Вы находитесь на странице: 1из 349

Discourse and Grammar

Studies in Generative Grammar 112

Editors

Henk van Riemsdijk


Harry van der Hulst
Jan Koster

De Gruyter Mouton

Discourse and Grammar


From Sentence Types to Lexical Categories

Edited by

Gnther Grewendorf
Thomas Ede Zimmermann

De Gruyter Mouton

The series Studies in Generative Grammar was formerly published by


Foris Publications Holland.

ISBN 978-1-61451-215-8
e-ISBN 978-1-61451-160-1
ISSN 0167-4331
Library of Congress Cataloging-in-Publication Data
A CIP catalog record for this book has been applied for at the Library of Congress.
Bibliographic information published by the Deutsche Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the Deutsche
Nationalbibliografie; detailed bibliographic data are available in the Internet
at http://dnb.dnb.de.
2012 Walter de Gruyter, Inc., Boston/Berlin
Typesetting: PTP-Berlin Protago-TeX-Production GmbH, Berlin
Printing: Hubert & Co. GmbH & Co. KG, Gttingen
Printed on acid-free paper

Printed in Germany
www.degruyter.com

Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Gunther Grewendorf and Thomas Ede Zimmermann

Part I. Semantic and Pragmatic Properties of Sentence Types


Implicatures in Discourse . . . . . . . . . . . . . . . . . . . . . . . . .
Nicholas Asher

11

Permission and Choice . . . . . . . . . . . . . . . . . . . . . . . . . .


Paul Portner

43

Promises and Threats with Conditionals and Disjunctions . . . . . . . .


Robert van Rooij and Michael Franke

69

Part II. Sentence Types and Clausal Peripheries


Revisiting the CP of Clefts . . . . . . . . . . . . . . . . . . . . . . . .
Adriana Belletti

91

Delimitation Effects and the Cartography of the Left Periphery . . . . . 115


Luigi Rizzi
Sentence Types and the Japanese Right Periphery . . . . . . . . . . . . 147
Mamoru Saito

Part III. Clausal Properties of Lexical Categories


On NPs and Clauses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

Zeljko
Boskovic
Discerning Default Datives: Some Properties of the Dative Case
in German . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Henk van Riemsdijk
How Phase-Based Interpretations Dictate the Typology
of Nominalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Leah Bauke and Tom Roeper
Scope and Verb Meanings . . . . . . . . . . . . . . . . . . . . . . . . . 321
Edwin Williams

Introduction
Gunther Grewendorf and Thomas Ede Zimmermann

Bringing together papers from various subfields of theoretical linguistics, this


volume gives a representative glimpse of current research on form and function
in grammar. Its overarching topic is as old as it is hot: the relation between the
major clause types as determined in syntax, and their canonical or idiosyncratic
roles in discourse as characterized in pragmatic terms. Though none of the
papers addresses this topic in its full breadth, they can all be seen to make their
specific contributions to it, scrutinizing the pertinent aspects of the grammatical
interfaces and elaborating detailed case studies.
The way from pragmatics to syntax leads via the semantic component of
grammar. The first part of this collection comprises three papers devoted to the
semantics/pragmatics interface. To a large part and with Gricean pragmatics
as a role model, contemporary pragmatics centers around the reconstruction
of rational communicative behavior as limited by the bounds of grammar, i.e.
primarily semantics.
Despite this general perspective, Asher shows that the dependence of scalar
implicatures on discourse structure cannot be captured by Gricean global rationality alone correction sequences being a simple case in point:
(1)

a.
b.

A: John either did some of the reading or he did some of the homework.
B: John did all of the reading, but youre right, he didnt do all of
the homework.

Due to the illocutionary role of Bs utterance (1b), one of the scalar implicatures
of (1a) gets cancelled, while the other one is preserved. This suggests that implicature calculation may depend on discourse structure a hypothesis that Asher
pursues and generalizes, ultimately developing a unified account of implicatures
as defeasible implications that also captures more intricate cases of embedded
implicatures that seemingly support localist approaches of implicature calculation.
Portners contribution provides an analysis of two related properties of imperatives: (i) their variation in discourse function and (ii) their licensing of free

Gunther Grewendorf and Thomas Ede Zimmermann

choice inferences. With regard to (i), it is argued that imperatives are semantically uniform, and that their wide range of interpretations is explained by two
factors: differences in the grounds for issuing a given imperative and the logical
relationship between the imperative and other commitments of the addressee.
Concerning (ii), the same ideas which are used to analyze permission, in combination with an alternatives semantics for disjunction and indefinites, are able
to explain free choice and related phenomena, such as Rosss Paradox and the
licensing of any.
Van Rooij and Franke address a puzzle concerning performative uses of
declarative sentences. As the following pair illustrates, threats may be expressed
by conditionals or corresponding disjunctions, in accordance with logically motivated paraphrase:
(2)

a.
b.

You will give me your wallet or I will punish you severely.


If you dont give me your wallet, I will punish you severely.

However, unlike its conditional counterpart (3a), a disjunctive statement like


(3b) does not lend itself to use as a promise:
(3)

a.
b.

If you give me your wallet, I will reward you splendidly.


You will not give me your wallet or I will reward you splendidly.

Using techniques from game theory to model the difference between promises
and threats, the authors provide a purely pragmatic explanation of the observed
distribution of use conditions. Retaining the logical equivalence between disjunctions and conditionals, the analysis exploits the difference in wording and,
in particular, an asymmetry created by mentioning and thus bringing to the fore,
different possibilities.
The second part of this volume deals with the question of how the constitution of sentence types can be related to properties of functional categories in the
clausal periphery. Based on a cartographic approach to sentence structure the
contributions of this part of the collection show that in left-headed languages,
the functional heads of the left clausal periphery have interpretive effects on the
interfaces that determine the semantic properties of scope-discourse configurations as well as the prosodic properties associated with information structure.
It is also shown that in a right-headed language like Japanese, the configuration
of the functional heads that are relevant to the determination of sentence types
form a right-peripheral mirror image of the left clausal periphery.
Rizzis contribution illustrates some of the results of cartographic studies
on the left clausal periphery and discusses the implications of the cartographic
approach for the study of the interfaces that connect syntax with the systems

Introduction

of sound and meaning. In particular, Rizzi shows that the functional heads assumed by the criterial approach not only have a syntactic function in attracting
elements to the periphery of the clause but also have interpretive effects on the
interfaces with meaning and sound: they signal the interpretive properties of
scope-discourse configurations and give instructions to the prosodic component
to yield the appropriate contour associated with the various discourse functions.
He then compares the cartographic approach, which assumes that complex syntactic configurations are derived by a simple computational system, with an
approach that proceeds from a simpler functional lexicon, assuming only a
single C-layer and admitting multiple adjunctions to TP. Rizzi argues that the
cartographic approach is conceptually and empirically superior to the simpler
approach in that the latter shifts crucial non-interpretive properties such as
positional and co-occurrence restrictions (including parametric variation) to the
interfaces, and is unable to deal with phenomena such as co-occurring multiple
C-particles or the distributional properties of topic and focus particles. By contrast, the cartographic approach simplifies the burden of the interface systems
by syntacticizing scope-discourse functions without enriching the computational system with specific mechanisms operative at the interfaces. Finally, Rizzi
shows that the criterial positions associated with interpretive properties such as
Force, Focus, Aboutness etc. have the effect of delimiting movement chains
(by Criterial Freezing) and thus offer new ways of approaching the classical
problem of locality.
Saito investigates the complementizer system of Japanese and claims that
this system is identical to the complementizer system of Spanish. He shows that
Japanese has three phonetically distinct complementizers the functions of which
correspond to the three (phonetically non-distinct) Spanish complementizers
posited by Plann (1982): a complementizer (to) that introduces paraphrases or
reports of direct discourse, a complementizer (ka) that selects questions, and a
complementizer (no) for embedded tensed propositions, which represent events,
states, or actions. Based on the observation that these complementizers, when
co-occurring, always appear in the order no-ka-to, Saito argues that their hierarchical organisation correlates with the functional structure that Rizzi (1997)
suggested for the Italian left clausal periphery: no represents the head of FinP,
ka is located in the head position of ForceP, and to is the head of a ReportP
that is located above ForceP and is not represented in Rizzis system. He then
shows that in Japanese, Topic heads can be (recursively) generated above Fin
and below Force, which leads him to the conclusion that the peripheral clausal
structure of Japanese can be represented as a right-peripheral mirror image of
the Italian left periphery:

4
(4)

Gunther Grewendorf and Thomas Ede Zimmermann

[. . . [. . . [. . . [. . . [TP . . . ] Fin (no)] (Topic*)] Force (ka)] Report (to)]

Based on a cartographic framework, according to which different syntactic positions in the functional structure of the clause overtly express different interpretations, Belletti shows that cleft constructions display interpretive variations
that correlate with variation in their functional structure. Her crucial assumption
is that for all clefts, the sentential complement selected by the copula represents
a reduced CP that lacks the highest ForceP layer. She then points out interpretive differences between subject and non-subject clefts. While the former can
be utilized as an answer to a question of information as well as an expression
of contrastive focalization, non-subject clefts can only have the latter option.
Belletti therefore assumes that a clefted subject that is interpreted as the (noncontrastive) focus of new information moves into the low vP-peripheral focus
position of the matrix copula, passing a left-peripheral EPP-position in the reduced CP complement. By contrast, clefted constituents that are contrastively
focused (subjects as well as non-subjects) move to the left-peripheral focus
position in the reduced CP complement of the copula. The fact that only subject clefts can undergo movement to the internal focus position of the matrix
clause is shown to be a consequence of independent locality principles: in the
presence of the EPP in the left periphery of subject clefts, object movement
is excluded due to relativized minimality; in non-subject clefts, where there is
no left-peripheral EPP, long movement of non-subjects can be excluded by the
phase impenetrability condition. Since wh-extraction from the reduced CP cannot utilize the embedded left-peripheral focus position due to criterial freezing,
wh-clefts are analyzed as extraposition of the embedded FinP combined with
remnant movement of the reduced CPs entire FocP. Finally, Belletti shows that
other structures such as the sentential complement of perception verbs share
crucial properties with the configuration of clefts.
The contributions in Part III of the volume deal with the interaction of lexical
elements and clausal functional categories, each of them revealing unexpected
parallels between clause structure and the internal structure in other, particularly
lexical categories, including the role of their peripheries.
Boskovic shows that there is a surprising interplay between the internal syntax
and semantics of the traditional Noun Phrase and clause-level phenomena. He
argues that there is a fundamental structural difference between languages with
articles such as English, and article-less languages such as Serbo-Croatian. On
the basis of several new generalizations, crucially related to the role of articles,
Boskovic demonstrates that this difference can be captured by the assumption
that article-less languages lack the category DP, which yields a typological distinction between DP languages and NP languages. The new generalizations that

Introduction

he adduces in support of this distinction involve phenomena such as negative


concord with complex negative constituents, quantifier scope, radical pro-drop,
number morphology, and the interpretation of possessives. He then explores consequences of the internal structure of the traditional nominal projection for the
internal structure of clauses, proceeding from the assumption that the structural
difference on the nominal level between languages with and without articles
has its counterpart on the clausal level. His crucial claim is that just like the
nominal structure is poorer in an NP language than it is in a DP language, the
structure of clauses is poorer in the former than it is in the latter in the sense
that the lack of DP in the nominal structure implies the lack of TP in the clausal
structure. That the clausal structure of article-less languages indeed lacks TP
is shown to receive independent support from crosslinguistic generalizations
involving phenomena such as subject expletives, asymmetries in the locality of
subject and object movement, and the Sequence of Tense phenomenon. The fundamental conclusion of this paper is that many clause-level phenomena cannot
be properly understood unless close attention is paid to the internal structure
and interpretation of nominal categories.
Van Riemsdijk presents a novel analysis of the apparent alternation between
datives and accusatives in German spatial prepositional phrases, which is known
to correlate with a locative vs. directional interpretation:
(5)

a.

Peter legt das Buch auf den


Tisch.
Peter puts the book on theACC table

b.

Tisch.
Das Buch liegt
auf dem
the book is-lying on theDAT table

Building on semantic investigations by Joost Zwarts, van Riemsdijk decomposes


directionality into source, route and goal, and goes on to argue for a domaindependent default case, which in oblique domains like PP and AP is the dative.
On this analysis, the accusative in directional PPs specifying goals or paths, come
out as depending on a (possibly implicit) measure phrase, whereas the dative
manifests itself in the elsewhere environments of locative or source PPs.
Bauke and Roepers paper analyzes nominal gerunds in English and argues
for a distinction between two types of nominal gerunds. Both types are shown to
project verbal functional structure, associated with an aspectual projection inside
the nominal compound. The distinction between the two types is argued to follow
from the position of the gerundive -ing affix. In one type of gerund this affix
is generated under the aspectual head, where it licenses a number of modifiers
(such as aspectual modifiers, non-sentential adverbial modifiers, prepositional
modifiers etc.). Further, Bauke and Roeper show that identifying the -ing affix as

Gunther Grewendorf and Thomas Ede Zimmermann

an aspectual marker allows for a natural explanation of why plural morphology


is absent from this type of nominal gerund. In the other type of gerund, the
affix is generated under a nominal node instead. This is argued to predict that
plural morphology is licensed on this type of gerund. Bauke and Roeper then
illustrate that this does not imply that aspectual structure is not projected in
this type of nominal gerund. On the basis of insights from Phase Theory it is
shown that access to the aspectual projection inside the gerund is simply blocked
by higher Phase-nodes. The authors outline further that both types of nominal
gerunds allow for incorporation and that the distinction between incorporated
and non-incorporated forms reduces to a symmetry-distinction according to
which a symmetric c-command relation between the head of the gerund and its
bare internal argument forces incorporation. In an additional step they derive
further support for the distinction between two types of nominal gerunds from a
comparison of the English constructions with their German counterparts, where
the two types of -ing affixes are realized as -ung and -en respectively.
Williams observes and investigates a surprising difference between verbs and
prepositions that shows in their interaction with certain quantifying elements
like only:
(6)

a.
b.

I only paid 5 dollars for it. I paid only 5 dollars for it.
/ I bought it only 5 dollars for it.
I only bought it for 5 dollars.

Generalizing these observations, Williams makes out a general pattern behind


the data in (6) due to an asymmetry in the scopal behavior among lexical categories: whereas verbs adhere to (7) and thus generally commute with quantifiers, no such generalization holds for prepositions:
(7)

No verb can have scope over its DP complement.

Williams then goes on to show how the asymmetry between verbs and prepositions is naturally accounted for within the Representation-theoretic architecture
of Williams (2003): like clauses, prepositional phrases may be subject to Level
Embedding.
The papers collected in this volume were selected from the contributions to
the conference 10 Years After, which took place in Frankfurt in June 2009, organized by the members of the Graduiertenkolleg Satzarten (Research Training
Group on Sentence Types), a temporary doctoral program supporting students
in theoretical and descriptive linguistics and specializing on the particular problems surrounding the distinction between the major clause types. The conference, which marked the end of the program, brought together a number of international experts who have advised and supported our students over a period of

Introduction

roughly 10 years (thence the title). In the name of our current and former students
and colleagues, we would like to take the opportunity to thank our advisors for
continuous support and encouragement, as well as the Deutsche Forschungsgemeinschaft (German Research Foundation) for their generous funding. Finally,
we would like to thank Benjamin Hubner, Jonas Metzler, Jacob Schmidkunz,
Christian Stidronski and Dina Voloshina for helping us with the preparation of
the manuscript.
It was our decision not to include any indices in this volume.
Frankfurt, November 2012
GG & TEZ

Part I.
Semantic and Pragmatic Properties
of Sentence Types

Implicatures in Discourse
Nicholas Asher

1.

Introduction

Implicatures are an interesting case study for the role of discourse in interpretation and in grammar generally. Scalar implicatures are the result of a defeasible
inference.
(1)

a.
b.

John had some of the cookies.


John had some of the cookies. In fact he had them all.

(1a) has the implicature that John didnt have all the cookies, this implicature
can be defeated by additional information, as in (1b). Nevertheless, scalar implicatures seem closely tied to lexical choice or structural factors, which has
led some authors, most notably Chierchia (2004) but also Fox (2007) and others to incorporate the generation of implicatures within the syntax-semantics
interface i.e., clearly at the core of grammar. However, scalar implicatures
share the characteristic of defeasiblility with inferences that result in the presence of discourse relations that link discourse segments together into a discourse
structure for a coherent text or dialogue call these implicatures discourse or
D-implicatures. I have studied these inferences about discourse structure, their
effects on content and how they are computed in the theory known Segmented
Discourse Representation Theory or SDRT.
This paper argues for the centrality of discourse structure in linguistic interpretation by detailing how discourse structure provides an important source of
information to computing scalar implicatures. Scalar implicatures have typically
received a Gricean treatment based on reasoning about the content of sentences
in isolation since the work of Larry Horn (1972). I show here how the Gricean
tradition has missed an important, indeed decisive, component in the calculation
of implicatures. This has general implications for the way discourse structure is
treated within grammar. I argue that discourse structure affects scalar implicature and that much the same procedures are operative in both. So if implicatures
are part of the grammar discourse structure is too. By integrating discourse and
implicature together, we get some nice consequences: at the theoretical level, we

12

Nicholas Asher

have a unified and relatively simple framework for computing all implicatures;
second, we have a clear account of the interaction of D-implicatures and scalar
implicatures in many cases; finally, we can capture the intuitions of so called
localist views about scalar implicatures, while making this compatible with a
broadly Neo Gricean framework; finally, since in my view discourse structure
triggers scalar implicatures, this goes some way towards explaining the variability of embedded implicatures noted recently (e.g., Geurts and Pouscolous
2009).
2.

D-implicatures

Lets examine some examples of D-implicatures.


(2)

a.
b.
c.
d.

John walked in. He poured himself a cup of coffee.


John fell. Mary pushed him.
We bought the apartment, *but weve rented it.
Il commence a` dessiner et peindre en 1943 , * frequente les ateliers
de sculpture * puis de peinture de l e cole des Beaux-Arts d Oran,
* o`u il rencontre Guermaz (ANNODIS corpus).

A presumption of relevance leads us to infer some link between elementary


discourse units or EDUs (clauses or subclausal units whose boundaries are either
sentence initial or marked by * in the examples above). These links introduce
relations taking these discourse units as arguments that are familiar even to the
non-linguist: some units elaborate or go into more detail concerning something
introduced in another constituent (these are Elaboration type relations) as in
(2d); some units form a parallel or a contrast with other units (such units are
linked by Parallel or Contrast), as in (2c); some units furnish explanations why
something described in another unit happened (Explanation) as in (2b); and
some units constitute a narrative sequence of events (Narration) as in (2a) or
(2d). Other discourse relations of interest for our purposes are indirect question
answer pairs (IQAP), which link responses to a prior question, Correction, where
a second discourse move revises the content of a first, and Alternation, which is
linked to certain uses of disjunction.
Some D-implicatures are encoded grammatically through the use of certain
grammatical constructions (like adverbial or purposive clauses, parentheticals
or left fronted temporal or spatial adverbials)1 or through discourse connectors
like as a result, because, but, . . . or the choice and sequencing of lexical items.
1. For a discussion of these, see for instance, Vieu et al. (2005)

Implicatures in Discourse

13

An example of a set of discourse relations triggered by the choice of verb and


complement comes in (2d), with the use of beginning with, followed by and
ending with. Sometimes, it is less clear what linguistic source triggers the inference of the discourse relation as in (2a)(2b) most likely, an as yet not fully
understood mix of lexical semantics and world knowledge.
While scalar implicatures can occur embedded under quantifiers and other
operators, many people have noted that these implicatures are less robust in
many contexts than the unembedded cases of scalar implicatures like (1a). Dimplicatures, on the other hand, robustly embed under quantifiers and other
operators (as well as other discourse relations).
(3)

a.
b.
c.
d.

If it was late, John took off his shoes and went to bed.
If it was late, John went to bed and took off his shoes.
If John drank and drove, he put his passengers in danger.
The CEO of Widgets & Co. doubts that the company will make a
profit this year and that (as a result) there will be much in the way
of dividends for shareholders this year.

In both (3a) and (3b), the D-implicature that there is a narrative sequence between
the two clauses in the consequent of the conditional survives under embedding,
and (3c) shows that this holds in the antecedent of a conditional as well. (3d)
shows that the causal relation of result holds when embedded under a downward entailing attitude verb. Discourse relations like explanation occur between
clauses under modals and other elements of discourse in a sort of quantifying
in way as in (4a) and (4d). Note that the paraphrase of (4a) is (4c) not (4b).
(4)

a.

John broke his leg. Sam told me (I think, Its likely) he slipped on
the ice.
b. #John broke his arm because [Sam told me (I think, its likely) he
slipped on the ice.]
c. John broke his arm. Sam told me (I think, its likely) its because he
slipped on the ice.

Thus, D-implicatures take scope underneath and across various syntactically


realized operators.

14

Nicholas Asher

3.

Discourse Structure and D-implicatures

To investigate D-implicatures further, we need a theory about the structure, the


construction and the interpretation of discourse structures with an emphasis on
the computation of D-implicatures. A theory of discourse structure must answer
to three tasks:
It must segment a text into EDUs.
It must provide attachment points for EDUs in a discourse structure.
It must provide one or more discourse relations between an EDU and its
attachment point(s).
SDRT (Asher 1993; Asher and Lascarides 2003) is a theory that does all of these
things but also provides an account of how to calculate D-implicatures.An SDRT
discourse structure or SDRS is a logical form for discourse with a well-defined,
model-theoretic semantics that has many equivalent representations as a sui
generis, quasi first order model like structure consisting of a set of labels and
assignments of formulas to labels (Asher and Lascarides 2003), as a DRS like
structure (Asher 1993) or as a term in intensional logic (Asher and Pogodalla
2010). An SDRS is the counterpart to a formula of intensional logic in Montague
Grammar for discourse interpretation.
To get an idea of what SDRSs look like consider the following text (5)
discussed at length in Asher and Lascarides (2003). The model-like SDRS is
given in (5) . Note that some EDUs, e.g., 2 and 5 , are grouped together into
complex discourse units, e.g., 6 .
(5)

1
2
3
4
5

(5)

A, F , Last, where:

John had a great evening last night.


He had a great meal.
He ate salmon.
He devoured lots of cheese.
He then won a dancing competition.

A = {0 , 1 , 2 , 3 , 4 , 5 , 6 , 7 }
F (0 ) = Elaboration(1 , 6 )
F (6 ) = Narration(2 , 5 ) Elaboration(2 , 7 )
F (7 ) = Narration(3 , 4 )
Last = 5
In SDRT we can abstract away from the details of the structure to get a
graph representation, which is relevant to computing discourse accessibility for

Implicatures in Discourse

15

anaphoric antecedents and sites for presupposition binding and accommodation


(again for details see Asher and Lascarides (2003)).
1
Elaboration
6
Narration

Elaboration
7
3

Narration

Notice that some discourse relations are represented as vertical arrows in the
graph whereas others are horizontal arrows; these correspond to two different
types of relations subordinating and coordinating relations, and these two
types of relations affect anaphoric and attachment possibilities differently.2 The
non-arrow lines show which discourse units are parts of more complex discourse
units.
3.1. Inferring D-implicatures
Inferring D-implicatures is a matter of defeasible and uncertain inference. Many
of the features used to infer discourse relations are only good indications of a
particular discourse relation or particular discourse structure; very few are in
and of themselves sufficient to deductively infer the relation or structure. Many
discourse connectives are for example ambiguous. In addition, many segments
may bear discourse relations to other segments despite the lack of discourse
connectives or known structural or lexical cues, as in (2a), (2b) or (5). To solve
this problem, my colleagues and I developed a non-monotonic logic, a logic for
defeasible inference, tailored to inferring D-implicatures.
The task of building such a logic is not completely trivial. Integrating nonmonotonicity in discourse interpretation is problematic, especially if this integration occurs at the level of contents or what is said. Reasoning over contents
non-monotonically requires finding a class P of preferred models, those with
the intended discourse relations, and computing validity or logical consequence
2. For a discussion, see for example Asher 2008; Asher and Lascarides 2003.

16

Nicholas Asher

with respect to P. Given that the language of information content is at least


that of first order logic, where the complexity of the computation of validity
and logical consequence is only recursively enumerable and that almost all nonmonotonic logics require some sort of consistency test over the formulas one is
using for the inference, the complexity of computing logical consequence with
respect to the class P of preferred models is not recursively enumerable i.e.,
computationally hopeless. This is not just a matter of implementation but one
of principle. We cannot assume that agents, with their limited computational
capacities, are able to solve a problem reliably which we can show mathematically to be incapable of having anything like what we would call an algorithm.
Attributing such computational capacities to agents shows that we have mischaracterized the problem: they are not computing logical consequence over
formulas of information content; either the language in which the computation
is done is somehow a simplification of the language of information content, or
they are computing something other than logical consequence, perhaps using
some sort of heuristic.
One could develop a heuristic for computing D-implicatures. In fact, Fox
(2007) as well as others have proceeded to do this for scalar and free choice
implicatures. The problem with taking this tack, however, lies in its verification.
In order to make sure that the heuristic is doing what it is supposed to, we
must check it in the way that computer programs are checked, via program
verification. In most if not all instances, this means translating the problem into
a logic and then checking that the result desired is in fact a logical consequence
of the programs translation and the input data. This leads us back to the task of
building a non-monotonic logic for D-implicatures.
SDRTs solution to this problem is to look at non-monotonic reasoning not
over contents but over logical forms. Roughly, instead of trying to compute the
non-monotonic consequences of a bunch of facts about the world, facts which
may be quantificationally complex and which leads to an unsolvable problem,
we compute the non-monotonic consequences of a discourse logical forms having a certain shape and of a segments having the lexical choices and structure
that it does. This means that we are trying to solve a logical consequence problem, not in the language of information content, but in a language for describing
discourse logical forms and features of discourse constituents. Asher and Lascarides (2003) develop such a language, which they call the glue language. The
non-monotonic logic adapted to this language is known as the glue logic or GL.
Asher and Lascarides (2003) show that the problem of logical consequence for
formulas of this language is in fact decidable.
This approach has an additional advantage. Computing implicatures, as well
see, is sensitive to the structure of discourse and of individual sentences. This

Implicatures in Discourse

17

is available at logical form, but not at the level of semantic content, which is
described in terms of sets of possible worlds. This brings the calculation of
implicatures into an area closer to the syntax-semantics interface, as well see.
GL uses axioms exploiting various resources to get the intended discourse relations to hold between discourse constituents. The general form of such axioms
is this:
General Form: (?( , , ) some stuff) > R( , , )
> is a weak conditional; some stuff is information about , and thats
transferred into the glue language from more expressive languages for other information sources like: compositional semantics, lexical semantics, pragmatic
maxims of conversation, generalizations about agent behavior in conversation,
and domain knowledge. The semantics of > was developed by Asher and Morreau (1991) in a first order non-monotonic logic known as common sense entailment. This is a logic for non-monotonic or defeasible reasoning based on a
weak conditional >. Originally devised to treat generics, I have used a version
of it restricted to a quantifier free description language, the glue language, to
calculate D-implicatures, and it is a relatively adaptable non-monotonic logic. It
has two parts: a basic, monotonic, conditional logic with a standard proof theory
 and consequence relation |=, and then a defeasible inference relation | and
a non-monotonic consequence relation | that make use of the basic logic. I
will use the glue logic version of common sense entailment to model both Dand S-implicatures.
Let me briefly recapitulate the basics of common sense entailment restricted
to a propositional language.
A modal generic frame F = W ,  where W is a non empty set of worlds and
: W P(W ) P(W ) is a selection function. A model A is constructed by
adding a valuation function.
A, w |= A > B iff (w, ||A||) ||B||
the standard clauses for the quantifiers and connectives, though the glue
language itself has quantifier free logical structure.
This semantics for > will be familiar to those accustomed to conditional logics,
and has a complete axiomatization.3
How do we pass from a notion of monotonic consequence to a non-monotonic
one? The idea of the non-monotonic consequence relation is to assume that
matters are as normal as possible given the information in the premises and
3. The axiomatization for the first order language can be found in Asher and Morreau
(1991).

18

Nicholas Asher

to then see what follows. Assuming matters are as normal as possible means
making > as much like the material conditional as possible moving from
if p then normally q to if p then q. There is both a proof theoretic method
for defining nonmonotonic consequence ( |) and a model theoretic method
( | ) for which a correspondence theorem is given in Asher (1995). The proof
theoretic method is less involved, and I briefly sketch it here. First, we define a
A extension of :
 {A :   A > }, if consistent.
, if not.
For each antecedent A of a > conditional derivable from , define a A
extension of  inductively relative to an ordering over antecedents of > statements derivable from the previous stage in the sequence with  ,0 = . Every
such sequence has a fix point. We now have the definition:
Definition 1.
 | iff for all orderings the fixpoint  of each 
extension sequence is such that   .
4.

A Gricean Account of Scalar Implicature

With this sketch of commonsense entailment, let us now turn to Grices picture
about implicatures. Grices view of implicatures is that they are calculated after
compositional semantics has finished its job via his famous general maxims of
conversation, quality, quantity and relevance. In principle the Gricean picture
tells the beginning of an attractive story for computing scalar implicatures (Horn
1972, Schulz and van Rooij 2004, Schulz 2007, Spector 2006). Lets take a look
at how the story is supposed to go for the following example (the implicature is
indicated by the ).
(6)

John or Susan came to the party  John and Susan didnt both come to
the party.

The derivation of the implicature in (6) uses Grices maxims of quality and
quantity via the following steps.
a. The speaker has said j or s, so she believes that j or s (quality).
b. It follows from the maxim of quantity that that the speaker does not
believe more than this relative to what she could have said (j, s, j
and s): in other words, she only believes j or s, i.e. she does not have
the belief that j, she does not have the belief that s and she does not
have the belief that j and s.

Implicatures in Discourse

19

c. Nor can she believe j or s, since given a. beliefs in these would


entail a belief in s and a belief in j respectively.
d. An axiom of Expertise allows us to infer that she believes not j
and s, (it applies only in cases where the doxastic implicatures in a.,
b. and c. are not contradicted).
Ive said that this is only the beginning of a story, since Grices maxims of quality
and quantity are not precise enough really to license any inferences. In addition,
these maxims should clearly be formalized within a non-monotonic logic. GL
provides a very simple yet precise reconstruction.
Since GL works with descriptions of logical forms, we have the means to
write down the fact that a certain relation holds between two formulas, namely
that is an alternative that could have been said in the given discourse context
instead of . Being able to express and to define this set of alternatives is crucial
to the enterprise of formalizing S-implicatures.
step a: follows defeasibly in GL if we assume the defeasible principle of
Sincerity (Asher and Lascarides 2003):
Say > B
To derive step b, we need a measure of informativeness. Lets suppose its
logical entailment, and that Alt( , ) holds only if is a strictly stronger
formula (entailing but not entailed by) than .
 , 
 Alt( , ) ((Say Say ) > B )
Step b will follow in GL, provided Alt( j s) = {j, s, j s}.
step c: doxastic reasoning is closed under  and so this follows.
Expertise about an issue gives us the stronger form of the implicature:
(Alt( , ) Say Say ) > B( )
To get the weaker implicature from (6) that the speaker doesnt believe that s,
j or that j s, we simply turn all the relevant > statements that instantiate the
schemas in steps and the appropriate instance of the schemas in steps a and
b above into statements. Without Expertise, we have only one fixed point,
in which the desired conclusions hold. The stronger implicatures necessitate a
slightly more complex axiom to maintain consistency of the sincerity implicature
with the expertise implicatures. Expertise, together with the other axioms and
the assumed facts, yields 2 different fixed points, one in which Bj, one in
which Bs. Since non-monotonic consequence is defined as those entailments
from all of the fixed points, the desired strong implicature follows:
(7)

B(j s) B(j s)

20

Nicholas Asher

The derivation of the implicatures in GL is nonmonotonic and thus the desired


implicatures follow defeasibly from the premises and the principles as I have
formalized them.4
However, the GL derivation of the Gricean implicature is valid only if the
alternativeness relation Alt is restricted as above. If the set of alternative contains
all of the strictly stronger logical entailments, then a great many distinct fixed
points arise and GL predicts no implicature can be drawn. This is a formalization
of the so called symmetry problem (Kroch 1972; Jasinskaya 2002; Block 2009).
(s j) (j s)  s j but its not the case that s j  (s j) (j s).
So using the principle in step b, we get B((s j) (j s)).
Since by hypothesis B(s j), propositional logic and the K axiom for B
yield: B((j s)).
Expertise cannot fire without yielding a doxastic inconsistency. Putting this
derivation together with the standard one yields: B(j s) B((j s))
By similar reasoning, we could conclude:
Bj B(j) in one fixed point
Bs B(s) in another fixed point
The conclusion is that there is no particular scalar implicature.
The success of the Gricean programme depends on a hidden assumption about
what other more informative things could have been said in the particular discourse context.
As Block notes, the Gricean programme itself says nothing about relevant
sets of alternatives. We would need an additional element of the pragmatics to
do this. Or it might be specified lexically and be part of the grammar (Fox and
Katzir 2011). For example, we might say that alternatives are defined lexically
for scalar items, for adjectives beautiful, stupendous, gorgeous,. . . , for nouns
genius, idiot, . . . , for connectives (, ) (here it depends as to whether >, ,
etc. are also considered connectives), and for quantifiers (no, some, many, most,
all). Such lexically specified alternatives seem reasonable but we need to know
how alternatives compose together. One might also try to fix the alternatives
for an utterance via a question under discussion approach, but this has many
difficulties of its own, not the least of which is determining what the question
under discussion is.

4. In particular Sauerlands (2004) restriction is not needed. This is desirable, since this
allows for a straightforward cancellation of the implicatures if the explicit semantic
content contains for example an explicit denial of one of the implicatures.

Implicatures in Discourse

21

Even though the problem of the specification of alternatives remains unresolved, we can still comment on the role of GL in the grammar. GL reasons over
logical forms and so has access to the structure of the predication and of lexical
choice. But more generally, reasoning over logical forms as GL does is preferable to approaches that reason directly over preferred models of information
content (Van Rooij and Schulz 2004, 2006), because it leaves the semantics of
the alternatives intact. Van Rooij and Schulz use a set of alternatives to induce
a partial order on worlds, Information update with will pick those worlds
that are minimal with respect to the ordering. Intuitively, these worlds are just
those that make true but not any more informative response to the question
under discussion.5
The problem with the model minimization technique is its side effects. Minimization tells us that only one person came to the party given this question, and
that seems unwarranted. The mechanism of minimization is too powerful. Here
are some other untoward consequences.
(8)

Assume a problem set with 10 problems on it.


How many problems did John do on the problem set?
John did some of the problem set  B (John did more than 1 problem
of the problem set).

(9)

a.
b.

Who did John kiss at the party? (Alternatively: Did John kiss all of
the girls at the party?)
John kissed some of the girls at the party  John kissed two girls
at the party.

The minimization technique and ordering given above predicts implicatures


that dont seem born out in practice. One option is to restrict the ordering of
minimal elements in the ordering to some relevant set of alternatives, but now
we have injected some syntactic or at least extra-semantic restrictions into the
procedure. GLs reasoning over logical forms and explicit sets of alternatives
brings the set of alternatives to the fore front. The semantic minimization in

5. More technically, we define the ordering as follows. Let be some sentence and P a
partition.
w, w p P (w, w |= ((w p w p) w w))
The minimization of (6) generates a natural ordering on worlds. The worlds that are
minimal w.r.t the ordering are those where just John comes to the party and no one
else or where just Susan and no one else comes to the party.

22

Nicholas Asher

effect is computationally very complex and nets us nothing in addition to the


simpler approach based on reasoning about logical forms.6
5.

Localist Accounts of Scalar Implicatures

The Gricean account of scalar implicatures is a rationalist one with special


cognitive assumptions about cooperativity that provide a derivation of scalar
implicatures. But weve seen that such an account cannot derive S-implicatures
without the help of some sort of extra assumptions about alternatives to what was
said. It seems reasonable that the set of alternatives is at least partially conventionally determined. But at what level are these alternatives computed? Griceans
claim that the alternatives are computed pragmatically after truth conditional
semantics has finished its job. Given a standard view of what truth conditions
are, this means that Griceans do not have access to the fine structure of a
sentences meaning; the particular way the truth conditions has been erased in
the semantic value. A contrasting, localist theory claims implicatures are, like
presuppositions, conventionally determined by the lexicon and computed during compositional semantic interpretation. In contrast to Griceans, localists have
access to the fine structure of meaning in computing implicatures, in particular
in computing the set of alternatives upon which the derivation of implicatures
depends. This makes them closer to the technical apparatus of the previous
section.
The localist computation of implicatures goes beyond a particular take on
the generation of alternatives, however. To quote Chierchia (2004):
The claim is that there are situations in which (standard) implicatures are by default present and situations in which they are by default absent, and such situations
are determined by structural factors.

That is, implicatures result from structural properties of the grammar, not from
any pragmatic inferences based on Gricean maxims. This leads to a recalibration
of the vision of pragmatics within the grammar, putting it much closer to the
core. But as we will see, the notion of structure that is required for generating
implicatures goes far beyond that of sentential syntax.
6. There is another technical problem with minimization. These updates need to be defined relative to very particular models that contain the relevant worlds. They must
not bring any information extraneous to the question at hand all the unmentioned
background facts stay as they were. In the model minimization framework, without
constraints on the sentences for which the models are provided, this constraint cannot
be satisfied in general, and the minimization problem is unsolvable.

Implicatures in Discourse

23

The principal motivation for the localist approach is the presence of embedded implicatures, which can present problems for Neo-Gricean approaches.
(10)

John did the reading or some of the homework.

This implicates John didnt do the reading and some of the homework. It also
implicates that he didnt do all of the homework. But it doesnt implicate that
(John did the reading or all of the homework). Since Griceans compute implicatures only on whole utterances or full sentences, its not clear how to get the
second implicature.
While Chierchias example has impressed some linguists as decisive (for
example Danny Fox), the difficulty for Griceans with (10) depends once again
entirely on the set of alternatives chosen. If one of the given alternatives to (10)
is in fact John did the reading or all of the homework, we have trouble using the
Gricean strategy, but if we rule out somehow this alternative and have instead
the set of alternatives consisting just of
(11)

John did the reading and some of the homework.

(12)

John did all of the homework.

we would end up with the right predictions. A Gricean could adopt instead a
localists computation of alternatives, while nevertheless maintaining a broadly
pragmatic approach to the derivation of implicatures.
Inspired by Chierchias localist meaning clauses, I define in the Appendix
sets of alternatives using the recursive structure of the logical form and lexically
stipulated alternatives. We can then use Common Sense Entailment to formalize
the broadly Gricean reasoning to derive implicatures of the sort that Chierchia
claims to hold. Moreover, it is an account that is compatible that Griceans can
accept, as weve seen that the computation of the relevant set of alternatives for
implicatures is not something that is forthcoming from Gricean principles alone
but is rather extraneous to it.
The main problem, however, is the Chierchia inspired calculation yields us
implicatures that dont fit the facts. Consider first the exhaustivity implicature
based on disjunction. The epistemic reading of the disjunction is often prominent.
(13)

John did the homework or the reading.

For many interpreters (13) just conveys that the speaker doesnt know which
of these two alternatives is correct, but in fact both could be. The disjointness

24

Nicholas Asher

implicature predicted by Chierchia and the account in the Appendix seems to


be triggered only in certain discourse contexts.
The computation of alternatives in the Appendix also undergenerates with
respect to observed implicatures. For instance, it predicts that one should not
get the standard implicatures (the exhaustivity implicature for or, for instance)
inside the scope of downward entailing operators. But this is wrong for the
antecedents of conditionals, which are downward entailing. Consider (16), and
its paraphrase (17).7
(16)

If you take cheese or dessert, you pay $ 20 ; but if you take both there
is a surcharge.

(17)

If you take only a cheese dish or only a dessert, the menu is 20 euros;
but if you take both there is a surcharge.

or
(18)

If John owns two cars, then the third one outside his house must be his
girlfriends.

(19)

If one person reads my book, Ill be happy.

(20)

If you want more food, you can order either the biryani or the stuffed
naan.

(21)

? If you want more food, you can only order, either the biryani or the
stuffed naan.

Further, the standard implicatures in the consequent of a conditional (which


is upward entailing) are much less strong than predicted given the system of
alternatives in the Appendix.

7. The Chierchia inspired computation of alternatives isnt convincing either on its own
for other downward entailing operators, even for the example with doubt or not
believe. Prosody here I think is essential to getting the implicature:
(14)

I dont believe that John has read MANY philosophy books.

(15)

I dont believe that JOHN has read many philosophy books. (implicature less
strong)

But this issue goes beyond the scope of this paper.

Implicatures in Discourse

(22)
(23)

25

If Mary comes to the party, John or Sam will be happy.


? If Mary comes to the party, only John or only Sam will be happy.

Finally the embedded implicatures which motivate the localist approach are
acknowledged to be less clear cut than originally supposed. Consider
(24)

Every student read some of the books.  Every student didnt read all
of the books. (No students read all of the books)

Judgements vary concerning the predicted implicature of (24), depending on


how the implicature is presented. Chemla (2009) and Geurts and Pouscoulous
(2007) have experimental results to show that such embedded implicatures are
less strong than the simple ones.
Finally, the fragility of embedded scalar implicatures contrasts with other
implicatures like the free-choice implicatures or discourse based implicatures
that freely embed in non-DE contexts:
(25)

You may take an apple or a pear.  a p

(26)

Every student may take an apple or a pear.  Every student may take
an apple and every student may take a pear.

(27)

John slipped on the snow and fell. (Implicature is that the falling is the
result of the slipping.)

(28)

Every student slipped on the snow and fell. (Implicature is that the
falling is the result of the slipping.)

6.

Structural Problems with Localist and Gricean Accounts

As Larry Horn made clear many years ago, scalar implicatures depend, at least
in part, on scales associated with lexical items. The challenge is to determine
how these scales create a manageable ordered set of alternatives for sentences
that contain them. But while there are, most likely, scales lexically associated
with determiners like some and all and modals like can and most, experimental
research is less clear that adjectives like full, bald also support alternatives in
arbitrary contexts. For open class words, the actual values and perhaps even the
presence or the activation of the scale for the purposes of calculating implicatures
is dependent on discourse context.

26
(29)

Nicholas Asher

a.
b.
c.
d.

A: Do you like him?


B: Why, yes, I do.
A: I mean, do you LIKE him like him or just like him?
B: Oh, Im not sure.

In the first question-answer pair with a yes/no question, the lexical scale associated with like isnt really operative or needed to understand the exchange to
an alternative question where the scale is explicitly invoked.
The dependence of implicatures on discourse contexts surfaces in other
places too. Consider, the defeasibility of implicatures and their relation to ordinary semantic entailments. What a localist grammar produces is a pair of
contents, the first element of which is the narrow semantic content of the
discourse, the content given by lexical and compositional semantics, and the
second element of which is the strengthened meaning containing both the narrow content and the implicature. Ordinary semantic entailments are understood
as product entailments of the pair, while implicatures are understood as entailments of the second element. According to localists, the defeasibility of the
implicature is done externally to the meaning computation. Those implicatures
that arent inconsistent with established facts in the common ground or the
narrow semantic content continue to be operative as the discourse content is
computed.
This view of implicatures gives us the wrong results. Consider the following
example of a sentence (30a) generating the embedded implicature in (30b).
(30)

a.
b.

John either did some of the reading or he did some of the homework. 
John didnt do all of the reading; John didnt do all of the homework; and he didnt do some of the homework and some of the
reading.

And now consider the following dialogue.


(31)

a.
b.

A: John either did some of the reading or he did some of the


homework.
B: John did all of the reading, but youre right, he didnt do all of
the homework.

What happens to the implicature (30b) in this discourse context? To my ears,


(31a), (31b) has the implicature that John did all of the reading and some of the
homework. In fact this is an implicature of (31b). But this implicature overrides
the implicature of (31a) that John didnt do some of the homework and some

Implicatures in Discourse

27

of the reading. Thus, implicatures interact in important ways with discourse


moves like the corrective move given by B. We need to integrate S-implicatures
into discourse content and structure and then compute the appropriate update of
that content after taking into account Bs corrective move. This implies that we
cannot calculate implicatures merely at a sentential level, as most Griceans and
localists have assumed. We need a much more finegrained view of the discourse
context to calculate implicatures properly.8
Another problem with localist and Gricean accounts of implicatures is that
implicatures are not always cancellable. Sometimes they are required for discourse coherence, in which case they are not cancellable.
(33)

a.
b.
c.

John has an even number of children. He has four. (Implicature is


that he has exactly 4.)
#John has an even number of children (1 ). He has three (children)
(2 ).
John has an even number of children (1 ). He has at least three
(children) (2 ). John has an even number of children. He has four
(children).

A Gricean or a localist like Chierchia should predict that (33a)(33c) are OK,
since the implicature to the stronger, exactly meaning of three should be
blocked. However, it is not, and (33b) is infelicitous. Once again, I believe
this stems from an interaction of discourse structure and implicatures: there is
a particular sort of elaborative move going on in the second clauses of (33),
which accounts for the freezing of the implicature. Only a framework like GL
even has a hope of handling such examples.
A final indication that something is amiss with current accounts of implicatures is their fragility. As Chemla (2009) notes, localist theories predict that
8. We can continue this pattern with more complex embedded examples.
(32)

a.
b.

A: Some of the students did some of the reading or some of the homework.
B: At least one student did all of the reading, but otherwise youre right.

It would seem that Bs correction still leaves many of the implicatures of his original
statement intact; hes still committed to the implicature that Some of the students
didnt do all of the homework and that some of the students didnt do all of the
reading and some of the reading. We need a more finegrained notion of implicature
revision in the face of corrections. Contrast also (32b) with (32c):
(32)

c.

At least one of those students did all of the reading.

28

Nicholas Asher

(34b) should not have the implicature below, making a stark contrast between
(34a) and (34b).
(34)

a.
b.

John didnt read all of the books.  John read some of the books.
No student read all of the books. ?  All of the students read some
of the books.

For localists, the predicted implicature of (34b) is No student read some book
or some students read some of the books, which is weaker than the implicature
tested by Chemla. However, (34b) is equivalent to:
(34)

c.

All the students didnt read all of the books.

and this intuitively (and on a localist theory) implicates that all the students read
some of the books. So equivalent meanings seem to yield distinct implicatures!
This seems to indicate strongly that implicatures depend not only on semantic
content but of something else in addition.
Interestingly, D-implicatures are not closed under arbitrary first order equivalences either. Consider the logical equivalence in (35a). If D-implicatures were
computed on deep semantic content and hence closed under first order equivalences, we would predict no difference between (35b) and (35c) since (35b),
where the relation of Explanation linking the two clauses is inferred, is perfectly
coherent in contrast to (35c), where no discourse relation is inferred:
(35)

a. Some one pushed him. CL Not everyone didnt push him.


b. John fell. Someone pushed him. (Explanation inferred)
c. #John fell. Not everyone didnt push him.

Few people think that D-implicatures are compositionally determined at the


syntax semantics interface or off pure semantics; so this failure of substitutivity,
which is equivalent to a failure of compositionality, is not much of a surprise. In
GL this failure comes about because D-implicatures are dependent on information about logical form and about the global discourse context, rather than just
semantic content. A simple hypothesis is that S-implicatures are also dependent
on logical form and the global discourse context, in particular the structure of
the discourse context, in order to explain their fragility. That is the claim I shall
defend in the next section.

Implicatures in Discourse

7.

29

Interactions between D-implicatures and S-implicatures

Lets recap. I have detailed a general mechanism for computing D implicatures


that also serves to capture scalar implicatures.9 I have also given arguments that
call into question the empirical adequacy both of Gricean and localist accounts
of implicature. Such accounts dont explain the fragility of S-implicatures nor
their occasional uncancelability. My suggestion is that Chierchia was right that
implicatures, in particular the set of alternatives from which they are generated,
are structurally determined; he was wrong, however, about restricting his attention to syntactic structure or the structures properties of logical forms of sentences. The relevant structure in question is discourse structure.10 This should
not come as a surprise; the behavior of other sorts of non-assertoric content,
like presupposed content, also has complex interactions with discourse structure (Asher and Lascarides 1998). As over 30 years of work on presupposition
has shown, it is unwise to try to compute presuppositions without examining
how the surrounding discourse context might affect these presuppositions. But
unlike presuppositions, a theory of S-implicatures needs not a theory of accommodation or binding, but of a theory of triggering. More specifically, we
must answer the question: when does a discourse context license or induce an
appropriate alternative set over which to compute (scalar) implicatures?
Discourse structure triggers additions to content, in particular structural relations like Contrast, Correction, Parallel, QAP, and various species of Elaboration. These relations together with prosody, which can signal a lexical choice,
induce structure preserving maps that can provide a set of alternatives and license an implicature. On the other hand, implicature inferences are sometimes
required to establish discourse relations. When the latter is the case, then I predict that the implicatures are not cancellable without affecting the coherence of
the discourse.
To build a case for my claim, I will look at examples that localists like
Chierchia, Fox and Spector (2008) have put forward for the robustness of localist
implicature computations. I will show that in a representative sampling of those
examples, it is the discourse structure that triggers the embedded implicature,

9. Fox (2007), Kratzer and Shimoyama (2002) and Alonso-Ovalle (2005) argue that free
choice implcatures should be treated with the mechanism for scalar implicatures. I
have a rather different take on free choice implicatures, but that would take us too
far afield here. See Asher and Bonevac (2005).
10. See Geurts (2009, 2010) for more support on this point.

30

Nicholas Asher

and in some cases the implicatures are required to maintain the discourse relation
established.11
Lets look at a class of examples in Chierchia, Fox and Spector (2008) that
involve the discourse relation of Correction.
(36)

a.
b.
c.

Joe didnt see Mary or Sue; he saw both. (only a clear exhaustive
interpretation of the disjunction).
It is not just that you can write a reply. You must.
I dont expect that some students will do well, I expect that all
students will.

(36a)(36c) all are only felicitous as corrections of assertions that are echoed
under the scope of the negation. The observation is that the echoic use of correction in (36) makes the embedded implicatures happen. For instance, because we
take the correction move in (36a) to correct the exhaustively interpreted assertion Joe saw Mary or Sue, we have to interpret the embedded clause exhaustively
as well. And then voil`a: we have an embedded implicature.
Asher (2002) (written a decade earlier) provides an analysis of discourse
relations in terms of a map from the source (the constituent to be linked to
the discourse structure) to a target (a discourse constituent that serves as an
attachment point). This map exploits prosodic cues and the logical structure of
the constituents, which can be displayed in a modified embedding or ME graph,
which extends the SDRS graphs I introduced in Section 3 with sentence internal
logical structure (see Asher (2002) for the details, which arent relevant here).
Such maps can also be made to serve an account of discourse triggered scalar
implicatures.12
My account of Correction involves the following constraint:
Correction( , ) only if K entails K and there is a map : such
that there is at least some element x of K ( (x)
x ) > K . The element x is
said to be the correcting element.
Lets look at how such a constraint works. In (36a), we assume an earlier constituent K of the form John saw Mary or Sue that is the target of the Correction
move. As is often the case with Corrections, the second clause in (36a) elaborates on the first. But the Correction move and its elaboration are coherent
only if we assume that an exhaustivity implicature is added to the content of the
assumed, antecedent constituent. That is, K is in this case John saw Mary or
11. For a discussion of more of their data, see Asher (forthcoming).
12. Cf. Schwarzschild (1998) for similar ideas about the interpretation of focus.

Implicatures in Discourse

31

Sue but not both, and the exhaustivity implicature is the target of the Correction.
Without the addition of the implicature, there would be nothing for (36a) to correct. The presumption of discourse coherence, and the interpretation of (36a)
as a correction move triggers, indeed requires, the presence of the embedded
implicature.
Notice that the implicature calculated is relative to the map . The set of
alternatives can be calculated as in the Appendix or simply using the map
itself. Furthermore, the inference of the S-implicature is triggered by the need
to establish the coherence of the discourse move, in this case Correction. Thus,
D- and S-implicatures are codependent; the need to calculate a D-implicature
triggers the calculation of the S-implicature, and it is the S-implicature that
supports the D-implicature. Relations of Parallel and Contrast work similarly
(Asher, forthcoming).
Recall the example I gave above of a non-cancellable implicature.
(33)

a.

John has an even number of children. He has four. (Implicature is


that he has exactly 4.)
b. #John has an even number of children (1 ). He has three (children)
(2 ).
c. John has an even number of children (1 ). He has at least three
(children) (2 ). John has an even number of children. He has
four (children).

These examples all exemplify the discourse relation of Elaboration. And in


fact, Elaboration makes the implicature non-cancellable. Clearly it is a better
Elaboration of having an even number of children to give the exact number,
than it is to simply say that he has some number greater than n. The presence
of the lexical choice between three and at least three together with the presence
of the Elaboration relation, and the parallel structure and prosodic prominence
on three which itself determines the structure preserving map buttresses this
conclusion. The parallel structure and the instantiation of the quantifier an even
number of children to a given number is now strongly a particular kind of
Elaboration, Instance(1 , 2 ), meaning that 2 furnishes an instance or witness
for a quantifier in 1 . In such an Elaboration, the instance is read as a specific
number (i.e. 3), which is inconsistent. The example sounds awkward, because
the surface cues for a particular kind of Elaboration, Instance, and the deep
semantics of the clauses come apart. When an explicit quantifier at least 3 is
used, Instance is not triggered; we have a general specification of the number of
Johns children, and everything remains consistent.

32

Nicholas Asher

The same observation of non-cancellability holds in fact for all the implicatures triggered by discourse structure. Consider once again (37) or (38); it would
be completely incoherent to cancel the implicature of the first clause given the
discourse environment provided by the second clause:
(37)

#Sam or Susan came (in fact both did), or both did.

(38)

#If most (in fact all) of the students do well, I am happy; if all of them
do well, I am even happier.

In general if the discourse structure requires the S-implicature for coherence, it


isnt cancellable except on pain of incoherence.
On the other hand, sometimes even S-implicatures conveyed by disjunction
dont arise when not needed to verify constraints. Txurruka and Asher (2007)
argue that disjunctions can play a special discourse role. They are defeasible
marks of the relation Alternation. Alternation introduces alternatives that carve
up the set of discourse possibilities relative to some topic, which can be introduced via a question or via a simple assertion. When the topic, however, already
contains the disjunction, the constraint that is a consequence of Alternation is
not met and so the relation of Alternation does not hold. Consider, for example:
(39)

a.
b.

Did John meet the Vice President or the President?


He must have met the Vice President or the President, since he got
the job.

(40)

a.
b.

Theres a lot of poop on the streets here.


Everyone in the village owns a horse or a donkey.

The relation in (40) is Explanation (41a), (41b). There is no alternation in either


(39b) or (40b) because the topic constraint on Alternation is not met. But on the
other hand, compare:
(41)

a.
b.

What kind of animal do people own in this village?


Everyone owns a horse or a donkey. (No one owns both.)

Given the topic established by (41a), the topic constraint on Alternation is met
and so we get the embedded implicature.
Question-answer pairs also trigger S-implicatures. A question induces a partition on the information state (Asher 2007; Groenendijk 2008). A complete answer picks out one cell in the partition; indirect answers (which stand in the IQAP
relation to the question they address) require reasoning or additional premises
to infer a complete answer. Sometimes the additional information comes from

Implicatures in Discourse

33

an implicature given by a structure preserving map from the response to the


question. IQAP can thus give rise to an overanswer in which we get more than
just a complete answer to the question; we get a more informative subset of the
element of the partition picked out. This is once again an instance of the general
strategy of being as informative as possible . . .
A particularly interesting case involves prosodically marked overanswers to
polar questions like the following.
(42)

a.
b.

Did John eat all of the cookies?


John ate SOME of the cookies.

(42b) is an overanswer to (42a). By itself (42b) doesnt provide enough information to compute an answer to the question. But the prosodic marking gives
rise to a structure preserving map, an ME graph in the language of Asher (1993)
or Asher (2002), from the response to the question. In this case the prosodically
marked some is mapped to all, and provides the relevant alternative set. The lack
of a full answer also triggers the S-implicature, and including the implicature
that John didnt eat all of the cookies together with (42b) provides a complete
answer to (42a). But (42b) also gives more information than just a simple no
would have.
A response that on its own fails to give a complete answer to a question can
also trigger embedded implicatures and ones that wouldnt be calculated from
standard lexical alternatives for some. Consider, for instance, the S-implicature
of (43b), which is that John believes that not many of the students passed the
exam, or the implicature of (44b) which is that everyone didnt read most of the
books. These implicatures follow given the mappings on ME graphs that map
some to many in (43b) and some to most in (44b).
(43)

a.
b.

Does John believe that many of the students passed the exam?
John believes that SOME of the students passed the exam.

(44)

a.
b.

Did everyone read most of the books?


Everyone read SOME of the books.

Notice that once again the addition of the implicatures in these responses gives
us complete answer to the questions they are paired with.
My approach makes predictions about when the implicatures should not
arise, even if the implicature is consistent with the information in the discourse
context.
(45)

Did some students go to the party?

(46)

Yes, some students went to the party.

34

Nicholas Asher

Here there is no prosodically marked element and no pairing of two distinct


elements on a scale. I predict the implicature not to arise.
Lets now look briefly at Wh-questions.
(47)

a.
b.

Who read all of the books on the reading list?


JOHN read SOME of the books.

This example follows the treatment of polar questions, except there are two
prosodically prominent elements in the response. It is the second that under the
ME graph mapping generates the implicature.
The interaction between questions and answers for generating implicatures
has an effect on how evidence for implicatures has often been gathered. As
argued by Geurts (2009), introspection is a biased method for implicatures. If
youre given
(48)

Some of the boys have a cough.

and you then ask yourself: does this implicate


(49)

Not all of the boys have a cough.

A question is suggested:
(50)

Do all of the boys have a cough?

You now get an overanswer or an indirect answer: IQAP (50), (48). Given the
suggested question, we predict the implicature to hold in this inference task.
The moral of this, however, is that these implicatures need not, and indeed are
predicted not to hold outside of this discourse context.
So how do implicatures arise in the absence of an inference task or in out
of the blue contexts? Sometimes simple prosody suggests something about the
discourse context. Roberts (1996) and Kadmon (2009) argue that prosodic information can tell us something about the question under discussion.
Robertss constraint: the focus semantic value of = the question under
discussion addressed by .
So
(51)

Who did Larry kiss?

(52)

Larry kissed NINA.

(53)

How much of the homework did John do?

(54)

He did SOME of the homework.

Implicatures in Discourse

35

Robertss constraint doesnt get us the desired implicatures. But we might imagine that something similar holds for certain lexical stressed elements.
A focussed scalar item x in an assertion can give rise to a QUD for which
there exists a structure preserving map : such that for some scale
S, (x) S x.
Using our constraint,
(55)

a.
b.
c.

John read SOME of the books.  the QUD:


Did John read all/many/most of the books?
 The implicature that John did not read all/many/most of the
books.

I predict that the implicatures will be much vaguer without specifying a particular discourse context.
Finally lets go back to implicatures within downward entailing contexts.
Recall that these were a problem for the more finegrained account of the set of
alternatives motivated by the localists.
(56)

None of the students answered all of the questions.

does not seem to have, at least in an out of the blue context, the implicature:
(57)

All of the students answered some of the questions

even though (56) is equivalent to


(58)

All the students didnt answer all of the questions.

(58) appears to have the predicted implicature, at least in the right discourse
context. For instance, if (58) is part of a Correction, the implicature seems to be
fine:
(59)

a.
b.

A: All of the students answer all of the questions.


B: No, all of the students DIDNT answer ALL of the questions.

There is a natural prosodic prominence to the second occurrence all in (58)


signalling a lexical choice that is the source of the disagreement, which gives rise
to a structure preserving map and the relevant implicature. (56) in this context
still less clearly has the implicature because the Correction is also supported by
the choice of none.

36

Nicholas Asher

8.

Conclusions

My approach to S-implicatures, which links them to D-implicatures, is quite


diferent from the view that implicature reasoning is triggered by default (the
position of Chierchia); my approach predicts that certain out of the blue inference tasks for embedded implicatures should be quite difficult, whereas the
default approach can only do so on general grounds of sentence complexity. But
then the default approach cant explain why in only certain discourse contexts
embedded implicatures are inferred, why when they are inferred, they are often
not cancellable. My approach is also distinct from extant Gricean approaches,
as these also ignore discourse structure and its effects in their description of the
reasoning mechanisms governing implicatures. They too will have troubles with
the issues I have raised in this paper for a Chierchia-like position.
Ive argued that there is a unified theory of implicatures. Implicatures, understood as defeasible implications, can arise from several sources: semantics,
discourse structure or prosody together with the logical form of what is said.
Ive looked at the first two sources here, and Ive argued that S-implicatures are
often parasitic on D-implicatures. The localist theory would benefit from having
the left hand side of the > statements giving the implicatures restricted to those
cases in which the implicatures are required for discourse coherence (though the
discourse context may be quite minimal). The theory makes several predictions:
implicatures can be cancelled even if theyre consistent with information in
the discourse context; implicatures can be uncancellable even in the face of
inconsistency (when mandated by discourse structure); embedded implicatures
(both negative and positive) require a more elaborate discourse setting to be triggered and so should be harder to get. This line of thinking also suggests a line
of empirical research: given the right discourse context, embedded implicatures
should follow as easily as the unembedded ones.
My argument here also has implications for the role of a theory of discourse structure in the grammar. If S-implicatures are taken to be part of the
grammatical system, then discourse structure and D-implicatures will be too.
Alternatively, one can throw all implicatures out of the core grammar, but then
grammar becomes much more impoverished and a lot less interesting. While Ive
used S-implicatures as a case study, the effects of D-implicatures are everywhere
from lexical composition and phenomena of coercion, in semantic composition
and the resolution of scope ambiguities up to purely semantic structures like the
temporal structure of texts. Whats also interesting is that D implicatures and
their effects, if Im right, play out at a particular level of the grammar, at the
level of logical form; they complete and augment literal content, but they are
best computed prior to the model-theoretic interpretation of the logical form.

Implicatures in Discourse

37

In fact, if computational tractability is an issue (and I believe that it should


be), then implicatures have to be computed at the level of logical form, since
the underlying non-monotonic reasoning involved at the level of full sentence
contents is completely intractable.

Appendix on a Particular Calculation of Implicatures


Consider a simplified language with and and or as truth functional connectives,
a set of generalized quantifiers or up and downward monotonic operators and a
finite set of predicates and individual constants. I assume that only predicates,
operators and quantifiers are associated in the lexicon with scales, not individual
constants. A scale for a predicate yields a scale for an atomic formula in an
obvious way (I like John  {I love John, I adore John, I dislike John, I hate
John}), and these make up alt( ) for an atomic formula. S-alt( ) denotes
the formula that is the strongest logical alternative to amongst the formulas
in this set, in the spirit of Chierchia. For a formula with a mon quantifier,
S-Alt( ) = S-Alt(Q) ; if is of the form of a binary operator O( , ), Salt( ) = S alt(O)( , ). The axioms below are expressed in a subjectivist
semantics and can be seen as a more detailed working out of the maxim of
quantity given in the paper. To have a subjective epistemic logic means that
these axioms are meant to describe a space of worlds that are a subset of the
doxastic possibilities of an agent, in our case the speaker.
I assume that the material on the left hand side of the > statements below is
a description in GL of something that has been said by the speaker.
> S-alt( ), for atomic .
for mon operators Q:
Q( ) > (S-alt(Q)( ) Q((S-alt( ))))
> (S-alt( ) S-alt( ) S-alt( ))
for mon operators DE:
DE( ) > (DE(S-altDE( ) ( )))
A special case of a DE operator:
> (S-alt ( ))
> (S-alt( ) S-alt( ) S-alt( ))(S-alt( ) typically is ).
Fact 1.
If the basic scales are linear and if is positive, then S-alt( ) exists
and the implicatures of are well defined.

38

Nicholas Asher

Proof is by induction on the complexity of . Base case for atomic formulas


is given by our assumptions. So assume that , have linear scales associated
with them, and let Q be monotonic operator. The scale associated with Q is by
assumption linear, and so there is an S-alt(Q). The induction hypothesis gives
the rest. Similarly for and .
Lets return now to the Chierchia examples that supposedly present troubles
for Griceans.
(10)

John did the reading or some of the homework.

(60)

John did all of the reading or all of the homework.

This account straightforwardly yields the implicature that John didnt do both
the reading and some of the homework, and he didnt do all of the homework
for (10). Given the scales for quantifiers, no implicatures for the disjuncts are
predicted; the only implicature for (60) is that John didnt do both all of the
reading and all of the homework, as desired.
Multiple implicatures are in principle also not a problem.
(61)

a.
b.

Someone at the party will be smoking or someone will be drinking 


Its not the case that someone at the party will be smoking and
that someone will be drinking and not everyone will be smoking
and not everyone will be drinking.

Lets consider (61a). Since the disjunction has wide scope, we deal with that
first using the recursion, we get
s d > (s d) S-alt(s) S-alt(d)
Since this is consistent with the context, we infer using DMP:
(s d) S-alt(s) S-alt(d)
S-alt(s) = Everyone at the party will be smoking and
S-alt(d) = Everyone at the party will be smoking
We predict the desired implicatures. Similarly an example like (62) can be shown
to have the implicature in (63):
(62)

Every boy did some of the homework.

(63)

Every boy didnt do all of the homework.

Implicatures in Discourse

39

Negative implicatures work similarly:


(64)

I doubt that John has read many philosophy books.  I believe that he
has read some philosophy books.

Suppose we simply calculate the bits inside the doubt context via the DE rule
as Chierchia suggests:
(65)

I doubt John read some philosophy books. |= I doubt John read many
philosophy books.

So we get as an implicature
(66)

I doubt (John read some philosophy books)

(67)

I believe (John read some philosophy books)

References
Alonso-Ovalle, Luis
2005
Distributing the disjuncts over the modal space. In: Leah Bateman and
Cherlon Ussery (eds.), North east linguistics society, vol. 35. Amherst,
MA.
Asher, Nicholas
1993
Reference to Abstract Objects in Discourse. Dordrecht: Kluwer Academic Press.
Asher, Nicholas
1995
Commonsense Entailment: A Logic for Some Conditionals. In:
Gabriella Crocco, Luis Farinas del Cerro and Andreas Herzig (eds.),
Conditionals: From philosophy to computer science, 103147. New
York: Oxford University Press.
Asher, Nicholas
2002
From Discourse Micro-structure to Macro-structure and back again:
The Interpretation of Focus. In: Hans Kamp and Barbara Partee (eds.),
Current Research in the Semantics/Pragmatics Interface, Vol. 11:
Context Dependence in the Analysis of Linguistic Meaning. Amsterdam: Elsevier Science Ltd.
Asher, Nicholas
2007
Dynamic Discourse Semantics for Embedded Speech Acts. In: Savas
Tsohatzidis (ed.), John Searles Philosophy of Language, 211244.
Cambridge: Cambridge University Press.

40

Nicholas Asher

Asher, Nicholas
2008
Troubles on the Right Frontier. In: Peter Kuhnlein and Anton Benz
(eds.), Proceedings of Constraints in Discourse 2005. Amsterdam:
John Benjamins.
Asher, Nicholas
2012
Implicatures and Discourse Structure. Forthcoming in Lingua.
Asher, Nicholas and Daniel Bonevac
2005
Free Choice Permission is Strong Permission. Synthese 145(3): 303
323.
Asher, Nicholas and Alex Lascarides
1998
The Semantics and Pragmatics of Presupposition. Journal of Semantics 15: 239299.
Asher, Nicholas and Alex Lascarides
2003
Logics of Conversation. Cambridge: Cambridge University Press.
Asher, Nicholas and Michael Morreau
1991
Commonsense Entailment: A Modal Theory of Nonmonotonic Reasoning. On: Proceedings of the 12th International Joint Conference
on Artificial Intelligence. San Francisco: Morgan Kaufmann.
Asher, Nicholas and Sylvain Pogodalla
2010
SDRT and Continuation Semantics. LENLS 2010, Tokyo, Japan.
Block, Eliza
2009
Gricean Implicature. Paper presented at the Michigan Pragmatics
Workshop.
Chemla, Emmanuel
2009
Universal Implicatures and Free Choice Effects: Experimental Data.
Semantics and Pragmatics 2(2): 133.
Chierchia, Gennaro
2004
Scalar Implicatures, Polarity Phenomena and the Syntax/Pragmatics
Interface. In: Adriana Belletti (ed.), Structures and Beyond. Oxford:
Oxford University Press.
Chierchia, Gennaro, Danny Fox, and Benjamin Spector
2008
The Grammatical View of Scalar Implicatures and the Relationship
between Semantics and Pragmatics, draft.
Fox, Danny
2007
Free Choice Disjunction and the Theory of Scalar Implicatures. In: Uli
Sauerland and Penka Stateva (eds.), Presupposition and Implicature
in Compositional Semantics, 71120. NewYork: Palgrave Macmillan.
Fox, Danny and Roni Katzir
2011
On the characterization of alternatives. Natural Language Semantics
19(1): 87107.
Geurts, Bart
2009
Scalar Implicature and Local Pragmatics. Mind and Language 24(1):
5179.

Implicatures in Discourse

41

Geurts, Bart
2010
Quantity Implicatures. Cambridge: Cambridge University Press.
Geurts, Bart and Nausicaa Pouscoulous
2009
Embedded Implicatures?!? Semantics and Pragmatics 2(4): 134.
Groenendijk, Jeroen
2008
Inquisitive Semantics. Proceedings of SALT XVIII The University of
Massachusetts at Amherst: Amherst, MA.
Horn, Laurence
1972
The semantics of logical operators in English. PhD thesis, UCLA.
Horn, Laurence
2006
The Border Wars: a neo-Gricean perspective. In: Ken Turner and Klaus
von Heusinger Where Semantics meets Pragmatics. Dordrecht: Elsevier.
Jasinskaja, Katja
2002
Relevance and Other Constraints on the Quantification Domain of
only. Proceedings of the Workshop on Information Structure in Context, IMS Stuttgart.
Kadmou, Nirit
2009
Contrastive topics and the focal structure of quations. Semantics
archive. URL: semanticsarchive.net/Archive/jMzOTczN/Kadmonms-2009-TOPIC-FOCUS.pdf.
Kratzer, Angelika and Junko Shimoyama
2002
Indeterminate pronouns:The view from Japanese. In:Yukio Otsu (ed.),
The Proceedings of the Third Tokyo Conference on Psycholinguistics,
125. Tokyo: Hituzi,
Kroch, Anthony
1972
Lexical and inferred meanings for some time adverbs. Quarterly
Progress Report of the Research Laboratory of Electronics 104: 260
267.
Roberts, Craige
1996
Information Structure in Discourse: Towards an Integrated Formal
Theory of Pragmatics. In: Jae Hak Yoon and Andreas Kathol (eds.),
OSU Working Papers in Linguistics 49: Papers in Semantics, 91136.
The Ohio State University Department of Linguistics.
Sauerland, Uli
2004
On Embedded Implicatures. Journal of Cognitive Science 5: 107137.
Schulz, Katrin and Robert van Rooij
2006
Pragmatic Meaning and Non-Monotonic Reasoning: The Case of Exhaustive Interpretation. Linguistics and Philosophy 29(2): 205250.
Schwarzschild, Roger
1998
GIVENness, AvoidF and other Constraints on the Placement of Accent. Natural Language Semantics 7(2): 141177.

42

Nicholas Asher

Spector, Benjamin
2006
Aspects de la pragmatique des operateurs logiques. PhD dissertation,
Universite Paris 7.
van Rooij, Robert and Katrin Schulz
2004
Exhaustive Interpretation of Complex Sentences. Journal of Logic,
Language and Information 13(4): 491519.
Schulz, Katrin
2007
Minimal Models in Semantics and Pragmatics: Free Choice, Exhaustivity, and Conditionals. Ph.D. Dissertation, University of Amsterdam.
Txurruka, Isabel and Nicholas Asher
2007
A discourse-based approach to Natural Language Disjunction (revisited). In: Michel Aurnague, Kepa Korta and Jesus Mari Larrazabal
(eds.) Language, Representation and Reasoning. University of the
Basque country Press.
Vieu Laure, Myriam Bras, Nicholas Asher and Michel Aurnague
2005
Locating adverbials in discourse. Journal of French Language Studies
15(2): 173193.

Permission and Choice*


Paul Portner

1.

Introduction

Functions of imperatives
Imperatives can be used to perform a variety of intuitively distinct speech acts,
for example ordering, advising, requesting, and permitting (Schmerling 1982;
Davies 1986; Han 1998; Schwager 2005a; Portner 2007; among many others):
(1)

a.
b.
c.
d.

Sit down right now! (Order)


Talk to your advisor about this. (Advice)
Help me! (Request)
Have a piece of fruit, if you like. (Permission)

As Davies (1986) argues, it is not productive to focus on the details of the


categories which happen to have names in English. Rather, we need to understand
the nature of the variation: its source, limits, and effects. These distinctions are
relevant to grammar. We see this, for example, with the fact that particles may
limit the range of meanings available, as shown by the following data from
Badiotto, due to Poletto and Zanuttini (2003), and German, from Grosz (2009a):
(2)

a.

You need to eat well, so you can grow up to be big and strong.
(Advice)
M`ange-l ma!
eat-it
ma

* I have presented versions of this paper at the conference 10 Years After at the
University of Frankfurt, the University of Chicago, the Ohio State University, the
University of Pennsylvania, MIT, and the third conference Semantics and Philosophy
in Europe in Paris. I am thankful for the input Ive had from many people, at these
venues and elsewhere, including Chris Barker, Nate Charlow, Kai von Fintel, Irene
Heim, Chris Kennedy, Tony Kroch, Jason Merchant, Peter Pagin, Miok Pak, Craige
Roberts, Magdalena Schwager, and Raffaella Zanuttini.

44

(3)

Paul Portner

b.

We cant let the food go to waste. You have to finish it, even if you
dont want to. (Order)
M`ange-l mo!
eat-it
mo

a.

Iss *blo/ *JA/ ruhig


den Spinat! Das stort
mich
eat BLO JA RUHIG the spinach that disturbs me
nicht. (Permission)
not
Eat blo/JA/ruhig the spinach! That doesnt disturb me.
Iss blo/ JA/ *ruhig den Spinat! Sonst wirst du
eat BLO JA RUHIG the spinach or.else will.be you
bestraft. (Order)
punished
Eat blo/JA/ruhig the spinach! Or else youll be punished.

b.

One basic issue is whether any of these differences are semantic in nature. It
is tempting to analyze the permission imperative in (1d) as different from the
others, as it can be paraphrased with a possibility modal:
(4)

a.
b.
c.
d.

You must sit down right now! (Order)


You should talk to your advisor about this! (Advice)
Wont you please help me? (Request)
You may have a piece of fruit! (Permission)

I will refer to (1d) as a permission imperative and the others as requirement


imperatives.
Choice phenomena
An intuitive way of describing the function of a permission sentence is to say
that it offers the addressee a choice s/he didnt have before. When a permission
sentence offers more than one choice at the same time, we call it a free choice
sentence. As is well-known, free choice sentences can be made with disjunction
or an indefinite (including special indefinites like any).
The free choice inference refers to the fact that a requirement or permission sentence (made with a modal or imperative, or by others means) implies
that each disjunct, or each entity described by an indefinite, corresponds to a
permitted option.

Permission and Choice

(5)

45

Free choice inference


a. You may take an apple or an orange. You may take an apple./You
may take an orange.
b. Take a piece of fruit! You may take this apple./You may take that
pear.

Related is Rosss paradox, the lack of licit inference from a permission sentence to disjunction:
(6)

Rosss paradox
a. You may take an apple.  You may take an apple or an orange.
b. Take an apple!  Take an apple or an orange!

The paradoxical aspect can be seen from the comparison with declaratives,
where p entails (p q). I lump all this together under the label choice phenomena. A key testing ground for analyses of permission sentences will be how
well they fit into our understanding of choice phenomena.
There has been a great deal of research on free choice in modal sentences.
We may classify it into several major approaches:
1. Traditional assumptions
Choice phenomena come about on the basis of fairly traditional semantic
values and Gricean reasoning (Aloni and van Rooij 2004; Schulz 2005).
2. Alternatives
Choice phenomena come about because the semantics introduces each alternative separately, one way (Zimmermann 2000; Geurts 2005) or another
(Kratzer and Shimoyama 2002; Simons 2005; Menendez-Benito 2005; Alonso-Ovalle 2006; Fox 2007; Aloni 2007).
3. The Andersonian reduction
Choice phenomena come about because permission is defined, in the tradition
of Anderson (1956), as something like If p, then things are ok (Asher and
Bonevac 2005, Barker 2010).
4. Dynamic semantics
Choice phenomena come about because of the dynamic semantics associated with particular elements, for example deontic may (van Rooij 2008) or
epistemic might (Ciardelli et al. 2009).
We also have a divide between those who think that free choice with disjunction
is a conversational implicature (Kratzer and Shimoyama; Menendez-Benito;
Alonso-Ovalle; Aloni and van Rooij; Schulz), a matter of semantics (Geurts; Simons; Aloni; Barker; Ciardelli et al.), both (Fox), or something else (van Rooij).

46

Paul Portner

It will not be possible to consider all of these analyses in detail. My hope is to


shed new light on the problem of free choice by focusing on choice phenomena
as they occur with imperatives. While imperatives have been discussed in connection with choice phenomena, it is typically assumed that they are implicitly
modal sentences, and so dont have anything special to teach us. I will argue, in
contrast, that a treatment of imperatives which doesnt assume that they contain
a modal operator serves as the basis for an insightful treatment.
Main claims of this paper
It is the goal of this paper to argue for the following hypotheses:
There is no semantic difference between requirement and permission imperatives.
Differences in function among imperatives mostly depend on the grounds
upon which the imperative is issued.
True permission imperatives are derived from the logical relation between
the imperative and the context to which it is added.
Choice phenomena with imperatives follow as a special case of the analysis
of permission.
2.

Background on the Semantics of Imperatives

There are two main approaches to the semantics of imperatives: the modal
theory and the dynamic theory.
1. The modal theory proposes that imperatives contain a modal operator, so that
an imperative is very close in meaning to certain sentences containing must
or should. (Han 1999, to appear; Schwager 2005a; Aloni 2007; Grosz 2009a).
Within the overall modal approach, various authors may treat the so-called
modal element proposed as more or less similar to regular modals, and at
some point we might better call this the modaloid theory, a less attractive
term to be sure, but perhaps appropriately so.1
1. If you want to treat the imperative modal as a purely dynamic modal, similarly to
the treatment of epistemic modals in Groenendijk et al. (1996) and the treatment of
expressions of expectation in Veltman (1996), Id consider that an implementation
of the dynamic theory. See Rooij (2008) and Portner (2009) for discussion of how
this might be done. The adherents of what I call the modal theory assume that the
modal in question falls under a standard (static) analysis of modals, such as Kratzers
(1981, 1991, 2012).

Permission and Choice

47

2. The dynamic theory claims that the meaning of imperatives consists (entirely, or virtually so) in the way they affect the discourse context (Portner
2004, 2007; Mastop 2005; in a sense Lewis 1979). The dynamic theory of
imperatives is really a part of the dynamic theory of clause types (or of sentence mood, if you prefer that terminology). Imperatives are one of the three
major clause types, alongside declaratives and interrogatives (Sadock and
Zwicky 1985), and we should aim for an explanation for why these three
are universal (Portner 2004). Assertion is commonly analyzed in terms of
Stalnakers concept of common ground (Stalnaker 1974, 1978), and asking a
question has been analyzed in terms of a second discourse component, what
Ginzburg calls the Question Under Discussion Stack (Ginzburg 1995a, b;
Roberts 1996). Parallel to these, Portner (2004) proposes that imperatives are
interpreted as contributing to the addressees To-Do List.
The central theoretical claim of this paper is that the dynamic approach can
explain, in a simple and natural way, both the variation in function among imperatives and choice phenomena.
Outline of the dynamic analysis of imperatives
Portner (2004, 2007) argues that the meaning of imperatives can be given within
a dynamic framework as follows:
(7)

Pragmatic function of imperatives


a. The To-Do List function T assigns to each participant in the
conversation a set of properties T ( ).
b. The canonical discourse function of an imperative clause imp is to
add [[ imp ]] to T (addressee). Where C is a context of the form
CG, Q, T : C + imp =
CG, Q, T [addressee/(T (addressee) {[[ imp ]] })]

The To-do List is similar to ideas in Lewis (1979), Han (1998), Roberts (2004),
and Mastop (2005). Whats different is the Ordering pragmatics for imperatives. In particular, the To-Do List functions to impose an ordering on the worlds
compatible with the Common Ground, and this ordering determines what actions
an agent is committed to taking (Portner 2004):

48
(8)

Paul Portner

Ordering pragmatics for imperatives


a. Ordering of worlds:
For any w1 , w2 CG and any participant i, w1 <i w2 iff for some
P T (i), P(w1 )(i) = 1 and P(w2 )(i) = 0, and for all P  T (i), if
P  (w2 )(i) = 1, then P  (w1 )(i) = 1.
b. Agents commitment:
For any participant i, the participants in the conversation mutually
agree to deem is actions rationaland cooperative to the extent that
those actions in any world w1 CG tend to make it more likely

that there is no w2 CG such that w2 < i w1 .

By contributing to the addressees To-do List, an imperative affects the ordering


of worlds, which in turn guides how the addressees actions will be judged and
understood. The Common Ground and To-do List are parallel at the discourse
level to the modal base and ordering source in the sentential semantics of modals.
3.

Variation in the Function of Imperatives

Given the above dynamic analysis of imperative meaning, we can turn to the
variation in function of imperatives. Note that (7) applies to all imperatives, so
the variation cannot be because some imperatives contribute to the To-do List,
while others do not. Rather, in what follows, Ill argue that variation in function
should be explained as follows:
1. Subtypes of requirement imperatives are characterized by the grounds which
justify issuing the imperative.
2. Some imperatives which are intuitively described as giving permission are
actually requirement imperatives, also characterized by the grounds which
justify their being issued.
3. True permission imperatives are characterized by the fact that they contradict
something else in the To-do List; permission readings may be encoded via a
presuppositional element.
3.1. The Grounds for Issuing an Imperative
The kinds of illocuationary acts performed by requirement imperatives differ in
characteristic ways in terms of the grounds which the speaker has for issuing
them.
(9)

a.

Speakers authority Order


Sit down right now!

Permission and Choice

b.
c.

d.
e.

49

Helps the addressee achieve her desires Suggestion


Go for a nice walk!
Helps the addressee achieve her desires in a situation where the
speaker has authority to prevent the act Invitation
Have a piece of fruit!
Helps the addressee achiever her goals Advice
Speak to your adviser about this!
Helps the speaker achieve her desires or goals, without authority
Request
Let me taste that!

The above descriptions of each speech act are just suggestive, and firm definitions
may be impossible. But they show the point: types of requirement imperative
differ in the grounds on which they are offered, notably in the presence or absence
of speakers authority, the identity of the individual whose priorities are being
advanced, and the nature of the priorities being advanced.
(10)

With/without invoking authority, helps participant achieve priority


of type . . .

Portner (2007) shows that these distinctions are relevant to semantic/pragmatic


theory. First, (2)(3) show that language can mark such differences explicitly.
Second, the various subtypes of imperatives parallel the subtypes of priority
modal meanings:
(11)

a.
b.

Sit down right now! (order)


Noah must sit down right now, or hell be punished. (deontic)

(12)

a.
b.

Go for a nice walk! (suggestion)


You should go for a nice walk. (bouletic)

(13)

a.
b.

Speak to your adviser about this! (advice)


Noah should speak to his adviser about this. (teleological)

And third, one imperative or modal will constrain the meanings which are possible for subsequent imperatives or modals. Portners (2007, p. 356) psycho
boss provides one example:
(14)

Be there at least two hours early. Then have a bite to eat.

Example (14) is odd because both sentences must be interpreted as expressing


the same pragmatic function. Since the first sentence expresses an order, the
second is interpreted that way as well, even though on its own it could easily be

50

Paul Portner

a suggestion. Thus, the boss comes across as excessively authoritarian, ordering


an employee when to eat.
The differences among interpretations of priority modals are standardly analyzed in terms of the choice of ordering source (Kratzer 1981, 2012). In parallel
to this, Portner (2007) analyzes the differences among the imperatives in terms
of subsets to the addressees To-do List. For example, an imperative which is issued on the grounds of the speakers authority is added to a characteristic subset
of the To-do List (orders). This partitioning of the To-do List allows one to
define the links between the interpretation of imperatives and modals in context.
See Portner (2007) for details.
The grounds for permission
Semanticists often analyze permission imperatives as being based on the addressees desire (e.g., Wilson and Sperber 1988; Han 1998; Portner 2004;
Schwager 2005b). Consider an invitation imperative, like (9c), issued because it
will help the addressee achieve her desires, and without invocation of speakers
authority. In such a case, the addressee is free to refuse the imperative, and thus
it will only lead to a requirement if the addressee wants it to be a requirement.
This is a lot like permission, even more so if the context is one in which it is
assumed that the speaker has the authority to prevent the act in question.
Wilson and Sperber (1988) have an interesting and, to my mind, convincing
version of this view of permission. They describe permission as the situation
where a speaker imposes a requirement in order to overcome the reluctance of
the addressee to undertake an action in his or her own interests. (More recently,
Schwager 2005b has developed a similar idea in terms of the modal analysis
of imperatives.) Suppose that you aim to be a polite guest, and as a result have
Do not take the fruit on your To-do List. I could then say (9c), in order to put
a contradictory requirement onto your To-do List. Let us assume for a moment
that the To-do List must remain consistent. In that case, you could either refuse
the imperative (No thanks) or accept it while removing the politeness-based
abstinence.2 For practical purposes, you were given permission to have the fruit,
because it was up to you whether to make it a requirement or retain not having
fruit as a requirement. But technically the invitation is a requirement imperative,
since at each point you had one or the other requirement. The pretense of al-

2. Note that this treatment is in the same spirit as Kamps (1973) analysis of permission
as removing a prohibition. However, it does more than this: it removes the prohibition by adding a contradictory requirement. Well return to Kamps insights about
permission in Section 3.2. below.

Permission and Choice

51

ways following a requirement serves a social function, describable in Politeness


Theory as maintaining the hosts negative face (Brown and Levinson 1987).
3.2. Logical Relations between Imperative and To-do List
Theres a sense of permission which cannot be explained in the above terms.
This is the case where the permission-granting sentence is accepted, but its
content is not required (in the broad sense of following from the To-do List).
The following examples show non-imperative permission sentences which do
not impose a requirement:
(15)

You are allowed to have an apple, but you are not required to.

(16)

A: You may have an apple.


B: Thank you /OK. (Doesnt take apple.)

In contrast to explicit performatives and may, its not so easy to use imperatives
this way. In (17)(18), if you dont want the apple, you usually have to refuse.
(17)
(18)

#Have an apple, but youre not required to.


A: Have an apple!
B: #Thank you /OK. (Doesnt take apple.)
Thank you, not right now. (Doesnt take apple.) [Refusal]

The refusal here points to an analysis of the kind sketched for (9c) above.
A conditional can work better in creating permission:
(19)

Have an apple, if you like, but youre not required to.

(20)

A: Have an apple, if you like.


B: Thank you /OK. (Doesnt take apple.)

If you like suggests the addressee-desire analysis (Schwager 2005b).


One way an imperative can give permission in this strict sense is via a choice
sentence:
(21)

A: Have an apple or a pear.


B: Thank you /OK. (Doesnt take apple, but does take pear.)

(22)

A:
B:

Stay in or go out, either way.


OK.

Here, having an apple is not required. But even then, the overall disjunction
gives a requirement:

52
(23)
(24)

Paul Portner

#Have an apple or a pear, but youre not required to do either.


A: Have an apple or a pear.
B: #Thank you /OK. (Doesnt take apple or pear.)

The right context can also produce a permission reading in the strict sense. If B
ends up bringing beer and no wine, she has not failed to do as requested:
(25)

A: Please bring some beer to tomorrows party.


B: But I have some good wine at home.
A: Then sure, bring wine!

The bottom line is that permission imperatives are severely restricted, and only
come about in specific constructions and contexts.
Permission via an inconsistent To-do List
Let us work with the following Lewis-style example (Lewis, 1979):
(26)

Monday carry rocks! Tuesday carry rocks! And Wednesday carry rocks!
[. . . Tuesday comes along.] Take tomorrow off!

After the Master utters the first three imperatives, the Slaves To-do List becomes
{K(x,m), C(x, mo), C(x, tu), C(x, we)}.3 This To-do List implies that the Slave
must not kill the Master (not explicitly stated, but surely assumed) and must
carry rocks each of the three days.
Then on Tuesday, the Master gives permission to take Wednesday off. Thus,
the To-do List becomes {K(x, m), C(x, mo), C(x, tu), C(x, we), C(x, we)}.
This To-do List is inconsistent. Given (8a), it implies that the Slave does as
well as he can: he either carries rocks Wednesday, or he doesnt. Either way,
he can make four of the five propositions in the To-do List true. See Figure 1.
(More accurately, after having worked Monday and Tuesday, and not killing the
Master, either working on Wednesday or taking the day off makes it the case that
there is no better-ranked world.) Intuitively, the inconsistency of the To-do List
represents the fact that carrying rocks on Wednesday was formerly required, but
3. On the analysis of Portner (2004, 2007), the variable x should be abstracted
over, to produce a property, and should be restricted to the addressee: [ x : x =
addressee(c) . C(x, mo)]. The restriction to the addressee is responsible for the fact
that the property is added to the addressees To-do List, as opposed to someone elses.
I dont insert the x or restriction, both for simplicity, and to indicate the fact that
the points under discussion here dont depend on the decision to treat imperatives as
properties, as opposed to sets of worlds.

Permission and Choice


K(x,m),C(x,mo),C(x,tu),C(x,we)

K(x,m),C(x,mo),C(x,tu),C(x,we)

K(x,m),C(x,mo),C(x,tu),C(x,we)

K(x,m),C(x,mo),C(x,tu),C(x,we)

53

Figure 1. The choices on Wednesday

is no longer required. Crucially, killing the Master is still not an option, because
K(x, m) is consistent with both options; the ordering pragmatics thus solves
the problem of permission raised by Lewis (1979).4,5
This analysis of permission has some similarity to that of Kamp (1973).
Kamp proposes that permission is the removal of a standing prohibition. In the
present treatment, the previous requirement (in this case, Carry rocks Wednesday!, though it could be stated as a prohibition, Dont take Wednesday off!)
is not removed, but it no longer has the effect of producing a requirement, once
it is contradicted by another entry on the To-do List. At that point, it simply
corresponds to one permitted option. (Please note, though, that I do not mean
to imply that requirements are never removed from a To-do List. As well see
below, such retraction does occur, but it is not necessary to produce permission.)
Whats on the To-do List?
Example (27) poses a minor problem.
(27)

Carry rocks every day! [. . . Weeks pass . . . ] Tomorrow, take the day off!

It seems that after the first imperative is uttered, the To-do List is
{K(x, m), d[C(x, d)]}. Adding C(x, we) to this set leads to a situation where never working again is permitted. The solution to this problem is to expand the To-do List from {K(x, m), d[C(x, d)]} to one more
like like {K(x, m), d[C(x, d)], C(x, mo), C(x, tu), C(x, we), C(x, th), C(x, fr)}.
4. Mastops (2005) dissertation is suggestive of an analysis of choice phenomena in
imperatives similar to the one offered here, but because his views on permission
imperatives are unclear, it is difficult to say whether he endorses the same perspective.
5. It has been suggested to me (Peter Pagin, p.c.) that the analysis cannot account for the
cruel dictator who purposefully imposes a contradictory set of laws, so that everyone
is always susceptible to punishment. Im not certain that such a use of imperatives
would be considered pragmatically competent, but if it is, this kind of dictator would
be defining the pragmatics of imperatives differently from (7). The alternative would
be that an agent takes actions in world w which tend to make it that case that, for every
world compatible with the common ground, w is at least as highly ranked. Doing so
is impossible when the To-do List is contradictory.

54

Paul Portner

There might be interesting ways to implement this, for example by giving the
quantifier scope over a force operator. But I prefer an uninteresting way: the
Slave knows that he is to work on Monday, on Tuesday, on Wednesday, etc. So
the To-do List is expanded by inference. Once C(x, we) is added to this set,
we represent permission to take Wednesday off (or not), but still require work
on the other days. Note that this solution is essentially the one Kratzer (1977)
applies to the pros and cons of striding and flying; van Rooij (2000) draws on
relevant entailments for a similar purpose.
Definitions of permission and requirement
In Figure 1, we can say that not carrying rocks Wednesday is permitted, because
some best-ranked worlds are ones in which C(x, we) is satisfied.
(28)

S is a permission sentence in context c if an utterance of it in c results,


as a matter of its conventional meaning, in a context c in which some
best-ranked worlds are in [[ S ]] :



w[w CG w [[ S ]] w  [w CG w  <i w]]

We can say not killing the Master is required, because all best-ranked worlds are
ones in which K(x, m) is satisfied.
(29)

S is a requirement sentence in context c if an utterance of it in c


results, as a matter of its conventional meaning, in a context c in which
all best-ranked worlds are in [[ S ]] :



w[(w CG w [w  CG w <i w]) w [[ S ]] ]]

These descriptions make the limit assumption, for simplicity. A more precise and
general version of (28) would say that S is a permission sentence iff [[ S ]] is a
good possibility, with respect to CG and T (addressee), in the terms of Kratzer
(1981, 1991). Likewise, (29) would say that S is a requirement sentence iff [[ S ]]
is a weak necessity, with respect to CG and T (addressee). This way of thinking
about things lets us consider the status of propositions which meet the criteria for
other grades of modality in Kratzers system. For example, (simple) possibility
would describe a situation where S-worlds and S-worlds alternate endlessly
in the ordering, as in a case where theres too much uncertainty concerning the
effects of ones actions.According to (8), an agent doesnt have the right to pursue
a possibility which is not a good possibility. (If you think that this is wrong, and
that an agent can rationally and cooperatively pursue a simple possibility of this
kind, the prediction can be changed by modifying (8b)). These points exemplify

Permission and Choice

55

nicely how the theory of modality can be brought to bear within the dynamic
approach to imperatives, even though no modal is syntactically present.
Permissibility is a gradable concept, as illustrated in (30):
(30)

It is more permissible to take an apple than to take an apple and a pear.

The ordering pragmatics based on the To-do List allows us to use Kratzers
definition of comparative possibility to analyze graded permissibility. If the Todo List contains Dont take an apple and Dont take a pear, (30) will be
true. Ultimately, however, well need to integrate our analysis of the gradability
of modal expressions into the general theory of gradability, perhaps based on
degrees, so there is much more work to be done.
Permission vs. retraction
The analysis as developed so far captures those cases in which an addition to
the To-do List in conflict with an existing directive results in permission. But as
noted above, it takes a specific context for real imperatives to do this. Heres a
case where it doesnt work:
(31)

Bring beer to the party tomorrow! Actually, bring wine!

If the second imperative merely gives permission, we would expect that its ok
for the addressee to bring beer and no wine. However, we would say that the
addressee did not comply with what was requested if he brings beer. We can
describe this as the retraction of an existing requirement, and the imposition of
a new one.
Of course, retraction also happens with declaratives:
(32)

Hes angry. Actually, hes in pain.

There is a difference between imperatives and declaratives, however. Given that


the common ground should remain consistent, we expect that one of a pair of
contradictory declaratives must be retracted, at least partially. However, as weve
seen, a contradictory To-do List represents a perfectly acceptable situation in
which multiple options are permitted, and theres no immediate motivation for
retraction.
As mentioned above, some languages provide resources which allow the explicit marking of whether an imperative expresses permission or a requirement.
Grosz (2009a) has the most semantically detailed discussion that Im aware of.
He points out that the particles JA and blo disambiguate imperatives towards

56

Paul Portner

a command-type reading, and ruhig disambiguates towards a permission-type


reading:6
(33)

a.

b.

Iss *blo / *JA / ruhig


den Spinat! Das stort
eat BLO JA RUHIG the spinach that disturbs
mich nicht.
me not
Eat blo/JA/ruhig the spinach! That doesnt disturb me.
Iss blo / JA / *ruhig
den Spinat! Sonst wirst du
eat BLO JA RUHIG the spinach or.else will.be you
bestraft.
punished
Eat blo/JA/ruhig the spinach! Or else youll be punished.

We may say that JA and blo may only be used in requirement sentences, as
defined above, and ruhig only in permission sentences. Grosz (2009b) admits
that this analysis captures the relevant facts with imperatives and modals.7
These German data show that it is possible to explicitly mark an imperative
as introducing a requirement or a permission. I would like to suggest that English affords itself of this possibility as well. Specifically, the absence of any
marking indicates that the imperative should be a requirement sentence, while
the presence of any of a number of expressions, for example initial or, please,
go ahead, by all means, and if you like, as well as intonation or contextual evidence, may indicate that it is a permission sentence. We can make this proposal
more explicit by proposing phonologically null particles analogous to JA and
ruhig:
(34)

a.
b.

A root sentence S containing REQ presupposes that the context in


which S is used is one in which it is a requirement sentence.
A root sentence S containing PERM presupposes that the context
in which S is used is one in which it is a permission sentence.

English imperative sentences normally contain REQ, unless PERM is indicated


by linguistic material, intonation, or context.8 Thus, the second sentences in (31)
6. These forms are not always discourse particles, and we are to ignore the other uses.
Furthermore, unstressed ja is assumed to be a different discourse particle.
7. Grosz (2009b) suggests that there is a problem for the dynamic analysis based on the
occurrence of these particles in purpose clauses, but the existing literature (see, for
example, Bach 1982; Jones 1991; Johnston 1999; Whelpton 2002) does not support
his assumption that a covert modal is present.
8. Another implementation of these ideas would have only one particle, with its absence
implicating the other interpretation.

Permission and Choice

57

contains REQ, presupposing that the context is one in which the the addressee is
being placed under a requirement to bring wine.Assuming that the addressee will
bring just one kind of drink, the only way for this presupposition to be satisfied
is for bring beer to be retracted. In this way, the presupposition provides a
motivation for retraction similar to that observed with the declarative in (32).
In contrast to (31), an example like As second imperative in (25), repeated
here, would contain PERM:
(35)

A: Please bring some beer to tomorrows party.


B: But I have some good wine at home.
A: Then sure, bring wine!

The presupposition of PERM is straightforwardly satisfied in this context (assuming that B will not bring both wine and beer). Thus, after this sequence, B
would be in compliance with As request whether he brings beer or wine.
A third situation worth considering occurs when the PERM is present, but
the context is one in which the imperative would impose a requirement. The
following is modeled on Groszs (2009a) example (18):
(36)

A:
B:

Do I turn left here?


Sure, turn.

If something like Continue straight ahead is on As To-do List, Bs imperative


will will yield permission, not a requirement the correct result. But if nothing
on the To-do List is incompatible with turning left, it will yield a requirement,
and while this result is compatible with the presupposition of PERM, it is not
empirically correct. The permission reading of (36) can be explained as a scalar
implicature, however, considering that REQ is stronger than PERM.9 That is,
the presence of PERM implicates that REQ would not be appropriate. In order
for REQ not to be appropriate, Continue straight ahead (or another direction
incompatible with turning left) must be accommodated.
To summarize the results of this section, the dynamic theory allows a uniform
account of the meaning of imperatives. The distinction between permission
and requirement imperatives is essentially pragmatic, having to do with the
relationship between the imperative and the other entries in the To-do List. As a
result, whenever an imperative is constrained to function in one way or the other
(as a permission sentence or as a requirement sentence), this must be encoded
in the pragmatics, for example as a presupposition.
9. In any context in which the presupposition of REQ+S is met, that of PERM+S is also
met. We could also modify (34b) so as to include the presupposition that S is not a
requirement sentence.

58

Paul Portner

4.

Choice Phenomena

4.1. Disjunction and Choice


Given the above analysis of permission, we can also give an explanation of free
choice disjunction in imperatives and of Rosss paradox. In (37), (a) implies that
its ok to take an apple, and its ok to take a pear; this is the free choice inference.
(Its probably not ok to take both.) In (38), (b) does not follow from (a), in some
sense; this is Rosss paradox.
(37)

a.
b.

Take an apple or a pear!


You may take an apple. / You may take a pear.

(38)

a.
b.

Take an apple! 
Take an apple or a pear!

I follow the alternatives camp in analyzing disjunction as creating alternatives


within a Hamblin semantics. Crucially, these alternatives are typically, and perhaps always, exclusive (Menendez-Benito 2005; Alonso-Ovalle 2006, 2008;
Aloni 2007).10 Thus (37) denotes the set:
(39)

{( x[y[A(y) T (x, y)] y[P(y) T (x, y)]], ( x[y[P(y)


T (x, y)] y[A(y) T (x, y)]]}

Or for short: {T (x, a), T (x, p)}

10. The literature on alternative semantics differs on whether exclusivity is a semantic


or pragmatic phenomenon. For example, Menendez-Benito takes the position that
exclusivity is introduced in the semantics, Alonso-Ovalle assumes that it is an implicature, and Aloni does not take a stand on the issue.
You may wonder how to analyze cases where disjunction seems to be inclusive. I
know of two possibilities:
(i)

(ii)

Alternatives are not automatically made exclusive. Exclusivity is merely an


implicature. (This is the traditional analysis, transferred into alternative semantics; see Alonso-Ovalle 2006).
Alternatives are always exclusive, but they may appear to be non-exclusive
due to the workings of situation semantics. In particular, each alternative is
the set of minimal (exemplifying) situations where only one disjunct is true,
but this doesnt rule out that these are part of larger situations in which both
disjuncts are true. (This is Kratzers idea; see Menendez-Benito 2005 for discussion.)

Employing (ii) would require transferring the analysis of imperatives into situation
semantics, as needs to be done anyway, if you believe in situation semantics.

Permission and Choice

59

The rule specifying the discourse function of imperatives should be modified to


allow for the denotation to be a set of properties:
(40)

Pragmatic function of imperatives (ver. 2)


The canonical discourse function of an imperative clause imp is to add
every member of [[ imp ]] to T (addressee). Where C is a context of
the form CG, Q, T :
C + imp =
CG, Q, T [addressee/(T (addressee) [[ imp ]] )]

Because the alternatives are exclusive, an utterance of (37) results in an inconsistent To-do List which gives the (minimal) ordering of worlds illustrated in
Figure 2. In this setting, the addressee will behave correctly by taking either an
apple or a pear.
T(x,a),T(x,p)

T(x,a),T(x,p)

Figure 2. Both choices ok

(Of course if K(x, host) is already on the To-do List, none of this will permit
killing the host, but I leave this out of the figure for simplicity.)
In many contexts, it will be ok to take neither an apple nor a pear. A natural
situation of this kind occurs when the imperative is designed to overcome the
politness-based reluctance of addressee to impose on the speaker (Wilson and
Sperber 1988). That is, we assume an initial To-do List {T (x, a) T (x, p)}.
In this context, (37) leads to the To-do List {T (x, a) T (x, p), T (x, a)
T (x, p), T (x, a) T (x, p)} and to the order in Figure 3.11
T(x,a),T(x,p)

T(x,a),T(x,p)

T(x,a),T(x,p)

Figure 3. All three choices ok

Negation
Negative imperatives do not involve exclusive alternatives:
(41)

Dont take an apple or a pear!

11. In German, this type of reading seems to be marked by ruhig plus disjunction, as
opposed to JA in the case illustrated by Figure 2. Accounting for these cases will
require extending the definitions of permission and requirement to sentences which
express multiple alternatives, and then getting the scope relations right. Thanks to
Elena Herburger for discussing this contrast.

60

Paul Portner

This case can be handled in either of two ways. If exclusivity is a scalar implicature, then we would expect it not to be generated in this environment (following
Alonso-Ovalle 2006). In the present framework, the denotation will then be:
(42)

{y[A(y) T (x, y)], y[P(y) T (x, y)]}

Adding both alternatives to the To-do List results in best-ranked worlds being
ones in which the addressee takes neither. Moreover, on this approach worlds
in which the addressee takes just an apple are automatically ranked as better
than ones in which he takes both an apple and a pear, a point which proves an
advantage in the case of the addressee who cant help eating an apple at the
beginning of the party.
If exclusivity is introduced in the semantics, negation must prevent it by
collapsing alternatives before exclusivity is applied:12
(43)

[[ not ]] (A) = {{w : p[p A w p]}}

In that case, (41) denotes the set (containing the set) of worlds in which the
addressee takes neither an apple nor a pear, also resulting in the right prohibition.
Rosss paradox
Intuitively, the solution to Rosss paradox comes from the observation that a
master who says (38a) would not necessarily endorse (38b), since the latter would
permit an action not permitted by the former. But a master who says (38b) would
have no problem endorsing (38a). We want to say that an imperative warrants
another imperative iff adding the latter to a To-do List which already contains
the former never changes the ordering of worlds. A more general statement, in
which regular entailment falls out as a special case, is the following:
(44)

For any sentences , :


warrants =def for every context c (in which c + is defined),
c + = (c + ) + .

When we say that (38a) does not entail (38b), what we mean is that the former
doesnt warrant the latter. It doesnt warrant it because the disjunction gives the
12. Aloni (2007, fn.12) identifies the need to collapse alternatives in this context. Note that
Alonis approach involves a modaloid imperative operator, and predicts that disjunctive imperatives are ambiguous between choice-offering and alternative-presenting
readings. As she notes, the latter are marginal (at best), a fact which does not seem
amenable to her pragmatic explanation (a pragmatic preference for stronger interpretations, p. 88).

Permission and Choice

61

addressee an additional choice. In contrast, (38b) does warrant (38a). Warrant


encompasses Kamps (1973) notion of p-entailment, a concept also important
for van Rooij (2008). This definition of warrant is equivalent to one of Veltmans
(1996) definitions of validity in update semantics (specifically, his valid2 ), as
pointed out by van Rooij. Veltman, however, does not have a treatment of imperatives. And while van Rooij extends p-entailment to commands, he doesnt
provide a general definition. I use the term warrant, because the p stands for
permission (we want something more general) and because entailment is
not felicitious for sentences to which we dont intuitively attribute truth or falsity.
4.2. Indefinites and Choice
Free choice imperatives with indefinites work the same way as those with disjunction.
(45)

a.
b.

Pick a card!
{P(x, c) : c card}

Figure 4 gives the ordering, with choices taken to be exclusive. (This is some
kind of trick where the addressee can see the cardss faces.)
P(x,A)

P(x,2)

P(x,3)

P(x,4)

P(x,5)

P(x,6) . . .

Figure 4. All 52 choices ok

Lack of time precludes me from getting into the complexities of indefinites


in detail, as the literature is too extensive for me to do justice to. See Kratzer and
Shimoyama (2002) and Menendez-Benito (2005) for discussion of indefinites
in alternative semantics.
I would like to point out that Kadmon and Landmans (1993) analysis of free
choice any, as extended to modal permission and command sentences by van
Rooij (2008), will explain the licensing of any in imperatives. According to this
analysis, any is a domain widener, and is licensed when widening strengthens the
claim made; van Rooijs addition to this picture is that warrant (i.e., p-entailment)
is a relevant kind of strengthening. Note the following fact:
(46)

If D D , [[ imp ]] = {P(x, c) : c D}, and [[ imp ]] = {P(x, c) : c


D }: imp warrants imp .

Hence, any is predicted to be licensed in imperatives.

62

Paul Portner

5.

Conclusion and Next Steps

The range of interpretations displayed by imperatives can be explained in terms


of the dynamic theory without postulating any semantic ambiguity. Most subtypes are characterized by the grounds which justify issuing the imperative,
and such distinctions are formalized as subsets of the To-do List. Permission
imperatives (in the strictest sense) arise when the imperative adds a property
which is inconsistent with the rest of the To-do List, and it is possible to force a
permission interpretation via presupposition. Choice phenomena fall out as the
special case when a single imperative adds multiple inconsistent properties to
the To-do List.
There are two significant projects which should be tackled next:
1. Conditional imperatives. As we saw in (20), a conditional imperative can
easily produce a permission sentence. In order to bring this data into the
analysis, we need an account of conditional imperatives within the dynamic
theory. We will need to work with an analysis of conditionals which allows
for an interaction with dynamic aspects of meaning, such as that of Isaacs
and Rawlins (2008).
2. Modals. Modal sentences also show choice phenomena. Previous work on
choice phenomena with imperatives has assumed that they should be reduced
to choice phenomena with modal sentences, thus supporting the modal theory
of imperatives. My suggestion would be to do the opposite, explaining choice
phenomena observed in modal sentences in terms of dynamic (or performative) aspects of their meaning which parallel the dynamic meaning of
imperatives.13 Such an approach predicts that choice interpretations will not
be equally present with all modal sentences (in contrast to imperatives, which
always show choice phenomena), but rather will depend the presence of a
performative component of meaning. There is data suggesting that this pre13. A potential problem is brought up by Kamp (1973). He argues that (i) is performative
but does not give rise to free choice (his (13), p. 67; thanks to Chris Barker for
pointing out Kamps example):
(i)

You may pillage city X or city Y . But first take counsel with my secretary.

However, Im not convinced that this sentence is performative in the required sense.
On its own, (i) does not permit any action before consulting with the secretary,
the vassal had better not pillage either city. (Of course, in a different sense, it is
performative, in that it is guaranteed to be true on the basis of the fact that the
speaker is the King.) Only once the secretary is consulted can the vassal add the legal
kind of pillaging to his To-do List.

Permission and Choice

63

diction is correct. While many modal sentences show choice readings, none
of the following do:
(47)

a.

b.

c.

At the top of the mountain, you can find snow or ice to make
drinking water.
 At the top of the mountain, you can find snow to make drinking
water, and you can find ice to make drinking water.
A: Can any of the students speak Chinese or Japanese?
B: Chou Wei-han can. 14
 Chou Wei-han can speak Chinese and Chou Wei-han can speak
Japanese.
All of our students should take logic or stats.
 All of our students may take logic and all of our students may
take stats. (The syntax and semantics students should take logic,
while the phonology and socio students should take stats. The
former should not take stats, and the latter should not take logic.)

As far as I can tell, these are not examples in which choice interpretations
are cancelled or dispreferred; they simply do not have them. Previous work
has either assumed that all modal sentences will show choice readings, or has
focused on particular modal elements (e.g., van Rooij 2008; Ciardelli et al.
2009), not aiming to cover other cases. We do not have much understanding at
all of when choice readings occur and when they do not. Regardless of whether
the suggestion to relate choice phenomena to performativity is correct, in order
to develop a general theory which covers both the cases which show choice
phenomena and those which do not, more empirical work will be needed.
References
Aloni, Maria
2007

Free choice, modals, and imperatives. Natural Language Semantics


15: 6594.
Aloni, Maria and Robert van Rooij
2004
Free choice items and alternatives. In: G. Bouma, I. Kramer, and
J. Zwarts (ed.), Proceedings of KNAW Academy Colloquium: Cognitive foundations of interpretation, 525. Amsterdam: Edita KNAW.

14. Compare the non-elliptical reply, B: Chou Wei-han can speak Chinese or Japanese,
which does license the choice inference.

64

Paul Portner

Alonso-Ovalle, Luis
2006
Disjunction in alternative semantics. Doctoral dissertation, University
of Massachusetts, Amherst.
Alonso-Ovalle, Luis
2008
Innocent exclusion in an alternative semantics. Natural Language Semantics 16: 115128.
Anderson, Alan Ross
1956
The formal analysis of normative concepts. Technical Report 2, U.S.
Office of Naval Research.
Asher, Nicholas and Daniel Bonevac
2005
Free choice permission is strong permission. Synthese 145: 303323.
URL http://www.jstor.org/stable/20118599.
Bach, Emmon
1982
Purpose clauses and control. In: Pauline Jacobson and Geoffrey K.
Pullum (eds.), The nature of syntactic representation, Synthese Language Library 15, 3558. Dordrecht, Boston, and London: D. Reidel
Publishing Co.
Barker, Chris
2010
Free choice permission as resource-sensitive reasoning. Semantics and
Pragmatics 3(10): 138.
Brown, Penelope and Stephen C. Levinson
1987
Politeness: Some universals in language usage. Cambridge: Cambridge University Press.
Ciardelli, Ivano, Jeroen Groenendijk and Floris Roelofsen
2009
Attention! Might in inquisitive semantics. The proceedings of SALT
19. CLC Publications.
Davies, Eirlys
1986
The English imperative. London: Croom Helm.
Fox, Danny
2007
Free choice disjunction and the theory of scalar implicatures. In: Uli
Sauerland and Penka Stateva (eds.), Presupposition and implicature
in compositional semantics, 71120. Palgrave-Macmillan. Ms., MIT.
Geurts, Bart
2005
Entertaining alternatives: Disjunctions as modals. Natural Language
Semantics 13: 383410.
Ginzburg, Jonathan
1995a
Resolving questions, part I. Linguistics and Philosophy 18: 459527.
Ginzburg, Jonathan
1995b
Resolving questions, part II. Linguistics and Philosophy 18: 567609.
Groenendijk, Jeroen, Martin Stockhof and Frank Veltman
1996
Coreference and modality. In: Shalom Lappin (ed.), The handbook of
contemporary semantic theory, 179213. Oxford: Blackwell Publishers. URL citeseer.ist.psu.edu/article/groenendijk96coreference.html.

Permission and Choice


Grosz, Patrick
2009a

Grosz, Patrick
2009b

65

German particles, modality, and the semantics of imperatives. In:


S. Lima, K. Mullin, and B. Smith (eds.), The proceedings of NELS
39, GLSA.
Modal particles redux. Ruhig, blo, and JA: A case study in modal
matching. Unpublished MIT Generals paper.

Han, Chung-Hye
1998
The structure and interpretation of imperatives: mood and force in
Universal Grammar. Doctoral dissertation, University of Pennsylvania.
Han, Chung-Hye
1999
Deontic modality, lexical aspect and the semantics of imperatives.
Linguistics in morning calm 4. Seoul: Hanshin Publications.
Han, Chung-Hye
to appear
Imperatives. In: Claudia Maienborn, Klaus von Heusinger and Paul
Portner (eds.), Semantics: An international handbook of natural language meaning. Mouton de Gruyter.
Isaacs, James and Kyle Rawlins
2008
Conditional questions. Journal of Semantics 25: 269 319.
Johnston, Michael J. R.
1999
A syntax and semantics for purposive adjuncts in HPSG. In: Robert
Levine and Georgia Green (eds.), Studies in contemporary phrase
structure grammar, 80118. Cambridge: Cambridge University Press.
Jones, Charles
1991
Purpose clauses: Syntax, thematics, and semantics of English purpose
constructions. Kluwer.
Kadmon, Nirit and Fred Landman
1993
Any. Linguistics and Philosophy 16: 353422.
Kamp, H.
1973
Free choice permission. Proceedings of the Aristotelian Society, N.S.
74: 5774.
Kratzer, Angelika
1977
What must and can must and can mean. Linguistics and Philosophy 1: 33755.
Kratzer, Angelika
1981
The notional category of modality. In: Hans-Jurgen Eikmeyer and
Hannes Rieser (eds.), Words, worlds, and contexts, 3874. Berlin: de
Gruyter.
Kratzer, Angelika
1991
Modality. In: Arnim von Stechow and Dieter Wunderlich (eds.), Semantik/semantics: An international handbook of contemporary research, 639650. Berlin: de Gruyter.

66

Paul Portner

Kratzer, Angelika
2012
Modals and conditions. New and revised perspectives. Oxford: Oxford
University Press.
Kratzer, Angelika and Junko Shimoyama
2002
Indeterminate phrases: The view from Japanese. In: Y. Otsu (ed.), The
proceedings of the third tokyo conference on psycholinguistics, 125.
Tokyo: Hituzi Syobo.
Lewis, David K.
1979
A problem about permission. In: E. Saarinen, R. Hilpinen, I. Niiniluoto, and M. Provence Hintikka (eds.), Essays in honour of Jaakko
Hintikka, 163175. Dordrecht: Reidel.
Mastop, Rosja
2005
What can you do: imperative mood in semantic theory. Doctoral dissertation, University of Amsterdam.
Menendez-Benito, Paula
2005
The grammar of choice. Doctoral dissertation, University of Massachusetts, Amherst.
Poletto, Cecilia and Raffaella Zanuttini
2003
Making imperatives: Evidence from central Rhaetoromance. In:
Christina Tortora (ed.),The syntax of Italian dialects, 175206. Oxford
University Press.
Portner, Paul
2004
The semantics of imperatives within a theory of clause types. In:
Kazuha Watanabe and Robert B. Young (eds.), Proceedings of semantics and linguistic theory 14, 235252. Cornell University Linguistics
Department: CLC Publications.
URL http://semanticsarchive.net/Archive/mJlZGQ4N/.
Portner, Paul
2007
Imperatives and modals. Natural Language Semantics 15: 351383.
Portner, Paul
2009
Modality. Oxford: Oxford University Press.
Roberts, Craige
1996
Information structure in discourse: Towards an integrated formal theory of pragmatics. In: Jae-Hak Toon and Andreas Kathol (eds.), Papers
in semantics, OSU Working Papers in Linguistics, vol. 49, 91136.
Columbus: Department of Linguistics, the Ohio State University.
Roberts, Craige
2004
Context in dynamic interpretation. In: Laurence Horn and Gregory
Ward (eds.), Handbook of pragmatics, 197220. Oxford and Malden,
MA: Blackwell.
van Rooij, Robert
2000
Permission to change. Journal of Semantics 17: 119145.

Permission and Choice

67

van Rooij, Robert


2008
Towards a uniform analysis of any. Natural Language Semantics 16:
297315.
Sadock, Jerrold M. and Arnold Zwicky
1985
Speech act distinctions in syntax. In: Timothy Shopen (ed.), Language
typology and syntactic description, 155196. Cambridge: Cambridge
University Press.
Schmerling, Susan
1982
How imperatives are special, and how they arent. In: R. Schneider,
K. Tuite and R. Chameltzy (eds.), Papers from the Parasession on
Nondeclaratives. Chicago Linguistics Society, The Univ. of Chicago.
Schulz, Katrin
2005
A pragmatic solution for the paradox of free choice permission. Synthese 147: 343377. URL http://www.jstor.org/stable/20118661.
Schwager, Magdalena
2005a
Interpreting imperatives. Doctoral dissertation, University of Frankfurt.
Schwager, Magdalena
2005b
Permitting permissions. In: Judit Gervain (ed.), Proceedings of the
tenth ESSLLI student session.
Simons, Mandy
2005
Dividing things up: The semantics of Or and the modal/Or interaction.
Natural Language Semantics 13: 271316.
Stalnaker, Robert
1974
Pragmatic presupposition. In: M. Munitz and P. Unger (eds.), Semantics and philosophy, 197213. New York: New York University Press.
Stalnaker, Robert
1978
Assertion. In: P. Cole (ed.), Syntax and semantics 9: Pragmatics, 315
332. New York: Academic Press.
Veltman, Frank
1996
Defaults in update semantics. Journal of Philosophical Logic 25: 221
261.
Whelpton, Matthew
2002
Locality and control with infinitives of result. Natural Language Semantics 10: 167210.
Wilson, Deirdre and Dan Sperber
1988
Mood and the analysis of non-declarative sentences. In: J. Dancy,
J. Moravcsik and C. Taylor (eds.), Human agency: Language, duty
and value, 77101. Stanford CA: Stanford University Press.
Zimmermann, Thomas Ede
2000
Free choice disjunction and epistemic possibility. Natural Language
Semantics 8: 255290.

Promises and Threats with Conditionals


and Disjunctions
Robert van Rooij and Michael Franke

1.

No disjunctive promise

In propositional logic, formulas A P and A P are equivalent, and our intuitions about their natural language counterparts also, for the most part, support
this equivalence. However, as it is often the case where two expressions are
logically equivalent, the pragmatics of conversation poses restrictions on the acceptability of one that do not seem to apply to the other. In particular, in the case
of conditionals and disjunctions an interesting pragmatic difference surfaces in
the context of inducements, where the speaker is trying to influence the behavior of the hearer by conditional promises and threats: whereas the conditional
statements in (1a) and (1b) can be a threat and a promise, used respectively to
induce the hearer to hand over her wallet, the alleged disjunctive equivalents in
(1c) and (1d) are both preferably read as threats.1
(1)

a.
b.
c.
d.

If you dont give me your wallet, I will punish you severely.


A P
threat
If you give me your wallet, I will reward you splendidly.
AR
promise
You will give me your wallet or I will punish you severely.
AP
threat
? You will not give me your wallet or I will reward you splendidly.
A R
threat

This is astonishing, in particular because if we expect (1b) and (1d) to be logically


equivalent, we could also expect (1d) to be a promise, just as (1b) is. But, if
felicitous at all, (1d) rather reads as an inducement not to hand over the wallet,
and as a threat that the hearer will be punished with a splendid reward if he
does not comply. Since especially the possibility of a punishment by a splendid

1. Notice that in (1) letters P and R stand for punishment and reward respectively.

70

Robert van Rooij and Michael Franke

reward is awkward, the preferred threat-reading of (1d) may therefore lead to


verdicts of pragmatic infelicity.
The question that we would like to address in the following therefore is: what
makes it so that a disjunction of the form A P is preferably read as a threat
and not as a promise? The answer we give is that, from the point of view of
a rational speaker who is concerned with the efficacy of her inducements, an
alleged disjunctive promise, such as in (1d), is a suboptimal, i.e., irrational,
conversational move. Therefore, the preferred way of interpreting, i.e., rationalizing, a disjunctive inducement given by a rational speaker is as a threat, even
if that goes against contextual assumptions of hearer-desirability.
2.

Relation with Previous Accounts

The pragmatic puzzle we are trying to account for here may be familiar from
discussions of so-called pseudo-imperatives (see van der Auwera 1986; Bolinger
1979; Clark 1993; Lawler 1975 for early contributions). Pseudo-imperatives
(shortly, pis) are mixed mood sentences where an imperative clause is followed
by conjunction and or disjunction or and a declarative sentence. In other
words, a pi is a sentence of the form:
(2)

a.
b.

Do A, and X will happen/be the case/be done.


Do A, or X will happen/be the case/be done.

Interestingly, pis behave quite similar to the sentences in (1):


(3)

a.
b.
c.
d.

Close the window and I will kiss you.


Close the window and I will kill you.
Close the window or I will kill you.
? Close the window or I will kiss you.

threat Do A and R
promise Do A and P
threat Do A or P
threat Do A or R

What is peculiar is that we can find examples of conjunctive pis which read as
conditional threats (3a) and examples which read as conditional promises (3b),
depending on whether we assume that the hearer wants the declarative second
conjunct to be realized. But, for disjunction, only threat-readings, so to speak,
are possible: any second disjunct is either construed as hearer-undesirable or
else the whole disjunctive pi seems pragmatically infelicitous.
Several issues are worth the linguists attention here. Firstly, it needs to be
explained how a conjunction and can obtain a kind of conditional reading in
the first place, especially one in which the illocutionary force typically associated
with an imperative clause cancels out. Secondly, it needs to be explained how a

Promises and Threats with Conditionals and Disjunctions

71

disjunctive pi cannot function as a promise.2 It is the latter problem that this paper
deals with, including but not restricted to pis. In our exposition, we will focus on
the contrast between conditionals and disjunctions of declarative sentences of the
form in (1). This carries over to pis if we may assume that (i) conjunctive pis have
conditional readings, and that (ii) imperatives have some descriptive content that
refers to a hearer action or a hypothetical state of affairs. Our aim, then, is to be
as linguistically sober as possible: treating clauses as denoting propositions and
remaining as conservative as possible. In our analysis of conditionals, negation
and disjunction, we would like to explore to what extent rationales of influencing
others behavior by promises and threats alone can explain the contrast between
conditionals and disjunctions.
3.

Promises and Threats as Strategic Commitments

An account of conditionals and disjunctions as possible promises and threats


requires us to establish sufficient transparency of our view of the basic notions
of promise and threat. In line with both linguistic and game-theoretic analyses,
we take these to be special kinds of speech acts with which the speaker commits
herself to a particular course of events.3 Following the seminal work of Schelling
1960, we will explore how such commitment can be used strategically in order
to influence the hearers behavior, and what difference it may make to strategic
commitments whether they are of conditional or disjunctive form. Towards this
end, we will first look briefly at commitment-based analyses of speech-acts in
Sections 3.1. and 3.2. After that we will zoom in on the rationality requirements
of strategic commitments in Section 3.3.
3.1. Speech-acts and Speaker Commitment
In linguistic pragmatics, Gazdar (1981), for instance, proposes to analyze speech
acts in general as functions that change a context in such a way that it alters the
speakers commitments. An assertion that is true, for example, would come
out as an update function on the context of utterance such that after the utterance
2. Attempts to solve the first problem have been made by, for instance, Culicover and
Jackendoff (1997); Franke (2008); Jayez and Dargnat (2009); Russell (2007) and
Schwager (2006). Explanations of the second problem feature of disjunctive pis have
been attempted, for instance, by Clark (1993); Franke (2008); Schwager (2006) and
van der Auwera (1986).
3. Whether it is in her power to actually bring the event to which she commits herself
about, or whether it is only a doxastic commitment is not crucial for our concern.

72

Robert van Rooij and Michael Franke

(and its uptake) the speaker is committed, at least if prompted and within certain
limits, to defend the truth of (c.f. Brandom 1983; Hamblin 1970). Under this
view, promises and threats can be regarded as speaker commitments as well.
For instance, Gazdar (1981: p.69) writes: A promise that is a function that
changes a context in which the speaker is not committed to bringing about
into one in which he is so committed.4
What is a threat, and what is the difference between a threat and a promise?
Psychologically speaking, one can feel threatened by another person, just because one fears that this other person might be harmful. Notice that according to
this conception, one can feel threatened by someone without this person having
done anything. However, in this paper we will adopt a more operational notion. A
threat is a commitment by one person intended to change another persons future
behavior. But how, then, does a threat differ from a promise? Also a promise
involves a commitment of the speaker often with the intention to change the
hearers future actions. However, the major difference is that whereas in case
of a threat, the commitment has negative consequences for the other, with a
promise these consequences are positive.5
Still, there is a further difference between promises and threats. According
to Searle and Vanderveken (1985: p. 193), unlike in promises, no obligation
is involved in threatening.6 That means that although by uttering a threat the
speaker commits herself to carrying out a sanction if needed, she is strangely
enough not obliged to do so in the same way that she would be in case of a
promise. Clearly, if, say, Jones threatens to punish Smith in case he giggles,
then, after giggling, Smith will not insist on his punishment, and, more importantly, also cannot lay claim to a social obligation that he be punished. The case
of a promise, obviously, is essentially different in this respect. This is an interesting point to which we will return later: from the speakers point of view, a
threat is cheaper in expectation than a promise, all else being equal, because, as
Searle and Vanderveken put it, threatening is not as institutionally dependent
as promising (p. 193).

4. There are further felicity conditions that one would have to consult if a flawless
conceptual characterization of promises and threats qua speech act was at stake (the
locus classicus is Searles (1969) exemplary analysis of a promise). This is not crucial
though for any of our present concerns.
5. This does not exclude that we sometimes use the terms in a slightly misleading way,
as in If you lend me your wallet, I promise you I wont hurt you.
6. This point is corroborated by empirical data reported by Verbrugge et al. (2004).

Promises and Threats with Conditionals and Disjunctions

73

3.2. Conditionality of a Commitment


In order to be effective as an inducement, promises and threats should not be
made unconditionally: intuitively, the sentences in (1) do not commit the speaker
to severe punishment or splendid rewards come what may, but only in case the
hearer did or did not hand over her wallet; a promise of a splendid reward comewhat-may will not convince anyone to hand over his wallet, and neither will the
threat (if you can call it such in the first place) that one is to be severely punished
independent of ones own choices. For our present purposes, it is actually a
mood point to ponder whether such inducements are essentially conditional
commitments or commitments to conditionals.7 What ultimately counts for the
efficacy of an inducement by promises and threats is that the hearer will get to
understand that a certain reward or punishment is conditional upon his choice
of action.
From this point of view, the presumption of rationality of strategic commitments explains the fact that conditional promises and threats readily receive
biconditional, so-called conditional-perfection readings.8 A conditional threat
also implicates a promise, and the other way around. The threat that you will
be punished if you do not hand over your wallet is effective only to the extent
that (it is commonly understood that) the speaker promises that you will not be
punished if you do hand over the wallet. An analogous argument applies to inducements by promises, of course. Hence, under the assumption that the speaker
is rational, i.e., concerned with the maximal efficacy of her statements also on
the level of what is implicated, we should enrich a conditional promise or threat
to a biconditional reading. Obviously, the same rationale applies to exclusive
readings of disjunctions: also a threat such as (1c) is read exhaustively; and so
should, all else being equal, a promise like (1d), if it was felicitous after all. We
will come back to this point also later in our analysis in Section 4.

7. Stalnaker 2006, for instance, argues that under commitment-based analyses of assertion conditional assertions and assertions of conditionals come down to essentially
the same thing.
8. For theoretical assessments of reading if as if and only if see, for instance, Geis
and Zwicky (1971); Horn (2000); van der Auwera (1997) and references therein.
Empirical research supporting the strong availability of an if and only if-reading in
conditional promises and threats is reported by Fillenbaum (1986) and van CanegemArdijns and van Belle (2008).

74

Robert van Rooij and Michael Franke

3.3. Strategic Commitments


The view that conditional promises and threats are strategic commitments of the
speaker with the aim of influencing the hearers choice of action in a gametheorylike situation, is a centerpiece in the analysis of human interaction that Thomas
Schelling became famous for (Schelling 1960).9 In sequential games it is normally the agent who acts first who has the advantage. The idea of making a
strategic commitment is to seize the initiative, even if your action follows that
of the other. On Schellings account, a commitment of the speaker is modelled
as a pruning of the game tree: by a binding public announcement the speaker ostensibly excludes some of her action alternatives in a sequential game (cf. Klein
and OFlaherty 1993). Look at the simple game in Figure 1: the opening move
is the hearers choice of A or A and subsequently the speaker performs some
action X or Y . The commitment if A, then X , for example, is modelled as
pruning (at least) the branch w2 from the game (and possibly also, if conditional
perfection is taken into account, the branch w3 ).

v1
X

v2
Y

w1

w2

w3

w4

Figure 1. Sequential game with potential for commitment

A strategic commitment is a commitment in this sense with the intent to


influence the hearers behavior. Obviously, by the prospect of a splendid reward,
for instance, the speaker can persuade the hearer into performing an action
that the speaker likes, but the hearer dislikes. Such a strategic commitment is
subject to a number of obvious rationality constraints, in particular whether the
commitments are credible, beneficial and efficacious (see Klein and OFlaherty
1993, for a more thorough formal account).
9. It is here also that promises and threats differ from assurances and warnings: they
might have the same effect, but only the former two involve commitments.

Promises and Threats with Conditionals and Disjunctions

75

Credibility
The rationality constraint that received most attention in the economic literature
is that a threat or promise should be credible. To punish or reward someone else
can be costly. This is obvious if you promise to give some money, but the (future)
consequences of punishing (e.g. killing) somebody if she doesnt perform the
desired action can be very costly as well. As already observed by Schelling
(1960: p. 177), from this it immediately follows that threats costs more when
they fail, while promises cost more when they succeed. But given that effective
threats and promises involve only conditional commitments, a major issue arises
as to whether the threat or promise was credible (see Hirschleifer 2001).
Above, we have suggested to model commitments as pruning of a game tree.
But the idea that a formerly possible action is entirely excluded from a players
choice set just by public announcement that she will not choose to play so,
is unrealistically strong. In reality, the option of playing Y after A, even after
having said if you do A, I will do X remains. The problem is that as long as
the option Y is still available in principle, it might still be chosen, and, more
strongly even, it might even be rational to choose it.
Suppose that you made the conditional threat or promise to do X , if the
hearer performed A. That means that you are committed to do X after the hearer
performed A. But, irrespective of anger or gratitude, if you like doing Y more
than doing X , i.e., if you prefer outcome w2 over outcome w1 in Figure 1, then
it is (only) rational to deviate from your previous commitment. In case of a
threat, harm is already done, and in case of a promise, you already have what
you desired. Why would you stick to your commitment?
There is an obvious reason why you should carry out your commitment under
these circumstances after all, if there is a good chance that you will be engaged
with the other person in similar circumstances again in the future. Carrying
out the commitment strengthens your reputation, while not carrying it out only
destroys it. But if reputation cannot be brought into the picture, the only way in
which threats and promises like in (1) can be credible is when they are (seen to
be) costless. And this also makes intuitive sense. Take for instance the threat in
(1a). If the threatener is seen to be desperate or irrational enough, one cannot
rule out that he wont kill you if you dont give your wallet: perhaps he simply
doesnt care about the possible consequences if things dont go as he desires.
And indeed, to make your threat look credible, it is not unwise to act (as if you
are) irresponsible.
It is clear that a lot can (and has been) theorized about credibility of inducements by conditional promises and threats from a game-theoretic point of view.
For the purposes of this paper, however, it is sufficient to simply assume that

76

Robert van Rooij and Michael Franke

promises and threats are credible: the involved obligations are binding. More
precisely even, our analysis proceeds from the assumption that the speaker believes that her statements will be believed; it is inessential whether this belief is
actually correct.
Benefit and efficacy
Since we are concerned mostly with the speakers perspective, we will concentrate, instead, on the other rationality constraints of promises and threats: that
strategic commitments should be beneficial for the speaker (in expectation), and
that they should be efficacious (in expectation).
Take a conditional promise like if you do A, I will reward you with R with
which the speaker commits herself to a reward R after the hearer has performed
A. Naturally, the benefit for the speaker of having A performed should exceed
the speakers detrimental cost of paying the reward R in order to count as a
rational inducement. (This also entails that the speaker should not, of course,
promise to reward performance of an action that the speaker would prefer not
to have performed.) It is also clear that a strategic promise is only efficacious to
the extent that the promised reward R is a sufficient incentive for the hearer to
perform an otherwise dispreferred action A. So, while for the speaker the danger
of having to pay the cost of the reward R must not exceed the expected gain of
the action A (benefit), for the hearer the gain of obtaining R must exceed the
loss in performing A (efficacy). Thus, if you do A, I will reward you with R
is a rational and effective promise just in case both the speaker and the hearer
prefer A R above A R.
Similar considerations apply to conditional threats. Uttering a threat if you
do A, I will punish you with P commits the speaker to punishing the hearer
if he does not perform A. This is beneficial only if the chance that the hearer
performs A and the associated benefit of that for the speaker exceeds the possible
cost that P might have if the hearer does not perform A. In order to be efficacious,
the danger of being punished by P when A is not performed must outweigh the
loss of performing A. So, while for the speaker the danger of having to punish
the hearer by P must not exceed the expected gain of having A performed
(benefit), for the hearer the danger of being punished by P must exceed the loss
of performing A (efficacy). Thus, if you do A, I will punish you with P is a
rational and effective threat just in case both the speaker and the hearer prefer
A P above A P.
An analogue story holds for disjunctive threats of the form You do A or I
will punish you with P. In this case, the speaker commits himself to punish the
hearer who does not perform A with P. This is beneficial just in case the speaker

Promises and Threats with Conditionals and Disjunctions

77

expects that the benefit of action A taking place exceeds the cost of performing
P in case the hearer abstains from A. The threat is efficacious in case the hearer
fears P more than the harm he expects from performing A. Thus, You will do
A or I will punish you with P is a rational and effective threat just in case both
the speaker and the hearer prefer A P above A P.
All these arguments seem completely straightforward and intuitive. The problem is that a seemingly equally straightforward argument can be given for why,
and when, disjunctive promises of the form You do A or I will reward you
with R make sense. What would be wrong with a disjunctive promise, if both
speaker and hearer would prefer A R to A R? As far as our argument
goes so far, there is nothing that would suggest that disjunctive promises are
any different from a conditional promise. The disjunctive promise would be
beneficial just in case the speakers loss of A not being performed exceeds the
cost of giving the reward R, and it would be efficacious iff the reward R would
be bigger for the hearer than the cost of performing A. The problem is that a
disjunctive promise of the form You do A or I will reward you with R has
exactly the same preference structure as the conditional promise If you do A, I
will reward you with R. Still, the latter is acceptable, but the former is not (as a
promise). The problem is why? What is the difference between a conditional and
a conjunctive promise that makes the former a reasonable strategic commitment
and thereby a feasible move in dialogue, but not the latter?
4.

Disjunctive Promises are Risky Inducements

In response to this question, we suggest to scrutinize more carefully the speakers


expected utility by taking into account that the speaker cannot know for sure
what preferences and beliefs the hearer has. Doing so, we propose that disjunctive promises like (1d) are suboptimal inducements because they are the only
inducements in (1) that contain a risk of inefficiency in the light of the speakers
uncertainty that cannot be compensated.
More concretely, the main idea, to be spelled out in this section, is that, even
where semantically equivalent, there is a minor linguistic difference between a
conditional A X and a disjunction A X : the former mentions the possibility A, while the latter mentions the possibility A. We will assume that
mentioning raises salience, however slightly, and that mentioning will therefore
increase the probability with which the speaker expects the mentioned possibility to be realized. This then affects the speakers expected utility of choosing
a given inducement. Clearly, for (1b) and (1c), mentioning a speaker-desirable
option only increases the speakers expected utility and is unproblematic. Alter-

78

Robert van Rooij and Michael Franke

natives (1a) and (1d), on the other hand, mention a speaker-undesirable option
which slightly decreases expected utility. This puts the efficacy of these statements at risk, given that the speaker cannot be certain about the hearers actual
preferences and beliefs. However, a conditional threat like (1a) can compensate
for this risk by committing to a stronger punishment, which is cheap in expectation, as explained in Section 3.1. Committing to a stronger reward is not cheap
in expectation, and therefore a disjunctive promise cannot compensate for the
risk of inefficacy.
In order to spell out this idea, we first have to enlarge on the assumption that
mentioning an alternative slightly raises the speakers expectation of realization
(Sections 4.1. and 4.2.). Secondly, we have to spell out the structure of the
speakers uncertainty (Section 4.3.) and how this all affects the expected utility
of inducements (Section 4.4.).
4.1. Mentioning in Disjunctions and Conditionals
The standard view of the effect of an assertion that (and its acceptance) is
that it eliminates all possibilities from the common ground where is not true
(Stalnaker 1978). But even before an assertion is accepted, the fact that the
assertion was made and understood had already another effect on the common
ground: the possibility that is or might become true is brought to the (joint)
attention of the participants of the conversation, in particular, to the hearer
(c.f. de Jager 2009; Swanson 2006). For simple assertions with the content It is
raining this extra effect is negligible because it is a side-effect of the acceptance
of the assertion anyway. For more complex assertions, however, this extra effect
does not fall out as a consequence of the assertion by itself. To see that this is so
also for disjunctions and conditionals look at the simple question-answer pairs
in (4).
(4)

Who (of John and Mary) came to the party?


a. John.
b. John or John and Mary.
c. John, if not also Mary.

Although all three answers are semantically equivalent,10 they certainly do not
convey the same idea. Whereas (4a) implicates that Mary did not come, answers
(4b) and (4c) both implicate that it is also possible that Mary might have come
together with John. The intuitive reason why is because otherwise the speaker
10. This holds for logical disjunction and material implication, but also for other standard
analyses of the conditional, such as (variably) strict implication.

Promises and Threats with Conditionals and Disjunctions

79

would not have mentioned this latter possibility, despite the semantic equivalence
with the answer in (4a) (cf. Gazdar 1979; Schulz and van Rooij 2006).
In general, from just eliminating those possibilities in the common ground
where or are true, the extra effect of bringing - and -possibilities
to the attention does not follow. In fact, it has been argued that a very important
purpose (among others) of a disjunctive claim is to bring its disjuncts to the
attention (c.f. Geurts 2005; van Rooij 2005; Zimmermann 2000). Similarly, the
antecedents of (indicative) conditionals are normally associated with a
speaker presupposition that be possible (e.g. Stalnaker 1975). Taken together,
whatever the concrete mechanisms at work, it is fair to say that, on top of their
semantic meaning, it is important to the way that disjunctions and conditionals
are processed in discourse that these constructions mention, or are about, certain
states of affairs.
4.2. Mentioning, Salience, and Priming
But now suppose that a conditional or a disjunction such as in (1) is uttered
where the truth of the antecedent or first disjunct is under the control of the
hearer. What effect does it have to mention an action under hearer control in
a game-like setting where the speaker wants to induce a certain action in the
hearer?
First of all, if the hearer has not been aware of it at all, then just mentioning
an action will inevitably make him aware of it. To make the hearer aware of a
possibility that the speaker does not want to be realized might therefore just be
a very dumb move in conversation because it would put the wrong ideas into the
hearers head (de Jager 2009; Franke and de Jager 2011). In other words, if the
hearer is (possibly) unaware of some action A that the speaker does not want to
have performed, then it is, intuitively speaking, a deficient inducement strategy
to mention it in the first place unless it is strongly and credibly discredited, such
as by a threat of sanction or similar.
But what if an action is mentioned that the hearer is (most likely) already
aware of? Even then, it may seem prima facie suboptimal to bring such a possibility to the hearers attention in case the speaker doesnt want this possibility
to be(come) true. Mere mentioning makes an option salient, and to increase the
salience of a choice option for the hearer simply means that, from the speakers
point of view, the probability with which this option is chosen increases, even if
only very slightly. Think of marketing and advertisement: you want your product
name to be ubiquitous, you want it to be the first thing that comes to mind when
consumers make a decision (c.f. Nedungadi 1990). But also in more on-the-spot
decision making: the salience of choice options matters especially when these

80

Robert van Rooij and Michael Franke

have a strong desire-raising component (compare the discussion of incentive


salience by Zhang et al. 2009 and references therein).
Taken together, this seems to suggest that when trying to manipulate hearer
behavior the speaker should not mention and thereby raise the salience of
speaker-undesirable, yet hearer-desirable options without at the same time discrediting those. In other words, raising the salience of a choice option A stands
a mild chance of priming the hearer into choosing A (if not actual, then at least
in speaker expectation). In strategic inducements this increase in the probability
that the hearer performs hearer-undesirable A leads to a decrease in expected
utility. This decrease, intuitively speaking, must be compensated by highlighting
negative aspects or consequences of A, or otherwise the strategic inducement
might rather lead to performance of A than to abstinence from it. This is possible for threats that mention speaker-undesirables, but not for promises. It is this
intuitive argument that the remainder of this section spells out in more detail.
4.3. Towards a more Realistic Game Model
We would like to compare in particular the benefit and efficacy of conditional
and disjunctive promises and threats in the light of a sufficiently realistic game
model that incorporates the speakers natural uncertainty about the concrete
preferences and beliefs of the hearer. Since performance of an action A is at
stake, together with a potential reward or punishment thereafter, the minimal
sequential game model we should consider is (something like) the one given in
Figure 2: the hearers choice is between A and A, and subsequently the speaker
may choose to either reward the hearer (R), punish him (P) or stay neutral and
abstain from both (N ).
To be entirely precise, there should really be a set of rewards {R1 , R2 , . . . }
(and similar for punishments). The hearer would prefer some of the rewards
over others and, in the simplest case, the more the hearer desires a reward,
the more costly it is for the speaker to give. We would then have to ponder
whether this set is reasonably infinite, and which properties the hearers and
speakers preference-ordering have and how these are related to each other.
We will sidestep these details here and simply consider some reward R that is
speaker-costly but hearer-desirable, and some punishment P that is also speakercostly and hearer-undesirable.
More specifically, we will assume that the speakers preferences are qualitatively of the following form:
w 2 > w3 , w 1 > w5 > w6 , w 4 .

Promises and Threats with Conditionals and Disjunctions

81

v1

v2

w1

w2

w3

w4

w5

w6

Figure 2. Game model for punishing and rewarding hearer actions

This captures the intuition that the speaker mostly cares about whether action A
is performed. Subordinate to her preference for A, she would prefer to remain
neutral over punishing and rewarding. In contrast to that, we should assume that
the hearer prefers A over A, but that the reward R and the punishment P that we
consider are potentially efficacious, so that they outweigh the hearers preference
about A. More precisely, the hearers preferences are then qualitatively given as:
w4 > w1 > w5 > w2 > w6 > w3 .
In other words, the hearer (is assumed by the speaker) to value most the reward, and prefers a neutral outcome over a punishment. Subordinate to these
preferences is his preference of performing A over performing A.
It is not necessary, but also not desirable to specify the hearers preferences
any further than that. This is because a speaker will never be able to know for
sure how exactly the hearer will value a promise or a reward. Our modelling here
adopts the speakers perspective and takes her natural uncertainty into account. In
other words, the model assumes that the speaker believes the hearers preferences
are qualitatively as specified above, but that the speaker does not know for certain
how strongly, for example, w4 is preferred over w1 .
Similar remarks then also apply to the speakers beliefs about the hearers
beliefs. Here it is most natural to suppose that the speaker believes that the
hearer expects, all else being equal, a neutral outcome. We could even go as
far as saying that the hearer might not even be aware of possible punishments
and threats and that it is only when pointed out to him that he accommodates
these possibilities into his decision-making. To keep matters simple here, we
will refrain from representing such a sequential game with possibly unaware
players (c.f Feinberg 2005; Heifetz; Meier and Schipper 2009). For the present

82

Robert van Rooij and Michael Franke

purpose it suffices to assume that the speaker believes that the hearers beliefs
are qualitatively as follows:
w2 , w5  w1 , w3 , w4 , w5 .
The idea is that the speaker again does not know precisely which probabilistic
beliefs the hearer holds, but she does believe that, barring any speaker commitment, the hearer considers action N substantially more likely than either reward
or punishment.
4.4. Risk of Strategic Inducements
Given the speakers natural uncertainty about the hearers precise preferences and
beliefs, it turns out that disjunctive promises are risky, and therefore suboptimal
in expectation, in a sense that threats and conditional promises are not. To
see what is at stake, we need to compare the statements in (1) one by one
as committing strategic inducements against the background of the speakers
uncertainty as described in the previous section. Let us look at threats and
promises in turn and let us ask what update effects these statements would have
on the hearer and how this affects the speakers assessment of her expected
utility of uttering these statements.
As for their semantic update effect, both the conditional threat A P in (1a)
and the disjunctive threat A P in (1c) are semantically equivalent and denote, if
taken as binding, the set {w1 , w2 , w3 , w6 }. If we also take conditional perfection,
respectively exclusive readings of disjunctions, into account the impact of these
threats is an update that leaves only outcomes {w2 , w6 }.
Still, in line with our reasoning above, there should be a small difference
between the conditional and the disjunctive threat. Whereas the conditional
threat slightly increases the probability of w6 (in the expectation of the speaker),
the disjunctive threat slightly increases the probability of w2 . This is because the
conditional mentions A and so the speaker will assume a slight increase in the
chance that the hearer will play this option. For the disjunctive threat rather the
hearer choice A is given a slightly higher probability. That means that mentioning
the speaker-desirable action A in the disjunctive threat actually has a slight
increasing effect on the expected utility of that statement, as compared to the
conditional threat that mentions A. However, this slightly detrimental effect
of mentioning the speaker-undesirable action is relatively harmless, because the
speaker believes that the hearer prefers w2 over w6 and the speaker believes
that the hearer considers w2 much more likely than w6 . It is therefore not likely
that the conditional threat would not be efficacious despite the fact that it might
slightly increase the chance of performance of A.

Promises and Threats with Conditionals and Disjunctions

83

Moreover, and more importantly, the speaker can compensate for the risk of
a conditional threat by choosing a sufficiently stronger punishment. As noted in
Section 3.1., this need not decrease the speakers expected utility, because threats
are (relatively) cheap in expectation in that the (possibly costly) punishment is
not (as) socially binding as in the case of a promise.
This is different for disjunctive promises. Again, the conditional promise
A R in (1b) and the disjunctive promise A R in (1d) are semantically
equivalent. Their update effect is to eliminate outcomes w2 and w3 , and additionally, if perfection and exclusive readings are taken into account, restrict the
options under consideration to {w1 , w5 }. Once more, we also attest a difference
from mentioning different alternatives: whereas the conditional promise slightly
increases the speakers expected utility because it mentions the desirable option
A and thus increases the probability that w1 is realized, the disjunctive promise
slightly decreases the expected utility by mentioning the undesirable option A
and thereby increasing the probability of w5 . However, unlike with threats, this
latter decrease is more risky from the point of view of an uncertain speaker: the
problem is that although w1 is assumed more hearer-desirable than w5 , the latter
is naturally assumed substantially more likely. If the speaker is uncertain about
the extent to which the hearer prefers w1 over w5 , mentioning the undesirable
option puts the efficacy of the inducement at risk.
But could the speaker not compensate this risk, as she could with conditional
threats, by promising a higher reward, so as to make sure that w1 is sufficiently
preferred over w5 ? She probably could, but not necessarily without sacrificing
even more on expected utility. Promises are costly when efficacious and more
institutionally dependent than threats. To the extent that the speaker would like
to invest on the promise to compensate risk, the statements expected utility
decreases, because any stronger reward would only be more costly and thus
further decrease the speakers expected utility. This is then the main difference
between conditional threats and disjunctive promises: although both mention
a speaker-undesirable option, which is risky under uncertainty, threats, but not
promises, can be pumped up cheaply, so to speak, to compensate for the risk.
Taken together, we argue that disjunctive promises are suboptimal, because
they emphasize the wrong alternative and cannot compensate for any negative
effects that this might have. It is the combination of natural speaker uncertainty,
priming by mentioning and the asymmetry of when punishments and rewards are
speaker-costly that explains why disjunctive promises are a deficient inducement
strategy.

84

Robert van Rooij and Michael Franke

5.

Conclusion

In this paper we explained promises and threats as strategic commitments of


the speaker. We discussed the constraints that such commitments have to obey
to be rational. We saw that from the speakers point of view, a threat is cheaper
in expectation than a promise. The main idea presented in this paper was that
disjunctions like (1d) preferably get a threat-reading because their use as a
disjunctive promise is a risky, and therefore suboptimal strategic inducement.
This explanation involved the idea that mere mentioning a possibility raises
its salience. This helps to explain the difference in acceptability between the
conditional promise A R and the disjunctive promise A R. The idea that a
threat is cheaper in expectation than a promise, on the other hand, explains why
the disjunctive promise is also more risky than the conditional threat A P.
This explanatory strategy certainly raises a number of concerns. Perhaps
the most pressing is the question whether the alleged suboptimality of a disjunctive promise is something that is checked on-the-spot, every time anew a
speaker would like to influence a hearer. We emphatically do not subscribe to
this obviously nonsensical view. Rather we suggest here that the suboptimality
of disjunctive promises is a force that informs language organization, not ad
hoc choice of formulation. Certain locutions and grammatical constructions,
and not others, are conveniently and conventionally used for certain discourse
functions, and not others. It is at this level of functional organization that, we
suggest, evolutionary pressures have weeded out disjunctions as a viable vehicle
of making promises.
References
Bolinger, Dwight
1979
Is the imperative an infinitive? In: Bolinger, Dwight (ed.), Meaning
and Form, 152182. New York: Longman.
Brandom, Robert
1983
Asserting. Nous 4: 637650.
Clark, Billy
1993
Relevance and pseudo-imperatives. Linguistics and Philosophy 16:
79121.
Culicover, Peter W. and Ray Jackendoff
1997
Semantic subordination despite syntactic coordination. Linguistic Inquiry 28(2): 195217.
de Jager, Tikitu
2009
Now that you mention it I wonder...: Awareness, Attention, Assumption. PhD thesis, Universiteit van Amsterdam.

Promises and Threats with Conditionals and Disjunctions


Feinberg, Yossi
2005

85

Games with incomplete awareness. Research Paper No. 1894, Stanford University.
Fillenbaum, Samuel
1986
The use of conditionals in inducements and deterrents. In: E. C. Traugott, A. ter Meulen, J. S. Reilly and C. A. Ferguson (eds.), On Conditionals, 179195. Cambridge University Press.
Franke, Michael
2008
Pseudo-imperatives and other cases of conditional conjunction and
conjunctive disjunction. In: C. Fabricius-Hansen, and W. Ramm (eds.),
Subordination versus Coordination in Sentence and Text From a
Cross-Linguistic Perspective, Studies in Language Companion Series
(SLCS), 255279. Amsterdam, Philadelphia: John Benjamins.
Franke, Michael and Tikitu de Jager
2011
Now that you mention it: Awareness dynamics in discourse and decisions. In: Anton Benz, Christian Ebert, Gerhard Jager and Robert van
Rooij (eds.), Language, Games, and Evolution, 6091, Heidelberg:
Springer.
Gazdar, Gerald
1979
Pragmatics: Implicature, Presupposition, and Logical Form. New
York: Academic Press.
Gazdar, Gerald
1981
Speech act assignment. In: Joshi, A. K., Webber, B., and Sag, I. A.
(eds.), Elements of Discourse Understanding, 6483. Cambridge University Press.
Geis, Michael L. and Arnold M. Zwicky
1971
On invited inferences. Linguistic Inquiry 2(4): 561566.
Geurts, Bart
2005
Entertaining alternatives: Disjunctions as modals. Natural Language
Semantics 13: 383410.
Hamblin, Charles L.
1970
Fallacies. Methuen, London.
Heifetz, Aviad, Martin Meier and Burkhard C. Schipper
2009
Dynamic unawareness and rationalizable behavior. Unpublished
manuscript.
Horn, Laurence R.
2000
From if to iff : Conditional perfection as pragmatic strengthening.
Journal of Pragmatics 32: 289326.
Hirschleifer, Jack
2001
Game-theoretic interpretations of commitment. In: R. M. Nesse (ed.),
Evolution and the Capacity for Commitment, 7793. Russell Sage
Foundation.

86

Robert van Rooij and Michael Franke

Jayez, Jacques and Mathilde Dargnat


2009
One more step and youll get pseudo-imperatives right. In: Torgrim
Solstad and Arndt Riester (eds.), Proceedings of Sinn und Bedeutung
13, 247260, University of Stuttgart.
Klein, Daniel B. and Brendan OFlaherty
1993
A game-theoretic rendering of promises and threats. Journal of Economic Behavior and Organization 21(3): 295314.
Lawler, John M.
1975
Elliptical conditionals and / or hyperbolic imperatives: Some remarks
on the inherent inadequancy of derivations. In: Papers from the 11th
Regional Meeting. Chicago Linguistic Society, 371382. Chigaco Linguistic Society.
Nedungadi, Prakash
1990
Recall and consumer consideration sets: Influencing choice without
altering brand evaluations. Journal of Consumer Research, 17: 263
276.
Russell, Bertrand
2007
Imperatives in conditional conjunction. Natural Language Semantics
15: 131166.
Schelling, Thomas C.
1960
The Strategy of Conflict. Harvard University Press, Cambridge, Massachusetts.
Schulz, Katrin and Robert van Rooij
2006
Pragmatic meaning and non-monotonic reasoning: The case of exhaustive interpretation. Linguistics and Philosophy 29: 205250.
Searle, John
1969
Speech Acts: An Essay in the Philosophy of Language. Cambridge
University Press.
Searle, John and Daniel Vanderveken
1985
Foundations of Illocutionary Logic. Cambridge University Press.
Schwager, Magda
2006
Interpreting Imperatives. PhD thesis, Johann-Wolfgang-Goethe Universtitt zu Frankfurt am Main.
Stalnaker, Robert
1978
Assertion. In: P. Cole (ed.), Syntax and Semantics, vol. 9, 315332.
New York: Academic Press.
Stalnaker, Robert
1975
Indicative conditionals. Philosophia 5(3): 269286.
Stalnaker, Robert
2006
Conditional assertions and conditional propositions. Paper presented
at the Formal Epistemology Workshop, UC-Berkeley.

Promises and Threats with Conditionals and Disjunctions


Swanson, Eric
2006

87

Interactions with Context. PhD thesis, Massachusetts Institute ofTechnology.


van Canegem-Ardijns, Ingrid and William van Belle
2008
Conditionals and types of conditional perfection. Journal of Pragmatics 40: 349376.
van der Auwera, Johan
1986
Conditionals and speech acts. In: E. C. Traugott, A. ter Meulen, J. S.
Reilly, and C. A. Ferguson, (eds.), On Conditionals, 197214. Cambridge University Press.
van der Auwera, J.
1997
Conditional perfection. In: A. Athanasiadou and R. Dirven(eds.), On
Conditionals Again, 169190. John Benjamins.
van Rooij, Robert
2005
A modal analysis of presupposition and modal subordination. Journal
of Semantics 22: 281305.
Verbrugge, Sara, Kristien Dieussaert, Walter Schaeken and William Belle
2004
Promise is debt, threat another matter: The effect of credibility on the
interpretation of conditional promises and threats. Canadian Journal
of Experimental Psychology 58(2): 106112.
Zhang, Jun, Kent C. Berridge, Amy J. Tindell, Kyle S. Smith and J. Wayne Aldridge
2009
A neural computational model of incentive salience. PLoS Computational Biology 5(7): 114.
Zimmermann, Thomas Ede
2000
Free choice disjunction and epistemic possibility. Natural Language
Semantics 8: 255290.

Part II.
Sentence Types and Clausal Peripheries

Revisiting the CP of Clefts*


Adriana Belletti

1.

Introduction

This paper reconsiders the analysis of the CP of cleft sentences developed in previous works (Belletti 2008, 2009), with the main aim of going deeper into some
of its aspects, and widening its empirical coverage. The main issues addressed
in this perspective concern in particular: the way the locality of syntactic computations conditions the syntax and interpretation of clefts; the detailed analysis
of the shape of the CP domain in clefts; aspects of the interpretation of clefts in
combination with closely related structures, such as the clausal complements of
perception verbs.
The analysis is framed in cartographic terms (Cinque 2002 ed.; Rizzi 2004
ed.; Belletti 2004 ed.; Cinque and Rizzi 2010, and related work), according
to which dedicated positions in the functional structure of the clause overtly
express different interpretations in different syntactic positions. This richness in
the functional structure can have crucial consequences for the relevant locality
principle operating in syntax, whose specific operation is considered here in the
domain of clefts. A further consequence of the approach is that it also allows
one to explicitly identify and characterize both the content and the size of the
functional domain: the CP space of clefts is analyzed in some detail in this
perspective.

* I thank here the people from whose comments and remarks this article has specially
benefited, in different presentations in different occasions: Paola Beninc`a, Liliane
Haegeman, David Pesetsky, Luigi Rizzi, Dominique Sportiche, Edwin Williams, Ede
Zimmermann.

92

Adriana Belletti

2.

Some Bbackground

2.1. The Essential Features of the Assumed Analysis


The crucial insight of the proposed analysis is twofold: on the one hand it is
assumed that a general property of cleft sentences is that the copula selects
as its sentential complement a reduced CP1 which lacks (at least) the highest
ForceP layer, assuming the articulated analysis of CP schematized in (1) (Rizzi
1997 and much related work):2
(1)

[ForceP . . . [TopP [FocP [TopP . . . [FinP [TP

On the other hand, such reduced CP complement may have a partly different
status in subject and non-subject clefts (referred to with the general term object
clefts, henceforth). Only in subject clefts the reduced CP complement is (/may
be) endowed with an EPP feature, the formal way to represent the predication
relation which is expressed by this type of clauses. The proposal is summarized
in (2):
(2)

a.

b.

Subject clefts:
the complement of the copula is/may be a reduced CP containing
an EPP feature
Object clefts:
the complement of the copula (always) is a reduced CP with no EPP
feature

1. See Belletti (2008, 2009) for more details. Throughout I will use the general label
CP to refer to sentential (small) clausal complements. Where relevant, I will specify
some of the possible labels of the reduced CP. That the copula in clefts selects a
CP containing the clefted constituent is a proposal sharing similarities with classical
analyses such as those in Ruwet (1975), Kayne (1994) and related work; see also
Clech-Darbon, Rebuschu, Rialland (1999) for an overview of this type of approach
to the analysis of clefts.
The paper will only address the analysis of argument clefts, where the clefted constituent is a DP, either a subject or a direct object (with occasional reference to clefted
PPs).
2. Only some of the positions assumed in the left periphery from much work in the
literature are represented in (1); specifically, only the discourse related positions of
Focus andTopic which will be referred to in some of the discussion below are indicated
in (1). See also Rizzi (2004); Beninc`a and Poletto (2004); Haegeman (2004); Bocci
(2004); Grewendorf (2005); Mioto (2003); Bianchi and Frascarelli (2009), for some
items of a rich relevant literature dealing with the map of the left periphery and its
interpretable positions in the sense of Chomsky (1995, 2005).

Revisiting the CP of Clefts

93

In (2) the fundamental distinction between the two types of clefts is stated.
According to this proposal, subject clefts have one option more than object
clefts: their reduced CP complement may be endowed with an EPP feature.
As far as its size is concerned, the CP complement of the copula is thus a
small CP in both subject and object clefts, since it lacks some layers. Technically, it can be defined as a small clause (Starke 1995) in the case of subject
clefts, as only in this case a predication relation holds between the clefted subject
and the rest of the (relative) clause following (Stowell 1983; Burzio 1986; Moro
1997; Rothstein 2000). Examples of a subject and an object cleft are given in
(3a) and (3b), from French and Italian respectively:
(3)

a.
b.

S: Cest
it is
O: E
(it) is

Marie qui a
parle
Marie that
has spoken
MARIA che (i ragazzi) hanno incontrato
Maria that the boys
have met

According to the proposal in (2), subject clefts in which the copula selects a small
clause CP and object clefts in which the copula selects a reduced CP, correspond
to the schematic structures in (4) following. For convenience, the shortcut be
indicates the copula and che indicates the complementizer, the realization of
C in Italian clefts. The crucial hypothesis in (4) concerning the complementizer,
is that in clefts it is the realization of the Fin head.3 It is left open for now, the
amount of reduction the CP undergoes (3.2 for relevant discussion).
(4)

a.
b.

Subject clefts (selecting a CP small clause):


T . . . be [CP Force . . . . . . EPP. . . . . . [FinP che [TP S . . . O/(PP)]]]]
Object clefts (selecting a reduced CP):
T . . . be [CP Force . . . . . . . . . [FinP che [TP S . . . O/(PP)]]]]

The different analysis corresponds to the possibly different interpretations that


subject clefts may allow in contrast to object clefts. The following section takes
up the interpretation issue.4

3. Only finite clefts are considered here.


4. That the position occupied by a subject in the CP is different from the one occupied
by an object is not a unique property of the CP of clefts. See Haegeman (1996) for
similar conclusions in relation to the position of a subject and the position of an object
in Germanic (Dutch) V2.

94

Adriana Belletti

2.2. The Interpretation of Subject and Object Clefts


Although a cleft sentence always instantiates a form of focalization, as has generally been observed in different languages (Kiss 1998; Abels and Muringi 2005)
its nature and interpretation may differ in subject and object clefts, as the different writing of the clefted constituent in (3) suggests. In the case of the subject
cleft in (3a), the clefted subject can be interpreted as the focus of new information, while the clefted object in (3b) only allows for a contrastive/corrective
focus interpretation, indicated with capital letters throughout (in order to make
the different interpretations immediately recognizable). This is clearly revealed
in the context of question-answer pairs. In languages such as French a cleft
(possibly reduced, leaving the relative predicate unpronounced) can typically be
utilized as the answer to a question of information on the identification of the
subject. This type of answering strategy, however, cannot be utilized to identify
the object (Belletti 2008, 2009 and references cited therein, bearing also on
relevant L1 and L2 acquisition data):
(5)

Q
A

(6)

Q
A

Qui (est-ce qui) a parle?


Who spoke?
Cest Jean (qui a parle).
It is Jean (who spoke).
Quest-ce que tas achete ( /Quas-tu achete)?
What have you bought?
( *) Cest un livre (que jai achet
e).
It is a book.

Similarly for a clefted PP:


Q
A

Avec qui es-tu sorti?


With whom did you go out?
( *) Cest avec Jean (que je suis sorti).
It is with Jean.

Of course, there is nothing wrong with the form of the clefts in (6), which are
perfectly grammatical sentences in French, in particular in their non-reduced
form. What is wrong, is just their use as an answer, in the question-answer context. They cannot answer a question, which is simply asking for the identification
of the object, with no other presupposition implied. The clefts in (6) are perfect

Revisiting the CP of Clefts

95

in, e.g., a contrastive/corrective context as the one in (7), which illustrates the
direct object case5 :
(7)

a.

b.

Context:
On ma dit que hier tas achete un journal.
They told me that yesterday you have bought a newspaper.
Correction:
Non, cest UN LIVRE que jai achete.
It is a book that I have bought.

In contrast to object clefts which only allow for a contrastive/corrective focus


interpretation of the clefted constituent, subject clefts, beside the new information focus interpretation illustrated in the French example in (5), also allow for
a contrastive/corrective interpretation in the appropriate context. (8) illustrates
this point:
(8)

a.

b.

Context:
On ma dit que Marie a parle.
They told me that Marie has spoken.
Correction:
Non, cest JEAN qui a parle.
No, it is JEAN that/who has spoken.

Hence, subject clefts have one interpretive option more. Ultimately, this further
option should be ascribed to the possibility for subject clefts to also admit the
same syntactic analysis as object clefts.
How can the difference between subject and object clefts be expressed? Why
do subject clefts and object clefts differ in their interpretative possibilities in the
described way?
Following the cartographic analysis in Belletti (2004, 2009), I assume the
existence in the low part of the TP clausal map of a vP periphery containing a
Focus position, characteristically dedicated in different languages to host purely
new information constituents. For instance, this low Focus position is assumed
to host post-verbal subjects carrying new information in a language like Italian.
According to this analysis, the post-verbal subject in the answer in (9b), fills the

5. Leaving the relative predicate unpronounced is less felicitous in this case. I leave
open here the issue as to what extent exactly the relative predicate of a cleft can be
left unpronounced. An issue which clearly deserves attention in future research.

96

Adriana Belletti

dedicated low Focus position and it is thereby interpreted as the new information
asked for in the question in (9a). This assumed analysis is schematized in (9c,d):6
(9)

a.
b.

Chi ha parlato?
Who spoke?
Ha parlato Gianni.
has spoken Gianni

c.

[CP . . . [ TP . . . [TopP . . . [FocP [TopP . . . vP]]]]]

d.

[CP . . . [ TP pro . . . ha parlato. . . [FocP Gianni . . . [vP . . . ]]]]

Verb (or part of the verb phrase) movement, yields the verb-subject order, with
the subject linearly appearing post-verbally.
Going back to the interpretive issue under discussion, with the clefted
constituent (possibly) new information focus in subject clefts vs contrastive/
corrective focus in object clefts, a crucial idea is that the Focus position utilized
is different in the two cases. It is the vP peripheral low Focus position in the
matrix clause vP periphery of the copula, in the case of new information subject
clefts; it is the high left peripheral position in the reduced CP complement of the
copula, in the case of contrastive/corrective object clefts and also in subject
clefts when they are interpreted/used contrastively/correctively, a possibility
seen in (8). In the former case, the clefted subject is interpreted as focus of new
information, and, in a language like French in which clefts are typically used in
answers, the structure can be used as an answer to a question on the identification
of the subject; in the latter case, the clefted object is interpreted as contrastive/
corrective focus, as is typically the case for left peripheral focalization, and a
cleft cannot be used to answer a question concerning the identification of the

6. For a more detailed discussion of the various implications of this analysis in the
cartographic perspective as well as of the existence of different answering strategies
to the same questions across languages and their acquisition in L1 and L2, see the
references of my previous work quoted. As is also discussed there, the vP periphery
also containsTopic positions illustrated in (9c) in the text, along parallel lines as the CP
clause external left periphery. A post-verbal subject with a Topic/given interpretation,
typically associated with a downgrading intonation, should fill this low position,
accordingly. I will not elaborate on this any further here, as it would take the discussion
too far apart and continue to only indicate the relevant vP peripheral Focus position
in the derivations. The reader is again referred to the reference quoted (in particular
Belletti 2004) for detailed elaboration on this point.

Revisiting the CP of Clefts

97

object.7 Hence, coherently with cartographic guidelines, in new information


subject clefts, the subject is interpreted in exactly the same position as the new
information post-verbal subject in a null-subject language like Italian.8 Similarly, a contrastive/corrective focus object is interpreted in the left peripheral
position, as contrastive/corrective focalized constituents typically are. Thus, a
direct reason is provided as to why object clefts cannot be used as answers to
questions for the identification of the object, even in a language of the French
type, which widely adopts clefts as answering strategies: the new information
focus position is the low focus position, where only the subject can be located,
not the high left peripheral one, where the object (or a contrastive/corrective
subject) is found.

7. For (classical) references on (syntactic) focalization, see Chomsky (1977); Kiss


(1998) and Rizzi (1997); Cruschina (2006, 2010); Bocci (2004); Belletti (2004) in
the frame of the discussion differentiating low and high focus.
In the analysis of contrastive/corrective clefts assumed, these clefts share a crucial
similarity with simple left peripheral focalization of the type in i., as the same focus
position is utilized in both cases:
i.

MARIA ho salutato (non Gianni).


Maria I have greeted (not Gianni).

However, the semantics of the two structures cannot be completely assimilated. The
contrastive/corrective focalization of clefts is more constrained than simple contrastive/corrective focalization as in i: a quantified expression cannot be focalized
through a contrastive/corrective cleft, whereas it can be contrastively/correctively
focalized in a simple sentence:
ii.

a.
b.

*Non e` NESSUNO / e` TUTTI che ho incontrato.


It is NOBODY/EVERYBODY that/whom I met.
NESSUNO /TUTTI ho incontrato.
NOBODY/EVERYBODY I met.

Thanks to P. Beninc`a and to E. Zimmermann for pointing out this type of contrasts,
which seems to indicate the non quantificational status of the Focus position involved
in clefts. The cartographic consequences of this conclusion will not be developed
here. For ease of exposition, I am going to continue to assume that the Focus position
involved is the same in both simple focalization and (object) clefts, and that the interpretive limitations shown in clefts should ultimately be derived from their overall
semantics, only partly linked to the type of focalization that clefts express.
8. On the possible correlation between new information subject clefts and the nonnull subject nature of the language, see Belletti (2008, 2009), also partly based on
experimental results from Brazilian Portuguese reported in Guesser (2007).

98

Adriana Belletti

The proposed analyses are given in the schematic derivations in (10).


(10)

a.

Subject clefts (new information)

[TP Ce . . . T [ FocP [vP etre [CP Force [CP . . . [ EPP Jean [FinP qui[ a parle]]]]]]]] 9
b.

Object clefts (correction/contrast)


. . . be [CP Force . . . [FocP . . . [FinP che [TP S . . . . . . O(/PP)]]]]. . .

If this type of analysis provides a straightforward characterization of the detected


different interpretive possibilities of subject and object clefts, the obvious question to ask is: why is it so? Why is it that only subject clefts allow for an analysis
as in (10a)? In a more explicit and formal way, the question amounts to asking:
if the copula can take as its complement a small clause as in (10a), with the
subject moved to the EPP position in CP and then moving further to the vP
peripheral new information focus position of the copula in the matrix sentence,
why couldnt exactly the same derivation also be at work in the case of object
clefts, with the object moving to the EPP position and then to the matrix vP peripheral focus position? The following section is devoted to provide an answer
to this fundamental question.
2.3. A Locality Explanation
There is indeed a direct principled reason as to why only subject clefts can
undergo the derivation in (10a). From the perspective of the interpretation then,
there is a principled reason as to why only the subject can be the focus of
new information in the vP periphery of the matrix copula, as in (10a). This
reason is the locality of syntactic derivations, expressed through the operation
of Relativized Minimality (RM, Rizzi 1990, 2004). Ultimately, the account is
in terms of intervention, along the following lines.
Suppose that in the CP small clause the object were to move from its merge
position within the relative predicate into the EPP position. As illustrated in
(11), the subject would be crossed over, giving raise to a straight violation of
RM, as both the subject and the EPP position, as well as the object, are positions
of the same type with regard to the principle; simply put, in terms of the A/A
9. I explicitly assume that the EPP position within the CP is not a criterial position in the
sense assumed for the subject position of TP, the position labeled Subj in Rizzi and
Shlonsky (2007), following Cardinaletti (2004), from which movement is blocked
under the criterial freezing principle motivated in their work.

Revisiting the CP of Clefts

99

distinction, they may be assumed to be all A positions. Hence, EPP cannot be


satisfied by the object, as indicated in the schematic derivation in (11):

(11)

[TP . . . [. . . [FocP . . . [vP be [CP Force [CP EPP [FinP che [TP S . . . . . . O ]]]]]. . .
This in turn implies that the object cannot move to the vP peripheral focus
position, and consequently cannot be interpreted as the focus of new information.
Hence, an object cleft cannot function as a possible answer to a question of
information, but only a subject cleft can, as the contrasts in (8) have shown.
Related to the informational/discourse issue, one may ask whether it could
not be possible for the object to directly move to the vP peripheral focus position
in the matrix clause, in a structure where the CP is just a reduced CP with no
EPP feature, like the one in (10) b. The corresponding derivation schematically
illustrated in (12) is, however, also ruled out on locality grounds:
(12)

[ TP . . . . . . [ FocP [vP be [CP Force . . . [FocP . . . [FinP che [TP S . . . O. . .]]]. . .


Phase theory (Chomsky 2005) explicitly rules out the possibility of such a long
direct movment with no intermediate steps (within the CP), with the embedded
CP sent to spell out. However, no intermediate step is possible in this case as
the reduced CP complement of the copula does not contain an escape hatch
edge position which could be used to exit the CP. If the CP is reduced at the
level of FocP, as is likely to be the case (3.2 below), it would not contain a relevant edge position different from the criterial/interpretable Spec/FocP position,
from which movement is excluded in principle through any version of criterial
freezing (Rizzi 2006; Boskovic 2007). Furthermore, should the vP peripheral
Focus position be considered an A position, the long movement would again be
ruled out as an intervention effect induced by the embedded subject.
In conclusion, in object clefts the copula cannot select for a CP small clause
with an active EPP feature to be checked by the object for principled locality
reasons linked to intervention, nor can the object of the relative predicate contained within the reduced CP move directly to the vP peripheral new information
focus position in the matrix clause, again for ultimately locality reasons related
to phase theory and, possibly, intervention again. The result is that it is for principled reasons that the analysis in (10a) may only concern subject clefts, whereas
object clefts solely correspond to the an analysis along the lines in (10b). This

100

Adriana Belletti

has the desired consequences for the different discourse value of subject and
object clefts discussed.
In the next section new issues and some further consequences derivable from
the proposed account are presented.
3.

Some Further Theoretical and Descriptive Issues

3.1. On Extraction from the Reduced CP


Is extraction from the reduced CP complement of the copula always excluded
or are there other movements possible? For instance, could movement affect the
object in the relative predicate and move it into the matrix CP? The answer to
this question is straightforwardly no: as in the case just discussed in (12), direct
movement from inside the relative predicate of the cleft into the matrix clause
is excluded on locality grounds, no matter where the moved object lands, be it
inside the matrix clause (e.g. into the vP peripheral focus position as discussed
in relation to (12)) or in its left periphery.
The question, however, deserves special further attention, as different kind of
sentences apparently implicating movement of the object appear to be possible.
In sentences such as (13), movement of the object seems able to affect the object
from the clefted position, the left peripheral focus position of the reduced CP.
In the well-formed Italian sentences in (13), the clefted object seems to be able
to undergo wh-movement (13a) or focus movement (13b) into the left periphery
of the matrix clause, from the focus position of the cleft:
(13)

a.
b.

Chi e`
Who is
GIANNI
Gianni

che Maria ha salutato ?


that Maria has greeted
e` che Maria ha salutato
is that Maria has greeted

At first sight, a possible derivation of (13a,b) is the one given in (14), modulo
the different landing site of the moved DP in the two cases, which are identified
with a single position in (14) to make the representation easier to read:
(14)
[CP [FocP / Wh [TP . . . [vP e` [CP Force . . . [FocP chi/GIANNI. . . [ FinP che [TP S. . . ]]]. . .
The derivation in (14) is, however, highly problematic. If, as illustrated in (14),
movement can affect the peripheral focussed phrase from the reduced CP, as this
CP only contains interpretable criterial positions by hypothesis, such movement
should not be possible at all. Under the criterial freezing approach mentioned

Revisiting the CP of Clefts

101

above, a derivation like (14) should not be a viable option. I follow here Rizzis
(2010) recent proposal on the question. According to this proposal, in cases
like those in (13) movement does not actually affect the DP within the Focus
Phrase as in (14), precisely because this movement would be in clear violation
of criterial freezing. Movement to the matrix left periphery may rather target
here a larger constituent corresponding not just to the DP in Focus, but the
whole FocP. Moving a larger constituent than the one which satisfies a criterion
(here the Focus criterion in the embedded reduced CP) is compatible with the
freezing principle (Rizzi 2006, 2010 for detailed discussion). In order to make
this movement possible, Rizzi assumes that first the relative predicate of the CP
is extraposed, and then the whole remnant phrase containing the Focus Phrase,
is moved to the relevant position into the matrix CP. The relevant steps of the
assumed derivation are illustrated in (15):
(15)

[CP [FocP/Wh [TP . . . [vP e` [FocP chi/GIANNI . . . [FinP che [TP S. . . ]]]
Extraposition of relative predicate10
[CP [FocP/Wh [TP . . . [vP e` [FocP chi/GIANNI . . . <FinP>] [FinP che [TP S. . . ]

Movement of the remnant:


[CP [FocP/Wh [FocP chi/GIANNI . . . <FinP> ] [TP . . . [vP e` <FocP>] [FinP che [TP S. . . ]

In conclusion, movement is possible from the reduced CP of clefts; however,


movement may only apparently take place from the focus position; in compliance
with criterial freezing, it is the whole Focus Phrase containing the focussed
clefted object which moves. This answers the question raised at the beginning
of this section.
Before concluding this discussion, a further issue deserves some close attention, related to the analysis just proposed. A crucial step of this analysis is the
extraposition of the relative predicate of the cleft; it is thanks to this operation
that the subsequent remnant movement of the whole Focus Phrase can be performed, yielding the apparent effect of movement of just the focussed clefted
object. But why should extraposition occur? Rizzi (2010) shows that the extraposition process is possible in (object) clefts, and provides examples as those in
(16) in Italian to show this possibility. In the case of (16b), the adverb oggi is
interpolated between the clefted object and the relative predicate:
10. For the sake of clarity, the extraposition operation is indicated as rightward movement.
A different analysis in the spirit of Kayne (1994) can be formulated in terms of
leftward movement and subsequent movement of the remnant. Various possibilities
come to mind as to the possible landing site of the extraposed constituent, but they
would take the discussion too far afield and are left open for future research.

102
(16)

Adriana Belletti

a.
b.

E
(it) is
E
(it) is

Gianni
Gianni
Gianni,
Gianni,

che devo
that I have
oggi, che
today, that

incontrare oggi.
to meet
today
devo incontrare.
I have to meet
(Rizzi 2010: 7577)

The important question to ask is whether the extraposition of the relative predicate is just a option, as (16b) shows, so that it is made appeal to when the freezing
principle is at stake, or whether the process is somehow more general, possibly
inherently linked to the cleft computation. The latter possibility would be much
more interesting, as it would allow one to conclude that nothing special should
concern structures such as those in (14) where movement from the reduced CP
is performed, as far as extraposition is concerned. Converging evidence to the
effect that extraposition of the relative predicate of clefts is not an isolated phenomenon is interestingly indicated by the following contrasts in West Flemish,
brought to my attention by Liliane Haegeman (who has provided the examples):
(17)

a.

Het is Val`ere niet geweest die dat gezegd heeft.


it
is Val`ere not been
die that said
has
b. *Het is Val`ere die dat gezegd heeft niet geweest
c. *Het is Val`ere niet die dat gezegd heeft geweest

The ungrammaticality of both (17b,c) indicates that, irrespective of the position


of the negation, the relative clause predicate cannot be left in situ in the cleft (here
a subject cleft), but must be extraposed to a position which ends up following
the clause final past participle of the copula. It is tempting to interpret these West
Flemish data as a clear indication of the widespread occurrence of extraposition
in clefts, possibly a process which always occurs in these structures. I leave open
to further elaboration the question as to why this should be the case and what
information/discourse or purely formal factors tightly linked to clefts should
trigger the extraposition process.11

11. That extraposition of the relative predicate occurs in clefts is proposed on semantic
grounds in Percus (1996), see also Hedberg (2000). Relevant in this regard is also
the consideration of the nature of the subject of the matrix clause in clefts it in
English whose expletive nature has been recently put into question in various
respects in Reeve (2010); see also Moro (1997) for relevant converging reasoning
on French ce of French clefts. It is tempting to somehow relate, at least in part, the
necessity of extraposition to the presence of the quasi-expletive in clefts. A research
program that I will undertake in future work.

Revisiting the CP of Clefts

103

3.2. How Much Reduced is the CP?


In order to answer the question of the amount of reduction the CP of clefts
manifests, the discussion will be centered on object clefts, that it is to say on
contrastive/corrective clefts.12
As for the reduced CP of object clefts (and contrastive/corrective subject
clefts), one question to ask is whether the reduced CP contains an active Topic
position or not, given the assumed left peripheral map. Even if the CP is reduced
at the level of FocP, still the Topic position which is located below FocP should
be present in the structure. We start by considering the contrasts in (18), which
clearly indicate that clefting a Topic is not an available option. Not even if the
Topic is a contrastive Topic as in (18b) (in italics, for convenience). Only a
corrective Focus ((18a)) ca be expressed in this structure. Interestingly, in the
very same discourse/pragmatic context, a contrastive Topic (in the sense of
Bocci (2009), Beninc`a (2001), realized in a Clitic Left Dislocation structure) is
possible as shown in (18c), which although similar to (18b), crucially, does not
contain the copula:
(18)

A.
B.

Gianni ha comprato un tavolo e una sedia.


Gianni has bought a table and a chair.
a.
No, e` il TAVOLO che ha comprato; (la sedia lha avuta in
regalo).
No, it is the TABLE that he has bought (the chair, they
offered it to him).
b. *No, e` il tavolo che lha comprato, la sedia lha avuta in
regalo.
No, it is the table that he he has bought it, the chair, they
offered it to him.
c.
No, il tavolo lha comprato, la sedia lha avuta in regalo.
No, the table he has bought it, the chair, they offered it to
him.

Hence, despite the possible availability of the position in the structural space,
the Topic position in the reduced CP selected by the copula cannot host a clefted
constituent. This is coherent with the often observed fact that clefting strictly
12. The natural assumption is made that the CP of subject clefts has the same shape,
modulo presence of the EPP feature. As both subject and object clefts can be contrastive/corrective as discussed, all other things being equal, reference to contrastive/
corrective clefts makes the discussion wider in principle (than just considering new
information subject clefts only).

104

Adriana Belletti

correlates with focalization, as it ultimately is a form of focalization. In order


to express this crucial property of clefts, it can be proposed that the copula in
clefts selects the embedded Focus head, which is then active and needs to be
realized/checked; a constituent must then be attracted into its Spec. Hence, in
the reduced CP of the copula one constituent must be focalized. This is precisely
the clefted constituent.13
Once the selection requirement is fulfilled, one may wonder whether some
other constituent can realize the Topic position, which should be present below
the Focus head.A sentence like (19) indicates that this possibility can be realized:
(19)

EMARIA che il libro lha comprato (non Gianni).


It is MARIA that the book she has bought it (not Gianni).

Let us comment on (19) a bit further and consider first, the respective order of the
constituents in Focus and Topic, and second, the position of the complementizer
in the CP.
As far as the respective order of the Focalized and theTopicalized constituents
is concerned, (19) displays the order with Focus higher than Topic. The opposite
order is totally excluded, as indicated in (20):
(20)

*E Il libro, MARIA che lha comprato (non Gianni).


It is the book, MARIA that she has bought it (not Gianni).

The total exclusion of (20) strongly suggests that the Topic position above Focus
is not present at all, thus arguing for the hypothesis that the reduction of the CP
complement of the copula does not include positions higher than Focus. This
conclusion is coherent with the idea just discussed that the copula selects Focus.
As for the co-occurrence of the corrective/contrastive focus and the left
dislocated topic in the left periphery, (19) corresponds to a sentence like (21),
where the same corrective/contrastive focalization does not involve clefting but
takes place in a simple declarative:
(21)

MARIA il libro lha comprato (non Gianni).


MARIA the book she has bought it (not Gianni).

13. In new information subject clefts, in which the relevant focus head is the vP peripheral
one in the matrix clause, it can be proposed that the selection relation is re-established
by the movement of the copula into the matrix T, as the embedded CP does not contain
an active Focus head in this case, but rather the active head is the one with the EPP
feature.

Revisiting the CP of Clefts

105

Notice however, that in (21) the respective order of the corrective/contrastive


focalized phrase and the left dislocated topic phrase can be reversed as illustrated
by the perfect status of (22):
(22)

Il libro, MARIA lha comprato (non Gianni).


The book, MARIA has bought it (not Gianni).

The possibility of (22) contrasts with the complete ungrammaticality of (20).


The strong impossibility of the order Topic Focus in (20), can then indeed be
considered a direct consequence of the reduced space available in the reduced
CP of clefts. Thus, given the two Top positions surrounding the Focus head in a
full fledged left periphery, we can conclude that the reduced CP of clefts does
not contain the highest one above Focus. This in turn strongly suggests that the
CP is reduced at the level of the projection of the left peripheral Focus.14
Let us now turn to the second point raised. There is an obvious question
which needs to be addressed in connection with (19): if on the one hand, the
fact that the order Focus-Topic is instantiated in the CP of clefts entitles us to
assume that the CP is reduced at the level of the Foc layer, as suggested on the
basis of the contrast between (19) and (20), on the other hand, the position of
the complementizer is not the one expected in the well-formed (19). As che
has been analyzed here as the realization of Fin, all things being equal, the
complementizer should be the last element in the CP, linearly following both the
Focalized and the Topicalized constituents; it is however situated between Foc
and Top in (19). Although a fully structured answer to this descriptive question
would go beyond the aim of this article, the following considerations can be
offered to indicate a line of interpretation for this otherwise unexpected word
order. Building on the proposal in Belletti (2008), it is tempting to suggest that
the unexpected position of the complementizer in (19) follows from the fact
that the complementizer moves from the Fin head into the highest head in the
CP; as proposed, this movement is assumed to always affect complementizers
of this type.15 In a full fledged CP, the highest head position is the Force head.
14. Reduction can be expressed in terms of the truncation hypothesis first developed in
Rizzi (1993/94). In Belletti (2009: Chapter 11) I have proposed, following Kayne
(2005) and his extension of Rizzis truncation idea, that elements filling the highest
Spec position in the clause, the edge are left unpronounced. If the Focalized
element in the reduced CP of clefts fills indeed the highest Spec position, hence the
edge, one should conclude that the principle of unpronounceability could not affect
elements in Focus; this might follow from some version of full interpretation.
15. This movement should hold in languages where the same C element realizes both the
content of Fin and the content of Force. See the reference quoted for further details.

106

Adriana Belletti

In the reduced CP of clefts that we are investigating, it corresponds to the Focus


head. Hence, once moved into the focus head, the complementizer ends up in the
position between Focus and Topic. It can be speculated that movement from Fin
into the highest head is generally triggered by the necessity to check selection: in
a declarative full fledged CP, the selectional properties are generally expressed
by the Force head; in the CP of clefts they are expressed by the Focus head, as
the copula selects Focus, as proposed above.
Converging evidence that C always fills the highest head, thus locally expressing the selectional properties of the matrix verb, is provided by the following
paradigm in (23):
(23)

a.

Penso che il libro, MARIA, a Gianni non glielo abbia ancora dato
(non Francesca).
I think that the book MARIA, to Gianni has not yet given it to
him (not Francesca).
b. ?Penso il libro, che MARIA, a Gianni non glielo abbia ancora dato
(non Francesca).
I think the book that MARIA, to Gianni has not yet given it to
him (not Francesca).
c. *Penso il libro, MARIA che a Gianni non glielo abbia ancora dato
(non Francesca).
I think the book, MARIA that to Gianni has not yet given it to
him (not Francesca).
d. *Penso il libro, MARIA a Gianni che non glielo abbia ancora dato
(non Francesca).
I think the book, MARIA to Gianni that has not yet given it to
him (not Francesca).

In all the sentences in (23), the two Topic positions and the Focus position in
the left periphery of the complement of the verb pensare (think) are realized.
(23a) is a perfect sentence where the expected word order with che in the highest Force is realized. (23b) is possible, although slightly marginal, presumably
because the (declarative) force of the complement clause is realized in a non
canonical head (the Topic head). (23c), however, where the complementizer
follows both the high Topic and the contrastive/corrective peripheral Focus, is
not just marginal, it is plainly ungrammatical. (23d), where the complementizer
follows lowest Topic as well, presumably remaining in Fin, is also excluded. It is
tempting and reasonable to interpret this ungrammaticality as due to selection:
the complementizer is located too far from its selecting verb in (23c,d). If this
interpretation is on the right track, nothing special to clefts needs to be said on
the position of the complementizer in (19), unexpected at first sight, but actually

Revisiting the CP of Clefts

107

in line with the regular syntax of the complemenizer, in combination with the
selectional properties of the copula, and the reduced size of the CP.
3.3. The Similarity with Other Structures
3.3.1.The CP Complement of Perception Verbs: Pseudorelatives and Clefts
The shape of the reduced CP complement attributed here to subject clefts finds
a close analogue with the one of pseudorelatives, the sentential complement
of perception verbs (Belletti 2008, 2009). Here, I will just review the main
features of the similarity shared by the two structures and then point out some
interpretive constraints which follow directly from the general account proposed.
The interpretive constraints manifest themselves when new information subject
clefts and pseudorelatives are combined in a Question-Answer exchange.
The analysis proposed for pseudorelatives, essentially updates the analysis
originally proposed by Guasti (1993), under the articulated conception of the CP
domain adopted here. According to this analysis, a sentence like (24a) contains
a pseudorelative complement of the perception verb vedere (see), which is a
reduced CP with an EPP feature, the CP small clause illustrated in (24b):
(24)

a.
b.

Ho visto Maria che parlava con Gianni.


I have seen Maria that spoke to Gianni.
Ho visto [CP . . . [EPP Maria [FinP che [TP (pro) parlava () con Gianni]]]].

For concreteness, in (24) I assume that Maria is moved to the EPP position in
the CP from inside the TP of the predicate, with movement taking place from
the postverbal position (as is always the case in Italian, see Rizzi and Shlonsky
2007 and references cited therein), indicated with in (24). As pointed out
in the references quoted, a sentence like (24a) can constitute an answer to both
question (25a) and question (25b) below; this is so because, differently from
subject clefts, pseudorelatives do not necessarily require the focalization of the
subject argument, as the focus of new information can also be constituted by
the whole clause in pseudorelatives. (25c), illustrates the interpretation of the
pseudorelative as an answer to (25a). It is clear from (25c), the close similarity
of this type of pseudorelative with subject clefts:
(25)

a.
b.
c.

Chi hai visto (che parlava con Gianni)?


Whom have you seen (that spoke to Gianni).
Che cosa hai visto?
What have you seen?
Ho visto [Foc Maria [vP [CP . . . [EPP <Maria>[FinP che [TP (pro) parlava () con Gianni]]]]]]

108

Adriana Belletti

As discussed in Guasti (1993) and Rizzi (2000), in the pseudorelative in (25c)


there is direct perception of Maria. A cleft, on the other side, implies a peculiar semantics which provides a unique identification explicitly expressed by
the focussed argument. In subject clefts, the (uniquely) identified argument is
the subject. One may now ask: can a subject cleft constitute an answer to a
question containing a pseudorelative? In looking for an answer to this question,
the language to look at is, once again, French, where a cleft constitutes a privileged way to answer a question on the subject, as discussed in 2.1. The relevant
question-answer pair is given in (26):
(26)

Q: Qui as-tu vu qui pleurait?


Whom have you seen who was crying?
A: *Cest Marie (que jai vu qui pleurait)
It is Marie (whom/that I saw who was crying)

The answer in (26) is not possible, with the indicated meaning16. This is expected
under the proposed analysis, as the subject cleft answer implies the vP peripheral
focalization in the matrix clause containing the copula, which in turn inevitably
yields the violation of Relativized Minimality illustrated in the derivation in
(27), due to the presence of the intervening subject of the perception verb (je,
in the example).
*
(27)

[TP . . . [FocP [vP be[C [EPP Maria] [FinP que[TP jai vu [CP [EPP <Marie>]
[FinP qui [TP . . . pleurait. . . ]]]]]]]]]

In contrast, in a context different from the question-answer one, a sentence like


(28), where a contrastive subject cleft is present, is perfectly well-formed. This
is also expected, as focalization is here contrastive left peripheral focalization,
so that intervention by the subject of the perception verb does not matter.
(28)

E MARIA che lui ha visto che piangeva.


(It) is MARIA that he has seen that was crying.
[TP . . . [vP be[CP [FOC MARIA] [FinP che [TP lui ha visto [CP [EPP < Maria>
[FinP che [TP piangeva. . . ]]]]]]]]]

16. Thanks to Dominique Sportiche for providing the judgment and also for clarifying
his intuition in the following terms: (26) is not just the answer to the question asked; a
different prosody would make the sentence possible, but this would mean changing the
presuppositions, which is not the situation the question-answer pair in (26) refers to.

Revisiting the CP of Clefts

109

3.3.2.A Possible Different Fin: -ing


Let us now ask whether there is any analogue of the small clause CP of (subject)
clefts, in which an element different from a complementizer is present under
Fin. I submit here the proposal that a possible candidate be the -ing gerundival
ending found in the complement of perception verbs in English examples like
those in (29) (thanks to D. Pesetsky for pointing out the possible connection):
(29)

I saw John working with his sister.

Interestingly, a -ing clause is also possible as a subject cleft of the type illustrated
in (30). A cleft like (30) can possibly function as a new information subject cleft:
(30)

(Who was making this noise?)


It was John working with his sister.

Once again the complement of a perception verb and subject clefts appear to be
strictly related.
If the proposal is on the right track, the derivation of the -ing cleft in (30)
should involve both V-to-T-to-C/Fin and the familiar movement of the subject
into the EPP position of the CP small clause, and, from there, into the vP peripheral Focus position of the matrix copula (plus the movement of the copula into
the matrix T), as in the derivation 1 in (31) below. Alternatively, in the case of a
reduced CP with no EPP feature, the sentence should involve direct focalization
in the reduced left periphery, the derivation 2 in (31)17 :
(31)

1
[. . . [ FocP . . . [vP be [CP FOC // EPP [FinP -ing [ TP John T [work . . . ]]]]]]]
2

4.

Conclusion

This article has focussed on two main points: the investigation of the way in
which the locality of syntactic computations conditions the syntax and the interpretation of clefts; and the way in which the CP domain is shaped in clefts.
17. A possible third alternative analysis exists, in which the whole -ing clause moves into
the new information focus position. This would correspond to a reading in which the
whole -ing clause, not just the subject is the focus of new information, a possibility
available for this type of clefts, similarly to pseudorelatives, as noted in the discussion
surrounding (25).

110

Adriana Belletti

The two aspects have been shown to correlate closely, as the locality ban against
intervention according to Relativized Minimality, derives the different interpretations of subject and object clefts, under the crucial assumption that the CP
space, syntactically analyzed in cartographic terms, is reduced in clefts. The
possibly different discourse value of clefts is thus ultimately a consequence of
the operation of the general locality principle and of the specific shape of the
syntactic structure of clefts.
The shape of the CP space of clefts is not isolated, other structures such as the
sentential complement of perception verbs share the same type of configuration;
furthermore, in both the complement of perception verbs and in clefts the Fin
head can host different type of material beside the complementizer; we have
speculated that the -ing ending be a possible realization of this head which closes
up the CP domain. Potentially problematic derivations have also been discussed,
among which a prominent one is the apparent possibility of extraction out of
the interpretable criterial positions of the CP of clefts, for which an account
making reference to extraposition of the cleft sentence has been suggested.
Rather than being just an option made use of in the special extraction conditions,
the hypothesis has been entertained that the extraposition process may turn out to
be inherently linked to clefts. This hypothesis, which has also sometimes been
proposed on semantic grounds, will be further closely investigated in future
work.
References
Abels, Klaus and Muriungi, Peter
2005
The Focus particle in Ktharaka: Syntax and Semantics. Lingua 118:
687731.
Belletti, Adriana
2009
Structures and Strategies (Answering strategies: New information
subjects and the nature of clefts, Chapter 10). New York: Routledge.
Belletti, Adriana
2008
The CP of Clefts. Rivista di Grammatica Generativa 33: 191204.
Belletti, Adriana
2004
Aspects of the low IP area. In: Luigi Rizzi (ed.), The Structure of
CP and IP. The Cartography of Syntactic Structures, Vol. 2, 1651.
Oxford: Oxford University Press.
Belletti, Adriana
2004 ed.
Structures and Beyond. The Cartography of Syntactic Structures,
Vol. 3. New York: Oxford University Press.

Revisiting the CP of Clefts


Beninc`a, Paola
2001

111

Syntactic Focus and Intonational Focus in the left periphery. In: Guglielmo Cinque and Giampaolo Salvi (eds.), Current Studies in Italian
Syntax Offered to Lorenzo Renzi, 3964. Amsterdam: Elsevier.
Beninc`a, Paola and Poletto, Cecilia
2004
Topic, Focus and V2: Defining the CP Sublayers. In: Luigi Rizzi (ed.),
The Structure of CP and IP. The Cartography of Syntactic Structures,
Vol. 2, 5275. New York: Oxford University Press.
Bocci, Giuliano
2004
Contrastive focalization on topics and preverbal subjects in Italian.
Rivista di Grammatica Generativa 29: 359.
Bocci, Giuliano
2009
On Syntax and Prosody in Italian. PhD dissertation. University of
Siena.
Bianchi, Valentina and Mara Frascarelli
2009
Is Topic a root phenomenon. Iberia.

Boskovic, Zeljko
2007
On Successive Cyclic Movement and the Freezing Effect of Feature Checking. In: Jutta Hartmann, Veronika Hegedus and Henk van
Riemsdjik (eds.), The Sounds of Silence. North Holland: Elsevier.
Burzio, Luigi
1986
Italian Syntax. A Government-Binding Approach. Dordrecht: Reidel.
Cardinaletti, Anna
2004
Towards a Cartography of Subject Positions. In: Luigi Rizzi (ed.),
The Structure of IP and CP. The Cartography of Syntactic Structures,
Vol. 2, 115165. New York: Oxford University Press.
Chomsky, Noam
1977
On Wh-Movement. In: Adrian Akmajian, Peter Culicover and Thomas
Wasow (eds.), Formal Syntax, 71132. Academic Press.
Chomsky, Noam
1995
The Minimalist Program. Cambridge Mass: MIT Press.
Chomsky, Noam
2005
On Phases. In: Robert Freidin, Carlos Peregrn Otero and Maria Luisa
Zubizarreta (eds.), Fundational Issues in Linguistic Theory. Cambridge, Mass: MIT Press.
Cinque, Guglielmo
2002 ed.
Functional Structure in DP and IP. The Cartography of Syntactic
Structures, Vol. 1, New York: Oxford University Press.
Cinque, Guglielmo and Luigi Rizzi
2010
In: Bernd Heine and Heiko Narrog (eds.), The Cartography of Syntactic Structures. The Oxford Handbook of Linguistic Analysis, 5165.
Oxford/New York: Oxford University Press.

112

Adriana Belletti

Clech-Darbon, Anne, Georges Rebuschi and Annie Rialland


1999
Are there cleft sentences in French? In: Georges Rebuschi and Laurice
Tuller (eds.), The Grammar of Focus. Amsterdam/Philadelphia: John
Benjamins.
Cruschina, Silvio
2006
Informational focus in Sicilian and the left periphery. In: Mara Frascarelli (ed.), Phases of Interpretation, 363385. Berlin: Mouton de
Gruyter.
Cruschina, Silvio
2010
Fronting as Focalisation in Sicilian. In: DAlessandro Roberta, Ledgeway Adam and Ian Roberts (eds.), Syntactic Variation. The Dialects
of Italy, 247260. Cambridge: Cambridge University Press.
Grewendorf, Gunther
2005
The discourse configurationality of scrambling. In: Joachim Sabel
and Mamoru Saito (eds.), The Free Word Order Phenomena, 75136.
Berlin: Mouton de Gruyter.
Guasti, Maria Teresa
1993
Causative and Perception Verbs. A Comparative Study. Torino: Rosenberg and Sellier.
Guesser, S.
2007
Soggetto nullo e focalizzazione del soggetto in Portoghese Brasiliano.
Master thesis; University of Siena.
Haegeman, Liliane
1996
Verb second, the split CP and null subjects in Dutch finite clauses.
GenGenP 4.2: 133175.
Haegeman, Liliane
2004
Topicalization, CLLD and the Left Periphery. In: Benjamin Shaer,
Werner Frey and Claudia Maienborn (eds.), Proceedings of the Dislocated Elements Workshop ZAS Berlin, November 2003, 158192.
Berlin: ZAS.
Hedberg, N.
2000
The referential status of clefts. Language 76(4): 891920.
Kayne, Richard
1994
The Antisymmetry of Syntax. MIT Press, Cambridge Mass.
Kayne, Richard
2005
On Parameters and on Principles of Pronounciation. In: Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster (eds.), Organizing Grammar. Linguistic Studies in Honor of Henk
van Riemsdijk, 289299. Berlin: Mouton De Gruyter.

Kiss, Katalin E.
1998
Identificational Focus versus Information Focus. Language 74(2):
245273.

Revisiting the CP of Clefts


Mioto, Carlos
2003
Moro, Andrea
1997
Percus, Orin
1997
Reeve, M.
2010
Rizzi, Luigi
1990
Rizzi, Luigi
1993/94
Rizzi, Luigi
1997
Rizzi, Luigi
2000

Rizzi, Luigi
2004

Rizzi, Luigi
2004 ed.
Rizzi, Luigi
2006

Rizzi, Luigi
2010

113

Focalizaca o e quantificaca o. Revista Letras (Curtiba) 61: 169189.


The Raising of Predicates. Cambridge: Cambridge University Press.
Prying open the cleft. In: Kiyomi Kusumoto (ed.), Proceedings of
NELS 27, 337351. Amherst, MA: GLSA.
The syntactic structure of English clefts. Lingua, doi: 10.1016/j.
Relativized Minimality. Cambridge, MA: MIT Press.
Some notes on linguistic theory and language development: The case
of root infinitives. Language Acquisition 3: 371393.
The Fine Structure of the Left Periphery. In: Liliane Haegeman (ed.),
Elements of Grammar, 281337. Dordrecht: Kluwer.
Direct perception, government and thematic sharing. In: Luigi Rizzi
Comparative Syntax and Language Acquisition, 189210. London/
New York: Routledge.
Locality and the Left Periphery. In: Adriana Belletti (ed.), Structures
and Beyond. The Cartography of Syntactic Structures, Vol. 3. New
York: Oxford University Press.
The Structure of CP and IP. The Cartography of Syntactic Structures,
Vol. 2. New York: Oxford University Press.
On the Form of Chains: Criterial Positions and ECP Fffects. In: Lisa
Lai-Shen Cheng and Norbert Corver (eds.), Wh-Movement: Moving
on, 97134. MIT Press.

Some consequences of Criterial Freezing. In: Peter Svenonius (ed.),


Functional Structure from Top to Toe. New York: Oxford University
Press.
Rizzi, Luigi and Ur Shlonsky
2007
Strategies of Subject Extraction. In: Uli Sauerland and Hans-Martin
Gartner (eds). Interfaces + Recursion = Language?, 115160. Mouton De Gruyter.
Rothstein, Susan
2000
Predicates and Their Subjects. Kluwer.

114

Adriana Belletti

Ruwet, Nicolas
1975
Starke, Michal
1995

Stowell, Tim
1983

Les phrases copulatives en francais. Recherches Linguistiques 3: 143


191.
On the Format of Small Clauses. In: Anna Cardinaletti and Maria
Teresa Guasti (eds.), Small Clauses, Syntax and Semantics, Vol. 28,
237269. New York: Academic Press.
Subjects across categories. The Linguistic Review 2.3: 285312.

Delimitation Effects and the Cartography


of the Left Periphery
Luigi Rizzi

1.

Introduction

Half a century of formal syntactic studies has brought to light the complexity
and richness of syntactic structures. The cartography of syntactic structures is
the line of research which addresses this topic: it is the attempt to draw maps
as precise and detailed as possible of syntactic configurations. The cartographic
projects started, over a decade ago, as an attempt to provide fine descriptions of
certain zones of the syntactic tree in some Romance and Germanic languages,
but they immediately showed a universal dimension, and were quickly extended
to many other language families.
In this paper I would like to illustrate some of the results of the cartographic
studies in connection with the left periphery of the clause, and discuss the
implications of this line of research for the study of the interfaces connecting
syntax with the systems of sound and meaning. In the last part I will show how
cartographic maps of the left periphery interact with classical topics of syntactic
research such as the theory of locality and the freezing effects.
2.

The Criterial Approach to Scope-Discourse Semantics

Economy conditions on movement in the Minimalist Program limit the application of movement to configurations in which it yields some effect on the
interpretive interface (Reinhart 2005). This is rather straightforward for one important class of cases of movement, displacing elements to the initial periphery
of the clauses, and yielding A -movement chains:
Consider some typical A -constructions in English, such as the following
clauses:
(1)

a.
b.

Which book should you read ?


The book which you should read

is here.

116

Luigi Rizzi

c.
d.

This book, you should read .


(It is) THIS BOOK (that) you should read

(not Bills book).

It is intuitively clear what movement to the front does in these cases. In (2a) it puts
an interrogative operator, the wh-phrase, in its proper scope position to yield the
appropriate logical form for which x, x a book, [you should read x]; similarly,
the moved element is interpreted as a relative operator (the book x such that [you
should read x]) in (2b), and analogous operator-variable structures are assigned
to exclamatives, comparatives and other constructions involving the left periphery of the clause. So, in these cases the relevant syntactic element which book,
etc. receives two interpretive properties: the argumental thematic role of patient
of read, and the role of interrogative operator taking scope over the whole clause.
Cases (2c) and (2d) are slightly different: here the preposed element expresses
the discourse-related property of being the Topic, or the (contrastive) Focus
of the structure, respectively, properties relevant for the information packaging of the expression (Vallduv 1992), and its usability in discourse. Putting
together all these cases under a synthetic label, I will follow Chomsky (2004)
and call the interpretive properties associated to the initial position properties of
scope-discourse semantics: the scope of operators and the discourse-related
properties expressing the informational articulation of the structure. So, these
cases of movement connect two positions: one dedicated to argumental semantics, the thematic role, or, more generally, the position in which an element is
semantically selected (or S-selected: for instance, a time adverbial is S-selected
by a T head, an aspectual adverbial is S-selected by the appropriate Asp head,
etc. in an approach to adverb syntax like the one developed by Cinque 1999);
and one dedicate to scope-discourse semantics.
What does it mean for a position to be dedicated to a certain interpretation?
In the case of argumental semantics this is straightforward and uncontroversial:
argumental roles are assigned by certain lexical heads, typically verbs, to their
immediate dependents. So, the verb read assigns the role patient to its complement and the role agent to its specifier (perhaps through the mediation of
a light verb v). As for scope-discourse semantics, I would like to assume an
approach which generalizes the same basic mechanism to it: there is a dedicated system of functional heads, typically in the left periphery of the clause,
which assigns to its dependents such properties as scope position and scope
domain of such and such type of operator, Topic and Comment, Focus
and Presupposition, etc. So the assignment of both kinds of interpretive properties is uniformly a matter of head-dependent relations. This structural view
of scope, topicality and focus is sometimes called the criterial approach to
scope-discourse semantics:

Delimitation Effects and the Cartography of the Left Periphery

(2)

117

The criterial approach: there is a system of dedicated functional heads,


typically expressed in the initial periphery of the clause, attracting elements to their Specs and signalling to the interface components the
basic scope-discourse properties.
(Rizzi 1991, 1997)

This amounts to saying that sentences (1) should have representations like the
following, with heads such as Q, Top, etc., which attract the wh-operator, the
Topic, etc., to their specifier (in these representations I have expressed the gap
left by movement as a full unpronounced copy of the moved phrase, as in the
copy theory of traces):
(3)

a.
b.
c.
d.

Which book
The book which
This book
THIS BOOK

Q
R
Top
Foc

should you read <which book> ?


you should read <the book>
you should read <this book>
you should read <this book> not Bills book

In English these heads are silent, they do not correspond to any overt morpheme
(except for the fact that Q in main questions attracts the functional verb, in a
case like (3a) the modal, so that the position ends up being filled in this case).
There is straightforward comparative evidence which makes the hypothesis
expressed by (2)(3) immediately plausible: in many languages this system of
left peripheral heads is pronounced, expressed by overt particles:
(4)

a.

b.

(5)

a.

b.

Ik weet niet [wie of [Jan


gezien heeft]].
I know not who Q Jan
seen
has
(Dutch varieties, Haegeman 1996)
gfundn hot]].
Der Mantl [den wo [dea Hons
found has
The coat which R the Hans
(Bavarian, Bayer 1984)
` s`e [
Un
d`o [ d`an
I
heard that snake
(Gungbe, Aboh 2004)
` s`e
Un
[d`o [d`an
I
heard that snake
(Gungbe, Aboh 2004)

lo y`a
[K`of h`u
`]]].
the TOP Kofi killed it
lo w`e [K`of h`u ]]].
the FOC Kofi killed

(4a) shows that the Q head is overtly expressed as of (if) in certain (non-standard)
varieties of Dutch. Relative clauses are marked in Bavarian by the special complementizer wo, homophonous to the wh-element for where, and which can
co-occur with the relative operator (the D-pronoun den in (4b) its Spec.The

118

Luigi Rizzi

topic and focus heads are expressed by the particles y`a and w`e , respectively, in
Gungbe, as shown in (5).
The A criteria were originally formulated as representational principles in
terms of the GB framework: a phrase endowed with the appropriate A feature
must be in a Spec-head configuration with a left-peripheral head endowed with
the same feature; the configuration must be met at LF uniformly across languages
as it is crucial for the interpretation of such structures; particular languages may
require its satisfaction already at S-structure, a matter of parametric variation;
in these cases, movement to the left periphery will have to be overt (Rizzi 1991).
In derivational terms more congenial to the minimalist conception of syntactic computations, the syntactic role of the criteria can be recast as the capacity
that criterial heads have of attracting phrases endowed with matching criterial
features (see Aboh 2007 for an implementation along these lines):
(6)

XF attracts XPF to its Spec, for F = Q, R, Top, Foc, Excl, . . .

I leave open here the question of whether attraction involves a system of uninterpretable features, as in the detailed implementation proposed by Pesetsky
and Torrego (2001) (see Rizzi 2006, 2009 for a discussion of the possible role of
purely formal, uninterpretable counterparts of criterial features in an approach
to A syntax based on criteria). I will also leave open here the issue of the in
situ strategy for wh and other A constructions, which may involve covert movement or the establishment of a pure Agree relation from the criterial head to the
relevant criterial phrase.
In addition to their strict syntactic role of triggering movement, criterial heads
have another and equally important function: they are visible to the interface
systems of sound and meaning to signal in a transparent manner the interpretive
properties of scope-discourse configurations. So, for instance, the criterial head
Top goes with an interpretive instruction of the following kind:
(7)

[XP [Top
YP
]]
Topic
Comment

I.e., my Spec is to be interpreted as the topic, and my complement as the comment made about it, where a topic expresses a referent selected among those
familiar from context, and about which a comment is made (with the possibility of further subspecifications, characterizing a possible family of topics:
Beninc`a and Poletto 2004; Frascarelli and Hinterhoelzl 2007; Bianchi and Frascarelli 2009; see Rizzi 2010b for a more detailed discussion of the interpretive
properties of Topics). Similarly for the Foc head:

Delimitation Effects and the Cartography of the Left Periphery

(8)

119

[XP [ Foc
YP
]]
Focus
Presupposition

I.e., my Spec is to be interpreted as focus and my complement is the presupposition, where focus means that XP provides the value of a variable in the
presupposition, a value that is assumed to be new information for the interlocutor. Here too, there can be additional specifications which define different
types of foci, for instance contrastive focus as opposed to simple new information focus. E.g., when I utter (1d), I assume not only that you know that you
should read something (presupposition), and I convey the information that the
thing you should read is this book (new information), but I also assume that this
new information falls outside what I take to be your natural expectations, with
which the new information I am providing is contrasted (so, this contrastive
focal structure is very naturally concluded by a negative tag explicitly excluding the information imputed to the interlocutors system of beliefs, i.e., in the
case of (1d), Bills book). Clearly, languages assign distinct focal properties to
different positions, in ways that are parametrised in part. In Romance languages
like Standard Italian, the clause-initial focal position is dedicated to contrastive
focus, while in the Sicilian dialects and in the regional varieties of Italian spoken
in Sicily the initial focus position can be used for simple new information focus
(Cruschina 2008), hence be used to answer a simple wh-question.
Criterial heads also guide the interpretation of the sentence at the interface
with the sound system, much as they guide the interpretation at the interface
with semantics and pragmatics. Consider for instance the careful experimental
study of the pitch contour of Italian Topic and Focus constructions conducted by
Bocci (2009). The following figures (Giuliano Boccis courtesy) illustrate the
two contours:
(9)

Pitch contour of Topic Comment (from Bocci 2009)


A Michelangelo (Top), Germanico vorrebbe presentare Pierangela.)
To Michelangelo (Top), Germanico would want to introduce Pierangela.

450
400

Pitch (Hz)

300
200
100
H+
a mi he

+L*
lan

L-L%

A Michelangelo
0

H+ +L*

de lo der

ma

H+ +L*
ni ho vo

Germanico
Time (s)

H+

r be pre zen

vorrebbe

+L*
ta

presentare

H+ +L*
re pje

L-L%

ran de la

Pierangela
3.7942

120

Luigi Rizzi

(10)

Pitch contour of Focus Presupposition (from Bocci 2009)


A MICHELANGELO (Foc) Germanico vorrebbe presentare Pierangela.
TO MICHELANGELO (Foc) Germanico would want to introduce
Pierangela.

550
500
400

Pitch (Hz)

300
200
100
L+ +H*
a

mi

he

lan

Lde lo

A MICHELANGELO
0

L*
der

ma

L*
ni ho

Germanico

vo rb be pre

vorrebbe
Time (s)

zen

ta

presentare

L*
re pje

ran

L-L%
de la

Pierangela
3.61823

The pitch contours of Topic and Focus are distinct (with contrastive Focus receiving a higher prominence than Topic, Bocci 2009), and the contours of Comment
and Presupposition are sharply different, with a complete flattening of the contour of the presupposition in (9), which contrasts with the highly articulated
contour of the comment in (8). Again, we can think of the criterial heads as
giving instructions to the prosodic component to yield the different types of
contours associated with the different discourse functions, much as they do in
the interpretive components on the meaning side, as in the model that Bocci
(2009) adopts and develops.
In conclusion, criterial heads have a syntactic function, attracting elements
to the periphery of the clause, and interpretive functions on both interfaces with
sound and meaning, triggering certain semantic-pragmatic routines expressing
the informational organization of the clause, and the assignment of specific pitch
contours which will make the relevant interpretive properties salient and immediately detectable from the speech signal (see also the discussion in Frascarelli
2000).
This way of looking at the expression of scope-discourse properties has
several advantages over imaginable alternatives.
On the one hand, rather than assuming a proliferation of different devices for
the expression of scope-discourse semantics, it assumes a uniform mechanism
to hold across languages: scope-discourse properties are expressed by a system
of dedicated functional heads acting upon their immediate dependents, much as
thematic properties are expressed by a system of lexical heads (or low functional
heads of the v type). An elementary but salient parametric property here has to
do with whether such heads are overt, expressed by pronounced morphemes,

Delimitation Effects and the Cartography of the Left Periphery

121

or not (essentially the same kind of elementary spell-out parametrisation which


has been assumed for Case systems ever since the Chomsky-Vergnaud theory
of syntactic Case in GB); there are of course other parameters of familiar kinds
(possibly, overt or covert movement to the Specs of criterial heads, certain
permissible variations in the hierarchical order, etc.), but the fundamental computational mechanisms and configurations dealing with the syntax-pragmatic
interface can be assumed to be essentially uniform across languages.
Moreover, movement to a Topic or Focus position is like other kinds of movement, internal Merge triggered by a search activated by certain morphosyntactic
features, like object movement in passive, wh-movement, clitic movement and
the like. There is no need to postulate a separate, special category of prosodically driven movement, hence no reason to enlarge the class of computational
options admitted by Universal Grammar. And there is no need to postulate a direct link between intonational structure and pragmatics: the connection is fully
mediated by syntax, through the interface role of criterial heads. The criterial
approach makes things fully parallel for scope-discourse semantics and argumental semantics: in the latter case, it is generally assumed that the connection
between the thematic role that a nominal expression receives and the position in
which it is pronounced are mediated by syntax, without the need of establishing
any additional, syntax-independent, link between interpretive roles at LF and
positions of pronunciation at PF. So, in terms of the criterial approach we can
stick to the simplest assumption for both argumental and scope-discourse semantics: representations of sounds and meanings are fully mediated by syntax,
no independent and extra-syntactic connecting path needs to be postulated.
3.

The Cartography of Syntactic Structures

In many languages, Topic and Focus can co-occur, often in a fixed order, illustrated by the following Gungbe example:
(11)

. . . d`o K`of y`a g`ankpa m`e w`e kp`on`on


le su- `
. . . that Kofi Top PRISON IN Foc policemen Pl shut him
do.
there
(Gungbe: Aboh 2004)

Hence, cartographic issues arise at this point: we want to know what global configurations the left periphery of the clause can assume, what properties remain
constant across languages and what other properties are submitted to parametric
variation.

122

Luigi Rizzi

The observation that dedicated scope-discourse positions can co-occur with


ordering restrictions was one of the triggering factors of the cartographic
projects. Around the mid-nineteeneighties the clausal structure was assumed
to result from the hierarchical organisation of three X-bar layers headed by the
verb, the inflection node, and the complementizer, respectively:
(12)

[CP . . . C . . . [IP . . . I . . . [VP . . . V . . . ]]]


(Chomsky 1986)

But it soon became very clear that this picture was oversimplified. Examples
like (11) show that more elements can co-occur in the left peripheral pre-subject
position of the clause; this fact per se may still be expressed with a simple
representation like (12) through the assumption that multiple adjunctions to the
IP are allowed, as in a traditional line of analysis of topicalisation. But this
assumption is at odds with various properties:
a. the rigidity of certain orders as in (11), while a multiple adjunction analysis
would typically predict free ordering;
b. the presence in many languages of overt Top and Foc heads, whose distributional properties must be deferred to the interface systems in an adjunction
analysis;
c. the uniqueness of Topic and Focus in cases like (11): again, under an adjunction analysis one would expect free recursion, unless some special constraint
is added to the adjunction mechanism, and/or to the interface systems.
Similar considerations of word order, co-occurrence of different kinds of elements (adverbials and other elements), and properties of the morpho-syntax
had already led to the splitting of the I node into more elementary components
(Pollock 1989, Belletti 1990). The special attention to such constraints led to the
conclusion that each layer of (12) is an abbreviation for a much richer structural
zone. The cartography of syntactic structures is the attempt to build detailed
maps of each structural zone.
If the structural zones may be complex, the syntactic atoms are remarkably
simple and uniform: the structures are built through successive applications of
Merge, through which a head is combined with a complement and a specifier,
thus projecting a phrase. If the fundamental geometry of the building block is
always the same, the rich articulation of the structures is due to the richness of
the inventory of functional heads: the functional heads are much more numerous
than one would have assumed twenty five years ago, and the possible functional
structures, resulting from the combinations of functional heads according to
their selectional properties, are correspondingly more articulated.

Delimitation Effects and the Cartography of the Left Periphery

123

So, if we take the magnifier and draw maps expressing the fine details of
structures, we find the same basic configurations and shapes that we observe
through a first-pass, naked eye survey. In this sense, syntactic structures are
like crystals, with the same shapes showing up at the macroscopic and microscopic levels.
Cartographic projects started with detailed descriptions of the IP and CP systems in some Romance and Germanic languages (Rizzi 1997, 2004b; Cinque
1999, 2002; Belletti 2004, 2009; Grewendorf 2002; Haegeman 2003, 2006), but
they quickly showed a general dimension, triggering much work on different languages and language families: Finno-Ugric (Puskas 2000), Semitic (Shlonsky
2000), Slavic (Krapova and Cinque 2004), West African (Aboh 2004), Bantu
(Biloa 2008), Creole (Durrleman 2008), East-Asian (Tsai 2007; Endo 2007;
Saito 2012), Dravidian (Jayaseelan 2008), Austronesian (Pearce 1999), Classical languages (Salvi 2005), etc.
4.

The Fine Structure of the Left Periphery

Going back to the CP system, the motivations for splitting it into a sequence of
functional heads were twofold.
First, different C-like particles occupy distinct syntactic positions, as is
shown by their ordering with respect to other elements: for instance, in my
variety of Italian the finite declarative complementizer che must precede topics, while the infinitival (prepositional) complementizer di necessarily follows
topics (che Top /Top di). This suggests that che and di occupy distinct positions
in the C space (Rizzi 1997), while both being part of the complementizer system (the complementizer status of di is well motivated for reasons discussed in
Kayne (1983), Rizzi (1982) and much subsequent work).
(13)

a. *Penso, a Gianni, che gli dovrei parlare.


I think, to Gianni, that I should speak to him.
b. Penso che, a Gianni, gli dovrei parlare.
I think that, to Gianni, I should speak to him.

(14)

a.

Penso, a Gianni, di dovergli parlare.


I think, to Gianni, of to have to speak to him.
b. *Penso di, a Gianni, dovergli parlare.
I think of , to Gianni, to have to speak to him.

Che and di cannot co-occur, as they are specialized for finite and non-finite
clauses, respectively; but sometimes delimiting complementizer particles can

124

Luigi Rizzi

co-occur, before and after the topic-focus field, as particles mai and a in Welsh,
according to Roberts (2004) analysis:
(15)

Dywedais i [ mai r dynion fel arfer a


Said
I
C
the men
as usual C
[werthith y ci ]].
will-sell the dog
(Welsh: Roberts 2004)

Back to the Italian periphery, it can be observed that the interrogative complementizer se (if), marking yes/no embedded questions, seems to occupy an
intermediate position between che and di in that it can be both preceded and
followed by a topic:
(16)

a.
b.

Non so, a Gianni, se gli potremo parlare.


I dont know, to Gianni, if we could speak to him.
Non so se, a Gianni, gli potremo parlare.
I dont know if, to Gianni, we could speak.

On the other hand, se must be high enough to obligatorily precede a focus


position:
(17)

a.

Non so se proprio QUESTO volessero dire (e non qualcosaltro).


I dont know if exactly THIS they wanted to say (and not something else).
b. *Non so proprio QUESTO se volessero dire (e non qualcosaltro)
I dont know exactly THIS if they wanted to say (and not something else).

These simple distributional facts are hard to reconcile with a single layer
hypothesis for the C system, whereas they can be immediately accommodated
if a richer system is assumed: che delimits the complementizer system upwards,
di delimits it downwards, and se can appear in the middle, possibly preceded
and followed by topics (but necessarily higher than focus):
(18)

che . . . Top . . . se . . . Foc . . . Top . . . di

Much evidence of this sort can be gathered by pairwise comparisons of elements


(if A > B and B > C, then A must be in a higher position than C even if the two
elements cannot be found simultaneously in the same structure for independent
reasons, as che and di in Romance, specialised for finite and non-finite clauses,
respectively). But natural languages sometimes offer more direct evidence for

Delimitation Effects and the Cartography of the Left Periphery

125

complex hierarchical structure of functional heads: in some cases, distinct C-like


particles co-occur in rigid orders, e.g. que si (that if) in certain kinds of Spanish
indirect questions (Plann 1982; Suner 1994, etc.) in cases like the following:
(19)

Me preguntaron que si tus amigos ya te visitaron en Granada.


They asked me that if your friends had already visited you in Granada.
(Suner 1994: 349)

Cases like (19) straighforwardly confirm the higher part of map (18), arrived at
through indirect comparisons like (13)(16).
The opposite order if that is also found in some languages, e.g. Dutch varieties
in which the sequence wie of dat alternates with of in cases like (4a) (see (20a)).
Clearly, (the equivalent of) that is an unmarked, versatile complementizer form,
capable of occurring in the highest C position, and also, in cross-linguistically
variable manners, in lower positions: as the head hosting a preposed adverbial
clause in English varieties admitting (20b) (McCloskey 1992; Rizzi 2010b), as
the head hosting various types of left peripheral elements including topics in
the old southern Italian dialects discussed by Ledgeway (2003), as in (20c); it
can also be a focus head in Brazilian Portuguese (Mioto 1999), as in (20d), and
a marker of wh-exclamatives in Italian, as in (20e), etc.
(20)

a.

b.
c.

d.

e.

Ik weet niet [wie of dat [Jan


gezien heeft]].
I know not who Q that Jan
seen
has
(Dutch varieties, Haegeman 1994)
I think that, if they arrive on time, that they will be greeted.
Le mand`o a dire che tutte quille dinare che le voleva dare re de
Franza per larmata.
He sent (someone) to tell him that all this money that the king of
France wanted to give him for the army.
(Ledgeway 2003: 131)
A Joana acha que A MARIA (que) o Joao encontrou no cinema.
Joana thinks that MARIA Joao met in the cinema.
(Mioto 1999)
Che bel libro che ho letto!
What a nice book that I have read!

Clearly, the unmarked form of the complementizer can occur in different positions, sometimes simultaneously at the edge and within the complementizer
zone (Ledgeway 2003 argues that in cases of co-occurrence like (20c) the lower
occurrence can be analyzed as the spell out of the trace of che moving from Fin
to Force across intermediate head positions).

126

Luigi Rizzi

A striking case of multiple C-like particles allowed to co-occur is the no ka to


sequence uncovered in Japanese by Saito (2012) where to marks the reported
character of the indirect question (much as in the Spanish case (19)), ka expresses
interrogative force, and no is analyzed by Saito as a finiteness marker:
(21)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (no) ka (to)]


T.-top
he-gen sister-nom there-in was no ka to
minna-ni tazuneta.
all-dat inquired
Taroo asked everyone if his sister was there.
(Saito 2012: (41))

Given the head final character of the language the hierarchical order will be
. . . no] ka] to] which mirrors the hierarchical order Declarative/report interrogative force finiteness that is postulated for Romance. Saito (2012) also
provides evidence for a recursive topic field sandwiched in between Force and
Finiteness in Japanese:
(22)

[. . . [. . . [. . . [. . . [TP . . . ] Fin] (Topic*)] Force] Report]


(Saito 2012)

This structure is very close to the one originally proposed in Rizzi (1997) for
Italian/Romance, modulo headedness and other parametric differences such as,
in Saitos analysis, the non-occurrence of a peripheral focus position in Japanese.
All these cases of distinct C-like elements occurring in different positions and
sometimes allowed to co-occur in a rigid order provide straightforward evidence
for a rich cartographic representation of the C-system.
5.

Cartography vs. Simpler Representations

Are there alternatives to the view embodied in (18), (22) and similar representations? The main alternative amounts to insisting on more impoverished clausal
representations like (12), with basic ingredients such as:
a.
b.
c.

a single CP layer;
multiple adjunctions permitted to IP/TP;
interpretive systems made capable of interpreting the adjoined material as
topic or focus, and of expressing general or language specific co-occurrence
and ordering constraints.

Delimitation Effects and the Cartography of the Left Periphery

127

In essence, the main alternative to the cartographic representations would


amount to simplifying the syntactic representations and shifting to the interface systems much of the work that cartographic syntax does.
Actually, simplifying the syntax is probably a misnomer to refer to such
an approach. In a certain sense, syntax is equally simple with richer and more
impoverished representations, as it involves the same basic computational ingredients: binary merge of heads and phrases. What is simpler in this alternative
approach is the functional lexicon, and the resulting representations are more
impoverished, but the mechanism of syntactic computation does not vary.
Such a simple representations approach has several drawbacks, as the
previous discussion already made clear:
i.

It has no natural way to express cases with co-occurring multiple C-particles


(i.e., (15), (19), (21), etc.), which often can be interrupted by phrasal material, hence cannot be analyzed as single syntactic nodes: these facts are
simply inconsistent with a simple C node approach;
ii. In cases of distinct C-particles occurring in different positions with respect
to third elements, a simpler representations approach must assume complex constraints to hold on the adjunction mechanism. E.g., in order to
capture the pattern in (13), (14), (16) adjunction must be assumed to be
permissible to a non-finite CP and to a finite interrogative CP (to permit
(14a), (16a)), but not to a finite declarative CP (to exclude (13a)); and it
must be allowed to apply to a finite TP (to permit (13b), (16b)), but not to
a non finite TP (to exclude (14b)). No such complex system of constraints
is required by the approach assuming cartographic representations, which
can in fact dispense with phrasal adjunction altogether (much as in Kaynes
1994 restrictive approach to phrase structure), and analyzes movement to
topic, focus, etc. as movement to Spec.
iii. In languages with overt topic and focus particles, a simpler representations approach must introduce special mechanisms in the interface systems
to account for their distributional properties, ordering constraints, etc.
iv. More generally, it must endow the interpretive systems of mechanisms capable of capturing the observed variation: unique topics in some languages,
recursive topics in other languages. In other words, such an approach must
admit the possibility of expressing a parametrisation in the postsyntactic
interpretive system.
An approach assuming cartographic representations can capture the different
co-occurrence and ordering properties, as well as the observed variation, via a
fundamental, independently necessary mechanism: head-complement selection,
admitting a certain amount of parametrisation: the Force head can select a Topic

128

Luigi Rizzi

Phrase, which can be recursive or not (i.e., Top can in turn select a Topic Phrase,
as in Italian, or not, as in Gungbe); in many languages a Top head can select a
Focus Phrase, but not vice versa; in Italian a Foc head can select a Top phrase,
perhaps of a special kind (a familiarity topic, in the typology introduced by
Frascarelli and Hinterhoelzl 2007); Force, Top, Foc can all select a Fin Phrase,
and Fin always selects an IP/TP, much as T can select an Aspect Phrase, v
selects a VP, V can select a DP, etc.: the same fundamental formal mechanism
is used across the board to express invariant and variable properties in structure
building. The selecting heads are sometimes overt and sometimes null, much
as, e.g., certain tense, aspect, voice or case morphemes.
A related but distinct issue is the question of the further explanation of
the functional hierarchy that is observed: why is it that we typically find certain
orders, rather than others? As pointed out in Cinque and Rizzi (2010) it is
unlikely that the hierarchy may be an absolute syntactic primitive, unrelated to
other requirements or constraints: why should natural language syntax express
such a complex and apparently unmotivated primitive? It is more plausible
that the functional hierarchy may be rooted elsewhere. External factors such
as interpretive requirements may be relevant in some cases. For instance, it is
argued in Rizzi (1997) that the non-recursive character of FocP is linked to the
interpretive properties of the Focus Presupposition articulation: if FocP were
recursive, the lower FocP would end up being in the presuppositional part of the
higher FocP, and an interpretive clash would arise. No such interpretive problem
arises in the case of TopP (because the comment of a higher TopP can in turn
have a Topic Comment structure), hence TopP can be recursive, an attested
parametric option.
Another factor which may explain certain orders is the theory of locality. As
Abels (2010) argues, if in a language the TopP has island-creating properties
and the FocP does not (or such properties hold relative to the two types of
movement, as in the system of Starke 2001; Rizzi 2004), then the only possible
order is Top Foc because in the opposite order the focused element could not
reach the Spec of the FocP across the topic island. This is a perfectly fine mode
of explanation and, as far as I can see, it is fully consistent with the adoption
of cartographic representations. On the one hand, it has nothing to say on cases
in which movement is not involved (i.e., in the Force Fin system in cases like
(15), (19), (21)), which therefore require a rich functional system anyhow, with
ordering determined on some other basis (presumably interpretive requirements,
possibly the locality of selection, as suggested in Rizzi 1997). On the other hand,
in the cases which involve movement of topics, foci, wh, etc., the critical issue
for the divide between cartographic vs simpler representations has to do with
the nature and properties of scope-discourse movement, not with the primitive

Delimitation Effects and the Cartography of the Left Periphery

129

or derived nature of the ordering. If the criterial approach is on the right track,
as we have argued, a rich system of attracting heads is needed anyhow, quite
independently from the ordering issue, and it can very naturally be combined
with an approach a` la Abels to the ultimate, principled explanation of ordering.
So, the Occams razor argument that Abels gives is against the assumption of
a primitive templatic ordering, while, as far as I can see, it is fully consistent
with the criterial approach to scope-discourse semantics, which is the crucial
ingredient of the cartographic representations defended here.
Locality considerations may provide a principled explanation for certain hierarchical orderings of the functional sequence, as in Abels approach, and also
explain certain restrictions on the occurrence of left peripheral elements in certain structural environments, as argued in detail by Haegeman (2010): in her
analysis, the restrictions on the occurrence of a full-fledged left periphery in
various kinds of adverbial and complement clauses is a consequence of the fact
that such embedded constructions necessarily involve some kind of operator
movement which would be adversely affected by the intervention of other leftperipheral elements. Again, this kind of locality-based explanation is fully consistent with detailed cartographic representations, in fact it presupposes them.
In this kind of approach, the syntax-interpretation interface is fully transparent: there is a single kind of structural relation, the head-dependent relation which is read off syntactic representations by the interpretive systems, in
conjunction with the content of the criterial heads, to assign topicality, focus,
presupposition, etc., much as thematic role assignment works. The cartographic
representations exploit fundamental syntactic devices to transparently express
interpretive properties, thus simplifying the burden of the interface systems.
They represent an attempt to syntacticize scope-discourse semantics as much
as possible (Cinque and Rizzi 2010) without enriching the inventory of the
computational mechanisms needed on either side of the interpretive interface.
6.

Criterial Positions Delimit Chains

A criterial position delimits a movement chain in the sense that a phrase meeting
a Criterion cannot undergo further movement . This freezing effect is best illustrated by cases in which the same phrase contains two criterial features, e.g., a
Q feature in the wh-specifier of a nominal expression and a (contrastive) focus
feature in the lexical restriction, as in:
(23)

[quantiQ LIBRIFoc ]
How many BOOKS

130

Luigi Rizzi

A priori one could expect the phrase to move to one criterial position, satisfying
the requirement of one feature, and then continue to move to a higher criterial
position to satisfy the other feature. But this never happens, and the phrase gets
stuck in the lower criterial position. In the following case, the phrase (23) moves
to the Q head of the embedded question, and then cannot further focus move to
the main clause, e.g. in a cleft configuration:
(24)

a.

Non ho capito [[ quantiQ LIBRIFoc ] Q [ dobbiamo leggere ]], non


quanti articoli.
I didnt understand how many BOOKS we have to read, not how
many articles.
b. *E [ quantiQ LIBRIFoc ] che non ho capito [ Q [ dobbiamo leggere ]], non quanti articoli
It is how many BOOKS that I didnt understand we should read,
not how many articles.

The only options for a well-formed outcome, given (24a), are focalization in
situ (possibly obtained by an Agree relation with a Foc head in the periphery of
the main clause) as in (24a), or pied-piping of the whole indirect question to the
focus cleft position:
(25)

E [[ quantiQ LIBRIFoc ] Q dobbiamo leggere ] che non ho capito ,


non quanti articoli.
It is how many BOOKS we should read that I didnt understand, not
how many articles.

Here movement does not undo the criterial configuration, which remains intact
in the C-system of the indirect question, but the whole configuration is piedpiped to a higher criterial position, the focus position of the cleft (Belletti 2009).
This kind of pattern justifies a freezing principle of the following kind (Rizzi
2006; Rizzi and Shlonsky 2007; see also Lasnik and Saito 1992 for an earlier
characterisation of the freezing effect, Boskovic 2005 for an analysis in terms
of deactivation, and Rizzi 2010a, b for discussion):
(26)

Criterial Freezing: an XP meeting a Criterion is frozen in place.

One way of conceptualizing this principle is to assume that as soon as a criterial


configuration is formed, the configuration is handed over to the interface systems
for interpretation, and becomes unavailable to further syntactic computation.
This principle, among other things, reduces the amount of reconstruction that
is needed (i.e., the necessity of reconstructing a criterial phrase into a lower
criterial position will never arise), thus facilitating computation by the grammar

Delimitation Effects and the Cartography of the Left Periphery

131

and the parser. Criterial Freezing can thus be looked at it as an economy principle
favoring efficiency in syntactic computations.
7.

The Subject Criterion

In previous work, I have proposed that movement to subject position is criterial


in nature. Classical EPP can be restated in terms of assuming a criterial head
Subj(ect), obligatory part of the clausal backbone, much as T, etc. This head
expresses an interpretive relation between its specifier and its complement, as
criterial heads do in general. In this particular case, the specifier of Subj, the
subject, is interpreted as the argument about which the event referred to by
the clause is presented. In this sense, an active and a passive sentence differ, as
the same event is presented as being about the agent or the patient, respectively:
(27)

a.
b.

Mario ha insultato Piero.


Mario insulted Piero.
Piero e` stato insultato da Mario.
Piero was insulted by Mario.

As originally observed by Calabrese (1986), a pro subject in a null subject


language picks out an aboutness subject, at least in cases in which ambiguity
might arise. So, consider the following continuation of discourse:
(28)

Subito dopo, pro e` andato via.


Immediately after that, pro left.

If (28) is uttered immediately after (27a), it is understood that Mario left; if


(28) is uttered immediately after (27b), it is Piero who left (referring to the
non subject in (27), the object or the by-phrase, would require using the overt
subject pronoun lui (he) in (28)).
Aboutness is a property that subjects have in common with topics (Reinhart
2005). So, if an aboutness subject and a topic are present, they can both be
picked up by pro in the following discourse continuation:
(29)

a.
b.

Piero, Mario lo ha insultato.


Piero, Mario insulted him.
Subito dopo, pro e` andato via.
Immediately after that, pro left.

132

Luigi Rizzi

In (29b), pro can pick up both the topic Piero (in the Clitic Left Dislocation
construction) and the subject Mario as antecedent, perhaps with a preference
for the latter, possibly for proximity reasons.
As for the structural expression of the Subject Criterion, I have assumed,
following Cardinaletti (2004), that the Criterial head Subj is part of the obligatory
structure of the IP, in a position higher than T and immediately lower than the
lowest head of the C-system Fin. A plausible candidate of a language with
an overt morphological realization of Subj is a language with subject clitics
occurring in between the subject DP and the predicate starting with the inflected
verb, such as the Northern Italian Dialects (Poletto 2000; Manzini and Savoia
2005):
(30)

a.

b.

El fio el
mangia l
pom.
The boy subj eats
the apple
(Milan: Rizzi 1986; Poletto 2000; Manzini and Savoia 2005, etc.)
Le ragazze le son venute.
The girls subj have+3pl come.
(Florence: Brandi and Cordin 1989)

The syntactic role of the Subj head is to delimit the structure of the IP zone,
and attract a nominal expression to its Spec; its interface role is to trigger the
aboutness interpretation, imposing the interpretation of the event expressed
by the predicate as being about the subject.
Capitalizing on the formal similarity (or identity) of the Subj head with the
determiner in cases like (30), one may speculate that Subj is a D-like head
which has in common with its counterpart in the extended projection of NPs
the affinity with a nominal projection; more specifically, both clausal and
nominal Ds would have the capacity to attract a nominal expression to their
Specs (here I am thinking in particular of the derivational processes overtly
triggered in D-final languages, analyzed by Cinque (2005) as involving the
attraction of the NP to the Spec of D, with or without pied-piping of additional
material).
Different kinds of evidence can be provided in favor of the view that movement to subject position is determined by the attraction of nominal features by
the relevant functional head. This approach can provide an explanation to a curious asymmetry observed long ago by Ruwet (1972). In French, an adnominal
complement of the object can be pronominalized by the clitic pronoun en, and
the remnant of the object DP can be moved to subject position in passive:
(31)

a.

Jean a publie [la premi`ere partie de ce roman] lannee passee.


Jean published the first part of this novel last year.

Delimitation Effects and the Cartography of the Left Periphery

b.

c.

133

Jean en a publie [la premi`ere partie ] lannee passee.


(de ce roman : en = pro-PP)
Jean of-it published the first part last year.
[La premi`ere partie ] en a e te publiee lannee passee.
The first part of-it was published last year.
(Ruwet 1972)

En can also pronominalize the NP in a DP introduced by an indefinite quantifier


like [trois NP] in (32a); but in this case, the remnant [trois ] cannot be further
moved to subject position, as in (32c):
(32)

a.

Jean a publie [trois romans] lannee passee.


Jean published three novels last year.
b. Jean en a publie [trois ] lannee passee.
(romans : en = pro-NP)
Jean of-them published three last year.
c. *[Trois ] en ont e te publies lannee passee
Three of-them have been published last year.

This prohibition is limited to movement to subject position: A movement to the


left periphery can move the DP remnant after en extraction:
(33)

[Combien ] il en a publies lannee passee?


How many he of-them published last year?

A natural analysis of the impossibility of (32c) is that in this case, after the pro-NP
en has been extracted, the DP does not contain any more an active (attractable)
component specified [+N], hence the attraction to Spec Subj cannot take place
(here I am assuming, with much recent work, that even if we adopt the copy
theory of traces, according to which a copy of en remains expressed within the
object DP in (32), a trace cannot be attracted and moved to a higher attractor).
No problem arises in (31), because the remnant DP still contains a nominal part,
headed by partie, which can be attracted in (31c); nor with A movement as in
(33), because here the attracting feature is [+Q], which is specified in combien
in the remnant DP, not [+N].
Another classical observation which is amenable to the conception of subject
movement adopted here is Kosters generalization that subject sentences dont
exist (Koster 1978). Koster noticed that putative sentential subject clauses as
in (34) are better analyzed as sitting in topic position; that they may not be in
the canonical subject position is suggested by different kind of evidence, e.g.,
the fact that they dont naturally allow I to C movement:

134
(34)

Luigi Rizzi

a. That he will come proves this.


b. *What does that he will come prove?

If finite clauses are not [+N] (as is suggested by the fact that they dont bear
Case in many languages), they cant be attracted to Spec-Subj, hence they cannot
appear in the canonical subject position (while they can be topicalized, focus
moved, etc.).
If this analysis is on the right track, the Criteria are not solely involved in
the triggering of A movement: also the core case of A-movement, movement
targeting the subject position, is criterial in nature. So, the notion of Criteria
cuts across the A/A distinction. On the other hand, the A/A distinction cannot
be entirely obliterated: among other things, A-movement clearly is more local
than typical A -movement. For instance, movement to Top can easily skip intervening DPs, while movement to Subj is strictly local: only the closest DP can
be attracted by Subj:
(35)

a. Mary, I believe that Bill should talk to .


b. *Mary was believed that Bill should talk to

The difference has to do with the different nature of the attracted feature. A
criterial A head such as Top or Foc attracts an expression endowed with a
matching feature, so, if in (35a) neither I nor Bill are endowed with [+Top],
they can be skipped by Mary moving to Top, under a featural formulation of
Relativized Minimality (Starke 2001; Rizzi 2004). On the other hand, if Subj
attracts a [+N] element, it will inevitably go for the closest such element, and
the different nominal interveners in (35b) will block movement of Mary.
Why does Subj not attract a more selective feature analogous to Top, Foc,
etc.? Presumably this has to do with the inherent weakness of the interpretive
effect of the Subject Criterion: Spec-Subj does not express, per se, a particular
informational property (given or new information), or a particular quantificational property, it only expresses pure aboutness. As such, the notion has no
specific scope-discourse featural content to operate on, it just exerts an attraction on any nominal. Because of locality, the attracted nominal will inevitably
be the closest one.
So, the stricter locality of movement to subject w.r.t. A movement may well
be a consequence of the weakness of the interpretive content associated to the
subject position.
This state of affairs may also be connected to the existence of expletive
subjects. An expletive is a nominal expression which cannot carry any scopediscourse featural specification (it cant be topical, nor focal, as it doesnt have
any referential content, nor can it express any kind of quantification), so it cannot

Delimitation Effects and the Cartography of the Left Periphery

135

undergo A criterial movement, and we typically dont find expletives in topic,


focus, or Q positions (but see Beninc`a and Poletto 2004 on a possible case of
expletive topic in Old Italian). But an expletive is specified [+N], and as such
it can be attracted to Spec-Subj. In this case, the aboutness interpretive routine
cannot be triggered, as there is no referent about which the event is presented, and
the sentence receives what is sometimes called a presentational interpretation.
8.

Subject-Object Asymmetries: ECP Effects as Criterial Freezing

The combined effect of the Subject Criterion and Criterial Freezing provides
an explanation for the well-known subject-object asymmetries in extraction
processes, instantiated by the that-trace effect in English:
(36)

a. *Who do you think [that [ subj will come]]?


b. Who do you think [that [Mary subj will meet

]]?

The traditional analysis involves the Empty Category Principle (ECP), requiring
a certain type of government relation (proper government) to be satisfied by
traces. But the ECP is hard to accommodate with the guidelines of the Minimalist
Program, as it does not have a natural status within the principled typology of
UG principles assumed by minimalism: neither can it be naturally construed as
an economy principle, nor as a principle enforced by some requirement of the
interface systems.
The system proposed here provides a simple alternative to an ECP-based
analysis of subject-object asymmetries (Rizzi 2006; Rizzi and Shlonsky 2007).
If there is a Subject Criterion, further movement of the subject will be generally
blocked by Criterial Freezing. So, in the derivation of (36a), who has to move to
Spec-Subj in order to satisfy the Subject Criterion in the embedded clause, as
the Criterion cannot be satisfied in any other way in this case; but then who will
be stuck there, and will be disallowed to move further and undergo extraction
by Criterial Freezing:
(37)

Subject extraction is blocked by Criterial Freezing.

Various kinds of empirical evidence have been produced to show that this alternative is more satisfactory than an ECP approach (Rizzi and Shlonsky 2007).
For instance, the ECP approach does not offer a natural analysis of the fortrace effect, the fact that a subject is not extractable across the prepositional
complementizer for in English:

136
(38)

Luigi Rizzi

a. *Who would you prefer [for [


b. Who do you work for ?
c. *Who would you prefer [for [

to win]]?
subj to win]]?

It is hard to see how the subject trace could fail to be properly governed by the
complementizer for, particularly in view of the fact that the minimally different preposition for licenses extraction, as in (38b). In terms of the alternative
envisaged here the ill-formedness of representation (38c) is straightforward:
for is a member of the C-system (presumably a realisation of Fin: Rizzi 1997),
so the whole IP system will be developed under it, including the SubjP layer.
Then (38b) involves further movement of an element which satisfies the Subject
Criterion, which determines a violation of Criterial Freezing, much as in the
that-trace configuration in (36a).
If the complementizer is dropped in (36a), (38a), extraction becomes possible, either because complementizer drop actually involves the truncation of an
important structural chunk, including the Subj layer, so that no freezing effect is
determined; or, more plausibly, in the more indirect and elaborate way proposed
in Rizzi and Shlonsky (2007), which assimilates subject extraction to simple
movement in subject questions like Who will come?
Subject extraction is notoriously harder than object extraction across languages, but it is not banned altogether. Languages normally invent strategies to
make it possible to form a question or other A -constructions on an embedded
subject (Rizzi and Shlonsky 2007). One typical such strategy is what Rizzi and
Shlonsky (2007) call a skipping strategy: the language uses an expletive-like
element to formally satisfy the Subject Criterion, and this allows the thematic
subject to escape the freezing effect and remain accessible to extraction. A systematic utilisation of the skipping strategy is observed in null subject languages.
Consider the following comparative generalisation:
(39)

Null subject languages are not sensitive to that-trace effects.


(Perlmutter 1970; Taraldsen 1978; Pesetsky 1981; Rizzi 1982, 1990;
Nicolis 2005, 2008).

In fact, the word by word equivalents of sentences like (36a) are fully acceptable
in null subject languages like Italian, Spanish, Rumanian, etc. In my original
work on the null subject parameter, I proposed that subject extraction in these
languages always proceeds from a lower position, while the canonical subject
position (the EPP position) is filled by an expletive occurrence of the null pronoun pro. Hence (40a) has a representation like (40b).

Delimitation Effects and the Cartography of the Left Periphery

(40)

a.

b.
c.

137

Chi credi che verr`a?


Who do you think that will come?
(Rizzi 1982; 1990)
Chi credi [che [pro verr`a ]]?
Who do you think that will come?
Chi credi [che [pro subj verr`a ]]?
Who do you think that will come?

In the analysis of Rizzi (1982), pro offered a device to avoid leaving a trace in a
non-properly governed position like the subject position, hence in an illegitimate
position according to the ECP. The analysis can now be immediately transposed
to the Criterial Freezing approach. The representation is (40c), and pro formally
satisfies the Subject Criterion and is frozen there, thus making the thematic
subject accessible to extraction from a lower, non-criterial position. This is just
one device that languages may use to satisfy the Subject Criterion without having
to move the thematic subject to the freezing position (Rizzi and Shlonsky 2007;
and see Taraldsen 2001; Endo 2008 on the role of certain clause final particles
in Japanese to license subject movement; and the discussion in Miyagawa 2004
on ways of satisfying the EPP).
What lower position is the thematic subject extracted from in (40)? In the
original analysis along the lines indicated in (40a)(40b), subject extraction in
Romance null subject languages was made contingent upon the existence of
free subject inversion; i.e., the following, possible in Italian, Spanish, etc.,
was considered an intermediate step in the derivation of (40a).
(41)

Credo che verr`a Gianni.


I believe that will come Gianni.

But this particular assumption is unnecessary and problematic, both theoretically


and empirically. Chao (1981) had already observed that this line of analysis
is difficult to maintain for Brazilian Portuguese, which does not allow a free
inversion process (at least not as freely as the European null subject languages),
and still admits free violations of the that-trace effect. Nicolis (2008) analyzes
several Creole languages which admit a free violation of that-trace and disallow
subject inversion, and observes that the lack of a dependency between these two
properties also emerges from Gilligans (1987) survey study (three languages in
his sample, Basque, Papiamento and Yoruba, permit free violations of that-trace
and have no subject inversion).
There are also theoretical reasons for assuming that (41) and, more generally,
subject inversion structures, are not an intermediate step in the derivation of
(40c). Belletti (2001, 2004) proposed that the core case of subject inversion is

138

Luigi Rizzi

in fact a device for focusing the thematic subject, which is moved to a low focus
position, specified in the periphery of the vP. Null subject languages not allowing
subject inversion can thus be seen as lacking this particular focalization device.
As the low focus position expresses a scope-discourse property, it is natural to
analyze it as a criterial position. But then, an inverted subject would fall under
Criterial Freezing, and should not be movable from the low focus position.
In fact, the skipping strategy permitting a free violation of that-trace does
not require a passage through the low focus position: the essential ingredient is
the availability of expletive pro in the language, the device formally satisfying
the Subject Criterion and permitting the A extraction of the thematic subject
from a lower position which does not have to be the low focus position (and
cannot be the low focus position, under Criterial Freezing). In fact, Brazilian
Portuguese, and the Creole languages studied in Nicolis (2008) (Berbice Dutch,
Cape Verdean, Mauritian, Papiamentu) all admit a null expletive, hence the free
violation of that-trace is expected (see also Menuzzi 2000 for an argument that
in Brazilian Portuguese the subject is indeed extracted from a position lower
than Spec-Subj); on the other hand, the variety of Jamaican Creole described in
Durrleman (2008) does not allow null expletives in embedded clauses (only in
root contexts), and it also disallows that-trace violations: the connection between
null expletives and subject extraction is thus confirmed by this case (the case of
Haitian Creole is complicated by the residual presence of a kind of French que
qui rule in the language: see Nicolis 2008, for discussion).
9.

Conclusion

This paper focused on A -chains, the kind of chains which associate two types of
interpretive properties to expressions: properties of semantic selection (thematic
properties for arguments) and properties of scope-discourse semantics. I have
adopted the criterial approach to scope-discourse semantics, which traces back
the assignment of discourse and informational functions to familiar syntactic
ingredients, head-dependent relations, and I have shown how this view naturally leads to the study of the cartography of syntactic structures: the study of
the complex syntactic configurations created by a simple computational system,
operating on a richly structured functional lexicon. Looking at the structure of
the initial periphery of the clause, I have compared an analysis involving rich
cartographic representation, and one based on simpler representations, assuming a single C-layer, and admitting multiple adjunctions to TP. The latter simpler
representations approach shifts much of the descriptive burden to the interface
systems, not only interpretation proper, but also positional and co-occurrence

Delimitation Effects and the Cartography of the Left Periphery

139

restrictions, and the related parametrisation. On the contrary, the approach based
on cartographic representations stretches the role of syntax in the effort of syntacticizing as much as possible the interfaces with sound and meaning, assuming
syntactic representations which carry interpretive properties on their sleeves.
I have tried to show that the cartographic representations compare favorably to
the syntactically less elaborate alternatives. I have then turned to the issue of
delimitation. Movement is delimited, in the sense that it must start and finish
in particular structural positions: in fact, the positions dedicated to particular
interpretive properties have the effect of delimiting the movement chains, and
delimitation theory is establishing itself as a new chapter of the classical topic of
locality. I have tried to show that, by assuming principles of upward delimitation,
or freezing, we can explore new generalizations, and envisage novel explanatory accounts for much-studied phenomena, such as the classical subject-object
asymmetries in extraction processes
References
Abels, Klaus
2010
Aboh, Enoch
2004
Aboh, Enoch
2007

The Italian Periphery: A View from Locality. Ms., University College


London.
The Morphosyntax of Complement-Head Sequences. New York: Oxford University Press.
Information structuring begins with numeration. Ms., ACLC, University of Amsterdam.

Bayer, Josef
1984
Comp in Bavarian. The Linguistic Review 3: 209274.
Belletti, Adriana
1990
Generalized Verb Movement. Turin: Rosenberg & Sellier.
Belletti, Adriana
2001
Inversion as Focalization. In: Aafke Hulk and Jean-Yves Pollock
(eds.), Subject Inversion in Romance and the Theory of Universal
Grammar, 6090. Oxford: Oxford University Press.
Belletti, Adriana
2004
Aspects of the Low IP Area. In: Luigi Rizzi (ed.), The Structure of CP
and IP, 1651 (The Cartography of Syntactic Structures 2.). Oxford:
Oxford University Press.
Belletti, Adriana
2009
Structures and Strategies. London: Routledge.

140

Luigi Rizzi

Beninc`a, Paola and Cecilia Poletto


2004
Topic, Focus and V2: Defining the CP sublayers. In: Luigi Rizzi (ed.),
The Structure of CP and IP, 5275 (The Cartography of Syntactic
Structures 3.). Oxford: Oxford University Press.
Bianchi, Valentina and Mara Frascarelli
2009
Is Topic a Root Phenomenon? Ms., University of Siena, University of
Roma 3.
Biloa, Edmond
2008
The Cartography of the Left Periphery in Tuki. Ms., University of
Yaounde I, Cameroon.
Bocci, Giuliano
2009
On Syntax and Prosody in Italian. Doctoral dissertation, University of
Siena.

Boskovic, Zeljko
2005
On the Operator Freezing Effect. Ms., University of Connecticut.
Brandi, Luciana and Patrizia Cordin
1989
Two Italian Dialects and the Null Subject Parameter. In: Osvaldo Jaeggli and Kenneth Safir (eds.), The Null Subject Parameter, 111142.
Dordrecht: Kluwer.
Calabrese, Andrea
1986
Some Properties of the Italian Pronominal System: An Analysis Based
on the Notion of Thema as Subject of Predication. In: Harro Stammerjohann (ed.), Tema-Rema in Italiano, 2536. Tubingen: Gunter
Narr.
Cardinaletti, Anna
2004
Towards a Cartography of Syntactic Positions. In: Luigi Rizzi (ed.),
The Structure of CP and IP, 115165 (The Cartography of Syntactic
Structures 2.). Oxford: Oxford University Press.
Chao, Wynn
1981
Pro-drop languages and nonobligatory control. University of Massachusetts Occasional Papers in Linguistics 7: 4674.
Chomsky, Noam
1986
Barriers. Cambridge, Mass: MIT Press.
Chomsky, Noam
2004
Beyond Explanatory Adequacy. In: Adriana Belletti (ed.), Structures
and Beyond, 104131 (The Cartography of Syntactic Structures 3.).
New York: Oxford University Press.
Cinque, Guglielmo
1999
Adverbs and Inflectional Heads. Oxford, NewYork: Oxford University
Press.
Cinque, Guglielmo
2005
Deriving Greenbergs Universal 20 and its Exceptions. Linguistic Inquiry 36(3): 315332.

Delimitation Effects and the Cartography of the Left Periphery

141

Cinque, Guglielmo and Luigi Rizzi


2010
The Cartography of Syntactic Structures. In: Bernd Heine and Heiko
Narrog (eds.), The Oxford Handbook of Linguistic Analysis, 5165.
Oxford, New York: Oxford University Press.
Cruschina, Silvio
2008
Discourse-related Features and the Syntax of Peripheral Positions: A
Comparative Study of Sicilian and other Romance Languages. Doctoral Dissertation, University of Cambridge.
Durrleman, Stephanie
2008
The Syntax of Jamaican Creole A Cartographic Perspective. (Linguistics Today 127.) Amsterdam/Philadelphia: John Benjamins Publishing Company.
Endo, Yoshio
2007
Locality and Information Structure A Cartographic Approach to
Japanese (Linguistics Today 116.) Amsterdam/Philadelphia: John
Benjamins Publishing Company.
Frascarelli, Mara
2000
The Syntax-Phonology Interface in Focus and Topic Constructions in
Italian. Dordrecht, Kluwer Academic Publishers.
Frascarelli, Mara and Roland Hinterholzl
2007
Types of Topics in German and Italian. In: Susanne Winkler and Kerstin Schwabe (eds.), On Information Structure, Meaning and Form,
87116. Amsterdam: Benjamins.
Gilligan, Gary
1987
A Cross-linguistic Approach to the Pro-drop Parameter. PhD Dissertation, University of Southern California.
Grewendorf, Gunther
2002
Left Dislocation as Movement. In: Simon Manck and Jennifer Mittelstaedt (eds.), Georgetown University Working Papers in Theoretical
Linguistics, 3181.
Haegeman, Liliane
1994
An Introduction to Government-Binding Theory. Oxford: Blackwell.
Haegeman, Liliane
2003
Notes on long adverbial fronting in English and the left periphery.
Linguistic Inquiry 31: 640649.
Haegeman, Liliane
2006
Argument Fronting in English, Romance ClLD and the Left Periphery.
In: Raffalla Zanuttini, Hector Campos, Elena Herburger and Paul Portner (eds.), Negation, Tense and Clausal Architecture: Cross-linguistic
Investigations, 2752. Washington DC: Georgetown University Press.
Haegeman, Liliane
2010
Adverbial Clauses, Main Clause Phenomena, and the Composition of
the Left Periphery. Ms., University of Gent.

142

Luigi Rizzi

Jayaseelan, K. A.
2008
Topic, Focus and Adverb Positions in Clause Structure. Nanzan Linguistics 4: 4368.
Kayne, Richard
1983
Connectedness and Binary Branching. Dordrecht: Foris Publications.
Kayne, Richard
1994
The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press.
Koster, Jan
1978
Why Subject Sentences dont Exist. In: Samuel J. Keyser (ed.), Recent
Transformational Studies in European Languages, 5364. Cambridge,
Mass.: MIT Press.
Krapova, Iliyana and Guglielmo Cinque
2004
On the Order of wh-phrases in Bulgarian Multiple wh fronting. Ms.,
University of Venice.
Lasnik, Howard and Mamoru Saito
1992
Move Alpha: Conditions on its Application and Output. Cambridge,
Mass.: MIT Press.
Ledgeway, Adam
2003
Il sistema completivo dei dialetti meridionali: la doppia serie di complementatori, Rivista Italiana di Dialettologia 27: 89147.
Manzini, Maria R. and Leonardo Savoia
2005
I dialetti Italiani e romanci. Alessandria: Edizioni dellOrso.
McCloskey, James
1992
Adjunction, Selection and Embedded Verb Second. Ms., UCSC.
Menuzzi, Sergio
2000
That-trace effects in Portuguese. Forum Lingustico 2.1: 1339. Florianopolis, UFSC.
Mioto, Carlost
1999
A periferia esquerda no portugues brasileiro. Ms., Universidade Federal de Santa Catarina, University of Siena.
Miyagawa, Shigeru
2004
On the EPP. MITWPL 49: 201236.
Nicolis, Marco
2005
On pro drop. Ph.D. dissertation. University of Siena.
Nicolis, Marci
2008
The Null Subject parameter and correlating properties: the case of
Creole languages. In: Theresa Biberauer (ed.), The limits of syntactic
variation, 271294. Amsterdam: John Benjamins.
Pearce, Elizabeth
1999
Topic and Focus in a head-initial language: Maori. In: Carolyn Smallwood and Catherine Kitto (eds.), Proceedings of AFLA vi. The SixthMeeting of the Austronesian Formal Linguistics Association, held at

Delimitation Effects and the Cartography of the Left Periphery

143

the University of Toronto April 1618, 249263. Toronto: University


of Toronto Dept. of Linguistics.
Perlmutter, David
1970
Deep and Surface Structure Constraints in Syntax. New York: Holt
Rinehart & Winston.
Pesetsky, David
1982
Complementizer-trace Phenomena and the Nominative Island Condition. The Linguistic Review 1: 297343.
Pesetsky, David and Esther Torrego
2001
T-to-C Movement: Causes and Consequences. In: Michael Kenstowicz
(ed.), Ken Hale:A Life in Language, 355426. Cambridge, Mass.: MIT
Press.
Plann, Susan
1982
Indirect Questions in Spanish. Linguistic Inquiry 13: 297312.
Poletto, Cecilia
2000
The Higher Functional Field. Evidence from Northern Italian Dialects. New York, Oxford University Press.
Pollock, Jean-Yves
1989
Verb Movement, Universal Grammar and the Structure of IP. Linguistic
Inquiry 20: 365424.
Reinhart, Tanya
2005
Interface Strategies. Cambridge, Mass.: MIT Press.
Rizzi, Luigi
1982
Issues in Italian Syntax. Dordrecht: Foris Publications.
Rizzi, Luigi
1986
Null Objects in Italian and the Theory of pro. Linguistic Inquiry 17(3):
501557.
Rizzi, Luigi
1991
Residual Verb Second and the Wh Criterion. Geneva Working Papers on Formal and Computational Linguistics. Republished in Rizzi
(2000).
Rizzi, Luigi
1997
The Fine Structure of the Left Periphery. In: Liliana Haegeman (ed.),
Elements of Grammar, 281337. Dordrecht: Kluwer.
Rizzi, Luigi
2000
Comparative Syntax and Language Acquisition. London: Routledge.
Rizzi, Luigi
2004a
Locality and Left Periphery. In: Adriana Belletti (ed.), Structures and
Beyond. The Cartography of Syntactic Structures, vol. 3, 104131.
New York: Oxford University Press
Rizzi, Luigi (ed.)
2004b
The Structure of CP and IP (The Cartography of Syntactic Structures 2.). Oxford: Oxford University Press.

144

Luigi Rizzi

Rizzi, Luigi
2006

Rizzi, Luigi
2009

Rizzi, Luigi
2010a

Rizzi, Luigi
2010b

On the Form of Chains: Criterial Positions and ECP Effects. In: Lisa
Cheng and Norbert Corver (eds.), Wh Movement: Moving on, 97133.
Cambridge, Mass.: MIT Press.
Movement and Concepts of Locality. In: Massimo Piattelli-Palmarini,
Juan Uriagereka and Pello Salaburu (eds.), Of Minds and Language:
A Dialogue with Noam Chomsky in the Basque Country, 155168.
Oxford, New York: Oxford University Press.
On some Properties of Criterial Freezing. In: E. Phoevos Panagiotidis
(ed.), The Complementizer Phase: Subjects and Operations, 1732.
Oxford, New York: Oxford University Press.

Topic, Focus and the Cartography of the Left Periphery. Ms., University of Siena.
Rizzi, Luigi and Ur Shlonsky
2007
Strategies of Subject Extraction. In: Hans-Martin Gartner and Uli
Sauerland (eds), Interfaces + Recursion = Language? Chomskys
Minimalism and the View from Syntax-Semantics, 115160. Berlin:
Mouton de Gruyter.
Roberts, Ian
2004
The C-system in Brythonic Celtic Languages, V2 and the EPP. In:
Luigi Rizzi (ed.), The Structure of CP and IP, 297328 (The Cartography of Syntactic Structures 2.). Oxford: Oxford University Press.
Ruwet, Nicolas
1972
Theorie syntaxique et syntaxe du Francais. Paris: Seuil.
Saito, Mamoru
2012
Sentence Types and the Japanese Right Periphery. In: this volume.
Salvi, Giampaolo
2005
Some Firm Points on Latin Word Order: The Left Periphery. In: Katalin
Kiss (ed.), Universal Grammar and the Reconstruction of Ancient
E.
Languages, 429456. Berlin: Mouton de Gruyter.
Shlonsky, Ur
2000
Subject Positions and Copular Constructions. In: Hans Bennis, Martin
Everaert and Eric Reuland (eds.), Interface Strategies, proceedings of
the colloquium, Amsterdam, 2426 September 1997, 325347. Amsterdam: Royal Netherlands Academy of Arts and Sciences.
Suner, Margarita
1994
V-movement and the Licensing of Argumental Wh Phrases in Spanish.
Natural Language and Linguistic Theory 12, 335372.

Delimitation Effects and the Cartography of the Left Periphery


Starke, Michal
2001

145

Merge Dissolves into Move. Doctoral dissertation, University of


Geneva.

Taraldsen, Tarald
1978
On the NIC, Vacuous Application and the That-Trace Filter. Bloomington: Indiana University Linguistics Club.
Taraldsen, Tarald
2001
Subject Extraction, the Distribution of Expletives, and Stylistic Inversion. In: Aafke Hulk and Jean-Yves Pollock (eds.), Inversion in
Romance, 163182. Oxford: Oxford University Press.
Tsai, Dylan
2007
Left Periphery and Why-How Alternations. Ms., National Tsing-Hua
University, Taiwan.
Vallduv, Enric
1992
The Informational Component. (Outstanding Dissertations in Linguistic). New York: Garland.

Sentence Types and the Japanese Right Periphery*


Mamoru Saito

1.

Introduction

In this paper, I examine the complementizer system of Japanese and present a


preliminary hypothesis on the structure of the Japanese right periphery. I discuss
three complementizers, to, ka, and no. Examples of their occurrences are shown
in (1).
(1)

a.

b.

c.

Taroo-wa [CP Hanako-ga Ziroo-ni atta to] omotteiru.


H.-nom
Z.-dat met to think
T.-top
Taroo thinks that Hanako met Ziroo.
Taroo-wa [CP Hanako-ga dare-ni atta ka] tazuneta.
H.-nom
who-dat met ka inquired
T.-top
Taroo asked who Hanako met.
Taroo-wa [CP Ziroo-ni atta no]-o kookaisiteiru.
T.-top
Z.-dat met no-acc regret
Taroo regrets that he met Ziroo.

To is often considered the complementizer for embedded propositions as it


appears in the CP complements of verbs such as omou think and yuu say.
However, I first show that it is employed for paraphrases of quotes in the sense
of Plann (1982) or reports of direct discourse in the sense of Lahiri (1991).
Given this, I re-examine no and argue that it instead is the complementizer for
embedded propositions. The analysis to be proposed is summarized in (2).1

* This paper was presented at the International Conference on Sentence Types: Ten
Years After, held at the University of Frankfurt on June 2628, 2009. I would like to
thank the audience, especially NicholasAsher, Adriana Belletti, Gunther Grewendorf,
Paul Portner, and Peter Sells, for helpful comments.
1. This is an extension of the analysis proposed in Kuno (1973, 1988). It becomes clear
in the following pages where I depart from his analysis.

148

Mamoru Saito

(2)

a.
b.
c.

to is the complementizer for paraphrases or reports of direct


discourse.
ka is the complementizer for questions.
no is the complementizer for propositions.

A well-known peculiar property of to is that it can follow questions as in (3).


(3)

kuru
Taroo-wa Ziroo-ni [CP dare-ga kare-no ie-ni
who-nom he-gen house-to come
T.-top
Z.-dat
ka to] tazuneta.
ka to inquired
Lit. Taroo asked Ziroo that who is coming to his house.

A similar phenomenon is observed in Spanish, as discussed in Plann (1982).2


In (4a) que heads a propositional CP but it precedes embedded questions in
(4b)(4c).
(4)

a.

b.

c.

Saba
que coria.
knew-3sg que run-3sg
He knew that he was running.
Te preguntan que para que quieres
el prestamo.
you ask-3pl
que for what want-2sg the loan
They ask you what you want the loan for.
Penso
que cuales
seran
adecuados.
thought-3sg que which ones would be appropriate
He wondered which ones would be appropriate.

Examining examples of this kind in detail, Plann proposes that que is ambiguous between complementizers for propositions and for paraphrases of quotes.
According to her analysis, Spanish has three distinct complementizers as in (5).
(5)

a.
b.
c.

que for paraphrases of quotes


Null [+Q] C for questions
que for propositions

What I propose in this paper is that Japanese has an identical complementizer


system. Japanese in fact provides explicit evidence for Planns analysis as the
three complementizers have distinct phonetic realizations: To corresponds to
que in (5a), ka is the [+Q] C, and no is the counterpart of que in (5c).
2. Thanks are due to Kensuke Takita for pointing out the relevance of Plann (1982) and
the parallelism between the Japanese to and the Spanish que.

Sentence Types and the Japanese Right Periphery

149

In the second part of the paper, I examine the co-occurrence restrictions on


the three complementizers, to, ka and no. In addition to the ka-to sequence
observed in (3), there are examples with all three complementizers, as shown
in (6).
(6)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (no) ka (to)]


he-gen sister-nom there-in was no ka to
T.-top
minna-ni tazuneta.
all-dat inquired
Taroo asked everyone if his sister was there.

The three complementizers always appear in the order no-ka-to, and this suggests
the recursive CP structure in (7).
(7)

[CP [CP [CP [TP . . . ] no] ka] to]

Then, I consider the distribution of thematic topics and present evidence that
they can appear in CPs headed by to or ka, but not in CPs headed by no. This
leads to the hypothesis that there is a Topic head, located between no and ka as
in (8), that hosts thematic topics in its Spec.
(8)

[CP [CP [CP [CP [TP . . . ] no] e[+TOPIC] ] ka] to]

The similarity between (8) and the structure of the Italian left periphery proposed
in Rizzi (1997) is evident. His proposal is shown in (9).
(9)

[Force [(Topic*) [(Focus) [(Topic*) [Finite [TP . . . ]]]]]]

If ka is Force and no is Finite, Japanese is identical to Italian except for the presence of to and the absence of Focus.3 I suggest then that the Japanese periphery
is comparable to Italian with the addition of the highest C, to, which is equivalent
to the Spanish que as a marker of paraphrase or report of direct discourse.
This conclusion, if correct, provides additional evidence for the universality of
the structure of the left/right periphery.
In the following section, I discuss the parallelism between to and the Spanish
que in some detail. As noted above, they can both take question CPs as complements. Rivero (1994) presents examples where imperatives follow que in
support of Planns (1982) analysis. Kuno (1988) notes similar facts in Japanese
3. The C heads in Japanese appear in the mirror image of Italian because of the headparameter. As discussed below in Section 4, it is argued in Kuroda (1988) and Saito
(2007) that Japanese allows multiple thematic topics. The basic pattern of focusing
in Japanese is like English: Any phrase with stress is interpreted as focus in situ.

150

Mamoru Saito

and argues that the complement of to can be a blended discourse, which starts
out as a regular embedded sentence and ends with a verb that expresses a request.
I re-examine those facts and show that they too provide supporting evidence for
Planns analysis. In Section 3, I turn to no. Kuno (1973) argues that the CPs
headed by this complementizer typically carry factive presuppositions in the
sense of Kiparsky and Kiparsky (1970). I first show that the distribution of no
is much wider than his analysis implies. Then, I argue that it should be considered the complementizer for propositional complements. Section 4 concerns
the structure of the Japanese right periphery. I discuss the co-occurrence restrictions on the three complementizers, to, ka and no, and also the distribution of
thematic topics. This leads to the hypothesis on the structure of the Japanese
right periphery alluded to above. Section 5 concludes the paper.
2.

To as a Complementizer for Paraphrases of Direct Discourse

In this section, I examine cases where to takes questions and expressions of


request as complements and argue that it is a complementizer for paraphrase
or report of direct discourse just like que in Spanish. Section 2.1. concerns
examples in which to follows questions and Section 2.2. deals with those in
which it follows expressions of request.
2.1. The Parallelism of to and que with Respect to Question Complements
Another well-known property of to is that it is employed for direct discourse as
well as indirect discourse. It follows a direct quotation in (10a), while it marks
an indirect discourse in (10b).
(10)

a.

b.

Hanako-ga, Watasi-wa tensai da, to itta /omotta


H.-nom
I-top
genius is
to said/thought
(koto)
fact
(the fact that) Hanako said/thought, Im a genius.
Hanako-ga [CP zibun-ga tensai da to] itta /omotta
self-nom genius is to said/thought
H.-nom
(koto)
fact
(the fact that) Hanako said/thought that she is a genius

A CP complement as in (10b) is usually considered to represent a proposition.


This is because to heads the CP complements of typical bridge verbs, as noted

Sentence Types and the Japanese Right Periphery

151

above. On the other hand, a different complementizer ka is required for CP


complements when the matrix verbs select questions. This is illustrated in (11).
(11)

a.

b.

c.

kuru]
Taroo-wa [CP [TP Hanako-ga kare-no ie-ni
H.-nom
he-gen house-to come
T.-top
to/*ka] omotteiru.
to ka think
Taroo thinks that Hanako is coming to his house.
Taroo-wa Ziroo-ni [CP [TP Hanako-ga kare-no ie-ni
T.-top
Z.-dat
H.-nom
he-gen house-to
kuru] *to/ka] tazuneta.
come to ka inquired
Taroo asked Ziroo if Hanako is coming to his house.
kuru]
Taroo-wa [CP [TP Hanako-ga kare-no ie-ni
H.-nom
he-gen house-to come
T.-top
*to/ka] siritagatteiru.
to ka want to know
Taroo wants to know if Hanako is coming to his house.

The matrix verbs, tazuneru inquireand siritagatteiru want to know, in (11b)


(11c) select questions, and their CP complements must contain ka. It then looks
like to is a [Q] C while ka is a [+Q] C.
However, as noted above, the [+Q] C ka can be followed by to in some cases.
To embeds a yes/no question in (12a) and a wh-question in (12b).
(12)

a.

b.

Taroo-wa Ziroo-ni [CP [CP [TP Hanako-ga kare-no


H.-nom
he-gen
T.-top
Z.-dat
ie-ni
kuru] ka] to] tazuneta.
house-to come ka to inquired
Lit. Taroo asked Ziroo that if Hanako is coming to his house.
Taroo-wa Ziroo-ni [CP [CP [TP dare-ga kare-no ie-ni
who-nom he-gen house-to
T.-top
Z.-dat
kuru] ka] to] tazuneta.
come ka to inquired
Lit. Taroo asked Ziroo that who is coming to his house.

As the matrix verb tazuneru inquireselects for a question CP as shown in (11b),


it appears that the verb and the question complementizer ka enter into selectional
relation across to in (12). It is then tempting to hypothesize that to is unspecified
with respect to [Q] and is transparent for the purpose of selection. But this
cannot be the correct analysis because not all verbs that select questions allow

152

Mamoru Saito

the ka-to sequence. (13a)(13b) show that (12a)(12b) become ungrammatical


when siritagatteiru want to know is substituted for tazuneta inquired.
(13)

kuru]
a. *Taroo-wa [CP [CP [TP Hanako-ga kare-no ie-ni
H.-nom
he-gen house-to come
T.-top
ka] to] siritagatteiru.
ka to want to know
Lit. Taroo wants to know that if Hanako is coming to his house.
b. *Taroo-wa [CP [CP [TP dare-ga kare-no ie-ni
kuru]
T.-top
who-nom he-gen house-to come
ka] to] siritagatteiru.
ka to want to know
Lit. Taroo wants to know that who is coming to his house.

It is then necessary to examine which matrix verbs allow the ka-to sequence to
find out what is going on in examples such as (12).
At this point, Planns (1982) analysis of Spanish que mentioned above becomes quite relevant. The examples in (4b)(4c), where que takes a question
CP complement, are repeated below as (14a)(14b).
(14)

a.

b.

Te preguntan que para que quieres


el prestamo.
you ask-3pl
que for what want-2sg the loan
They ask you what you want the loan for.
Penso
que cuales
seran
adecuados.
thought-3sg que which ones would be appropriate
He wondered which ones would be appropriate.

Plann notes that only a subset of those verbs that select for question CPs allow
the presence of que. (15) shows some cases where que cannot occur.
(15)

Ya
supieron /entendieron /recordaron
(*que)
already found out-3pl/understood-3pl/remember-3pl
que
por que lo habas hecho.
why
it had-2sg done
They already found out/understood/remembered why you had done it.

Examining more relevant examples, she draws the generalization that que can
take a question CP as a complement only when the matrix verb is a verb of
saying or thinking, that is, a verb that is compatible with direct quotation. Based
on this, she goes on to propose that the que-headed CPs express paraphrases of
direct discourse in this case.

Sentence Types and the Japanese Right Periphery

153

Planns generalization and analysis are directly applicable to Japanese. A


partial list of the matrix predicates that allow the ka-to sequence, that is, a toheaded CP with a question complement, is given in (16a). On the other hand,
the predicates in (16b) are incompatible with the ka-to sequence.
(16)

a.

ka-to:

b. *ka-to:

kiku ask, situmonsuru question, yuu say,


sakebu scream, omou think
tyoosasuru investigate, hakkensuru discover,
rikaisuru understand, siranai dont know

The predicates in (16a) are verbs of saying and thinking, and those in (16b) are
not. The former can occur with direct quotes, and the latter cannot, as illustrated
in (17).
(17)

a.

Taroo-wa, Boku-ga soko-ni ikimasyoo ka to itta.


T.-top
I-nom
there-to shall go
ka to said
Taroo said, Shall I go there?
b. *Taroo-wa, Dare-ga soko-ni ikimasu ka to siranai.
T.-top
who-nom there-to go
ka to not know
Lit. Taroo doesnt know, Who is going there?

Thus, to can take a question CP as a complement in exactly the same context


as que. Then, it too should be analyzed as a complementizer for paraphrases
of direct discourse.4 In the following subsection, I examine another peculiar
property of to discussed in Kuno (1988). I show that it provides further evidence
for this analysis of to, and also for the parallelism between to and que.
2.2. Kuno (1988) on blended and quasi-direct discourse
Kuno (1988) examines examples such as (18a), and argues that a to-headed CP
complement can represent a blended discourse, which starts out as indirect
and shifts to direct.
(18)

a.

Taroo-wa zibun-no uti-ni


kite kure
to Ziroo-ni
T.-top
self-gen home-to come for me to Z.-dat
itta.
said
Lit. Taroo said to Ziroo that come to self s house.

4. It seems that complementizers of this kind are quite widespread. See, for example, Jayaseelan (2008) for relevant discussion on Malayalam, and Grewendorf and
Poletto (2009) for a similar phenomenon in Cimbrian, a German dialect spoken in
northeastern Italy.

154

Mamoru Saito

b.

Taroo-wa, Boku-no uti-ni


kite kure, to Ziroo-ni
T.-top
I-gen
home-to come for me to Z.-dat
itta.
said
Taroo said to Ziroo, Come to my house.

He assumes that the predicate of the embedded clause of (18a) represents a


direct discourse as it expresses a request. On the other hand, the initial part of
the clause must be indirect because zibun self takes the matrix subject Taroo
as its antecedent. If it were a direct quotation of Taroos utterance, the first
person pronoun boku I should occur instead of zibun as shown in (18b). In
this subsection, I argue that Kunos blended discourse is indirect discourse,
and that the grammaticality of examples such as (18a) is indeed predicted by
the analysis of to as a complementizer for paraphrases of direct discourse.
Although Kuno analyzes the embedded clause of (18a) as blended discourse, he also points out that the direct part cannot be a direct quotation of
Taroos utterance. Note first that expressions of request vary in form in accordance with the degree of politeness, as illustrated in (19).
(19)

a.

b.

#Taroo-wa, Boku-no uti-ni


kite kure, to
T.-top
I-gen
home-to come for me to
Ito-sensei-ni itta.
I.-Prof.-dat said
Taroo said to Prof. Ito, Come to my house.
Taroo-wa, Watasi-no uti-ni
oide itadakemasu ka
T.-top
I-gen
home-to come for me (polite) ka
to Ito-sensei-ni itta.
to I.-Prof.-dat said
Taroo said to Prof. Ito, Could you to come to my house?

(19a) is not an appropriate utterance of Taroo, a student, to his teacher, Prof.


Ito, because kite kure come for me is a non-polite, neutral expression. (19b)
shows what Taroo would actually say in this context. What Kuno observes is
that the judgments, interestingly, are reversed when direct discourse is turned
into blended discourse. (20) confirms this observation.
(20)

a.

kite kure to
Taroo-wa zibun-no uti-ni
T.-top
self-gen home-to come for me to
Ito-sensei-ni itta.
I.-Prof.-dat said
Lit. Taroo said to Prof. Ito that come to self s house.

Sentence Types and the Japanese Right Periphery

155

b. *Taroo-wa zibun-no uti-ni


oide itadakemasu ka to
T.-top
self-gen home-to come for me (polite) ka to
Ito-sensei-ni itta.
I.-Prof.-dat said
Lit. Taroo said to Prof. Ito that could you come to my house?
(20a) contains the non-polite, neutral form, kite kure come for me, and is
perfectly grammatical. On the other hand, (20b) with the polite form is not.
Kuno concludes then that the direct part of blended discourse is not precisely
direct but only quasi-direct.
Kuno goes on to discuss why polite expressions are not allowed in blended
discourse. His answer is that this is because polite forms of verbs do not appear
in embedded clauses, as shown in (21).
(21)

kaimasita]
hon]-o
a. *Watasi-wa [NP [ kinoo
yesterday bought (polite) book-acc
I-top
yomimasita.
read (polite)
I read the book I bought yesterday.
b. Watasi-wa [NP [ kinoo
katta]
hon]-o
I-top
yesterday bought (neutral) book-acc
yomimasita.
read (polite)
I read the book I bought yesterday.

The sentences in (20) are polite expressions as the matrix verb is in the polite
form. Yet, the verb in the relative clause must be in the neutral form as the
contrast between (21a) and (21b) indicates. Kunos analysis is that the polite
form of the expression of request in (20b) is excluded for the same reason.
This analysis suggests that blended discourse is after all indirect discourse.
This is so because it patterns with embedded clauses while direct discourse is
known to have matrix properties. Then, the remaining question is why to can
embed a sentence expressing a request. This is mysterious if a CP headed by to
stands for a proposition. The following English example is totally out:
(22)

*John said to Mary that (please) come to his house.

However, the answer is straightforward given the analysis of to proposed in the


preceding subsection. To, unlike that, is a complementizer for paraphrases of
direct discourse. The direct discourse that is paraphrased can be an expression
of request as well as a question. Hence, the analysis predicts correctly that an

156

Mamoru Saito

expression of request can appear as the complement of to just as a question


can.
It was argued in the preceding subsection that to is just like que in Spanish.
Then, it is predicted that blended discourse is observed in Spanish as well.
This prediction is indeed borne out. Rivero (1994) presents examples in which
que takes imperative complements in support of Planns (1982) analysis. One
of her examples is shown in (23a), together with its direct discourse counterpart
in (23b).
(23)

a.

b.

Dijo
que a no molestarle.
said-3sg que to not bother-him
He said not to bother him.
Dijo,
A no molestarme!
said-3sg to not bother-me
He said, Dont bother me!

In (23a), the embedded object clitic can co-refer with the matrix subject, and in
this case, the embedded clause must represent indirect discourse despite the fact
that it is an imperative. Riveros conclusion is in fact identical to the one drawn
above for blended discourse in Japanese. As que can be a complementizer
for paraphrases of direct discourse, it is not at all surprising that it can take
imperatives as complements. Thus, the comparison of Kuno (1988) and Rivero
(1994) leads to another parallelism between to and que.
3.

No as a Complementizer for Propositions

It was shown above that to can be a complementizer for paraphrases of direct


discourse exactly like que.According to Plann (1982), que is ambiguous between
a complementizer for paraphrases and a complementizer for propositions. In this
section, I argue that to is specialized for the former function. More precisely, I
argue that there is a division of labor between to and another complementizer
no: to is for paraphrases and no is for propositions.
3.1. Clausal Complements Designating States, Events, and Actions
An example of que with an embedded proposition was shown in (4a), repeated
below as (24).
(24)

Saba
que coria.
knew-3sg que run-3sg
He knew that he was running.

Sentence Types and the Japanese Right Periphery

157

In a parallel context, the complementizer no is employed in Japanese, as shown


in (25).5
(25)

Taroo-wa [CP [TP Hanako-ga soko-ni iru no]-o sitteita.


T.-top
H.-nom
there-in is no-acc knew
Taroo knew that Hanako was there.

This raises the possibility that to, unlike que, is unambiguously a complementizer for paraphrases of direct discourse, and that no is the complementizer for
propositions. In this section, I argue that this is indeed the case.
Discussing the distributions of to and no, Kuno (1973) presents a rough
generalization that no is associated with a factive presupposition in the sense of
Kiparsky and Kiparsky (1970) while this is never the case with to.6 However, the
distribution of no is much wider than this suggests. Partial lists of the predicates
that take to-headed CP complements and those that appear with no-headed CPs
are shown in (26).
(26)

a.

b.

c.

predicates that take to-headed CP complements


omou think, kangaeru consider, sinziru believe, yuu say,
sakebu scream, syutyoosuru claim, insist, tazuneru inquire,
kitaisuru expect, kanziru feel
predicates that take no-headed CP complements
wasureru forget, kookaisuru regret, miru see, matu wait,
tamerau hesitate, kyohisuru refuse, ukeireru accept,
kitaisuru expect, kanziru feel
predicates that take no-headed CP subjects
akiraka clear, kanoo possible, kantan easy,
muzukasii difficult, taihen big deal

Typical factive verbs such as wasureru forget and kookaisuru regret take CP
complements headed by no. But they are clearly a minority in (26b).
Then, what would be the proper characterization of the distributions of to
and no? First, the predicates in (26a) are all verbs of saying and thinking. They
are indeed all compatible with direct quotation. A couple of examples are given
in (27).

5. No is nominal in nature and requires a Case marker when it heads a CP in an argument


position. I present one of Murasugis (1991) arguments in the following subsection
that it should still be considered a complementizer rather than a noun.
6. He also considers koto, which has a similar, though not identical, distribution as no.
I do not discuss it here as it is fairly clear that it is a noun.

158
(27)

Mamoru Saito

a.

b.

Taroo-wa, Boku-no uti-ni


atumatte
T.-top
I-gen
house-at gather
sakenda.
screamed
Taroo screamed, Gather at my house!
Hanako-wa, Watasi-ga Taroo-ni au,
H.-top
I-nom
T.-dat
meet
Hanako insisted, I will go see Taroo.

kure, to
for me to

to syutyoosita.
to insisted

It seems then that to always serves as a complementizer for paraphrases of direct


discourse when it heads a complement CP.
The no-headed CPs that appear with the predicates in (26b), on the other
hand, all seem to represent events, states, or actions. For example, one regrets
a past event or a present state, sees a present event or state, waits for a future
event or state, and hesitates to perform an action. The same can be said of those
no-headed CPs in the subject position. What can be clear is the existence (or
non-existence) of an event or state in the past, present, or future. Similarly, what
can be easy or difficult is to perform a certain action. In short, those CPs headed
by no seem to always represent propositions. This is consistent with Kunos
observation that factive verbs take no-headed CPs and never to-headed CPs as
complements. Only sentences that express propositions can be presupposed to
be true. Hence, those verbs are compatible only with no-headed CPs.
Then, there is a division of labor between to and no: to is the complementizer for paraphrases of direct discourse while no is the complementizer for
propositions. An additional piece of evidence for this can be found in data from
child language. As discussed extensively in the acquisition literature, the overgeneration of no in relative clauses is widely observed with 24 year olds. The
following examples are from Murasugi (1991):
(28)

a.

b.

[ohana motteru *no] wanwa (2;6)


flower have
no doggie
the doggie that is holding flowers
[buta-san tataiteru *no] taiko (2;11)
Mr. Pig is hitting no drum
the drum that the pig is playing

These examples are ungrammatical in adult Japanese with no. Murasugi examines the properties of the overgenerated no in detail, and argues that it is
a complementizer. According to her analysis, relative clauses are TPs in adult
Japanese. However, children at one point hypothesize that they are CPs, just
like English relative clauses, and hence, place no in their head positions. They

Sentence Types and the Japanese Right Periphery

159

only later discover that there is no position for a complementizer in Japanese


relatives and cease to overgenerate no.
One question that arises with this analysis is why no, and not to, is overgenerated in the head positions of relative clauses. Murasugi (2009) addresses
this question, referring to Schachters (1973) observation that many languages
employ the same complementizer in relative clauses and clefts. No appears in
Japanese clefts as shown in (29).
(29)

[CP Nimotu-ga
todoita no]-wa Nagoya-kara da.
package-nom arrived no-top N.-from
is
It is from Nagoya that a package arrived.

Then, given Schachters generalization, it is not surprising that children overgenerate no. But one may ask further why it is that no, and not to, is employed
in clefts and childrens relative clauses. And for this, the analysis of to as a complementizer for paraphrases of direct discourse provides a clear answer. The
subject CP of a cleft sentence expresses a proposition and is not a paraphrase
of direct discourse. Hence, no must be employed. There is simply no way for to
to appear in this context. Similarly, a relative clause does not paraphrase a direct discourse. Then, children could not overgenerate to in relative clauses. This
account holds if to is never a complementizer for propositions and is employed
exclusively as a complementizer for paraphrases of direct discourse, as argued
here.
3.2. The Nominal Nature of no and its Complementizer Status
I argued above that no is the complementizer for propositions in Japanese. A
CP headed by no requires Case when it is in an argument position as noted in
Footnote 5, and it is often called a nominalizer in part for this reason. Although
whether no is a complementizer or a noun does not affect the overall discussion
in this paper, I would like to briefly comment on its complementizer status in
this subsection.
The Case property of no just mentioned clearly indicates that it is nominal in
nature.7 However, it does not provide decisive evidence that it is a noun. First,
it is known that complementizers vary with respect to their Case properties. For
example, as discussed in detail in Stowell (1981), English CPs headed by that
do not appear in typical Case positions like the object position of a preposition,
7. Another relevant fact is that the predicates that take no-headed CP complements correspond roughly to those in English that take gerunds as complements. See Rosenbaum
(1967) for detailed discussion on the latter.

160

Mamoru Saito

but there is no such restriction with question CPs. Relevant examples are shown
in (30).
(30)

a. *They talked about [CP that Mary is a genius].


b. They talked about [CP whether Mary is a genius].

Thus, Stowell concludes that that-headed CPs cannot be Case marked while
question CPs can be. Note that question CPs only allow Case and do not require
Case. The following examples are perfectly grammatical though the embedded
CPs are not in Case positions:
(31)

a.
b.

They wonder [CP whether Mary is a genius].


(cf. *They wonder it)
It was debated [CP whether Mary is a genius].
(cf. *It was debated it)

A parallel observation can be made with to-headed CPs and question CPs in
Japanese. Thus, the former cannot appear in the object position of a postposition
but the latter can, as shown in (32).
(32)

a. *Karera-wa [CP Hanako-ga soko-ni iku to]-nituite


they-top
H.-nom
there-to go to-about
kangaeta.
considered
They thought about Hanako going there.
b. Karera-wa [CP Hanako-ga soko-ni iku beki
they-top
H.-nom
there-to go should
ka]-nituite kangaeta.
ka-about considered
They thought about whether Hanako should go there.

However, no adds to the paradigm in the case of Japanese. That is, to resists
Case, ka allows Case, and no requires Case, as shown in (33).8
(33)

a.

Karera-wa [CP Hanako-ga soko-ni iku to](*-o) omotta.


they-top
H.-nom
there-to go to-acc thought
They thought that Hanako was going there.

8. The accusative -o is often omitted in colloquial style. But the contrast between (33b)
and (33c) is quite clear. The former is grammatical without -o in any register.

Sentence Types and the Japanese Right Periphery

b.

c.

Karera-wa [CP Hanako-ga doko-ni iku beki


they-top
H.-nom
where-to go should
kentoosita.
discussed
They discussed where Hanako should go.
Karera-wa [CP Hanako-ga soko-ni iru no]*(-o)
they-top
H.-nom
there-in is no-acc
They felt that Hanako was there.

161

ka](-o)
ka-acc

kanzita.
felt

Thus, the three-way distinction in (34) obtains.


(34)

a.
b.
c.

to cannot appear in a Case position.


ka can appear in a Case position.
no must appear in a Case position.

It seems difficult to account for this based on the categorial difference between
a complementizer and a noun. The whole paradigm instead seems to reflect the
lexical properties of the specific items.
Stronger arguments for the complementizer status of no are presented in
Murasugi (1991). One of them is based on childrens overgeneration of no,
discussed above. The relevant examples in (28) are repeated below in (35).
(35)

a.

b.

[ohana motteru *no] wanwa (2;6)


flower have
no doggie
the doggie that is holding flowers
[buta-san tataiteru *no] taiko (2;11)
Mr. Pig
is hitting no drum
the drum that the pig is playing

In order to examine the category of the overgenerated no, Murasugi considers


three possibilities; the genitive Case marker, a pronoun, and a complementizer,
which are all homophonous and realized as no. The genitive no appears after
any NP or PP within a nominal projection, as illustrated in (36).
(36)

a.

b.

c.

Taroo-no hon
T.-gen
book
Taroos book
Hanako-no yooroppa-e-no ryokoo
H.-gen
Europe-to-gen trip
Hanakos trip to Europe
midori-iro-no
kuruma
green-color-gen car
a green car

162

Mamoru Saito

The pronoun no, which roughly corresponds in meaning to one in English, is


observed in examples like (37a)(37b).
(37)

a.

b.

mizukasii no
Difficult one
a difficult one
[Taroo-ga katta] no
T.-nom bought one
the one that Taroo bought

Murasugi first excludes the genitive no by observing the overgeneration pattern


in Toyama Dialect, where the genitive is no as in most other dialects but the
pronoun and the complementizer are realized as ga. As the Toyama Dialect
speaking children overgenerate ga as in (38), the no in (35) cannot be the genitive
Case marker.
(38)

[anpanman tuitoru *ga] koppu (2;11)


(a character) attach
ga cup
a cup that is pictured with anpanman

Then, she presents an argument that it is not a pronoun either. Note first that
if the relative clauses in (35) and (38) are headed by the pronoun no /ga,
they must be NPs. This means that the genitive no is required between those
relatives and the head noun. Murasugi shows through an experimental study
that those children who overgenerate no /ga in relative clauses never fail to insert
the genitive no after an NP modifying an N. Then, if the relative clauses are
indeed NPs, the children must insert the genitive no after those relatives, but
they never do. She concludes then that the overgenerated no /ga cannot be a
pronoun and hence must be a complementizer.
This argument against the analysis of the overgenerated no /ga as a pronoun
suggests simultaneously that the complementizer no /ga cannot be a noun.
Suppose that the children overgenerate the complementizer no /ga in relative
clauses as Murasugi argued. If the complementizer no /ga is a noun, then
the relative clauses must be NPs. Then, again, the children must insert the
genitive no after those relative clauses. Since they do not, it is clear that the
children do not consider the complementizer no /ga a noun. What the children
overgenerate must be no /ga of the category complementizer. This constitutes
indirect but strong evidence that the complementizer no /ga is not a noun but
is indeed a complementizer in adult grammar as well.

Sentence Types and the Japanese Right Periphery

4.

163

Preliminary Notes on the Japanese Right Periphery

It was argued in the preceding sections that Japanese has the three complementizers in (39).
(39)

a.
b.
c.

to is the complementizer for paraphrases of direct discourse in the


sense of Plann (1982).
ka is the complementizer for CPs that represent questions.
no is a complementizer for CPs that represent propositions.

In Section 4.1., I consider the hierarchical relation among those complementizers


and suggest, following Hiraiwa and Ishihara (2002), that no is the Finite head.
Then, in Section 4.2., I argue that there is a Topic head above the Finite no and
below ka, which I consider to be a Force head. This leads to the conclusion that
the Japanese right periphery is remarkably similar in structure to the Italian left
periphery discussed in Rizzi (1997).
4.1. CP Recursion at the Right Periphery
As noted above, an embedded CP in Japanese can contain a sequence of the
complementizers ka and to. Another example is shown in (40).
(40)

Taroo-wa [CP kare-no imooto-ga soko-ni ita ka (to)]


he-gen sister-nom there-in was ka to
T.-top
minna-ni tazuneta
all-dat inquired
Taroo asked everyone if his sister was there.

The number of complementizers in a single CP is not limited to two. There are


in fact cases where all three complementizers appear. (40), for example, can
have no preceding ka, as in (41).
(41)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (no) ka (to)]


T.-top
he-gen sister-nom there-in was no ka to
minna-ni tazuneta
all-dat inquired
Taroo asked everyone if his sister was there.

This example instantiates three kinds of complementizer sequences, no-ka, kato, and no-ka-to. Whenever there are multiple complementizers, their order is
fixed in this way. This suggests that Japanese CPs can have the recursive structure
in (42).

164
(42)

Mamoru Saito

[CP [CP [CP . . . no] ka] to]

The structure in (42) undoubtedly has a semantic basis. Ka can select a CP


headed by no as a question can be formed on a proposition. To, in turn, can
select a CP headed by ka as to embeds a paraphrase of a direct discourse and
the direct discourse can be a question, as discussed in detail in Section 2. One
case that (42) allows but is missing is the no-to sequence. (43) is ungrammatical
with no.
(43)

Taroo-wa [CP kare-no imooto-ga soko-ni ita (*no) to]


he-gen sister-nom there-in was
no to
T.-top
omotta.
thought
Taroo thought that his sister was there.

This too is expected because a no-headed CP represents a proposition while to


embeds a paraphrase of direct discourse. Other orderings of complementizers
are plausibly excluded in similar ways. For example, the ka-no sequence is illicit
as the content of a proposition cannot be a question.
Given the hierarchy in (42), it is tempting to compare it with the structure
of the Italian left periphery proposed in Rizzi (1997). As noted at the outset of
this paper, he proposes the structure in (44).
(44)

[Force [(Topic*) [(Focus) [(Topic*) [Finite [TP . . . ]]]]]]

Ka, being the question marker, is plausibly a Force head. To occupies a higher C
position that does not appear in (44). Let us call this C Report, following Lahiri
(1991). Finally, no is analyzed as Finite in Hiraiwa and Ishihara (2002). As it
occupies the lowest C position in (42), this fits the hierarchy in (44) well.9 Further, childrens overgeneration of no in relative clauses discussed above receives
a straightforward interpretation under this analysis. It is unclear what force relative clauses have, and it is unlikely that children consider relative clauses ForceP.
On the other hand, it is not surprising if children produce them as FiniteP. Then,
they would overgenerate no in the head position. In the remainder of this subsection, I introduce another piece of suggestive evidence from Matsumoto (2010)
for this anaysis of no.
Matsumoto examines the types of sentential complements no can take and
shows that they are more limited when compared with ka and to. In particular,
9. Thanks to Adriana Belletti for pointing out the relevance of Hiraiwa and Ishihara
(2002) in this context. I do not discuss their argument here because it is based on an
attractive and yet controversial analysis of Japanese clefts.

Sentence Types and the Japanese Right Periphery

165

she argues that the complement of no must be headed by a morphologically


overt T. She first notes that there are modal-like words that do not inflect for
tense. Daroo it is probably the case that in (45) is one such element.
(45)

a.

b.

Taroo-wa soko-ni iru daroo.


T.-top
there-at is daroo
Taroo probably is there.
Taroo-wa soko-ni ita daroo.
T.-top
there-at was daroo
Taroo probably was there.

The complement of daroo can be in present or past, but daroo itself does not
carry tense. And interestingly, sentences headed by daroo can be embedded
under ka or to, but not under no. This is shown in (46).
(46)

a.

b.

c.

Taroo-wa [CP ame-ga huru (daroo) ka] kangaeta.


T.-nom
rain-nom fall
daroo ka considered
Taroo considered whether it would rain.
Taroo-wa [CP ame-ga huru (daroo) to] omotta.
rain-nom fall
daroo to thought
T.-top
Taroo thought that it would rain.
Taroo-wa [CP ame-ga huru (*daroo) no]-o kitaisita.
rain-nom fall
daroo no-acc expected
T.-top
Taroo hoped that it would rain.

The ungrammaticality of (46c) with daroo is likely to be due to its incompatibility with no. The example is fine without daroo. Further, as Matsumoto points
out, it becomes grammatical also when the formal noun koto is substituted for
no as in (47).
(47)

Taroo-wa [ame-ga huru (daroo) koto]-o kitaisita.


T.-top
rain-nom fall
daroo N-acc expected
Taroo hoped that it would rain.

Although koto literally means matter, state or fact, it has little semantic
content in this context.10 Thus, there is basically no difference in meaning between (46c) and (47). It seems then that (46c) is out because no, specifically,
cannot take a clausal complement headed by daroo.
On the basis of observations like this, Matsumoto concludes that no can
only take clausal complements that are headed by Tense. This is expected if
10. Kuno (1973) considers it a complementizer, as noted in Footnote 6.

166

Mamoru Saito

no is Finite, because Finite is by definition the C that is closely related with T.


Matsumotos discussion thus provides suggestive evidence that no is in fact the
Finite head.
4.2. The Position of Topic in the CP Structure
The hypothesis arrived at so far on the Japanese right periphery is shown in (48).
(48)

[. . . [. . . [. . . Finite (no)] Force (ka)] Report (to)]

In this subsection, I consider how Topic fits into this structure. More specifically,
I argue that Topic heads can be generated above Finite and below Force as in
(49).
(49)

[. . . [. . . [. . . [. . . Finite (no)] (Topic*)] Force (ka)] Report (to)]

A classical analysis of topic and focus in Japanese is found in Kuno (1973).


Any stressed phrase receives focus interpretation in situ in this language. But
Kuno discusses one case where focus interpretation seems to arise in a specific
position in the sentence. Let us consider the examples in (50) for illustration.
(50)

a.

b.

Hanako-ga sono hon-o


yonde ita.
H.-nom
that book-acc reading was
Hanako was reading that book.
Hanako-ga sono hon-ga
suki da.
H.-nom
that book-nom like
It is Hanako that likes that book.

(50a) can be a neutral description of a past progressive event. On the other


hand, (50b) only has the interpretation with focus on the subject Hanako.11
Examining examples of this kind extensively, Kuno proposes the generalization
that a matrix-initial nominative phrase receives focus when the predicate is
stative. The generalization is on matrix clauses and it does not apply to embedded
clauses. Thus, Hanako need not be interpreted as focus when (50b) is embedded
as in (51).
(51)

katta (koto).
Taroo-ga [NP [Hanako-ga suki na] hon]-o
H.-nom
like
book-acc bought fact
T.-nom
(the fact that) Taroo bought a book that Hanako likes.

11. Kuno (1973) refers to this kind of focus as exhaustive-listing focus.

Sentence Types and the Japanese Right Periphery

167

Kuno (1973) also discusses topics that are marked by the particle -wa, and notes
that they can receive two distinct interpretations. Taroo in (52a), for example,
can be interpreted as a thematic topic or as a contrastive topic.
(52)

a.

b.

Taroo-wa sono hon-o


yonda.
T.-top
that book-acc read
A. As for Taroo, he read that book. (thematic topic)
B.
Taroo read that book, but I dont know about the other
people. (contrastive topic)
Taroo-ga sono hon-wa yonda.
T.-nom that book-top read
Taroo read that book, but I dont know about the other books.
(contrastive topic)

He observes in addition that while the contrastive topic interpretation is always


available, only a sentence-initial wa-phrase can be construed as a thematic topic.
The object is marked by -wa in (52b), and it can only be a contrastive topic. It
must be placed at the sentence-initial position as in (53) to receive the thematic
topic interpretation.
(53)

Sono hon-wa Taroo-ga yonda.


that book-top T.-nom read
A. As for that book, Taroo read it. (thematic topic)
B. Taroo read that book, but I dont know about the other books.
(contrastive topic)

Kuno notes that here too, sentence-initial means matrix-initial. The topic in the
initial position of the relative clause in (54) cannot be construed as a thematic
topic.
(54)

katta (koto).
Taroo-ga [NP [Hanako-wa suki na] hon]-o
H.-top
like
book-acc bought fact
T.-nom
Taroo bought a book that Hanako, though probably not the others,
like. (contrastive topic)

Building on Kunos observations, Heycock (1994) argues that there is no sentence-initial focus position in Japanese and proposes to analyze the focus interpretation discussed above in the mapping from syntax to information structure.
This analysis is extended to thematic topics in Heycock (2008). The strongest
piece of evidence for this approach is that those focus and thematic topic interpretations are matrix phenomena. If there were focus and topic positions in
Japanese, then they would be expected to occur in embedded clauses as well as

168

Mamoru Saito

matrix clauses. This is indeed the case in Italian, as the following example from
Rizzi (1997) shows:
(55)

Credo
che a Gianni, QUESTO, domani, gli dovremmo
I believe that to Gianni this
tomorrow we should
dire.
say
I believe that we should say this to Gianni tomorrow.

In this example, a Gianni and domani are topics and questo is a focus in the
complement CP. On the other hand, if foci and thematic topics are represented
in the information structure, it is not surprising that they occur only in matrix
clauses.
Heycocks argument is well taken for the obligatory focus interpretation
of sentence-initial nominative phrases as it is indeed observed only in matrix
clauses. It was shown in (51) that it is not observed in a relative clause. Other
types of embedded clauses exhibit the same pattern as shown in (56).
(56)

a.

b.

suki na no]-o
Taroo-ga [CP Hanako-ga sono hon-ga
H.-nom
that book-nom like
no-acc
T.-nom
wasureteita koto.
forgot
fact
The fact that Taroo forgot that Hanako likes that book.
suki da to]
Taroo-ga [CP Hanako-ga sono hon-ga
H.-nom
that book-nom like
to
T.-nom
sinziteiru koto.
believe
fact
The fact that Taroo believes that Hanako likes that book.

In these examples, focus interpretation is not forced on the embedded subject


Hanako. However, it has been known that there are cases where thematic topics
occur in embedded clauses. Typical examples are shown in (57).
(57)

a.

Taroo-ga [CP Hanako-wa zibun-no uti-ni


kuru to]
T.-nom
H.-top
self-gen home-to come to
sinziteiru koto.
believe
fact
A. (The fact that) Taroo believes that as for Hanako, she is
coming to his house. (thematic topic)
B.
(The fact that) Taroo believes that Hanako, though probably
not the others, is coming to his house. (contrastive topic)

Sentence Types and the Japanese Right Periphery

b.

169

Taroo-ga [CP Hanako-wa zibun-no uti-ni


ita to]
T.-nom
H.-top
self-gen home-at was to
itta koto.
said fact
A. (The fact that) Taroo said that as for Hanako, she was at his
house. (thematic topic)
B.
(The fact that) Taroo said that Hanako, though probably not
the others, was at his house. (contrastive topic)

In both of these examples, Hanako can easily be construed as a thematic topic


although it is embedded within a to-headed CP. Examples like these are often
just mentioned in footnotes as exceptions. But they seem to provide an important
clue for the analysis of thematic topics. First, the embedded CPs in (57) clearly
represent indirect discourse as the reflexive zibun self refers or at least can refer
to the matrix subject. Hence, the interpretive property of these examples cannot
be attributed to the fact that to can be a marker of direct quotation. Secondly,
these examples are after all very similar to the Italian (55), which shows that
embedded CPs can contain positions for foci and in particular topics.12
The contrast between (54) and (57) indeed suggests that there is a specific
position for thematic topics. (54) shows that thematic topics cannot occur in
relative clauses, which I assume are TPs. (57), on the other hand, suggests
that they can be located within CPs headed by to, which are the largest CPs
according to the recursive CP structure in (48). Then, if thematic topics occupy
the Spec position of a Topic head, the following structure accommodates the
data discussed so far:13
(58)

[. . . [TopicP Thematic Topic [Topic [TP . . . ] Topic]] Report (to)]

The position for thematic topics is outside TP, and hence they cannot occur
in relative clauses. But they can be present in CPs headed by to because the
position is contained within to-headed CPs. Given this kind of reasoning, it
should be possible to pinpoint the location of the Topic head by examining
whether thematic topics are possible in other types of CPs. Although the relevant
data require subtle judgment in some cases, they indicate that the Topic head is
located just above Finite and just below Force.
12. Heycock (2008), for example, does take this kind of exceptions seriously and suggests
that a detailed comparison with embedded verb-second in German may prove fruitful.
13. S.-Y. Kuroda assumed over the years that thematic topics are located in CP Spec.
Thus, the proposal made here is a refinement of his analysis. See in particular Kuroda
(1988) for relevant discussion.

170

Mamoru Saito

First, examples such as those in (59) show that thematic topics cannot be
licensed within no-headed CPs.
(59)

a.

b.

kuru
Taroo-ga [CP Hanako-wa zibun-no uti-ni
H.-top
self-gen home-to come
T.-nom
no]-o wasureteita koto.
no-acc forgot
fact
(The fact that) Taroo had forgotten that Hanako, though probably
not the others, was coming to his house. (contrastive topic)
Taroo-ga [CP Hanako-wa zibun-no uti-ni
hairu
T.-nom
H.-top
self-gen house-to enter
no]-o mita koto.
no-acc saw fact
(The fact that) Taroo saw Hanako, though not the others, enter
his house. (contrastive topic)

In both (59a) and (59b), the contrastive topic interpretation is forced on the
embedded subject Hanako. This indicates that the Topic head is not contained
within a CP headed by no, or more straightforwardly, a FiniteP.
On the other hand, the following examples suggest that the thematic interpretation of topics is possible within ka-headed CPs:
(60)

a.

b.

kata
ka]
Taroo-ga [CP Hanako-wa zibun-no hon-o
H.-top
self-gen book-acc bought ka
T.-nom
tazuneta koto.
inquired fact
A. (The fact that) Taroo asked if as for Hanako, she bought his
book. (thematic topic)
B.
(The fact that) Taroo asked if Hanako, though probably not
the others, bought his book. (contrastive topic)
Taroo-ga [CP Hanako-wa zibun-no uti-ni
kuru no
T.-nom
H.-top
self-gen home-to come no
ka] siritagatteiru koto.
ka want-to-know fact
A. (The fact that) Taroo wants to know if as for Hanako, she
is coming to his house. (thematic topic)
B.
(The fact that) Taroo wants to know if Hanako, though
probably not the others, is coming to his house. (contrastive
topic)

Sentence Types and the Japanese Right Periphery

171

It seems then that the Topic head is located within a ka-headed CP. This leads
to the more refined CP structure in (61).
(61)

[CP . . . [CP . . . [CP thematic topic [C [CP [TP . . . ] Finite (no)] Topic]]
Force (ka)] Report (to)]

(61) predicts correctly that thematic topics can occur in CPs headed by to or ka,
but not in CPs headed by no or TPs.
Further, there is evidence that the Topic projection is recursive just as in
Italian. As noted above, Kuno (1973) proposed a generalization that only a
sentence-initial wa-phrase can be construed as a thematic topic. This is consistent with (62), where only the subject can receive thematic interpretation.
(62)

Hanako-wa (kyonen) Teruabibu-e-wa itta.


H.-top
last year Tel Aviv-to-top went
A. As for Hanako, she went to Tel Aviv, but I dont know about the
other places. (Hanako-thematic, Tel Aviv-contrastive)
B. Hanako went to Tel Aviv, but I dont know about the other people
and the other places. (Hanako-contrastive, Tel Aviv-contrastive)

Teruabibu-e to Tel Aviv is not sentence-initial, and it can only be a contrastive


topic. However, Kuroda (1988) points out that multiple thematic topics are possible when the second topic is preposed over the first. (63) confirms this observation.
(63)

Teruabibu-e-wai [Hanako-wa (kyonen) ti itta]


went
Tel Aviv-to-top H.-top
last year
A. As for Tel Aviv, Hanako went there, but I dont know about the
other people. (Tel Aviv-thematic, Hanako-contrastive)
B. As for Hanako, she went to Tel Aviv, but I dont know about the
other places. (Tel Aviv-contrastive, Hanako-thematic)
C. As for Tel Aviv and as for Hanako, she went there. (Tel Avivthematic, Hanako-thematic)
D. Hanako went to Tel Aviv, but I dont know about the other places
and the other people. (Tel Aviv-contrastive, Hanako-contrastive)

(63) is four-ways ambiguous as indicated: each of the two topics can receive
thematic or contrastive interpretation. The interpretation that is important here
is the one in C, where both Teruabibu-e to Tel Aviv and Hanako are construed
as thematic topics. This shows that multiple thematic topics can occur in a single

172

Mamoru Saito

clause contrary to Kunos generalization. Given the analysis presented above, it


implies that the Topic projection can be recursive.14
The discussion above indicates that the CP system of Japanese is remarkably
similar to that in Italian. Rizzis (1997) proposal in (44) for Italian is repeated
below in (64).
(64)

[ . . . Force [ . . . (Topic*) [ . . . (Focus) [ . . . (Topic*) [ . . . Finite


[TP . . . ]]]]]]

The structure for Japanese in a parallel format is as in (65).


(65)

[ . . . [ . . . [ . . . [ . . . [TP . . . ] Finite] (Topic*)] Force] Report]

There are only two differences aside from the linear order. One is the presence
of the Report head in Japanese, as discussed in detail in Section 2. It seems
clear that there is a parameter here. Spanish and Japanese have it, but Italian
and English do not. The other is the absence of the Focus head in Japanese.
For this also, there is likely to be a parameter. That is, languages may vary with
respect to the presence/absence of the Focus head within the C system. It would
be much too hasty to propose a concrete hypothesis on the possible variations
in the left/right periphery just on the basis of (64) and (65). Nevertheless, the
preliminary investigation in this paper suggests that the CP structure is fairly
rigid across languages with the locus of variation in Report, Focus, and possibly
Topic.
5.

Conclusion

In this paper, I have examined the complementizer system of Japanese and presented a preliminary hypothesis on the structure of the Japanese right periphery.
I first proposed that to is not a complementizer for embedded propositions as
widely assumed, but is a complementizer for paraphrases or reports of direct
discourse just like que in Spanish. I showed that Planns (1982) analysis of que
is directly applicable to this complementizer. I then argued that no, which Kuno
14. See Saito (2007) for detailed discussion of examples like (63). It is suggested there
that thematic topics are licensed clause-initially, and the interpretation in B obtains
when the contrastive topic is scrambled over the clause-initial thematic topic. One
question that remains is why the two wa-phrases in (62) cannot both be in Spec
positions of Top heads and be construed as thematic topics. Although I do not have
a clear account for this, I suspect that some sort of crossing constraint is at work,
preventing the subject topic from occupying the Spec position of the higher TopicP.

Sentence Types and the Japanese Right Periphery

173

(1973) associates with factivity, has a wide distribution and should be considered the normal complementizer for embedded propositions. As noted above,
these descriptive results provide explicit evidence for Planns (1982) proposal
on Spanish. She proposes that Spanish has three complementizers; que for paraphrases, null C for questions, and que for propositions. Those three are not only
present but have distinct phonetic realizations, to, ka and no, in Japanese.
In the second part of the paper, I first considered examples where to, ka and
no co-occur, and suggested that the three complementizers are hierarchically
organized as in (48), repeated below in (66).
(66)

[. . . [. . . [. . . Finite (no)] Force (ka)] Report (to)]

I then re-examined the distribution of thematic topics, and showed that they are
not limited to the matrix-initial position as widely believed. I argued that they
occur not only in to-headed CPs as sometimes observed but also in ka-headed
CPs. This led to the hypothesis that there is a Topic projection located above
FiniteP and below ForceP. Based on Kurodas (1988) observation that multiple
thematic topics are possible, I proposed finally that the Japanese right periphery
has the structure in (67).
(67)

[ . . . [ . . . [ . . . [ . . . Finite (no)] (Topic*)] Force (ka)] Report (to)]

As repeatedly noted, this is quite similar to the structure of the Italian left periphery proposed in Rizzi (1997). Further work is required to discover the precise
structure of the Japanese right periphery. But this paper has demonstrated that
it is quite rich, much more so than has been assumed, and that it is comparable
to Spanish and Italian. Then, it seems fairly clear that its investigation can contribute fruitfully to the research project initiated by Rizzi (1997) on the universal
properties and possible variations in the left/right periphery.
References
Grewendorf, Gunther and Cecilia Poletto
2009
The Hybrid Complementizer System of Cimbrian. In: Vincenzo
Moscati and Emilio Servidio (eds.), Studies in Linguistics Vol. 3: Proceedings XXXV Incontro di Grammatica Generativa, 181194. Centro
Interdipartimentale di Studi Cognitivi sul Linguaggio, Universita di
Siena.
Heycock, Caroline
1994
Focus Projection in Japanese. In: Merc`e Gonz`alez (ed.), Proceedings
of the North East Linguistic Society 24, 157171. Department of Linguistics, University of Massachusetts, Amherst.

174

Mamoru Saito

Heycock, Caroline
2008
Japanese -Wa, -Ga, and Information Structure. In: Shigeru Miyagawa
and Mamoru Saito (eds.), The Oxford Handbook of Japanese Linguistics, 5483. New York: Oxford University Press.
Hiraiwa, Ken and Shinichiro Ishihara
2002
Missing Links: Cleft, Sluicing and no da Construction in Japanese.
In: Tania Ionin, Heejeong Ko and Andrew Nevins (eds.), The Proceedings of Humit 2001, 3554 (MIT Working Papers in Linguistics 43.).
Cambridge, Mass: Department of Linguistics and Philosophy, MIT.
Jayaseelan, K. A.
2008
Topic, Focus and Adverb Positions in Clause Structure. Nanzan Linguistics 4: 4368.
Kiparsky, Paul and Carol Kiparsky
1970
Fact. In: Manfred Bierwisch and Karl E. Heidolph (eds.), Progress in
Linguistics, 143173. The Hague: Mouton.
Kuno, Susumu
1973
The Structure of the Japanese Language. Cambridge, Mass.: MIT
Press.
Kuno, Susumu
1988
Blended Quasi-Direct Discourse in Japanese. In: William J. Poser
(ed.), Papers from the Second International Workshop on Japanese
Syntax, 75102. Stanford: CSLI Publications.
Kuroda, Sige-Yuki
1988
Whether We Agree or Not: A Comparative Syntax of English and
Japanese. Linguisticae Investigationes 12: 147.
Lahiri, Utpal
1991
Embedded Interrogatives and Predicates that Embed Them. Ph.D. dissertation, MIT.
Matsumoto, E.
2010
Quotation Expressions and Sentential Complementation in Japanese.
B.A. thesis, Nanzan University.
Murasugi, Keiko
1991
Noun Phrases in Japanese and English: A Study in Syntax, Learnability, and Acquisition. Ph.D. dissertation, University of Connecticut.
Murasugi, Keiko
2009
What Japanese-Speaking Childrens Errors Tell us about Syntax. Presented at GLOW in Asia VII, EFL University, Hyderabad.
Plann, Susan
1982
Indirect Questions in Spanish. Linguistic Inquiry 13: 297312.
Rivero, Maria-Luisa
1994
On Indirect Questions, Commands, and Spanish Quotative Que. Linguistic Inquiry 25: 547554.

Sentence Types and the Japanese Right Periphery


Rizzi, Luigi
1997

175

The Fine Structure of the Left Periphery. In: Liliane Haegeman (ed.),
Elements of Grammar, 281337. Dordrecht: Kluwer Academic Publishers.
Rosenbaum, Peter S.
1967
The Grammar of English Predicate Complement Constructions, Cambridge, Mass.: MIT Press.
Saito, Mamoru
2007
Semantic and Discourse Interpretation of the Japanese Left Periphery.
Presented at the Sound Patterns of Syntax Workshop, Ben-Gurion
University, Beer Sheva.
Schachter, Paul
1973
Focus and Relativization. Language 49: 1946.
Stowell, Timothy A.
1981
Origins of Phrase Structure. Ph.D. dissertation, MIT.

Part III.
Clausal Properties of Lexical Categories

On NPs and Clauses*

Zeljko
Boskovic

With a few exceptions, it is standardly assumed that languages without articles


have a null D; i.e. the difference between (1) and Serbo-Croatian (SC) (2) is
assumed to be PF-based, the D being null in SC.
(1)

The stone broke the window.

(2)

Kamen je razbio prozor.


stone is broken window

(SC)

In Boskovic (2008a) I argue that there is a fundamental structural difference


in the traditional Noun Phrase (TNP) of English and article-less languages like
SC, which can be captured if DP is not even present in the TNPs in (2) (see
also Fukui (1988); Corver (1992); Zlatic (1997); Chierchia (1998); Cheng and
Sybesma (1999); Lyons (1999); Willim (2000); Baker (2003), among others for
no-DP analyses of at least some article-less languages). My main argument for a
fundamental difference in the structure of TNP in languages with and those without articles concerns a number of syntactic and semantic generalizations where
the presence/lack of articles in a language plays a crucial role. In this paper I
will strengthen the argument by adding a number of new generalizations that
were not discussed in Boskovic (2008a). Several of these generalizations involve
a surprising interplay between TNP internal syntax/semantics and clause-level
phenomena, which shows that many clause-level phenomena cannot be properly
* This paper has benefitted a great deal from comments and help with judgments by
numerous linguists. The material from the paper was presented in seminars at the
University of Connecticut, and talks at Nanzan University, University of Sarajevo,
University of Novi Sad, University of Frankfurt, University of Sao Paolo, University
of Nova Gorica, University of Venice, University of Leiden, Hungarian Academy of
Sciences, Utrecht University, Goethe Universitat, University of Paris 8, University of
Michigan, FDSL 8 (University of Potsdam), Moscow Student Conference on Linguistics 5, GLOW 33 (University of Wrocaw), MayFest 2010 (University of Maryland),
Syntax Fest 2010 (Indiana University), and the Linguistic Summer School in the
Indian Mountains 5. This work was partially supported by NSF grant BCS-0920888.

180

Zeljko
Boskovic

understood without paying close attention to the internal structure and interpretation of TNPs. I will also explore consequences of the internal structure of TNP
for the internal structure of clauses under the assumption that the two have parallel structure. Taking the TNP/Clause parallelism hypothesis and the NP/DP
parameter seriously leads to the conclusion that just like the structure of TNP is
poorer in NP languages than in DP languages, the structure of clauses should be
poorer in NP languages than in DP languages. I will start by briefly summarizing
the generalizations from Boskovic (2008a). I will then discuss a number of new
NP/DP generalizations and then turn to the issue of TNP/Clause parallelism
and consequences of the structure of TNP and TNP internal processes for the
structure of clauses and clause-level processes.
1.

Generalizations from Boskovic (2008a)1

1.1. Left Branch Extraction


Languages differ regarding whether they allow left-branch extractions (LBE)
like (3)(4).
(3)

*Expensive /Those i

he saw

[ti cars]

(4)

Skupa /Ta i
expensive/that

je vidio
is seen

[ti kola]
car

(SC)

(5)

Doroguju /Tu i
expensive/that

on videl
he saw

[ti masinu]
car

(Russian)

Noting a correlation with articles, Uriagereka (1988), Corver (1992) and Boskovic (2005) establish (6).
(6)

Only languages without articles may allow LBE examples like (4).

To illustrate, Boskovic (2005) notes that Bulgarian and Macedonian, the only
two Slavic languages with articles, differ from most other Slavic languages
1. The generalizations from Sections 1 and 2 (the reader should note that most of them
are one-way correlations) are still in the process of verification against additional
languages. Future research will undoubtedly discover exceptions to many of the
generalizations discussed below. However, even if the generalizations turn out to be
only strong tendencies, that will still call for an explanation. Note also that a weaker
version of the main claim made in this paper would be that some languages without
overt articles do not have DP. The stronger (and more interesting) position is that this
holds for all languages without overt articles, not just those discussed in the paper.

On NPs and Clauses

181

(e.g. SC, Russian, Polish, Czech, Ukrainian, Slovenian) in disallowing LBE.


Within Romance, Latin, which did not have articles, differs from Modern Romance, which has articles, in that it had LBE. Mohawk, Southern Tiwa, and
Gunwinjguan languages (see Baker 1996) as well as Hindi, Bangla, Angika,
and Magahi also allow LBE and lack articles.2
(7)

a. *Novata i ja prodade Petko [ti kola].


car
new
it sells
Petko
b. Novata kola ja prodade Petko.

(Macedonian)

A particularly strong confirmation of (6) is provided by Finnish. As discussed


in Laury (1997), Colloquial Finnish has developed a definite article (Dal Pozzo
(2007) suggests that it may also be developing an indefinite article). Significantly,
Franks (2007) observes that LBE is allowed only in literary Finnish, which does
not have articles. Thus, (8a) is acceptable only in literary Finnish.
(8)

a.

Punaisen
red-acc
b. ?*Punaisen
red-acc

ostin
buy-pst-1sg
ostin
buy-pst-1sg

auton. (literary Finnish, poetic style)


car-acc
(sen) auton.
(spoken Finnish)
the
car-acc

Language change can often take a good amount of time. What we are witnessing
in Finnish is rather fascinating from this perspective: the emergence of the article
has led to a pretty much instantaneous loss of LBE.
Another argument regarding language change comes from the history of
Greek. Ancient Greek underwent a change from an article-less to an article
language. Thus, while Homeric Greek was an article-less language, Koine Greek
was a full-blown article language. Taylor (1990) has conducted an investigation
of what she refers to as split wh-phrases (involving extraction of just the wh2. I focus on adjectival LBE (demonstratives are adjectives in Slavic LBE languages,
see below), ignoring possessor extraction. The reason for this is that several accounts
of the ban on AP LBE in article languages leave a loophole for possessor extraction
to occur in some languages of this type (see Boskovic 2005: 4). Thus, Hungarian,
which has articles, allows possessor extraction, although it disallows adjectival LBE,
which is what is important for our purposes (see, however, den Dikken (1999), who
suggests that Hungarian possessive extraction may actually involve a left dislocationtype configuration with a resumptive pronoun).
(i)

a.
b.

*Magas(-ak-at) latott
lany-ok-at.
tall-pl-acc
saw-3sg girl-pl-acc
cf. Magas lany-ok-at latott.
Tall girls, he saw.

182

Zeljko
Boskovic

word out of a wh-phrase) and split NPs in the history of Ancient Greek and
observed a very significant drop in the number of split wh-phrases/NPs in the
Homeric and the post-Homeric period. While not all split wh-phrases and split
NPs involve LBE, many of them do, which makesTaylors results very significant
in the current context. Taylor has examined the following texts and periods for
Homeric Greek and Koine Greek:
1. Homeric period: Homer Iliad and Odyssey (8th century BC)
2. Koine period:
New Testament corpus (1st century AD).
Taylors corpus contains 68% of split wh-phrases and 25% of split NPs for
the Homeric period, which, as noted above, was an article-less language. On
the other hand, the corpus for Koine Greek, an article language, contains only
15% of split wh-phrases and 0% of split NPs.3 Given that many cases of split
wh-phrases/NPs involve LBE, these facts strongly confirm the generalization
in (6).
Before proceeding, let me note that for the purpose of (6) and other generalizations below, I take articles to be unique, i.e. occur once per TNP. The i ending
in SC (9) is then not considered to be an article (see Despic (2011) for relevant
discussion of this element).4
3. The definite article came into general use in the Classical Greek period (which comes
in between the Homeric and the Koine Greek period), though it was likely fully
established only in the Koine Greek Period. In this respect, it is worth noting that
the percentage of split wh-phrases/NPs is significantly lower in the Classical Greek
Period than in the Homeric period, though higher than in the Koine Greek period (see
Taylor 1990).
4. A word is in order here regarding Modern Greek. Androutsopoulou (1998) claims that
Greek allows what appears to be AP LBE. My informants, however, uniformly reject
examples like (i) (in fact, even Androutsopoulou notes that speakers have difficulty
accepting such examples).
(i)

to kokkino idha forema.


the red
saw dress
I saw the red dress.

Notice also that to, which is traditionally considered to be an article, can appear
on more than one element in a TNP (the so-called polydefinite construction), which
may cast doubt on its article status. Mathieu and Sitaridou (2002) suggest that this
type of articles in Greek are actually agreement markers. More importantly for
oi (2008), who treat to as a true article, analyze
our purposes, Lekakou and Szendr
polydefinite constructions as involving multiple full DPs with nominal ellipsis. Under
the ellipsis analysis, (i) may be analyzable as involving full DP movement, not LBE,
with ellipsis of the NP in the fronted constituent. The analysis makes (i) (and Greek

On NPs and Clauses

(9)

novi /nov
crveni
auto
new-def./new-indef. red-def. car

183

(SC)

Furthermore, it should become clear from the discussion below that what is
important for the generalizations given here is the presence/absence of definite,
not indefinite articles in a language, given that indefinite articles have often been
argued to be located below DP even in languages like English that clearly have
DP (see, e.g., Bowers 1987; Stowell 1989; Chomsky 1995; Boskovic 2007c). In
fact, Slovenian, which uncontroversially has indefinite but not definite article, in
all relevant respects patterns with article-less languages (see Boskovic 2009a).
Thus, it allows LBE.
(10)

Visoke je videl studente.


tall
is seen students

1.2. Adjunct Extraction from NP


Consider adjunct extraction from TNP, which English disallows (see Chomsky
1986a).
(11)

a. Peter met [NP girls from this city]


b. *From which city i did Peter meet [NP girls ti ]?

Observing that SC and Russian allow extraction of adjuncts out of TNPs while
Bulgarian does not allow it, Stjepanovic (1998) argues for (16). Note that Slovenian, Polish, Czech, Ukrainian, Hindi, Bangla, Angika, and Magahi, all articleless languages, pattern with SC and Russian, while Spanish, Icelandic, Dutch,
German, French, Arabic, and Basque, which have articles, pattern with English.5
(12)

Iz
kojeg grada i je Petar sreo [djevojke ti ]?
from which city
is Peter met girls

(SC)

(13)

Iz
kakogo goroda ty vstrechal [devushek ti ]?
from which city
you met
girls

(Russian)

more generally) fully consistent with the LBE generalization. It is also worth noting
that Androutsopoulou (1998) treats (i) in terms of remnant DP fronting. It is shown
in Boskovic (2005) that such an analysis cannot be applied to AP LBE in true AP
LBE languages like SC. If Androutsopoulous analysis of (i) is correct we may then
be dealing here with a different phenomenon from AP LBE in languages like SC
(recall, however, that the grammaticality status of (i) is highly controversial.)
5. See Ticio (2003) for Spanish and Fortmann (1996) for German. ((11b) is actually
acceptable in Spanish, where the relevant phrase is an argument, as Ticio shows (see
Ticio for relevant tests)).

184

Zeljko
Boskovic

(14)

*Ot koj
grad i Petko [srestna momiceta ti ]?
from which city Petko met
girls
(Bg, Stjepanovic. 1998)

(15)

a. *En donde robaron [una estatua t]?


in where stole
a
statue
(Spanish, Ticio 2003)
b. *Fra hvaa borg ser u stelpur?
from which city see you girls
(Icelandic)

(16)

Only languages without articles may allow adjunct extraction out of


TNPs.

1.3. Scrambling
There is also an important correlation between articles and the availability of
scrambling.6
(17)

Only languages without articles may allow scrambling.

SC, Russian, Polish, Czech, Slovenian, Latin, Japanese, Korean, Turkish, Hindi,
Chukchi, Chichewa, and Warlpiri all have scrambling and lack articles. Particularly interesting are Slavic and Romance. Bulgarian, e.g., has noticeably less
freedom of word order than SC. Also, all modern Romance languages have
articles and lack scrambling, while Latin lacked articles and had scrambling.
It is also worth noting Lakhota, Mohawk, and Wichita, which are also related
languages. The latter two lack articles and have more freedom of word order
than Lakhota, which has articles.
6. By scrambling I mean the kind of movement referred to as scrambling in Japanese,
not German, whose scrambling is a very different operation with very different
semantic effects from Japanese scrambling. One of the defining properties of scrambling for the purpose of (17) is taken to be the existence of long-distance scrambling
from finite clauses, which German lacks (for German, see also Boskovic (2004a) and
Grewendorf (2005)).
One needs to be careful here regarding the usage of the term scrambling in the literature, since the term is often used for ease of exposition when an author wants
to remain uncommitted regarding the nature of the movement involved. From this
perspective, many potential counterexamples to the scrambling generalization can be
easily explained away. Consider, e.g., Albanian and Greek, which are sometimes said
to have scrambling. However, an object that is fronted to a sentence initial position in
these languages either has to be clitic doubled or contrastively focused and adjacent
to the verb (see here Kallulli 1999). This indicates that object fronting involves either
clitic left dislocation or focus movement, not what is referred to as scrambling in
Japanese.

On NPs and Clauses

185

1.4. Negative Raising


I now turn to a generalization regarding negative raising (NR), where negation
can be taken to be either in the matrix or the embedded clause of (18). The
embedded clause option is confirmed by the strict clause-mate NPIs in (21).
That these items require negation is shown by (19), while (20) shows that nonNR verbs like claim disallow long-distance licensing of these items. Since they
require clause-mate negation, negation must be present in the embedded clause
of (21) when the NPIs are licensed.
(18)

John does not believe that Mary is smart.

(19)

a. *John left until yesterday.


b. John didnt leave until yesterday.
c. *John has visited her in at least two years.
d. John hasnt visited her in at least two years.

(20)

a. *John didnt claim [ that Mary would leave [NPI until tomorrow]]
b. *John doesnt claim [that Mary has visited her [NPI in at least two
years]]

(21)

a.
b.

John didnt believe [ that Mary would leave [NPI until tomorrow]]
John doesnt believe [that Mary has visited her [NPI in at least two
years]]

Before establishing the NR generalization, note that for the purpose of the
generalization I confine myself to negative raising from finite clauses and use
as the relevant diagnostics the ability of NR to license strict-clause mate NPIs.
A crosslinguistic check of the availability of NR under these conditions reveals
the following:
(22)

Negative raising is disallowed in languages without articles.

SC, Czech, Slovenian, Polish, Russian, Ukrainian, Turkish, Korean, Japanese,


and Chinese lack articles and NR (i.e. strict clause-mate NPI licensing under
NR). On the other hand, English, German, Spanish, French, Portuguese, Romanian, and Bulgarian have both articles and NR (i.e. allow strict clause-mate
NPI licensing under NR, see Boskovic (2008a)). In light of this, we may in fact
be dealing here with a two-way correlation, which would strengthen (22) to (23).
(23)

Languages without articles disallow NR, and languages with articles


allow it.

186

Zeljko
Boskovic

Interestingly, even in languages where the NPI test fails negation is interpretable
in the lower clause: SC (24) has the atheist (non-agnostic) meaning Ivan believes God doesnt exist (the same holds for Korean, Japanese,Turkish, Chinese,
Russian, Polish, and Slovenian).
(24)

Ivan ne vjeruje da bog postoji.


Ivan neg believes that God exists

This suggests that lower clause negation interpretation and strict NPI licensing
under NR should be divorced (contrary to the standard practice, where the two
are correlated), with a three-way split among verbs: (a) negation interpreted
in the lower clause and strict NPIs licensed under NR (possible only for some
verbs in languages with articles) (b) negation interpreted in the lower clause,
strict NPIs not licensed (c) no NR at all.
1.5. Superiority and Multiple wh-Fronting
MWF languages differ regarding whether they show Superiority effects (strict
ordering of fronted wh-phrases) in examples like (25)(26). It turns out that there
is a correlation between Superiority effects with multiple wh-fronting (MWF)
and articles, given in (27).
(25)

a.

Koj kogo vizda?


who whom sees
Who sees whom?
b. *Kogo koj vizda?
(Bulgarian)

(26)

a.
b.

(27)

Ko koga vidi?
who whom sees
Koga ko vidi?

(SC)

MWF languages without articles do not display superiority effects in


examples like (25)(26).

MWF languages without articles do not show Superiority effects. This is the case
with SC, Polish, Czech, Russian, Slovenian, Ukrainian, and Mohawk. MWF
languages that show Superiority effects all have articles. This is the case with
Romanian, Bulgarian, Macedonian, Basque, and Yiddish. Hungarian is an exception (it has articles and no superiority), which however does not violate (27).7
7. Interestingly, Watanabe (2003) suggests that Hungarian traditional definite article
is not a D-element, which casts doubt on its DP status. (For relevant discussion of
Hungarian MWF, see Boskovic (2007a).)

On NPs and Clauses

187

1.6. Clitic Doubling


Another generalization concerns clitic doubling, where Slavic again gives us
a useful clue. Clitic doubling is allowed only in two Slavic languages, Bulgarian and Macedonian, which also have articles.8 Slavic languages without
articles, like SC, disallow it. In fact, all clitic doubling languages I am aware
of (Albanian, Macedonian, Bulgarian, Greek, Somali, Spanish, French (some
dialects), Catalan, Romanian, Hebrew, Dutch (some dialects)) have articles. We
then have (29).
(28)

(29)

a.

Ivo go napisa pismoto.


Ivo it wrote letter-the
Ivo wrote the letter.
b. *Ivan (*ga) napisa pismo.
Ivan
it
wrote letter

(Bulgarian/Macedonian)

(SC)

Only languages with articles may allow clitic doubling.

1.7. Adnominal Genitive


Willim (2000) notes English, Arabic, Dutch, German, and Catalan, all article
languages, allow two nominal genitive arguments, i.e. both the external and the
internal argument can be genitive, where the genitive is realized via a clitic/
suffix or a dummy P (30). The same holds for Portuguese, Basque, French,
Greek, Hebrew, Icelandic, Macedonian, Bulgarian, Spanish, Welsh, Maltese,
Maori, Samoaon, Swedish, all article languages. On the other hand, Willim
notes article-less languages Polish, Czech, Russian, and Latin disallow two
lexical genitives. (Note that word order does not matter in (31). In languages
of this type the external argument is generally realized via a PP headed by an
analogue of English by (a semantically contentful P) or an inherent oblique
Case, see (32).) The same holds for SC, Ukrainian, Quechua, and Turkish. This
leads to (33).9
8. Note that the doubled NP in (28a) is in situ, it is not right-dislocated; note also that
true clitic doubling is associated with a definiteness/specificity effect.
9. (33) concerns only nominal arguments, not possessives, and disregards inherent Case,
as in SC lisavanje (depriving) sina (son, gen) njegovog (his, gen) nasledstva (inheritance, gen) depriving the son of his inheritance (see Zlatic 1997), where the
second genitive is inherent. (The second NP remains genitive even when the casemarker is a verb, as in On lisava sina njegovog nasledstva He is depriving the son of
his inheritance. I am not concerned with this type of lexically specified cases here.)
I also ignore for obvious reasons languages such as Japanese which allow multiple
identical case marking constructions. (The same holds for languages like Estonian,

188
(30)

Zeljko
Boskovic

a.

b.

(31)

(32)

a. *odkrycie Ameryki
Kolumba
discovery America-gen Columbus-gen
Columbus discovery of America
ma
b. *znicen
R
barbaru
destruction Rome-gen barbarians-gen
The barbarians destruction of Rome
a.

b.

(33)

Hannibals
Eroberung Roms
Hannibal-gen conquest Rome-gen
Hannibals conquest of Rome
(German)
lavaluacio
de la comissio dels resultats
the evaluation of the comitte of the results
the committees evaluation of the results
(Catalan)

(Polish)

(Czech)

odkrycie Ameryki
przez Kolumba
discovery America-gen by
Columbus
the discovery of America by Columbus
ma
znicen
R
barbary
discovery Rome-gen barbarians-instr
the destruction of Rome by the barbarians

Languages without articles do not allow transitive nominals with two


lexical genitives.

1.8. Superlatives

Zivanovi
c (2008) notes that Slovenian (34) does not have the reading where
more than half the people drink beer. It only has the reading where more people
drink beer than any other drink though it could be less than half the people.
(34)

Najvec ljudi
pije pivo.
most
people drink beer.
More people drink beer than drink any other beverage. (Plurality reading, MR)
*More than half the people drink beer.
(Majority reading, PR)

English most gives rise to both readings, though in different contexts. German
MOST also has both readings.

which allows multiple genitives on adjectives that does not arise through case concord
with a noun (the noun can be non-genitive in such cases).)

On NPs and Clauses

(35)

189

Die meisten Leute trinken Bier.


the most
people drink beer.
More than half the people drink beer.
More people drink beer than any other drink. (with focus on beer.)

Zivanovi
c notes English, German, Dutch, Hungarian, Romanian, Macedonian,
and Bulgarian, which have articles, allow the majority reading (the same holds
for Basque and Arabic). The reading is disallowed in Slovenian, Czech, Polish, SC, Chinese, Turkish, and Punjabi, which lack articles and allow only the
plurality reading (the same holds for Hindi, Angika, and Magahi).10 We then
have (36) (I set aside cases where the majority reading is expressed with a noun
like majority).
(36)

Only languages with articles allow the majority superlative reading.

1.9. Head-Internal Relatives and Locality


There is a locality distinction among languages with head-internal relatives
(HIR). HIRs in Japanese, Korean, Quechua, Navajo, and Mohawk are island
sensitive, while those in Mojave and Lakhota are not (Basilico 1996; Watanabe
2004; Baker 1996). Interestingly, the former lack articles, while the latter have
them. This leads to (37).
(37)

Head-internal relatives display island-sensitivity in article-less languages, but not in languages with articles.

Grosu and Landman (1998) show that there is also a semantic difference at
work here, in particular, HIRs are restrictive in languages with articles and
maximalizing in those without articles.
1.10. Polysynthetic Languages
Baker (1996) observes the following generalization regarding polysynthetic languages.
(38)

Polysynthetic languages do not have articles.

10. The following context enforces the majority reading for (the past tense version) of
(35): Suppose people at a dinner were allowed more than one beverage. 60% of the
people had a beer and 75% of the people had a glass of wine.

190
2.

Zeljko
Boskovic

Additional Generalizations

I now turn to new generalizations that were not discussed in Boskovic (2008a).
(Anticipating the discussion in Sections 34, from now on I will refer to languages with articles as DP languages, and to languages without articles as NP
languages.)
2.1. Focus Morphology
In some languages, negative constituents have overt focus morphology (see
(39)). Such morphology is often realized through the presence of focal elements
like even, also, or too (SC has two series of negative constituents, a negative
concord series and an NPI series, both of which contain even), and sometimes
through obligatory emphatic (focus) stress, as in Greek.11
(39)

n+i+ko
i+ko
neg+even+who even+who noone/anyone

(SC)

While in DP languages negative constituents may but do not have to have a focus
marker, in NP languages they have a focus marker. This holds for SC, Russian,
Polish, Ukrainian, Lithuanian, Hindi, Chinese, Japanese, Korean, Finnish,Yakut,
Lezgian, Kannada, Quechua, Mansi, Latin, Persian, Turkish, and Kazakh.12 This
leads to (40).
(40)

Negative constituents must be marked for focus in NP languages.

2.2. Negative Concord with Complex Negative Constituents


It is well-known that in some negative concord languages, the negative concord reading is unavailable with complex negative constiutents (NCIs). This is
illustrated below with examples from Italian. (41a)(4b) show that Italian is
11. Another option, which may be realized in Slovenian, may be obligatory focus movement of the negative constituent.
12. Boskovic (2009c) argues that in languages with both negative concord and NPI series,
the two are derived from the same underlying items, which means it suffices for
one of these to have a focus marker to meet (40). There is a bit of a complication
with Persian. Persian negative concord series contains hic. Hic is analyzable as air
one+ciy, an emphatic particle which I assume is focus related. Hic was borrowed into
Turkish and Kazakh, which I assume can be analyzed in the same way. A potential
counterexample to the generalization in (40) is Georgian. Georgian does have a
focus marker in the existential quantifier series, but not in NCIs. I leave a detailed
examination of Georgian for future research.

On NPs and Clauses

191

a negative concord language. However, the negative concord reading becomes


unavailable with multiple NCIs if one of the NCIs is a complex element.
(41)

a.

b.

c.

Non ho
visto nessuno /nessuno studente.
neg have seen nobody/no
student
I didnt see anybody/any students.
(negative concord only)
Nessuno ha letto niente.
nobody has read nothing
(negative concord or double negation)
Nessuno student ha letto nessun libro /niente.
no
student has read no
book/nothing
(double negation only)

It turns out that DP languages differ with respect to whether the double negation
reading is forced in examples like (41c). Thus, the reading is forced in Italian,
Spanish, West Flemish, and French. However, Brazilian Portuguese, Basque,
Hebrew, and Romanian still allow the negative concord reading. On the other
hand, NP negative concord languages all allow the negative concord reading in
examples like (41c). This is, e.g., the case with SC, Russian, Polish, Ukrainian,
Japanese, Korean, and Turkish. I am in fact unaware of any negative concord
NP language that would disallow the negative concord reading. We are then led
to the following generalization.
(42)

The negative concord reading may be absent with multiple complex


negative constituents only in DP negative concord languages.

2.3. Quantifier Scope


Consider now (43).
(43)

Someone loves everyone.

The example is ambiguous: Everyone can take either narrow or wide scope
in (43). I will refer to the latter reading as the inverse scope reading. A number
of languages disallow the inverse scope reading in the unmarked order for the
subject, verb, and object (SVO in SVO languages, and SOV in SOV languages).
Thus, inverse scope is allowed in English, Spanish, Brazilian Portuguese, Macedonian, and Hebrew, but it is impossible in German, Basque, Dutch, Icelandic,
Bulgarian, Welsh, Romanian, Japanese, Korean, Turkish, Persian, Hindi, Bangla,
Chinese, Russian, Polish, Slovenian, Ukrainian, and SC. Focusing on the latter
group of languages, while the first seven are DP languages, other languages in

192

Zeljko
Boskovic

this group are all NP languages. In fact, I do not know of any NP language that
productively allows inverse scope in such examples.13 We then have (44).
(44)

Inverse scope is unavailable in NP languages in examples like (43).

2.4. Radical Pro-drop


I now turn to the phenomenon of radical pro-drop, which I define as productive
argumental pro-drop of both subjects and objects in the absence of rich verbal
agreement. This type of pro-drop differs from pro-drop in languages like Spanish, where pro-drop is licensed by rich verbal morphology. As a result, since
Spanish has subject but not object agreement, pro-drop is allowed only with
subjects in Spanish. Radical pro-drop is allowed in Japanese, Chinese, Korean,
Kokota, Turkish, Hindi, Wichita, Malayalam, Thai, Burmese, and Indonesian,
all of which are NP languages.14 In light of this, we have the generalization
in (45).
(45)

Radical pro-drop is possible only in NP languages.

2.5. Number Morphology


Gill (1987), who considers only a few languages, suggests a potential correlation between obligatory number morphology and the availability of articles. The
phenomenon we are looking at here is the possibility of having examples like
(46), where the N can be interpreted as plural in the absence of plural morphology. (47) divides languages into two groups, where one group has languages
that at least optionally can lack number morphology with at least some Ns (i.e.
where some or all countable Ns can receive plural interpretation without the

13. Certain quantifiers always require wide scope. Such quantifiers tend to be interpreted
with wide scope even in NP languages. What I am concerned with here is quantifiers
that do not require wide scope, i.e. whether the inverse scope is productively available
for all quantifiers in a given language in this type of examples.
14. See also Tomioka (2003) (and Saito (2007) and Neeleman and Szendr
oi (2007) for
different perspectives). Turkish seems to combine Spanish type and radical pro-drop.
A potential problem that is being investigated is Cheke Holo. If Cheke Holo indeed
turns out to be an exception, (45) would simply be a strong tendency. (It is worth
noting here that Brazilian Portuguese is not classified as a radical pro-drop language,
since it does not have fully productive subject drop. In fact, since pronoun objects
could incorporate into V, the availability of a fully productive subject pro-drop is
crucial here.)

On NPs and Clauses

193

presence of number morphology), and the other group contains languages that
have obligatory plural morphology (on either D or N).15
(46)

(47)

Susumu-ga
hon-o
yonda.
Susumu-nom book-acc bought
Susumu bought a/the book/books.

(Japanese)

No obligatory number morphology: Japanese, Korean, Chinese, Hindi,


Bangla, Malayalam, Mohawk, Dyirbal, Warlpiri, Warrgamay, KukuYalanji, Indonesian, Vietnamese.
Obligatory number morphology: Russian, SC, Ukrainian, Hebrew, Portuguese, German, Bulgarian, Polish, Hungarian, Spanish, Romanian,
French, Slovenian, Finnish, Bulgarian, Swahili, Greek, Dutch, Italian,
Latin, Ossetic, Kannada, Macedonian, Somali, Estonian.

While the second group comprises both NP and DP languages, all languages in
the first group are NP languages.16 We then have the generalization in (48).
(48)

Number morphology may not be obligatory only in NP languages.

2.6. Focus Adjacency


The next generalization deals with the question of whether languages that have
focus movement require adjacency with the verb. It is well-known that some
languages require movement of focalized elements. Such languages differ regarding whether the fronted focalized element has to be adjacent to the verb. I
illustrate this for Bulgarian (49) and SC (50). (Capital letters indicate contrastive
focus.)
(49)

a. *Kartinata
Ivan podari
na
painting-the(foc) Ivan give-as-a-present-pst-3sg to
Maria.
Maria
Ivan gave Maria the painting as a present.
(Lambova 2004)
b. Kartinata podari Ivan na Maria.

15. I ignore here TNPs involving numerals, since numerals by their very nature express
number.
16. The NP/DP status of Vietnamese is somewhat controversial; see, however, Cheng
(in preparation) for arguments that Vietnamese lacks true articles, hence should be
classified as an NP language.

194
(50)

Zeljko
Boskovic

Jovana
(Petar)
savjetuje.
Jovan-acc Petar-nom advises
Petar is advising Jovan.

(Stjepanovic 1999)

It turns out that Basque, Hungarian, Bulgarian, Armenian, Greek, Catalan, Romanian, Macedonian, Italian, Spanish, and Albanian are subject to the adjacency requirement. This is not the case with Slovenian, Russian, SC, Polish,
Chinese, and Nupe.17 The adjacency languages are all DP languages, while the
non-adjacency languages are all NP languages. We may then have here another
NP/DP generalization.
(51)

Elements undergoing focus movement are subject to a verb adjacency


requirement only in DP languages.

2.7. Interpretation of Possessors


Partee (2006) observes that English (52) presupposes that Zhangsan has exactly
three sweaters. On the other hand, there is no such exhaustivity presupposition
in Chinese (53).18

17. Turkish is also subject to the adjacency requirement. However, Sener (2006) provides
convincing evidence that Turkish actually does not have focus movement. Rather,
focalized elements in Turkish remain in their base position, where they are subject to
a prosodic requirement that focalized elements be exhaustively parsed into the same
intonational phrase as the verb. The adjacency requirement in Turkish is therefore
phonological, not syntactic in nature (in fact, it affects both contrastively focused
elements and elements that bear simple new information focus, a state of affairs that
is not found with focus movement, which typically affects only the former in the case
of focalized non-wh-phrases.) Notice also that Sener (2010) argues that all elements
that are interpreted as old information (both topics and discourse anaphoric elements)
must undergo movement out of vP in Turkish, this in fact being the only movement
that the language has, which leaves only focalized elements next to the verb. Since
Turkish does not have focus movement, it is irrelevant for the phenomenon under
consideration here.
18. There is more than one option for word order regarding Chinese possessors. I focus here on the possessor-numeral order (in the absence of a demonstrative); see
Partee (2006) for discussion of the full paradigm. (Constructions where the posessor precedes the numeral are definite, while those where the possessor follows the
numeral are indefinite and roughly correspond to English three sweaters of Johns.
I exclude such examples from the discussion here since I am focusing on definite
possessor phrases. For this reason I also focus on article+possessor constructions in
DP languages that allow the two to co-occur.)

On NPs and Clauses

(52)

Zhangsans three sweaters

(53)

[san jian maoxianyi]


Zhangsan de
Zhangsan DEPOSS three CL sweater
Zhangsans three sweaters

195

NP languages I have checked so far, Russian, SC, Turkish, Japanese, Korean,


Hindi, Bangla, Malayalam, and Magahi, all pattern with Chinese in this respect.
(Partee notes this for Russian.) Spanish, Brazilian Portuguese, Italian, Basque,
Hebrew, Dutch, and Arabic, on the other hand, pattern with English. This leads
to (54).
(54)

Possessors may induce an exhaustivity presupposition only in DP languages.

It is worth noting in this respect that Lyons (1999) argues that the DP projection
is responsible for the presupposition of uniqueness/exhaustivity. It is then not
surprising that the presupposition is lacking in article-less languages, which
lack DP.19
2.8. Classifiers
Cheng (in preparation) examines languages with obligatory classifier systems
and notes a correlation with absence/presence of articles, given here in (55).
(55)

Obligatory numeral classifier systems occur only in NP languages.

In other words, if a language has an obligatory classifier system, it does not


have DP. What this generalization suggests is that there is an incompatibility
between a classifier system, i.e. ClP, and DP. This in turn can be interpreted as
support for Cheng and Sybesmas (1999) proposal that classifiers do the job of
D in languages like Chinese. In other words, since ClP and DP basically do the
same job, a language cannot have both. Note, however, that we are not simply
19. Lyons in fact argues that there is no grammaticalized definiteness in Chinese, i.e.
he also argues that Chinese (and other article-less languages) lack DP. Chinese (and
other article-less languages) do have some definiteness effects. However, he argues
definiteness effects found in such languages represent a semantico/pragmatic notion
of definiteness as identifiability (which seems to correspond to what Partee calls familiarity), correlated with topichood. According to Lyons, such non-grammaticalized
definiteness fails to involve the presupposition of uniqueness/exhaustivity (i.e. nongrammaticalized definiteness normally lacks the uniqueness/exhaustivity presupposition).

196

Zeljko
Boskovic

dealing here with a label difference, where DP is replaced by ClP. Cheng and
Sybesma show that ClP is very low in the structure. This means that the source
of definiteness is lower in the structure in NP than in DP languages. In work
in preparation I take advantage of this to account for the well-known fact that
number (more precisely, plurality) interacts with definiteness in ClP languages
like Chinese but not in DP languages like English (the reason for this being
that the projection that is responsible for plurality is higher than ClP (the source
of definiteness in Chinese), but lower than DP (the source of definiteness in
English).
2.9. Second Position Clitics
Another generalization concerns the type of clitics a language has (see also
Migdalski (2010) and Runic (2011)). Languages typically have either verbal
(i.e. V-adjacent) clitics or so-called second position clitics.20 Languages that are
standardly assumed to have second position clitics include a number of Slavic
languages (SC, Czech, Slovak, Slovenian, Hucul Ukrainian, and Sorbian), Latin,
Ancient Greek, Pashto, Tagalog, Ngiyambaa and Warlpiri (and a number of
north-central Australian languages), which interestingly all lack articles.21 This
leads us to the generalization in (56).
(56)

Second-position clitic systems are found only in NP languages.

Slavic and Romance are again quite informative: while a number of Slavic
languages have second-position clitic systems, Bulgarian and Macedonian are
glaring exceptions. As for Romance, Latin had second-position clitics, while
Modern Romance languages lack them. The history of Greek provides a rather
strong confirmation of (56). Thus, Taylor (1990) shows that 90% of enclitics
in the Homeric period, when Greek did not have articles, were in the second
position; this simple second position cliticization system broke down in the
later stages (i.e. article stages), like Koine Greek.
Additional generalizations will be discussed in Section 5; for additional generalizations that cut across the DP/NP line, the reader is also referred to Herdan (2008), Marelj (2008), Boeckx (2003a), Runic (2011), Despic (2011), and
Boskovic (2009d). Taken together, these generalizations provide strong evidence
that there is a fundamental difference between TNP in languages like English
20. I am simplifying here the actual state of affairs. Note that true second-position clitics
are not simply enclitics (i.e. not all enclitics are second-position clitics). I refer the
reader to Boskovic (2001) and references therein for discussion.
21. Uto-Aztecan languages are currently being investigated in this respect.

On NPs and Clauses

197

and article-less languages like SC that cannot be reduced to phonology (overt


vs phonologically null articles). If DP is posited for both, we need to make
a radical principled distinction between D in English and D in SC. Appealing to phonological overtness will not work since English, e.g., disallows LBE
(*Fast, he likes cars), adjunct extraction from TNP, and scrambling even with
null D. Moreover, the above generalizations deal with syntactic and semantic,
not phonological phenomena.
It is often assumed that the TNP should be treated in the same way in articleless languages and languages like English for the sake of uniformity. This argument fails on empirical grounds in light of the above generalizations: it is simply
a fact that there are radical syntactic and semantic differences between the two
there is no uniformity here. Most importantly, as shown in Boskovic (2008a),
Boskovic and Gajewski (in press), and Section 4, these differences (i.e. all the
generalizations discussed above) can be deduced if there is DP in the TNP of
English, but not languages like SC. Moreover, the NP/DP analysis provides a
uniform account of these differences, where a single difference between the two
types of languages is responsible for all of them. It is extremely hard to see how
this can be accomplished under a uniform DP analysis. In fact, I contend that a
universal DP analysis cannot even be entertained seriously until it can be shown
that the analysis can also provide a principled, uniform account of the above
generalizations.
3.

D-like Items in Article-less Languages

It should also be noted that traditional D-items do not exhibit the behavior that is
standardly associated with D-items in article-less languages. Let us take a look
at SC as a representative of NP languages in this respect. Although SC does
not have articles, it does have lexical items like that, some, and possessives.
However, such items behave like adjectives in SC both morphologically and
syntactically (see Zlatic 1997 and Boskovic 2008a).22 In contrast to English
D-items, they clearly have the morphology of adjectives (57), occur in typical
adjectival positions like the predicate position of a copula (58), allow stacking
up (59), and often (though not always) fail to induce Specificity effects that
English D-items induce (60). Another interesting quirk is that SC possessives
22. The point of the following discussion is to demonstrate that the SC items in question
behave differently from their English counterparts; we would not necessarily expect
that the items in question will exhibit the same behavior in all NP languages or rule
out the possibility that in some DP languages some of the items under discussion
could exhibit some of the properties of the SC items in question.

198

Zeljko
Boskovic

cannot be modified by adjectives (61), which follows if adjectives cannot modify


adjectives given that SC possessors are actually adjectives.
(57)

a.

b.

tim
thoseFEM.PL.INST
djevojkama
girlsFEM.PL.INST
tih
thoseFEM.GEN.PL
djevojaka
girlsFEM.GEN.PL

nekim
mladim
someFEM.PL.INST youngFEM.PL.INST

nekih
mladih
someFEM.GEN.PL youngFEM.GEN.PL

(58)

a. *This book is my.


b. Ova knjiga je moja.
this book is my

(59)

a. *this my picture
b. ta moja slika
this my
picture

(60)

O
kojem piscu je procitao [svaku knjigu / sve knjige /
about which writer is read
every book / all books /
(tu) tvoju knjigu ti ]?
that your book
*About which writer did he read every book/all books/this book of
yours?

(61)

*bogati susjedov konj


rich
neighbors horse

They also have some freedom of word order. While English D-items must precede adjectives, SC allows adjectives to precede some D-items (see Boskovic
2007b for some interpretational effects regarding the order of adjectives and
possessors).
(62)

Jovanova skupa
slika
vs. skupa
Jovanova slika
Jovans
expensive picture
*expensive Jovans
picture

(63)

bivsa Jovanova kuca vs. Jovanova bivsa kuca


*former Jovans
house
Jovans
former house

Notice, however, that the order of SC adjectives and D-items is not completely
free. Thus, both adjectives and possessives must follow demonstratives.

On NPs and Clauses

(64)

a.
b.

ova
this
ova
this

skupa
expensive
Jovanova
Jovans

199

kola /?*skupa ova kola


car
slika /?*Jovanova ova slika
picture

These ordering restrictions follow straightforwardly from the semantics of the


elements in question (see Boskovic and Hsieh 2012 regarding Chinese). Semantically, it makes sense that possessives and adjectives should be able to occur
in either order. The most plausible semantics for possessives is modificational
(see e.g. Partee and Borschev 1998 and Larson and Cho 1999).
(65)

Partee and Borschev (1998) (Ri is a free variable) [[ Marys ]] =


x.[Ri (Mary)(x)]

(66)

Larson and Cho (1999) [[ to Mary ]] = x.[POSS(j,x)]

Given the standard assumptions that adjectives are also of type <e,t> and that
there is a rule of intersective Predicate Modification, compositional semantics
imposes no restrictions on the order in which possessives and adjectives may
be composed. On the other hand, the situation is different with demonstratives.
Kaplan (1977/1989) argues that demonstratives are markers of direct reference.
In other words, demonstrative noun phrases pick out an individual of type e. The
individual is picked out at least partially as a function of its predicate complement
phrase. Thus, a demonstrative element like that is a function of type e,t>,e>.
Once a demonstrative has mapped a nominal element to an individual, further
modification by predicates of type <e,t> is impossible. Hence, semantic composition requires both adjectives and possessives to be composed before demonstrative determiners. In other words, semantic composition allows possessives
to be composed either before or after modifying adjectives, while demonstratives must be composed after both adjectives and possessives.23 This perfectly
matches the actual facts regarding the ordering of the elements in question in SC.
It is worth noting in this respect that English counterparts of the unacceptable
examples in (64) are significantly worse than the SC examples. This follows if
the English examples have the semantic violation we have discussed as well as a
syntactic violation (violations of the requirement that DP must be projected on
top of TNP and whatever is responsible for the incompatibility of articles and
possessives in English).

23. Note that the above account readily extends to non-restrictive adjectives under Morzyckis (2008) analysis, where non-restrictive adjectives are also treated as having
type <e,t> and required to be interpreted inside the determiners.

200
(67)

Zeljko
Boskovic

a. **expensive this car


b. **Johns this picture

The proponents of the universal DP analysis (Basic 2004; Rappaport 2000;


Pereltsvaig 2007) account for (64) by placing the demonstrative in DP, which is
located above the projection where possessives and adjectives are located. ( P
is a projection where adjectives are generated, with multiple adjectives requiring
multiple Ps.)
(68)

[DP Demonstrative [PossP Possessive [ P Adjective [NP

(Basic 2004)

(68) accounts for (64), but it fails to capture the relative freedom of the adjectives/possessives order in SC and the SC/English contrast in this respect. Furthermore, Despic (2009, 2011, in press) provides conclusive evidence against
(68) based on the following SC/English contrasts.24
(69)

a.
b.

Hisi latest movie really disappointed Kusturicai .


Kusturicai s latest movie really disappointed himi .

(70)

ga i je zaista razocarao.
a. *Kusturicin i najnoviji film
Kusturicas latest
movie him is really disappointed
Kusturicai s latest movie really disappointed himi .
je zaista razocarao
b. *Njegov i najnoviji film
his
latest
movie is really disappointed
Kusturicu i .
Kusturica
Hisi latest movie really disappointed Kusturicai .

Despic notes that (69) can be accounted for if, as in Kayne (1994), English
possessives are located in the Spec of PossP, which is immediately dominated
by DP, the DP preventing the possessive from c-commanding anything outside
of the subject. The contrast between English and SC then follows if the DP is
missing in SC. (Following Boskovic (2005) Despic in fact treats SC possessives
as NP adjuncts, on a par with adjectives; see Section 4.1).
Significantly, Chinese and Japanese behave just like SC in the relevant respect
(see Boskovic and Hsieh (2912), Cheng (in preparation) and Takahashi (2011)

24. The examples in the rest of this subsection assume a neutral (i.e. non-focused) intepretation of the relevant nouns/pronouns. (Since contrastive focus affects binding
relations it is important to control for it. I have also avoided using relational nouns
like father since at least for some speakers they involve irrelevant interfering factors;
see Takahashi 2011).

On NPs and Clauses

201

for further discussion of Chinese and Japanese respectively), which provides


strong evidence for the no-DP analysis for these languages.
(71)

a. *Tai -de zuixinde dianying rang Li-An i hen shiwan.


he-gen newest movie
make Li-An very disappointed
His latest movie really disappointed Li-An.
b. *Li-An i zuixinde dianying rang ta i hen shiwang.
Li-An newest movie
make he very disappointed
Li-Ans latest movie really disappointed him.
c. *Kurosawai -no saisin-no eega-wa hontoo-ni karei -o
Kurosawa-gen latest-gen movie-top really
him-acc
rakutans-ase-ta.
disappoint-cause-past
Kurosawas latest movie really disappointed him
d. *Karei -no saisin-no eega-wa
hontoo-ni Kurosawai -o
he-gen latest-gen movie-top really
Kurosawa-acc
rakutans-ase-ta.
disappoint-cause-past
His latest movie really disappointed Kurosawa.

Despic also shows that demonstratives and adjectives do not change anything
in SC, which provides strong evidence that demonstratives, possessives, and
adjectives should be treated as multiple adjuncts/specs of the same phrase. Since
demonstratives and adjectives do not introduce an extra projection, they do not
prevent the possessive from c-commanding the co-indexed elements in (72).
(72)

a. *[NP Ovaj [NP Kusturicin i [NP najnoviji [N film]]]] ga i je


this
latest
movie him is
Kusturicas
zaista razocarao.
really disappointed
This latest movie of Kusturicai really disappointed himi .
b. *[NP Brojni
[NP Kusturicini i [NP filmovi ]]] su ga i
Kusturicas
numerous
movies
are him
razocarali.
disappointed

It should, however, be noted that the application of Despics test shows that
functional structure is not completely lacking in SC TNPs. Thus, while demonstratives and adjectives do not bring in additional projections, non-adjectival
numerals which assign genitive of quantification do bring in an additional projection. Despic observes that these elements confine the c-command domain of

202

Zeljko
Boskovic

possessives, allowing them to co-refer with other elements without causing a


binding violation. The contrast between (72) and (73) provides strong evidence
that additional phrasal structure is present above the possessive only in (73),
involving a numeral.
(73)

[QP Pet [NP Dejanovih i [NP prijatelja ]]] je doslo na njegovo i


five
friendsGEN is come to his
DejansGEN
vencanje
wedding
Five of Dejans friends came to his wedding.

Notice that Chinese and Japanese classifiers pattern with SC numerals; they
apparently also introduce an additional projection into the structure.
(74)

a.

b.

You wu-ge Zhangsani -de pengyou lai


canjia tai -de
have 5-cl Zhangsan-gen friend
come attend he-gen
hunli.
wedding
Five of Zhangsans friend came to his wedding.
Go-nin-no Johni -no tomodachi-ga karei -no
five-cl-gen John-gen friends-nom he-gen
kekkonsiki-ni kita.
wedding-dat came
Five of Johns friends came to his wedding.

It is worth noting here that Saito, Liu, Murasugi (2008) argue that only Chinese
has a ClP, classifiers in Japanese being NP-adjuncts. The above data provide evidence against this conclusion. Just like Chinese classifiers, Japanese classifiers
confine the c-command domain of possessives, which indicates that they also
project a ClP above NP.
4.

Some Deductions of the NP/DP Generalizations

I now turn to explanations for the generalizations from Sections 1 and 2 under
the DP/NP analysis. I will first briefly summarize the account of a couple of
representative generalizations given in Section 1 from Boskovic (2005, 2008a)
and Boskovic and Gajewski (in press), referring the reader to Boskovic (2008a)
for the deduction of other generalizations from Section 1. I will focus here
on the phenomena that are relevant to the clause-level syntax, namely leftbranch extraction (I will also suggest a modification of my original analysis

On NPs and Clauses

203

of the phenomenon), which involves extraction of TNP-internal elements to


the clause level (two other locality phenomena, adjunct extraction out of TNPs
and islandhood with HIRs, will also be addressed), and two surprising cases
of an apparent interaction between TNP-internal syntax/semantics and clauselevel phenomena, namely the generalizations concerning the interpretation of
superlatives and negative raising. The generalizations in question confirm the
importance of TNP-internal syntax and semantics for clause-level phenomena. I
will then propose a deduction of two generalizations from Section 2, namely (45)
and (48) (suggestions for deductions of two other generalizations, (54) and (55),
have already been given in Section 2), and explore its consequences for the
clausal structure of NP languages.
4.1. Back to Left Branch Extraction: The Phase Analysis
In Boskovic (2005) I gave two deductions of (6). Here, I will summarize only
one of them, the one based on the Phase-Impenetrability Condition (PIC), which
says only the Spec of a phase is accessible for phrasal movement outside of
the phase (so, XP movement from phase YP must proceed via SpecYP). On a
par with Chomskys (2000) claim that CP but not IP is a phase, I assumed in
Boskovic (2005) that DP is a phase, but NP isnt. Given the PIC, XP can then
move from DP only if it moves to SpecDP. There are two more ingredients of
the analysis: the traditional claim that AP is NP-adjoined and the anti-locality
hypothesis (the ban on movement that is too short), which is deducible from
independent mechanisms and argued for by many authors (e.g. Boskovic 1994;
1997; Saito and Murasugi 1999; Ishii 1999; Abels 2003; Grohmann 2003; Ticio
2003; Boeckx 2005; Jeong 2006).25 Like most other approaches, the version of
anti-locality adopted in Boskovic (2005) requires Move to cross at least one full
phrasal boundary (not just a segment). AP then cannot move to SpecDP in (75)
due to anti-locality. Given the PIC, it cannot move directly out of DP either, as
in (76). Anti-locality/PIC thus prevent AP extraction from DP, banning AP LBE
in English (they dont ban all movement from DP, who do you see [DP t[NP friends
of t]] is still allowed)
(75)

*[DP APi [D D [NP ti [NP

(76)

*APi [DP [D D [NP ti [NP

25. Among other things, anti-locality accounts for the ban on short subject topicalization
and zero subject null operator relatives (Boskovic 1994, 1997), the that-trace effect
(Ishii 1999), the ban on movement of the phase complement (Abels 2003), and the
patterns of extraction of arguments out of DPs (Grohmann 2003, Ticio 2003).

204

Zeljko
Boskovic

The impossibility of adjunct extraction out of TNP in English can be accounted


for in the same way as the impossibility of AP LBE, given that NP adjuncts are
also adjoined to NP. Moreover, the PIC/anti-locality problem does not arise in
SC, which lacks DP.
I would, however, like to suggest here a modification of one aspect of my
earlier analysis of LBE. As discussed in Boskovic (2005), SC disallows deep
LBE, i.e. LBE out of a complement of a noun (the same holds for Polish, Czech,
and Russian).
(77)

On cijeni
[NP [N [ prijatelje [NP pametnih
friends
smart
he appreciates
[NP studenata]]]]].
students
He appreciates friends of smart students.
b. ?*Pametnihi on cijeni [NP [N [ prijatelje [NP ti [NP studenata]]]]].
a.

What this shows is that an NP above an LBE-ing NP has the same effect on
LBE as a DP above an LBE-ing NP does in English; they both block LBE. This
can be accounted for if NP is a phase even in NP languages. (77b) can then be
accounted for in exactly the same way as (3) (*Expensive, he saw cars), with the
higher NP blocking LBE for the same reason DP does it in the English example.
As noted in Boskovic (2010), strong evidence that this suggestion is on the right
track concerns Abelss (2003) generalization that the complement of a phase
head is immobile. Thus, Abels observes that an IP that is dominated by a CP,
a phase, cannot undergo movement. This in fact follows from an interaction of
the PIC and anti-locality, with the PIC requiring IP movement through SpecCP,
and anti-locality blocking such movement because it is too short. Now, if NP is
indeed a phase in NP languages we would expect that an NP complement of a
noun cannot undergo movement. Zlatic (1997) observes genitive complements
of nouns indeed cannot be extracted in SC.
(78) ?*Ovog studenta
sam pronasla [NP knjigu ti ].
book
this student-gen am found
Of this student I found the/a book.
The impossibility of deep LBE and the immobility of genitive complements of
nouns thus fall into place if NP is a phase in article-less languages. They are
both ruled out in exactly the same way.26
26. It should be noted here that nominal complements that bear inherent case can be
extracted. However, they also allow deep LBE. The correlation between the two

On NPs and Clauses

205

Notice furthermore that it is not necessary to posit crosslinguistic variation


regarding phasehood. If we simply assume that the highest phrase in a TNP
counts as a phase there is no real variation in the phasehood of the TNP between
NP and DP languages; the real source of variation lies in the amount of structure
a TNP has in DP and NP languages (see Boskovic 2012). That this analysis is
on the right track is confirmed by the genitive of quantification construction,
where, as discussed in Section 3, even SC projects functional structure above
NP (see (73)). Significantly, nominal complement movement is possible in this
context.
(79)

Ovog studenta
sam pronasla [QP ti mnogo /deset
this studentGEN am found
many/ten
[NP knjiga ti ]].
books

This is expected if it is the highest phrase in a TNP that functions as a phase.


QP rather than NP then functions as a phase in (79). In contrast to (78), where
the phasal edge is SpecNP, genitive NP in (79) can move to the phasal edge,
SpecQP, without violating anti-locality.
4.2. Back to Head Internal Relatives
Turning now to (37), Watanabe (2004) argues that languages differ regarding
the licensing mechanism employed in HIRs. He argues that some languages
employ unselective binding, which is not subject to locality, while others employ
movement/feature checking, which is subject to locality, i.e. intervention effects.
Given (37), the former should be employed in DP languages, and the latter in NP
languages (I depart here from Watanabe). Significantly, Bonneau (1992) argues
for independent reasons that the D that comes with a HIR is the unselective
binder of its head (he makes the proposal for Lakhota). Since the D is missing
in article-less languages, the island-insensitive binding option is unavailable in
these languages.
phenomena thus still holds. I discuss inherent case complements, as well as LBE out
of other phrases, in Boskovic (2010).

Cime
ga je [pretnja ti ] uplasila?
i
what-instr him is threat
scared
The threat of what scared him?
ga je pretnja [ti smrcu]
uplasila?
(ii) ?Kakvom i
what-kind-of him is threat
death-instr scared
Of what kind of death did a threat scare him?
(i)

206

Zeljko
Boskovic

4.3. Back to Negative Raising


Boskovic and Gajewski (in press) explain (23) by highlighting a similarity in
the interpretation of definite plurals and NR predicates. A common analysis of
negative raising attributes to certain predicates (negative raising predicates) an
excluded middle presupposition (EMP), where A believes that p presupposes
A believes that p or A believes that not p. As a presupposition, the EMP survives negation. Then, in A does not believe that p the assertion and the EMP
presupposition together entail A believes that not p.
Now, Gajewski (2005, 2007) argues that the EMP is the hallmark of constructions that can be semantically analyzed as distributive plural definite descriptions, rather than universal quantifiers. To illustrate the EMP of definite
plural NPs, Bill saw the boys implies Bill saw all the boys; Bill didnt see the
boys implies he saw no boys not merely not all, with a universal scoping over
negation, which Gajewski attributes to the EMP and which is analogous to the
lower clause negation reading with negative raising (compare Bill didnt see the
boys with the negation of a universal quantifier: Bill didnt see all the boys).
Returning to negative raising, sentence-embedding predicates are standardly
treated as universal quantifiers over accessible worlds. Gajewski (2005) argues
that having the EMP, negative raising predicates should be treated as plural
definite descriptions, which serve as arguments of the predicates contributed by
their propositional complements.
In Boskovic and Gajewski (in press) we assume that sentence-embedding
predicates combine a modal base (set of accessible worlds) with a quantificational element. The quantificational element may be either a universal quantifier
or a definite article. If a modal base combines with the definite article, the result is a negative raising predicate. Given this, if a language lacks the definite
article, it lacks the necessary material to assemble a negative raising predicate.
It follows that negative raising is possible only in DP languages.27

27. Recall that even languages disallowing strict NPI licensing under negative raising
allow negative raising negation interpretation. Boskovic and Gajewski (in press)
suggest that this is a pragmatic effect capturable in an approach like Horn (1989),
who argues that the lower clause understanding is a case of inference to the best
interpretation. Significantly, Gajewski (2005) shows this approach cannot explain
strict NPI licensing under negative raising (more specifically, it cannot create the
anti-additive environments needed for the licensing), which his semantic account
can do.

On NPs and Clauses

207

4.4. Back to Superlatives


Boskovic and Gajewski (in press) also give a deduction of (36) based on Hackls
(2007) proposal that most should be analyzed as the superlative of many (most
= many-est). Szabolcsi (1986) and Heim (1985, 1999) argue that -est can move
independently to take scope. Hackl shows that if we allow movement of -est in
most we can derive both the majority (MR) and the plurality (PR) reading. PR
corresponds to the comparative superlative reading discussed by Szabolcsi and
Heim and analyzed as -est taking clausal scope. Hackl shows that MR can be
derived if the -est of most stays inside the DP, taking scope below the article.
The ingredients of Hackls analysis are given below:
A. many has a modificational meaning of type <d,e,t>,<e,t>, unlike other
gradable adjectives, like tall, whose denotation is type <d,<e,t:
(80)

[[ many]](d)(N) = x.[N(x) & |x|d]

B. The superlative is a degree quantifier (cf. Heim 1999). C is the set of contextually relevant alternatives and D is a relation between degrees and individuals
(81)

a.
b.

[[-est]](C)(D)(x) is defined only if x C & y [y = x & y C] &


y C [ d D(d)(y)]
[[-est]](C)(D)(x) = 1 iff y C [y = x max{d:D(d)(x)} >
max{d:D(d)(y)}]

C. most = many + -est. -est is generated in the degree argument position of


many, namely SpecAP. Due to a type mismatch, -est must QR.
(82)

most = [AP

[DegP -estC ]
[A many] ]
d,<e,t,<e,t <d,e,t>,<e,t> mismatch!

When it moves, -est must target a node of type <e,t>. One option is local
adjunction to NP. Or, -est can move out of the TNP completely (Szabolcsi
1986, Heim 1999).
(83)

Bill owns (the) most Pere Ubu albums.


a. Bill owns [ DP (the) [NP -est [NP [AP t many ] [NP PB albums]]]]
b.

Bill [ -est [ owns [ DP (the) [NP [AP t many ] [NP PB albums]]]]

208

Zeljko
Boskovic

QR landing site Reading


TNP-internal: Majority (Bill owns more than half of the PB albums)
TNP-external: Plurality (Bill owns more PB albums than any relevant alternative individual does)
Movement out of TNP yields the plurality reading: when -est lands beneath
the subject, this establishes that individuals will be compared on the number
of albums owned. Hackls achievement is in showing that TNP-internal scope
yields the majority reading. The key to achieving this result is interpreting nonidentity of pluralities as non-overlap.
(84)

[ -esti [ ti many] PB albums ]

(84) then denotes a predicate true of a plurality of PB albums if it contains


more PB albums than any other non-overlapping plurality of PB albums. The
pluralities of PB albums that contain more PB albums than any non-overlapping
pluralities of PB albums are precisely those that contain more than half the
PB albums. A covert existential determiner quantifies over these, yielding the
majority reading.
Returning to the DP/NP parameter, in article-less languages the only option
that would yield the majority reading is adjunction to NP. Notice, however, that,
in contrast to DP languages, where DP is the argument, in NP languages, NP
is an argument. Chomskys (1986a) (see also Boskovic 2004b and McCloskey
1992) ban on adjunction to arguments then rules out local scoping of -est in NP
languages (but not in DP languages, where NP is not an argumental category),
ruling out the MR reading.28 Only long-distance movement of -est, which yields
the PR reading, is then available in NP languages (see Boskovic and Gajewski
(in press) for discussion of other superlatives.)
(85)

-EST movement in NP languages


a. Bill owns [ NP - est [NP [AP t many ] [NP PB albums]]]
b.

Bill [ - est [ owns [ NP [AP t many ] [NP PB albums]]]

28. This does not contradict the earlier assumption that APs are NP-adjoined. In Boskovic
(2005), I interpret the ban on adjunction to arguments derivationally. WhenAP adjoins
to NP in the SC counterpart of I like green cars, NP has not yet been merged as an
argument; when covert -est movement applies, NP is already an argument.

On NPs and Clauses

209

4.5. Back to Radical Pro-drop and Number Morphology


I now turn to the deduction of the pro-drop generalization in (45), which concerns
another clause-level phenomenon, namely licensing of pro-drop. It turns out that
the generalization can be captured in the same way as the number generalization
in (48). I will therefore start with the latter. To capture the generalization, I adopt
the condition in (86).29
(86)

The number feature of D must be morphologically realized.

What I mean by (86) is that an Agree relation that involves the number feature of
D must have morphological realization. Following Longobardi (1994), there is a
feature checking relation between D and N, which includes Agree for the number
feature. The relation must be morphologically realized. There are three ways of
doing this: realizing it on D, as in French and colloquial Brazilian Portuguese,
on N, as in English, or on both N and D, as in Bulgarian (I assume that what
counts here is the singular-plural opposition, which means that the lack of a
marker indicates singular in (87c); see also footnote 15).
(87)

a.
b.
c.

livr]
[l@
the-sg book
grad-@t
city-the
the book

[le
livr]
the-pl book
gradove-te
city(pl)-the(pl)
the books

(French)
(Bulgarian)

(86) captures (48) by requiring morphological realization of number morphology in DP languages, leaving it up to the morphological properties of the language/relevant lexical items to determine whether number morphology will be
realized in NP languages.
(86) also deduces the radical pro-drop generalization, providing a uniform
account of the generalizations in (48) and (45). In the case of phonologically null
pro, number morphology cannot be realized on either D or N. Yet, (86) requires
its realization. I suggest that this is done via verbal morphology. In other words,
(86) forces the presence of rich verbal morphology with pro-drop in DP lan-

29. (86) can be actually generalized to include -features in general, including the person
feature. However, since the person feature is hardly ever present in D (in fact, it is
possible that it is not present in non-pronominal TNPs), there may not be much
empirical difference between adopting (86) as it is and generalizing it. However, the
latter would obviously be conceptually more appealing.

210

Zeljko
Boskovic

guages, giving it the appearance of licensing pro-drop by verbal morphology.30


The licensing condition is irrelevant in NP languages because (86) is itself
irrelevant, languages in question not having the DP layer.31
A question that arises here is why there is a difference in the requirement of
morphological realization of number morphology between D and N. I suggest
that we are dealing here with a more general difference, where this type of
licensing requirements can hold only of functional elements.32

30. It appears that we now cannot force morphological realization of the Agree relation
between D and N (for number), if the number of D will be later morphologically
realized through rich verbal morphology. In other words, it appears that we may allow
number morphology not to be realized in DP languages with rich verbal morphology.
Consider (i).
(i)

T/v

What is relevant here is the timing of the relevant relationships. D and N enter into
an Agree relation before any relationship between a DP external and a DP internal
element is established. We can easily capitalize on this by requiring (86) to satisfy
Pesetskys (1989) Earliness Principle. This would require (86) to be satisfied as early
as possible, which means after the D-N relation is established in (i). Alternatively,
we can appeal here to cyclic spell-out, assuming that D and N are sent to spell-out
together before any TNP external elements are merged with it. (86) would then have
to be satisfied within this spell-out unit. (D and N will be sent to spell-out together if
we assume that DP is a phase and either that the whole phase is sent to spell-out or
that the edge of a phase is not sent to spellout but the edge contains the Spec but not
the head of the phase (this would still allow successive cyclic phrasal movement, but
head movement would be pushed outside of the syntax, as in Boeckx and Stjepanovic
(2001) and Chomsky (2001), among others).
31. The underlying assumption is that pronouns are NPs, not DPs, in NP languages, as
Fukui (1988) argues (see also Tomioka 2003; Boskovic 2008a; Despic in press, 2011;
Runic 2011).
32. Note incidentally that this assumption (more generally, assuming that only functional
categories are subject to syntactic licensing requirements) suffices to deduce the
radical pro-drop generalization. Let us assume that pro-drop is subject to a -licensing
requirement, as standardly assumed, and that only functional categories can be subject
to syntactic licensing requirements. Since pro-drop involves DP-drop in DP languages
it is subject to the -licensing requirement, which means that radical pro-drop is
disallowed in DP languages. The requirement cannot be imposed in NP languages,
since pro-drop involves NP drop in such languages, and NP is a lexical category.
(This type of analysis can be easily restated under the PF deletion/argument ellipsis
approach to radical pro-drop; see also Tomioka 2003 and Cheng in preparation for
relevant discussion under the ellipsis account of radical pro-drop).

On NPs and Clauses

211

A question that arises then is whether all functional categories should be


subject to this kind of a morphological realization requirement, i.e. whether
the morphological realization requirement can be generalized to hold for all
functional categories and all features. In the attempt to generalize (86), consider
first the Agree relation that D and N are involved in.
(88)

D (unvalued, interpetable #)

N (valued, interpretable #)

Following standard assumptions, the number of nouns is valued and interpretable. Capturing the intuition that D agrees with N in number, the number
feature of D is lexically unvalued, its value being determined through agreement with N. Following Sauerlands (2004) arguments that the number feature
of nominals should be interpreted in a high position,33 I assume that the number
feature of D is also interpretable.34 Given this, we may be able to generalize (86),
keeping the effects of (86) discussed above with respect to verbal morphology, as
in (89), where iK stands for an interpretable feature, and F for a functional head.
(89)

iK of F must be morphologically realized.

(89) may turn out to be too strong. For example, assuming that force is semantically interpretable and that it is encoded in C, different force specifications may
now require different morphological realizations of C, which seems too strong.
(89) can then be weakened as follows.
(90)

Unvalued iK of F must be morphologically realized.

Given the plausible assumption that C is lexically valued for the force feature,
the problem raised above is resolved. Notice, however, that depending on how
topicalization and focalization are treated, (90) may need to be stated in terms of
PF realization more generally. If topicalization and focalization are treated on
a par with Pesetsky and Torregos (2007) treatment of wh-movement, the topic
and focus heads (which attract topicalized and focalized phrases) would have
an unvalued interpretable topic/focus feature, which would receive its value
from the topic/focus phrase. This would make (90) relevant to topicalization/
focalization. Now, in most languages topicalized and focalized phrases do not
33. I am departing here from details of Sauerlands proposal, which places the feature in
question not only higher than N, but also higher than D.
34. I am following Pesetsky and Torrego (2007) and Boskovic (2009b) in not lexically
associating interpretability and valuation, as in Chomsky 2001, so that both uninterpretable and interpretable features can be either valued or unvalued. See the works
in question for empirical and conceptual arguments for this position.

212

Zeljko
Boskovic

morphologically differ from their non topic/focus counterparts (in contrast to


wh-phrases). However, topicalization and focalization are typically accompanied by a PF reflex, a pause in the case of topicalization and contrastive stress
in the case of focalization (note that focus movement typically affects only contrastively focused phrases, not phrases bearing simple new information focus).
I assume that this suffices to satisfy (90), with morphological realization now
taken to be PF realization.
I turn now to a rather interesting consequence of (90). Suppose that there is
an Agree relation between T and V for the tense feature, as argued by Pesetsky
and Torrego (2007). I will follow Pesetsky and Torrego regarding the exact
features involved in this Agree relation. They implement the Agree relation as
in (91), where T has an unvalued interpretable tense feature, and V has a valued
uninterpretable tense feature. (The underlying assumption here is that tense is
interpreted in T. However, the exact value of the tense feature of T depends on
the verb with which it co-occurs.)
(91)

T (unvalued

iTense)

V (valued uTense)

Since the tense feature of T is interpretable and unvalued (90) requires morphological realization of tense. In English, tense is generally morphologically
realized, except in the Present Tense, where the only morphology, -s in the third
person singular, is a -feature, not a tense feature. Interestingly, Enc (1991)
argues that precisely in this case there is actually no tense in English. In other
words, she argues that there is no Present Tense in English, the relevant constructions lacking tense.35
It should be noted, however, that generalizing (86) does not necessarily require morphological realization of tense. Even if we keep the T-V feature checking relation from (91),36 there is a way of generalizing (86) that keeps its effects
for number morphology but does not require morphological realization of tense.
Thus, we can generalize (86) as in (92), where proxy values are defined in (93),
the intuition here being that percolation, i.e. passing on, of values of interpretable
features from lexical (L) to functional (F) categories requires morphological realization.
(92)

Proxy values must be morphologically realized.

(93)

Proxy: unvalued iK of F which receives its value from iK of L.

35. Alternatively, the opposition -ed vs may suffice here, where the lack of -ed would
indicate Present Tense.
36. The relation can of course be changed in such a way that it would not be affected
by (90).

On NPs and Clauses

213

The Agree relation between T and V from (91) now does not require morphological realization of tense, since the relevant feature of V is uninterpretable.
However, (92) still captures the effect of (86) regarding the presence of number
morphology within TNP.37
Suppose, however, that we adopt (90), which requires morphological realization of tense given (91), as the correct generalization of (86). From this
perspective, the lack of morphologically-present tense in e.g. Chinese (I am
putting aspect aside here) would then imply quite generally the lack of TP in
Chinese. There are in fact some pretty strong arguments that there is no need to
posit TP for Chinese, since temporal interpretation can be easily derived from
aspect or temporal adverbs (see Lin 2003, 2005 and Smith and Erbaugh 2005,
among others38 ).
37. It does not, however, capture its effect regarding the impossibility of radical pro-drop
in DP languages, unless we assume that pronouns have both D and N. Recall, however
that there is an alternative way of deducing the radical pro-drop generalization (cf.
footnote 32), i.e. (92) is not necessarily needed for this generalization.
38. Lin (2003, 2005) and Smith and Erbaugh (2005) argue that temporal interpretation
is derived from aspect even in examples without morphologically realized aspect
markers. Consider (i)(ii), where (i), involving a telic verb, can only have past interpretation, and (ii), involving an atelic verb, can only have present interpretation.
(i)

Ta dapuo yi-ge hua


ping.
he break one-Cl flower vase
He broke a flower vase.
(only past)
(ii) Wo xiangxin ni.
I
believe you
I believe you.
(only present)
The gist of the analysis pursued by Lin (2003, 2005) and Smith and Erbaugh (2005)
is that temporal interpretation comes from aspect, past from perfective and present
from imperfective, with the default aspect (which is what we have in (i)(ii)) for
telic events being perfective, and for atelic events imperfective. Smith and Erbaugh
implement this via (iii).
(iii) Temporal schema principle:
In a zero-marked clause, interpret a verb constellation according to the temporal schema of its situation type, unless there is explicit or contextual information to the contrary.
They argue that if an event is taking place at the speech time, it is inconsistent with
the telic meaning that indicates that the event has an endpoint. Then, telic verbs without aspectual markers or adverbs cannot appear with present interpretation. They
argue that bare telic verbs cannot be interpreted as future tense because past tense

214

Zeljko
Boskovic

Interestingly, there are NP languages that have very rich verbal morphology;
yet, they seem to lack tense morphology. This is, e.g., the case with SC. Consider
the SC paradigm for present and aorist.
(94)

Present Tense for to love: volim, volis , voli, volimo, volite, vole
Aorist for to love: volih, voli, voli, volismo, voliste, volise
Infinitive for to love: voli-ti
(SC)

There is quite a bit of morphology in (94). However, it is all -related, there is no


overt tense morphology. There is a very surprising difference with English here:
Although SC is way richer with respect to verbal morphology than English, it
is actually poorer than English with respect to tense morphology. On the other
hand, there is tense morphology in Bulgarian (see Scatton 1984), a DP language
that is otherwise closely related to SC. Could we be dealing here with another
difference between DP and NP languages? In the next section I will explore this
possibility.
Before discussing the issue, let me briefly note that we may now be in a
position to deduce (at least partially) the Lobeck (1990)/Saito and Murasugi
(1990) generalization that only agreeing functional heads may license the ellipsis

is simpler than future tense in that the latter contains modal interpretation. (What
is relevant here is a pragmatic principle they adopt which favors interpretation that
requires the least additional information). Finally, if the event represented by a verb
is not bounded, the default interpretation is present because the event is on going and
temporally open.
It should be noted that when they are present, aspectual markers and temporal adverbs
determine the temporal interpretation of the sentence. This is illustrated for the latter
in (iv).
(iv)

a.

b.

c.

Ta zuotian
hen mang.
he yesterday very busy
He was very busy yesterday.
Ta xianzai hen mang.
he now
very busy
He is very busy now.
Wo mingtian hen mang.
I
tomorrow very busy
I will be very busy tomorrow.

The case for deducing temporal interpretation from aspect/adverbs in Chinese thus
seems quite strong. (Note also that Hu et al. 2001 argue against the finite/non-finite
distinction for Chinese; see also Lin 2010 for the lack of Tense in Chinese.)

On NPs and Clauses

215

of their complement, a generalization that has been quite extensively appealed


to in the literature on ellipsis but has eluded deeper understanding.39
Roughly following Merchant (2008), suppose ellipsis involves marking of the
functional head F whose complement is elided with an additional interpretable
feature X that indicates ellipsis, which in the spirit of (89)/(90) requires additional morphological realization, not otherwise found on the head F. This is
accomplished by having F undergo Agree with another, morphologically realized element, which then undergoes movement to SpecXP to avoid undergoing
deletion (since the element in question realizes the additional feature, it must
escape undergoing ellipsis).40
5.

No TP in Article-less Languages

Returning now to the issue of whether NP languages have TP, positing a difference in the availability of TP between DP and NP languages can in fact be easily
justified theoretically. Suppose we assume that DP is the counterpart of IP, not
39. More precisely, Lobeck (1990) and Saito and Murasugi (1990) note that functional
heads can license ellipsis of their complement only when they undergo Spec-Head
agreement (SHA). Thus, (i) shows that tensed INFL, s, and +wh-C, which according
to Fukui and Speas (1986) undergo SHA, license ellipsis, whereas the non-agreeing
functional categories the and that do not.
(i)

a.
b.

John liked Mary and [IP Peter i [I did ti like Mary]] too.
Johns talk about the economy was interesting but [DP Bill [D s talk
about the economy]] was boring.
c. *A single student came to the class because [DP [D the student]] thought
that it was important.
d. John met someone but I dont know [CP who i [C C John met ti ]].
e. *John believes C/that Peter met someone but I dont think [CP [C C/that
Peter met someone]].
40. This way of looking at the Lobeck/Murasugi and Saito generalization requires agreement with a morphologically realized element. Since agreement with PRO does
license ellipsis this leads us to Hornsteins (1999) approach to PRO, where PRO is
actually a copy of a moved element. Assuming the relevant condition is checked
when the moved element is still within the projection of the head X, there will be no
problems with respect to the licensing condition on ellipsis discussed above if the
moving element is itself morphologically realized. This approach will likely require
considering some phonologically null elements to arise through PF deletion of overt
elements (see in this respect Takahashi 1997, who argues for an analysis along these
lines for several constructions that were traditionally assumed to involve null operator
movement), with the condition in question checked before the deletion.

216

Zeljko
Boskovic

CP as is often assumed, which is not implausible given that SpecDP is the host
of the counterpart of movement to SpecTP in examples like Johns destruction
of the painting under Chomskys (1986b) analysis. Suppose furthermore that
we take the TNP/Clause parallelism hypothesis seriously, where the lack of DP
in a language would imply the lack of its clausal counterpart, namely TP (assuming with Chomsky 1995, 2000, 2001 that TP stands for the IP of the GB
framework). It would then follow that TP should be absent in NP languages (or
perhaps it is weak, as in Tsai 2008).41 This fits some NP languages rather nicely.
For example, the assumption can capture Chinese and the surprising case of
SC, which we have seen lacks tense morphology although it has very rich verbal morphology (see in this respect Paunovic 2001 for arguments that SC does
not have grammaticalized tense (i.e. TP), temporal interpretation being derived
from aspect and mood). There are, however, NP languages that are traditionally
considered to have tense morphology, like Japanese and Turkish. The Turkish
case is actually quite controversial. A number of authors have argued that what
has been traditionally considered to be tense markers in Turkish are in fact aspect
and/or modal markers. Thus, the aorist {-Ar/Ir} is argued to be an aspect and
modal marker by Yavas (1981, 1982b) and Giorgi and Pianesi (1997), {-mIs}
is treated as an aspect and a modal marker by Slobin and Aksu (1982), {-DI}
is also treated as an aspect and modal marker by Taylan (1988, 1996, 1997),
{-AcAK} is also treated as a modal marker by Yavas (1982a), and {-Iyor} is
treated as an aspect marker by Giorgi and Pianesi (1997). The relevant state
of affairs is actually also not completely clear even in Japanese, which Fukui
(1988) argues lacks TP (see also Whitman 1982 as well as Shon et al. 1996 and
especially Kang 2012 for a similar view for Korean).
It is, however, important to notice that analyzing the traditional tense morphology in Japanese and Turkish as actual tense morphology does not necessarily
force us to posit TP for these languages. Consider again (91). In (91), tense is
represented in two different structural places, T and V, and interpreted only in
the former. Suppose, however, that the tense feature of V is interpretable in a
language. In such a language there would be no semantic need for T (as far as
temporal interpretation is concerned), since temporal interpretation would come
from the verb. It is then possible that Japanese and Turkish do have temporal
verbal morphology. However, since the tense on the verb is interpretable, the
41. It is important to bear in mind that adopting a no-DP analysis of article-less languages
does not require adopting a no-TP analysis for such languages, i.e. the absence of
DP in a language does not have to be correlated with the absence of TP, which means
that if it turns out that (some) article-less languages do have TP, the no-DP analysis
of such languages will not be invalidated.

On NPs and Clauses

217

languages can still be considered to lack T (the morphology itself would be part
of the morphologically complex verb, a suggestion that was actually made by
Fukui (1988) for Japanese). The state of affairs can in fact be nicely captured
within the system developed in Osawa (1999), who also argues that languages
differ with respect to the presence of TP (in fact, he argues that the property in
question can be affected by historical change). Parallel to the line of research
pursued by Higginbotham (1985), who argues that nouns have an open position,
Osawa argues that verbs have an open event position which must be saturated
through binding. In TP languages, the event position is bound by T. On the other
hand, Osawa argues that in languages lacking TP the event position is bound
by a temporal/aspectual affix on the verb (in a language like Chinese, where
adverbs affect temporal interpretation, the event position can also be bound by a
temporal adverb). Osawas analysis can be considered to be an implementation
of the above suggestion that in some languages the tense feature of the verb is
interpretable, given that on Osawas analysis in languages where a verbal affix
binds the event position of the verb there is no need for T to accomplish temporal interpretation. What is important for our purposes is that a mere presence of
temporal verbal morphology does not necessarily require positing a dedicated
TP projection, as already pointed out by Fukui (1988) (see also Whitman 1982).
Granted that assumption, a question still arises regarding other TP-related
effects, such as movement to SpecTP. Such movement can be easily reanalyzed
as movement to a projection other than TP, which would not be surprising if
the traditional IP is split at least to some extent. Research on quantifier float,
V-movement, and multiple subject positions quite clearly shows that a simple
structure where the only A-related phrase above the projection where a subject
gets a -role is TP is clearly wrong. The simple TP-over-vP structure cannot
account for the fact that it is possible to float a quantifier in between the and
the surface position of a subject which does not undergo A -movement (see
Boskovic (2004b) for crosslinguistic data to this effect), or that there are many
languages where the verb is lower than in Spanish but higher than in English
(this is e.g. the case with most Slavic languages, see Boskovic (2001), and with
French infinitives, see Pollock (1988)); if the Spanish verb is in T, and the English
verb is in v, where would then the verb be in these languages if all we have above
vP is TP? Multiple subject position languages like Icelandic also very clearly
have two subject positions above the -position of the subject (see Bobaljik and
Jonas 1996; Bobaljik and Thrainsson 1998; Jonas 1996; Vangsnes 1995), which
again cannot be accommodated under the simple TP-over-vP structure. It thus
seems quite clear that there should be more A-related structure above vP; TP
alone is not enough. Since what I am arguing here is merely that one layer of
clausal structure is missing in NP languages, there is still room for accommodat-

218

Zeljko
Boskovic

ing A-movement of a subject in no-DP/no-TP languages. The point made here


is simple, the presence of tense morphology on the verb or A-movement of a
subject does not necessarily require positing a TP projection. But then, how can
one argue that TP is present/absent in a language? Fortunately, there are better
tests for the presence of TP that can be conducted to test the hypothesis that the
TP projection is missing in article-less languages. I will consider the issue by
focusing on SC and Japanese/Korean as representatives of article-less language
(for discussion of Japanese and Korean, see also Fukui 1986, 1988; Fukui and
Takano 1998; Fukui and Sakai 2003; Kang 2012; and Shon et al. 1996).
5.1. Subject Expletives
First, one property of the TP projection is the traditional EPP, which requires
filling SpecTP. When nothing moves to SpecTP, the position is often filled by an
overt expletive. In fact, it appears that the only function that an overt expletive
like there has in English is to satisfy the EPP. Japanese, Korean, and SC do not
have overt expletives. In fact, article-less languages seem to lack uncontroversial
overt subject expletives like English there (see here Cheng in preparation).
If this is true, the question is why that would be the case.42 The hypothesis
42. Franks (1995) notes potential candidates for overt expletives in several Slavic languages. However, the elements in question are either used to indicate discourse focus
or emphasis, as indicated by Czech (i), where the overt expletive (v)ono is not possible, or are located in SpecCP, as argued quite conclusively by Lindseth (1993) for
Upper Sorbian wone, or function as an event pronominal located in the CP field, as
shown with a number of convincing arguments by Progovac (2005) for Serbian to.
In other words, we are not dealing with true subject expletives here.
(i)

*(v)ono prsi,
ale svt
it
is-raining but sun

slunce.
is-shining

The only potential counterexample to the claim that article-less languages lack true
overt expletives I am aware of involves Finnish (see Holmberg and Nikanne 2002).
For the sake of argumentation, I will assume below that article-less languages do not
have overt expletives, leaving a detailed investigation of Finnish for future research.
I merely note here that at least in some cases the element in question (sita ) does
not seem to be located in SpecIP since it precedes the question particle -ko and the
focus-particle -han (recall also that at least colloquial Finnish has developed a true
definite article, so we may be dealing here with a system undergoing a change).
(ii)

a.

Sitako ovat teidan lapset


jo
kaikki kayneet uimassa?
exp-q have your children already all
been
swimming
Have your children already all been swimming?

On NPs and Clauses

219

pursued here, that article-less languages do not have the TP projection, provides
a straightforward explanation for this state of affairs, given that expletives are
introduced to satisfy the EPP, a property of the TP projection (Fukui 1986, 1988
also uses the lack of subject expletives in Japanese to argue against the presence
of (syntactically active) TP in Japanese). On the other hand, the lack of overt
expletives in article-less languages is a real mystery if such languages do have
TP. Consider what could be appealed to in this scenario. First, if a language
has Spanish-style pro-drop (which SC does have), where overt pronouns are
typically used for emphasis, we might not expect to find an overt expletive in such
a language, since expletives cannot be used for emphasis. It is actually not clear
that this explanation works for Spanish-style pro-drop languages since Galician
Portuguese, which has this type of pro-drop, does have an overt expletive (see
Franks 1995). At any rate, even if the account can be made to work, it would
extend only to the Spanish-style pro-drop. In languages that have pro-drop of
the Japanese type (what I called radical pro-drop, which we have seen is in fact
limited to article-less languages), overt pronouns are not used only for emphasis,
so the above consideration regarding Spanish-style pro-drop would not apply
to them. We are then still left with the question of why article-less languages
lack overt expletives. One could assume here that the EPP is parameterized (see
McCloskey 1997 and Wurmbrand 2006). We would then not expect to find overt
expletives in all article-less languages. But we would still certainly expect to find
it in some, if not many, languages of this type. The upshot of the above discussion
is that if article-less languages indeed consistently lack overt expletives, which
seems to be the case, this will provide very strong evidence that such languages
lack TP. The no-TP hypothesis provides a straightforward explanation for the
lack of overt expletives in article-less languages, which otherwise represents a
real mystery.
5.2. Subject-Object Asymmetries
Another relevant test concerns well-known subject-object asymmetries of the
kind found in English. One such asymmetry concerns the that-trace effect, where
an object, but not a subject can be extracted across a clause-mate that.43

b.

Sitahan
ei nykyaa n puhuta vakoilusta.
exp-prtcl not nowadays talk-pass espionage-abl
We dont talk about espionage these days, do we?
(Holmberg and Nikanne 2002: 95)
43. For ease of exposition, I am simplifying here the actual state of affairs. The same
holds for the discussion regarding extraction out of subjects/objects directly below.

220
(95)

Zeljko
Boskovic

a. Whoi do you think that John saw ti ?


b. *Whoi do you think that ti saw John?

Another asymmetry concerns extraction out of subjects/objects. As is wellknown, English allows extraction out of objects, but not subjects.
(96)

a. *Who did friends of see you?


b. Who did you sees friend of?

Turning now to Japanese/Korean and SC, these languages do not display the
subject-object asymmetries in question (see here Shon et al. 1996 regarding
Korean. (97) actually cannot be tested in Japanese since the language disallows
scrambling of -ga marked phrases.)
(97)

a.

b.

I
mokcang-i i
Chelswu-ka [ti caknyen-kkaci-man hayto
this meadow-nom Chelswu-nom last year-until-just
kwaswuwen-i-ess-ta-ko] malha-yess-ta.
orchard-be-pst-dc-comp say-pst-dc
Chelswu said that this meadow used to be an orchard just until
last year. (Shon et al 1996)
Nwu-ka i [ne-nun [ti nay cacenke-lul
who-nom you-top
my bicycle-acc
hwumchie-ka-ss-ta-ko sayngkakha-ni?
steal-go-pst-dc-comp think-q
Who do you think took my bicycle away?

(98)

da ti voli Mariju?
Ko i tvrdis
who you-claim that
loves Marija
Who do you claim loves Marija?

(99)

[OP[ Mary-ga
t yonda-no]-ga aikarana yorimo John-wa
Mary-nom read that-nom is obvious than
John-top
takusan-no hon-o
yonda].
many-gen book-acc read
John read more books [than Mary read ] is obvious. (Japanese,
Takahashi 1994)
i tvrdis
Ciji
da [ti otac] voli Mariju?
(SC)
whose you-claim that father loves Marija
Whose father do you claim loves Marija?

(100)

(SC)

It should be noted, however, that the asymmetries may also be missing in article
languages. They are indeed tests for movement to SpecTP, since they affect only

On NPs and Clauses

221

subjects in this position (see Stepanov (2001a, b) for evidence that extraction
is crosslinguistically disallowed out of subjects that must raise to SpecTP).
However, since subjects do not have to always move to SpecTP in all article
languages, such asymmetries can also fail to surface in article languages. One
such language is Spanish, which does not show the that-trace effect and allows
extraction out of subjects, but crucially only out of postverbal subjects, which
do not move to SpecTP. Extraction is impossible out of preverbal subjects,
which do move to SpecTP (see Gallego and Uriagereka 2007). The question is
then whether the above subject-object asymmetries are ever found in articleless languages. I am in fact not aware of any article-less language that would
exhibit such asymmetries.44 If it indeed turns out that such asymmetries are
never found in article-less languages we will have here very strong evidence
that such languages lack TP. Since the asymmetries arise only in the case of
subjects that move to SpecTP, the lack of such asymmetries is straightforwardly
captured if the languages in question lack TP.
To summarize, if article-less languages indeed turn out to consistently lack
true subject expletives and consistently fail to exhibit subject-object asymmetries, as seems to be the case, we have here strong evidence that article-less
languages lack TP.45 This in turn follows once the TNP/Clause parallelism hypothesis is taken seriously, given that TP is the clausal counterpart of DP, which
is missing in article-less languages.46
44. Thus, SC, Russian, Palauan, Chinese, Japanese, Turkish, Hindi, and Navajo all allow
extraction out of subjects. Note that non-colloquial Russian disallows subject extraction across the counterpart of that; however, it also disallows object extraction, which
means that it does not exhibit a subject-object asymmetry here.
45. It should, however, be noted that it is certainly not out of question that any movement
of the subject, not just movement to SpecTP, could lead to the problems noted in the
text regarding extraction from/of subjects. Since TP-less languages may still have
subject movement, as discussed above, the problems in question could still arise in
TP-less languages. Under this analysis, article-less languages (i.e. languages without
TP) would only be much less likely to exhibit subject-object asymmetries than DP/TP
languages.
46. In principle, V-to-T head movement is another relevant phenomenon, since it should
be missing in article-less languages. It is, however, rather difficult to test whether such
movement is indeed absent in article-less languages for several reasons: assuming
that the traditional IP should be split, even if TP is missing there would still be head
positions above vP and below CP where a verb could move in article-less languages.
We may still be able to run the relevant test if TP is the highest head in the Split
IP field. Assuming that the finite verb in Romance languages like French, Spanish,
and Italian moves to T, we would then expect to find article-less languages with

222

Zeljko
Boskovic

5.3. Nominative Case


An obvious question arises at this point. If article-less languages lack TP, what is
responsible for nominative Case assignment in such languages? One possibility
is that a head other than T is responsible for this. As an illustration, if IP is
split into TP and AgrsP, it may be that AgrsP is responsible for nominative case
assignment, at least in languages lacking TP. Nominative case in Turkish, an
article-less language, indeed correlates with (and has been argued to be assigned
by) agreement (see George and Kornfilt 1981, Kornfilt 1984, 2005, 2006). Under
this analysis it can be assumed that a head other than T assigns nominative case
in all languages (C has also been occasionally assumed to be the source of
nominative Case), or that TP and non-TP languages differ in this respect, a head
other than T assigning nominative only in the latter. The variation would not
be particularly surprising given that it has been argued that nominative can be
assigned by different heads even within a single language (see Harley 1995,
Sigursson 1996, 2000, McGinnis 1998, Alexiadou 2003, and Boeckx 2003b
for Icelandic; and Tada 1993, Takahashi 1996, and Saito and Hoshi 2000 for
Japanese). There is, however, another possibility that does not require positing
more than one source for assignment of structural nominative case. It is possible
that nominative case is not a structural case in non-TP languages. If correct, this
will allow us to maintain T as the sole source of structural nominative licensing
crosslinguistically. Saito (1985) has in fact argued that Japanese -ga is not a
structural case (licensed by tense; see also Fukui 1986, 1988). It is in fact very
easy to justify this claim; Japanese -ga in many respects simply does not behave
like regular structural nominative case. One well-known respect in which -ga
departs from regular structural nominative concerns the well-known multiple
-ga construction, where non-subjects receive -ga in addition to the subject. One
relevant example, together with its English counterpart, is given below.
(101) *Civilized countries, male, the average life span is short.

V-movement, which, however, would not go as high as in French/Spanish/Italian


(unless the landing site of the movement is C). Most Slavic languages indeed seem
to behave like this, with the verb able to move higher than in English but not as
high as in French/Spanish/Italian. Notice also that many relevant languages are SOV
languages, where it is quite difficult to determine whether the verb undergoes Pollockstyle head movement (notice also that Koizumis (1994) analysis on which Japanese
verb moves to C is consistent with the no-TP movement analysis; it is however far
from clear that there is any kind of V-movement in Japanese, see in this respect Fukui
and Sakai (2003), who argue against V-movement to either T or C in Japanese).

On NPs and Clauses

(102)

223

Bbunmeikoku-ga
dansei-ga heikinzyumyoo-ga
Civilized countries-nom male-nom average lifespan-nom
mizikai.
is short
It is civilized countries that men, their average lifespan is short in.
(Kuno 1973)

Moreover, Fukui and Sakai (2003) observe that what gets -ga does not have
to be a syntactic constituent, and it does not have to be an NP (PPs and some
clauses such as those headed by -ka Q can also get -ga), again a non-standard
behavior from the point of view of standard assumptions regarding structural
nominative.
The well-known operation of ga/no conversion, where a subject of what
should be a finite clause fails to get -ga, instead getting genitive from a higher
noun, is another illustration of non-standard behavior of -ga.
(103)

Taroo-ga/
-no it-ta
tokoro
Taroo-nom/ -gen go-past place
the place where Taroo went

It is well-known that in contrast to objects marked for structural accusative,


caseless objects cannot scramble. The fact that -ga marked NPs cannot scramble, as noted by Saito (1985), is significant in this respect, given that -ga marked
elements here pattern with caseless objects, not with objects marked with accusative case.
(104)

a.

John-ga
dare(-o) nagutta no?
John-nom who-acc hit
Who did John hit?
b. Dare-o John-ga nagutta no?
c. *Dare John-ga nagutta no?

(Saito 1985: 267)

(105) *Sono okasi-ga


John-ga
[oisii to] omotte iru (koto)
that candy-nom John-nom tasty that think
fact
John thinks that that candy is tasty.
(Saito 1985: 210)
The above discussion shows that Japanese -ga quite clearly does not behave like
regular structural nominative case. There are, however, article-less languages
where traditional nominative does not exhibit such exceptional behavior. Unless we want to fall back on the possibility of more than one nominative case
licensor (or non-T nom case licensor), there is only one possibility for such
languages: nominative case in such languages is assigned by default. We are

224

Zeljko
Boskovic

then making a prediction: nominative case in article-less languages will either


exhibit exceptional behavior (like Japanese -ga) or it will function as default
case. As is well-known, default case for English and French is accusative, as
shown by the accusative case on the pronoun in (106) (in an out-of-the-blue
context).
(106)

Me /*I intelligent?!

A survey of which case functions as a default case in article-less languages


reveals that nominative is indeed the default case in Polish, Russian, Slovenian,
SC, Turkish, Hindi, and Korean. In fact, I am not aware of any article-less
languages where a case other than nominative functions as a default case. In
light of the above discussion, I conclude that the presence of nominative case
does not argue against the no-TP analysis of article-less languages.
Before turning to a rather strong argument for the lack of TP in article-less
languages in the next section, I will briefly mention one potential argument for
the lack of TP in Japanese. Tanaka (2002) and Nemoto (1991) provide strong evidence that A-movement across CP boundaries is possible in Japanese. Tanakas
examples involve raising to object out of finite CPs. (107 a)(107b) give the
base-line data, with (107b) being the relevant example. (107d), where stupidly
modifies the matrix V, shows the accusative subject moves into the matrix clause
(in contrast to the nominative subject in (107c)). (108) and (109) argue against
an alternative, control analysis. (108a) illustrates the well-known fact that the
Proper Binding Condition holds for movement in Japanese (see Saito 1992),
though it is irrelevant for control (109). The fact that (108b) patterns with (108a)
rather than (109) indicates that the construction under consideration involves
movement (into the matrix clause) rather than control. (111b), where what used
to be the embedded clause subject binds an anaphor in the matrix subject, shows
that raising to object can be followed by A-scrambling to the sentence initial
position of the higher clause, which confirms the raising to object analysis (i.e.
it shows the relevant movement involves A-movement, as expected under the
raising to object analysis). Finally, (110) shows raising to object can oocur even
out of +wh clauses, which are uncontroversially CPs.
(107)

a.

b.

John-ga
[Bill-ga
baka-da-to]
omot-teiru.
John-nom [Bill-nom fool-cop-comp] think-prog
John thinks that Bill is a fool.
[ti baka-da-to]
omot-teiru.
John-ga
Bill-oi
John-nom Bill-acci [ti fool-cop-comp] think-prog
John thinks of Bill as a fool.

On NPs and Clauses

225

c. *John-ga
[Bill-ga
orokanimo tensai-da-to]
John-nom [Bill-nom stupidly
genius-cop-comp]
omot-teiru.
think-prog
Stupidly, John thinks that Bill is a genius.
d. John-ga
Bill-o i
orokanimo [ti tensai-da-to]
John-nom Bill-acci stupidly
[ti genius-cop-comp]
omot-teiru.
think-prog
John thinks of Bill stupidly as a genius.
(108)

a. *[[Bill-ga
ti katta-to]j
[sono-hon-o i [John-ga
tj
[[Bill-nom ti bought-comp]j [the book-acci [John-nom tj
itta]]].
said]]]
[That Bill bought ti ]j , the booki , John said tj .
John-ga
Bill-o i
tj omot-teiru.
b. *[ti baka-da-to]j
[ti fool-cop-comp]j John-nom Bill-acci tj think-prog
[ti as a fool]j , John thinks of Billi tj .

(109)

[proi gakko-ni iku yoo-ni]j


John-ga
Bill-ni i
tj
school-to
go
John-nom
proi
in-order-to]j
Bill-dati tj
meizita.
ordered
John ordered Bill to go to school.

(110)

John-ga
Bill-o
baka-ka-to kangaeta.
John-nom Bill-acc fool-q-comp consider
John wonders if Bill was a fool.

(111)

a. ??Otagaii -no sensei-ga


karera-o i [ti baka-da-to]
each otheri s teacher-nom them-acci [ti fool-cop-comp]
omot-teiru.
think-prog
Each otheri s teachers think of themi as fools.
b. Karerai -o otagaii -no
sensei-ga
ti [ti baka-da-to]
themi -acc each otheri s teacher-nom ti [ti fool-cop-comp]
omot-teiru.
think-prog
Themi , each otheri s teachers think of ti as fools.

Turning now to Nemoto (1991), her examples involveA-movement out of control


CPs (the fronted element binds the anaphor in (112a)).

226
(112)

Zeljko
Boskovic

a.

b.

John to Bobi -o
otagaii -no titioyaj -ga [CP proj ti
John and Bob-acc each others fathers-nom
rikaisiyoo to] kokoromita.
understand C attempted
John and Bob, each others fathers attempted to understand.
*Otagaii -no titioya-ga [CP pro John to Bob-o rikaisiyoo to] kokoromita.

Regarding (112), Nemoto (1991) assumes that A-movement cannot skip CP/TP
pairs, following Chomsky (1986a). It must then be that either CP or TP is missing
in (112). Nemoto then argues that since CP is clearly present in (112) it must be
that the embedded clause lacks TP. The analysis also straightforwardly extends to
(107b)/(110)/(111b). The possibility of A-movement out of CPs in Japanese is
suggestive of a rather strong argument for the no-TP analysis. While a question
still arises how to block A-movement out of CPs where such movement is not
possible, what is important here is that such movement is in principle possible
in Japanese.47
I now turn to a rather strong argument for the lack of TP in article-less
languages, which involves the Sequence-of-Tense phenomenon.
5.4. Sequence of Tense
Consider English examples in (113) and (114). (113) is ambiguous between the
non-past/simultaneous and the anteriority reading. As for (114), the time of the
alleged illness must contain not only the time of Johns believing, but also the
utterance time (see, e.g., Sharvit 2003).
(113)

John believed that Mary was ill.


Non-past/simultaneous reading: Johns belief: Mary is ill (time of the
alleged illness overlaps Johns now)
Anteriority reading: Johns belief: Mary was ill (the time of the alleged
illness precedes Johns now)

(114)

John believed that Mary is ill.


The time of the alleged illness contains the time of Johns believing
state and the utterance time

47. The issue, however, arises not only in Japanese but in other article-less languages
as well. It is possible that a projection from split IP other than TP (which would be
present in article-less languages) is to blame for this. Another rather straightforward
possibility is that an A-movement that is blocked is driven by a feature that the CP
in question also has (the movement would then be blocked by Attract Closest).

On NPs and Clauses

227

The above illustrates typical behavior of a language that exhibits the Sequenceof-Tense phenomenon (SOT). It is well-known that not all languages exhibit
SOT. Thus, SC (115) has the non-past reading. More generally, the simultaneous
reading of English examples like (113), where the embedded clause is in the
Past Tense, is expressed with a structure where the embedded clause is in the
Present Tense in SC. Furthermore, (116) can only have the anteriority reading
in SC.48
(115)
(116)

Jovan je vjerovao da je Marija bolesna.


Jovan believed that Mary is ill.
Jovan je vjerovao da je Marija bila bolesna.
Jovan believed that Mary was ill.

(nonpast/simultaneous)
(only anteriority)

English and SC can then be taken to illustrate the behavior of SOT and non-SOT
languages respectively. Is the variation between SOT and non-SOT languages
arbitrary? I am not aware of any proposals in the literature to the effect that
it is not. A preliminary crosslinguistic investigation of how languages behave
regarding SOT shows the following language division:
(117)

a.
b.

SOT languages: English, Dutch, Modern Greek, Spanish, French,


German, Italian
non-SOT languages: Russian, Polish, Czech, SC, Romanian,
Hebrew, Japanese, Korean, Hindi, Turkish, Malayalam, Bangla,
Angika

There is a generalization lurking behind the language division in (117). While


DP languages are found in both the SOT and the non-SOT group, NP languages
are uniformly located in the non-SOT group. This leads me to posit the generalization in (118). (In fact, if it were not for Hebrew and Romanian we would have
a two-way generalization here. It is possible that a more careful examination of
Hebrew and Romanian,49 or a more fine-grained distinction between SOT and
non-SOT languages, will yield a true two-way correlation.)
48. This does not appear to be the case in all non-SOT languages. I therefore take the
availability of the non-past reading for examples like (114)(115) as the more reliable
test for the presence/absence of SOT in a language.
49. It could be relevant that the definite article in both Hebrew and Romanian is an affix.
As discussed in Boskovic (2008b) and Despic (2011), affixal articles behave differently from non-affixal articles in some rather surprising respects (e.g. the distinction
determines the behavior of a language with respect to wh-islands, see Boskovic
(2008b)). It is, however, unclear why this would matter here.

228
(118)

Zeljko
Boskovic

Languages without articles do not show Sequence of Tense.

This surprising generalization, which is another illustration of a surprising interaction between NP-level and clause-level phenomena, falls into place rather
nicely under the approach pursued here, where article-less languages lack TP.
If TP is needed to impose Sequence of Tense, it follows that languages without
articles will fail to exhibit Sequence of Tense, since they lack TP.50
50. I-Ta Hsieh (p.c.) observes that the generalization can in fact be quite straightforwardly
deduced under the Stowell (1993, 1995a, b)/Kusumoto (2005) approach to SOT. In
this approach, a predicate has an argument slot for time; the past tense morpheme in
an SOT language like English functions as a time variable which receives its value
from higher operators like the phonetically null anteriority operator PAST. (In other
words, the past tense morpheme is like a polarity item that needs to be licensed by
PAST.) The anteriority meaning is then introduced by the operator PAST, which I
assume is located in T, not the past tense morpheme itself. On the anteriority reading
of (113), there is a PAST operator in both the matrix and the embedded T, as a result
of which the two past tense morphemes receive their values from different operators.
On the simultaneous reading of (113), on the other hand, PAST is present only in the
matrix T, hence both past tense morphemes are licensed by the same operator; the
attitude verb here quantifies-in over the time variable introduced by the embedded
past tense morpheme. In NP languages, due to the lack of TP, the operator PAST
is unavailable. As a result, such languages cannot have an English-like past tense
morpheme, which semantically merely introduces variables and is licensed by PAST.
Elements that these languages have instead of English past must carry a lexically
specified meaning and contribute to the temporal interpretation on their own by saturating the time argument slot of the predicate (as for their location, one possibility
is AspP). When these elements are embedded in the complement of an attitude predicate that is anchored with the past tense, since they are not variables they cannot be
quantified-in by an intensional verb like say. As a result, the simultaneous reading for
(113) is unavailable in the case of past-under-past in an NP-language. (As noted by
I-Ta Hsieh (p.c.), the elements in question can be treated as generalized quantifiers
over time, which, with the assumption that the topic time is saturated by the utterance
time in the root context and by the matrix event time in the embedded context, can
quite straightforwardly yield the simultaneous reading for pres-under-past in nonSOT languages.) Let me finally point out that while an NP language cannot have a
PAST operator due to the lack of TP, a DP language can still lack the operator as an
idiosyncratic property, which is consistent with the one-way nature of the generalization in (118) (recall that some DP languages lack SOT).
Khomitsevichs (2007) system can also be quite straightforwardly modified to deduce (118). On Khomitsevichs analysis SOT results from a series of Agree relations
through which the embedded verb is valued for the tense feature by the matrix verb.
Modifying the original system, this can be implemented as a V-T-V series of Agree

On NPs and Clauses

229

5.5. Subject Reflexives


Finally, I note here a very interesting generalization in the context of a noTP analysis of article-less languages noted by Despic (2011). Despic examines
subject reflexive constructions like (119) and observes there is a generalization
regarding languages that allow such constructions. In particular, he observes
such constructions are allowed only in the subset of article-less languages (e.g.
Japanese, Chinese, Korean, Thai, Vietnamese, Tamil), which leads him to posit
the generalization in (120).
(119) *John thinks that himself will leave.
(120)

Only languages without articles may allow subject reflexives, i.e. examples like (119).

The gist of Despics account of (119) is that TP closes the binding domain for
anaphors (more precisely, TP dominated by CP closes it; Despic states his analysis in terms phases under Chomskys (2008) approach to phases, where a CP/TP
pair works as a phase51 ). As a result, examples like (119) can only be possible in
languages that lack TP.52 If correct, Despics generalization/analysis provides a
rather strong argument for the no-TP analysis of article-less languages.53
6.

Conclusion

Based on a number of crosslinguistic generalizations, including new generalizations regarding radical pro-drop, number morphology, negative constituents,
negative concord, second-position clitics, and focus-V adjacency, where the
presence/absence of articles in a language plays a crucial role, I have argued
that there is a fundamental difference between TNP in languages with artirelations under Chomskys (2001) definition of the PIC, where V can Agree with T
in its CP complement. Due to the lack of T(P), the higher V however cannot Agree
with the lower V in an article-less language without violating the PIC.
51. Working within this system, Kang (2012) argues that CP in Korean does not work as
a phase (since it does not dominate TP, Korean lacking TP). Her argument is based
on her claim that Korean does not have successive cyclic movement via SpecCP (she
observes that several standard diagnostics for such movement fail in Korean).
52. See Despic (2011) for details of the analysis and other factors that are involved.
53. Note that Migdalski (2010) offers a deduction of the second-position clitic generalization in (56) which crucially ties it to (118) (i.e. the current hypothesis that article-less
languages lack TP), whereby a second-position clitic system cannot occur in a TP
language (see also Condoravdi and Kiparsky 2002 for relevant discussion).

230

Zeljko
Boskovic

cles like English and article-less languages like SC that cannot be reduced to
phonology (overt vs phonologically null articles) since the generalizations in
question involve syntactic and semantic, not phonological phenomena. In particular, I have argued languages with articles and article-less languages differ
in that the latter do not have DP. Given that a number of these generalizations
involve surprising interactions between clause-level and TNP-level phenomena, pursuing the TNP/Clause parallelism hypothesis, I have then explored the
possibility that the structural difference between languages with and without
articles on the TNP-level has its counterpart on the clausal level; in other words,
I have explored the possibility that just like the structure of TNP is poorer in
NP languages than in DP languages, the structure of clauses is poorer in NP
languages than in DP languages. Taking the TNP/Clause parallelism hypothesis seriously in fact naturally leads to the conclusion that the lack of DP in
a language implies the lack of its clausal counterpart, which I assume is TP. I
have offered initial evidence that article-less languages indeed lack TP (but see
footnote 41). The evidence came from crosslinguistic generalizations involving
phenomena such as subject expletives, subject-object asymmetries regarding locality of movement, subject reflexives, and Sequence of Tense. The discussion
in this paper is in some respects reminiscent of the traditional configurational
vs non-configurational languages distinction, given that many article-less languages belong to what used to be called non-configurational languages (see
here the scrambling correlation in (17), scrambling being one of the central
characteristics of non-configurational languages). Over the years we have seen
conclusive evidence that non-configurational languages do have structural hierarchies. While nothing in this paper challenges that conclusion, the discussion
here leads to the conclusion that traditional non-configurational languages (more
precisely, article-less languages) may have a bit flatter (i.e. a bit less) structure
than configurational languages (more precisely, article languages) on both the
clausal and the TNP level.54
Finally, I will briefly compare the position taken in this paper with the possibility explored in Fukui (1986, 1988), Fukui and Takano (1998), and Fukui
and Sakai (2003) that Japanese lacks syntactically active functional structure.
Putting aside for the moment the two obvious differences in the scope of these
works and the current work (the current work deals with all article-less lan54. I emphasize here that although I am arguing that article-less languages are structurally
poorer than article languages on both the clausal and the TNP level I am not arguing
that article-less languages completely lack functional structure in clauses and TNPs, a
point which has sometimes been misinterpreted in the literature addressing Boskovic
(2005, 2008a).

On NPs and Clauses

231

guages and does not deny the presence of all functional structure), the issue
I would like to focus on is the distinction between lacking certain projections
vs having these projections as syntactically inactive. As noted in Fukui (1988),
it is actually very hard to tease apart these two options (the works cited above
mostly leave the issue unresolved); in fact many of the generalizations discussed
in this work would be compatible with a weaker position that article-less languages have DP and TP, which are however syntactically inactive. This would
mean that these projections could not be involved in any syntactic phenomena
(such as movement or agreement) in article-less languages; they would merely
serve as place holders for certain lexical elements. Such an analysis would still
allow for the placement of verbal morphology, such as e.g. Japanese -ru, under
T (though, as discussed in Section 5 and Fukui (1988), this is not really necessary). However, Fukuis (1988) point is that as far as syntax is concerned, this
analysis does not seem to differ from the analysis where the relevant projections
would be completely lacking. There are, however, still some differences between
the two analyses. Thus, it appears that to account for Despics (2009, 2011, in
press) binding facts, it is not enough to posit a DP which cannot be targeted by
movement or be involved in an Agree relation in article-less languages; even
without these properties such a projection should still close off the c-command
domain of possessors, thus failing to account for the binding properties of SC,
Japanese, and Chinese possessors discussed in Section 3. I also refer the reader
to Boskovic (2008a) for deductions of the generalizations from Sections 12
(see also the discussion in Section 4), some of which appear not to be consistent
with postulation of syntactically inactive DP in article-less languages. More
generally, if the inactivity is confined to syntax, which means that DP (and TP)
would be active in phonology and semantics, the semantic generalizations from
Sections 1, 2 and 5 will raise questions, given that DP (and TP) would then be
present in the semantics of article-less languages. Let me finally note that the
reason for deactivation (or lack of) DP and TP in Japanese from Fukui (1986,
1988), Fukui and Takano (1998), and Fukui and Sakai (2003), who relate it to
the lack of agreement in Japanese, cannot be extended to many of the languages
discussed here, since many article-less languages clearly have agreement (this,
e.g., holds for all Slavic article-less languages). Nevertheless, in spite of the
differences noted above, in many respects the current work follows the line of
research that originated with Fukui (1986).

232

Zeljko
Boskovic

References
Abels, Klaus
2003

Successive cyclicity, anti-locality, and adposition stranding. Doctoral


dissertation, University of Connecticut, Storrs.
Alexiadou, Artemis
2003
On nominative case features and split agreement. In: Ellen Brandner
and Heike Zinsmeister (eds.), New Perspectives on Case Theory, 23
52. Stanford: CSLI Publications.
Androutsopoulou, Antonia
1998
Split DPs, Focus, and Scrambling in Modern Greek. In: Emily Curtis, James Lyle and Gabriel Webster (eds.), Proceedings of the 16 th
West Coast Conference on Formal Linguistics, 116. Stanford: CSLI
Publications.
Baker, Mark
1996
The Polysynthesis Parameter. New York: Oxford University Press.
Baker, Mark
2003
Lexical categories: Verbs, nouns, and adjectives. Cambridge: Cambridge University Press.
Basilico, David
1996
Head position and internally headed relative clauses. Language 72:
498532.
Basic, Monika
2004
Nominal subextractions and the structure of NPs in Serbian and English. MA Thesis, University of Troms.
Bobaljik, Jonathan D. and Diane Jonas
1996
Subject positions and the roles of TP. Linguistic Inquiry 27: 195236.
Bobaljik, Jonathan D. and Hoskuldur Thrainsson
1998
Two heads arent always better than one. Syntax 1: 3771.
Boeckx, Cedric
2003a
Islands and chains: Resumption as stranding. Amsterdam: John Benjamins.
Boeckx, Cedric
2003b
Intricacies of Icelandic agreement. Ms., University of Maryland, College Park.
Boeckx, Cedric
2005
Some notes on bounding. Ms., Harvard University, Cambridge, Mass.
Boeckx, Cedric and Sandra Stjepanovic
2001
Heading toward PF. Linguistic Inquiry 32: 345355.
Bonneau, Jose
1992
The structure of internally headed relative clauses. Doctoral dissertation, McGill University, Montreal.

On NPs and Clauses

233

Boskovic, Zeljko
1994
D-structure, -criterion, and movement into -positions. Linguistic
Analysis 24: 247286.

Boskovic, Zeljko
1997
The syntax of nonfinite complementation: An economy approach.
Cambridge, Mass.: MIT Press.

Boskovic, Zeljko
2001
On the nature of the syntax-phonology interface: Cliticization and
related phenomena. Amsterdam: Elsevier Science.

Boskovic, Zeljko
2004a
Left branch extraction, structure of NP, and scrambling. In: Joachim
Sabel and Mamoru Saito (eds.), The free word order phenomenon: Its
syntactic sources and diversity, 1373. Berlin/New York: Mouton de
Gruyter.

Boskovic, Zeljko
2004b
Be careful where you float your quantifiers. Natural Language and
Linguistic Theory 22: 681742.

Boskovic, Zeljko
2005
On the locality of left branch extraction and the structure of NP. Studia
Linguistica 59: 145.

Boskovic, Zeljko
2007a
A note on wh-typology. In: Proceedings of FDSL 6: 159170.

Boskovic, Zeljko
2007b
On the clausal and NP structure of Serbo-Croatian. In: Proceedings
of FASL 15: 4275.

Boskovic, Zeljko
2007c
On the locality and motivation of Move and Agree: An even more
minimal theory. Linguistic Inquiry 38: 589644.

Boskovic, Zeljko
2008a
What will you have, DP or NP? In: Proceedings of NELS 37: 101114.

Boskovic, Zeljko
2008b
On the operator freezing effect. Natural Language and Linguistic Theory 26: 249287.

Boskovic, Zeljko
2009a
The NP/DP analysis and Slovenian. In: Proceedings of the Novi Sad
Generative Syntax Workshop 1, 5373. Novi Sad: Filozofski fakultet
u Novom Sadu.

Boskovic, Zeljko
2009b
Unifying first and last conjunct agreement. Natural Language and
Linguistic Theory 27: 455496.

Boskovic, Zeljko
2009c
Licensing negative constituents and negative concord. In: Proceeding
of NELS 38: 125138.

234

Zeljko
Boskovic

Boskovic, Zeljko
2009d
On relativization strategies and resumptive pronouns. In: Proceedings
of FDSL 7: 7992.

Boskovic, Zeljko
2010
Phases beyond clauses. Ms., University of Conecticut, Storrs.

Boskovic, Zeljko
2012
Now Im a phase, now Im not a phase: On the variability of phases
with extraction and ellipsis. Ms., University of Connecticut, Storrs.

Boskovic, Zeljko
in press
On unvalued uninterpretable features. In: Proceedings of NELS 39.

Boskovic, Zeljko
and Jon Gajewski
in press
Semantic correlates of the DP/NP parameter. In: Proceedings of
NELS 39.

Boskovic, Zeljko
and I-Ta Chris Hsieh
2012
On word order, binding relations, and plurality within Chinese NPs.
Ms., University of Connecticut, Storrs.
Bowers, John
1987
Extended X-bar theory, the ECP, and the left branch condition. In:
Megan Crowhurst (ed.), Proceedings of the 6 th West Coast Conference
on Formal Linguistics, 4762. Stanford: CSLI Publications.
Cheng, H.-T. Johnny
in preparation Null arguments in Mandarin Chinese and the DP/NP parameter. Doctoral dissertation, University of Connecticut, Storrs.
Cheng, Lisa L.-S. and Rint Sybesma
1999
Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry
30: 509542.
Chierchia, Gennaro
1998
Reference to kinds across languages. Natural Language Semantics 6:
339405.
Chomsky, Noam
1986a
Barriers. Cambridge, Mass.: MIT Press.
Chomsky, Noam
1986b
Knowledge of Language: Its Nature, Origins, and Use. New York:
Praeger Publishers.
Chomsky, Noam
1995
The minimalist program. Cambridge, Mass.: MIT Press.
Chomsky, Noam
2000
Minimalist inquiries. In: Roger Martin, David Michaels and Juan
Uriagereka (eds.), Step by step: Essays on minimalist syntax in honor
of Howard Lasnik, 89155. Cambridge, Mass.: MIT Press.
Chomsky, Noam
2001
Derivation by phase. In: Michael Kenstowicz (ed.), Ken Hale: A life
in language, 152. Cambridge, Mass.: MIT Press.

On NPs and Clauses

235

Chomsky, Noam
2008
On phase. In: Robert Freidin, Carlos Peregrn Otero and Maria Luisa
Zubizarreta (eds.), Foundational Issues in Linguistic Theory. Essays
in Honor of Jean-Roger Vergnaud, 133166. Cambridge, Mass.: MIT
Press.
Condoravdi, Cleo and Paul Kiparsky
2002
Clitics and clause structure. Journal of Greek Linguistics 2: 139.
Corver, Norbert
1992
On deriving certain left branch extraction asymmetries: A case study
in parametric syntax. In: Proceedings of NELS 22: 6784.
Dal Pozzo, Lena
2007
The Finnish noun phrase. Doctoral thesis, Universit`a Ca Foscari di
Venezia.
Despic, Miloje
2009
On the structure of Serbo-Croatian NP Evidence from binding. In:
Proceedings of FASL 17: 1732.
Despic, Miloje
2011
Syntax in the Absence of Determiner Phrase. Doctoral dissertation,
University of Connecticut, Storrs.
Despic, Miloje
in press
On two types of pronouns and so-called movement to D in SerboCroatian. In: Proceedings of NELS 39.
Dikken, Marcel den
1999
On the structural representation of possession and agreement: The
case of (anti-)agreement in Hungarian possessed nominal phrases. In:
Istvan Kenesei (ed.), Crossing boundaries: Advances in the theory of
central and eastern European languages, 137178. Amsterdam: John
Benjamins.
Enc, Murvet
1991
On the absence of the present tense morpheme in English. Ms., University of Wisconsin, Madison.
Fortmann, Christian
1996
Konstituentenbewegung in der DP-Struktur: Zur funktionalen Analyse
der Nominalphrase im Deutschen. Tubingen: Max Niemeyer Verlag.
Franks, Steven
1995
Parameters of Slavic morphosyntax. New York: Oxford University
Press.
Franks, Steven
2007
Deriving discontinuity. In: Proceedings of FDSL 6.5: 102120.
Fukui, Naoki
1986
A theory of category projection and its applications. Doctoral Dissertation, MIT, Cambridge, Mass.

236

Zeljko
Boskovic

Fukui, Naoki
1988

Deriving the differences between English and Japanese. English Linguistics 5: 249270.
Fukui, Naoki and Margaret Speas
1986
Specifiers and projection. In: Naoki Fukui, Tova Rapoport and Elizabeth Sagey (eds.), MIT Working Papers in Linguistics 8, 128172.
Department of Linguistics and Philosophy, MIT, Cambridge, Mass.
Fukui, Naoki and Yuji Takano
1998
Symmetry in syntax: Merge and demerge. Journal of East Asian Linguistics 7: 2786.
Fukui, Naoki and Hiromu Sakai
2003
The visibility guideline for functional categories: Verb-raising in
Japanese and related issues. Lingua 113: 321375.
Gajewski, Jon
2005
Neg-raising: Polarity and presupposition. Doctoral dissertation, MIT,
Cambridge, Mass.
Gajewski, Jon
2007
Neg-raising and polarity. Linguistics and Philosophy 30: 289328.

Gallego, Angel
J. and Juan Uriagereka
2007
Sub-extraction from subjects: A phase theory account. In: Jose Camacho, Nydia Flores-Ferran, Liliana Sanchez, Viviane Deprez and Mara
Jose Cabrera (eds.), Romance Linguistics 2006, 149162.Amsterdam:
John Benjamins.
George, Leland M. and Jaklin Kornfilt
1981
Finiteness and boundedness in Turkish. In: Frank Heny (ed.), Binding
and Filtering, 105127. Cambridge, Mass.: MIT Press.
Gill, David
1987
Definiteness, noun phrase configurationality, and the count-mass distinction. In: Eric J. Reuland and Alice G. B. ter Meulen (eds.), The
representation of (in)definiteness. Cambridge, Mass.: MIT Press.
Giorgi, Alessandra and Fabio Pianesi
1997
Tense and Aspect: From Semantics to Morphosyntax. Oxford: Oxford
University Press.
Grewendorf, Gunther
2005
The discourse configurationality of scrambling. In: Joachim Sabel and
Mamoru Saito (eds.), The free word order phenomenon: Its syntactic
sources and diversity, 75135. Berlin: Mouton de Gruyter.
Grohmann, Kleanthes
2003
Prolific domains: On the anti-locality of movement dependencies. Amsterdam: John Benjamins.
Grosu, Alexander and Fred Landman
1998
Strange relatives of the third kind. Natural Language Semantics 6:
125170.

On NPs and Clauses


Hackl, Martin
2007
Harley, Heidi
1995

237

On the grammar and processing of proportional quantifiers: most versus more than half. Ms., Pomona College.
Subjects, events and licensing. Doctoral Dissertation, MIT, Cambridge, Mass.

Heim, Irene
1985
Notes on comparatives and related matters. Ms., University of Texas.
Heim, Irene
1999
Notes on superlatives. Ms., MIT, Cambridge, Mass.
Herdan, Simona
2008
Degrees and amounts in relative clauses. Doctoral dissertation, University of Connecticut, Storrs.
Higginbotham, James
1985
On semantics. Linguistic Inquiry 16: 547593.
Holmberg, Anders and Urpo Nikanne
2002
Expletives, subjects, and topics in Finnish. In: Peter Svenonius (ed.),
Subjects, expletives, and the EPP, 71106. Oxford: Oxford University
Press.
Horn, Laurence
1989
A natural history of negation. Chicago: University of Chicago Press.
Hornstein, Norbert
1999
Movement and control. Linguistic Inquiry 30: 6996.
Hu, Jianhua, Haihua Pan and Liejiong Xu
2001
Is there a finite-nonfinite distinction in Chinese? Linguistics 39: 1117
1148.
Ishii, Toru
1999
Cyclic spell-out and the that-trace effect. In: Sonya Bird, Andrew
Carnie, Jason D. Haugen and Peter Norquest (eds.), Proceedings of
the 18th West Coast Conference on Formal Linguistics, 220231.
Somerville, Mass.: Cascadilla Press.
Jeong, Youngmi
2006
The landscape of applicatives. Doctoral dissertation, University of
Maryland, College Park.
Jonas, Diane
1996
Clause structure and verb syntax in Scandinavian and English. Doctoral Dissertation, Harvard University, Cambridge, Mass.
Kallulli, Dalina
1999
The comparative syntax of Albanian. Doctoral dissertation, University
of Durham.
Kang, Jungmin
2012
To have TP or Not: Evidence from Successive-cyclic movement via
Spec CP. Ms., University of Connecticut, Storrs.

238

Zeljko
Boskovic

Kaplan, David
1977/1989

Demonstratives. In: Joseph Almog, John Perry, and Howard Wettstein


(eds.), Themes from Kaplan, 481563. Oxford: Oxford University
Press.

Kayne, Richard
1994
The antisymmetry of syntax. Cambridge, Mass.: MIT Press.
Khomitsevich, Olga
2007
Dependencies across Phases: From Sequence of Tense to Restrictions
on Movement. Utrecht: LOT publications.
Koizumi, Masatoshi
1994
Phrase Structure in Minimalist Syntax. Doctoral Dissertation, MIT,
Cambridge, Mass.
Kornfilt, Jaklin
1984
Case marking, agreement, and empty categories in Turkish. Doctoral
dissertation, Harvard University, Cambridge, Mass.
Kornfilt, Jaklin
2005
Agreement: The (unique and local) syntactic and morphological licenser of subject Case. In: Joao Costa and M. C. F. Silva (eds.), Studies
on agreement, 141172. Amsterdam: John Benjamins.
Kornfilt, Jaklin
2006
Agr in Turkish as an expression of categorial features. In: Meltem
urk (eds.), Proceedings of the workshop on
Kelepir and Balkz Ozt
Altaic formal linguistics (WAFL2), 2146. MITWPL, Department of
Linguistics and Philosophy, MIT, Cambridge, Mass.
Kuno, Susumu
1973
Constraints on internal clauses and sentential subjects. Linguistic Inquiry 4: 363385.
Kusumoto, Kiyomi
2005
On the Quantification over Times in Natural Language. Natural Language Semantics 13: 317357.
Lambova, Mariana
2004
On information structure and clausal architecture: Evidence from Bulgarian. Doctoral dissertation, University of Connecticut, Storrs.
Larson, Richard and Sungeun Cho
1999
Temporal adjectives and the structure of possessive DPs. In: Sonya
Bird, Andrew Carnie, Jason D. Haugen and Peter Norquest (eds.),
Proceedings of the 18th West Coast Conference on Formal Linguistics,
299311. Somerville, Mass.: Cascadilla Press.
Laury, Ritva
1997
Demonstratives in interaction. The emergence of a definite article in
Finnish. Amsterdam: Benjamins.

On NPs and Clauses

239

Lekakou, Marika and Krista Szendr


oi
2008
Polydefinites in Greek: A close appositive analysis. Ms., Meertens
Institute and UCL.
Lin, Jo-Wang
2003
Temporal reference in Mandarin Chinese. Journal of East Asian Linguistics 12: 254311.
Lin, Jo-Wang
2005
Time in a language without tense: The case of Chinese. Journal of
Semantics 23: 153.
Lin, Jo-wang
2010
A tenseless analysis of Mandarin Chinese revisited: A response to
Sybesma 2007. Linguistic Inquiry 41: 305329.
Lindseth, Martina
1993
The syntactic status of overt non-referential pronouns in Upper Sorbian. Ms., Indiana University.
Lobeck, Anne
1990
Functional heads as proper governors. In: Proceedings of NELS 20:
348362.
Longobardi, Giuseppe
1994
Reference and proper names: A theory of N-movement in syntax and
Logical Form. Linguistic Inquiry 25: 609665.
Lyons, Christopher
1999
Definiteness. Cambridge: Cambridge University Press.
Marelj, Marijana
2008
Probing the relation between binding and movement: Left branch extraction and pronoun-insertion strategy. In: Proceedings of NELS 37:
7386.
Mathieu, Eric and I. Sitaridou
2002
Split wh-constructions in Classical and Modern Greek. Linguistics in
Potsdam 19: 143182.
McCloskey, James
1992
Adjunction, selection, and embedded verb second. Ms., University of
California, Santa Cruz.
McCloskey, James
1997
Subjecthood and subject positions. In: Liliane Haegeman (ed.), Elements of grammar, 197235. Dordrecht: Kluwer.
McGinnis, Martha
1998
Locality and inert case. In: Proceedings of NELS 28: 267281.
Merchant, Jason
2008
Variable island repair under ellipsis. In: Kyle Johnson (ed.), Topics in
ellipsis, 132153. Cambridge: Cambridge University Press.

240

Zeljko
Boskovic

Migdalski, Krzysztof
2010
On the loss of Tense and verb-adjacent clitics in Slavic. Presented at
DiGS 12, Cambridge University.
Morzycki, Marcin
2008
Nonrestrictive modifiers in nonparenthetical positions. In: Louise McNally and Christopher Kennedy (eds.), Adverbs and adjectives: Syntax, semantics and discourse, 101122. Oxford: Oxford University
Press.
oi
Neeleman, Ad and Krista Szendr
2007
Radical Pro Drop and the Morphology of Pronouns. Linguistic Inquiry
38: 671714.
Nemoto, Naoko
1991
Scrambling and conditions on A-Movement. In: Dawn Bates (ed.), The
Proceedings of the 10th West Coast Conference on Formal Linguistics,
349358. Stanford: CSLI Publications.
Osawa, Fuyo
1999
The relation between Tense and Aspect: The emergence of the T system. UCL Working Papers in Linguistics 11: 521544.
Partee, Barbara
2006
A note on Mandarin possessives, demonstratives, and definiteness. In:
Betty J. Birner and Gregory Ward (eds.), Drawing the Boundaries of
Meaning: Neo-Gricean Studies in Pragmatics and Semantics in Honor
of Laurence R. Horn, 263280. Amsterdam: John Benjamins.
Partee, Barbara and Vladimir Borschev
1998
Integrating lexical and formal semantics: Genitives, relational nouns,
and type-shifting. In: Robin Cooper and Thomas Gamkrelidze (eds.),
Proceedings of the SecondTbilisi Symposium on Language, Logic, and
Computation, 229241. Tbilisi: Center on Language, Logic, Speech,
Tbilisi State University.

Paunovic, Zeljka
2001
Aspectual-temporal relations in Serbo-Croatian verbal morphology.
Essex Graduate Student Papers in Language and Linguistics 3: 171
196.
Pereltsvaig, Asya
2007
On the universality of DP: A view from Russian. Studia Linguistica
61: 5994.
Pesetsky, David
1989
The Earliness Principle. Ms., MIT, Cambridge, Mass.
Pesetsky, David and Esther Torrego
2007
The syntax of valuation and the interpretability of features. In: Simin
Karimi, Vida Samiian and Wendy Wilkins (eds.), Phrasal and Clausal
Architecture: Syntactic derivation and interpretation: In honor of
Joseph E. Emonds, 262294. Amsterdam: John Benjamins.

On NPs and Clauses

241

Pollock, Jean-Yves
1989
Verb movement, Universal Grammar, and the structure of IP. Linguistic
Inquiry 20: 365424.
Progovac, Ljiljana
2005
A syntax of Serbian: Clausal Architecture. Bloomington: Slavica Publishers.
Rappaport, Gilbert
2000
Extraction from Nominal Phrases in Polish and the theory of determiners. Journal of Slavic Linguistics 8: 159198.
Runic, Jelena
2011
Clitic Doubling in Non-Standard Serbian and Slovenian Dialects. Ms.,
University of Connecticut, Storrs.
Saito, Mamoru
1985
Some asymmetries in Japanese and their theoretical implications.
Doctoral dissertation, MIT, Cambridge, Mass.
Saito, Mamoru
1992
Long distance scrambling in Japanese. Journal of East Asian Linguistics 1: 69118.
Saito, Mamoru
2007
Notes on East Asian argument ellipsis. Language Research 43: 203
227.
Saito, Mamoru and Hiroto Hoshi
2000
The Japanese light verb construction and the minimalist program. In:
Roger Martin, David Michaels, and Juan Uriagereka (eds.), Step by
step: Essays on minimalist syntax in honor of Howard Lasnik, 261
295. Cambridge, Mass.: MIT Press.
Saito, Mamoru, Jonah T.-H. Lin and Keiko Murasugi
2008
N-ellipsis and the structure of Noun Phrase in Chinese and Japanese.
Journal of East Asian Linguistics 17: 247271.
Saito, Mamoru and Keiko Murasugi
1990
N-deletion in Japanese. In: Javier Ormazabal and Carol Tenny (eds.),
University of Connecticut Working Papers in Linguistics 3: 87107.
Department of Linguistics, University of Connecticut, Storrs.
Saito, Mamoru and Keiko Murasugi
1999
Subject predication within IP and DP. In: Kyle Johnson and Ian
Roberts (eds.), Beyond principles and parameters, 167188. Dordrecht: Kluwer.
Sauerland, Uli
2004
A comprehensive semantics for agreement. Ms., UniversitatTubingen,
Tubingen.
Scatton, Ernest
1984
A Reference grammar of modern Bulgarian. Cambridge, Mass.: Slavica Publishers.

242

Zeljko
Boskovic

Sener, Serkan
2006

Sener, Serkan
2010
Sharvit, Yael
2003

Multiple wh-questions in Turkish: An investigation of in-situness. In:


Simona Herdan and Miguel Rodrguez-Mondonedo (eds.), University
of Connecticut Working Papers in Linguistics 14, 131170. Department of Linguistics, University of Connecticut, Storrs.
(Non-)peripheral matters in Turkish syntax. Doctoral dissertation,
University of Connecticut, Storrs.

Embedded tense and universal grammar. Linguistic Inquiry 34: 669


681.
Shon, Young-Sook, Young-Yeah Hong and Kyung-Sun Hong
1996
The functional category IP in Korean reconsidered. Studies in Generative Grammar 6: 351384.

Sigursson, Halldor A.
1996
Icelandic finite verb agreement. Working Papers in Scandinavian Syntax 57: 146.

Sigursson, Halldor A.
2000
The locus of Case and agreement. Working Papers in Scandinavian
Syntax 65: 65108.
Slobin, Dan I. and Ayhan Aksu Koc
1982
Tense, aspect and modality in the use of the Turkish evidential. In: Paul
J. Hopper (ed.), Tense and aspect: Between semantics and pragmatics,
185200. Amsterdam/Philadelphia: John Benjamins.
Smith, Carlota S. and Mary S. Erbaugh
2005
Temporal interpretation in Mandarin Chinese. Linguistics 43: 713
756.
Stepanov, Arthur
2001a
Cyclic domains in syntactic theory. Doctoral dissertation, University
of Connecticut, Storrs.
Stepanov, Arthur
2001b
The End of CED? In: Karine Megerdoomian and Leora Anne Barel (eds.), Proceedings of the 20th West Coast Conference on Formal
Linguistics, 524537. Somerville, Mass.: Cascadilla Press.
Stjepanovic, Sandra
1998
Extraction of adjuncts out of NPs. Presented at the Comparative Slavic
Morphosyntax Workshop.
Stjepanovic, Sandra
1999
What do second position cliticization, scrambling, and multiple whfronting have in common. Doctoral dissertation, University of Connecticut, Storrs.

On NPs and Clauses


Stowell, Tim
1989

Stowell, Tim
1993
Stowell, Tim
1995a

Stowell, Tim
1995b

243

Subjects, specifiers, and X-bar theory. In: Mark R. Baltin and Anthony
S. Kroch (eds.), Alternative conceptions of phrase structure, 232262.
University of Chicago Press.
Syntax of tense. Ms., University of California, Los Angeles.
The Phrase Structure of Tense. In: Johan Rooryck and Laurie
Zaring (eds.), Phrase Structure and the Lexicon, 277291. Dordrecht:
Kluwer.
What is the meaning of the present and past tenses? In: Pier Marco
Bertinetto, Virginia Bianchi, James Higginbotham and Mario Squartini (eds.), Temporal Reference: Aspect and Actionality. Vol. 1: Semantic and Syntactic Perspectives, 381396. Torino: Rosenberg and
Sellier.

Szabolcsi, Anna
1986
Comparative superlatives. In: Naoki Fukui, Tova Rapoport and Elizabeth Sagey (eds.), MIT Working Papers in Linguistics 8, 245266.
Department of Linguistics and Philosophy, MIT, Cambridge, Mass.
Tada, Hiroaki
1993
A/A partition in derivation. Doctoral dissertation, MIT, Cambridge,
Mass.
Takahashi, Daiko
1994
Minimality of movement. Doctoral dissertation, University of Connecticut, Storrs.
Takahashi, Daiko
1996
On antecedent-contained deletion. In: Elizabeth Laurencot, Rhang Lee
and Myung-Kwan Park (eds.), University of Connecticut Working Papers in Linguistics 5, 6580. Department of Linguistics, University
of Connecticut, Storrs.
Takahashi, Daiko
1997
Move F and null operator movement. The Linguistic Review 14: 181
196.
Takahashi, Masahiko
2011
Some theoretical consequences of Case-marking in Japanese. Doctoral dissertation, University of Connecticut, Storrs.
Tanaka, Hidekazu
2002
Raising to object out of CP and phases in derivation. Ms., University
of Minnesota, Minneapolis.

244

Zeljko
Boskovic

Taylan, Eser Erguvanl


1988
On the Expression of Temporal Reference in Subordinate Clauses in
Turkish. In: Sabri Koc (ed.), Studies on Turkish Linguistics, 333351.

Ankara: ODTU.
Taylan, Eser Erguvanl
1996
On the Parameter of Aspect in Turkish. In: Ahmet Konrot (ed.), Proceedings of the 6th International Conference on Turkish Linguistics,
153168. Anadolu University, Eskisehir.
Taylan, Eser Erguvanl
1997
Turkcede Gorunus: Zaman ve Kiplik Iliskisi: {-DI} Bicimbirimi. [Aspect in Turkish: tense and modality relation: the DI morpheme.] In:
Deniz Zeyrek and Su kriye Ruhi (eds.), XI. Dilbilim Kurultay Bildiri Egitim Fakultesi, Yabanc Diller Egitimi
leri, 111. Ankara: ODTU
Bolumu.
Taylor, Ann
1990
Clitics and configurationality in Ancient Greek. Doctoral dissertation,
University of Pennsylvania.
Ticio, Emma
2003
On the structure of DPs. Doctoral dissertation, University of Connecticut, Storrs.
Tomioka, Satoshi
2003
The semantics of null arguments in Japanese and its cross-linguistic
implications. In: Kerstin Schwabe and Susanne Winkler (eds.), Interfaces, 321339. Amsterdam: John Benjamins.
Tsai, W.-T. Dylan
2008
Tense anchoring in Chinese. Lingua 118: 675686.
Uriagereka, Juan
1988
On government. Doctoral dissertation, University of Connecticut,
Storrs.
Vangsnes, jstein Alexander
1995
Referentiality and argument positions in Icelandic. Working Papers in
Scandinavian Syntax 5: 89109.
Watanabe, Akira
2003
Indeterminates and determiners. Ms., Presented at the Workshop on
Altaic in Formal Linguistics, MIT, May 2003.
Watanabe, Akira
2004
Parametrization of quantificational determiners and head-internal relatives. Language and Linguistics 5: 5997.
Whitman, John
1982
Configurationality parameters. Ms., Harvard University, Cambridge,
Mass.

On NPs and Clauses


Willim, Ewa
2000

245

On the grammar of Polish nominals. In: Roger Martin, David Michaels


and Juan Uriagereka (eds.), Step by step: Essays on minimalist syntax
in honor of Howard Lasnik, 319346. Cambridge, Mass.: MIT Press.

Wurmbrand, Susi
2006
Licensing Case. Journal of Germanic Linguistics 18: 175236.
Yavas, Feryal
1981
On the Meaning of the Tense and Aspect Markers in Turkish. Doctoral
Dissertation, University of Kansas.
Yavas, Feryal
1982a
Future reference in Turkish. Linguistics 20: 411429.
Yavas, Feryal
1982b
The Turkish Aorist. Glossa 16: 4053.
Zlatic, Larisa
1997
The structure of the Serbian Noun Phrase. Doctoral dissertation, University of Texas at Austin.

Zivanovi
c, Saso
2008
The varieties of most: On different readings of superlative determiners.
In: Proceedings of FDSL 6.5: 337354.

Discerning Default Datives:


Some Properties of the Dative Case in German*
Henk van Riemsdijk

1.

Some Remarks on Default Case

It would be hard to deny that default case exists, as a phenomenon. Exclamations


consisting of a noun or noun phrase, for example, tend to occur in the nominative
case, at least in those languages that have a system of overt case marking.
Take shit!, or its German counterpart Scheisse!. In all likelihood there is
no silent or deleted clause in which these nominal expressions are embedded.
These forms do not show any overt case marking, but this cannot be due to
the absence of a clausal environment, for as soon as the nominal expression is
somewhat more elaborate, the nominative case manifests itself, as the following
German examples show. In the absence of a clause containing a case assigner,
the nominative case shows up as the default case.
(1)

der Hammer!
the hammer
great!
ein Holler!
a
elderberry
thats crazy!

* This paper was presented at the workshop entitled Sentence Types: Ten Years After,
which took place at the Johann Wolfgang Goethe Universitat in Frankfurt/M on June
2628, 2009. I wish to thank Gunther Grewendorf and Ede Zimmermann and their
colleagues for organizing this wonderfully inspired and inspiring conference and
the audience for interesting discussion. Most of the materials discussed are based
on earlier work, in particular Van Riemsdijk (2007; to appear). I thank Josef Bayer,
Aniko Csirmaz, Joe Emonds, Riny Huijbregts, Viola Schmitt and, last but not least,
Memo Cinque, whose 2005 workshop on PP-structure in Venice rekindled my interest
in the topic discussed here.

248

Henk van Riemsdijk

so
ein verdammter Schlendrian!
such a
damned
rut
such a damned rut!
so
ein Wahnsinn!
such an insanitiy
such madness!
It is, perhaps, less obvious that default case manifests itself also with other cases.
The present paper discusses a number of interesting ways in which the dative
case shows default behavior, with brief remarks also on default accusatives and
genitives.
One reason why default case has not attracted the interest that it deserves
is probably that its existence creates a fairly serious problem for the case filter,
one of the cornerstones of case theory (cf. Vergnaud 2008) which in turn was
one of the cornerstones of Government and Binding Theory (Chomsky 1981).
Take the paradigm example that case theory was introduced for: the passive. The
idea was that passive morphology absorbs or suppresses the verbs power to
assign accusative case to its direct object. Therefore the object must move to the
subject to pick up nominative case, for if it remains without case it falls prey to
the case filter and the sentence is ruled out. This line of reasoning presupposes
that there are no other ways in which the object could receive case. But there are.
In fact, inside the DP the equivalent of passive can avail itself of of-insertion,
or it can be moved to the prenominal position, where it picks up genitive case:
(2)

a.
b.

the destruction of the city


the citys destruction

Why cant of-insertion be used to rescue the object of a passive verb, or, for that
matter a DP that has failed to undergo raising, from the case filter?
(3)

a. *It was destroyed of the city.


b. *It seemed of the city to be destroyed.

There are potential answers to these question. For example, it may very well be
correct to say that of-insertion and genitive assignment are processes that apply
automatically inside DPs only. In a sense, of and genitive case are two sides of
the same coin, and we may well say that the genitive is the default case in the
nominal domain. For an interesting elaboration of such an idea, see Kagan and
Pereltsvaig (2009).
There is, however, a second way in which a case-less direct object might be
rescued: it might get the default nominative, as discussed in (1) above. Indeed,

Discerning Default Datives: Some Properties of the Dative Case in German

249

the default nominative does show up in more complete clausal contexts, such
as hanging topic left dislocation (HTLD):1
(4)

Der HansNOM ich glaube nicht dass ich ihnACC mag.


the John
I
believe not that I
him
like
John, I dont think I like him.

In other words, what prevents the following example from being rescued by
default nominative realization?
(5)

*It was destroyed the cityNOM .

Undoubtedly, answers can be provided for this problem as well. In general, it


would seem that the domain in which a specific default case imposes itself must
be defined in precise ways.
The present paper presents two studies on the default dative case.2 In Section 2, I present arguments to show that the dative that shows up in locational
adpositional phrases is not an assigned case but a default case. And in Section 3
I discuss ways in which the overt expression of the dative case is sometimes
obligatory but can, in other cases, be waived. The way I will refer to this set
of phenomena is to say that certain uninflected nominal forms are incompatible
with contexts in which dative case is assigned, but that there are other contexts
in which this dative incompatibility is suppressed.

1. Occasionally, case attraction does occur in HTLD constructions, hence the following
alternative to (4) is not unacceptable:
Den HansACC , ich glaube nicht, dass ich ihnACC mag.
In the corresponding contrastive left dislocation, accusative case is obligatory:
Den HansACC /*Der HansNOM denACC glaube ich nicht, dass ich mag.
For details, see Vat (1997).
2. The ideas and proposals presented here are intentionally presented in a rather theoryneutral way. For some proposals on the integration of case and case theory in a
minimalist framework, see Pesetsky and Torrego (2004, 2007), among others. There
are some ideas there that point in a similar direction as in my own proposals, while
in other respects the approaches appear to be quite divergent. Mostly, however, the
empirical domains covered as well as the type of questions asked are rather disjoint.
Hence I abstain from discussing their views in the present article.

250
2.

Henk van Riemsdijk

Pure Route Ps take the Accusative, the Dative is the Elsewhere Case3

2.1. The Questions and the Program


In case marking languages such as German, objects of adpositions tend to be
case-marked. One of the most important subregularities is generally formulated
as follows: spatial adpositions govern the dative when they are purely locative
but the accusative when they are directional. This dual case marking behavior
is illustrated in (6).
(6)

a.
b.

Peter legt
Peter puts
Das Buch
the book

Tisch.
das Buch auf den
the book on theACC table
liegt
auf dem
Tisch.
is-lying on theDAT table

Similar patterns are found in other Indo-European languages that have preserved
(part of) the case system. The following table, which I borrow from Zwarts
(2005), shows how the case system of Proto-Indo-European has syncretized to
the four-way case system of present day German.
(7)

Spatial
meaning

Proto-IE
Nominative
Vocative
Genitive
Dative
Instrumental

German
Nominative
Genitive

Dative
source
location
extent
goal

Ablative
Locative
Accusative

Accusative

The dominant view is that adpositions that allow both variants have the property
referred to as Doppelrektion (dual case government) (cf. Abraham 2003). The
alternative I wish to explore here (cf. Van Riemsdijk 2007) is that the choice of
the accusative is dependent on certain specific factors, while the dative case is
the elsewhere or default case.

3. This section presents an abbreviated version of the materials discussed in Van


Riemsdijk (2007).

Discerning Default Datives: Some Properties of the Dative Case in German

251

The main question that I will address in this section, then, is the following.
What factors determine the choice of the dative and the accusative case in
spatial PPs (in German)?
My argumentation will proceed in the following steps:
The notion of direction must be split up into two distinct subcomponents:
ROUTE and SOURCE/GOAL (Section 2.2).
ROUTE is a major contributor to the choice of the accusative case (cf. Zwarts
2005; 2006) (Section 2.3).
ROUTE can be identified as the main factor determining the accusative case
in GOAL-PPs (Section 2.4).
2.2. Decomposing DIR
Much work on the internal structure of spatial PPs has assumed that there are
separate positions for location and direction, cf. in particular Van Riemsdijk
(1990), Koopman (2000), Den Dikken (2010), Huijbregts and Van Riemsdijk
(2001; 2007) and many others. Others have proposed structures far richer than
these (cf. among others Noonan 2005, Svenonius 2010). In (8) I give the relatively simple structure argued for in Huijbregts and Van Riemsdijk (2001; 2007).
(8)

Pmax
P
N

PDIRo
PLOCo

No

This type of structure is insufficient to handle contrasts like the following.


(9)

a.

b.

Dach
Die Schnecke kroch auf das
the snail
crept on theACC roof
hinauf /hinab /hinuber.
up/down/across
The snail crept up/down/across onto the roof.
Dach hinauf /hinunter.
Die Schnecke kroch das
the snail
crept theACC roof up/down
The snail crept upward/downward along the roof.

252

Henk van Riemsdijk

The three variants of (9a) correspond to the three motions depicted in (10). In
each case (the top of/upper side of) the roof is the endpoint, the terminus of
the motion, while the three possible postpositional elements corresond to the
orientation of the path or route along which the snail moves. In these examples,
an endpoint of the motion is specified: some position on the roof.
(10)

hinab
hinber

hinauf

In (11), which corresponds to the example (9b), on the other hand, we have
an upward or downward motion without an explicitly indicated starting point
or endpoint. The roof is the ground in relation to which the upward motion is
defined, but there is no implication as to whether the motion takes place on the
top side or the bottom side of the roof.
(11)

The contrast between the two examples shows quite straightforwardly that the
prepositional element serves to pinpoint the goal of the motion while the postpositional element denotes the orientation or the path of the motion. In the
remainder of this article I will use the following terminology and abbreviations:
place/location:
motion/orientation/path/route:
source/goal:

LOCATION
ROUTE
SOURCE
GOAL

LO
RO
SO
GO

Discerning Default Datives: Some Properties of the Dative Case in German

253

And I will minimally modify (8) to accommodate RO as in (12).


(12)

Pmax
PSOGOo

P
PROo
P
PLOo
N

No

2.3. Joost Zwarts Generalization


Table 1 represents the list of adpositions that Zwarts gives in his (2006) paper.
Table 1.
DATIVE

ACCUSATIVE

DAT & ACC

aus out of
auer outside
bei near
entgegen against
gegenuber opposite
nach to
von from
zu at, to

durch through
entlang along
gegen against
um around

an on
auf on
hinter behind
in in
neben next to
u ber over
unter under
vor in front of
zwischen between

It should be noted that Zwarts ignores the postpositional elements in circumpositional PPs, despite the fact that these elements can occur independently, as
shown in examples like (9) above. Zwarts groups these adpositions according
to the parameters discussed above in the following way:

254

Henk van Riemsdijk

Table 2.
Locative prepositions

Source
Route

Directional
prepositions

Goal

DATIVE
an
auf
bei
gegenuber
hinter
in
mit
neben
u ber
unter
vor
zwischen
aus
von
(entlang

on (hanging)
on (standing)
near
opposite
behind
in
with
beside
over, above
under
in front of
between
out of
from
along)

entgegen
nach
zu

against
to
to

ACCUSATIVE

durch
entlang
u ber
um
an
auf
gegen
hinter
in
neben
u ber
unter
vor
zwischen

through
along
over
around
onto
onto
against
(to) behind
into
(to) beside
over
(to) under
(to) in front of
(to) between

And he draws the following conclusion (in Zwarts 2006), correctly in my view.
DATIVE case goes with locative or source adpositions
ACCUSATIVE case goes with route or goal adpositions
We see immediately that the dative-accusative divide does not correspond to the
locative-directional distinction, but that SOURCE-Ps pattern with LOCATIONPs while ROUTE-Ps pattern with GOAL-Ps. In the next subsection I will try
to argue that both RO-Ps and GO-Ps impose a kind of measure phrase (MP)
interpretation on the PP, which causes the accusative case to show up.4
4. I return to the unexpected datives with GO-Ps (entgegen, nach, zu) in subsection 2.4.3.

Discerning Default Datives: Some Properties of the Dative Case in German

255

2.4. A Proposal
The basic observation on which the proposal below rests is this:
(13)

Pure ROUTE PPs take the accusative

This is true for the adpositions durch (through), entlang (along), u ber
(over), um (um) as well as post-positional elements of the type found in (9).
The complete list of these is given in (14).

(14)
auf




hin
ein (< in)

her
u ber

unter

upwards

inwards
across

downwards

away from X
towards X

Note that the deictic prefix hin- /her- does not denote a source or a goal, but an
orientation. Orientations are properties of paths (RO).
Taking this observation as a point of departure, we may formulate the following hypotheses.
1. It is the ROUTE component that is responsible for the accusative in the
GOAL-PPs (as opposed to the SOURCE-PPs);
2. The object of a ROUTE-P functions like a Measure Phrase (MP);
3. GOAL-Ps imply an (implicit or explicit) ROUTE component, SOURCE-Ps
do not;
4. Any datives showing up are not governed cases but represent the default
case in oblique domains.
The rest of this section will be devoted to some arguments in favor of these
hypotheses.
2.4.1. ROUTE-DPS as Measure Phrases5
A major insight underlying my line of argumentation has its roots in some
diachronical observations and their interpretation by Joost Zwarts:
The behaviour of the dative, straddling the line between locative and directional
uses, can partially be understood historically. The dative and accusative in presentday German PPs evolved out of the richer case system of Proto-Indo-European
(PIE) (Beekes 1995; Blake 1994; Fox 1995). The present-day dative is a syn5. The material presented in this section profited greatly from discussion with and input
from Aniko Csirmaz and Viola Schmitt.

256

Henk van Riemsdijk

cretism of three distinct cases in PIE. It covers the PIE dative, which was a nonspatial case for recipients and benefactives and is still used as such in German and
many other Indo-European languages. However, it has absorbed the locative case
(that was used for location) and the ablative case (for sources). The accusative in
PIE was used for goals (like an allative), which is still reflected in its PP use, but
it was also used for extents, which is very similar to our route use here (what is
called perlative or translative in local case terminologies).
(quoted from Zwarts 2006 emphasis mine, HvR)

At this point, some clarification of terminology is in order. The extents that


Zwarts talks about in the above quote, and which he likens to the notion of
route, are often referred to as measure phrases. It is, indeed, easy to see that
(nominal) measure phrases are characterized by the accusative case in German
and other Indo-European languages. Extent expressions in turn play an important
role in the literature on event structure, where they are often referred to as
(situation) delimiters (cf. Tenny 1994). The link between (situation) delimiters
and accusative case is argued for extensively in Csirmaz (2006) on the basis of a
variety of Hungarian constructions. Tenny uses measure for scales, including
paths, hence measure phrases are not necessarily delimited. Therefore, in view
of the reasoning below, and as suggested to me by Aniko Csirmaz, I adopt the
term delimiters.
Accusative is indeed the case for (delimiting) MPs (henceforth DMPs) in
German:
(15)

a.

b.

ganze Nacht geschlafen.


Er hat die
he has theACC whole night slept
He slept all night.
zuruckgelegt.
Sie hat 2 km
she has 2 kmACC covered
She covered 2 km.

Let us now apply some criteria for DMP-status.

Discerning Default Datives: Some Properties of the Dative Case in German

257

Adjectival amount/degree modifiers


Modifiers like whole and half can be used to modify DPs that denote an extent,
but not DPs that denote a goal.
(16)

a.
b.

c.

d.

e.

Er hat die halbe /ganze Nacht geschlafen.


he has the half/whole night slept
Er ist den halben /ganzen Berg
hinauf gegangen.
he has the half/whole
mountain up
gone
He went up half the mountain/the whole mountain.
(extent maximalized, goal pragmatically implied)
Er ist den halben /ganzen Berg
hinauf gerannt.
he has the half/whole
mountain up
run
He covered half the distance up the mountain/the whole distance
running.
*Er ist auf den halben /ganzen Berg
hinauf
he has onto the half/whole
mountain up
gegangen.
gone
He has gone up onto half the mountain/the whole mountain.
(GO)
Er ist auf den Berg
halb /ganz hinauf gegangen.
he has onto the mountain wholly
up
gone
He has gone halfway/completely up onto the mountain.

As (16b) shows, the DP preceding a pure RO-P acts just like the MP in (16a) in
this respect. (16c) shows that a RO-PP that is used as an adjunct acts identically.
In (16d), however, we have a typical GO-PP whose DP cannot be modified by
extent-denoting modifiers. However, the postpositional route component of such
a circumpositional phrase can be modified by the corresponding extent-adverb,
as shown in (16e). The same behavior can be observed in temporal cases that
are expressed by the figurative use of spatial Ps:
(17)

Sie hat die halbe /ganze Nacht hindurch geschlafen.


she has the half/whole night through slept
She slept throughout half the night/the whole night.

258

Henk van Riemsdijk

Non-distributive universal quantification


Universal quantifiers that quantify over an extent can only be interpreted nondistributively.
(18)

Wir haben alle Tage (des


Urlaubs) geschlafen.
we have all days (of-the holiday) slept
We slept the whole holiday (all days of it).
*We slept each and every day of the holiday.

The same effect can be observed with spatial PPs:


(19)

a.

b.

Ich bin alle Stufen der


Treppe hinauf gegangen.
I
have all steps of-the stairs up
gone
I went up the whole stairs (all steps of it).
Ich bin auf alle Stufen der
Treppe hinauf
I
have onto all steps of-the stairs up
gegangen.
gone
(#) I went up each and every step of the stairs.

Similarly for adnominal spatial PPs:


c.

Alle Stufen der


Treppe hinauf gibt es
ein
all steps of-the stairs up
is
there a
Gelander.
banister
Up the whole stairs (all steps of it) there is a banister.
(#) Up each step of the stairs there is a banister.

Ambiguity with adverbs like again


Adverbs like again can lead to ambiguity in eventive contexts (cf. von Stechow
1996):
(20)

John closed the door again.


(i) the event of Johns closing the door has occurred previously
(ii) the state of the door being closed has existed previously

This ambiguity is also found in Goal-PPs (as opposed to Source PPs), as argued
in Nam (2004: (24)(26)).

Discerning Default Datives: Some Properties of the Dative Case in German

(21)

a.
b.

(22)

a.
b.

259

John drove to New York again.


ambiguous: event repetition or state restitution
John drove from New York again.
unambiguous: repetitive meaning only
John sent the book to New York again.
ambiguous
John sent the book from New York again.
unambiguous

The same effect is indeed found with other diagnostics for telicity of events:
(23)

a.
b.
c.

Mary ran (for ten minutes /*in ten minutes).


Mary ran to the store (*for ten minutes /in ten minutes).
He ran from the library (for ten minutes /*in ten minutes).

(23c) is an atelic activity like (23a): no time-frame and no delimited route; (23b)
however is a telic event (an accomplishment) that typically takes a delimiting DP.
For now we conclude that there is considerable evidence that the DP-object of
a pure RO-postposition is a DMP. The asymmetry between source-PPs and goalPPs that was also evidenced by the examples will be pursued in Subsection 2.4.2.
The conclusion we have reached is not without its problems, however. First,
RO-PPs can take an additional explicit DMP, as shown in (24).
(24)

a.

b.

300 m den Berg


hinauf macht einen total
300 m the mountain up
makes one totally
fertig.
finished
300 m up the mountain exhaust you completely.
Ganz den Berg
hinauf ist wohl
zu weit.
wholly the mountain up
is presumably too far
All the way up the mountain is presumably too far.

Upon consideration, however, these examples are unproblematic since DMPs


can co-occur quite generally, as shown in (25), and each one of the MPs takes
the accusative case.
(25)

Er hat die ganze Nacht nur zwei Stunden geschlafen.


he has the whole night only two hours
slept
He slept only two hours during the whole night.

260
(26)

Henk van Riemsdijk

Sie hat zwei Stunden (lang) 20 km /h aufrechterhalten


she has two hours
(long) 20 km/h keep-up
konnen.
could
She was able to keep up 20 km/h for two hours.

It is true that such examples are in conflict with the Unique Path Constraint
of Goldberg (1991) and/or the Single Delimiting Constraint of Tenny (1994),
where both impose the uniqueness of delimiters associated with eventualities.
The conclusion must be that the delimiters in (25) or (26) modify distinct entities (Csirmaz p.c.): eventuality time and topic time in (25) and distinct scales
(duration and speed) in (26).
Does this help for (24)? Perhaps. We would have to say that there are two
distinct DMPs, one associated with the extent of the path (sc. 300 m), and one
specifying the actual distance that suffices to exhaust the climber. These two
DMPs and their interpretations happen to be coextensive in this case.
A second potential problem seems to arise when we look at the Japanese
accusative-dative alternation (Fukuda to appear, Kuno 1973, Sugamoto 1982),
which may appear to counterexemplify the connection between goal and accusative. Take (27) from Fukudas paper.
(27)

Gakusei-ga yama-o /ni


nobor-ta.
studentNOM mountainACC/DAT climbPERF
Students climbed the mountain.

Apparently the goal (mountain) can be expressed either by the dative or the
accusative. But in actual fact, as Fukuda argues, there is an important difference
between the two variants having to do with the notion of path:
(28)

Gakusei-ga kaidan-o /#ni nobor-ta.


studentNOM stairsACC/#DAT climbPERF
Students climbed the stairs.

(29)

Gakusei-ga yane-#o /ni


nobor-ta.
studentNOM roof#ACC/DAT climbPERF
Students climbed on the roof.

As Fukuda puts it, stairs is naturally interpreted as a path while in (29) the roof
is the endpoint of the climbing, but the roof does not define the path. Hence this
may actually turn out to be supporting evidence. Consider again the examples
in (9), repeated here for convenience and with the addition of a third variant
with a preposition only.

Discerning Default Datives: Some Properties of the Dative Case in German

(30)

a.

b.

c.

261

Die Schnecke kroch auf das


Dach
the snail
crept on theACC roof
hinauf /hinab /hinuber.
up/down/across
The snail crept up/down/across onto the roof.
Die Schnecke kroch das
Dach hinauf /hinunter.
the snail
crept theACC roof up/down
The snail crept upward/downward along the roof.
Dach.
Die Schnecke kroch auf das
the snail
crept onto theACC roof

The b-example is a case where, unlike example (29), the roof is the path. Hence
we might expect a corresponding example in Japanese to be felicitous with both
case endings, the accusative corresponding to the b-example and the dative to the
interpretation in which the roof is the endpoint of the snails climbing activity:
the roof is reduced to a point without extension. This seems to be confirmed.
(31)

Katatsumuri-ga yane-o /ni


nobor-ta.
snailNOM
roofACC/DAT climbPERF
The snail crept upward on/onto the roof.

Similarly, (27) is actually ambiguous in the intended way. What remains is the
question as to why the goal-interpretation is associated with the dative case. I
will assume that in Japanese, other than in German, the DMP, which is implicitly
associated with a goal-PP, does not have the force to impose its accusative case
to the object of the goal-P. This actually appears to be a more general property
of Japanese: even though the language does have an overtly marked accusative
case, (adverbial) MPs are caseless.6
Observe, finally, that we have to say that ROUTE is a measure, an extent,
but an extent with an orientation. This is so because if it did not have an orientation, den Berg hinauf (up the mountain) and den Berg hinunter (down the
mountain) would have the same meaning, which they do not. This may well
be true for MPs in other contexts as well. Time is intrinsically oriented in cases
like he slept three hours. Also, presumably, he covered 2 km implies directed
locomotion.

6. Thanks to Yukinori Takubo (p.c.) and Masayuki Oishi (p.c.) for helpful discussion
of the subtle nuances in (27)(29) and (31). Thanks to Oishi-sensei as well for
confirming that Japanese MPs, provided they are adverbial, do not have any overt
case, unless they are used contrastively.

262

Henk van Riemsdijk

The apparent problems for our conclusion, that the object of a pure route-P is
a delimiting DP and hence expected to show up in the accusative case, may thus
be solvable and, what is more, may even turn out to support the hypothesis.7
2.4.2. GOAL vs. SOURCE
In this section I will suggest that the difference between SO-PPs and GO-PPs
can be traced back to the role of the RO-component: I propose that in GO-PPs
there is an implied RO-component which is lacking in SO-PPs. My ideas on
this issue are admittedly quite speculative and the evidence is rather suggestive.
Still, I feel that this is a line of reasoning that is worth pursuing.
The main idea is this. If you move towards an endpoint, a GOAL, it makes
sense to specify the distance in space or time. If you move away from some
SOURCE, this is much less obvious. In other words, we always focus on the
way ahead, not on the path already covered. The distance ahead when we move
away from a source is always indeterminate, unless a GOAL is specified in
addition.
Consider the following examples.
(32)

a. ?I walked 500 m out of the parking lot.


b. ?I walked the whole way out of the parking lot.

Confronted with (32), the question arises what these sentences mean exactly. The
picture below suggests three possible interpretations. Consultation with about
a dozen speakers of English and German has revealed a considerable diversity
of judgments as to which interpretation is the one that imposes itself.

7. Note also that our conclusion sheds new light on the old issue of whether English
ago is a true postposition taking a DP-complement, or whether it is an intransitive
preposition taking an obligatory MP.
(i) a. two nights ago
b. *all nights ago
c. the whole night ago
Similarly with German her (ago):
(ii) a. zwei Nachte her
(two nights ago)
b. *alle Nachte her
(all nights ago)
c. *Nachte her
(nights ago)
d. die ganze Nacht her
(the whole night ago)
What my line of reasoning suggests is that this is a non-issue to the extent that
delimiting DPs associated with adpositions are both objects of that adposition and
measure phrase like modifiers.

Discerning Default Datives: Some Properties of the Dative Case in German

263

(33)

Similarly, (32) seems to be rather indeterminate as to what exactly is implied.


Interestingly, the most prominent meaning seems to be the one in which my
walking reaches the confines of the parking lot, turning the SO-PP into an
implied GO-PP. Indeed, with GO-PPs no such difficulties seem to arise, as shown
by the examples in (34) that are each directly representable by the corresponding
pictures in (35).
(34)

a.
b.
c.

(35)

a.

We walked 500 m up the slope.


We walked 500 m towards the house.
We walked 500 m into the parking lot.

500m

b.

c.

Furthermore, while it is not entirely impossible, it is often very difficult to express


the orientation of the ROUTE explicitly with SO-PPs. Consider the following
examples.

264
(36)

Henk van Riemsdijk

Aus dem Haus heraus /??hinauf leckten Flammen.


out the house out/up
flared flames
Out of/up out of the house flared flames.

When the semantically empty copy heraus is used, there is no indication as to


the orientation of the ROUTE. When the orientation is made explicit, as with
hinauf (up), the sentence is clearly quite degraded. Similarly for the following
example.
(37)

Aus dem Panzerfahrzeug heraus /??herauf /??herunter kamen


out the armored-vehicle out/up/down
came
brennende Soldaten.
burning
soldiers
Out of/up out of/down out of the armored vehicle came burning soldiers.

I conclude, tentatively, that there is reason to believe that implied MPs (implied
ROs) can be associated with GO-PPs but not, or only with difficulty, with SOPPs. And I conjecture that it is the implied RO-component that is responsible
for the accusative that is assigned in GO-PPs.
2.4.3.What about the DATIVE?
The remaining question at this point is: why do we get the dative case in locationPPs and in source-PPs? The answer I would like to suggest is that there is no
positive property that LO- and SO-PPs share that is responsible for the dative
case. Instead, building on earlier work, I propose that the dative case manifests
itself in these LO/SO-PPs simply as a consequence of the fact that the dative
is the elsewhere or default case in oblique domains such as PPs. This is what I
argued in van Riemsdijk (1983). In other words, when there is a positive reason
for the accusative to be assigned, such as the presence of a DMP-component,
that is what happens. When there is no such factor, the dative automatically
shows up.
In support of this claim, I will summarize some of the main arguments
presented in my (1983) article.

Discerning Default Datives: Some Properties of the Dative Case in German

265

The overwhelming majority of DP-complements to adjectives (also an


oblique domain) shows up in the dative. (38) presents some examples:
(38)

der seinem Vorgesetzten a hnliche /treue /unbekannte /


similar/loyal/unknown/
the hisDAT boss
verhasste /ergebene /gleichgultige Mann
hated/devoted/indifferent
man
the man similar/loyal/unknown/odious/devoted/indifferent to his
boss

A much smaller set of adjectives takes the genitive, modulo those that are Measure Phrase like, such as keinen Heller wert (worth not a penny) that, not
unexpectedly, take the accusative.
Possessive datives
When the adnominal genitive is absorbed by the possessive adjective, the
possessor shows up in the dative case (cf. van Riemsdijk 1983: (42), 244):
(39)

[dem Mann] [sein] Vater


his
father
theDAT man
the mans father
Mannes] Vater
b. [des
father
theGEN man
the mans father
c. *[des
Mannes] [sein] Vater
theGEN man
his
father
the mans father
d. *[dem Mann] Vater
father
theDAT man
the mans father

a.

Appositive DPs to obliquely case-marked DPs can show up in the dative


Appositives to obliquely case-marked DPs may either agree in case with the
nominal head of the complex DP or take the dative, deviating from the agreement
pattern that is obligatory in non-oblique contexts (cf. Leirbukt 1978; Winter
1966 examples cited from these sources in van Riemsdijk 1983: (42)(51),
245247).
First consider appositives to (oblique) genitives. (40) contains an adnominal
genitive DP whose nominal appositive appears in the Dative. (41) is an example
of a genitive assigend by the preposition wegen (because of ) whose apposition
again shows up in the dative. Finally, the verb sich annehmen (attend to) in (42)

266

Henk van Riemsdijk

assigns a genitive to its internal argument, and the appositive to that argument is
realized with the dative case. In all these cases the dative case on the appositive
DP is perfectly grammatical (though an agreeing genitive is also possible).
(40)

Sie war im Besitz


zweier Kleidungsstucke der
she was in possession twoGEN pieces-of-clothing theGEN
Ermordeten,
einem Persianermantel und einem roten
murdered-woman, aDAT fur coat
and aDAT red
Kimono . . .
kimono . . .
She owned two pieces of clothing of the murdered woman, a fur coat
and a red kimono.

(41)

Die Hauptgestalt, Amos Comenius, war schon dem


the main-character, Amos Comenius, had already to-the
Knaben Kokoschka . . . teuer gewesen wegen
seines
boy
Kokoschka . . . dear been
because-of hisGEN
Orbis Pictus, dem
alten Lehrbuch
in Bildern.
Orbis Pictus, theDAT old schoolbook in pictures
The main character, Amos Comenius, had already been dear to
Kokoschka when he was still a boy because of his Orbis Pictus the
old pictorial schoolbook.

(42)

Endlich hat sich ein kompetenter Mechaniker meines


finally has refl. a
competent
mechanic
myGEN
Wagens angenommen, einem hierzulande
seltenen
car
attended-to
a
in-this-country rare
russischen Modell.
Russian
model
Finally a competent mechanic has attended to my car, a Russian model
that is rare in this country.

Turning now to oblique accusatives, that is, accusatives assigned by a preposition, we see in (43) that here too the dative appositive is acceptable.

Discerning Default Datives: Some Properties of the Dative Case in German

(43)

267

Der Konig kam aber


ohne
Krone
und
the king came however without crownACC and
den
Zepter,
wichtigsten
Symbolen seiner Macht
scepterACC , theDAT most-important symbols of-his power
und Wurde.
and dignity
But the king arrived without crown and scepter, the most important
symbols of his power and dignity.

In non-oblique contexts, datives are always excluded, as is shown by the appositive to a direct object accusative in (44) and the appositive to a nominative subject
in (45). In other words, in non-oblique contexts case agreement is obligatory.
(44)

Ich besuchte dann Herrn Muller, *unserem /unseren


I
visited
then Mr.ACC Muller, ourDAT /ourACC
Vertreter
in Pforzheim.
representative in Pforzheim
I then visited Mr. Muller, our representative in Pforzheim.

(45)

alter Mann, *einem /einer


Im
Haus wohnte ein
in-the house lived
anNOM old man, oneDAT /oneNOM
der
a ltesten Bewohner der
Stadt.
of-the oldest inhabitants of-the city
In the house lived an old man, one of the oldest inhabitants of the city.

Oblique part-whole constructions in Warlpiri (data from Ken Hale)


In Warlpiri, part-whole relationships between body parts and the body are generally rendered by means of agreeing DPs. With the grammatical cases (absolutive
and ergative) as well as with the dative, agreement is obligatory. But with oblique
cases such as the allative case, agreement is optional, and when the DPs do not
agree, the possessor of the body part (the whole) shows up in the dative, as in
(50). Observe that Warlpiri marks datives with an additional dative agreement
marker on the auxiliary (examples cited from van Riemsdijk 1983: (52)(56),
248249).
(46)

Kurdu
ka
wanti-mi
rdaka
ngulya-kurra.
childABS pres fallNONPAST handABS holeALL
The child falls into the hole with its hand.

268

Henk van Riemsdijk

(47)

Maliki-rli ka
kurdu
yarlki-mi
rdaka.
dogERG
pres childABS biteNONPAST handABS
The dog bites the child in the hand.

(48)

Maliki-rli ka
kurdu
yarlki-mi
kartirdi-rli.
dogERG
pres childABS bitesNONPAST mouthERG
The dog bites the child with its mouth.

(49)

Kurdu
ka-rla
maliki-ki yarnka-rni
ngirnti-ki.
childABS presDAT dogDAT go-forNONPAST tailDAT
The child goes for the dogs tail.

(50)

a.

b.

Yumangi ka
langa-kurra yuka-rni
maliki-kurra.
fly
pres earALL
enterNONPAST dogALL
The fly flies into the dogs ear.
Yumangi ka-rla
langa-kurra yuka-rni
maliki-ki.
fly
presDAT earALL
enterNONPAST dogDAT
The fly flies into the dogs ear.

It is on the basis of these considerations that I claim that the dative found in
spatial PPs needs no separate explanation: it is simply the default case.
2.4.4. Some Residual Cases
There is a relatively small (and diminishing) number of (mostly non-locative)
German adpositions that govern the genitive most of these are denominal or
deadjectival and possibly reducible, at least in part, to an analysis in terms of
an actual or silent N. And there is a very small group of, again mostly nonlocative, adpositions that govern the accusative: fur (for), ohne (without),
wider (against). I will not address these cases here. Instead I will limit myself
to some brief remarks on five spatial adpositions that are interesting in that,
while not fitting completely into the general pattern described above, they do,
at least in part, show properties that are in line with my proposal.
um
entlang

(around) is a pure RO-P that takes the accusative, as it should, though


(other than the other RO-Ps) it is prepositional.8
(along) is a RO-P, but it occurs in a variety of frames:
as postposition it occurs with the accusative case, as it should;
as preposition it takes the the genitive case;

8. The issue of prepositional vs. postpositional order among spatial Ps is one that I have
not considered in the present paper.

Discerning Default Datives: Some Properties of the Dative Case in German

269

for some speakers (not for the present author) it can also be a pre-or
postposition that takes the dative case, but this type of construction
is on the way out and tends to be replaced by the prepositional
dative: am See entlang (at the lakeDAT along); instead
with dative case, entlang can be used as a locative preposition:
entlang dem See stehen grosse Villen (along the lake stand large
villas)
In other words, entlang is gradually sliding into the general and regular pattern.
entgegen (towards) is a GO-P that (exceptionally) takes the dative case;
zu
(to) is a GO-P that (exceptionally) takes the dative case, but (perhaps
significantly) this preposition seems more resistant to DMPs than the
other GO-Ps: ??300 m zu mir (300 m to me), ??ein Stuckweit zu
seiner Mutter (a part-of-the-way to his mother);
nach
(to) is a GO-P that takes the dative case, but (significantly) we
only know this by inference from the temporal use (after); nach is
used exclusively with place names without articles that do not overtly
express any case:
nach Berlin (to Berlin)
*nach der Hauptstadt (to the capital)
*nach dem Berlin das Du mir beschrieben hast (to the Berlin that
you described to me)
*nach Peter (to Peter)
*nach dem Pazifik (to the Pacific)
nach Den Haag (to The Hague) vs. *nach Dem Haag (to theDAT
Hague) (cf. im Haag (in theDAT Hague) vs. in Den Haag (in The
Hague))
In other words, we might as well say that spatial nach takes the accusative.9
2.5. Conclusion
I have argued that the accusative in German spatial PPs can be fully attributed
to the delimiting measure phrase character of the DP in ROUTE-PPs and that
the accusative in GOAL-PPs is due to the presence of an (implicit) DMP. The
datives that show up in LOCATIVE-PPs and SOURCE-PPs are manifestations
of the more general principle that dative is the default case in oblique domains.

9. Section 3 takes up the issue of non-case-inflected, not overtly case marked nominal
elements that occur in dative contexts in greater detail.

270
3.

Henk van Riemsdijk

Dative Incompatibility Suppression10

3.1. First Preliminaries


If dative marked DPs in PPs get their marking not from case assignment but as a
virtue of the default manifestation of the dative case, questions arise as to whether
this case manifestation is of equal force as that found in true case assignment situations. This section investigates a set of phenomena that, though not well understood in some ways, does shed some light on the issue. In brief, there is a set of uninflected nominal pro-forms that cannot survive in dative case assignment contexts but can, provided certain other conditions are met, survive in (oblique) adpositional contexts. I take this contrast to be another indication that case marking
in oblique and non-oblique domains is quite different, a conclusion that appears
to be consonant with the conclusions reached in Section 2 above in a general
sense, though working out the precise details will be seen to be far from trivial.
German possesses a series of nominal quantificational pro-forms that can
remain uninflected in nominative and accusative domains, but not when they
are assigned dative case, cf. Gallmann (1997: p67f).
(51)

Hans hat viel(-es) / alles


/ wenig(?-es) / nichts
Hans has much
/ everything / little
/ nothing
gegessenACC .
eaten

(52)

Viel(es) / alles
/ wenig(?-es) / nichts hatNOM uns
us
much
/ everything / little
/ nothing has
u berzeugt.
convinced

There is much syncretism in the paradigms, but in particular the above examples
with viel show that while case inflection is possible in these contexts, the uninflected form is tolerated, that is, it is compatible with the requirements imposed
by nominative/accusative domains.
The following examples, however, show that in dative contexts it is only the
inflected form that yields a grammatical result. I will call this effect dative
incompatibility.
(53)

Das widersprichtDAT vielemDAT / allemDAT / wenigemDAT


that contradicts
much
/ everything / little

10. This section presents an abbreviated and partly improved version of van Riemsdijk
(to appear).

Discerning Default Datives: Some Properties of the Dative Case in German

271

(54)

*Das widersprichtDAT viel / allerlei


/ etwas
/
that contradics
much / all-kinds-of-things / something /
nichts / wenig.
nothing / little

(55)

*Was widersprichtDAT das?


that
what contradicts
What does that contradict?

Some of these words including allerlei, etwas, nichts, was lack a dative form
altogether and consequently they are not tolerated as objects of a verb that
governs the dative case.
However, if the governing element is a preposition, this dative incompatibility disappears (modulo the case-less r-pronouns, cf. Gallmann (1997); van
Riemsdijk (1978)).
(56)

mitDAT vielemDAT / beiDAT allemDAT / vonDAT wenigemDAT


with
much
/ with
everything / of
little

(57)

mitDAT
with
was /
what /

viel / allerlei
/ etwas
/ nichts /
much / all-kinds-of-things / something / nothing /
wenig
little

(58)

beiDAT
with
was /
what /

viel / allerlei
/ etwas
/ nichts /
much / all-kinds-of-things / something / nothing /
wenig
little

These data, which I will refer to as dative incompatibility suppression, essentially admit two types of interpretations:
A. The prepositional dative is a different case from the indirect object or verbgoverned dative. That is, we could distinguish grammatical vs. oblique dative case. We can then say that uninflected pro-forms are compatible with
the oblique dative but not with the grammatical dative case, following van
Riemsdijk (1983, 2007).
B. Dative (in-)compatibility is contextually determined: the pro-forms in question are inherently incompatible with dative case, but prepositions suppress
this inherent feature. This is what, following van Riemsdijk (to appear), I will
argue for in the next section (Section 3.2).

272

Henk van Riemsdijk

3.2. An Argument for a Context Dependent Solution


A first piece of evidence for hypothesis B comes from an examination of the
position of adpositions. German, which is generally considered to be prepositional, has postpositional and circumpositional PPs as well. Here are some
postpositions that govern the dative:
(59)

entgegen (towards), entlang (along), nach (according-to), zufolge (as


a consequence of), zuliebe (benefactive for), zuwider (contra)

None of these tolerate any of the case neutral proforms illustrated in (51)
through (58):
(60)

a. *allerlei entgegen DAT


*nichts nach DAT
*was zufolge DAT
*viel zuwider DAT
*viel entlang DAT

(towards all kinds of things)


(according to nothing)
(as a consequence of what)
(opposed to much)
(alongside much)

Second, Bayer and Bader (2007)11 observe that the phonetic weight and the morphological complexity of adpositions may play a role, though there is considerable variation in the judgments. Heavier prepositions yield results significantly
worse than those in (56)(58).
(61)

a.

% wahrend DAT nichts


% wegen DAT allerlei
% dank DAT was

(during nothing)
(because of all kinds of things)
(thanks to what)

It should be pointed out, however, that the examples in (60) are considerably
worse than those in (61). Furthermore, nach, which is monosyllabic, morphologically simplex and not in any obvious sense denominal, deverbal or deadjectival,
can occur both as a preposition and as a postposition; as a preposition governing
the dative it is used in prepositional objects and as a temporal adposition meaning after; as a postposition it means according to; in these various uses, we
get a clear contrast: the preposition nach tolerates uninflected pro-forms, but
the postposition nach does not.
(62)

Nach was hast Du gesucht?


after what have you searched
What were you looking for?

11. They credit Andrew McIntyre with the observation.

Discerning Default Datives: Some Properties of the Dative Case in German

(63)

(64)

273

Nach was kommt das Intermezzo?


after what comes the Intermezzo
After what comes the Intermezzo?
*Was nach
soll
eine Eiszeit bevorstehen?
what according-to should an ice-age be-imminent
According to what is an ice-age supposed to be imminent?

I conclude from these facts that the factor preposition vs. postposition is the
major determinant of whether dative incompatibility is suppressed. Postpositions and verbs both govern leftward in German, as opposed to prepositions.
This suggests that it is the direction of government that plays a decisive role in
whether or not dative incompatibility suppression is active.
This interpretation of the facts is confirmed when we look at adjectives.
Adjectives in German can take DP complements which they govern leftward,
(cf. van Riemsdijk 1983).The case, as argued in van Riemsdijk (1983) and briefly
alluded to above (Section 2.4.3), is mostly the dative case, cf. example (38). In
such a left-governing context, we find no suppression of dative incompatibility,
in other words, the uninflected pro-forms are not tolerated.
(65)

*Nur wenige Leute sind nichts verfallen DAT .


only few
people are nothing addicted
Only few people are addicted to nothing.

(66)

*Was a hnlich DAT sind denn diese Felsformationen?


what similar
are prt. these rock-formations
What are these rock formations similar to?

This observation supports our conclusion that the direction of government determines whether uninflected pro-forms are tolerated under dative government.
Pro-forms such as etwas, nichts, viel, allerlei, was, wenig are inherently specified as nominative/accusative and hence as incompatible with contexts in which
the dative case is required. This dative incompatibility is suppressed, under
non-canonical (that is, left-to-right) case government.12 However, direction of
government is not the only factor determining whether dative incompatibility
suppression (DIS) occurs. Morphological complexity and phonological weight
may also block DIS.
12. I am using the term non-canonical here in view of the overall head-final character of
German syntax: O-V, DP-A, DP-P or P-DP. I am leaving aside the internal structure
of the DP here. If genitive phrases are DPs (and not PPs), they occur both to the
left and to the right of N. Dative phrases in DPs only occur to the left of the N in
possessive constructions, as briefly discussed in 2.4.3.

274

Henk van Riemsdijk

The latter observation invites the conjecture that DIS is an interface phenomenon: uninflected pro-forms survive throughout the syntactic derivation but
the derivation crashes at the PF interface unless the uninflected nominal form
is immediately preceded by a light preposition.
The interpretation of DIS as a PF-interface effect is also supported by considerations related to its interaction with head movement. In their clause final
position verbs cause the dative incompatibility effect, as shown in Section 3.1.
Observe now that the dative incompatibility effect persists (that is, no DIS effect
arises) when the (finite) verb undergoes Verb Second in root clauses, as shown
in the examples (53)(55). Hence, it is not the case that Verb Second creates a
situation of non-canonical (left-to-right) case government as discussed above
in connection with PRE-positions. This might at first sight be interpreted as an
indication that the DIS effect is truly syntactic. That would be the wrong conclusion, however, since it is predicted by phase theory and in particular phasal
transfer. By the time the finite verb is raised to C, the vP phase has become inaccessible to syntactic operations, with the exception of its head (the finite verb)
and its edge, due to the Head Constraint,13 currently referred to as the Phase
Impenetrability Condition (PIC). In phase theory parlance, the inaccessible elements of vP have already been transferred to spell-out. Therefore Phase Theory
in combination with the proposal that the DIS effect applies at the PF-interface,
that is, at spell-out, correctly predicts that Verb Second does not affect dative
incompatibility.
At first sight, this line of reasoning might be thought to apply to the
preposition-postposition alternation as well, which would be a problem as prepositions, unlike postpositions, do exhibit the DIS effect, as shown in Section 3.1.
Before answering this question, however, we need to know how the prepositionpostposition alternation is accounted for. However, this issue remains largely unsolved. On the one hand, a uniform underlying head-final analysis for languages
like Dutch and German would suggest that adpositions are base-generated in
final position and are fronted to yield prepositions. On the other hand, these
languages might be considered to be underlyingly mixed-headed. Furthermore,
it might also be the case that the adpositional head remains inert and that its
dependents move around.14 There is little question, however, that the relationship between pre- and postpositional structures is a matter that does not involve
domains larger than the (extended projection of the) PP itself. While (pace
13. See van Riemsdijk (1978).
14. See Corver (1997) for enlightening discussion of the corresponding question in Dutch
Adjective Phrases. Corver argues that in AP it is the head that moves, rightward in
his proposal.

Discerning Default Datives: Some Properties of the Dative Case in German

275

Chomsky 2008) the PP may well constitute a phase, as argued in van Riemsdijk
(1978), there is every reason to assume that the position of the adposition within
the PP is settled by the time the PP is transferred to spell-out. Hence there is no
reason why the headedness (or quite simply the left- or right adjacency of the
adposition to its complement) should not feed the DIS effect.
These considerations, which support the interpretation of DIS as an interface effect, will be central to our discussion of the behavior of was (what) in
transparent free relative clauses, which I turn to in the following section.
3.3. Was in Transparent Free Relatives
One of the prominent and important properties of headless or free relatives
(FRs) is the matching effect. The FRs wh-word or -phrase, though arguably in
the complementizer position of the relative clause (cf. Groos and van Riemsdijk
1981; van Riemsdijk 2006a), must match the case requirements of the matrix
clause. The matching requirement is sensitive not to the abstract case imposed
in a certain context, but by the surface form of the word or phrase in question.15
The relative pronoun was (what) is syncretic between the nominative and the
accusative case. Hence, an FR introduced by was in German can simultaneously
satisfy a matrix nominative requirement and an accusative requirement in the
relative clause or vice versa, as shown in the following two examples.
(67)

Ich esse ACC was da


ist NOM .
I
eat
what there is
I eat what is there.

(68)

Was ich koche ACC muss gegessen werden NOM .


what I
cook
must eaten
be
What I cook must be eaten.

With datives this does not work. For discussion about the facts, see Groos and
Van Riemsdijk (1981) and Grosu (2003, 2007).
(69)

*Dieses Bild
gleicht DAT was Du gezeichnet hast ACC .
have
this
picture resembles what you drawn
This picture resembles what you have drawn.

15. This property of the matching effect is, of course, fully in line with our earlier conjecture that the case phenomena we are looking at here are situated at the interface.

276

Henk van Riemsdijk

(70)

*Ich kaufe ACC was dieses Bild


gleicht DAT .
I
buy
what this
picture resembles
I buy what this picture looks like.

But the DIS effect is active. A (matrix) preposition suppresses dative incompatibility on the relative pronoun was that initiates the FR. This is fully expected
in view of what we have said about matching in FRs above.
(71)

Die Kinder spielen mit DAT was sie bekommen haben ACC .
have
the children play
with
what they received
The children are playing with what they got.

(72)

Die Kinder spielen mit DAT was verteilt


wurde NOM .
the children play
with
what distributed was
The children are playing with what was distributed.

This is not the place to argue in detail how FRs, and in particular the matching
effect, should be dealt with in current syntactic theory. Let me just point to
the analysis that I have argued for extensively elsewhere, the so-called graft
analysis (cf. van Riemsdijk 2006b) in which the wh-word (here was) is remerged
twice, once from its base position into the SpecCP position of the relative clause,
an instance of internal merge, and once into the direct object position of the
matrix verb, an instance of a combination of internal and external merge which
I refer to as graft (the example illustrated here is (67)).
(73)

Such an analysis predicts that was displays (case) properties both of the matrix
clause and of the relative clause.

Discerning Default Datives: Some Properties of the Dative Case in German

277

Let us now turn to a very special type of FRs, so-called transparent free
relatives (TFRs). These are illustrated by examples such as the following.
(74)

John lost what according to the dictionary are called his marbles

My analysis (cf. van Riemsdijk 2006b and references cited there) for TFRs is
that in an example like this, his marbles and not what is grafted into the matrix
clause to account (among many other properties) for the fact that lose ones
marbles is a local idiom. In other words, it is the predicate nominal that enters
into a matching relationship with the matrix clause. A graft analysis treats such
TFRs as shown in (75).
(75)

Grosus (2003 and elsewhere) has argued against my graft analysis on various
grounds, which space prevent me from discussing here. His alternative analysis
is to assume that TFRs are just like regular FRs and that, due to the predication
relation between was and the predicate nominal, the relevant information about
the predicate nominal is passed along to was and in this way becomes accessible
to the matrix clause. For reasons I have addressed elsewhere (cf. van Riemsdijk
2006b, 2006c), I believe Grosus position is quite untenable.
In a more recent paper (2007), however, Grosu presents a new argument
against my graft analysis which he undoubtedly thinks is the coup de grace.
Indeed, in my analysis of TFRs the DIS effect should not be found since it is the

278

Henk van Riemsdijk

predicate nominal and not was that is grafted into the matrix structure. However,
the DIS effect does obtain.16
(76)

Er wohnt in DAT was man einen Huhnerstall ACC nennen


he lives in
what one aACC chicken-coop
call
konnte.
could
He lives in what one may call a chicken coop.

(77)

Sie spricht mit DAT was ich einen Idioten ACC nennen
she speaks with
what I
aACC idiot
call
wurde.
would
She is talking with what I would call an idiot.

Grosu also observes that this effect is not found in non-prepositional contexts,
as expected:
(78)

*Er hat was man eine merkwurdige Idee ACC nennen


idea
call
he has what one a
strange
konnte viel Aufmerksamkeit geschenkt DAT .
could much attention
given
He has devoted considerable attention *(to) what one might call a
strange idea.

Grosus conclusion, not unexpectedly, is that was in TFRs must be a kind of


relative clause head just like the was in regular FRs, as has been his claim all
along (cf. Grosu 2003).
I believe Grosu is, again, on the wrong track here. In fact TFRs show the
same behavior that we observed above when governed leftward by a postposition
or an adjective (cf. van Riemsdijk to appear). As the following examples show,
postpositions and (leftward governing) adjectives do not suppress the dative
incompatibility effect.
(79)

*Hans ging was man einen kraftig


gebauten Mann ACC
man
Hans went what one aACC powerfully built
nennen konnte entgegen DAT .
call
could towards
Hans went towards what one might call a powerfully built man.

16. The examples are from Grosu (2007: 111).

Discerning Default Datives: Some Properties of the Dative Case in German

279

(80)

*Susanne ist was man als ihre beste Freundin ACC


Susanne is what one as her best woman-friend
bezeichnen kann zufolge DAT 27 Jahre alt.
characterize can according-to 27 years old
Susanne is, according to what one can characterize as her best womanfriend, 27 years old.

(81)

*Gregor war was in der DDR die Stasi ACC genannt


Gregor was what in the GDR the Stasi
called
wurde hinlanglich bekannt DAT .
was
abundantly known
Gregor was well known to what in GDR-times was called the Stasi
(= Staatssicherheitsdienst = State Security Service).

We must be careful here, since In these examples it is impossible to tell whether


the ungrammaticality is caused by the dative incompatibility effect due to was
or by the (grafted) predicate nominal. On the one hand, was cannot suppress
dative incompatibility because it is governed by a leftward governing adjective,
on the other hand the predicate nominal has a non-matching case.
In order to make the argument stick, let us look at adjectives that (leftward)
govern the accusative case. They yield grammatical output and they do so in
precisely those cases where the predicate nominal must be assumed to be shared
(in some way) by the matrix. And indeed, it is the predicate nominal that induces
a case matching conflict.
(82)

Dieses Haus ist was man keinen Heller ACC zu nennen


dime
to call
this
house is what one no
pflegt wert ACC .
uses worth
This house isnt worth what they usually call a dime.

(83)

*Dieses Haus ist was kein Heller NOM zu heissen scheint


this
house is what no dime
to be-called seems
wert ACC .
worth
This house isnt worth what seems to be called a dime.

(84)

Was man den Kudamm ACC zu nennen pflegt entlang ACC


what one the Kudamm
to call
uses along
gehe ich gern spazieren.
go
I
gladly walk
Along what they usually call the Kudamm, I like to go for a walk.

280

Henk van Riemsdijk

(85)

*Was der Kudamm NOM zu heissen scheint entlang ACC


what the Kudamm
to be-called seems along
gehe ich gerne spazieren.
go
I
gladly walk
Along what seems to be called the Kudamm, I like to go for a walk.

Clearly, was does not play any role here since its syncretic properties would
predict that in each case both examples are grammatical. In other words, as far
as case matching is concerned, was is out of the picture since not it but the
predicate nominal is grafted into the matrix structure.
So far so good, but then, why do Grosus examples with prepositions (76)
and show (77) the DIS effect? In line with our earlier findings, the DIS effect is a
pure interface effect. Uninflected pro-forms such as was, nichts etc. suppress the
(dative) case inducing requirement on (non-canonically governing) adpositional
elements immediately to their left in the string. And since this effect takes place
at spell-out, it takes place after linearization, as it must in a graft analysis. Hence,
Grosus argument, though interesting in itself, is without force.
Needless to say, the discussion about the DIS effect in this section raises
many questions, many of which I cannot address here. Nevertheless, I will point
out a number of these questions.
First, if dative is the default case in PPs, it is ipso facto not assigned. But then,
how can we maintain that uninflected nominal proforms, and in particular was,
have the power to suppress the (dative) case inducing requirement on (noncanonically governing) adpositional elements immediately to their left in the
string? The answer, I assume, is to be found in the notion of oblique domain
that has been one of the cornerstones of the idea that the dative case is the
default case in oblique domains. Clearly oblique domains must be syntactically
detectable as such. The oblique domains that have played a role in the present
discussion are the PP and the AP. Another way to put things is to say that P and A
mark their domain as oblique. Formulating the DIS effect more precisely, then,
we need to say that uninflected nominals suppress the oblique marking effect of
prepositions (and adjectives17 ). Importantly, this does not turn a preposition into
a structural case marker assigning, say, nominative or accusative. Otherwise we
would expect the same dative incompatibility effect that we find in the domain
of verbs.
But, and this is a second question, why is the DIS phenomenon subject to a
directionality (of government?) effect? Here I can only speculate. What appears
to be at issue is Germans largely head final but in some cases mixed headedness.
17. Given that adjectives take DP complements to their left only, this addition is vacuous.

Discerning Default Datives: Some Properties of the Dative Case in German

281

Verbs take their nominal complements on the left and so do adjectives, while
adpositions are ambivalent. Noun phrases are the domain of the genitive case,
which has been left out of consideration here.18
For case government, adjectives pattern with verbs, while nouns do not govern any case (modulo genitives). This makes P the only head that (sometimes)
governs rightward. I suspect that, being exceptional, this is a recessive property
and that it is this property that is at the origin of dative incompatibility suppression and is responsible for the fact that the DIS effect is relegated to the
(PF-)periphery.
Third (and last), why is the DIS effect triggered by function words like
viel, wenig, allerlei, etwas, nichts and was (much, little, all-kinds, some
(thing), nothing, what)? These words can occur as full nouns, but are generally regarded as pro-forms or function words. They waver in terms of their
categorial status. First, they occur as quantificational elements in the functional
structure of (extended) nominal, adjectival and adpositional projections:
(86)

a.
b.

c.

viel Geld, wenig Gluck, allerlei Probleme, etwas Wein


much money, little luck, all kinds of problems, some wine
etwas unbequem, wenig besser, nichts Neues, was Gutes
somewhat uncomfortable, little better, nothing new, something
good
etwas u ber dem Durchschnitt, nur wenig unter der 20%-Marke
a bit above the average, only little below the 20% mark

These words occur as functional elements in the functional projections of N,


A and P. Following the tenet of Categorial Uniformity (cf. Grimshaw 1991,
2005; van Riemsdijk 1990, 1998),19 this means that these elements, at least in
their functional use, are seriously underspecified in their categorial identity. This
categorial versatility20 , in all likelihood due to underspecification, may well be
18. Given that adpositions also govern the genitive case, sometimes, why is there not a
phenomenon of genitive incompatibility (suppression)? There are only few adpositions that govern the genitive. Furthermore, these adpositions in many cases govern
the dative alternatively. In fact the genitive may simply be on its way out. Cf. Der
Dativ ist dem Genitiv sein Tod (the dative is the genitives death) by Bastian Sick,
Kiepenheuer & Witsch Verlag, Cologne, 2005). The very title constitutes an example
of a dative possessive construction that replaces the traditional but somewhat archaic
genitive.
19. The Categorial Identity Thesis (CIT) states that functional heads are categorially
non-distinct from the lexical head that projects them.
20. Indeed, even when used independently, they can occur in different functions, e.g. as
adverbials.

282

Henk van Riemsdijk

at the bottom of the deviant, defective behavior of these uninflected elements in


terms of their case properties.
4.

Summing up

In this paper I have argued the following main points.


Default case is domain dependent;
The Default case in oblique domains (including PP and AP) is the dative case;
The accusative that characterizes pure paths and goals in spatial PPs is due
to the presence of an (explicit or implicit) delimiting measure phrase;
The apparent alternation between dative and accusative in spatial PPs must
be reinterpreted as an alternation between the DMP-induced accusative and
the default dative;
P and A are oblique domain inducers; the oblique domain inducing effect
of (light) prepositions (as opposed to postpositions) can be suppressed by
uninflected nominal elements in their (immediate) domain;
This dative incompatibility suppression (DIS) effect is a PF-interface phenomenon.
(i)

a.

b.

Das Gewitter
hat etwas nachgelassen.
the thunderstorm has a-bit abated
The thunderstorm has abated a bit.
Peter ist viel herumgekommen in der Welt.
Peter is much around-come
in the world
Peter has travelled a lot around the world.

In addition, was has a tendency to take on the function of a complementizer


as it does in relative clauses in many variants of Bavarian (wo being the more general
option).
(ii)

de Leid, de was vui


Geid
hobm, . . .
the people who that much money have
the people who have much money

And possibly, the intermediate clause initial was in partial movement constructions in German also has the status of complementizer.
(iii) Was glaubst du was Peter meint wohin wir fahren sollen?
what believe you what Peter thinks where we drive should
Where do you believe Peter thinks we should drive?

Discerning Default Datives: Some Properties of the Dative Case in German

283

References
Abraham, Werner
2003
The myth of doubly governing prepositions in German. In: Erin Shay
and Uwe Seibert (eds.), Motion, Direction and Location in Languages.
In honor of Zygmunt Frajzingier, 1938. Amsterdam: John Benjamins.
Bayer, Josef and Markus Bader
2007
On the syntax of prepositional phrases. In: Andreas Spath (ed.), Interface and Interface Conditions, 157180. Berlin: Mouton de Gruyter.
Beekes, Robert Stephen Paul
1995
Comparative Indo-European linguistics: an introduction. Amsterdam/Philadelphia: John Benjamins.
Blake, Barry J.
1994
Case: Cambridge textbooks in linguistics. Cambridge/New York:
Cambridge University Press.
Chomsky, Noam
1981
Lectures on government and binding. Dordrecht: Foris Publications.
Chomsky, Noam
2008
On phases. In: Robert Freidin, Carlos Otero and Maria Luisa Zubizarreta (eds.), Foundational Issues in Linguistic Theory. Essays in
Honor of Jean RogerVergnaud, 133166. Cambridge, MA: MIT Press.
Corver, Norbert
1997
The internal syntax of the Dutch extended adjectival projection. Natural Language and Linguistic Theory 15: 289368.
Csirmaz, Aniko
Kiss (ed.), Event Structure
2006
Accusative case and aspect. In: Katalin E.
and the Left Periphery: Studies on Hungarian, 159200. Dordrecht:
Springer.
Dikken, Marcel den
2010
On the functional structure of locational and directional PPs. In:
Guglielmo Cinque and Luigi Rizzi (eds.), Mapping Spatial PPs: The
Cartography of Syntactic Structures, Vol. VI, 74126. Oxford/New
York: Oxford University Press.
Fox, Anthony
1995
Linguistic Reconstruction: An Introduction to Theory and Method.
Oxford: Oxford University Press.
Fukuda, Shin
to appear
Object case and event type: Accusative-dative object case alternation
in Japanese. In: Proceedings of the 33rd annual meeting of the Berkeley
Linguistic Society. Oakland: University of California at Berkeley.

284

Henk van Riemsdijk

Gallmann, Peter
1997
Zur Morphosyntax und Lexik der w-Worter. In: Arbeitspapiere des
Sonderforschungsbereichs 340, Sprachtheoretische Grundlagen fur
die Computerlinguistik. Tubingen: Universitat Tubingen.
Goldberg, Adele Eva
1991
It cant go down the chimney up: Paths and the English resultative.
In: Laurel A. Sutton, Christopher Johnson and Ruth Shields (eds.)
Proceedings of the 17th annual meeting of the Berkeley Linguistic
Society, 368378. Oakland: University of California at Berkeley.
Grimshaw, Jane
1991
Extended projection. Ms., Brandeis University.
Grimshaw, Jane
2005
Extended Projection. In: Jane Grimshaw: Words and Structure, 169.
Stanford: CSLI Publications.
Groos, Anneke and Henk C. van Riemsdijk
1981
Matching effects in free relatives: a parameter of core grammar. In:
Adriana Belletti, Luciana Brandi and Luigi Rizzi (eds.), Theory of
Markedness in Generative Grammar. Proceedings of the 1979 GLOW
Conference, 171216. Pisa: Scuola Normale Superiore.
Grosu, Alexander
2003
A unified theory of standard and transparent free relatives. Natural
Language and Linguistic Theory 21: 247331.
Grosu, Alexander
2007
Direct versus indirect approaches to Transparent Free Relatives.
In: Gabriela Alboiu, Andrei A. Avram, Larisa Avram and Daniela Isac
(eds.), Pitar Mos: A Building with a View. Papers in Honour of Alexandra Cornilescu, 107123. Bucharest: Bucharest University Press.
Huijbregts, Riny and Henk C. van Riemsdijk
2001
Location and locality. In: Elena Anagnostopoulou and Marc van Oostendorp (eds.), Progress in Grammar, 123. Amsterdam: Meertens
Institute.
Huijbregts, Riny and Henk C. van Riemsdijk
2007
Location and Locality. In: Simin Karimi, Vida Samiian and Wendy
Wilkins (eds.), Clausal and phrasal architecture: syntactic derivation and interpretation: A Festschrift for Joseph E. Emonds, 339364.
Amsterdam: John Benjamins.
Kagan, Olga and Asya Pereltsvaig
2009
Syntax and semantics of bare NPs: objects of na-sja verbs in Russian
(abstract of talk). In: CSSP2009. Paris.
http://www.cssp.cnrs.fr/cssp2009/abstracts/Kagan.pdf.
Koopman, Hilda
2000
The syntax of specifiers and heads. London: Routledge.

Discerning Default Datives: Some Properties of the Dative Case in German

285

Kuno, Susumu
1973
The structure of the Japanese language. Cambridge, MA: MIT Press.
Leirbukt, Oddleif
1978
Dativische Appositionen bei akkusativischem Bezugswort im
Deutschen. Linguistische Berichte 55: 117.
Nam, Seungho
2004
Goal and Source: Their syntactic and semantic asymmetry. Ms. Los
Angeles.
Noonan, Maire
2005
Spotlight on spatial PPs: Evidence from German shadows. Ms.
McGill, Montreal.
Pesetsky, David and Esther Torrego
2004
Tense, case and the nature of syntactic categories. In: Jacqueline
Gueron and Jacqueline Lecarme (eds.), The Syntax of Time, 495537.
Cambridge, MA: MIT Press.
Pesetsky, David and Esther Torrego
2007
The syntax of valuation and the interpretability of features. In: Simin
Karimi, Vida Samiian and Wendy K. Wilkins (eds.), Phrasal and
Clausal Architecture: Syntactic Derivation and Interpretation. In
Honor of Joseph E. Emonds, 262294. Amsterdam: John Benjamins.
Riemsdijk, Henk C. van
1978
A case study in syntactic markedness: the binding nature of prepositional phrases. Lisse: The Peter de Ridder Press, later published by
Foris Publications Dordrecht and currently by Mouton de Gruyter,
Berlin.
Riemsdijk, Henk C. van
1983
The case of German adjectives. In: Frank Heny and Barry Richards
(eds.), Linguistic categories: auxiliaries and related puzzles, 223252.
Dordrecht: Reidel.
Riemsdijk, Henk C. van
1990
Functional prepositions. In: Harm Pinkster and Inge Genee (eds.),
Unity in Diversity: papers presented to Simon C. Dik on his 50th
birthday, 229241. Dordrecht: Foris Publications.
Riemsdijk, Henk C. van
1998
Categorial feature magnetism: The endocentricity and distribution of
projections. The Journal of Comparative Germanic Linguistics 2: 1
48.
Riemsdijk, Henk C. van
2006a
Free relatives. In: Martin Everaert and Henk C. van Riemsdijk (eds.),
The Blackwell Companion to Syntax, Vol. II, 338382. Oxford: Blackwell.

286

Henk van Riemsdijk

Riemsdijk, Henk C. van


2006b
Grafts follow from Merge. In: Mara Frascarelli (ed.), Phases of Interpretation, 1744. Berlin: Mouton de Gruyter.
Riemsdijk, Henk C. van
2006c
Towards a unified theory of wh- and non-wh-amalgams. In: Yubun
Suzuki, Mizuho Keizo and Ken-ichi Takami (eds.), In search of the
essence of language science: Festschrift for Professor Heizo Nakajima
on the occasion of his 60th birthday, 4359. Tokyo: Hitsuji Shobo.
Riemsdijk, Henk C. van
2007
Case in spatial adpositional phrases: the dative-accusative alternation
in German. In: Gabriela Alboiu, Andrei A. Avram, Larisa Avram and
Daniela Isac (eds.), Pitar Mos:A Building with aView. Papers in Honor
of Alexandra Cornilescu, 265283. Bucharest: Bucharest University
Press.
Riemsdijk, Henk C. van
(to appear)
A note on dative incompatibility suppression. In: Anna Cardinaletti,
Giuliana Giusti and Nicola Munaro (eds.), Festschrift for Guglielmo
Cinque. Oxford: Oxford University Press.
Sugamoto, Nobuko
1982
Transitivity and objecthood in Japanese. In: Paul J. Hopper and Sandra A. Thompson (eds.), Studies in Transitivity, 423447. New York:
Academic Press.
Svenonius, Peter
2010
Spatial prepositions in English. In: Guglielmo Cinque and Luigi Rizzi
(eds.), Mapping Spatial PPs: Cartography of Syntactic Structures,
Vol. VI, 127160. Oxford/New York: Oxford University Press.
Tenny, Carol
1994
Aspectual Roles and the Syntax-Semantics Interface. Dordrecht:
Kluwer.
Vat, Jan
1997
Left Dislocation, Connectedness and Reconstruction. In: Elena Anagnostopoulou, Henk C. van Riemsdijk and Frans Zwarts (eds.), Materials on Left Dislocation, 6792. Amsterdam: John Benjamins.
Vergnaud, Jean-Roger
2008
Letter to Noam Chomsky and Howard Lasnik on Filters and Control,
April 17 1977. In: Robert Freidin, Carlos P. Otero and Maria Luisa
Zubizarreta (eds.), Foundational Issues in Linguistic Theory: Essays
in Honor of Jean-Roger Vergnaud, 315. Cambridge, MA: MIT Press.
Von Stechow, Arnim
1996
The different readings of wieder again: a structural account. Journal
of Semantics 13: 87138.

Discerning Default Datives: Some Properties of the Dative Case in German


Winter, Werner
1966
Zwarts, Joost
2005
Zwarts, Joost
2006

287

Vom Genitiv im heutigen Deutsch. Zeitschrift f. Deutsche Sprache 22:


2135.
The case of prepositions: Government and compositionality in German PPs. Ms. Radboud University, Nijmegen and Utrecht University.
Case marking direction: The accusative in German PPs. In: Jackie
Bunting, Sapna Desai, Robert Peachey, Chris Straughn and Zuzana
Tomkova (eds.), Proceedings from the 42nd Annual Meeting of the
Chicago Linguistic Society, 93107. Chicago: Chicago Linguistic Society.

How Phase-Based Interpretations Dictate


the Typology of Nominalizations
Leah Bauke and Tom Roeper

0.

Introduction and Outline

A classic conundrum is the relation of inflectional morphology to derivational


morphology: why are they sometimes the same and what properties carry over
from inflectional to derivational structure? To be explicit, why, for instance, do
we have a progressive morpheme -ing in nominalizations like (1) and -ing in a
result form like (2) and in the intermediate form (3):
(1)

Johns singing songs beautifully.

(2)

the drawings

(3)

Johns singing of songs beautifully.

An expression like his singing beautifully carries the activity reading of inflectional -ing, but no progressive meaning. The result nominals carry no activity
whatsoever:
(4)

His strange singing surprised everyone.

A parallel set of theoretical questions arise for Phase theory. A number of papers,
beginning with Fu, Roeper, and Borer, have argued for the presence of a VP
within the DP, using ellipsis, adverbs and aspectual evidence, and it is assumed
in most subsequent analyses (Barrie 2006; Alexiadou (et al.) 2007; Sichel 2009).
Under this assumption, two Phases arise:

290
(5)

Leah Bauke and Tom Roeper

DP

= Ph2
D

-ing

AspP

Spec

= Ph1
Asp

-ing

Under the Strong Minimalist Thesis, Transfer to Interpretation is called for at


both Phases: DP and vP (or ASP-P, which we will assume).
We will argue that the pattern of interpretation and the exclusion of aspectual
information follows from the SMT applied at these two points. The argument entails that one see English as containing two -ing morphemes, which prima facie
may seem surprising, but is reinforced by cross-linguistic comparison, because
in German one meaning is captured with -ung and the other with -en. We will
argue that the internal structure of gerunds includes verbal structure, but that the
nature of Phase interpretation will eliminate a set of meanings linked to Aspect.
Thus we argue that deep principles of syntax affect morphology. Evidence from
the periphery of grammar for fundamental principles is the strongest evidence
that they are real.
In particular the derivation introduces Phase boundaries when DP is projected
which, under the Strong Minimalist Thesis, blocks access to material inside
lower VP Phase boundaries. The SMT requires interpretation at the Phase Edge.
If the verb is raised into the DP, then lower Aspectual nodes are no longer
visible, under the assumption that the vP is a Phase. We thus reproduce within
nominalizations a variety of Phase-level restrictions found elsewhere in syntax
and the interpretive restrictions of the Strong Minimalist Thesis.
The execution of this perspective involves a classic mode of argument, found
in long-distance wh-movement. If the verb+object remain in the lower vP, they
are interpreted there, together with possible aspectual information, as in:
(6)

the mowing of the lawn in two hours

Under minimalist theory, if the verb+object moves out of the lower Phase to
the higher Phase, then they are only interpreted at a later point, at which the
lower Aspectual information is no longer available because the Phase is over.
In order to capture these Phase properties correctly, we therefore argue that
aspectual information is only preserved when the interpretation occurs in the

How Phase-Based Interpretations Dictate the Typology of Nominalizations

291

first (i.e. lower phase) vP-Phase but blocked when the interpretation occurs in the
second, (i.e. higher) DP Phase. The Phase-Head (e.g. CP) is not transferred, only
the Phase-complement (e.g. IP). In order to execute this mechanism, the lower
verb+object must move into the Phase-Head, which remains for interpretation
until the next Phase, while the Phase-complement undergoes Transfer. This is
just like a wh-word moving into a lower clause CP, before cyclic movement, and
avoiding Transfer of the Phase-complement. Therefore we argue that the lower
verb+object (e.g. mow lawn) moves out of the VP into the ASP node, when VP
is transferred.
A natural question to ask is: why would the verb+object (i.e. mow lawn)
move as a unit? If the object occupies the clitic position, advocated in Keyser
and Roeper (1992), which is a part of a verb and the launching site for incorporation, then they would naturally move together, and be available for incorporation.
Thus, before movement the argument is moved into the clitic position, or alternatively, the THEME is directly projected into the clitic position and forced out
if it is a Phrase not a Head. If the ultimate incorporation occurs within a DP, then
the event is pluralizeable as well, and we predict the possibility of such forms:
(7)

the lawn-mowings

And we predict the impossibility of:


(8)

*the lawn-mowings in two hours

because the Aspect information is too low and could not be interpreted in the
first Phase without the verb present.
These are the extreme cases and many intermediate cases occur with varying
degrees of grammaticality which we will discuss. What happens if the object is
not incorporated, but still in the DP? Then we find that of-insertion occurs to
provide case, as argued classically, pluralization is still possible, but aspect is
still ruled out:1
(9)

a. the mowings of the lawn (were interminable)


b. *the mowings of the lawn in two hours

We will now present this argument in greater detail and show its connection to
other theories.
1. The form in (9b) is often marked ?? or even * because it appears that it could belong
to the vP level phase. We argue that semantics indicates that it is a part of the DP-level
phase and therefore receives a Manner or Style reading rather than a simple direct
object reading.

292
1.

Leah Bauke and Tom Roeper

The Basic Structures

As we have shown, contrary to standard assumptions in the literature (cf. e.g.


Grimshaw 1990, Alexiadou, Iordachioaia and Soare 2009) nominal gerunds
allow for pluralization in their incorporated as well as in their non-incorporated
forms:
(10)

a.
b.
c.
d.

Teds /The cutting of (the) grass


Teds /The grass-cutting
Teds /The cuttings of (the) grass
Teds /The grass-cuttings

The corresponding structures for the forms in (10) are the following, where we
assume the two Phases mentioned above and the movement to the Phase Head:
(11)

a.

DP
NP

Ted D
s

NP
Spec

N
N

V+AspP
cutting/*s

AspP
DP
the grass

Asp
Asp

VP

cutting V
cut

DP
the grass

How Phase-Based Interpretations Dictate the Typology of Nominalizations

293

DP

b.
DP

DP

TED

NP
Spec

N
N

AspP

V+AspP
grass-cutting/*s

VP
grass-cut

Asp
Asp
ing

VP
NP

VP

grass V

NP

cut

grass

DP

c.
D
the

DP
Spec

N
V+AspP
cut

AspP
n

DP/NP

Asp

ing/s (the grass) Asp


cut

VP
V

DP

cut

the grass

294

Leah Bauke and Tom Roeper

DP

d.
D
the

DP
Spec

N
V+AspP
grass-cut

AspP
n

Asp

VP

ing/s grass-cut NP
grass

VP
V

NP

cut

grass

While (10c)/(11c) and (10d)/(11d) show that nominal gerunds can be pluralized,
these forms cannot be modified by aspectual PPs or adverbial phrases despite
the fact that they contain aspectual structure, which follows naturally from the
assumption that a Phase boundary blocks access to the lower verbal functional
structure:2
(12)

the grass-cuttings /cuttings of (the) grass


*[PP with a scythe]
*[PP for hours]
*[PP in an hour]
*[AdvP immediately]
*[AdvP unfortunately]
*[and Bobbys doing so too]

Also, nothing about everyday meaning rules out the illicit modifications as is
illustrated by the example in (13) where not even under a multiple event reading
the aspectual modifier for hours is licensed:
(13)

*The shootings of Jews for hours in the holocaust did not bother the
participants.

2. Note incidentally that this is absolutely compatible with the assumption that adjunction is Late or at least Later Merge as noted for instance in Boeckx (2008), Chomsky
(1993) and Lebaux (1988).

How Phase-Based Interpretations Dictate the Typology of Nominalizations

295

Here again the Phase boundary below the nominalizing -ing suffix blocks access
to the aspectual phrase that is adjoined below the nominal node.
In effect, then, we argue for a distinction between two types of -ing affixes.
One is an aspectual affix that is generated in a lower vP Phase where also
aspectual modifiers are licensed and one is a nominalizing affix that is merged
as the Phase head of a higher Phase.This N-head can thus host plural features and,
by virtue of constituting a Phase boundary, it blocks access to lower functional
structures, in line with the PIC. Further support for making a distinction between
two types of -ing in English comes from German, where the aspectual -ing
structures are nominalized infinitives and the the nominal -ing structures are
-ung nominalizations.
2.

A Closer Look

2.1. Non-Incorporated Non-Plural Nominal Gerunds


Abney (1987) in his seminal analysis of different types of gerunds argues that
gerundives, (i.e. verbal gerunds of the acc-ing, poss-ing and potentially also
pro-ing type) contain verbal structure that is adjoined in syntax, while nominal
gerunds (i.e. -ing of gerunds) do not contain any verbal functional structure
above V and the -ing affix is a lexical affix here that is adjoined in morphology.
The verbal properties Abney identifies for these gerunds, namely that they tolerate particles and that they cannot be passivized without passive morphology
(cf. ibid. 21417) are not further discussed in his account and do not have any
impact on his structural classification. Siegel (1997) basically follows Abneys
representation and focusses on gerundives, saying little about nominal gerunds.
Alexiadou, Iordachioaia and Soare (2009) capture the distinction between nominal and verbal gerunds in terms of boundedness and argue for nominal functional
projections above VP in their account, hence following Abneys classification of
nominal gerunds as not containing any verbal functional structure, yet, in this
account -ing attachment is a syntactic rather that a morphological process.
Van Hout and Roeper (1998) and Fu, Roeper and Borer (2001) on the other
hand point out that complex event nominals or argument structure nominals in
the sense of Grimshaw (1990) and thus also nominal gerunds might have the
external distribution of derived nominals or NPs just as Abney indicates
but that they also show a significant amount of verbal properties, indicating that
they contain verbal functional structure and in particular an aspectual projection
(which according to Alexiadou, Iordachioaia and Soare (2009) is only present
in verbal gerunds). Evidence for the fact that verbal structure is licensed comes
from the following examples:

296
(14)

Leah Bauke and Tom Roeper

a. Teds cutting of the grass in an hour


b. Teds cutting of the grass with a scythe /immediately
c. Teds cutting of the grass and Bobbys doing so too
d. *Teds cutting of the grass unfortunately

(14a) illustrates that nominal gerunds license (even telic!) aspectual modifiers,
which are most likely adjoined to the AspP in (11a). Furthermore, non-sentential
adverbial and prepositional modifiers are licensed, as (14b) shows, which again
points to verbal functional structure in these types of nominals. The fact that sentential modifiers are not licensed (cf. (14d)) follows naturally from the structure
in (11a) where TP or higher functional projections are missing. Finally, (14c)
indicates that the appropriate non-tensed form of the do-so anaphor is licensed
as well, which is yet another indication that TP is missing but verbal functional
structure below that and in particular AspP is projected.
Finally, when comparing the forms in (14) to the forms in (15) below, what
prima facie looks like a counter-argument to the analysis suggested here actually
provides further support for the assumptions made:
(15)

a. ?Johns cutting of the lawn for hours but never finishing it was a
problem.
b. Johns cutting the lawn for hours but never finishing it was a problem.

When the DP object stays in situ as it does in (15) b. the aspectual reading is
more naturally available. In (15) a., however, the object raises and the meaning
of the DP is fixed in the 2nd Phase, therefore the first vP/aspectual Phase cannot
be accessed. Notice that in (11) a. a nominal Phase does occur, however, it is
on top of the aspectual node with its -ing head. This is the second Phase, which
adds definiteness that can be seen as creating an implicature of completeness as
is familiar and standardly assumed for cases like the following:
(16)

John ate the pie.

Hence, the subtle distinctions between (15a) and (15b) corroborate the assumption that the -ing affix is an aspectual affix also in the nominal gerunds, instead
of undermining it.
2.2. Incorporated Non-Plural Nominal Gerunds
The question that immediately arises from the discussion in the previous section is, whether the incorporated forms of nominal gerunds also contain verbal
functional structure or whether the incorporation site is low in the tree, resulting

How Phase-Based Interpretations Dictate the Typology of Nominalizations

297

in the structures that Abney (1987), Siegel (1997) or Alexiadou, Iordachioaia


and Soare (2009) suggest for nominal gerunds, i.e. excluding any verbal functional structure above V(P). A comparison of incorporated and non-incorporated
nominal gerunds to the corresponding verbal structure allows for the following
conclusion:
(17)

a.
b.
c.

Teds grass-cutting
Teds cutting of the grass
Ted cuts the grass

Following the line of reasoning in Kratzer (1994), van Hout and Roeper (1998)
point out that in (17c) the eventuality variable is closed off via existential closure
by the tensed T head, which leads to the event interpretation of this structure. The
incorporated nominal in (17a), on the other hand, is ambiguous between an event
and a result reading, which is the result of generic binding of the event variable.3
The interpretational difference between the verbal forms and the incorporated
nominals that follows from the different licensing properties of the event variable
(i.e. existential closure vs. generic binding), leads van Hout and Roeper (1998)
to the conclusion that the incorporated element is a non-maximal projection,
i.e. a head. This head is base generated in the abstract clitic position (ACP) of
the verb, originally identified in Keyser and Roeper (1992) for (among others)
verb-particle constructions. Under this analysis clitics are base-generated in a
position to the right of the verb:
(18)

V
V

ACP

play dumb/chess/out

similarly: lose out, stand out, hold up, . . .


Note: the category of the item in the ACP can be P, N or A.
In the verbal/sentential domain the clitics move to LF covertly, but in the nominal
domain the clitic is incorporated overtly, leading to the structure in (19) (cf. also
Keyser and Roeper 1997):

3. In fact, existential closure is not an option in (17b) either, because here (as shown in
Section 2.1) TP is not projected either. So the event interpretation in this structure
might also arise from generic binding.

298

Leah Bauke and Tom Roeper

(19)

V(P)
N
grass

V(P)
V

cut

grass

Van Hout and Roeper (1998) go on arguing that the structure in (19) then is
incorporated into a nominalizing affix, which means that in this approach there
is no indication for verbal functional structure on top of the incorporated verbal
form. While this account provides a natural explanation for the interpretational
difference between (17a) and (17c) it remains silent about why the form in
(17b) can get an event interpretation just like the form in (17a) but not the result
interpretation of that latter form. One possible answer is that the variation for the
form in (17a) is due to the fact that the incorporated element is not an argument
of the verb. In fact, Harley (2009) argues that the incorporated element can be
analyzed either as an unanalyzed root with incorporation being triggered by a
case feature or it can be analyzed as having undergone category change to a
nominal category prior to -ing attachment. The distribution of the incorporated
element then is strongly distributional, basically mirroring the effect of the ACP
in van Hout and Roeper (1998). Though Harleys analysis offers a solution for
getting to grips with the two readings of (17a) that correlate with the argument/
non-argument status of the incorporated element, additional problems arise. First
of all, the status of the case-feature that triggers incorporation is left unclear in
her account, thus leaving open the question why overt LF-movement leads to the
two interpretational variants in (17a) while one of these is blocked in the nonincorporated form in (17b). Furthermore, Harleys account, just like van Hout
and Roepers, does not provide for any verbal functional structure on top of the
VP and in both analyses the incorporated forms result from head movement, an
assumption that is not unproblematic in modern minimalist theorizing.
Barrie (2006) avoids the problem of head movement and argues for phrasal
movement of the internal argument, which is forced by a symmetric c-command
relation between the verb and its complement, thus following a weak-antisymmetry approach in the spirit of Moro (2000):

How Phase-Based Interpretations Dictate the Typology of Nominalizations

(20)

VP
V
washing

299

VP
N
glass

N
glass

VP
V
washing

N
glass

Here the verb and the internal argument, i.e. a bare N, are in mutual c-command
and thus cannot be linearized (cf. Kayne 1994). This symmetric c-command
relation is dissolved by adjoining the N in the specifier of VP. So, this approach
avoids the problems of head movement but still does not provide an accurate
explanation for the interpretational variants attested for (17a), nor does it license verbal functional structure on top of VP. In fact, Barries analysis does
not provide any information on the status of the -ing suffix, which makes the
verbal projection and the lack of functional projections on top of it even more
problematic and leaves the external nominal distribution of the gerund totally
unaccounted for.
When, in analogy to the non-incorporated forms, the -ing suffix is analyzed
as an aspectual affix the incorporation of the nominal argument can still be
determined by a symmetric c-command relation between that argument and
the verb, resulting in the structure in (11b). Under this account the properties
exhibited in (21) follow naturally:
(21)

a. Teds grass-cutting and Bobbys doing so too


b. Teds grass-cutting with a scythe
c. Teds grass-cutting for hours
d. *Teds grass-cutting in an hour
e. Teds grass-cutting immediately
f. *Teds grass-cutting unfortunately
g. Tedi enjoyed PROi rock throwing

Just like for the non-incorporated forms, the relevant form of the do-so anaphor
is licensed (cf. (21a)) and non-sentential adverbial and prepositional modifiers
are licensed (cf. (12b) and (21e)), while the sentential modifier in (21f) is again
illicit. The contrast between (21c) and (21d) illustrates that only atelic modifiers
are licensed, which is expected because the incorporated element is a bare N that
is not quantized. If it were quantized, it would not be in a symmetric c-command
relation to the verb thus not necessitating incorporation in the first place. (21g)
finally shows that even control is possible which again underlines the eventuality

300

Leah Bauke and Tom Roeper

interpretation of this form, because this is where the PRO-form is expected to


be generated and licensed.
However, the incorporated and the non-incorporated form of the nominal
gerund do not pattern exactly alike. It has been indicated already that the incorporated nominal alternates between an event and a result interpretation. When
interpreted as eventive, the incorporated nominal gets a Kind reading while the
non-incorporated form is Specific:
(22)

Teds rock-throwing kind

(23)

Teds throwing of the rocks specific

So, the incorporation here establishes a Kind reading, which is in line with
observations originally made in Williams (1981) that incorporated nouns are not
arguments, but rather Manner phrases, which nonetheless absorb the THEME
argument projection of the verb. Thus we have:
(24)

John likes opera-singing.

which indicates a preference for a style, but it is still incompatible with


THEMES:
(25)

a. *John likes opera-singing of Verdi.


b. *John likes opera-singing of songs.

but if it is explicitly manner adverb incorporation becomes possible:


(26)

John likes opera-style-singing of songs.

Thus, incorporation refers to Kinds, and differs from object projection via an
of-phrase.4
Incidentally, the Kind vs. Specific distinction offers another way to motivate
movement by meaning. If we assume that the incorporation position can have a
Kind-feature projection, then it could serve as a motivation for movement rather
than an abstract kind of case. We will not explore the question further here.
4. Note, that this cannot be captured under Harleys account either, where the ungrammaticality of (i) is left unexplained:
(i)

*the truck-driving of Fords

Keyser and Roeper (1992) argue that the head is moved into the clitic position from
the argument position after the THEME has been satisfied. Alternatively, one can
allow the verb to project the THEME theta-role to the clitic position. In any case, as
the example above shows it is not a pure adjunct.

How Phase-Based Interpretations Dictate the Typology of Nominalizations

301

2.3. Non-Incorporated Plural Nominal Gerunds


As has been pointed out in the previous section, nominal gerunds show strong
evidence for the existence of verbal functional structure on top of the VP node
and particularly for an aspectual projection this being the place where the
-ing affix is generated. It has also been pointed out that some forms are actually ambiguous between an event and a result reading. Non-incorporated plural
nominal gerunds do not seem to display this ambiguity, which raises some questions about their internal structure, but justifies the fact that incorporation is not
obligatory.
In light of Grimshaws (1990) analysis of complex event nominals, pluralized forms of nominal gerunds are completely unexpected and should not be
grammatical. As the examples in (10c)/(11c) illustrate, however, these forms
are attested. In fact, Roodenburg (2006) already suggests that the availability
of plural marking is parameterized with Romance languages allowing for pluralized argument structure nominals while in Germanic languages they are not
well-formed. This, however, does not account for the plural marker on the English forms. As shown in Alexiadou, Iordachioaia and Soare (2009), (based on
insights from Iordachioaia and Soare (2008) on nominalizations of Romanian
infinitives and supines) what Roodenburg describes as inter-language variation
rather is an intra-language variation phenomenon. Alexiadou, Iordachioaia and
Soare (2009) describe the distinction between nominal and verbal gerunds in
these terms, arguing that only the former can be pluralized and do not display
any verbal functional structure on top of the VP node.
Essentially they argue for a distinction in terms of boundedness, with the
aspectual projection in verbal gerunds and the classifier projection in nominal
gerunds being in complementary distribution, thus reflecting the distinction between inner and outer aspect. In fact, the structure that Alexiadou, Iordachioaia
and Soare (2009) suggest for verbal gerunds is on a par with the structure suggested for nominal gerunds in (11c) above. The main motivation for excluding
verbal functional structure from nominal gerunds in Alexiadou, Iordachioaia
and Soare (2009) is that adverbial modifiers are illicit in these structures. This
is illustrated in the following sample sentences (cf. ibid.):
(27)

a. *The carefully restoring of the painting took six months.


b. The prompt answering of the question surprised the critics.

However, as pointed out for complex event nominals other than -ing of nominals
in Fu, Roeper and Borer (2001) and as discussed in Section 2.2 for nominal
gerunds, adverbial modification is possible in these structures, as long as these
modifiers are generated in a post-head position in the right periphery. This is still

302

Leah Bauke and Tom Roeper

expected to hold for nominal gerunds that contain verbal functional structure
below a nominalizing node. The fact that adverbial modifiers are licit only in
the right periphery and adjectival modifiers show up only in the left periphery
is what naturally follows from the configuration in (11c). Hence, the criterion
of adjectival vs. adverbial modification in the left periphery is neither sufficient
to rule out verbal functional structure in nominal gerunds nor does it provide an
argument for the existence of verbal functional structure in verbal gerunds:
(28)

a. *his /him carefully restoring the painting . . .


b. his /him restoring the painting carefully . . .
c. ?his /him restoring carefully the painting . . .
d. his restoring of the painting carefully

The question that remains, however, is why the forms in (10c)/(11c) license
plural markers. The examples in (29) show that pluralized nominal gerunds do
not show the characteristics that have been pointed out for their non-pluralized
counterparts in Sections 2.1 and 2.2.
(29)

a.
b.
c.
d.

*the /Teds cuttings of the grass immediately


*the /Teds cuttings of grass for hours /in an hour
*the /Teds cuttings of grass with a scythe
*the /Teds cuttings of grass and Bobbys doing so too

Neither the do-so anaphor, nor non-sentential adverbial modifiers or aspectual


modifiers are licensed.
These data become less puzzling when we observe a subtle distinction. Nominal Event-plurals do not denote an event but either a result (30a) or a plurality
of individualized and distinguishable events (30b):
(30)

a.
b.

Result:
the cuttings of grass
Event Variation:
The illegal shootings of the deer in the forest happened in very
different circumstances, so different fines were levied.
The renderings of the murder in court testimony were sharply at
odds.

The distinction between a result interpretation and that of a plurality of individualized events rests on the nature of the direct object in so far, as a direct object
that is a definite description allows for the plurality of individualized events
interpretation, while a bare noun object does not:

How Phase-Based Interpretations Dictate the Typology of Nominalizations

(31)

a.
b.

303

the roastings of coffee > different roasts, e.g. strong and mild
coffee
the roastings of the coffee > same coffee roasted more than once

The same effect can be observed when the direct object is a plural form:
(32)

a.
b.
c.
d.
e.
f.

the screenings of movies


the killings of journalists
the firings of guns
the snatchings of cell phones
the bombings of Hiroshima and Nagasaki
the trouncings of Germany at major tournaments

These stand in contrast to the forms in (33) and (34):


(33) *?the mowings of the grass
where different styles of mowing are imaginable, but not normal, hence either
a * or ? judgement.
(34) ??the shootings of the journalists
Being THEMES requires the DPs to be generated inside the same Phase as
the gerund form. This, however, is not the case. The affix -ing, which is not
an aspectual but a nominalizing affix here, is generated in a higher Phase (cf.
(11c)), where plural morphology attaches to this nominalizing Phase-head. This
nominal Phase then blocks access to the lower verbal functional structure and
in particular to Spec, Asp, where the THEME DP is projected.
Notice, however, that the forms in (30)(32), which are not THEMES and
which are not definite DPs are licit. This strongly implies that aspectual structure
is projected in these nominal gerunds as well and that it is just not accessible,
as is also evidenced by the inadmissibility of the aspectual modifiers in the
examples in (29). Thus once the first Phase is complete, the aspectual features are
inaccessible, but the THEME argument is marginally accessible, either because
it raises via the verbs clitic position as a bare noun or because the verb still
projects a THEME from the higher DP, reaching into an already completed
Phase, hence the sense of marginality or ungrammaticality.
This is supported by the distinction between inner and outer morphology
presented in Marantz (2007). At the level of inner morphology a little x-head is
merged with a category-neutral root, which is what corresponds quite closely
to the structure in (11a) and (11b), where -ing is an aspectual affix. At the level
of outer morphology a category changing affix is merged with an already cat-

304

Leah Bauke and Tom Roeper

egorized head, which corresponds to the nominalizing affix in (11c). Again,


following the analysis in Marantz, the nominalizing -ing affix on a par with
other category changing little x-heads constitutes a Phase head. In line with
standard assumptions of Phase theory (Chomsky 2008) only the Edge of the
Phase is active for further computation. The inaccessibility of lower functional
projections and in particular the aspectual projections for modification follows
from this quite naturally. The adjunction of aspectual modifiers, like all adjunction, is an instance of Later-Merge (cf. Boeckx 2008a), and upon Spell-Out of
the Phase complement the Edge Property makes the lower functional projections
inaccessible for adjunction just as much as it blocks the licensing of THEME
DPs as internal arguments.
2.4. Incorporated Plural Nominal Gerunds
In 2.3 it has been shown that the apparent ambiguity between an event and a result
interpretation of singular nominal -ing of gerunds is dissolved upon pluralization. While the singular forms are ambiguous between those forms discussed in
2.1 and 2.2 where the affix is an aspectual affix and those discussed in 2.3 where
the affix is a true nominalizing affix, the plural forms can be clearly dissociated
from the former. Pluralization is only possible in those cases, in which the -ing
affix is a nominalizing affix. It is thus expected that the same diagnostics apply
to incorporated pluralized nominal gerunds as well. As illustrated in (35) these
are (somewhat unexpectedly) abundant:
(35)

grass-cuttings, beer-brewings, coffee-roastings, chair-stringings,


movie-screenings, gravel-pilings, paper-writings, deer-shootings,
wood-splittings, forrest-clearings, picture-framings,
appointment-plannings, stone-carvings, church-burnings,
blood-testings, river-bendings, network-programmings,
road-crossings, interview-schedulings, lap-timings, peace-makings,
lip-kissings, pipe-sealings, rock-throwings, ghost-sightings,
heart-monitorings, mine-stoppings, stove-heatings, inmate-beatings,
gun-firings, book-readings, news-reportings, city-bombings . . .

None of these plural forms allows for a single-type event interpretation. Instead
these forms (just like their non-incorporated counterparts in 2.3) are all interpreted either as a result or as a plurality of individualized events. As expected,
without the plural marker on the -ing suffix the forms can still be interpreted as
results, but an event reading, which actually corresponds to the aspectual -ing
affix, is recoverable:

How Phase-Based Interpretations Dictate the Typology of Nominalizations

(36)

a.
b.
c.
d.
e.

305

the picture-framing (carefully)


the road-crossing (by bike)
the stove-heating (with wood)
the chair-stringing (for hours)
Johns pipe-sealing (and Bills doing so too)

This is not surprising because the lower verb+object allows incorporation as


well.
In non-inflected NN incorporations, on the other hand, it is not possible, to
recover an event interpretation, which follows naturally from the fact that verbal
functional structure is not involved in these forms:
(37)

a.
b.
c.
d.

*the movie screen for hours


*Dicks gun fire and Georges doing so too
*the stove heat with wood
*the lap-time accurately

The incorporated plural forms of -ing of gerunds do not license aspectual modifiers, non-sentential adverbial modifiers, prepositional modifiers or the do-so
anaphor either, nor is control of PRO licit here:
(38)

a.
b.
c.
d.
e.

*the movie-screenings for hours


*the stove-heatings with wood
*the lap-timings accurately
*Dicks gun-firings and Georges doing so too
*Jacki enjoyed PROi rock-throwings

As argued for the non-incorporated forms in 2.3, these modifiers are not licensed, because lower verbal functional structure is not accessible to material
adjoined by Later-Merge, in a higher cycle. Once again the Edge Property of
the Phase-inducing nominalizing -ing affix blocks the lower projections which
again determine the crucial contrast between an exclusive result interpretation
for incorporated mass terms and that of a plurality of individualized events for
bare N incorporations.
In effect, the pluralized forms show the same variation between the incorporated forms and their non-incorporated counterparts with respect to the interpretation of the direct objects as the singular forms, where the -ing affix is an
aspectual affix. If the object DP is incorporated it is not an argument and it gets
a Kind-reading.
Our analysis comports well with other recent approaches to morphology. Just
like in the non-plural cases incorporation can be seen as forced by a symmetric

306

Leah Bauke and Tom Roeper

c-command relation between the verb and its complement (cf. Barrie 2006;
Moro 2000; Kayne 1994) that is resolved by Comp to Spec roll-up. As a result,
the V+N complex is merged with the nominalizing -ing affix as a complex head
and thus can escape the lower Phase.
If the DP is not incorporated though, it moves to AspP separately. From there,
however, it cannot move together with the V-head to the nominalizing node,
which is in a higher Phase. This is why these forms are blocked and why only
non-THEME DPs can be merged in this position. So, here the higher Phase,
which is instantiated by the nominalizer -ing, does not only block aspectual
modifiers but also prevents THEME objects in Spec,Asp in the non-incorporated
forms.
(39)

*DP
D

AspP

V+Asp n

DP/NP

Asp'

ing/s THEME Asp

VP
V

DP
THEME

Hence, the higher Phase has the effect of allowing objects only when they are
incorporated, which has the effect of creating a set of DP-events whose aspectual
structure is completive by virtue of the implications of the DP itself. This is why
the multiple event reading is available in addition to the result reading. The
aspectual structure in the lower Phase, however, cannot be accessed here, nor
can it be accessed in the non-incorporated plural cases:
(40)

a. *the grass-cuttings in an hour


b. *the cuttings of the grass in an hour

In sum, the two Phase analysis, based on the SMT and the Phase-head Phasecomplement distinction, which is linked in turn to a verbal clitic position allowing Verb+object to move as a unit, provides a syntactic analysis with independent
roots but which, as we have demonstrated, provides a semantic (interepretive)
analysis of subtle aspectual behavior in nominalizations.

How Phase-Based Interpretations Dictate the Typology of Nominalizations

3.

307

A Closer Look at German

The English nominal -ing of gerunds discussed in the various subsections of


Section 2 correspond to two different types of nominalizations in German: the
nominalized infinitive in -en and nominalizations in -ung.
(41)

a.

b.

(42)

a.

b.

das Spalten
des
Holzes
the split-en-inf of-the wood
the splitting of the wood
das Holzspalten
the wood-split-en-inf
the wood-splitting
die Spaltung des
Holzes
the split-ung of-the wood
the splitting of the wood
die Holzspaltung
the wood-split-ung
the wood-splitting

Just like their English counterparts, both types of nominalizations in German


can be incorporated and the distinction between the two types in German pattern
quite closely with the characteristic features of the two types distinguished above
for English.
Alexiadou, Iordachioaia and Soare (2009) point out in their analysis of argument supporting nominalizations that the German nominalized infinitive corresponds to English verbal gerunds. We argue that this is rather the structure
for those nominal gerunds, where the -ing affix is an aspectual affix and where
verbal functional projections below the nominalizing node are accessible for
modification by phrases adjoined by Later Merge. German -ung nominalizations, on the other hand, correspond to English nominal gerunds where the -ing
affix is a nominalizer that induces a Phase and thus blocks access to lower functional projections in the complement domain of the Phase head. We now explore
the correspondence in depth.
3.1. Nominalized Infinitives
Much like their English counterparts, nominalized infinitives in German can be
formed from transitive (cf. (43a)) and intransitive verbs (cf. (43b)) just as well
as from ditransitive (cf. (43c)) or reflexive verbs (cf. (43d)):

308
(43)

Leah Bauke and Tom Roeper

a.

b.

c.

d.

das Kussen
the kiss-en-inf
the kissing
das Laufen
the run-en-inf
the running
das Geben
the give-en-inf
the giving
das Rasieren
the shave-en-inf
the shaving

As far as the modificational properties are concerned, German nominalized infinitives again reflect the same pattern that has been outlined for English gerunds
that host an aspectual affix. Prepositional modifiers and non-sentential adverbial modifiers are licensed in incorporated and non-incorporated nominalized
infinitives in German:
(44)

a.

b.

(45)

a.

b.

das Mahen
des Rasens
mit einer Sense
the mow-en-inf the lawn-gen with a
scythe
the mowing of the lawn with a scythe
das Rasenmahen
mit einer Sense
the lawn-mow-en-inf with a
scythe
the lawn-mowing with a scythe
das Mahen
des Rasens
gestern /heute abend
the mow-en-inf the lawn-gen yesterday/this evening
the mowing of the lawn yesterday/this evening
gestern /heute abend
das Rasenmahen
the lawn-mow-en-inf yesterday/this evening
the lawn-mowing yesterday/this evening

Aspectual modifiers pattern in German nominalized infinitives also on a par


with their English -ing of counterparts.
(46)

a.

das Mahen
des Rasens
in zwei Stunden / fur
the mow-en-inf the lawn-gen in two hours
/ for
zwei Stunden
two hours
the mowing of the lawn in two hours/for two hours

How Phase-Based Interpretations Dictate the Typology of Nominalizations

b.

309

das Rasenmahen
fur zwei Stunden / *?in zwei
the lawn-mow-en-inf for two hours
/
in two
Stunden
hours
the lawn-mowing for two hours/in two hours

As has been observed above for nominal gerunds in English, in the incorporated
forms only atelic aspectual modifiers are licit. This follows from the fact that the
incorporated element is not quantized and thus does not license a telic reading
(cf. e.g. Borer 2005).
Another parallel between the German and English structures can be observed
when looking at control phenomena:
(47)

a.

b.

des Rasens.
Jacki bevorzugt das PROi Mahen
Jack prefers
the PRO mow-en-inf the lawn-gen
Jack prefers the mowing of the lawn.
Jacki bevorzugt das PROi Rasenmahen.
Jack prefers
the PRO lawn-mow-en-inf
Jack prefers (the) lawn-mowing.

Once again, nominalized infinitives in German pattern like nominal gerunds


with an aspectual -ing suffix in English. None of the German forms in (43)
(47) can be pluralized, which is what is expected as well.
Another interesting fact about the German structures is that progressive forms
in this language show the same affix that is used in the nominalized infinitives
(cf. Barrie 2006) and these progressive forms allow for incorporated and nonincorporated variants as well:
(48)

a.
b.

beim Mahen des Rasens


while mowing the lawn
beim Rasenmahen
while lawn-mowing

This in turn can be seen as yet another indication that German nominal infinitives host verbal functional structure below the N-node, and in particular an
aspectual projection in which the distinction not only between quantized and
non-quantized direct objects but also between progressive and non-progressive
is reflected.
As for the incorporated structures of the nominalized infinitives in German,
it is worth noticing that only accusative marked direct internal arguments can
be incorporated. External arguments cannot be incorporated:

310
(49)

Leah Bauke and Tom Roeper

a.

Der
The
b. das
the
c. *das
the

Mann lauft.
man runs
Laufen
des Mannes / des Mannes Laufen
run-en-inf the man-gen / the man-gen run-en-inf
Mannlaufen
man-run-en-inf

In the nominalized form the genitive marked external argument can precede or
follow the nominalization, but incorporation is not possible, this is only licit for
direct internal arguments such as Marathon in (50):
(50)

a.
b.

c.

Der Mann lauft einen Marathon.


the man runs a
marathon
das Laufen
des Marathons
(des Mannes)
the run-en-inf the marathon-gen the man-gen
the running of the marathon of/by the man
das Marathonlaufen
des Mannes
the marathon-run-en-inf the man-gen
the marathon-running of/by the man

Indirect internal arguments cannot be incorporated either, neither in German


nominalized infinitives nor in English -ing of nominal gerunds:
gibt dem Jungen das Geschenk.
Der Mann
the man-nom gives the boy-dat the present-acc

(51)

a.

(52)

b.

das
the
c. das
the
d. *das
the

(53)

a. The man promises the boy a present.


b. the promising of a present to the boy
c. the present-promising to the boy
d. *the boy-promising of the present

Geben
des Geschenks an den Jungen
give-en-inf the present-gen to the boy-acc
Geschenkgeben
an den Jungen
present-give-en-inf to the boy-acc
Jungegeben
des Geschenks
boy-give-en-inf the present-gen

In both languages the direct internal argument (which is marked for accusative
Case in German) can be incorporated into the nominal form, but the indirect
internal argument is not licensed as an incorporation under nominalization. This
is what is expected under dynamic antisymmetry in the sense of Moro (2000).

How Phase-Based Interpretations Dictate the Typology of Nominalizations

311

The verb and the internal argument are in a symmetric c-command relation
that violates the linear correspondence axiom of Kayne (1994). This symmetryrelation is resolved by complement specifier roll-up (cf. Barrie 2006) and thus
leads to LCA compliance at Spell-Out. External arguments and indirect internal
arguments, however, are not generated under symmetry, hence, they should not
be available for incorporation in the first place. The same logic applies to the
reflexive structures in (54):
(54)

a.
b.
c.
d.
e.

Der
the
das
the
*das
the
das
the
*das
the

Berg
spiegelt sich im
Wasser.
mountain reflects self in the water
Spiegeln
(des Berges)
im
Wasser
reflect-en-inf (the mountain-gen) in the water
Wasserspiegeln
water-reflect-en
Sich-Spiegeln
(des Berges)
im
Wasser
self-reflect-en-inf (the mountain-gen) in the water
Bergspiegeln
im
Wasser
mountain-reflect-en-inf in the water

In (54a) the verb spiegeln is a reflexive form that can be nominalized without
incorporation as in (54b).5 This reflexive form is the only one that is available
for incorporation and it blocks incorporation of any other arguments (cf. (54c)
and (54e)). Again this is the logical consequence from the reflexive being the
only constituent that is available for a symmetric c-command relation in the first
place. All other constituents are embedded under V asymmetrically and thus
need not move to a higher projection to break symmetry in order to be LCA
compliant upon Spell-Out.
Further evidence for this claim comes from the derived structure with particles in English, where no incorporation is possible if there is a particle, although
the thematic role remains a THEME:
(55)

a. apple-picking
b. *apple-picking up cf. [pick apples up]

Thus our account of this morphological operation fits the symmetry diagnostic
for movement in syntax.

5. Note that das Sich-Spiegeln des Berges im Wasser is also fine.

312

Leah Bauke and Tom Roeper

3.2. Nominalizations in -ung


In sharp contrast to the nominalized infinitives discussed in the preceding section
the formation of -ung nominals in German is much more restricted. Interestingly,
restrictions apply to all types of verbs:
(56)

a. *die
the
b. *die
the
c. *die
the
d. *?die
the

Kussung
kiss-ung
Laufung
run-ung
Gebung
give-ung
Rasierung
shave-ung

Regardless of whether the verb is transitive (56a), intransitive (56b), ditransitive


(56c) or reflexive (56d), nominalization is not possible, as the examples above
illustrate. While it still needs to be determined, what actually rules out the forms
in (56), it is to be noted that a simple blocking operation from result nominals
such as those in (57) does not qualify as a case in point:
(57)

a.
b.

der
the
der
the

Kuss
kiss
Lauf
run

While these forms, that might potentially block the forms in (56a) and (56b), do
exist, there are no zero derived result nominals for the forms in (58) that could
lead to blocking effects:
(58)

*die Singung, die Schreibung, die Liebung, die Jagung, die Gehung,
die Sehung, die Essung, die Kratzung, die Kommung, . . .

Transitivity on the other hand seems to be a vital criterion for the formation
of -ung nominals. All intransitive verbs lack this type of nominalization (while
they are fine as nominalized infinitives as has been shown in Section 3.1). As the
examples in (59) illustrate, however, transitivity is a necessary but not a sufficient
criterion for nominalization in -ung. The transitive verb schreiben cannot be
nominalized while the prefixed verbs beschreiben and ausschreiben can:
(59)

schreiben *die Schreibung


beschreiben die Beschreibung
ausschreiben die Ausschreibung

How Phase-Based Interpretations Dictate the Typology of Nominalizations

313

Quite remarkably though, whether the prefix can be stranded under nominalization or not does not play any role for the availability of the nominalized
form:
(60)

a.
b.

Er schreibt den Auftrag aus.


*Er ausschreibt den Auftrag.
Er beschreibt den Weg.
*Er schreibt den Weg be.

This transitivity sensitivity suggests that, much like English -ing of nominal
gerunds where the -ing affix is a nominalizing affix, verbal structure is involved
in the forms in (59). Again, paralleling the characteristics of their English counterparts and in stark contrast to German nominalized infinitives, German -ung
nominals can be pluralized:
(61)

a.

b.

(62)

a.

b.

die Beschreibungen des Weges


the describe-ung-pl the way-gen
directions
die Ausschreibungen des Auftrags
the bid-ung-pl
the contract-gen
the bidding of the contract
die Beschreibungen der Wege
the describe-ung-pl the way-gen-pl
directions
die Ausschreibungen der Auftrage
the bid-ung-pl
the contract-gen-pl
the bidding of the contracts

What the examples in (61) and (62) show is that in German the distinction between what has been identified as a result interpretation in English and that of
a plurality of individualized events depends on the nature of the direct object
as well and is determined here by whether the direct object is marked for plural or not. This distinction naturally gets lost under incorporation, because, as
expected, only one form is available for incorporation.
(63)

a.

b.

die Auftragsausschreibung /-en


the contract-gen bid-ung /-pl
the contract-bidding/s
die Wegbeschreibung /-en
the way describe-ung /-pl
the direction/s

314

Leah Bauke and Tom Roeper

However, the fact that the non-incorporated forms are sensitive to this distinction
is yet another indication that verbal structure is involved in these forms.
Not surprisingly, though, the -ung nominalizations do not license non-sentential adverbial modifiers or prepositional modifiers nor aspectual modifiers
neither in their incorporated nor in their non-incorporated variants:
(64)

a. *die
the
b. *die
the
c. *die
the
d. *die
the

Spaltung
splitt-ung
Spaltung
splitt-ung
Spaltung
splitt-ung
Spaltung
splitt-ung

des
the
des
the
des
the
des
the

(65)

a. *die
the
b. *die
the
c. *die
the
d. *die
the

Holzspaltung
wood-splitt-ung
Holzspaltung
wood-splitt-ung
Holzspaltung
wood-splitt-ung
Holzspaltung
wood-splitt-ung

Holzes
wood-gen
Holzes
wood-gen
Holzes
wood-gen
Holzes
wood-gen

gestern
yesterday
in zwei Tagen
in two days
fur zwei Tage
for two days
mit der Axt
with an axe

in zwei Tagen
in two days
fur zwei Tage
for two days
gestern
yesterday
mit der Axt
with an axe

Just like for the English -ing of nominal gerunds, this inaccessibility of the
lower verbal functional projections can be explained by the fact that the -ung
affix is generated under a category changing, nominalizing node on top of the
embedded functional structure that induces a Phase.Adjunction understood as an
operation of Later Merge is thus not possible for structures that are not located on
the Phase Edge. Since the functional projections are in the complement domain
of the nominalizing node, these will thus not be accessible. Interestingly, -ung
nominalizations can be interpreted reflexively only via a PROarb reading (cf.
also Sichel 2009), which can be seen as yet another indication that a Phase is
involved which cannot be accessed after nominalization:
(66)

a.
b.

die
the
das
the

Anmeldung
register-ung
Anmelden
register-en

der
the-gen
der
the-gen

Gaste
guests
Gaste
guests

How Phase-Based Interpretations Dictate the Typology of Nominalizations

(67)

a. *die
the
b. das
the

Sich-Anmeldung
self-register-ung
Sich-Anmelden
self-register-en

der
the-gen
der
the-gen

315

Gaste
guests
Gaste
guests

The ungrammaticality of the form in (66a) is even more surprising and significant, when taking into account that incorporation in -ung nominals is not
limited to direct internal arguments. In fact, indirect arguments of ditransitives
or reflexives and modifiers can be incorporated:
(68)

a.
b.

(69)

a.

b.
(70)

a.
b.

die Stadtfuhrung
city-guide-ung
Er fuhrt sie durch die Stadt.
He guides them through the city.
die Wasserspiegelung (*des Berges) /?des Berges
the water-reflect-ung (the mountain-gen)
Wasserspiegelung
Der Berg
spiegelt sich im
Wasser.
the mountain reflects self in the water
die
the
Der
the

Flussbiegung
river-bend-ung
Fluss biegt sich.
river bends self

As the examples in (67)(69) illustrate, German -ung nominals can incorporate


more freely than their nominalized infinitive counterparts. However, incorporation of an element that is not a direct internal argument is only possible, if
the incorporated element is a first sister of the incorporating head (as argued
in Roeper (1978), Harley (2009)). Any intervening direct internal argument is
illicit:
(71)

a.
b.
c.

Er fuhrt die Touristen durch die Stadt.


He guides the tourists through town
durch
die Stadt
seine Fuhrung der Touristen
his
guide-ung the tourists-gen through town
seine Stadtfuhrung
(*der Touristen)
his
town-guide-ung (the tourists-gen)

If the direct internal argument tourists is present in the incorporated nominal,


the structure is ungrammatical. Only in the absence of that argument can the
PP argument realized as town be incorporated into the -ung nominal. The same

316

Leah Bauke and Tom Roeper

principle is at play in (69) and most likely also in (70), where the argument Fluss
is an internal argument of the unaccusative verb bend. Thus, the examples show
that for incorporation there still needs to be a symmetric c-command relation
between the incorporated element and the nominalized form. Intervening arguments that are closer to the -ung form block incorporation if they are realized
overtly. This is perfectly in line with the motivation for incorporation being a
symmetry relation (cf. e.g. Moro 2000).
In contrast to the relatively unconstrained incorporation possibilities illustrated in (68)(71) there is a relatively strict limitation on the incorporated
element. Similar to the pattern found in nominal root compounds (cf. Bauke
2009), the incorporated element in -ung nominals must be specified for plural,
for genitive case or it must be a bare stem. Nominalized infinitives, on the other
hand, are much less restrictive here:
(72)

a.

die
the
b. *die
the
c. *die
the

Straenkreuzung
street-pl-cross-ung
Straekreuzung
street-sg-cross-ung
Strakreuzung
street-stem-cross-ung

(73)

a.

Straekreuzen
street-sg-cross-en-inf
Straenkreuzen
street-pl-cross-en-inf

b.

das
the
das
the

There is an interpretational difference between the forms in (73a) and (72b).


The former refers to an event of crossing one (or more) street(s), while the latter indicates an event of crossing more than one street. This distinction cannot
be captured in (72), where the only grammatical form available leads to a result interpretation. This is what is expected under a Phase-based approach. Only
nominalizations in -ung involve a higher Phase. Inflectional marking must therefore be fixed before the DP Phase level, i.e. before incorporation, because the
Edge Property will once again block access at the lower level. In the nominalized
infinitives a lower Phase is involved.
4.

Consequences

It has been argued that two types of nominal -ing of gerunds need to be distinguished in English. Both of these types project verbal functional structure on

How Phase-Based Interpretations Dictate the Typology of Nominalizations

317

top of a V-node, and in particular an aspectual projection. The distinction between the two types of gerunds identified rests on the projection site for the -ing
affix. When this affix is projected under the aspectual head licensing of aspectual modifiers, of non-sentential adverbial and prepositional modifiers, of the
anaphor do-so and of a PRO-element is what is expected and attested. Sentential modifiers are not licensed, which can likewise be explained by the absence
of a licensing TP or higher projection. Naturally, nominal gerunds of this type
cannot be pluralized either, because the -ing suffix is not a nominal affix and
thus unable to host nominal inflectional morphology.
When the -ing affix is projected under the nominal node instead, plural morphology is licensed. This does not mean, however, that this type of nominal
gerund does not project verbal functional structure below the nominalizing
node. On the contrary, the sensitivity to the quantized nature of the nominalized
verbs internal argument the sensitivity to non-incorporated THEME DPs and
the distinction between Kind- vs. Specific-interpretations point into the opposite
direction. This functional structure is not accessible for those modifiers that are
licensed with the other type of nominal gerund, however, because the nominalizing node is a Phase boundary whose Edge Property blocks accessibility of
projections in the complement domain of the Phase-head.
Both types of nominal gerunds allow for incorporation structures and the distinction between the incorporated and non-incorporated forms of the respective
types of nominal gerunds reduces to a symmetry-distinction. If the direct internal argument is licensed in an of-clause the nominalized verb and its argument
enter into an asymmetric c-command relation and incorporation is blocked. In
the absence of such a projection, a bare internal argument enters into a symmetric c-command relation that is dissolved by moving this argument from the
complement to a higher specifier position.
In sum, this paper follows the tradition that syntactic principles should apply
in the lexicon. We have argued that the abstract notion of Phase and the SMT
in fact predict exactly where subtle interpretive differences linked to aspect can
occur. It is precisely the ability of a theory to predict seemingly peripheral data
which illustrates its strength.

318

Leah Bauke and Tom Roeper

References
Abney, Steven Paul
1987
The English Noun Phrase in Its Sentential Aspect. Doctoral Dissertation. Massachusetts Institute of Technology.
Artemnis, Alexiadou, Gianna Iordachioaia and Elena Sorare
2009
Plural marking in argument supporting nominalizations. In: P. Cabredi-Hofherr and B. Laca (eds.), Layers of Aspect. Stanford, CA: CSLI
Publications.
Barrie, Michael
2006
Dynamic antisymmetry and the syntax of noun incorporation. PhD
dissertation. University of Toronto.
Bauke, Leah
2009
Nominal root compounds in German. Ms. University of Massachusetts, Amherst.
Boeckx, Cedric
2008
On the locus of asymmetry in UG. Ms. Harvard University.
Boeckx, Cedric
2008a
Bare Syntax. Oxford: Oxford University Press.
Borer, Hagit
2005
Structuring Sense Vol I+II. Oxford: Oxford University Press.
Chomsky, Noam
1970
Remarks on nominalization. In: Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English Transformational Grammar, 184
221. Waltham, MA: Georgetown University Press.
Chomsky, Noam
1993
The Minimalist Program. Cambridge, MA: MIT Press.
Chomsky, Noam
1995
Bare Phrase Structure. In: Gert Webelhuth (ed.), Government and
Binding and the Minimalist Program, 383440. Cambridge, MA:
Basil Blackwell.
Chomsky, Noam
1995
Minimalist inquiries: the framework. In: Roger Martin, David
Micheals and Juan Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax, 89155. Cambridge, MA: MIT Press.
Chomsky, Noam
2001
Derivation by Phase. In: M. Kenstowicz (ed.), Ken Hale: A life in
language, 152. Cambridge, MA: MIT Press.
Chomsky, Noam
2008
On Phases. In: R. Freidin, C. Otero and M.-L. Zubizarreta (eds.),
Foundational Issues in Linguistic Theory, 133166. Cambridge, MA:
MIT Press.

How Phase-Based Interpretations Dictate the Typology of Nominalizations

319

Fu, Jinqui, Thomas Roeper and Hagit Borer


2001
The VP within process nominals: evidence from adverbs and the VP
anaphor do-so. Natural Language and Linguistic Theory 19(3): 549
582.
Grimshaw, Jane
1990
Argument Structure. Cambridge, MA: MIT Press.
Harley, Heidi and Rolf Noyer
1998
Licensing in the non-lexicalist lexicon: nominalizations, vocabulary
items and the encyclopaedi. In: Heidi Harley (ed.), Papers from the
University of Pennsylvania/MIT Roundtable on Argument Structure
and Aspect, 11938 (MIT Working Papers in Linguistics 32). Cambridge, MA: MIT Press.
Harley, Heidi
2009
Merge conflation and head movement. In: Proceedings of the North
Eastern Linguistic Society 34: 23954. Amherst MA: GSLA Publications.
Iordachioaia, Gianna and Elena Sorare
2008
Two kinds of event plurals: evidence from Romanian nominalizations.
In: O. Bonami and P. Cabredo Hoffher (eds.), Empirical issues in
syntax and semantics 7: 193216. (http://www.cssp.cnrs.fr/eiss7).
Kayne, Richard
1994
The antisymmetry of syntax. Cambridge, MA: MIT Press.
Keyser, Samuel J. and Thomas Roeper
1992
Re: The abstract clitic hypothesis. Linguistic Inquiry 23: 89125.
Keyser, Samuel J. and Thomas Roeper
1997
Anti-symmetry and leftward movement in morphology. Ms. MIT and
UMass.
Kratzer, Angelika
1994
Severing the External Argument from its Verb. In: J. Rooryck and
L. Zaring (eds.), Phrase Structure and the Lexicon, 109137. Dordrecht: Kluwer.
Lebaux, David
1988
Language Acquisition and the Form of Grammar. Doctoral dissertation. University of Massachusetts, Amherst. Reprinted 2000. Philadelphia/Amsterdam: John Benjamins Publishing Company.
Marantz, Alec
1997
No Escape from Syntax: Dont Try Morphological Analysis in the
Privacy of Your Own Lexicon. University of Pennsylvania Working
Papers in Linguistics: Proceedings of the 21st Annual Penn Linguistics
Colloquium 4(2): 20125.
Marantz, Alec
2001
Words. Ms. MIT.

320

Leah Bauke and Tom Roeper

Marantz, Alec
2007
Phases and Words. Ms. NYU.
Moro, Andrea
2000
Dynamic Antisymmetry. Cambridge, MA: MIT Press.
Roeper, Thomas and Angeliek Van Hout
1999
The impact of passive -able and middle: Burzios generalization and
feature-movement in the lexicon. In: Liina Pylkkanen, Angeliek van
Hout and Heidi Harley (eds.), Papers from the University of Pennsylvania/MIT Roundtable on the Lexicon, 185211 (MIT Working
Papers in Linguistics 35). Cambridge, MA: MIT Press.
Roeper, Thomas, William Snyder and Kazuko Hiramatsu
2002
Learnability in a Minimalist framework: Root compounds, merger,
and the syntax-morphology interface. In: I. Lasser (ed.), The process
of language acquisition. Frankfurt: Peter Lang.
Roeper, Thomas and William Snyder
2005
Language learnability and the forms of recursion. In: A. M. Di Sciullo (ed.), UG and external systems, 155169. Amsterdam: John Benjamins.
Roodenburg, Jasper
2006
The role of Number within Nominal Arguments. Rutgers University:
LSRL 36.
Sichel, Ivy
2009
New evidence for the structural realization of the implicit external
argument in nominalizations. Linguistic Inquiry 40: 712723.
Siegel, Julia
1997
Gerundive Nominals and the Role of Aspect. In: J. Austin and A. Lawson (eds.), Proceedings of the Fourteenth Eastern States Conference
on Linguistics, 170179. Ithaca: CLC Publications.
Van Hout, Angeliek and Thomas Roeper
1998
Events and aspectual structure in derivational morphology. In:
Heidi Harley (ed.) Papers from the University of Pennsylvania/MIT
Roundtable on Argument Structure and Aspect, 175200 (MIT Working Papers in Linguistics 32). Cambridge, MA: MIT Press.
Williams, Edwin
1981
Argument structure and morphology. The Linguistic Review 1: 81
114.

Scope and Verb Meanings*


Edwin Williams

Many prepositions show scopal interaction with quantifiers like only but no verb
does:
(1)

a.
b.

John can box only with his left arm. =/ John can box with only his
left arm.
John only lost / sold /deified / reified /reanimated /transmogrified
his left arm.
= John lost /sold /deified /reified / reanimated / transmogrified only
his left arm.

Since there are thousands of verbs this is a curious finding but I believe there is
a good explanation of it.
The Representation Theory (RT) model of the grammatical system embeds
direct objects and complement clauses under different regimes for a variety of
reasons which are detailed in Williams (2003) and listed in Section 4 below.
The clausal regime is called Level Embedding (under the Level Embedding
Conjecture) and the direct object regime is called Cogeneration. But in that
earlier work I did not specify the regime for other phrase types. Here I will
suggest that Prepositional Phrases are embedded under the Level Embedding
regime, and I will support that proposal with predictions of the scopal behavior