10–11) and (13–14)
probably do reflect a choice and thus go with (1–5), but (15–17) are again
probably NEUTRAL, since the positioning of modifiers before heads is
obligatory. (12) and (18), finally, are not relevant here. Is this very helpful?
Certainly not as useful as for English, since Focus in the broad sense of
‘newsworthiness’ is so strongly associated with initial P1 position in
Nootka that it virtually always adheres to that position (except in the Con-
trastive Focus construction), and there is virtually nothing to choose
between – only what particular item in the message is to fill it.19 In general,
it looks as if the structure of Nootka is so strongly determined by dis-
course-pragmatic factors that there is little to be found that parallels the
complex interaction between such ‘triggers’ and the often conflicting re-
quirements of sentence-internal syntax displayed by English.
5. A minimalistic functionalism?
The question remaining, then, is whether the two perspectives compared in
the previous section are compatible. I would claim that they are actually
not, either as a unitary model embracing both, or as reflecting two separate
modules on the same ‘ontological’ plane. The relationship is – and must
remain – one of complementarity. This is in part because the variable-plus-
restrictor (‘predicate calculus’) format of the pattern grammar has no direct
process correlate – at least not one involving experientially interpretable
meaning. I have proposed equating the processes involved with White-
headian prehensions. These are situated within real contexts, in a world of
entities rather than variables and of complex embedded communicational
purposes rather than discrete digital choices of ‘function’. More impor-
tantly, ‘emics’ cannot be matched directly to ‘etics’. Purposes are the
domain of the latter; the former define the various means at hand to express
them. Given the typological distinction that this chapter has illustrated be-
tween languages with a tighter or looser linkage of lexicon and grammar
(and between languages whose grammar reflects greater or lesser dis-
course-pragmatic vs. lexical control of grammar), one surely cannot expect
there to be a universally valid division of labour between a grammar mod-
ule and a discourse module. What we need is not so much a module or a
grammar of discourse, but a set of (rational) principles allowing us to link
specific communicational intentions to the means provided by the abstract
categories of grammar (cf. Itkonen 1983: 177).
Perhaps Hengeveld’s proposed reformulation of the basic architecture
of FG can be understood in just these terms. Presumably the idiosyncracy
of Nootka, from this perspective, has to do with the relative transparency or
fluidity of the expression level in this language, which matches interper-
sonal level choices much more directly (e.g. ascriptive act decisions as to
what is to be the main predicate) than in more familiar European lan-
guages. However the FG model is extended, reformulated or reinterpreted
in the future, we can at least require of it that it should be able to account
for languages of the Nootkan type in which morphosyntactic indication of
function is rather minimal and where inference from context is constantly
required.20 If FG is to accommodate signed languages within its overall
framework the same could be said – the pragmatically determined fluidity
with which these articulate the flow of communication into discrete infor-
mation packets is indeed reminiscent of Nootka.
What in effect I have been arguing for in this chapter is a kind of func-
tionalist minimalism, parallel to the Minimalist Program of the generativists
(Chomsky 1993). Having shaken off the ‘superfluous’ baggage of transfor-

mations and the deep-versus-surface structure distinction and thus returned
towards the central nature of the linguistic sign as binding form and content,
minimalists continue their search for ever more economic generalizations as
to the universals of language patterning. Which is fine, as long as neither
they nor we persist in the mistake of thinking this has anything much to do
with real psychological processes. We, as functionalists, however, should not
be complacent in our belief that we, by default, as it were, must be the
guardians of the ‘true’ key to Process. If we are to do so, we must first strip
our process view of language of all its hidden pattern trimmings, just as some
generativists (though perhaps not Chomsky himself) have been willing to
abandon the process, or strategy, interpretation of their evolving models. As
regards the process interpretation of the FG model in particular, I propose
that we limit our representations of sentence meaning to simple layered
structures that take experiential ‘prehensions’ as the baseline.21
As regards the specific question I started out with, how to handle Focus
in Nootka and where in general it should be positioned in the FG model (on
some higher pragmatic/illocutionary level or in a separate interfacing dis-
course module), I can only propose that Focus should be part of both the
pattern and the process representations of utterances in Nootka, but that
this does not necessarily hold for all languages. (Nor, as mentioned above,
is Focus in the process sense of cumulative etic triggers necessarily equiva-
lent to the corresponding pattern sense of emic coding choice.) In some
languages, like Yukagir, Focus is highly grammaticalized within the pat-
terning of its morphosyntax, and in others it is not grammaticalized at all,
focality being borne solely by the ‘uncoded’ or analogue ‘etics’ of its pros-
ody (Nootka is somewhere in between in this respect).
As to where in the model Focus should be introduced, it looks as if it
needs to be admitted on process representations at the very outset, i.e. at
the outset of ‘concrescences’ that produce determinate utterances from un-
differentiated, holistic ‘intentions’ and work towards greater and greater
differentiation en route towards their final ‘satisfaction’ (= matching of in-
tention and output). Using the standard formalism (that of Dik 1989), this
actually means locating Focus at the lowest predicational level (albeit per-
colated down from the highest, pragmatic level), since choices there may
have an immediate influence on choice of structure at ensuing levels. It
should be borne in mind that process representations are by necessity lan-
guage-specific.
As regards pattern interpretations or representations (i.e. in the tradi-
tional FG model), it really does not matter very much, since ‘everything
comes out in the wash’ in the expression rules anyway, as long as the rele-
vant functions are marked somewhere. In Hengeveld’s revised model one
would presumably start with Focus choices made at the interpersonal level
(where the factors affecting its assignment are obviously ‘at home’, with
direct access to the communicative context), but it is also possible that dif-
ferent kinds of Focus are relevant to distinct levels. It may turn out that
both contrastive constituent focus and the kind of ‘newsworthiness’ focus
leading to choice of first constituent in Nootka are determined from the
outset at the ‘interpersonal level’, whereas the actual articulation of the ut-
terance into successive predicate and referential phrases realizing these
choices must be accomplished (as process proceeds generally ‘from left to
right’) by reference to pattern possibilities belonging to the other two levels
(for example activating the contrastive focus template or summoning and
positioning the ‘telic’ affix -a).
This may be a disappointing conclusion for those who would like a de-
finitive answer to the question of ‘where pragmatics belongs’ in the FG
model. But perhaps we should not be so fixated upon the interrelation be-
tween grammar and discourse after all, but think rather in terms of Pattern
versus Process, a distinction that cuts across this divide, since Pattern in-
cludes the more ritualized/grammaticalized aspects of pragmatics as well
as core clausal grammar. It is not yet clear how exactly Hengeveld’s new
‘architecture’ is to be understood in this respect. One possibility for ac-
commodating both perspectives in one model (which appears to be
Hengeveld’s intention) would be, as hinted at above, to interpret it as pure
Pattern on the vertical axis, but as Process on the horizontal axis (‘left-to-
right’ as he puts it). This at least avoids the problem of psychologically
suspect ‘percolation’ processes – as long as all relevant triggering factors
are potentially there ‘in the top left-hand corner’ from the start.22
Whatever the way forward, I suggest that the goal must be to bring the
Pattern and Process perspectives on FG into line in a relation of comple-
mentarity which does not blur over the essential distinction between them.
There is still much mileage to be derived from applying standard FG to the
abstraction of generalizations across linguistic pattern at all levels of its
layered template (including grammaticalized higher-level ones), thus
achieving still greater typological adequacy. Elaborating a specifically
Process variety of FG does not, it seems to me, require further theoretical
or formal sophistication – on the contrary, we need to simplify what we al-
ready have in the Pattern model. Meanwhile, let us at least avoid
ontologically suspect hybrid models. When talking about Pattern let us do
just that, but when talking about Process let us make it quite clear that we
are talking about something else. This does not preclude eventually elabo-
rating a single model that can be interpreted both ways. What we do need
is clear principles for (and constraints on) how to interpret such a model in
the one or the other way.
Notes
1. Cf. Dik (1989: 13) for the possibility of interpreting FG as a production
model that achieves psychological adequacy and which is said to “lay out
recipes for construing linguistic expressions from their basic building
blocks”. Grammar he sees from this point of view as tripartite (production
and comprehension plus a common store of elements and principles). In this
context he approved (1989: 13, fn. 9) of Nuyts’s proposal to give a proce-
dural interpretation to FG, although his own approach was different, being
closer to the predicational representations of standard FG. Thus Dik (1988)
suggested that the representations used by a hypothetical logic component
could be the same as those of the underlying propositional structure of sen-
tences, whereas Nuyts (1992) argues that deeper cognitive representations
are needed that cover both (see Anstey this volume: 38ff).
2. Harder’s ‘process vs. product’ dichotomy is not quite the same as my ‘proc-
ess vs. pattern’ one. For example, his representation of the phrase the old
elephant as: (1 (prop: old (ent: elephant))) is to be understood as a set of
nested instructions to produce the phrase in question, which would then be
the static ‘product’ of the procedural representation. By ‘pattern’ I mean the
grammar as such, the relational template of coding choices which constrains
the processes of expression in a given language. This is not a ‘product’ (ex-
cept in the historical sense), but rather a high-level abstraction across norms
of (communicative) behaviour.
3. See in particular Bolkestein (1998) for discussion of the question of the
level of the revised FG model at which the pragmatic function of Focus
should be assigned.
4. Hannay proposes several basic types of utterance ‘mode’ at this level, which
he labels TOPIC, ALL NEW, REACTION, NEUTRAL and PRESENTA-
TIVE. The last-mentioned, for example, corresponds to ‘thetic’ as opposed
to ‘categorial’ judgments typically expressed by special constructions for
introducing a NewTop, whereas REACTION mode in English triggers Fo-
cus in P1, and TOPIC mode Topic in P1. In general, the choice of mode
determines what will fill the P1 slot (in English). So for Hannay the initial
choice in a processual interpretation of FG concerns which entity – or other
semantic element – is to go in initial P1 position in the accruing utterance.
As Mackenzie interprets this (Mackenzie 1996: 144) these modes may be
represented as operators on the level of the illocution, whose effect perco-

lates through the hierarchy of levels, constraining the assignment of
pragmatic functions to elements of the utterance. Nuyts (1992: 281), too, as-
sumes that the first thing that has to be decided when mapping an
underlying cognitive ‘Singular State of Affairs’ onto an FG clause structure
is the choice of the basic predicate (and its terms). The factors he sees be-
hind this choice include higher-level matters of illocution, politeness and
attitude, etc. (as well as the availability in the lexicon of the language of
predicate frames corresponding to that SoA). However, he does not elabo-
rate on how, within his own cognitive interpretation of FG, such complex
choices are actually to be implemented.
5. More specifically, Hengeveld’s reformulated ‘interpersonal level’ can easily
be interpreted in terms of Whiteheadian complex subjective aims (e.g. illo-
cutions embedded within discourse moves), and propositional prehensions,
which combine indicative prehensions (Hengeveld’s ‘referential acts’) and
predicative conceptual prehensions (Hengeveld’s ‘ascriptive acts’). The
downward layering at this level is what is essentially new in Hengeveld’s
reformulation. This allows for the alignment of an ascriptive act at the in-
terpersonal level with an entity of any order at the representational level
below it, not just a zero-order predicate. In the Whiteheadian framework of
‘eternal objects’ (here ‘predicates’) and ‘nexūs’ (here read ‘entity types’)
this is to be welcomed, since the same eternal objects may be involved in
nexūs of various degrees of abstraction. The left-to-right restrictions on de-
cisions in production at the interpersonal level that Hengeveld calls for are
in fact largely provided for by the Whiteheadian theory of the successive
stages of the ‘concrescence’, as applied to linguistic utterances (cf. Fortes-
cue 2001: Appendix 2).
6. The phenomenon is by no means limited to polysynthetic languages like
Eskimo, and can be illustrated for Japanese, for example, as in the following
structure from Shibatani (1990: 248), where the affixed nominal tyuu ‘mid-
dle of’ is attached to an entire clause:
i. [Yamada-san ga tyuukosya o hanbai]-tyuu

Mr. Yamada NOM used.car ACC sell
′in the middle of Mr. Yamada’s selling used cars’
Compare even in English the man I saw at the station yesterday’s face, with
wide phrasal scope of genitive -s.
7. ‘TEL(IC)’ is a ubiquitous affix that, according to Jacobsen (1993: 250), sig-
nals ‘finitehood’ of a predicate, i.e. its relative independence as a clause,
not just the second part of a serializing construction, which in Jacobsen’s
Role and Reference terms is a ‘nuclear cosubordination’ construction in
which the second predicate shares any clausal mood marking of the preced-
ing one. Nakayama prefers not to use the term ‘finite’, since it smacks of
morphosyntactic finiteness of form (pers. comm.). He describes it as “mak-
ing the event feel more immediate and punctuated” and points out that it
plays an important role in discourse organization. Mithun glosses it as
‘momentaneous’, and Swadesh as ‘now, at that time’. In general, I have
used Nakayama’s glosses in the examples. Jacobsen’s characterization of
the affix’s principal function seems to me convincing, as distinguishing
clausal from nuclear cosubordination in the case of ambiguous ‘absolutive’
(uninflected) forms of predicates with no marking of illocutionary mood (or
subordination or modality) following initial predicates that may well show
such marking. Note that the CONDIT(IONAL) mood is used here to mark
habitual action.
8. But Focus in the sense of non-contrastive ‘newsworthiness’, as in the sen-
tences above, can also be assigned to (at least) object terms, in which case
they can be put in initial position, without these special relational predi-
cates, as in:
i. c’aak t’a:ps-at-a-qu:
river dive-sinking.into(water)-TEL-3.COND.INFER
‘He would dive into the river every once in a while’
A similar distinction between contrastive stress and merely ‘most newswor-
thy’ emphasis on a given entity referred to can be observed in most
languages. In English, for example, one would use respectively either a cleft
construction (or simply strong stress on a constituent) and (other) prosodic
means of emphasis. In Danish Sign Language the former is also expressed
either by emphasis (an emphatic production of a sign) or by something like
a cleft construction, as opposed to raised eyebrows marking a referent (or
any other segment of an utterance) as the most newsworthy part (Engberg-
Pedersen 1991: 64).
9. Despite the general polysynthetic character of this language, Nootka word-
forms can only accommodate a single verbal stem, typically marking no
more than one participant, so that transitive clauses are often split up into
two parts, one indicating the subject’s action, another involving the preposi-
tion-like transitive focusing predicate ‘do with respect to’ mentioned above
to indicate the object. The study of cross-linguistic variation in the density
of packaging of information in single clause units is still in its infancy (al-
though it is being broached in various ways, e.g. in terms of alternative
event structure construals – by cognitive linguists like Langacker – and of
the distinction between ‘verb and satellite framed’ languages – by psycho-
linguists like Slobin – and in terms of cosubordinate clause chaining within
Role and Reference Grammar). Some languages containing ‘serial verb’
constructions (like Chinese and Japanese) seem to fall somewhere between

English and Nootka as regards constraints on how much information can be
packaged in a single clause. The issue is complicated by the further ques-
tions of how (and how much) new information is distributed in successive
clauses (for example, one such unit per clause, as Chafe 1994: 109ff. sug-
gests) and whether semantically obligatory serial verb constructions in
Papuan languages, for instance, really represent more than one predicate per
clause (as opposed to a single complex one).
10. Typically of Nootka, neither 3rd person subject nor object are overtly
marked when context can supply them, as here. Nakayama glosses the ‘in-
verse’ marker as ‘SHIFT’, since, as he points out, its function is broader
than that suggested by the former term.
11. As in kusanartu-mik qaja-liur-puq (handsome-INSTR kayak-make-
3SG.INDIC) ‘he made a handsome kayak’. There is no way the affix here
could be construed as ‘enclitic’, by the way – it must follow the ‘incorpo-
rated’ head. In West Greenlandic there are even a few affixes, for example
-nirar- ‘say that’, which can take as base an entire proposition with its own
tense or modality marker (stripped of only person/mood inflection), as in
(i); though again its position is fixed
i. imiq nillir-sima-nirar-paa
water be.cold-PERF-say.that-3SG/3SG.INDIC
‘He said that the water had been cold’
12. Engberg-Pedersen points out to me that sentences such as these are com-
pletely parallel to similar utterances in Danish Sign Language (pers.
comm.).
13. It is not that traditional FG has any particular problem with languages that
have only two parts of speech (as can be argued is the case for Nootka),
namely nominals and verbals, which are distinguished almost entirely by
morphological criteria, i.e. by the type of inflectional and derivational proc-
esses to which they are subject respectively (Dik 1989: 163 specifically
mentions neighbouring Salishan languages in this context). What is more
problematical is the association of such a wide array of lexical stems with
both verbal and nominal (and other) uses. Term predicate formation in
Dik’s sense (also corresponding derivations from semantically adjectival
and adverbial elements) would on the one hand have to apply across a high
percentage of all sentences and, on the other, a wide array of potential SoAs
would have to be associated with individual lexical entries.
14. Deciding what is to be the main predicate given a complex communica-
tional intent can also be problematical with languages which do not display
the same ambiguity of major word classes as Nootka. Thus consider the ini-
tial choices in Japanese watashi wa kono hoo ga ii ‘I like this one’, in which
watashi, which corresponds to the English subject I, is marked as Topic (or
Theme), the inner grammatical Subject is kono hoo (‘this one’) and the
predicate is ii ‘good’. The standard FG treatment must start with basic
predication this one is good and somehow attach (as for) me at a later stage,
say as Theme at the pragmatic function assignment level, even though it is
clearly central to the basic communicational intention (the propositional in-
put to be expressed). In the following French sentence, it seems even more
difficult to decide on the basic predication: j’ai ma voiture qui est en panne
‘my car has broken down’, where the Topic is actually me and the domain
the whole predication (a kind of possessive cleft). Should the possessive
predicate be included in the underlying structure or introduced at the prag-
matic level? It is evidently a matter of focus, involving the presupposition
that the speaker has a car but with the focus actually on the whole sentence
as the essential new information (Lambrecht 1994: 14). Note that thetic,
presentational propositions like this are often expressed with the help of
clefts or dummy locative subject constructions, effectively splitting the
proposition into two – in Whiteheadian terms this corresponds to expressing
the indicative and the propositional prehensions behind a judgment as if
they were separate propositions (they are indeed separate ‘speech acts’), and
it would be natural for a process model to reflect this two-tiered prehen-
sional structure at the deepest (initial) level.
15. Note that Nakayama’s ‘predication’ corresponds roughly to Dik’s State of
Affairs, but this is for him not a matter of the lexicon at all (Nakayama
1997: 94ff.).
16. The predicate-based formalism of the original model does need to be ex-
tended, however, to allow for predicate-level restrictors (for the ‘restrictive’
lexical suffixes of Nootka at least) of the kind I argued for in connection
with Koyukon and in line with Hengeveld’s independently motivated pro-
posal for predicate variables (Hengeveld 1992: 31ff.).
17. The distinct focus construction illustrated in (6a) and (6b) above (which
highlights an argument for essentially contrastive purposes) does at all
events need to be marked – presumably on the pragmatic-function assign-
ment level of the traditional model.
18. The case is parallel: THETIC or PRESENTATIVE would be part of the
outermost bracketing – reflecting the act of embedding the inner proposition
with its NewTop within a broader contextual nexus – i.e. much the same af-
fect as is achieved by Hannay’s discourse management mode of that name.
19. The scope of that focus is in part determined by the presence or absence of
the TEL suffix mentioned above.
20. Dik (1989: 8ff.) himself stressed that much of communication is implicit
and not overtly coded, and this evidently is something that can vary more
from language to language than we often imagine. The reason Nootka ap-
pears to be close to the extreme here is the high degree to which purely
discourse-pragmatic factors directly determine coding choices (by contrast,
English employs to a greater degree subtle, analogue differences of intona-
tion to articulate such matters). Mackenzie (1998) follows Dik’s cue in his
recent exploration of the ontogenetic expansion of the ‘holophrase’, taken
as the essential item under Focus in an information ‘chunk’ that initially
fills the P1 position. This meshes nicely with Hannay’s emphasis on P1 as
the locus of initial message management choices. If English itself could
then be seen as in some sense also the phylogenetic result of a long histori-
cal development where P1 became gradually freed up for other items than
those under Focus (but with the initial situation remaining in adult holo-
phrases), one might want to say that Nootka represents a language that is
still much closer to its holophrastic roots (not for nothing is it also ‘polysyn-
thetic’). P1 is still reserved for Focus (in some sense) in Nootka in virtually
all its utterances types – here is positioned the one item of an information
unit that cannot just be dropped. This contrasts, by the way, with equally
polysynthetic West Greenlandic (which is basically SOV rather than VSO
like Nootka), where there is no P1 with that function – in that language Fo-
cus often adheres to final position in the clause (except in its contrastive
focal construction).
21. Note that I am discussing a level of representation corresponding to (poten-
tially) conscious experience. This I see as essentially a matter of inferential
processes (whether on-line or automatized), both lexically-cued and gram-
mar-cued (see Givón 1995: 364ff.). These may well in turn involve basic
operations such as ‘search and retrieve’, ‘establish new node’, ‘connect
node’, etc., of the kind Givón proposes – or indeed Nuyts’s (1992: 272ff.)
sentencing (i.e. the breaking up of information into clause-sized units) and
the like – but this is surely at a level of assumed cognitive activity below
what is accessible to conscious experience. Moreover, they are not relevant
to minimalizing the FG model as such. Nuyts’s cognitive component actu-
ally consists of declarative rather than procedural knowledge (it is a
situational network consisting of representations, i.e. – like the FG compo-
nent– represents pattern), the claim being that such knowledge is relevant to
both logic (process) and to conceptual knowledge (pattern). The processes
meant to implement it are only vaguely adumbrated, however. The kind of
basic processes I am concerned with are more like Givón’s referent pointing
(a matter of attracting attention), easily understandable as complex concres-
cences and describable in prehensional terms (in this case involving an
indicative prehension).
22. In general, interpreting the vertical axis of Hengeveld’s revised model as
universal process does not seem warranted. As regards Nootka, for exam-
ple, a left-to-right elaboration of an interpersonal level structure would re-

quire very little reference to expression-level categories (parts of speech and
syntactic phrase units) at all. All the relevant triggers for expression are ar-
guably there at the interpersonal level (interacting with the lexicon). The
only involvement of the representational level in production would in turn
be in decisions to construe the propositional content in terms of first-order
or higher-order terms and/or bare predicates – in other words as nominal or
verbal expressions (this, as we have seen, being determinate and overt if in-
flectional suffixes are present). Languages can no doubt be typologized as
regards the degree of relevance of the two lower levels for produc-
tion/comprehension decisions. These would, at all events, appear to be more
language-specific than those at the interpersonal level.
References
Chafe, Wallace
1994 Discourse, Consciousness, and Time. Chicago: University of Chi-
cago Press.
Chomsky, Noam
1993 A minimalist program for linguistic theory. In: Ken Hale and Samuel
Jay Keyser (eds), The View from Building 20, 1–52. Cambridge MA
and London: MIT Press.
Dik, Simon C.
1988 Concerning the logical component of a natural language generator.
In: Michael Zock and Gerard Sabah (eds), Advances in Natural Lan-
guage Generation, vol.1, 73–91. London: Pinter.
1989 The Theory of Functional Grammar, Part 1: The Structure of the
1997 The Theory of Functional Grammar, Part 2: Complex and Derived
Constructions. Berlin and New York: Mouton de Gruyter.
Engberg-Pedersen, Elisabeth
1991 Lærebog i tegnsprogs grammatik. København: Døves Center for To-
tal Communikation.
Fortescue, Michael
1992 Aspect and superaspect in Koyukon: An application of the Func-
tional Grammar model to a polysynthetic language. In: Michael
Fortescue, Peter Harder, and Lars Kristoffersen (eds), 99–141.
1998 Language Relations Across the Bering Strait. London and New
York: Cassell.
2001 Pattern and Process: A Whiteheadian Perspective on Linguistics.
Amsterdam and Philadelphia: Benjamins.
Fortescue, Michael, Peter Harder and Lars Kristoffersen (eds)
1992 Layered Structure and Reference in a Functional Perspective. Am-
Givón, Talmy
1995 Functionalism and Grammar. Amsterdam and Philadelphia: Benja-
mins.
Hannay, Mike
1991 Pragmatic function assignment and word order variation in a func-
tional grammar of English. Journal of Pragmatics 16: 131–155.
delphia: Benjamins.
Harder, Peter
1992 Semantic content and structure in Functional Grammar. On the se-
mantics of “Nounhood”. In: Michael Fortescue, Peter Harder and
Lars Kristoffersen (eds), 303–327.
Hengeveld, Kees
1992 Parts of speech. In: Michael Fortescue, Peter Harder, and Lars
1997 Cohesion in Functional Grammar. In: John Connolly, Roel Vismans,
Christopher Butler, and Richard Gatward (eds), Discourse and
Pragmatics in Functional Grammar, 1–16. Berlin and New York:
Mouton de Gruyter.
Itkonen, Esa
1983 Causality in Linguistic Theory. London and Canberra: Croom Helm.
Jacobsen, William
1993 Subordination and cosubordination in Nootka: Clause-combining in
a polysynthetic verb-initial language. In: Robert Van Valin, Jr. (ed.),
Advances in Role and Reference Grammar, 235–274. Amsterdam
Lambrecht, Knud
1994 Information Structure and Sentence Form. Cambridge: Cambridge
University Press.
1996 Functional Grammar and the analysis of English. In: Jürgen Klein
and Dirk Vanderbeke (eds), Anglistentag 1995 Greifswald, 135–148.
Tübingen: Niemeyer.
Mithun, Marianne
1987 Is basic word order universal? In: Russell Tomlin (ed.), Coherence
and Grounding in Discourse, 281–328. Amsterdam and Philadel-
phia: Benjamins.
1999 The Native Languages of North America. Cambridge: Cambridge
University Press.
Nakayama, Toshihide
1994 Phrasal suffixation in Nootka. In: Osahito Miyaoka (ed.), Languages
of the North Pacific Rim, 263–272. Sapporo: Hokkaido University
Publications in Linguistics No. 7.
1997 Discourse-pragmatic dynamism in Nuu-chah-nulth (Nootka) mor-
phosyntax. Ph.D. dissertation, University of California at Santa
Barbara.
Nuyts, Jan
Quine, Willard V.O.
1960 Word and Object. Cambridge MA.: MIT Press.
Shibatani, Masayoshi
1990 The Languages of Japan. Cambridge: Cambridge University Press.
Strawson, Peter F.
1974 Subject and Predicate in Logic and Grammar. London: Methuen and
Co.
Swadesh, Morris
1936 Nootka internal syntax. International Journal of American Linguis-
tics 9 (2–4), 77–102.
Functional Discourse Grammar and language
production
1. Introduction
One of the central requirements placed on a Functional Grammar is that
“… such a grammar must also aim at psychological adequacy, in the sense
that it must relate as closely as possible to psychological models of linguis-
tic competence and linguistic behaviour’ (Dik 1997a: 13). The FG model is
presented in a quasi-productive mode (Dik 1997a: 57), but at the same time
sees itself as essentially generative – indeed the primary purpose of the
computer implementation of FG (Dik 1992) was to test and enhance its
generativity. Recognizing that to confuse generation with production is a
category error, FG has insisted that it is not a model of language produc-
tion. Yet the demands of psychological adequacy have encouraged
speculation on the relation between the model, in its various manifesta-
tions, and the ever more sophisticated models of language production
currently available. The purpose of this chapter is to examine the potential
of Functional Discourse Grammar for a further rapprochement between FG
and a psycholinguistic consensus on language production.
Jackendoff (1997: 7–8) has argued that there are three possible positions
on the relation between grammar and the processes of speech production
and perception:
(a) one can deny any relationship, insulating grammar from psycholin-
guistic findings (the traditional generativist position);
(b) one can maintain that processing mechanisms can ‘consult’ or ‘in-
voke’ a declarative grammar;
180 J. Lachlan Mackenzie
(c) or one can claim that the processor embodies the grammar, i.e. that
grammar is itself procedural.
I believe that there are certain developments in current FG that suggest

that greater psychological adequacy is achievable through a combination of
Jackendoff’s positions 2 and 3. More specifically, I will explore the possi-
bility that Hengeveld’s (this volume) interactional and expression levels
can be modelled as procedural, with the intermediate (and in some cases
by-passed) representational level functioning as a declarative module that
is consulted by the processing mechanisms.
2. FG as a declarative grammar
A central tenet of the functionalist stance is that linguistic form results
from a complex of choices. For example, there is in certain languages a
choice whether or not to apply the Subject function to a first argument,
which is reflected in English in the active vs. passive voice (Dik 1997a:
248–250). The choice would appear to be the speaker’s. However, this
raises an apparent problem for FG, and other functionalist approaches that
stress ‘choice’: psycholinguistic findings strongly suggest that most of
these choices are simply not accessible to the language-user’s awareness.
As Levelt (1989: 21) puts it, “A speaker doesn’t have to ponder the issue of
whether to make the recipient of GIVE an indirect object (as in John gave
Mary the book) or an oblique object (as in John gave the book to Mary)”.
This is known as ‘cognitive impenetrability’: many aspects of language
processing are automatized, and not accessible to choice.
The conclusion must be that, when FG invokes psycholinguistic expla-
nations, it is trying to make understandable why the automatic processes of
language are as they are. This applies as much to the hearer’s processing of
utterances as to the speaker’s production of them. To take an example, the
explanation offered for LIPOC is that “[i]t is easier [sc. for the hearer,
JLM] to perceive, process, and store complex information when this infor-
mation is presented in chunks of increasing internal complexity” (Dik
1978: 212). We must beware, however, of assuming that the speaker is ac-
tually choosing to alleviate the hearer’s interpretive task: language users do
not consciously apply linguistic principles in their effort to express them-
selves. Surely the explanation offered by Dik must be understood as
historical and selectionist: of all possible orderings of information, the one
that has sedimented into automatic processing is that which, everything
FDG and language production 181
else being equal, conforms with LIPOC. Note that there is nothing anti-
functional about this understanding of ‘choice’. On the contrary, automatic
processing can itself be seen as functional, since it certainly contributes to
efficiency, freeing the speaker’s and hearer’s minds to concentrate on the
subject-matter of the discourse.
Applying these insights to the construction of a model of FG, we will

need to regard the various generalizations made within FG as constraints
upon the cognitively impenetrable processes of language production. What
would then distinguish a functional grammar from a non-functional one
would be that the constraints are not arbitrary. Rather, they are understand-
able a posteriori in terms of efficient interpersonal communication. The
position that suggests itself, then, is one that is compatible with Jackend-
off’s second option: an FG is declarative and takes the form of a set of
constraints on processing. Across billions of communicative acts, these
constraints have proved their worth. It is the functionalist’s task to identify
them and to provide explanations for their existence.
3. FG as a model of language production

As mentioned in Section 1, some recent work has been moving FG in the
direction of Jackendoff’s third position. The main exception to the domi-
nant interpretation of FG as simply a declarative grammar is to be found in
the work of Jan Nuyts, who has forcefully presented his Functional Proce-
dural Grammar (Nuyts 1992). He has argued that a number of matters
traditionally regarded in FG as being part of grammar proper should be as-
signed to non-grammatical cognition. Questions such as the assignment of
pragmatic functions should in his view be handled in conceptual and/or
textual components, and in any case fall outside the grammar. Another im-
portant consequence of Nuyts’s work concerns matters such as the
distinction between operators and satellites, or the distribution of informa-
tion over main and subordinate clauses. For both phenomena, their
treatment in the grammar is entirely dependent upon their ultimate formal
properties: operators correlate with grammatical, and satellites with lexical
expression; similarly, there is no simple or direct correlation between cog-
nitively nuclear information and main clauses, or between cognitively
subsidiary information and satellite clauses. Nuyts argues that these
distinctions are also more plausibly understood as resulting from cognitive
(pre-linguistic) operations. He takes Jackendoff’s third position in propos-
ing a model of clause production. He nevertheless refers to it as a grammar,

not least because the processes described culminate in a modified,
slimmed-down version of FG, including a rather orthodox expression rule
component. He has thus clarified the stages that take us from the first
communicative intention through to the representation of the predication.
Bakker and Siewierska (this volume) have now provided an existence
proof of the possibility of regarding the expression rules, too, as following
production. They assume a fully specified underlying representation of the
clause which is static and declarative in style. But the expression rules they
propose can and should be seen as dynamic, working top-down, depth-first,
and - crucially, as in actual language production, left-to-right.
The trend towards a re-interpretation of (at least part of) FG as a pro-
duction model is reflected, too, in Hengeveld’s (this volume) call for a
conversion of FG from a bottom-up grammar taking the predicate frame as
its starting point and adding layers in an upward fashion to a top-down
grammar, “starting with the communicative intention and ending with the
articulation of the linguistic expression”. This call cannot be divorced from
the increasing interest in reconciling Functional Grammar and discourse
analysis. This urge to include discourse considerations in the overall model
arises from a desire to achieve ‘pragmatic adequacy’ rather than ‘psycho-
logical adequacy’. Nevertheless, we are aware that psychologists, too, have
been moving away from the examination of individual utterances towards
considering ‘arenas of language use’, as Clark (1992) has it, and the entire
range of activities that occur in those arenas. There may be less distinction
between ‘psychological’ and ‘pragmatic’ adequacy than in the early days
of FG.
The relationship between discourse and grammar can, I believe, be seen
in three different ways, all of which are being tried out in current work.
Firstly, we may seek to model discourse and grammar in separate com-
partments of our theory; we will then try to link them through an interface
(this is essentially the position taken in work by Kroon 1997 and by Vet
1998). Secondly, we may examine the proposition that discourse is struc-
tured in analogy to clausal grammar – this is how I read Chapter 18 of Dik
(1997) and, in different ways, the earlier and recent work of Hengeveld
(1997, this volume) and Moutaouakil (this volume). Thirdly, and this is the
position that seems most consistent with a production-mimicking approach,
we can see discourse production as a dynamic process occurring in real
time and the expression of the clause as a similarly real-time process.
Clark (1996: 285) has pointed out that “Utterances are often viewed as
the prerogative of speakers – products that speakers formulate and produce
on their own. Nothing could be further from the truth”. Dik (1997a: 8)
similarly stresses that we are concerned with verbal interaction. At this
level conscious decisions are involved: not the choice whether to assign
Object function to a Recipient or not, but a split-second determination of
what move to make at a particular juncture in the joint activity of commu-
nication: to contest what the other has said, to crack a joke to defuse a
tricky situation, etc. The choice of move is a complex matter involving, as
discourse analysts have shown us, a number of goals at once, not least that
of keeping the conversation going in a mutually co-operative manner.
4. Acts and subacts

Let us, with Hengeveld (this volume), take the speaker’s strategic move as
the point of departure for analysis. This ‘minimal free unit of discourse’
consists of a number (n ≥ 1) of acts. Under normal circumstances, their se-
quence in cognition will be reflected by their sequence in expression. In
this view, the schema Theme, Predication, Tail familiar from Dik’s early
work (1978: 130) will be reinterpreted as reflecting a single move consist-
ing of three acts: the Theme corresponds to an act that introduces a
discourse referent, the Predication to a cognitively subsequent proposition
involving that referent in some way, and the Tail to a final act, correcting
or amplifying the preceding act, possibly as a result of self-monitoring.
Each Act corresponds in principle to a single intonation unit, although
many extraneous factors may upset the biuniqueness. In a language like
English the intonation unit generally contains one particularly accented and
pitch-changing (tonic) syllable, which helps to mark a possible transition
point for the other conversation partners. This accented syllable is in most
cases also a guide to the communicatively most salient aspect of the act. In
FG this corresponds to the assignment of the pragmatic function Focus.
The Focus is thus the essential component of each act; i.e. there is no act,
no matter how brief, without a Focus.
Let us consider an utterance such as (1):
(1) Oh my God, it’s on fire, my hair.
This utterance will, let us assume, reflect one move by the speaker. The
utterance will most likely manifest three intonation contours, each with a
tonic syllable (God, fire, hair respectively). The move can correspondingly
be seen as containing three sequenced acts which rather naturally reflect
the succession of cognitive events: shock; perception; identification. Using,

but adapting, Hengeveld’s proposals, we may represent the outline of this
move as (2):
(2) (M1: [(A1), (A2), (A3)])
where the numbering of the acts corresponds to their sequence in cognition.
Acts come in two basic types. Simplex acts (like A1) consist only of
their Focus. (Hengeveld this volume: 11 provides the example of the ex-
clamation Damn!.) Simplex acts do not involve the grammar in any way,
but are passed directly from the interaction component to the expression
component. Hengeveld assumes that such acts involve a consultation of the
lexicon. In contrast, I should wish to claim that the words uttered in a sim-
plex act do not come from the lexicon, but are furnished directly by the
expression component. The lexicon contains predicates, and is therefore an
essential support for the representational component. But where the latter is
not involved, the lexicon is also not in play. Simplex acts, which possibly
represent a perpetuation of more primitive forms of communication, i.e.
animal ‘calls’, instruct the expression rules to activate formulae appropriate
to the communicative function of the act. Speakers may of course differ
within a speech community in which formulae they prefer. Consider how
various acts of exclamation offer a range of ready-made expressions, each
with its own social implications:
(3) agreement Yes, Absolutely, You said it, Too right, …

disagreement No, No way, Absolutely not, …
regret Sorry, Pardon me, Excuse me, …
We also find acts of solidarity, covering various greetings and valedictions,

but also echoes and vocatives. Echoes (where the speaker, for one of a va-
riety of reasons, repeats some fragment of the interlocutor’s previous
speech) involve the intention to (re-)express an element of the communica-
tive context; the lexicon need not be consulted. And vocatives similarly
draw upon a very specific part of that context, the identity of the interlocu-
tor, with the expression rules applying, where that is relevant for the
language being spoken, the appropriate vocative marking. Another promi-
nent type of simplex act is the act of organization, which is oriented to
keeping the conversation going: the forms used for this purpose are gener-
ally highly formulaic, ranging from interjections like ah and oh to more
complex forms that mimic the complexity of lexical formulations (techni-

cally speaking, to be more specific, …).
Complex acts differ from simplex acts in consisting of one or more
subacts: in (2) above, A2 (expressed as it’s on fire) and A3 (expressed as my
hair) are complex acts. One of these subacts carries the pragmatic function
Focus that characterizes the act as a whole. The subacts in question are
subacts of reference (R) or of ascription (T): both reference and ascription
are thus seen as activities carried out by the speaker in interaction. If a
complex act contains only one subact, this subact then necessarily carries
the Focus function (as in A3). Where a complex act contains more than one
subact (as in A2), one of these carries the Focus function.
All the acts and subacts discussed here are initiated and organized at a
pre-linguistic, conceptual level. Whereas simplex acts are sent directly to
the expression component, complex acts, which I take to be definitional
of the more sophisticated communication systems of the human being,
have a more complex history. All complex acts involve the representa-
tional component, and may also involve the lexicon, although not
necessarily so. Thus both (4a) and (4b) result from complex acts, each in-
volving a subact of reference and a subact of ascription, but only (4a)
invokes lexical material:
(4) a. The president smokes.

b. He does.
Hengeveld (this volume) points out that the mapping relation between
the interactional and representational components is not one-to-one. Al-
though there are default correlations between an ascriptive subact and a
predicate (T ↔ f), and between a referential subact and a term (R ↔ p, e,
x), other possibilities are available: ‘predicate nominals’ thus reflect the
mapping of an ascriptive subact to a term (T ↔ x).
A striking characteristic of Hengeveld’s proposal is that the valency
structure rightly posited for the representational component, i.e. (f1) (x1), is
also present in the interactional component: within the communicated con-
tent C, we find the representation … (T1) (R1) …, which is explained as
involving ‘the speaker who ascribes properties to entities’. If the interac-
tional component is to be viewed as reflecting production processes carried
out in real time, however, an alternative view suggests itself. Let us assume
that the Focus of each act corresponds to the temporally first cognitive ele-
ment activated in the preparation of an utterance. After all, it is the very
communication of that Focus that justifies saying anything at all. Indeed, in
appropriate circumstances, determined by such matters as urgency, or fa-

miliarity between the interlocutors, the Focus is all that need be
communicated. In the holophrastic speech of infants, too, it is focal infor-
mation that is expressed; it is only as they develop a ‘theory of mind’, i.e. a
social competence, that they come also to express non-focal (typically topi-
cal) information (cf. Levelt 1999: 85). In this view, then, the cognitive
assembly of an act commences with the identification of a Focus: this is the
first (and possibly only) subact.
Associated with any complex act is a fundamental stance: either a giv-
ing stance, where the speaker is offering information or goods/services to
the addressee, or a demanding stance, in which the speaker is eliciting in-
formation or goods/services from the addressee (cf. Halliday 1994: 68–69).
The stance will be determined by the speaker’s initial motivation for lin-
guistic communication, and it is this stance towards the Focus that
determines the basic ‘illocutionary’ status of the act as a whole. Thus a giv-
ing stance correlates with an assertion, a demanding-information stance
with a question and a demanding-goods/services stance with an order or
request. Rather than associating the illocution with the entire act, as
Hengeveld does, I would therefore prefer to indicate the speaker’s stance
on the subact in Focus.
This subact may be supported by further subacts, in line with the
speaker’s awareness of her participation in the joint project of communica-
tion with the addressee. Here she draws upon the sources of knowledge
available in the communicative context (Dik’s ‘pragmatic information’,
1997a: 10) to establish a framework in which her Focus will affect the ad-
dressee in the desired way. The more purely cognitive relations between
the supportive subacts and the Focus subact are familiar from existing
work in FG: the supportive subacts may be Topics of various kinds (Given,
Resumed, etc.) or possibly Settings; other subacts may be inspired by the
desire to reflect the (power) relationship between the interlocutors, such as
those that result in indirect formulations, attitudinal adverbials, etc.
The representation that suggests itself will thus differ in various ways
from that proposed by Hengeveld (this volume) for the interactional com-
ponent. The representation will be incremental, i.e. the left-to-right
ordering of acts and subacts will be designed to reflect actual sequence in
cognition. An outline suggestion for the representation of (1) above would
appear as follows:
(5) (M1: (express A1: SHOCK)Foc, (assert A2: (SA2:1: (T: FIRE)Foc), (SA2:2: (R:
<unspecified, nonhuman>)Top), (assert A3: SA3:1: (R: HAIR)Foc)
The various elements in capital letters stand for activated emotions and
concepts; the operators in bold print indicate the type of act. As soon as the
process represented in (5) begins, the expression component is set into ac-
tion. This is in line with “a basic consensus in the language production
literature” (Levelt 1999: 88), namely the notion of incremental production.
This says “that the next processing component in the general flow of in-
formation can start working on the still incomplete output of the current
processor. A processing component will be triggered into action by any
fragment [Levelt’s emphasis] of its characteristic input. As a consequence
the various processing components are normally simultaneously active,
overlapping their processes as the tiles of a roof”.
Whereas simplex acts can be sent directly to the expression component
(as outlined above), complex acts call upon the declarative grammar, which
may be modelled much as is proposed in Hengeveld’s representational
component. There, the units corresponding to the various subacts, Focus
and non-Focus, will generally, as with A2 in (5), be grouped around a
predicate into the familiar propositional form. The predicate is selected
from the Fund, i.e. the lexicon expanded by the various derivational proc-
esses permitted by the language. The selection of the predicate, i.e. of the
State of Affairs, may well reflect the speaker’s perspective upon her con-
ceptualization, as with the selection of buy versus sell for a commercial
transaction; the perspective may also manifest itself in syntactic function
assignment. As Levelt (1999: 91–92) points out, the perspective may be
influenced by pragmatic factors attributable to the ongoing discourse, but
the speaker does have some freedom in this regard.
5. The representational and expression components

The entire layering system of the FG representational component now
comes into play, setting down the ground rules for the linguistic formula-
tion of the cognitive processes that are to be expressed. They form a
complex of constraints upon the translation of “intention into articulation”
(Hengeveld this volume), constraints that are partially universal and par-
tially language-specific. The universal constraints cover such matters as
argument structure, the ordering of semantic functions, the relative scope
of the layers; the language-specific constraints relate to the distribution of
information over operators and satellites, the presence or absence of tense,
of Subject assignment, of evidentiality, etc. Further constraints derive from
the Fund: certain predicates strongly impose a certain argument structure,
such as English put, which in its literal sense calls for specification of
Agent, Goal and Location. In Levelt’s (1999: 95) formulation, “[i]n a way
grammatical encoding is like solving a set of simultaneous equations. Each
lemma [in FG, ‘predicate’ JLM] requires particular syntactic constraints
from its environment and the emerging syntactic structure should simulta-
neously satisfy all these constraints”.
If the representational component is seen as a set of constraints on proc-
essing, it is no longer necessary to assume that every utterance has a full
propositional form. The grammar need simply be consulted for what is nec-
essary to the task at hand, or as Hengeveld (this volume) has it, “it is no
longer necessary to assume that every discourse act contains a proposi-
tional content”. Thus an act such as A3 in (x) above, containing only one
referential subact, undergoes only those constraints that relate to term
structure, including, in those languages that require this, a semantic or syn-
tactic function that determines its case form. In this way, it is possible to
treat so-called fragmentary utterances without recourse to rules of deletion.
The complete clause then emerges as a system for supporting the Focus,
providing a context within which the addressee can understand it as well as
possibly satisfying other communicative goals of the speaker’s.
The interpretation of Hengeveld’s model that I am proposing is a com-
bination of Jackendoff’s second and third options: the overall grammar is
procedural (position 3), but in the construction of complex acts, it needs to
consult, indeed conform to, a declarative grammar (position 2). The advan-
tage of a procedural approach is that FG will accord better with the
“psychological models of linguistic competence and linguistic behaviour”
which Dik (1997a: 13) wishes the practitioners of FG to bear in mind. In-
cremental processing in the production of utterances works on the
assumption that the concepts that lie behind a complex act are not all avail-
able at once. Speakers characteristically do not wait until all the concepts
are “in” before starting to speak. The various levels of activity, conceptual-
izing, formulating and articulating, as Levelt (1989) has it, work in parallel,
with conceptualization having a head start over formulating, and formulat-
ing over articulating. However, speakers are not free to express their ideas
exactly as they occur to them: the declarative grammar, with its conven-
tional principles, will often mean that particular parts of a message are put
into a buffer before they can be expressed.
Clausal structure appears to have exactly this buffering function. At least,
this seems to be plausible for English, a language with rather fixed word or-
der, by which I mean that the syntactic template plays an important role in
expressing the relations imposed by the declarative grammar. But what of a
language with a so-called ‘free word order’? By this we understand that the
order of constituents is much more strongly influenced by their cognitive
status as New, Given, Inferred, etc. For such languages the temporal order in
which the concepts become available may be much more directly reflected in
word order, with the declarative grammar serving to ensure above all that the
semantic relations between the constituents remain transparent to the ad-
dressee. In such languages we might expect the Focus to be expressed first,
in line with its cognitive priority; an example of this situation is discussed in
depth by Fortescue (this volume). Yet even in language with relatively ‘free’
word order, there is often one syntactic position, frequently not P1, which is
reserved for the Focused constituent. In Turkish (Van Schaaik 2001: 45) this
is the immediately preverbal position; in Hungarian (De Groot 1989: 105) it
is the position between Topic and Verb.
The non-initial positioning of Focus in actual utterances, as against its
cognitive priority, may also be understandable from a processing viewpoint.
The cognitive identification of a Focus concept, which is typically new in-
formation, and the selection of an appropriate lexicalization of that concept,
both require time. By contrast, associated information that is already acti-
vated and thus highly accessible (i.e. Topics and Settings) is immediately
available for expression. The speaker who expresses accessible information
first allows herself more time for the identification and lexicalization of the
Focus. Hannay (1991) has shown that even for a syntactically rather rigid
language like English, the relative positioning of Topic and Focus is depend-
ent upon what he calls the ‘message mode’: in an urgent situation, the
Reaction mode encourages initial placement of the Focus, with optional
back-up from a following Topic; where there is a highly accessible Topic,
the Topic mode induces a postponement of the Focus to a later syntactic po-
sition. It may indeed be advisable to add a specification of the message mode
to the representation of utterances in the interactional component.
One consequence to be drawn from the preceding discussion is that the
expression component is no longer, as it is in orthodox FG, the only place
where the order of constituents is determined. Its function now emerges as
that of balancing the competing demands of the interactional component,
which ‘wishes’ to see each component of its emerging message expressed
as soon as possible, of the representational component (as in traditional
FG), which requires an unequivocal reflection of its demands, and of the
expression component itself, which has its own language-specific rules and
regulations. These are the ‘simultaneous equations’ to which Levelt (1999:
95) refers.
6. Some examples
Let us conclude by sketching out a couple of examples of how things might
work.
If I feel hunger, I can conceive of grapes as something I want. Under
certain circumstances, for example if I am surrounded by willing slaves, it
will be sufficient for me to utter a complex act with a single and inevitably
Focused subact of reference:
(6) Grapes!
to be analysed at the interactional level as (7)
(7) (M1: (request A1: (SA1:1: (R: GRAPES)Foc)))
Here only that part of the lexicogrammar is involved that deals with
subacts of reference. The lexicon offers the countable noun GRAPE and the
declarative grammar (of English) requires that the referential subact be
specified for at least definiteness and number; the expression rules react
appropriately to the operators indefinite and plural. The operator request
triggers the appropriate intonation contour. The rest lies with the interpre-
tive abilities of my slaves.
Under less utopian circumstances, I will have to devise a strategy to
achieve my purpose. The focal reference to ‘grapes’ now appears in the
context of this overall strategy. Let us assume that the strategy I have se-
lected is one of moderate politeness, and is associated with the formula
Can I have …. This kind of unit is reminiscent of the expression of simplex
acts in being a ready-made formula, probably learned and stored as such
(cf. Nattinger and Decarrico 1992). The following representation suggests
itself, where the Move takes the operator pol for politeness:
(8) (pol M1: (request A1: (SA1:1: (R: GRAPES)Foc)))
This is passed directly to the expression rules, which fit the two elements
Can I have and grapes, the latter (only) having passed through the repre-
sentational component, into an acceptable template:
(9) Can I have grapes?
This example is a good instance of how any one utterance can contain a
mixture of elements that come directly from the interactional component

and elements that have passed through the representational component. In
this way we are able to account for the fact that (9) functions in the same
way as (7), but differs only in degree of politeness. At the same time, (9)
remains parsable: if so desired, a language user can refer it back to the de-
clarative grammar, as appears to happen when a speaker, in an effort to be
hyperpolite, creates a baroque variant of the socially recognized formula:
(10) If you could possibly see your way clear to providing me with some grapes.
An example in which the representational component is more thor-

oughly implicated is (11), where capitalization indicates the tonic syllable:
(11) Did you EAT the grapes?
Here, let us assume, the speaker is aware of the addressee and the grapes:
both are given information, and the Focus is on eat. The initial structure in
the interactional component could therefore be:
(12) (M1: (inquire A1: (SA1:1: (T: EAT)Foc)))
which would yield the utterance (13), a highly unlikely utterance in the cir-
cumstances at issue here:
(13) EAT?
The representational component, however, will not be ‘satisfied’ with (13),

since the lexical entry for eat contains an Agent and a Patient. The lexicon
also contains the information that eat without a Patient has a different, here
irrelevant sense ‘have one’s meal’. The representational component there-
fore demands specification of the Patient, yielding (14), which would be a
possible utterance in the given situation:
(14) EAT the grapes?
In other words, in creating the representation in the interactional compo-

nent, the speaker heeds the declarative rules of the language, giving:
(15) (M1: (inquire A1: (SA1.1: (T: EAT)Foc), (SA1.2: (R: GRAPES))))
As was mentioned, the lexicon also recognizes the Agent of the eating as
an argument, which forms another, slightly less binding, constraint on the
speaker’s communicative intention as represented in the interactional
component:
(16) You EAT the grapes?

(17) (M1: (inquire A1: (SA1.1: (T: EAT)Foc), (SA1.2: (R: GRAPES)), (SA1.3
(R: ADDRESSEE))))
Finally, the representational component of English has rules regarding the

finiteness of propositions and the tense of predications which can also be
imposed upon the translation of the intention into expression. It is these
that are involved in the production of (11). More precisely, both the struc-
ture (17) and a representational structure (18) in which the relevant
operators are specified will form the input to the expression rules:
(18) (finite p1: (past e1: (f1: eat [V] (d1x1: [-S, +A])AgSubj (d1x2: f2: grape [N])Go)))
The final structure of (11) results from the expression rules’ reacting to a
collaboration between the on-line interactional component and the declara-
tive representational component:
(a) the speaker decides to produce an act of inquiry;

(b) the speaker focuses this act on EAT, as shown in (12) above;
(c) this activates the lexical entry eat and implicates the representational
component, which (as ever) imposes finiteness, tense, valency and
syntactic function assignment;
(d) the speaker now starts to formulate the utterance, gradually enriching
the interactional representation to take account of the requirements of
the representational component:
i. did in utterance initial position helps to signal the act of in-
quiry (along with the intonation contour) and satisfies the
representational component’s demand for finiteness and tense;
ii. you satisfies the representational component’s demand for
valency and subject assignment;
iii. EAT, with tonic syllable, expresses the Focus of the utterance;
iv. the grapes satisfies the representational component’s demand
for valency.
All three possible formulations impart the Focus (diii); (14) arises from
also heeding (div), (16) from also heeding (dii) and (div), and (11) from
heeding all the demands of the representational component.
7. Conclusion
I have taken as my point of departure Hengeveld’s (this volume) proposal
for the architecture of a Functional Discourse Grammar. With a view to
enhancing the psychological adequacy of FG and realizing the ambition of
bringing FG closer to a production model, I have suggested that the
interaction and expression levels can be reformulated as operating
incrementally in real time. At the interactional level, moves consist of se-
quences of acts, which in turn consist of sequences of subacts; at the
expression level, too, there is a sequence of constituents. Speakers typically
activate the expression level before completing the underlying act at the
interactional level. The representational level is regarded as a declarative
component that can be called upon to constrain these two real-time proc-
esses where lexical items from the fund are involved. A major contribution
of this level is to provide supportive material against which the Focus of
the Speaker’s message can be understood, encouraging and often constrain-
ing the speaker to provide more information than that Focus alone. It is
hoped that this approach can offer something to the understanding of the
ontogenesis of multi-word speech in infants and to the study of the im-
plicitness of much conversational speech.
References
This vol. Towards a speaker model of Functional Grammar.
Clark, Herbert H.
1992 Arenas of Language Use. Chicago: University of Chicago Press.
1996 Using Language. Cambridge, New York and Melbourne: Cambridge
University Press
ward (eds)
Dik, Simon C.
1978 Functional Grammar. Amsterdam: North-Holland.
1992 Functional Grammar in Prolog: An Integrated Implementation for Eng-

lish, French, and Dutch. Berlin and New York: Mouton de Gruyter.
1997a The Theory of Functional Grammar. Part 1: The Structure of the
1997b The Theory of Functional Grammar. Part 2: Complex and Derived
Constructions. Berlin and New York: Mouton de Gruyter.
Fortescue, Michael
This vol. The complementarity of the process and pattern interpretations of
Functional Grammar.
Groot, Casper de
1989 Predicate Structure in a Functional Grammar of Hungarian. Dor-
drecht: Foris.
Halliday, Michael A.K.
1994 An Introduction to Functional Grammar. 2nd edition. London: Ar-
nold.
Hannay, Mike
1991 Pragmatic function assignment and word order variation in a Func-
Hengeveld, Kees
1997 Cohesion in Functional Grammar. In: John Connolly, Roel M. Vis-
mans, Christopher S. Butler and Richard A. Gatward (eds), 1–16.
Jackendoff, Ray S.
1997 The Architecture of the Language Faculty. Cambridge MA and Lon-
don: MIT Press.
Kroon, Caroline
John Connolly, Roel M. Vismans, Christopher S. Butler and Richard
A. Gatward (eds), 17–32.
Levelt, Willem J.M.
1989 Speaking: From Intention to Articulation. Cambridge MA and Lon-
don: MIT Press.
1999 Producing spoken language: A blueprint of the speaker. In: Colin M.
Brown and Peter Hagoort (eds). The Neurocognition of Language,
83–122. Oxford: Oxford University Press.
Moutaouakil, Ahmed
This vol. Discourse structure, the generalized parallelism hypothesis and the
architecture of Functional Grammar.
Nattinger, James R. and Jeanette S. Decarrico
1992 Lexical Phrases and Language Teaching. Oxford: Oxford University
Press.
Schaaik, Gerjan van

2001 The Bosphorus Papers: Studies in Turkish Grammar 1996–1999. Is-
tanbul: Boğaziçi University Press.
Comment clauses, Functional Discourse
Grammar and the grammar-discourse interface
Peter Harder
1. Introduction
Hengeveld (this volume) introduces a new format for the incorporation of
discourse into the linguistic model of Functional Grammar, in which the
‘upward layering’ and the ‘modular’ approaches are integrated (FDG). In
this chapter I would like to take up the basic idea of a three-level integrated
model and use one of Hengeveld’s examples to discuss both the advantages
of his proposal and the reasons why a moderate change of the model in the
‘modular’ direction might improve the model’s capacity to bring out rela-
tions between discourse and grammar.
Hengeveld’s most radical break with previous FG models is perhaps in
the top-down derivational process, and in the motivation for that choice in
the form of an explicit adoption of a theory of speech production matching
the linguistic model point by point.
At the interpersonal level, the initial choice is an explicitly non-
grammatical entity, the move, defined in strictly interactional terms as cor-
responding to a single communicative intention; after that stage, however,
all choices reflect coded choices. After the top interpersonal level, there-
fore, all the purely code-free communicative choices have been made.
The representational level is conceived as consisting of entities of dif-
ferent orders which are described by the clause. Propositions (third-order
entities) are the top layer; speech acts are now above the scope of the rep-
resentational component. The main element of coding choice here is,
therefore, how to represent the referential and ascriptive choices made in
the interpersonal component. The expression level is described as consist-
ing of constituent structure, from paragraphs to words; I shall assume,
198 Peter Harder
however, that linear templates are part of the model, but have merely been
left out for purposes of simplification.
2. Interface rather than underlyingness

A valuable feature of FDG is that it represents a movement in the direction
of looking at complex constructions in an ‘interface’ perspective rather in
terms of a ‘depth’ perspective where actual linguistic expressions end up as
the ‘surface’. In FDG, we have three different analyses of the same linguis-
tic chain, representing three different sets of facts about it. Not all levels
are involved in every case; but in standard cases they are. Moreover, under
specifiable conditions there may be correspondences or discrepancies that
create different types of relationships between the three levels. The linguist
using the model to describe an utterance will have his task defined in terms
which make good sense from both a formal and a functional point of view:
he has to look at interactive organization, representational organization,
and organization of the expression side, and see how they interlock.
An example of how the interface perspective is brought out in FDG is
the element of ‘downward layering’, in which the interpersonal level is ex-
panded to cover communicated content, reference and ascription instead of
leaving that to the representational component. The functions of reference
and ascription are standardly but not inevitably carried out by means of
noun phrases and verbs; and the separation between the interpersonal acts
and their linguistic representation makes possible a much more adequate
account of the complex relationship between sentence organization and en-
tity categories than previous models.
Combined with the top-down perspective, the simultaneous operation of
all three levels makes it possible to see a descriptive procedure as a set of
choices involving all levels, with an anchor outside the grammar itself in
the form of the speaker's communicative intention. From a discourse point
of view, this is clearly an advance in adequacy over a model that begins the
execution of a communicative intention by choosing a basic predicate.
Comment clauses and FDG 199
3. The grammar-coding perspective vs the intention-coding

perspective
Although the layout is well motivated in terms of adequacy from the dis-
course viewpoint, it raises a fundamental question of the natural
ontological relations between grammatical choices and communicative
choices. Should we expect the speaker’s ‘decision tree’ of communicative
choices in making an utterance to be identical to the decision tree of mak-
ing grammatical choices?
It is noteworthy that over the past decades, models of grammar, whether
generative or more functional, have become increasingly lexically based in
their descriptive procedure, with the verb as the central source of those de-
pendencies that determine grammatical options. This trend suggests that if
we want to describe how grammatical choices are structured, there are fac-
tors that favour beginning with lexical elements – rather than placing them
at the end of a long hierarchical procedure. On the other hand, it is natural
to assume that there is a form of isomorphism in the sense that unless the
speaker can adapt his choice process to the choices offered by the code, it
is difficult to see how he could produce adequate utterances. Slobin and
Berman’s ‘thinking-for-speaking’ research (cf. Slobin 1987; Berman and
Slobin 1994), although focusing on other aspects of the issue, also suggests
that we need to provide descriptions where the two sides match up.
Until we know more, however, a number of different solutions are
possible: apart from point-by-point matching of linguistic and communi-
cative choices, one might imagine a first phase of communicative choices
gradually blending into a later phase of coding choices, or a process in
which the interface between communicative aims and coding options is
present from the beginning until the end. For reasons described else-
where, I think the linguistic hierarchy has an inherent order which
reflects a stage of ‘compilation’ rather than ‘execution’, or ‘recipe’ rather
than ‘cooking’ (cf. Harder 1996: 214ff.), where you have to relate opera-
tors to operands before you can execute the program, in which case there
would always be a simultaneous bottom-up and top-down logic; cf. also
Fortescue (1992: 122) on the bottom-up ‘centrifugal principle’ that seems
to apply to expression rules.
Whatever the exact truth may be, I believe that an explicit grammar-
discourse interface is an advantage in a framework that seeks the integra-
tion of discourse into a functional linguistic theory; a point-by point
matching process would then not be excluded, but it would emerge as a
special case. I think such a model would be in keeping with the central part
200 Peter Harder
of the aims stated in Hengeveld (this volume), even if it differs in the way
it conceives of the three strata.
Taking my point of departure in an example used by Hengeveld, I am
going to illustrate the issue with an example from English grammar, that of
so-called comment clauses.
4. The hedged performative and the report of emotional state in

FDG
To demonstrate the kind of complex phenomena that are brought into the
purview of the model by FDG, Hengeveld (this volume: 15) provides an
analysis of the ambiguity between the Spanish sentences (1) and (2):
(1) (Me) temo que Juan esté enfermo

to.me I.am.afraid that Juan is.SUB ill
(2) Me temo que Juan está enfermo
to.me I.am afraid that Juan is.IND ill
Both translate into ‘I am afraid that Juan is ill’, but in the indicative version
the subclause contains the main message, which is reflected in its status on the
representational and the interpersonal levels: the content of the subclause con-
stitutes the content of a referential act made on the interpersonal level:
(A1:[DECL(P1)Sp(P2)Addr(C1: [... (R1) ...] (C1))] (A1))

(c2)
(CL1)
This illustrates the advantages of the three simultaneous levels in FDG: the
interactive choice made by the speaker simultaneously affects the morpho-
syntactic expression and the linguistic representation of the coded content,
by inserting a different content (c2) as constituting the main representational
element.
In this respect the model represents an extension of the advantages of
earlier FG versions over a traditional grammatical description. From a tra-
ditional grammar point of view, the difference is simply a difference in
terms of mood, located in a paradigmatic slot in the Spanish verb; and tra-
ditional discussions about indicative and subjunctive mood were conducted
in terms of notional differences such as that between ‘real’ and ‘unreal’,
which were very imprecise and often difficult to apply to concrete cases
without a great deal of ‘semantics’ in the wrong sense of the word.
As demonstrated in depth in a number of publications by Hengeveld in

relation to modality, the layered model is useful precisely as a way to do
two things at once: to relate a grammatical choice like the indicative-
subjunctive contrast systematically to those aspects of sentence meaning
that it affects, and to show what the difference is in different types of cases.
This particular case requires access to more than pre-discourse layering.
Unless there were three simultaneous levels, with relations explicitly indi-
cated, and with the possibility of interpersonal factors moving ‘downward’,
this could not be captured.
But the question is how best to configure and distribute the discourse in-
formation and the grammatical information over the model. Consider now
the English translation of the two Spanish clauses, which can receive either
of the two readings:
(3) I am afraid that John is ill.
To describe this sentence adequately and to capture its ambiguity, we

need to be able to assign either of the FDG descriptions to it, rather than
one for each mood, as in Spanish.
Because of the orientation of FDG towards matching the speaker’s
choices, with the point of departure in his communicative intention, it is
logical to expect that the differences in communicative intention will be
clearly expressed at the interpersonal level; and in terms of the basic fea-
tures of Hengeveld’s model as laid out in the introduction, this is also the
aim. Quoting selectively from Hengeveld’s description (this volume: 5) of
what goes on at the interpersonal level, we see that
[a]t the interpersonal level a central unit of analysis is the move (M), ... the
vehicle for the expression of a single communicative intention of the
speaker.
In order to achieve his communicative intention, the speaker executes
one or more discourse acts (A), defined in Kroon (1995: 65) as ‘the small-
est identifiable units of communicative behaviour’. A move consists of one
central act, which may be supported by one or more subsidiary acts...In or-
der to build up the communicated content the speaker may have to execute
one or more ascriptive acts (T) and one or more referential acts (R).
Hengeveld, moreover, leaves open the possibility of including larger

structures such as turns and exchanges where necessary. The model as de-
scribed in the passages quoted above seems to me well suited to bring out
the difference between the two readings. The question I raised above
202 Peter Harder
comes in when we look at what information is provided by the actual de-

scriptive format. In Hengeveld’s schematic version of the ‘interpersonal
level’, the information reflects the familiar grammar-oriented structuring
principle, covering participants in the discourse seen as ‘arguments’ of the
illocutionary frame. The ‘discourse act’ information quoted above is not
immediately visible; instead we see a format that is tailored to cover the
generic features of declarative sentence types.
Because of that, the interpersonal level suggested for the Spanish in-
dicative version, cf. (2) above, would seem to cover the subjunctive version
as well: an act coded by a declarative sentence with a speaker, an addressee
and a content which involves an act of reference performed by means of
the subclause (which serves as the second argument of the verb temo). The
difference, it appears, is only reflected in the substitution of ‘communi-
cated content’ on the representational level and its relation to the expres-
sion level.
This appears to me less than ideal as an illustration of the aim of inte-
grating discourse structures explicitly into the model. Ultimately, it may be
a matter of notational variants, but it seems to me that if the difference be-
tween the two readings reflects an interpersonal choice, i.e. the choice
between ‘saying what one feels’ and ‘conveying a mitigated piece of bad
news’, the interpersonal level should also be the central point at which the
difference is visible in the model.
‘Hedging’ would appear to be a good example of a subsidiary discourse
act attached to a central act with face-threatening potential (with the two
together constituting a move, according to the description given by Henge-
veld). As such, hedging should be represented in the decision tree for
‘realization of intentions’ (cf., e.g. Brown and Levinson 1987: 60, showing
the decision tree of someone who is considering a face-threatening act).
Therefore the interpersonal level should specify two different choices, per-
haps describable as follows (S = speaker, H = hearer, C = communicated
content, T = ascription, R = reference):
(a) Communicative intention: Relieve one’s mind by entering into ‘trou-

bles talk’, realized by Move: ‘report of emotional situation’, divided
into a main act (A1): ‘S tell H of C = emotional state’ (T = fear, R1 =
S) + subact (a1): ‘specify object of emotion’ (= R2)
(b) Communicative intention: Do one’s duty towards H, realized by
Move: conveying relevant ‘bad news’, divided into a main act (A2):
S tell H of C (T = ill, R = J) + subact (a2): mitigate by showing sym-
pathy ‘S tell H of C (T = fear, R1 = S)’
` I would like to proceed to a revision also in the representational level,

which can be seen as the reverse of Hengeveld’s ‘downward layering’ in-
novation of the interpersonal level: just as the interpersonal level in
Hengeveld’s model is extended downwards so that C, T and R overlap with
items on the representational layer, I think representation should cover all
linguistically represented meaning (including that which codes interper-
sonal choices).
One example of why this would make sense is the analysis of the sen-
tence I apologize as on the one hand a performative utterance or on the
other hand as the answer to a question such as What do you do when you
realize you’ve hurt someone? A linguistic model that aims to specify rela-
tions between grammar and discourse should provide a level where it is
specified that the sentence codes the same represented content in both
cases – as well as indicate that under appropriate contextual conditions, the
sentence acquires the status of a direct coding of an interpersonal act rather
than of informative content. An illustration might be:
The performative reading:
Interpersonal Level Act: apology:

S apologize towards H
(No C)
Layer of linguistically represented A-coding:
content Decl (pres (apologize (S))
The informative reading:
Interpersonal level Act: Informative:

S tell H C
C= T, R
Layer of linguistically represented C-coding:
content Decl (Pres (apologize (S))
Figure 1. Performative and informative readings of I apologize.
The details of this representation are very preliminary, of course. The

point is to bring out even more clearly than in Hengeveld's version what I
see as a major advantage of the three-tier system, namely the possibility of
matching structures at different levels of analysis in order to stress the
similarities and differences (a perspective that has recently been in the
204 Peter Harder
focus of interest of the Danish Functional Grammar group under the

heading ‘iconicity and structure’).
Let us look at the analysis of the ambiguous English sentence (3) above.
In terms of linguistically represented content, there is presumably no ambi-
guity in English: under both readings, it is a declarative clause with two
argument positions, the latter being filled by an embedded clause. In the
FDG analysis there is no level at which this identity of coded content is ap-
parent, which is why (cf. above) I would like to extend the representational
level to accommodate the linguistically represented content of I am afraid
in its interactive function.
If we do this, the interesting facts about the ambiguity can be brought
out by interfacing the two different interpersonal readings with the same
linguistically represented content. It is generally assumed (cf. Hannay and
Mackenzie 1996: 114) that the main information is standardly put in main
clauses, yielding an unmarked ‘iconic’ matching relation between the in-
terpersonal level and the level of linguistic representation. If we see the
information in the subclause as a subact in its own right (it may be some-
what implausible, but not, I think, out of the question; I propose it for
purposes of illustration rather than as a hypothesis), this yields the ‘inter-
face’ analysis of the main features of the ‘emotion report’ reading shown in
Figure 2 below.
In English, there is no linguistic trigger of the discrepant second read-
ing; it is simply a fact of usage. In Spanish, where the same structural
discrepancy is found, the indicative está provides a coding explanation, be-
cause it overrules the motivation for a subjunctive that emerges from the
matrix verb, declaring everything with its scope to be ‘real’ – which is dis-
crepant with the standard influence of a mental matrix verb.
A: ‘Express emotion’
Interpersonal Intention: Move: Main act: Subact:

level Relieve one’s Express S tell H of C Refer to object of
mind by emotional emotion
entering into state (C = state of (R2 = J’s illness)
’troubles talk’ being afraid
T = afraid
R1 = S)
Level of Decl (pres Arg2:
linguistically (fear,(S, arg2))) That (pres(ill, J))
represented
content
Level of Main Cl (…) Sub-Cl (…)

expression
B. ‘Hedged performative’:
Interpersonal Communicative Move: Main act: Subact:

level intention:
Explain to H Convey S Inform H that Mitigate by sym-
why J will not relevant C pathy
be available bad (C= R) S express C to-
news R = J’s illness wards H (C = own
T = ill feelings, T =
R´ = J afraid, R´´ = H)
Level of lin- Decl (pres
guistically (afraid (S, arg2)))
represented
content Arg2:
That (pres (ill,
J))
Expression Sub-Cl (…) Main Cl (…)
level
Figure 2. Analysis of I am afraid that John is ill.

206 Peter Harder
5. Comment clauses and the coding of hedges in English

Even if English does not have a mood distinction for this purpose, it pro-
vides its own way to code the ‘hedged performative’ reading unambig-
uously, i.e. by means of a so-called ‘comment clause’ construction (cf.
Quirk et al. 1985: 1112). This construction involves removing the matrix
clause from its fronted position, putting it at the end or in mid-sentence po-
sition:
(4) John is ill, I am afraid.

(5) John, I am afraid, is ill.
Both possible positions exemplify the interaction of discourse principles

and coding principles. FG provides slots for extraclausal elements both be-
fore and after the clausal positions, and the ‘tail’ position is generally
recognized as a home for what may, iconically, be called ‘afterthoughts’ in
relation to clause content. The clause-internal mid-sentence position is also
used by ‘sentence adverbials’ of various kinds, including modal and attitu-
dinal adverbials. In languages like German or Danish the mid-sentence
position is also the home of particles that code attitudinal status. Such par-
ticles are absent in English, and the similarity between the content of
certain particles in German and Danish and certain types of comment
clauses in English reveals something about the coding pattern of English.
When comment clauses occupy mid-sentence position they are standing at
the beginning of a potential grammaticalization path that has not been fol-
lowed in English: certain Danish particles have historically developed out
of comment-clause-like constructions (cf. Davidsen-Nielsen 1996: 298).
Thus the removal of the matrix clause ‘chunk’ to these positions is one way
of bringing about a more motivated relationship between its interpersonal
status and the way it is linguistically represented.
The representational level needs to contain the matrix clause in order to
capture this relationship. First of all, if we remove I am afraid from the
representational layer, we lose the possibility of explaining what qualifies
represented content of that kind to code a hedge, namely that it is an
indication of sympathy with the addressee’s feelings. Moreover, it follows
from the coding principles above that the content of I am afraid ... achieves
a slightly different status in each case, while its relation to the subact of
‘hedging’ on the interpersonal level is constant. In one case it aligns itself
(roughly speaking) with attitudinal elements in the clause, in another it
aligns itself with extraclausal ‘dislocated’ elements. In spite of the differ-
ence in structure, the English mechanisms share with the Spanish mood
distinction the element of undermining the power of the matrix clause over
the subclause – something which is important in understanding the rela-
tionship between grammar and discourse categories in such cases (cf.
Bolinger 1968).
The absence of that is also significant: that can be understood as an op-
erator which (like ‘declarative’) takes scope over the entire proposition (cf.
Harder 1995), reifying it into an entity (while ‘declarative’ uses it to con-
vey a piece of information). Therefore the absence of that when the matrix
clause is in the comment clause position is different from cases where that
is optionally absent. By being preposed without the complementizer, the
subclause is effectively promoted into a full declarative clause.
The realignment of matrix and subclause does not bring about a full
parallelism between discourse status and grammatical status. Distributional
tests would still class it with matrix clauses. This is not so obvious with I
am afraid, however, so for ease of illustration let us use the formal version
I fear instead. In sentence (6):
(6) John is ill, I fear.
the verb fear is prima facie without a second argument, and the only
way of bringing about a canonical dependency situation is by having a
grammatical interpretation under which John is ill remains embedded in
relation to the verb fear, while at the same time I fear is treated analogi-
cally with subclauses in terms of discourse status.
An investigation of this field, I suspect, will lead to the need for many
subdivisions of the representational layer of organization with many differ-
ent links to both the expression and the interpersonal levels. The work of
pursuing this angle will also require a reconsideration of the status of the
sentence from a discourse perspective. Hansen (2001: 128) takes up this
issue, quoting an example from German from Franck (1985: 234):
(7) Das war / also im Jahre 1907 / bin ich geboren.

that was / then in:the year 1907 / was I born
‘approx. So that was in 1907 I was born.’
Here we see a constituent with what Franck called ‘double-bind’ status,

serving both in a sentence begun in the previous instalment and a sentence
completed in the following instalment. Hansen’s conclusion is that the sen-
tence should perhaps be replaced as the relevant unit of analysis in favour
208 Peter Harder
of something smaller than sentences. As suggested in Section 1 above, the

problem might also be that from the point of view of online production,
there is no guarantee that the relevant units will match grammatical units
point by point.
6. Final remarks
As pointed out in Hengeveld (this volume), there are many problems that
need to be solved in order to have a fully operational Functional Discourse
Grammar. The suggestions I make raise the additional problem of how pre-
cisely to individuate the discourse categories that I would like to place in a
more spectacular position in the interpersonal tier; pending that, it will al-
ways be unclear how much empirical substance there is in the interfaces of
the model. However, problems of that kind need to be faced by anyone
who would like to relate the categories of the code with categories of inter-
action. In spite of the problems, I see the three-tier model as a clear step
forward in the quest to integrate discourse and grammar from a functional
point of view.
The main point I have defended is that if the interface potential of the
model is to be fully realized, it would be an advantage to have a level at
which linguistically represented content is fully specified, to be interfaced
with an interpersonal level focusing more explicitly on interrelations be-
tween discourse elements. In cases where discourse and clausal organiza-
tion go hand in hand, this will mean that information is systematically
duplicated at the two levels – but rather than being a disadvantage, this is
an intuitively obvious way of marking iconicity between discourse and
grammar when it occurs. And in cases where there are discrepancies be-
tween structuring of linguistic content and structuring of discourse, these
can be read straight off the model.
References:
Berman, Ruth A. and Slobin, Dan I.
1994 Relating Events in Narrative: A Cross-linguistic Developmental
Study. Hillsdale NJ: Erlbaum.
Bolinger, Dwight
1968 Postposed phrases: An English rule for Romance subjunctive. Cana-
dian Journal of Linguistics 14: 3-30.
Brown, Penelope and Stephen C. Levinson

1987 Politeness. Some Universals in Language Usage. (Studies in Interna-
tional Sociolinguistics 4.) Cambridge: Cambridge University Press.
Davidsen-Nielsen, Niels
1996 Discourse particles in Danish. In: Elisabeth Engberg-Pedersen,
Michael Fortescue, Peter Harder, Lars Heltoft and Lisbeth Falster Ja-
kobsen (eds), Content, Expression and Structure: Studies in Danish
Functional Grammar, 283–314. Amsterdam and Philadelphia:
Benjamins.
Fortescue, Michael
1992 Aspect and superaspect in Koyukon: An application of the Func-
tional Grammar model to a polysynthetic language. In: Michael
Fortescue, Peter Harder and Lars Kristoffersen (eds), Layered Struc-
ture and Reference in a Functional Perspective, 99–141. Amsterdam
Franck, Dorothea
1985 Sentences in conversational turns. A case of syntactic “double bind”.
In: Mario Dascal (ed.), Dialogue, 233–245. Amsterdam and Phila-
delphia: Benjamins.
delphia: Benjamins.
Hannay, Mike and J. Lachlan Mackenzie
1996 Effective Writing in English. A Resource Guide. Groningen: Marti-
nus Nijhoff.
Hansen, Maj-Britt Mosegaard
2001 L’importance de l’analyse des interactions pour l’étude grammati-
cale de la langue. Revue Romane 36. 115–131.
Harder, Peter
1995 Subordinators in a semantic clause structure. In: Betty Devriendt,
Louis Goossens and Johan van der Auwera (eds), Complex Struc-
tures. A Functionalist Perspective, 93–118. Berlin and New York:
Mouton de Gruyter.
1996 Functional Semantics. A Theory of Meaning, Structure and Tense in
Hengeveld, Kees
der Auwera and Louis Goossens (eds), Ins and Outs of the Predica-
tion. 53–66. Dordrecht: Foris.
25: 127–157.
210 Peter Harder
Kroon, Caroline
1995 Discourse particles in Latin. A study of nam, enim, autem, vero and
Richard A. Gatward (eds), Discourse and Pragmatics in Functional
Grammar. 17–32. Berlin and New York: Mouton de Gruyter.
Quirk, Randolph, Sidney Greenbaum, Geoffrey N. Leech and Jan Svartvik
1985 A Comprehensive Grammar of English. London and New York:
Longman.
Slobin, Dan
1987 Thinking for speaking. In: Berkeley Linguistics Society: Proceedings
of the Thirteenth Annual Meeting. 435–445. Berkeley CA: Berkeley
Linguistics Society.
Functional Grammar and the dynamics of
discourse
María de los Ángeles Gómez-González
1. Introduction
From its inception to the present day (see Anstey, this volume) Functional
Grammar (FG) has striven to satisfy the three criteria of functional ade-
quacy it set for itself, i.e. pragmatic adequacy, psychological adequacy and
typological adequacy.1 Accordingly, a prime concern for FG co-workers
has been to provide linguistic descriptions with a strongly universalist bent
so as to not only relate as closely as possible to the rules and principles
governing verbal interaction, but also to dovetail with psychological mod-
els of linguistic competence and linguistic behaviour (Dik 1997: 13, 17).
To better suit these demands, in particular to span the perceived gap be-
tween discourse/cognition and grammar (Mackenzie and Keizer 1991;
Gómez-González 1998b, 2001), a number of FG representatives have felt it
necessary to work towards a discourse- and cognitively oriented expansion
of the standard model, which is, broadly, a lexico-semantically based,
three-staged bottom-up layered framework converting lexical elements and
terms into pragmatico-semantic constructs, converted in turn into prepho-
netic strings through the operation of expression rules (Dik 1997: 58ff.;
Hengeveld 1997). Crucially, in this standard model conceptual unity is en-
tailed between discourse and grammatical analysis by treating each
utterance as a mini-discourse (cf. also Moutaouakil, this volume).
By contrast, Kroon (1997), Bolkestein (1998), Steuten (1998a, b), Vet
(1998), Van den Berg (1998), Connolly (1998, this volume) and Fortescue
(this volume) assume that grammar and discourse are in essence inc-
ommensurable, and accordingly assign corresponding modules to
grammar, which is static and out-of-time, on the one hand, and to dis-
course, which is dynamic and ongoing in time, on the other. A different
position is taken by Nuyts’s (1992, this volume) Functional Procedural
212 María de los Ángeles Gómez-González
Grammar (FPG) and Mackenzie’s (1998, 2000, this volume) Incremental

Functional Grammar (IFG), which draw closer attention to work in lan-
guage production and comprehension. These represent two alternative,
though compatible top-down attempts to accommodate grammatical de-
scription to the dynamic nature of discourse organization, conceived in
both programmes as an interactive, flexible process that unfolds in real
time in ‘chunks’ or in mentally digestible units. FPG claims that linguistic
structures emerge from abstract cognitive factors or conceptualizations
which are fundamentally different from linguistic (lexico-semantic) or-
ganization, yet are still strongly language-bound (based on predicate-
argument constructs). The whole process of language production is claimed
to be a matter of gradually singling out packets of information which get
coded in single utterances (rarely full clauses, especially in speech), using
mid-to-short-term memory to adapt long-term conceptual knowledge to the
actual communicative situation and to the process of language production.
And it is only the last part of this process, the single utterance, that gram-
mar actually deals with (Jackendoff 1987, 1990). IFG, in turn, linking up
with research into starting points (MacWhinney 1977), posits an incre-
mental or gradual left-to-right build-up of chunks of information
(utterances or discourse acts), from minimally a P1 in Focus position, the
first increment or subact (the holophrase) which exploits the primacy effect
or prominence of first mention, to rightward expansions or linear se-
quences of subacts, with late, possibly last placement of the Focus in
conformity with the recency effect, according to which information offered
last is best retained (cf. Gómez-González 2001). Concomitant intuitions are
found in Bakker and Siewierska’s (this volume) rendering of expression
rules as dynamic, and working top-down, depth-first, and left-to-right as in
actual language production.
Hengeveld’s Functional Discourse Grammar (FDG; this volume) seems
to accommodate the insights gleaned from the aforementioned broad-brush
accounts. With a view to constructing a new architecture for FG, FDG re-
veals itself as a top-down hierarchical modular approach, in which entities
of a purely pragmatic Interpersonal Level (IL) at the top (acts/moves of ref-
erence and ascription) and purely semantic entities of varying orders at a
Representational Level (RL) next down are mapped by expression rules
(ER) onto the Expression Level (EL) at the bottom, i.e. the phonological
string that eventually triggers articulation. Speakers – Hengeveld claims –
draw on the cognitive component, covering (long-term) knowledge including
communicative and linguistic competence, at each of the three levels, while
the communicative component, containing (short-term) linguistic co(n)textual
FG and the dynamics of discourse 213
information, is fed by the interpersonal and expression levels and simultane-

ously feeds the representational level in order to enable later reference to
earlier acts and expressions.
This discourse- and cognitively oriented expansion of FG (see also
Harder, Connolly, Verstraete, this volume) parallels the trend of other
functionalist approaches, notably Systemic Functional Grammar (SFG) and
West Coast Functionalism (WCF; e.g. Cognitive Grammar, Construction
Grammar, Emergent Grammar, Conversational Analysis, etc.),2 as well as
the more formally-oriented proposals of e.g. Lexical Functional Grammar
(LFG), Generalized Phrase Structure Grammar (GPSG) or Head-Driven
Phrase Structure Grammar (HDPSG). Some common ground thus seems to
be emerging, seeking speaker-oriented models that treat the speech produc-
tion process as running from intention to articulation (Levelt 1989),
thereby assigning discourse and cognition corresponding levels of repre-
sentation. In these emergent approaches linguistic expressions (and by
implication universals) seem to arise out of the interplay of language-
specific properties (e.g. lexico-grammatical semantics, constructional
meaning) and factors that are not specifically linguistic, such as constraints
on attention and memory, the dynamics of discourse, co(n)textual charac-
teristics, channel restrictions etc., as processed by the human cognitive
system, despite the limitation that at present there is no full-blown answer
as to how this system works or what it looks like.
Possibly to some extent sprouting from this limitation, there are aspects
of FDG – as Hengeveld concedes – that deserve further elaboration in or-
der to become a fully-fledged framework. Indeed, in line with Bakker and
Siewierska (this volume), one misses a more detailed account of the role of
the lexicon in the model, of the processes whereby semantic representa-
tions come into existence, and of the reasons why left-to-right expansions
are only assumed for the interpersonal level when we have every reason to
suspect that these restrictions equally affect the entire production process.
Furthermore, FDG claims that in the coding of his/her communicative in-
tention, the speaker relies on entities of the third, second, first or zero order
(i.e. propositional contents (p), states of affairs (e), entities or individuals
(x), and properties (f)), but at the same time the possibility is admitted that
all entity types may also be expressed in a non-hierarchical way (i.e.
through lexical items). One could find in this last procedure an argument to
support the view that, instead of a threefold hierarchical ordering of IL, RL
and EL, only the latter is required (see Nuyts and Fortescue, this volume,
for concomitant suggestions to minimalize the model), with the lexicon and
the grammar forming a continuous spectrum of symbolic assemblies or
constructions, apparently in conformity with evidence reported from lan-

guage processing, language acquisition, aphasia and categorization (Bates
and Goodman 1997).3 As a corollary, there would be no need to posit un-
derlying structures or derivations, nor (expression) rules to link the
different components.
Different substantiations of this kind of non-hierarchical non-modular
framework are subscribed to by the exponents of WCF. Roughly, these ap-
proaches, which are, as has been remarked, in many respects analogous to
FDG, (a) defend a materialist model of perception with an empiricist func-
tionalist and universalist orientation, and (b) postulate real-time dynamic
programmes that discard the division between a core grammar and a pe-
riphery grammar, under the assumption that all constructions, whether
relatively recurrent or highly idiosyncratic, stand on an equal footing for
description, resorting to, among other things, (c) non-discrete categories or
prototype theories, (d) monostratal or ‘surface level’ syntactic descriptions,
and (e) making no sanctioned distinction between grammatical knowledge,
knowledge of language use and other sorts of knowledge under the as-
sumption that “knowledge of language is knowledge” (Goldberg 1995;
Langacker 1987, 1991, 2001b; Noonan 1999, Kay and Fillmore forthc.).
This latter stance (e) apparently flies in the face of Hengeveld’s outline
of FDG (this volume), in which Cognition and the Communicative Context
are presented at the sides of the schema as two separate, though connected
modules. This representation could lead to the assumption – counterin-
tuively in my view – that Cognition (i.e. (long-term) knowledge) and
communicative context (i.e. (short-term) linguistic co(n)textual information)
can be dissociated or can operate independently of each other. Instead, we
have every reason to suspect that in encoding and decoding strategies the
communicative context is filtered through and interpreted by the speaker’s
cognitive system: the latter encompasses the former as it were, both with a
certain amount of independence but nonetheless coordinated, the underlying
rationale being that processing proceeds simultaneously on multiple time
scales, and with respect to numerous parameters. Moreover, it seems that one-
way arrows running from and towards Cognition, the Communicative Con-
text and the three levels of grammar in different guises infuse the model
with a rigid one-way directionality that impoverishes the supposedly inter-
actionist (bi-directional or multidimensional) dynamicity that the model
strives to confer to the cognition-discourse-grammar relationship. In sum, a
modification of this design is here called for to better symbolize
Hengeveld’s claim that “the three levels interact with a cognitive component
and with a communicative component” (this volume: 3).
Also in this volume, Cornish, Inchaurralde and Hengeveld himself

stress the need for fleshing out the Cognition and Communicative Context
components of FDG, but without moving away from our main object of
study: the linguistic message. Of interest in this context is Mackenzie’s
(this volume) suggestion that the interaction and expression levels should
be viewed as incrementally operative in real time, while the representa-
tional level would act as a buffer between these two real-time processes,
serving to organize the material into linguistic form. Convergent with this
re-interpretative spirit and with the ongoing discussion on the discourse-
and cognitively oriented expansion of FG (and other approaches), I would
argue for an Incremental Discourse Cognitive Grammar (IDCG), as out-
lined in Figure 1 below.
Gathering inspiration from Hengeveld’s FDG, Nuyts’s FPG, Macken-
zie’s IFG, and cognitive grammar (CG; Langacker 1987, 1988, 2000,
2001b), IDCG pictures the connection between discourse expressions and
cognition as a dynamic real-time process in which information is incremen-
tally presented in coherent ‘packets’ of digestible size.
Broadly identifiable with intonation units,4 information packets are rep-
resented as framed (or co(n)textually delimited) bi-dimensional events,
comprising both conceptualization (embracing IL and SL) and expression
(EL), each with various co-ordinated planes. EL includes discourse expres-
sions of all sorts, segmental, suprasegmental, and paralinguistic or kinesic.
Conceptualization incorporates the interlocutors’ apprehension of the
ground (including the speech event itself, the speaker (S) and addressee
(A), and their immediate spatio-temporal circumstances), on the one hand,
and the current discourse space (CDS), on the other, that is, the mental
space comprising the elements and relations construed as being shared by S
and A at a given moment in the flow of discourse. Therefore, in this model
CDS embodies speakers’ shared knowledge, which in turn includes the
context of speech involving physical, mental and socio-cultural variables.
At the centre of the speech context is the ground, where S and A are en-
gaged in a co-operative viewing of some facet of any world (real or
imagined).
Figure 1. An outline for an IDCG

In this process both S and A, whether playing an active or reactive role,

have to deal with gradual or incremental conceptualizations and expres-
sions concerning their focus of attention (symbolized as →). In other
words, speakers’ decoding and encoding strategies involve time-
constrained conceptualizations and expressions of lexico-grammatical con-
structs that are contextually relevant, that is, which respond to text-building
and speech management strategies (i.e. information structure, textual pro-
gression, genre characteristics, conversational routines, etc.) A finer-
grained elaboration of IDCG lies well beyond the scope of this chapter.
Space constraints oblige us to concentrate on the implications that this
model may bring to the dynamics of discourse. Section 2 revises what in an
IDCG are felt to be the most relevant analytical tools to explore this issue,
namely the notions of choice, perspective and prominence as manifested in
three important attention-related dimensions: Topic, Focus, and the Theme
zone. The chapter closes with a summary (Section 3).
2. The dynamics of discourse

Effective communication is an essentially strategic encoding and decoding
process, in which both S and A, whether playing an active or reactive role,
are continuously monitoring choices as to what is being said, what they
want to say, and how they want to say it, generally following two basic co-
operative principles: ‘be orderly’ and ‘be consistent with the existing topic
framework’ (Grice’s 1975 maxims of relevance and quantity). Inherent to
the notion of choice is therefore the fact that all languages offer (mini-
mally) distinct available options at different levels – such as those
exemplified in (1) and (2) – which derive their distinctiveness fundamen-
tally from their syntagmatic properties or the associative relationships
holding among their constituents, on the one hand, and from their paradig-
matic potential on the other, that is, from the oppositive or contrastive
relationships established with other units. It is these two different kinds of
relationships, paradigmatic and syntagmatic, that make one option more or
less appropriate as a motivated choice for a particular discourse co(n)text.
(1) a. and ∨there ⏐ on the /table ⏐ where my ∨newspaper had been ⏐ was my
packet of \biscuits ⏐ (LIBMSECGPT03: 046-47)5
b. and my packet of biscuits was there, on the table, where my newspaper
had been.
(2) a. ⏐ there were \three of us in the \car ⏐ \all °rather \nervous ⏐ (LIBMSE-
CAPT04: 018)
b. in the car (there) were three of us, all rather nervous.
c. three of us were in the car, all rather nervous.
d. three of us were all rather nervous, in the car.
In what follows it will be argued that, even though truth-conditionally

equivalent or describing precisely the same objective situation, different
discourse expressions like those in (1) and (2) above correspond to differ-
ent conceptual organizations that imply different directions of mental
scanning, i.e. different paths of mental access, in the conceptualizers’
minds (both S and A) when constructing a full conception of a given SoA.6
In other words, different discourse expressions establish different view-
points (or vantage points) and determine different perspectives (global or
local) on the experience being constructed. For one thing, discourse in-
volves a succession of attentional frames, each of which represents the
scene being ‘viewed’ and acted on by S and A at a given instant, and it is
this viewing process as perspectivized by S and A that determines the
‘camera angle’ or ‘camera movement’ of discourse, structuring it into sepa-
rate attentional units or episodes (see also note 3). One particular framing
may be unmarked, or prototypical, in contrast with an array of compara-
tively marked alternatives, but all may be familiar and conventionally
sanctioned in specific situations.7
A major dimension of mental accessing is the degree of discourse
prominence accorded to various elements within a conceptualization, most
importantly topical, focal and thematic prominence. We will here not go
into the nitty-gritty of the categories of Topic, Focus, Theme – and other
related concepts – as a detailed analysis has been given elsewhere (Gómez-
González 2001). Rather, in the following subsections we shall offer a
sketchy presentation of these notions – for the most part coinciding with
FG accounts, although in some occasions departing from or complement-
ing them, especially in the case of Theme – so as to demonstrate how these
different kinds of prominence affect the Focus of attention of discourse in
keeping with the newly posited IDCG.
2.1. Topic
Dik’s view (1997: 312) will be endorsed that topicality concerns the status
of those entities ‘about’ which information is to be provided or requested in
the discourse. However, an expansion of the FG notion of Discourse Topic
(D-Topic) is suggested here which will entail a relationship rather than an
entity or a thing. As the reader will remember, Dik (1978: 19) explains that
Topic is “the entity ‘about’ which the predication predicates something in
the given setting”. A threefold implication follows: (a) only one Topic per
predication is allowed; (b) this function can only be assigned to terms (i.e.
discourse referents/participants, mainly Subjects or Objects), and (c) topi-
cal elements are endowed with different degrees of referential accessibility
in terms of being (i) Given Topics (GivTop), (ii) SubTopics (SubTop), (iii)
Resumed Topics (ResTop), or (iv) New Topics (NewTop): the first presen-
tation of a D-Topic is a NewTop,8 and an entity so introduced becomes a
GivTop if re-introduced; likewise, an entity which is mentioned, temporar-
ily neglected and later revived is a ResTop, while an entity inferred from
other entities is labelled SubTop.
This account of D-Topics in terms of entities endowed with different
degrees of referential accessibility raised a number of problems explained
in detail elsewhere (Gómez-González 1998b, 2001: 156–168), most impor-
tantly:
(a) the assumption that only entities qualify for D-topical status, while a
person’s pragmatic information also includes the SoAs in which
those entities play a role;
(b) the apparent identification of Topic with predication-internal entities,
as opposed to Theme and Tail (both predication-external), for it
seems that P1/P2 assignment neither is a discrete distinction nor can
be upheld as the only means of distinguishing these categories;
(c) the difficulties involved in identifying or distinguishing the four
Topic types.9
In an attempt to surmount at least some of the aforementioned compli-

cations, I will here resort to a more comprehensive, more cognitively-
oriented notion of D-Topic, from Kemmer (1995: 58):
a prominent conceptualisation which acts as a kind of cognitive anchoring

point; other [...] conceptualisations are brought into the discourse by virtue
of their perceived ties to the reference point.
Crucially D-Topics of this kind, whether local or global, entail relation-

ships rather than things or entities.10 This interpretation reflects the fact that
connected discourse comprises a succession of attentional frames – a series
of intonation units mapped onto clauses, or parts of clauses – each profiling
a relationship (Langacker 2000, 2001a, b; Gómez-González 2001). At the
basis of this relationship, an expression evokes a certain body of concep-

tual content, its ground or base. This directs attention to some particular
substructure – the profile – construed as the relationship (subsuming the
entity or entities) that the expression designates or refers to. By way of
illustration, (1a) and (1b) above profile an abstract presentational
relationship, (1b), (2c), and (2d) one of spatial attribution, and (2a) one of
‘existence’ (see Sections 2.3 and 2.4 for further reference to these alternate
relationships).
A D-Topic therefore refers schematically to the relationship (or thing)
profiled. It has a prospective, or forward-looking potential by virtue of sig-
nalling that the profiled relationship or entity will function as a conceptual
reference point for purposes of interpreting a subsequent proposition or es-
tablishing mental contact with another reference point (its target) within –
as is most likely – or outside its cognitive schema, or dominion, that is to
say:
... any system of concepts related in such a way that to understand any of
them you have to understand the whole structure in which it fits; when one
of the things in such a structure is introduced into a text, or into a conversa-
tion, all of the others are automatically made available. (Fillmore 1982:
111)
In addition, D-Topics have a retrospective or backward-looking dimen-

sion, in the sense that they normally refer to information that is already
accessible in the discourse co(n)text. This is implied by the very notion of
building a coherent or cohesive structure for, when structure is added, it is
usually anchored to what has already been built, by virtue of the conceptual
overlap. It follows that – as implied in FG – D-Topics tend to be endowed
with the features of identifiability and activation. Identifiability accounts
for the difference between referents for which the speaker assumes a file
has already been opened in the discourse register and those for which such
a file does not yet exist (Chafe 1976). What counts for the linguistic ex-
pression of this cognitive distinction is whether or not the speaker is able to
pick out a referent from among all those which can be designated by a par-
ticular linguistic expression and identify it as the one which the speaker has
in mind, provided that, given the appropriate discourse context, a referent
is more or less permanently stored in the long-term memory of S and A,
and can be retrieved without difficulty at any particular time. A referent
may be identifiable either because it has been mentioned in a discourse (i.e.
through anaphoric reference) or because it is either visible or otherwise sa-
lient in the speech setting or because it is ‘inalienably possessed’ or other-

wise anchored in the individuality of one of the interlocutors (i.e. through
deictic reference, e.g. those ugly pictures or the woman in the green hat
over there, your left leg, or my sister's second ex-husband). All instances of
identifiability and their expression by a single grammatical category as-
sume the existence of a cognitive schema, or frame.
Activation, on the other hand, evokes Chafe’s (1987: 22) idea “that our
minds contain very large amounts of knowledge or information, and that
only a very small amount of this information can be focused on, or be ‘ac-
tive’ at any one time”. Accordingly, three following degrees of activation
are posited:
(a) active, designating “currently lit up” information that is, as Chafe
has it, “in a person's focus of consciousness at a particular moment”;
(b) semiactive (or accessible), designating information that is in a per-
son’s peripheral consciousness or in their background awareness as a
result of: (i) deactivation from an earlier state (i.e. textual accessibil-
ity); (ii) inference from a cognitive schema or frame, that is, from
some other active or accessible element in the universe of discourse
(i.e. inferential accessibility); and (iii) presence in the text-external
world referent (i.e. situational accessibility).
(c) unused, designating inactive information that is “currently in a per-
son's long-term memory, neither focally nor peripherally active”.
Note that, although D-Topics normally entail some degree of identifi-

ability and/or accessibility, not all identifiable or accessible items become
D-Topics, for the latter must also show a prospective dimension (see note
13). The problem arises how to identify these cognitively oriented D-
Topics. For this sketchy presentation it suffices to assume that D-Topics
generally profile relationships normally anchored in the situational and lin-
guistic co(n)text, in conformity with natural paths of mental accessing (see
notes 5 and 14). This grounding process occurs in successive attentional
frames (intonation units with corresponding Focus spans), whose arrange-
ment is basically linear and affected by the order of presentation (in
principle what is presented first will profile what follows, see Section 2.3),
and which are superimposed on structural elements, whose arrangement is
both linear and hierarchical. There is a natural tendency for attentional
frames to coincide with grammatical constituents, but it is only a tendency,
not an inviolable principle. Conceptual organization need not be directly
mirrored in grammatical constituency, for grammar is a tool for building up
to – and symbolizing – complex conceptualizations, but it is not to be iden-

tified with those conceptualizations.
Further, this conception of D-Topic does not always point to one par-
ticular element in a proposition, or to one particular constituent. For one
thing, D-Topics also allow for degrees of topical relevance and degrees of
activation and identifiability that are incrementally built up in the scaffold-
ing of discourse, not to forget the elusiveness and fuzziness implicit in the
criteria upheld to assign D-topical status, i.e. ‘aboutness’, ‘activation’, or
‘identifiability’.
2.2. Focus
Here we shall depart somewhat from FG practice, which attaches Focus
status to those pieces of information which represent the “relatively” most
important or salient information (Dik 1978: 19; 1997). To my mind, this
description is better captured under the title ‘Focus of attention’ (Section
2.4) – equivalent to the FG account of focality11 – because discourse infor-
mation may become prominent (‘important’ or ‘salient’) not only as a
result of its focal status, but also, among other things, for its topical quality
and its thematic (vs. rhematic or final) arrangement, each of these dimen-
sions involving corresponding lexico-grammatical, phonological and
paralinguistic expressions (cf. Cornish this volume).
In this chapter Focus represents a purely phonological term, equivalent
to Focus accent. It designates intonational prominence in an informational
unit (see note 3), which may be associated not only with information pack-
aging but with other linguistic dimensions as well (e.g. illocution, string-
based deaccenting) and which may also be a consequence of non-linguistic
effects. As indeed stated in FG, Focus accent is taken to represent the pro-
sodic means whereby – in combination with morphosyntactic devices such
as morphological markers, word-order variation and special lexico-
grammatical constructs – the syntactic domain of Focus (the Focus do-
main) is expressed. Accordingly, a semantic element belonging to the
Focus component of a pragmatically structured proposition is said to be in
Focus or focal (as opposed to being ‘in the presupposition’ or ‘presupposi-
tional’), regardless of whether the constituent coding it carries an accent or
not. For example, in I went to the movies as a reply to What did you do last
night?, both the (relatively) unaccented constituents went and to together
with the accented constituent movies are in Focus (Lambrecht 1994: 209)
[emphasis mine].12 And, again as remarked in FG, Focus accent may fall
either upon New (inactive) information or upon Given (active) information.
Examples given by Siewierska (1991: 174) are Sonia and Joyce came to
help me. Sonia worked like mad, but Joyce was horribly slow; or I heard
that Peter got married. Peter’s married. How amazing! I don’t believe it!
[emphasis mine].
In principle, I see no objection to Dik’s (1997: 315ff.) distinction be-
tween New Focus (NewFoc) and Contrastive Focus (ContrFoc), broadly
evoking the widely agreed differentiation between broad and narrow Focus
(Selkirk 1984; Lambrecht 1994; Ladd 1996).13 However, I would contend
that besides assigning ContrFoc a number of functionalities (Parallel Focus
and Counter-presuppositional Focus, which comprises Replacing Focus,
Expanding Focus, Restricting Focus and Selecting Focus), a further subdi-
vision within NewFoc should be also posited in order to account for the
structural and attentional domains of two construction-types that are recur-
rent across languages, namely utterance-Focus (UttFoc) and predicate-
Focus (PredFoc) (cf. Lambrecht 1994; Vallduví and Engdahl 1996; Lam-
brecht and Polinsky 1997).
In a UttFoc construction (e.g. (3) below) both the Subject and the predi-
cate are in Focus, i.e. the Focus extends over the entire proposition. The
purpose of the assertion is to express a proposition which is linked neither
to an already established D-Topic nor to a presupposed open proposition.14
In other words, crucial in UttFoc constructions is the absence of a presup-
position attached to either the Subject or the Predicate, and resultingly the
identity of assertion and Focus (cf. Kuno’s 1972 ‘neutral description’; Al-
lerton and Cruttenden’s 1979 ‘all-new utterance’; and Kuroda’s 1972
‘thetic sentence’). UttFoc has been attributed to:
(a) impersonal expressions, that is, events, states, situations or facts in

which no referential entity is present and therefore nothing can be
said about it;
(b) presentative constructions, which introduce entities, but fail to report
an event about them;
(c) any state of affairs presented as a compact whole representing noth-
ing but new information.
It is the absence of a presupposition and the identity of assertion and

Focus that distinguishes UttFoc in (3) both from the PredFoc construction
in (4) and the ContrFoc in (5) below. In the PredFoc category the Predicate
is in Focus and the Subject (and possibly some other argument or adjunct)
is within the presupposition: the purpose of the assertion is to pragmati-
cally predicate some property of an already established discourse referent;
whereas in the ContrFoc category, the Subject (or some other argument) is
in Focus and the predicate is within the presupposition (examples are taken
from Lambrecht and Polinsky 1997 [emphasis in original]).
(3) UTTERANCE FOCUS (UttFoc):

Context sentence: Why didn’t she come work today?
Sentence: Her SON is sick.
(4) PREDICATE FOCUS (PredFoc):
Context sentence: Why didn’t she come to work today?
Sentence: Her son had an ACCIDENT.
(5) CONTRASTIVE FOCUS (ContrFoc):
Context sentence: Why didn’t she didn’t come to work today?
Sentence: Her SON is responsible.
Of importance here is that this threefold distinction of Focus types in

terms of UttFoc, PredFoc and ContrFoc seems to fit in well with FG work,
that is, Hannay’s (1991, 1998) three message modes, i.e. Grounding, Neu-
tral and Reaction, which also attach corresponding attentional and
structural properties to these three utterance types (see Section 2.4).
2.3. The Theme zone

For compelling evidence and reasons explained in detail elsewhere
(Gómez-González 1998a, 2001), behind which is the premise that the
speaker’s choice of point of departure, or Theme, represents a strategic
move in natural language (Mackenzie 1998, 2000, this volume) giving ori-
entation or anchoring for what is to follow in the background of a
co(n)text, the Theme zone (to be underlined in the following examples)
will be taken to extend as far as an initial ‘experiential’ constituent, the ex-
periential Theme (ET), i.e. (a) Subject, (b) Process/Verb, (c) Comple-
ment/Object or (d) Circumstance – all with realizations ranging from
words, phrases to (parts of) clauses – including any (non-restrictive) post-
modification.15 Cross-cutting illocutionary moods and Theme choices, a
thematic arrangement will be unmarked when ETs correspond to the first
element of each mood type, and it will otherwise be marked. Accordingly,
following Halliday (1994), we shall assume that in English the unmarked
Theme of declaratives is the Subject (e.g. Little Bo-peep has lost her sheep,
1994: 43),16 the WH-element in exclamatives (e.g. how cheerfully he seems
to grin, 1994: 47), and in interrogatives, either an auxiliary (carrying the
expression of polarity) plus the Subject, in yes/no questions (e.g. can you
find me an acre of land?, 1994: 48), or the WH-element in WH-questions
(who killed Cock Robin?, 1994: 48). All other thematic choices will be
marked – e.g. (1a), (2b) above in contrast with the other unmarked coun-
terparts.
In addition, also relevant for the ongoing discussion is the fact that the
Theme zone may house different functional loads. I dub ‘extended multiple
Themes’ (EMTs) those cases in which ETs are preceded and/or followed
by elements with a textual or an interpersonal orientation, which I shall call
‘textual Themes’ (TTs) and ‘interpersonal Themes’ (ITs) respectively, un-
der the assumption that the latter mark a boundary between the Theme
zone and the Rest (or Rheme). (6) – (9) below illustrate different kinds of
EMTs:
(6) a. ⏐ a —journalist ∨convoy ⏐ of about \fifteen ∨cars ⏐ was a\ssembled ⏐ and —

lights ∨flashing ⏐ it was ∨dusk ⏐ we —tore ‘down the \road ⏐ ‘holding —white
∨flags ⏐ —out of our \windows ⏐ (LIBMSECA04: 080) (ET) (TTÊTÊT)
b. ⏐ yet —she for her /treasure will en—dure ⏐ and ∨tremble ⏐ and —so find —
peace ⏐ that \passeth our under\standing ⏐ (LIBMSECH04: 033) (TTÊTÊT)
c. ⏐ /well ⏐ _em ⏐ \those that \come from agricultural \families \will ‘work ⏐ ∨on
the ‘land⏐ (LSECJPT06: 279–80) (TT^TTÊT)
(7) a. ⏐ and of \course they ‘lived in ∧ ‘segregated ∨quarters ⏐ which was \maybe O‘K
in the be∨ginning ⏐ but ∧ \later there was a ∨doorman ⏐ on the ∧ ∨women's build-
ing ⏐ ‘and they ∧ just —weren't a‘llowed to ∧ ‘visit at \all ⏐ (LIBMSECJ06: 016–
017) (TTÎTÊT)
b. ⏐ because ‘obviously Su∨dan is ⏐ ‘by and large a ∨Muslim country ⏐
(LIBMSECC06: 297) (TTÎTÊT)
(8) a. ⏐ West∨morland for e°xample ⏐ became par\ticularly ∨passionate ⏐ when
\
talking about the °influence ⎯television re∨porting from Viet°nam ⏐ had \had on
the ∨White °House ⏐ in the °late \sixties ⏐ (LSECAPT03: 030) (ET^TT)
b. ⏐ —one figure inci‘dentally ⏐ you \may find \useful for ⏐ ∨tax ‘purposes ⏐
(LIBMSECF02: 064) (ETÎT)
(9) a. ↑and in ⎯Mara\dona ⏐ without \doubt ⏐ they °had the \star of the ∧
°compe°tition ⏐ (LIBMSECJPT01: 065) ) (TTÊTÎT)
b. ⏐ the \Shi'ites in∨side ⏐ are \not the ‘hard cases of the _terrorist \world ⏐ in‘deed
∨
most of them ⏐ ‘probably did ‘little /more ⏐ than ‘be in the —wrong /place ⏐ at
the _wrong \time ⏐ (LIBMSECA07: 028) (TTÊTÎT)
c. ⏐ thus one /third for e‘xample ⏐ was ∨written as ⏐ —three _vertical \strokes ⏐
↓with the ∨mouth symbol placed a\bove them ⏐ (LIBMSECD03: 040)
(TTÊT^TT ⎯the two TTs being of different kinds⎯)
The excerpts above reveal a number of most important things. The first
is the vast richness entailed by the Theme zone, in terms of function (hous-
ing discourse markers, discourse connectors, vocatives, attitudinal markers,
circumstantial specifications, and/or entities of various orders), syntactic
realization (with the potential of having complex realizations, expanded
through relationships of subordination, coordination or apposition) and
volumewise (having the possibility of being packed within one intonation
unit or extending over more than one unit). Incidentally, the same potential
applies to the Rest (or Rheme). For this reason, here both the Theme and
Rheme zones are viewed not as a discrete locations, with fixed boundaries,
but rather as incremental and flexible zones entailing different kinds of
wave-like or pulse-like points of prominence. Theme profiles an orienta-
tional relationship as a result of the recency effect: discourse scaffolds
normally ground what is to be said on the previous context or the situ-
ational context. By contrast, Rheme exploits the recency effect, i.e. the
saliency of what is said last, in conformity with the principles of End Focus
(the late placement of focal accent) and End Weight (the last mention of
weighty units). Both kinds of prominence conspire to place active dis-
course information within the Theme zone and comparatively less active
material in the Rheme, following the principle of Functional Sentence Per-
spective, i.e. the Given-before-New array of information. But, again, these
are only tendencies, not inviolable principles.
Now, to go back to the Theme zone and to summarize, its flexibility
admits: (a) recursion (paratactic, hypotactic or appositive) within each of
its three orientational subfields, ET, IT, TT – as shown in (6) above – and
as a corollary, (b) either a P1 realization within a larger intonation frame
(e.g. (2a), (6b), (7a), (7b)), or a P2 realization, that is, a broadly ‘clause-
external’ realization, in which case Themes either have one attentional do-
main of their own (e.g. (7a), (7b), (8a), (8b), (9c)) or otherwise extend over
more than one frame (e.g. (1a), (6a), (6c), (9a), (9b)).
The presence/absence of intonational integration at the beginning and/or
end of an utterance implies the presence/absence of corresponding Focus
spans or message peaks (Bolkestein 1998; Hannay 1994). Hence, Themes
that are not marked off intonationally result in their phonological and con-
ceptual ‘compression’, thereby profiling the subsequent discourse
relationship as a single, complex characterization: more has to be squeezed
into a single, limited span of processing time, resulting in a somewhat less
articulated realization. By contrast, detached Theme zones are incremen-
tally coded as separate attentional gestures, thereby enhancing their
cognitive salience and that of the elements within their scope, if only by
being accorded intonational prominence, more processing time and thus a

fuller realization. The two implementations are quite distinct both experi-
entially and at the level of motor activity. And, most importantly, they
show that – as noted above – different structural configurations involve
some differences in conceptual organization.
In keeping with this, another issue that we will consider before closing
this section is the cognitive repercussions of the different arrangements of
the orientational subfields, i.e. ET, IT, TT, within the Theme zone. Here
we shall assume that the order of expression always induces a conceptual
ordering of the notions symbolized, as one facet of our dynamic apprehen-
sion of linguistic meaning. But, needless to say, not every facet of dynamic
conceptualization is necessarily correlated with expression order; neither
does the sequence of presentation in discourse necessarily coincide with
the chronological ordering implied.
The data handled in Gómez-González (1998a, 2001) reveal that EMTs
tend to display the organization TTÎTÊTÎT^TT – as shown in excerpts
(6) – (9) above and Figure 2 below – with the constraint that, if TTs and/or
ITs occur in the pre-ET subfield, then items of the same kind do not recur
in the post-ET subfield (for a classification of the types of TTs and ITs see
Gómez-González 2001: 110, 180–185). Further, an utterance may mini-
mally consist of one kind of Theme only, be it TT, IT, or ET, which will
therefore be in Focus (e.g. Tomorrow?, Fine and Great) as possible reac-
tions to Maria is coming for a visit).
y
x
2 1 H 1 2
(TT) ^ (IT) ^ ET ^ (IT) ^ (TT)
PRE-ORIENTATION INTERNAL POST-ORIENTATION

ORIENTATION
narrowest scope
experiential meaning
interpersonal meaning
textual meaning
widest scope
Figure 2. The Theme zone’s maximal expansion

The unmarked array of EMTs illustrated in Figure 2 represents y (TT),

x (IT) and H (ET) as being arranged in relation to their scope of influence,
suggesting that ET, IT (1) and ET (2) have increasing scopes of influence,
each within the span of the next. Thus ETs profile a backward-looking and
forward-looking relationship pointing to the message itself, but with ITs,
this profiling involves in addition the interlocutors (S and A), while with
TTs the profiling expands to the global logical organization of discourse.
All other arrays are claimed to be marked, or less frequent, or otherwise
determined by item-specific mobility conditions. Note, for example, the
flexibility of however, a textual item, which may appear (a) in the thematic
pre-orientational field (e.g. However, the man didn´t come), (b) in the the-
matic post-orientational field (The man, however, didn´t come), (c)
crammed within an EMT (The man, however, the one we saw yesterday,
didn´t come), or (d) outside the Theme zone (The man didn´t come, how-
ever).
Consistent with the unmarked array of the Theme zone, three further
orientations will be attributed to EMTs: the high-low orientation, the low-
high orientation and the combined type (cf. Smits 2002). High-low orienta-
tions are those profiled when TTs and/or IT occur in the pre-orientational
field, normally in decreasing scopal spans, in order to provide a textual and
interpersonal ground according to which the subsequent discourse is to be
interpreted. Illustrations can be found in the EMTs in (6) and (7) above. By
contrast, low-high orientations obtain when TTs and/or ITs occur in the
post-orientational zone, as shown in (8) above. Normally arrayed in in-
creasing scopal spans, and in corresponding intonation units, post-
orientational ITs and/or TTs separate ETs from what we have dubbed the
Rest, thereby distributing the focus of attention between both chunks of in-
formation, the ET and the Rest, which are zoomed in on and accordingly
receive cognitive salience (more processing time and thus a fuller realiza-
tion) and focal prominence. In mixed orientations – as the label suggests –
the functional capacities of both high-low orientations and low-high orien-
tations are combined (as in (9) above).
In the following section we shall expand the view that the different
kinds of orientation profiled by the Theme zone, whether representing
marked and unmarked choices, entail different instructions for the camera
angle of discourse.
2.4. Focus of attention

Resulting from the merging of attentional framing and structural con-
straints, linguistic representations are here viewed as instructions to modify
the S’s and A’s Focus of attention or current discourse space in particular
ways (cf. Hannay 1991, 1998; Harder 1996; Vallduví and Engdahl 1996;
Gómez-González 2001). These instructions – and naturally the correspond-
ing expressions – are here regarded as interpersonally-oriented textual
choices that are cognitively rooted, on the basis of their broadly bi-
dimensional functionality, i.e. their perspective-creating capability, and
their action as discourse markers, packaging information into digestible
units, both aspects influencing the way reality is monitored (encoded or de-
coded) by the human (S’s and A’s) cognitive system.
In unmarked cases, the instructions given by discourse expressions tend
to seek for sameness and usually reflect a static viewing: a constant per-
spective that entails instructions of continuity of discourse participants, of
referential predictability and of foregrounded or discourse-focused mate-
rial. By contrast, as a deviation from the norm, marked discourse
expressions are natural symptoms of a subjective viewpoint and signal an
invitation for a dynamic viewing, entailing, among other things: a topic
change, a change in focus of attention to formerly backgrounded (or de-
focused) material, and digressions or turns within the structure of the text
(Fox 1985; Givón 1985b, 1987, 1988; García 1994: 337).
By way of illustration, we would argue that (1a) above, a Subject-verb
inversion, in contrast with its non-inverted counterpart (1b), and the there-
existential construction in (2a), with a non-existential marked (inverted)
version (2b) and two unmarked alternatives between two different kinds of
expansions or tails (2c, 2d), entail different conceptual organizations or di-
rections of mental scanning. In each case the Theme zone, whether or not it
carries intonational prominence (Focus), provides an orientational frame
with reference to which rightward expansions or increments of discourse
are to be processed, thereby instructing both S and A to focus their atten-
tion accordingly.
In (1a) the Theme zone entails a locational path of access, nested from
more generic spatial reference (i.e. focal There) to more detailed location,
on the table, a focal spatial modifier that houses an equally focalized par-
ticipant, my newspaper, in the relative elaboration, thereby empathizing
with or providing a profiled relationship to the narrator via possession, the
possessor, my, and the possessed, newspaper. The stage being set through
this thematic spatial and empathic grounding, it then specifies what is
found in that location. Thus, the camera angle of the scene shifts to my
packet of biscuits, which is also focal, thereby establishing the postverbal
Subject as a prominent participant that is introduced in discourse in order
to, in this particular case, close a story about biscuits (the global D-Topic:
a stranger eats the narrator’s biscuits), exploiting the recency effect, or
prominence of last mention. Alternatively, (1b) lacks the presentative ef-
fect and simply profiles a spatial relationship attributing a location to my
biscuits (a local D-Topic), again by means of a zooming-in strategy
whereby the focus of attention shifts from a more generic to a more spe-
cific location.17
The presentative strategy recurs in (2b) – also a spatial Subject-verb in-
version – and in (2a), although in the latter case it is implemented
differently through the thematization of existential there. In the first incre-
ment of (2a) the Theme zone profiles an abstract setting, a presentational
frame through which the Subject (the three of us), speaker-empathic
through personal reference, receives focal prominence and exploits the re-
cency effect to become the target of attention over the successive
attentional frames, thereby becoming the D-Topic of the subsequent dis-
course spans. First it is assigned a location (in the car), then a state of mind
(all rather nervous), and finally attention zooms in on one of its members
(the third reporter), through reference chains coding Given information, as
shown in (10) below, an expanded version of (2a). (2c) and (2d) lack this
presentative potential – as was the case with (1b): a reference point (the
three of us) (the local D-Topic) is first assigned a spatial grounding and
then a mental grounding in (2c), and vice versa in (2d).
(10) ⏐ there were \three of us in the \car ⏐ \all °rather \nervous ⏐ the ∨third
re‘porter ⏐ was with the —Washington \Post ⏐ a /war corre‘spondent ⏐ for
\
twenty ∨years ⏐ who'd \covered ‘Viet\nam ⏐ the —Washington ∨Post ‘man
‘said ⏐ he ∨hoped ⏐ that ∧ at an —army ∨checkpoint ⏐ just \before the ‘final
\stretch ⏐ to ‘Suchi toto ⏐ they would stop us from ‘going \through ⏐
\ \
(LIBMSECAPT04: 018ff.)
To summarize, it can be argued that in contrast to the canonical varie-

ties, inversions, like preposings or left detachments, display a leftward
movement within the clause which has a substantial effect, whereas a dif-
ferent effect arises from the rightward movement entailed by existential-
there constructions, or right detachments and extrapositions. Inversions
normally act as instructions to place active, particularly inferrable, ele-
ments clause-initially, which in English are exempt from Subject status and
are thereby unexpected in their discourse role as point of departure, creat-

ing an effect of emotivity, typically over major discourse shifts. This effect
is also contributed to by the marked position of the Subject, which is
thereby signalled to be unexpected as well: in principle some other poten-
tial referent would have been more likely or expected. The Subject referent
does not need to be ‘brand new’, but merely less salient/familiar than the
fronted element. The presentative nature of inversions resides in their fore-
grounding of the speaker’s syntactic point of departure, which (a) usually
receives intonational prominence (ContrFoc),18 (b) precedes a clausal pre-
supposition, and (c) serves a variety of communicative purposes (viz. to
solve problems of encoding, to act as a spatio-temporal framework-setting
device for the introduction of new participants in discourse, to establish a
cohesive tie, or to create an effect of emphasis). These wide-ranging dis-
course functions make inversions particularly appropriate for literary styles
or for relatively vivid accounts of an event, as well as for expository and
argumentative discourse, in which case their role is normally to contribute
to the scaffolding of the S’s explanation or argumentation.
Alternatively, existential-there constructions lack a presupposition of
existence and, in the unmarked cases, place in P1 a Stager, i.e. a scene-
setting spatial or temporal adverbial, which acts as a presentative device
which, in combination with BE or another existential verb, creates an exis-
tential/presentational framework, bringing to the rhematic zone a (lexically
and informatively) ‘heavier’19 Existent that receives focal prominence
(UttFoc), as dictated by the principles of End Focus and End Weight. This
array of information contributes to facilitating the processing of there-
constructions: placed last, heavy rhematic constituents do not disturb the
processing of the thematic part, which at the same time provides an exis-
tential-presentational framework for grounding the most informative
material. Generally, the point of introducing this informative material (the
Existent) by means of a there-construction is to develop it as local D-Topic
over the ensuing discourse span(s).
Because of their functional features, there-constructions are particularly
suited for informative and descriptive discourse seeking spatio-temporal
locations for focal or discourse-prominent information (entities and situa-
tions). The side-effect is often to silence the agency of a State of Affairs,
imbuing these constructions with a depersonalized quality, and thereby
permitting speakers to avoid claims about the responsibility for the asser-
tion in question, a precondition of objective discourse. In keeping with this,
existential-there constructions are, for processing reasons, scarce in active
dialogic discourse. The notional Subject is delayed until the end of the con-
struct, which puts the A’s short-term memory under strain and decelerates
the process of comprehension, an undesirable effect for dialogic discourse,
which normally works on an immediate-response basis.
Lastly, to round off this section, unmarked high-low, low-high and
combined orientations within EMTs can also be associated with corre-
sponding instructions for discourse continuity, discourse discontinuity and
both. High-low-oriented ETMs (e.g. (6) and (7) above) show how dis-
course continuity is ensured by firstly providing a logico-conjunctive peg
on which speaker’s attitudes and the core message – normally in this order
– is to be hung, whether viewed through one attentional window or more.
An alternative discontinuous low-high strategy – as in (8) – assigns focal
and cognitive prominence to the profiled relationship, normally encoded in
the ET. Detached from the Rest (or Rheme) by means of IT and/or TT, the
ET specifies a domain of knowledge, a ‘mental address’, and the proposi-
tion by the following clause is ‘delivered’ to that address, i.e. integrated
into the domain of knowledge centred on the (local) D-Topic.
In most cases, not only ETs, but also post-field ITs and/or TTs, are
framed within their own intonation units, which are accorded focal promi-
nence, more processing time and a fuller realization. Both strategies and
their corresponding discourse effects are present in the combined type
(shown in (9)).20
Owing to their different orientations, EMTs tend to be used either in
constructive discourse, which often requires the help of logico-conjunctive
signposts to develop their specialized contents, or in dialogic discourse,
where speakers can freely express their points of view and have to gain or
maintain the discourse floor. Besides, both constructive and dialogic dis-
course tend to involve the type of State of Affairs preferred by EMTs, i.e.
material processes in the sense of Halliday (1994: 106–107), which, be-
cause of their informative and objective nature, readily accept logico-
conjunctive connectors and interpersonal hedging.
3. Conclusion
In this chapter we have attempted to offer a cognitively-oriented extension of
Hengeveld’s (this volume) FDG, dubbed IDCG in line with the recent trends
observed within and outside FG - mostly Mackenzie’s IFG, Nuyts’s FPG and
Langacker’s CG. Space constraints have forced us to give but a cursory ex-
position of what the model as a whole would look like, to zoom in, in
subsequent sections, on different aspects of the dynamics of discourse.
The process of discourse production and comprehension has been por-

trayed as involving attentional framings superimposed on structural domains,
following the shifts of the S’s and A’s focus of attention. The progression of
this viewing movement is realized incrementally, or in chunks, through cog-
nitive-structural directions or instructions to focus attention on what is felt to
be prominent in discourse. We have seen that very high on the scale of dis-
course prominence are the notions of Topic, Focus and Theme. It is to be
hoped that, its conciseness notwithstanding, this piece of work – part of cur-
rent research – may at least act as a catalyst for further discussion.
Notes
1. I am grateful to Lachlan Mackenzie, Knud Lambrecht, Ronald Langacker
and Francisco Gonzálvez-García for contributing data, comments, and/or
suggestions to this chapter. The research reported here was supported by the
Xunta de Galicia, grant number pgidt00pxi20402PR, research project
‘Análise do discurso na lingua inglesa: Aspectos sincrónicos e contrastivos
con referencia ó galego o mailo castelán’.
2. WCF can be taken to align with the functional paradigm insofar as (a) it
admits that the primary function of the language is to serve as a vehicle of
communication and (b) it strives to account for those aspects of grammar
that are non-arbitrary in terms of functional principles, that is, relating them
to language use (for details see Butler 2003).
3. Carroll et al. (1971) and Goldberg, Sethuraman and Casenhiser (forthc.)
show that children begin to learn the associations between form and mean-
ing on two levels, i.e. verb-centred categories and abstract argument-
structure constructions (especially with such recurrent verbs as do, make,
get, go, etc.). Likewise, Bencini and Goldberg’s (2000) experimental re-
search suggests that individuals are likely to sort out sentences in terms of
argument-structure constructions rather than basing themselves on the lexi-
cal semantics of the matrix verb, which gives evidence for the psychological
or cognitive reality of constructions in language users’ minds.
4. An intonation unit can be characterized as being typically demarcated by
pauses or breaks in timing, by acceleration and deceleration, by changes in
pitch level and terminal pitch contours, and as being constrained to a length
of four words in English, oftentimes coinciding with clauses, though many
others are parts of clauses (Chafe 1994: 69). This characterization suggests
a cognitive constraint on how much information can be fully active in the
mind at one time. Givón (1995: 358) calls this the ‘one new or informative
chunk-per-clause constraint’, and he argues that in many cases it is often
sufficient to mention only this chunk (Givón 1989: 209; cf. Du Bois 1987:
819). It is no less true, however, that we often cram more than one clause or
concept into intonational frames, or stretch out a single clause into more
than one frame.
5. Unless otherwise stated, data are taken from the Lancaster IBM Spoken
English corpus (LIBMSEC). For further details on this corpus see Taylor
and Knowles (1988) and Gómez-González (2001: 192–206).
6. Langacker (2001a) proposes the following list of natural paths, or cogni-
tively natural ordering of the elements of a complex structure:
- speech time (word order, expression order in general)

- locational path of access (nested or chained)
- temporal sequence of events
- successive points along any scale
- chain of reference point of relationships
- empathy (speaker > hearer > other; human > animate > inanimate;
etc.)
- chain of cause-effect relationships
- clausal organization (main clause > subordinate clause)
- viewpoint chain
- partonomy (hierarchy of whole-part relations)
- grammatical relations (subject > object > other)
He contends that, ceteris paribus, there is a tendency for natural paths to

coalign, and for the sequences in which this applies to be more basic and
easier to process (i.e. unmarked), although there are also cases in which this
tendency is overridden by other factors.
7. Here ‘markedness’ refers to probabilities of choice within a linguistic sys-
tem. Linguistic choices normally skew, in that one or some of their terms
are marked or less probable, in contrast with unmarked, or more prototypi-
cal or recurrent items. Further, with Kies (1988: 74), we shall take the view
that marked options have: (a) comparatively lower frequency of occurrence,
(b) comparatively higher structural complexity; and (c) more restricted dis-
tribution in definable environments (Givón’s 1993: 178 condition of
‘discourse distribution’). However, it should be borne in mind that marked
patterns are not always less frequent – less expected from the addressee’s
point of view – than their unmarked counterparts, for they are affected by
such factors as text type (Siewierska 1988: 12), or by matters of position
and context (Mithun 1987: 313). A case in point is discourse-initial sen-
tences, which belong to highly marked situations and therefore give rise to
the so-called markedness reversal, i.e. to a local affinity with marked con-
structions (Fox 1987; Pu and Prideaux 1994).
8. Mackenzie and Keizer (1991: 194) and Hannay (1991: 138) consider New-
Tops as a subcategory of Focus, i.e. Presentative Focus.
9. In this connection Mackenzie and Keizer (1991: 187) argue, for example,
that not all GivTops need be contextually given, or introduced into the dis-
course by means of a NewTop, but can also be situationally given
(Situationally GivTops) or generally given (Generally GivTops): e.g. Watch
out! The ceiling is caving in!). Similarly, it seems that not all inferrable
elements are inferred from a GivTop or fulfil a SubTopic function, for many
inferrable elements may also act as NewTops. These may be new at the con-
textual level but given or inferrable with regard to the addressee`s general or
situational pragmatic information: e.g., What did you see in the circus?
Well, there was an elephant that amazed us with his tricks... .
10. Note, however, that D-Topic relationships may also hold between nominal
expressions.
11. In FG ‘focality’ encompasses “those pieces of information which are the
most important or salient with respect to the modifications which S [...]
wishes to effect in P [....], and with respect to the further development of the
discourse” (Dik 1997: 312).
12. Worthy of mention here is Lambrecht’s (1994: 97) differentiation of activa-
tion accent and Focus accent in that a point of prosodic prominence is not
necessarily an indicator of either a Focus relation or inactiveness of a refer-
ent. It may be one or the other, or both at the same time. An activation
accent expresses temporary cognitive states of discourse referents and may
fall on a constituent expressing a presupposed proposition (even if it is not
in Focus): e.g. the underlined constituents in I saw Mary and John yester-
day. She says hello, but he’s still angry at you. vs. I saw Mary yesterday.
She says hello (John was very busy that morning) [emphasis mine]).
13. If only that the label ‘NewFoc’ does not seem to be very felicitous, since
non-contrastive Focus need not fall on New information. Besides, both
kinds of Focus, NewFoc and ContrFoc, entail a certain degree of newness or
newsworthiness, albeit of different kinds. Hence both kinds could deserve
the label ‘NewFoc’.
14. We shall endorse Lambrecht’s (1988: 1) definition of ‘presupposition’ and
‘assertion’:
[…] [presupposition entails] the proposition or set of propositions
which the speaker assumes the hearer considers true (believes,
knows) and is aware of at the time of utterance and which is rele-
vant in the context of utterance. […] [By contrast, assertion is what
is] added to or superimposed on the pragmatic presupposition by
an utterance [the proposition which the hearer is expected to know
as a result of hearing a sentence].
Of importance here is to bear in mind that ‘givenness’ and ‘presupposition’

evoke two different concepts. Broadly speaking, givenness is a discourse
notion referring to the informational status of the constituents of a message
as determined by the speaker’s view of the situational and linguistic
co(n)text and indicated by attenuated morphosyntactic and phonological
forms (e.g. pronominalization, definiteness, lower pitch, weaker stress, etc.).
Conversely, presupposition is a logico-semantic notion realized by sentence
form which involves a ‘proposition’ (i.e. a potential sentence having the ca-
pacity of being true or false), whose assumability is required for the success
of the message (Jackendoff 1972: 276–278). Accordingly, a proposition
may convey Given information, but need not be presupposed (e.g. It can't
be true, as an answer to I saw the man, where it stands as given information,
but the proposition is not presupposed). Or, vice versa, informationally new
items may occur within a presupposition (e.g. What the duke gave to my
aunt was that teapot, where it is presupposed that the duke gave something
to someone (my aunt), but the identity of that something is presented as
news). An ‘open proposition’, in turn, refers here to a kind of proposition
“which contains one or more variables, and it represents what is assumed by
the speaker to be salient [in the presupposition] in the discourse [...]. The
variable of the open proposition is instantiated with the focus of the utter-
ance” (Ward 1985: 5).
15. For alternative demarcations of the Theme-Rheme partitions see e.g. Halli-
day (1994), Berry (1996), and Dik (1997).
16. Subject has special status as a point of attachment to what has already been
constructed and a point of access to what is currently being constructed
(Chafe 1994), hence its tendency to be given rather than new in discourse.
This is consonant with Langacker’s (1998, 1999) view that Subject and Ob-
ject are the first and second conceptual reference points accessed in building
up to the full conception of a profiled relationship: Subject is a normally ini-
tial participant invoked for purposes of arriving at the conceptualization of a
profiled relationship, while an Object is a target evoked in the context of
conceptualizing a relationship in which the Subject participates. Langacker
concludes that, since the Subject anchors the conception of a relationship
encompassing the Object, both tend to coalign with Given vs. New informa-
tion respectively.
17. Observe the awkwardness of the ‘zooming out’ strategies below ⎯ note
that there here is not to be confused with existential there:
i. and my packet of biscuits was on the table, where my

newspaper had been, there.
ii. and on the table, where my newspaper had been, there
was my packet of biscuits.
18. The marked ‘heavy’ Themes move an informational Focus to the front
(ContrFoc, in LIBMSEC normally uttered on a high level), and at the same
time, the Subject, which would not be in focus in CWO counterparts, also
acquires focal status (in LIBMSEC usually with Falling tones) and is
thereby put ‘on stage’.
19. In LIBMSEC the rhematic area of there-constructions was found to be lexi-
cally and informatively four times heavier than the thematic part (8 vs 2).
20. According to the results reported for LIBMSEC, most EMTs are obtained
by the presence of a prefield TT, thereby having a high-low or combined
orientation, that is, structural or conjunctive Themes (e.g. and, but), linking
hypotactic or paratactic clause complexes, or continuatives (e.g. em, well,
so), punctuating an exchange and staging discourse turns.
References
Anstey, Matthew
This vol. Functional Grammar from its inception.
This vol. Towards a speaker model of Functional Grammar.
Bates, Elizabeth and Judith C. Goodman
1997 On the inseparability of grammar and the lexicon: evidence from ac-
quisition, aphasia and real-time processing. Language and Cognitive
Processes 12: 507–584.
Bencini, Giulia M.L. and Adele E. Goldberg
2000 The contribution of argument structure constructions to sentence
meaning. Journal of Memory and Language 43: 640–651.
Berry, Margaret
1996 What is Theme? A(nother) Personal View. In: Margaret Berry,
Christopher Butler, Robin Fawcett and Guowen Huang (eds), Mean-
ing and Form: Systemic Functional Interpretations. Norwood, NJ:
Ablex. 1–64.
Butler, Christopher S.
2003 Structure and Function: A Guide to Three Major Structural-
Functional Theories. Amsterdam and Philadelphia: Benjamins.
Carroll, John B., Peter Davies, and Barry Richman
1971 Word Frequency Book. New York: Houghton Mifflin.
Chafe, Wallace L.
1976 Givenness, contrastiveness, definiteness, Subjects, Topics and point
of view. In: Charles Li (ed.), Subject and Topic, 26–56. New York:
Academic Press.
1987 Cognitive constraints on information flow. In: Russell S. Tomlin
(ed.), Coherence and Grounding in Discourse, 21–51. Amsterdam
1994 Discourse, Consciousness, and Time: The Flow and Displacement of
Conscious Experience in Speaking and Writing. Chicago: University
of Chicago Press.
Connolly, John H.
1998 Information, situation semantics and Functional Grammar. In Han-
nay and Bolkestein (eds), 167–189.
This vol. The question of discourse representation in Functional Discourse
Grammar.
Connolly, John H., Roel Vismans, Christopher Butler, and Richard Gatward (eds)
Cornish, Francis
This vol. Focus of attention in discourse.
Dik, Simon C.
1978 Functional Grammar. Amsterdam: North Holland.
1997 The Theory of Functional Grammar. Part 1: The Structure of the
Du Bois, John W.
1987 The discourse basis of ergativity. Language 63: 805–855.
Fortescue, Michael
This vol. The complementarity of the process and pattern interpretations of
Functional Grammar.
Fox, Barbara A.
1987 Morphosyntactic markedness and discourse structure. Journal of
Pragmatics 11: 359–373.
Givón, Talmy
1989 Mind, Code and Context: Essays in Pragmatics. Hillsdale NJ: Erl-
baum.
1993 English Grammar. (Vol. 2) Amsterdam and Philadelphia: Benja-
mins.
1995 Functionalism and Grammar. Amsterdam and Philadelphia: Benja-
mins.
Goldberg, Adele E.
1995 Constructions: A Construction Grammar Approach to Argument
Structure. Chicago and London: The University of Chicago Press.
forthc. Argument realization: The role of constructions, lexical semantics

and discourse factors. In: Mirjam Fried and Jan-Ola Östman (eds),
Construction Grammar(s): Cognitive and Cross-Language Dimen-
sions. Amsterdam and Philadelphia: Benjamins.
Goldberg, Adele E., Nitya Sethuraman and Devin Casenhiser
forthc. Learning argument structure generalizations. Ms. University of Illi-
nois and University of California at San Diego.
Gómez-González, María de los Ángeles
1998a A corpus-based analysis of extended multiple Themes in Present-day
English. International Journal of Corpus Linguistics 3: 81–114.
1998b Aspects of Topic and topicality in Functional Grammar. Working
Papers in Functional Grammar 65. Amsterdam: University of Am-
sterdam.
2001 The Theme-Topic Interface: Evidence from English. (Pragmatics and
Beyond New Series 71.) Amsterdam and Philadelphia: Benjamins.
Grice, H. Paul
1975 Logic and conversation. In: Peter Cole and Jerrold L. Morgan (eds),
Studies in Syntax 3: Speech Acts. 41–58. New York: Seminar Press.
nold.
Hannay, Mike
1991 Pragmatic function assignment and word order variations in a Func-
1994 The Theme zone. In: Ronny Boogaart and Jan Noordegraaf (eds),
Nauwe Betrekkingen. 107–117. Amsterdam: Stichting Neerlandistiek
VU and Münster: Nodus.
1998 Message modes in a modular FG. Paper delivered to the 8th Interna-
tional Conference on Functional Grammar. Amsterdam: Vrije
Universiteit. 6–11 July 1998.
delphia: Benjamins.
Harder, Peter
English. (Trends in Linguistics. Studies and Monographs 87). Berlin
This vol. Comment clauses, Functional Discourse Grammar and the grammar-
discourse interface.
Hengeveld, Kees
1997 Cohesion in Functional Grammar. In John H. Connolly Roel M.
Vismans, Christopher S. Butler and Richard A. Gatward (eds), 1–16.

Inchaurralde, Carlos
This vol. Behind the scenes: Cognition and Functional Discourse Grammar.
Jackendoff, Ray
1972 Semantic Interpretation in Generative Grammar. Cambridge MA
and London: MIT Press.
1987 Consciousness and the Computational Mind. Cambridge MA: MIT
Press.
1990 Semantic Structures. Cambridge MA: MIT Press.
Kay, Paul and Charles J. Fillmore
forthc. Construction Grammar. Stanford: CSLI Publications.
Kemmer, Suzanne
1995 Emphatic and reflexive –self: Expectations, viewpoint and subjectiv-
ity. In Dieter Stein and Susan Wright (eds), Subjectivity and Sub-
jectivisation, 55–82. Cambridge: Cambridge University Press.
Kies, Daniel
1988 Marked Themes with and without pronominal reinforcement, their
meaning and distribution in discourse. In: Erich M. Steiner and
Robert Veltman (eds), Pragmatics, Discourse and Text, 47–75. Lon-
don: Pinter Publishers.
Kroon, Caroline
Richard A. Gatward, 17–32.
Ladd, D. Robert
1996 Intonational Phonology. Cambridge: Cambridge University Press.
Lambrecht, Knud
1994 Information Structure and Sentence Form: Topic, Focus and the
Mental Representations of Discourse Referents. Cambridge: Cam-
Lambrecht, Knud and Maria Polinsky
1997 Typological variation in sentence-Focus constructions. Chicago Lin-
guistic Society 33. 1–20.
Langacker, Ronald W.
1987 Foundations of Cognitive Grammar. Stanford: Stanford University
Press.
1998 Topic, Subject, and possessor. Linguistic Notes from La Jolla 19: 1–28.
1999 Assessing the cognitive linguistic enterprise. In: Theo Janssen and
Gisela Redeker (eds), Cognitive Linguistics: Foundations, Scope,
and Methodology, 13–59. (Cognitive Linguistics Research 15.) Ber-
lin and New York: Mouton de Gruyter.
2000 A dynamic usage-based model. In: Michael Barlow and Suzanne

Kemmer (eds), Usage-Based Models of Language, 1–63. Stanford:
CSLI Publications.
2001a Dynamicity in grammar. Axiomathes 12. 7–33.
2001b Discourse in cognitive grammar. Cognitive Linguistics 12. 143–188.
1989 Speaking: From Intention to Articulation. Cambridge MA: MIT
Press.
2000 First things first: towards an Incremental Functional Grammar. Acta
This vol. Functional Discourse Grammar and language production.
Mackenzie, J. Lachlan and M. E. Keizer
215.
MacWhinney, Brian
1977 Starting points. Language 53. 152–168.
Mithun, Marianne
1987 Is basic word order universal? In: Russell Tomlin (ed.), 281–328.
Moutaouakil, Ahmed
This vol. Discourse structure, the Generalized Parallelism Hypothesis and the
architecture of Functional Grammar.
Noonan, Michael
1999 Non-structuralist syntax. In: Michael Darnell, Edith Moravcsik, Fritz
Newmeyer, Michael Noonan and Kathleen Wheatley (eds), Func-
tionalism and Formalism in Linguistics, 11–31. Amsterdam and
Nuyts, Jan
This vol. Remarks on layering in a cognitive-functional language production
model.
Pu, Ming-Ming, and Prideaux, Gary D.
1994 Coding episode boundaries with marked structures: A cross-
linguistic study. Canadian Journal of Linguistics 39: 283–296.
Selkirk, Elisabeth O.
1984 Phonology and Syntax: The Relationship between Sound and Struc-
ture. Cambridge, MA: MIT Press.
Siewierska, Anna
1988 Word Order Rules. London: Croom Helm.
1991 Functional Grammar. London: Routledge.
Smits, Aletta
2002 How writers begin their sentences: Complex beginnings in native
and learner English. Doctoral dissertation, University of Amsterdam.
Steuten, Ans A. G.
1998a A Contribution to the Linguistic Analysis of Business Conversations
within the Language/Action Perspective. Doctoral dissertation, Uni-
versity of Delft.
1998b Structure and coherence in business conversations. In: Mike Hannay
Taylor, Lita J., and Gerald Knowles
1988 Manual of Information to Accompany the LIBMSEC Corpus: The
Machine-Readable Corpus of Spoken English. Unit for Computer
Research on the English Language Bowland College, University of
Lancaster. Bailrigg: Lancaster.
Tomlin, Russell (ed.)
1987 Coherence and Grounding in Discourse. Amsterdam and Philadel-
phia: Benjamins.
Vallduví, Enric, and Elizabeth Engdahl
1996 The linguistic realization of information packaging. Linguistics 34:
459–519.
Van der Auwera, Johan, and Louis Goossens (eds)
1987 Ins and Outs of the Predication. Dordrecht: Foris.
This vol. The problem of subjective modality in the Functional Grammar
model.
Vet, Co
1998 The multilayered structure of the utterance: About illocution, modal-
ity and discourse moves. In Mike Hannay and A. Machtelt Bolke-
Ward, Gregory Louis
1985 The Semantics and Pragmatics of Preposing. Michigan: University
Microfilms International.
The problem of subjective modality in the
Functional Grammar model
Jean-Christophe Verstraete
1. Introduction
The purpose of this chapter is twofold.1 I will first propose a number of
modifications to the analysis of subjective modality in the Functional
Grammar model (as developed in Hengeveld 1987, 1988, 1989, Dik 1997),
using the English modal auxiliaries as a test case. I will then confront the
alternative view of subjective modality resulting from these modifications
with the architecture for FG proposed in Hengeveld (this volume). I will
show how the proposed modifications can be dealt with more easily in the
modular, top-down architecture than in the traditional FG model, but I will
also argue that some further change is required in the model, more particu-
larly concerning the optionality of layers at the representational level.
1.1. The subjective-objective distinction

The first part of this chapter will be devoted to a critical examination of the
distinction between subjective and objective modality in the traditional FG
model. I will first propose a number of modifications to the criteria used to
support the subjective-objective distinction. It is often argued that objective
modality can be questioned and hypothesized, whereas subjective modality
cannot. I will show that this is not entirely correct: conditionality and inter-
rogation are not excluded for subjective modality, but interact with it in a
special way. I will argue that this interaction can be explained as a conse-
quence of the fundamentally interpersonal nature of subjective modality.
On the basis of these modifications, I will then re-assess the position of
deontic modality with respect to the subjective-objective distinction. In the
FG model, as in many other frameworks, deontic modality is excluded
from the subjective category. On the basis of the functions of deontic
244 Jean-Christophe Verstraete
modals and their behaviour in reaction to the criteria of conditionality and

interrogation, I will show that this exclusion is not justified, and that
deontic modality can be subjective just like epistemic modality.
In the final section, I will argue that what sets deontic modals apart
from their epistemic counterparts is not their potential for subjective status,
but rather the type of domain over which they operate: subjective epistemic
modality deals with the truth of propositions, whereas subjective deontic
modality deals with the desirability of actions. This functional distinction is
also grammatically reflected in the terms of the feature of tense: subjective
epistemic modality operates over tensed SoAs, whereas subjective deontic
modality operates over tenseless SoAs.
1.2. Implications for the FG model(s)

In the second part of this chapter, I will confront this alternative analysis of
subjective modality both with the traditional FG model in Hengeveld
(1989) and Dik (1997) and with the alternative architecture for FG pro-
posed by Hengeveld (this volume), showing how the alternative analysis of
subjective modality proposed in this chapter can be accommodated more
easily in that alternative architecture. The modular separation between the
interpersonal and the representational components allows for a subjective
analysis of deontic modality in parallel with epistemic modality, independ-
ently of the fact that subjective epistemic and subjective deontic modality
are associated with different types of domains. This difference in domain
between epistemic and deontic modality, on the other hand, can be dealt
with in terms of the top-down organization of the model, as an example of
how choices within the interpersonal component have repercussions on
choices within the representational component.
To round off, I will also discuss one problem that is not dealt with in
Hengeveld’s (this volume) proposed model. In spite of the modular separa-
tion between the interpersonal and the representational components, the
model still implicitly assumes that the full set of representational layers is
relevant for every main clause. In order to adequately model subjective de-
ontic modality, however, it is necessary to recognize optionality of layers
in the representational component.
The problem of subjective modality in the FG model 245
2. The analysis of modality in FG

I will first briefly outline the most important features of the analysis of
modality in FG, as it has been developed in the work of Hengeveld (1987,
1988, 1989) and Dik (1997).
2.1. Three categories

The basis of the FG analysis of modality is a distinction between three dif-
ferent categories, which is itself based on a distinction between different
functional layers in the clause with which the modal categories can be as-
sociated. More particularly, Hengeveld (1988, 1989) and Dik (1997)
distinguish between the categories of subjective, objective and inherent
modality.
Inherent modality includes “all those linguistic means through which S
can characterize the relation between a participant in an SoA and the poten-
tial actualization of that SoA” (Hengeveld 1988: 233). In terms of the
traditional tripartition of epistemic, deontic, and dynamic modality (see, for
instance, Palmer 1990), FG’s inherent category includes the dynamic mo-
dals of ability and volition – the traditional participant-oriented modals – as
well as those uses of deontic modality “in which it is reported that some
participant in a state of affairs is under the obligation or has received per-
mission to perform in that state of affairs” (Hengeveld 1988: 234). These
types are exemplified respectively in examples (1), (2) and (3), in which
the modal expressions in question have been underlined:
DYNAMIC, ABILITY
(1) Well, certainly marriage has changed him. It mellowed him to the extent
that he lost his drive and his motivation. In the midst of all that, he thought
2
to himself, “Hey, wait a minute. I can still play this game.” (CB)
DYNAMIC, VOLITION
(2) They were servicing AWACs planes for the Royal Saudi Air Force and com-
pleting contract work on a controversial air defense network for Saudi
Arabia known as the Peace Shield. Boeing won't say how many of its em-
ployees are still there, but says those who stayed did so voluntarily. (CB)
DEONTIC
(3) But Ramadan means more than just physical deprivation. It has spiritual
and moral obligations, too. Followers must refrain from bad thoughts,
words and actions, perform special acts of charity and spend even more
time than usual in worship. (CB)
Objective modality includes “all those linguistic means through which S

can evaluate the actuality of an SoA in terms of his knowledge of possible
SoAs” (Hengeveld 1988: 233). The relevant knowledge here can be either
epistemic “knowledge of possible situations obtaining in S’s conception of
reality or of a hypothesized situation” (Hengeveld 1988: 234) or deontic
“knowledge of possible situations relative to some system of moral, legal
or social conventions” (Hengeveld 1988: 234). Accordingly, the objective
category includes both epistemic and deontic modality, exemplified in (4)
and (5):
EPISTEMIC
(4) Is it possible that all human beings on earth today are descended from a
single woman? (CB)
DEONTIC
(5) The bandages they wrapped around his face were loose enough to allow him
to breathe, yet tight enough not to slip. “You must not move a muscle,”
Ram Das warned him. “That would cause a riot of terror. We’re going to
cover you with saffron dust to show we’re carrying a corpse to the holy
river.” (CB)
Subjective modality, finally, involves “all those linguistic means

through which S can express his commitment with regard to the truth of a
proposition” (Hengeveld 1988: 233). This applies only to epistemic modal-
ity, exemplified in (6):
EPISTEMIC
(6) Michael Jackson must have enjoyed his wedding in May, because he decid-
ed to have another one at his California ranch. (CB)
2.2. Theoretical motivation

Within the FG framework, the distinction between subjective, objective
and inherent modality is theoretically motivated in terms of an association
with different functional layers in the structure of the clause (Hengeveld
1989, compare Foley and Van Valin 1984, partly also Lyons 1977).3
The starting point for the FG account of the different categories of mo-
dality is the distinction between a propositional and a predicational layer in
the structure of the clause. According to Hengeveld (1989), this is
grounded in the distinction between a representational and an interpersonal
function in the clause (Halliday 1994), which take care respectively of the
conceptualization of the situation referred to in the clause and the interac-
tive intentions of the speaker towards the interlocutor. In this sense, the
predicational and propositional layers represent two different functional
perspectives on the traditional notion of State of Affairs (SoA): the predi-
cation is the representational perspective on the SoA, as a description of
the situation to which the speaker refers in the clause, whereas the proposi-
tion is the interpersonal perspective on the SoA, as the propositional
content for a particular communicative act performed by the speaker in the
clause.
In terms of this distinction between interpersonal and representational
functions of the SoA, a first distinction in the domain of modality is made
between subjective modality on the one hand and objective and inherent
modality on the other hand. Subjective modality serves to express the
speaker’s commitment, and is therefore an interpersonal category, taking
care of the speaker’s interactive positioning with respect to the SoA as the
object of his speech act. In the FG model, this is accommodated theoreti-
cally by analyzing subjective modality as an operator of the propositional
layer. Objective and inherent modality, on the other hand, do not have any
interpersonal function, and are therefore associated with the predicational
layer.
Within the non-interpersonal domain, a further distinction is made be-
tween objective and inherent modality on the basis of their divergent
positions with respect to the SoA. Inherent modals serve to evaluate the ac-
tualization of the SoA from a perspective internal to the SoA, more
particularly by indicating a particular type of relation (for instance ability
or willingness) between one of the participants and the SoA, whereas ob-
jective modals evaluate it from an external perspective. In the FG model,
this is accommodated theoretically in terms of different positions with re-
spect to the predicational layer: the SoA-external function of objective
modals is reflected in their position as operators with scope over the predi-
cation, whereas the SoA-internal function of inherent modals is reflected in
their position as categories internal to the predication.
2.3. Criteria
Hengeveld (1988) argues that the functional distinction between subjective
and objective modality is also reflected in their grammatical behaviour, and
adduces a number of grammatical criteria to which subjective and objective
types of modality react differently. In addition to tense, polarity and prag-
matic appropriateness of questioning the source of information, which will

not be dealt with in this chapter, Hengeveld also mentions conditionality
and interrogation: “Objectively modalized predications can be questioned,
subjectively modalized ones cannot”; “Objectively modalized predications
can be hypothesized in a conditional sentence, subjectively modalized ones
cannot” (Hengeveld 1988: 236).
2.4. Summary
Table 1. The FG analysis of modality
FUNCTION LAYERING CATEGORY MEMBERS CRITERIA
Interpersonal Prop. operator Subjective Epistemic -{interr,

cond}
Representational Pred. operator Objective Epistemic, +{interr,

Deontic cond}
Representational Pred.-internal Inherent Deontic, +{interr,

Dynamic cond}
3. Criteria: refinement and explanation

In order to demonstrate the need for an alternative analysis of the subjec-
tive-objective distinction, I will first show that the criteria used to
distinguish between subjective and objective modality, especially condi-
tionality and interrogation, are not entirely accurate as they are formulated
in Hengeveld (1988). Subjective modals are not excluded from conditional
and interrogative constructions: rather, using subjective modals in these
constructions leads to modifications in their interpretation that do not occur
with objective modals.
These descriptive refinements, however, do not imply a rejection of
conditionality and interrogation as criteria for the subjective-objective dis-
tinction. The semantic modifications observed for subjective modals in
these constructions provide an important key to their fundamentally inter-
personal status: the semantic effects of interrogation and conditionality can
be explained in terms of the interpersonal function of subjective modality
described in Hengeveld (1988: 233), viz. the fact that they serve to encode
positions of commitment with respect to the SoA described in the utter-

ance.
3.1. Conditionality
Hengeveld (1988: 236) proposes to use conditionality as an absolute crite-
rion for the subjective-objective distinction, and argues that objective
modality can occur in the protasis of conditional constructions, whereas
subjective modality cannot.
This absolute interpretation of the conditionality criterion may work for
the distinction between modal adjectives and adverbs,4 but it is not an ade-
quate account of the behaviour of modal auxiliaries in conditional contexts:
subjective modals can occur in conditionals, but their interpretation is af-
fected by the conditional context. Subjective modals in the protasis of
conditional constructions invariably become echoic (see also Palmer 1990:
182): they do not express the current speaker's opinion, but echo an opinion
that has been voiced in or is implied by the preceding discourse. For in-
stance,
(7) In distilling a statement of theme from a rich and complicated story, we

have, of course, no more encompassed the whole story than a paleontologist
taking a plaster mold of a petrified footprint has captured a living bronto-
saurus. A writer (other than a fabulist) does not usually set out with theme
in hand, determined to make every detail in the story work to demonstrate
it. Well then, the skeptical reader may ask, if only some stories have themes,
if those themes may be hard to sum up, and if readers will probably dis-
agree in their summations, why bother to state themes? (CB)
In (7), the epistemic position expressed by may is crucially not the posi-
tion of the ‘sceptical reader’ who uses the conditional construction, but an
opinion voiced by some other speaker which the ‘sceptical reader’ echoes
in his conditional construction without committing himself to it. Thus, sub-
jective modals are not excluded in conditional contexts, but their
interpretation is affected by the conditional.
This semantic effect does not occur with non-subjective modals like the
modal of ability in (8): the interpretation of can’t in the conditional in (8) is
not different from its interpretation in non-conditional contexts.
(8) Phillips: Oh, I think it’s just the–the same thing as the sign that they had
during the campaign, that everything else was secondary to the economy. If
Bill Clinton can’t deal with the economy in the next year or year and a half,
he’s really going to be in trouble in–in the whole breadth of his administra-
tion. (CB)
3.2. Interrogation
Hengeveld (1988) argues that objective modality can be questioned,
whereas subjective modality cannot. Again, I agree that this may be the
case when subjective and objective modality are realized as modal adverbs
and adjectives respectively,5 but it is certainly not the case for modal auxil-
iaries. Subjective modal auxiliaries can occur in interrogatives just like
objective ones, as shown in examples (9) and (10):
(9) THOMPSON: If the trials are successful, might this be used as the first line
of treatment of cancer? THORPE: Well that’s exactly where we are going.
We are hoping that this will replace conventional chemotherapy. (CB)
(10) I mean maybe we’ve got no choice if we want to do it but I mean erm do you
think it’s possible to do anything worthwhile in history lessons or can this
work really only take place in somewhere like civics social studies sociol-
ogy politics? (CB)
As with conditionality, the difference between subjective and objective

modals is not a matter of acceptability in interrogative contexts, but of the
way in which the interpretation of the modal is affected by the interrogative
construction. With subjective modality, the commitment encoded by the
modal undergoes a shift of orientation under the influence of interrogation.
In declarative contexts, the responsibility for the position is assigned to the
speaker, but in interrogative contexts like (9) or (10), the responsibility for
the position encoded by subjective modals like epistemic might and can is
no longer taken by the speaker, but is assigned to the interlocutor in the
next turn: “Are you committed to this position?” rather than “I am commit-
ted to this position”.
For non-subjective modals, on the other hand, interrogation does not af-
fect the interpretation of the modal. Since the interpretation of non-
subjective modality does not require any assignment of responsibility, it
does not undergo any shift of orientation under the influence of interroga-
tion. For instance, if we look at examples like (11) or (12):
(11) They’ve already worried the charts and won many admirers with the single
Fun for Me. Can they do it live? It’s time to find out. (CB)
(12) First-time buyers should ask: Who is responsible for a title search and
abstract? Who will provide title insurance? Must the termite inspection be
paid before closing (in which case you will need a receipt), or can it be paid
at settlement? (CB)
Unlike the subjective modals in (9) and (10), the non-subjective uses of
can and must in (11) and (12) do not undergo any shift of orientation in in-
terrogative contexts. They do not imply any responsibility or a position of
commitment, and therefore they are not affected by interrogation. Rather
than taking part in the speaker-interlocutor transfer of responsibility, dy-
namic can and deontic must in (11) and (12) belong to the content with
respect to which positions of responsibility are taken or transferred. With
must in (12), for instance, the obligation or necessity denoted by must is
part of the content about which speaker and interlocutor negotiate epis-
temically (“Do you think it is obligatory or necessary to pay the termite
inspection before closing?”; see further in Section 5.2.). The same goes for
can in (11): the ability denoted by the modal is part of the content that is
subject to epistemic negotiation (“Do you think they are able to do the
same thing live?”).
3.3. Explanation of the behavioural distinctions

The descriptive modifications proposed in the previous two sections do not
invalidate the basic idea that conditionality and interrogation can be used
as criteria for the subjective-objective distinction. The behaviour of subjec-
tive and objective modality in reaction to conditionality and interrogation is
indeed different, only not in terms of unacceptability but in terms of the ef-
fect on the interpretation of the modal. In this section, I will show how
these semantic effects do not merely serve as criteria to differentiate be-
tween subjective and objective uses of modal auxiliaries, but that the
difference in behaviour can actually be linked back to the functional basis
of the subjective-objective distinction proposed in Hengeveld (1989). The
effects of echo interpretation for conditionals and shift of orientation for
interrogatives can be explained as a consequence of the fact that subjective
modals take part in the interpersonal organization of the clause: they serve
to encode positions of commitment with respect to the SoA described in
the utterance.
The fact that subjective modality becomes echoic in the protasis of con-
ditional constructions is a consequence of the clash between the function of
the subjective modal, which is to encode a position of commitment by the
speaker, and the function of the conditional marker, which is to suspend the
speaker’s commitment with respect to the proposition in its scope. Condi-
tionals are special contexts in that they operate with suppositions rather
than assertions (Declerck and Reed forthcoming): the typical function6 of
the conditional marker if is to suspend the commitment of the speaker with
respect to the proposition in its scope (Dancygier 1998) and thus to treat it
as a mere supposition in the context of the apodosis. For objective modals,
this suspension of commitment is unproblematic, because objective modal-
ity does not encode any type of speaker commitment. For subjective
modals, however, the suspension of speaker commitment in conditionals
clashes with the very function of subjective modality, which is to encode
the speaker’s commitment with respect to the proposition. This clash leads
to an echoic reinterpretation, which could be regarded as the interpreta-
tional compromise between the two conflicting functions in the modal and
the conditional: the position encoded by the modal is no longer a position
of the current speaker, because the conditional context suspends such posi-
tioning, but resumes a position taken in the preceding discourse, usually by
another speaker. That is the only way a modal can still encode a position in
a context which does not allow such positioning by the current speaker.
The fact that subjective modality undergoes a shift in orientation in an
interrogative context is again a consequence of its position-encoding func-
tion. A position of commitment necessarily implies responsibility for that
position, and that is what the declarative-interrogative contrast operates on.
The declarative allows the speaker to take the responsibility for the posi-
tion encoded in the subjective modal in his own turn, whereas the
interrogative allows the speaker to transfer this responsibility to the inter-
locutor in the next turn (Davies 1979). Unlike subjective modals, objective
modals do not serve to encode any positions of commitment, and therefore
do not undergo any shift of orientation under the influence of interrogation.
Interrogation can only interact with modality if the function of the modal
implies assignment of responsibility in discourse.
4. The position of deontic modality

On the basis of the modified criteria, the position of deontic modality in
the overall model for modality in FG can now be reconsidered. In
Hengeveld (1988), deontic modality is included in the objective and in-
herent categories, but crucially not in the subjective category. FG is not
the only framework to treat deontic modality differently from epistemic
modality: deontic modality has often been set apart in the analysis of the
subjective-objective distinction. Foley and Van Valin (1984), for instance,
do not recognize any subjective-objective ambiguity for deontic modality

at all, and treat the deontic modals as ‘core’ operators (more or less the
equivalent of FG’s predicational operators) together with the dynamic
modals. Even Lyons (1977), who seems to recognize both subjective and
objective types for deontic modality, still differentiates his analysis of sub-
jectivity by associating subjectivity with different functional layers (in
terms of Hare’s 1970 tropic-neustic-phrastic framework) depending on the
epistemic or deontic nature of the modal. Halliday (1970) is in fact the
only one who recognizes a subjective function for deontic modality that is
fully parallel with the subjective function of epistemic modality. In this
section, I will try to defend this position both on the basis of the function
of deontic modals (Section 4.1) and on the basis of their behaviour in reac-
tion to some of the criteria for subjectivity (Section 4.2). In the next
section, I will show that the difference between epistemic and deontic mo-
dality does not lie in their potential for subjective status, but in the domain
over which they operate.
4.1. The functions of deontic modals: inherent, objective and

subjective
4.1.1. Types of deontic modality
In terms of function, there are at least three types of deontic modality that
can be distinguished in English, illustrated in (13) – (16) below:
(13) “I need to see Izzy,” I said. “I told you, she's sound asleep. Deeply
asleep.” “May I see for myself?” “You may not, you shit-sucking liar! You
cheat. You coward. You sit!” I pulled back the heavy dining room chair and
sat before the typewriter. (CB)
(14) What we want is for the right honourable gentleman to use the full weight of
his office. We are getting tired of a cosmetic approach an oversanguine ap-
proach. There is a crisis and he must act now <ICE-GB:S1B-056 #87-
89:1:F>
(15) But Ramadan means more than just physical deprivation. It has spiritual
and moral obligations, too. Followers must refrain from bad thoughts,
words and actions, perform special acts of charity and spend even more
time than usual in worship. (CB)
(16) But to reach orbit an object must accelerate to a speed of about 17,500
miles per hour (28,000 kilometres per hour, called satellite speed or orbital
velocity) in a horizontal direction; and it must reach an altitude of more
than 100 miles (160 kilometres), in order to be clear of the atmos-

phere.<ICE-GB:W2B-035 #39:1>
In examples (13) and (14), the function of the deontic modal is to ex-
press the speaker’s commitment to the permission or obligation encoded in
the modal: may and must in these examples can be paraphrased as “I do not
allow you to …” and “I oblige him to …”. An example like (15), on the
other hand, does not express the speaker’s commitment to the obligation
signalled by the modal (“I want the followers to …”), but rather reports on
the existence of a particular obligation without necessarily committing the
speaker to it. Unlike in (13) and (14), the speaker is not the deontic source
of the obligation or permission, but merely the one who describes the exis-
tence of such an obligation, which may itself originate from another
source. An example like (16), finally, is different from both (13) and (14)
and from (15) in that no deontic source is involved at all. Must in (16) does
not express an obligation originating from a deontic source (irrespective of
the question whether this source is identical with the speaker or not), but
denotes a necessity that is inherent in the situation:7 the state of things is
such that if the object in question is to reach orbit, a speed of 17,500 mph is
required.
The FG analysis of modality also recognizes functional ambiguity for
deontic modals, and accounts for it in terms of the distinction between ob-
jective and inherent modality. Examples like (13) and (14) would be
classified as objective, i.e. expressing an evaluation of the actuality of the
SoA on the basis of the speaker’s “knowledge of possible situations rela-
tive to some system of moral, legal or social conventions” (Hengeveld
1988: 234). Examples like (15), on the other hand, would be classified as
inherent, i.e. “report[ing] that some participant in a state of affairs is under
the obligation or has received permission to perform in that state of affairs”
(Hengeveld 1988: 234).
4.1.2. An alternative to the FG categorization of the types
I do not believe that the objective-inherent dichotomy is an adequate way

to account for the functional distinction between (13) and (14) on the one
hand and (15) on the other hand. In this section, I will argue that in terms
of function it is more plausible to analyse (13) and (14) as subjective in
parallel with epistemic modality rather than as objective. Examples like
(15), on the other hand, should be analysed as the objective counterparts of
such subjective uses rather than grouping them together with the dynamic
modals of ability and volition in the inherent category. Examples like (16),
finally, which are not explicitly dealt with in the FG framework,8 are the
most plausible ‘deontic’ candidates for the inherent category, because of
their functional similarity with the dynamic modals of ability and volition.
The FG analysis of examples like (13) and (14) as representatives of the
objective category implies that the function of the modal here is to provide
an evaluation of the actuality of the SoA in terms of the speaker’s knowl-
edge of conventions or morals. I do not think that this is an adequate
description of the function of may and must in (13) and (14), as also argued
by Goossens (1996: 49–50): in these examples, the speaker does not evalu-
ate the actuality of the SoA, but expresses his commitment to the
permission or obligation. It is the speaker who wants the SoA to be actual-
ized: the speaker is not evaluating descriptively but acting interpersonally.
Functionally, therefore, this use comes closer to the category of subjective
modality, which “expresses [the speaker’s] commitment” (Hengeveld
1988: 233). Given the function of examples like (13) and (14), I see no rea-
son to exclude deontic modality from the subjective category: deontic
modals can be subjective just like epistemic ones, the only difference being
that the commitment in question does not concern the truth of propositions,
but the desirability of actions (see further in Section 5).
For examples like (15), the functional characterization proposed by FG
is entirely adequate. Examples like these do not involve the speaker’s
commitment to the obligation: the speaker is the one who reports on the ex-
istence of an obligation for some participant to act but does not coincide
with the deontic source of this obligation. In spite of the adequacy of the
functional characterization for such examples, however, I do not think that
they should be grouped together with dynamic modals of ability and voli-
tion as members of the inherent category, as is the case in the FG analysis.
There are two important arguments for distinguishing examples like (15)
from the modals of ability and volition. On the one hand, the obligation
expressed by must in (15) still implies some external source, whereas the
dynamic modals of ability and volition never involve any external source
but are entirely internal to the SoA. In this sense, (15) naturally groups to-
gether with (13) and (14) in contrast with the dynamic modals of ability
and volition: (13), (14) and (15) always imply some external source (which
may or may not coincide with the speaker), whereas the dynamic modals of
ability and volition do not involve any source at all. The second reason for
keeping must in (15) apart from the dynamic modals of ability and volition
is that within the deontic system there is a another use of must which does
naturally group together with the modals of ability and volition in that it
does not involve any external source. Must in (16) is similar to the dynamic
modals and different from (13), (14) and (15) because it is not related to
any deontic source: must in (16) does not express an obligation coming
from an external source but simply expresses the existence of a necessity
internal to the SoA.
I believe that we can do justice to both these arguments by redistribut-
ing examples like (15) and (16) over the objective and inherent categories:
(15) should be regarded as objective rather than as inherent, and (16)
should be included in the inherent category. Including examples like (15)
in the objective category does justice to their natural affinity with subjec-
tive modality: in both cases, the modal is related to a deontic source, and
the difference between the two relates to the question whether the deontic
source coincides with the speaker (subjective) or not (objective). On the
other hand, the ‘gap’ that is left by taking examples like (15) out of the in-
herent category is more adequately filled by examples like (16): grouping
these examples together with the dynamic modals of ability and volition
does justice to their shared SoA-internal function.9
Thus, a consideration of the various functions of deontic modality leads
to the following alternative proposal for the subjective-objective-inherent
matrix:
Table 2. The functions of deontic modality
FG Alternative
Subjective (13), (14)
Objective (13), (14) (15)
Inherent (15) (16)
4.2. Criteria applied to deontic modality

A subjective analysis of deontic modals like (13) and (14) in parallel with
epistemic modality is further justified by their behaviour in reaction to the
various criteria that can be used to distinguish between subjective and ob-
jective modality. Subjective deontic modals are just as much speaker-
hearer-related as their subjective epistemic counterparts, and therefore
show the same behaviour with respect to conditionality and interrogation.
4.2.1. Conditionality
As already argued in Section 3.1, subjective epistemic modality becomes

echoic in conditional constructions: it does not express a position taken by
the speaker who uses the conditional, but echoes one voiced by another
speaker in the preceding discourse context. With subjective deontic mo-
dality, conditionality can have the same effect – although because of the
almost systematic ambiguity with non-subjective functions this is not
automatic as it is for epistemic modals. For instance, if we compare (17)
with (18):
(17) The key stumbling block remained Republican insistence on a Medicare

premium increase. Mr Clinton argued that Medicare increases were not nec-
essary to meet demands for a balanced Budget. If America must close down
access to quality education, a clean environment and affordable health care
for our seniors in order to keep the government open, then that price is too
high,” Mr Clinton said in vetoing the temporary spending Bill. (CB)
(18) I suffer from acne but also burn very easily - can I use sun oils? It is always
best to tan gradually, but especially so if you have a sensitive skin. If you
must be exposed to sun that may burn you, it may be wise to use a sun-
screen preparation. (CB)
In (18), the non-subjective deontic modal is unproblematic in the condi-

tional: it does not echo any position taken by another speaker, but simply
expresses that if a situation occurs that requires exposure to sun, it is wise
to use a sun-screen preparation. In (17), on the other hand, the subjective
deontic modal is echoic: must in the protasis does not simply express the
existence of necessity or obligation, but echoes another speaker’s commit-
ment to such an obligation. This is evident from the fact that the
conditional in (17) is rhetorically directed against those who are committed
to Medicare premium increases. The speaker in (17) takes up the Republi-
cans’ position in his conditional protasis without committing himself to it:
the must in (17) is clearly a must that is attributed to the Republicans, and
not a position to which Clinton himself subscribes.
4.2.2. Interrogation
In Section 3.2, it was already argued that interrogation interacts with sub-
jective modality, in that the responsibility for the position encoded by the
modal is transferred to the interlocutor’s next turn in an interrogative con-

text. The same applies to subjective deontic modality, as illustrated in the
following examples:
(19) Dr Gwyn Adshead thank you very much for joining us. And please may I
call you Gwyn? Certainly. Thank you. That's a good start. What are the dif-
ferences generally speaking between a counsellor a therapist and a
psychiatrist? (CB)
(20) You’ve got to be there by nine o’clock in the morning at the latest. You’ll be
crossing the main refugee routes. Shouldn’t be too bad.” Must I leave my
platoon, sir? At this moment?” Stop arguing and get down there. It’s no-
body’s fault but yours that you speak fluent German. You know perfectly
well every linguist’s name is listed.” (CB)
Both in (19) and in (20), the deontic modals express some position of
commitment to the permission and the obligation to carry out the action in
question, but the responsibility for this commitment is not the speaker’s.
Rather, the speaker transfers this responsibility to the interlocutor’s next
turn, i.e. asks whether the interlocutor is committed to it: “Will you allow
me to call you Gwyn?” and “Are you obliging me to leave my platoon?”.
For non-subjective deontic modality, on the other hand, the modals do not
take part in the speaker-interlocutor exchange: if a modal does not express
any position of commitment, there is no responsibility to be assigned. An
example like (21) below cannot be paraphrased as “Do you want the
money to be repaid to A?” like (19) or (20), but should rather be inter-
preted as “Do you think it is obligatory/necessary to repay the money?”.
What is exchanged between speaker and interlocutor in this example is not
a position of deontic commitment, but rather a position of epistemic com-
mitment to the existence of necessity or obligation (see further in Section
5.2).
(21) [From a letter asking advice about legal matters] Can I use a portion of
the proceeds from the first sale effect improvements to the second prop-
erty? If so, must it be repaid to A later? If a contract of sale has been
signed by all parties and my aunt was to die, will that contract still be
binding? (CB)
4.3. Conclusion
If we look at the function of deontic modals in examples like (13) or (14),
and their behaviour in conditionals and interrogative contexts, there is no
reason to exclude deontic modality from the subjective category. Deontic

modals can be subjective just like epistemic modals, the only difference be-
ing the domain of the modality. Whereas epistemic modals express
positions concerning the truth of propositions, deontic modals express posi-
tions concerning the desirability of actions. Within the traditional FG
conception of layering, it is probably this distinction of domains that has
been the motivation for the exclusion of deontic modality from the subjec-
tive category, as I will argue in Section 6.
5. Subjective deontic versus subjective epistemic modality:

different domains
As I tried to show in the previous section, deontic modality can be subjec-
tive just like epistemic modality, both in terms of the function of encoding
positions of commitment, and in terms of the behaviour in reaction to the
different criteria for subjectivity. In this section, I will argue that the differ-
ence between the two categories lies in the domain over which they operate
rather than in their subjective status. Subjective epistemic modality en-
codes positions of commitment to the truth of propositions, whereas
subjective deontic modality encodes positions of commitment to the desir-
ability of actions. I will show that this functional difference is also
reflected in the grammatical behaviour of the categories, more particularly
in terms of the category of tense: subjective epistemic modality operates
over tensed domains, whereas subjective deontic modality operates over
tenseless domains.
5.1. Tensed versus tenseless domains

The distinction between tensed and tenseless domains has traditionally
been associated with the distinction between the basic mood types of in-
dicative and imperative (Bolinger 1967, 1977; Lyons 1977: 746–747;
Bolkestein 1980: 41).
(22) John is aware of the problems.

(23) Give me the money!
In indicative clauses like (22), the SoA represented is always located

with respect to the temporal zero-point (the time of encoding and/or decod-
ing by speaker and/or hearer, Declerck 1991b): the category of tense
locates the actualization of the SoA in the past, present or future relative to
the temporal zero-point. In imperative clauses like (23), on the other hand,
there is no tense marking and the SoA is consequently not located with re-
spect to the temporal zero-point. It might be objected that imperatives are
not tenseless but have an inherently future orientation, but this ‘future’ ori-
entation cannot be considered a tense the way future tense is in the
indicative mood. Future tense establishes a relation to the speaker’s here-
and-now by locating the actualization of an SoA at a point in time which is
later than the time of utterance, but the so-called ‘future’ orientation in
imperatives cannot reasonably be said to locate the actualization of the
SoA at a particular point in time later than the moment of utterance. As an
object of the speaker’s desire or will, the SoA in an imperative is not
‘located’ but rather a purely ‘virtual’ concept (Bolinger 1968, 1977).
Virtuality is very different from location in the future: location in the future
still entails some relation with the world of the speaker’s here-and-now,
which can be evaluated in terms of truth and falsity, whereas virtuality
implies no relation at all with the world of the here-and-now and is
therefore outside the realm of truth and falsity.
The same contrast between tensed and tenseless domains that distin-
guishes indicative from imperative mood also distinguishes subjective
epistemic modality from subjective deontic modality. This becomes espe-
cially clear if we look at the present-perfect contrast for the main verb
following the subjective modal auxiliary, which has a different functional
value depending on the epistemic or deontic nature of the modal. For in-
stance, if we compare (24) with (25):
(24) John must surely be aware of the problems.

John must surely have been aware of the problems.
(25) Jack must give me the money, or I'll kill him.
Jack must have given me the money (by ten), or I'll kill him.
In subjective-epistemic (24), the function of the contrast between present

and perfect main verb is to locate the actualization of the SoA differently
with respect to the temporal zero-point: perfect locates it in the past relative
to the zero-point, whereas present locates it in the non-past. Thus, subjec-
tive epistemic modality operates over a tensed domain, and is therefore
paradigmatically equivalent to the bare indicative mood. In subjective-
deontic (25), on the other hand, the contrast between present and perfect
does not serve to locate the actualization of the SoA differently relative to
the temporal zero-point. Both with present and with perfect main verbs, the
SoA is a desired and therefore still virtual SoA just as in imperative
clauses: the perfect in this case does not locate in the past, but simply indi-
cates that actualization of the virtual SoA denoted by the predication is
desired by a specific point in time (here lexically realized in by ten). Thus,
subjective deontic modality operates over tenseless domains and is there-
fore paradigmatically equivalent to the imperative mood: its SoA is not
located in time with respect to the temporal zero-point, but is simply an
unlocalized, virtual SoA.
5.2. Subjective versus non-subjective deontic modality

The paradigmatic equivalence between the imperative mood and subjective
deontic modality has sometimes been challenged on the basis of the fact
that deontic modals can have features which are normally excluded for im-
peratives, like tense marking and propositional attitude markers
(Bolkestein 1980: 36–47). In this section, I will show that these arguments
apply only to nonsubjective (objective or inherent) uses of deontic modal-
ity, but crucially not to subjective uses. More particularly, I will argue that
what is interpersonally at issue in structures with nonsubjective deontic
modals is the subjective epistemic modalization realized in the indicative
mood: this is what explains the presence of typically epistemic features
such as tense and propositional attitude marking in such structures.
(26) You must ask him if he fancies me and love him and ask him why he says
he’d phone me that often, ask him that, you must say, right yeah but don’t
tell him that I told you to ask him, yeah (CB)
(27) The 16-strong Scottish squad is free to travel south and whoop it up.
Johnston’s ex-Rangers pals - six are named in the party - are likely to take
advantage, but the proviso is they must be back at Motherwell by 2.30 pm
on Sunday to work off their excesses.
The deontic modal must in (26) above is subjective: it serves to encode the
speaker’s commitment to the desirability of the action in question, and the
SoA with respect to which the speaker expresses this commitment is not
located relative to the temporal zero-point but still ‘virtual’. This subjective
use of deontic modality can unproblematically be regarded as a paradig-
matic equivalent of the imperative, which equally expresses the speaker’s
commitment to the desirability of the action and equally operates on tense-
less SoAs. In this sense, it is not a coincidence that subjective deontic must
in (26) naturally alternates with imperative structures in one and the same
stretch of discourse.
The deontic modal must in (27) is objective: it does not express the
speaker’s commitment to the desirability of the action, but merely reports
on the existence of an obligation originating from another deontic source.
Structures with this type of deontic modality behave very differently from
structures with subjective deontic modality.
(28) The Scottish squad had to be / will have to be back at Motherwell by

2.30 pm on Sunday to work off their excesses.
(29) Fortunately, the Scottish squad must be back at Motherwell by 2.30 pm on
Sunday to work off their excesses.
(30) Must the Scottish squad be back at Motherwell by 2.30 pm on Sunday?
First of all, they are not tenseless but tensed: the issue in these structures
is no longer the speaker’s commitment to the obligation, but rather the ex-
istence of the obligation in question. This existence can be located in the
past, present or future just like any other SoA, as shown in (28). Secondly,
structures like (27) also allow the expression of propositional attitude
markers like probably, possibly, fortunately or sadly (Bolkestein 1980: 40–
42), as shown in (29). Again, the object with respect to which the attitude
is expressed is the existence of the obligation rather than any speaker’s
commitment to such obligation. Finally, objective deontic modality be-
haves differently from its subjective counterpart in reaction to
interrogation. The responsibility transferred to the interlocutor’s next turn
in (30) is not deontic (“Do you want the players to …?”) as would be the
case for subjective deontic uses like (26) but epistemic (“Is it the case that
the players are obliged to …?”).
What these features show is that structures with objective deontic mo-
dality are subjectively modalized in epistemic terms, by the indicative
mood. What is interpersonally at issue in these structures is not deontic
commitment to desirability of actions but rather epistemic commitment en-
coded by the indicative mood: the speaker’s or interlocutor’s commitment
to the truth of a proposition about the existence of an obligation (“Is it or is
it not the case that this obligation exists?”). The presence of subjective
epistemic modalization in such structures can explain the availability of
tense: as shown in the previous section, subjective epistemic modality op-
erates on tensed SoAs. The availability of propositional attitude markers is
also a typically epistemic feature: propositional attitudes can only be ex-
pressed in epistemic utterances, where the truth of a proposition is at stake,
but not in subjective deontic utterances where the desirability of an action
is at stake (Verstraete 2000, see also Moutaouakil 1996: 212–213). The be-
haviour of interrogation, finally, can also be linked up with the subjective

epistemic modalization encoded in the indicative. What the interrogative
transfers to the interlocutor in (30) is the responsibility for an epistemic
rather than a deontic position, typically paraphrased with an epistemic
predicate (“Do you think it is the case that the players are obliged to …?”
rather than “Do you want the players to …?”).
In this sense, the objections raised in the literature against the paradig-
matic equivalence between deontic modality and the imperative mood are
valid only for nonsubjective deontic modality: structures with nonsubjec-
tive deontic modals are epistemically modalized by the indicative mood,
and can therefore not be regarded as paradigmatically equivalent to impera-
tives. But this argument crucially applies only to nonsubjective deontic
modality. Subjective deontic modality does not show any of these features:
it does operate on tenseless SoAs, does not allow expression of proposi-
tional attitude markers10 and the speaker-interlocutor transfer effected by
the interrogative is deontic rather than epistemic. In this sense, the argu-
ments against paradigmatic equivalence with the imperative cannot be used
for the deontic category as a whole, but should also take into account the
functional diversity within the deontic category.
5.3. Summary of the model

Table 3 below summarizes the most important differences between the
traditional FG treatment of modality and the alternative proposal put
forward in this chapter.
The central point of divergence between the two proposals is the possi-
bility of subjective status for deontic modality, but this also has
repercussions for the overall organization of the model. At the level of
speaker-hearer interaction, this implies that there is a functional dichotomy
between utterances where speaker and interlocutor negotiate about the de-
sirability of actions, and those where speaker and interlocutor negotiate
about the truth of propositions. This bifurcation is also continued at the
representational level, in the sense that the material negotiated in the inter-
action is different depending on the subjective-deontic or subjective-
epistemic nature of the negotiation. The relevant parameter here is the
category of tense, with its function of locating the SoA with respect to the
speaker’s temporal zero-point: subjective-deontic modality operates on
bare SoAs, which are not located with respect to the temporal zero-point,
whereas subjective-epistemic modality operates on tensed SoAs, because
location with respect to the temporal zero-point is a necessary prerequisite

for arguing about truth or falsity (see Halliday 1994: 75).
Table 3. The standard FG model and the proposed alternative
Interpersonal Representational
FG model Epistemic Tense Deontic
Dynamic
Alternative Epistemic Tense Deontic
Indicative Dynamic
Deontic X
Imperative
6. Implications for the FG model(s)

In the final section of this chapter, I will confront my analysis of subjective
modality both with the traditional layered model of the clause as it is pre-
sented in Hengeveld (1989) and Dik (1997), and with the alternative model
proposed in Hengeveld (this volume). I will argue that the exclusion of de-
ontic modality from the subjective-modal category (Hengeveld 1988,
1989) is a natural consequence of the way the theory of layering has been
formulated so far: given the assumption that the full set of layers is present
in every (main) clause, the difference in domain between epistemic and de-
ontic modality naturally excludes the possibility of subjective deontic
modality. I will also show, however, that the modular top-down organiza-
tion proposed in Hengeveld (this volume) is better suited to dealing with
subjective deontic modality. On the one hand, the modular separation be-
tween the interpersonal and representational components allows one to
dissociate the question of the subjective status of epistemic and deontic
modality from the question of the type of domain over which the modals
operate: the domains of the modal operators are a representational issue,
whereas the subjective status of the operator is an interpersonal issue that
can be dealt with in its own right. On the other hand, the connection be-
tween the interpersonal distinction [subjective epistemic vs subjective
deontic] and the representational distinction [tensed SoAs vs tenseless
SoAs] can still be incorporated in the model as an example of top-down
organization, where representational distinctions are steered from the inter-
personal component.
6.1. Theory-internal problems with subjective-deontic modality

The exclusion of deontic modality from the subjective category in earlier
FG proposals can probably be attributed to the understanding of layering
that is implicit in the model presented in Hengeveld (1989) and Dik (1997).
Given the difference in domain between epistemic modality (the truth of
propositions) and deontic modality (the desirability of actions), it is a logi-
cal step within the traditional FG version of layering to exclude deontic
modality from the subjective category, as I will try to show in this section.
The basic difference between subjective epistemic modality and (what I
claim to be) subjective deontic modality is a difference in domain. As
shown in the previous section, subjective epistemic modality expresses
commitment to the truth of propositions, whereas subjective-deontic mo-
dality expresses commitment to the desirability of actions. In grammatical
terms, this difference in domain is reflected in the presence or absence of
tense in the non-modalized part of the utterance: subjective epistemic mo-
dality operates over tensed SoAs, located with respect to the speaker’s
temporal zero-point, whereas subjective deontic modality operates over
tenseless, virtual SoAs. This is a descriptively relevant distinction that will
probably be recognized in most analyses, but the question is how to incor-
porate it in the model. The decision within FG to locate the ‘highest’ types
of epistemic and deontic modality at different sides of the tense operator
can be regarded as one way to do justice to the distinction in domains –
epistemic modality at the proposition level, with scope over the tense op-
erator, and deontic modality at the predication level, without scope over the
tense operator.
I fully agree that the distinction between subjective epistemic and deon-
tic modality in terms of tense should somehow be included in the model,
but the question is whether association of the modals with different layers
is the right way to do this. In the FG conception of layering, association
with different layers necessarily also implies different status for operators,
and this makes it impossible to analyse both epistemic and deontic modal-
ity as subjective. If the highest type of deontic modality is associated with
the predication, and the highest type of epistemic modality is associated
with the proposition, it is impossible to do justice to the functional and be-
havioural similarities between the two categories that point towards a
shared subjective status. The reason is that the full set of layers is consid-
ered to be relevant for every type of (main) clause, more or less in the form
of a template: in this sense, utterances with subjective deontic modality as
an operator over the predicational layer still have to contain a propositional
layer with an associated slot for the subjective epistemic operator,11 so that
the deontic modal cannot be analysed as subjective in the same way as
epistemic modals.
Thus, the assumption that every main clause contains a propositional
layer is the most important obstacle to the inclusion of subjective deontic
modality in the model. The presence of a propositional layer need not be
regarded as a universal characteristic of clauses, however: this idea proba-
bly derives from the assumption implicit in many frameworks that every
utterance is fundamentally about knowledge. Following Halliday (1975,
1994) and McGregor (1997), I believe that the subjective-deontic domain
of action and the subjective-epistemic domain of knowledge constitute a
fundamental dichotomy in the system, such that in terms of subjective mo-
dalization any utterance belongs either to the deontic domain or to the
epistemic domain, i.e. that any utterance argues either about action or about
knowledge. This also implies that only epistemic utterances contain a pro-
positional layer. Subjective deontic structures lack all propositional
characteristics (McGregor 1997: 216–217), both functionally (because they
argue about virtual action) and grammatically (because they are tenseless):
what subjective deontic modality operates on is not propositions but ‘unlo-
calized’ SoAs, tenseless ‘virtual’ SoAs which are considered by the
speaker in terms of desirability.
Thus, one alternative to the traditional FG analysis is to give up the im-
plicit assumption that every main clause structure has a propositional
layer.12 In this way the distinction in tense between subjective epistemic
and subjective deontic modality can be incorporated in the model (in terms
of the presence and absence of a propositional layer in the structure of the
clause) in a way that does not a priori exclude the parallel analysis of the
two modal categories as subjective. As I will show in the following section,
however, implementing this alternative proposal requires two types of
modification in the traditional formulation of the FG model: a modular
separation between the interpersonal and the representational structures of
the clause, and optionality of layering in the representational structure.
6.2. The usefulness of modularity and top-down organization

The two most important features in Hengeveld’s (this volume) alternative
architecture for the FG model are modularity and top-down orientation.
Modularity means that the continuous layering from terms to illocutionary
frames as seen in the traditional model is abandoned in favour of a distinc-
tion between different modules dealing with different aspects of the
organization of the utterance. There is one module that takes care of inter-
personal organization, and another that takes care of matters of
representation, much as in frameworks like Halliday’s (1994) Systemic
Functional Grammar and McGregor’s (1997) Semiotic Grammar. Top-
down orientation, on the other hand, implies that in this modular frame-
work the interpersonal module, which takes care of speaker-hearer
interaction, is primary in that it can determine choices in the representa-
tional module. In this section, I will show how both of these features are
directly reflected in the behaviour of subjective modality as it has been
analysed in this chapter.
As explained in the previous section, the main problem with the tradi-
tional layered model was the assumption that the full set of layers is
considered to be present in any type of utterance. In this perspective, asso-
ciation with different layers for the highest types of epistemic and deontic
modality necessarily implies different status. If the highest type of deontic
modality is associated with the predication, this implies that it cannot have
the same subjective status as the highest type of epistemic modality, since
every utterance – including those with the ‘highest’ type of deontic modal-
ity – always contains a propositional layer with a slot for precisely this
subjective-epistemic modal operator. As argued in the previous section, I
fully agree with the descriptive motivation – the divergent behaviour of
tense – for associating the highest types of epistemic and deontic modality
with the proposition and the predication, respectively, but I do not agree
that this should imply a different status for the two categories of modality.
In a modular system, like that of Halliday (1994), McGregor (1997) or
Hengeveld (this volume), the representational association of the modal
categories no longer has any influence on their interpersonal status, be-
cause representational and interpersonal functions belong to separate
modules. The fact that epistemic and deontic modality have different do-
mains (tensed versus tenseless SoAs) is a purely representational matter
that can be dealt with in the representational component and does not inter-
fere with the question of subjective status. The subjective status of
epistemic and deontic modality, on the other hand, is an interpersonal issue
in its own right that can be determined on the basis of function (encoding
positions of commitment) and behaviour (in reaction to conditionality and
interrogation) and is not related to representational questions at all. The
connections that do exist between the interpersonal and representational
components, finally, can be captured in the top-down orientation of the
model. The representational choice between tensed and tenseless SoAs is
steered from the interpersonal component by the choice between subjective
epistemic and subjective deontic modality. Table 4 below provides a sche-

matic representation of how the analysis of subjective modality proposed in
this study can be dealt with in terms of the modular architecture proposed
by Hengeveld (this volume).
In sum, we can say that the descriptive problems discussed in the first
part of this chapter can generally be accommodated more easily in a modu-
lar architecture with top-down orientation between interpersonal and
representational components than in the traditional FG model. Because the
modular approach separates the interpersonal status of an operator from
possibly divergent representational associations, it also removes the most
important obstacle to the inclusion of deontic modality in the subjective
category: association with different layers at the representational level does
not necessarily have implications for the interpersonal status of epistemic
versus deontic modals. On the other hand, the top-down orientation still al-
lows one to incorporate the link between the epistemic-deontic distinction
at the interpersonal level and the tensed-tenseless distinction at the repre-
sentational level, in the sense that this representational choice is steered by
a choice at the interpersonal level.
Table 4. Modularity and top-down organization for subjective modality in Eng-

lish
Interpersonal module
Meaning Speaker’s commitment
Echo-effect in conditionals
Subjective status of
Form
the modal
Interlocutor-transfer in interrogatives
Epistemic (truth) and deontic (desirability)
Representational module
Meaning ‘Located’ propositions ‘Virtual’ actions

Domain
Of the modal
Form + Tense - Tense
There is one problem, however, which still remains unaddressed in the

model proposed by Hengeveld (this volume), and that is the problem of op-
tionality of layering at the representational level. The top-down orientation
of the model can easily deal with the fact that the interpersonal choice be-
tween epistemic and deontic modality leads to a representational choice
between tensed SoAs and tenseless SoAs, but this still requires a way for
the representational module to deal with the fact that some structures are
tensed and others are not. Assuming that the distinction between tensed and
tenseless SoAs corresponds to the distinction between propositions and
predications in the representational module,13 it is necessary to explicitly
include optionality in the representational layering. Optionality of the pro-
positional layer is a necessary prerequisite for adequate modelling of
structures with subjective deontic modality.
Notes
1. I would like to thank Bert Cornillie, Kristin Davidse, Renaat Declerck, Pat-
rick Goethals, Kees Hengeveld, Peter Lauwers, Bill McGregor, Jan Nuyts,
William Van Belle, Dieter Vermandere, and an anonymous referee for more
general discussions about modality and/or comments on a previous draft of
this chapter. A preliminary version of this chapter was presented at the 9th
International Conference on Functional Grammar in Madrid. I would like to
thank the members of the audience for their remarks and criticisms. Thanks
also to the late Machtelt Bolkestein for making available a copy of Bolke-
stein (1980).
2. Examples taken from the Cobuild corpus are marked with CB, and exam-
ples taken from the ICE-GB corpus are marked with their standard ICE-GB
text code. The relevant modal is always underlined.
3. See Verstraete (2001a) for an overview of the distinction between subjec-
tive and objective modality in terms of various theories of layering.
4. As shown by Nuyts (1992, 1993), however, the difference between modal
adjectives and adverbs cannot be explained in terms of the distinction be-
tween objective and subjective function, but must be related to three
interacting parameters, viz. evidentiality, discourse functionality and per-
formativity (Nuyts 1993).
5. In fact, this was the original context where this criterion was introduced; see
Jackendoff (1972) and Bellert (1977).
6. It is important to note that it is this commitment-suspending function of the
conditional marker rather than the presence of the conditional marker as
such that leads to an echo effect. As shown in Declerck and Reed (2001),
not all clauses introduced by if have the function of suspending the

speaker's commitment: there are for instance what they call performative
conditionals, where the if-clause contains a performatively used performa-
tive verb. Accordingly, such conditionals will not trigger any echo effect for
subjective modals, since it is only the commitment-suspending function of if
that leads to echo interpretation for subjective modals.
7. In fact, this use of deontic modality is more typically realized by periphras-
tic modals like have to rather than modal auxiliaries like must (Palmer 1990:
113–116, Declerck 1991a: 376–377).
8. See, however, the more descriptively oriented typology in Hengeveld
(forthcoming).
9. In this respect it is not surprising that Palmer (1990: 113) lists such exam-
ples as dynamic necessity.
10. That is, of course, without change of interpretation. Adding a propositional
attitude marker to a subjective deontic expression is always possible, but
necessarily shifts it from the subjective to the objective category. It is in fact
a common rhetorical strategy to add propositional attitude markers to orders
or prohibitions, in order to present the obligation as somehow existing inde-
pendently of the speaker, as in Unfortunately, you cannot enter this
building.
11. As I have shown in Section 5.2, such configurations do occur, but crucially
only in structures with nonsubjective deontic modality, which are interper-
sonally epistemic utterances about the existence of necessity or obligation.
12. There have been a number of proposals in the same direction within the FG
tradition, for instance in Moutaouakil’s (1996) argument that only declara-
tive clauses contain a propositional layer and in Hengeveld’s (1990)
argument that imperative clauses lack a propositional layer. If the proposi-
tional layer is tied to the grammatical feature of tense, however, neither of
these proposals is entirely adequate: epistemic interrogatives are tensed just
like their declarative counterparts, and subjective-deontic declaratives are
tenseless just like their imperative counterparts.
13. In the traditional model, the distinction between proposition and predication
was motivated in terms of the interpersonal-representational distinction
(Hengeveld 1989: 127–131). In Hengeveld’s new proposal (this volume),
what used to be the propositional layer now belongs both to the interper-
sonal component (as the ‘C’ variable) and to the representational component
(as the ‘p’ variable). This implies that the proposition-predication distinc-
tion is no longer motivated by the interpersonal-representational distinction
but requires a new motivation. The structural property of presence vs ab-
sence of tense is probably a good candidate for this purpose: location with
respect to the temporal zero-point is a necessary prerequisite for epistemic
negotiation about truth or falsity, as argued by Halliday (1994: 75).
References
Bellert, Irene
1977 On semantic and distributional properties of sentential adverbs. Lin-
guistic Inquiry 8. 337–350.
Bolinger, Dwight
1967 The imperative in English. In: To Honor Roman Jakobson, 335–362.
The Hague: Mouton.
1977 Is the imperative an infinitive? In: Dwight Bolinger, Meaning and
Form, 152–182. London: Longman.
1980 Problems in the Description of Modal Verbs. Assen: Van Gorcum.
Dancygier, Barbara
1998 Conditionals and Prediction: Time, Knowledge and Causation in the
Grammar of English if-then Constructions. Cambridge: Cambridge
University Press.
Davies, Eirian
1979 On the Semantics of Syntax. Mood and Condition in English. Lon-
don: Croom Helm.
Declerck, Renaat
1991a A Comprehensive Descriptive Grammar of English. Tokyo: Kaitaku-
sha.
1991b Tense in English. Its Structure and Use in Discourse. London: Rout-
ledge.
Declerck, Renaat and Susan Reed
2001 Conditionals. A Data-driven Analysis. Berlin and New York: Mou-
ton.
Dik, Simon
Foley, William and Robert Van Valin
1984 Functional Syntax and Universal Grammar. Cambridge: Cambridge
University Press.
Goossens, Louis
1996 English Modals and Functional Models: A Confrontation. (Antwerp
Papers in Linguistics 86.) Antwerp: University of Antwerp.
Halliday, Michael A. K.
1970 Functional diversity in language as seen from a consideration of mo-
dality and mood in English. Foundations of Language 6: 322–361.
1975 Learning How to Mean. Explorations in the Development of Lan-
guage. London: Longman.
1994 An Introduction to Functional Grammar. London: Arnold.
Hare, Richard
1970 Meaning and speech acts. Philosophical Review 79: 3–24.
Hengeveld, Kees
der Auwera and Louis Goossens (eds), Ins and Outs of the Predica-
tion, 53–66. Dordrecht: Foris.
1988 Illocution, mood and modality in a functional grammar of Spanish.
Journal of Semantics 6: 227–269.
25: 127–157.
1990 The hierarchical structure of utterances. In: Jan Nuyts, A. Machtelt
Bolkestein and Co Vet (eds), Layers and Levels of Representation in
Language Theory. A Functional View, 1–23. Amsterdam and Phila-
delphia: Benjamins.
forthc. Mood and modality. In: Geert Booij, Christian Lehmann, and
Joachim Mugdan (eds), Morphology: A Handbook on Inflection and
Word Formation. Berlin and New York: Mouton.
Jackendoff, Ray
1972 Semantic Interpretation in Generative Grammar. Cambridge, MA:
MIT Press.
Lyons, John
1977 Semantics. Vol. 2. Cambridge: Cambridge University Press.
McGregor, William
1997 Semiotic Grammar. Oxford: Oxford University Press.
Moutaouakil, Ahmed
1996 On the layering of the underlying clause structure in Functional
Grammar. In: Betty Devriendt, Louis Goossens and Johan van der
Auwera (eds), Complex Structures. A Functionalist Perspective,
201–227. Berlin and New York: Mouton.
Nuyts, Jan
1992 Subjective vs. objective modality: What is the difference? In: Mi-
chael Fortescue, Peter Harder and Lars Kristoffersen (eds), Layered
Structure and Reference in a Functional Perspective, 73–97. Am-
1993 Modality and the layered representation of conceptual and linguistic
structure. Linguistics 31, 933–969.
Palmer, Frank
1990 Modality and the English Modals. London: Longman.
2000 Attitudinal disjuncts and illocutionary force in clause combining – a
response to Bill McGregor. Functions of Language 7: 117–131.
2001 Subjective and objective modality: interpersonal and ideational func-

tions in the English modal auxiliary system. Journal of Pragmatics
33: 1505–1528.
Remarks on layering in a cognitive-functional
language production model
Jan Nuyts
1. Introduction
In this chapter I will take another look at the phenomenon of the layering of
qualificational categories, in language and beyond. The layered representa-
tion of the clause is a core concept in the framework of Functional
Grammar (henceforth FG) as developed by Dik (1989b, 1997) and
Hengeveld (1989, this volume). In this chapter I will adopt a basic ap-
proach to linguistic theorizing which is in a few respects substantially
different from that advocated in ‘standard FG’ (as represented in Dik
1997), viz. what I have called a cognitive-pragmatic or cognitive-functional
approach (Nuyts 1992a, 2001a). This approach shares with FG a strongly
functionalist orientation in the analysis of linguistic structure. But it devi-
ates from mainstream FG in adopting a much more radically cognitive
orientation, and, to the extent that there are cognitive assumptions relating
to (dimensions of) FG (cf. Dik 1987a, 1989), in adopting a different view
on the relationship between linguistic structures in the grammar and con-
ceptual structures figuring in human thought. Correspondingly, a
grammatical framework based on these principles – what I have called a
Functional Procedural Grammar – has to assume more levels of representa-
tion and processing, some of which (notably the conceptual ones) involve
degrees of abstractness which go well beyond what would commonly be
admitted in FG. Moreover, the present framework is much more proce-
durally oriented, assuming that producing linguistic expressions is a
strongly interactive and flexible process.1
As a consequence, the present framework leads to a view of the layering
of qualifications which is substantially different from the standard FG con-
276 Jan Nuyts
ception, both in terms of its position in a model of language production and

in terms of its precise format. This chapter, then, will present a sketch of
the basic elements of layering in the cognitive-functional perspective
adopted here, and I will discuss the how and why of some of the most ob-
vious contrasts with the standard FG view of layered representation.2
My repeated reference to the ‘standard conception’ of FG is important.
In the last few years some FG scholars have proposed modified concep-
tions of FG, which at least in certain respects constitute steps towards a
more cognitive and procedural approach to grammar. Probably the first
such proposal was Mackenzie’s (1998, 2000). More recent, but completely
different from Mackenzie’s, is what is proposed in Hengeveld (this vol-
ume). Whether Hengeveld’s proposal has consequences for the issues to be
discussed here is not obvious to me as yet. Hengeveld’s concerns are pri-
marily towards an integration of the discourse level into grammar, and this
clearly does cause some changes in the role and format of the layered sys-
tem in FG, inter alia in terms of its applying at or spreading out over
multiple levels in the model. But, at least at face value, the fundamental is-
sues which I will be discussing below remain unaffected. Whether this first
impression is accurate is to be decided on the basis of a future elaboration
of the new trend(s), however, and is therefore beyond the scope of the pre-
sent chapter.
I will assume basic acquaintance with the principles of the (classical)
FG layered proposal. I will use Hengeveld’s (1989) version as my refer-
ence, but the points to be made are equally applicable to Dik’s (1997)
(slightly different) version. For ease of reference, Figure 1 offers a sum-
mary overview of the essence of the system.
clause
interpersonal (E1: [π4 ILL: (S) (A) (π3 X1: [proposition] σ3) σ4])
representational π2 e1: [π1 pred (x1) ... (xn) σ1] σ2
predication
Figure 1. Basic elements of layered clause structure in Functional Grammar
Predicate operators (π1) and satellites (σ1): additional properties of the SoA
π1: qualificational aspect ((im)perfective, inchoative, progressive, etc.), predicate
negation
σ1: additional participants, manner, spatial orientation
Remarks on layering 277
Predication operators (π2) and satellites (σ2): setting of the SoA

π2: tense, quantificational aspect (iterative, semelfactive, etc.), objective modality
(epistemic and deontic), polarity
σ2: spatial and temporal setting, setting relative to other SoAs (cause, co-
occurrence, condition, reason, purpose)
Proposition operators (π3) and satellites (σ3): validity of the propositional content
π3: subjective modality (epistemic, boulomaic), evidentiality (inferential, quotative,
experiential)
σ3: propositional attitude, validity of proposition (source, evidence, motivation,
condition)
Illocution operators (π4) and satellites (σ4): communicative strategy of the speaker
π4: mitigation or reinforcement of illocutionary force
σ4: manner of speech act, communicative setting of speech act
2. A basic assumption: Linguistic and conceptual structure are

substantially different
I should start by briefly repeating a basic assumption which is critical for
understanding the present cognitive-pragmatic view of layered representa-
tion, concerning the nature of the human conceptual apparatus and its
relation to the linguistic systems.3 Pace Dik’s (1987, 1989a) view that FG
predications might serve as the coding device for conceptual representa-
tion, there is good evidence for assuming that conceptualization is non-
linguistic.
Probably the strongest argument for this assumption derives from a
combination of empirical observations about the structural and functional
nature of sets of semantically related expression types (henceforth called
‘semantic paradigms’), and basic functionalist views (also accepted in FG)
on what a grammar can and cannot do. Firstly, among (the numerous ex-
amples of) such semantic paradigms are the alternative expressions of
epistemic modality (Nuyts 1993a, 2001a), or the alternative expressions of
a basic action scene such as the ‘commercial event’ (Fillmore 1977, 1985,
Kay 1996). The variants in such paradigms can be shown to be due to vari-
ous functional factors which are essentially independent of the basic
semantic categories underlying the paradigm, including (quite prominently)
matters of perspectivization and information structure. From a language
production perspective, this means that all those variants can be considered
to originate in one common conceptual notion or cluster, the differences
between them emerging in the course of the production process due to, in-
ter alia, contextualization procedures adjusting the information to the local
278 Jan Nuyts
communicative situation. Secondly, the most important basic functionalist

principle (also a standard assumption in language psychology, cf. Bock et
al. 1992, Bock and Levelt 1994) is the implausibility of transformational
operations in a grammar. In this vein, the linguistic variants in semantic
paradigms are so fundamentally different that they cannot be derived from
each other in the grammar. Consequently, none of the variants can be used
to represent the basic semantic category. This unavoidably leads to the con-
clusion that the semantic category must have a format different from its
linguistic realizations.
The implications of all this go further than just the claim that conceptu-
alization is non-lexical (i.e. that it does not use lexical elements occurring
in natural languages). It also extends to the claim that its principles of or-
ganization are fundamentally different from linguistico-semantic organ-
ization. Thus, even a model such as Jackendoff’s (1983, 1987, 1990) Con-
ceptual Semantics – which postulates conceptual structures which are non-
lexical yet still strongly language-bound, i.e. based on the principle of
predicate-argument organization – faces problems with semantic para-
digms such as those mentioned above (Nuyts 2001a: 296ff.).
None of this should come as a surprise to functionalist researchers: The
functional requirements for a central information processing and storage
system, which conceptualization is, are completely different from the func-
tional requirements of a communication system such as language. So it is
only natural that they should have a completely different shape and organi-
zation.4
3. Layering is a conceptual, but not a linguistic phenomenon

If conceptualization is non-linguistic, what does it look like? There is no
full-blown answer at present, but the evidence mentioned in Section 2 does
uncover at least one major property of conceptualization: it must be lay-
ered. In fact, the argument in Section 2 simultaneously forces one to
conclude that notions such as epistemic modality are not specifically lin-
guistic, but basically conceptual (Nuyts 2001a). And this reasoning applies
in equal measure to other qualificational categories (most of which also in-
volve ‘semantic paradigms’), and by extension to the principle that there is
relative semantic scope between all these qualificational categories, which
is at the heart of the concept of layered representation in FG.
Perfectly in line with this conclusion is the observation (Nuyts 1998)
that there is a ‘mismatch’ (albeit one clearly motivated by functional prin-
ciples) between qualificational categories and linguistic forms, in the sense

that there is no one-to-one relation between them. Space prevents me from
elaborating this here (see for details Nuyts 2001a: 305ff.), but in strongly
generalized terms the mismatch comes in two types.
One type is involved in the phenomenon of semantic paradigms intro-
duced above: one qualificational category is expressed in a range of
different form types. There are two subtypes of this. One qualification as a
whole can be expressed in alternative forms: for example, epistemic
modality codes into adverbs, adjectives, auxiliaries, and mental state and
5
similar predicates. Or different dimensions of one qualification can on oc-
casion be expressed in two forms, as for instance when a negative polar
epistemic evaluation is simultaneously expressed by one form denoting the
epistemic scaling and one denoting the negative polarity (cf. I doubt vs I
don’t think, or it is doubtful vs it is improbable vs it is not probable). Such
one-meaning-to-multiple-forms mappings are to a large extent due to the
multi-functionality of language use, i.e. to the interaction of the semantic
category with other functional factors pertaining to its linguistic expression
(such as, very prominently, information structure), and the different de-
mands those functional dimensions impose on structure (Nuyts 2000).
The other type of mismatch works the opposite way: one linguistic form
is used by two or more qualificational categories. This, too, comes in two
subtypes. Two qualifications may be expressed simultaneously in one
form: I think, for example, expresses epistemic likelihood and subjectivity
(and as Perkins 1983 shows, such combinations are rather common). Alter-
natively, two or more qualifications may on different occasions use one
and the same form: past tense, for instance, normally marks time, but can
sometimes express epistemic modality, viz. when it is used as a weakener
on other epistemic expressions, as in the ‘past’ modals in English (e.g. may
vs. might) or in ‘performative past’ mental state predicates (i.e. certain uses
of I thought). These kinds of mismatch are clearly not accidental or unsys-
tematic. The former subtype is no doubt due to the frequent co-occurrence
of the two qualifications in certain usage conditions (for example, subjec-
tivity and epistemic uncertainty frequently go hand in hand in spontaneous
conversation). And the latter subtype is a by-product of the principle of ex-
pressibility, viz. the long-term striving for qualificational categories to find
expression, whereby expression forms are not created out of the blue, but
are initially ‘borrowed’ from other meanings on the basis of the principle
6
of (most probably) metonymic transfer.
The mismatch between meanings and forms not only underscores the
need to separate a conceptual system of qualifications from a linguistic sys-
280 Jan Nuyts
tem of qualificational expressions. It obviously also implies that there is a

discrepancy between the two. One may take this to mean that there must be
two layered systems, which differ in more or less substantial ways: one in
grammar and one in conceptualization. However, the observations rather
force one to rethink the role of layered representation in grammar in more
fundamental ways. As argued in Nuyts (1998, 2001a), there is no need for
a layered system in the cognitive procedures dealing with linguistic struc-
tures at all. Layering as such is a purely semantic issue, viz. a matter of
semantic scope. But grammar deals with forms, not with meanings. And
where forms respond to semantic properties, as is often the case in the be-
haviour of qualificational expressions, this should be handled in terms of
the procedural interaction between linguistic structure and the layered sys-
tem in conceptualization, rather than by duplicating the latter in grammar. I
am not arguing that layering plays no role in grammar: quite on the con-
trary, its effects on grammatical/syntactic structure and processing are
tremendous. But this is precisely the right way to state the issue: we are
dealing with effects, not with an inherent feature. To put it metaphorically:
layering is an ‘emergent’ property of linguistic structure (more or less in
Hopper’s 1998 sense).
Some effects of the semantics of layering certainly do become inherent
in grammar, i.e. when they are structuralized or grammaticalized in linguis-
tic form. But this process leads to different results for different
qualificational expression types, so that it cannot be stated in generalized
terms anyway (contrary to what is implied by the postulation of one global
layered system in grammar). Thus, if we stick to our example of epistemic
modality, in languages such as English, Dutch or German there are consid-
erable differences between the grammatical (auxiliary), lexical (adverbial)
and predicational (adjectival, verbal) expressions. This is true in terms of
which distinctions they code: the adverbs and adjectives allow a fairly sub-
tle rendering of all positions on the epistemic scale (even more so when
combined with modifiers), but the auxiliaries and verbal predicates only
allow a vague reference to a few areas on the scale (and they hardly allow
additional modifiers). And this is equally true of the behaviour of the forms
in these expression types, for example in the range of word-ordering possi-
bilities. Moreover, all these features can be stated in terms of the traditional
dimensions of grammatical description, i.e. which forms are available, how
can they be combined, and how can they be ordered in an utterance, in
view of the basic properties of the ‘parts of speech’ involved. There is no
need for any additional constructs such as a layered system. Once again,
there is no doubt that the linguistic properties of qualificational forms are
motivated by the conceptual properties of the qualifications they express.

But that is a matter of diachronic effects of meaning on form, and of syn-
chronic procedural links between forms and the basic conceptual
dimensions they express.
4. Basic properties of the layered system in conceptualization

But then, if we try to model the layered system in conceptualization, can
the layered system postulated in FG not serve as a source of inspiration?
The foregoing already implies that the format of ‘entities’ in the FG system
cannot be taken over (i.e. lexical structures such as the predicate and argu-
ments, or the distinction between operators and satellites). Yet maybe the
semantic notions involved in the system and the principles of their organi-
zation could be maintained? In some respects the answer may be positive,
but in some other respects it appears to be negative. In the following I will
try to sketch what layering in conceptualization may involve, with refer-
ence to some of the major features of the layered system in FG.
4.1. The gradual nature of the layered system, and the rationale
behind it
The layered system in conceptualization is (unlike the FG system) most
probably a gradual system. This is apparent in observations regarding the
semantic scope effects that arise when combining different qualificational
expressions in an utterance: one can always determine a semantic domi-
nancy hierarchy between the qualifications involved, which is moreover
stable across various tokens of the combination. This is also true of qualifi-
cations which in the FG system are on one level, e.g. the predicational one.
7/8
The following examples are self-explanatory:
(1) a. John went skiing a few times [quant. aspect] last year [time]
b. It happened last year that John went skiing a few times
c. *It happened a few times that John went skiing last year
(2) a. John must [deontic modality] go skiing next year [time]
b. There is an obligation for John to go skiing next year
c. *It will happen next year that John must/is obliged [at the time of
speaking] to go skiing
(3) a. John may [epistemic modality]9 go skiing next year [time]
b. There is some kind of likelihood that John will go skiing next year
c. *It will happen next year that John may go skiing
282 Jan Nuyts
In terms of this kind of reasoning, then, one arrives at the – very incom-
plete10 – gradual conceptual hierarchy of qualifications in (4):
(4) evidentiality
> epistemic modality
> deontic modality
> time
> quantificational aspect
> qualificational aspect
> state of affairs
This hierarchy (and hence the linguistic-semantic observations on which it

is based) is clearly not accidental or arbitrary: there is a straightforward
deeper rationale behind it, which ties in with very basic aspects of human
cognitive functioning. (This rationale is implicitly or subliminally con-
tained in, or closely approximated by, some aspects of the motivation for
the layered system in FG, but it got blurred in the FG account due to the
interference of other dimensions which are irrelevant at a conceptual level;
see Section 4.4 below.) Climbing up the cline in (4) correlates with a grad-
ual widening of the perspective on the State of Affairs (henceforth SoA).
The cline develops from, at the lower end, qualifications which further
specify internal features of the SoA, to, at the higher end, qualifications
which provide a global assessment of the status of the SoA. Correspond-
ingly, qualifications low in the system require no or hardly any information
other than knowledge of the SoA itself, while qualifications high in the
system are predominantly or exclusively based on information external to
the SoA. In even more basic terms, the system relates to the tension be-
tween perception and interpretation in cognitive functioning. Climbing up
the system in a way involves a decreasing role for direct perception of the
SoA, and an increasing role for interpretation and creative involvement on
the part of the speaker. In principle, the basic level of the conceptual repre-
sentation of the SoA is closest to, i.e. has been acquired through or should
be accessible for, direct perception of the world. Any instance of percep-
tion is obviously necessarily local, restricted to what is within the direct
reach of the perceptive system (whichever is involved) at one point in time.
The higher one climbs up in the hierarchy, the more the qualifications ap-
pear to be concerned with specifying aspects of this SoA which are beyond
the locality of immediate perception, and are thus dependent on, or aim at,
abstraction and generalization, and/or on perceptions or information out-
side the SoA proper.
For example, qualificational aspect requires the perception of an SoA

and its actual state of development, no more and no less. Quantificational
aspect already requires repeated perceptions of SoAs and a generalization
over them, e.g. in terms of the uniqueness of one perceived SoA relative to
others, or the similarity of different perceived SoAs such that this ‘type’ of
SoA can be characterized as having a certain frequency, or as being ge-
neric, etc. ‘Situation in time’ requires linking the perception of an SoA to
the (in itself very complex) perception of natural cycles of the sun and the
moon and/or the even more abstract estimation of numbers of such cycles
using conventional, hence completely abstract, definitions of time (e.g. the
calendar). (Situation in space is similar: determining that one is ‘in Paris’,
for example, goes far beyond mere immediate perception of one’s where-
abouts.) And epistemic modality is not at all a matter of direct perception
of an SoA any more, but of abstract deductive reasoning from perceptions
of other SoAs, or generalizations over them, to the tentative postulation of
a possible, hypothetical SoA (which should be accessible to perception if
the hypothesis is to be verified, of course).
Obviously, the fact that the system of layering of qualifications ties in
with such very basic features of human perception and information proc-
essing lends strong further support to our assumption that we are dealing
here with a basic conceptual phenomenon, not with a specifically linguistic
11
one.
4.2. Stacks in the system

The foregoing, however, does not mean that the FG concept of ‘stacks’ in
the layered system is completely inaccurate, even at the conceptual level.
In fact it is possible to distinguish, superimposed upon the gradual hierar-
chy, groups of qualifications which share certain semantic properties
(although the groups do not fully correspond to those in the FG system).
One distinction corresponds perfectly to that between the predicate-level
and predication-level qualifications in FG: in the system in (4), this line
should be drawn between quantificational and qualificational aspect. Quali-
fications below it further specify the internal structure of the SoA:
qualificational aspect, for instance, concerns the state of development of
the SoA (and manner of action – cf. note 10 – could be characterized in a
similar way). Above this line, qualifications no longer affect the internal
structure of the SoA, but situate the SoA as a whole in the world: this ap-
plies to quantificational aspect, which concerns the frequency of the SoA,
and it applies even more clearly to temporal (and spatial) situation. Another
284 Jan Nuyts
semantic division in (4) – one not present in the FG system, however – lies
in between time and deontic modality. Deontic and epistemic modality and
evidentiality (and emotional attitude, if it is included in the system) all in-
volve (different) kinds of speaker attitudes, i.e. explicit statements of
degrees to which the speaker is committed (in different ways) to the SoA.
Qualifications from time downward cannot be called ‘attitudinal’, however
(cf. Nuyts 2001a: 344ff.).
These stacks do not bring any new ‘logic’ into the system: as will ap-
pear from their definition above, they fit perfectly into the general rationale
provided in Section 4.1. This also means that the borders between them do
not constitute radical breaks (for the rationale implies graduality). Consider
the middle stack. Quantificational aspect does not change the properties of
the SoA as such, but nevertheless it still draws fairly closely on those inter-
nal properties, since it requires a comparative check whether there is
identity between instances of a potentially recurrent SoA. Temporal and
spatial qualification, however, involve a specification of or situation in the
external dimensions of time and space of an SoA, for which purpose the
internal structure of the SoA matters much less. If we jump to the upper
stack, deontic modality brings in the issue of speaker commitment, which
naturally arises due to the increased role of interpretation that emerges
from the clash between the SoA and other, external knowledge. But in a
way, deontic modality also still situates the SoA, viz. in the social world
12
and its moral values, without putting its reality status as such at issue.
Epistemic modality, however, does not situate the SoA any more, but
rather broaches the question whether it needs to be situated at all. Clearly,
then, the breaks in the system are no more than somewhat more drastic
qualitative jumps culminating from an accumulating number of smaller
jumps underneath them, which all fit into the same general logic.
4.3. Explaining scope extensions in terms of the rationale behind the

system
The rationale behind the layered system discussed in Section 4.1 also natu-
rally explains the differences in the scope of qualifications, not only
relative to each other, but also over domains of the qualified SoA. That is,
in the present account scope is a natural by-product of the logic inherent in
the system. FG deals with the scope of qualifications over domains of the
SoA by relating the qualifications to different hierarchical levels in clause
structure (the predicate, the predication, etc.). This is an essential dimen-
sion of the FG account which I have neglected so far. However, this kind
of construction is probably as superfluous in syntax as is the semantic

hierarchy of qualifications itself. For it follows directly from the layered
system at the conceptual level. Obviously, if low-level qualifications fur-
ther specify internal aspects of the SoA, they will naturally affect the
relation between the participants in the SoA. Since this relation usually gets
expressed as the predicate in a linguistic expression, they appear to be
predicate modifiers. If medium-level qualifications situate the SoA as a
whole, they affect not only the relation between the participants, but also
the participants themselves. So in linguistic expression they will usually
appear to affect the predicate plus the core participants, plus the expres-
sions of the low-level qualifications, of course. Finally, high-level
committing qualifications affect whatever there is about the SoA, including
its external situation, so it is only normal that their expressions will usually
also affect the situating operators and satellites.
Note that I am relativizing these statements with the labels ‘usually’ or
‘normally’. In fact, the situation need not be as described. Corpus data
(Nuyts 2001a: Section 2.4, 3.4 and 4.4) show that a qualification such as
epistemic modality, for example, sometimes only affects a subpart of a lin-
guistic expression, such as a single constituent or a parenthetical insertion.
Moreover, it has been observed that most or all qualificational dimensions
which can operate at the sentence level can also occur within the noun
13
phrase (the term operators in FG – cf. Rijkhoff 1990). Our present ap-
proach allows for a straightforward explanation. Depending on how the
speaker, in view of the communicative circumstances, decides to organize
the chunk of information he aims to express, a relation in conceptualization
that is qualified somehow (e.g. epistemically) can be linguistically coded,
not in the main pattern of an utterance, but in some subdomain (a noun, a
constituent, a parenthetical, etc.). In such cases, then, the qualificational
expression can hardly be considered to be attached to – in the epistemic
case – the predication in utterance syntax. This observation offers another
good reason to consider the extension of scope over domains of the SoA to
be a matter of the layered system in conceptualization, and not in syntax.
4.4. Qualifications and dimensions of interaction and discourse

In spite of certain correspondences between the FG account of layering and
the present approach, many aspects of the organization in the FG system
are entirely absent here. I have argued elsewhere (cf. Nuyts 1992a: 196f.,
1992b) why the distinction in FG between an interpersonal and a represen-
286 Jan Nuyts
tational dimension in language use (Hengeveld 1989) is not very well mo-
tivated. My objections (which I will not repeat here) partly boil down to an
intrinsic problem with the concept of separate strata in linguistic organiza-
14
tion correlating with separate language functions (whatever these are).
This clearly involves a departure from Dik’s (1986) view that there is a
multiple mapping from functions to forms, such that there is no one-to-one
relation between the two dimensions. I wholeheartedly endorse Dik’s origi-
nal view (cf. Section 3 above), but to go into this matter here would lead us
astray.
More relevant in the present context is that, if the layered system ap-
plies at the conceptual and not at the linguistic level, notions such as
‘speech event’ or ‘narrated event’ (which Hengeveld correlates with the in-
terpersonal and representational dimensions, respectively) are not at stake
anyway. A conceptual structure is neither a narrated nor a speech event, it
is just a coding of derivations from percepts (SoAs), including a marking
of their status (qualifications). That is, a conceptual system, including the
layered system (in a way – see Section 5 below), just represents. To illus-
trate this directly in terms of qualifications: it is hard to see why deontic
modality, i.e. whether a speaker considers something good or bad, or desir-
able or necessary, would be representational, while an evidential category
such as inference would be interpersonal. Both just participate in the con-
15
ceptual representation of reality, no more and no less.
Thus, the layered system in FG to a considerable extent integrates inter-
actional – and correlated with this, discursive – elements, not only in the
rationale offered for the organization of the system, but also in terms of
concepts and notions actually figuring in the system. But none of these be-
long in a conceptual system of layering, even if some of them do relate to
dimensions of the layered system in conceptualization. Let me explain
these cryptic statements in some more detail.
Firstly, the FG layered system fully integrates the matter of illocution-
ary force. This is justified in an account of the behaviour of qualificational
forms in the grammar, since speech act markers clearly interact with and
have properties comparable to expression devices for qualifications. But
illocutionary force as such is not a conceptual category. It is certainly not a
qualification of an SoA: it is a matter of the speaker’s plans and intentions
with an utterance vis-à-vis the hearer in the actual communicative situa-
tion, in relation to his/her deeper (non-linguistic) intentions with respect to
the world (as relevant at the conceptual level). Thus, unlike qualifications
which assess the position of an SoA in the world, illocutionary force is
only relevant in connection with the process of producing linguistic expres-
sions in an actual communicative situation. It is, in fact, a major element

driving this process.
Certainly, illocution is related in very profound ways to the conceptual
system, and even to certain qualificational dimensions there, i.e., probably
not accidentally, the high-level ones, belonging to the ‘attitudinal stack’
(see Section 4.2). For one thing, illocution shares with these qualifications
the matter of performativity. In fact, quite like illocution, speaker attitudes
16
are bound to the hic et nunc of the speaker. Of course, what is performed
in qualification and in illocution are quite different, i.e. a conceptual opera-
tion of determining an attitude towards an SoA vs a linguistic act towards
an interlocutor. Nevertheless, the ‘performing entity’ is no doubt the same
in both (cf. Section 5 below).
For another thing, the sources of illocutionary force are obviously also
conceptual, and in this respect too, there are links with (the same)
qualificational categories. Information questions ensue from gaps or
unclarities in conceptual knowledge, i.e. essentially from epistemic
uncertainty, and the intention to fill or resolve them. And orders or requests
for action, or promises, probably result from an epistemic assessment of
the (non-)existence and a deontic assessment of the (un)desirability and/or
(non-)necessity of an SoA, and the intention to do something about that.
This may explain why illocutionary markers and deontic and epistemic
markers are often closely related. It also explains why performative episte-
17
mic expressions do not occur in questions. Thus, a speaker can deal with
an informational gap in two ways, depending on what (s)he aims to
achieve. Either (s)he can communicate his/her own view, with an epistemic
assessment, of how it might be filled, leading to a declarative with an
epistemic expression. Or (s)he can ask the hearer to fill it, leading to a
question thematizing the gap or unclarity as such. But then, there is no use
for an epistemic expression. In fact, if there is such an expression in other
than a speech act-modifying function, then this is automatically part of the
thematized gap, hence descriptive (cf. note 16).
A second and even more complex issue is how to deal with conditional,
temporal or causal relations, relations of purpose, reason, motivation, etc.
in linguistic expressions. FG fully integrates these in the layered system, as
satellites at the different levels. But conceptually, these dimensions are ob-
viously quite different from ordinary qualifications of SoAs: they concern
relations between SoAs. Thus, even as satellites they usually take the form
of full subordinate clauses, although the same links exist between syntacti-
cally independent clauses. So these relations extend beyond the domain of
a single utterance or SoA. How such relations get expressed, then, is a dis-
288 Jan Nuyts
cursive or interactional issue, relating to how the speaker organizes his/her

discourse in view of his/her communicative purposes.
Still, it is quite obvious that such clause connections do relate to the hi-
erarchy of qualifications, though probably in very complex ways. For
example, temporal connections are unavoidably based on the temporal
situation of each of the related SoAs. Or consider conditionals. Performa-
tive epistemic expressions do not occur in a conditional protasis (Nuyts
2001a: Sections 2.3, 3.3 and 4.3). The explanation may be similar to that
for the absence of performative epistemic forms in questions. In the prota-
sis, a speaker brings up a hypothesis about how some conceptual gap might
be filled. But (s)he is not conjecturing how (s)he expects the gap to be
filled; in that case an epistemic assessment would be to the point. Rather,
(s)he is stating one possible way the gap might be filled, with the aim of
pointing out (in the apodosis) what consequences that might have, irrespec-
tive of its probability. Hence, there is no use for an epistemic expression in
the protasis. This analysis does show that conditionals have an intimate
link with the domain of epistemic qualification. Nevertheless, this and
other dimensions of SoA linking are not part of the layered system as
18
such. How combining SoAs in relation to the layered system in concep-
tualization does work in a language production system is a matter that goes
beyond the scope of the present chapter.
With reference to the current concerns in FG with discourse structure,
the foregoing obviously implies that in the present cognitive-functional
framework, discourse organization as such does not belong in grammar ei-
ther, i.e. in the cognitive system for processing linguistic structures (I am
using the term ‘grammar’ in this chapter in this narrow sense, obviously). I
am not certain where or how in Hengeveld’s (this volume) new proposal
the lexicon comes in. But assuming that at least the ‘expression level’ is
fully lexical, since a complete unit there is a paragraph, I take it that dis-
course organization is ultimately considered to be entirely part of linguistic
structure and processing. In the current cognitive-functional framework,
however, grammar is a device solely for utterance processing. Surely,
grammatical processing is critically determined by discourse factors, but
that is a matter of processing the effects of discourse on the internal consti-
tution of a single utterance. Discourse organization as such is handled in
several steps in Functional Procedural Grammar. But central to it is a sys-
tem (the ‘situational network’) which performs a first adaptation of long-
term conceptual knowledge to the actual communicative situation. This
system still works in terms of conceptual structures, unlike grammar
proper. But unlike the central conceptual system, it uses mid- to short-term
memory, and in some respects it already adapts conceptual structures to the

19
process of language production. Unfortunately, going further into this
debate would lead me too far astray from my present concern with layering
20
(but see De Schutter and Nuyts 1983, Nuyts 2001a: 273ff).
5. A few more thoughts about conceptual layering

Let me briefly make a few more final observations pertaining to the status
of qualifications – and especially the high-level, attitudinal ones – in con-
ceptualization, which reveal complexities of the issue of layering which
clearly extend far beyond what belongs in a linguistic grammar.
5.1. Control
If attitudinal qualifications (like speech acts) are ‘performed’ (see Section
4.4), what is it that ‘performs’ them? If climbing up the qualificational hi-
erarchy involves an increasing role for creative involvement of the speaker
(see Section 3.1), what causes that involvement? If attitudinal qualifica-
tions all involve an explicit statement about types of speaker commitment
to the SoA, and can be expressed with or without speaker commitment to
the qualification itself (the performativity vs descriptivity issue – see note
16), where does that commitment come from? All these questions strongly
point in the direction of a concept which has a long tradition in AI but
which apparently must be assumed in cognitive theories of natural intelli-
gence as well: they all suggest the existence of a ‘control unit’ which
steers, coordinates and supervises at least some of the operations of the
cognitive system. And this no doubt includes in a quite direct way the op-
erations leading to attitudinal qualifications of information about the world
(SoAs): they result from the control system’s evaluative comparison of the
chunk of knowledge (the SoA) that is to be qualified to other knowledge
stored somewhere in the conceptual system. This is not the context in
which to elaborate much on this notion of control and its implications (for
discussion of its links to other highly interesting notions such as con-
sciousness and attention see Nuyts 1992a, 2001a: 357ff.). But it is well
known that this element has pervasive effects on all kinds of perceptive and
behavioural systems in human cognition, witness the continuing discus-
sions about automaticity vs control in many areas of cognitive processing.
As such, it stands beyond all those individual cognitive systems, including
the linguistic. But if attitudinal qualifications are so closely linked to the
290 Jan Nuyts
operations of this unit, then those qualifications (and the system they are
part of) can hardly be less independent of the various task-specific percep-
tual and behavioural systems, including the linguistic one.
5.2. An SoA is specified for all qualifications, conceptually, but not

linguistically
Speakers rarely mention more than one or a few qualifications of an SoA in
an utterance. But that does not mean they have no notion of the status of
that SoA in terms of all other qualificational dimensions. What is men-
tioned in a specific utterance is a matter of what is relevant for the hearer in
the very specific, local communicative situation. But knowing about an
SoA is not so local, of course: it means anchoring it in one’s long-term
knowledge about the world. In line with Chafe’s (1994: 129) adage that
“consciousness cannot function without being oriented in space, time, soci-
ety, and ongoing background events”, then, it is very plausible to assume
that for each SoA in conceptual knowledge all qualificational dimensions
in the layered system must be present and set for a certain value. Knowl-
edge about events and objects is probably not fully anchored in one’s
conceptual system until all dimensions of its status are clear, and if some of
those dimensions are not clear, the control unit will strive to resolve the
gaps as soon as possible. ‘Anchoring’ an SoA thus means knowing how it
relates to other conceptual information, its frequency (aspect), its spatial
and temporal situation, its social value (deontic modality), its reality status
(epistemic modality), and how one got to know about it (evidentiality). Of
course, there is no reason to burden the linguistic system with this: gram-
mar only needs to deal with what actually appears in the linguistic
expression.
5.3. Qualifications as representations and as operations

In grammar, a qualificational expression is unavoidably always a (lexical)
label, a representation. To be sure, it is processed in grammar, but even in
that process it remains a label: grammar processes lexical elements, and
those are representational things. Conceptually, however, giving qualifica-
tion the status of a label or a representation will not do: as is implied in
Section 5.1, qualifying something is an operation, a process. The matter is
complicated, however. Let me lift a corner of the veil.
A few times, I have mentioned the distinction between performative and
descriptive uses of expressions of high-level qualifications such as epis-
temic modality. In conceptual terms, the difference can be stated as
follows. Performative expressions linguistically render the conceptual out-

put of operations of the central control unit directly. But descriptive,
commitment-less expressions do not result from an operation of the control
system, but render qualifications which have been represented and stored
21
as part of the speaker’s knowledge of the world. This does not mean,
however, that the distinction can simply be grasped in terms of an opposi-
tion between an operation and a stored representation. For, obviously, an
operation over an SoA still results in a value being assigned to the SoA;
and, as was implied in Section 5.2, this value must even be placed in long-
term storage. What really differentiates qualifications with speaker com-
mitment from those without, then, is this: the former, after having been
executed, remain ‘validated’, i.e. they remain coded within the layered sys-
tem, and this is probably the case as long as the speaker is not forced to
reassess them. This does not apply to the latter. Descriptive qualifications
involving the speaker’s own former views originally involved validated
qualifications in the layered system, which have become invalidated by in-
formation acquired later. But for some reason they have not simply been
erased from memory, but have been stored as knowledge about one’s for-
mer ideas. Descriptive qualifications involving another speaker’s views, on
the other hand, are acquired purely through perception, either because one
has been informed about them, or because one has inferred them from
other observations. So they immediately enter the conceptual system as
knowledge about facts in the world, i.e. as a representation of an SoA.
When a speaker brings up a performative qualification in communica-
tion, however, the ‘operational’ status of the qualification, probably
unavoidably, becomes acute again. Let us take the example of epistemic
modality again. A speaker will probably only bring up his/her degree of
epistemic commitment to an SoA in a discourse if the status of the SoA in
this regard is not obvious, to him/herself, or because the hearer turns out to
hold a different view, or because there is otherwise new information rele-
vant for the speaker’s view. Thus, one may reasonably assume that an SoA
in long-term storage in conceptualization is normally certain knowledge.
That is, the epistemic slot in the layered system over it is set for certainty.
Consequently, its epistemic status is not an issue and is not brought up, at
least not by the speaker, when talking about the SoA. When knowledge is
epistemically less than certain to the speaker, however, it is inherently un-
stable, and (s)he will keep trying to resolve the uncertainty. This means
that (s)he keeps processing information potentially relevant to the SoA,
thereby epistemically reassessing it. (One may actually wonder whether
uncertain information is normally stored for more than a short to interme-
292 Jan Nuyts
diary period of time at all. Similarly, one may wonder whether clearly
counterfactual information, i.e. information marked for ‘certainty that not’,
makes it into the long-term store of world knowledge, except for rare,
highly significant cases, such as ‘that job of a lifetime which one did not
get’.) Of course, even if the speaker’s knowledge about an SoA is certain,
if a hearer brings in an alternative view, the speaker is again forced to re-
calculate his/her view in function of what the hearer adds in terms of
background information for the SoA. Whatever the details, it is obvious
that it is precisely when talking about one’s commitment to an SoA – i.e.
precisely when epistemic or other attitudinal elements enter the grammar
— that the operational character of these high-level qualifications is cru-
cial. Yet, the grammar is hardly the place to handle those high-level
operations as such – in fact, it does not have the infrastructure for them.
6. Conclusion
The point of this chapter can be summarized as follows. Linguistic theories
of grammar have always shown a tendency to deal with whatever they en-
counter in linguistic data in the grammar proper. As long as they deal with
structural phenomena, that is fine. But when they start dealing with seman-
tic and pragmatic phenomena, they quickly run into the problem of
overburdening the grammar with constructs which do not belong there, but
belong in other areas of cognition, such as conceptualization. That is, they
create grammars with an ‘overcapacity’. I suspect that the layered system
in FG – as well as the current tendency in FG to render discourse organiza-
tion in grammar – is a case in point.
Notes
1. The same is probably true of understanding linguistic expressions. But I will
mainly adopt a production perspective here.
2. The arguments and views presented in this chapter are developed in much
more detail, and are moreover underpinned by substantial amounts of ex-
perimental and corpus data, in Nuyts (2001b).
3. I have elaborated the arguments for this assumption at length elsewhere
(Nuyts 1990, 2001a), so for reasons of space I will not repeat them here.
4. The foregoing actually does not imply that conceptualization must be imag-
istic (cf. Dik’s ‘perceptual representations’ or Jackendoff’s ‘3D
representations’). An action scene such as the ‘commercial event’ involves
many dimensions which are entirely non-visual (e.g. possession, trading

conventions, etc.), and epistemic modal evaluations are by definition en-
tirely non-perceptual. What all of this means for the nature and (no doubt
highly complex) organization of the human conceptual systems is beyond
the present chapter (but see Nuyts 2001b for discussion).
5. The FG distinction between subjective and objective epistemic (and deon-
tic) modality is conceptually problematic, as I have argued in Nuyts (1992b,
2001a, 2001). So I will not draw this distinction in the remainder of this
chapter. Even if one does draw it, however, the arguments made in this
chapter remain intact, since each subtype will still attract at least two alter-
native expression forms.
6. On the pervasiveness of this principle of ‘borrowing’ and its links to dia-
chronic processes such as subjectification and grammaticalization, see
Nuyts (1998, 2001a).
7. An asterisk in these examples obviously refers to a semantic anomaly, not
(necessarily) to a grammatical one.
8. It is more difficult, if not impossible, to give an example combining a per-
formative epistemic and deontic qualification. But that has little to do with
the hierarchical relation between them. This combination – quite like a
combination of performative epistemic and evidential forms or of performa-
tive deontic and evidential forms, for that matter – is hard to get for reasons
that have to do with the basic nature of these qualifications as ‘attitudinal’
or ‘committing’ categories (see below) and the limits this imposes on con-
ceptual processing. See Nuyts (2001a) for an explanation.
9. As mentioned before, I do not adopt the distinction between subjective and
objective modality here. But for those who do accept this distinction in FG,
example (3) is based on the assumption that in current FG modal auxiliaries
would classify as objective (i.e. belonging at the predication layer). The ar-
gument actually works equally well with epistemic predicative adjectives,
which are no doubt considered objective.
10. Obviously, there are quite a few qualificational categories missing from this
list, at least including (inter)subjectivity and mirativity, boulomaic or emo-
tional attitude, volition, and kinds of spatial situation and direction. The
reason for omitting them is that their precise position in the system or their
exact status relative to other categories is not clear to me at present. Notions
such as dynamic modality and manner of action probably fit under the label
of one of the types of aspect.
11. The very fundamental nature of the layered system obviously also explains
why this system has such a pervasive effect on linguistic development. It is
well known that the ontogenesis of modal meanings (dynamic, deontic,
epistemic) largely follows one, cross-linguistically recurrent path, which
corresponds to climbing up the qualificational hierarchy in (4) (see, for in-
294 Jan Nuyts
stance, Stephany 1993 and Hickmann et al. 1993). Similarly, the diachronic
evolution of qualificational expressions in general is known to follow the
cline in the layered system (cf., for example, Traugott’s many illustrations
of the principle which she dubs ‘subjectification – Traugott 1989, 1995,
1997, among others – and which clearly correlates with ‘gaining height’ in
the layered system in (4) – cf. Nuyts 2001a). For acquisition, the reason for
the correlation is obvious. Increasing reliance on information external to the
SoA proper and increasing abstractness and generalization obviously lead to
increasing complexity involved in determining the status of the SoA. And,
presumably, higher complexity correlates with greater acquisitional diffi-
culty, hence later acquisition. For diachronic evolution this explanation is
less evident: why should complexity in an individual mind correlate with
order of long-term development in a linguistic community? Yet in the bio-
logical world evolution also appears to lead to increasingly complex
organisms. So it is tempting to assume that even for diachrony there is
something to the matter of the complexity of the qualificational categories,
even if it is not immediately apparent in which way.
12. The SoA can be either real or non-real: cf. It is a good thing that you did
that vs. You should do that soon. However, this reality status as such is not
at issue, but is a ‘precondition’ for the deontic evaluation.
13. Probably the parallelism between the two levels goes very far, in the sense
that qualificational expressions at both levels are at least roughly subject to
the same organizational principles, microstructurally (i.e. their ordering be-
ing determined by semantic scope) and macro-structurally (in terms of the
general rationale behind the organization). But showing this is beyond the
present chapter.
14. Representational vs interpersonal will probably not do for that purpose
anyway: see Nuyts (1992a: 26–64, 1993b).
15. If anything, deontic modality would then appear to have more to do with an
interpersonal or social dimension than inference, although that is exactly the
opposite of what the FG account implies. But in a way, everything concep-
tual is social or interpersonal. This is clearly true of ‘good’ or ‘bad’, but the
same applies, for example, to what one considers fast or slow, or large or
small (i.e. kinds of aspectual qualifications). Mind is a thoroughly social
phenomenon, but that observation helps little to structure the layered sys-
tem.
16. This is reflected in the fact that all attitudinal qualificational expressions
show a difference between performative and descriptive uses, i.e., respec-
tively, between uses in which the speaker expresses his/her own current
attitude towards the SoA and uses in which a speaker reports on someone
else’s, or his/her own but former, attitude without thereby committing
him/herself to the SoA at the time of speaking. Compare I think he is the
murderer and John thinks he is the murderer or At that moment I thought he

was the murderer. A similar distinction exists in the use of speech act verbs.
But it does not exist in expressions of qualifications from time downwards.
It is actually important to stress that the layered system in (4), and the dis-
cussion relating to it, only concerns the performative ‘version’ of deontic
and epistemic modality and evidentiality. The status of descriptive uses is a
completely separate story. I will briefly return to this matter in Section 5 be-
low.
17. If they do, they lose their epistemic status and become speech act modifiers:
Dutch misschien ‘maybe’, for instance, can figure in a question, but then it
does not express epistemic uncertainty, but acts to turn the question into a
tendentious one. See Nuyts (2001a).
18. This relation between qualifications and clause linking immediately ex-
plains why there are so many diachronic and synchronic connections
between qualificational and discourse markers, as observed, for example, by
Traugott and König (1991), and why diachronic developments even of dis-
course markers do fit into the pattern of continuing subjectification, as
Traugott calls it, i.e. of climbing up the ladder of the qualificational hierar-
chy: they do so by virtue of the relation of discourse markers to
qualifications in the layered system.
19. The difference between Hengeveld’s handling of discourse and the account
in the present framework could be summarized as follows. Hengeveld
seems to adopt an ‘accumulation’ concept of discourse production: the steps
taken at the interpersonal and representational levels accumulate at the ex-
pression level. Functional Procedural Grammar adopts a ‘tear apart and
code into language’ concept: large clusters of information going into what
will ultimately emerge in the hearer’s mind as a coherent discourse are pre-
sent in the speaker’s conceptualization and early pre-linguistic stages of
communicative processing only. The whole process of language production
is a matter of gradually singling out small chunks of information which get
coded in single utterances (albeit in discursively adequate ways). Grammar
actually only deals with the last part of that process, the single utterance.
20. It may be relevant to refer to the fact that Systemic Functional Grammar
(e.g. Halliday 1994), too, deals with discourse organization in the semantic
stratum, and not in the lexico-grammar, for good reasons, which have been
elaborated at length in the Systemic literature.
21. This does not necessarily mean that, as part of one’s conceptual representa-
tion of the world, descriptive qualifications are not distinguished from the
facts about the world which they qualify. Maybe they do maintain a special
status, for knowing about their qualificational status is obviously necessarily
part of one’s knowledge about them. So they may form a sort of intermedi-
ate level between the representation of the SoA and the hierarchy of
296 Jan Nuyts
qualifications with speaker commitment. But they are in any case fully sub-
ject to the latter, as is obvious from the fact that in linguistic expression they
can be affected by any qualification in the layered system with speaker
commitment. A descriptive epistemic qualification, for instance, can be
qualified by the speaker for aspect, time, deontic modality, epistemic mo-
dality, etc.
References
Bock, Kathryn and Willem J.M. Levelt
1994 Language production: Grammatical encoding. In: Morton Ann
Gernsbacher (ed.), Handbook of psycholinguistics. San Diego: Aca-
demic Press, 945–984.
Bock, Kathryn, Helga Loebell and Randal Morey
1992 From conceptual roles to structural relations: Bridging the syntactic
cleft. Psychological Review 99: 150–171.
Dik, Simon C.
1986 On the notion “functional explanation”. Belgian Journal of Linguis-
tics 1: 11–52.
1987 Linguistically motivated knowledge representation. In: Makoto Na-
gao (ed.), Language and Artificial Intelligence, 145–170. Amster-
dam: North-Holland.
1989a Towards a unified cognitive language. In: Frieda J. Heyvaert and F.
Steurs (eds), Worlds behind Words, 97–110. Leuven: Leuven Uni-
versity Press.
1989b The Theory of Functional Grammar, Part 1. Dordrecht: Foris.
1977 Topics in lexical semantics. In: Roger W. Cole (ed.), Current Issues
in Linguistic Theory, 76–138. Bloomington: Indiana University Press.
1985 Frames and the semantics of understanding. Quaderni di Semantica
6: 222–254.
1994 An Introduction to Functional Grammar. London: Arnold.
Hengeveld, Kees
25: 127–157.
Hickmann, Maya, Christian Champaud and Dominique Bassano
1993 Pragmatics and metapragmatics in the development of epistemic
modality. First Language 13: 359–389.
Hopper, Paul J.
1998 Emergent grammar. In: Michael Tomasello (ed.), The New Psychol-
ogy of Language, 155–175. Mahwah: Erlbaum.
Jackendoff, Ray
1983 Semantics and Cognition. Cambridge, MA: MIT-Press.
1987 Consciousness and the Computational Mind. Cambridge, MA: MIT-
Press.
1990 Semantic Structures. Cambridge, MA: MIT-Press.
Kay, Paul
1996 Intraspeaker relativity. In: John J. Gumperz and Stephen Levinson
(eds), Rethinking Linguistic Relativity, 97–114. Cambridge: Cam-
1998 The basis of syntax in the holophrase. In: Mike Hannay and Machtelt
Bolkestein (eds), Functional Grammar and Verbal Interaction, 267–
296. Amsterdam and Philadelphia: Benjamins.
2000 First things first: Towards an incremental Functional Grammar. Acta
Nuyts, Jan
1990 Linguistic representation and conceptual knowledge representation.
In: Jan Nuyts, Machtelt Bolkestein and Co Vet (eds), Layers and
Levels of Representation in Language Theory, 263–293. Amsterdam
1992a Aspects of a Cognitive-Pragmatic Theory of Language. Amsterdam
1992b Subjective vs. objective modality: What is the difference? In: Mi-
chael Fortescue, Peter Harder and Lars Kristoffersen (eds), Layered
Structure and Reference in a Functional Perspective, 73–98. Am-
1993a From language to conceptualization: The case of epistemic modality.
Chicago Linguistic Society 29, vol. 2: 271–286.
1993b On determining the functions of language. Semiotica 94: 201–232.
1998 Layered models of qualifications of states of affairs: Cognition vs.
typology? In: Johan van der Auwera, Frank Durieux and Ludo Le-
jeune (eds), English as a Human Language, 274–284. München:
Lincom Europa.
2000 Tensions between discourse structure and conceptual semantics: The
syntax of epistemic modal expressions. Studies in Language 24:
103–135.
2001a Epistemic Modality, Language and Conceptualization: A Cognitive-
Pragmatic Perspective. Amsterdam and Philadelphia: Benjamins.
298 Jan Nuyts
2001b Subjectivity as an evidential dimension in epistemic modal expres-

sions. Journal of Pragmatics, 33: 383-400.
Perkins, Michael R.
1983 Modal Expressions in English. London: Pinter.
Rijkhoff, Jan
1990 Toward a unified analysis of terms and predications. In: Jan Nuyts,
A. Machtelt Bolkestein and Co Vet (eds), Layers and Levels of Rep-
resentation in Language Theory, 165–191. Amsterdam and Phila-
delphia: Benjamins.
Schutter, Georges de and Jan Nuyts
1983 Toward an integrated model of a functional grammar. In: Simon C.
Dik (ed.), Advances in Functional Grammar, 387–404. Dordrecht:
Foris.
Stephany, Ursula
1993 Modality in first language acquisition: The state of the art. In: Nor-
bert Dittmar and Astrid Reich (eds), Modality in Language
Acquisition, 133–144. Berlin and New York: De Gruyter.
Traugott, Elizabeth C.
1989 On the rise of epistemic meanings in English: An example of subjec-
tification in semantic change. Language 65: 31–55.
1995 Subjectification in grammaticalisation. In: Dieter Stein and Susan
Wright (eds), Subjectivity and Subjectivisation, 31–54. Cambridge:
Cambridge University Press.
1997 Subjectification and the development of epistemic meaning: The
case of promise and threaten. In: Toril Swan and Olaf J. Westvik
(eds), Modality in Germanic Languages: Historical and Compara-
tive Perspectives, 185–210. Berlin and New York: De Gruyter.
Traugott, Elizabeth C. and Ekkehard König
1991 The semantics-pragmatics of grammaticalization revisited. In: Eliza-
beth C. Traugott and Bernd Heine (eds), Approaches to Grammatic-
alization, 189–218. Amsterdam and Philadelphia: Benjamins.
Discourse Structure, the Generalized Parallelism
Hypothesis and the Architecture of Functional
Grammar
Ahmed Moutaouakil
1. Introduction
One of the most extensively discussed issues in the recent FG literature is
the way in which the current sentence-oriented model can be extended to
handle supra-sentential phenomena. In order to come to grips with this
problem, two kinds of solutions have been proposed: an ‘extrinsic’ solution
consisting in combining FG with other theories (e.g. Gulla 1997) and an
‘intrinsic’ solution intended to achieve the extension in question by inter-
nally transforming the FG apparatus itself. Within the latter line of thought,
two approaches can be distinguished: an expanding ‘upward layering’ ap-
proach (Hengeveld 1997; Moutaouakil 1998, among others) involving a
continuum of successively larger units, and a modular approach (Kroon
1997) according to which sentential and supra-sentential (discourse) phe-
nomena are to be dealt with in two separate modules.
Hengeveld (this volume) proposes a model of Functional Discourse
Grammar (FDG) which integrates these two approaches. My aim in this
study is to further argue for Hengeveld’s claim and show that the two ap-
proaches in question should indeed not be considered incompatible. In the
light of this assumption and in the perspective of a new architecture for FG,
I will suggest a general frame for describing and explaining discourse phe-
nomena which is both hierarchical and modular. This frame differs in
certain substantial as well as organizational aspects from Hengeveld’s
model of FDG but it goes in the same direction and strives to achieve the
same theoretical goal, simplification and unification of the theory of FG.
Elaborating on the idea put forward in Dik (1997b) that the hierarchical
structure and the functional relations postulated for the clause can be pro-
300 Ahmed Moutaouakil
jected onto the discourse level, I will push a step forward the structural
parallelism hypothesis advocated in earlier works (Rijkhoff 1992, Mou-
taouakil 1993) by assuming that the different discourse categories (from
word to text) can be said to be underlain, although with various degrees of
surface explicitness, by one and the same archetypal, hierarchically organ-
ized structure whose (quantitative and/or qualitative) actual realization is
regulated by structural and typological parameters. It will be hypothesized,
in line with Moutaouakil (1999), that the representation of this structure
can be perfectly ‘transmodular’ in the sense that its different parts (i.e.
levels and layers) can be represented in separate but interacting modules.
2. Discourse and Discourse Categories

The claim that will be advocated here presupposes a re-examination of the
notion ‘discourse’ as well as the notion ‘discourse categories’.
2.1. On defining discourse

As is well known, discourse is one of the notions that are used rather
polysemically in the linguistic (and the paralinguistic) literature. I do not
intend, here, to explore the different meanings that the term ‘discourse’
covers in traditional and modern linguistic theories. Rather I will restrict
myself to the use of this term within the FG community. According to
Mackenzie and Keizer (1991), ‘discourse’ is understood in two main ways:
it may be seen as (a) the product of text-creating activity, or (b) the ongo-
ing text-creating process itself. They rightly point out that FG takes the
former view, and in current FG literature ‘discourse’ is indeed being used
primarily to refer to supra-sentential (textual) stretches (cf. Dik 1997b:
379).
In the remainder of this chapter, I adopt the ‘product of text-creating ac-
tivity’ view but with the difference that I suggest that we call ‘discourse’
any ‘complete communicative unit’, i.e. any utterance fulfilling a commu-
nicative purpose in a given setting or, in other words, any utterance with
both a content and a communicative intention. According to this concep-
tion, any utterance (be it a single word or a mere interjection) can be taken
as a discourse if and only if it serves to achieve a certain communicative
goal, whatever its length, as we will see below.
Discourse structure, the generalized parallelism hypothesis, and FG 301
Worthy of notice is that, in this view, the concept of discourse covers

not only the linguistic expression proper but also its contextual correlates.
In this sense, we can say that FG, being a pragmatically oriented grammar,
has always been a discourse grammar although it has been, in general, ac-
tually restricted to sentential discourses. This means that the problem now
being extensively discussed in the FG literature should be reformulated: the
required extension should take place not from a sentence FG to a discourse
FG but rather from a sentential discourse FG to a textual discourse FG.
2.2. Discourse categorization

If we adopt the view advocated above, we can distinguish between four
discourse categories: text, clause, term-phrase and word, yielding:
TEXT
(1) DISCOURSE = CLAUSE
TERM-PHRASE
WORD
The discourse categorization visualized in (1) calls for a number of pre-

liminary remarks. Firstly, the term ‘text’ is taken here to designate any
coherent combination of (sequences of) simple or complex clauses. In this
sense, it replaces the traditional term ‘discourse’, which will henceforth be
used as a generic term referring, as mentioned above, to any complete
communicative unit (including a text).
Secondly, as rightly pointed out by Dik (1997b: 379), Natural Language
Users (hereafter NLUs) typically communicate with each other by means
of full texts. In these cases, hierarchical relationships arise between the
four discourse categories, as visualized in the following hierarchy, which
we may call the Discourse Categories Hierarchy (DCH):
(2) Discourse Categories Hierarchy (DCH):

Text > Clause > Term-phrase > Word
However, NLUs can also communicate by using single (simple or com-

plex) clauses:
(3) a. It's cold in here.

b. Can you pass the salt, please?
And perfect communication can take place with one-constituent utter-

ances such as the following:
(4) a. Some tea!

b. Cheers!
As shown in Moutaouakil (1993, 1996) and extensively demonstrated in

Mackenzie (1998), ‘holophrastic’ constructions like (4a–b), contrary to
what is commonly asserted, are complete stretches of discourse rather than
reduced sentences.
In sum, Clause, Term-phrase and Word may display two different uses:
‘integrated’ and ‘free’ uses. In the former case, they organizationally func-
tion as shown in hierarchy (2); in the latter case, they behave as
autonomous complete discourse units. It will be shown in Section 3 that the
difference between the two uses is reflected in the internal structure of
these discourse categories.
A third preliminary remark concerns a particular type of constructions
referred to as ‘Extraclausal constituent + Clause’ constructions in FG. Here
are some illustrative examples :
(5) a. Mary, can you help me?

b. Well, we can now continue the lecture.
c. As for John, he will come tomorrow.
d. She is a nice girl, your neighbour.
The problem constructions like (5a–d) pose relates to their status. In this
respect, two approaches can, it seems to me, be suggested. First, one can
conceive of the construction at hand as a distinct full-fledged discourse
category which one may call ‘Expression’ (Cuvalay 1997) or ‘Sentence’
(Moutaouakil 1988, Dik 1997b). This is indeed, as far as I know, the posi-
tion commonly taken so far in the FG community (cf. Dik 1978, 1989,
1997b; Moutaouakil 1988, 1989, 1998; Cuvalay 1997, among others). If
one adopts this approach, one can position this discourse category ‘Sen-
tence’ between text and clause, as a supra-clausal entity, which yields the
following alternative DCH:
(6) Discourse Categories Hierarchy (DCH)

Text > Sentence > Clause > Term-phrase > Word
Second, a closer look at the pragmatic and structural properties of extra-

clausal constituents, as discussed in Dik (1997b), reveals that they are not
necessarily associated with clauses. Rather, they may co-occur with any
discourse category, i.e. with clauses or whole texts, but also with term-
phrases or words. Moreover, some extraclausal constituents seem to co-
occur only with discourse categories other than clauses. To take an exam-
ple, ‘topic shifters’, ‘push and pop markers’ and ‘finalizers’ typically
function as textual (more particularly conversational) discourse markers.
Within an approach that takes these facts into account, it becomes difficult
to speak of ‘Sentence’ as a distinct discourse category.
Leaving the question open, I will assume in the rest of this chapter that
both approaches are tenable and that hierarchies (2) and (6) are thus
equally relevant. For the sake of simplicity, however, I will refer by DCH
to hierarchy (2) rather than to hierarchy (6).
A fourth and final remark is that a text can be divided into sub-parts that
are commonly called, depending on the text type, ‘episodes’, ‘passages’,
‘moves, etc. Since these result from a strictly internal sub-categorization of
text and otherwise have no existence, these entities cannot be taken as full-
fledged discourse categories. That is why they are not mentioned in DCH.
3. Discourse Structure
One of the recent tendencies in FG has been to try and establish a structural
parallelism between the different discourse categories. In the remainder of
this chapter, I will show that one of the logical endpoints of such a ten-
dency is the postulation of a universal abstract archetypal discourse
structure whose actualization in these discourse categories takes place ac-
cording to certain parameters.
3.1. Generalized Parallelism Hypothesis

By the ‘Generalized Parallelism Hypothesis’ (GPH), I refer to the assump-
tion that there is a certain structural parallelism between all the different
discourse categories discussed above. In the evolution of this idea, two
types of structural parallelism may be distinguished which are perceptible
in the last decade of FG work: a surface parallelism and an underlying par-
allelism.
3.1.1. Surface structural parallelism
The source of the assumption that certain discourse categories display a

surface formal similarity can, in my opinion, be traced back to Dik's (1989)
formulation of ordering principles (especially the ‘Principle of Cross-
Domain Harmony’) which he conceives of as governing word order within
both the clause and the term-phrase.
In a similar vein, De Groot (1990) argues for an approach based on the
idea that word structure (mainly the position of affixes with respect to the
stem) can be described in the same way as clause structure, i.e. in terms of
placement rules, ordering principles and patterns.
3.1.2. Underlying structural parallelism
As for the assumption of underlying structural parallelism, we may say that

it has been developed and extended in three steps.
In his insightful work on the Noun Phrase, Rijkhoff (1992) puts forward
the idea that Predication and Term display a structural parallelism to the
extent that they both consist of three layers: a quality layer, a quantity layer
and a locality layer. According to Rijkhoff's analysis, the common structure
of Predication and Term can be represented as follows:
(7) [Loc-Op [Quant-Op [Qual-Op [Nucleus] Qual-Sat] Quant-Sat] Loc-Sat]
In Moutaouakil (1993), the parallelism assumption is pushed a step fur-

ther. Starting from constructions like (8a–b) and (9) whose peculiarity is to
exhibit modalized terms (i.e. terms with subjective, exclamative or voli-
tional modality), Moutaouakil observes that a parallelism can also be
established between the term and the layer at which this type of modality is
supposedly located, i.e. the proposition layer:
(8) a. French Marie travaille avec quel enthousiasme!

‘With what enthusiasm Mary works!’
b. What a nice girl I saw yesterday!
(9) a. Standard Moroccan Arabic kana l-marhumu kariman

was the blessed-nom generous-acc
‘The hopefully to-be-blessed was generous.’
He then suggests adding to the existing term layers a fourth one which
he calls the ‘modality layer’. This yields structure (10), assumed to under-
lie both Term and Proposition :
(10)
[Mod-Op Mod-Sat]
[Loc-Op [Quant-Op [Qual-Op [Nucleus] Qual-Sat] Quant-Sat] Loc-Sat]
The next and final step in the development of the parallelism hypothesis
comes with Dik’s (1997b) proposal that Clause and Text (Discourse in his
terminology) can be said to display a strong similarity, both on the struc-
tural level and the relational level. The basic idea is that intra-clausal
layered structure as well as intra-clausal functional relations can be pro-
jected onto Text. According to this view, the structure which can be taken
as common to both Clause and Text may be represented as follows:
(11)
[Ill-Op [Mod-Op Mod-Sat] Ill-Sat]
It should be noted, finally, that the works (Hengeveld 1997, Cuvalay

1997 and Moutaouakil 1998 among others) in which the suggestion is
made to deal with supra-clausal phenomena in terms of an expanding up-
ward-layering approach (where Text is conceived of as a supra-clausal, or
indeed a supra-sentential, layer rather than constituting a separate module)
can somehow also be regarded as contributions to the development and the
extension of GPH in the sense that they all postulate, though to different
degrees, a certain structural similarity between clause, sentence and text.
3.2. Archetypal Discourse Structure

Before it can be taken as underlying the various discourse categories dis-
cussed in the previous section, structure (11), requires, it seems to me,
some further refinements and enrichments on both the constituency and the
relational levels. Here are some suggestions.
According to the typology proposed in Dik (1997b: 384–386), greetings
and leave-takings, summonses, addresses and certain polite expressions

such as those exemplified in (12a–d) form a subclass of extraclausal con-
stituents whose function is to manage interaction, to create and maintain
“the interactional conditions which must be fulfilled for a discourse event
to be implemented”:
(12) a. Good morning, sir, what time is it?

b. Hey there, where are you going?
c. John, can you give me your pencil?
d. Excuse me, could I have a cigarette?
We can also consider as pertaining to this subclass expressions like You

know, listen, look etc., which function, in their grammaticalized use (as in
(13) for instance), as mere interactional devices:
You know
(13) Listen , let's go to the theatre this evening.
Look
It is clear that, given the status and the function of these expressions,
structure (11), as it stands, cannot handle them: they can be located in none
of the five available layers. In order to do justice to this kind of expression,
I would suggest enriching structure (11) with a third interpersonal layer
which we may call the ‘Interactional layer’. Such an enrichment results in
the following structure:
(14)
[Inter-Op [Ill-Op [Mod-Op Mod-Sat]Ill-Sat]Inter-Sat]
Some remarks are in order here. First, the added constituent is a real full
layer in the sense that it involves, as do the other layers, both an interac-
tional operator (Inter-Op) and an interactional satellite (Inter-Sat), slots
intended to host particles like Hi!, Hey! as well as the quasi-
grammaticalized expressions exemplified in (12). Second, in structure (14),
the interactional layer is the outermost and highest one, taking all the other
layers in its scope. This is indeed what the data suggest, witness the follow-
ing contrast:
(15) a. Hey there, frankly, I don't like your hat!

b. *Frankly, hey there, I don't like your hat!
It becomes clear from (15a–b) that interactional expressions preferably oc-

cupy the absolute initial position preceding all the other constituents
including illocutionary satellites (such as frankly), which corroborates the
assumption that they pertain to a layer higher than the illocutionary one.
In the discussion of the extension of FG, the problem of the number and
the nature of the relevant units (or layers) has received more attention than
the relations holding between these units. Yet the notion ‘structure’ im-
plies, as is well known, constituency as well as relationships. In order to
give a complete and precise picture of discourse structure, we must also
take into account the various relations it involves.
In FG, relational structure subsumes two main types of relations: func-
tional and referential relations. Since the FG definition and characterization
of these two types of relations are largely well known, I will not go into
this matter in any more detail. Let us only keep in mind the following three
salient facts. First, the two types of relations differ from each other in the
sense that functional relations hold within the discourse structure itself, i.e.
between its constituents, whereas referential relations obtain between the
structure as a whole and what it refers to. We can speak, therefore, of
strictly ‘internal’ relations as opposed to ‘external’ relations. Second, as re-
peatedly pointed out and emphasized in Dik (1997a and 1997b), the world
within which the discourse refers is a mental ‘Discourse Model’ which
represents either S's conceptualization of the outside world or a purely
fictitious world. Third, as we will see below, not all functional relations are
involved in all the discourse categories distinguished so far.
Still on constituency, Dik (1997b: 415–422) draws attention to the im-
portance of “the global discourse decisions” that S takes in building up a
discourse, such as choosing a discourse type, a discourse style, a discourse
world, etc. The importance of such decisions, as Dik convincingly shows,
lies in the fact that they co-determine the internal structure of (the sub-part
of) the discourse they take in their scope. This obviously means that these
decisions must be represented in the underlying discourse structure. As re-
gards the way in which this can be done, I think that the following general
reflections may be of some relevance.
The exhaustive list of the discourse decisions in question, as established
by Dik, includes those which relate to the choice of discourse event (in par-
ticular the deictic centre), discourse genre (conversational, narrative,
argumentative, etc.), discourse style (formal/informal, polite/familiar, etc.),
discourse world, discourse illocution, discourse tense and discourse topics.

In the section he devotes to discourse coherence, Dik (1997b: 433–435)
mentions ‘Frame’, which he describes as the “organized knowledge con-
cerning what can be done and said within a given institutional setting”.
Given this definition, it is possible, it seems to me, to consider Frame as a
special, highly institutionalized discourse type. This reinterpretation is sup-
ported by the fact that Dik (1997b: 435) exemplifies this notion with a
sonnet, which he mentions as a discourse genre. In the same vein, given the
close interdependence between discourse genre and discourse world (cf.
Dik 1997b: 418) and in order to reduce the number of layers to a minimum,
these two notions can be collapsed into the generic notion ‘discourse type’
and taken as forming together one and the same layer. This collapsing is
supported by the fact that Dik (1997b: 422) does not mention discourse
world when he speaks about “the highest brackets”, which he restricts to
discourse event, discourse genre and discourse style.
In Dik's view, all discourse decisions are subsumed by a single generic
category of ‘discourse settings’. As regards underlying representation, it is
worthy of notice that some of the decisions in question can be coded in the
already existing layers of structure (14), as is the case for illocutionary,
temporal and topical decisions. Consequently, only the remaining features,
i.e. discourse event, discourse type and discourse style, can be included in
the category settings. The values this category can take are, thus, the fol-
lowing:
EVENT
(16) SETTING = TYPE
STYLE
In fact, (16) does not tell the whole story about the underlying represen-
tation of the category at hand, for two major problems remain to be solved.
They can be formulated as follows: first, where must SETTING be located
in the underlying discourse structure; second, what is its exact status there?
Concerning the location problem, it is clear that SETTING should stand at
the opening of this structure. This is indeed Dik's view. In (1997b: 422) he
conceives of the features represented in (16) as opening up “the highest
brackets” involving the discourse as a whole. As for the problem of the
status of SETTING, I think that one of the most reasonable ways to deal
with it is by adding a third level – which we may call by default the “rhe-
torical level” – to the existing ones, i.e. the Interpersonal and the
Representational levels. At this level, three – at least – hierarchically or-
ganized layers may be distinguished, in descending order for the discourse

event, the discourse type and the discourse style. With the addition of this
level the full underlying discourse structure becomes something like (17):
(17) Archetypal Discourse Structure (ADS)
Rhetorical level [Ev-Op [Typ-Op [Sty-Op Sty-Sat] Typ-Sat] Ev-Sat]
Interpersonal [Inter-Op [Ill-Op [Mod-Op Mod-Sat]Ill-Sat]Inter-Sat]

level
Representational [Loc-Op[Quant-Op[Qual-Op [Nucleus] Qual-Sat]Quant-Sat]Loc-Sat]

level
The main argument in favour of such an approach is that the three sub-
categories of SETTING codetermine, as mentioned above, the form as well
as the content of the subsequent discourse and should therefore be consid-
ered to function as operators just like the operators of the other, well
established layers. In the same vein, although it seems to be relatively less
easy to find linguistic expressions which function as discourse type or dis-
course style satellites, we cannot conclude that satellites of this kind do not
exist.1 In any case, satellites are optional constituents and their existence
can therefore not be taken as a crucial defining feature of layerhood.
Dik (1997b) points out that the discourse decisions under discussion
here at hand are ‘global’ in the sense that they affect not a single clause but
rather a whole text. This means that the added rhetorical level – or at least
some of its layers – is typically a textual level. We will return to this issue
in Section 3, where we will discuss the parameters regulating the actualiza-
tion of structure (17).
3.3. Unified representation or modular representation

As regards the underlying representation of the discourse structure dis-
cussed above, we can opt for one of two procedures, which I propose to
label the ‘unified procedure’ and the ‘modular procedure’.
According to the procedure adopted so far in the FG tradition, all the
(semantic and pragmatic) properties are coded in the same underlying
structure, which is represented in the Grammatical Module of the Model of
Natural Language User (MNLU). It is this structure that Expression Rules
operate on. On the other hand, according to a recent tendency (cf. Kroon
1997, Bolkestein 1998, Vet 1998, Van den Berg 1998, Liedtke 1998), a
separate module should be added to the MNLU. In fact, I think that there is
some ambiguity on the nature and the function of this additional module,
probably attributable to the fact that ‘discourse’ and ‘pragmatics’ are not
always systematically differentiated. In Kroon’s view, it is conceived of, as
far as I can judge, as a ‘discourse module’ (a text module in our terminol-
ogy) intended to handle supra-sentential phenomena. In this sense, it stands
in contrast with a ‘sentence module’ whose task it is to take care of prop-
erly intra-sentential phenomena. In Vet’s view, the added module is quite
different. It is a pragmatic module intended to deal with the contextually
determined properties of linguistic expressions – whatever their length, and
even extending to whole texts) – and in particular with speech acts (or non-
literal illocutions).
In Moutaouakil (1999), a proposal along the lines of Vet’s view is made
which can be further developed as follows. We can say that the two repre-
sentational procedures described above are theoretically equivalent in the
sense that they both fit in with the principles and the organization of FG. It
is clear that the definition and the categorization of discourse opted for
here, in particular when viewed in the light of GPH, leave no room for the
distinction between a ‘discourse (or text) module’ and a ‘sentence module’.
All the discourse categories distinguished in hierarchy (2) (or hierarchy
(6)) are dealt with in the same way, according to either the unified or the
modular procedure. In the latter case, once the added module is understood
as a pragmatic (and not exclusively a textual) module, the opposition be-
tween ‘upward layering’ and modularity is neutralized and the two
approaches can thus go perfectly hand in hand.
Two kinds of organization of the MNLU are possible. In one of these,
two modules contribute together to taking care of the relevant underlying
properties of any complete communicative unit whatever its category (text,
clause, term-phrase etc.). In the grammatical module, a (grammatical) un-
derlying structure represents the semantic and structural features, whereas
in the pragmatic module (in the sense of Vet) a (pragmatic) underlying
structure encodes the contextual speaker-oriented features such as illocu-
tionary force, subjective modality and pragmatic functions. These two
underlying structures are taken together as inputs, possibly with underlying
structures from other (epistemic, logical, social etc.) modules, to expres-
sion rules which deliver the final syntactic form.
MNLU
PRAGMATIC MODULE GRAMMATICAL MODULE
Grammatical
Structure
Pragmatic structure
Expression rules
Syntactic form
Figure 1. One organization of the MNLU
MNLU
PRAGMATIC MODULE SEMANTIC MODULE
Pragmatic Semantic structure
Structure
GRAMMATICAL MODULE
Expression rules
Syntactic
Syntactic form
form
Figure 2. An alternative organization of the MNLU

Alternatively, pure semantic properties (semantic and perspectivizing

functions, operators and satellites pertaining to the representational level,
etc.) can be represented in a separate semantic module. In this case, the ex-
pression rules constituting an autonomous grammatical module operate on
the information coded in the two underlying structures represented in the
pragmatic and the semantic modules. These two types of organization are
visualized in Figures (1) and (2) respectively.
Notice that in Hengeveld’s proposal (this volume), the representational
and the interpersonal entities are regarded as hierarchically organized
modules whereas here they are assumed to be, together with the rhetorical
entity, merely layered levels of the same structure which are located in
separate but related modules.
4. Archetypal Discourse Structure Actualization

Structure (17) can be said to be ‘archetypal’ in the sense that it is postu-
lated as a generic abstract structure whose actualization in stretches of
discourse is regulated, as we will assume, by – at least – a structural and a
typological parameter.
4.1. The structural parameter

As was pointed out above, discourse categories can have two kinds of use,
‘free’ and ‘embedded’. It will be shown here that the actualization of struc-
ture (17) in these discourse categories depends on the way in which they
are used.
4.1.1. ADS actualization in free use
The basic idea which will be developed in what follows is that Arche-
typal Discourse Structure (ADS) is fully actualized in texts, whereas its
explicit actualization in the lower discourse categories is dependent on the
(decreasing) hospitality of these categories.
Examination of the internal structure of the discourse categories distin-
guished so far shows that, in their free (non-embedded) occurrences, they
differ from each other in their capacity to host ADS. As one might expect,
hosting capacity diminishes as we run from the top to the bottom of the
DCH. The text is thus the most hospitable category and the word is the least
hospitable. In fact, one may say that ADS is potentially present in any
normal communicative event, whatever the discourse category through
which it is mediated. The difference between the discourse categories in
question resides, therefore, rather in how they house ADS. In general, the
full, explicit actualization of ADS takes place in the text; in the other dis-
course categories, only a part of ADS is expressed, the other part being
implicitly taken care of by the context/situation.
As may be deduced from the characterization of verbal interaction given
in Dik (1997b: 409), the discourse category through which a complete
communicative event takes place in the most explicit way is the text. This
is why NLUs speak in texts rather than in isolated sentences. We may thus
expect the full actualization of structure (17) to obtain optimally in a whole
text. This indeed emerges clearly from the contrastive examination of three
typologically quite different texts: Brunhoff's fairy-tale The story of Babar
the little elephant (already analysed from the point of view of Topic-Focus
assignment by Mackenzie and Keizer 1991), Najib Mahfouz's novel Han
Al-Halili ('Al-Halili Quarter') and Diderot's Jacques le fataliste.
It should be noticed here that the other discourse categories (clause,
term-phrase and word), as pointed out in many works (Dik 1997a, Mou-
taouakil 1993, and, in particular, Mackenzie 1998), cannot be said to be
less communicative. They also can be used to carry a complete piece of in-
formation, with the difference, however, that part of this information
remains implicit, i.e. is transmitted by contextual and/or situational means.
In other words, the difference between the discourse categories at hand
does not reside in their communicative capacity proper but rather in the de-
gree of explicitness of their communicativity.
Concerning the actualization of structure (17) in the discourse catego-
ries other than text (clause, term-phrase and word), data suggest that it
becomes more restricted as we run from the top to the bottom of DCH. As
regards this restriction, it will become clear through the following
examination of the internal structure of the clause, the term and the word
(where they stand alone as performing autonomous and complete speech
events) that (a) this is, in fact, a consequence of the decreasing hosting
capacity of these discourse categories, (b) it may affect the layering, the
operator values and/or the relations and (c) it operates according to a
certain directionality, i.e. from higher to lower levels and layers.
When used as a complete discourse unit, a clause can generally display
all the layers of the interpersonal and representational levels as well as,
though to a lesser extent, as (a) and (c) predict, those of the rhetorical level
(which typically occur in larger stretches of discourse). As regards func-
tional relations, pragmatic functions are clearly text-based notions (cf. Dik
1997a: 309–338), although they may be expressed within the clause.
As shown above, term structure can be said to run parallel to clause
structure. However, the layering of term structure typically does not go be-
yond the lower layer in the interpersonal level, i.e. the modal layer. It is
hard to speak of a ‘term illocutionary layer’, at least as far as explicit term
structure is concerned. The values of the modal operator are more restricted
in the term than they are in the clause: I have shown elsewhere (Mou-
taouakil 1993) that only volitional and emotional subjective modality
distinctions can occur in a modalized term (excluding evidential distinc-
tions).
In languages with a concatenative morphology, word formation is
achieved, as is well known, through an affixation (prefixation, suffixation,
infixation or circumfixation) rule which applies to a stem. Approached in
terms of the GPH, word structure can be conceived of as a partial actualiza-
tion of structure (17): on a stem standing as a nucleus, prefixes and suffixes
build hierarchically organized layers. As one might expect, the actualiza-
tion of structure (17) in the word is more restricted than it is in the other
discourse categories. First, affixes can express all the features of the repre-
sentational level through a quality layer (Republican, Chinese, panelling,
anti-social etc.), a quantity layer (hypercritical, supernatural, overdressed,
subhuman etc.), and a locality layer (subway, transatlantic, pre-marital,
post-classical, unfair, etc.). Word layering can attain the interpersonal
level, as in the case of words with pejorative affixes such as: (misleading,
malodorous etc.). However, it cannot go beyond the modal layer: I am
aware of no affixes which can be said to express illocutionary features.
Second, not all the operator values may obtain at the word level. For in-
stance, the modal values are restricted, it seems to me, to pejorative
features. Third and more importantly, words do not have, as one might ex-
pect, the same hosting capacity as terms and clauses. A complex word can
hardly contain more than two prefixes. This means that it would be diffi-
cult to find words involving more than two layers at once.
A final word on the text-clause parallelism: the structural similarity be-
tween the clause and the text is commonly described in terms of the
projection of the structure of the former onto the latter. The facts examined
above clearly show that the projection process – if we can speak of such a
process at all – should be conceived of as taking place in the reverse direc-
tion, i.e. from the text to the clause. With the approach proposed here we
no longer need the notion of projection: the similarity between the different
discourse categories is not due to a structural projection from one category
to another; it rather resides in the fact that they all borrow their internal or-
ganization from one and the same archetypal structure. It seems to me that
the parallelism phenomena at hand can be accounted for in a psychologi-
cally more adequate way in the light of the assumption that NLUs organize
their discourse according to a common archetypal communication structure
than if we consider that they do so by projecting the structure of a given
discourse category, especially since there is no agreement about the dis-
course category which is the source of this projection (clause or text).
Interestingly enough, this permits us to avoid the undesirable ‘clause-
centricity’ of FG.
4.1.2. ADS actualization in integrated use
As mentioned above, the discourse categories lower than the text can be
used in two ways: they can constitute complete communicative units or oc-
cur as (embedded) parts of larger categories. In the latter case, words can
constitute parts of terms which stand as parts of a clause; a clause can be
embedded in another clause yielding a complex clause; (simple and com-
plex) clauses, when coherently sequenced, form a text. In the previous sub-
section, we were concerned with the free use of the discourse categories;
our aim in what follows is to examine the way in which ADS is actualized
in these discourse categories when they are integrated into each other, fo-
cusing particularly on the embedding of clauses into clauses and clauses
into texts.
In the FG framework, several works have been devoted to embedding
phenomena (cf. Bolkestein 1990, Moutaouakil 1987, Hengeveld 1996 and
Dik 1997b, among others). Once reinterpreted in terms of the actualization
of ADS, the basic ideas advocated in these works can be reduced to the fol-
lowing general assumptions:
(a) The actualization of ADS is more restricted in integrated (embedded)

parts than it is in integrating (embedding) parts;
(b) Integrated parts ‘inherit’ features (layers or operator values) from in-
tegrating parts;
(c) The inheritance process is a matter of degree.
As regards assumption (a), it is now well established in FG circles that

the subordinate part of a complex clause need not be a complete clause: as
argued in Hengeveld (1996), it can be a clause or a proposition or even a
predication, depending on the nature of the superordinate predicate or on

the semantic function it bears.
Assumption (b) is in fact a generalization of the claim advanced in Dik
(1997b: 421) that “the Tense operator specifications on successive clauses
are inherited from the temporal value which has been fixed for the whole
discourse unit”. In fact, successive clauses forming the same (phase of a)
text inherit not only the temporal value but all the operator values (illocu-
tionary, modal, aspectual etc.) globally fixed for this (phase of the) text.
So, the schema which represents the temporal specifications inheritance in
Dik (1997b: 421) becomes, once it has been generalized, something like
(18) :
(18) α((X). (Y). (Z)...)
where α stands for any operator value and where X, Y, Z symbolize a se-
quence of clauses.
Assumption (c) means that the inheritance process can be partial or to-
tal. In other words, the embedded parts can inherit either some operator
values only or all of them. In the former case, the embedded parts (i.e. the
clauses of a text for example) can be ultimately reduced, at the underlying
structure level, to nothing but their nuclear predications. So, the nucleus of
(a phase of) a highly homogeneous text, i.e. a text with neither deictic-
centre change nor discourse-type shift, can consist of merely the set of the
nuclear predications of its constituent clauses.
The re-examination of embedding phenomena in the light of the param-
eterized actualization of ADS allows us to approach these on a new basis.
First, it becomes possible to redefine embedding as a linkage of two dis-
course units U and U' where a part of the ADS as actualized in U' is
specified in U. To put it another way, U' is embedded in U if some values
of the ADS actualized in U' are inherited from U. This definition permits us
also to handle those cases where embedding is not expressed by any formal
marker (i.e. without a subordinator).2 It furthermore makes it possible to
characterize coordination as a linkage of discourse units with actualization
of the same (part of) ADS. Second, it may be possible to establish a more
precise typology of embedded constructions on both quantitative and quali-
tative grounds: it becomes feasible to classify these constructions with
respect to the degree of embedding (which is calculated on the basis of the
number of the layers specified in the matrix) and to the values of the em-
bedded layers.
4.2. The typological parameter

The third parameter is ‘typological’ in the sense that it implies that the ac-
tualization of ADS takes place according to three factors: language type,
discourse type and discourse style.
4.2.1. ADS and language types
The examination of data from several typologically different languages

(namely English, French, Standard Modern Arabic as well as Moroccan,
Egyptian and Syrian colloquial Arabics) allows us to make the following
general observations. All the languages investigated display the three levels
distinguished in ADS as well as the layers these levels subsume. They dif-
fer from each other, however, with respect to the values of these layers.
Here are some examples illustrating the point.
Firstly, Modern Standard Arabic as well as the various Colloquial
Arabics examined seems to be more interpersonally oriented than the other
languages. The data suggest that Arabic displays a relatively greater num-
ber of vocative particles and expressions (cf. Moutaouakil 1989).
Moreover, it is a well known fact that “Egyptians always speak by implica-
ture” (and I think that this feature can be generalized to all Arabic
languages); if this observation is correct, we can say that Egyptian is a lan-
guage with a rich illocutionary layer – at least in comparison to the other
languages examined. We can say in the same vein that Arabic is one of the
more highly modalizing languages in the sense that it displays a rather rich
set of subjective modality values both on the clause and the term levels (cf.
Moutaouakil 1993, 1999).
Secondly, at the representational level, we often find in the Arabist lit-
erature the claim that Arabic, in comparison with Indo-European
languages, is an aspectual rather than a temporal language. If this claim
turns out to be tenable, we may hypothesize that there are languages with
rich aspectual quality and quantity layers in contrast with languages with a
rich temporal locality layer.
And as a last example, Dik (1997a: 181–182) points out that languages
differ in the number of demonstratives (from a minimum of two to a
maximum of well over twenty) as well as in the types of distinctions these
elements express. This again can be taken as a criterion with which to clas-
sify languages with respect to the locality layer in the term.
In addition, languages differ not only in the number and the kinds of the
layer values but also in the means by which these values are expressed.
Some languages use grammatical means; others use lexical ones. For in-
stance, in Arabic, illocutionary and subjective modality values are
mediated by morphological means (particles, special morphemes, etc.)
rather than lexically. Keeping in mind that grammatical and lexical means
are underlyingly represented by operators and satellites respectively and
generalizing this observation to all the layers involved in ADS, we may
speak of ‘operator-prominent languages’ in contrast to ‘satellite-prominent
languages’.
A large part of the FG literature has been devoted to the selection and
the assignment, in different languages, of perspectivizing functions as well
as to the study of Topic and Focus types and sub-types. The results of this
work can be reinterpreted in terms of typological actualizations of the rela-
tional part of ADS.
4.2.2. ADS and discourse types
The realization of ADS in an actual discourse is codetermined by the type

of that discourse. Let us concentrate on narrative discourse, whose main
distinguishing features (extensively discussed in Moutaouakil 1998) are as
follows.
In narrative texts, the interpersonal level, as one might expect, is much
less activated than it is in other text types. Consider, for example, that in a
‘pure’ narrative text with no deictic-centre change, the fixed global illocu-
tion is what Dik (1997b: 419) calls a “default Declarative Illocution”. What
is more, in this genre, implicature phenomena hardly occur. Likewise,
highly neutral narrative texts (‘récits’ in Benveniste’s 1966 terminology)
have the prominent feature that they rarely involve subjective modality dis-
tinctions: the narrator is, in this type of text, reduced to a mere ‘paper’
entity (to use Barthes’s 1976 metaphor).
At the representational level, on the other hand, the more frequently ac-
tivated temporal and aspectual values are Past and Perfective respectively
(unless the so-called ‘Historical Present’ is used). And as regards prag-
matic functions, given that the information conveyed in a narrative text is
typically new (unshared), the type of Focus selected is, as one might ex-
pect, Completive Focus.
4.2.3. ADS and discourse style
The notion ‘style’ is defined in Dik (1997b: 417) as “a sequence of choices

... of means of expression which have consistent values along such polar
dimensions as Formality-Informality, Politeness-Familiarity and Concise-
ness-Redundancy”. Dik points out that the choice of a given style is
strongly determined by the choice of the discourse type in the sense that
not all kinds of styles can fit in with all discourse types. As for the impact
of the style on the discourse, the definition given above seems to restrict it
to formal means of expression. In fact, if we represent the style as a layer
operator, it becomes clear that it plays, like any operator, a twofold role: on
the one hand, it combines with other operators to trigger the rules responsi-
ble for the formal realization of the discourse; on the other hand, it
codetermines the values of the operators it has in its scope, i.e. the opera-
tors of lower layers.
Let us take as an example the polar dimension Politeness-Familiarity.
The actualization and the formal realization of ADS in polite discourse in-
volve special operations such as the selection of particular interactional
devices, the extensive activation of the indirect illocutionary values, the
choice of specific lexical items as well as of certain forms of predicates and
terms. Notice that the three typological factors under discussion can inter-
act with each other in an interesting way in the sense that certain languages
handle certain discourse types and styles better than others do. The most
well known example is the case of so-called ‘polite languages’ such as, for
example, Japanese and, to a less extent, colloquial Egyptian.
4.2.4. ADS as a communicative universal
In the FG framework (cf. Dik 1997a: 7), linguistic universals are defined
within an approach which conceive of natural languages as particular solu-
tions to the communication problem, i.e. to the problem of establishing
“high-level communicative relationships between human beings”. It is
clear that ADS can be considered as one of the common linguistic proper-
ties which fit optimally into this functional conception of linguistic
universals. Notice that, given its nature, its components and its organiza-
tion, ADS is compatible with functional theories in which universals are
conceived of as Formal-Functional correlations.
The following arguments can be taken as militating in favour of the
universality of ADS. Firstly, structure (17) reflects, as pointed out above,
the main components and the organization of the communication event it-
self, which typically involves three operations: (a) representing some State
of Affairs occurring in some possible world and constituting the message
that S intends to communicate; (b) establishing relationships between S and
A on the one hand and between S and the content of this message on the
other; (c) choosing the discourse type and the discourse style in which (a)
and (b) are to be delivered. Secondly, from the FG point of view, the ac-
quisition of language is approached, as is pointed out in Dik (1997a: 7), in
terms of “its development in communicative interaction between the
maturing child and its environment”. Here again, the notion
‘communicative interaction’ is central. This would enable us to think of
ADS as representing a relatively advanced stage of the child’s progressive
mastery of linguistic communicative abilities. And thirdly, it becomes clear
from the facts discussed in the previous sub-sections that ADS can be
assumed to underlie, with quantitative and qualitative parametrical
variance, natural discourses of various categories, types and styles in
different types of languages (and probably all language types). Moreover,
if we are able to assume that any communicative process involves the three
levels of ADS, it becomes not unreasonable to hypothesize that a structure
with similar components and organization is also at work in non-verbal
(pictorial, musical,3 etc.) communication systems.
In sum, ADS can be viewed as a universal structure in which language
types make quantitative and/or qualitative choices. In another, stronger
formulation, language types partially result from different choices in ADS.
Within each language type, the actualization of this structure is regulated
by further factors relating to discourse type, discourse style and discourse
category.
5. Conclusions
One of the possible solutions that can be proposed to the problem of ex-
tending the current clause/sentence FG model and transforming it into a
more text-oriented model is built on four related assumptions that presup-
pose each other: (a) that the notion ‘discourse’ covers all kinds of
utterances expressing a complete communicative event, (b) that discourse
can be formally mediated through four formal categories: text, clause,
term-phrase and word (or perhaps five if we take sentence as a full-fledged
discourse category), (c) that these categories are underlain, at different de-
grees, by one and the same archetypal structure, and (d) that this
hierarchically organized structure fits perfectly into a modular FG. This so-
lution can be said to have the following major advantages. First of all, it
permits us to unify the FG approach to the different discourse categories.
Second, it makes it possible to avoid two theoretically undesirable options:
(a) an unnecessarily recourse to other theories and (b) a costly and not al-
ways very convincingly justified multiplication of the modules within
MNLU on the other hand. Third, it allows us to redefine and extend the
central notion of embedding and to further refine the typology of embed-
ded clauses. Fourth, the perhaps most important gain is that this solution
provides the universal part of FG with a tool permitting us to describe and
explain in a more principled and unified way the similarities and the differ-
ences between natural language types as well as those between types of
natural discourse. Moreover, Dik’s (1997b: 415) insightful idea that a text
“can in many ways be likened to a piece of music” leads to the assumption
– hopefully to be verified by further research – that, if the transposability of
ADS turns out to be feasible, the solution proposed here can be viewed as
paving the way for a General Functional Theory (GFT) whose main task
would be to describe the general structure of the communicative process
and to account for its actualization in verbal and non-verbal modes of in-
teraction as well as providing particular theories for the various
communicative systems. Within such a general theory, FG would stand as
a member of a subset of linguistic theories to be compared to and evaluated
in relation with the theories of the non-linguistic subset. In this view, ADS
would be taken as the actualization in natural language of a more abstract
archetypal communication structure constituting one of the primitives of
this all-encompassing functional theory.
In this study, we have mainly been concerned with the underlying
(pragmatic and semantic) side of the structural parallelism between the dis-
course categories. I think that it would be of great interest to verify, in
future research, the extent to which a similar parallelism can be said to
also hold (within the grammatical module) at the surface (morphosyntactic)
level as well as the extent to which the latter parallelism can be taken as
resulting from a projection of the former.
Notes
1. In arguing for the relevance of the discourse-type layer, Hengeveld (1997)
reports that in some languages (Turkish, Tauya and Krongo) certain verb
forms only occur in narrative texts. As for the discourse-style layer, its rele-
vance is evidenced, for example, by the fact that, in ‘polite’ languages such
as Japanese, politeness is coded by both grammatical and lexical means
which are to be underlyingly represented as Politeness operators and satel-
lites respectively.
2. By ‘non-formal embedding’, I refer to the kind of linkage that is found in
constructions like Paul said: “John is an excellent linguist.” where John is
an excellent linguist is an embedded clause functioning as a (Goal) second
argument of the matrix predicate said. More generally, this is the type of
embedding involved in (parts of) texts where successive clauses inherit
some of their (illocutionary, modal, temporal, etc.) features from globally
fixed values. It becomes clear that any formally oriented approach to em-
bedding phenomena fails to properly describe the structure of this type of
construction.
3. For the structural parallelism between a text and a piece of music, see Dik
(1997b: 415). In the same vein, J. Amjad (p.c.) has pointed out to me that
the structure of a musical composition is quite similar to the ADS advocated
here.
References
Barthes, Roland
1976 S/Z. Paris: Points.
Benveniste, Emile
1966 Problèmes de linguistique générale. Paris: Seuil.
Berg, Marinus van den
1990 Sentential complements in Functional Grammar: embedded predica-
tions, propositions, utterances in Latin. In: Jan Nuyts, A. Machtelt
Bolkestein and Co Vet (eds), Layers and Levels of Representation in
Language Theory: A Functional View, 71–100. Amsterdam and
ward (eds)
Cuvalay, Martine
1997 The Arabic Verb. A Functional Approach to Verbal Expressions in
Classical and Modern Arabic. Berlin and New York: Mouton de

Gruyter.
Dik, Simon C.
1978 Functional Grammar. Amsterdam: North-Holland.
1997a The Theory of Functional Grammar. Part 1: The Structure of the
1997b The Theory of Functional Grammar. Part 2: Complex and Derived
Groot, Casper de
1990 Morphology and the Typology of Expression Rules. In: Mike Han-
nay and Elseline Vester (eds), Working with Functional Grammar:
Descriptive and Computational Applications, 187–201. Dordrecht:
Foris.
Gulla, J. Atle
1997 Combining Functional Grammar and Rhetorical Structure Theory for
Discourse Representation. In: John H. Connolly, Roel M. Vismans,
Christopher S. Butler and Richard A. Gatward (eds), 75–89.
delphia: Benjamins.
Hengeveld, Kees
1996 The internal structure of adverbial clauses. In: Betty Devriendt,
Louis Goossens and Johan van der Auwera (eds), Complex struc-
tures: A functionalist perspective, 119–147. Berlin and New York:
Mouton de Gruyter.
1997 Cohesion in Functional Grammar. In: John H. Connolly, Roel M.
Vismans, Christopher S. Butler and Richard A. Gatward (eds), 11–
16.
Kroon, Caroline
Liedtke, Frantz
and Machtelt Bolkestein (eds), 107–128.
Mackenzie, J. Lachlan and Keizer, M. Evelien
215.
Moutaouakil, Ahmed
1987 Term-predication in Arabic: A functional approach. Paper. Univer-
sity Mohamed V, Rabat.
1988 Essais en Grammaire Fonctionnelle. Rabat: SMER.
1989 Pragmatic Functions in a Functional Grammar of Arabic.
Dordrecht: Foris.
1993 Reflections on the layered underlying representation in Functional
Grammar. Paper, University Mohamed V, Rabat.
1996 On the layering of the underlying structure in FG. In: B. Devriendt,
L. Goossens and J. van der Auwera (eds), 201–227. Berlin and New
1998 Benveniste’s récit and discours as Discourse operators in Functional
Grammar. In: Mike Hannay and A. Machtelt Bolkestein (eds), Func-
tional Grammar and Verbal Interaction, 25–42. Amsterdam and
1999 Exclamation in Functional Grammar: Sentence type, illocution or
modality? Working Papers in Functional Grammar 69. Amsterdam:
University of Amsterdam.
forthc. Préliminaires à une théorie fonctionnelle de discours. Premier collo-
que sur la Grammaire Fonctionnelle. Mohammedia.
Rijkhoff, Jan
2002 The Noun Phrase. Oxford: Oxford University Press.
Vet, Co
1998 The multilayered structure of utterance: About illocution, modality
and discourse moves. In: Mike Hannay and A. Machtelt Bolkestein
(eds), 1–24.
Towards a speaker model of Functional
Grammar
Dik Bakker and Anna Siewierska
1. Introduction
According to the standard model of an FG grammar as presented in Dik
(1997: 58ff.) a linguistic utterance is constructed in three stages.1 First of
all the relevant lexical elements and terms are selected from the Fund, if
necessary after predicate and/or term formation (Stage I). Then from these
basic elements, an underlying representation is constructed in an inside-out
fashion, starting with the layer of the nuclear predicate and ending with the
pragmatic (clausal) layer. In the process, the corresponding operators and
semantic, syntactic and pragmatic functions are assigned to each layer
(Stage II). And finally, once this pragmatico-semantic construct has been
completed, this is converted into a prephonetic string by the expression
rules (Stage III).
The Functional Discourse Grammar model (FDG) as presented in
Hengeveld (this volume) proposes an alternative for the organization of
Stages I and II above, and indeed an extension of the latter. An important
distinction between FDG and the standard model is the separation of
pragmatic and semantic information into two separate levels. The prag-
matic top layer of the underlying representation in the standard model has
been separated from the now purely semantic Representational Level (RL)
and added as the inner layer to an organizationally higher and purely prag-
matic Interpersonal Level (IL). In addition, the interpersonal level has two
higher, discourse-related layers, viz. the Act and the Move. The informa-
tion contained in the IL and RL is constructed in a top-down, outside-in
fashion rather than the inside-out direction (upward layering) followed
throughout in the standard model. The result is that pragmatics goes before
326 Dik Bakker and Anna Siewierska
semantics, and semantics before expression, which is an improvement from

the perspective of the principle of functional explanation (Dik 1986). Fur-
thermore, the scope between the layers is respected in the sense that
choices are determined on the basis of scope relations.2 The relevant part of
the information on the interpersonal level, mainly reference and ascription,
is mapped onto the representational level, and is then expressed from there.
Purely pragmatic information is expressed directly from IL, without inter-
mediate semantic mapping. Although Hengeveld’s model does not discuss
the role of the lexicon in any detail, we assume that both the interpersonal
and the representational levels have access to it via the Cognition module.
As opposed to the pragmatic and semantic levels, the FDG model does
not contain any new proposals for the organization of the expression rules
(ER) in a direct sense; they simply map IL and RL entities onto the Expres-
sion Level (EL), i.e. the morphophonological string that will eventually
trigger articulation. However, several aspects of the model suggest that the
overall structure of the standard expression component should be reconsid-
ered in order to fit the rest of the model. Firstly, the expression rules now
take their input from two levels rather than just from the traditional fully
specified underlying representation, as is the case in the standard model.
This considerably complicates the interaction between the respective com-
ponents of the grammar. Secondly, the expression level communicates with
both the Cognition module and the Communicative Context. The Cognition
module contains the linguistic competence, i.e. the lexical and grammatical
knowledge of the language user, which is also represented in the standard
model, although there the grammar is mainly presupposed. However, this
module also represents communicative competence and all other informa-
tion needed during the formulation and encoding of the speaker’s intentions.
The interaction with the Communicative Context, not represented in the
standard model at all, should be considered vis-à-vis expression in more de-
tail as well. Hengeveld’s model suggests that the Communicative Context
module is one outlet for expressions. Thus, after having been uttered, ex-
pressions also go there in their raw form, i.e. in the shape of the output of
the expression rules. This is to take care of metalinguistic operations, where
anaphoric reference is made to forms rather than meanings. Finally, the
model seems to assume a hierarchical organization for the (pre-phonetic)
output of the expression rules, going from paragraph to sentence to clause to
the different types of constituents and further down. The traditional expres-
sion rules and their output lack any form of constituency at all.
In this chapter, we wish to consider in some detail the implications of
the grammar model proposed by Hengeveld, henceforth FDG, for the ex-
Towards a speaker model of FG 327
pression rules. We will take as our point of departure the proposals relating
to the expression component made in Bakker (2001) rather than the stan-
dard expression rules as presented in Dik (1997). The major aspects of this
dynamic model of expression will be presented in Section 2. Section 3 will
briefly discuss linguistic modelling in general. Furthermore, we give a
critical discussion of some aspects of the FDG model in this light. Sugges-
tions are given to clarify some of our points. In Section 4 we will look at
the implications of FDG for our revised expression rules. We will integrate
the two models, and simulate the production of a short stretch of spoken
discourse on the basis of this integral dynamic model. Finally, in Section 5
we draw some conclusions for the overall organization of an F(D)G gram-
mar and its further development.
2. Overview of the dynamic model of expression

In their standard form, the FG expression rules come in three phases. First,
starting out from a fully specified underlying representation, which con-
tains the predicates and the relevant functions and operators, the
grammatical elements are generated. This gives us both the independent
grammatical morphemes such as determiners, auxiliaries and prepositions,
and the respective kinds of bound morphemes, such as tense markers and
agreement suffixes. In the second phase, all lexical and morphological ma-
terial is ordered into a linear string. In the third phase this string is given its
final phonological shape.
In Bakker (1994; 1999) it is shown that there are two major shortcom-
ings to the way the expression rules are traditionally organized. Firstly the
expression component as it stands undergenerates, i.e. it cannot produce
certain forms that actually occur in languages. This endangers the descrip-
tive adequacy of an FG grammar in a rather obvious sense. A clear, and
ubiquitous example of undergeneration is provided by constructions in
which form and order interact. Examples (1) from Breton (Borsley and
Stephens 1989) and (2) from Dutch are a case in point.
(1) a. Ar vugale ne lenn-ont ket levrioù

the children PCL read-3PL not books
b. Ne lenn ket ar vugale levrioù
PCL read not the children books
‘The children do not read books.’
(2) a. Jij speel-t goed vandaag!

you play-2SG well today
‘You are playing well today!’
b. Vandaag speel je goed!
today play-ø you well
‘Today you are playing well!’
c. Ik zie dat jij vandaag goed speel-t!
I see that you today well play-2SG
‘I see that you are playing well today!’
In the (a) examples, with SV order, we find a 2nd person singular subject
agreement marker on the verb. In the (b) examples, with VS order, there is
no such marker; the verb appears as a bare stem. In the complement clause
of (2c), with its more or less fixed SOV order, the agreement marker is al-
ways expressed. If the expression rules first generate the grammatical
forms and only then establish linear order, as in the standard model, there
is no way to determine whether the suffix should be generated or not. The
only solution would be to look at the factors that determine the eventual
constituent order, in this case pragmatic aspects of the subject constituent,
the availability of another potential P1 filler, and the level of embedding.
But this would imply ‘prerunning’ the ordering rules in some way or
other.3
A second problem that we encountered with the expression rules in their
current form is that they overgenerate, i.e. they will produce all kinds of
forms that actually do not occur in any existing language. At first sight,
overgeneration seems to be a less serious problem than undergeneration:
we may simply assume that the theory will prevent such ‘impossible’ ex-
pressions from arising at all because they express ‘impossible’ underlying
representations. However, under the assumption that there is a niche in the
grammar which allows for a certain amount of autonomy within morpho-
syntax there might indeed exist certain (universal or typological)
constraints on what can be a well-formed expression in languages which
are not directly reducible to functional criteria (cf. Croft 1995). A linguistic
theory should make these constraints explicit, much as it should be descrip-
tively adequate in the more obvious sense of generating the forms that do
occur.4 But overgeneration is not only a theoretical issue. An overgenerat-
ing expression component puts into question the learnability of a FG
grammar, and as such endangers the cognitive adequacy of the theory.
In Bakker (2001) a model is proposed for the expression component
that tries to repair these shortcomings. In correspondence with the standard
model, the point of departure is a fully specified underlying representation

(UR), with all the predicates and operators present. The morphosyntactic
templates and the corresponding placement rules are also retained, albeit in
another guise. A major contrast with the standard approach is that our
model is dynamic rather than static in the sense that the top-down, left-to-
right order in which elements of the corresponding URs are expressed
plays a decisive role in determining the final shape of the expression. It is
shown that at least some of the problems that the original model runs into
can be solved by this dynamicity. It provides both a set of constraints on
what may and may not be found in expressions and a way of getting the
right information at the right place during expression. An almost logical
implication of this dynamicity is the integration of phase I (form genera-
tion) and phase II (linearization) into one process, interleaving form and
order operations.
Another aspect that distinguishes the dynamic model from the standard
approach is that the templates necessary for the expression of underlying
representations are combined into a tree-like structure as in Figure 1 below.
Although such a hierarchical organization of templates has always been
possible in theory – and may even have been implied; see also Hengeveld
(this volume: Figure 6) – it has never to our knowledge been explicitly in-
troduced as such, nor has any other form of constituent structure been
posited within FG. Although Figure 1 may be seen as the traditional (static)
representation of the constituent structure underlying the corresponding
expression, in our case it attempts to code the history of the process leading
to that expression. Thus, at any level in the tree, the corresponding expres-
sion pattern (‘template’) is selected on the basis of the actual information
contained in some specific part of the UR. The respective elements of that
part of the UR are then distributed over the slots of the pattern on the basis
of slot specific selection rules (‘placement rules’). For each (filled) slot in
the pattern this process is repeated recursively until no further distribution
is possible. This recursive process of slot filling and template selection is
represented statically in Figure 1 for the (simplified) underlying representa-
tion in (3).
(3) [Decl E1: [X1: [Pres e1: [Progr [expand [V]

(Def Pl x1: tree [N]: (Prox Sg x2: theory [N])POSS)ZERO,SUBJECT ] ] ] ] ]
‘The trees of this theory are expanding.’
In Figure 1 below, templates are found as a row of categories at the re-

spective levels of expansion. P1 VFIN VINFIN on the second row, for
example, is arguably an appropriate template for the clause level.5 The ap-
propriate template for the actual filler of the P1 position, the subject term in
this case, turns out to be DET NOUN NOMRESTR.
MAIN CLAUSE
P1 VFIN VINFIN
TERM(SUBJECT) [IMPRF,PRES,PL,3] VERB [IMPRF, GERUND]
/are/ /growing/
DET NOUN NOMRESTR
[DEF] [PL]
/the/ /trees/ TERM
ADPOS [PROX,SG] NOUN

[POSS]
/of/ /this/ /theory/
Figure 1. Expression tree with nested templates
This process of alternating template selection and slot filling provides

us with a tree in which functional categories (the ones in bold print) and
formal categories alternate. The process bottoms out in a row of nodes with
lexical and grammatical categories, such as NOUN and ADPOS, or a clus-
ter of grammatical elements, such as [PROX, SG]. By organizing the
distribution of information in this way, we attribute a dynamic character to
the process of tree construction, in which the sequence of slot filling, and
therefore time, plays a crucial role. This makes it possible to postpone the
determination of the final form of both predicates and grammatical ele-
ments until the moment at which they are eventually expressed. Elements
that have influence on the form of elements that will be expressed before
them, e.g. because they assign case to them, have to be available in time
only in as far as their assigning (functional) aspects are concerned, not nec-
essarily in their final form. Although it is not clear to us at what stage of

the expression process phonological details of the respective lexical and
grammatical entities are available, we will nevertheless give our terminal
elements a (pseudo)phonological shape indicated by forward slashes, but
will furthermore remain agnostic about the phonology.6 In underlying
clauses we will follow the practice of assuming that a more abstract repre-
sentation of predicates is present, indicated by the uses of single quotation
marks.
We will organize this process of dynamic tree construction by applying
the following principles. We would like to stress that the model almost
completely ignores the phonological level. Far from assuming that the
phonological module may simply be stuck on as a separate component to
the model presented here, we nevertheless think that the phonology might
be interleaved with the morphosyntax in the same fashion in which we in-
terleave syntax and morphology below.7
Principle 1: Constituent structures are developed top-down from the
material in the underlying representation. Thus, first the higher syntactic
constituents, such as noun phrases, are developed, then the individual free
morphemes, next the bound morphemes, and finally their phonological
form. In this way, and in contrast to the standard model, the explanatory
hierarchy of FG as discussed in Dik (1986) is respected, which runs from
syntax to morphology to phonology.
Principle 2: Development takes place from left to right. This is the
‘natural’ order in which linguistic forms are uttered in the first place. Left-
to-right ordering may be expected to shape and have shaped language over
time, at least to some extent, and more than any other order.
Principle 3: Development works depth-first. This means that of any two
contiguous elements to be expressed, at whatever syntactic or morphologi-
cal level, the leftmost one will be completely expanded up to its terminal
forms before the rightmost one is considered, and this applies recursively.
An implication of this is that only a fraction of the complete information
which is necessary for the production of the whole utterance will be avail-
able at any one time. This considerably reduces the problem for short-term
memory which is inherent to breadth-first development. In combination
with Principle 2, this may solve a number of problems of the type illus-
trated in example (1) above.
Principle 4: For any node N in the tree, all features found on a direct
path from that node to the top node are available for inheritance by N, at
least in principle. Trivially, this implies that overt primary operators are
available for every node in the tree. However, a distinction will be made
between ‘raw’ UR material (functions, π and ω operators, and inherent fea-

tures of predicates such as Animacy and Gender) on the one hand, and µ
operators on the other hand. In order to be accessible for inheritance, it will
be assumed that any UR feature has to be transformed into a µ operator,
either in a one-to-one or a more-to-one fashion, as in the case of portman-
teau operators. An implication of this is that only those features of URs
qualify as such if they are turned into a µ operator at some stage during ex-
pression. Apart from this there may be universal or language-(type)-
specific downward barriers that make features inaccessible to lower nodes.
Barriers to inheritance may be either functional or formal in nature, i.e. ei-
ther determined by a layer or substructure in the UR or determined by
some syntactic or morphological boundary, i.e. a specific node in the tree.
Principle 5: Operators may percolate, i.e. move upwards to higher
nodes. For percolation there may be universal and language-(type)-specific
upward barriers that make them inaccessible to higher nodes in the tree.
Barriers to percolation are only formal in nature.
These five principles should contribute to the (cognitive and typologi-
cal) adequacy of the dynamic model. They have two interesting implica-
tions for the global working of the model. On the one hand, linguistic
forms are produced precisely in the order in which they are uttered by the
speaker in a live setting, thus giving the model a real-time procedural fla-
vour. On the other hand, several constraints follow from the principles,
which restrict the formal power of the expression component. Indeed, it
may now be determined what information has to be available at what stage
in the tree expansion process. In this way, it may be determined more pre-
cisely what information necessarily has to come in from the UR as opposed
to arising in the course of expression, and thus, which elements have predi-
cational status in the language under consideration and which ones do not.
Furthermore, these constraints provide a means of determining the order in
which grammatical material has to be available in the first place. For ex-
ample, a case-marking postposition is needed on some node in the tree
before the elements on which the corresponding case is expressed are cre-
ated. Since grammatical forms may come into existence at several stages in
the expansion process, we also get an instrument to distinguish between
partially and fully grammaticalized elements. The former will appear rela-
tively early (‘high’) and the latter relatively late (‘low’) in the constituent
structure. Arguably, this distribution may reflect the length of their dia-
chronic path through the language, typically leading from a full predicate
to a fully grammatical entity, and their relative role in the syntax and mor-
phology.
The constraints implied by principles 1–5 above determine the informa-

tion flow through the expression tree that is dynamically constructed by
them. It is crucial what functional and formal information may be available
at which node in the tree during tree expansion. In order to accommodate
this information, we think that nodes should have the following internal
structure, depicted in Figure 2.
LABEL
Functional aspects
CONFIGURATION
FUNCTIONAL FEATURES
Formal aspects
FORMAL FEATURES
SUBCATEGORIZATION
Figure 2. Structure of a node
A node has the following fields associated with it. The label (Lab)
represents the category of a slot in a template, such as P1, Subject or Vin-
fin. The configuration (Config) is a specification of some part of the
underlying representation that will be expressed by this node. It cannot be
just any part of the UR; it should be part of the configuration of the mother
node of the one under consideration, for example the specification of the
subject term, a temporal satellite, the head of a term, or the main verbal
predicate. Typically, such descriptions will contain one or more predicates;
in those cases we will call them functional configurations. A configuration

may also be the description of a grammatical element, such as an auxiliary,
a demonstrative, or a tense suffix. In that case we will call them formal
configurations. The functional features (FncFtrs) are the primary µ opera-
tors relevant for the slot under consideration. They may be derived from
operators, functions and lexical elements of the Config, or they are directly
inherited from the FncFtrs field of the mother node. Examples of functional
features are Number, Tense and Animacy. Formal features (FrmFtrs) are
the auxiliary µ operators relevant for this slot. They may be derived from
the Config and the FncFtrs, be inherited from the mother node, or percolate
upwards from a daughter node. Examples of formal features are Case and
Finiteness. Finally, the Subcategorization (SubCat) is a template consisting
of nodes for the functional categories into which Config is to be split up.
The appropriate template is selected from the set of templates of the lan-
guage on the basis of the concrete value of the functional part of the node,
via more or less general rules. A template may also be inserted on the basis
of lexical information, typically from the head of the structure in Config,
for cases where the default node specifications would lead to the selection
of the wrong syntactic environment. This is an instance of lexical priority.
All these aspects together represent what in the standard model of the
expression rules are phase I and phase II. The templates and placement
rules of phase II are replaced by the subcategorizations and the functional
configurations of our model, respectively. Furthermore, formal configura-
tions and both feature sets replace the formation rules of phase I.
To round off this short sketch of the dynamic model of the expression
component, we give the complete derivation of a very simple sentence. It is
the expression of the underlying representation in (4).
(4) [decl E1:[X1:[pres e1:[smart [A] (prox pl x1:girl [N])ZERO,SUBJECT ]]]]

‘Those girls are smart.’
The first step in the expression process is the selection of a node that will
express the complete UR of (4). In order to cater for (4), this node needs to
have a more or less maximum representation of an UR for its Config field,
with variables for all relevant functions, operators and predicates and all
layers present. After its selection, the Config field will be unified with the
UR of (4), and the respective variables will get the corresponding values.
Example (5) below gives a simplified version of this top node before unifi-
cation with the UR and example (6) after unification. Elements in upper
case such ‘TENSE’ indicate free variables; lower case elements such as
‘pres’ represent the values for these variables after they are bound with the
corresponding values from the UR.8
(5) NODE 1 (uninstantiated)

Lab: sentence
Config: [ILLOC [[TENSE [ PRED CAT ARG1 ]]]]
FncFtrs: ILLOC, TENSE, CAT, NUMBER.SUBJECT, PERSON.SUBJECT
(6) NODE 1 (partially instantiated)
Lab: sentence
Config: [decl E1:[X1:[pres e1:[smart [A]
(prox pl x1: girl [N])ZERO,SUBJECT ]]]]
FncFtrs: decl, pres, [A], pl, [-sp,-hr]
The last step for this node is the selection of the right template for its
SubCat field. The information in the Config field is instrumental in this.
The result is given in (7).
(7) NODE 1 (fully instantiated)

Lab: sentence
Config: [decl E1:[X1:[pres e1:[ smart [A]
(prox pl x1: girl [N])ZERO,SUBJECT ]]]]
FncFtrs: decl, pres, [A], pl, [-sp,-hr]
SubCat: p1, subject, vfin, mainpred
Node 1 ends up being fully instantiated because there are no variables left
to be bound. If this were to be the case, then the only way left for these
variables to be bound by a value would be via percolation from lower
nodes. In a fully expanded tree there may be no unbound variables left.
The next step in the expression of the UR in (3) is the development of
the p1 position in its subcategory. Typically, there will be several alterna-
tive versions for this node in the grammar, e.g. with a Focus, Topic or Sub-
ject term for its configuration, but also with adverbials. In the concrete case
of (3), the UR element chosen will be the Subject. Filling the p1 slot with
the Subject constituent from the Config field will trigger a term template
for the filler of the SubCat of Node 2. The next step will then be the expan-
sion of the Determiner node of this template, assuming that this is the first
slot. Figure 3 below gives the complete flow of the expansion of Node 1,
up to the expression of the slot for the main predication, mainpred.
Note that here, the plural morpheme s is introduced by node 6, i.e. via a
rule. Our model is neutral towards the type of morphology, i.e. it could just
as well handle a word-based approach which has the inflected forms in the
lexicon.
This should serve for a brief overview of the dynamic expression com-
ponent. In Section 4 we will see to what extent it has to be adapted to fit
the multi-level FDG model.
1: SENTENCE
1 10 11 14 15 20
2:P1 7:VFIN 9:MAINPRED
2 3 4 9 12 13 16 19
3:DET 4:NOMHEA 8:VERBFORM 10:ADJHEAD
5 6 7 8 17 18
/those/ /are/
5:NOUN 6:NUMSFX 11:ADJECTIVE
/girl/ /s/ /smart/
Figure 3. Full tree expansion
3. Linguistic modelling and FDG

When discussing a model in linguistics, the central – indeed the first –
question is what is it that we are modelling. In general, a model seeks to
give an operational, explicit and non-ambiguous version of a theory. It is
through a model that a theory may be subjected to experimentation and
testing. For the present discussion, we will distinguish between two types
of linguistic models: the grammar model and the model of the language
user. A grammar model represents the respective components a theory dis-

tinguishes in the grammar of a language and the relations between these
components. Furthermore, it provides formal representations for the re-
spective rules and structures. It abstracts away from any speech situation
and perceives the language and its grammar as an abstract, formal con-
struct. The grammar model shows in what way the data structures (the
static elements of the grammar, such as the lexical items and morphologi-
cal and syntactic rules) lead to underlying semantic and syntactic
representations and these to actual linguistic expressions. Usually, these
models are restricted to the sentence as the highest level of representation
and production. There may be a fixed order in the flow of information
through the respective grammar components. Often, the ‘later’ components
not only expand the structures generated by the ‘earlier’ components but
also operate as filters on ill-formed structures. In principle, such models are
bidirectional in the sense that they relate the lexicon and underlying repre-
sentations to expressions and vice versa via ‘left-to-right’ or ‘right-to-left’
applications of the rules of the grammar. However, the traditional interpre-
tation of such models is from underlying representations to expressions; for
this reason they are often called generative. The grammar model is usually
thought to represent the linguistic competence of the ‘ideal speaker-hearer’.
It does not deal with typical performance aspects such as speech errors and
hesitations. This, indeed rather abstract model is the most common one
found in the linguistic literature. The model of FG presented in Dik (1997:
58) is an example of this type. Sells (1985) presents such models for Gov-
ernment-Binding Theory (1985: 24), Generalized Phrase Structure
Grammar (1985: 79) and Lexical Functional Grammar (1985: 137), respec-
tively. Although, as a model of the theory, such models should have
descriptive adequacy, they are not required to be psycholinguistically ade-
quate.
In Dik (1997: 1) the ultimate goal of a linguistic theory is formulated as
the construction of a model of the language user, a model that can ap-
proximate the communicative performance of human beings when using
their linguistic competence. It might therefore be assessed with regard to its
psychological adequacy. Such a model should not only contain linguistic
knowledge. It should also deal with all kinds of other knowledge, with be-
liefs, prejudices, etc. Furthermore, it crucially takes into consideration the
various types of memory: working memory, semantic memory and epi-
sodic memory. While a grammar model is neutral, a model of the language
user should distinguish between speaking and hearing, given that these two
modes of linguistic behaviour differ in many respects. A model of the
speaker should give a precise representation of the process of language

production, running from the initial, prelinguistic intention to the eventual
utterance, while specifying all relevant intermediate steps in the right order.
This information flow is mainly top-down, as in the generative grammar
model. However, psycholinguistic research strongly suggests that the pro-
duction of an utterance is not a purely linear process but an incremental
one, managed by several more or less independent modules which work in
parallel, at least partially, rather than consecutively. It is above all speech
error data which are suggestive of this parallelism, as well as the occur-
rence of hesitations, pauses, repetitions, etc. A model of the speaker should
cater for the extremely complex data flow during such a process as well as
for the ‘performance’ phenomena related to it. Finally, this kind of model
should include a representation of the speaker’s knowledge about the con-
versational partner(s). Hearers mainly use the same knowledge resources as
speakers do. However, the procedural aspects of a model of the hearer are
rather different from those of speaking. The hearing mode interleaves bot-
tom-up processes with top-down strategies, such as the construction of
cohorts of phonologically similar strings, or the priming of semantically
related entities during recognition. Another example of this is the choice of
the most promising syntactic structure at an early stage of speech analysis
via probabilistic devices. In the (rare) case of erroneous intermediate
choices, backtracking mechanisms are employed. One of the differences
between speaker and hearer that the respective models should make clear is
the often observed gap between a language user’s passive knowledge and
active use of a language. Eventually, both models should make clear how
first and second language acquisition take place.9
Hengeveld contrasts the FDG model sharply with the grammar model in
Dik (1997). The latter is seen – mistakenly, we think – as a speech produc-
tion model which lacks psychological adequacy, and therefore runs counter
to the adequacy requirements of the theory (cf. Dik 1997: 13f.). Dik (1989)
makes it clear, however that an FG model of the speaker necessarily has
other properties than the standard model of the theory and then sketches
such a Model of the Natural Language User.10 Hengeveld (this volume) ob-
serves that FDG is an improvement on the FG grammar model precisely
with respect to psychological adequacy. This, and the fact that the model
could be described in terms of the decisions a speaker makes during lan-
guage production, is highly indicative of the fact that FDG should be seen
as a model of the speaker rather than of the grammar. However, when we
take a closer look, it seems to have aspects of both models. We will briefly
mention these, and then assume for the rest of the discussion that a speaker
model is the ultimate goal of the enterprise. We will not assume that this
should be a full psychological model of the speaker; this might better be
constructed from the perspective of psychological modelling, and by a psy-
chologist. The FG version would be a linguistic model of the speaker
which is as much as possible in harmony with psycholinguistics.
Just like other generative models of grammar, FDG (Hengeveld this
volume) shows how complete (‘fully specified’) underlying representations
are first constructed and then fed to the expression rules which generate the
corresponding forms. Layered formal representations are introduced for the
respective intermediate structures, which are an extension of the traditional
underlying clause formalism first introduced into FG in Hengeveld (1989).
A novelty vis-à-vis the standard model is the introduction of hierarchical
tree-like structures suggested for the output of the expression rules. Most of
the structures at the respective levels are static and ready made, i.e. nothing
is said about how they come into existence. No explicit claims are made as
to their psychological status, i.e. whether they should be seen as mental
representations or just as formalisms. The latter possibility therefore seems
to be the more likely.11 Only for the representation at the pragmatic level is
an outside-in history claimed: Move goes before Act and this goes before
Communicated Content. Interestingly, though there are modules for Cogni-
tion and Communicative Context in the model, there is a separate box
labelled ‘Grammar’ around the pragmatic (IL), semantic (RL) and expres-
sion (EL) levels. This suggests that, apart from the lexicon, which is
contained in Cognition, the grammar gets a separate status outside the lat-
ter, with the inclusion of the projection rules and the expression rules
which link IL, RL and EL. Linguistic competence, however, is located
within Cognition. So it may be safe to assume that the three levels are just
levels of representation, while the rules that create them and link them are
stored in Cognition. Probably the label ‘Grammar’ should then be inter-
preted as ‘created by the grammar’.
On the other hand, FDG has several properties that are suggestive of a
model of the language user, more specifically of the speaker. Firstly, in
contrast to the standard grammar model, which takes the lexicon as a point
of departure for the generation of sentences via the respective sets of rules,
FDG introduces Cognition. This is the complete knowledge store of the
speaker, of which the lexicon and the rest of the grammar are just substruc-
tures. Indeed, Cognition controls the generation process at all stages. It
introduces long-term memory to the model, or rather semantic memory.12
Short-term memory (or rather working memory) is introduced via the
Communicative Context module. This stores both the formal representa-
tions of the Moves and (certain aspects of?) their expression. Interestingly,
in FDG the Content of the respective Acts is not stored in the Communica-
tive Context, but only influenced by it. Furthermore, there is a link from
this short-term device to the long-term Cognitive module. This feedback
loop accounts for the speaker’s monitoring function. A second point that
gives FDG the flavour of a speaker model is that it takes the Move as its
highest entity of processing, which brings discourse notions into the arena.
The feedback loop via working memory adds to this discourse orientation
by making aspects of previous utterances available to current or following
ones. This caters for problems such as long-distance anaphoric and
metalinguistic reference. Thirdly, in FDG a distinction is made between the
pragmatic and semantic modules. The output of the former partially feeds
into the latter, but both modules have their own, in principle parallel con-
tribution to make to the final output via independent access to the
expression rules. This parallelism introduces the possibility of testing the
model with respect to its performance characteristics.
Assuming, then, that FDG should be perceived as a model of the
speaker rather than of the grammar, we may make some further observa-
tions which in our view should be taken into consideration for future
developments of FDG.
Firstly, there is the role of the Communicative Context. In Hengeveld
(this volume: 3) this is supposed to be “the (short-term) linguistic informa-
tion derivable from the preceding discourse and the non-linguistic,
perceptual information derivable from the speech situation”. The arrows in
Hengeveld's Figure 6 make it clear that the information currently at the
speaker’s IL and EL levels is included in this information. If we are right in
assuming that ‘short-term’ in the quote above should be read as ‘contained
in short-term memory’, then the point is that this information can only be
very restricted, i.e. correspond to a few seconds of the verbal interaction at
most.13 It is not likely that this will always be the complete intended Move
at hand, including its Central Act and possible subordinate Acts as sug-
gested in Figure 6. The simple reason is that full recursion must be
assumed for Acts since they will get semantic content via referential and
ascriptive acts which in their turn are expressed by recursive linguistic enti-
ties such as NPs, PPs and Ss. Since there seems to be no upper limit to the
length and complexity of a Move, the information contained at the IL and
EL levels, which are thought to feed into the Communicative Context, will
regularly exceed the maximum capacity of short-term memory.
We think that it may in fact be more realistic to assume that what is
contained in the IL, RL and EL levels is actually in working memory, at
least partially, with the inclusion of the rules that operate on them. The
complete set of materials necessary for the construction of these entities is
stored more permanently in semantic memory. Alongside the grammar and
the lexicon, this is both the ‘full story’ that is being told and the scripts,
scenarios or mental models that are relevant for it (cf. Schank and Abelson
1977; Johnson-Laird 1983). A central planning mechanism sends coherent
bits of this material to working memory for linguistic expression, where it
is operated upon in three stages, leading to the representations at the IL, RL
and EL levels. In its EL form, it is finally send to the articulator, which
runs it as a set of instructions for the speech apparatus. The feedback loop
from Communicative Context to Cognition would then mean that whatever
information is held in working memory at any one time is fed back to se-
mantic memory for more permanent storage, or at least certain aspects of it.
In this respect it is interesting that there is no feeding arrow from the
semantics of the RL level to the communicative context. Rather, RL gets
part of its input from the Communicative Context, probably anaphoric in-
formation above all. Since Communicative Context is related to short-term
memory only, this must be very local information. This would imply that
no (direct) information about the precise semantics of an utterance finds its
way back to semantic (sic!) memory, but only information about the prag-
matics and the form. This is not a very probable state of affairs. The
extensive literature on recall experiments shows that hearers and speakers
retain the content of what they hear, read or say rather than the form. See
Brandsford and Franks (1971), Schweller et al. (1976) and Johnson-Laird
et al. (1974) for classical experiments that refute the recollection of syntac-
tic structure, precise words and parts of speech, respectively. But even
though it is the meaning of an utterance rather than anything else that is
remembered, it is not necessarily the concrete linguistic meaning in terms
of the meaning representation as given by an Underlying Clause, for in-
stance. Rather, it is a broad interpretation of the meaning, the gist or
intention of what was said. Therefore, speakers will not normally remem-
ber whether they have said (8a) or (8b).
(8) a. The window was open.

b. The window was not closed.
Taking a different perspective on things, one might argue that the con-
tents of what is being said are stored in semantic memory even before
speaking starts, and that there is no need for them to be fed back. This may
be true to some extent for a coherent and structured story or event in the
past that has been remembered and that is being told. However, in sponta-
neous conversation completely ‘new’ things may be said that speakers do
remember later on. Also, telling ‘old’ stories may have a restructuring ef-
fect on the way they are remembered and told on a later occasion. Thus, it
seems to be necessary to have feedback from the semantics of an utterance
to semantic memory, too.
In the light of the above we would suggest that the FDG model be re-
fined in such a way that what resides in working memory and what in
permanent storage is determined more precisely. A possible scenario is the
following. Assuming that all basic material – both building blocks and
rules – are located in semantic memory, an utterance is then constructed on
three more or less independent levels: IL, RL and EL. For all three levels
construction starts out from semantic memory, which sends (parts of) the
contributions of pragmatics, semantics and morphosyntax to working
memory at certain intervals. These intervals and the amount sent are related
to the capacity of working memory and to what is already present in it.
This process follows the hierarchy pragmatics > semantics > form. Accord-
ingly, some bits of the pragmatics are first sent to working memory, say
some aspects of the Move and the Act. When this has arrived, semantics
starts filling in the relevant parts, while pragmatics moves further on in the
process, e.g. to the first subordinate Act. When semantics is finished with
the first chunk and is ready to start working on the second chunk it gets
from pragmatics, the expression rules kick in and start working on the first
chunk of RL material. Apart from the fact that a great deal of the material
to be uttered by the speaker is in semantic memory in some form or other
before any utterance is made at all, we assume that it is IL material and
above all RL material that is fed back to semantic memory. There it is used
by an autonomous process that constructs a version of the total discourse,
to the extent that what is being said is a complete paraphrase or a more or
less new version of what was already residing there in some form or other
anyway. Only very little of the EL material is fed back for permanent stor-
age. Only this, and what is left in working memory, is available for
monitoring processes.
As to the shape and size of the contents of working memory at any one
time, we should probably best think of entities such as Chafe’s intonation
unit (Chafe 1987; and see Crystal 1975 for an earlier version of it) or Hal-
liday’s information unit (Halliday 1994).14 Indeed, in studies of spoken
language it has been frequently observed that (spontaneous) speech gener-
ally proceeds via relatively short contributions, which may contain one or
more given elements and rarely introduce more than one element that is
new to the conversation. Such units typically take 1 to 2 seconds at most,

and are embedded between intermediate pauses (cf. Givón 1975: 202f.).
The most common unit is of the Topic-Comment type, prototypically ex-
pressed in Subject-Predicate form. Quite often however, utterance units do
not consist of the complete intended Move or Act, and may also be ‘in-
complete’ from the perspective of GrammarN.15 This incompleteness may
be based on processing constraints. It may also be caused by the fact that
part of the topical information is thought by the speaker to be redundant. In
the latter case, this ties in neatly with the discussion in Mackenzie (1998),
which is a plea for adapting FG such that it also produces holophrases like
those in (9) below.
(9) Tomorrow.
To Mila.
Liam too.
Five of these, please.
The green one, if possible.
Skating.
Indeed, many utterances that appear to be ‘incomplete’ from the per-

spective of a sentence grammar turn out to be complete from the
perspective of a discourse grammar. Speakers, it is assumed, only produce
‘complete’ sentences in order to create more cohesion in their discourse
when the speech situation calls for this. In such cases they opt for what
Hannay (1991) calls the Topic Mode. However, in informal discourse
speakers will try to limit themselves as often as possible to new, focal in-
formation, thus producing chunks that need only the P1 (clause-initial)
position of the traditional FG template for their expression. The latter is
above all the case in the Reaction Mode, where speakers are frequently tak-
ing turns in the conversation. Such chunks may fit into working memory in
their entirety, which makes for ease of processing in the highly interactive
and competitive set-up of a face-to-face conversation. On the other hand, it
is precisely the size of working memory which puts constraints on what
can be maximally dealt with during speech production in the first place. So,
in the case of weightier Moves and Acts, spoken language typically comes
in Sub-Acts and Sub-Clauses, or even parts of these, i.e. mere fragments of
GrammarN complete clauses. Quite frequently, in spontaneous speech con-
secutive Sub-Clauses do not even combine to form GrammarN-complete
and well-formed ones. We think this is also a strong indication of ‘incom-
pleteness’ at the higher levels of linguistic planning, of the role of feedback
mechanisms, and of the parallel, incremental nature of the process of

speech production in general.
Example (10), from Chafe (1987: 23), is a case in point. In (10), a hy-
phen indicates lengthening of the preceding segment; two dots indicate a
brief temporal break; three dots indicate a pause; the acute accent shows an
intonation peak. Each line would be an intonation unit in Chafe's terms.
(10) a. … A-nd .. he would come into cláss

b. … a-t .. uh- you know three or f
c. … precísely one minute after the hóur,
d. or something like thát
e. … a-nd he- .. wou-ld .. immédiately open his … nótes up
The proposal in Mackenzie (2000) for an incremental model of FG

(IFG) should be seen in the light of the above remarks.
A last point we would like to make here is related to the nature of the
representations of the information given at the respective levels in
Hengeveld (this volume). For the RL level the usual Underlying Clause
(UC) structure is assumed, cut off at the propositional layer. For EL we
propose the structures introduced in Section 2; the link with the other levels
of representation will be worked out further in the next section. However,
some aspects of the IL representation remain unclear to us. The outer lay-
ers – Move, Act and Illocution – are formally of the Underlying Clause
(UC) type. Their contents are variables, operators, and references to
speech-act participants, i.e. they are formal in nature. The inner layer, un-
der variable C, has referring acts (with the variable R) and ascription acts
(with the variable T) as its contents. These acts get their semantic content
only after projection on the RL level, in the form of predicates of the lan-
guage, plus the corresponding variables, operators and functions. The point
now is how this predicational content should be determined. This problem
relates directly to the way in which mental representations are made in se-
mantic memory in the first place, both of the knowledge to be reported in
general and more specifically of the communicative context. If linguistic
relativism is assumed, as is suggested by the stepwise lexical decomposi-
tion philosophy proposed within FG ever since Dik (1978), then the
predicates of the language concerned make a considerable contribution to
the ‘stuff which thoughts are made of’. In addition, Dik (1989) proposes to
take UCs as the representational formalism for those types of knowledge
which are non-pictorial in nature. In other words, FG promotes a linguisti-
cally motivated type of knowledge representation. If this were indeed valid,
then projection from IL to RL would in fact be trivial: the R and T vari-

ables would simply be linking devices between UC-like mental
representations and newly made underlying clauses, via discourse variables
of the right type.
However, such an extreme relativistic position seems to be untenable.
There are a number of arguments, both logical and experimental in nature,
which point towards a non-linguistic, abstract representation of knowledge.
We will mention but a few. If language were to be the code in which (con-
ceptual) knowledge was stored, it would not be clear how knowledge about
language itself (or rather, a specific language) could be acquired by the
pre-linguistic brain. There must be some built-in mode of operation, neces-
sarily prelinguistic, with which to bootstrap the brain to thinking and
knowledge acquisition. There must therefore be some kind of thinking be-
fore language, and indeed without language altogether, since the
congenitally deaf do think, and not in a dramatically different way from the
non-deaf (cf. Furth 1966). Furthermore, there is no evidence for differences
in the way conceptual knowledge on the one hand and knowledge acquired
through perception on the other are stored and processed in the mind. The
same goes for declarative knowledge versus procedural knowledge. For
most types of knowledge that we possess, we do not remember how it was
acquired, nor can we determine by ourselves how it is internally repre-
sented. Furthermore, we seem to reason and to process it all in much the
same way. This all points in the direction of a unified form of representa-
tion, both for the knowledge that is linguistically expressible, like most
declarative knowledge, and knowledge that is not, or only partially so, like
procedural and perceptual knowledge. For further arguments we refer to
Clark and Clark (1977) and Fodor (1983) and references therein.
In what follows, our assumption will be that the planning of utterances
is largely pre- and indeed non-linguistic, i.e. it proceeds in terms of ab-
stract, language-independent concepts. This means that the mapping from
IL to RL involves translation from such abstract concepts to predicates of
the language concerned. The crucial link for this translation process lies in
the meaning definitions of the predicates in the lexicon. We will assume
that these are constructed on the basis of such abstract concepts, possibly
mixed with language-specific elements. Arguably, the ongoing discourse –
the complete Communicative Context – as represented in semantic mem-
ory is coded in the same, more or less language-independent fashion.16
This concludes our critical discussion of the more general aspects of the
FDG model. In the next section we will see how FDG, with the provisions
made above, may be linked to our dynamic model for the expression rules.
4. The FDG model and expression

We will now try and fit the FDG model and our expression module to-
gether. We will do this on the basis of example (9), repeated here in a
slightly simplified form under (11):
(11) a… A-nd .. he would come into cláss

b.… a-t .. uh- three or f
c. … precísely one minute after the hóur,
d…. or something like thát
These four intonation units are part of a longer fragment, which comprises
40 units in all. In its turn, this is part of a tape-recorded conversation during
an informal dinner party. The participants are telling each other stories
about the university teachers they had. In the 8 units that precede (11a) the
speaker has introduced a certain (male) teacher she particularly remembers.
We will assume that the information in that introduction is represented in
the speaker’s semantic memory in an activated, readily accessible form. In
that stretch of text, the teacher has been introduced as a topic; he has been
referred to several times pronominally. Furthermore a typical ‘school’
script has probably been activated, from which all kinds of default aspects
can be derived, such as the article-less use of ‘class’ in (11a). Finally, the
speaker will assume that her audience is in the same ‘frame of mind’. Let
us now try to trace the possible flow of information within the FDG model
as sketched above, with special attention to the expression of the respective
utterances.
The whole story of 40 intonational units, which is not interrupted by
anything other than assertive noises and laughter, consists of a series of
Moves. All of them express the communicative intention of the speaker to
inform the audience about this special teacher. We shall interpret stretch
(11a) – (11d) as one Move, which we will call M6. It consists of a central
Act, A6. M6 is about the fact that the teacher in question always manages to
arrive shortly after the moment at which the class should start. We will as-
sume that this specific piece of recollection is represented in the semantic
memory of the speaker in an abstract, non-linguistic way, as an inference
made over a set of observations over time concerning the behaviour of this
teacher. M6 starts with a pause, prolonged by the lengthened discourse con-
tinuity marker And and another short pause. During this pause the contents
of M6 are planned, and part of its expression form is calculated in working
memory. This will give rise to the first intonation unit, which is part of A6.
(12) … A-nd .. he would come into cláss. [previously (11a)]
According to FDG, after the mental material for M6 has been gathered,
the first step in this process will be the decision that this is going to be an
Informing move. Although Hengeveld (this volume) does not give an indi-
cation in this direction, we will assume that there will be an M operator
INFRM for this.17 Then the Act is constructed. The illocution chosen is
DECL(arative), the default choice under the INFRM operator. It is fol-
lowed by the variables for the discourse participants PS and PA and the
actual Content C6. Together this is a paraphrase of the intention ‘I tell you
that C6’. C6 will be filled in by a number of referring and ascriptive acts Ri
and Tj. The order in which this happens is determined by various compet-
ing pragmatic principles. Good candidates are Topic Continuity and Task
Urgency (Givón (ed.) 1983; Givón 1987, 1988). Both are scalar variables
which determine the formal expression of referring and property-assigning
entities in terms of constituent order and other devices, such as special con-
structions. Topic Continuity favours topical and other ‘old’ and
presupposed entities; Task Urgency favours focal and other ‘new’ and un-
predictable entities. For our discussion, our point of departure is that these
forces work on the interpersonal level, above all in determining the order in
which referring and ascribing acts take place. This pragmatically deter-
mined ‘underlying’ order is reflected only indirectly in the final syntactic
constituent order. It is reflected in the availability at the RL level and the
constituent-order rules of the language concerned.
More concretely, Topic Continuity, Task Urgency and possibly other
forces determine the order in which the respective Ris and Tjs will enter C6.
In the case of (12) the unmarked surface form is an indication that three
elements have been entered into C6 initially, and probably in the (default)
order Topical > Focal: the topical referent for the teacher (R1); the focal
referent for the notion CLASS (R2); and the third element for the ascriptive
notion GO INTO (T1).18 In the case of R1, instead of an abstract predicate,
the existing discourse variable for the teacher, say x1, will represent the re-
ferring act. In the case of referents which are new to the discourse, new
variables are introduced. Something like the representation under (13) may
then be in working memory. Note that here and in later representations we
leave out bits of the FG formalisms that seem to be less relevant for our
discussion.
(13) (M6: [INFRM [(A6: [DECL (P1)S (P2)A

(C6: [R1(x1)TOP R2(x8:CLASS)FOC T1(f1: GO INTO)])])]])
Now semantics will kick in and start working on this structure. Seman-
tic representations will be created for the referents of R2 and T1 by
searching the lexicon and by adding the operators which are relevant given
the grammar and the different types of background information such as the
ongoing discourse and the script. We will not go into the details of this un-
doubtedly very complicated process of translating cognitive structures into
semantic ones, precisely because we lack a representation for the former,
and a theory about the translation process. The result in working memory
will be as in (14).19
(14) (p1: [(past e1: [hab f1: come [V] (def x1)AG ] (gen sg x8: class [N])ILLAT )])
This will be passed on to the expression component. In parallel, parts of

(13) have been handed to the expression component as well. This might be
something like the information in (15).
(15) [DECL (P1)S (P2)A (C6: [(x1)TOP (x8)FOC (f1)])]
Here, unlike in the version introduced in Section 1 where we used the

traditional multi-layered underlying clause, we are dealing with two struc-
tures for expression, (14) and (15). Several options are now open. One is to
merge these two structures into one structure, which will give us something
like the traditional underlying clause. The variables shared by the two
structures should make this possible. However, this would mean reintro-
ducing the upward layering procedure. Also, in order to be combinable, the
two structures should be complete in the sense of the traditional underlying
clause. It seems to be more interesting to see whether we could also main-
tain the parallelism in the model during expression by keeping the two
structures apart. This can be accomplished by extending the Config field of
the top node – and other nodes for that matter – such that it can hold both
the pragmatic structure of type (15) and the semantic one of type (14). The
top node for the expression will then look like (16).
(16) NODE 1 (uninstantiated)

Lab: sentence
Config: [ILLOC PS PA CONTENT], TENSE [PRED CAT ARG1]
FncFtrs: ILLOC, TENSE, ASPECT, CAT, NUMBER.SUBJECT,
PERSON.SUBJECT
FrmFtrs: VARIABLE.SUBJECT
After unification with the structures in (14) and (15) we get the node in
(17).

Lab: sentence
Config: [decl (P1)S (P2)A (C6: [(x1)TOP (x8)FOC (f1)])],
[ [(past e1: [hab f1: come [V] (def x1)AG ]
(gen sg x8: class [N])ILLAT ) ] )
FncFtrs: decl, past, hab, [V], NUMBER.SUBJECT, PERSON.SUBJECT
FrmFtrs: VARIABLE.SUBJECT
Note that the values for number, person and the variable of the Subject
have not been bound. This was not possible because no Subject has yet
been assigned. In contrast to the position taken in the standard theory of
FG, we think that grammatical functions should not be assigned to underly-
ing representations, but by the expression rules themselves. Like many
authors, we see the phenomenon of Subject as the outcome of a grammati-
calization process, arguably involving left-dislocated topics. The result of
this process is a set of formal relations in morphosyntax, a subset of which
is obligatory for those languages which have Subject. Therefore, syntax
seems to be the right place for assigning this function to one of the con-
stituents that qualify for it. The choice is typically restricted to arguments
of the main predicate, and motivated on the basis of a language-dependent
set of aspects of constituents, such as their semantic function and a number
of other properties, e.g. animateness or person. However, cross-
linguistically, and maybe not surprisingly, it seems to be the case that in
the great majority of utterances in spoken discourse, Subject is assigned to
the constituent which is also the Topic.20 Subject assignment tends to have
effects at the highest level of the sentence: among these are constituent or-
der, agreement and several types of Same-Subject marking or non-
expression in coordinated and subordinated clauses. Therefore, the relevant
choices are necessarily made at the highest level of the syntactic structure,
i.e. at the top node of the expression. For the English example in (17), in
fact only one constituent is available for the assignment of Subject func-
tion: the Ag argument, which happens also to have Topic function. This
leads to the binding of the Subject-related variables in (17), which gives us
(18).

Lab: sentence
[(past e1: [hab f1: come [V] (def x1)AG,SUBJECT]
FncFtrs: decl, past, hab, [V], sg, 3
FrmFtrs: x1
The next step is the selection of the right template, for which the actual
contents of Config are scanned. This has been done in (19), where three
slots are introduced. Note that there is a position for the subject rather than
a general P1 position, since it is ‘known’ that the subject will land here in
this case. The other positions are for the verbal cluster and the satellite.

Lab: sentence
[(past e1: [hab f1: come [V] (def x1)AG,SUBJECT]
FncFtrs: decl, past, hab, [V], sg, 3
FrmFtrs: x1
SubCat: subj, vc, sat
This will now be expressed further via the process sketched in Section
1. This will lead to the expression of (12), on the basis of the partial tree
depicted in Figure 4 below (the numbers next to the arrows give the order
of creation; we have left out the upward-pointing percolation arrows). Note
that the tree below is in fact completely ‘flat’, i.e. there are no ‘unneces-
sary’ nodes and branches. We think that structure is represented in the first
place on the RL and IL levels and is not necessarily reflected iconically
and without further arguments in the structure of morphosyntax. Further
elaboration of the theory of expression should make clear whether more
formal structure and more formal categories are necessary to represent the
linguistic facts beyond some (relative or absolute) minimum.21
SENTENCE
1 4 9
SUBJ VC SAT
2 5 7 10 12
PRO AUX PRED PREP N

3 6 8 11 13
/he/ /would/ /come/ /into/ /class/
Figure 4. Partial expression tree; stage 1
After (12) – previously (11a) – has been articulated, working memory

will contain this tree, or at least such a substructure of it as to support the
rest of the syntactic entity under construction. The same will be assumed
for the IL and RL structures of (14) and (15). There is a short pause now, in
which new pragmatic and semantic material is being processed, corre-
sponding to (20).
(20) … a-t .. uh- you know three or f
This unit is complicated and particularly interesting. It starts with a

preposition, which is an indicator of a temporal satellite, as we know on
hindsight. This preposition is articulated, but in a hesitant, prolonged way,
apparently because the speaker is not sure about the temporal details of the
satellite. Therefore, extensions will be made at the IL and RL levels. We
will represent these as follows. First, an update will take place on the struc-
ture in (13). An extra referential act R3 will be added at the IL level, being
the cognitive version of the temporal satellite under creation. This leads to
(21); the new part is printed in bold.

(C6: [R1(x1)TOP R2(x8:CLASS)FOC T1(f1: GO INTO)
R3(x9:TIME)FOC])])]])
After completion of this, the IL level will be updated with the semantic
version of the temporal satellite. This gives us (22); again, the new bit is in
bold. Separately, the relevant part of the IL extension will be sent to the
expression component. This will give us (23). In both cases we will effect
these changes not in an independent structure, but directly in the Config
field of the relevant node of the expression, in this case Node 1. So we will
have the following updated structures there.
(22) (p1: [(past e1: [hab f1: come [V] (def x1)AG ] (gen sg x8: class [N])ILLAT )
(def x9)TEMP )])
(23) [DECL (P1)S (P2)A (C6: [(x1)TOP (x8)FOC (x9)FOC (f1)])]
Expression continues from Node 1 with the new element that was added
to the RL structure. After this update, a new node is created to express the
new satellite. After this Node 15 has been fully specified it will look like
(24).

Lab: sat
Config: [(x9)FOC ],
[ (def x9)TEMP]
SubCat: prep
Thus, the tree structure of Figure 5 is incrementally extended with a

new satellite constituent SAT for its rightmost constituent. Only the PREP
position will be expressed since there is no more material in this term.
SENTENCE
1 4 9 14
SUBJ VC SAT SAT

2 5 7 10 12 15
PRO AUX PRED PREP N PREP

3 6 8 11 13 16
/he/ /would/ /come/ /into/ /class/ /a-t/

It turns out now that the hesitation was not long enough, and has not led
to a solution of the search problem. It is therefore followed by three addi-
tional pauses of different types: silence; the prolonged uh- and the
meaningless interjection you know. The first two belong structurally to
move M6. We think, however that the you know bit should be seen as a new
Move, M7, pushed on top of whatever is in working memory. Arguably,
whatever was there is ‘saved’ in the background, for later elaboration.
The new move M7 is a very simple one. The contents of its Act, A7, are
a sole illocutionary operator. Its function is to recapture the attention of the
hearers which may be flagging because of the long pause. The structure of
M7 is as in (25) below. It will be sent directly to the expression component,
after which a lexicon search will provide its form.
(25) (M7: [INFRM [(A7: [ATT])]])
For this side step a new top node is created, which gets the pragmatic
structure of (25) as the only filler of its Config field. The ATT operator
will get expression via a special interj(ection) node, which searches the
lexicon for a node of the grammatical category ‘interjection’.
(26) NODE 1’ (fully instantiated)

Lab: sentence
Config: [att]
FncFtrs: att
SubCat: interj
NODE 2’ (uninstantiated)
Lab: interj
Config: [FORM, [INTERJ], TYPE ]
FncFtrs: TYPE
NODE 2’ (partially instantiated)

Lab: interj
Config: [FORM, [INTERJ], att ]
FncFtrs: att
NODE 2’ (fully instantiated)

Lab: interj
Config: [ynow , [INTERJ], att ]
FncFtrs: att
SubCat: ynow
In the meantime, the memory search for the value of the temporal satel-
lite has been completed. Copies of the saved IL and RL structures are
popped back to working memory again. Our (tentative) analysis in terms of
FDG is that IL is updated as in (27).
[R1(x1)TOP R2(x8:CLASS)FOC T1(f1: GO INTO)
R3(x9: 3+ MINUTE)FOC])])]])
Recall that expression had stopped at the SAT node as given in Figure 5
above, of which the preposition has been expressed. Control returns to this
Node 15. Its Config field will be extended as follows. Again, the new ele-
ments are represented in bold characters.
Lab: sat
Config: [(x9)FOC ],
[(def (3 OR 4) x9: minute [N])TEMP]
SubCat: prep
Two things will happen after this. On the one hand, the syntactic struc-
ture of Figure 5 will be further expanded with the new elements on node
15. This gives us the expression tree in Figure 6 below.
SENTENCE
1 4 9 14
SUBJ VC SAT SAT

2 5 7 10 12 15 21
PRO AUX PRED PREP N PREP NUM OP NUM

3 6 8 11 13 16 18 20 22
/he/ /would/ /come/ /into/ /class/ /a-t/ /three/ /or/ /f/
At the same time, the new information on Node 15 will percolate upwards
to the top node. As a result, the Config field of Node 1 will contain the fol-
lowing RL level information:
(def (3 OR 4) x9: minute [N])TEMP] ))])
When the rest of the satellite is finally expressed the speaker interrupts
herself because she wants to revise her initial estimate. This is an example
of self-monitoring, which is represented in the model. Information from the
working memory is fed back to semantic memory and added to the overall
discourse structure after evaluation. It is this evaluation which leads to the
instruction to stop expression, and reconsider. A new planning period starts
with a new pause. It is followed by a corrected version of the contents of
referential act R3.
(30) … precísely one minute after the hóur,
This is the result of an incremental update of the three levels. We first

give the updated IL and RL levels, the former without an attempt at a con-
ceptual representation. (31) contains the new referential act R4. Note that,
although the act is new, the variable that it refers to is that of R3, i.e. x9.
This variable stands for the mental notion under discussion.

(C6: [R1(x1)TOP R2(x8:CLASS)FOC T1(f1: GO INTO)
R3(x9: 3+ MINUTE)FOC
R4(x9: … )FOC])])]])
Correspondingly, the Config field of Node 15 will again be updated

with the contents of the time satellite. This gives us the following. Note
that again only the material represented in bold will be expressed.

Lab: sat
Config: [(x9)FOC ],
(def (1: precise [A]) x9: minute [N]:
(def sg x10: hour [N])POST)TEMP]
SubCat: prep num n x
Accordingly, the tree of Figure 6 will be further expanded. Expression

will continue at the rightmost satellite node again. Figure 7 shows the re-
sults; only a partial tree is given this time.
Again, the new information percolates upwards to Node 1. This time,

however, it is not added to the Config structure, but replaces part of it. The
new contents of the top node will be as in (33).22
(def (1: precise [A]) x9: minute [N] : (def sg x11: hour [N])POST))TEMP]))])
SAT
15
23 26 28
PREP NUM OP NUM NUM N SAT

16 18 20 22
/a-t/ /three/ /or/ /4/ 24 25 27 29 30 31
/precisely/ /one/ /minute/ /of/ /the/ /hour/
After (30) – previously (11c) – has been uttered, the monitor appears to
have been active again. The very specific version of the latest temporal sat-
ellite is relaxed somewhat in the last contribution to the current act, (11d),
now renumbered as (34).
(34) or something like thát
For this speech act, which we interpret as an alternative for the com-
plete time satellite we give only the RL update. Syntactic expansion
proceeds again from the top node, via a third SAT node. The final result in
the Config of Node 1 is as in (35).
(def (1: precisely) x9: minute [N]: (def sg x10: hour [N])POST)TEMP] )
OR (indef sg x11: [-S,-A,-M,-F]: (x9)COMPAR ) TEMP )])
This terminates the Act, the Move and our discussion of the integration
of the FDG model and our dynamic model of the expression rules.
4. Conclusions
In our contribution to this volume we have done three things. First, we
gave a short introduction to our dynamic model for the FG expression
rules. Then we looked critically at the Functional Discourse Grammar
model as proposed by Hengeveld (this volume). We made some sugges-
tions for the further elaboration of FDG. Finally, we tried to amalgamate
our slightly adapted version of FDG and the expression component into
one dynamic model of a discourse-oriented FG grammar of the Speaker. A
short example from spoken language served as a rough sketch of the work-
ings of this model.
Most of the problems that we ran into have been solved, at least to some
extent. Many aspects should be worked out in much greater detail, and in
the light of more empirical material, preferably recorded and annotated
versions of informal spoken conversation. We mention just a few of the
major problematic issues that spring to mind.
There is the – old – point of keeping track of the material that has been
expressed, and the related question of which underlying elements of the IL
and RL levels may be expressed more than once. This is of much greater
importance in a dynamic model such as the one given above than in the
traditional grammar model which deals with the ideal speaker-hearer. Re-
lated to this point is the requirement that the model should provide an
answer to what may, or must be repeated after hesitations and self-
corrections, and how corrections affect the eventual representations.
A second point is the precise nature of the interaction between the re-
spective kinds of memory, and the way information is stored there, what is
retrievable, how it may be extended, and under what conditions. A sub-
problem is the way discourse is represented in the speaker’s memory. This,
in turn, calls for a deeper insight into the way speakers monitor their own
speech process. Underlying this is the fundamental, and extremely complex
problem of finding an acceptable way of representing abstract concepts.
Undoubtedly, there are many problems that have been overlooked by us
altogether. These can not be solved without state-of-the-art knowledge of
the intricacies of human mental models of information processing. Fortu-
nately, within the Cognitive Linguistics (CL) enterprise there are many
hundreds of linguists, psychologists and computer scientists working in this
area. CL seems to be one of the obvious scientific arenas FG should turn to
for discussion, suggestions and solutions about these and other matters.
A dynamic speaker model is not only a research goal in its own right.
We are confident that the combined work on such a construct will put the
FG researcher in a better position to test all kinds of hypotheses about spo-

ken language. A rather concrete aspect of the integrated model presented
above, one which could introduce an interesting line of research to the the-
ory is the selection of the right subcategorization for the pragmatic and
semantic material offered to the expression component, i.e. the semantics-
syntax interface. The constraints the model presents may help us see better
how cognitively or semantically coherent stretches of input to the expres-
sion rules control and shape morphosyntax, with the obvious connections
to grammaticalization. We might also find out how certain grammatical-
ized patterns might in their turn ‘pull’ on semantics via the feedback
mechanisms, thus to some extent shaping the rest of the planning process.
Eventually, this should lead to a better understanding of the behaviour of
the language user, and to a better explanation of why languages are the way
they are.
Notes
1. The authors wish to thank Matthew Anstey for a number of valuable com-
ments and interesting points of discussion with regard to an earlier version
of the text. We hope we have done justice to them.
2. Actually, outside-in expansion of layers (or left-to-right as it is called by
Hengeveld) is only assumed for the interpersonal level. Nothing is said
about the coming into existence of semantic representations. Although we
think that here, too, the hierarchy among the respective layers has some
predictive value in the outside-in sense (cf. Bakker 1994 for more on this),
there can be no direct correspondence between the universal order in which
underlying clauses are structured and language-specific surface orders of
constituents.
3. Of course we could, if only for the sake of argument, take an extreme func-
tional perspective, denying all autonomy to morphosyntax. It might then turn
out to be possible in the grammar of individual languages to determine the
combination of all underlying factors which will eventually lead to certain
order patterns. These factors could be made available as µ operators to the
Phase I expression rule that determines the form of the finite verb. Unless we
see the form of the agreement marker as the expression of this conglomerate
of factors, i.e. not only of person/number aspects of the Subject, but also as
information about its pragmatic status, this would indeed boil down to pre-
running the expression rules in a more or less disguised way.
4. This requirement can be made explicit by formulating the role of the ex-
pression component such that a grammar of language L is descriptively
adequate if it generates precisely the set of well-formed expressions of L.

And a theory of grammar is descriptively adequate if it allows for all and
only the forms and structures that may occur in human language as such.
5. This is just an example of a possible tree expansion. We have no intention
here to develop an optimal constituency theory for English or FG for that mat-
ter. That will be a separate and very elaborate exercise, which should be done
on the basis of an in-depth formal morphosyntactic analysis of a typologically
well-founded sample of languages. Much can be gained from the insights of
other functionally based theories which have developed a morphosyntax, such
as Role and Reference Grammar (Van Valin and LaPolla 1999).
6. Certain types of speech errors make it clear that phonological information
must be available at a relatively early stage of the production of an utter-
ance, such as the frequently observed reversal of initial consonant clusters
between two words, either accidental or on purpose (‘spoonerisms’).
7. It would be a challenge to implement intriguing examples such as (1) and
(2) below (from Nespor and Vogel 1986: 238), where the presence or ab-
sence of flapping in American English depends on the strength of the
semantic or pragmatic link between the two sentences: Turn up the heat. I
am freezing (… hea[ſ]) vs. Turn up the heat. I am Frances (*… hea[ſ]).
Technically speaking, ours is by far not the only way the strategy envisaged
here might be implemented. However, in this form we follow models in the
psycholinguistic literature, such as Kempen and Hoenkamp (1987) and
Levelt (1989), thereby adding to the cognitive adequacy of the FG model.
8. For a fully worked out implementation of the dynamic model, a feature-
value notation will be assumed including a set of logical operators and a
unification mechanism. Such a formal system is necessary to make the rep-
resentations explicit and unambiguous, and to define the operations on them
in a straightforward way. A proposal for such a system may be found in
Bakker (1989; 1994: 263–296). Since these are mainly technicalities which
may obscure the current discussion we will stick to a more informal version
here. In the representation of nodes we will leave out the fields that have no
value.
9. Typically, the models for speaker and hearer are sketched more or less in-
dependently, although they use the same linguistic and non-linguistic
knowledge resources. However, since speakers obviously hear and monitor
their own output a realistic model of the speaker should at least contain a
feedback loop to the hearer model.
10. Hesp (1990) is a critical discussion of the MNLU proposal from the per-
spective of the psychology of language. The author shows that the MNLU
model as introduced in Dik (1989) cannot work, and is for a number of rea-
sons not in harmony with the models and experiments found in psycho-
logical literature. See our discussion below and Nuyts (this volume).
11. Note that Dik (1989) proposes the FG underlying representations as candi-
dates for mental representations.
12. Long-term memory is assumed to be split up between semantic memory for
relative static bodies of knowledge such as the grammar and lexicon, and
episodic memory for the storage of events, e.g. what happens during a con-
versation. Instead of short-term memory we will use the wider and more
dynamic notion of working memory. This holds the information directly re-
lated to the utterance, typically for a very short time, but it also includes –
and runs – the processes which operate on that information. See Carroll
(1994: 49f.) for a more detailed discussion of the several types of memory
relevant for language processing.
13. In this context the magical ‘7 plus or minus 2 items of information’, stem-
ming from Miller (1956), should be mentioned.
14. In working memory, data and processes are in competition for space and
time. In this connection it is tempting to think of grammaticalization as a
method to economize on the resources of working memory.
15. We will use the notion GrammarN for the normative standard grammar of
the language, as it is used in writing and in formal speech. Good examples
of short and incomplete utterances in informal discourse may be found in
the contributions to Chafe (1980), in which versions of the renowned Pear
Story are discussed. DuBois (1987), also taking the Pear Film as a point of
departure, contains interesting observations about the utterances typically
used during recall sessions of this short film. More particularly, he focuses
on the distributions of nominal and pronominal (or zero) NPs in the expres-
sion of grammatical relations. A recent survey of spoken English may be
found in Biber et al. (1999), where it is observed that more than a third of
the utterances in the spoken English section of the British National Corpus
are non-clausal, with an average length of just under 2 words.
16. For an early discussion of abstract predicates in FG meaning definitions see
Siewierska (1993), who discusses proposals in Jackendoff (1990). See Fa-
ber and Mairal Usón (1999) for a recent discussion of meaning and the
lexicon in FG which also takes abstract concepts in meaning definitions as a
point of departure.
17. Typically, in this type of story-telling discourse, each Move will be Inform-
ing by default. The insertion of continuation markers such as And and And
then may well be the expression of this operator.
18. We will follow the convention of distinguishing abstract, language-
independent concepts from the related English words by using capital letters
for the former. For the rest, no attempt will be made to say anything about
the way mental concepts might be represented.
19. In order not to burden the discussion with matters irrelevant to it, we follow
the FG convention of representing adpositions via semantic functions. In
Bakker and Siewierska (2002), and following Mackenzie (1992), we defend

the position that some adpositions should be seen as predicates rather than
grammatical elements.
20. Givón (1983), basing himself on text counts, gives an estimate of between
80% and 90% of the cases in running text.
21. This has, of course, a direct bearing on the (sub)theory of syntactic catego-
ries. To date, FG has not developed any theory of the kind; notions such as
Noun Phrase and Prepositional Phrase are used in a more or less theory-
independent fashion (cf. Rijkhoff 1992). Our point of departure will be that
a syntactic category is in the first place the prototypical formal correlate of
some underlying element: e.g. an NP is the prototypical expression form of
a term. To the extent that (formal) generalizations are possible over such
categories, they may be assigned a language-independent, or universal
status. It is unclear to us at the moment to what extent syntactic categories
sui generis will turn out to be necessary.
22. Of course, there are different ways to represent corrected underlying ele-
ments. For example, we could merge instead of replace, and maintain the
old, rejected material. Then we could have something like (i):
i. [(def (3 OR 4) (1: precise [A]) x9: minute [N]:
(def sg x10: hour [N])POST)TEMP]
We think, however, that our replacement solution is closer to the intuition
that the speaker will most probably remember the latest version as part of
the discourse structure. The fact that he made a correction is possibly re-
membered as such in the episodic memory, which also stores a number of
details about the conversation, but not as a part of the story as it is told.
References
Bakker, Dik
1989 A formalism for Functional Grammar expression rules. In: John H.
Connolly and Simon C. Dik (eds). 45–64.
1994 Formal and Computational Aspects of Functional Grammar and
Language Typology. Amsterdam: IFOTT.
1999 FG Expression Rules: From templates to constituent structure’.
Working Papers in Functional Grammar 67.
2001 The FG expression rules: A dynamic model. Revista Canaria de
Estudios Ingleses 42. 15–53.
2002 Adpositions, the lexicon and expression rules. In: Ricardo Mairal
Usón and Maria Jesús Pérez Quintero (eds). 125–177.
Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad and Edward Fine-
gan
1999 Longman Grammar of Spoken and Written English. Harlow: Long-
man.
Borsley, Robert D. and Janig Stephens
1989 Agreement and the position of subjects in Breton. Natural Language
and Linguistic Theory 7: 407–428.
Brandsford, John D. and Jeffrey J. Franks
1971 The abstraction of linguistic ideas. Cognitive Psychology 2: 331–
350.
Carroll, David W.
1994 Psychology of Language. Pacific Grove: Brooks/Cole Publishing
Company.
Chafe, Wallace L.
1987 Cognitive constraints on information flow. In: Russell Tomlin (ed.).
21–51.
Clark, Herbert H. and Eve V. Clark
1977 Psychology and Language. New York: Harcourt Brace Jovanovich.
Connolly, John H. and Simon C. Dik (eds)
1989 Functional Grammar and the Computer. Dordrecht: Foris.
Croft, William
1995 Autonomy and functionalist linguistics. Language 71: 490–532.
Crystal, David
1975 The English Tone of Voice. London: St. Martin.
Dik, Simon C.
1978 Stepwise lexical decomposition. Lisse: Peter de Ridder.
1986 On the notion “functional explanation”. Belgian Journal of Lin-
guistics 1: 11–52.
1989 FG*C*MNLU: Functional Grammar computational model of the
natural language user. In: John H. Connolly and Simon C. Dik (eds).
1–28.
1997 The Theory of Functional Grammar, Part I. Berlin and New York:
Mouton de Gruyter.
Du Bois, John
1980 The discourse basis of ergativity. Language 63: 805–855.
Faber, Pamela and Ricardo Mairal Usón
1999 Constructing a Lexicon of English Verbs. Berlin and New York:
Mouton de Gruyter.
Fodor, Jerrold A.
1983 The Modularity of the Mind. Cambridge MA: MIT Press.
Furth, Hans
1966 Thinking without Language. New York: Free Press.
Givón, Talmy
1975 Focus and the scope of assertion: Some Bantu evidence. Studies in
African Linguistics 6: 185–205.
1987 Beyond foreground and background. In: Russell Tomlin (ed.), 175–
188.
1988 Pragmatics of word order: Predictability, importance and attention.
In: Michael Hammond, Edith Moravcsik and Jessica Wirth (eds),
Studies in Syntactic Typology, 243–284. Amsterdam and Philadel-
phia: Benjamins.
Givón, Talmy (ed.)
1983 Topic Continuity in Discourse: Quantitative Cross-Language Stud-
ies. Amsterdam and Philadelphia: Benjamins.
nold.
Hannay, Mike
1991 Pragmatic function assignment and word order variation in a Func-
1998 Functional Grammar and Verbal Interaction. (Studies in Language
Companion Series 44). Amsterdam and Philadelphia: Benjamins.
Hengeveld, Kees
25: 127–157.
Hesp, Cees
1990 A critique of FG-CNLU. Working Papers in Functional Grammar
Jackendoff, Ray
1990 Semantic Structures. Cambridge MA: MIT Press.
Johnson-Laird, Philip N.
1983 Mental Models. Cambridge: Cambridge University Press.
Johnson-Laird, Philip H., Clive J. Robins and L. Velicogna
1974 Memory for words. Nature 251: 704–705.
Kempen, Gerard and Eduard Hoenkamp
1987 An incremental procedural grammar for sentence formulation. Cog-
nitive Science 2: 201–258.
1989 Speaking. From Intention to Articulation. Cambridge MA: MIT
Press.
1992 English spatial prepositions in Functional Grammar. Working Papers
in Functional Grammar 46. Amsterdam: University of Amsterdam.
2000 First things first: towards an incremental Functional Grammar. Acta
Mairal Usón, Ricardo and María Jesús Pérez Quintero (eds)
2002 New Perspectives on Argument Structure in Functional Grammar.
Berlin and New York: Mouton de Gruyter.
Miller, George A.
1956 The magic number seven, plus or minus two. Psychological Review
63: 81–97.
Nespor, Marina and Irene Vogel
1986 Prosodic Phonology. Dordrecht: Foris.
Rijkhoff, Jan
1992 The noun phrase. Doctoral dissertation: University of Amsterdam.
Schank, Roger C. and Robert P. Abelson
1977 Scripts, Plans, Goals and Understanding. Hillsdale: Erlbaum.
Schweller, Ken G., William F. Brewer and D.A. Dahl
1976 Memory for illocutionary forces and perlocutionary effects of utter-
ances. Journal of Linguistic and Verbal Behavior 15: 325–337.
Sells, Peter
1985 Lectures on Contemporary Syntactic Theories. Stanford: CSLI.
Siewierska, Anna
1993 Semantic functions and theta-roles. Working Papers in Functional
Grammar 55. Amsterdam: University of Amsterdam.
Tomlin, Russell (ed.)
1987 Coherence and Grounding in Discourse. Amsterdam and Philadel-
phia: Benjamins.
Van Valin, Robert D. and Randy LaPolla
1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge
University Press.
Epilogue
Kees Hengeveld
1. Introduction
In the first chapter of this volume I presented a basic outline of Functional
Discourse Grammar (FDG),1 pulling together various strands of research in
FG over the last decade. Basic features of FDG not shared by earlier ver-
sions of Functional Grammar are (i) its top-down organization, (ii) an
architecture which is both modular and hierarchical,2 (iii) the presence of a
structural3 module as a separate component of the underlying clause struc-
ture, and (iv) the location of the cognitive and contextual components. My
initial presentation of FDG focused on just these characteristics, ignoring
many other issues. The subsequent articles raise quite a number of these
issues, and in this chapter I will try to address some of them.4 Obviously,
many topics will remain untouched: FDG is a research programme rather
than a fully-fledged theory.
The main questions that I will touch upon here are the following: (i)
What is FDG a model of; (ii) How does FDG operate; (iii) What do the
cognitive and communicative components look like; (iv) What are the rele-
vant operator and modifier slots in FDG; (v) Where are semantic, syntactic,
and pragmatic functions located in FDG; (vi) What is the position of the
fund in FDG? The first three questions concern the architecture of FDG it-
self. The last three concern the incorporation of existing FG components
into FDG. After considering the six issues one by one, I will present a re-
vised and more detailed version of FDG in the final section of this paper.
366 Kees Hengeveld
2. A model of encoded intentions and conceptualizations

Since the introduction of the layered structure of the clause in FG
(Hengeveld 1988, 1989; Dik 1997), the question has been raised (e.g. by
Bolkestein 1992 and Harder 1996) whether this structure should be inter-
preted as a representation of the communicative process itself, or as a
representation of the linguistic units put to use in that process. These two
interpretations have become known as the process and pattern views on
underlying representations. In this volume Fortescue raises the same issue
explicitly in relation to FDG, and concludes that it is a process model. An-
stey, Bakker and Siewierska, Harder, and Nuyts all arrive at the same
conclusion.
It is important to note that this is not how I intend FDG to be taken. I
conceive of FDG as a pattern model, i.e. as a model that represents linguis-
tic facts. The process interpretation given to it by the aforementioned
authors is probably due to two facts. First of all, FDG is presented in
Hengeveld (this volume) as the grammatical component of a wider theory
of verbal interaction (cf. Dik 1997) and as such interacts with a conceptual
and a contextual component. These latter components, however, are not
part of the grammar as such, even though many grammatical phenomena
can be studied much more fruitfully if such components are assumed to ex-
ist. Secondly, in FDG, as in FG, the patterns of language are described as
reflecting the process of communication. This, however, does not mean to
say that FDG is a model of that process.
The latter point can be clarified starting from a familiar and generally
accepted underlying representation in FG:
(1) (e1: [(f1: collapseV (f1)) (x1: houseN (x1)Ø)Pat] (e1))

‘the collapsing of a house’
The representation in (1) contains three types of variable from the re-
presentational level: (e) for states-of-affairs, (f) for relations and properties,
and (x) for spatial objects. These variables are defined in terms of entity
types in the external, extra-linguistic world. Yet nobody would claim that
the formula in (1) represents part of the external world, as it would if it
were an ontological representation. Rather, (1) represents a linguistic unit
in terms of its function of designating part of the external world. This fact
has been obscured by informal terminological usage. It is common to say
that (1) represents a state of affairs, whereas it would be more appropriate
to say that (1) represents a linguistic unit in terms of its ideational function.
Epilogue 367
Thus, (1) is a linguistic representation based on non-linguistic factors.

Similar considerations apply to the representation of discourse acts.
Consider the following underlying representation:
(2) (A1: [DECL (P1)Sp (P2)Addr

(C1: [(T1: collapsed (T1)) (R1: house (R1))] (C1))] (A1))
‘The house collapsed.’
This representation contains five types of variables from the interper-

sonal level: (A) for discourse acts, (P) for discourse participants, (C) for
communicated contents, (T) for ascriptive acts, and (R) for referential acts.
These variables are defined in terms of the units relevant to the unfolding
speech event. Now, in exact parallel to what was said in relation to the rep-
resentation of states of affairs, this representation should not be interpreted
as a direct representation of the actual communicative situation, but rather
as representing a linguistic unit in terms of its function in communication.
Thus, (2) does not represent a discourse act, as it would if it were an action
theory representation, but should be interpreted as the formalization of a
linguistic unit in terms of its interpersonal function.
This is precisely why FDG can be called a functional model of lan-
guage: it captures the structure of linguistic units in terms of the world they
describe and the communicative intentions with which they are produced,
i.e. in terms of their representational and interpersonal functions.
3. The dynamic construction of linguistic expressions

A pattern model of language is not necessarily a static one. In this volume
the contributions by Bakker and Siewierska, Harder, and Mackenzie stress
the importance of a dynamic implementation of the model. The same issue
is raised in Bakker (1999; 2001) and Mackenzie (1998; 2000).
For a model of grammar a dynamic interpretation entails an implemen-
tation that mirrors the language production process in individual speakers.
Again, this does not mean that the grammatical model is a model of the
speaker. Rather, the model is assumed to be more effective, the more
closely it resembles this language production process. For the implementa-
tion of FDG this means that the various levels operate simultaneously,
though with a slight delay applying from higher to next lower levels. That
is, as soon a certain decision at the interpersonal level allows the selection
of elements at the representational level, the latter level becomes active.
368 Kees Hengeveld
And structural choices are made as soon as sufficient information is avail-

able from the interpersonal and representational levels. One could visualize
this provisionally as in Figure 1, anticipating a more elaborate picture in
Section 8.
Interpersonal Level activated

Representational Level activated
Structural Level activated
→→ time axis →→
Figure 1. The dynamic interpretation of FDG
The dynamic approach faces an important problem that touches upon

the basic principles of FG. The dynamically produced order of constituents
at the structural level is linear. But the underlying semantic representations
in FG are generally assumed to be unordered. As a result, in a dynamic im-
plementation of the model either the semantic representations have to be
specified before expression starts, or the semantic representations have to
be linearly ordered too. In this volume, Bakker and Siewierska opt for the
latter solution. This is a rather drastic departure from current FG practice,
and seems to suggest that differences in the structural organization of lan-
guages reflect differences in their semantic organization. Phenomena
which contradict such an approach are the existence of cataphora, and of
reflexive pronouns preceding their antecedents.
I will not attempt to formulate a complete alternative solution for this
problem here. A promising possibility involves a strict separation between
lexemes on the one hand, and the frames in which they occur on the other,
as proposed in García Velasco and Hengeveld (2002). Once these two ele-
ments are dissociated from each other, one might argue that expression
starts at the moment that frames have been selected, and the first lexeme is
inserted into the appropriate slot. Expression then evolves parallel to the
insertion of further lexemes into the remaining slots.
A full account of dynamic expression should not stop at the structural
level. A structural representation is part of the underlying representation of
discourse acts, but is not the same as the acoustic, orthographic, or signed
output. This output requires a separate component within the overall
model. The place of the output in relation to the FDG model will be pre-
sented at the end of the next section.
Epilogue 369
4. The conceptual and contextual components

In my initial presentation of FDG (Hengeveld this volume) the cognitive
component was defined rather sketchily, and far too restrictively, as one
which defines the communicative and linguistic competence of a speaker
and his/her knowledge of the world. Several authors in this volume, but
particularly Nuyts, have drawn attention to the necessity of a further elabo-
ration of this component.
A dynamic implementation of FDG, as discussed in the previous sec-
tion, begs for a ‘driving force’ to trigger the creation of linguistic
expressions. This driving force is embodied by a conceptual component,
within which communicative intentions develop and combine with suitable
conceptualizations. The conceptual component is not part of the grammati-
cal model, but in actual language use serves as a trigger for the grammar to
operate. Anstey (2002) presents a formal representation of the conceptual
level which basically mirrors the underlying linguistic representation, but
abstracts away from purely linguistic distinctions, and thus generalizes, for
instance, across operators and satellites.
Within the conceptual component there is an important distinction be-
tween communicative intentions on the one hand, and conceptualizations
on the other. This is reflected at the grammatical level in the presence of
interpersonal and representational levels. In his contribution to this volume,
Connolly makes exactly the same distinction for the contextual component,
incorporating insights from Discourse Representation Theory. He proposes
a detailed and convincing format for the description of the contextual com-
ponent, distinguishing between the contextual description and the
interpersonal description of the communicative context. It is important to
note that in Connolly’s view, which I share, the contextual component is
interpreted as a discourse domain (cf. Vet 1986) and thus contains a de-
scription of the knowledge shared by the interlocutors. This interpretation
of the contextual component is different from the one advocated in An-
stey’s contribution to this volume. He assumes different discourse models
for speaker and addressee.
Together, the proposals by Anstey for the conceptual component and
Connolly for the contextual component allow for a systematic treatment of
certain linguistic phenomena that can only be understood in terms of a
wider theory of verbal interaction. Chief among these is the question of
pronominalization, which is the topic of Cornish’s contribution to this vol-
ume. At the conceptual level, referents do not change: coding differences
do not reflect differences in perception. To simplify matters a bit, referents
370 Kees Hengeveld
are encoded pronominally when they are available within the contextual
component, and nominally when they are not. The importance of the inter-
action between the conceptual and contextual modules is also stressed by
Gómez-González in her contribution to this volume.
On the basis of the discussion in the previous section and in this one we
can give a more detailed presentation of FDG, shown in Figure 2, as the
grammatical component of a wider theory of verbal interaction. In this Fig-
ure, boxes represent components and levels, and circles represent
operations. The main components are the conceptual component, the con-
textual component, the acoustic component,5 and the grammatical compon-
ent. The grammatical component distinguishes an interpersonal, a
representational, and a structural level. The vertical arrows indicate the fol-
lowing: the conceptual component drives the grammatical component, both
as regards the formulation of the interpersonal and of the representational
level; the underlying representation resulting from this operation is en-
coded at the structural level;6 and interpersonal choices together with the
structural configuration determine the phonetic properties of the utterance.
The other way round, the result of these operations feeds the conceptual
component through the contextual component. The horizontal arrows indi-
cate that all grammatical and articulatory levels feed this contextual
component, creating possible antecedents at the representational level. But
the interpersonal level draws on the contextual component too, for instance
with respect to bystander deixis.
Epilogue 371
Conceptual Component
Formulation
Grammatical Component
Interpersonal Level
Representational Level
Contextual Component
Encoding
Structural Level
Acoustic Component
Articulation
Expression Level
Figure 2. Main components of FDG within a wider model of verbal interaction

372 Kees Hengeveld
5. Operators and modifiers

The FDG-model introduces a number of additional layers and as a result
creates a number of additional potential slots for operators and modifiers.
There are two areas where the additional slots seem to be particularly help-
ful: (i) the distinction between communicated (C) and propositional (p)
contents which results from upward layering at the representational level
offers new opportunities to account for speaker-bound and non-speaker-
bound modalities; (ii) the introduction of referential acts (R) through
downward layering at the interpersonal level provides a new perspective on
the layered structure of term phrases.
With respect to the first issue, one should note that in FDG proposi-
tional contents (as opposed to communicated contents) are situated at the
representational level, rather than at the interpersonal level. This calls for a
redefinition of modal categories: the distinction between communicated
and propositional contents creates an additional level of analysis. In fact,
this is a welcome addition since it helps to distinguish between modal, and
in particular evidential, categories which are restricted to main clause (i.e.
speaker-bound) uses, and modal, and in particular epistemic, catego-
ries which may be used in embedded clauses and then express the
propositional attitude of a main-clause participant, rather than the proposi-
tional attitude of the speaker. In his contribution to this volume, Verstraete
does indeed locate speaker-bound modalities at the interpersonal level, i.e.
as operators at the level of the communicated content.
With respect to the second issue, consider the following full representa-
tion of a referential act (R) in which reference is made to a first order entity
and in which an ‘Ω’ indicates an operator slot and a ‘µ’ a modifier slot.
(3) (Ω R1: [(Ω x1: [(Ω f1: LexemeN (f1): µ (f1))] (x1): µ (x1))] (R1): µ (R1))
In this representation of referential acts in FDG the same number of

layers, operators and modifiers can be distinguished as in Rijkhoff (1992),
but now they can also be provided with a systematic underpinning in terms
of the functions of the various layers. Thus, f-operators concern the nature
of the property (f) itself, x-operators concern properties of the set of desig-
nated entities, and R-operators situate the referential act in the actual
communicative situation.
For modifiers within term phrases similar things may be said. Consider
the following series of examples:
Epilogue 373
(4) a a poor driver

b a poor man
c Poor guy!
In (4a) the property of being a driver is modified by the adjective, which

does not state anything about the driver entity itself. One might well talk
about a poor driver who is very rich or even about a rich poor driver.
Bolinger (1967) calls the type of modification in (4a) ‘reference modifica-
tion’. In (4b) the adjective selects a subclass of the set of entities defined
by the head noun. This is the type of modifier typically treated in the FG
literature. In Bolinger’s terms this is a case of referent modification. In
(4c), finally, the adjective embodies the subjective evaluation by the
speaker of the entity referred to. The functional differences between these
three types of modifiers are reflected in structural differences in various
languages, as has for instance been demonstrated for Spanish in Berniell
(1995). The characterizations given to the various uses of the adjective
poor in (4) furthermore tie in nicely with their interpretation as f-modifier,
x-modifier, and R-modifier respectively.
6. Functions
As Anstey notes in his contribution to this volume, the functional hierarchy
of influence (pragmatics > semantics > syntax) is instantiated in the exis-
tence and ordering of the interpersonal, representational and structural
levels respectively. Similarly, these levels provide straightforward slots for
pragmatic, semantic and syntactic functions.
With respect to pragmatic functions Cornish and Mackenzie provide ar-
guments in this volume for locating pragmatic functions at the inter-
personal level. An important reason to situate pragmatic functions at the
interpersonal level is that the selection of a predicate at the next level
down, the representational level, is sensitive to the information status of
constituents (see e.g. Bolkestein and Risselada 1985; 1987). Semantic
functions, as in earlier versions of FG, are situated at the representational
level. They are part of the frames which are used to build up a semantic
representation. Finally, FDG offers a new location for syntactic functions: I
propose to situate these at the structural level (cf. also Bakker and Siewier-
ska this volume). In this way, syntactic function assignment can be seen as
the outcome of a process in which both pragmatic factors (at the interper-
sonal level) and semantic factors (at the representational level) are taken
374 Kees Hengeveld
into consideration. Expression rules sensitive to these factors determine

which constituents come out as Subject or Object, where this is relevant.
This proposal has consequences for the conception of syntactic functions in
FG: here they are considered to be purely grammatical rather than semantic
notions. The perspectival flavour of syntactic functions is then the product
of more basic pragmatic and semantic notions which trigger their appear-
ance, rather than being part of their meaning.
7. The fund
In the initial presentation of FDG I did not say anything about the location
of the fund in FDG, as noted by Gómez-González and Bakker and Siewier-
ska in their contributions to this volume. Here I can only present a first
sketch of this part of the model.
The basic idea is that for every level within the model the fund contains
the set of basic units which are used to build up that level. Basic units are
language-specific inventories. The fund of a given language contains at
least the sets of basic units given in Table 1:
LEVEL BASIC UNITS

Interpersonal Illocutionary Frames
Lexemes
Operators
Representational Predication Frames
Lexemes
Operators
Structural Templates
Morphemes
Acoustic Prosodic Patterns
Sounds
Table 1. Basic units stored in the fund
García Velasco and Hengeveld (2002) propose a new organization of

the fund in which predicate frames are split into lexemes on the one hand
and predication frames on the other. They give arguments to show that
from the perspectives of psychological, pragmatic and descriptive ade-
quacy such an approach is called for. Predication frames and lexemes,
together with the relevant operators, are the building blocks of the repre-
sentational level. A similar approach may be applied at the interpersonal
Epilogue 375
level: acts are based on an illocutionary frame; there are certain lexemes
that have an interpersonal content only (personal names, pronouns, inter-
jections);7 and the interpersonal level has its own set of operators. The
structural level draws on a set of (word order) templates8 and on a set of
grammatical morphemes. This is what Anstey calls a ‘syntacticon’ in his
contribution to this volume. Finally, the acoustic level contains at least
prosodic patterns and sounds, but conceivably also syllabic and word pat-
terns.
Note that this organization of the fund into components corresponding
to the various levels is crucial if one strives for a dynamic interpretation of
FDG. It allows for expression to start as soon as sufficient information has
come in. This information may come from as high up as the interpersonal
level and go as deep down as the acoustic level. For instance, the selection
of a particular illocutionary frame at the interpersonal level may be suffi-
cient to trigger a certain prosodic pattern at the acoustic level.
8. Summary: expanding the model

By way of summary, Figure 3 gives an expanded version of the FDG
model which incorporates the various adaptations that have been discussed
in the previous sections, and which were triggered by the many interesting
issues raised in the various contributions to this volume.
376 Kees Hengeveld
Conceptual Component
Formulation
Frames,
Lexemes, Interpersonal Level
Operators
Grammatical Component
Representational Level
Contextual Component
Encoding
Templates,
Morphemes
Structural Level
Acoustic Component
Articulation
Prosodic
Patterns,
Sounds
Expression Level
Figure 3. An expanded version of FDG

Epilogue 377
Notes
1. The use of the label FDG does not mean to suggest a radical departure from
the basic principles of FG. However, since the top-down organization of
FDG does imply that all existing components of FG have to be re-
interpreted, it is convenient to have a separate label to refer to this new re-
search enterprise.
2. In this volume, however, Moutaouakil presents another version of FG which
is both modular and hierarchical.
3. I have substituted the term ‘structural level’ for the term ‘expression level’
which was used in the first chapter of this volume, since expression in-
volves more than just structure. I will come back to this issue in Section 3.
4. I am grateful to Matthew Anstey, Annerieke Boland, Lachlan Mackenzie,
and Gerry Wanders for discussion of several of the topics dealt with in this
chapter.
5. Of course the acoustic component is only relevant for spoken language. For
sign language this would be a sign component, for written language an or-
thographic component.
6. Note that the structural level in fact has to be split up into a morphosyntactic
and a phonological level.
7. See e.g. Mackenzie's and Moutaouakil's contributions to this volume.
8. The difference between frames and templates is that frames are unordered
semantic configurations whereas templates are ordered syntactic configura-
tions.
References
Anstey, Matthew
2002 Layers and operators revisited. Working Papers in Functional
Grammar 77.
Bakker, Dik
1999 Functional Grammar expression rules: From templates to constituent
structure. Working Papers in Functional Grammar 67.
2001 The FG expression rules: A dynamic model. Revista Canaria de
Estudios Ingleses 42: 15–53.
Berniell, Silvia
1995 La posición del adjetivo en Español: Un análisis en el marco de la
Gramática Funcional. MA thesis. University of Amsterdam.
Bolinger, Dwight
1967 Adjectives in English: Attribution and predication. Lingua 18: 1–34.
378 Kees Hengeveld
1992 Limits to layering: Locatability and other problems. In: Michael
Fortescue, Peter Harder and Lars Kristoffersen (eds), Layered Struc-
ture and Reference in a Functional Perspective, 387–407. Amster-
Bolkestein, A. Machtelt and Rodie Risselada
1985 De tekstuele funktie van valentie: Latijnse drieplaatsige predikaten in
kontekst. Tijdschrift voor Taal- en Tekstwetenschap. 5 (2): 161–176.
1987 The pragmatic motivation of syntactic and semantic perspective. In:
J. Verschueren and M. Bertuccelli-Papi (eds), The Pragmatic Per-
spective, 497–512. Amsterdam and Philadelphia: Benjamins.
Dik, Simon C.
1997 The Theory of Functional Grammar. Part I: The Structure of the
Clause. Kees Hengeveld (ed.), 2d rev. ed. Berlin and New York:
Mouton de Gruyter.
García Velasco, Daniel and Kees Hengeveld
2002 Do we need predicate frames? In: Ricardo Mairal Usón and María J.
Pérez Quintero (eds), New Perspectives on Argument Structure in
Functional Grammar, 95–123. Berlin and New York: Mouton de Gruyter.
Harder, Peter
Hengeveld, Kees
1988 Layers and operators. Working Papers in Functional Grammar 27.
25 (1): 127–157.
this vol. The architecture of a Functional Discourse Grammar.
Machtelt Bolkestein (eds), Functional Grammar and Verbal Interac-
tion, 267–296. Amsterdam and Philadelphia: Benjamins.
2000 First things first. Towards an Incremental Functional Grammar. Acta
Rijkhoff, Jan
1992 The noun phrase: a typological study of its form and structure. Ph.D.
dissertation. University of Amsterdam.
Vet, Co
1986 Pragmatic approach to tense in Functional Grammar. Working Pa-
pers in Functional Grammar 16.
Index of names
Abelson 76, 341 Brown, G. 131

Amjad 322 Brown, P. 202
Anderson 75, 149 Brown, R. 59
Anstey vii, x, 23, 170, 211, 358, 366, Brunhoff 313
369, 373, 375, 377 Bunt 59
Aoyama 127, 133 Butler 19-20, 25, 31, 55, 73, 233
Asher 101, 113 Butterworth 83
Atkinson 79 Buyssens 56
Austin 78 Campbell 56
Bach 31, 57 Carnap 57
Baddeley 79 Carroll 233, 360
Bakhtin 94, 110 Casenhiser 233
Bakker x, 29, 34, 55, 57, 182, 212- Chafe 173, 220-221, 233, 236, 290,
213, 325, 327-328, 358-359, 366- 342-343, 360
368, 373-374 Cheshire 137
Barthes 318 Chomsky 30, 39, 57, 168
Bartlett 76 Clark, E. 345
Bates 214 Clark, H. 16, 182, 345
Bauer 34 Clark, K. 94
Bellert 269 Comrie 34
Ben Achour 40 Connolly viii, 25, 91, 98, 101, 103,
Bencini 233 108, 211, 213, 369
Benveniste 20, 318 Contini-Morava 118
Berg 52 Cornillie 269
Berman 199 Cornish viii, 117, 119, 128, 136-137,
Berniell 375 139, 215, 222, 369, 373
Berry 236 Coulthard 5, 92, 106-107
Biber 360 Crevels 18
Birner 123, 145 Croft 328
Bock 278 Cruttenden 223
Boland 52, 56, 377 Crystal 342
Bolinger 207, 259-260, 375 Cuvalay 302, 305
Bolkestein 2, 17, 25, 27, 44, 55, 58- Dahl 32
59, 93, 117, 127, 132, 142, 152, Dancygier 252
170, 211, 226, 259, 261- 262, Davidse 269
269, 310, 315, 366, 373 Davidsen-Nielsen 206
Borsley 327 Davies 252
Brandsford 341 De Groot 55, 189, 304
Broadbent 79 De Jong 40
380 Index of names
De Schutter 25, 289 Glover 137

Decarrico 190 Goethals 269
Declerck 15, 251, 259, 269-270 Goldberg 118, 214, 233
Devlin 101, 110 Gómez-González ix, xi, 25, 117,
Devriendt 25 131, 144, 211-212, 218- 220, 224,
Dietz 40 227, 229, 234, 370
Dik vii, x, 2, 5, 24-43, 45, 50- 51, Gómez Soliño 18
55-60, 73, 78, 89-93, 95, 103, Gonzálvez-García 233
106, 111, 117, 120, 125, 127, 131, Goodman 214
133, 138-139, 141- 142, 151-154, Goossens 25, 255
160, 162-163, 168, 170, 173-175, Grice 49, 146, 217
179-180, 182-183, 186, 188, 211, Gulla 92, 96, 299
218, 222-223, 235-236, 243-245, Gundel 119, 129, 133, 137, 145
264-265, 275-277, 286, 292, 300- Halliday 29, 186, 224, 232, 236,
303, 305, 307-309, 313, 315-321, 246, 253, 263, 266-267, 270, 295,
325-327, 331, 337- 338, 344, 359- 342
360, 366 Hancher 91
Diver 118, 127 Hannay 2, 17, 25, 117, 123-124,
Dowty 59 126, 132, 140-141, 145, 152- 153,
Du Bois 234 164, 166, 170, 174-175, 189, 204,
Engberg-Pedersen 25, 56, 172, 173 224, 226, 229, 235, 343
Engdahl 223, 229 Hansen 207
Erteschik-Shir 120, 125-126, 139 Harder ix, 44, 53-54, 58, 151, 170,
Faber 25, 30, 41, 360 197, 199, 207, 213, 229, 366-367
Fairclough 94 Hare 253
Fillmore 32, 56, 76, 214, 220, 277 Hasan 103
Firth 104 Hedberg 133
Fodor 74, 84, 345 Helmbrecht 61
Foley 40, 58, 246, 252 Hengeveld vii-x, 2, 5, 7-8, 15, 23-
Fortescue viii-ix, 25, 34, 151- 152, 24, 26, 34, 40-53, 55-60, 73- 74,
154, 160-162, 171, 189, 199, 76, 78, 80, 82, 84, 89, 92- 93, 96,
211, 213, 366 108, 111, 117, 132, 139-140, 144,
Fox 229, 235 153-154, 160, 162-163, 165, 167,
Franck 207 169, 171, 174, 180, 182-188, 197,
Franks 341 199-203, 211-215, 232, 243-252,
Fraser 15 254-255, 264-267, 269-270, 275-
Fried 239 276, 285-286, 288, 295, 299, 305,
Furth 345 312, 315, 321, 325-327, 329, 338-
García, E 118, 130, 145, 229 340, 344, 347, 357-358, 365-369,
García Velasco 57, 368, 374 374
Garrod 76 Hesp 39, 73, 359
Gebruers 32 Hickmann 293
Gerrig 16 Hitch 79
Givón 175, 229, 234, 342, 347 Hoekstra 25
Index of names 381
Hoenkamp 359 Levinson 60-61, 202

Holquist 94 Lewis 58
Hopper 280 Liedtke 18, 46, 93, 310
Huffman 121-123, 126, 129, 144- Loftus, E. 75
145 Loftus, G. 74
Hymes 34, 56, 104 Longacre 27
Inchaurralde viii, 73, 76, 85, 215 Lucy 8
Itkonen 167 Lyons 18, 40, 54, 60, 246, 253, 259
Jackendoff 179-181, 188, 212, 236, Mackenzie viii-xi, 2, 17, 25, 44, 51,
269, 278, 292, 360 55-59, 117, 131, 171, 175, 179,
Jacobsen 171-172 204, 211-212, 215, 224, 232-233,
Jadir 18 235, 276, 300, 302, 313, 343-344,
Jakobson 27, 30 367, 373, 377
Johnson 32 MacWhinney 212
Johnson-Laird 341 Maes 146
Kamp 100-101 Mahfouz 313
Katz 57 Mairal Usón 30, 41, 56-57, 60, 360
Kay 214, 277 Mann 92
Keizer 15, 25, 58-59, 117, 131, 211, Marin 87
235, 300, 313 Martin 94
Kemmer 219 Martín Arista 25
Kempen 359 Martín Mingorance 25, 41
Kies 234 Mathesius 32
Kirsner 118 Matthews 29
Knowles 234 Miller 360
König 295 Minegishi Cook 86
Kress 94, 105 Minsky 76
Kroon 5, 17, 25, 44, 93, 95-96, 110, Mithun 154-155, 172, 234
182, 201, 211, 299, 309 Moutaouakil ix, 2, 25, 44, 182, 211,
Kuno 223 262, 270, 299-300, 302, 304-305,
Kuroda 223 310, 313-315, 317-318, 377
Kwee 55 Nakayama 154, 156, 158-159, 161,
Ladd 223 166, 172-174
Lakoff 76, 79 Nattinger 190
Lambrecht 119-121, 125-126, 131, Nespor 359
144, 174, 222-224, 233, 235 Nichols 31
Langacker 75, 129, 172, 214-215, Noonan 214
220, 232-234, 236 Nuyts ix, 25, 38, 73, 170-171, 175,
LaPolla 359 181, 212-213, 215, 232, 269, 275,
Laury 146 277-280, 284-285, 288-289, 292-
Lauwers 269 295, 297, 360, 366, 369
Leonard 127, 133-134 Olbertz 25, 55-56
Levelt 2, 7, 73, 180, 186-189, 213, Olson 79
278, 359 Ortony 76, 87
382 Index of names
Östman 146 Sinclair 5, 92, 106-107

Otheguy 118 Slobin 173, 199
Palmer 245, 249, 270 Smits 228
Pérez Quintero 25, 41 Sperber 118
Perkins 279 Stamper 103
Piťha 32, 34, 56-57 Stephany 293
Polinsky 144, 223 Stephens 327
Postal 57 Steuten 18, 40, 92-93, 96-97, 211
Prideaux 34, 235 Strang 56
Pu 235 Strauss 134-135
Quine 151, 178 Strawson 151
Quirk 206, 210 Swadesh 155, 159, 172
Radden 79, 87 Sweetser 8
Reed 251, 269, 271 Taylor 234
Reichling 32 Tesnière 56
Reid 119, 121-122, 130, 144 Thompson 92
Reinhart 126, 131, 145 Traugott 294-295
Reyle 100-101, 115 Tulving 75
Rijkhoff 25, 42, 285, 300, 304, 372 Vallduví 223, 229, 242
Rijksbaron 61-62, 64 Van Belle 269
Risselada 17, 375 Van Buuren 46
Rolland 40 Van den Berg 2, 18, 44, 93, 132,
Rosengren 144 140, 211, 310
Ruck 123 Van der Auwera 34, 44
Rumelhart 76, 87 Van Hoek 129, 149
Saffran 79, 87 Van Leeuwen 94, 105
Sánchez García 56 Van Schaaik 58, 189
Sanford 76, 87 Van Turennout 52
Schank 76, 341 Van Valin 40, 56-58, 60, 246, 252,
Schilperoord 83, 88 359
Schwartz 79, 87, 88 Vendler 59
Schweller 341, 365 Vermandere 269
Searle 40, 91, 108, 115 Verschueren 111
Selkirk 223, 242 Verstraete ix, 42, 213, 243, 262, 269,
Sells 337 374
Sethuraman 233, 239 Vester 25, 59-60
Seuren 58, 70 Vet 2, 17, 18, 40, 44, 59-60, 93, 97-
Shatz 79 98, 132, 182, 211, 310, 369
Shibatani 171 Vogel 359
Shiffrin 79 Wales 56
Siewierska x, 25, 29, 57-58, 117, Wanders 18, 377
122-123, 132, 145, 182, 212-213, Ward 123, 125, 145, 236
223, 234, 325, 360, 365, 366-368, Watters 35, 71
373-374 Weigand 40, 71
Index of names 383
Weijdema 59, 71 Wodak 94, 114

Weinrich 29, 57, 71 Xinzhang Yang 42, 71
Whorf 56 Yule 131
Wierzbicka 76, 88 Zacharski 133
Wilson 118
Index of subjects
ability 36, 245, 247, 249, 251, 254- ascription 7, 108, 185, 198, 202,
256 212, 326, 344
aboutness 120, 131, 222 aspect 40, 54, 79, 92, 118, 126, 131,
accessibility viii, 144, 175, 180, 189, 151, 153, 156, 183, 276- 277,
219-221, 282-283, 331, 346 281-284, 290, 293, 329, 358
acoustic 368, 370, 375, 377 assertion 123, 125, 186, 223, 231,
acquisition 56, 62-63, 214, 294, 320, 235
338, 345 attention viii-ix, 11, 26, 45, 48, 76,
act 5, 7-8, 10-11, 15-16, 43-44, 47, 79, 102, 117, 119, 121, 123-125,
76, 91, 95-97, 99, 102, 108-111, 127-130, 133, 135-138, 143, 145,
135, 152-153, 167, 171, 174, 183- 175, 212-213, 217-218, 220, 222,
186, 188, 190, 192, 200-203, 205, 228-230, 233, 289, 307, 346, 353,
215, 230, 233, 235, 247, 253, 255, 363, 369
277, 286-287, 294-295, 347, 351, attitude 109, 171, 186, 206, 226,
355-356, 367, 372 261-263, 270, 277, 287, 289, 292-
activation 75, 184, 193, 220-222, 294, 372
235, 319 bottom-up vii, 2, 182, 199, 211, 338
addressee ix, 5, 15, 77, 80, 83, 90, buffer 188, 215
108, 119, 121, 125, 127- 134, bystander 370
136-140, 143-145, 186, 188-189, categorical 119-120, 122, 139, 143-
191, 202, 206, 215, 234-235, 369 144, 149
adjacency 102, 109-111, 141 category 8, 10, 17, 28, 30-31, 46, 52,
affix 157-159, 165, 169, 171, 173 56-59, 78-79, 91, 107-108, 118,
animacy 157, 161, 234 153, 158, 161, 163-164, 167, 176,
antecedent 43, 53, 129 179, 198, 207-208, 214, 218-219,
apodosis 252, 288 221, 223, 233, 244-247, 252, 254-
apposition 57, 226 259, 263-268, 270, 275, 277-279,
archetypal 300, 303, 312, 315, 320 286-287, 293-294, 300-303, 305,
architecture vii-x, 23, 25, 50, 55, 73- 307-308, 310, 312-315, 320-321,
74, 80, 89, 167, 169, 212, 243- 329-330, 333-334, 350, 353, 361,
244, 266, 268, 299, 365, 378 372
argument 16, 28, 35-36, 43, 48, 59, chain 50, 52, 75, 80, 161, 165, 198,
120-121, 124, 127, 145, 157-158, 234
161-164, 166, 174, 180, 187, 192, classifying 8
202, 204, 207, 213, 223, 231, 263, clause ix, xiii, 7, 11, 15, 25, 36, 41-
270, 277- 278, 293, 309, 349, 358 42, 44, 56, 60, 89-90, 92- 93, 95,
articulation 2, 34, 52, 55, 73, 161, 108, 118-119, 121- 122, 127, 131-
164-165, 169, 182, 187, 212- 213, 132, 136, 139-145, 152, 156-157,
326 160-166, 169, 171-172, 175, 181-
Index of subjects 385
182, 188, 197, 204, 206-209, 230- conceptual 23, 39, 60, 75, 87, 98,
234, 244-246, 251, 264-266, 275- 152, 165, 171, 175, 181, 185,
276, 284, 288, 295, 300-302, 304- 211-212, 218, 220, 226-227, 229,
305, 309-310, 313-315, 317, 320, 236, 272, 275, 277-279, 281-283,
325-326, 328- 329, 339, 348, 365- 285-290, 293-297, 345, 355, 366,
366, 372 369-370
clause combining 41 conceptualization ix, 79, 84, 187-
cleft 172, 174, 296 188, 215, 218, 227, 236, 247,
clitic 118, 122, 127-129, 143, 145, 277-278, 280-281, 285-286, 288-
158 289, 291-292, 295, 297, 307
code 16, 44, 49, 153, 199, 206, 269, conditional 16, 248-249, 251, 256-
280, 295, 329, 345 257, 269, 287-288
cognition viii-x, xiii, 3, 17, 38- 39, configuration 333, 335, 370
50, 56, 74-80, 82-84, 93, 140, connectionism 75, 79
143, 151-152, 170-172, 175, 180- connective 123, 133, 141
181, 183, 185-189, 211-215, 219- constituent 7, 26-28, 35-36, 38, 46,
221, 226-229, 232-233, 235, 241, 52, 56-57, 90, 108, 119, 124, 126,
275-276, 280, 282, 288-289, 292, 131, 138, 140, 142- 143, 145,
296, 328, 332, 348, 351, 359, 365, 154-161, 166, 169, 172, 188-189,
369 197, 207, 217, 221-222, 224, 231,
cognitive component 3, 17, 74, 76- 235-236, 285, 302, 305-307, 309,
79, 82-84, 93, 175, 212, 214, 369 316, 326, 328-329, 331-332, 335,
cognitive grammar 215, 241 347, 349, 352, 358-359, 361, 368,
cognitive-functional 275-276, 288 373, 377
coherence 126, 142, 308 constituent order (see also word or-
cohesion 161, 220, 231, 343 der) 328, 347, 349
commitment 40, 133, 246-248, 250- content ix, 3, 5-6, 8, 10-11, 15, 37,
252, 254-255, 257-259, 261-262, 46, 73, 77, 82, 90, 97-99, 101,
265, 267-268, 270, 284, 289, 291, 111, 120, 139-140, 153, 160, 168,
295 185, 188, 198, 200-206, 208, 220,
communication x, 8, 13, 26, 47, 49, 247, 251, 277, 300, 309, 320,
80-83, 94, 103, 105, 167, 175, 340-341, 344, 372, 375
181, 183-186, 217, 233, 278, 291, context 2, 12, 17, 27, 36-37, 44, 49-
301, 315, 319-321, 366-367 51, 56, 60, 76, 82-83, 90, 93, 102-
communicative competence 3, 30, 103, 108, 110, 118-121, 123, 125-
74, 326 127, 136, 139- 140, 142, 145,
communicative context 12-13, 36, 153, 155, 157, 159-161, 163, 166-
44, 49-51, 60, 76, 82-83, 140, 167, 169- 170, 173, 184, 186,
169, 184, 186, 214, 341, 344, 369 188, 190, 214-215, 220, 226, 234-
complement(ation) 41, 129, 161, 328 236, 249, 252, 256-257, 269, 286,
computational 25, 39, 55 289, 313, 341, 344, 360, 369
concept 29, 32, 52, 56, 131, 165, continuity 45, 58, 91, 126, 229, 232,
189, 234, 260, 275-276, 278, 283, 346
286, 289, 295, 300
386 Index of subjects
contrastive 145, 155, 166, 169, 172, 310, 312-321, 327, 340, 342-349,
174-175, 217, 313 355, 357, 360, 367-369
control 30, 85, 161, 167, 289, 290, discourse act 2, 5, 8, 10, 44, 47, 54,
358 93, 97, 103, 188, 201-202, 212,
controller 130 367-368
conversation 52, 77, 92, 106, 135, discourse grammar 2, 152, 154, 160,
183-184, 220, 279, 341-343, 346, 301, 343
357, 360-361 discourse particle 2, 60
coordination 24, 27, 57, 139, 226, discourse setting 132, 308
316 D(iscourse-)Topic 218-220, 222-
copula 15, 160, 162 223, 230-232, 235
coreferentiality 60, 129 domain-specific 74
current discourse space 215, 229 downward layering 7-8, 171, 198,
declarative ix, 56, 75, 90, 131, 166, 203, 372
175, 179-182, 187-188, 190-193, driving force 369
202, 204, 207, 250, 252, 270, 287, dynamic viii, x, 2, 30, 50, 90, 151,
345 182, 212, 214-215, 227, 229, 241,
decoding 73, 83, 214, 217, 259 245, 251-252, 254-256, 270, 293,
definite 37, 58, 132, 134, 137, 141, 327, 329-332, 334-335, 345, 356-
144, 149, 157, 160 357, 359- 361, 367-369, 375, 377
deixis 20, 49, 60, 127, 134, 136, dynamic modal 245, 252, 254-256,
148-149, 221, 307, 370 293
delimiter 107-108 echoic 249, 251, 256-257
demonstrative 13, 53, 60, 127- 128, embedding 15, 31, 90, 94, 152, 165,
130, 133-134, 136-137, 146, 333 167, 171, 174, 204, 207, 312,
denotation 44, 120 315-316, 321-322, 328, 342, 372
deontic modal ix, 243-246, 252-270, emic 153, 161, 164-165, 168
281-282, 284, 286, 290, 294, 296 encapsulation 74
depth-first x, 182, 212, 331 encoding 7, 18, 46, 60, 73, 81-83,
diachronic 281, 293, 295, 332 127, 129, 130, 132, 145, 187, 214,
dialogue 18, 90-92, 94, 96, 104, 106, 217, 231, 248, 251-252, 259, 261,
108, 114, 141, 232 267, 296, 326
direct speech 16, 44, 53 encyclopaedic 75-76, 141
discourse vii-ix, 2-3, 5, 8, 10, 12, 26, entities 6, 8, 11-12, 53-54, 58-59, 75,
38, 43-44, 47, 49-52, 54, 59-60, 79, 122-123, 131-132, 135, 138,
74, 83, 89-113, 119-121, 123-128, 141, 151-152, 159, 167, 170-172,
130-134, 136, 138-146, 150, 152- 185, 197-198, 212-213, 218-220,
153, 155-156, 158-161, 163-169, 223, 226, 231, 281, 287, 302-303,
171-172, 174, 181-183, 187-188, 312, 318, 326, 331-332, 338, 340,
197-199, 201-203, 206-208, 211, 342, 347, 351, 372-373
213, 215, 217-222, 224, 226-237, episode 241
249, 252, 256, 261, 269, 276, 285, epistemic modal ix, 59, 125, 244-
288, 291-292, 295, 299-303, 305- 246, 249-252, 254-256, 258-270,
277-288, 290-293, 295-298, 310, formal feature 28, 334

372 formalization 26, 29, 31, 39-40, 49,
etic 155, 164, 168 53-54, 367
event 59, 90-91, 94, 103, 121, 123, formula 98, 190-191, 366
145, 156, 172, 215, 223, 231, 277, fragment 184, 187, 346
286, 292, 306-308, 313, 320, 341, Functional Discourse Grammar
367 (FDG) vii, ix, 2-4, 17, 24, 73-74,
evidential 187, 269, 277, 282, 284, 76-77, 79-80, 82-84, 89, 92-93,
286, 290, 293, 295, 298, 314, 372 95, 99, 103. 106, 117-118, 139-
exchange 5, 96, 106, 108, 237, 258 141, 148, 179, 197-198, 200-201,
exclamative 90, 304 204, 208-209, 212-215, 232, 299,
existential 123, 159, 230-231, 236 325-327, 335-336, 338-340, 342,
expression vii-viii, x, 2-3, 5-8, 10- 345-347, 354, 356-357, 365-377
11, 13, 15-17, 18, 25, 28- 29, 31, functional feature 231, 333
37, 44-49, 51, 53, 56, 60, 73, 75, Functional Lexematic Model 41
77, 83, 91, 100, 103, 108, 110- functional minimalism ix, 167
111, 120-122, 127, 131, 133, 135, Functional Procedural Grammar ix,
137, 140, 142, 144, 153-154, 157- 181, 212, 275, 288, 295
158, 162, 164-165, 167, 169-170, Fund x, 63, 76, 162, 164, 187, 325
180-187, 189-190, 192-193, 197- Generalized Parallelism Hypothesis
202, 205, 207, 211-212, 214-215, x, 241, 299, 303
220, 224, 227, 234, 262-263, 270, Generalized Phrase Structure
277, 279-280, 285-288, 290, 293, Grammar 213, 337
295, 301, 306, 310, 312, 319, generative 27, 31, 37-38, 57, 179,
325-330, 332, 334-335, 339, 341- 199, 337, 339
343, 345-362, 368, 375, 377 genre 90, 94, 99, 107, 217, 307, 318
expression rule 2-3, 29, 37, 45-47, Given Topic 219
53, 62, 74, 165, 169, 182, 184, grammaticalization 76, 206, 293,
190, 192, 199, 211-212, 310, 312, 298, 349, 358, 360
325-328, 334, 339- 340, 342, 345, head 33
349, 356-358, 361-362, 377 Head-Driven Phrase Structure
extraclausal xiii, 90, 206, 303, 305 Grammar 213
finite 56, 121, 127, 165, 172, 192, hedge ix, 206
358 hierarchy ix, 2-3, 5-6, 36, 38, 43, 48,
focality 125, 131, 138-139, 145, 165, 73, 77, 91, 93, 95, 102, 106, 109-
175, 185, 190, 218, 222, 226, 111, 138, 171, 199, 212-213, 221,
228-232, 237, 343, 347 234, 272, 281- 285, 288-289, 293,
Focus viii, 117, 119-121, 124, 127, 295, 299-303, 310, 326, 329, 331,
130-131, 138, 140, 142-144, 155, 339, 342, 358, 365, 373, 377
160-161, 164-166, 168-170, 172, holophrase 20, 175, 185, 212, 302
175, 183-189, 191-193, 212, 217- hypotaxis 226, 237
218, 221-224, 226-227, 229, 231, iconicity 52, 204
233, 235, 237-238, 240, 318, 322, identifiability 5, 94, 144, 201, 215,
335, 363 220-222
ideology 94, 107, 110 185, 206, 211, 215, 243, 263, 267,
illocution 5, 11, 18, 43, 46, 91, 99, 279-280, 285, 306, 313, 320-321,
116, 153, 164, 171, 186, 222, 287, 326, 340, 357, 366, 370-371
307, 318, 347 interface 2, 76, 93, 98, 153, 161,
imperative 90, 259-262, 270-271 182, 197-199, 204, 358
increment ix-x, 186-187, 212, 217, interjection 107, 109, 300, 353
226, 230, 338, 343-344, 355 interpersonal vii-ix, 3, 5, 7-8, 10- 11,
Incremental Discourse Cognitive 13, 15, 17, 41-45, 47-48, 51-52,
Grammar ix, 215 73, 75-77, 82-83, 91, 99, 103,
Incremental Functional Grammar 106, 109-112, 167, 169, 171, 176,
viii, 25, 212 181, 197-198, 200-204, 206-208,
indefinite 37, 136-137, 144, 190 213, 225, 227-228, 232, 243-244,
indexical 117, 127, 137-138, 146 246- 248, 251, 264, 266-270, 273,
individual vii, 2, 6, 31, 56, 75- 76, 276, 285-286, 294-295, 306, 312-
90, 92, 94, 129, 144, 151-153, 314, 318, 325, 347, 358, 367,
160-161, 173, 182, 289, 294, 331, 369-370, 372-373, 375
358, 367 interpretation 10, 27, 37, 39, 46- 47,
inference 50, 132, 136, 153, 161, 50, 58, 74-75, 83, 100, 119, 122,
167, 221, 286, 294, 346 125, 129-130, 136, 143, 151-154,
inferrable 124, 230, 235 163, 168, 170, 181, 188, 207, 219,
information viii-ix, 2-3, 5, 8, 12, 15, 248-251, 270, 282, 284, 337, 341,
29, 35-36, 40, 47-48, 51-53, 74- 366-369, 373, 375
76, 78-84, 90, 98, 102, 104, 109- interrogative 90, 156, 243-244. 248,
110, 120-122, 124-126, 131, 135- 250-252, 257-258, 261- 263, 267
137, 139-142, 144-145, 153, 167, intonation 175, 183, 190, 192, 215,
172, 174- 175, 180-181, 185-187, 219, 221, 226, 228, 232- 233,
189, 191, 201-202, 204, 207, 212- 342, 344, 346
215, 217-223, 226, 228-231, 233, inverse 157, 161, 164, 173
235-236, 247, 277-279, 282-283, knowledge 3, 37, 39-40, 44, 74-76,
285, 289-291, 294- 295, 312-313, 78-79, 84, 105, 119, 131, 175,
318, 325-326, 329-332, 334-335, 186, 212, 214-215, 221, 232, 246,
337, 340-344, 346, 348, 354, 356- 254-255, 266, 282, 284, 287-291,
360, 368, 373, 375 295-297, 308, 326, 329, 337, 339,
inherent modal 245-247, 254 344-345, 357, 359-360, 369
inheritance x, 315, 316, 322, 331 label 110, 228, 235, 290, 293, 309,
instruction 81, 90, 119, 125, 355 333, 339, 377
intention vii, 2-3, 5-6, 10-11, 34, 47, language user 39, 56, 118, 180, 191,
52, 73, 82, 153, 165, 168- 169, 233, 326, 336-337, 339, 358, 362
174, 182, 184, 187, 192, 197-198, layering vii, 40, 43-44, 47-48, 51-52,
201-202, 205, 213, 287, 300, 337, 59-60, 80-83, 91-92, 108, 111,
341, 346-347, 359 197, 203, 206-207, 246-247, 266-
interaction 25-26, 30-31, 34, 39, 44, 267, 269-270, 293, 304-308, 314,
49-50, 60, 73-74, 76, 79- 80, 90, 317-319, 321, 325, 332, 344
109-110, 115, 153, 166, 182, 184-
left-to-right ordering x, 10, 17, 169, 331, 337, 339-348, 351, 353-355,
171, 176, 182, 186, 212- 213, 357, 360-361
329, 337, 358 message 16, 49, 73, 81, 84, 93, 98,
level viii-x, 2-3, 5-8, 10-13, 15-17, 108, 118, 132, 140, 142, 153, 166,
20, 26, 40, 43-44, 46-47, 56, 73, 175, 188, 189, 193, 200, 215, 224,
76-77, 81-83, 92-93, 99-100, 102- 226, 228, 232, 236, 320
103, 105-106, 108, 110-111, 115, message management 132, 140, 142,
122-123, 127, 129, 133, 139-143, 153, 166, 175
145- 146, 151, 153-155, 160-161, metalinguistic 13, 16, 78, 326, 340
163-165, 167-171, 174-176, 180, modality 58, 206, 243, 245, 249-258,
183, 185, 190, 197, 200-206, 213- 260-261, 264, 266-270, 273, 292-
215, 227, 233, 235, 243, 263, 265, 293, 297-298, 314, 316, 322, 372
268-269, 276, 281-283, 285-286, modal auxiliary 243, 249-251, 260,
288, 295, 300, 305, 308-309, 312- 270, 273, 293
314, 316-318, 321, 325-326, 328- modifier 33
329, 331, 337, 339, 341, 344, 347, modularity vii, 2-3, 44, 51, 74, 78,
349, 351-352, 355, 358, 366-370, 84, 92-93, 153, 197, 212, 239,
372-375, 377 243-244, 264, 266-268, 299, 309-
lexeme 12-13, 16, 54, 61, 368 310, 321, 365, 377
lexical 6, 10-11, 25, 28, 30-31, 35- module viii, 36, 44, 49, 93, 97-99,
36, 45-46, 55, 57-60, 75-76, 78, 132, 140, 153, 160, 163, 167-168,
83, 93, 102, 108, 128- 129, 137- 180, 267-269, 305, 310, 312, 321,
138, 154, 158-162, 164, 166-167, 326, 331, 339, 346, 365
173-174, 181, 184-185, 191-192, monitor 356-357, 359
199, 211, 213, 233, 278, 280-281, monologue 90, 92
288, 290, 296, 318-319, 322, 325- mood 79, 118, 172-173, 200- 201,
327, 330, 334, 337, 344, 362 206-207, 224, 259-262, 271-272
Lexical Functional Grammar 213, morphology 18, 46, 74, 131, 163-
337 164, 173, 222, 314, 318, 327,
lexical priority 334 331-332, 335, 337
lexicology 41 move 5, 10-11, 13, 43-44, 50, 60, 92,
lexicon 25, 30, 35, 46, 51, 55, 76, 98-99, 104, 108-109, 136, 153,
78, 152-154, 160-162, 164, 166, 183, 197, 201-202, 224, 246, 332,
167, 171, 174, 176, 184- 185, 347, 353
187, 190-192, 213, 288, 326, 335, multimedia 95, 105
337, 339-340, 345, 348, 353, 360 narrative 2, 99, 126, 307, 318, 321
linearization 36, 329 negation 28, 153, 156, 161, 279, 281
LIPOC 51, 180 New Focus 223
marker 107, 134, 136, 145, 157- 158, New Topic 131-132, 219
173, 251, 269-270, 316, 328, 346, newsworthiness 155, 160-161, 163-
358 164, 166, 169, 172, 235
memory 74-75, 78-79, 85, 87-88, node 75, 175, 331-335, 348-349,
212-213, 220-221, 232, 288, 291, 352-356
noun phrase 25, 28, 35-36, 42, 56, polarity 224, 247, 277, 279
198, 285, 331 politeness 90, 106-107, 305, 307,
nucleus 35, 57-58, 92, 121, 144, 152, 319, 322
172, 181, 314, 316, 325 pragmatics 3, 10, 31-32, 35-38, 44,
object 32, 61, 142, 178, 183, 224, 47-50, 53-54, 57-58, 76, 90, 93,
236, 374 97-98, 116-117, 119, 124, 131-
objective modal 59, 243, 247-252, 132, 139-142, 145, 153, 156, 160-
256, 269, 272-273, 277, 293, 297 161, 164, 168, 170-171, 174, 181-
183, 185-187, 211-212, 219, 235,
operator 37, 40, 58, 190, 207, 247- 247, 292, 302, 309-310, 312-313,
248, 264-268, 306, 313-316, 319, 318, 321, 325-326, 328, 339- 340,
332, 347, 353, 360, 365, 372 347-348, 351, 353, 358, 359, 365,
overgeneration 328 373, 374
P1 position 5, 9-12, 14-17, 43, 46, predicate vii, 2, 7, 25, 29-32, 35- 36,
52, 82, 99, 104-108, 112, 124, 40, 43, 57, 58, 60, 76, 124, 151-
145, 166, 170, 175, 189, 200, 212, 167, 169, 171-174, 182, 185, 187-
219, 226, 231, 328-330, 333, 343, 188, 198, 223-224, 262, 276, 281,
348-352, 354- 355, 367 284, 315, 325, 332-333, 347, 349,
paragraph 7-8, 92, 133, 136, 288, 373-374, 378
326 predicate frame 2, 29-32, 35-36, 40,
parataxis 226, 237 57, 76, 152, 160-162, 164, 171,
parenthesis 285 182, 374, 378
participant 49, 108, 121-122, 124, predicate phrase 7
145, 156, 172, 229, 236, 245, predication 25, 31-33, 36-37, 40, 42-
254-255, 372 43, 57-60, 120, 122, 124, 126,
pause 344, 346, 351, 353, 355 131, 152, 154, 159-162, 164-165,
perception 19, 179, 183, 214, 282- 174, 182, 219, 247, 260, 265, 267,
283, 291, 345, 370 270, 276, 284- 285, 293, 315,
percolation x, 164, 165, 169, 332, 335, 374, 378
334-335, 350, 354 prehension 152, 165, 175
performative 15, 78, 200, 203, 205- preposition 121, 351, 354
206, 270, 279, 287-288, 290-291, presentative 123, 223, 230-231
293-294 presupposition 37, 174, 222-223,
perspective 23, 25, 46, 55, 74, 76, 231, 235-236
84, 147, 164, 167, 187, 198- 199, primacy effect 212
204, 207, 217, 229, 247, 267, process x, 2, 10, 12, 17, 26, 34, 46-
276-277, 282, 292, 299, 323, 326, 47, 53, 73, 79-83, 90, 93, 95, 151-
338, 341, 343, 358- 359, 372, 378 153, 162-164, 166-170, 174-176,
phonology 7, 31, 46, 60, 78, 103, 180, 182, 186, 194, 197, 199,
212, 222, 226, 236, 327, 330- 212-213, 215, 217-218, 221, 232-
331, 359 234, 238, 275, 277, 280, 286,
pitch 233, 236 289-290, 295, 300, 314-316, 320-
placement 189, 212, 226, 304, 329, 321, 325, 329-332, 334, 337, 339,
334
342-343, 345, 347-350, 357- 358, referential 6-8, 10-12, 15-16, 44, 47-
366-367, 373 48, 53-54, 60, 100, 169, 171, 185,
production viii, x, 2, 7, 17, 34, 46, 188, 190, 197, 200- 201, 219,
50-52, 60, 73-75, 78, 85, 87-88, 223, 229, 307, 340, 351, 355, 367,
151, 170-172, 179-182, 185, 187- 372
188, 192, 197, 212- 213, 233, referential phrase 7, 11-12, 16, 169
241, 275-277, 288-289, 292, 295, reflexive 8, 127-129, 145, 240, 368
327, 331, 337- 338, 343, 359, 367 relativism 344
profile 220-221, 228 representation vii, 5-7, 10, 15- 17,
prominence 15, 124-125, 184, 212, 25, 28, 37, 39, 43, 45-46, 48, 50,
217-219, 222, 226-233, 235, 318 52, 77, 79, 82-83, 87, 90, 95-102,
pronoun 28, 60, 118, 127-129, 132, 109-110, 114, 119, 132, 140, 142,
134, 137, 143, 145, 147, 158 151, 164-166, 170, 175, 182, 185-
properties 6, 27-28, 31, 38, 53, 56, 186, 189-192, 198, 200, 203-204,
74, 79, 84, 103, 108, 128, 130, 213- 214, 238, 267-268, 272, 275-
141-142, 144, 158, 181, 185, 213, 278, 280, 282, 286, 290-291, 295,
217, 224, 270-271, 276, 278, 280- 300, 308-309, 324-329, 331, 333-
281, 283-284, 286, 302, 309-310, 334, 337, 339, 341, 344-345, 347-
312, 319, 338-339, 349, 366, 370, 348, 355, 359, 366-370, 372-373
372-373 request 80-83, 186, 190
proposition 58-59, 96, 108, 120, 152, restrictor 152, 165
154, 164-165, 173-174, 182-183, Resumed Topic 132, 219
207, 220, 222-223, 232, 235-236, rheme 27, 222, 231, 237
246-247, 251, 262, 265, 267, 270, rhetorical 43, 51, 92, 101-102, 109-
276-277, 304, 315 110, 270, 308-309, 312- 313
prosody 11, 46, 120, 131, 145, 172, rigid 163, 189, 214
222, 235, 375 salience 34, 126, 131, 139, 158, 163-
protasis 249, 251, 257, 288 164, 183, 221-222, 226, 228, 231,
psychological adequacy 2, 26, 28, 235-236, 307
34, 38-39, 51-52, 86, 170, 179- satellite 42, 92, 165, 172, 181, 253,
180, 182, 193, 211, 337- 338 306, 333, 350-352, 354-356
qualification 279, 284-291, 293, 296 schema 76, 183, 214, 220-221, 316
recency effect 212, 226, 230 scope ix, 44, 59-60, 84, 90, 92, 94,
recursivity 162, 329, 340 101, 158, 171, 174, 187, 197, 204,
reference ix, 3, 8, 13, 15-16, 25, 43- 207, 217, 226-228, 247, 251, 265,
44, 49, 53-55, 91, 108, 117, 133- 276, 278, 280- 281, 284-285, 288,
135, 152, 169, 185, 190, 198, 202, 294, 306-307, 319, 326
212, 219-221, 229- 230, 234, 236, selection restriction 57, 78
276, 280-281, 288, 326, 340, 372- semantics ix, 3, 6, 10-11, 26-32, 34-
373 38, 46-48, 50, 54-55, 57-59, 75-
referent 38, 100, 119-139, 143- 144, 76, 86, 88, 100, 103, 114, 118,
157, 164, 172, 175, 183, 220-221, 141-142, 145, 153, 158, 161, 166,
224, 231, 235, 347, 373 170, 177, 187-189, 200, 212-213,
222, 233, 248-249, 251, 277-281,
283, 285, 292-295, 298, 309-310, strategy 39, 53, 128, 130, 138, 143,
312, 316, 321, 325-326, 337, 339- 155-156, 168, 183, 190, 217, 224,
342, 344-346, 348-349, 351-352, 230, 232, 270, 277, 359
355, 358-361, 365, 368, 373 structural 35-36, 38, 111, 118, 155,
semantic memory 75, 88, 337, 339- 161, 204, 221, 223-224, 227, 229,
342, 344-346, 355, 360 233, 234, 270, 277, 292, 300,
Semiotic Grammar 267, 272 302-305, 310, 312, 314, 321, 368,
sentence 2, 5, 8, 15, 17, 26-28, 30, 370, 373, 375
34, 36, 40-41, 44, 48, 51, 56, 58, style 90, 96, 98, 107, 182, 307-309,
91, 100-101, 111, 118- 121, 126, 317, 319-320
128, 132, 149, 154, 156-159, 164, subact 185-186, 188, 190, 202, 204,
168, 174, 198, 201-204, 206-207, 206, 212
223-224, 236, 248, 285, 301, 305, subcategorization 358
310, 320, 326, 334-335, 337, 343, subject 32, 142, 160-161, 174, 180,
348-350, 353 187, 223-224, 230-232, 236, 333,
serialization 156, 158-159, 161, 164 335, 349, 358, 374
short-term memory 79, 232, 288, subjective modal 242-244, 247-252,
331, 340-341, 360 255-257, 260, 264, 266-268, 270,
sign 29, 56, 135, 168, 172, 249 277, 310, 314, 317- 318
speaker vii, ix-x, 2-3, 5-6, 8, 10- 12, subsidiary 5, 96, 110, 181, 201- 202
15-16, 34, 44, 46-48, 50, 53, 60, SubTopic 126, 235
74, 77, 80, 83, 90, 94, 99, 108, syllable 183, 191-192, 375
118-119, 121-122, 127-128, 131, syntax 27, 31-32, 34-38, 47-48, 50,
133-134, 136-140, 143, 145, 155, 54-55, 57, 60, 74, 78, 100, 142,
174, 180, 183-189, 191-192, 198- 145, 158, 160-161, 164, 166, 187-
202, 213-215, 220, 224, 231-232, 189, 192, 214, 222, 226, 231, 280,
234-236, 247, 249-252, 254-259, 285, 310, 324-325, 331, 334, 337-
261-263, 265-266, 270, 277, 282, 338, 341, 347, 349, 351, 354, 365,
284-289, 291, 294, 295, 325-326, 373, 377, 378
332, 337-340, 342-343, 346, 351, Systemic Functional Grammar 94,
355, 357, 359, 361, 367, 369, 103, 213, 267, 295
372-373 Tail 206
speaker model vi, 193, 237, 325, template 153-154, 163, 169-170,
338, 340, 357 188, 190, 265, 329-330, 333, 335,
speech act 40, 43-44, 91-92, 95, 97- 343, 350
98, 102, 110, 174, 197, 247, 277, tense 40, 58, 79, 91, 106, 110, 118,
286-287, 289, 294- 295, 310, 356 173, 187, 192, 244, 247, 259,
speech situation 3, 11, 58, 336, 340, 261-263, 265-267, 270, 277, 279,
343 307, 327, 333
state of affairs 6, 30, 57, 60, 123, term 32, 35-37, 42, 57-60, 75, 90,
140, 213, 223, 245, 254, 282, 341, 110-111, 119-130, 132-133, 137,
366-367 141, 144-145, 152, 155, 162, 166,
story 57, 230, 249, 295, 308, 313, 172-173, 185, 188, 222, 266, 285,
340-341, 346, 361
288, 300-301, 304, 313-314, 317, 214, 277, 303-305, 307-310, 312,
325, 329, 333, 335, 352, 372 316, 321, 324-329, 331, 333-334,
text x, 13, 42, 78, 89, 94, 105, 118, 337, 339, 344, 347-349, 357-360,
121, 126, 135, 144, 147, 159, 217, 365-370
220-221, 224, 229, 234, 236, 269, unification 80, 299, 334, 349, 359
300-303, 305, 309-310, 312- 316, upward layering 2-3, 8, 43, 92-93,
318, 320, 346, 358 96, 197, 299, 310, 325, 348, 372
theme 27, 249 utterance 2, 6, 27, 44, 54, 90, 94, 97-
thetic 120, 122, 139, 143-144, 149, 98, 101, 116, 118-122, 124, 126-
165, 170, 174, 223 127, 131, 138-140, 143- 144, 150,
thinking-for-speaking 199 166, 169-170, 172, 183, 185, 188,
top-down vii, x, 2, 10, 12, 17, 73, 190-192, 198- 199, 203, 211-212,
117, 153, 161, 182, 197-199, 212, 223-224, 227, 235-236, 242, 248,
243-244, 264, 266-268, 269, 325, 251, 259, 265-267, 280-281, 285-
329, 331, 337, 365 288, 290, 295, 300, 324-325, 331,
Topic viii, 77, 91, 106-108, 110-111, 337, 341-343, 359-360, 370
117, 120-122, 124, 127, 131, 133, valency 78, 162, 185, 192
136, 143-144, 158- 159, 217, 229, variable 7-8, 30, 40, 43-44, 53, 59-
303, 346, 369 60, 152, 236, 270, 344, 347, 349,
topicality 91, 120, 122, 124, 138- 355, 366
139, 143, 145-146, 186, 218- 219, verbal interaction 25-26, 31, 34, 39,
222, 308, 343, 347 44, 48, 50, 60, 153, 182, 211, 313,
transaction 82, 96, 106, 108-109, 340, 366, 369-371
111, 187 viewpoint 58, 153, 189, 199, 229,
tree 10, 96, 199, 202, 329-332, 335- 234, 240
336, 350-352, 354-356, 359 vocabulary 74-78
typology 25, 30-31, 56, 59, 162-163, voice 94, 104, 110, 180
167, 169, 211, 270, 300, 305, 312, volition 245, 254-256, 293
316-319, 321, 328, 332 West Coast Functionalism 213
undergeneration 327-328 word order (see also constituent or-
underlying structure vii, 2, 26, 28, der) 122, 155, 188, 234, 304, 375
30, 34, 36-37, 39, 45, 47, 49-51, working memory 75, 79, 337, 339-
56, 58, 77, 98, 132, 142, 151-152, 340, 342-343, 346-348, 351, 353-
160, 163, 170- 171, 174, 182, 355, 360

A New Architecture For Functional Grammar 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

A New Architecture For Functional Grammar PDF

Загружено:

Авторское право:

Доступные форматы

A New Architecture for Functional Grammar

Library of Congress Cataloging-in-Publication Data

A new architecture for functional grammar / edited by J. Lachlan

Preface ......................................................................................... vii

The Architecture of a Functional Discourse

Remarks on layering in a cognitive-functional

Index of names .............................................................................. 379

This volume arises indirectly from the Ninth International Conference on

FDG brings grammar closer to cognitive science in introducing a

interpersonal component analyses the Move as a sequence of Acts, each of

ture of the word. Moutaouakil proposes a Generalized Parallelism Hy-

Our gratitude goes to Susanna Hop MA for her invaluable assistance

J. Lachlan Mackenzie María de los Ángeles Gómez-González

A Discourse Act; addressee FPG Functional Procedural

LIPOC Language Independent Ps speaker

ous levels of analysis are necessary. The model of a Functional Discourse

3. Outline of the model

Figure 1. Outline of FDG GRAMMAR

4. Levels and layers

4.2. The interpersonal level

Figure 2. The interpersonal level

Note that this representation is non-exhaustive, in the sense that there

4.3. The representational level

(p1: [(e1: [(f1) (x1)] (e1))] (p1))

Figure 3. The representational level

Note that, again, this representation is non-exhaustive. There are higher

4.4. The expression level

Figure 4. The expression level

This representation is just a simplified example of what the expression

expression possibilities, which lead to different expression units in their

Figure 5. Hengeveld (1989/1997)

The major differences at the interpersonal and representational levels,

4.6. Upward layering

Figure 6. Levels and layers

5. Making choices at the interpersonal level

(1) George, watch out for that tree!

Top-down decisions are of a more complex nature. These concern the

(A1: [EXPR (P1)Sp (P2)Addr (C1) ] (A1))

(A1: [VOC (P1)Sp (P2)Addr (C1: [ (R1) ] (C1))] (A1))

Now consider a case in which the speaker draws on the representational

(4) The Plaza Santa Ana is the best place to go.

(A1: [DECL (P1)Sp (P2)Addr (C1) ] (A1))

(A1: [DECL (P1)Sp (P2)Addr (C1: [ (R1) ] (C1))] (A1))

(6) I want to know your opinion.

(A1: [DECL (P1)Sp (P2)Addr (C1: [ (R1) ] (C1))] (A1))

(7) The Plaza Santa Ana is wonderful, don’t you think?

(A1: [DECL (P1)Sp (P2)Addr (C1: [ (T1) ] (C1))] (A1))

(A1: [DECL (P1)Sp (P2)Addr (C1: [ (T1) ] (C1))] (A1))

The examples just given serve to illustrate how linguistic expressions

6. Complex interactions between levels

(9) a. If you behave well, I’ll let you read my poems.

In (9b) the demonstrative that refers to the preceding move. In (10) it

(m1: [(a1: [(c1: [...(t1) (r1)... ] (c1))] (a1))] (m1))

Figure 7. Levels and layers GRAMMAR

6.2. Referents from the interpersonal at the representational level

(11) a. (Me) temo que Juan esté enfermo.

(12) The Morning Star ís the Evening Star.

6.3. Referents from the expression level at the representational level

2.1. Chronological Overview

Figures 1 and 2 provide an overview of these five periods.2

Figure 1. Functional Grammar publications from 1978 to 19983

FG0 Dik 1968. Coordination

Figure 2. Historical Overview of Functional Grammar