Culicover - Natural Language Syntax PDF

   
Natural Language Syntax

O T  L
General editors: Keith Brown, University of Cambridge; Eve V. Clark, Stanford University;
April McMahon, University of Edinburgh; Jim Miller, University of Edinburgh;
Lesley Milroy, University of Michigan
PUBLISHED
The Grammar of Words

An Introduction to Linguistic Morphology
Second edition
by Geert Booij
A Practical Introduction to Phonetics

Second edition
by J. C. Catford
Meaning in Language
An Introduction to Semantics and Pragmatics
Second edition
by Alan Cruse

by Peter W. Culicover
Principles and Parameters

An Introduction to Syntactic Theory
by Peter W. Culicover
A Semantic Approach to English Grammar

by R. M. W. Dixon
Semantic Analysis: A Practical Introduction

by Cliff Goddard
Pragmatics
by Yan Huang
Diachronic Syntax
by Ian Roberts
Cognitive Grammar: An Introduction

by John R. Taylor
Linguistic Categorization
Third edition
by John R. Taylor
IN PREPARATION
Translation: Theory and Practice

by Kirsten Malmkjaer
Peter W. Culicover
1
3
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trademark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Peter W. Culicover 2009
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2009
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
CPI Antony Rowe, Chippenham, Wiltshire
ISBN 978–0–19–923017–4 (Hbk.)

ISBN 978–0–19–923018–1 (Pbk.)
1 3 5 7 9 10 8 6 4 2
Contents
Preface xi
Acknowledgments xiii
List of Abbreviations xv
1. Overview 1
1.1. What is syntax? 1
1.2. The goals of linguistic theory 3
1.3. Where does syntactic theory fit in? 6
1.4. Simpler Syntax 7
2. Syntactic Categories 11
2.1. Traditional categories 11
2.1.1. Nouns 12
2.1.2. Verbs 19
2.1.3. Adjectives 22
2.1.4. Prepositions 24
2.1.5. Adverbs 25
2.1.6. Minor categories 27
2.2. Morphosyntax 28
2.2.1. Words and lexical items 28
2.2.2. The structure of the lexicon 30
2.2.3. Paradigms 34
2.2.4. More morphosyntactic properties 36
2.3. Heads and phrases 43
2.4. ∗ The theory of linguistic categories 44
2.4.1. Justifying categories 44
2.4.2. Universal categories 46
2.4.3. Tests for categories 48
2.4.4. A paradox resolved? 50
Exercises 52
Problems 56
Research questions 58
vi CONTENTS
3. Basic sentential structure 61

3.1. Methodological preliminaries 61
3.2. The simple sentence 64
3.3. Complements, arguments, and adjuncts 68
3.4. Grammatical functions 70
3.4.1. Structural grammatical functions 70
3.4.2. Tests for subject 71
3.5. Marking grammatical functions 72
3.5.1. Case 72
3.5.2. Case-marking patterns 74
3.5.3. Agreement 77
∗
3.6. Tests for constituency 79
3.6.1. Ellipsis 80
3.6.2. Proform replacement 81
3.6.3. Coordination 83
3.6.4. Displacement 84
Exercises 93
Problems 98
4. Phrasal Categories 103

4.1. X theory 103
4.2. The structure of the verb phrase 106
4.3. The structure of the noun phrase 109
4.4. Other phrasal categories 111
4.5. The English verbal sequence 112
4.5.1. Auxiliary verbs 112
4.5.2. Some generalizations 117
4.5.3. Accounting for the sequence 118
4.5.4. Have and be 121
4.6. Rule summary 122
4.7. ∗ Applications of strong X theory 123
4.7.1. IP and CP 123
4.7.2. DP 125
4.7.3. VP internal subjects 128
Exercises 129
Problems 131
CONTENTS vii
5. Conceptual structure and the lexicon 139

5.1. Overview 139
5.2. Correspondences 140
5.2.1. Concepts 140
5.2.2. Indices 143
5.2.3. Lexical entries 144
5.3. CS relations 146
5.4. Thematic roles and linking 147
5.4.1. Thematic structure 147
5.4.2. Linking to syntactic structure 150
5.5. Linking hierarchies 154
5.5.1. Intransitives 154
5.5.2. Oblique arguments 155
5.6. Computing correspondences 159
5.7. Selection 169
5.7.1. S-selection 169
5.7.2. C-selection and the theta criterion 172
5.8. ∗ Case 173
5.9. ∗ Modification 176
Exercises 178
Problems 180
6. Argument Correspondences 185

6.1. Canonical argument correspondences 185
6.2. Passive 185
6.2.1. Passive relations 185
6.2.2. Passive constructions 187
6.3. Applicatives and the dative alternation 191
6.4. Causative 195
6.5. Antipassive 197
6.6. Dummy subjects 199
6.7. ∗ Null pronouns and clitics 202
6.7.1. Null pronouns 202
6.7.2. Clitics 207
6.8. ∗ The transformational analysis of passive 209
6.8.1. Background 209
6.8.2. The classical analysis 210
viii CONTENTS
6.8.3. Structure preserving movement 212

6.8.4. Why passive? 214
6.8.5. Passives without movement 216
6.9. ∗ Theta criterion, EPP, and UTAH 221
Exercises 224
Problems 228
7. Complex clauses: raising and control 237

7.1. Infinitival complements 237
7.1.1. Subject control 239
7.1.2. Raising to subject 242
7.1.3. Object control 246
7.1.4. “Raising” to object 248
7.1.5. The case of expect 251
7.1.6. Gerundives 252
7.1.7. Summary: raising and control 253
∗
7.2. More correspondences 253
7.3. ∗ Raising as movement 258
7.3.1. Move NP in MGG 258
7.3.2. More raisings 260
7.3.3. Interactions of raising, passive, and control 264
7.4. ∗ Syntactic configuration and control 267
7.4.1. Uniformity 267
7.4.2. Case and PRO 269
7.4.3. Identifying the controller: c-command and MDP 274
7.4.4. Problems with MDP 277
7.4.5. The lexical representation of control 278
Exercises 281
Problems 285
8. Predication 299
8.1. Secondary predication 299
8.1.1. Predicates and antecedents 299
8.1.2. Predication and control 300
8.1.3. Resultative predicates 301
8.1.4. Correspondences 303
8.2. ∗ Small clauses 305
8.3. ∗ Secondary predication cross-linguistically 310
CONTENTS ix
Problems 315
9. A constructions 321
9.1. Questions 322
9.2. Types of wh-questions 329
9.2.1. Piedpiping and preposition stranding 329
9.2.2. In situ wh-questions 331
9.2.3. English wh-in-situ 335
9.2.4. Multiple wh-questions 337
9.3. Relative clauses 339
9.3.1. Relatives with gaps 339
9.3.2. Piedpiping in relative clauses 342
9.4. Constraints on chains 344
9.5. ∗ The theory of wh-movement 349
9.5.1. Basics of wh-movement 350
9.5.2. Feature discharge 353
9.5.3. Covert movement 355
9.5.4. Movement in relative clauses 358
∗
9.6. Topicalization 361
9.6.1. Basic structure 361
9.6.2. Topicalization as movement 362
∗
9.7. More on Constraints 365
9.7.1. Conditions and Barriers 365
9.7.2. Violability of constraints 369
9.8. Other A constructions 373
∗
9.8.1. Questions 373

9.8.2. Relatives 376
9.8.3. Clefts and pseudo-clefts 379
9.9. Summary 381
Exercises 382
Problems 386
10. Coreference and Binding 401

10.1. Coreference 401
10.2. Binding 404
10.2.1. Bound anaphors 404
10.2.2. Bound pronouns 407
10.2.3. Condition C 407
x CONTENTS
10.3. Quantification 409

10.4. ∗ Binding in CS and syntactic structure 411
10.4.1. The GB binding theory 411
10.4.2. CS- and GF-binding 414
10.4.3. Long distance anaphora 420
10.5. ∗ Reconstruction 422
10.5.1. A constructions and binding 422
10.5.2. The copy theory of movement 425
10.6. ∗ Crossover and anti-reconstruction 426
10.7. Summary 428
Exercises 428
Problems 430
11. Fragments 437

11.1. Bare argument ellipsis 439
11.1.1. Two approaches 439
11.1.2. Problems for a syntactic account of Bare
Argument Ellipsis 442
11.1.3. Reasons to believe syntax is involved in BAE 446
11.1.4. A resolution: indirect licensing 448
11.2. VP ellipsis and related constructions 450
11.2.1. The syntax of VP ellipsis 451
11.2.2. VP anaphora 454
11.2.3. The interpretation of ellipsis 456
11.3. Gapping 460
11.3.1. Basic gapping facts 460
11.3.2. Why gapping cannot be syntactic deletion 462
11.4. Summary 465
Exercises 465
Problems 466
Glossary 471
References 477
Index 485
Preface
The aim of this book is to provide an introduction to the study of natural

language syntax. Syntax is concerned primarily with how languages con-
figure strings of words and morphemes into sentences in order to express
meanings. Consequently there are two major foci that are developed hand
in hand in this book: (i) the syntactic and morphosyntactic devices that
languages use, and (ii) the conceptual structures that correspond to partic-
ular aspects of linguistic form. Not only are the forms emphasized but their
correspondences with meanings are.
The book is mainly about “how language works”, and what a syntactic
theory has to do in order to be able to account for how language works. But,
realistically, an introduction to syntax that focuses on how language works
has to take account of the profound influence of mainstream generative
grammar (MGG), that is, the Chomskyan tradition leading from Syntactic
Structures all the way to the Minimalist Program. So I have organized
the chapters accordingly. The beginning sections of each chapter work
through a range of descriptive issues, using a fairly neutral non-derivational
approach to isolate key syntactic and morphosyntactic properties and spec-
ify how they contribute to interpretation. The theoretical underpinnings of
this approach are spelled out in Simpler Syntax, which Ray Jackendoff and
I published in 2005. The later sections of each chapter, those starred with
an asterisk (∗ ), explore various theoretical issues, with a focus on evaluating
how MGG, using such devices as movement, deletion, and functional heads,
seeks to captures the correspondences between form and meaning that we
find in natural languages and that any syntactic theory has to account for.
I have organized the book so that it can be used for an undergraduate
or a graduate introduction to syntax. For an undergraduate introduction,
it is possible to work through just the unstarred sections, with perhaps an
occasional foray into a more technical starred section if student interest
warrants it. For students at this level, who may have no prior familiarity
with doing syntax, I have provided a number of Exercises and Problems at
the end of each chapter. The Exercises are for the most part intended to help
xii PREFACE
the student become comfortable with the technical aspects of describing the
structure of sentences of a natural language and their basic meanings. The
Problems are somewhat more challenging. Most of these are data-oriented,
requiring that the student identify some pattern in data, use data to support
or falsify a claim, or develop original data to support or falsify a claim.
For the graduate introduction, it is reasonable to presuppose that most
students have familiarity with much of the material in the unstarred sec-
tions. But the backgrounds of students can often be uneven. Thus, it can be
useful to ask students to read the unstarred sections either as a review or
to fill in whatever gaps there might be. The starred sections presuppose the
descriptive material and go into theoretical questions. There are two main
objectives of these starred sections: (i) to sketch out the essential concepts
and methods of mainstream syntactic theory, and (ii) to evaluate the ade-
quacy of this approach. Some of the Problems and Research questions at
the end of the chapters are designed to get more advanced students thinking
critically about these issues, and working out possible solutions. In addition,
many of the Research questions point the student to phenomena that are
not addressed in the text; these are for the most part open-ended questions
that may stimulate a student’s interest in research on syntactic issues beyond
the introductory course.
Many of the examples that are used in this book to illustrate various
technical points are drawn from English. At the same time, I have aimed at a
broad comparative perspective where that is practical, using data and analy-
ses from languages other than English. In presenting such data I sought
to maintain as consistent and transparent a glossing approach as possible,
based on the Leipzig Glossing Rules. In many cases, therefore, I have re-
glossed examples cited from the literature, where the Leipzig Glossing Rules
have not been followed.
Acknowledgments
I am deeply indebted to many people who have contributed to the design,

organization, and content of this book. There is no question that much of
what is good and useful in the book as it currently appears is due to their
influence.
My first and most profound debt is to Ray Jackendoff. Ray and I have
worked for many years together on research that culminated in Simpler
Syntax, and it is because of Simpler Syntax that I decided to write another
introductory syntax text. Simpler Syntax reflects our feeling that it is time
to go beyond mainstream approaches to syntactic analysis. The current text
reflects our view that it is important to teach students to view syntactic
phenomena from a Simpler Syntax perspective, too, and to contrast it with
other ways of capturing the relationship between form and meaning that
are found in the literature. But it was a considerable challenge to see how
to convey the Simpler Syntax perspective while at the same time providing
the student with the bigger picture, in which Simpler Syntax is part of an
inquiry into the proper form of a syntactic theory. Ray read every page of
the manuscript and made literally hundreds of suggestions about wording,
presentation, organization, and notation that have helped me get a lot closer
to these goals.
I am also deeply grateful to the Alexander von Humboldt Foundation
for a Research Award that made it possible for me to spend a year at the
University of Tübingen during 2006–2007, during which time I completed
and made initial revisions of the manuscript. I owe a debt of gratitude to
Erhard Hinrichs and Marga Reis for nominating me for the Humboldt
Award and for being my hosts while I was in Tübingen. And my deepest
thanks go to Dean John Roberts of the College of Humanities at Ohio State
for making it possible for me to take fullest advantage of the Humboldt
Award.
While I was in Tübingen a wonderful group led by Susanne Winkler
offered to work through a draft of the text with me and provide me with
comments and criticisms. I thank Susanne for her extensive and enormously
xiv ACKNOWLEDGMENTS
helpful comments on all of the chapters, and Melanie Henschke, Nora

Kaltenbach, Andreas Konietzko, Katharina Schmeh, and Andreas Wurtz
for their incisive questions, suggestions, and observations. These have led to
numerous substantial improvements.
After returning to Ohio State I was fortunate to be able to use an earlier
draft of this text in both an undergraduate course and a graduate course
during the same academic quarter. I want to thank the students in these
courses for their limitless patience with the challenges posed by the draft,
for their suggestions on how to improve it, for their ingenuity and good
nature in trying to make sense of often obscure formulations, and for their
questions, which were always useful and insightful.
Thanks also go to several anonymous reviewers for Oxford University
Press for their penetrating and constructive criticism of an early draft. One
reviewer in particular was very helpful in focusing on how I address mor-
phological structure and glossing, and I very much appreciate the criticism
and the advice. Thanks to Anastasia Smirnova, for checking references and
citations, and Jon Dehdari, for correcting typos and formatting inconsisten-
cies. And finally, thanks to my editor, John Davey, who as always has shown
unfailing patience and good sense as we work towards completion of yet
another project.
List of Abbreviations
A argument
A non-argument
ABS absolutive (case)
ACC accusative (case)
ADJ adjective
ADV adverb
AP adjective phrase
arb arbitrary (reference)
ART article (e.g. the)
AUX auxiliary
AVM attribute value matrix
C complementizer
CAT category
Comp complement
CONJ conjunction (e.g. and)
CP complementizer phrase
CS conceptual structure
DAT dative (case)
Deg degree (e.g. very)
DEM demonstrative (e.g. this)
DET determiner
DIR direction
DP determiner phrase
DU dual
e empty category
EPP extended projection principle
ERG ergative (case)
GB government binding (theory)
GEN genitive (case)
GF grammatical function
HPSG head-driven phrase structure grammar
IND indefinite
xvi LIST OF ABBREVIATIONS
INSTR instrumental (case)

INTR intransitive
IO indirect object
IP inflection phrase
IS information structure
LF logical form
LFG lexical-functional grammar
LOC location
M modal (e.g. could)
MASC masculine
MDP minimum distance principle
N noun
NOM nominative (case)
NP noun phrase
NPI negative polarity item (e.g. any)
O object
OBL oblique (case)
OP empty operator (in syntax)
P preposition
PART participle
PASS passive
PF phonetic form
PL plural (number)
POSS possessive
PP prepositional phrase
PREP prepositional (case)
PRES present tense
PROG progressive
PSR phrase structure rule
Q interrogative operator (in CS)
Q quantifier (e.g. every)
REL relative (feature)
S sentence
SG singular (number)
Spec specifier
SU subject
t trace
LIST OF ABBREVIATIONS xvii
TOP topic (feature)

TR transitive
UG Universal Grammar
UTAH Uniform thematic assignment hypothesis
V verb
VAUX auxiliary verb
VP verb phrase
VPISH VP internal subject hypothesis
WH interrogative (feature)
X X-bar
Ë theta
This page intentionally left blank
1
Overview
1.1. What is syntax?
The conventional answer to the question “What is syntax?” in theoretical

linguistics is something along the lines of, “It is the system that governs the
relationship between form and meaning in a language”. What this actually
means in practice is something that you will appreciate in some depth as you
work through this book, but at this point it is likely to be rather obscure. So
let us start with a concrete example.
Consider the notion of form. An expression in a language can be
described in a number of ways, all of which are valid. For example, the
expression biting dogs can be described as a string of sounds, which is its
“form” in the most concrete sense.
(1) [baytiNdOgz]
This form is often called phonetic form, or PF.

More abstractly, this expression can be described as a string of mor-
phemes –
(2) bite + -ing + dog + -s
– or as a string of words –
(3) biting dogs
– or as a phrase of the category noun phrase (NP) consisting of a sequence

of categories –
(4) [NP [V biting ] [N dogs ]].
For any given language, the particular way in which the categories may or
must be sequenced determines how the words will be ordered, which in turn
determines how the morphemes will be ordered, which in turn determines
2 1. OVERVIEW
its phonetic form, that is, how the sounds will be ordered sequentially in
time. The information given in (4), consisting of the categories of the words
and phrases and the ordering of words and phrases, falls within the domain
of syntax. So we see how syntax bears on the form. If one word or phrase
X precedes another word or phrase Y, then the sounds of X will precede the
sounds of Y.
Now consider the meaning. The string biting dogs has two meanings, one
in which the dogs bite (Biting dogs also bark a lot), and the other in which
they are bitten (Biting dogs is not much fun). So we have to provide two
semantic descriptions for this string of words. In doing so, we relate the same
concrete form to different meanings.
Part of the job of syntax is to provide enough information so that, given
a string and a syntactic description, it is possible to explain all of the
meanings of the string. This information, which has to do with categories
and phrasing, is abstract. By “abstract” we mean that it is invisible, in the
sense that we cannot see it or hear it. It does not correspond to anything
concrete in the string of sounds, or even in the string of morphemes and
words. The syntactic description of an expression concerns the categories
of the words, how the words are grouped into phrases, the categories of
the phrases, how they are grouped together, and perhaps invisible elements
that contribute to the meaning but not to the form. This description is a
syntactic structure.
So a phrase of a language, even a very simple phrase consisting of a
single word, has a phonetic form, a meaning, and a syntactic structure that
mediates between them. We will call this triple of a form, a meaning, and a
syntactic description a correspondence. The sum total of all of the syntactic
structures of a language comprises the syntax of the language. It is part of
what we know when we know a language.
Here is a simplified example of such a triple for the word bite.
(5) bites
 
FORM [bayt] 
 
 CATEGORY V 
  
SYNTAX  NUMBER SINGULAR  
 
 PERSON 3RD 
MEANING BITE
We use the capitalized boldface BITE here to symbolize the meaning of the
word bite.
1.2. THE GOALS OF LINGUISTIC THEORY 3
Owing to the fact that the number of possible correspondences in a

language is infinite, it is not possible to list all of them. Moreover, it would
not be particularly insightful to do so, even if it was possible, since there are
many regularities in a natural language that can and should be captured in
our description. These regularities, and not the individual correspondences
themselves, are what a native speaker of a language knows.
A grammar is a description of a language that makes explicit the knowl-
edge that speakers of a language have about their language. Part of what
speakers know is how to form phrases of each category, and how to form
sentences. For example, English speakers know that the word the precedes
the other words in a noun phrase, and does not follow them –
(6) a. the biting dogs (are vicious)

b. ∗ biting dogs the (are vicious)
The symbol “∗ ” before a string of words as in (6b) indicates that it is

not a possible sentence or phrase in the language under discussion – it is
ungrammatical in the language. The expression in (6a), on the other hand,
is a possible phrase in the language – that is, it is grammatical.
The grammatical knowledge in this case is general. It is not specific to
individual words, like the, biting, and dogs. It holds for all words of a
category. For example, not only must the word the precede dogs, as shown
in (6), but so must a, all, these, some, and most, all members of the category
determiner. And these words must precede any word of the same category as
dogs, such as cats, pigs, ambulances, ideas, etc., all members of the category
noun. In describing the knowledge of an English speaker, we must say that
the determiner precedes the noun.
1.2. The goals of linguistic theory1
In our study of syntax we are not simply interested in the description of a

particular language, or even in the description of a collection of languages.
Our overriding concern is to understand what the properties of human
languages are, and why they are that way. We informally refer to this
notion as How language works and why. Thus, our particular descriptions
1
Parts of this section are adapted from Culicover and Jackendoff 2005:
Chapter 1.
4 1. OVERVIEW
of particular phenomena are of fundamental importance, but are in

service of the broader objective. This perspective, one that relates the
description of particular languages to the theoretical question of how lan-
guage works and why, is central to generative grammar, as distinguished
from purely descriptive linguistics.
Generative grammar takes as central the view that what we are studying
is the instantiation of language in the human mind/brain, rather than an
abstract phenomenon that exists “in the community”, in a collection of
texts, or in some sort of Platonic space. The fundamental linguistic phenom-
enon is a speaker who produces an utterance that is understood by a hearer;
the fundamental question is what knowledge is present in the speaker’s and
hearer’s mind/brain that enables this interchange to take place? A language
exists “in the community” insofar as there is a community of speakers able
to participate equivalently as speakers or hearers. In other words, generative
grammar seeks a mentalistic account of language.
Unlike vocal communication systems in other primates, human language
is not limited to a relatively small number of isolated signals. Rather, a
speaker of a human language can create and understand an unlimited
number of different utterances, concerning an unlimited number of different
topics. This entails that a language user with a finite brain must have
a productive system for constructing new utterances (in both production
and perception) from finite knowledge. Crucially, this productive system is
unconscious knowledge. It is like the principles by which the visual system
constructs perception of the physical world, not like one’s knowledge of the
rules of a game or traffic laws.
It has been customary since Chomsky’s Aspects of the Theory of Syntax
(1965) to make a distinction between linguistic competence – the language
user’s knowledge of his or her language, and linguistic performance – the
processing strategies by which this knowledge is put to use. The theory of
competence is the linguist’s idealization of what the speaker’s knowledge
consists of, what we referred to above as the grammar. The goal of linguistic
description in generative grammar is to provide an account of the linguistic
competence of the native speakers of a language under investigation that as
accurately as possible accounts for the form/meaning correspondences of
that language.
The term “grammar” is conventionally used to refer both to the linguist’s
description of competence and the actual competence that is in the native
speaker’s head. The reasonable presumption is that the linguistic description
1.2. THE GOALS OF LINGUISTIC THEORY 5
of native speaker competence corresponds in some interesting way to the

knowledge that the native speaker has in his/her head.
To reiterate, generative grammar is a mentalistic theory. It is not con-
cerned just with form and meaning; it is concerned with the knowledge that
enables speakers to relate form and meaning in a productive and creative
way. From this mentalistic view, the question arises of how speakers acquire
their grammars. In particular, since grammar is unconscious, parents can-
not impart the rules to their children by explicit instruction. Rather, the
process of language acquisition must be understood in terms of the child
unconsciously constructing the grammar on the basis of linguistic and
contextual experience. However, this raises two further questions: What
sorts of experience does the child make use of, and, most crucially, what are
the internal resources that the child brings to bear on the construction of
a grammar based on the experience? The complexity of the achieved gram-
mar, as discovered by investigation in linguistic theory, demands that the
child be provided in advance with some guidelines along which to pursue
generalization. Such guidelines rule out logically possible but linguistically
impossible analyses of the child’s linguistic experience.
The generative tradition has taken as its most important goal the char-
acterization of these guidelines, calling them Universal Grammar (UG),
the language capacity, or the language faculty. The nature of UG has
been investigated by examining large-scale patterns of similarity across the
grammars of languages (spoken and signed), language acquisition by chil-
dren and adults, patterns of language loss and impairment, and historical
change due to drift and language contact, as well as through mathemati-
cal/computational modeling of all these phenomena.
To summarize to this point, the enterprise of describing the grammars of
particular languages serves the broader goal of understanding the nature
of Universal Grammar. It provides us with the means of exploring what
knowledge of language must be built into the learner and what knowledge is
acquired on the basis of experience. Some knowledge may well derive from
the internal structure of the learner, and not from experience, if there is in
fact no basis for it in experience. We would expect such knowledge to be
universal, holding across all languages and all speakers. Other knowledge
is demonstrably very specific to a given language and thus must be learned
on the basis of experience with that language. And crucially the learner’s
internal resources for learning language must be innate, for they precede
and enable learning.
6 1. OVERVIEW
One can further ask what aspects of these internal resources are specific
to language learning, and what parts are shared with other components of
other human – or primate – capacities. To the extent that some parts are spe-
cific to language, we are led to the claim that the capacity to acquire and use
human language is a human cognitive specialization, a claim that has been
central to generative grammar since its inception. We might distinguish the
child’s full internal resources for language acquisition, which include inter
alia various social skills, pattern recognition, categorization, identification
of correlations, and the capacity for imitation, from the language-specific
resources, calling the latter Narrow UG and the rest Broad UG. Then an
eventual goal of linguistic theory is to sort out Narrow UG from Broad UG.
Doing so, of course, may require a comparable account of the other aspects
of human cognition that make use of elements of Broad UG, an account
at present far beyond the horizon but very much a concern of cognitive
science. 2
1.3. Where does syntactic theory fit in?
Syntactic theory sits squarely in the middle of this general perspective. A

syntactic description of a given phenomenon in some language is situated
within a network of theoretical questions. The most fundamental question
is whether the syntactic description is something that could be (i) acquired
by a learner on the basis of experience or (ii) plausibly assumed to be part
of our innate knowledge of language. To get a feel for this question, let us
consider two examples at either end of the spectrum and one in the middle.
First, consider the fact that the word dog is a noun and its plural is dogs.
Clearly knowledge of these facts cannot be part of Universal Grammar
per se. There would be no way to predict, strictly on the basis of universal
principles, that the facts would be precisely these in any language. So they
are part of the knowledge acquired by the learner on the basis of experi-
ence with the language, through exposure to examples of correspondences
between form and meaning.
Second, consider a more complex fact about questions in English. As we
will discuss in considerable detail in Chapter 9, in an English question the
interrogative word or phrase appears in the initial position of the question
2
Cf. Pinker and Jackendoff 2005.
1.4. SIMPLER SYNTAX 7
and there is a “gap” in the position that corresponds to the function of this
initial word or phrase. In the following examples the direct object of the
verb is what, which appears in initial position, and there cannot also be a
phrase following the verb.
(7) a. What did Sandy say ___ ?
b. What do you think Sandy said ___ ?
c. What do you think I thought Sandy said ___ ?
∗
(8) a. What did Sandy say something?
∗
b. What do you think Sandy said something?
∗
c. What do you think I thought Sandy said something?
A theory of syntax must tell us that it is possible to form a question as

illustrated in (7). It must also tell us what other ways of making questions
there are, and whether there are certain logical possibilities that do not
exist. And ideally it should provide an explanation of why some logical
possibilities exist and others do not.
Consider finally the fact that the verb precedes the direct object in Eng-
lish. Clearly this is not part of our innate knowledge of language, since in
some languages the verb follows the direct object, or may appear on either
side. We might take this to be a contingent fact about the language, acquired
on the basis of experience. But there is a more general fact lurking here,
which is that many languages appear to have verb phrases built around
verbs. While the particular location of the verb in a given language clearly
has to be learned, it could be plausibly argued that the possibility of having
a phrase based on a verb and containing a direct object and other phrases is
something that is not a contingent fact about language but something quite
central to language in general. Precisely what the nature of this knowledge
is, and how it is represented in the mind of the learner, is of course the key
question.
1.4. Simpler Syntax
Interacting with the problem of how language is acquired is the question

of what precisely is acquired when we acquire a language. The linguist’s
grammar is a theory of what a native speaker’s knowledge consists of.
Different syntactic theories make different claims about the nature of this
knowledge. This book is organized around the perspective that a syntac-
tic description should be the simplest one that is capable of accounting
8 1. OVERVIEW
properly for the correspondence between form and meaning. This is the
perspective of Simpler Syntax (Culicover and Jackendoff 2005), and many
of the analyses in this book are based on those sketched out there and the
general approach.
We illustrate the Simpler Syntax perspective by briefly comparing one of
our analyses with an alternative. We go into more detail later. Consider the
following pair of sentences.
(9) a. Mary expects that she will win.

b. Mary expects to win.
The phrase that she will win is called a “sentential complement” of the verb
expect. In (9a), she can refer to Mary. Example (9b) can be paraphrased as
Mary expects that she (Mary) will win. Hence the two sentences may have
the same meaning.
Mainstream generative grammar (see, for example, Chomsky 1973) has
traditionally used this synonymy, and related facts, to motivate assigning
the same syntactic description to the two sentences. Since (9a) contains
a sentential complement, that she will win, so does (9b), if we apply this
methodology. The sentential complement of (9b) would then be to win. But
to win lacks an apparent subject. So in order to maintain a uniform syntactic
description of the two sentences, we must assume that there is an invisible
subject of to win in (9b) that refers to Mary, just like she does in (9a).
The alternative pursued in Simpler Syntax is to account for the synonymy
by positing rules of interpretation for (9b) without an invisible syntactic
subject, which produce the same meaning as the rule of interpretation
for (9a). 3
Thus, the question of simplicity comes down to this: Is it possible to
explain this form/meaning correspondence without assuming that there are
invisible subjects, and associated invisible syntactic structure? More gener-
ally, is it possible to account for all of the form/meaning correspondences
in natural languages without assuming invisible phrases and associated
invisible structure? If it is not possible, then the argument for the more
abstract structure is secure. But if it is possible, then Simpler Syntax argues
that the simpler alternative should be adopted.
3
Such an approach is also taken in contemporary non-mainstream approaches
such as Head-driven Phrase Structure Grammar (HPSG) – see Pollard and Sag
1994; and Lexical Functional Grammar (LFG) – see Kaplan and Bresnan 1982.
1.4. SIMPLER SYNTAX 9
A related difference between Simpler Syntax (and a number of other

syntactic theories) and mainstream generative grammar is that mainstream
generative grammar is derivational, while Simpler Syntax is not. A deriva-
tional theory is one that assumes that the observed position of syntactic
units may be different from their position in a more abstract representation.
For example, in mainstream approaches, the subject of the passive sentence
The students were arrested is the direct object of the verb in a more abstract
representation. Again, much of the motivation for this abstract structure is
uniformity of meaning. In this case, the students functions as the “logical
object” of arrested, just as it does in [The cops] arrested the students.
The derivational approach captures this sameness of meaning directly;
a non-derivational approach must capture it in other ways. In Simpler
Syntax, the position of words and phrases with respect to one another, and
their precise form, is dependent on the rules of a language that specify the
relationship between position and meaning directly. For example, for the
two sentences in (10) –
(10) a. The cops arrested the students.
b. The students were arrested.
there are two rules for positioning the phrase the students (and similar
phrases). One rule makes the students the direct object of arrested in a
position following arrested, as in (10a). The other rule puts the students
in subject position, as in (10b). On the derivational approach, there is one
rule that makes the students the direct object of arrested in a position fol-
lowing arrested for both sentences. Then (10b) is derived by a “movement”
that removes it from its position following arrested, and puts it in subject
position. The two approaches are equivalent in that they produce the same
structures, but they differ in the way in which they do it.
The approach taken in this book is to lay out the basic relationships that
a grammar must account for, and sketch out the Simpler Syntax analyses
that express these relationships in a more or less schematic way. We also
summarize the mainstream approach to the same phenomena, since much
of the terminology and the specifics of mainstream analyses are the lingua
franca in contemporary discourse about syntactic phenomena.
2
Syntactic Categories
This chapter is concerned with syntactic categories. Section 2.1 introduces

the traditional lexical categories, such as noun and verb, as well as the
minor categories, such as article, quantifier, and conjunction. An important
idea introduced in this section is that words of the same category may
substitute for one another in a given syntactic context. Section 2.2 looks at
the grammatical properties of words, such as number, case, and gender, and
introduces the notion of a morphological paradigm. Section 2.3 looks at
how a sequence of words forms a phrase of a given category, and highlights
the relationship between the category of the phrase and the category of the
head of the phrase. Section 2.4 explores some methodological issues in the
general theory of linguistic categories.
2.1. Traditional categories
The traditional lexical categories found in English are noun (1a), verb (1b),
preposition (1c), adjective (1d), and adverb (1e).
(1) a. dog, unicorn, truth, Mary, encouragement, plumber, . . . [noun]

b. reads, would, smile, crying, represented, . . . [verb]
c. in, on, over, under, against, . . . [preposition]
d. tall, big, prepared, amusing, . . . [adjective]
e. today, fast, quickly, upwards, often, . . . [adverb]
The conventional basis for deciding that a group of words are members of
a particular category is that these words can be substituted for one another
in all linguistic contexts without affecting grammaticality. The general prin-
ciple is this:
Substitution: The result of substituting a word of a category C for another word of

the same category does not change the grammaticality of the phrase or sentence in
which it appears, although it may render it odd in meaning or even nonsensical.
12 2. SYNTACTIC CATEGORIES
We develop the general idea of substitution and its relation to grammatical-

ity in section 2.1.1 where we discuss nouns, and then apply it to the other
traditional categories in sections 2.1.2–2.1.6. While substitution is often
useful for identifying or validating categories, it has limitations, as we see
in section 2.4.
2.1.1. Nouns
In general, if a sentence contains the word cat, it is possible to replace cat

with dog. In the following examples and throughout, the braces notation {}
signifies that the listed elements are alternatives that may appear in a given
position in a phrase. So (2a) is an abbreviation for the cat, this cat, that cat,
every cat.
 
 the 
 
this
(2) a. cat
 that 
 
every
 
 the 
 
this
b. dog
 that 
 
every
And we can have
(3) a. The cat is sitting on the mat.

b. The dog is sitting on the mat.
(4) a. I was petting the cat.
b. I was petting the dog.
and
(5) a. one very fat furry cat

two very fat furry cats
b. one very fat furry dog
two very fat furry dogs
Intuitively, then, cat and dog are members of the same category; it is hard
to imagine any grammatical context that can have one but not the other,
although there are certain combinations of words that we would not expect
to find, such as The dog meowed. We call the category containing cat and
dog noun, typically abbreviated as N.
2.1. TRADITIONAL CATEGORIES 13
Notation: Representing categories

The category of a noun is conventionally represented in several ways in a
description of the syntactic properties and structure of a string of words
and morphemes. One way is to bracket the word and label the bracket
with the category. This is called a labeled bracketing.
[N cat]
Naturally, in a complete description of a string of words, all of the words

would be labeled with their categories, as would the phrases that they are
part of.
Another way is to draw a diagram (which is part of a tree) in which the
label is shown above the word.
N
|
cat
We will give more complex examples of labeled bracketings and tree
diagrams as we consider more complex phrases.
Yet another way is to consider the category to be an attribute or feature
of the word. We then represent the word cat with a notation that says
“the category of this word is N”. This style of representation is called an
attribute value matrix (AVM).

cat
CATEGORY N
All of these notational conventions are used in syntax, and we use all of
them in this book.
The phrases in (2) show that there are certain contexts in which both cat
and dog may appear. This is to be expected if cat and dog are of the same
category. The sentences in (3)–(4) suggest that a phrase that contains dog
can appear wherever the same kind of phrase that contains cat appears. Of
course, in order to test this hypothesis fully we would have to look at a lot
more contexts. Finally, the phrases in (5) show that both cat and dog may
appear with the marker for the plural, -s.
It should be apparent even from these simple examples that there is an
implicit appeal to meaning in the application of substitution tests. For
example, in the case of (5), we are assuming that the -s that appears with
cat and dog is the same plural marker, which means “more than one”, and
not the possessive -s (written ’s) or the third person singular present verbal
inflection (as in speaks). We informally use the English spelling -s to refer
to this morpheme; its more technical name would be PLURAL , abbreviated
as PL. And we are assuming that this -s (or PL) is the same grammatical
morpheme whether it is realized phonetically as /z/ when it is attached to
dog or as /s/ when it is attached to cat. Another form of plural -s is /@z/, as
in beaches. This phenomenon, where a morpheme takes various phonetic
forms, is called allomorphy. The various forms that a morpheme takes are
called its allomorphs. When an allomorph of a morpheme has no phonetic
form, that allomorph is called a zero-allomorph.
If we are describing a word in terms of its morphological structure
and how that determines how it combines with other words to form a
phrase, what is important are the morphemes that make it up, and not
their allomorphs. The contrast between the morphological structure and the
allomorphy is illustrated in (6). The allomorphs that are easily distinguished
are marked in boldface.
(6)
word morphological phonetic form
structure
dog dog- SG /dOg/

dogs dog- PL /dOgz/
cat cat- SG /kæt/
cats cat- PL /kæts/
bush bush- SG /b2S/
bushes bush- PL /b2S z/
sheep sheep- SG /Sip/
sheep sheep- PL /Sip/
ox ox- SG /Aks/
oxen ox- PL /Aksn./
woman woman-SG /w2mn./
women woman-PL /wImn./
We expect that words that mean more or less the same thing, or that refer
to things of more or less the same type, or more generally have the same type
of meaning, will have the same grammatical category. Cat and dog are the
same type of thing, and so it is not surprising that we can use them both with
the, that we can count them and use them in the singular and plural, and so
on. Similarly, because we can use the, this, that, and every with dog and cat
in more or less the same way, we might suppose that they are members of
the same category. So, to a certain extent, we are using semantic intuitions
to judge that two words are of the same category.
Semantic intuitions of this sort are also at the basis for the notion that
the same categories hold across languages. It is of course impossible to
substitute a word of one language into a sentence of another language
while maintaining grammaticality. But the fact that words of two languages
mean the same thing suggests that the categories that they belong to are
the same. So English cat, German Katze, French chat, and Spanish gato are
all said to be members of the category noun, even without any possibility
of substitution for one another in any of the languages (e.g. I love my
∗
Katze
∗
chat ).
∗
gato
It is important to note that semantic properties do not always correlate
with syntactic and especially morphosyntactic properties. For example, the
nouns scissors and pants are morphologically plural, but semantically sin-
gular. Agreement with the verb is sensitive to the morphology, so we get My
pants are too short and not ∗ My pants is too short. On the other hand, we
may say that furniture is morphologically singular, but semantically plural:
My furniture is expensive, ∗ My furniture are expensive.
The examples of singular and plural nouns raise another important point.
We conventionally say that singular and plural nouns are all nouns, but
singular and plural nouns cannot freely substitute for one another; in fact,
there are only certain contexts in English (like after the and possessives)
where substitution is freely possible.

cat
(7) a. the
cats

cat
b. my
cats

cat
c. a ∗
cats
∗
cat
d. two
cats
What these examples show is that where there is a morphological mark-

ing, say for singular and plural number, members of the same category
are marked for the same property, and not substitutable for one another
without appropriate marking. Typically this state of affairs is called a mor-
phological paradigm. Where there is a paradigm, a single word has a number
of alternate forms whose distribution is governed by meaning differences (in

the case of singular and plural) and grammatical restrictions. The paradigm
is one way of verifying that two words are in the same category: if they are
in the same category, they should have alternating forms in the paradigm
for that category.
Notation: Representing paradigm properties

A common convention for representing the fact that two words are actu-
ally different paradigmatic forms of the same word is that of attributes or
features, which we introduced above, taking the category of a word to be
a feature of the word. Another example of a feature is NUMBER. For a
word like cat, the value of the feature NUMBER is SINGULAR ( SG ), while
for cats it is PLURAL ( PL ). Both words are members of the category N.
We can represent this information as follows:
N N
[NUMBER SG] [NUMBER PL]

cat cats
Alternatively, we may use an attribute value matrix (AVM) to show
all of the attributes in a uniform fashion. (So far we have NUMBER and
CATEGORY , but there are many more.)
   
cat cats
   
 CATEGORY N   CATEGORY N 
NUMBER SG NUMBER PL
In an AVM we list each feature and its corresponding value. A non-

linguistic case of an AVM would be a listing of personal information on
a driver’s license application, for example.
 
NAME Sandy Student
 DATE OF BIRTH Feb. 29, 1985 
 
 
 SEX Female 
 
 HEIGHT 160 cm 
 
 WEIGHT 
 53 kg 
 
 HAIR COLOR brown 
EYE COLOR brown
The features that appear in the AVM are the essential properties of what
is being described.
This discussion of categories thus far shows three things:

r Elements of the same category may substitute for one another.
r Paradigms play a role in defining what category a word is a member of.
r If two words cannot be substituted freely for one another, this does not mean
that they are not in the same category. They might simply be alternate forms of
a paradigm with different distributional properties. The intuition that this is the
case again depends on meaning, because we must know independently that the
two forms are actually variants of the same word.
There are other cases where “same category” does not mean “completely
free substitutability”. Intuitions about substitution immediately run up
against the fact that there are actually many contexts in which substituting
one word for another of the same apparent category results in some kind of
unacceptability. For instances, Cats meow is very natural but Dogs meow
sounds a little strange. In such a case we can say that there is nothing
linguistically wrong with Dogs meow, it’s just that dogs do not meow, so
the sentence is false, but it is not ungrammatical. By way of comparison,
the sentence Dogs don’t meow is completely normal, both syntactically and
semantically.
It is possible to construct more and more extreme violations of the nor-
mal relationship between a noun and a verb, but in each case we would not
want to say that the violation is due to the words not being of the proper
category. Here are some examples. In (8a), we attribute meowing to some-
thing that does not exist. In (8b), we attribute meowing to a class of human
beings, which is odd. In (8c), we attribute meowing to inanimate objects,
which is arguably impossible (but imaginable in some alternate universe in
which rocks behave like animate objects). And in (8d) we attribute meowing
to an abstraction, truth, which is impossible.
(8) a. Unicorns meow.

b. Plumbers meow.
c. Rocks meow.
d. Truth meows.
In each case, we say that the sentence is false because the property expressed
by the verb does not (and in some cases cannot) hold of the thing referred to
by the noun. The oddness of the examples in (8) is typically called semantic
anomaly.
Semantic anomaly must be distinguished from ungrammaticality.
Ungrammaticality occurs when there is something wrong with the
arrangement of the words and phrases of a sentence according to their

category and their morphological form. Semantic anomaly occurs when
there is something peculiar about the meaning. The meaning is determined
by the grammatical properties of a sentence, but it is not equivalent to it.
A sentence can be perfectly grammatical yet completely nonsensical. And it
can be ungrammatical but perfectly coherent in meaning.
If we substitute a word for cat that is not a noun, then we have a problem
of grammaticality.
(9) a.The cat is sitting on the mat.

∗
b. The furry is sitting on the mat.
∗
c. The on is sitting on the mat.
∗
d. The the is sitting on the mat.
∗
e. The is is sitting on the mat.
 
 ∗cats 
 
on’s
(10) two furry ∗

 ∗ the’s 

is’s
There is a close connection here between the syntactic facts and the seman-
tic facts. These expressions are also semantically anomalous because the
words that substitute for cat are incapable of referring to a definite object.
To summarize, the defining characteristics of nouns in English are the
following:
r they can appear immediately after the/this/every, etc.
r they can appear immediately after adjectives.
r they may participate in the singular/plural (number) paradigm (if they denote
things that can be counted, like dogs and cats).
Beyond this, nouns tend to have certain semantic properties. Our immedi-
ate intuition might be that nouns refer to things, but such an intuition is too
simple; we know that nouns can refer to places (New Orleans), times (tomor-
row), actions (swimming), and events (the recent football match), emotions,
ideas, intuitions, sentences, memories, and much else. What does seem to
be true of almost every noun if not all nouns is that what it refers to is in
principle quantifiable or is a set made up of quantifiable members. A noun
can refer to something that is countable (like dogs and cats), something that
is measurable but not countable (such as water or sincerity), or something
that is unique (like Albert Einstein or the US government). A noun can
refer to a particular collection (like humankind or furniture) or a particular
species (like the platypus).
Substitutability typically fails when a noun is in a context that requires

that it refer to something countable, and the noun to be substituted does
not, and vice versa. For example,
(11) a. every dog

∗
b. every sincerity
(12) a. much sincerity
∗
b. much dog
Again, we do not want to say that dog and sincerity do not belong to the
same category; the problem here is a semantic one, not a syntactic one.
2.1.2. Verbs
Verbs are words like talk, eat, run, and sing. We abbreviate this category as
V. Verbs typically express actions, relations, and properties. But we cannot
use this semantic property to define what a verb is, because there are words
of other categories that also express relations and properties. For example,
brother expresses a kinship relation, while on expresses a spatial relation.
But brother and on are not verbs in English. And tall expresses a property,
but it is not a verb.
In English and many other languages a verb is distinguished by the mor-
phological paradigm that it participates in. The English verbal paradigm is
summarized in (13).
(13)
B ARE FORM 3 RD PERSON PAST - ING FORM 1 - EN FORM 2
SING. PRESENT
talk talks talked talking talked

eat eats ate eating eaten
run runs ran running run
sing sings sang singing sung
The only verb that deviates from this pattern is be, which has three forms in
the present and two in the past.
1
In grammatical terminology, this form is called the progressive or the present
participle.
2
In grammatical terminology, this form is called the past participle.
(14)
B ARE FORM P RESENT PAST - ING FORM - EN FORM
be am, are, is was, were being been
We will look more closely at the verbal paradigm in our discussion of

morphosyntax in section 2.2.
As expected, it is often possible to substitute one verb for another. Some-
times the meaning becomes strange: I ate my dinner is quite natural but
I welded my dinner is not. But, as we have already seen, these failures of
substitution do not bear on whether the words are of the same grammatical
category – they are semantically anomalous.
Matters become more complex when we consider verbs that take different
numbers of arguments. An argument of a verb is a phrase that refers to some
thing, person, place, etc. that participates in the relation expressed by the
verb. As discussed in more detail in Chapter 5, there are
r verbs that take one argument (intransitive), such as die;
r those that take two arguments (transitive), such as eat;
r those that take three arguments (ditransitive), such as give;
r and those that take zero arguments, such as rain.
While it is often possible to leave out an argument of a verb, it is very

difficult if not impossible to use a verb with more arguments than it per-
mits. 3 So, if we substitute a one-argument verb for a two-argument verb, or
a two-argument verb for a three-argument verb, or a zero-argument verb
for a one- or two-argument verb, the result is decidedly ill-formed, because
there is at least one argument too many. In the following examples, we have
underlined the superfluous arguments.
(15) a. I ate my dinner.
b. I fell.
∗
c. I fell my dinner.
(16) a. Mary gave John the magazine.
b. Mary shredded the magazine.
∗
c. Mary shredded John the magazine.
(17) a. It rained.
∗
b. I rained.
3
The exception to this is called “coercion”, where a verb is forced into a par-
ticular syntactic context. An example is They looted me a television, meaning that
while they were looting they took a television to give to me. See Chapter 5 for more
discussion of coercion.
∗
c. It rained my dinner.
∗
d. I rained my dinner.
∗
e. It rained John my dinner.
∗
f. I rained John my dinner.
The fact that different verbs take different numbers of arguments is often
held to constitute evidence for syntactic subcategories. On this view, the
number of arguments that a verb takes is a syntactic property of the verb.
All verbs that take this number of categories fall into the same subcategory,
and all verbs taken together constitute the larger category verb (V).
To a considerable extent the grouping of verbs into subcategories is a
consequence of the semantic properties of the verb. If the meaning of a
verb is a relation involving at most two participants, then there is simply no
meaning that can be assigned to the extra argument in the starred exam-
ples in (15)–(17). In such cases, we may say that the number of syntactic
arguments exceeds the number of semantic arguments.
A second type of verbal subcategory concerns the auxiliary verbs, that is,
have and be and the modals will, can, etc. We call this category VAUX . The
auxiliary verbs contrast with main verbs such as eat, run, and advise in their
distribution. As the following examples illustrate, the form of a sequence of
verbs in English is restricted.
(18) a. I have visited NY many times.

b. I am visiting NY.
∗
c. I have visiting NY many times.
∗
d. I am visited NY.
The auxiliary verb have must be followed by a verb with the -ed form (the
“past participle”), and the auxiliary verb be must be followed by a verb with
the -ing form (the “progressive participle”). We return to a fuller analysis of
the restrictions on the English verbal sequence in Chapter 3.
A third subcategory consists of verbs that select infinitival or finite com-
plements.
(19) a. I expect that you will win.

I believe that you will win.
∗
I want that you will win.
b. I believe you to have won.
I expect you to win.
I want you to win.
c. ∗ I believe to have won.

I expect to win.
I want to win.
d. I persuaded Mary to leave.
I persuaded Mary that she should leave.
∗
I persuaded to leave.
∗
I persuaded that Mary should leave.
The verbs that take infinitival or finite complements fall into a number of
subcategories. Some take only finite complements, others only nonfinite
complements. Some allow a noun phrase before the infinitival or finite
complement, others require it, while others disallow it. In general it does
not appear to be possible to predict all of the properties of such verbs
on semantic grounds; those that cannot be must be part of the lexical
specification. We return to verbs of this type in Chapter 7.
2.1.3. Adjectives
Another lexical category in English is adjective (ADJ). Some examples are

tall, angry, old, irritating. Substitution tests for adjectives are revealing.
Adjectives typically precede the noun that they modify. This is called the
attributive use of adjectives –
 

 tall 

angry
(20) a(n) bear

 old 

irritating
– and they can also appear as the complement of a form of the verb be. This
is called the predicational use of adjectives.
 

 tall 

angry
(21) The bear is .

 old 

irritating
Typically, when a combination of adjective and noun fails, it is because
of a semantic anomaly or incompatibility. We use “#” in the following
examples to indicate such an anomaly.
(22) # the sincere tree
# the rational rock
# the blue truth
# the three-sided square
# the present(day) King of France
But there are some adjectives that cannot be used predicatively.
(23) a. the present(day) King of France

∗
The King of France is present(day).
∗
I consider the King of France present(day).
b. the alleged assassin
∗
The assassin was alleged.
∗
I consider the assassin alleged.
c. a perfect idiot
∗
The idiot was perfect.
∗
I consider the idiot perfect. 4
Since these appear to be adjectives in other respects, we may hypothesize

that the failures here are due to meaning.
Finally, an adjective that denotes an attribute that has quantity appears
in the paradigm exemplified in (24).
(24)
B ASE C OMPARATIVE S UPERLATIVE
tall taller tallest

old older oldest
angry angrier angriest
irritating more irritating most irritating
These adjectives may also be modified by intensifiers like very and so.
 
 tall
 

very angry
(25)
so 
 old 

irritating
The adjectives that cannot be used predicatively do not participate in the

comparative paradigm.
∗
(26) the more present King of France
∗
the more alleged assassin
∗
the more perfect idiot
The explanation appears to be a semantic one: these adjectives do not

denote a measurable property.
4
This sentence is acceptable under another interpretation of perfect.
2.1.4. Prepositions
Using substitution tests, we find that there is another class of words in

English that are not nouns, verbs, or adjectives. This class, called preposition
(P), is a closed class of words that are used to express place, time, manner,
and other aspects of events and actions.
 

 on 


 in 


 


 at 


 

 near 
(27) Mary was sitting under the Ferrari.

 


 next to 


 
 on top of 
 


 
 in front of 

behind
All of the words or word sequences in (27) pick out some location in
combination with the phrase the Ferrari. A similar set of words can be used
with time expressions.
 

 by 

 after 
 
(28) We’ll be gone before tomorrow.

 


 until 

during
It is not possible to switch most of these prepositions with those in (27); a
reasonable intuition is that the failure is due to semantic anomaly.
 
 after 
 
before
(29) # Mary was sitting the Ferrari. 5
 until 
 
during
Mary was sitting by the Ferrari.
 

 in 


 


 at 


 near 


 

under
(30) # We’ll be gone tomorrow.

 next to 
 on top of 
 


 


 
 in front of 
 

behind
The prepositions cannot be substituted for nouns, verbs, or adjectives, in

general.
5
Before can be used as a preposition of location in other contexts, such as I see a
strange face before me and The Ferrari came to a complete stop right before the finish
line.
(31) a. the book / ∗ the on

b. I am reading a book / ∗ I am aftering a book
c. the interesting book / ∗ the after book
However, some prepositions can be used predicatively, and perhaps idiomat-

ically in some cases, as in
(32) a. She’s really on.
b. One more step and you’re in.
c. You’re a little behind.
d. My horse is in front (∗ of).
e. The book you want is on top (∗ of)
But not all prepositions allow this use.

 
∗

 by 
 ∗ after 
 


∗ 

before
(33) She’s really ∗ .
 ∗ until 
 

 

 ∗ during 
 
at
2.1.5. Adverbs
The category adverb (ADV) is a problematic one, because it is not clear

on the basis of distributional evidence whether there is a single gener-
alized category or a number of more specialized ones. (See Ernst 2002
for an extensive treatment of adverbs and related constructions.) We will
take adverbs here to be individual lexical items that are used to modify
verb phrases or sentences. Many adjectives can be made into adverbs by
adding -ly.
(34) a. quick → quickly
b. necessary → necessarily
c. optional → optionally
d. dark → darkly
e. stupid → stupidly
etc.
Other adverbs, like fast and well, do not have -ly but have the same function
as the -ly adverbs.
Adverbs may express manner, direction, location, time, and other
attributes of an action or state of affairs. These notions may also be
expressed by using prepositional phrases.
(35) a. Sandy was walking at a steady pace.

b. Sandy jumped onto the table in a split second.
These prepositional phrases are not members of the syntactic category

adverb, although they may have the same grammatical and semantic func-
tion as adverbs. We say that these prepositional phrases, and the adverbs,
have adverbial functions. Later we will find it useful to refer to the class of
“adverbials” that contains the prepositional phrases and the adverbs.
Adverbs in English have the interesting property that they may appear in
a number of positions in a sentence, sometimes with subtle meaning differ-
ences. Consider the examples in (36) that illustrate the possible positions for
quickly.
(36) a. Quickly, Sandy jumped onto the table.
b. Sandy quickly jumped onto the table.
c. Sandy jumped onto the table quickly.
The examples show that quickly may appear in initial position (36a), imme-
diately before the verb (36b), or in final position (36c). But merely, which is
also typically classified as an adverb, may appear only before the verb.
(37) a. ∗ Merely, Sandy jumped onto the table.
b. Sandy merely jumped onto the table.
c. ∗ Sandy jumped onto the table merely.
Regrettably may appear in initial position or before the verb; it may appear
at the end only parenthetically.
(38) a. Regrettably, Sandy jumped onto the table.
b. Sandy regrettably jumped onto the table.
∗
c. Sandy jumped onto the table regrettably.
d. Sandy jumped onto the table, regrettably.
And when there is more than one verb in a sequence, the distribution of
adverbs become somewhat more complex. Not only are not all positions
allowed with all adverbs, but there are meaning differences. For example,
in (39), the adverb sadly can be a judgment by the speaker about “Sandy
should have confessed”, or about Sandy, or about the manner of confession.
(39) a. Sadly, Sandy should have confessed.
b. Sandy sadly should have confessed.
c. Sandy should sadly have confessed.
d. Sandy should have sadly confessed.
e. Sandy should have confessed sadly.
f. Sandy should have confessed, sadly.
(40) a. ?Quickly, Sandy should have confessed.

b. Sandy quickly should have confessed.
c. ?Sandy should quickly have confessed.
d. Sandy should have quickly confessed.
e. Sandy should have confessed quickly.
∗
f. Sandy should have confessed, quickly.
Problems 3 asks you to look in more detail at the effect of adverb position
on its interpretation with respect to the rest of the sentence.
2.1.6. Minor categories
Articles (ART) in English are the words the and a. The category demon-
strative (DEM) consists of this, that, these, and those. Quantifiers (Q) are
words such as every, all, each, and both. These categories are traditionally
distinguished on semantic grounds, since they have very different functions.
Substitution tests suggest that they are all of the same category, deter-
miner (DET). The following examples show that while members of these
categories can be substituted for one another, they can in general not be
used together in the same phrase. As always, this latter fact may be the
consequence of semantic incompatibility or redundancy, but in the absence
of a suitable semantic account, we take this to be a syntactic fact.
 

 the  

 every 

(41) a. this book

 


 that 

each
 

 these 

those
b. people

 all 

both
∗
(42) a. the a book
∗
b. the every book
∗
c. every the book
∗
d. every this book
Exceptions are that all and both can precede the articles and demonstratives:

the
(43) a. all books
these

the
b. both people
these
This fact is sometimes accounted for by treating these sequences as alter-

native forms of all of the/these and both of the/those, etc. Some adjec-
tives, like many, express quantity and are therefore semantically related to
quantifiers, but can appear with articles, e.g. the many supportive friends of
Sandy.
Another minor category contains the conjunctions (CONJ). And and or
are called coordinating conjunctions, because they are used with phrases of
the same type, e.g. Albert Einstein and Kurt Godel, to eat and drink, in and
out, Speak now or forever hold your peace. Subordinating conjunctions like
although, while, if, and because, are used to introduce sentences.
(44) although it is raining
while we were there
if it doesn’t rain
because we were angry
Some subordinating conjunctions also serve as prepositions.

before the concert
(45)
after the concert started
2.2. Morphosyntax
In this section we look at the relationship between the form of a word and
its syntactic properties, called morphosyntax. Morphology has to do with the
form of words; morphosyntax is concerned with the relationship between the
form of a word and its function and distribution in a phrase or sentence.
2.2.1. Words and lexical items
In English the forms he, she, they can only be the subject of a finite sentence.

He
(46) She called.
They
∗
he
We called ∗
(47) she .
∗
they
The forms him, her, them, on the other hand, cannot be subjects of a finite
sentence.
2.2. MORPHOSYNTAX 29
∗
Him
(48) ∗ called.
Her
∗
Them

him
(49) We called her .
them
We see that there is a strong connection between the form of the word and
its syntactic function in the sentence – certain forms must be subjects, and
certain forms cannot be. In English the correlation between the form and
grammatical function of nouns is restricted to the pronouns, but in some
languages it is much more general.
Another example of morphosyntax involves the marking of the verb in
the third person in English. If the subject is singular, the verb is marked
with -s; if it is plural, it is not marked.

sees
(50) a. Leslie ∗ me.
see

see
b. Leslie and Lee ∗ me.
sees
Observations such as these show that we have to distinguish between a word

as an individual element in a sentence, and the collection of words that form
a single paradigm. Informally, we think of a word as a unit of a language
defined by certain sounds. For example, there is the word pronounced /si/
(that is, “see”), and the word pronounced /siz/ (that is, “sees”). But, in some
sense, these two forms are two variants of a single more abstract element
which we call see (pronounced /si/).
In order to capture this distinction we define the notions of lexical entry
(or lexical item) and word differently. A lexical entry is an abstract object
that has a meaning and syntactic properties, such as CATEGORY. A word,
on the other hand, is the form that a lexical entry takes when it appears in an
actual syntactic context. The word inherits its category and other syntactic
properties, as well as its sound, from the lexical entry that it represents. In
the simplest cases, the word is the basic phonetic realization of the lexical
entry, with no modifications. So the lexical entry see has the form of the
word /si/. In terms of this distinction, see is a lexical entry, /si/ and /siz/ are
words. Moreover, the sequence of sounds /sæf/ could be a word of English,
since it sounds like an English word (it rhymes with “laugh”). It is not
because it does not correspond to any lexical entry.
Not all lexical entries correspond to words. There is a special class of
lexical entries that have grammatical functions associated with them, and
cannot stand alone but must be attached to something to form words. These
are inflectional morphemes. The set of inflectional morphemes for a syntactic
category (like V) constitute a particular type of morphological paradigm
which we call an inflectional paradigm.
One example of an inflectional morpheme is the marker of the third
person singular present tense in English, which takes the form /z/ when it
is attached to /si/, giving /siz/. The allomorph of this marker depends on
the form of the word it attaches to. It is /z/ when the word ends in a vowel
or a voiced stop (as in /siz/ for “sees” and /ridz/ for “reads”), /s/ when the
word ends in a voiceless stop (as in /rayts/ for “writes”), and /@z/ when the
word ends in a fricative (as in /rayz@z/ for “rises” or /bæS@z/ for “bashes”).
For convenience, we refer to this morpheme as 3.SG.PRES, indicating that
it marks the third person singular present.
There is another type of morphology that does not involve inflectional
paradigms but morphologically defined relationships between syntactic cat-
egories. It is customary to refer to this as derivational morphology. An
example of this type of morphology is given by the word derivational, which
is composed of deriv(e), -ation, and -al. Notice that the morphological
structure of derivational determines the syntactic category of the word:
r A word of the form V+-ation is a noun: derive ∼ derivation.
r A word of the form N+-al is an adjective: derivation ∼ derivational.
Derivational morphology contrasts with inflectional morphology, which

links the precise form of a member of a particular category to its syntactic
function.
2.2.2. The structure of the lexicon
The preceding discussion shows that what we see as a “word” in a phrase

may have a rather abstract linguistic description in terms of lexical entries
and a complex internal structure. For example, the word “sees” is the
realization of the lexical entry see and the lexical entry 3.SG.PRES.
The lexicon is the sum total of all of the lexical entries. It is the repository
of all of the information that we have about linguistic expressions that
cannot be explained in terms of other expressions. To take a simple example,
the word pig has a particular form (/pIg/), particular syntactic properties
(it is a noun), and a particular meaning (it refers to certain types of farm
animals). On the other hand, catch a pig has a form that is made up of the
form of the individual words, a syntactic structure that is determined by the
rules of English grammar, and a meaning that is the product of combining
these words with their meanings in this particular way. So we need to list
pig in the lexicon but not catch a pig.
(51) Lexicon, first version:
lexical entries
form
syntactic properties
meaning
It might appear from this simple example that the lexicon consists only of
words, like pig and catch. But we have already seen that some words are
comprised of paradigms, which specify which form of the word is to be
used for a particular function. So the lexicon must include not only words
but paradigms.
Moreover, we have seen that some words have complex structure and that
in some cases this structure is regular – for example, derivation is related to
derive in the same way that infestation is related to infest. Not only are the
forms related systematically but the meanings are, too, in that a derivation
is the result or act of deriving, while infestation is the result or act of
infesting. But there are many cases where a word has a clear morphological
relationship to another word, but the meaning is not totally predictable. For
instance, one meaning of animation is only loosely related to animate: She
spoke with great animation. And the morphological relationships between
words are restricted, so that not all apparently similar words may have the
same morphological structure. Contrast, for example, derive ∼ derivation
(∗ derival) and arrive ∼ ∗ arrivation (arrival).
So it seems reasonable that we would include in the lexicon not only
the actual words with their forms and meanings but their morphological
structure.
(52) Lexicon, second version:
lexical entries
form
morphological structure
meaning
It turns out, now, that some morphological structure is productive, in

the sense that it is possible to apply it to new instances, while other
morphological structure is frozen. A productive morpheme in English is

-ness; for any adjective, it is possible to make up a new noun by adding
-ness to it. In (53), the made-up words in the left column are supposed to be
adjectives and the words in the right column are the corresponding nouns.
(53) Adjectives Nouns
glarky glarkiness
bigarre bigarreness
halumph halumphness
gleek gleekness
floog floogness
Since the capacity to make up new words by adding -ness is an aspect

of our knowledge of English, we need to represent it somewhere in our
description of the grammar of English. The lexicon already contains real
words with -ness, of course: happiness, quickness, restlessness, etc. So there
is very clearly a link between the actual structures of some words and
this capacity to carry this structure over to the creation of new words.
We include this capacity in the lexicon, as well, by linking the properties
of morphological structure found in the lexical entries to word formation
rules.
(54) Lexicon, third version:
lexical entries
form
meaning
word formation
The word formation rules define an unlimited number of “possible” new

words, constrained only by what is phonologically possible in the language.
We do not think of word formation as actually being in the lexicon but as a
mechanism that determines what may be in the lexicon.
Having gone this far, we can now see that there are even more complex
expressions whose meaning is not entirely predictable. One class of cases
are those like kick the bucket, have a cow, blow one’s stack, go postal, take
advantage of, set store by, and thousands of others. Like words with complex
structure, the meanings of some of these expressions may be related to the
meanings of the parts, but the relationship is not entirely systematic and the
meaning of the entire expression is not predictable. Kick the bucket is par-
ticularly opaque (it means “die”), while go postal is somewhat transparent
if we are familiar with recent history. 6
Complex expressions such as these have varying degrees of idiomatic-
ity, that is, unpredictable meaning. We call such cases constructions. A
construction is a syntactically complex expression whose meaning is not
entirely predictable from the meanings of its parts and the way that they are
combined in the structure. We use the term idiom for the constructions that
are completely or highly opaque in meaning, such as kick the bucket or go
postal.
In general, constructions have the structure of normal phrases of the
language, but have special meaning properties. It again seems reasonable to
expand our lexicon to include constructions, with their meaning and struc-
ture, while leaving open the possibility that new constructions can come into
the language through the connection with the productive mechanisms for
constructing new phrases, that is, through the connection with the syntax
of the language.
(55) Lexicon, final version:

lexical entries
form
meaning
constructions
form
syntactic structure
meaning
syntax
word formation
The link between “syntactic structure” and “syntax” reflects the fact that
the structures of constructions in the lexicon are for the most part deter-
mined by the syntactic structures that are possible in the language more
generally.
6
The expression means “go crazy”, in reference to several notable outbursts of
violence by postal workers.
2.2.3. Paradigms
Let us consider once again the verbal paradigm in English to see how a
paradigm works. A verb in English has several forms. The verb see has
the forms “see”, “sees”, “seeing”, “saw”, and “(have) seen”. We take the
lexical item itself to be see, which we pronounce “see”. Some of the forms
of see are entirely predictable, some are not. When a form is predictable
from the morphological paradigm, we say that it is regular; when a form is
not predictable, it is irregular. So the form “seen” is not predictable as the
past participle (She has never seen Paris like this), nor is the form “saw” as
the past tense.
On the other hand, a verb like talk is completely regular: “talk”, “talks”,
“talking”, “talked”, and “(have) talked”. We want to capture the fact that
“saw” and “talked” are both past tense forms, even though one is irregular
and the other one is regular. We can do this if we think not about their
superficial form but about how they are composed of lexical items, that is,
their morphological structure. In both cases, there is a lexical item which
we will call PAST (tense) that is added to the verb. So “saw” is see-PAST
and “talked” is talk-PAST. Providing an abstract morphological analysis
allows us to show explicitly that the two forms play the same syntactic role
in sentences without getting bogged down by the fact that the form of one
is regular and the form of the other is irregular.
We represent the verbal paradigm in the form of a table, elaborating
and modifying the preliminary table in (13). In the left column, we list the
individual verbs. The cell where the inflectional morpheme column meets
the verb row indicates what form the combination of the two takes. When
the form is predictable, it is not necessary to list the form in the individual
cell – it is sufficient to specify at the top of the column how to construct the
predictable form given the basic form, called the root.
Here is a preliminary version of the table for the verbs see and talk.
We use the symbol ∅ to indicate “zero”, that is, phonetically nothing. The
hyphen that precedes a morpheme, as in -ing and -ed, indicates that the
morpheme follows what it is attached to – it is a suffix. A hyphen that
follows a morpheme indicates that the morpheme is a prefix. The blank cells
in this table are those where the form is predictable, while the cells that are
filled in are those that are not predictable. Crucially, when a cell is filled with
an irregular form, it takes precedence over the regular form. (The irregular
form is said to block the regular form.)
(56)
VERBAL root 3rd singular other -ing form past -en form
INFLECTION present present (progressive) (past
tense tense participle)
Inflectional 3. SG. PRES. PRES. PROG. PAST PERF /

morphemes (-s) (-0 ) (-ing) (-ed) PAST. PRT
(affixes): (-ed)
talk talk
see see saw seen
The full version of this table will of course have thousands of rows, one
for each verb in English. But most of them will be fully regular. Moreover,
since the 3.SG. PRES., the PRES., and the PROG. cells are always regular, for
practical purpose we only have to show the root and the last two columns.
Representing the morphosyntactic properties of words in this way
assumes that when a particular word reflects the zero form of a morpheme,
that morpheme is nevertheless present in the abstract analysis of the word.
For example, the morphological structure of the word “talk” as in “we
talk” is not simply talk, but talk-1.PL . PRES. The morpheme PRES. has no
overt phonological form in the plural. That is, it has a zero allomorph in the
plural. (Exercise 8 asks you to work with some allomorphs of an abstract
morpheme in a language other than English.)
A regular paradigm such as verbal inflection in English can be expressed
as a chart. But what makes it possible for us to construct such a chart? In
fact, what we know when we know a language is not simply what the various
forms are for each word but the general pattern of the paradigm. The
knowledge of each word and its forms allows us to immediately recognize
the grammatical function of the word in a sentence – e.g. talked is about an
event in the past, as is saw.
But we are also able to apply the paradigm to new words. Suppose I tell
you that the word wug is a verb and it means to scratch one’s neck. If I
show you a picture of someone scratching his neck, you would say “He is
wugging” or “He wugs” and, in the past, the action would be “He wugged”.
In other words, the regular paradigm is the description of the pattern of the
forms that we know, and also constitutes a rule that we may use to construct
new forms.
Not all categories of words are as regular as the verbs in English. The
English non-reflexive personal pronouns are completely irregular, in the
sense that there is no root form for any pronoun and none of the cells in
the paradigm is predictable. In (57) we give a table for the English personal
pronouns. We use the person, number, and gender properties of each lexical
item in the leftmost column to distinguish them from one another. We name
the columns with the grammatical functions of the individual words, using
the conventional terminology for grammatical case. For the irregular forms
there are no affixes corresponding to the various grammatical functions.
(57)
CASE NOM ( INATIVE )/ ACC ( USATIVE )/ GEN ( ITIVE )/ REFL ( EXIVE )

SUBJ ( ECTIVE ) OBJ ( ECTIVE ) POSS ( ESSIVE )
Lexical
items
1.SG I me my myself
2.SG you you your yourself
3.SG. MASC he him his himself
3.SG. FEM she her her herself
1.PL we us our ourselves
2.PL you you your yourselves
3.PL they them their themselves
2.2.4. More morphosyntactic properties
2.2.4.1. Number
Next we look a range of morphosyntactic categories, not all of which
are found in all languages. One that we have already discussed is number.
Languages that mark number typically mark singular and plural, a few
mark dual (exactly two). A language that has the dual form for nouns in
addition to the singular and plural is Slovene.
(58)
SG ( SINGULAR ) DU ( DUAL ) PL ( URAL )
volk-‘wolf’ volk volkova volkov

(a) wolf two wolves wolves
There are languages that mark not only singular, plural, and dual
but also trial (three) in the pronominal system. Harley and Ritter 2002
cite the following paradigm from Biak, an Austronesian language. There
is no distinction between TRIAL and PLURAL in the first and second
persons.
(59)
SINGULAR DUAL TRIAL PLURAL
1st exclusive aiá nu n’o

(“me, you and me”)
∗
1st inclusive ‘u ‘o
(“us, including them”)
2nd áu mu m’o
3rd animate i su s’o si
3rd inanimate i su s’o na
Some languages, such as Chinese and Japanese, do not mark number

on the noun at all. These languages contrast with languages like English,
where there is a number paradigm consisting of singular and plural forms.
In English, most of the forms are regular but some are not:
(60)
N UMBER SG ( SINGULAR ) PL ( URAL )
Lexical items -0 -s
dog -0 -s
ox -0 -en
woman -0 women
man -0 men
sheep -0 -N
As in the case of the verbal paradigm, we only need to specify the contents
of a cell when it is not predictable from the morphological rules of the
language.
The information in this table is associated with the form that words take
when they appear in a sentence. We have seen that there are various ways
of expressing this information, but all of them reflect the fact that there is a
difference between the abstract morphological structure of a word and how
it is pronounced. The forms in (61) show the structure for the nouns cat and
cats.
(61) N N
[NUMBER SG] [NUMBER PL]
cat SG cat PL
/ kæt / /s/
/ kæt /
The representation for cat given here indicates that there is no realization
for SG ; this can also be represented by linking SG to -∅.
(62) N
[NUMBER SG]
cat SG
/kæt/ /-∅/
The information at the top of these structure shows the grammatical prop-
erties of the words. The forms in the middle are the different lexical items
that constitute the word, while the forms on the bottom show the phonetic
spelling out of the combination of the noun and a number morpheme.
To the extent that a part of the actual word corresponds to a part of the
morphological structure, as in the case of cat-s, we can see the part-by-part
correspondence.
Here is how we would represent oxen using this kind of notation.
(63) N
[NUMBER PL]
ox PL
/Aks/ /n./
What about the paradigm for scissors? It is plural but there is no singular,
as the following shows.
(64)
NUMBER SG PL
Lexical items -0 -s
scissors -s
And for person, there are two plurals, one of which (people) is completely
unpredictable.

people
(65) We had a reservation for two .
persons
(66)
NUMBER SG PL
Lexical items -0 -s
person -0 people
person -0 -s
2.2.4.2. Case
Consider next the following expression in Russian:
(67) Russian
b’ela-ja čaik-a
white-FEM . SG. NOM seagull-FEM . SG. NOM
The line below the actual words is called the gloss. The gloss shows the
composition and translation of each word. In the case of b’ela-ja, for
example, the gloss consists of the root b’ela-, which means “white”, and
the ending -ja. The gloss FEM . SG. NOM indicates that this ending is used for
a feminine noun in the singular. NOM ( INATIVE ) is one possible value for
case, the morphosyntactic category that indicates the grammatical function
of a noun phrase in Russian and other languages. The notation “-” is
used to show how a word is made up of its parts; the notation “.” (as
in FEM . SG. NOM) is used to show that all of these properties are bundled
together in a single form and do not match up individually with part of the
word. 7
Russian has six cases: NOM ( INATIVE ), GEN ( ITIVE ), DAT ( IVE ),
ACC ( USATIVE ), INST ( RUMENTAL ), and PREP ( OSITIONAL ). So, in the
paradigm, the feature CASE has six possible values:
(68) a.
CASE NOM ACC GEN DAT INSTR PREP
š kol-
NUMBER SG škola školu školy škol’e školoj škol’e
PL školy školy škol školam školami školax
Compare this paradigm to the paradigm for the English personal pronouns
in (57). In English, none of the forms are predictable. But in Russian, a
7
For a comprehensive summary of the standard rules for glossing, see the
Leipzig Glossing Rules at www.eva.mpg.de/lingua/files/morpheme.html, as well as
Lehmann 1983 and Croft 2003.
particular phonetic form can be associated with the case per se. So it is
possible to state the case paradigm for the class of feminine nouns of which
škol- is a member much more generally:
(68) b.
CASE NOM ACC GEN DAT INSTR PREP
X-
NUMBER SG -a -u -y -’e -oj -’e
PL -y -y -0 -am -ami -ax
In other languages, such as Chinese, a noun has a single form regardless of

number or function.
2.2.4.3. Gender
As we mentioned just above, in some languages nouns are members of
classes, and membership in a particular class determines the form that the
noun takes for each cell in the paradigm. In languages with two or three
noun classes, the classes are typically gender classes: masculine, feminine,
and, if there is a third, neuter. These classes are called gender classes because
the forms that are used for nouns in a particular class are those that are used
for words that refer to animates that have biological gender. For example,
In French, the word for table has feminine gender, as indicated by the form
of the determiner la: la table “the table”. This is the same form that is used
for females, e.g. la fille “the girl”. The word for book is masculine: le livre
“the book”, because le is the same form that is used for males, e.g. le garçon
“the boy”.
It is important to recognize that gender is related to but not the same
as biological gender, or what we usually refer to as “sex”. There are two
biological sexes for living things, that is, animals and plants, namely male
and female. But inanimate things, and abstract things like ideas and beliefs,
and substances like water and wood, do not have sexes. However, in lan-
guages like French, all nouns have gender. Everything is either masculine or
feminine. This is not about the biology, it is about the way that the language
classifies the nouns.
Making the situation a bit more complicated is the fact that in some
languages there are three or even more noun classes. As the number of
classes gets larger, the connection with biological gender becomes more
tenuous. For example, Swahili has eleven noun classes; the class that a
particular noun is in depends in part on its physical or abstract properties.
Classes 1 and 2 contain for the most part nouns referring to humans, while
class 3 contains nouns referring to non-human non-animal things that have

life, such as trees, body parts, and supernatural phenomena. 8
English appears to expresses gender only in the words he/him, she/her,
his/her, and himself/herself. But this is not gender. In fact, we use he/him and
she/her, etc. to refer to people and animals, and the word that we use agrees
with the biological gender. He is used to refer to a single male individual,
and she to a single female individual. The pronoun it is used to refer to
things. English nouns do not have gender. But because all French nouns
have gender, the counterparts of he and she have to be used to refer to all
things according to their gender. Compare the following sentences.
(69) J’ai acheté un livre et il était
I have bought a book and it(masculine singular) was
cher.
expensive(masculine singular)
(70) J’ai acheté une maison et elle était
I have bought a house and it(feminine singular) was
chère.
expensive(feminine singular)
When we see the translation “it(masculine singular)” we might be tempted
to think that it is the same as he, and it is true that we would use the word il
in French to refer to a single male person, just as we would use him. But il
does not mean “him”, it means “him/it” and agrees with the gender class of
the noun, not just the biological gender. So, translating il as “him” and elle
as “her” in these sentences would be a mistake.
2.2.4.4. Representing lexical features in AVMs

We can collect all of the lexical and morphosyntactic features with their val-
ues and display them together as an attribute value matrix (AVM). Here is
the representation of the Russian word kniga “book”, a feminine noun with
NOMINATIVE case. For completeness we also put the phonetic properties
of the word in the matrix.
 
(71) kniga
 PHON 
 /kniga/ 
 
 CATEGORY NOUN 
 
 GENDER FEMININE 
CASE NOMINATIVE
The word kniga is actually composed of two parts, the root and the case
ending. We can use AVMs to represent this structure.
8
http://www3.iath.virginia.edu/swahili/swahili.html.
 
(72) kniga
 PHON 
 /kniga/ 
 
 CATEGORY NOUN 
 
 GENDER FEMININE 
CASE NOMINATIVE
   
knig- -a
 PHON /knig/   PHON /a/ 
   
   
 CATEGORY NOUN   CASE NOMINATIVE 
GENDER FEMININE GENDER FEMININE
As we develop the relationship between the lexicon and the syntactic repre-
sentation more fully, we will find AVMs to be a useful uniform device for
displaying the properties of words and phrases and their structure.
Summary: Frequently used glossing symbols
Category Symbol Meaning English example
Number SG singular a book, she

PL plural books, they
DU dual (‘the two of them’)
Person 1 first I, we, me, us
2 second you
3 third he, she, him, her, it, they, them
Gender F ( EM ) feminine she
M ( ASC ) masculine he
N ( EUT ) neuter it
Case NOM nominative I, he, she, they
ACC accusative me, him, her, them
DAT dative (‘to them’)
GEN genitive my, your, his, her, our, their
INST instrumental (‘with them’)
Tense PRES present goes
PAST past went
Aspect PROG progressive (is) going
PERF perfective (has) gone
2.3. HEADS AND PHRASES 43
At times we will use a less uniform and somewhat less explicit notation
for convenience and readability. For example, let us make the reasonable
assumption that the features of the parts of a word become the features of
the entire word (a relation called feature composition). And let us assume
that the phonetic representation of the parts of a word forms the phonetic
representation of the entire word through linear concatenation, so that, if
one part precedes another, its phonetic representation also precedes that of
the other. With these two assumptions, we can leave certain details out of
the representation, and use (73) instead of the more cluttered (72).
(73) kniga
   
knig- -a
 PHON /knig/   PHON /a/ 
   
   
 CATEGORY NOUN   CASE NOMINATIVE 
GENDER FEMININE GENDER FEMININE
2.3. Heads and phrases
As we have seen, a category consists of items that have the same distribu-
tional properties. Sameness of distribution can be seen not only for individ-
ual lexical items but for phrases. A phrase is a string of one or more words
that functions as a unit. In general, the grammatical behavior of a phrase is
determined by a lexical item that it contains, which is called the head. If we
consider first a phrase consisting of one word, e.g. a noun or a verb, we see
that the category of the word determines distribution of the phrase that it
defines within the sentence. In English, for example, a noun may function as
the subject or object of the sentence under certain conditions, while a verb
may function as the predicate or part of the predicate. In (74), the noun dogs
defines a phrase that functions as the subject, while bark and chase cats are
phrases that function as predicates (that is, they attribute a property to the
subject).
(74) a. Dogs bark.
b. Dogs chase cats.
The fact that bark and chase cats perform the same function in the sentence
suggests that they are of the same phrasal category. Since this phrase is
distinguished by the fact that it is based on a verb (bark or chase), it is called
a verb phrase. The verb is the head of the verb phrase. Correspondingly, the
noun dogs is the head of the phrase of category noun phrase. Henceforth,
we will use the abbreviations VP for verb phrase and NP for noun phrase.
There are phrases based on other categories as well, including AP (adjective
phrase) and PP (prepositional phrase).
Importantly, a phrase acquires not only the category of its head but other
properties. For example, if the head noun of an NP is plural, the phrase
is plural: the furry barking dogs. In general, syntactic phrases are headed,
in that the category of the phrase is determined by the category of one of
its parts. Typically, the phrase is of the same category as the head. Such
phrases are called endocentric. Phrases that do not contain heads of the
same category are called exocentric. A phrase that lacks a head entirely (if
such a thing is possible) is also exocentric.
2.4. ∗ The theory of linguistic categories
2.4.1. Justifying categories
The theory of linguistic categories is based on distribution and substitution.

However, it is obvious that two members of the same traditional category
do not necessarily have the same distribution. For example, only singular
nouns may have the determiners a and every.
(75) a dog
every dog
∗
a dogs
∗
every dogs
At the same time, members of different categories may have the same distri-
bution with respect to a particular test. For example, both verbs and prepo-
sitions select the objective or accusative form of the personal pronouns in
English.

him/∗ he
(76) see ∗
her/ she

him/∗ he
with ∗
her/ she
In spite of the sameness of distribution, we do not want to say that preposi-

tions and verbs are members of the same category.
2.4. THE THEORY OF LINGUISTIC CATEGORIES 45
Intuitively, it seems that in order to be members of the same category two

items must share all distributional properties. The fact that they share one
property, or several properties, does not appear to be sufficient. But this
very strict requirement runs up against the observation that members of the
same category do not always have the same distribution.
Verbs and adjectives share some properties, too, in English; for example,
the prefix un- can be attached to a verb or an adjective, but with very few
exceptions it cannot be attached to a noun or a preposition.
(77) untie [Verb]

unimaginative [Adjective]
∗
unafter [Preposition]
∗
undog [Noun] 9
While the general description of the properties of the category adjective

in English appears fairly straightforward, there are numerous problematic
cases. For example, an adjective modifies the noun that it precedes, as we
have seen. But nouns also have a modifying function when they appear
before another noun. Here are some examples.
(78) a. the birthday party

b. a glass window, rubber boots, a cotton dress
c. the Army training manual
d. a good career move
e. the film studio
f. a desk chair
g. an attack dog
These nouns that precede nouns have the function of describing or restrict-
ing the reference of the head – for example, a birthday party is a special
type of party. But birthday, glass, and so on do not distribute in other ways
like adjectives, in that they cannot be used predicatively and they cannot be
compared.
∗
(79) The party was birthday.
∗
I considered your party birthday.
∗
a more birthday party
etc.
9
The advertising slogan Uncola is an example of word play that takes advantage
of this restriction on the use of un-. The words unearth and unhorse are verbs, not
nouns, although their roots are nouns.
In fact, these words (birthday, glass, rubber, etc.) are not adjectives, but
nouns used restrictively. That is, they do not share the category member-
ship of adjectives, but they may have overlapping semantic functions with
adjectives. They arguably acquire this function in virtue of appearing in the
prenominal position, a property that they share with adjectives. This is a
very important point, because it highlights the fact that there is a difference
between syntactic category and semantic function. Failure to recognize this
distinction is likely lead to incorrect syntactic analyses in which two differ-
ent categories are conflated because they have some semantic functions in
common.
2.4.2. Universal categories
Substitution tests turn out to be fallible, because of paradigms and multiple

subcategories. The observation that certain categories overlap with one
another in certain respects suggests that the universal categories may not be
the unanalyzed Noun, Verb, Preposition, Adjective, etc. Rather, it has been
suggested that there are supercategories that may correspond to higher level
distributional properties. 10 For example, the set of words that may show
morphological marking for case (the nouns and adjectives in our traditional
terminology) are members of a single supercategory. Call this category
[+N]. Anything that is not in this supercategory is in the supercategory [−N]
(e.g. a verb). Similarly, the verbs and adjectives are members of a single
supercategory, because they take complements and assign case. Call this
category [+V]. Adjectives are members of both [+N] and [+V], while verbs
are [−N] and [+V].
Here is the full range of possibilities covered by the feature theory of
categories.
(80)
T RADITIONAL CATEGORY [N] [V]
Noun +N −V
Adjective +N +V
Verb −N +V
Preposition −N −V
10
The original proposal is due to Chomsky 1970. See also Jackendoff 1977.
Membership in the supercategories +V, −V, +N, and −N captures the

similarity in distribution of nouns and adjectives, of verbs and adjectives,
of verbs and prepositions, and of nouns and prepositions.
The purpose of the feature theory of categories is to capture the fact
that members of different traditional categories have similar distributional
properties. But the very simple account just outlined does not capture the
rich diversity of distributional overlap that is found in natural language.
For example, Jackendoff 1977 notes that both nouns and verbs may take
subjects, as seen in the following examples.
(81) a. The committee discussed the proposal.

b. the committee’s discussion of the proposal
The possessive in the nominalization bears the same relationship to the

noun that the subject does in the full sentence. On this basis, we might want
to say that there is a supercategory containing nouns and verbs, but not
adjectives and prepositions. But this cannot be captured using the feature
scheme given in (80).
While verbs and adjectives are both [+V] in this scheme, and verbs and
prepositions are [−N], only the verbs inflect for tense and number agree-
ment. Nouns, verbs, adjectives, and prepositions may take prepositional
phrase complements and sentential complements, suggesting that there is
a supercategory that contains all of these traditional categories.
(82) a. our anger at the decision

b. look at the screen
c. angry at the decision
d. from under the table
(83) a. our anger that they never called
b. angry that they never called
While nouns and prepositions do not appear at first glance to share any fea-
tures, noun phrases and prepositional phrases, as well as adjective phrases
and adverb phrases, appear in sentence-initial position in questions, while
verb phrases do not. We indicate the category of each phrase by labeling the
left bracket enclosing it.
(84) a. [NP Which students] were you talking to?

b. [PP To which of the students] were you talking?
c. [AP How tall] do you think she is?

d. [AdvP How fast] do you think he can run?
e. ∗ [VP Talking to which students] were you? 11
These observations suggest that there may be other supercategories

besides the ones that we originally introduced. But the cases that we have
examined thus far, which are fairly representative, suggest that there is a
problem using features to define categories that are based on the distribu-
tional properties of different types of phrases. The reason is that there are
many, perhaps an unlimited number, of distributional contexts that can be
evaluated for the purpose of categorization. For each such context, all or a
subset of the members of a traditional category may be compatible with the
context. Since many of the contexts are peculiar to a given language, and not
universal, this distributional diversity undermines the original assumption
that there is a small, finite number of universal categories that all languages
choose from.
2.4.3. Tests for categories
In the end, the issue about categories is whether it is possible to state

exceptionless conditions under which a word or phrase is or is not a mem-
ber of a given category. To do so would be to define the category, either
universally or with respect to a given language. But if no such conditions
can be given, then the very foundation of linguistic categories is called into
question. Along with this problem is the question of whether the conditions
are syntactic (distributional), morphological, or semantic, or a combination
of these. We use nouns to illustrate the point, although any traditional
category would serve.
There are basically three types of nouns in English: count nouns, such
as dog, mass nouns, such as sincerity, and proper nouns, such as Albert
Einstein. There are certain distributional tests on which these different types
of nouns behave differently, which would suggest that they are not members
of the same category, e.g.
11
This example is similar to utterances by Yoda (in the Star Wars movies), but is
not grammatical in English.
∗
(85) the dog the sincerity the Albert Einstein
∗
see dog express sincerity discuss Albert Einstein
∗ ∗
a lot of dog12 a lot of sincerity a lot of Albert Einstein13
∗
dogs sincerities Albert Einsteins14
While it is possible to find tests that group two of these noun types, it is
hard to find tests that group all three of them. Thus, the counterintuitive
conclusion seems to be that there is no category “noun”.
There does not appear to be a common semantic property that all nouns
share that can be used to distinguish them from all other categories. It is
true that nouns refer to things, and that proper nouns refer to particular
individuals. But there are many nouns that do not refer to things per se, so it
is not possible to reduce the category “noun” to a simple semantic criterion.
For instance, Christmas refers to a holiday and a time of the year, anger
refers to an emotional state, victory (over someone) refers to a relation, Paris
refers to a place, and so on.
Consider next morphology. Since some languages have overt morphology
for case and agreement and others do not, it is not possible to define
categories such as “noun” in terms of whether they show morphological
case, since there are many languages in which they do not. Moreover, as we
have noted, in languages such as Russian, adjectives show morphological
case as well as nouns. So we cannot use morphology as a universal basis for
defining the categories.
To summarize, it appears that neither syntactic distribution, semantic
properties, nor morphological properties are sufficient to define syntactic
categories for a single language, let alone across languages. At this point we
appear to have a paradox. On the one hand, it is clear that there are no nec-
essary and sufficient distributional or semantic conditions that will allow us
to define the familiar categories noun, verb, preposition, and adjective. On
12
This phrase is acceptable if we understand dog as a mass noun, as in that’s a
lot of dog you’ve got at the end of that leash, my friend.
13
This phrase is acceptable in the idiomatic I’ve been seeing a lot of Albert
Einstein lately. Its interpretation is literally: “I’ve been seeing Albert Einstein a lot
lately”.
14
This phrase is acceptable when it refers to the Albert Einstein family (We had
the Albert Einsteins over for dinner), when it refers to a number of people with the
name “Albert Einstein” (Every year all of the Albert Einsteins gather in Zurich), or
when it refers to multiple instances of the same person (It would be nice if there were
many Albert Einsteins in our physics department).
the other hand, it appears that nouns and verbs are universal in the world’s
languages, and that many if not all have prepositions (or postpositions) and
adjectives. So the question is, where do these universals come from?
2.4.4. A paradox resolved?
Let us try to resolve this paradox. It is true that there are no semantic
criteria that are sufficient to define the syntactic categories exactly. But it
still is plausible that the syntactic categories are in some way a reflection
of basic universal semantic categories, even though the syntactic categories
extend beyond these basic semantic categories. For example, the category
noun reflects the fact that there are physical objects, the category verb reflects
the fact that there are actions and other types of relations that involve
these objects, the category preposition (and postposition) reflects the fact that
there are spatial relations with respect to objects, and the category adjective
reflects the fact that they have properties.
At first glance it might seem that these observations do not really help
us solve the paradox, because they are not complete accounts of what sorts
of words are members of these categories. Not all nouns denote physical
objects, not all verbs denote events, not all prepositions involve spatial
representation, not all adjectives denote properties, and not all languages
have prepositions and adjectives.
Our paradox results from our attempt to make the semantic and the
syntactic categories match exactly. But we can get around the paradox if
we assume that each syntactic category in a language is defined not by the
properties of all of the words in the category but by a well-defined restricted
subset of the words in the category.
This subset consists of common words that are used in speech to young
children. Let us call this the concrete subset. The concrete subset con-
sists of the nouns that refer to things, the verbs that refer to physical
actions, the prepositions or other forms that refer to spatial representa-
tions, and the adjectives or other forms that refer to concrete properties of
things.
Crucially, on this view the semantic, distributional and morphosyntactic
properties of each category are defined strictly in terms of the concrete
subset. For example, on the basis of forms such as a dog, referring to
a single observable dog, the dog, referring to a single definite dog, and
more generally determiners such as my, this, some, etc., it is possible to

hypothesize that a preceding determiner is a marker for a noun phrase.
(86) a. Oh, I see a dog!
b. Can you point to the dog?
c. Look at the dogs!
d. There are some dogs in the pond!
Moreover, since dogs are physical objects, we might hypothesize that words
that denote similar concepts, like cat, cow, lion, penguin, and so on, are
members of the same category. On this basis is it immediately possible to
reliably hypothesize that for any word W, if we hear, “Oh, I see a W!”, or
“Can you point to the W?”, or “Look at the Ws”, that W is a noun. We can
form this conclusion for nonsense words as well –
(87) Oh, I see a wug! Can you point to the wug? 15
– and for words that do not actually correspond to anything in the physical
world.
(88) a. I have an idea.
b. You hurt my feelings.
The category of every word not in the concrete subset of nouns may thus
be defined in terms of the properties that can be identified on the basis of
the concrete subset. In other words, if we first hypothesize the category N
on the basis of these properties, we can account for the distributional and
morphological properties associated with this category in English without
requiring all nouns to refer to physical objects in the world.
The same preliminary category will be formed in every language, since all
languages are spoken in an environment where there are physical objects.
But in other languages there will be different distributional conditions that
define the category.
This theory of categories is a case of what is called a bootstrapping theory.
On the basis of basic semantic properties a category is formed and then
extended (“semantic bootstrapping” – see Pinker 1984). Then, on the basis
of the morphological and distributional properties of the words in this cat-
egory, the category is extended to new words (“syntactic bootstrapping” –
see Landau and Gleitman 1985).
15
The ability of children to correctly classify nonsense words such as wug and
apply number morphology to them was demonstrated in a classic study by Gleason
1958.
The idea of bootstrapping is attractive because it allow us to locate the

universality of syntactic categories in the semantics and break out of the
paradox that arises if we assume that it is a strictly syntactic universality.
This approach assumes that there are a few fundamental primitive semantic
categories (objects, actions, space, and properties) that underlie the forma-
tion of syntactic categories.
At the same time, syntax is distinct from semantics, and the categories
that are ultimately formed in syntax do not correspond exactly to semantic
categories, as we have seen. Especially when we consider the minor cate-
gories, we see that there are syntactic categories defined in distributional
terms that do not completely match semantic categories.
Moreover, it is possible to express the same semantic notion using expres-
sions of distinct semantic categories. For example, quantity in English can
be expressed by words such as every, each, all, and three (and other num-
bers), each of which has slightly different properties from the others (see
Problem 3).
In sum, while it is impossible to state conditions that precisely define
each of the syntactic categories of a language, it is possible to define the
categories in terms of the concrete subset of the words in the language, and
then use the properties of these words to define the categories further.
Exercises
1. Use the labeled bracket notation and the tree notation that we intro-
duced in section 2.1.1 to label the words in the following strings. The word
a is a member of the category that contains the.
(1) a. build a tall hill of bananas
b. deliver a huge pizza to Sandy
c. read a story about chimpanzees
d. buy a new outfit with the money
As an illustration, here is the answer for the first three words of example (a).
[V build] [ART a ] [ADJ tall] . . .
V ART ADJ . . .
build a tall
[§2.1.]
EXERCISES 53
2. In English, suffixes may be attached to words to create new words.

(This is called derivational morphology.) For example, the suffix -er may be
added to a verb to create a noun meaning “one who”; e.g. swim ∼ swimmer.
For each of the following suffixes, say what category it attaches to, what
category it creates, and what meaning it contributes. Give three examples
for each morpheme and two pieces of distributional evidence to show what
category it creates. Try to find examples in which the meanings of the related
words are systematically related to one another, as in the examples swim ∼
swimmer. Here is an example involving -er:
(0) swim ∼ swimmer

run ∼ runner
eat ∼ eater
-er creates nouns out of verbs (V+-er → N)
For a verb V it means “one who Vs”.
Use of the article (the swimmer) and the singular/plural alternation (swim-
mer/swimmers) show that the result of adding -er is a noun. The only appears
with nouns, and -s is the plural marker for nouns.
(1) a. -ness f. -ment
b. -less g. -al
c. -able h. -ly
d. -est i. -ity
e. -ing j. -ish
[§2.1.1.]
3. The English morpheme -er has more uses than those discussed in
Exercise 2, as shown by the following examples.
(1) a. Leslie is taller than Sandy.

b. Sandy runs faster than Leslie.
For these uses of -er, (i) say what its meaning is, (ii) what category it applies
to, and (iii) what category it creates. (Follow the model in Exercise 2.) Give
two pieces of distributional evidence for your answers in each of (ii) and
(iii) to justify the category or category that -er applies to and the category
or categories that it creates.
[§2.1.]
4. English has prefixes that may be attached to words to create new words,
like un-, which turns adjectives into adjectives (e.g. true ∼ untrue). The
meaning of un- is “not”.
For each of the following prefixes, say what category it attaches to, what
category it creates, and what meaning it contributes. Give three examples
for each morpheme and two pieces of distributional evidence to show what
category it creates. Try to find examples in which the meanings of the related
words are systematically related to one another. For a model of what your
answer should look like, see Exercise 2.
(1) a. re- f. dis-

b. anti- g. pre-
c. out- h. under-
d. counter- i. over-
e. mis- j. de-
[§2.1.]
5. Determine what lexical category or categories each of the following
words belongs to in English.
(1) a. awake
b. alone
c. womanly
d. trashy
e. seldom
You will need to construct example sentences to discover the distribution of

each word. Give as much evidence as possible for your answer, looking at
the distribution, morphology (inflections) and function of the words.
[§2.1.]
6. Using the table in (56) as a guide, add rows for the following English
verbs.
be
have
speak
sing
bring
dive
You should find that for the verb be the table needs to be made slightly more
complicated than it is in the text in order to accommodate all of the forms.
Moreover, there may be some disagreement among native speakers about
what the correct forms are for some of the other verbs.
[§2.2.3.]
EXERCISES 55
7. Add a row to (57) for the personal interrogative pronouns who, whom,
whose. Explain why you put each form where you did.
[§2.2.3.]
8. The following examples illustrate a case of allomorphic variation
in German. There are two abstract morphemes for grammatical number
shown here, SG and PL . List each of the distinct allomorphs of PL with its
distinguishing properties. (The vowel “ä” is the fronted counterpart of the
back vowel “a”.)
Word Gloss Meaning
Arbeiter worker-SG worker

Arbeiter worker-PL workers
Laden shop-SG shop
Läden shop-PL shops
Arm arm-SG arm
Arme arm-PL arms
Hand hand-SG hand
Hände hand-PL hands
Bild picture-SG picture
Bilder picture-PL pictures
Reise trip-SG trip
Reisen trip-PL trips
Mensch human being-SG human being
Menschen human being-PL human beings
Auto car-SG car
Autos car-PL cars
[§2.2.3.]
9. Using the feature notation introduced in the text for the English num-
ber paradigm (for example as in (60)), notate the adjective paradigm for tall
that comprises the basic adjective, the comparative, and the superlative. Call
the feature SCALE , and the values BASIC, COMP, and SUPER .
[§2.2.4.1.]
10. Provide glosses for the English pronouns in (1) below. An example is
given in (0) below to get you started. Use NOM for pronouns that function
as subjects, ACC for pronouns that function as objects, GEN for possessives,
and REFL for reflexive pronouns. If a form can be glossed in more than one
way, give all of the glosses.
(0) my
1.SG. GEN
(1) a. myself
b. us
c. your
d. himself
e. it
f. she
g. ourselves
h. her
i. I
[§2.2.4.]
11. Consider the following sentences.
(1) a. It was a giant hill of bananas that I built.
∗
b. It was a huge pizza to Sandy that I delivered.
c. It was a story about chimpanzees that I read.
∗
d. It was a new outfit with the money that I bought.
What conclusion do you draw from these judgments, given the category
assignments of Exercise 1? Explain your reasoning.
[§2.3.]
Problems
1. In sentence (1a) below up the street is a unit and in (1b) up the battery
is not. That is, they have different structures even though they superficially
look the same. The reasoning is based on the examples in (2)–(3).
(1) a. I charged up the street.
b. I charged up the battery.
∗
(2) a. I charged the street up.
b. I charged the battery up.
(3) a. It was up the street that I charged.
∗
b. It was up the battery that I charged.
Explain in your own words the reasoning behind this statement. That is,
explain how to arrive at this conclusion about the difference in structure
given the pattern of grammaticality and ungrammaticality in (2)–(3).
[§2.1.4.]
PROBLEMS 57
2. Make up your own examples that illustrate the structural difference

that is shown in Problem 1. The word up in (2b) is called a particle. Find
at least two more English particles. The key is to find a preposition, like up,
that may also function as a particle. Such a preposition/particle may appear
before or after a noun phrase, as the examples in Problem 1 show.
[§2.1.4.]
3. Based on their distribution and behavior with respect to substitution
tests, what category or categories do the following English words belong
to? Do they belong to one category, or are there several smaller categories
exemplified here? Give evidence to support your answers. The examples in
(2) provide some partial information to help you get started.
(1) each both

every many
any some
all three

each
(2) a. ∗ of my friends
every

three
b. these ∗ answers
some

all
c. My classes are ∗ some interesting.
∗
three
[§2.3.]
4. It has been suggested that the following instances of all and both are
alternative forms of all of and both of.
 

 the 

these
(1) a. all books

 my 

Terry’s
 

 the 

these
b. both books

 my 

Terry’s
Other quantifiers do not share this distribution.

 
 many   

   the
 some 
  

these
(2) a. two of books

   my 
 any 
   Terry’s 

each
 
 many   

   the
 some 
 

these
b. ∗ two books

 
 
 my 


 any 
 Terry’s
each
What do the following examples suggest about the relationship between

all and all of ? Does it suggest that they are variants of the same
both both
construction, or different (but synonymous) constructions? Justify your
answer.

all ∗
(3) a. I want to buy (of) them.
both

all ∗
b. the students, (of) whom were well behaved, . . .
both

all ∗
c. Which books did you try to buy (of).
both
(The notation ∗ (. . . ) here means that the string is ungrammatical when the
material in parentheses is omitted.)
[§2.4.]
Research questions
1. In the text we cited fall and rain as instances of 1- and 0-argument

verbs, respectively. We showed that fall cannot appear with a direct object,
and rain appears with neither object nor subject.
(1) a. ∗ I fell my dinner.

b. ∗ I rained.
c. ∗ It rained my dinner.
The following examples appear to be counterexamples.
(2) a. The package fell a great distance.

b. It rained cats and dogs.
How do you reconcile these two groups of examples? (Hint: What is the
proper way of specifying how many and which arguments a verb takes?
Also, the NPs in (2) cannot be replaced by what.
(3) a. ∗ What did the package fall?

b. ∗ What did it rain?
RESEARCH QUESTIONS 59
but

What distance
(4) a. did the package fall?
How far
b. How much did it rain?
The meaning of rain is explored further in Problem 4 of Chapter 5.)

[§2.1.2.]
2. In the text we saw some adjectives (present, alleged, perfect) that cannot
be used predicatively. Make a list of as many other adjectives you can think
of that cannot be used predicatively, and then try to determine whether they
have any properties in common that might be used to explain their behavior.
You might want to take into account the observation that the ones that we
have already noted cannot be quantified, e.g.
∗
(1) the more present King of France
∗
the more alleged assassin
∗
the more perfect idiot
[§2.1.3.]
3. What are the meaning differences, if any, between the following sen-
tences?
(1) a. Stupidly, Sandy is trying to stand on one foot.
b. Sandy stupidly is trying to stand on one foot.
c. Sandy is stupidly trying to stand on one foot.
d. Sandy is trying stupidly to stand on one foot.
e. Sandy is trying to stupidly stand on one foot.
f. Sandy is trying to stand stupidly on one foot.
g. Sandy is trying to stand on one foot stupidly.
Be as precise as you can about what the differences are between the mean-
ings of the adverbs when they are in different positions. Do these differences
help explain the different distributional properties of sadly in the text (see
example (39)).
[§2.1.5.]
4. Some adjectives may function as adverbs without the addition of -ly,
some must, and some may not.
 

 fast 


 ∗ 


 fastly 

slow
(1) a. She runs very .
 slowly 
 

 


 ?quick 

quickly
 
 tight 
 
tightly
b. You have to grab the handle ∗ .
 firm 
 
firmly
A. Make an inventory of common adjectives that must or may lack -ly.

B. Do the adjectives that may or must lack -ly share any semantic prop-
erties?
C. For those adjectives that may lack -ly, is there a meaning difference
depending on whether -ly is present?
D. What is the distribution of adjectives that lack -ly in the sentence,
compared to that of adjectives that have -ly?
[§2.1.5.]
5. In the text we suggested that there are no distributional, semantic, or
morphological criteria that can be used to unambiguously define member-
ship in specific lexical categories. Is this claim true for prepositions? Justify
your answer.
[§2.4.]
Section Exercises Problems Research questions
2.1. 1, 2, 3, 4, 5 1, 2 1, 2, 3, 4
2.2. 6, 7, 8, 9, 10
2.3. 11 3
2.4. 4 5
3
Basic sentential structure
In this chapter we lay out some basic structural properties of simple sen-
tences. We assume for this purpose that sentences contain phrases such as
noun phrases (NPs), verb phrases (VP), and prepositional phrases (PP),
although we have not worked out the internal structure of these phrases.
Once we have developed the basic sentential structure in this chapter, we
will turn in Chapter 4 to how these and other phrases are made up.
3.1. Methodological preliminaries
Contemporary theorizing about linguistic structure draws on the presump-

tion that this structure is abstract, that it is in the mind, and that speakers
are able to use this structure in order to form intuitions about whether a
string of sounds and words is a legitimate expression in their language.
This presumption stands in contrast to the notion that the structure is
somehow in the speech signal itself. To illustrate, let us consider a simple
phrase consisting of more than one word, like go home. If we examine the
physical speech signal, we find that in general it is impossible to find places
in the sound stream where the boundaries between words are marked. That
is, the space between go and home does not correspond to any recognizable
feature of the speech signal. So to the extent that it is correct to say that a
linguistic expression has internal structure, e.g. can be broken up into words,
this structure is abstract – it is not physically present in the speech signal. In
effect, we as speakers and hearers impose structure on the speech signal in
virtue of our knowledge of the grammar of the language. Structure is not in
the speech signal, it is in the mind.
This abstractness holds for all linguistic structure, and not simply the
level at which we analyze a string of sounds into a sequence of words.
62 3. BASIC SENTENTIAL STRUCTURE
Within the word there is recognizable morphological structure, as we saw

in Chapter 2. For example, dogs is made up of the morphemes dog and -s,
and unresponsiveness is made up of the morphemes respons(e), un-, -ive,
and -ness. But, again, there is nothing in the physical signal that marks the
boundaries between morphemes.
By the same token, syntactic structure is abstract – that is, it is not marked
in any overt way in the sound stream. Consider a simple sentence like (1).
(1) Many people like dogs.
There is nothing in the string to indicate that many people is a unit, or that
like dogs is a unit. The conclusion that there is structure, and precisely what
that structure is is determined on the basis of linguistic intuitions.
Some of these intuitions are of the type that we discussed in Chapter 2.
For instance, substitution of we for many people provides evidence that many
people is a unit, since it can be replaced by a one-word phrase without
affecting grammaticality. Substitution of sleep for like dogs provides similar
evidence that like dogs is a unit.
On the basis of such intuitions, it is possible to bracket the string of
words into the individual units. The consequence for (1) is given in (2). We
deliberately put aside at first any attempt to categorize the various units.
Notice that smaller units combine to form larger, more complex, units:

(2) [many][people] [like][dogs]
Just as we used labeled brackets in Chapter 2 to indicate the category of

a word, we can use labeled brackets to indicate the category of phrases. In
the following illustration, we used different-sized brackets and lines to show
how the brackets match up.
(3) S NP QUAN many Npeople VP Vlike NP Ndogs
(4) label . . . stands for . . .

S sentence
NP noun phrase
QUAN quantifier
N noun
VP verb phrase
V verb
3.1. METHODOLOGICAL PRELIMINARIES 63
Labeled bracketings can be difficult to read, even for relatively simple

sentences like Many people like dogs. Therefore we often find it useful to
represent the structures of phrases and sentences in terms of trees. Here is
the tree for the noun phrase many people.
(5) NP
QUAN N
many people
We draw a double line from the label NP to the label N to highlight the fact
that the noun phrase is built around a noun. The noun is called the head
of the NP. Notice how the labeled bracket that surrounds a word or phrase
corresponds to the label of the part of the tree that contains that word or
phrase. For example, QUAN is the label for many, and it appears above
many in the tree, while NP is the label for many people, and it, too, appears
above many people in the corresponding tree.
Here is the tree for like dogs. 1
(6) VP
V NP
like N
dogs
And here is a tree for the whole sentence.
(7) S
NP VP
QUAN N V NP
many people like N
dogs
1
Evidence that VP is a constituent of the English sentence is given in
section 3.6.
Terminology: trees
Here is some terminology for talking about trees:
r The diagram in (7) is called a phrase structure tree, or more simply a tree, or a
phrase marker.
r The labeled branch points in a tree, such as S and VP, are the nodes in the tree.
r A line from one node to another is called a branch.
r A node that has more than one branch is called a branching node.
r A node that has exactly two branches is called a binary branching node.
r The parts of a phrase that make up its structure are called its constituents.
r The tree in (7) says that an S has as its immediate constituents (or daughters) an
NP and a VP. S is said to immediately dominate NP and the VP. VP immediately
dominates V and an NP.
r The node that immediately dominates some node X is called the mother of X.
r Two nodes that have the same mother are sisters.
r S dominates every node in the tree that it is connected to it by a path of
immediate domination – hence in (7) the node S dominates what it immediately
dominates, namely the NP many people and VP, as well as what VP immediately
dominates, namely V and the NP dogs.
3.2. The simple sentence
Next we develop a way of stating explicitly what the structure of a phrase of

a given type is. We begin with some simple sentences in English.
(8) a. President Smith called.
b. The dog chased the cat.
Typically a simple sentence is about something. 2 What that thing is is

expressed by the subject, and what the sentence says about that thing is
expressed by the VP. Typically the subject is an NP, but it could be some-
thing else, and typically the predicate is a VP, in English. The VP may be
simple, consisting just of a verb as in (8a), or it may be more complex, as in
(8b).
General statements about the internal structure of constituents of each
type may be expressed in terms of phrase structure rules (PSR). A PSR says
2
Henceforth, we will speak informally of parts of a sentence referring to some-
thing even though there might not actually be a thing in the world that is referred to.
3.2. THE SIMPLE SENTENCE 65
what the possible make up of a phrase of a given category is, in terms of

its immediate constituents. For example, having established that an English
sentence (which we are calling S) may be composed of an NP and a VP, in
that order, we express this in terms of the following statement.
(9) S → NP VP
This is a first approximation of the rule for S.
Labeled bracketings, trees, and rules

Given part of a tree A
B C
the corresponding labeled bracketing is: [A B C ],
and the phrase structure rule that permits this structure is: A → B C.
Note that rule (9) says that, in an English sentence, an NP may precede
a VP. It does not say that the reverse order, that is, VP NP, is possible in
English. If such an order were possible, we would expect the following to be
a sentence of English, but it is not.
∗
(10) Chased the cat the dog.
What would be the form of a rule that allows for the order VP NP in a
language? See Exercise 7.
Rule (9) is not exhaustive, in that there are other ways in which a sentence
may be constituted. Most notably, it is possible to have an auxiliary verb
before the main VP, as in the following.
(11) a. President Smith will call.
b. The dog is chasing the cat.
We describe sentences of this type in terms of the two phrase structure rules
in (12).
(12) a. S → NP AUX VP
b. AUX → VAUX
We will see as we proceed that this formulation is considerably oversimpli-

fied and preliminary, but it will serve for the present.
This initial characterization of the simple sentence highlights the funda-
mental distinction between
r grammatical category
r grammatical function
r semantic role.
The phrase the dog is a phrase based on a noun. Thus, it is a noun phrase,
abbreviated NP. NP is the grammatical category of the phrase. An NP may
appear in a number of different places in a sentence. For example, it may
appear in the position where President Smith appears in (8a) and where the
cat appears in (8b). It may appear with a preposition, as in with the cat or to
President Smith. It may have a number of different grammatical functions,
such as subject and direct object. But it is always an NP.
An NP is also the subject of each of the sentences in (8). This means
that the NP occupies a particular position in the grammatical structure
of the sentence, appears in a particular order with respect to the other
constituents, and plays a particular role in the sentence. For example, an
NP that is the subject determines the form of the verb in the present tense,
e.g. the dog is/∗ are and the dogs ∗ is/are.
It may be possible for phrases of category other than NP to function as
the subject of a sentence in a language. For example, it is possible that doing
syntactic research is a VP and not an NP, but in any case it is the subject of
the sentence in (13).
(13) [Doing syntactic research] gives me a headache.
It is very important not to confuse the grammatical function with the

syntactic category. In English, if a phrase is an NP it should be able to
function as a subject, but if a phrase is functioning as a subject, as in (13),
that does not in itself mean that it is an NP. Non-NPs may function as
subjects simply in virtue of their meaning and position in the structure, as
in (13).
Finally, there is the semantic or thematic role associated with the phrase.
A subject may correspond to a number of different roles depending on the
semantic properties of the verb. Here are some examples in which the same
NP, the dog, is the subject of the sentence, yet has a different semantic role
in each case.
(14) a. The dog chased the cat.

[the dog is an Agent, it initiates the action]
b. The dog was chased by the cat.
[the dog is a Patient, it is affected by the action]
3.2. THE SIMPLE SENTENCE 67
c. The dog got sick.

[the dog is a Theme, it undergoes a change of state]
d. The dog received a special treat.
[the dog is a Recipient, it comes into possession of something]
As we will discuss in more detail in Chapter 5, the role assigned to a phrase

depends on the lexical properties of the verb and the grammatical function
of the phrase with respect to that verb.
The phrases called and chased the cat in (8) also have a grammati-
cal category and grammatical function. Their syntactic category is VP.
Their grammatical function in these sentences is that of predicate, which
is complementary to subject – the predicate expresses the property that
is attributed to the subject in the sentence. Together, the subject and the
predicate constitute the sentence. In English, the predicate of a main clause
must be a VP.
The sentence in turn expresses a proposition, that is, a complete idea.
Putting all of this together, we can say that an English sentence may be
composed of an NP and a VP, that the function of this NP is subject and
the function of the VP is predicate, and that the NP has a semantic role
determined by the VP. We summarize these relationships in the following
diagram. 3
(15) S: Proposition
NP: Subject VP: Predicate
V ...
At this point, it will be useful for us to examine in a more systematic

way what kinds of intuitions we consult in positing syntactic structure. In
the next section, we look at the different kinds of syntactic functions that
constituents of sentences may have. In sections 3.4–5, we consider the ways
in which the basic grammatical functions of phrases are distinguished in
languages. In section 3.6, we review some standard diagnostics that can
be used to argue that a phrase or a sentence has a particular internal
structure.
3
This diagram and the discussion leading up to it is inspired by the treatment of
basic English structure in Huddleston and Pullum 2002.
3.3. Complements, arguments, and adjuncts
Consider now the verb phrases called and chased the cat in (8). These
examples show that a VP may consist of a V alone or of a V followed by
an NP. We require two phrase structure rules, one for each type of VP.
(16) VP → V
(17) VP → V NP
Notice that the VPs described by these two rules share two properties. One
is that, in both cases, there is a V in the VP. And the other is that the V is
the first constituent in the VP. We express the fact that these two types of
VPs share these properties by abbreviating the two rules into one rule.
(18) VP → V (NP)
There are many other possible structures that VPs may have. Here are a
few of them.
(19) a. go to Chicago [VP V PP]
b. give Sandy a book [VP V NP NP]
c. put the groceries on the table [VP V NP PP]
d. think it will rain [VP V S]
What we see is that the VP contains a V in initial position, regardless of

what else it contains. These VPs are phrases that are all based on verbs.
Therefore, the verb is the head of the VP.
The other constituents of the VPs in these examples are called comple-
ments of the V because they satisfy semantic and syntactic requirements of
V – they “complement” the verb. The complement of eat is an NP that refers
to what is eaten. The complement of go is a PP that expresses direction. The
complement of think has the form of a sentence that expresses the content
of the thought. The verb put takes two complements, one of which is an NP
that refers to what is being put and the other of which is a PP that refers to
its location.
The following rules state that these are possible ways to form a VP, in
addition to (18).
(20) VP → V PP
VP → V S
VP → V NP PP
We put off attempting to abbreviate all of these rules for the Engllish VP
until Chapter 4, where we look at the internal structure of VP in more detail.
3.3. COMPLEMENTS, ARGUMENTS, AND ADJUNCTS 69
The arguments of a verb are the complements of that verb and its subject.
In contrast with arguments are adjuncts. An adjunct expresses a refinement
of the meaning of a phrase; it is not an integral component of it. So, when
the adjunct is omitted, the meaning of the phrase is less specific, but not
incomplete. No ungrammaticality arises. For example,
(21) a. I was sleeping.
b. I was sleeping in the kitchen.
c. I was sleeping soundly.
d. I was sleeping soundly in the kitchen.
In these examples, in the kitchen is an adjunct that specifies the location of

the event, while soundly is an adjunct that specifies the manner. Adjuncts
in sentences typically are used to denote notions such as time, location,
manner, means, purpose, and reason.
As the following examples show, many of the same syntactic and semantic
categories may be used as a complement or adjunct. The difference is
whether or not the phrase is required by the verb to complete the meaning.
The verb put takes as an obligatory complement an expression denoting
location; that is, put is a relation between an Agent (the one who puts), a
Theme (what is put), and a Location. It is not possible to say what you are
putting without saying where you are putting it.
On the other hand, a location may also specify where an event takes place
or a relation holds, in which case it is an adjunct.
(22) Complement: I put the book on the table.
∗
I put the book.
Adjunct: We rarely play cards on the table.
We rarely play cards.
In contrast to putting, it is not necessary to give the location of the card-

playing.
The preceding discussion raises the possibility that arguments are oblig-
atory and adjuncts are optional. But a broader examination of the facts
shows that there is no such simple correlation. Some arguments are obliga-
tory, while others are optional. For example, eat does not require a comple-
ment (23b), while devour does (24b).
(23) a. We were eating the pizza.
b. We were eating.
(24) a. We were devouring the pizza.
∗
b. We were devouring.
While adjuncts are typically optional, in some cases they are obligatory. For
example, the verb type does not require a manner adverb (25b), while the
verb word does (26b).
(25) a. Sandy typed the letter carefully.
b. Sandy typed the letter.
(26) a. Sandy worded the letter carefully.
∗
b. Sandy worded the letter.
3.4. Grammatical functions
3.4.1. Structural grammatical functions
The core grammatical functions (GFs) are subject (SU), object (O) and indi-
rect object (IO). The subject in English is usually the NP that immediately
precedes the VP –
(27) Sandy is sleeping.
SU
– the object is typically the NP that immediately follows the V when there is
no indirect object –
(28) Open the door!
O
– and the indirect object is the first NP when there are two argument NPs
following the V.
(29) Sandy gave Chris the money
IO O
Where the subject, object, and direct object appear in a sentence is a conse-
quence of how they are arranged in the structure of the sentence.
As we have already noted, the subject NP and the VP are the major
constituents of the sentence, while the object forms a VP with the V.
(30) S
NP1 VP
V NP2
3.4. GRAMMATICAL FUNCTIONS 71
Notice that the subject is the NP that is the sister of VP, and the direct object
is the NP that is the sister of V.
The indirect object is also a constituent of VP, and hence is dominated
by VP. We will assume for now and show later that it is also immediately
dominated by VP, so that there is no additional branching structure in VP.
That is, the structure is (31).
(31) S
NP1 VP
V NP3 NP2
The question of whether there is more branching structure in VP is one that

we take up in more detail in Chapter 4. In general, when there is no evidence
for a more complex structure, we assume that the structure is “flat”.
3.4.2. Tests for subject
It is important to show that a phrase actually bears a certain structural

function with respect to the sentence and is not simply in a particular
linear position with respect to the other constituents (for example, “at the
beginning of the sentence” or “immediately before the verb”). There are
several tests that can be used to demonstrate that something is a subject in
English.
First, the subject agrees with the verb in number. If the subject is singular
(for example Sandy) the verb is marked singular (for example is), and
similarly for plural (the students ∼ are).

is
(32) a. Sandy ∗ tall.
are

is
b. Sandy ∗ writing a book.
are

has
c. Sandy ∗ received a letter.
have

has
d. Sandy ∗ fallen.
have

is
e. It ∗ raining.
are
∗
is
(33) a. The students tall.
are
∗
is
b. The students writing a book.
are
∗
has
c. The students received a letter.
have
∗
has
d. The students fallen.
have
The verb does not agree with a topicalized NP, which also precedes it.

is
(34) Those students, Sandy ∗ friendly with.
are
Second, a pronoun that is identical in number and gender to the subject
appears in a tag question, adjoined to the sentence. We say that the pronoun
in the tag question agrees with the subject of the sentence.
(35) a. Sandy is unfriendly, isn’t s/he.
b. The students are unfriendly, aren’t they.

she
c. She is unfriendly, isn’t ∗ .
he
There can never be a tag question in which the pronoun agrees with a non-
subject NP.
∗
don’t you
(36) a. Susan likes you, .
doesn’t she

don’t you
b. You like Susan, ∗ .
doesn’t she
Third, in a question, the subject (if it is not an interrogative itself) appears
to the right of the inflected verb; a topic does not.
(37) a. Is Sandy friendly?
b. Which students is Sandy friendly with?
c. ∗ Is those students, Sandy friendly with?
All of these tests will help us to identify the subject regardless of where it
appears in the linear order in the sentence.
3.5. Marking grammatical functions
3.5.1. Case
The basic grammatical functions are expressed in some languages by case

marking and agreement, rather than constituent order. Japanese shows
3.5. MARKING GRAMMATICAL FUNCTIONS 73
case marking of its subject, object, and indirect object. The word order in
Japanese is very free, as long as the verb is in sentence-final position – we
give just a few possibilities here to illustrate.
(38) Japanese
a. Taro-ga Hanako-ni hana-o ageta
Taro-NOM Hanako-DAT flower-ACC gave
‘Taro gave a/the flower to Hanako.’
b. Taro-ga hana-o Hanako-ni ageta
Taro-NOM flower-ACC Hanako-DAT gave
c. Hanako-ni Taro-ga hana-o ageta
Hanako-DAT Taro-NOM flower-ACC gave
d. Hanako-ni hana-o Taro-ga ageta
Hanako-DAT flower-ACC Taro-NOM gave
NOM stands for nominative; it is the marker that typically goes on a subject.
ACC stands for accusative; it is a marker that typically goes on an object.
DAT stands for dative; it is a marker that typically goes on an indirect object,
translated here as the object of the preposition to.
As we have seen, English marks grammatical function by putting an NP
in a particular position in the syntactic structure. The subject is the sister of
VP, and the direct object is the sister of V. This structure is reflected in the
constituent order: normally the subject appears before the VP, and hence
before the V, and the direct object follows the V.
However, given how free constituent order is in Japanese, there may be
no need to distinguish subject and object in Japanese in terms of where they
are attached in the syntactic structure – the case marking is sufficient. If this
is so, then it may be that there is no VP in Japanese, in contrast to what we
find in English.
There are several ways in which case is expressed on an NP. It is often
marked on the head noun, as in the Japanese examples given here. 4 The case
marker follows the noun, and does not appear on the preceding modifiers.
(In the gloss, NEG. IMP means “negative imperative”, translated as don’t in
English.)
4
Since the head noun is final in the NP in Japanese, it could also be the case that
the case marker is a particle that is attached to the entire NP.
(39) a. sono hon-o

that book-ACC
b. Ano aoi mi -o taberu-na.
That blue berry-ACC eat-NEG. IMP.
‘Don’t eat those blue berries.’
In many languages, the case of the NP also appears on the other con-
stituents of the noun phrase, as illustrated in the following Russian exam-
ples. (The instrumental case marked INSTR is used with the preposition s
“with”.)
(40) a. moj-a interesn-aja knig-a

my-NOM interesting-NOM book- NOM
‘my interesting book’
b. s moj-ej interesn-oj knig-oj
with my-INSTR interesting-INSTR book-INSTR
In some languages, such as German, the case is not marked on the head
noun but on the other constituents of the noun phrase. 5 (The dative case
marked DAT is required with the preposition mit “with”.)
(41) a. mein interessantes Buch

my.NOM interesting.NOM book
‘my interesting book’
b. mit meinem interessanten Buch
with my.DAT interesting.DAT book
‘with my interesting book’
3.5.2. Case-marking patterns
The pattern of realizing arguments in terms of grammatical functions

found in English and most other European languages is called nominative-
accusative. English shows the nominative-accusative pattern in the pronoun
system. In this pattern, the subject of an intransitive sentence (a verb with
only a subject) and the subject of a transitive (a verb with a subject and an
object) have the same case form.
5
There are some contexts in German where the head noun also shows marking
for case.
 
(42)  He 
 
She
a. fell.
 We 
 
They
∗ 

 ∗ Him 
Her
b. ∗ fell.

 ∗ Us  
Them
   

 He 
  him 
 
She her
c. saw .
 We 
   us 
 
They them
  ∗ 
 He 
  
 ∗ he  
She she
d. saw ∗ .
 We 
  
 ∗ we  
They they
Non-pronominals in English do not show case. Nevertheless, the subject

of an intransitive sentence and the subject of a transitive are realized in the
same way, as the sister of VP that precedes VP. We use SUintr and SUtr to
distinguish the two types of subject.
(43) a. Sandy fell.
SUintr
b. Sandy is eating [an ice cream cone].

SUtr O
There is another pattern, which is called ergative-absolutive, or simply
ergative. In ergative languages, the subject of the intransitive has the same
form as the object of the transitive. If English were an ergative language, and
still used constituent order to mark grammatical function, the sentences in
(43) might look like this:
(44) a. Sandy fell.
SUintr
b. [An ice cream cone] is eating Sandy.

O SUtr
Since English uses order to mark grammatical function, a pattern in which
SUintr and O were marked in the same way would put them on the same side
of the verb.
Ergative languages with case marking display a pattern in which the sub-
ject of the intransitive and the object of the transitive are marked the same,
but they typically use morphology and not word order for this purpose.
The subject of the transitive is marked with the ergative case, while the
object of the transitive and the subject of the intransitive are marked with
the absolutive case.
Here are examples from Burushaski (Pakistan) and West Greenlandic.
The noun without any overt case morpheme is in the absolutive case. 6
The morphologically marked form of the noun phrase is in the ergative
case. As in Japanese, the verb is sentence-final. Note that in the transitive
sentences (45a, 46a), the ergative-marked phrase is the subject, while in the
intransitive sentences (45b, 46b) the absolutive-marked phrase is the subject.
(The gloss IND. TR in the West Greenlandic example means “indefinite
(object)/transitive”.)
(45) Burushaski 7
a. ne hír-e phaló bók-i
the.MASC man-ERG seed.PL . ABS sow.3 SG. MASC
‘The man planted the seeds.’
b. ne hir yált-i
the. MASC man.ABS yawn.3 SG. MASC
‘The man yawned.’
(46) West Greenlandic
a. Oli-p neqi neri-vaa
Oli.ERG meat.ABS eat-IND. TR .3 SG.3 SG
‘Oli eats meat.’
b. Oli sinippoq
Oli.ABS sleep.- IND. INTR .3 SG
‘Oli sleeps.’
[Manning 1996:2–3]
These examples show that languages have various ways to distinguish one
syntactic argument from another.
Here is another example, from Jiwarli, an Australian Aboriginal lan-
guage. 8 The noun without any overt case morpheme is in the absolutive
case. This form appears as the object of “see” in (47a) and as the subject of
“fall” in (47b). The marked form is in the ergative case.
6
The absolutive case in ergative languages virtually always lacks overt case
morphology (Iggesen 2005:91).
7
The glosses here are simplified versions of the originals.
8
From http://www.linguistics.unimelb.edu.au/research/projects/jiwarli/gramm.
case.html. We have changed the gloss to show that the same case forms appear in
the two examples.
(47) Jiwarli
a. Juma-ngku wuru nhanya-nyja
child-ERG tree.ABS see-PAST
‘The child saw the tree.’
b. Wuru warni-nyja
tree.ABS fall-PAST
‘The tree fell.’
This case-marking pattern is another one in which the same morphological

inflection appears on the object of a transitive (47a) and the subject of an
intransitive (47b).
Basic nominative/accusative and ergative/absolutive patterns
Case NOM ACC ERG ABS

Transitive SU OBJ SU OBJ
Intransitive SU SU
3.5.3. Agreement
One other way in which grammatical functions may be realized is through

agreement. In cases where there is agreement, the form of a word or phrase
depends on the properties of another word or phrase. We have already seen
a simple case of agreement, where the form of the verb in English depends
on the number of the subject. Another example involves the English tag
question, in which the form of the pronoun in the tag depends on the
properties of the subject of the sentence, as we saw in section 3.4.2.
In a language with verb agreement, the verb is marked in such a way
that it specifies certain properties of certain of its arguments. In Choctaw,
a Muskogean language of Oklahoma and Mississippi, verbs are marked for
subject, direct object, indirect object, and other phrases. There is a specific
position for each type of marker with respect to the verb root, as shown
in (48). (All of the Choctaw examples are taken from Davies 1986. In the
glosses, CONTR means “contrastive”.)
(48) Choctaw
a. chi- bashli -li -tok
2.ACC cut 1.NOM PAST
‘I cut you.’
b. Ano is- sa- hottopali -tok

I 2.NOM 1.ACC hurt PAST
‘You hurt me.’
(49) a. Alla towa ish- ı̃- pila -tok
child ball 2.NOM 3.DAT throw PAST
‘You threw the ball to the child.’
b. An- at- o iskali chim- a: -li -tok
I NOM CONTR money 2. DAT give 1. NOM PAST
‘I gave the money to you.’
Notice the markers attached to the verb for subject (2.NOM and 1. NOM ),
for object (2. ACC and 1. ACC ) and indirect object (2. DAT and 3. DAT ). If
there is no full NP the verb marking is interpreted as though it was a
pronoun, as in (48a). (This is very typical in languages where the verb is
marked for agreement.)
The Choctaw examples show that, if there are full NPs in a sentence,
their grammatical functions are identified through agreement between the
NPs and the verb. For example, a feminine singular NP will be identified
as the subject because there is a feminine singular subject marker on the
verb. Sometimes this type of marking is called “case marking”, even when
the NPs themselves are not overtly marked for case. The following exam-
ples from Tukang Besi, a language of Indonesia (Sulawesi), illustrate this
pattern.
(50) Tukang Besi
a. no-‘ita-‘e na kene-no te ana
3.NOM-see-3. ACC ART friend-3.POSS ART child
‘The child saw its friend.’
b. te kene-no no-‘ita-‘e te ana
ART friend-3.POSS 3.NOM-see-3ACC ART child
‘That child saw its friend.’
c. te kene-no no-‘ita-‘e te ana
ART friend-3.POSS 3. NOM-see-3.ACC ART child
‘That child saw its friend.’
[Donohue 1999:51, 60, 61, exs. 1, 31, 35]
d. no-wila
3.NOM-go
‘S/he went.’
[Donohue 2002]
Here, no- is the form attached to the verb for a third person subject, while
-‘e is the form for the third person object. Note how the third person NPs na
kene-no “its friend” and te ana “the child” agree with these forms. The word
3.6. TESTS FOR CONSTITUENCY 79
order is flexible, but the grammatical functions are constant. (Because both
NPs are third person, these examples are ambiguous.) In the last example,
there is no NP subject, and in this case the form attached to the verb is
interpreted as though it were a pronoun.
Marking of a different pattern is shown in the following examples, from
Konjo, a language of Uganda (Friberg 1991 cited in Blake 1994:124). Here,
the form used for the subject of an intransitive verb, -a in (51), is the same as
the form used for the object of a transitive verb, as shown in (52). Example
(53) shows that the first person singular form when it is subject of the
transitive is different from when it is the subject of the intransitive (ku- vs
-a). (The gloss INTR indicates “intransitive”.)
(51) Konjo
A’-lampa-a
INTR-go-1. ABS
‘I go.’
(52) Na-itte-a
3-ERG-see-l.ABS
‘S/he sees me.’
(53) Ku-itte-i balla’-na
l.ERG-see-3-ABS house-3.POSS
‘I see his/her house.’
(54) Na-itte-i balla’-ku
3-SG-see-3.ABS house-l.POSS
‘S/he sees my house.’
This is an ergative pattern.
3.6. ∗ Tests for constituency
As we have already pointed out, there is usually nothing overt to mark

the boundary between constituents, e.g. between V and NP in the VP,
although we do have intuitions about constituency based on meaning. But
because we have knowledge of the structure of the language, it is possible to
apply tests that reveal what the constituents are. A constituency test isolates
one of the constituents and highlights it against the rest of the structure,
thereby providing evidence that it is indeed a distinct unit within the larger
structure.
There are four standard types of constituency tests that make use of the
syntactic constructions of a language: (i) ellipsis, (ii) pro-form replacement,
(iii) coordination, and (iv) displacement. In addition, when there is morpho-

logical agreement and the form of one constituent depends on another, this
relationship typically holds within a constituent, and semantic intuitions
also provide hints about what the constituents are. Ideally, all tests should
be consistent with one another, to the extent that they are applicable. As
we proceed we will discuss some of the possible conclusions that one might
draw when some but not all tests are consistent.
3.6.1. Ellipsis
One type of constituency test involves what is conventionally called ellipsis,

often called deletion, which is the omission of part of a sentence under cir-
cumstances where its meaning can be determined from context or from the
structure of the sentence. The standard hypothesis is that only a constituent
can be omitted and not an arbitrary substring. For instance, in English it
is possible to omit a VP, using the VP ellipsis construction. We use the
strikeout notation to show the part of the sentence that has been omitted.
(55) a. The dog will chase the cat, and the horse will chase the cat too.
b. The dog will chase the cat, and the horse will chase the cat too.
A fundamental assumption of syntactic theory is that what is relevant to

determining the grammatical sentences of a language are the constituents
that make up sentences, and not the strings of words. If a string of words
is not a constituent, it should not be possible to omit it. In this case, the
string chase the does not correspond to a constituent, and this explains the
ungrammaticality of the following example.
∗
(56) The dog will chase the cat, and the horse will chase the cow.
It is important to use caution in applying a constituency test such as

ellipsis. If something can be omitted, that is evidence that it is a constituent.
But if it cannot be omitted, that is not evidence that it is not a constituent,
since it is also possible that the language simply does not provide a way to
omit that constituent. For example, in English it is impossible to omit an
object NP, parallel to VP ellipsis.
(57) a. The dog will chase the cat, and the horse will kick the cat.
b. ∗ The dog will chase the cat, and the horse will kick the cat.
On the other hand, there are examples that suggest that an NP can be
omitted in subject position.
(58) a. The dog chased the cat, and the dog kicked the cat.
b. The dog chased the cat, and the dog kicked the cat.
However, there are other ways to analyze this type of sentence that do not
require ellipsis of an NP; see section 3.6.3.
3.6.2. Proform replacement
Another standard test involves the replacement of a constituent with

a proform. A proform is a generic element that can be used to express
the same meaning as an arbitrarily complex expression. For example, a
pronoun (that is, a pro-noun) such as it, she, they, etc. can be used to refer
to anything that a full NP refers to as long as the pronoun and the full NP
agree in number and gender.
 

 a house. 

 a house with three bedrooms.
 

(59) I bought a house with three bedrooms that faces the woods.

 

 a house with three bedrooms that faces the woods in a suburb. 
 
...
 
 It 
∗ was on a cul-de-sac.
She
 ∗ They 
The fact that it can, in effect, replace these phrases suggests that they are all
constituents.
Proforms for VPs in English are do so and do it, which have slightly
different distributional properties.

did it
(60) I bought a house last year; I for investment purposes.
did so
Both of these pro-VPs are used to refer to voluntary actions, and hence
cannot be used for VPs headed by verbs like know, receive, live, and so on.

does it
(61) a. ∗ I know the answer, but no one else . [proform replacement]
does so
I know the answer, but no one else does. [ellipsis]
b. ∗ For my birthday I received a new computer, and I didn’t expect to

do it
. [proform replacement]
do so
For my birthday I received a new computer, and I didn’t expect to. [ellipsis]

do it
c. ∗ I used to live in NY but I don’t now. [proform replacement]
do so
d. I used to live in NY but I don’t live in NY now. [ellipsis]
Most VP proforms (but not do so) are composed of the verb do and an
NP. In order to question a VP in English, the NP is replaced by the wh-word
what (the interrogative correlate of it).
(62) What did you do?
Thus, do what is the interrogative pro-VP, paralleling do it, do something.

The form so used alone functions as a pro-S, as shown by the examples
in (63).
(63) I think that John is very wealthy, but no one else believes so. [So stands for
that John is very wealthy.]
The form there is a pro-adverbial referring to place, and then is a pro-

adverbial referring to time.
(64) a. We thought about going to Cannes for our vacation next year, but no one
goes there any more.
b. We thought about going to Cannes for our vacation next year, but decided
in the end to do something else then.
To see how proform replacement functions as a diagnostic for structure,

consider the ambiguous sentence in (65).
(65) I helped the student with a smile.
One meaning of the sentence is that I helped the student who had a smile;
the other is that I had a smile when I helped the student. On the first
meaning, the student with a smile is a constituent (66a), while on the second
meaning, the student and with a smile are separate constituents of VP (66b).
(66) a. VP
V NP
helped Det N PP
the student P NP
with a smile
b. VP
V NP PP
helped Det N P NP
the student with a smile

As we might expect, a pronoun such as her can substitute for the entire
NP the student with a smile in (66a), producing
(67) I helped her.
Her can substitute for the student in (66b).

(68) I helped her with a smile.
But her cannot substitute for the student in (66a) because it is not a con-
stituent in (66a). Hence (68) has only the meaning that I had a smile when I
helped the student, and not that I helped the student who had a smile.
3.6.3. Coordination
Coordination involves conjunctions, such as the English and and or. If two
constituents A and B are of the same category XP, then a constituent of
category XP can be formed by conjoining A and B. For example, Sandy and
Chris are NPs, and we can form the conjoined NPs Sandy and Chris and
Sandy or Chris. A conjoined NP appears in all syntactic positions where a
simple NP can appear and satisfies all of the tests that distinguish NPs from
constituents of other categories.
(69) a. [Sandy and Chris] just arrived.
b. I called [Sandy and Chris].
c. We were talking to [Sandy and Chris].
Similarly, we may have conjoined VPs (70a), PPs (70b), and APs (70c).
(70) a. I [[came home] and [sat down]].
b. We have lived [[in Chicago], [in Paris], and [in Moscow]].
c. Sandy was [[angry at Terry] but [ashamed of her own behavior]].
We express the possibility of having coordinate structures regardless of

the category by the following rule schema. It is a “schema” because it refers
to a range of categories, symbolized by X, and not just a single category.
(71) XP → XP CONJ XP
Here, CONJ is the category conjunction. The scheme says that any category
XP can be a coordinate structure containing XP-CONJ-XP.
Coordination also appears to apply when two constituents have the same
grammatical function but not necessarily the same syntactic category.
(72) Chris is angry and a bully.
Angry is an AP and a bully is an NP. If we want to say that only constituents

of the same category may be conjoined, then we are obliged to analyze a
bully in (72) as a VP with a phonologically elided head, so that the conjoined
structure would be
(73) Chris [[VP is angry] and [VP is a bully]].
Such an analysis requires that we formulate a principle that says under what
circumstances a head may be empty; note for example that the head of the
left conjoined phrase cannot be empty.
∗
(74) Chris [[VP is a bully] and [VP is angry]]
A first approximation of such a principle would be that the leftmost head

cannot be elided.
3.6.4. Displacement
3.6.4.1. Topicalization of NP
We use the term displacement to describe a situation in which a part of a
sentence is not in its canonical position. 9 A typical example is given in (75),
which exemplifies what is called topicalization.
(75) That cat, the dog was chasing ___.
Here we are presuming that there is such a thing as canonical structure, i.e.
structure that is most typical of a given language, and corresponding typical
positions for the various parts of a sentence based on their grammatical
functions. Grammatical functions such as subject and object are defined in
terms of this canonical structure. For example, the subject is the NP that
together with the VP forms a sentence, as stated by rule (9). The object is
the NP that together with the V forms a VP. One rule for VP is this:
(76) VP → V NP
9
We use this term here because it is more neutral than the more commonly used
movement, which carries with it the presumption of some computational operation
that transforms one syntactic structure into another. We address the question of
whether displacement is best characterized in terms of movement at various points
in this book.
When a direct object such as the cat does not appear immediately after its
verb, as in (75), we say it is in a non-canonical position. When a constituent is
in a non-canonical position, we say that is displaced. For (75), for example,
the intuition is that that cat is the object of chasing, although it is not in the
position where a phrase would normally acquire this function in English.
(Normally the verb chase selects an object; in this sentence it appears in
initial position.)
The natural question to ask is why the position occupied by that cat
in (75) cannot simply be an alternative canonical position for object. The
answer has two parts. First, there are an infinite number of such positions,
so it is impossible to define an alternative canonical position along these
lines. This point is demonstrated by examples such as the following.
(77) That cat, you said [the dog was chasing __]
That cat, I think [you said [the dog was chasing ___]]
That cat, they claim [I think [you said [the dog was chasing ___]]]
etc.
Since there is no bound on the distance that the initial constituent may be
located from the position immediately after the verb of which it is the object,
there is no way to list all of the configurations in which it might appear. And
this brings us to the second point, which is that non-objects may also appear
in this position. The examples in (78) show subjects in initial position other
than their canonical position.
(78) That cat, you said [___ was chasing the dog].
That cat, I think [you said [___ was chasing the dog]].
That cat, they claim [I think [you said [___ was chasing the dog]]].
etc.
We must conclude that that cat in these examples gets its grammatical
function in virtue of being linked to the empty position, that is, the position
immediately adjacent to the verb to the right of it in the case of (77) and to
the left of it in the case of (78).
The idea that there is a linking between the initial phrase and the empty
position is supported by the fact that the verb in the complement agrees with
it. As we have seen (section 3.5.3), agreement is the situation in which the
form of one word or phrase reflects properties of another word or phrase.
For example, (79a) shows that when the subject is singular (the cat), the
verb will have the singular form (was and not were). Similarly, in (79a), the
subject is plural (the cats) and the verb is plural (were and not was).

was
(79) a. That cat ∗ chasing the dog.
were
∗
was
b. Those cats chasing the dog.
were
The examples in (80) show that when the subject is moved, the agreement
pattern is the same, even though the subject and the verb are not next to
one another.

was
(80) a. That cat, you said [___ ∗ chasing the dog].
were
∗
was
b. Those cats, you said [ ___ chasing the dog].
were
This distribution is explained if the clause-initial NP is in a very real sense

functioning as the subject of the complement, regardless of its superficial
position.
The question of exactly how to express the linking between the actual
position of the NP and the position that corresponds to its function in for-
mal terms has been the subject of considerable debate in linguistic theory;
we postpone a more detailed discussion to Chapter 9. But the dependence
of the verb on the number of the NP even when the NP is not in subject
position constitutes evidence for a link between the two positions, however
we characterize it in formal terms.
Returning to the central point of this section, we can see that the pos-
sibility of displacement can be used as a diagnostic for structure. While
the boundaries between constituents cannot be seen when we look at a
sentence with canonical structure, our knowledge of the structure of the
language provides us with the intuition that these boundaries exist, and we
can use this knowledge to construct sentences in which the constituents are
separated from one another. For example, the fact that when that cat is the
object of the V chasing it may appear elsewhere in the sentence is evidence
that that cat is a phrase. So is the fact that doing syntactic research may be
separated from like in (81).
(81) Doing syntactic research, I really like ___ a lot.
The counterpart to the notion that constituents may be displaced is that

non-constituents, that is, arbitrary strings of words, cannot be displaced.
Consider the sentence that is the basis for (81).
(82) I really like doing syntactic research a lot.

It is impossible to create a well-formed sentence by putting the strings like

doing syntactic, or syntactic research a, or research a lot at the beginning of
the sentence.
∗
(83) Like doing syntactic, I really ___ research a lot.
∗
Syntactic research a, I really like doing ___ lot.
∗
Research a lot, I really like doing syntactic ___.
The reason for this is that these strings of words are not constituents, a
conclusion that conforms to our intuitions.
What we are doing when we are applying a constituency test is using one
type of sentence to make inferences about the structure of another type of
sentence. The topicalization construction is used to demonstrate that the
object of a sentence in which topicalization has not applied is a constituent.
The reasoning here is that the two sentences are essentially identical in
meaning and, in particular, the putative constituent in question (the object)
has the same function in the two sentences. Therefore, if it is demonstrably
a constituent in one, it should be a constituent in the other.
3.6.4.2. Topicalization of VP
Applying constituency tests to objects and subjects has a somewhat redun-
dant feeling, since we have very strong intuitions that they are constituents
quite independently of the tests. But, having established the logic of con-
stituency tests, we are able to apply them with some confidence in cases
where our intuitions are not as strong. Consider the VPs in the examples in
(8), repeated here.
b. The dog chased the cat.
The intuition that called and chased the cat are constituents of the same
type is supported by the fact that they can be substituted for one another
without affecting well-formedness.
(84) a. President Smith chased the cat.
b. The dog called.
But this intuition is based on meaning – since we understand the sentences,

we understand that the part that expresses the property attributed to the
subject in each case is what is underlined. If possible, we would like to have
a constituency test that is consistent with our intuition.
It is possible to apply topicalization to a VP in English, and the results
are interesting. First, consider a sentence like
(85) President Smith will call.
Under the appropriate discourse conditions, it is possible to place call in

sentence-initial position.
(86) (They said that President Smith will call, and) call, President Smith will ___,
(I have no doubt).
While this example shows that call can be topicalized, it does not show that
a VP can be – perhaps topicalization applies to a verb. The test comes with
an example in which the putative phrase is more complex.
(87) a. The dog will chase the cat.
b. (They said that the dog will chase the cat, and) chase the cat the dog will
___, (I have no doubt).
c. ∗ (They said that the dog will chase the cat, and) chase the dog will ___ the
cat, (I have no doubt).
Example (87b) shows that chase the cat is a phrase, and (87c) shows fur-
thermore that what is topicalized is not a V but a VP. (If a V alone could
topicalize, this example should be grammatical.) So we have evidence that
the phrasal category VP exists in English.
Consider next the examples that we started out with, those in (8). When
we apply topicalization to them, something strange happens.
b. ∗ (They said that President Smith called, and) called President Smith ___,
(I have no doubt).
(89) a. The dog chased the cat.
b. ∗ (They said that the dog chased the cat, and) chased the cat the dog ___,
(I have no doubt).
What is strange is that the VPs called and chased the cat should be topical-
izable, since they are VPs, but they are not.
There are two reasonable conclusions to draw from this. One is that
they are not VPs, and the other is that there is something about them that
prevents VP topicalization from applying to them. Without trying to resolve
the issue here, let us note that the problem that these VPs have is that their
verbs are marked with past tense, while in the well-formed cases there is
no tense marked on the verb itself. That is, in will call, the verb call lacks
any tense marking. The ill-formed examples here can be saved if we pull
the tense marking off the main verb and locate it on the auxiliary verb did,
which is left behind.
(90) a. . . . and call, President Smith did.

b. . . . and chase the cat, the dog did.
It appears, then, that topicalization applies only to a VP that lacks tense

marking. What is notable about did is that it appears only when the verb
appears in a construction where it cannot be marked for tense; topical-
ization is one such construction. We return to a formal and more detailed
discussion of this aspect of English syntax in Chapter 4.
3.6.4.3. Pseudo-cleft
A test that is related to topicalization is the pseudo-cleft construction, illus-
trated in (91).
(91) a. [What President Smith did] was [call].

b. [What the dog did] was [chase the cat].
This construction is formed by combining a phrase of the form what Presi-

dent Smith did or what the dog did, or what I said, a form of the verb to be (in
this case, was), and then an additional string, called the focus. Only a single
constituent can be focus in this construction, as shown by the following
examples.
(92) a. President Smith called [Sandy][a hero].

b. ∗ What President Smith called was [Sandy] [a hero].
[Cf. What President Smith called Sandy was [a hero].]
(93) a. President Smith gave [Leslie][a book].
b. ∗ What President Smith gave was [Leslie][a book].
[Cf. What President Smith gave Leslie was [a book].]
Thus, the result of this test in (91b) is consistent with the hypothesis that
English has a VP constituent.
Pseudo-clefts and topicalization support the conclusion arrived at in the
preceding section that there can be ellipsis of the head of a VP in a conjoined
structure. Consider the following example, where this ellipsis is indicated.
(94) Chris [sold [the motorcycle] [to the first person who called]] and [sold [the car]
[to a neighbor]].
The alternative, that the car to a neighbor is a constituent, is inconsistent

with the fact that it cannot undergo topicalization or pseudo-cleft.
(95) a. ∗ The car to a neighbor, Chris sold/tried to sell.

b. ∗ What Chris sold was the car to a neighbor.
On both of these tests, the car to a neighbor is not a constituent. An analysis

along the lines of (94) is consistent with the results of these tests, and with
the principle that the leftmost head in a series of conjoined phrases cannot
be elided.
These cases highlight the fact that the possibility of omitting parts of
a constituent may yield false constituency when applying coordination.
What appears to be a constituent at first glance may turn out not to be.
Converging tests make conclusions about constituency more secure, while
conflicting results suggest that other factors may be at play.
Consider finally the fact that auxiliary verbs such as will precede the
negative not, while main verbs such as called do not. The difference is shown
in (96).
(96) a. President Smith will not call.
∗
b. President Smith will call not.
∗
c. President Smith called not.
d. President Smith did not call.
The distribution of not is consistent with the distinction that we have made
between AUX and VP: not precedes VP and follows AUX. We represent the
distribution of not by extending our basic rule for the structure of S.
(97) S → NP AUX (not) VP
3.6.4.4. Wh-question
Another constituency test is the wh-question construction. In this construc-
tion, a wh-phrase appears at the left edge of its clause, and it is linked to its
canonical position.
(98) a. What did the dog chase ___ ?
b. What are you eating ___?
c. Who did you talk to ___?
We use this construction to show that an object is a constituent, consistent

with our argument based on topicalization. While the object the cat is not
a wh-phrase, it has the same grammatical function in the sentence The dog
chased the cat as what has in (98a). So we can substitute what for the cat,
and then see if it can appear in initial position in a wh-question. Or, we can
use other more specific phrases to substitute for the cat.

(99) What cat
Which cat did the dog chase ___ ?
What kind of cat
The evidence is consistent with the constituent structure that we have

hypothesized – it shows that what follows chase is a phrase.
In English a VP does not participate in the wh-question construction.
This is not evidence that English lacks a VP, since the wh-question test is
inapplicable for an independent reason: there are no wh-words that sub-
stitute for a VP. 10 An action in English can therefore not be questioned
directly; it is necessary to use the interrogative variant of do something, i.e.
do what. In contrast, APs, PPs and AdvPs have interrogative variants that
can be used to form wh-questions.
(100) a. How tall is Sandy ___ ?

b. On which table did you put the bananas ___?
c. How high did you jump ___?
Such movement tests confirm our intuitions that these kinds of phrases also
exist in English.
3.6.4.5. Disambiguation
Using displacement tests, it is possible to investigate the structure of
ambiguous sentences. If a given string of words has two (or more) possible
structures, then it should have two meanings, corresponding to the two
structures. Consider the following example.
(101) Sandy looked over the table.
One meaning of this sentence is that Sandy examined the table. The other
meaning is that the direction of Sandy’s gaze was above the table.
On the second meaning, over the table denotes a direction (of Sandy’s
gaze), and thus we might expect it to be a constituent. On the first meaning,
the table is the object of the verb, and the action is expressed by look over.
So in this case we might expect over the table not to be a constituent;
rather, the table is. Therefore, a constituency test that picks out over the table
should apply only with the first meaning. The following sentences show the
results.
10
In Chapter 9 we discuss the fact that phrases that contain a wh-word or wh-
phrase may undergo wh-movement under certain circumstances, a phenomenon
illustrated by (100). However, a VP that contains a wh-word cannot undergo wh-
movement. This is a puzzling fact in view of the fact that topicalization can apply to
VP. There is an exercise in Chapter 9 that asks you to explore the implications and
possible reasons for this.
(102) Over the table, Sandy looked.
This sentence only means that the direction of Sandy’s gaze was over
the table, not that Sandy examined the table. So the two structures are,
roughly,
(103) a. Sandy looked [over the table] (i.e. where Sandy looked)
b. Sandy looked over [the table] (i.e. what Sandy examined)
To confirm our intuition, we note that over can appear after table only with
the meaning “examine”.
(104) Sandy looked the table over.
When the verb appears with a particle, such as over, the particle may appear
in the position following the NP.
Another example of ambiguity is the following.
(105) Terry helped the student with a smile.
On one meaning, the student has a smile, and, on the other meaning, Terry
has a smile. If the student has a smile, then the student with a smile is a
constituent and should function as a unit in a constituency test. But, if Terry
has a smile, with a smile is not part of the noun phrase headed by student
and should not function as a unit with the student. The following examples
test this prediction.
(106) a. The student with a smile, Terry helped (and not the student with a
frown).
b. With a smile, Terry helped the student.
c. The student, Terry helped with a smile.
In (106a), the only possible meaning is that the student has a smile. In
(106b,c), the only possible meaning is that Terry has a smile. The two
structures are thus
(107) a. Terry helped [the student [with a smile]]
b. Terry helped [the student] [with a smile]
Furthermore, since (106b,c) are not ambiguous, these sentences provide

evidence that topicalization cannot apply to part of an NP, a fact that we
explore in greater detail in Chapter 9.
The general point of these examples is that a test for structure will often
apply differently to the two (or more) structures associated with a single
ambiguous sentence. When we apply the test, the ambiguity disappears.
EXERCISES 93
Exercises
1. Consider the following phrase structure tree.

(1) Z
X V
Z B C P M
Q
N R
T W
i. List all of the nodes that M immediately dominates.
ii. List all of the branching nodes.
iii. List all of the binary branching nodes.
iv. List the immediate constituents of V.
v. List all of the constituents of V.
vi. What is the mother of B?
vii. What is/are the sister/s of R?
[§3.1.]
2. Answer the following questions about the tree in (1). We have numbered
the nodes so that they can be distinguished from one another. (Don’t worry
if you don’t know what some of the labels mean – the goal here is to
understand the relations between nodes.)
(1) S
NP1 VP1
Det1 N1
V1 VP2
V2 NP2 PP
Det2 N2 P NP3
a. What nodes does VP1 dominate? immediately dominate?
b. What nodes does S immediately dominate?
c. What nodes does NP2 dominate? immediately dominate?

d. What nodes are the sisters of NP1 ?
e. What nodes are the sisters of V2 ?
f. What node is the mother of P?
g. What node is the mother of Det1 ?
[§3.1.]
3. Draw a tree for each of the labeled bracketed strings in (1). Don’t
worry if you don’t know what the labels mean – the exercise is simply about
converting labeled brackets into tree diagrams.
(1) a. [NP [Det the ] [Adj furry ] [N poodle ]]
b. [S [V let ] [NP [N them ]] [VP [V eat ] [NP [N cake ]]]]
c. [PP [NP [Q two ][N miles ][PP [P into ] [NP [Det the ] [N tundra ]]]]
[§3.2.]
4. A phrase structure rule specifies a possible constituent structure of a

phrase of a certain type. For each of the following phrase structure rules,
draw a tree that illustrates the corresponding constituent structure. If the
rule specifies more than one structure, draw one tree for each structure.
(Don’t worry if you don’t know what the labels mean – the exercise is
about understanding how to use the rules to produce tree diagrams with
the proper symbols.)
(1) a. NP → Det N
b. DP → Det NP
c. XP → XP Conj XP
d. VP → PP NP V
e. VP → (PP) (NP) V
f. PP → P (NP)
g. AP → Det (S) A
h. CP → Spec C
i. NP → PP N (Dem)
[§3.2.]
5. State the minimal rule or rules needed to produce each of the following
structures. (The rules will be different for each structure.)
(1) NP
N Adj
EXERCISES 95
(2) WP
X F PP Z W
(3) NP
PP N
NP P
PP N
(4) VP
VP NP V
NP V
(5) AP
Int A PP
P NP
Dem N PP
P NP
(Don’t worry if you don’t know what the labels mean – the exercise is
about understanding how to use the rules to produce tree diagrams with
the proper symbols.)
[§3.2.]
6. The following rules together describe a set of six trees. Draw all of the
possible trees compatible with these rules.
(1) a. XP → A X (YP)
b. YP → B Y (ZP)
c. ZP → C Z
(Hint: Make sure that if a tree contains a XP, YP, or ZP that it dominates
the proper constituents.)
[§3.2.]
7. Translate the following statements into PSRs. For example, if the state-
ment is “An S may consist of an NP followed by a VP”, the corresponding
rule is
(0) S → NP VP
(1) a. A PP may consist of a P followed by an NP.
b. A VP may consist of a V followed by an NP followed by another NP
followed by an S.
c. An NP may consist of an S followed by an AP followed by an N.
d. An S may consist of a VP followed by an NP.
[§3.2.]
8. For each of the bracketed constituents, say whether it is an argument
or adjunct and give reasons to support your answer.
(1) a. [Robin] loves [pizza] [for breakfast].
b. [Terry] [carefully] arranged [the flowers] [in the vase] [for the guests] [imme-
diately before the party].
c. [I] bet [you] [everything I own] [that our team will be national champions
this year].
[§3.3.]
9. Show that the underlined phrases in the following sentences are not
subjects, using the tag question test for subject that we introduced in this
chapter.
(1) a. John loves Mary.
b. The dog ate all the pizzas.
[§3.4.]
10. Are the following patterns accusative or ergative? Explain your
answer.
i. Inuit
(1) a. anguti-up tuktu taku-jaa
‘The man saw the caribou.’
b. tuktu niri-juq
‘The caribou was eating.’
[Johns 1987]
ii. Russian
(2) a. Ja pročital knigu.
‘I read a/the book.’
EXERCISES 97
b. Kniga v komnate.
‘The book is in the room.’
c. Ja daval knigu Ivan’e
‘I gave the book to Ivan.’
d. Ivan videt mnja.
‘Ivan sees me.’
iii. Nepali
(3) a. manis-le aymay dekh-yo
‘The man saw a woman.’
b. aymay-le manis dekh-yin
‘The woman saw a man.’
c. manis uphr-yo
‘The man jumped.’
d. aymay uphr-yin
‘The woman jumped.’
e. manis gho.r-ma go-yo
‘The man went to the house.’
f. aymay gho.r-ma go-yin
‘The woman went to the house.’
[Givón 2001:208]
[§3.5.]
11. If English were an ergative language, what would be the form of the
pronoun they/them in the position marked by ___ in each of the following
sentences?
(1) a. ___ are sleeping.
b. Do ___ like the music?
c. Sandy doesn’t know ___ very well.
d. Why are ___ talking so loud?
e. Put ___ on the table, please.
[§3.5.]
12. Explain the ambiguity of the following sentences and show how in
each case topicalization and pseudo-cleft disambiguates.
(1) a. Terry saw the elephant with a telescope.
b. Robin broke the bottle on the table.
c. Leslie stole the letter from King Henry VIII.
d. Lee sold the presents for the boss.
[§3.6.]
Problems
1. The following examples suggest that a subject in English may be

something other than an NP.
(1) a. Under the bed is a good place to hide the beer.
[cf. The closet is a good place to hide the beer.]
b. That you don’t speak Nahuatl is obvious.
[cf. Your inability to speak Nahuatl is obvious.]
Formulate a phrase structure rule or rules that will accommodate sentences
such as these. (Assume that the category of under the bed is prepositional
phrase (PP) and the category of that you don’t speak Nahuatl is S.)
[§3.3.]
2. In the present tense, Russian appears to lack the equivalent of the verb
to be. Here are some illustrative examples.
(1) a. on sčastliv
he.NOM happy
‘He is happy.’
b. ona doma
she.NOM at-home
‘She is at home.’
(2) a. on byl sčastliv
he.NOM be.PAST. MASC happy.MASC
‘He was happy.’
b. ona byla doma
she.NOM be.PAST. FEM at-home
‘She was at home.’
There are essentially two ways to account for this phenomenon. One is to
say that Russian has an “empty” or “invisible” form of be in the present
tense. The other is to say that a possible structure of a sentence is NP-AP. On
the basis of relative simplicity and generality, which of the two accounts do
you prefer? Explain your answer. (Hint: In each of the two alternatives, what
properties of the sentences follow from general rules and what properties
need to be stipulated?)
[§3.3.]
3. Many verbs have corresponding nominalizations, e.g. destroy/
destruction, complete/completion, fly/flight. Like verbs, the correspond-
ing nominalizations occur with arguments and adjuncts. Using pairs
PROBLEMS 99
of sentences and corresponding noun phrases as illustrated in (1) for

destroy/destruction, such as complete/completion, fly/flight, and similar cases
that you identify yourself, test whether the argument/adjunct distinction
correlates with optionality in the noun phrase.
(1) a. The army destroyed the village with a few tanks on Monday.
b. the army’s destruction of the village with a few tanks on Monday
[§3.3.]
4. In the text we noted that so is a pro-S, as illustrated by examples such
as
(1) I think that John is very wealthy, but no one else thinks so.
However, so cannot always substitute for an S.
(2) a. That John is very wealthy is painfully obvious.

∗
b. So is painfully obvious.
(3) a. It is obvious John is very wealthy.
∗
b. It is obvious so.
Consider a broader range of contexts in which S may appear, and formulate

as simple a description as you can of the conditions under which this pro-S
so may appear.
[§3.6.2.]
5. In the text we noted that auxiliary verbs in English precede not, while
main verbs follow not. Using this difference as a diagnostic, make up exam-
ples that show what kind of verbs the following are. Say explicitly what
each example shows about the verb you are examining. For example, the
examples in the text in (96) show that will precedes not, so it must be an
auxiliary verb, while called cannot precede not, and therefore it must be a
main verb.
(1) a. could f. must

b. would g. fight
c. said h. can
d. need i. find
e. feel j. is
[§3.6.4.]
Research questions
1. A hypothesis regarding the sentences in Problem 1 above is that the

italicized expressions in clause-initial position are not subjects but adjuncts
and the subject is invisible, which we notate here as [NP e]. That is, the
analysis would be
(1) a. Under the bed [NP e] is a good place to hide the beer.
b. That you don’t speak Nahuatl [NP e] is obvious.
What evidence can you bring to bear that will help decide between this
analysis and the one assumed in Problem 1? Consider what properties sub-
jects have in English, and whether these constituents have these properties.
Here are some examples to help you get started.
(2) a. Is under the bed a good place to hide the beer?
b. ??Is that you don’t speak Nahuatl obvious?
c. ?Is that you don’t speak Nahuatl as obvious to you as it seems to be to
everyone else?
d. ∗ How obvious is that you don’t speak Nahuatl?
∗
is
(3) a. Under the bed and in the pantry good places to hide the beer.
are

is
b. That you don’t speak Nahuatl and that I don’t speak Maori ∗ obvi-
are
ous.
[§3.4.]
2. Tag question formation is somewhat more complex than we suggested
in section 3.4.2. Try to state as explicitly as possible what the rule is for
creating a tag question, using the following data as a starting point. Your
statement of the rule can be informal, but should say clearly when we get a
tag like isn’t he and when we get a tag like is he.
(1) a. John is tall, isn’t he.
b. John isn’t tall, is he.
∗
c. John is tall, is he.
∗
d. John isn’t tall, isn’t he.
(2) a. None of the students were there, were(∗ n’t) they.
b. Most of the students were there, were∗ (n’t) they.
c. Not even John knows the answer, does(∗ n’t) he.
d. Even John knows the answer, does∗ (n’t) he.
[§3.4.2.]
3.1. 1, 2
3.2. 3, 4, 5, 6, 7
3.3. 8 1, 2, 3
3.4. 9 1, 2
3.5. 10, 11
3.6. 12 4, 5
4
Phrasal Categories
4.1. X theory
As we saw in our discussion of the structure of the simple S, phrases such

as VP are built around heads. In such case, we say that the phrase is a
projection of the head. In the case of the VP, the head is V. Examination of
other phrasal categories suggests that they too are projections of heads. For
example, an NP typically contains an N as its head, an AP contains an A as
its head, an AdvP contains an Adv as its head, and a PP contains a P as its
head. A phrase that contains a head of the same type is called endocentric.
A phrase that does not contain a head of the same type is called exo-
centric. An open question in syntactic theory is whether all phrases are
endocentric.
Phrases of different categories in a given language typically display cer-
tain features in common. For example, in Japanese all phrases are head-
final. Verbs appear at the end of their sentences (or clauses), as in (1a),
nouns appear at the ends of their phrases, as in (1b,c), and prepositions
follow their complements, as in (1d). The verb benkyooshite iru “is studying”
follows the direct object nihongo-o “Japanese-ACC”, while the preposition
(actually postposition) de “in” follows its complement Nihon “Japan” in
these examples.
(1) Japanese
a. Sumisu-san-wa Nihon-de nihongo-o benkyooshite iru
Smith-Mr.-TOP Japan-in Japanese-ACC studying is
‘Mr. Smith is studying Japanese in Japan.’
b. akai kuruma
red car
‘a red car’
c. chichi-ga kinoo yonda hon
father-NOM yesterday read book
‘a book which my father read yesterday’
104 4. PHRASAL CATEGORIES
d. Nihon-e iku kara

Japan-to go because
‘because I go to Japan’
In French, most phrases are head-initial. 1

(2) French
a. regardez le chat
V NP
look-at-IMP the cat
‘Look at the cat.’
b. le chat noir
the cat black
N Adj
‘the black cat’
c. dans la chambre
in the room
P NP
‘in the room’
d. si je vais au Japon
if I go to Japan
Conj S
‘if I go to Japan’
The tendency of many languages to be primarily either head-final or

head-initial suggests that there may be some degree of uniformity of struc-
ture in languages that is not accidental, but reflective of general principles
of organization of the language faculty. A number of typological studies
have shown that languages tend towards uniformity in the ordering of heads
and complements across categories. 2 However, complete uniformity is by no
means the rule.
Additional tendencies towards uniformity can be seen in the similarities
between sentences and NPs. The examples in (3) show a sentence with the
verb destroy and an NP headed by the related word destruction. Destruction
is called the nominalization of the verb destroy. The grammatical and con-
ceptual functions of the phrases in the sentence and the corresponding NP
are identical. In (3a) the inspectors is the subject and precedes the verb. The
phrase the laboratory is the object and follows the verb. The same pattern
appears in (3b), with morphological adjustments that reflect that it is an NP
and not a sentence.
1
However, some adjectives precede the noun.
2
Greenberg 1963 and Hawkins 1994 are among the best known.
4.1. X THEORY 105
(3) a. The inspectors destroyed the laboratory.

b. the inspectors’ destruction of the laboratory
In both cases, the inspectors is understood as the agent of the action and the
laboratory is understood as the patient.
The apparent similarity between the structures of sentences and noun
phrases have led syntacticians to formulate a theory of phrase structure in
which uniformity of structure is the rule rather than the exception. This view
is called X theory. X theory takes the structure of any phrase to be a set
of projections, all of which are based on the category of the head, as shown
schematically in (4). Spec refers to one or more specifiers (which precede
the head in English), and Comp to one or more complements (which follow
the head in English), X refers to an intermediate projection and X0 refers
to the head. 3 The highest projection XP is called the maximal projection
of X0 .
(4) XP
Spec X
X0 Comp
In its strongest form, X theory holds that phrasal structures are uniform
across categories and across languages, regardless of superficial appearances
to the contrary. The final section of this chapter, section 4.7, sketches out
some of the consequences of a strong version of X theory for the analysis
of the structure of sentences and noun phrases.
X theory reflects a view of phrase structure that takes the observed pat-
terns of phrase structure to follow from general principles of simplicity and
naturalness. On one interpretation of these principles, languages are simpler
to the extent that they have uniform structures. Other things being equal
they will tend towards uniformity of phrase structure. However, deviations
from complete uniformity are possible and result in greater complexity.
This complexity, measured in terms of deviations from uniformity, might
be expected to have consequences for processing, language acquisition, and
3
In the history of X theory, additional projections above XP were considered.
Each level in the structure was annotated by adding another “ ”, so that there could
be X , perhaps X and so on. The 0 on X0 indicates that it is the lowest element of
the phrase of type X, the one with zero“ ”. In contemporary theory, it is assumed
that there are at most three levels within the phrase, as illustrated in (4).
language change, on the assumption that there is in general a pressure to

avoid or reduce complexity. We refer to this view as weak X theory.
In our discussion of phrase structure in this chapter we will provide
descriptions of the various phrasal categories based on their superficial
form, adopting the weak version of X theory. We assume as we proceed that
the structure of a phrase is as flat as possible consistent with the evidence.
Thus, we assume only that phrases are endocentric, i.e. that the canonical
structure of an English phrase is (5), where XP is the phrase and X0 is the
head.
(5) XP
… X0 …
We will flesh out the details of the structure of each type of phrase as we
proceed.
4.2. The structure of the verb phrase
We introduced a simple version of the phrase structure rules for the English
VP in Chapter 3; now we consider the VP in more detail. A VP in English
has the general property that it begins with V. It can be intransitive –
(6) The bomb exploded.
– or it can be followed by NP, –

(7) I read [NP a book].
– or by PP, –
(8) a. We are looking [PP at TV].
b. We are sitting [PP on the couch].
– or by NP followed by PP.
(9) I put [NP a book] [PP on the couch].
It is also possible to have two NP arguments following certain verbs –

(10) Chris gave Sandy a present.
Hence a preliminary rule for VP is

PP
(11) VP → V0 (NP)
NP
4.2. THE STRUCTURE OF THE VERB PHRASE 107
However, there can be more than one PP following V, and there can be
adverbs as well in VP – these tend to appear towards the end of the VP.
(12)
 

 The bomb exploded 

I read a book in the living room during the commercial
.

 We were looking at TV 
 last night surprisingly
We were sitting on the couch
And the PPs and Advs can be intermixed. 4
(13) a. I read a book [PP in the living room] [Adv quickly] [Adv last night] [PP after
the game].
b. I read a book [Adv quickly] [PP in the living room] [Adv last night] [PP after
the game].
c. I read a book [Adv quickly] [PP in the living room] [PP after the game]
[Adv last night].
It is difficult to state a phrase structure rule that allows for PPs and
adverbs following V-NP, because there can be any number of each, and they
can appear in any order. Consider the rule in (14). We use the notation “∗ ”
on a category to indicate that there can be one or more of the category in
the sequence.
(14) VP → V0 (NP)(NP)(PP∗ )(Adv∗ )
This rule says that a VP consists of a V, possibly followed by one or two

NPs, possibly followed by one or more PPs and one or more adverbs. But it
does not account for the fact that the PPs and the adverbs can be mixed in
with one another.
In order to keep our exposition relatively simple, we will work with
(14) and extend it, keeping in mind the ordering problem just noted. The
problem can be resolved in several plausible ways, but exploring them in
detail would take us somewhat far afield, so we simply note them here.
(i) It is possible that rule (14) is essentially correct, and there is an additional
mechanism in the language that reorders or “scrambles” the constituents
of VP according to independent principles. 5 (Such “reordering” may be
related to what is referred to as “scrambling” in so-called free word order
languages.)
(ii) It is possible that the rule for VP is
4
The situation is even more complicated if we analyze last night to be an NP,
since then we appear to have the structure [VP V PP Adv NP PP].
5
For discussion of the ordering preferences in the English VP and the factors
that determine them, see Wasow 2002.
(15) VP → V0 (XP∗ )
and there are independent principles that determine the actual order in
which the various XPs may appear.
(iii) It is possible that there is richer structure in VP. The rules are essentially

PP
(16) a. VP → V (NP)
0
NP

PP
b. VP → VP
Adv
where each PP or Adv is contained within a different VP. The situation

illustrated here, whereby a phrase of a certain category contains a phrase of
the same category as in rule (16b), is called recursion.
As the following examples show, the verb in VP may be followed by a
sentence, which we notate as category S –
(17) I think [S (that) you made a mistake].
– or by NP followed by S –
(18) I told you [S (that) you made a mistake].
– or by PP followed by S.
(19) I mentioned to Sandy [S (that) you made a mistake].
VP-final S is somewhat preferred, but it can precede adverbs and PPs.

(20) a. I mentioned belatedly [that it was my birthday].
(?) I mentioned [that it was my birthday] belatedly.
b. I mentioned belatedly to Sandy [that it was my birthday].
(?) I mentioned [that it was my birthday] belatedly to Sandy.
c. I told Sandy (belatedly) (over the phone) [that it was my birthday].
(?) I told Sandy [that it was my birthday] (belatedly) (over the phone).
The examples with (?) are not grossly ungrammatical but infelicitous to
some extent.
An extension of rule (14) that places S in VP-final position is the follow-
ing.
(21) VP → V (NP)(NP)(PP∗ )(Adv∗ )(S)
This is the rule that we will assume as we proceed, for concreteness. (But
note that this rule is unable to account for the fact that S may also precede
other constituents of VP.)
4.3. THE STRUCTURE OF THE NOUN PHRASE 109
The fact that there is a preference for VP-final S is important, because it

suggests that the ordering within VP must be accounted for in terms other
than simply a phrase structure rule that specifies the sequence of categories.
Note that S is a complement of the V in these cases and often alternates
with NP, yielding a pattern illustrated in (22).
(22) a. (?) I mentioned [S that my name is “Robin”] to Sandy.
b. I mentioned to Sandy [S that my name is “Robin”].
c. I mentioned [NP my name] to Sandy.
d. ?I mentioned to Sandy [NP my name].
Here the preferred position of the NP is before the PP (compare (22c,d)),

and the preferred position of the S after the PP (compare (22a,b)). But if
the NP is more complex, its appearance after the PP is not problematic –
compare (23) and (22d).
(23) I mentioned to Sandy [NP the name of my lawyer].
And if the NP is a pronoun, it is excluded from the position after the PP.
∗
(24) I mentioned to Sandy [NP it ].
Facts such as these suggest that the ordering preferences in VP are deter-
mined at least in part by the syntactic complexity or “weight” of the con-
stituents and not by their syntactic categories or grammatical functions. To
summarize this point, we have developed an analysis of the VP that assumes
minimal branching structure. Using phrase structure rules to account for
the order of constituents in VP runs into difficulties, because the ordering
possibilities go beyond what a PSR is able to express.
Section 4.7 discusses an alternative perspective on phrase structure in
which there is maximal branching structure, as contrasted with the minimal
branching structure that we assume here. In the next few sections, we look
at the internal structure of other phrasal categories.
4.3. The structure of the noun phrase
Here are a few examples of English NPs.

(25) a. poodles
b.
furry poodles

these
c. furry poodles
the

these
d. furry poodles that I own
the
e. a poodle in the park
Superficially, the English NP satisfies the following generalizations:

(i) It is a projection of N.
(ii) A determiner or demonstrative, such as the/these, must be initial (e.g. the
poodle, not ∗ poodle the).
(iii) Adjectives typically precede the N (e.g. furry poodle, not ∗ poodle furry).
(iv) Determiners, quantifiers, and possessives precede the adjectives and the head
N in English.
(26) a. many (black) poodles
b. ∗ black many poodles

my
c. (many) poodles
Sandy’s

my
d. ∗ many poodles
Sandy’s
(v) PPs and other complex modifiers follow the N. For example,
(27) a. the poodle [S that I own]
∗
b. the [S that I own] poodle
c. a poodle [PP in the park]
∗
d. a(n) [PP in the park] poodle
e. the poodle [VP sitting by the door]
∗
f. the [VP sitting by the door] poodle
g. the poodle [VP shaved by the groomer]
∗
h. the [VP shaved by the groomer] poodle
The modifier that I own is called a relative clause and is of the category S; we
consider its internal structure and relationship to the head N in Chapter 9.
Sitting by the door and shaved by the groomer are VPs; they are called
reduced relative clauses when they are used to modify NPs, because they
paraphrase full relative clauses of the form which is VP, e.g. the poodle which
is sitting by the door.
On the basis of the generalizations in (i)–(v) we can characterize in a
preliminary way the English NP in terms of the following rules, recognizing
that it is possible to have more than one Adj before the N and more than
one PP or relative clause after the N.

Det
(28) NP → (Adj∗ ) N0 (PP∗ )(VP∗ ) (S∗ )
NP’s
We use the ∗ notation for what follows the head N, since, as in the case of
VP, there can be more than one constituent of each type following the head.
4.4. OTHER PHRASAL CATEGORIES 111
Again, this rule does not capture the full range of ordering possibilities of
the constituents that follow the head.
4.4. Other phrasal categories
Other phrasal categories are PP, AP, and AdvP. PP appears to have a head-
complement structure (29a,b), as well as some quantificational and descrip-
tive modifiers (29c), but no determiners (29d). Moreover, some prepositions
can be intransitive – they can appear without complements (29e,f).
(29) a. There was a dog [in the room].
b. We walked [into the room].

halfway
c. We walked two feet into the room .
all the way
d. I was [[this ∗ (high)] off the ground].

halfway
e. We walked two feet in , and stopped.
all the way
f. Put the book [down].
We will assume that the structure of the PP is as follows, where halfway, this
high, etc. are degree phrases (DegP).
(30) a. PP
DegP P0 NP
b. PP → (DegP) P (NP)
0
We see here, as in the case of NP and VP, that the head precedes the
complement.
AP and AdvP are closely related, are based on the same adjectival roots,
and have essentially the same structure. However, AdvP does not allow the
full range of complements that AP does.
(31) a. Sandy was very angry (at Chris).
Sandy walked into the room angrily (∗ at Chris).
b. Sandy was angry (that Chris was late).
Sandy walked into the room angrily (∗ that Chris was late).
The pre-head specifiers of AP and AdvP are degree terms such as very, how,
too, so, and this/that (for example He was this angry!). We categorize them
here as Deg.

PP
(32) AP → (Deg) A0
S
(33) AdvP → (Deg) Adv0
Phrases such as very furry poodles, where very furry is an AP, show that
what precedes N is AP∗ and not simply Adj.
(34) a [AP very furry] poodle
And an AP can follow the head N, as shown by a poodle soaking wet from
the bath.
(35) a poodle [AP soaking wet from the bath]
These observations suggest that rule (28) should be modified as in (28 ).

Det
(28 ) NP → (AP∗ ) N0 (AP∗ ) (PP∗ )(VP∗ ) (S∗ )
NP’s
These modifications, while motivated by the examples that we have consid-

ered, raise additional problems, which are explored in Research question 1
at the end of this chapter.
4.5. The English verbal sequence
4.5.1. Auxiliary verbs
We turn now to the English verbal sequence, which occupies a special place
in the study of phrase structure. In Chapter 3 we summarized the basic
structure of the English sentence in terms of the following rules.
(36) S → NP VP
(37) S → NP VAUX VP
These rules gloss over a number of details. The most important are (a) the
restrictions on the sequence of verbal elements, (b) the marking of finite
tense, and (c) the verbal morphology. In this section we consider these
matters in greater detail.
4.5.1.1. Describing the verbal sequence

The English verbal sequence displays several noteworthy properties.
(i) The sequence may contain a number of verbal elements, including

a modal (such as will, can, must), auxiliary have, and auxiliary be. For
example,
4.5. THE ENGLISH VERBAL SEQUENCE 113
(38) Sandy would have been sleeping at that time.

(ii) Each element determines the form of the verb that follows it. In partic-
ular
a. A modal requires that the verb that follows it have the bare form (see (39a)).
b. Perfect have requires that the verb that follows it be a past participle (see (39b)).
c. Progressive be requires that the verb that follows it be a present participle (see
(39c)).
d. Passive be requires that the verb that follows it be a past participle (see (39d)).
 

 ∗leave  
leaves
(39) a. Sandy will ∗ .

 ∗ left 

leaving
∗ 

∗ leave 

leaves
b. Sandy has .

 left 

∗
leaving
∗ 

∗ leave 

leaves
c. Sandy is ∗ .
 left 
 
leaving
∗ 

 ∗ leave  
leaves
d. Sandy was by Terry.

 left 

∗
leaving
(iii) The modal, have, be of the progressive, and be of the passive must
appear in the strict order given by (40), as exemplified in (41).
(40) Modal have beprogressive bepassive
(41) a. ∗ Sandy is having left. [cf. Sandy has been leaving.]
b. ∗ Sandy has would leave. [cf. Sandy would have left.]
c. ∗ Sandy is canning go. [cf. Sandy can go; Sandy can be leaving.]
(iv) Tense is marked on the first verb in a verbal sequence.
(42) a. Sandy leaves tomorrow.

is leaving
b. Sandy ∗ tomorrow.
be leaves
On the basis of these patterns, Chomsky (1957), in his seminal Syntactic
Structures, proposed that the English verbal sequence consists of the fol-
lowing, where Tense may be Present or Past.
(43) Tense (Modal) (have +en) (be +ing) V . . .
In his original analysis, Tense, +en, and +ing are attached to the element
immediately to the right of them, through the application of a transfor-
mational rule called Affix hopping. A transformational rule takes a tree
structure that meets certain conditions and changes it into another tree
structure. A sequence of one or more transformations is called a derivation.
(We will have much more to say about transformations and derivations in
subsequent chapters.)
The main advantage of separating the affixes and then attaching them
is that it captures the generalization that each affix appears on the word
to the right of the verb that is responsible for its appearance. Suppose, for
example, that the initial sequence is Past run. Affix hopping produces the
sequence run+Past, which is realized as ran. If the initial sequence is Present
[be +ing] run, Affix hopping produces the sequence be+Present run+ing,
that is, is running.
(44) Present be +ing run
If the initial sequence is Present will [be +ing] run, Affix hopping produces
the sequence will+Present be run+ing, that is, will be running.
(45) Present will be +ing run
And so on.
4.5.1.2. Structure in the verbal sequence

Affix hopping in general is no longer accepted as part of the analysis of
the English verbal sequence, although we will see that it may continue to
play a useful role if restricted to the distribution of finite tense. The rigid
restrictions on the sequence of verbal forms suggests that the modals and
the verbs have and be select the form of the verbs that follow them. We
will have much more to say about selection in Chapter 5. The general idea
that we work with here is that a head, in virtue of its lexical specification,
combines only with a complement that has a certain property or properties.
Assuming that verbs are heads of VPs, we can characterize the selection
in this case in terms of the morphological properties of the complement of
each verb in the sequence. First, the following is a possible phrase structure
rule for English. It allows for each auxiliary verb to take a VP as its
complement.
(46) VP → V0 VP
The verb that follows an auxiliary verb is the head of the VP complement
of that auxiliary verb. For example, be running will have the structure
(47) VP
V0 VP
be V0 …
running
Second, for each VAUX , we must specify the properties of its complement
VP. Furthermore, since a VP is a projection of V, the complement VP and
its head share certain properties. (This relationship between the head and
its projection is called the head feature principle. 6 ) In particular, if the VP is
required to have a property, such as perfect or progressive morphology, the
VP passes its properties down to its head. For be running we then have the
following.
(48) progressive be selects a VP[PROG. PART]. The head of this VP is

V[PROG. PART].
The structure is
(49) VP
V0 VP
[PROG. PART]
be
V0 …
[PROG. PART]
running
Similarly, the selectional properties of the other auxiliary verbs are
(50) perfect have selects a VP[PAST. PART]. The head of this VP is V[PAST. PART].
passive be selects a VP[PAST. PART]. The head of this VP is V[PAST. PART].
6
The head feature principle was originally formulated in Head-driven Phrase
Structure Grammar (HPSG); see Pollard and Sag 1994.
If there is more than one auxiliary verb in a verbal sequence, the structure
is correspondingly more complex. The structure in (51) illustrates for have
and progressive be
(51) VP
V0 VP
[PAST.PART]
have V0 VP
[PAST.PART] [PROG.PART]
been V0
[PROG.PART]
running
Consider next the modals, such as will and can. There is a detail that
needs to be addressed before we move on, which is that unlike have, be, and
main verbs, the modals do not show agreement in the present tense. The
English verbal paradigm distinguishes this form overtly: do ∼ does, have ∼
has, am ∼ is, sleep ∼ sleeps. But there is no such alternation in the case of
the modals.

will
(52) a. Sandy ∗ leave.
wills

can
b. Kim ∗ swim.
cans
Given that the modal forms are fixed, it is reasonable to ask whether there
is any evidence that they are in fact marked for finite tense. The evidence
that bears on this question involves reported speech. If John says “I am not
feeling well” and we want to report later what John said, we can quote John
literally, or we express what John said as a sentential complement. Notice
what happens to the form of am when we do this.
(53) a. John said “I am not feeling well.”

b. John said that he was not feeling well.
The present tense in a report about past speech shifts to the past tense.
We can use this observation to analyze what happens when there is a
modal in the reported speech.
(54) a. Mary said “I will never vote for that guy again.”
b. Mary said that she would never vote for that guy again.
(55) a. Mary asked “Can you help me with my homework?”
b. Mary asked whether I could help her with her homework.
For will and can, then, it appears that there are corresponding present and
past forms, will/would and can/could. There are some complexities that arise
when we consider other modals, but since these do not affect the main point
we leave them to Problem 2.
Let us now consider the interaction between modals, tense, and other
verbs. We have already observed that the verb that immediately follows a
modal is in the “bare” form. That is, it lacks morphological inflection, as
seen in (56).
 
 ∗leave 
 
leaves
(56) She will ∗ .
 ∗ leaving 
 
left
This suggests that modals select VPs with no morphological features.
Moreover, in addition to points (i)–(iv) above we have the following
observation.
(v) When a sentence with a modal is negated with not, not appears between
the modal and the VP.
(57) a. Sandy will leave.
b. Sandy will not leave.
∗
c. Sandy not will leave.
∗
d. Sandy will leave not.
4.5.2. Some generalizations
Let us put what we have observed about English verbal sequences with
several other observations. We have seen that
r only the first verb in the sequence is tensed;
r there is a strict order of verbal elements: modal precedes have, which precedes be,
which precedes the main verb;
r modal precedes not in negative sentences.
Note now that when there is no modal, tensed have or tensed be precedes
not.
(58) a. I have not been thinking about the terms of the proposal.
b. We are not staying in this rat-infested dump.
This fact suggests that have and be are also instances of VAUX , and they may
be tensed. Note also that in questions the part of the sequence that appears
before the subject NP is in fact the tensed VAUX .
(59) Will Sandy leave?
Has Sandy left?
Is Sandy leaving?
This construction is called subject-AUX inversion or simply inversion. The

rest of the sequence is precisely what we find in a non-inversion construc-
tion.
 
 ∗leave 
 
leaves
(60) a. Will Sandy ∗ .

 ∗ left 

leaving
∗ 

∗ leave 

leaves
b. Has Sandy .

 left 

∗
leaving
∗ 

∗ leave 

leaves
c. Is Sandy ∗ .
 left 
 
leaving
∗ 

 ∗ leave  
leaves
d. Was Sandy ∗ .

 left  
leaving
We can capture these facts by simply putting tensed VAUX before the subject
NP. We use the notation [TENSE] to indicate that VAUX is inflected.
(61) S → NP VAUX [TENSE] VP
4.5.3. Accounting for the sequence
Having established that a modal is properly analyzed as being inflected for

tense, we look again at the fact that only the first element is marked for
tense. But if tense is marked on an auxiliary verb, how do we account for its
appearance on the main verb when there is no auxiliary verb? Crucially, the
main verb cannot precede not, it must follow not –
∗
(62) Sandy left not.
[cf. Sandy didn’t leave.]
– and, in inversion, the main verb cannot precede the subject.

∗
(63) Left Sandy?
[cf. Did Sandy leave?]
We can’t have one rule that puts tense on the auxiliary verb and another
rule that puts tense on the main verb, because that would cause a conflict in
sequences that have auxiliaries.
∗
(64) Sandy will leaves.
It has been argued in the literature that the least problematic solution is
one that retains Affix hopping. 7 Our variant of this analysis assumes that
what precedes VP is not simply tensed VAUX but the node AUX that may
contain just the element TENSE, as well as an uninflected auxiliary verb, as
in (65).
(65) a. S → NP AUX (not) VP
b. AUX → TENSE (VAUX )
Sandy [AUX PAST will ] leave
TENSE then “hops” on the verbal element immediately to the right of it. If
there is an auxiliary verb, TENSE hops onto it –
(66) Sandy [AUX PAST will ] leave
– and, if there isn’t, it hops out of AUX onto the main verb.
(67) Sandy [AUX PAST] leave
When VAUX is have, the VP that is selected will have the fea-
ture [PAST. PRT ], and when VAUX is be, the VP will have the feature
[PROG. PART ]. Because VAUX may be a modal, have, or be, any of these
may function as the AUX and therefore precede not and undergo inversion.
In section 4.5.4 we look at several other uses of have and be and their
distribution.
Now we come to a problem, and an interesting solution. We have seen
that inversion puts AUX before the subject. If AUX happens to lack an
auxiliary verb, the structure will be something like the following.
(68) [AUX PAST ] [NP Sandy] [VP leave ]
By assumption, TENSE hops onto the verbal element that immediately

follows it. But, in this case, there is no such verbal element. If something
7
Lasnik 1999.
intervenes between TENSE and the verb, TENSE cannot hop. The problem
is, since TENSE is an affix, there must be something that TENSE attaches to.
The solution to this problem is based on the observation that do appears
as an auxiliary verb in English exactly in those contexts where there is no
verb that can have TENSE hopped onto it. One such context is shown in
(68); others are those where the verb is simply missing, or the VP is not
adjacent to AUX.
The generalization about do is that do has to appear in AUX when TENSE
is not immediately adjacent to a verb, so that there is something for TENSE
to hop on to. This phenomenon is called do support – if there is no verbal
element, then it is necessary to insert do to “support” TENSE. (69) shows
some of the possibilities for attachment of TENSE. Note in particular the
introduction of do in (69d–f).
(69) a. Sandy [AUX PRESENT will] leave → Sandy will[PRESENT ] leave

‘Sandy will leave.’
b. Sandy [AUX PAST be] sleeping → Sandy be[PAST ] sleeping
‘Sandy was sleeping.’
c. Sandy [AUX PAST will] have left → Sandy will[PAST ] have left
‘Sandy would have left.’
d. Sandy [AUX PRESENT] not smoke → Sandy [AUX PRESENT do] not smoke
→ Sandy do[PRESENT ] not smoke
‘Sandy does not smoke.’
e. [AUX PAST] Sandy leave → [AUX PAST do] Sandy leave
→ do[PAST ] Sandy leave
‘Did Sandy leave.’
f. They said Sandy would leave, and [VP leave] Sandy [AUX PAST ]
→ They said Sandy would leave, and [VP leave] Sandy [AUX PAST do]
→ They said Sandy would leave, and [VP leave] Sandy do[PAST ]
‘They said Sandy would leave, and leave Sandy did.’
Thus, do will appear if not appears between AUX and VP, if AUX is moved
away from VP, if VP is moved away from AUX, or if the V is absent.
do support
S → NP AUX (not) VP
AUX → TENSE (VAUX )
1. Insert do if TENSE is not adjacent to a verbal element.
2. TENSE hops onto the adjacent verbal element (Affix hopping).
Exercise 4 asks you to derive a number of verbal sequences with and without
do.
4.5.4. Have and be
Have and be function not only as auxiliary verbs but as main verbs.
(70) a. You have a cold.
b. You are not well.
In the case of main verb have we get do support (71a), but not in the case of
main verb be (71b).
(71) a. Do you have some money?
b. ∗ Do you be well?
Inversion shows that main verb have also functions as an auxiliary verb in
AUX in British English and in limited cases in American English (72a,b),
while main verb be must be in AUX – compare (72c) and (71b).
(72) a. (Brit.) Have you some money?
b. Have you any idea how much that sofa costs?
c. Are you well?
The distribution of not is consistent with this structure.

(73) a. I haven’t a clue how much that sofa costs.
b. You aren’t really well.
Since have and be can function as AUX with respect to inversion and still
select complements as though they were heads of VP, one might assume that
they are restructured from VP into AUX when there is no VAUX in AUX, as
sketched in (74).
(74) S S
NP AUX VP ⇒ NP AUX VP
TENSE VAUX … TENSE VAUX …
However, a more direct solution that does not require a special restructuring
operation is to simply say that have and be correspond to VAUX in AUX, and
that the complements that they select are not in VP but attached directly to
S. For example, if the complement of be is an AP such as well, the structure
would be that of (75).
(75) S
NP AUX AP
TENSE VAUX well
be
The key is to specify in the lexical entry of be that it is always VAUX and the
lexical entry of have that it is VAUX under certain circumstances.
The correspondence that specifies the position of the auxiliary verb is the
following.
(76) The leftmost VAUX in the sequence appears in AUX.
In order for such a solution to work, the AUX will have to be treated as
a head with the selectional properties of the auxiliary verb. The auxiliary
verb must be able to select its complement (e.g. AP in the case of be) even
if both are dominated by S and not VP. We assume both points here. Note
also that, on this analysis, the complement is not a VP.
4.6. Rule summary
The following are the phrase structure rules for English that we have devel-
oped thus far.
(77) S → NP AUX (not) VP
AUX → TENSE (VAUX )
VP → V0 (NP)(NP)(PP∗ )(Adv∗ )(S)

Det
NP → (AP∗ ) N0 (AP∗ ) (PP∗ )(VP∗ ) (S∗ )
NP’s
PP → (DegP) P0 (NP)

PP
AP → (Deg) A0
S
AdvP → (Deg) Adv0
These rules specify the position of the head in the phrase and the relative
ordering of specifiers, arguments, and adjuncts. Moreover, they suggest that
phrases have flat structure, in the sense that all of the constituents are
sisters of one another. The flat structure hypothesis is the weakest version of
4.7. APPLICATIONS OF X THEORY 123
X theory, one in which there are only two levels of structure, X0 and XP. It is
also the simplest hypothesis, in the sense that it makes minimal assumptions
about structure. The alternative view, that phrases are maximally struc-
tured, is taken up in section 4.7.
4.7. ∗ Applications of strong X theory
As noted in section 4.1, mainstream generative grammar has adopted a

very strong version of X theory, expressed in terms of the X schema (4),
with the additional assumption that all branching is binary, so that Spec
and Comp each consist of exactly one constituent. This assumption of uni-
form binary branching is a standard assumption of mainstream approaches
(but not, for example, in the approach taken in this book; see Chapters 3
and 4).
(4) XP
Spec X
X0 Comp
In this section we explore some of the consequences of these assumptions,

without necessarily endorsing them.
4.7.1. IP and CP
We consider first the question of what the structure of S is if we assume

uniform endocentricity of all phrases, in conformity with strong X theory.
Assume that the X schema in (4) applies to S. Furthermore, assume that all
branching is binary. Based on our discussion in section 4.5, AUX is a plau-
sible candidate for the head of the sentence, where Spec is the subject and
Comp is the VP. 8 Since AUX contains the tense inflection, it is standardly
referred to as Infl, or I0 .
8
On a strict application of X theory, the auxiliary verbs have and be will always
originate as heads of VP; hence I0 will contain at most Tense and the modal.
(78) IP
NP I
I0 VP
tense (M)
Assuming strong X theory, it is standardly assumed that the comple-

mentizer that is also a head and that its projection is a complement S. The
complementizer is referred to as C0 , and its projection is CP.
(79) CP
Spec C
C0 IP
that NP I
I0 VP
tense (M)
We consider evidence for this structure in Chapter 9. The most important

type of evidence is that Spec of CP appears to be the position in which wh-
phrases appear in wh-questions such as What are you looking at, while C0 is
the position of the inverted auxiliary.
The structure in (78) is assumed in mainstream syntactic theory to be cor-
rect for languages that show inflection in finite sentences. A major difference
between such languages and English is that it appears that, in languages
such as French and German, all inflected verbs behave as though they are
in I0 , in that they precede negation and undergo inversion. For example, in
French we find the following.
(80) French
a. Marie ne regarde pas la télévision.
Marie NE look-at NEG the television
Lit. ‘Marie watches not the television.’
‘Marie isn’t watching television.’
b. ∗ Marie ne pas regarde la télévision.

Marie NE NEG look-at the television.
c. Regarde-t-elle la télévision?
watches-she the television
‘Does she watch television?’
In comparison, in English only the tensed auxiliary verbs precede negation

and undergo inversion.
The similarities between the behavior of main verbs in French (and other
languages) and auxiliary verbs in English can be captured by taking the
inflection, that is, TENSE, to be the head of the sentence. Since V is the head
of VP, assuming that structure (78) is uniform across all languages leads to
an analysis in which the verb is raised from VP to I0 by an operation called
V-raising, illustrated in (81).
(81) IP IP
NP I ⇒ NP I
I0 VP V0 + I0 VP
V0 … …
4.7.2. DP
On the assumption that (78) is the correct structure for IP(=S), and
assuming that it is important to maximize uniformity of structure across
categories, a similar structure becomes plausible for the NP. Recall
that the subject of IP is external to I0 – it is a sister of the first
node above I0 . The complement of I0 is VP, which contains V and its
complements.
By analogy, the specifier of an NP would be external to a head whose
complement is the noun and its complements. Call this head D(et)/D0 ; then
what we have been thinking of as an NP is a DP, that is, a determiner
phrase.
(82) DP
… D
D0 NP
Spec N
N0 …
The head D0 , the counterpart of the head I0 , is a member of the class of

functional heads, which also includes C0 . This class is contrasted with the
lexical heads, that is, N0 , V0 , A0 , and P0 . The projection DP is called a
functional projection, as are IP and CP.
The functional head D0 is the position of a number of pronominal ele-
ments, including the possessive inflection ’s, the determiners the and a, and
demonstratives such as this and these.
(83) DP
DP D
Kim D0 NP
’s Spec N
N0 PP
painting P DP
of Sandy
DP
D0 NP
this Spec N
N0 PP
painting P DP
of Sandy
This structure does raise certain difficult questions, though, which we can
only note here. First, while the subject position (that is, Spec of IP) must be
occupied in the sentence, the corresponding position (that is, Spec of NP)
does not have to be occupied in NP. Why is this, and when is it possible for
a position to be empty? The answer given in the framework of strong X
theory is that some heads have features associated with them that require
that there be a matching constituent in their Spec. One such head would be
the genitive ’s, which requires that there be a DP in Spec. By contrast, this
has features that require Spec to be empty.
Second, what is the status of adjectival modifiers? Are they heads, or
are they specifiers, or are they adjuncts? The most uniform analysis treats
them as heads. If they are heads, and precede the NP, then they too must
be heads of projections that can be complements of D0 , as illustrated
in (84).
(84) DP
D
D0 AP
this A0 NP
expensive Spec N
N0 PP
painting P DP
of Sandy
Moreover,
r what is the internal structure of a proper noun phrase, such as Sandy? Is Sandy an
N0 , and, if so, what is the structure of the DP?
r what is in the Spec of NP?
r what is the internal structure of expressions such as very expensive that contain
degree modifiers?
r how are the head features of N0 transmitted to DP, and to what extent are the
head features of A0 transmitted to DP?
r what is the internal structure of a determinerless phrase such as expensive paint-
ings? Is it a DP?
For discussion of a range of questions bearing on the internal structure of

DP, see Abney 1987 and the papers in Alexiadou and Wilder 1998.
4.7.3. VP internal subjects
Introduction of IP allows MGG to address a puzzle concerning the appli-

cation of strong X theory to the VP. Although the strong X schema
suggests that VP should have a specifier, there is no obvious candidate
for this position. At the same time, the subject of a sentence (in English
at least) is in the specifier position of IP. If all phrases conform to the
X schema, then VP should allow a DP in its specifier position. But, on
the face of it, this appears to be impossible, as examples like the following
show.
∗
(85) Sandy will Terry read the book.
The problem is that Terry has no role in this sentence; it is completely

superfluous.
A solution that reconciles the requirement that VP obey the X schema
with the ungrammaticality of (85) starts from the observation that,
although the subject of a sentence is in Spec of IP, it gets its semantic role
from the verb. For example, in Sandy will read the book, Sandy is the Agent
of read. But Sandy is not in the VP headed by read, so it is not clear how it
gets this role. (Of course, this problem arises precisely because it is assumed
that there is an IP above VP, in contrast to the flatter structure that we
assume earlier in this chapter.)
One possible resolution of this problem that has been widely assumed is
that there is some rule of predication by which the role that the verb must
assign to its subject is transmitted from the VP to the Spec of IP. Such a rule
is needed in any analysis in which the subject is external to the VP, including
the one given earlier in this chapter, where the subject may be the sister of
a VP that contains a number of auxiliary verbs, e.g. Sandy would [VP have
[VP been [VP sleeping]]].
But the solution in MGG is to say that the subject of the sentence is a
constituent of the VP, and in fact that it is the specifier. This step would
solve the problem of making VP conform to the X schema, as shown in
(86).
EXERCISES 129
(86) IP
Spec I
I0 VP
DP V
Sandy V0 DP
read the book
This analysis is called the VP internal subject hypothesis (VPISH).

There are a number of problems raised by this analysis. Not the least is
that the subject of a sentence in English does not appear to the right of the
I0 head but to the left of it in a normal declarative. So, if this analysis is
correct, the DP in Spec of VP must move to the Spec of IP. Why does it
move, and what happens if it does not move? Moreover, what happens if
there are several auxiliaries, each of which is the head of its own VP? In
such cases, there are presumably many Spec positions (one for each VP),
and the subject must move up through each of them. Such questions do
not arise in an account that does not assume a strict X analysis of the
sentence. But given the prominent position of the VPISH analysis, these
are questions that we need to recognize. We return to them in Chapter 6.
Finally, assuming a uniform X configuration for all projections, the struc-
ture in (86) generalizes to DP. The possessive is the Spec of NP, and raises
to Spec of DP, where it shows the possessive morphology, as in (83).
Exercises
1. Some of the following verb phrases are compatible with rule (14) in the
text and some are not. Say which are and which are not, and why. To do this,
you must show that a particular phrase fits or fails to fit the requirements
of this rule.
(1) a. (We should really) [VP give back the money].
b. (I) [VP bet Sandy ten dollars that it would rain].
c. (Pat) [VP hung a copy of the Mona Lisa on the wall].
d. (Sandy) [VP said in a very loud voice that it was time to leave].
e. (Kim) [VP tried very hard to ignore what was happening].

f. (You never should have) [VP put the food back in the refrigerator].
[§4.2.]
2. The head of the verb phrase is underlined in the following sentence:
(1) Cathy put a picture of Elvis on the wall of her bedroom last night.
i. Explain why we say that put is the head.
ii. Is on the wall of her bedroom a complement or an adjunct? Why?
iii. Is last night a complement or an adjunct? Why?
[§4.2.]
3. Some of the following noun phrases are compatible with rule (28) and
some are not. Say which are and which are not, and explain your answer.
To do this, you must show that a particular phrase fits or fails to fit one or
more requirements of this rule.
(1) a. two left feet
b. people angered by the decision
c. expensive paintings of rural scenes
d. space, the final frontier
e. a great place to work
f. an unforgettable little black poodle
[§4.3.]
4. Show the derivation of the following verbal sequences in English. To

do this, give the original sequence and show each step, following the model
of (69) in the text.
(1) a. would have been sleeping
b. could not be reading
c. might have seen
d. did not realize
e. . . . and [VP suffer] they did
[cf. . . . and [VP suffer] they will
f. . . . and they do.
[cf. . . . and they will.]
[§4.5.]
5. Following the pattern of (51) in the text, draw trees for the ver-
bal sequences in (1a–d) in Exercise 4. Show all verbal features such as
[PAST. PRT ] as appropriate.
[§4.5.]
PROBLEMS 131
6. Put brackets around every phrase in each of the following sentences

and label it with its category. Use [S . . . ] for the category S, and [NP . . . ] for
the category NP, and so on for the other categories. Remember, there have to
be matching left and right brackets, and the label goes on the left bracket.
For example, [NP my picture [PP of [NP Sandy]]]. (Hint: You might find it
easier to draw the tree first, and then translate it into labeled brackets.)
(1) a. Your resistance to advice is legendary.

b. Cathy put a picture of Elvis on the wall of her bedroom last night.
c. The quick brown fox jumped over the lazy dog.
d. We introduced a simple version of the phrase structure rules for the English
VP in Chapter 3.
e. You have a very good reason for your distrust of George.
[§4.6.]
7. Use the rules in (1), and draw the trees for each of the phrases in (2).
Assume the categories given in (3).

Det
(1) NP → (AP∗ ) N0 (AP∗ ) (PP∗ )(VP∗ ) (S∗ )
NP’s
PP → (DegP) P0 (NP)
(2) a. my dog
b. pictures of my dog
c. Sandy’s pictures of Robin’s dog
d. Sandy’s stories about Robin’s pictures of my dog
e. my stories about Robin’s pictures of Sandy
(3) NP: my, your, Sandy’s, Robin’s
N0 : dog, Sandy, stories, pictures
P0 : of, about
[§4.6.]
Problems
1. Here is an example of an NP in Niuean.
(1) Niuean
e tau manu kula fulufuluola e:
A BS C PL bird red beautiful that
‘those beautiful red birds’
[Kahnemuyipour and Massam 2002]
The marker A BS C is “a portmanteau morpheme, which indicates the case

of the [NP] as well as whether it is common or proper (where proper includes
pronominal)”.
Formulate a phrase structure rule or simple set of phrase structure rules
that will account for the structure of the Niuean NP, using the rules for
English as a guide.
[§4.3.]
2. The simple description of the English verbal sequence assumes that
a modal is always tensed. We have argued that there are present and past
forms for will and can, which can be seen when they are put into reported
speech in the past, e.g.
(1) a. Susan said “I will leave.”

b. Susan said she would leave.
There does not appear to be a past tense form for the English modal must;
in order to express the past of must the periphrastic form had to is required.

must
(2) a. Susan said “I leave.”
have to
∗
musted
b. Susan said she leave.
had to
What is the best way to express this fact in formal terms in the grammar of
English? (Hint: What must happen when the modal must appears in AUX
with PAST Tense? Where is this stated in the grammar?)
[§4.5.]
3. In section 4.5.4 we suggested that the structure of a sentence like (1) is
(2).
(1) My poodle is furry.

(2) [S NP [AUX is] [AP furry]]
Such a structure lacks a VP constituent. Revise the rules in (77) to accom-

modate this possibility, as well as cases in which the complement of be is
NP or PP.
[§4.5.]
4. The following are passages from Shakespeare’s Henry VIII. There are
four passages in boldface that in some way differ in grammatical form
from Modern English, identified as (a)–(d). Describe each difference in
grammatical terms, that is, in terms of what the relevant elements are, where
PROBLEMS 133
they appear, what their syntactic properties are, and what rules of contem-
porary English do or do not apply to them. You will have to refer to the
structure of the English verbal sequence developed in section 4.5. (Don’t get
distracted by the apostrophes – they are spelling differences, not syntactic
differences.)
(1) a. would all other women

Could speak this with as free a soul as I do!
b. My lords, I care not, so much I am happy
...
c. But, I beseech you, what’s become of Katharine,
The princess dowager? how goes her business?
d. Thou hast the sweetest face I ever look’d on.
Sir, as I have a soul, she is an angel.
[§4.5.]
5. We noted in the text that the complementizer that is required when a

sentential complement is in subject position or topicalized. State as simply
as you can the conditions under which that can be omitted. In order to
do this you will need to make up your own examples in which sentential
complements appear in different positions in the sentence. We’ve given you
a head start in (1) – the judgments are ours and yours might be different.
(The notation “∗ (that)” means that the sentence is not grammatical when it
lacks the that.)
∗
(1) a. (That) the world is flat, I doubt.
∗
b. (That) the world is round bothers me.
c. It bothers me (that) the world is round.
d. It bothers me very much ?(that) the world is round.
e. I proved (that) the world is round.
f. I proved conclusively ?(that) the world is round.
[§4.7.1.]
6. Consider whether there is evidence for the internal structure of IP

given in (78) (in particular I ) in terms of movement, deletion, or pro-
form replacement, using examples to test each possibility. Then, consider
whether there is evidence for or against this structure using coordination
tests.
(78) IP
NP I
I0 VP
tense (M)
[§4.7.1.]
Research questions
1. State as simply as you can a generalization that accounts for the pattern
seen in (1).
∗
(1) a. a [soaking wet from the bath] poodle
b. a [soaking wet] poodle
∗
c. a poodle [soaking wet]
d. a poodle [soaking wet from the bath]
Can this generalization be formulated just in terms of phrase structure

rules? Note also the grammaticality of the following examples. Do they
simplify or complicate your account? Why?

someone
(2) a. I met ∗ nice yesterday.
a person

anything
b. Do you have ∗ cheaper?
a room

somewhere
c. I want to go ∗ warm.
to a place
[§4.4.]
2. The rules that we have stated for the English verbal sequence do not
account for sentences such as the following.
(1) a. I can’t believe you did that.
b. You shouldn’t have said anything.
c. Won’t you please sit down.
d. Don’t they know that it’s raining.
The phenomenon illustrated here is called negative contraction.

Work out a modification of the rules to account for such sentences.
Most importantly, you must specify where the contracted form of negation
appears in the structure, and provide motivation for locating it in this

position.
[§4.5.]
3. It is traditional in generative grammar to view not in English as being
responsible for do support because it intervenes between I0 and the verb.
Consider the following data, which complicates the issue.

will
(1) a. Sandy not call.
did

will certainly
b. Sandy ∗ call.
did never

certainly
c. Sandy called.
never
A. Explain why this data is problematic for the view that anything that
intervenes between Tense and V automatically triggers do support.
B. What adjustments would you have to make in the analysis proposed in
the text in order to accommodate these facts?
[§4.5.]
4. Given the analysis of the English verbal sequence discussed in the text,
what is the status of the infinitive? Here are some relevant examples.
(1) a. to leave
b. to have left
c. to be leaving
d. to have been leaving

can
e. ∗ to leave
will

not to leave
f. I expect Sandy .
?to not leave
A. Suppose that to is a non-finite variant of AUX. Assuming this, how

should we change the rule that specifies the content of AUX?
B. What has to be said about the distribution of not in order to account
for example (1f)? Compare the distribution of not in this case with its dis-
tribution in finite sentences. Is there a natural way to capture the difference
that goes beyond simply stating the facts? (Hint: You may have to develop
a very different analysis of the structure of the verbal sequence in order
to account for the behavior of not, and this account may have interesting
consequences.)
[§4.5.]
5. The do support analysis given in the text is a version of the classical

analysis of Chomsky 1957. A number of alternatives have been explored in
the literature over the years, with the goal of eliminating Affix hopping and
accounting for the appearance of do on the basis of general principles rather
than a specific rule.
Assume for this problem that do is a modal. Formulate the simplest set
of rules that will account for the distribution of do. Make sure that when do
is present it is marked for TENSE. You will have to specify the conditions
under which do is deleted, since it does not appear before auxiliary have and
be.
[§4.5.]
6. An appealing alternative to the do support analysis in the text is one

that eliminates Affix hopping and assumes that all verbs (auxiliaries and
main verbs) are introduced into the structure already marked for TENSE.
One instantiation of such an analysis assumes the following.
(1) S → NP VP[TENSE]
Here, [TENSE ] is a feature of VP, and therefore V will have the feature
[TENSE ].
Such an analysis turns out to be difficult to work out in detail, because
of the fact that a tensed auxiliary verb has a different distribution from a
tensed main verb. VP may be headed by a main verb or by a VAUX . In the
latter case, V may precede not, and may precede the subject NP in cases of
inversion.
Work out a set of rules and conditions to govern the distribution of
TENSE and do without assuming Affix hopping. How does your analysis
compare to the do support analysis that assumes Affix hopping? (This is a
particularly challenging problem.)
[§4.5.]
7. A view that is commonly encountered in mainstream syntactic theory
is that a single phrase structure for each phrasal category is basic to all
languages. For example, it has been proposed that the DP has the following
basic structure.
(1) [D0 [ Quan0 [ Adj0 [ N0 ]]]]
where all branching is binary and to the right. On this view, other possible
orders must be derived from this basic structure by movement of con-
stituents to the left. By assumption, heads can only attach to heads. For
example, the order D0 –Quan–N–Adj would be derived by adjoining the N

to the left of Adj.
i. Given the assumption of fixed initial structure and order, how would you
derive
(2) N0 D0 Quan0 Adj0
ii. Does the assumed structure and leftward movement derivation explain
why it is that the following are not possible orders? Explain.
∗
(3) N0 Quan0 D0 Adj0
∗ 0
N Adj0 D0 Quan0
∗
Adj0 N0 D0 Quan0
∗
Quan0 N0 D0 Adj0
∗
Quan0 D0 N0 Adj0
[§4.7.2.]
Section Exercise Problems Research questions
4.1.
4.2. 1, 2
4.3. 3 1
4.4. 1
4.5. 4, 5 2, 3, 4 2, 3, 4, 5, 6
4.6. 6, 7
4.7. 5, 6 7
5
Conceptual structure
and the lexicon
5.1. Overview
The primary function of human language is to express meaning. A speaker

of a language who intends to express a meaning knows how to make sounds
that will convey this meaning to another speaker of that language through
the particular form of the utterance. 1 Syntax is the system that mediates the
correspondence between form and meaning. Syntacticians study what this
mapping consists of for particular languages, and what it may consist of in
principle.
The description of the sound part of the sound/meaning mapping is
relatively accessible (although by no means trivial), owing to the fact that
it is possible to say in physical terms what the sounds are. This is the
province of phonetics. However, a speaker’s knowledge of the sounds of a
language goes well beyond the physical properties of the sounds. Speech
sounds are perceived as distinct segments in the continuous speech stream.
For each language there is an intricate abstract structure that accounts for
how sounds are related to one another and how they are strung together to
form words. Moreover, the sound of a language involves accent, stress, and
intonation, which are laid over the speech sounds. All this is the province of
phonology, which has important connections with syntax.
On the other side of the mapping we have meaning. Unlike phonology,
meaning has no direct physical correlates. We can ascertain the meaning of a
linguistic expression only by making use of our intuitions about the things,
1
Sign languages express meaning through gesture, and many languages also
convey meaning through writing. For simplicity of exposition we will refer here only
to the correspondence between sound and meaning.
140 5. CONCEPTUAL STRUCTURE AND THE LEXICON
properties, and relationships that it refers to. The most basic intuition in the
domain of meaning that we can appeal to is that two expressions A and B
have the “same meaning”. A more subtle intuition is that the meaning of
A is part of the meaning of B. When such intuitions hold, we can represent
the relationships between meanings by using identical symbols for identical
meanings or identical parts of meanings. For example, if the symbol for
the meaning of book is BOOK, then this symbol should appear in the
representation of the meaning of expressions such as my book, my books,
this book, I read a book, the book that I read, bookstore, and so on.
The symbols are not the meanings themselves, just as symbols for speech
sounds are not the sounds themselves. They stand for the meanings. Hence
they provide us with the means for exploring the relationships between
meanings, and, moreover, the relationships between meanings and the
sounds that are used to express them.
The particular symbolic system that we will use here to represent mean-
ings is a simplified version of Conceptual Structure (CS). 2 We will be con-
cerned particularly with developing a representation for meaning in which
we can describe how the meaning of a linguistic expression and its gram-
matical structure are related to one another. Having established the basic
picture of how meaning and grammatical structure are related, we can begin
to develop accounts of why and how certain linguistic expressions have the
same or related meanings. As we suggested in Chapter 1 and will see in more
detail as we proceed, accounting for sameness of meaning has been the
most powerful methodological force in the development of contemporary
syntactic theory.
5.2. Correspondences
5.2.1. Concepts
The theory of conceptual structure reflects a mentalistic approach to mean-

ing. It assumes that the relationship between a linguistic expression and
what exists in the world is mediated by concepts in the mind, that is,
2
Developed in Jackendoff 1983, Jackendoff 1990a, Jackendoff 2002 among other
references.
5.2. CORRESPONDENCES 141
meanings. To see how this works, consider first the following linguistic
expressions and what they refer to in the world.
(1)
Linguistic expression Reference in the actual world
Peter Culicover The author of this book

the present King of France [no reference]
a unicorn [no reference]
George Washington the (deceased) rst President
of the United States
As can be seen even from these few examples, what a linguistic expression
means must be distinguished from what it refers to (or does not refer to) in
the actual world. The present King of France and a unicorn have the same
actual reference, since they do not refer to anything in the world, but they
do not have the same meaning. George Washington refers to someone who
does not currently exist.
One way that philosophers have dealt with this problem is to take a non-
mentalistic approach, in which the meaning of an expression is formulated
in terms of the correspondence between the expression and the state of
affairs in some actual or possible world, without involving the mind. 3 On
this view, the reference of expressions like the present King of France and a
unicorn are not what they pick out in the actual world but what they pick
out in some possible world in which these individuals exist. The thing that is
picked out by the present King of France in some world is different from the
thing that is picked out by a unicorn, even though neither has a reference
in the actual world. In this way it is possible to distinguish between the
meanings of non-referring expressions.
Our approach will be different. We assume that meanings are composed
of concepts. There are basic concepts, and there are complex concepts that
have structure. The meaning of a word or a phrase is its conceptual structure
(CS). The conceptual structure stands between the linguistic expression and
what it refers to in the world.
We focus primarily on the relationship between the linguistic expression
and its corresponding concept. That is, we are interested in how the syn-
tactic structure of an expression determines its conceptual structure. Our
concern is with the contribution that language makes to determining the
3
Stalnaker 2003.
meaning that we have in our minds, not with how these meanings are related
to our real or imaginary experience.
An expression refers to something in the world if the concept that
corresponds to it picks out something in the world. A concept may or
may not correspond to something in the actual world, or in some pos-
sible or imaginary world, or in the future. In the table in (2) we repre-
sent the CS representations in CAPS to distinguish them from linguistic
expressions.
(2)
Linguistic expression Conceptual Structure Reference in the actual
world
Peter Culicover PETER-CULICOVER the author of this book

the present King of PRESENT-KING-OF- [no reference]
France FRANCE
a unicorn UNSPECIFIED- [no reference]
UNICORN
To the extent that two expressions have different CS representations, they

are referentially distinct as far as the mental representation of the ideas is
concerned. This does not mean that they actually refer to different things in
the world. For example, the philosopher Gottlieb Frege’s famous example
of the Morning Star and the Evening Star showed that two linguistic expres-
sions can have the same reference in the world even though the speaker of
the language conceives of them as different. The table in (3) illustrates.
(3)
Linguistic expression Conceptual Structure Reference in the actual
world
Morning Star MORNING-STAR the planet Venus

Evening Star EVENING-STAR the planet Venus
the planet Venus SPECIFIED-PLANET- the planet Venus
VENUS
Where two meanings are different, there are different concepts.

On the other hand, it is possible to use two distinct linguistic expres-
sions with the same meaning, e.g. bachelor and unmarried man, or cellar
and basement (for many speakers). In simple cases we call this synonymy;
alternatively, we say that a single concept corresponds to two linguistic
expressions.
The CS representation is the representation of the meaning that a speaker

intends to convey by using a particular linguistic expression, or that the
hearer constructs on the basis of apprehending a particular linguistic
expression. The question of whether the meanings are linked to the same or
different things in the world is, for us, independent of the intended meaning
and not a matter of language.
5.2.2. Indices
The few examples that we have looked at thus far give a very incomplete
picture of how meanings are represented as CS. Expressions like George
Washington or the Morning Star correspond to unique concepts in CS,
and they are therefore represented by unique symbolic representations, i.e.
GEORGE-WASHINGTON and MORNING-STAR, respectively.
If two distinct and unique things have the same name, we have to resort
to special means of distinguishing them. If, for example, there are two
people named Mary Smith, then we will have to use indices or numbers
to distinguish them in CS, say · and ‚, or 35 and 1066. We may call the
concept corresponding to one of these people MARY-SMITH-· or MARY-
SMITH· , or even MARY-SMITH-35, and that corresponding to the other
MARY-SMITH-‚, or MARY-SMITH‚ , or MARY-SMITH-1066.
Similarly, if we say “This book (pointing) is expensive but this book
(pointing to another book) is not”, we have to represent the (concepts
corresponding to the) two books as distinct in the meaning of the sentence.
What is important here as far as the meaning is concerned is not that the
linguistic expressions are the same or different, it is that the CS indices are
different.
Moreover, it doesn’t matter whether the concept actually corresponds
to something real or something imaginary – as far as the language is
concerned, there is no distinction. Thus, Santa Claus in language cor-
responds to SANTA-CLAUS„ in conceptual structure with some index
that distinguishes him from Superman, Batman, and Robin, all fictional
characters.
Technically, we do not even need to use the letters MARY-SMITH in
order to distinguish the concepts as long as each distinct concept has a
distinct index. However, we will find it more convenient to spell out the con-
cepts than to use indices or numbers to distinguish them. For consistency,
the notation that we use in this book for reference is superscript Greek
letters.
5.2.3. Lexical entries
Let us consider now the expression a book. If someone says “I just bought
a book” we know what they mean although we may not know what specific
object they are referring to. In this case, again, we can use an index to
mark the reference of this expression, again not being particularly con-
cerned about what this index is as long as it does not lead us into con-
fusion. But, no matter what book is being referred to, or even if there is
no actual book, speakers of English know certain things about the word
book:
r that it is pronounced a certain way (/bVk/);
r that it is a noun and therefore can combine with other words to form a noun
phrase in a certain way;
r that it expresses a certain concept, namely BOOK.
This knowledge is part of the lexical information associated with book,

along the lines discussed in Chapter 2. We introduced the attribute-value
matrix (AVM) there as a uniform notation for representing the phonological
and grammatical information associated with a linguistic expression. Now
we add the meaning information in the form of a CS representation.
A basic AVM for book is as follows. It has entries for the phonological
information (PHON), the syntactic information (SYNTAX), and the con-
ceptual structure (CS).
 
(4) PHON /bvk/
 
 SYNTAX [CATEGORY N] 
CS BOOK
We call the totality of knowledge of this sort the lexicon of a language. As we

can see in the simple case of book, the lexicon expresses the correspondences
between the sounds, the grammatical properties, and the meanings of the
basic elements of the language. These elements are the words and other
formatives, such as bound morphemes – the lexical items. To the extent that
more complex expressions, such as idioms, also have particular meanings
associated with them that cannot be accounted for in terms of their struc-
ture, they must also be included in the lexicon.
The lexical entry of the word book specifies how the sound sequence
corresponds to the syntactic structure and the meaning. (5) shows how a
typical correspondence is represented in the lexicon.
(5) PHONETICS … /bVk/ …
SYNTAX … [N book] …
CS … BOOK …
When a lexical item is part of a more complex expression, it is necessary to

represent which part of the sound and syntactic representation corresponds
to which part of the meaning. In such cases we may use arrows and numeri-
cal subscripts, or both, as shown in (6), to keep things straight, especially if
the spelling does not do the job.
(6) PHONETICS … /bVk/1 …
SYNTAX … [N book]1 …
CS … BOOK1 …
The correspondences that are expressed in the lexicon for individual words
form part of the correspondences for larger expressions, with the addition
of subscripts, referential indices, and other notations to help us keep track
of them.
Just as individual words are represented in the lexicon as correspondences
between sound, syntactic properties, and meaning, so are idioms. Consider
the idiom kick the bucket, which means “die”. The phonological form of the
idiom consists of the three words kick, the, and the bucket. The entire idiom
is a verb phrase, which we represent in the syntax part of the correspon-
dence. And the meaning is DIE(. . . ).
For practical reasons we typically omit the part of correspondences
involving sound in what follows, and focus on the part that involves syntax
and CS. Example (7) shows the correspondence for kick the bucket. We use
subscripts to show how the parts of the syntactic representation correspond
to the parts of the phonetic representation.
(7) PHONETICS /kIk1D@2b2kIt3/
SYNTAX [VP[Vkick1] [NP[Detthe2] [Nbucket3] ] ]
CS DIE (…)
5.3. CS relations
Along with distinguishing the reference of noun phrases in a sentence,

the most important function of conceptual structure is to represent the
(conceptual) relations between what the noun phrases refer to. Consider
the following examples.
(8) a. Mary is writing a book.

b. Mary gave John a book.
c. This cheese stinks.
d. It’s raining.
The relation expressed by (8a) is one that holds between two objects, the
person Mary and some book. We represent this relation in CS as WRITE
with two arguments, MARY· and BOOK‚ .
(9) WRITE(MARY· , BOOK‚ )
In the lexical entry for WRITE we represent the information that the
first argument is the writer, and the second argument is the thing written.
These are the thematic roles (or θ-roles) of the arguments. We call the set
of thematic roles associated with an expression the thematic structure of
the expression. In any language there is at least one way to express the
thematic structure by linking the parts of the CS representation to parts
of the syntactic structure. We focus on what the thematic roles are in sec-
tion 5.4. Our knowledge of the particular relation WRITE encompasses
knowledge about how to express the writer and the thing written in a
5.4. THEMATIC ROLES AND LINKING 147
sentence; we discuss the representation of this aspect of lexical knowledge in

section 5.5.
Consider next (8b–d). Just as (8a) expresses a relation between two things,
example (8b) expresses a relation between three things.
(10) GIVE(MARY· , BOOK‚ , JOHN„ )
Example (8c) expresses a relation involving one thing.
(11) STINK(CHEESE‰ )
And example (8d) expresses a relation involving no arguments at all, at least

none that appear syntactically.
(12) RAIN()
(See Problem 4 for exploration of the possibility that rain may have a more
complex conceptual structure that expresses the fact that when it is raining,
something is coming down from the sky.)
With this background we are on the way to being able to represent
the meanings of linguistic expressions and to link the two. The critical
components that we have developed thus far are the concepts that express
reference (in the real or imaginary world), and the concepts that express
relations between referential concepts. Now we have to consider how the
lexical representation of a relational concept distinguishes the arguments
from one another.
5.4. Thematic roles and linking
5.4.1. Thematic structure
It is commonplace in contemporary syntactic theory to take the arguments

involved in a relation to be distinguished from one another in terms of
their syntactic representation. Consider, for example, (8a). As we noted
earlier, part of our knowledge about write is that it expresses a relation
between a writer and the thing written. The syntactic representation of this
knowledge is that the verb write assigns to its subject the meaning com-
ponent WRITER and to its object the meaning component WRITTEN. A
syntactic representation incorporating the thematic information might look
like (13).
(13) S
NP:WRITER VP
Mary V NP:WRITTEN
write a book
<WRITER,WRITTEN>
Here, the thematic information associated with the verb is represented in

the structure, and the corresponding roles are marked on the noun phrases.
While this approach may be roughly adequate for practical purposes, it
has the problem of introducing into the syntactic structure elements that
are strictly speaking components of the meaning. The thematic structure
associated with the verb is part of its lexical entry, and this information
determines the corresponding meaning when the verb is used in a sentence.
Given that we have a way to express meaning independently of syntactic
structure, we do not need to resort to this makeshift way of individuating the
arguments by marking the roles on the noun phrases. The first step in sepa-
rating out the two is to formulate a representation of the meaning of write in
the lexicon in which the properties of the arguments are explicitly expressed.
In the representation in (14), the variables X and Y are placeholders that, in
a complete representation of some relation of writing between a writer and
a thing written, are filled by CS concepts.
(14) write
SYNTAX [CATEGORY V]
CS WRITE(WRITER:X,WRITTEN:Y)
So the representation of Sandy writes a book will have the concepts for
Sandy and book in place of X and Y.
(15) CS WRITE(WRITER:SANDY,WRITTEN:BOOK)
The designation WRITER:X is a special case of the more general

AGENT:X, which is an entity capable of initiating action. The designation
WRITTEN:Y is a special case of the more general PATIENT:Y, which is
anything that is acted on in the course of an action. 4 Since the special
properties of AGENT:X and PATIENT:Y in the case of WRITE can be
determined from the special properties of WRITE, it is typically sufficient
to represent the CS of WRITE as follows.
4
PATIENT is sometimes called THEME, but we distinguish the two notions.

(16) SYNTAX [CATEGORY V]
CS WRITE(AGENT:X,PATIENT:Y)
Similarly, we represent the CS of GIVE, STINK, and RAIN as in (17)–

(19). In (17) we introduce the role Recipient, which is the individual that
comes into possession of a Theme when a transfer of possession takes place.
(17) give
SYNTAX [CATEGORY V]
CS GIVE(AGENT:X,THEME:Y,RECIPIENT:Z)
(18) stink
SYNTAX [CATEGORY V]
CS STINK(THEME:X)
(19) rain
SYNTAX [CATEGORY V]
CS RAIN
A Theme is something that is not physically affected by an action but whose

location or properties are specified by the sentence. The object that changes
Summary: the basic thematic roles

The thematic roles can be divided into two sets. One has to do with
actions and the other has to do with movement and location.
Action
Agent Initiator of an action. Capable of acting volitionally.
Patient Thing physically affected by an action.
Instrument Thing used in the carrying out of an action.
Movement/location/possession
Theme Thing that is in a location or state, or changes location
or state.
Source Initial location or state of a change.
Goal Final location or state of a change.
Location Location of a Theme.
Recipient Individual that comes into possession of something.
(A recipient is a special case of Goal when what
changes is possession.)
Other
Experiencer Individual in a perceptual or cognitive state, such as
seeing or knowing.
possession in an act of giving is a Theme, as is an object whose location

is specified (“The book is on the table.”) or an object whose property is
asserted (“That cheese stinks.”).
5.4.2. Linking to syntactic structure
Next, we have to specify how each argument gets linked to syntactic struc-
ture. While the representations that we have already given say that write is
a verb, that in itself is not sufficient to say what the syntactic correspon-
dences are. Without further analysis, we do not have an account of why
the meaning “John wrote the book” is not expressed as ∗ The book wrote
John. The situation is further complicated by the fact that RAIN has no
arguments.
In order to account for the correspondences, we assume the CS represen-
tations and the syntactic representations as described by independent rule-
governed structures (for the latter, see Chapter 4). We then state the linking
rules that connect them. These rules specify what part of a CS corresponds
to what part of a syntactic structure.
The linear ordering of these components of the structure in a sentence of
English is specified by the syntax of English. Hence the CS arguments of
a verb like write are going to be linked with the subject NP and an object
NP in VP. More specifically, the Agent role may be linked to Subject and
the Patient role is linked to Object. The type of correspondence that results
from these linking rules is given in (20).
(20) SYNTAX S
NP VP
V NP
write
CS WRITE(AGENT:X, THEME:Y)
Notice the important difference between the grammatical function and

the thematic role. The examples in (21) show that a subject may correspond
to a range of thematic roles, depending on the verb.
(21) a. Sandy is tall. [Sandy is Theme]

b. Sandy is writing the book. [Sandy is Agent]
c. Sandy has received a letter. [Sandy is Recipient]
d. Sandy has fallen. [Sandy is Theme]
e. It rained. [It has no role]
f. It is obvious that we should leave. [It has no role]
It is a matter of some controversy whether the grammatical functions are

syntactic primitives, or whether they are entirely reducible to independent
properties of syntactic structure. The view that we take here is that they are
primitives and that they correspond to different syntactic and morphologi-
cal forms in different ways across languages; see Chapter 3, section 3.4. But
it should be kept in mind that there is a considerable amount of research
that is devoted to explaining the differences between subjects and objects
purely in terms of syntactic configuration.
Certain redundancies can be eliminated in our description of the corre-
spondences between syntax and CS, so that the linking rules themselves
can be stated very simply. We do not have to state all of the information
about every verb individually, since many verbs follow general patterns.
Thus, there is considerable information that does not have to be specified
in the lexical entry for a verb like write, because it is “default” information.
Default information is what holds in the normal case. For one thing, we
do not have to say that the Agent of write corresponds to NP1 in the
configuration shown in (20). It is sufficient to say the following when there
is an Agent and a Patient in the CS:
r The Agent corresponds to Subject (in the default case).

r A general property of English is that the subject is the left sister of the VP.
r The Patient corresponds to Object.
r A general property of English is that the object is expressed as a right sister of V.
These statements are called default linking rules. However, when the syn-
tactic correspondence for a part of CS for a particular lexical item is not
predictable from general rules – and such cases do exist – then the lexical
entry must contain a specific linking rule.
We first illustrate the full correspondence for (20) in terms of the gram-
matical functions (GFs – see Chapter 3, section 3.4.1). The Subject GF
corresponds to the NP immediately dominated by S while the Object GF
corresponds to the NP immediately dominated by VP.
(22) SYNTAX S
NP VP
V NP
PHON write
GF Subject Object
CS WRITE(AGENT:X, THEME:Y)
Example (22) shows the correspondences between three distinct repre-

sentational levels. We say that a triple consisting of a CS, a set of GFs,
and a syntactic structure is licensed if each of the three representational
levels plays its proper role. That is, the correspondence is licensed if the
syntactic structure has the correct constituents and morphemes in the right
places, if there are an appropriate number of GFs, if the CS has the requisite
arguments and relations, and if everything is properly linked.
The lexical entry itself pairs the verb write with its CS representation.
This is the circled part of the correspondence in (22). The remainder of the
correspondence is taken care of by default linking rules that hold generally
across the language. The realization of Subject as the NP that corresponds
to AGENT:X is one such correspondence, the realization of Object as the
NP that corresponds to PATIENT:Y is another.
The linking of the relation WRITE with the simple verb write in a verb
phrase is also a default. As we will see, there are ways to map CS onto a
syntactic structure that do not link the CS relation directly to a single verb.
But, in any case, where the verb goes in the VP is a default property of the
syntax of the language, so we do not have to specify it in our correspondence
rules.
As a first approximation, we state the following correspondence rules that
yield (22) from (16).
(23) L INKING (D EFAULT )
Agent ↔ Subject
Theme/Patient ↔ Object
(24) GF CORRESPONDENCE (D EFAULT ) (E NGLISH )

a. Subject ↔ [S NP . . . ]
b. Object ↔ [VP . . . NP . . . ]
This default linking produces the following correspondence for Mary writes
a book. The concepts MARY and BOOK in CS determine the lexical form
of the Subject and Object NPs, respectively.
(25) SYNTAX S
NP VP
Mary V NP
write book
GF Subject Object
CS WRITE(AGENT:MARY, THEME:BOOK)
This representation is arrived at by taking the CS representation for write

and plugging the representations for Mary and book into the appropriate
argument slots in CS.
Other languages have different GF correspondence rules. In Russian, for
example, the GF of a phrase is marked not by its position in the linear order
but by its morphological form (its case). The examples in (26) are all Russian
sentences that mean “Mary writes a book”. The gloss NOM indicates that a
word has the nominative case form, and the gloss ACC that a word has the
accusative case form. The symbol š is a voiceless palatal. (For comparative
purposes, note that the nominative form of book is kniga and the accusative
form of Mary is Mariju.)
(26) a. Mari-a pišet knig-u

M.-NOM writes book-ACC
b. Mari-a knig-u pišet
M.-NOM book-ACC writes
c. pišet knig-u Mari-a
writes book-ACC M.-NOM
d. knig-u Mari-a pišet

book-ACC M.-NOM writes
e. knig-u pišet Mari-a
book-ACC writes M.-NOM
f. pišet Mari-a knig-u
writes M.-NOM book-ACC
Here are the Russian default GF correspondence rules.
(27) GF CORRESPONDENCE (D EFAULT ) (RUSSIAN )

a. Subject ↔ NP-NOM
b. Object ↔ NP-ACC
Notice that these rules do not specify configuration or linear order the way
that the rules for English do.
There are many alternatives to the default linking rules, in English and
in other languages. An important part of the description of the syntax of
a particular language is an account of how the GF correspondence rules
work in that language.
5.5. Linking hierarchies
5.5.1. Intransitives
An intransitive verb is one that has only one argument, e.g.
(28) a. Sandy called.

b. The treefell.
crying
c. Kim is flying .
dying
Whatever the thematic structure of the corresponding CS relation is, since

there is only one CS argument, it must correspond to Subject. But, as we
saw in our default thematic linking rule (23), when there are two arguments,
the Agent corresponds to the Subject and the Theme corresponds to the
Object.
This suggests that there are two hierarchies, one for the thematic roles
and one for the grammatical relations, that are linked to one another. The
following illustrates this linkage.
5.5. LINKING HIERARCHIES 155

Agent Subject
⇓ ↔ ⇓
Theme/Patient Object
In the θ-role hierarchy the Agent is higher than the Theme/Patient. In the
GF hierarchy, Subject is higher than Object. The highest available role is
mapped to the highest available GF. If there are two roles, the mapping is
as in (23). If there is only one role, it is mapped to Subject. (29) schematizes
the linking hierarchy.
5.5.2. Oblique arguments
As we have seen, the default marking of CS arguments in many languages is

by expressing them as subjects and direct objects. There are also non-default
ways for expressing arguments syntactically. We refer to these arguments as
oblique arguments. English uses prepositions to mark oblique arguments. 5
Here is a pair of sentences in English that illustrates the different grammat-
ical devices.
(30) a. Chris went into the plane.
b. Chris entered the plane.
Both of these sentences express the same semantic relation: enter

means “go into”, which we represent in CS as GO(AGENT:X,GOAL:
INTO(Y)), where GOAL is the role for the direction of the movement. In
the case of enter, the GOAL corresponds to the Object, but in the case of
go into, the GOAL is expressed by the PP.
The lexical entry for go corresponds to GO and that of into corresponds
to INTO(), as shown by the subscripts.
5
In languages that use morphological case to express the grammatical argu-
ments, oblique arguments appear with cases other than nominative/accusative or
ergative/absolutive. One of the most common such cases is DATIVE case. For
example, the subject of xolodno “(to be) cold” in Russian is in the DATIVE
case.
(i) Mne xolodno
1SG-DAT cold
‘I am cold.’
This type of non-standard case marking is sometimes called quirky case. See section
5.8.
(31) SYNTAX S
NP VP
V PP
go P NP
into
GF Subject
CS GO(AGENT:X, GOAL:INTO (Y))
Because go into and enter are synonymous, the lexical entry for enter must
also correspond to GO((. . . ),INTO()).
(32) SYNTAX S
NP VP
V NP
enter
GF Subject Object
CS GO(AGENT:X, GOAL:INTO (Y))

While enter takes a direct object, go into does not. Yet the two express
the same CS representation. The following lexical entries abbreviate the
correspondences shown in (31) and (32). We do not mention how the CS
arguments are linked to the syntax, because this is taken care of by the
default linking rules.
(33) go

SYNTAX [CATEGORY V]
CS GO(AGENT:X,GOAL:Y)
5.5. LINKING HIERARCHIES 157
(34) into
! "
SYNTAX [CATEGORY P]
CS INTO(Z)
(35) enter
! "
SYNTAX [CATEGORY V]
CS GO(AGENT:X,GOAL:INTO(Y))
The example of enter shows that a single verb can have a complex CS
representation. The inverse may also be true: a simple CS representation
may correspond to several words, including a verb. Consider look for, which
has the meaning SEEK, e.g.
(36) a. Sandy was looking for a solution to the problem.

b. Sandy was seeking a solution to the problem.
In the case of look for the second CS argument does not correspond
to a direct object but to an oblique argument. In this case there is no
correspondence between the preposition itself and any particular part of
the meaning.
(37) SYNTAX S
NP VP
V PP
look P NP
for
GF Subject
CS SEEK(AGENT:X, THEME:Y)
But when the verb is seek, the Theme role corresponds to the Object func-
tion.
(38) SYNTAX S
NP VP
V NP
seek
GF Subject Object
Thus, there is a sense in which there is a lexical item look . . . for that
has the meaning SEEK. This lexical item plays the semantic role of a
verb, but it is not a verb – it is a verb plus a preposition. It is like the
idiom kick the bucket that we discussed in section 5.2.3, in the sense that
it is syntactically and phonetically complex but corresponds to a simple
meaning.
The evidence in (39) shows that for NP can be separated from look. Hence
it is a PP constituent.
(39) a. Sandy looked very intently for the money.

b. For the money, Sandy was looking very intently (but not for the receipts).
c. Only for the money was Sandy looking very intently.
In the case of look for, the preposition for functions as a marker for the
second argument. That is, the preposition identifies the NP that corresponds
to the Theme role. Since this argument is inside of a PP, it does not bear the
Object GF, which is reserved for the sister of V.
Many verbs in English take oblique arguments. The prepositions that
mark these objects have varying degrees of semantic transparency, but
the oblique arguments are in PPs, as shown by the fact that they can be
separated from the V, like for the money.
(40) a. Sandy relies a lot on Leslie.
Sandy turned into a monster.
5.6. COMPUTING CORRESPONDENCES 159
Sandy believes only in the tooth fairy.

Sandy can’t possibly agree with an outcome like that.
b. On Leslie, Sandy seems to rely a lot.
Into what kind of a monster does Sandy seem to have turned.
It is only in the tooth fairy that Sandy believes.
With an outcome like that, Sandy can’t possibly agree.
5.6. Computing correspondences
We have seen how to represent the correspondences between syntactic

structure (SS) and conceptual structure (CS). Now we need to state the
correspondence rules explicitly.
The correspondence rules are statements about the conditions under
which a string of sounds, a syntactic structure, and a conceptual struc-
ture go together. To see how this works, consider the following three
examples. In these examples, the string of sounds and the syntactic struc-
ture constitute the representation of the sentence Sandy runs. The rep-
resentations in (41)–(43) match three possible conceptual structures with
this sentence, RUN(AGENT:SANDY), WALK(AGENT:SANDY), and
WALK(AGENT:LESLIE). We note the fact that a given correspondence
is not valid by putting an X on the link.
(41) SYNTAX S
NP VP
Sandy V
runs
GF Subject
CS RUN(AGENT:X)
(42) SYNTAX S
NP VP
Sandy V
PHON
runs
GF Subject
CS WALK(AGENT:SANDY)
(43) SYNTAX S
NP VP
Sandy V
PHON
runs
GF Subject
CS RUN(AGENT:LESLIE)
Only the first correspondence, (41), is correct. This correspondence

says that Sandy runs means “Sandy runs”. But the correspondence in (42)
says that Sandy runs means “Sandy walks”, and the correspondence in
(43) says that Sandy runs means “Leslie runs”, both of which are incorrect.
The function of the correspondence rules is to say what sentences mean.
So the correspondence rules must mark only the correspondence in (41) as
correct.
It is easy to see what the problem is in these examples. In the lex-

icon, the words run and walk are paired with the CS representations
RUN(AGENT:X) and WALK(AGENT:X), respectively. Moreover, Sandy
corresponds to the concept SANDY and Leslie to LESLIE, and not
the other way around. So a sentence in which the verb is run but
the CS relation is not RUN(AGENT:X) will fail to satisfy the lex-
ical correspondence. More generally, in order for the correspondence
for a sentence to be completely correct, each of the lexical correspon-
dences must be correct. When there is a mismatch, the correspondence
fails.
Examples such as these suggest how to verify that a given pairing of
a sentence and a meaning correspond to one another. For each word in
the sentence, we must find a corresponding element in the CS, and vice
versa. We do this on the basis of the lexical correspondences, at least for
these simple cases. When we find a match, we mark it by drawing an arrow
between the corresponding elements, or using some equivalent means of
representing the correspondence.
Let us do this for (41). We start with the pair in (41) and the lexical
correspondences summarized in (44).
(44) Lexicon:
Sandy run
CS SANDY RUN(AGENT:X)
If X is SANDY, then SANDY is the Agent of run. Recall that there is a

linking rule that links the Subject of a verb to the Agent argument in CS.
So, using the correspondence for Sandy, we draw a line from the NP to
Subject.
(45) GF Subject
CS RUN(AGENT:SANDY)
Since run is the verb, it is the head of a VP. The syntactic rules for English
tell us what the syntactic structure is.
(46) SYNTAX S
NP VP
PHON run
GF Subject
CS RUN(AGENT:SANDY)
Since Subject corresponds (in English) to the NP that is the sister of VP, we
insert Sandy, which corresponds to SANDY, into this position.
(47) SYNTAX S
NP VP
Sandy V
PHON
runs
GF Subject
CS RUN(AGENT:SANDY)
Since all of the parts are matched, the sentence corresponds completely to
the meaning. If we paired up the sentence with one of the other meanings
that we have considered, the correspondence will not be complete.
Now let us work out a somewhat more complicated example, Terry chases
Robin. We start with the lexical correspondences.
(48) Lexicon:
Terry Robin chase
TERRY ROBIN CHASE(AGENT:X, PATIENT:Y)

TERRY is X and ROBIN is Y, and each CS argument corresponds to a GF.

(49) GF Subject Object
CS CHASE(AGENT:TERRY, PATIENT:ROBIN)
Next, we determine the correspondence for CHASE and build the syntactic
structure.
(50) SYNTAX S
NP VP
V NP
PHON chase
GF Subject Object
Finally, we determine the correspondences between TERRY and the subject

NP and ROBIN and the object NP.
(51) SYNTAX S
NP VP
V NP
PHON Terry chase Robin
GF Subject Object
The arrow from Subject to the NP Terry integrates the correspondence of

Subject with this subject configuration and TERRY with the form Terry,
and similarly for the NP Robin.
This method of matching up the parts of the sentence and the parts of
the CS representation illustrates how to check whether a given sentence has
a given meaning, or how to figure out how to express a certain meaning.

But we can also check to see what a sentence means. To do this, we use the
correspondences, such as those in (44), and the linking rules in a slightly
different way.
Suppose that we have the sentence Sandy runs, with the syntactic struc-
ture in (52).
(52) SYNTAX S
NP VP
Sandy V
runs
This syntactic structure is determined by the phrase structure rules of the

language and by the lexical categories of the words Sandy and runs.
To construct the correspondence for the tree we apply the correspondence
rules for each of the parts. Using the correspondence for runs, we construct
the CS RUN(AGENT:X) and draw a line to show the correspondence
between run and RUN. We also draw a link between Sandy and the Subject
GF because, in the syntactic structure, the NP Sandy bears a relationship to
the sentence that marks it as Subject.
(53) SYNTAX S
NP VP
Sandy V
runs
GF Subject
CS RUN(AGENT:X)
Next, we draw a line between Subject and AGENT. We know that we can
draw this line because of the general linking rule that says that the Subject
corresponds to the AGENT.
(54) SYNTAX S
NP VP
Sandy V
runs
GF Subject
CS RUN(AGENT:X)
Finally, we replace X with SANDY, since we have this correspondence in

(44). This yields the correspondence in (55).
(55) SYNTAX S
NP VP
Sandy V
PHON
runs
GF Subject
CS RUN(AGENT:SANDY)
The operations that are used to construct the corresponding CS representa-

tion from the syntactic structure are summarized in the box.
In most cases, the steps of Build CS will have to be applied over and over
until everything is linked. For example, if the Subject is not Sandy but a
more complex NP, such as a friend of Sandy, then it would not be possible
to look up the CS of the NP in the lexicon. Rather, it would have to be
constructed, and the constructed CS would form a part of the larger CS.
Let us work out the correspondence for A friend of Sandy runs. We

begin with the assumption that friend (of) has the CS representation
FRIEND(X). In order to make the notation more readable, when we have
worked out all of the correspondences within a constituent, we will leave
the arrows out and just mark the whole constituent as corresponding
to a piece of the CS representation. This does not mean that the inter-
nal correspondences disappear – we just don’t show them. So we will
go from
(56) Full form:

SYNTAX NP
N PP
friend P NP
of Sandy
CS FRIEND(X) SANDY
to
(57) Simplified form:

SYNTAX NP
N PP
friend P NP
of Sandy
CS FRIEND(SANDY)
This simplification is done simply in order to make the diagrams more

readable.
The individual lexical correspondences for A friend of Sandy runs are
given in (58); we ignore the interpretation of a in order to keep the example
as simple as possible.
(58) Lexicon:
Sandy run friend
CS SANDY RUN(AGENT:X) FRIEND(X)
In (59) we show the step-by-step construction of the correspondence. The

syntactic structure is on the left, and the CS representation is built up on
the right as each link is constructed.
(59) a. Syntactic structure

S
NP VP
N PP V
friend P NP runs
of Sandy
b. Lexical correspondences of friend and Sandy
S FRIEND(X)
SANDY
NP VP
N PP V
friend P NP runs
of Sandy
c. Simplification
S FRIEND(SANDY)
NP VP
N PP V
friend P NP runs
of Sandy
d. Add lexical correspondence of run

S
FRIEND(SANDY)
RUN(AGENT:X)
NP VP
N PP V
friend P NP runs
of Sandy
e. Add links to GFs

Subject
S
FRIEND(SANDY)
RUN(AGENT:X)
NP VP
N PP V
friend P NP runs
of Sandy
f. Simplification
S
RUN(AGENT:FRIEND(SANDY))
NP VP
N PP V
friend P NP runs
of Sandy
As can be seen, a good strategy for actually constructing an SS–CS

correspondence for a sentence is to start at the bottom of the SS with the
individual words and work from left to right, building up the CS as you go.
There are a number of exercises at the end of this chapter that give you some
practice in doing this.
5.7. SELECTION 169
5.7. Selection
5.7.1. S-selection
Let us consider once again the CS representation for Mary writes a/the book.
We include the thematic information.
(60) WRITE(AGENT:MARY,PATIENT:BOOK)
The representation in (60) reflects the knowledge that the relation WRITE
holds between two entities, MARY and BOOK. The preliminary CS rep-
resentation for write captures the fact that the two arguments are Agent
and Patient, but there is more that we know. We know, for example, that
the Patient is something that has writing on it. Similarly, the relation EAT
holds between an animate Agent and food, the relation SAY holds between
a human Agent and something that has linguistic content, the relation
THINK holds between an animate entity and an idea, and so on.
The specification of the features of CS arguments is sometimes called
selection or s(emantic)-selection. The specification of the categories of the
syntactic arguments is called subcategorization or c(ategorial)-selection. In
many cases these are equivalent, in the sense that for every CS argument
there is usually a corresponding syntactic argument, and vice versa.
In order to capture this knowledge we add to the CS representation those
selectional properties that can be reasonably assumed to be part of the
lexical entry of a word. For example, for EAT, a property of the AGENT
argument is that it is [ANIMATE] and a property of the PATIENT argu-
ment is that it is [FOOD]. But we want to say that the PATIENT does not
have to be expressed, which we notate by putting angled brackets around
the argument.
(61) eat
[CS EAT(AGENT:X[ANIMATE],PATIENT:<Y> [FOOD])]
Similarly, the CS representations of SAY and THINK contain selectional
information about the arguments.
(62) say
[CS SAY(AGENT:X[HUMAN],THEME:Y[LINGUISTIC])]
(63) think
[CS THINK(AGENT:X[ANIMATE],THEME:Y[PROPOSITION])]
A sentence can be grammatical but its meaning may be such that certain
selectional requirements are not met. Consider the following sentences.
(64) a. Two days elapsed.
b. ∗ George elapsed.
(64b) is strange because elapse s-selects a period of time as its subject, while
George refers to a human being, not a period of time. Thus, there is a
clash in CS between the requirement of ELAPSE given in (65a) and the
CS representation of (64b) in (65b).
(65) a. elapse
! "
SYNTAX
[CAT V]
CS ELAPSE(THEME:[TIME-PERIOD])

[TIME-PERIOD]
b. ELAPSE(THEME: )
GEORGE[HUMAN]
The features [TIME-PERIOD] and [HUMAN] are incompatible. This clash
is called a selection restriction violation.
S-selection does not mean that an argument has to have the specified
properties. We can say things like The vending machine ate my dollar or The
computer says that it can’t find the file, which appear to violate the selectional
requirements of eat and say. What does s-selection mean for sentences in
which an argument does not have the properties specified in the CS? There
are three possibilities.
i. First, the argument in question can be understood as having the specified

properties in some extended sense. Thus, we may think of a computer as
being human in a particular respect, which is that it can say things and
perhaps even think.
ii. Second, we may understand the sentence as meaning something possible
but strange. This would be the case, for example, with sentences like those
in (66).
(66) a. Mary is reading her shoe.
b. Mary is eating her shoe.
c. Mary is drinking her shoe.
In each case, we must imagine the object in question as having the relevant
property; for example, if Mary is reading her shoe, then the shoe must
have writing or at least meaningful marks on it. If Mary is eating her
shoe, perhaps it is made of chocolate, or perhaps she cooked it until it
was soft enough to eat. If Mary is drinking her shoe, perhaps it was made
5.7. SELECTION 171
of something organic and she put it in a blender and liquefied it. This
imagining is a variety of what is called coercion. 6
Coercion is a powerful tool, because it allows us to represent the typical
interpretation of a lexical item without ruling out the possibility that it can
combine lawfully in atypical ways.
iii. The third possibility is that the argument cannot be understood as
having the specified properties. In such cases, coercion is very difficult, if
not impossible, and then we have the intuition of a strong violation of
a s-selection requirement. In normal discourse, such violations may be
perceived as nonsense, but in the proper context they may be perceived
as having poetic effect, if the coercion is extended sufficiently far. A few
examples are given in (67).
(67) a. Mary is writing the coleslaw.
b. Mary is eating the s-selection.
c. Mary is drinking her sincerity.
We would not want to say that sentences such as these are necessarily impos-
sible in English (that is, ungrammatical), but we would want to characterize
the circumstances under which they can be understood as referring to a
situation in the world as requiring special coercion at best. 7
The examples in (68) shows that s-selection and c-selection are not iden-
tical, in that there can be CS arguments that do not correspond to syntactic
arguments. 8 These are called implicit arguments.
 

 writing 
 reading 
 
(68) Sandy is .

 eating  

 

drinking
The verb write can be used intransitively, as can verbs such as read, eat,
and drink. But these intransitive uses have the meaning that there is some-
thing with the appropriate properties that is written, read, eaten, and
drunk. These examples mean Sandy is writing something, Sandy is reading
6
Pustejovsky 1995.
7
So, for example, we might understand (67a) as referring to the unlikely but not
impossible situation in which the coleslaw is arranged in such a way that it spells out
words, similar to what is sometimes claimed about tea leaves and coffee grounds.
And (67c) is parallel to the familiar metaphorical expression swallow one’s pride.
8
There are also syntactic arguments that do not correspondence to CS argu-
ments. See Problem 1 and Research question 1.
something, and so on. Thus, the CS representation has two arguments, but
there is only one syntactic argument, the Subject.
5.7.2. C-selection and the theta criterion
Let us turn now to c-selection, which is the selection of an argument

on the basis of its syntactic properties. C-selection has to occur just in
case the category of the complement cannot be entirely predicted from
its meaning. For example, if the lexical entry of a verb corresponds to a
CS-representation with two arguments, but the verb can be intransitive,
like write, read, eat, and drink, the lexical entry must mention this fact
in some way. We have represented this by stating in the lexical entry of
the verb that the argument that corresponds to the Object GF need not
correspond to anything in the syntactic structure – see (61). There are other
and more complex departures from the default mapping between CS, GF,
and syntax that we have to be able to specify, for example, oblique argu-
ments (section 5.5); all of these contribute to the syntactic description of a
language.
While some CS arguments do not have to correspond to syntactic argu-
ments, it appears that quite often in English there is a one-to-one match.
Consider the following.
(69) a. Mary gave John a magazine.
∗
b. Mary gave John.
∗
c. Mary gave a magazine. 9
(70) a. The dog is chasing the cat.
∗
b. The dog is chasing.
(71) a. I put the groceries on the table.
∗
b. I put the groceries.
∗
c. I put on the table.
The intuition that ideally there is a one-to-one match between the number
of syntactic arguments and the number of semantic arguments that of a
verb has been captured in various ways by syntactic theories. The most
well-known formulation is Government Binding theory’s theta criterion
9
In the context of a collection it is possible to use give with no arguments, as in I
gave at the office, and with one argument, as in I gave to the United Way and I gave
$10, how much did you give? But even in this sense it is impossible to have just the
indirect object: ∗ I gave the United Way.
5.8. CASE 173
(Chomsky 1981). If a verb has two CS arguments, it has two θ-roles and
therefore two syntactic arguments, according to the theta criterion. The
theta criterion in its strongest form says that every referring grammatical
argument must correspond to a θ-role, and every θ-role must correspond to
a referring grammatical argument. 10
Since, as we have seen, there are cases in which a θ-role does not corre-
spond to a grammatical argument, the second part of the theta criterion
appears to be too strong. But the first part appears to be correct – there
cannot be a referring syntactic argument in a sentence that does not corre-
spond to a θ-role. 11 But two arguments can correspond to the same θ-role,
as we see in examples such as the following. In (72a) the NPs my car and
it both denote the location of the scratch, while in (72b) the NPs Terry and
herself both denote the Agent of behave.
(72) a. My car has a scratch on it.

b. Terry was behaving herself.
We do not assume the theta criterion in this book. It has been very
influential in the development of syntactic theory, however. We examine
some of the consequences of assuming it in our discussion of so-called “pro-
drop” in Chapter 6, and in our discussion of “control” in Chapter 7.
5.8. ∗ Case
In this section we consider the role of case marking in the linking between
CS and syntactic arguments. In a language with nominative and accusative
case, the default linking is the following.
(73) D EFAULT GF CORRESPONDENCE (N OMINATIVE / ACCUSATIVE LAN -

GUAGE ):
a. Subject ↔ NP-NOM
b. Object ↔ NP-ACC
10
Certain arguments, such as the subject of It’s raining, do not have a meaning –
they are “dummy arguments”. These arguments do not have θ-roles. We discuss
dummy arguments further in Chapter 6.
11
An analogy to the theta criterion is the game of musical chairs, where the
chairs are the roles and the players are the syntactic arguments. If there are more
syntactic arguments than there are roles, at least one of the arguments is without a
role, which makes the correspondence ill-formed.
Compare this to the default GF correspondence for a language like Eng-

lish in which the grammatical functions are marked in terms of syntactic
configuration.
(24) GF CORRESPONDENCE (D EFAULT ) (E NGLISH )
a. Subject ↔ [S NP . . . ]
b. Object ↔ [VP . . . NP . . . ]
We assume that in languages that mark case, the general hierarchy for
mapping between GFs and θ-roles holds. This hierarchy has the conse-
quence that if there is only one CS argument, it will correspond to Subject
and therefore be expressed as NOMINATIVE in the default case. If there are
two CS arguments, the first will be NOMINATIVE and the second will be
ACCUSATIVE . 12

Agent Subject
(29) ⇓ ↔ ⇓
In section 5.5.2 we discussed the fact that in English certain CS arguments
are expressed not as Object but as the complement of a preposition. These
are the oblique arguments. In some languages, what we have called oblique
arguments would be expressed not by ACCUSATIVE case but a different
case, say, DATIVE or INSTRUMENTAL , or by PPs with particular case-
marked NPs. In Russian, for example, NOMINATIVE and ACCUSATIVE
are the canonical cases corresponding to Subject and Object, but other
cases may be assigned depending on the idiosyncratic lexical properties
of the verb. Predicates expressing obligation and necessity, permission and
possibility, and mental and emotional states assign DATIVE case to their
subjects. Some examples are given in (74).
(74) Russian
a. Mne nado by čitat’ ešče mnogo knig o
1SG-DAT necessary to- read still more book-GEN. PL about
Tolstom.
Tolstoy-INSTR
‘I really ought to read more books about Tolstoy.’
b. Vam ne sleduet tak govorit’.
2SG-DAT not should thus to-speak
‘You should not say such things.’
c. Mne veselo bylo sredi vas.
1sG - DAT cheerful be-PAST. SG. NEUT among 2PL - GEN
‘I enjoyed myself among you.’
12
We leave open the question of how ergative languages fit into this picture.
5.8. CASE 175
d. Mne žal’ vaš-u sestr-u.

1 SG-DAT sorry your-ACC sister-ACC
‘I am sorry for your sister.’
This type of case marking falls under the general term quirky case because
it does not fit the canonical pattern. Quirky case is also found in Icelandic,
where a case other than nominative can be assigned to the subject. Even
though a subject has quirky case it still can be the antecedent of a reflexive
pronoun, which is one of the hallmarks of subjecthood (see Chapter 10).
An example of a reflexive pronoun in English is himself ; the corresponding
Icelandic form is sina.
(75) Hani elskar bókina sínai
he-NOM love.3SG book self-GEN
‘He loves his book.’
(76) Honumi finnst bókin sini skemmtileg13
him(DAT ) find.PAST book self-GEN amusing
‘He is amused by his book.’
[BarDhal 1997]
Moreover, a case other than accusative can be assigned to the object, in

Russian and in Icelandic.
(77) Russian
a. on ždet otvet-a.
3SG.NOM wait.for-3SG. PRES the.answer-GEN
‘He is waiting for the answer.’
b. On ne zanimalsja russk-im
3SG. MASC. NOM NEG study-PAST-3 SG. MASC Russian-INST
jazyk-om.
language-INST
‘He didn’t study Russian.’
(78) Icelandic 14
a. Eg hjálpaDi honum.
1SG.NOM help.PAST 3SG. DAT
‘I helped him.’
b. GuDrún saknar Haraldar.
Gudrun.NOM miss.PRES Harold-GEN
‘Gudrun misses Harold.’
Finally, here are some examples from Russian that show that prepo-
sitional phrases take NPs with particular case marking. S “with” takes
13
The gloss of this sentence is a slightly simplified version of the original.
14
Zaenen et al. 1990.
instrumental case, v “into” takes accusative case, and iz “out of” takes
genitive case.
(79) Russian
a. Ja govorila s nim.
1SG - NOM speak-PAST-3 SG. FEM with 3SG. MASC-INST
‘I spoke with him.’
b. Ona vošla v komnatu.
3SG. FEM-NOM go.into-PAST-3 SG. FEM into room-ACC
‘She went into the room.’
c. Ona vyšla iz komnaty.
3SG. FEM -NOM go.out.PAST-3 SG. FEM out.of room-GEN
‘She went out of the room.’
To summarize, there are essentially two ways to map CS arguments

into syntactic structure. One is to assign the arguments to grammati-
cal functions, Subject and Object, which then correspond to structure
or case (or perhaps both) according to default linking rules. The other
is to associate the arguments with particular prepositions or case, or
both, either by general rule or as specified by the lexical entry of the
verb.
5.9. ∗ Modification
Thus far we have left out all aspects of meaning besides arguments
and relations. A more complete account of CS must also mention mod-
ifiers of a constituent such as those expressed by prenominal adjec-
tives and by place, time, and manner adverbials. In the case of noun
phrases, we have to distinguish definiteness and indefiniteness. For
verb phrases we have to be able to link tense and aspect to the relevant
parts of CS.
At this point we must introduce more complexity into our account of CS
in order to be able to accommodate these additional aspects of meaning.
Already implicit in our discussion is the idea that a CS representation, like
a sentence, is a structured representation in which the pieces fit together
in certain regular ways. In the simpler cases we have seen that a CS may
consist of a relation with zero or more arguments. Each argument has a
certain type. In the simple examples that we have been looking at all of the
arguments are entities in the sense that they are (real or imaginary) objects
5.9. MODIFICATION 177
that we can refer to. Such entities may be concrete, e.g. persons or pieces of
furniture, or they may be abstract, e.g. numbers, places, times, and so on. 15
There are, however, other components of meaning. For example, in a red
lamp on a table, the word red denotes a property of the object, while on
a table denotes a place. Both red and on a table modify the head lamp.
The determiner a expresses indefiniteness. We represent the modification
by separating it from the arguments, as in (80).
(80) LAMP;INDEF, PROPERTY:RED, PLACE:ON(LOC:TABLE;INDEF)
For modifiers of a verb or sentence, we use a similar structure. Consider

the following two sentences.
(81) a. John was sitting in the kitchen.

b. John put the tarantula next to Mary.
In (81a) in the kitchen refers to the place where John was sitting.
SIT has only one argument, the sitter, and the state of affairs that
SIT(AGENT:JOHN) represents can be said to hold in the place referred
to by in the kitchen. We represent this information in (82).
(82) SIT(AGENT:JOHN); PLACE:IN(LOC:KITCHEN;DEF)
In (81b) next to Mary refers to the place where the tarantula ends up. On
the other hand, PUT takes three arguments, the putter, the thing put, and
the place where it was put. We represent this information in (83).
(83) PUT(AGENT:JOHN,PATIENT:(TARANTULA;DEF),PLACE:NEXT-
TO(LOC:MARY))
In (82) PLACE goes after “;”, since it is not an argument, while in (83) it
is one of the arguments and appears within the parentheses associated with
PUT(. . . ).
Much of this detail will not be directly relevant to our syntactic analyses
and we will omit it elsewhere in this book. What is crucial is that we are
able to show how the syntactic structure corresponds to the CS represen-
tation and, where appropriate, how two different syntactic structures may
correspond to the same CS representation.
15
See Jackendoff 2002.
Exercises
1. Specify in as much detail as you can the facts that should be captured
in a statement about the form/meaning correspondence for the following
sentences. In particular, what aspects of the form need to be accounted for
in each sentence and what aspects of the meaning need to be accounted for?
We’ve provided an example to give you some idea of what is involved.
(0) I fell.
The meaning is FALL(THEME:ME). The Theme corresponds to the subject
NP I, which precedes the verb. The subject is marked for case – the form is I
and not me. The verb is an irregular past tense – it is fell and not ∗ falled.
(1) a. She saw him.

b. They are visiting relatives.
c. Someone’s there.
d. We sold her beer.
[§5.1.]
2. Give approximate CS representations for the following sentences,
focusing on the arguments and the relations between them. The assignment
of the same index in the string to two phrases indicates that they are
intended to refer to the same thing.
(1) a. Maryi is proud of herselfi.
b. Johni thinks hei won.
c. Maryi saw a frog under heri chair.
Assume that CS of proud is PROUD(EXP:X[ANIMATE],THEME:(Y)),

and that the CS for think and see are as in the text.
[§5.2.2.]
3. Using (17)–(19) in the text as models, state the lexical entries for the
following verbs. Use the role Agent for actors, Experiencer for perceivers,
Patient for things that are acted on, and Theme for things that are otherwise
involved in a relation. If a verb allows for more than one possibility, state
each one separately.
(2) a. walk e. observe
b. throw f. discuss
c. eat g. marry
d. destroy h. collapse
[§5.3.]
EXERCISES 179
4. Using (22) in the text as a model, draw the Syntax/GF/CS diagrams for
the following verbs. Indicate the thematic roles in the CS representation, and
the links to GFs. Use the role Agent for actors, Experiencer for perceivers,
Patient for things that are acted on, and Theme for things that are otherwise
involved in a relation.
(1) a. kiss f. chew
b. fall g. drink
c. run h. snow
d. believe i. hear
e. show j. giggle
[§5.4.]
5. Using (22) in the text as a model, draw the Syntax/GF/CS diagrams
for the following verbs. Indicate the thematic roles in the CS representation,
and the links to GFs. The preposition in parentheses marks an oblique
argument.
(1) a. think (about) ‘contemplate’ [CONTEMPLATE(EXP:X,THEME:Y)]
b. bring (about) ‘cause’ [CAUSE(AGENT:X,THEME:Y)]
c. depend (on) ‘trust’ [TRUST(AGENT:X,THEME:Y)]
d. live (in) ‘inhabit’ [INHABIT(AGENT:X,LOC:Y)]
[§5.5.]
6. For each of the following sentences, say whether the unacceptability
is due to s-selection or c-selection (without s-selection). (We use # here
to indicate semantic strangeness.) If there is coercion in a particular case,
explain what is going on.
(2) a. #The book ate a pizza.
b. #I mailed to the office.
c. #Sandy thinks the rock.
d. #Leslie squeezed that it was raining.
e. #Lee repaired.
[§5.5.]
7. Using the description of the correspondences for go into and enter as a
model, draw the correspondences for go from and exit.
[§5.5.2.]
8. Using (59) in the text as a model and Summary: Constructing CS rep-

resentations as a guide, show how the CS representations for the following
sentences are derived.
(1) a. Sentence: Sandy dislikes baseball.

CS: DISLIKE(EXP:SANDY,THEME:BASEBALL)
b. Sentence: Sandy loaned Leslie a bicycle.
CS: LOAN(AGENT:SANDY,THEME:BIKE,RECIP:LESLIE)
[Hint: You will have to list the individual correspondences first, and then
show how to put them together to arrive at the complete CS representation.]
[§5.6.]
Problems
1. Here are some cases in which the number of syntactic arguments is

greater than the number of CS arguments.
(1) a. Mary behaved herself very well.
b. John availed himself of the free coffee.
Evidence that shows this is that it is impossible to replace the reflexives in

these examples with referring NPs.
(2) a. ∗ Mary behaved John very well.
b. ∗ John availed Mary of the free coffee.
Work out the lexical entries for behave (oneself) and avail (oneself), show-
ing the correspondence between the syntactic structure and the CS repre-
sentation.
[§5.7.]
2. The following examples suggest that instructions may violate the theta
criterion, in that it is possible to leave out syntactic arguments that corre-
spond to CS arguments of the verb.
(1) a. Insert into slot A.
[Cf. ∗ I inserted into slot A.]
b. Ignite while lifting.
[Cf. ∗ I ignited while lifting.]
c. (Note on a package.) Mary: show to John when you see him.
[Cf. ∗ Mary showed to John when she saw him.]
d. Roll into balls 3 inches in diameter and flatten with a knife.
[Cf. ∗ I rolled into balls 3 inches in diameter and flattened with a knife.]
What is going on here? Is the theta criterion inapplicable, or is there some

special way that it is being satisfied? Explain.
[§5.7.]
PROBLEMS 181
3. In section 5.7.1 we discussed the fact that under some circumstances a

concept may temporarily acquire a property from the verb of which it is an
argument. One such example is Mary ate her shoe, where shoe is understood
as edible. Using the lexical entry for eat in (61), we can see how this works.
First, we start with eat and shoe separately.
(1) eat ↔ EAT(AGENT:X,PATIENT:<Y>[FOOD])
shoe ↔ SHOE[CLOTHING]
Then we put the CS representation for shoe in for the Y argument. We merge
the [FOOD] property of Y and the [CLOTHING] property of SHOE.
(2) eat shoe ↔ EAT(AGENT:X,PATIENT:SHOE[FOOD, CLOTHING])
[FOOD] and [CLOTHING] are incompatible, so one has to go. If we

get rid of [FOOD], then Mary ate her shoe means that she literally
ate her shoe. On the other hand, if we get rid of [CLOTHING], then
shoe is coerced into being food, perhaps in virtue of being made out of
chocolate, etc.
Analyze the remaining sentences in (66) and those in (67) in a similar way.
[§5.7.]
4. We have been speaking of rain as a 0-argument verb because it appears
with a dummy subject.
(1) It is raining.
But the following examples show that rain can appear with a richer set of
arguments.
(2) a. It is (literally) raining walnuts.
b. The tree is raining walnuts (on my head).
c. Walnuts are raining from the tree (on my head).
∗
d. Walnuts are raining.
A. Formulate a lexical representation for rain that allows for these exten-
sions.
B. Given your formulation in A, how would you account for the cases in
which rain appears with a dummy subject?
[§5.7.]
5. Take a walk and go for a walk both mean walk.
(1) a. Robin and Terry walked.
b. Robin and Terry took a walk.
c. Robin and Terry went for a walk.
While there are subtle meaning differences, all three have the same basic
CS structure. Formulate the lexical entries for the idioms in b and c so
that correspondences between the syntactic structures and the conceptual
structures are clearly represented.
[§5.7.]
6. State as concisely as you can the rule or rules for assigning case to
English pronouns. Assume that the case on she, he, we, they is necessarily
NOM . My, your, etc. have GEN case. The pronouns him, her, us, them are
ACC case. You can be NOM , but may also be ACC . (Hint: Is it necessary to
state syntactic conditions for the assignment of all cases, or can one or more
be assigned by default?)
[§5.8.]
Research questions
1. The examples in (1) show syntactic arguments that do not correspond

to CS arguments.
(1) a. Leslie slept the sleep of the innocent.
b. Sandy died a violent death.
c. Robin smiled a sly smile.
These sentences paraphrase sentences with adverbials.

(2) a. Leslie slept innocently.
b. Sandy died violently.
c. Robin smiled slyly.
A. Are these verbs optionally transitive, or are they a class of idioms?

In either case, explain the correspondence between the syntactic struc-
ture in which they appear and their CS representations. Assume that
the adverbs like innocently are represented as MANNER (e.g. MAN-
NER:INNOCENT).
B. Is there some common property or set of properties that determines
which verbs can be used in this way and which cannot? Here are some
additional examples to get you started, but you should find more of your
own in order to develop and test a hypothesis.
(3) a. laugh (a happy laugh)
b. cough (a warning cough)
c. dream (a frightening dream)

d. wink (a sly wink)
e. frown (an unhappy frown)
C. Can you see a way to relate your analysis to examples such as the
following?
(4) a. I drank a quick cup of coffee,
b. I ate a fast hamburger for lunch.
c. Terry is a beautiful dancer. [=Terry dances beautifully]
d. Robin is a light sleeper.
[§5.7.]
5.1. 1
5.2. 2
5.3. 3
5.4. 4
5.5. 5, 6, 7
5.6. 8
5.7. 1, 2, 3, 4, 5 1
5.8. 6
5.9.
6
Argument Correspondences
6.1. Canonical argument correspondences
In Chapter 5 we saw a number of examples showing how CS arguments

are linked to grammatical functions and syntactic structures. We repeat the
default linking here.

Agent Subject
⇓ ↔ ⇓
On the default linking, a single CS argument corresponds to the Subject
grammatical function. When there are two arguments, the canonical cor-
respondence is one in which the Agent corresponds to Subject and the
Theme/Patient corresponds to Object. Other correspondences are possible,
as well, such as Experiencer and Theme, in the case of verbs like know, and
Goal and Theme, in the case of verbs like receive.
In this chapter we look at a range of other ways in which CS arguments
can be expressed syntactically. All but one involve arguments of a single CS
relation; one, the causative, involves arguments of two CS relations.
6.2. Passive
6.2.1. Passive relations
Here is a typical active/passive pair in English.

(2) a. Kim kissed Sandy.
b. Sandy was kissed (by Kim).
The passive construction is characterized by the fact that the NP that

would normally correspond to the Object of the verb in the canonical
186 6. ARGUMENT CORRESPONDENCES
correspondence becomes the Subject, while the NP that would become the
Subject of the verb in the canonical correspondence becomes an oblique
argument, if it is expressed. The NPs Kim and Sandy have the same thematic
roles in the two sentences; only the syntax is different. For this reason, Kim
in (2) is sometimes called the logical subject and Sandy the logical object.
We will use this terminology here.
Passive
r Logical object corresponds to Subject.
r Logical subject corresponds to oblique argument, or is not expressed.
In his seminal analysis, Chomsky 1957 pointed out the following proper-
ties of the passive construction.
(3) a. The passive participle following a form of to be occurs only with a transitive
verb.
b. V in the passive cannot be followed by a direct object. 1
c. An agentive by-phrase can occur only if the sentence is passive.
d. The selectional restrictions on subject and object of the active are mirrored
by the selectional restrictions on the by-phrase and subject of the passive,
respectively.
These properties are illustrated by the following examples. (“#” indicates

that the sentence is grammatical but is semantically anomalous.)
(4) a. <It/∗ Kim> was slept. [∗ was slept is an impossible passive]
∗
b. Kim was seen Sandy. [was seen cannot take a direct object]
∗
c. Kim saw Sandy by Chris. [by Chris cannot appear in the active2 ]
d. #Kim ate the sincerity. [“eating sincerity” is equally strange in
#Sincerity was eaten by Kim. the active and passive.]
Chomsky concludes (p. 43), “This inelegant duplication, as well as the

special restrictions involving the element be+en, can be avoided only if we
deliberately exclude passives from the grammar of phrase structure, and
reintroduce them by a rule . . . .”
We discuss the kind of rule that Chomsky had in mind in section 6.8.
Putting aside technical matters, we can see that the passive is an alternative
1
With caveats for examples like Sheila was sent flowers. In this case, it is the
indirect object that does not follow the verb.
2
This sentence is grammatical if by Chris means near Chris.
6.2. PASSIVE 187
way in which the CS arguments correspond to the syntactic structure.

We provide a description of this correspondence in terms of (1). The key
property of the passive is that the CS argument that canonically corresponds
to the highest GF (the logical subject) is instead suppressed or shunted off to
an oblique argument. Thus, the next CS argument on the hierarchy becomes
the one that corresponds to the highest GF argument, which appears syn-
tactically as subject. The Subject GF in English corresponds to the NP that
is the left branch of S. (5) illustrates.
(5)
SYNTAX S
NP VP
V PP
[PASSIVE]
P NP
seek
(=‘sought’) by
GF Subject Object
The Object GF is simply not associated with any argument and not realized
syntactically in the passive.
6.2.2. Passive constructions
An additional feature of the passive construction is that the verb has the
past participle form, and in a full clause it follows a form of the verb to be.
It resembles an adjective phrase in its distribution, as shown in (6). Example
(6a) shows that both types of phrase are complements of be. Example (6b)
shows that they can both be heads of topicalized predicates. Example (6c)
shows that they can both be used as heads of post-nominal modifiers. And
example (6d) shows that they can both be prefixed with un-.
(6) a. Sandy was seen by Kim.
Sandy was happy with Kim.
b. . . . and seen by Kim Sandy was.

. . . and happy with Kim Sandy was.
c. Anyone seen by Kim should report to the official’s tent.
Anyone happy with Kim should report to the official’s tent.
d. seen/unseen by Kim
happy/unhappy with Kim
So a more accurate realization is the following, where the notation

[PASSIVE ] indicates that the verb has passive participle morphology.
(7)
SYNTAX S
NP VP
be VP
[PASSIVE]
V PP
[PASSIVE]
P NP
seek
(=‘sought’) by
GF Subject Object
Notice that the VP complement of be is marked [PASSIVE] in virtue of the

fact that its head is [PASSIVE]. As we discussed in Chapter 3, section 3.5.1,
heads share not only their category but their grammatical properties with
their phrases. 3
The analysis of the passive in (7) assumes a phrase structure rule that has
to be added to what we arrived at in Chapter 3.
(8) VP → V VP
Whether a verb may appear in this configuration is a lexical property of the

verb. In English, the verbs be and get appear with passive VP complements.
3
This characteristic of phrases has been treated more or less informally in main-
stream generative grammar. It has been formalized in Head-driven Phrase Structure
Grammar (HPSG).
6.2. PASSIVE 189
! "
VP
We represent the verbs in the following AVMs. The value for
PASSIVE
COMPS indicates that the verb selects a complement VP whose head has
the property [PASSIVE]
(9) be (passive)
  
CATEGORY V
 SYNTAX  ! "
VP
COMPS
PASSIVE
(10) get (passive)
  
CATEGORY V
 SYNTAX  ! "
VP
COMPS
PASSIVE
The related construction exemplified by have someone arrested also makes

use of passive VPs. This use of have has a lexical entry in which an NP and
passive VP are complements.
(11) have (passive)

  
CATEGORY V
 SYNTAX  ! "
VP
COMPS NP
PASSIVE
This approach to the passive construction extends naturally to other lan-

guages. In languages where GFs are linked to case marking and not syntac-
tic configuration, the same relation holds. The following shows it schemat-
ically for a nominative/accusative language. OBL indicates an oblique case,
i.e. neither NOM nor ACC. Notice that the object GF is inert – nothing is
linked to it either from CS or syntax, just as in the English passive. F is a
CS relation that corresponds to V, which we show by coindexing them.
(12) a. SYNTAX NP−NOM NP−OBL V [PASSIVE]
GF Subject Object
CS F(AGENT:X, THEME:Y)
b. SYNTAX NP−NOM V [PASSIVE]
GF Subject Object
Here are some examples from Russian that illustrate this type of passive
construction. INSTR is the instrumental case, translated here as “by” but
also used to mark an NP as an instrument, translated as “with”.
(13) a. Amerik-u otkryl Kolumb
America-ACC discovered Columbus-NOM
‘Columbus discovered America.’
b. Amerik-a byla otkryta Kolumb-om
America-NOM was discovered Columbus-INSTR
‘America was discovered by Columbus.’
Notice that the word order in the two sentences is identical; the grammatical
functions are indicated by the case marking. 4
A similar picture is presented by Japanese –
(14) Japanese
a. Sensei-ga Taroo-o sikat-ta
teacher-NOM Taroo-ACC scold-PAST
‘The teacher scolded Taroo.’
b. Taroo-ga sensei-ni sikar-are-ta
Taroo-NOM teacher-OBL scold-PASSIVE - PAST
[Tsujimura 1996]
– and Latin.
(15) Latin
a. Mı̄litēs hanc provinciam dēfendērunt
troops-NOM this- ACC province-ACC defended-3 PL
‘The troops defended this province.’
b. Haec provincia ā mı̄litibus dēfensa est
this-NOM province-NOM by troops-ABL defended-NOM is
‘This province has been defended by troops.’
[Blake 1994:73]
4
All other orderings of the phrases in both sentences are also possible
under appropriate discourse conditions, because Russian is a “scrambling”
language.
6.3. APPLICATIVES AND THE DATIVE ALTERNATION 191
It is possible to have passives in ergative languages as well as in nomina-

tive/accusative languages. (See Chapter 5, section 5.8, for the distinction.)
In both cases, the logical object is realized as the syntactic subject, while the
logical subject is realized in an oblique case. Hence the passive is intransi-
tive.
Recall that in an ergative language, the same case is used for the subject
of the intransitive and the object of the transitive. This means that, if the
logical subject is suppressed, the logical object will be the syntactic subject
of an intransitive. Hence the logical object will have the same case in the
active and the passive. But the logical subject will have a different case.
Schematically, the active and the passive in the ergative display the fol-
lowing patterns.
(16) Active
logical object ↔ syntactic object ↔ -ABS
logical subject ↔ syntactic subject ↔ -ERG
Passive
logical object ↔ syntactic subject ↔ -ABS
logical subject ↔ suppressed or oblique
Here is an illustrative example from Nepali; the case ABS is the one used for
direct objects and intransitive subjects (adapted from Givón 2001:II:146f.).
In (17a) we see the ergative-absolutive case marking, where absolutive
appears on the direct object. In (17b), which is passive, the same NP is
now the subject of the passive, and therefore gets absolutive case because
the passive is intransitive. But now the logical subject gets the oblique case.
(17) Nepali
a. Raj-le Ram-lay mar-yo
Raj-ERG Ram-ABS kill-PAST /3 SG. MASC
‘Raj killed Ram.’
b. Raj-dwara Ram-lay mar-i-yo
Raj-OBL Ram- ABS kill-PASS - PAST /3 SG. MASC
‘Ram was killed by Raj.’
6.3. Applicatives and the dative alternation
The applicative construction is one in which a CS argument that is canon-

ically expressed as an oblique argument is expressed instead as a direct
argument, in particular as a direct object. In some languages, such an
alternation is systematic and morphologically marked on the verb. That is,

as in the case of the passive, when the (normally) oblique object is expressed
as a direct object, the verb reflects that this has happened. Some examples
from Chichewa are given in (18)–(21). 5
(18) Chichewa
a. Mbidzi zi-na-perek-a msampha kwa nkhandwe.
zebras SP-PAST-hand-ASP trap to fox
‘The zebras handed the trap to the fox.’
b. Mbidzi zi-na-perek-er-a nkhandwe msampha.
zebras SP-PAST-hand-to-ASP fox trap
‘The zebras handed the fox the trap.’
(19) a. Ndi-na-tumiz-a chipanda cha mowa kwa mfumu.
1 SG. SUBJ - PAST-send-ASP calabash of beer to chief
‘I sent a calabash of beer to the chief.’
b. Ndi-na-tumiz-ir-a mfumu chipanda cha mowa.
1 SG. SUBJ - PAST-send-to- ASP chief calabash of beer
‘I sent the chief a calabash of beer.’
(20) a. Fisi a-na-dul-a chingwe ndi mpeni.
Hyena SP - PAST-cut-ASP rope with knife
‘The hyena cut the rope with a knife.’
b. Fisi a-na-dul-ir-a mpeni chingwe.
hyena SP-PAST-cut-with-ASP knife rope
(21) a. Msangalatsi a-ku-yend-a ndi ndodo.
entertainer SP-PRES-walk-ASP with stick
‘The entertainer is walking with a stick.’
b. Msangalatsi a-ku-yend-er-a ndodo.
entertainer SP-PRES-walk-with- ASP stick
‘The entertainer is walking with a stick.’
[Baker 1988:229, 230, 238, 260]
In each of the (a) examples of (18)–(20), there is a direct object and an

oblique object, which is introduced by a preposition (kwa “to”, ndi “with”).
In the (b) examples, there are two objects. The first object corresponds
to the oblique object, while the second object is the direct object. And in
the (b) examples, the verb is marked with the applicative marker -er/-ir. In
(21) the applicative makes an oblique argument into a direct object, with
no alternation involving another argument of the VP. It is important to
5
The gloss SP is a prefix that is used when there is a full NP subject. ASP is an
aspectual marker.
6.3. APPLICATIVES AND THE DATIVE ALTERNATION 193
notice that the applicative is not restricted to verbs indicating change of

possession.
We illustrate the function of the applicative in the correspondences in
(22)–(23). Note that the CS argument structure is the same in both cor-
respondences. The CS argument that is expressed as an oblique argument
in the nonapplicative is expressed as a direct argument in the applicative,
with associated morphology on the verb. Because there can be two objects,
we have to add a SecondObject to our inventory of possible grammatical
functions.
(22) SYNTAX S
NP VP
V0 NP PP
P0 NP
GF Subject Object
CS F(h1:X,h2:Y,h3:Z)
(23) SYNTAX S
NP VP
V0 NP NP
[APPLIC]
GF Subject Object SecondObject
CS F(h1:X,h2:Y,h3:Z)
In other languages, similar alternations are found that involve particular

lexical items and lack morphological marking. In the following examples
of what is called the dative alternation in English, the complement of the
preposition has the role of being the Recipient (or Beneficiary) of the action;
this individual comes into possession of the Theme, which is expressed by
the direct object.
 

 gave 


 showed 


 


 loaned 


 


 sold 


 sent 

the money to Sandy
(24) Chris mailed .

 
 Sandy the money

 took 


 


 brought 


 



awarded 

 assigned 
 
bequeathed

built a house for Sandy
(25) Chris bought .
Sandy a house
found
 
 donated 
 
the money to Sandy
(26) Chris presented ∗ .

 pushed 
 Sandy the money
committed
 
 constructed 
 
purchased a house for Sandy
(27) Chris ∗ .

 created 
 Sandy a house
invented
These cases are not applicatives, although they resemble them. Most
importantly, the alternation found in a language like English is not system-
atic, as it is in Chichewa. In English it is a lexically restricted alternation.
For example, it is not possible in English to have two objects where one
object is an Instrument and the other is the Theme. Compare the English ∗I
cut a knife the rope (from I cut the rope with a knife) with the Chichewa (20).
The dative alternation can be represented by having two related lexical
entries for verbs that participate in it. The entries for give are shown in (28).
Note the coindexing of the Goal argument with the object of the preposition
to in (28a) and with the first Object NP in (28b).
(28) a. 
give1 
CATEGORY V
 SYNTAX 
 COMPS 
NP [PP to NP1 ]
CS GIVE(AGENT/SOURCE:X,THEME:Y,GOAL:Z1 )
b. 
give2 
CATEGORY V
 SYNTAX 
 COMPS NP1 NP 
CS GIVE(AGENT/SOURCE:X,THEME:Y,GOAL:Z1 )
6.4. CAUSATIVE 195
While there is no systematic rule for deriving the English double object
construction, there are some identifiable regularities that may permit some
simplification of the lexical entries and a statement of the possible relation-
ships between them. See Research question 1.
6.4. Causative
Like the applicative, the causative construction can be found both as a reg-
ular alternation in some languages and as a lexically governed alternation.
English has lexical causatives, exemplified by the following.
(29) a. Sandy melted the ice.
b. Terry broke the window.
(29a) means that Sandy caused the ice to melt. (29b) means that Terry
caused the window to break.
It is notable that while there is no causative verb ∗ to fall meaning “to
make (something) fall”, there is a verb drop with causative meaning “to
cause to fall”. Similarly, there is no verb ∗ to large, but there is a verb enlarge
that has the causative meaning “to cause to become large”. The point is
that the causative alternation in English is a lexical one. Causative verbs
exist only for some nouns, verbs, and adjectives. Their form is idiosyncratic
(although there are some regularities) and therefore must be represented
explicitly in the lexical entry of each causative verb, along with the corre-
sponding meaning.
Moreover, the causative alternation in English involves only intransitives.
We can say
(30) Sandy caused Terry to break the window.
but there is no verb ∗ embreak which means to cause someone to break

(something).
∗
(31) Sandy embroke Terry the window.
In contrast, there are languages in which the causative is general and

morphologically regular, and applies to transitives and intransitives. In
Japanese, for example, a form of the morpheme -(s)ase is appended to the
root of another verb to make it causative. If the state or event that is caused
has an agent subject, then in the causative the subject NP is expressed as
a dative object. Example (32a) shows the dative for an intransitive, and
example (32b) for a transitive.
(32) Japanese
a. Suzuki-san-wa musume-ni daigaku-e ik-ase-ta
Suzuki-Mr.-TOP daughter-DAT college-to go-CAUSE - PAST
‘Mr. Suzuki made his daughter go to college.’
b. Chichi-wa imooto-ni piano-o naraw-ase-ta
father-TOP younger sister-DAT piano-ACC learn to play-CAUSE - PAST
‘Father made younger sister learn to play the piano.’
c. Watashi-wa reizooko-de miruku-o koor-ase-ta
I-TOP refrigerator-in milk-ACC freeze-CAUSE - PAST
‘I froze milk in the refrigerator.’ (Literally ‘I made milk freeze . . . ’)
[Makino and Tsutsui 1986:387–8]
In Nepali, a causative marker āū ∼ ā is adjoined to the verb root to create a

causative verb.
(33) Nepali
a. Transitive:
mā kān gar-chu
I work do-PRS.1SG
‘I do the work.’
b. Causative of Transitive:
mā kān gar-āū-chu
I work do-CAUS - PRS.1 SG
‘I cause someone to do the work.’
c. Transitive:
shyām-le d.hoka khol-a
Shyam-ERG door open-PST.3 SG
‘Shyam opened the door.’
d. Indirect Causative:
rām-le shyām-bat.a d.hoka khol-ā-yo
Ram-ERG Shyam-INSTR door open-CAUS - PST.3 SG
‘Ram made Shyam open the door.’
[Wallace 1979:1]
We do not give the correspondences for the causative here because they
depend on certain points that will be developed in Chapter 7. There are two
problems in Chapter 7 that ask you to formulate the correspondences for the
Japanese causative illustrated here, and for the English lexical alternation
exemplified by melt and break.
6.5. ANTIPASSIVE 197
6.5. Antipassive
We have seen that the passive permits suppression of the logical subject
(the highest CS argument) or makes it an oblique argument. There are
also grammatical devices in natural languages for suppressing or making
oblique the logical object, that is, the CS argument that would canonically
be expressed as a direct object. Antipassive is sometimes called detransi-
tivization, because it makes a transitive into an intransitive while holding
the subject constant.
Here are some examples of antipassives. Notice that the thematic roles
are the same, but the grammatical functions as marked by case are different.
In (34a) we see the active, with the Subject corresponding to ergative case
(marapai-thu “woman-ERG”) and the Object corresponding to absolutive
case (ithirr “seed.ABS”). In (34b), on the other hand, Subject corresponds
to marapai (“woman.ABS ”) and ithirr-ku “seed-DAT” is in an oblique case.
The thematic structure is the same in the two cases, but the grammatical
correspondences are different.
(34) Kalkatungu
a. Marapai-thu rumpa-mi ithirr matyamirla-thu
woman-ERG grind-FUT seed-ABS grindstone-ERG
‘The woman will grind the seed with the grindstone.’
b. Marapai rumpa-yi-mi ithirr-ku matyamirla-thu
woman-ABS grind-ANTIPASS - FUT seed-DAT grindstone-ERG
‘The woman will grind seed with the grindstone.’
[Blake 1994:50]
A similar picture is shown by the data from Chukchee in (35). In the

Chukchee example, the suffix on the verb agrees with the phrase marked
with the absolutive case. In the antipassive example (35b), the instrumental
is optional, just like the by-phrase is in the English passive. 6
(35) Chukchee
a. Paaček-a kimitP-@n ne-nìPetet-@n
youth-ERG load-ABS 3.SUB(T)-carry-3SG. OBJ
‘(The) young men carried away the load’
b. Paaček-@t Ø-ine-nìPetet-ÈPet kimitP-e
youth-PL ( ABS ) 3.SUB ( I )- AP-carry-3PL . SUB ( I ) load-INSTR
‘(The) young men carried away a load’
[Kozinsky et al. 1988:652]
6
Chukchee shows agreement morphology for the subject of the transitive
(SUB ( T )) and the subject of the intransitive (SUB ( I )).
Taking our analysis of the passive as our model, we can schematically

analyze the antipassive as follows. Note the similarities between this formu-
lation and that of the passive in (12b).
(36) SYNTAX NP−ABS NP−OBL V [ANTIPASSIVE]
GF Subject Object
(12) b. SYNTAX NP−NOM NP−OBL V [PASSIVE]
GF Subject Object
As in the case of the applicative, there are non-systematic lexical alterna-

tions in English that exemplify the type of relationship that is found in the
antipassive. However, English does not have the antipassive construction.
Crucially, the English examples do not involve morphological marking of
the verb and are strictly lexical.
In (37) we see examples of the sort that we noted in Chapter 5 that show
that for some verbs it is possible to simply omit the direct object, which
conveys the meaning that the corresponding CS argument is unspecified.
(37) a. Kim was eating cereal.
Kim was eating. (i.e. eating something)
b. Sandy was drinking beer.
Sandy was drinking. (i.e. drinking something)
c. Chris was reading the newspaper.
Chris was reading. (i.e. reading something)
d. Terry was cooking dinner.
Terry was cooking. (i.e. cooking something)
e. Marty was writing a letter.
Marty was writing. (i.e. writing something)
6.6. DUMMY SUBJECTS 199
f. Leslie was chewing the gum.

Leslie was chewing (i.e. chewing something)
Other verbs that allow missing objects are watch, paint, type, hammer, climb,
clean, chew, nurse.
There are many verbs in English that do not permit this option.
∗ 

 burning. (i.e. burning something) 


 ∗ covering. (i.e. covering something) 


 


 ∗
opening. (i.e. opening something) 


 ∗ seeing. (i.e. seeing something) 

(38) Kim was ∗ fixing. (i.e. fixing something)
∗
 


 consuming. 


 (i.e. consuming something)



 down down 
 ∗
 writing out . (i.e. writing out something) 


up up
It appears that whether a verb allows for a missing object is lexically

determined. However, the phenomenon need not be entirely idiosyncratic –
there may be semantically defined subclasses of verbs determined by their
meaning that do or do not permit this option. (See Research question 2.)
In addition, there are some lexical alternations involving a single verb
where the logical object argument is realized either as a direct object or an
oblique object, with a subtle meaning difference.
(39) a. The dog was chewing (on) the book

b. We clutched (at) the rope.
c. The chimp climbed (up) the tree.
d. I was reading (in) the newspaper.
Such lexical alternations resemble what is found in the antipassive. But

the alternations found in English contrast sharply with that of the passive,
which is available to virtually any verb that takes a direct object; the passive
is not lexically governed. (See Research question 3.)
6.6. Dummy subjects
Thus far, we have seen constructions, such as passive and antipassive, in

which CS arguments are not realized as syntactic arguments. Some lan-
guages also allow for the possibility that there is a syntactic subject but
no CS argument that corresponds to it. In English, for example, there must
be an overt grammatical subject in a finite clause. When this requirement
comes into contact with the absence of a CS argument that may correspond
to Subject, the Subject function is realized by a dummy or expletive subject.

Schematically this correspondence takes the form shown in (40).
(40) SYNTAX S
NP VP
…
it V
GF Subject
CS F(…)
The correspondence shown in (40) illustrates the fact that Subject in English
must correspond to something in the syntactic structure, even if there is no
CS element that corresponds to Subject.
The dummy subjects in English are it, as in (40), and there. Dummy
subjects can arise in a number of ways:
(i) There are no CS arguments. The weather verbs such as rain, snow, thunder
lack CS arguments. A sentence in which such a verb is the main verb has
the dummy subject it.
 

 raining 

snowing
(41) It’s .
 thundering 
 
hot in here
The word it does not refer to anything, as contrasted with it in a sentence

like It tastes funny.
(ii) The CS argument is syntactically displaced. This happens in English
when a clause that receives the subject thematic role is realized as a con-
stituent of VP.
(42) a. [That you are going to drop this class] bothers me.
b. It bothers me [that you are going to drop this class].
(43) a. [That some significant failures occurred] is obvious.
b. It is obvious [that some significant failures occurred].
6.6. DUMMY SUBJECTS 201
(44) a. [That you didn’t win] is a shame.

b. It is a shame [that you didn’t win].
This construction is called extraposition or it-extraposition.

Similar displacement occurs with NPs in the existential construction, in
which case the dummy subject is there. The existential interpretation is often
very difficult to get when the subject is not displaced.
(45) a. A fly is in my soup.
b. There is a fly in my soup.
∗
(46) a. Many reasons are for the failures.
b. There are many reasons for the failures.
(47) a. A lot of people were displaced.
b. There were a lot of people displaced.
A particularly striking fact about the existential construction is that the

verb agrees with the NP that follows it, e.g.

is
(48) a. There ∗ a fly in my soup.
are

∗
is
b. There flies in my soup.
are
Agreement normally holds between the subject and the verb. There are a
number of ways that we might imagine to account for the apparent excep-
tionality of this construction with respect to agreement; you are asked to
explore them in Research question 5.
(iii) The verb requires the dummy subject. In English, verbs of appearance
and occurrence have this property.
 

 seems 

appears
(49) It that there was no leadership.
 happens 
 
turns out
These are not instances of extraposition, as shown by the fact that the
sentential complement cannot appear in subject position.
 

 seems 

appears
(50) ∗ That there was no leadership .
 happens 
 
turns out
Since these verbs take a single Theme complement, the fact that this com-
plement cannot be the subject either must be stipulated, or explained in
terms of the semantic properties of the verb.
One of the most striking characteristics of dummy subjects in English that

will prove to be very important in subsequent analyses is that they must be
subjects. That is, there are no dummy objects. (Object can be suppressed, but
Subject cannot.) Each of the following examples has a dummy NP (there or
it), and each sentence is thematically complete, in the sense that all of the
thematic roles correspond to a phrase. Example (51c), for example, means
“Sandy has fallen”. The dummy it is in object position, and contributes
nothing to the meaning.
(51) a. ∗ A fly in my soup is there.
b. ∗ That you are going to drop this class bothers me it.
c. ∗ Sandy has fallen it.
The reason why this is important is that there are certain constructions in
which it appears that the dummy NPs can appear as non-subjects, e.g. I
believe there to be a problem; I believe it to be hot in here. If there must be a
subject, that fact places severe constraints on how we analyze this sentence.
We explore this issue in Chapter 7.
6.7. ∗ Null pronouns and clitics
There are basically two ways in which a CS argument can get its reference
in the world. One is through the form of the NP that corresponds to it, e.g.
Sandy or the person standing in the corner. The other is from the context.
In the latter case, English typically uses a pronoun like she, it, or them.
In this section we look briefly at two other ways in which languages mark
contextually determined reference.
6.7.1. Null pronouns
In some languages, an argument that would appear in English as a subject is

simply not expressed. That is, there is no syntactic NP argument associated
with the argument in the position where we would otherwise expect an
argument to appear. One such language is Italian.
(52) a. Ho mangiato gli spaghetti
have-1. SG eat-PAST. PART. the-M.PL spaghetti-M.PL
‘I ate the spaghetti.’
6.7. NULL PRONOUNS AND CLITICS 203
b. Non vuole leggere questo libro

not want-3SG read-INF this book
‘He doesn’t want to read this book.’
c. Abitano qui.
live-3 PL here
‘They live here.’
When there is an overt NP, there is agreement between the NP and the verb.
For instance, in (53a) Gianni and vuole are both third person singular, while
in (53b) le grandi firme and abitano are both third person plural.
(53) a. Gianni non vuole leggere questo libro
G. not want-3 SG read-INF this book
‘G. doesn’t want to read this book’.
b. Le grandi firme abitano qui.
the-PL large-PL company-PL live-3 PL here
‘Great companies live here.’
Interestingly, the inflection on the verb that shows up when there is no

overt subject NP is what we would expect given the lexical properties of the
noun that expresses the meaning of what is being referred to contextually.
That is, the agreement is with the morphological form that would be used if
the object in question were actually referred to using an NP. For example,
the word for “pants” in Italian is plural, as it is in English. It is also
masculine. When we refer to a pair of pants in Italian, the agreement is
masculine plural, even if no one has mentioned pants.
(54) Sono chiari
be-3 PL expensive-M . PL
‘They are expensive’ [pointing to a pair of pants]
[cf. Sono chiari, i pantaloni. ‘They are expensive, the pants.’]
But if we were referring to a house, which is casa, feminine singular, we

would say
(55) È cara
be-3 SG expensive-F. SG
‘It is expensive’ [pointing to the house]
[cf. È cara, la casa. ‘It is expensive, the house.’]
Crucially, the agreement here cannot depend on the semantic properties of

the physical objects, since whether they are singular or plural, or masculine
or feminine, is a morphological and therefore a lexical fact.
An example that illustrates this phenomenon clearly is even found in
English. Imagine that you pick up a pair of scissors and observe that the
scissors are very sharp. The word scissors is morphologically plural but
conceptually singular. You turn to someone nearby and say, “These are
really sharp.” Although no one has actually mentioned scissors, you use the
plural for pronoun agreement, not the singular, because scissors is lexically
plural. Similarly, you could say “These are really expensive” about a pair of
pants that both of you can see.
In order to understand how this works, we must first figure out how
agreement works. These last examples from Italian suggest that a CS argu-
ment that is not expressed overtly in syntax has features, such as number,
that correspond to the morphological features of the verb. The Subject GF
corresponds to the features of the CS argument but not to any part of
the syntactic structure. The thematic structure determines which argument
corresponds to the Subject GF, while the features associated with this argu-
ment determine the morphological features of the verb. We represent this
situation schematically as follows.
(56) SYNTAX … V [FEATURES]
GF Subject
CS F(h:X [FEATURES], …)
If the CS argument corresponds to an overt NP, we can maintain the

same schema, but link X directly to an NP in the syntax. Note that this NP
is not linked to the Subject GF. It supplies meaning and reference to the
Subject.
(57) SYNTAX NP … V [FEATURES]
GF Subject
Agreement in this case is mediated by the CS features, which are shared by

the NP and the V.
Let us consider now what happens when there are lexical features that
disagree with the CS features, as in the case of English scissors and the
Italian examples cited above. For such a case, we must link [FEATURES]
in the lexical entry of the word that corresponds to the concept in CS to the
verb, as shown in (58).
(58)
SYNTAX NP … V [FEATURES]
GF Subject
Lexicon word
SYNTAX CATEGORY N
FEATURES FEATURES
CS X
Both the syntactic features and the semantic CS features are connected
to the morphological features. The lexical features are relevant when they
override the semantically determined features, such as number and gender.
Even when there is no noun phrase subject, the verb V can have the agree-
ment features specified by the lexical entry that corresponds to the argument
X.
We should note that morphological agreement can also be achieved in
the standard derivational approach. On this approach, Italian and sim-
ilar languages have pronominal subjects, just like English. However, the
pronominal subjects in Italian are invisible. The invisible pronominal sub-
ject is typically called pro (“little pro”). (Languages that have pro subjects
are called pro-drop languages.) Pro may have any of the person, number, and
gender features associated with nouns and noun phrases. Research question
7 explores how the correspondences would be implemented by assuming
that Italian has the empty pronoun pro.
There are languages in which verbs agree with objects as well as subject.
Consider the following Swahili data; OM means “object marker” and is used
for third person arguments. Agreement here consists in the fact that in (59a),
for example, OM is attached to the verb and there is also an NP referring to

the Theme argument.
(59) Swahili
a. Na-mU-on-aa Maria.
1-TNS - OM-saw Maria
‘I saw Maria.’
b. Na-mU-on-aa.
1-TNS - OM-saw
‘I saw her.’
[Woolford 2000]
As in the case of pro-drop subjects, we represent the correspondence as

consisting of three parts. The thematic structure determines the grammati-
cal function that corresponds to the CS argument, the verbal morphology
which argument has this grammatical function, and the overt NP gives its
reference.
(60) SYNTAX S
… V−OM NP
GF Object
CS F(…, h:X[FEATURES], …)
Finally, there are languages such as Japanese and Korean in which argu-
ments other than subject may be null, and there is no verbal morphology
that marks the agreement. Here is a Japanese example that illustrates the
point.
(61) kyonen hon-o chuumon-sita ga, mada Ø uketotteinai
last year book-ACC order-PAST but yet Ø receive-PROG-NEG
‘(I) ordered a book last year but haven’t received (it) yet.’
[Adapted from Kayama 2003]
There is no overt pronoun that corresponds either to the subject I or the

object it. And it is less than fully acceptable to replace the Ø with an overt
pronominal, such as sore-o “it/that (ACC )”. It is generally assumed that in
such languages, the reference of the null arguments is determined by the
discourse context.
6.7.2. Clitics7
In some languages, arguments that would be expressed as pronouns in

English correspond to clitics. A clitic is a word that cannot stand alone but
must be attached to a word. Clitics typically appear in a position where the
corresponding full phrase does not occur. Direct object clitics in French, for
example, occur to the left of the finite verb, while full direct objects follow
the verb.
(62) a. Jean a lu tous les articles.

J. has read all the articles
‘Jean has read all the articles.’
b. ∗ Jean a lu les.
J. has read them
‘Jean has read them.’
c. Jean les a lus.
J. them has read
‘Jean has read them.’
The form les is a pronominal clitic, meaning “them”, that must be adjoined
to the left of the verb.
That the clitics are attached to the verb is shown by the fact that, unlike
a freestanding pronoun, a clitic cannot be used when there is no verb.
(63) French
a. Qui as-tu vu? – ∗ Le/∗ la/∗ les
who did-you see? him her them
b. Qui as-tu vu? – Lui/Elle/Eux
who did-you see him her them
Lui, elle, and eux are not clitic pronouns but freestanding pronouns, like
those in English.
Evidence that the clitic is not a freestanding word is also shown by the
fact that a clitic cannot be contrastively stressed. (We use regular capital
letters to indicate emphatic stress.)
(64) French
a. Jean préfère CELLE-là.
John prefers that one
‘John prefers THAT one.’
7
The summary in this section draws extensively from data in van Riemsdijk
1999.
b. Jean la préfère.
John her prefers
‘John prefers HER.’
c. ∗ Jean LA préfère.
Compare with English: John prefers HER, not HIM.

Moreover, clitics cannot be conjoined, unlike freestanding pronouns.
(65) French
a. Je connais Jean et Marie.
I know John and Mary
b. ∗ Je le et la connais.
I him and her know
(66) I know him and her very well.
Finally, there is a fixed order in which clitics must appear when there is more
than one. The order does not reflect the grammatical functions uniformly.
For example, in French, the clitic order is me le “to-me it” but le lui “it
to-him”.
(67) French
a. Jean donnera le livre à moi seul.
John will-give the book to me alone
b. ∗ Jean donnera à moi le livre.
John will-give to me the book
(68) a. ∗ Jean le me donnera.
John it to-me will-give
b. Jean me le donnera.
John to-me it will-give
‘John will give it to me.’
(69) a. Jean donnera le livre à Marie.
John will-give the book to Mary
‘John will give the book to Mary.’
b. Jean le lui donnera.
John it to-her will-give.
c. ∗ Jean lui le donnera.
John to-her it will-give
Thus, it appears that a clitic pronoun is not a phrasal constituent

but a morpheme adjoined to the verb. Yet it serves the function of
expressing a CS argument. Research question 8 asks you to formalize
6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE 209
the correspondences for the French pronominal clitics reviewed in this

section.
6.8. ∗ The transformational analysis of passive
6.8.1. Background
The foregoing sections of this chapter deal with the ways in which CS argu-
ments relate to syntactic arguments, and how these syntactic arguments are
distinguished from one another. This section is concerned with the classical
transformational approach to the CS–syntax correspondence.
In the descriptions that we have given in this chapter thus far, we have
shown the correspondence directly. For example, we say that in the passive
construction, the logical subject is suppressed and therefore the logical
object is linked to the syntactic subject.
(1) LINKING ( DEFAULT )

   

 Agent 
  Subject 
 
⇓ ↔ ⇓
 Theme/Patient 
   Object 
 
(7)
SYNTAX S
NP VP
be VP
[PASSIVE]
V PP
[PASSIVE]
P NP
seek
(=‘sought’) by
GF Subject Object
(9) be (passive)
  
CATEGORY V
  ! "
 SYNTAX  VP 
COMPS
PASSIVE
Transformational approaches to the passive (and other constructions)

proceed from the assumption that the CS corresponds only to a single
uniform syntactic structure. Proceeding from this assumption, it is typically
necessary to transform the syntactic structure that corresponds to CS into
one that more closely corresponds to the observed order of words and
phrases in the language. In this section we look at the motivations for such
an analysis, and some of the consequences.
6.8.2. The classical analysis
We look first at the classical analysis of the passive in transformational

grammar. This analysis was proposed originally by Chomsky 1957, who
argued that a phrase structure account of the passive construction in Eng-
lish was redundant in that it replicated aspects of the account of the cor-
responding active. This redundancy could be eliminated by tying both the
active and the passive to a common syntactic representation, in terms of
which selection restrictions and other properties are accounted for. In our
analysis, this common representation is CS; lacking CS as part of his theory,
Chomsky was led to argue that the common representation was a syntactic
one. This is a fundamental point.
This common syntactic representation came to be known as Deep Struc-
ture, and later as D-structure. The key property of D-structure is that it
determines the thematic argument structure of a sentence. If two sentences
have the same argument structure, they have the same D-structure. To
the extent that the two sentences differ in terms of their observed syntac-
tic structure (called Surface or S-structure), it is then necessary to posit
transformations that derive the different syntactic structures from the same
D-structure. Such a view of syntax has come to be called derivational. It
is to be contrasted with non-derivational or monostratal theories. A deriva-
tional theory assumes at least two levels of syntactic representation (at least
D-structure and S-structure), while a non-derivational theory assumes only
one.
Non-derivational theories of syntax seek to state correspondence rules

that relate the apparent syntactic structure of a linguistic expression and
its meaning, along the lines that we have been exploring thus far. When
two different syntactic structures correspond to the same meaning, there
are two sets of correspondence rules, one for each syntactic structure. For
example, for the English active we have posited that there is a correspon-
dence between the logical subject, the Subject GF, and the NP that is the
sister of VP. In the English passive, the logical object corresponds to the
Subject GF, which is realized syntactically as the sister of VP, as shown
in (7).
Derivational theories, on the other hand, have sought to simplify the
syntax–meaning correspondence by making it more or less uniform: same
meaning, same (underlying) syntactic structure. 8 But with this simplifica-
tion comes complexity in another dimension: the syntactic structure that
corresponds to the meaning is not the superficial syntactic structure of the
sentence. So there has to be abstract syntactic structure that accounts for
the meaning, and that structure must be transformed into the superficial
syntactic structure of the sentence.
The derivational approach to the active–passive relates different syntactic
argument positions to one another. The non-derivational approach that we
have sketched out in the first part of this chapter relates CS arguments
to different syntactic positions. Consider the derivational account of the
English sentences in (2), repeated here.
(2) a. Kim saw Sandy.

b. Sandy was seen (by Kim).
In a derivational approach, the D-structure is assumed to have the essential

characteristics of (2a) with respect to the syntactic representation of the
subject and object. 9 In the earliest transformational accounts of passive,
this structure was literally transformed into the structure of (2b) by a set of
operations on the phrase structure tree, sketched in (70).
8
The earliest version of this idea appears in Katz and Postal 1964, and hence is
called the Katz-Postal Hypothesis.
9
In a derivational theory, it is quite possible and very often the case that the
D-structure does not correspond to the derived structure of any sentence.
(70) S
NP VP
Kim V NP
saw Sandy
NP VP
Sandy V VP
was V PP
see+en P NP
by Kim
This derivation shows how each NP is moved to a different position, and

was, by, +en, and the PP are inserted. The formal statement of the passive
transformation was essentially the following.
(71) (NP1 ) V NP2 ⇒ NP2 be V+PASSIVE (by NP1 )
6.8.3. Structure preserving movement
A rule such as (71) is powerful enough to construct new trees out of old
trees, by creating structure and inserting lexical items. It quickly became
apparent that such rules are much more powerful than is necessary for
the description of natural language phenomena. An effort was launched
to constrain transformations, both in terms of what operations they can
perform and under what conditions they apply. The history of mainstream
generative grammar can be understood in large part as a succession of
proposals about how to formulate such constraints; it has seen its most
recent realization in the Minimalist Program (Chomsky 1995).
The standard view in mainstream generative grammar at this point is that
transformations are structure preserving. Either they do not build structure,
or they only build structure that is independently justified by the phrase
structure rules of the language. This assumption is called the structure
preserving hypothesis. 10
How the structure preserving hypothesis is observed depends on our
assumptions about how syntactic structures are constructed. In earlier
forms of mainstream syntactic theory, empty structures are posited into
which constituents are moved. Under this approach, the standard account
of passive is one in which the object NP moves into an empty subject posi-
tion, along the following lines. (The logical subject is either unexpressed, or
an oblique argument marked by the preposition by.)
(72) IP
[ ] I
I VP
V VP
be V+en NP
Note that if the NP that moves is interpreted as the object of V prior to the
movement, all of the selectional restrictions that hold in the active will hold
in the passive. In other words, the D-structure is a syntactic encoding of the
thematic information that is contained in the CS representation.
Another interpretation of the structure preserving hypothesis is that the
structure is built up piece by piece by putting words and phrases together
to get larger phrases. Pieces of a structure can be removed and attached
externally to the structure. Each step in this process has to observe the
conditions imposed by the phrase structure of the language. 11 The main
10
The structure preserving hypothesis was originally proposed by Emonds 1970.
11
This view is essentially that of the Minimalist Program (Chomsky 1995). The
basic operation of putting pieces together is called merge, and the operation of
attaching an already merged piece of structure is called external merge.
advantage of this approach in the Minimalist Program is that it does not

require that there be a D-structure level of representation – it allows for the
possibility that structures are built up piece by piece from the bottom up,
and that “movement” occurs at the point at which the structure satisfies the
conditions (as in (72)).
6.8.4. Why passive?
A derivation along the lines of (72) raises a number of questions, some of

them internal to the theory, and others having to do with the way in which
we account for passive in general. Among the theory-internal questions are
the following.
(i) Why does the NP have to move in English?
The simple answer is, if the NP does not move, the resulting sentence is
ungrammatical.
∗
(73) It was seen Sandy.
∗
Was seen Sandy.
But the theoretical question is, why is it ungrammatical?

(ii) Can the NP move in structures where there is no be . . . +en?
(iii) Why is it be . . . +en that is responsible for the movement, and not some other
marker? (And what’s going on with get passives?)
We can only summarize here the answers given to these questions. 12 Con-
sider first question (i). In minimalist formulations, there are in principle two
kinds of reasons why something would have to move to a certain position:
(A) The empty subject position has a property that the object NP lacks, so that it
must move to this position in order to have this property satisfied.
(B) The empty position lacks a property that the NP has, which triggers the move-
ment.
In either case, failure to have the property satisfied results in ill-formedness.

In this particular situation, the passive morphology renders the verb
12
There are many other questions to be asked, which are of a more technical
nature, e.g. (iv) “Why does the NP move to the subject position? Could it move
elsewhere? Could it move down?” In order to discuss such questions we will have to
develop more technical apparatus than is available to us at this point, so we put off
discussion of them until subsequent chapters.
intransitive (just as an adjective is intransitive); this is the key to the answer

to question (i), which involves a number of assumptions. One set has to do
with the assignment of case.
r An intransitive verb does not syntactically select a direct object.
r Selection of a direct object can be formulated in syntactic terms in terms of the
assignment of case to the object NP, or licensing the case on the NP.
r By assumption, all NPs must have case assigned to them, even if case is not
morphologically realized. It is abstract case. (This is called the case filter.)
r The consequence of the verb being passive, then, is that the direct object cannot
remain in its D-structure position because this is not a case position. It is not a
case position because the passive verb is intransitive.
A second set of assumptions has to do with agreement between the overt

subject position (Spec of IP) and the inflection I0 , through which case is
licensed.
r A sentence is IP, the maximal projection of I0 .
r In general a head and its specifier agree. This means that the head determines or
constrains the properties of the specifier.
r Case assignment is an agreement relation between an NP and a head.
r The subject position is a case position, since it is the specifier of the inflection I0 .
Finally, the assumption that there can be structure preserving movement

allows for the possibility that the case of an object NP that is not in a case
position will be licensed in the subject position.
r The NP must move to the specifier of IP in order to have a case assigned to it.
These assumptions produce the derivation in (74). Here we are assuming

that the direct object NP has a case that has to satisfy the requirements of
the head I0 under agreement. This is called case licensing.
(74) IP
[ ] I
ca
se
I VP
V+en no case NP
case
The derivational analysis of the passive has been extremely influential in

the development of syntactic theory. If we hold to the assumption that the
absence of a direct object in the passive is due to movement along the lines
illustrated in (74), then many consequences follow. For example, the passive
VP in (75) does not overtly display an NP that could have been moved from
the object position of released.
(75) Recently released from prison, Sandy immediately looked for a job.
According to the derivational analysis, the phrase recently released from

prison must be derived from recently released NP from prison by moving
NP to a subject position. Since there is no such visible NP and no such
visible position, both must be assumed to exist, but invisibly. Let us call this
invisible NP PRO; we then have the derivation in (76).
(76) IP
[ ] I
I0 VP
V VP
(be) release+en PRO
We take up derivations of this type in more detail in Chapters 7 and 9. In

the next section we consider whether movement is in fact the right way to
describe passive in general.
6.8.5. Passives without movement
Challenges to the derivational approach to passive come from languages in

which the motivation for a movement analysis is weak or non-existent. If it
turns out that we have to analyze such passives as direct correspondences
with CS, then such an approach is also more plausible for the English
passive.
One such type of passive is that of Manggarai, a language in the Bima-
Sumba subgroup of Indonesian languages. In this language there is no
passive morphology, but there is marking of the agent as an oblique

argument.
(77) Manggarai
a. Aku cero latung-k
1S fry corn-1S
‘I fry/am frying corn.’
b. [Latung hitu] cero l-aku-i
corn that fry by-1S-3S
‘The corn is (being) fried by me.’
[Arka and Kosmas 2005]
Manggarai is a language in which grammatical roles are marked by word

order, so it can be argued that latung hitu “the corn” in (77b) is the subject,
hence l-aku-i “by me” is an oblique argument. Since the verb is not marked
with passive morphology, it is hard to argue that it is comparable to the
intransitive adjectival or participial that occurs in English. Hence there
seems to be no basis for the argument that the object moves to subject
position because it lacks abstract case.
It is of course possible to claim that cero has passive morphology on it
in (77b), but it is invisible. But we should avoid the temptation to posit
invisible elements unless we absolutely need them. Otherwise we risk con-
structing theories and analyses that are untestable and thus impossible to
confirm.
As we have seen, a particular property of English and similar languages
is that the logical subject may be suppressed, or a passive VP may contain
an oblique argument that can be used to realize the logical subject. Other
languages using other syntactic devices to suppress the logical subject do
not also allow for the possibility of expressing it as an oblique argument.
These constructions are passives, in the particular sense that the logical
subject is suppressed. Two well-known cases are se in Spanish and si in
Italian.
(78) Spanish
a. Mis amigos comieron la torta.
my friend-PL eat-PAST-3 PL the cake
‘My friends ate the cake’.
b. Se comió la torta (∗ por mis amigos).
SE eat-PAST-3 SG the cake by my friends
‘The cake was eaten.’
(79) Italian
a. In Italia tutti mangiano spaghetti.
in Italy everyone eat-PRES -3 PL spaghetti-PL
‘In Italy everyone eats spaghetti.’
b. In Italia si mangia spaghetti (∗ per tutti).
in Italy SI eat-PRES-3 SG spaghetti- PL ( by everyone)
‘In Italy spaghetti is eaten.’
The English translations given here are passive, but they may also be
impersonal: “Someone ate the cake”, “People eat spaghetti”. Note that
in (79b) the singular verb does not agree with the plural spaghetti,
showing that spaghetti is the object and not the subject. The impor-
tant point is that these constructions are other ways of suppressing the
highest CS argument without syntactic movement and without passive
morphology.
It is also possible in some languages to suppress the highest argument
even when there is only one argument. We have seen that the passive
in English can be used when the Agent or Experiencer in a transitive is
unknown or irrelevant, as in “The cake was eaten” or “Terry was observed”.
We would translate it in the active as “someone”, i.e. “Someone ate the
cake”, “Someone observed Terry”. This same indeterminacy can be used
in intransitives in German and Dutch, producing so-called impersonal pas-
sives. (German wurde and Dutch wordt are inflected forms of the verbs wer-
den “to become” and worden “to become” used to mark the passive in these
languages.)
(80) German
a. Es wurde getanzt.
it be(come).PAST.3 SG dance-PAST. PART.
‘There was dancing.’
b. Es wurde viel gelacht.
it be(come).PAST.3 SG much laugh-PAST. PART.
‘There was a lot of laughing.’
(81) Dutch
a. Er wordt voor de konig geknield.
it is before the king kneel-PAST. PART
Lit. ‘It is kneeled before the king.’
‘One kneels before the king.’
b. Er wordt in deze kamer vaak geslapen.
it is in this room often sleep-PAST. PART
Lit. ‘It is often slept in this room.’
‘This room is often slept in.’
[Perlmutter 1978:168]
These impersonal passives appear to have the structure in (82).

(82) IP
[ ] I
cas
e
I0 VP
V0 VP
wurde V+en
getanzt
The empty position in the German impersonal passive can be filled by es

“it”, or by some other, topicalized constituent. Thus, the examples in (83)
are possible.
(83) a. Es wurde heute getanzt.
it was today dance-PAST. PART
b. Heute wurde getanzt.
today was danced
A straightforward correspondence for (83a) is as shown in (84). The
Agent is unspecified, and since there is no other CS argument, no CS
argument corresponds to Subject. But, by assumption, the syntactic subject
position must be filled. If a non-NP does not appear in this position, then
Subject is expressed as es by default.
(84) SYNTAX S
es VP
V0 VP
[PASSIVE]
werd- V0 …
[TENSE] [PASSIVE]
getanzt
GF Subject
CS DANCE(AGENT:X)
Such an analysis requires an account of how non-NPs can appear in the

initial, so-called “topic” position. Assume that in German and Dutch
declarative clauses there is a requirement that some constituent precede the
tensed verb. This requirement can be expressed as the following schematic
correspondence.
(85) SYNTAX S
XP VP
V0 …
[TENSE]
GF Subject
CS F(…)
The correspondence has three main features. First, it states that there is an
XP immediately before the tensed verb. Second, it links this XP with the
Subject GF, so that if there is nothing in the initial position, it will be filled
by a dummy NP. (In case something other than the logical subject corre-
sponds to Subject, the logical subject appears in VP with nominative case.)
And, third, it establishes this structure in the context of a non-interrogative
CS representation. This correspondence captures the fact that German and
Dutch are so-called “V2” languages, that is, languages in which the tensed
V must appear in second position in a declarative.
A question that arises with respect to impersonal passives is why English
cannot also have an impersonal passive, with an it or there subject.
∗
(86) <It/∗ There> was danced today.
This question is left for Problem 2.

Let us consider the derivational alternative, in which the passive is derived
through movement. The facts about the German and Dutch impersonal
passive suggest that, if the derivational analysis of the passive is correct,
the object appears in subject position in English not because of case but
because of some property of the Spec of IP in English that requires it to be
filled. On such an approach, German might differ from English in one of
several ways. One plausible possibility is that German, like English, has a
requirement that Spec of IP must be filled, but would have ways of filling it
in addition to movement of the object NP, i.e. by inserting the dummy NP
6.9. THETA CRITERION, EPP, AND UTAH 221
es “it” or by moving a non-NP to this position. Such an analysis would have

nothing to do with case per se, but would rather be formulated in terms of
the possible occupiers of the empty Spec of IP position. However, if an NP
with nominative case is moved into this position, its case can be licensed in
this position.
Recapitulating, a derivational approach in terms of case is not available
either for languages in which the passive is not morphologically marked,
e.g. the Spanish and Italian se/si-passives, and the Manggarai passive. 13 If
the verb is not detransitivized, there is no reason to argue that the direct
object is not properly case-marked. The fact that there are such passives
suggests that the central property of the passive is that the logical subject is
unspecified, and does not correspond to a syntactic argument, as illustrated
in (12).
A reasonable conclusion is that movement may account for some cases of
the active/passive alternation, but that it is not able to account for the full
range of possibilities found in natural languages. The alternative, that the
constructions are formulated in terms of correspondences crucially involv-
ing GFs, makes it possible to capture the observed variants transparently
and systematically.
6.9. ∗ Theta criterion, EPP, and UTAH
In Chapter 5, section 5.7.2, we introduced the theta criterion, which is

a central principle of mainstream syntactic theory, particularly Govern-
ment Binding (GB) Theory (Chomsky 1981) and subsequent work. As we
suggested,the theta criterion has certain consequences for syntactic analy-
ses. In particular, it leads to the conclusion that there are real syntactic
arguments that are phonetically empty. In this section we summarize a
number of related principles that will be relevant in the analyses that we
develop in the remainder of this chapter and elsewhere in this text.
The theta criterion stipulates that there is a correspondence between the
θ-roles associated with a head and the syntactic arguments of that head. We
formulate it here as follows.
13
See Haspelmath 1990 for discussion.
Theta criterion
i. Every θ-role is assigned to one and only one argument.
ii. Every argument is assigned one and only θ-role.
As we noted in Chapter 5, it follows from the theta criterion that if there

is a θ-role associated with a head, there must be a syntactic argument that
corresponds to that θ-role even if we cannot see it. So it follows that adjuncts
containing a passive, as in (87) –
(87) After having been arrested by the police, . . .
– must have an invisible object (call it PRO) that is assigned the Theme θ-
role and moves to the subject position, as illustrated in (88).
(88) After PROi having been arrested ___i by the police, . . .
Dummy subjects and the passive in other languages discussed in this

chapter suggest another principle. Recall that in the case of extraposition
there is a dummy subject; we repeat examples (42)–(44) from section 6.6
that illustrate this construction.
(42) a. [That you are going to drop this class] bothers me.
b. It bothers me [that you are going to drop this class].

(44) a. [That you didn’t win] is a shame.

b. It is a shame [that you didn’t win].
Dummy subjects occur because the requirement that there must be a subject
in English clearly must be satisfied whether or not there is an NP in subject
position that is assigned a θ-role. In the case of extraposition the θ-role is
assigned to the extraposed S.
(89) It bothers me that you are going to drop this class.
Exp Theme
Similarly, the subject position of seem must be filled by it when there is a
sentential complement. The examples in (50), repeated here, and (90) show
that this position is not thematic – it cannot contain a meaningful phrase,
such as an S (e.g. that there was no leadership) or an NP (e.g. Sandy). Only
the dummy it is possible.
6.9. THETA CRITERION, EPP, AND UTAH 223
 

 seems 

appears
(50) ∗ That there was no leadership .
 happens 
 
turns out

It
(90) ∗ seems that there was no leadership.
Sandy
This requirement that there must be a subject in English is called the
Extended Projection Principle, or EPP in mainstream analyses. This is
because it extends another principle, called the Projection Principle (see
Chomsky 1981; Chomsky 1986). The Projection Principle requires that all
grammatical features associated with lexical items be satisfied by all syn-
tactic representations in which they appear. This requirement is of course
straightforward if there is only one syntactic structure, but if there is a
derivation consisting of several representations, then things become more
complex. In particular, the Projection Principle requires that if a head
assigns a θ-role, then in any structure containing that head there must be
a corresponding argument that the θ-role is assigned to. This means that if
something moves to a position where it does not get assigned a θ-role, there
must be an invisible copy of it left behind.
In the case of dummy subjects, there is no θ-role assigned to the subject.
So in order to ensure that there will always be a subject in every structure,
even when there is no θ-role, it is necessary to extend the Projection Princi-
ple. Hence the “Extended” Projection Principle.
The EPP is typically implemented through an agreement feature on the
head in mainstream generative grammar (e.g. Chomsky 1995). In the case of
dummy subjects, it is assumed that I0 has a feature that must agree with the
NP feature of a phrase in Spec of IP. We call this feature [NP], as illustrated
in (91).
(91) IP
[ ] I
I0 …
[NP]
This feature is called the “EPP feature” of I0 . It follows from EPP that there
must be a constituent in Spec of IP in English that also has the feature [NP].
By assumption, an NP has the feature [NP], so this position may be filled
either by a full NP or a dummy NP.
(92) IP
NP I
[NP]
I0 …
[NP]
The last principle of mainstream analyses that we consider here is called

UTAH: Uniformity of Theta-Assignment Hypothesis. UTAH is the hypoth-
esis that, when two expressions have the same thematic structure, they
have the same syntactic structure. UTAH turns out to be a very powerful
hypothesis, since, if we assume it, we must posit syntactic arguments and
syntactic structure strictly on the basis of the thematic structure. So, it
follows from UTAH that (87) has an invisible object, since the thematic
structure of arrested by the police is the same as that of The police arrested
(someone).
UTAH is by no means universally assumed, but it is widely assumed
and it has been very influential in the formulation of a number of widely
accepted mainstream analyses of various syntactic phenomena. In this book
we do not assume UTAH. It is possible to capture the fact that two different
syntactic structures (such as active and passive) correspond to the same
thematic structure through correspondence rules which are needed anyway,
without having to posit syntactic transformations.
Exercises
1. Here are some sentences from Dyirbal. What kind of case-marking

pattern or patterns are shown in these examples? (By “pattern” we are refer-
ring to nominative/accusative, ergative/absolutive, passive or anti-passive,
etc.) Justify your answer. We’ve glossed the cases with X and Y for obvious
reasons.
(1) nguma yabu-nggu bura-n

father(X) mother-Y see-NONFUT
‘Mother saw father.’
(2) yabu nguma-nggu bura-n
mother(X) father-Y see-NONFUT
‘Father saw mother.’
EXERCISES 225
(3) nguma banaga-nyu

father(X) return-NONFUT
‘Father returned’
(4) yabu banaga-nyu
mother(X) return-NONFUT
‘Mother returned’
[§6.1.]
2. Demonstrate that what we are calling the “subject” of the English

passive is actually a subject, and not an object that for some reason appears
in initial position (e.g. as a consequence of topicalization). In order to do
this you first need to identify what properties English subjects have that
distinguish them from non-subjects, and then demonstrate that the subject
of a passive behaves in the same way.
[§6.2.]
3. What do the examples in (2) tell us about the constituent structure of a

sentence containing a by-phrase in a sentence such as (1)?
(1) Sandy was attacked by a cougar.
(2) a. They said that Sandy would be attacked by a cougar, and attacked by a
cougar she was.
b. They said that Sandy would be attacked by a cougar, and she was.
c. ∗ ?They said that Sandy would be attacked by a cougar, but she was by a
leopard.
d. A: Sandy was attacked.
B: ∗ ?Was she by a leopard?
Your precise answer will depend on whether or not you agree with the
judgments in (2c,d).
[§6.2.]
4. The basic property of the English passive that must be captured by

any analysis is that the subject of the passive has the same θ-role as the
object of the corresponding active. Verify that this property holds generally
by comparing the active and passive versions of sentences containing the
following verbs. In some cases you may find that it is difficult to say exactly
what the θ-role is, but you should be able to show that it is the same for
the object of the active and the subject of the passive by describing their
function. (Make up a pair of active and passive sentences for each verb in
order to test your intuitions.)
(1) a. kiss f. receive

b. believe g. enter
c. show h. evacuate
d. chew i. terrify
e. drink j. paint
[§6.2.]
5. Construct examples to determine which of the following verbs in

English allow for the double object construction. Each verb should appear
in a sentence of the form NP-V-NP1 -[PP to NP2 ] and in a sentence of the
form NP-V-NP2 -NP1 , e.g. Sandy gave a book to Kim and Sandy gave Kim a
book.
(1) a. throw f. transport
b. ship g. carry
c. email h. drag
d. fax i. pitch
e. Fedex j. transmit
There may be disagreement among speakers about some of these, but not
about others. What do you conclude from this? Why should some there be
some verbs that everyone agrees on, and others where there is disagreement?
[§6.3.]
6. For each of the following groups of sentences, say which argument-

changing relation is exemplified, and give your reasons. In each case, say as
precisely as possible what the morphological and syntactic differences are
between the examples that are relevant to the process.
(1) Turkish
a. Hasan öl-dü.
‘Hasan died.’
b. Ali Hasan-i öl-dür-dü.
‘Ali caused Hasan to die / Ali killed Hasan.’
(2) Swahili
a. Ni-me-lim-a shamba.
‘I have cultivated the plantain.’
b. Ni-me-m-lim-i-a Musa shamba.
‘I have cultivated the plantain for Musa.’
(3) Tukang Besi
a. No-lagu na mia.
‘The people are singing.’
EXERCISES 227
b. No-pa-lagu=’e na mia.
‘They made the people sing.’
(4) a. No-ja’o na bangka=’u.
‘Your boat is wrecked.’
b. No-pa-ja’o=ke na bangka=’u kene baliu.
‘They wrecked your boat with axes.’
(5) a. Ku-manga te ika.
‘I ate some fish.’
b. No-pa-manga=aku te ika.
‘She had me eat fish.’
(6) Tukang Besi
a. No-tu’o te kau kene baliu.
‘He chopped the tree with an axe.’
b. No-tu’o=ako te baliu te kau.
‘He used the axe to chop the tree.’
[§6.4.]
7. Demonstrate that the dummy subjects it and there in English are

really subjects, and not simply fixed forms affixed to the beginning of
the sentences in which they appear. In order to do this you first need to
identify the properties of non-dummy subjects that distinguish them from
non-subjects, and then demonstrate that it and there behave in the same
way.
[§6.5.]
8. Here are additional examples of the antipassive construction that was
introduced in [§6.5]. For each group of sentences, isolate and identify the
ergative, absolutive, and oblique case morphology, the antipassive marker
on the verb, and any verbal agreement if it exists. (Note that, in the
Chukchee examples, there is a phonological alternation in the verbal root
q@rir ∼ rer.)
(1) Chukchee
a. @tl@g-e qora-N@ q@rir-nin.
‘The father looked for the deer.’
b. @tlg-@n ena-rer-gPe.
‘The father did some searching.’ (i.e., searched for something)
[Kozinsky et al. 1988, cited by Kroeger 2004:293]
(2) Greenlandic Eskimo
a. Angut-ip arnaq unatar-paa.
‘The man beat the woman.’
b. Angut arna-mik unata-a-voq.

‘The man beat a woman.’
c. Angut unata-a-voq.
‘The man beat someone.’
[Sadock 1980:306; Baker 1988:131, cited by Kroeger 2004:293]
[§6.5.]
∗
9. Using syntactic trees, show the transformational derivation of the
following passive sentences. In each case, explain why the movement takes
place in terms of case. Assume that the basic structure is as given in (74) in
the text. (For the position of the by-phrase, see Exercise 3.)
(1) a. Sandy was attacked by a cougar.
b. Kim has been arrested.
c. We should never have been admitted into that club.
[§6.8.]
Problems
1. Work out the syntax–CS correspondence for the Spanish/Italian pas-

sives using se/si exemplified in (78)–(79) in the text. State the correspon-
dence generally: for any verb V expressing a CS relation F, under what
circumstances does se/si+V appear in the syntax? (Hint: The only argument
of F that is relevant is the logical subject).
[§6.2.]
2. Work out the correspondence for the German/Dutch impersonal pas-
sives exemplified in (80)–(81) in the text. State the correspondence gener-
ally: for any verb V expressing a CS relation F, under what circumstances
does the dummy subject with passive morphology on the verb appear in the
syntax?
[§6.2.]
∗
3. We have discussed the transformational analysis of the English passive
in which the direct object moves to Spec of IP because it fails to get a case.
One way of implementing this analysis is to say that the direct object has
a case associated with it, but its case cannot be “checked” by the passive
verb, which is intransitive. For present purposes, let us assume that there
PROBLEMS 229
are two ways in which case is checked: a lexical head checks the case of
its complement, and a functional head, such as I0 , checks the case of its
specifier.
What would a comparable syntactic analysis of the applicative look
like? In order to answer this question, you must be explicit about the
following:
r Is there movement? If so, what moves?
r Where does it move to?
r Why does it move?
r What is the role of the applicative morphology with respect to case assign-
ment/checking? Exactly how does it work (that is, what is the syntactic config-
uration in which it applies and what effect does it have)?
Illustrate your analysis using the following pair of sentences.
(1) Chichewa
a. fisi a-na-dul-a chingwe ndi mpeni.
hyena(1) SUBJ(1)-PAST-cut-ASP rope with knife
b. fisi a-na-dul-ir-a mpeni chingwe.
hyena(1) SUBJ(1)-PAST-cut-APPL-ASP knife rope
[§6.3.]
4. In the text it was claimed that an extraposed S is a constituent of the
VP. One alternative is that it is a sister of IP. Another is that it is a sister of
the inner VP that contains the V and its arguments, e.g.
(1) a. It [VP bothers me [CP that you didn’t win]].

b. [IP [It [VP bothers me]][CP that you didn’t win]].
c. It [VP [VP bothers me][CP that you didn’t win]].
Find syntactic evidence (e.g. constituency tests that isolate the VP), in order
to choose among these possibilities.
[§6.5.]
5. We suggested in the text that the subject it in English extraposition is a
dummy subject.

An alternative is that it in such examples is a pronoun that refers to the

proposition denoted by the extraposed clause, as it does in
(2) We knew that some significant failures occurred and it bothered us.
The following examples suggest that the dummy subject analysis is correct.
Explain why.

That
∗
(3) a. is obvious [that some significant failures occurred].
This
∗
b. [That some significant failures occurred], it is obvious.
[§6.5.]
6. On the assumption that the passive is accounted for in terms of a

correspondence, how do we capture the difference between English on the
one hand and German and Dutch on the other with respect to impersonal
passives? Precisely where does the difference lie?
[§6.8.]
Research questions
1. One way of thinking about the English dative alternation is that it is

constructional. The idea of a construction is that there are certain aspects of
meaning that are associated with the overall syntactic configuration and not
with any particular lexical items. In this case, the constructional approach
assumes that the structure [VP V NP1 NP2 ] has the interpretation associated
with it that NP1 comes into possession of NP2 in virtue of the action
denoted by V with respect to NP2 .
The meaning of the verb must be consistent with the constructional
meaning. On the constructional view, it is not strictly speaking true that
some verbs govern the dative alternation and others don’t. Rather, some
verbs can more readily be understood as denoting actions that can give rise
to this interpretation. Only those that can may be used with two objects.
Since different speakers may have different judgments with respect to the
suitability of a particular action functioning in this way, we might expect
differences among speakers regarding some of the judgments given in (24)–
(27) in the text. This is particularly the case with the benefactive for, where
it may take some imagination to see the connection between the action and
the intended interpretation.
The following are some judgments about verbs in English that allow
or disallow the double object construction as a paraphrase of for. Is
there a pattern that accounts for which verbs will allow this construction
and which ones will not? Feel free to add more verbs to test your
hypothesis.
(1) whistle Sandy a tune

sing Sandy a song
hum Sandy a tune
croon Sandy a tune
chant Sandy a lullaby
intone Sandy a blessing
play
 Sandy  a tune
 boil 
(2) broil Sandy an egg
 fry 
(3) paint Sandy a picture
sketch Sandy a house
scribble Sandy a picture/a note
doodle Sandy a cartoon
jot Sandy a note
scrawl Sandy a poem
outline Sandy the plans
 

 build 

construct
(4) ∗ Sandy a castle
 ∗ create 
 
concoct
(5) ∗ mow Sandy the lawn
[§6.3.]
2. The examples in (37) in the text show some verbs in English that allow
their objects to be omitted, while the examples in (38) show some verbs that
do not allow their object to be omitted. Is the ability of a transitive verb to
omit its object predictable? This is a somewhat open-ended puzzle, because
of the large number of verbs that would have to be checked in order to test
a hypothesis. Start with a hypothesis, e.g. about the completeness of the
activity that the VP denotes, or whether or not the object of the activity is
“conventional” in some sense, etc., apply that hypothesis to the cases in (37)
and (38), and proceed from there.
[§6.5.]
3. In the text we noted that some verbs can be used with direct or oblique
objects, e.g.
(1) to chew (on) the book

A. What is the meaning difference between the two cases illustrated in

(1)?
B. Does the meaning difference that you identify in A show up in other
cases where the verb appears with a direct object or with an oblique object
of the form [PP on NP]? Or is there a range of meanings associated with the
oblique object?
C. We noted that there are other prepositions that also permit a verb to
take an oblique object argument instead of a direct object, e.g. at, up, and
in. Do these prepositions all have the same function as on does? Is it possible
to predict which preposition will be used on the basis of the meaning of the
verb or the NP?
[§6.5.]
4. What is the constituent structure of a sentence in which there is the

subject? Start with the following examples.
(1) a. There is a serious problem with your proposal.

b. With your proposal there is a serious problem.
c. . . . and a serious problem there was with my proposal.
d. They said that there would be a serious problem with my proposal, and a
serious problem with my proposal there was.
e. ∗ They said that there would be a serious problem in my proposal, and be a
serious problem with my proposal there will.
[cf. . . . and be President of the company I will.]
(Hint: There are some really difficult issues here, the data is not cooperative,
and there may not be a clean solution.)
[§6.6.]
5. In the text we noted that agreement doesn’t work the same way in
existential sentences as it does in regular sentences. In particular, we get
the following pattern.

is
(1) a. There ∗ a fly in my soup.
are
∗
is
b. There flies in my soup.
are
There are several ways that we could account for these facts, while main-
taining the view that agreement is between the subject and the verb. We’ve
summarized them informally here; try to make them precise.
i. There is singular or plural depending on whether the NP that follows the

copula is singular or plural. (How do we represent the relationship between
there and the NP?)
ii. Agreement is between subject and verb, but the definition of what consti-
tutes “subject” is a little more complicated than we thought it was. (What
would be the definition of “subject”?)
iii. There is inserted into the structure by a transformation, and agreement
is taken care of before this transformation applies. (What does this transfor-
mation look like and how is agreement taken care of ?)
[§6.6.]
6. The agreement facts involving there that are discussed in Research

question 5 are further complicated by the fact that there’s can appear with
plural NPs.
 

 a lion 


 lions 

(1) There’s lots of lions in the closet.

 


 a lion and a tiger 

two lions and three tigers
The plurals are impossible with there is, though. But conjoined singular
NPs are possible with there is.
 

 a lion 


 lions
∗ 

(2) There is ∗ lots of lions in the closet.

 


∗a lion and a tiger 

two lions and three tigers
And the plural cannot be the subject of is.

 

 A lion 


 Lions
∗ 

(3) ∗ is in the closet.
Lots of lions

 


 ?A lion and a tiger 

∗
Two lions and three tigers
Which, if any, of the proposals in Research question 5 most easily accom-

modate facts such as these?
[§6.6.]
7. Assume for this exercise that the Italian sentences in (52)–(53) in the
text have an empty pro subject. Formulate the agreement rule for subjects
and the correspondence rule that relates the properties of the subject to the
properties of the corresponding CS argument. (Hint: Part of the trick here
is to isolate the morphology of the verb so that the agreement rule can refer
to it.)
[§6.7.]
8. Consider examples (62)–(69) in the text of French pronominal clitics.
Work out a set of correspondences that connect each clitic to the corre-
sponding CS argument. Here is a first approximation that handles just the
clitic les “them”.
(1) SYNTAX VP
V …
les−V+
GF Object
CS F(…, Ë:· [NUMBER:PLURAL])
The clitic corresponds to Object. V+ is the root of the verb, which may or
may not be lexical – it may be an auxiliary verb, literally corresponding to
English have or be. (Therefore, this verb does not necessarily correspond to
the CS relation F.)
An important observation about this type of object clitic is that it
excludes the possibility that there is also a full NP direct object. How is
this generalization captured in this analysis?
[§6.7.]
9. Assume that the object clitics in French exemplified in (62)–(69) in the

text are generated in the same configuration as full NP. They then have to
move to the verb in order to produce the observed surface order. State this
movement rule as precisely as you can. How is the observed order of clitics
captured in this approach?
[§6.7.]
10. In some dialects of Spanish, it is possible in certain cases to have both
a clitic and an overt NP. For example:
(1) a. Lo vimos a el
him see.PST.3PL a him-ACC
‘We saw him.’
b. Juan le dio el libro a Maria
John to-her give.PST.3SG the book a Mary
‘John gave the book to Mary.’
This phenomenon is called clitic doubling. How would you characterize this
phenomenon in terms of a syntax-GF-CS correspondence?
[§6.7.]
11. The following sentence is called a pseudo-passive.
(1) Robin was spoken to by the guard.
The NP that becomes the subject of this pseudo-passive is not a direct

object but an oblique object. There is a problem with pseudo-passives in
the transformational account that involves case assignment – isolate the
problem and say what it is. How would you solve this problem in a move-
ment account? How would you handle the problem using a correspondence
rule?
[§6.8.]
12. The following are grammatical sentences in English.
(1) a. Sandy took unfair advantage of Pat.

b. Pat was taken unfair advantage of.
c. Unfair advantage was taken of Pat.
A. Discuss these sentences in terms of the movement analysis of passive,

where the object NP moves to Spec of IP in order to have case assigned to
it. Specifically, is there a way to account for the two passives in terms of
this analysis? Be explicit in your answer, draw trees in order to be explicit,
and be clear about what needs to be assumed in order to make things work
out.
B. What would the analysis be in terms of a CS–syntax correspondence?
Are the problems noted in A avoided if we try to account for the two
passives this way? Again, be explicit in your answer.
C. What would the solution look like if the passive is triggered not by case
but by the EPP feature of I0 ?
[§6.9.]
6.1. 1
6.2. 2, 3, 4 1, 2
6.3. 5 3 1
6.4. 6
6.5. 7, 8 4, 5 2, 3
6.6. 4, 5, 6
6.7. 7, 8, 9, 10
6.8. 9 6 11
6.9. 12
7
Complex clauses: raising
and control
7.1. Infinitival complements
In Chapter 4 we arrived at the following rule to summarize the constituent

structure of VP.
(1) VP → V (NP) (NP) (PP∗ ) (Adv∗ ) (S)
One very important omission in this rule is the infinitival complement,

which is illustrated in (2).
(2) a. George tried to run for President.

b. George appeared to run for President.
c. George persuaded Al to run for President.
d. George believed Al to have run for President.
An infinitival complement is a verb phrase in which the bare form of the

verb is preceded by to. We assume for convenience that to is an inflection on
the verb; to run will be represented in the structure as in (3).
(3) V
to run
We refer to the infinitival VP as VPINF .

Given such examples, it is natural to extend our rules along the following
lines.

NP
(4) VP → V (NP)
VPINF
This rule will accommodate the patterns V-VPINF and V-NP-VPINF that are
seen in (2).
238 7. COMPLEX CLAUSES: RAISING AND CONTROL
Whether a particular verb appears with particular complements is a

lexical property of the verb, as shown by the following ungrammatical
examples. Try and appear do not permit NP complements, while persuade
and believe require them.
∗
(5) a. George tried Al to run for President.
∗
b. George appeared Al to run for President.
∗
c. George persuaded to run for President.
∗
d. George believed to run for President.
On the basis of such observations, we can specify the subcategorization

properties of each verb with respect to the various complements. A few
verbs in the table in (6) are left for you to check in Exercise 1 (the ones
marked in italics).
(6)
Verb VP complement? NP VP complement?
try VPINF
appear VPINF
persuade NP VPINF
believe NP VPINF
want VPINF NP VPINF

force NP VPINF
promise VPINF NP VPINF
manage VPINF
tell NP VPINF
expect VPINF NP VPINF
forget VPINF
say VPINF
A very important fact about infinitival complements is that in many cases

they allow for paraphrases in the form of a tensed sentence. Here are some
examples.
(7) a. i. George appeared to run for President.
ii. = It appeared that George ran for President.
b. i. George persuaded Al to run for President.
ii. = George persuaded Al that he (Al) should run for President.
c. i. George believed Al to have run for President.
ii. = George believed that Al ran for President.
But sentential paraphrases are not always possible; cf.
7.1. INFINITIVAL COMPLEMENTS 239
(8) a. i. George tried to run for President.

∗
ii. George tried that he would run for President.
b. i. George forced Al to run for President.
∗
ii. George forced Al that he (Al) should run for President.
In considering the interpretation of infinitival complements, observe that
part of the knowledge that we have about the meaning of the infinitival con-
struction is that a particular individual satisfies the thematic requirements
of the infinitival verb. For example, in the case of try we know that the
“tryer” is also the person running for President, and in the case of force we
know that the “forcee” (the Patient of force) is also the person running for
President, and not the “forcer” (the Agent of force). This relation, called
control, holds for a given verb whether or not a sentential paraphrase along
the lines of (8) is possible. We look more closely in the following sections at
what determines how the thematic requirements of the infinitival predicate
are satisfied.
7.1.1. Subject control
Let us begin with the verb try. As just noted, the identity of the individual
that is understood as the subject of the infinitival complement is the subject
of try. This is true no matter what the infinitival complement is, as a
consideration of the following examples shows.
(9) a. George tried to run for President.
b. Al tried to be charismatic.
c. Ted tries to look honest.
d. Bill tries not to think about the past.
In none of these cases can we understand the subject of try to be exerting

an effort so that someone else has the property denoted by the infinitive.
For example, in (9a) George did not try to bring it about that someone else
ran for President, and in (9c) Ted cannot be trying to bring it about that
someone else looks honest. These are cases of obligatory control.
Consider now sentence (10).
∗
(10) George tried to elapse.
There is a selection restriction violation in this case – see Chapter 5, sec-

tion 5.7.1. Intuitively, the semantic clash in this case is the same as the one
in (11).
(11) a. ∗ George elapsed.

[TIME-PERIOD]
b. ELAPSE(THEME: )
GEORGE[HUMAN]
It is as though George is the subject of elapse in (10), even though elapse

does not appear to have a subject in this case. One way to capture this
relationship between the two sentences is to say that in the CS of (10)
there are two identical GEORGE arguments. One GEORGE is the Agent
of TRY and the other is an argument of ELAPSE. Since these are the same
individual, we represent them by putting the same referential index on the
two CS arguments. We put the representation GEORGE in just once, for
readability – the sameness of reference is indicated by the identity of the
indices (see Chapter 5, section 5.2.2).
(12) TRY(AGENT:GEORGE[HUMAN]· , ELAPSE(THEME:·[TIME-
PERIOD]))
Since · is the same individual as GEORGE[HUMAN]· and ELAPSE

(·[HUMAN]) is semantically anomalous, the clash between [TIME-
PERIOD] and [HUMAN] that we see in (11b) also occurs in (13).
(13) TRY(AGENT:GEORGE[HUMAN]· , ELAPSE(THEME: ·

[TIME-PERIOD]
))
[HUMAN]
The problem is that the bound argument · is represented simultaneously as

[TIME-PERIOD] and [HUMAN].
A similar clash occurs when we use a temporal expression as the subject
of try.
∗
(14) Two days tried to elapse.
While two days can be the subject of elapse, it cannot be the subject of try,
which requires that its subject refer to an animate object. The semantic clash
appears in the Agent argument of TRY in (15).

TWO-DAYS[TIME-PERIOD] ·
(15) TRY(AGENT: ,ELAPSE(THEME: ·
[ANIMATE]
[TIME-PERIOD]))
In general, whether or not there is a semantic clash, we interpret the

infinitival complement of try as having an argument that is identical in
reference to the subject of try. This argument of the infinitival complement
is the one that would be realized as a subject if the infinitival complement
were restated as a tensed complement.
The correspondence between the syntactic structure in which try takes

an infinitival complement and the CS representation is shown schematically
in (16).
(16) SYNTAX S
NP VP
V0 VP
...
try V0
GF Subject [ Subject ]
CS TRY(AGENT:X·,F(θ:·, . . .))
This schema shows that the argument of the infinitival clause that would
be realized as subject does not correspond to anything in the syntax. It
is mapped to the Subject of the complement, which is not mapped into
anything in the syntactic structure, because the complement is an infinitival
VP and not an S. Moreover, this argument is coindexed with the Agent
argument of try, as shown by the index · on the Agent of try and in the
relevant argument position of the complement.
These are the two main characteristics of subject control. The subject of
try in this case is called the controller of the unrealized argument.
Subject control
r The argument of the complement that corresponds to Subject is not syntacti-
cally realized.
r The unrealized argument is coindexed with the CS argument that corresponds
to the Subject of the higher verb.
As we will discuss at greater length in section 7.4, there is also a strictly

syntactic treatment of control that captures the correspondence outlined
here. On this alternative analysis, the infinitival complement actually is a
sentence with an invisible subject NP, called PRO, which corresponds to the
argument expressed as · in the CS representation. On such an approach, the
superficial syntactic structure is not that given in (16), but (17); note how ·
in the CS corresponds to PRO in the syntax.
(17) SYNTAX S
NP VP
V0 S
NP VP
try
PRO V0 ...
CS TRY(AGENT:X·, F(θ:·, . . .))
We will discuss in detail the motivations for and theoretical consequences

of such an approach to control in section 7.4.
7.1.2. Raising to subject
Next we look at (2b), which contains the verb appear. Here is the sentence
again.
(2) b. George appeared to run for President.
This sentence looks just like a sentence with try: there is an infinitival com-
plement of the verb that follows the main verb try or appear. However, there
are important differences between try and appear. First, we can paraphrase
the sentence with appear so that it has a sentential that-complement and a
dummy it-subject, but we cannot paraphrase the sentence with try in this
way.
(18) a. It appears that George ran for President.
b. ∗ It tried that George ran for President.
Second, while intuitively the subject of try is an Agent (in this case, someone
who tries), the subject of appear is not an Agent. In fact, the subject of
appear has no role with respect to appear, and can even be a dummy NP,
such as there and it in (19).
(19) a. There appeared to be a problem.
cf. It appeared that there was a problem.
b. It appeared to be raining.
cf. It appeared that it was raining.
c. It appeared to be obvious that George would win.
cf. It appeared that it was obvious that George would win.
The subject of appear may be part of an idiomatic expression (such as the

devil (be in the details) in (20)).
(20) a. The devil is in the details.
b. The devil appears to be in the details.
c. It appears that the devil is in the details.
∗
d. The devil tried to be in the details.
On the other hand, as with try, if the subject of a predicate is odd because of
a selectional violation, then the subject of appear is odd when the predicate
is infinitival.
(21) a. #George elapsed.
#George appeared to elapse.
b. #The dog dispersed.
#The dog appeared to disperse.
This generalization also accounts for the oddness of examples such as

(21). If George cannot be the subject of elapse, then George cannot be
the subject of appear to elapse. The close connection between the subject
of appear and the infinitival complement is shown as well by examples in
which a predicate is grammatically dependent on the subject. Consider the
following examples.
(22) a. Kim and Sandy are friends.
∗
b. Kim is friends.
c. Kim and Sandy appear to be friends.
∗
d. Kim appears to be friends.
(23) a. Kim is friends with Sandy.
b. Kim appears to be friends with Sandy.
Here we see that there are two uses of the predicate friends. The examples
in (22a,b) show that friends takes a plural subject. This pattern is replicated
in (22c,d), where a plural subject of appear is grammatical, while a single
subject is not. The grammaticality of (22b) can be accounted for if the
subject of appears is in some sense the subject of to be friends. As seen in
(23), it is possible for friends to take a singular subject, but only when it also
has a complement PP of the form with-NP. But then the singular subject of
appear is also possible.
So it seems that in some respects try and appear are the same, while in
others they are different. How do we account for these facts? The key is
to observe that dummy NPs like it and there may only appear when they
are the subjects of certain predicates. As the pairs of sentences in (19)–
(20) show, we get there appeared to be . . . when the paraphrase is it appears
that there was . . . , and we get it appeared to be . . . when the paraphrase is it
appeared that it was . . . . The expression The devil is in the details has a non-
literal interpretation, which is preserved even when the devil is the subject
of seem and the infinitival complement is be in the details. In other words,
something can be a subject of appear to VP whenever it can be a subject of a
that-complement containing VP in it appears that . . . .
Thus, it appears that in all respects the Subject GF of the infinitival
predicate is expressed syntactically in the “wrong place” as the subject of
the verb appear. We show this relationship in the following correspondence.
(24) SYNTAX S
NP VP
V0 VP
appear V0 ...
GF Subject [ Subject ]
APPEAR(F(θ:X, . . .))
The CS argument of the complement that corresponds to the Subject GF

of the infinitival complement is linked to the Subject GF of appear, which
is expressed syntactically as the sister of VP.
This schema shows that the syntactic subject of appear does not cor-
respond to a CS argument of APPEAR. Rather, the syntactic subject of
appear is linked to the Subject GF of the predicate. In fact, APPEAR has
no arguments in CS that correspond to NPs – it is what we call an operator.
It has one argument, which corresponds to a clause.
This infinitival construction with verbs like appear is often called raising
to subject, because the subject of appear satisfies all of the semantic require-
ments that the infinitival predicate places on its subject.
Raising to subject
• The Subject of the complement corresponds to the Subject of the higher, that
is, the raising verb.
• The Subject of the higher, that is, the raising verb does not correspond to a CS
argument of that verb.
Comparing raising and control

There is no control in the case of a raising verb like appear. When there
is raising, there is just one semantic argument. This argument appears as
the subject of the raising verb (see (24)).
APPEAR(RUN(AGENT:X, . . . )) (for appear to run)
When there is control, as in the case of try, there are two distinct CS
arguments that refer to the same thing. These are X· and · in the CS
representation, e.g.
TRY(AGENT:X· , RUN(AGENT:·, . . . )) (for try to run)
The transformational analysis of raising to subject, which we discuss in

section 7.3, captures these facts by assuming that the infinitival predicate
has a subject that is moved by a syntactic transformation to the subject posi-
tion of the higher verb. The syntactic structures and the correspondences
with CS are identical for the infinitival and that-S complement.
(25) SYNTAX S
[ ] VP
V0 S
appear NP VP
V0 ...
CS APPEAR(F(θ:X, . . .))
Summary: Tests for “raising”

The signs of “raising” are that an NP in an argument position of some
verb V1 bears no semantic relationship to V1 ; rather, it behaves semanti-
cally as though it is the subject of a verb V2 lower in the structure. This
is shown by the following properties:
• The NP gets a θ-role from V2 .
• The NP satisfies the subject selectional restrictions of V2 .
• The NP may be a dummy subject that is licensed by V2 .
• The NP may be part of an idiomatic expression that includes V2 .
7.1.3. Object control
Consider next (2c).

(2) c. George persuaded Al to run for President.
Unlike try and appear, persuade takes an NP complement (the object) and a
VP complement. We know that Al is a syntactic object of persuade because
it can be replaced with a pronoun that has accusative case.

him
(26) George persuaded ∗ to run for President.
he
The meaning of persuade when it takes an infinitival complement involves
getting an individual to undertake an action. This individual has two
semantic roles – being acted upon (as Patient) by the subject of persuade,
and undertaking the action in the role of its Agent. Thus, the infiniti-
val predicate of persuade must denote an action, and the object of per-
suade must refer to an animate entity that can serve as the Agent of this
action.
Our intuition about the meaning of a sentence like (2c) is that the object in
fact has two semantic roles associated with it. It is understood as the subject
of the infinitival complement, in this case, the Agent of run for President,
and it is understood as the object of persuade, that is, a Patient, in the sense
of someone or something that is acted on and affected by the subject of
persuade.
The following examples show what happens when these requirements are
not met.
 

 the sky to be blue 


 


 2 + 2 to equal 4 

the chair to break
(27) #George persuaded .

 it to rain 


 


 there to be an explosion 

the devil to be in the details
Because of the dual roles, the understood subject of the infinitival com-
plement is the entity denoted by the direct object of persuade. This relation
is similar to that in the case of try, except that in this case it is the object and
not the subject that determines the missing subject. We call this relation
object control.
Object control
• The argument of the complement that corresponds to Subject is not syntacti-

cally realized.
• The unrealized argument is coindexed with the CS argument that corresponds
to the Object of the higher verb.
We represent the correspondence between syntactic structure and CS in

this case as with the other verbs. What is different in the case of persuade is
that the NP object corresponds to two CS arguments.
(28) SYNTAX S
NP VP
V0 NP VP
persuade V0 ...
GF Subject Object [Subject ]
CS PERSUADE(AGENT:X, PATIENT:Z·,F(θ:·, . . .) )
The CS of a sentence that has object control will have a coindexed argu-
ment in the complement.
(29) PERSUADE(AGENT:GEORGE, PATIENT:AL· , THEME:RUN(AGENT:·))

Because the object of the verb persuade determines the reference of the
bound argument of the complement, this construction is object control. The
controller in this case is the direct object of the higher verb (e.g. persuade).
Note once again that there are two arguments marked with ·, although only
one is realized syntactically.
7.1.4. “Raising” to object
Finally, let us consider the verb believe.
(2) d. George believed Al to have run for President.
This case resembles object control syntactically, in that there is an NP

complement and an infinitival complement. But in this case the object of
believe does not bear a role with respect to believe. We can show this by
observing that while there is a “believer”, there is no “believee”. Because of
the absence of a thematic role, the object of believe may be a dummy NP or
part of an idiom, parallel to the subject of appear. These examples contrast
dramatically with the object control examples in (27).
 

 the sky to be blue 


 


 2 + 2 to equal 4 

the chair to break
(30) George believed .

 it to rain 


 


 there to be an explosion 

the devil to be in the details
In other words, the object of believe has the selectional properties only of
the subject of the infinitival complement. Consistent with this is the fact
that when this NP is an expression that refers to something, it satisfies the
semantic requirements of the complement as though it was the subject.
 
∗

 Al 
 two days to have elapsed  
(31) George believes .


∗ 
to have dispersed 
dog
 the 
dogs
Just as ∗ Al elapsed violates a selectional requirement of elapse, so does

∗
George believes Al to have elapsed, and similarly for the other examples.
If we substitute persuade for believe we get a different pattern of judg-
ments, because the object of persuade must satisfy selectional restrictions of
both persuade and the infinitival predicate.
 
∗

 Al 
 ∗ two days to elapse  
(32) George persuaded .


∗ 
to disperse 
dog
 the 
dogs
Believe behaves the same with respect to friends as does the raising to
subject verb appears, except that in this case, the NP in question must be
the object of believe.

Terry and Robin
(33) a. I believe ∗ to be friends.
Terry
b. I believe Terry to be friends with Robin.
Because of the fact that the object of believe has the properties of the
subject of the infinitive, this construction has been referred to as raising to
object. In earlier transformational accounts, the subject of the complement
was transformationally moved from the subject position of an infinitival
complement S into the object position of believe. In contemporary accounts,
there is no raising. There are two types of non-raising analysis.
(i) The apparent object of believe is represented as the actual syntactic
object of believe but has properties that correspond to the subject of the
complement; this alternative is sketched in (34).
(34)
SYNTAX S
NP VP
V0 NP VP
believe V0 ...
GF Subject Object [Subject ]
CS BELIEVE(EXP:X, THEME:F(θ : Y, . . .))
(ii) The apparent object of believe is really the syntactic subject of the
complement, although it turns out to behave in all respects as though it is a
syntactic object. This alternative is illustrated in (35).
(35) SYNTAX S
NP VP
V0 S
believe NP VP
V0 ...
CS BELIEVE(EXP:X, THEME:F(θ:Y, . . .))
This is the standard analysis in Government Binding theory (see Chomsky

1981).
In a theory that assumes a structure like (35) it is necessary to explain
how it is that the subject of the infinitival complement has the grammat-
ical character of a direct object of the higher verb believe in all respects.
This is a very important issue and forms the basis for much of contem-
porary mainstream syntactic theory. We consider it in greater detail in
section 7.3.2.
Raising to object
• The argument of the complement that corresponds to subject is syntactically

realized as the object of the higher verb.
• The object of the higher verb has no semantic function with respect to that
verb.
We conclude our comparison of raising and control by noting the behav-

ior of passive infinitival complements in the two constructions. In the case
of raising, the active and passive are essentially synonymous.
(36) a. The police seem to have arrested Robin.
b. Robin seems to have been arrested by the police.
(37) a. Terry believes the police to have arrested Robin.
b. Terry believes Robin to have been arrested by the police.
The reason for synonymy in these cases is that the only difference between
the active and the passive is how the CS arguments are realized syntactically.
But, in the case of object control, the active and the passive are not
synonymous.
(38) a. The police tried to arrested

Robin.
?be
b. Robin tried to arrested by the police.
get
(39) a. Terry persuaded the policeto arrest
Robin.
?be
b. Terry persuaded Robin to arrested.
get
The reason is that the CS of the two are different. The controlled argument
is different in the active and the passive, so the controller is different, too.
And, since the controller has a CS role, the meaning is different in the active
and the passive.
7.1.5. The case of expect
A complication in the picture that we have presented here is the verb expect.
Consider the following sentences.
(40) a. George expected to win.

b. George expected Al to win.
On the one hand, expect looks like a subject control verb, like try. We
understand the Experiencer of win to be GEORGE in (40a). But on the
other hand, expect looks like a raising to object verb, like believe. We
understand the Experiencer of win to be AL in (40a). So which is it?
The answer is, “both”. There are two variants of expect that take infini-
tival complements, one of each type. Both are paraphrasable by that-
complements.
(41) a. George expected that he (George) would win.

b. George expected that Al would win.
In this respect, expect is similar to believe. However, believe has only one
variant. While expect allows for subject control, believe does not.

expects
(42) George ∗ to be famous.
believes
Finally, because it is a raising to object verb, expect has the same super-
ficial structure as persuade, which is an object control verb. Exercise 4 asks
you to show that expect is actually a raising to object verb, not an object
control verb.
7.1.6. Gerundives
Before concluding this discussion of infinitival complements we note some

basic properties of another type of non-finite construction, the gerundive.
Some examples are given in (43).

into
(43) a. I talked Terry taking a picture of me.
out of
b. I talked to Terry after taking a picture of the scene.
c. Terry considered spending the night at a local hotel.
d. I began speaking German to the children when they were young.
e. Taking a picture of the scene turned out to be a good idea.
In (43a) the Agent of taking a picture is Terry. This sentence looks like a case

of object control, where talk into selects a complement of a particular
out of
form. Notice that there cannot be an infinitival complement here.

∗ into
(44) I talked Terry to take a picture of me.
out of
In (43b) we have what appears to be subject control – the person taking

the picture cannot be Terry. Since after taking a picture is an adjunct, it is
not selected as an argument of talk.
Example (43c) appears to be a case of subject control, where the comple-
ment is selected by consider. Consider resembles expect in that it does not
require control.
(45) Terry considered (= thought about) Robin spending the night at a local hotel.
In comparison, (43d) is a case in which subject control is required.

∗
(46) I began Robin speaking German to the children when they were young.
Begin allows not only a gerundive complement but an infinitival comple-

ment.
(47) I began to speak German to the children when they were young.
Finally, (43e) is an example of arbitrary control.

We do not work out the analysis of gerundives in detail in this book. But
it should be kept in mind that any theory of control must account not only
for the distribution and interpretation of infinitivals but the distribution and
interpretation of gerundives.
7.2. MORE CORRESPONDENCES 253
7.1.7. Summary: raising and control
The simple syntactic structure of infinitival complements in English is cap-

tured by the rule in (4), repeated here.

NP
(4) VP→ V (NP)
VPINF
This simple syntactic structure masks a more complex set of relations. There
are two types of verb that take an infinitival complement alone, and two
types of verb that take an NP and infinitival complement.
i. Subject control verbs, like try, select infinitival complements. The subject of try
is the controller of the unrealized subject of the infinitive.
ii. Subject raising verbs, like appear, select infinitival complements. The subject
of appear is in all respects except syntactic configuration the subject of the
infinitive.
iii. Object control verbs, like persuade, select NP and infinitival complements. The
object of persuade is the controller of the unrealized subject of the infinitive.
iv. Raising to object verbs, like believe, select NP and infinitival complements. The
object of believe is in all respects except syntactic configuration the subject of
the infinitive.
The remainder of this chapter is concerned with complexities in the

analysis of raising and control. Section 7.2 refines the correspondences
sketched out in section 7.1 to reflect the fact that the controlled or “raised”
argument is not always the logical subject of the complement but can be
any argument that could be realized as a syntactic subject. In sections 7.3
and 7.4 we summarize the classic transformational approach that assumes
maximal syntactic uniformity in order to capture these correspondences
and the generalizations about them. Several of the problems at the end of
this chapter look at how control is expressed cross-linguistically.
7.2. ∗ More correspondences
Consider again example (40a), an example of subject control with expect.
(40) a. George expected to win.
The lexical entry of this variant of expect will specify that its subject takes
the Experiencer role, and that it takes an infinitival VP complement. As a
first approximation, we state these properties as follows in an AVM. Note

that we do not mention the coindexed argument in the complement.
(48) expect
   
!
CAT
V "
 SYNTAX  VP1  
 COMPS 
 TENSE INF 
CS EXPECT(EXP:Xa , THEME:F1 (. . . ))
This AVM shows through the subscripts that the infinitival complement
VP1 corresponds to the proposition F1 (. . .) that bears the Theme role in
the CS representation. This Theme role can also be realized as a sentential
complement, which is expressed by another lexical entry for expect.
(49) expect
 
CATEGORY V
 SYNTAX 
 COMPS S1 
CS EXPECT(EXP:X, THEME:F1 (. . . ))
By default linking, the Experiencer corresponds to Subject of expect. Our
current default linking rule does not accommodate the Experiencer role, so
we have to extend the rule slightly, so that either an Agent or an Experiencer
ranks higher than a Theme.
(50) LINKING ( DEFAULT ) 
  

 Agent/Experiencer 
 
 Subject 

⇓ ↔ ⇓

 Theme 
 
 Object 
We return to our main issue, that of control. There is a CS argument of

the infinitival complement that is not realized syntactically and is linked
to the argument that corresponds to the subject of expect. So far we have
accounted for this by designating the logical subject of F as ·, which would
indicate that it is bound by X· . For example, in the case of (40a) it would be
the Experiencer of win.
(51) EXPECT(EXP:X· ,THEME:WIN(EXP:·, . . . ))
Here, the EXP role of EXPECT and the EXP role of WIN have the same
index ·.
The problem with this approach is that we do not actually know which
argument of F corresponds to the syntactic subject, since F itself can corre-
spond to an active or a passive VP. If it is active, its logical subject will be
the unrealized subject, as in (52a). But if it is passive, its logical object will
be the unrealized subject, as in (52b).
(52) a. Sandy expects [VP to vote for George].

b. Sandy expects [VP to be elected].
In the first case, Sandy is the Agent of vote, and, in the second case, Sandy
is the Theme of elect. So we cannot say where this missing argument will be
in the thematic organization of F, the embedded predicate. All we can say is
that the missing argument is the argument that corresponds to the Subject
GF of the infinitival complement, which is not expressed syntactically.
So a second approximation would be to say in the lexical entry of expect
that the CS representation of the complement must contain an argument
that is coindexed with the controlling argument. In this case, we say that the
argument of the complement is bound by the controlling argument.
(53) EXPECT(EXP:X· ,THEME:F(. . . ,θ:·,. . . ))
But it is not sufficient to stipulate that there is a bound argument. We must

require in addition that this argument is actually mapped to the Subject
grammatical function associated with the complement. The following chart
shows the correspondences for expect.
(54) SYNTAX S
NP VP
V0 VP
expect V0 ...
to V
GFs Subject [Subject ]
CS EXPECT(EXP:X·, THEME:F(. . . , θ:·, . . .))
What this diagram says is that the bound argument that corresponds to
Subject of the infinitival complement is not realized syntactically, and the
relation that has the Theme role is realized as an infinitival VP.
We represent the properties of this variant of expect by expanding the
AVM in (48) as follows. The AVM has the same information as the diagram
in (54). The modification that we make is that we add an entry for the
Subject of the VP complement and assign an index to it that is the same as
the index of the bound argument in the corresponding CS representation,
in this case, ·3 .
(55) expect
  
CAT V

  
  SUBJECT [3] 
 SYNTAX  COMPS VP2 [SYNTAX 
  TENSE INF 
 
 
 SUBJECT [ 1] 
CS EXPECT(EXP:X1 · ,THEME:F2 (. . . , θ:·3 , . . . ))
A number of aspects of the syntactic configuration in which expect
appears are taken care of by default, although we have put the full infor-
mation into this AVM. First, the Experiencer role X1 maps by default into
the subject SUBJECT[1], by extension of the default linking rule introduced
earlier. Second, an infinitival VP in English maps into the form [VP [to
V] . . . ]. As the AVM shows, this VP is the syntactic realization of a CS
relation in which the argument that corresponds to Subject is bound to
the Experiencer. Finally, the Subject of an infinitival VP is not realized
syntactically, again by default.
These default realizations are part of the syntax of the language; they are
the properties that define how the language works in the typical case.
Default syntactic realizations for English
• The Agent or Experiencer corresponds to Subject (by the default linking

hierarchy).
• An infinitival complement is realized as [VP [to V] . . . ].
• The Subject of the infinitival complement is not realized overtly.
By assumption, a grammatical sentence is produced just in case all of

the conditions stated in the AVM (or, alternatively, represented in the
schematic diagram) are satisfied. If for some reason ·3 in the complement
is not mapped to Subject in a particular case, the structure fails to meet
the requirements of this AVM. This can happen if ·3 is a direct object
and the complement VP is not in the passive. Consider the following CS
representation.
(56) EXPECT(EXP:AL· ,THEME:ELECT(AGENT: PEOPLE, THEME:·))
The meaning represented here is that Al expects that the people will elect
him. If the complement is infinitival but not passive, then · corresponds to
the direct object of elect. This argument must be realized overtly.
(57) Al expects the people to elect ∗ (him).
The sentence without an overt object is ungrammatical because in English

only the subject of a non-finite complement is allowed to be both bound
and unrealized. A bound non-subject will be realized as a pronoun.
Consider next the AVM for try.
(58) try  

CATEGORY V

 SYNTAX  
  SUBJECT [3] 
 COMPS VP2 
 TENSE INF 
CS TRY(AGENT:X1 · , THEME:F2 (. . . , θ:·3 ))
Not surprisingly, the AVM for try is just like the one for the subject control
case of expect. The only difference between expect and try is that it is more
natural to embed the passive as the complement of expect that for try.

expected
(59) George to be elected by a huge majority.
?tried
The reason for the difference lies with the semantics of the two verbs.
Try presupposes that the unrealized argument of the complement is an
Agent that is responsible for the action expressed by the complement. This
produces a conflict when there is another Agent, either implicit or expressed
by a by-phrase, which happens when the complement is passive. A similar
conflict arises whenever the predicate denotes a state of affairs that cannot
be directly brought about by an Agent, e.g.

expected
(60) George to be have grandchildren.
?tried
Here is the AVM for appear

(61) appear
  
CATEGORY V

  
  SUBJECT [1] 
 SYNTAX  COMPS VP2 
 
 
 SUBJECT [1] 
CS APPEAR(F2 (. . . , θ:X1 , . . . ))
The argument of the infinitival complement that matches the subject GF of

the complement is realized as the subject of the verb appear, which gives rise
to the appearance of raising. Notice that this argument can be the subject
of a passive.
(62) George appears to have been elected by a huge majority.
Finally, consider the AVMs for persuade and believe.

(63) persuade
 
 
CATEGORY V

   
 
 SYNTAX  
SUBJECT [3]
 COMPS NP VP  
 
2
TENSE INF  
 
 
 OBJECT [1] 
CS PERSUADE(AGENT:X,PATIENT:Y1 · ,THEME:F2 , (. . . , θ:·3 , . . . ))
(64) believe
  
CATEGORY
V
  
  SUBJECT [1] 
 SYNTAX  COMPS NP VP2 
 
 
 GF2 [ 1] 
CS BELIEVE(EXP:X,THEME:F2 (. . . , θ:Y1 , . . .))
The AVM for persuade says that the Patient, Y1 , is realized as Object, and
that this argument binds the unrealized argument in the infinitival comple-
ment that corresponds to subject of the complement. The AVM for believe
says that the argument in the infinitival complement that corresponds to the
grammatical subject is realized as the Object of believe.
This concludes our summary of the various types of verbs that take
infinitival complements. In the next section, we consider how the properties
of these verbs are represented if we assume that the argument of the infini-
tival complement that corresponds to the subject grammatical function is
actually a syntactic subject, that is, if it is represented as a subject in the
syntactic configuration.
7.3. ∗ Raising as movement
In this section we look at the analysis of “raising” in terms of movement.
7.3.1. Move NP in MGG
Consider the construction that we have referred to as “raising to subject”.

(65) a. Robin appears to prefer yogurt.
b. It appears (that) Robin prefers yogurt.
The generalization that we observed in this case is that except for its superfi-
cial syntactic position, the subject of appear acts in every respect as though
it is the subject of the infinitival complement.
7.3. RAISING AS MOVEMENT 259
This generalization is captured in MGG (mainstream generative gram-

mar) by moving the subject from the infinitival complement of seem to the
subject of appear; hence it is called raising to subject. It is assumed that
appear has an empty subject as a lexical property. (We use S here instead of
IP/CP in order to simplify the diagrams.)
(66) S
[e] VP
V0 S
appear NP VP
Robin
If the complement is inflected with finite tense, then the subject remains in
the complement and the empty subject of appear is realized as it.
(67) S
it VP
V0 S
appear NP VP
Robin
But if the complement is an infinitival S, the subject of the complement must

raise to the higher subject position.
(68) S
[e] VP
V0 S
appear NP VP
Robin
∗
(69) It appears Robin to like yogurt.
The usual questions arise: Why does the subject move? What exactly goes
wrong if it doesn’t move when it should? Where exactly can it move to? How
far can it move? and so on.
In MGG there is a precedent for answering such questions, namely,
the derivational analysis of the passive construction in terms of case (see
Chapter 6, section 6.8). If we simply stipulate that appear does not assign
a case to the subject of the complement, then it will have to move in order
to have case assigned to it elsewhere in the structure. In fact, in MGG the
two constructions are accounted for by the same rule, move NP or simply
move ·. The prediction is that, whatever the limitations are on passive, the
same ones will hold for raising. We will see this very clearly in the next
section.
7.3.2. More raisings
Recall the MGG analysis of expect. The transitive verb expect assigns
accusative case to its NP complement. When the complement is infiniti-
val, expect “exceptionally” assigns accusative case to the subject of the
infinitive.
(70) a.
VP
case
V0 NP
expect
b.
VP
V0 case
IP
expect NP I
Now, what happens when a verb is passive? We have seen that it fails to
assign the objective case. This means that passivized (be) expected does not
assign case to its complement. Hence it does not exceptionally assign case
to the subject of the infinitival complement. (71) illustrates.
(71) a. VP
V0 NP
[PASSIVE]
no case
expected
b. VP
V0 IP
[PASSIVE]
NP I
expected
no case
So, the NPs in (70)–(71) do not have case assigned to them, and they must
move to a case position. Hence we will get the following derivations.
(72) a. [e] was expected [an explosion] → [an explosion] was expected ___
b. [e] was expected [IP [there] to be an explosion] →
[there] was expected [IP ___ to be an explosion]
This derivation is parallel to (73).
(73) [e] appeared [IP [there] to be an explosion] →

[there] appeared [IP ___ to be an explosion]
On the MGG analysis, these derivations are in fact identical in all respects
except for the reason why the subject of the complement fails to get case
assigned to it. In both cases, it can’t get case from the lower clause, because
the non-finite inflection cannot assign case to its subject. In the (71b)
construction, it cannot get case from the higher verb, because the verb
expected is passive, while in the raising case it is because appeared is not
a case-assigning verb.
Let’s return now to the questions raised in the preceding section.
(i) Why does the subject move?

(ii) What happens if it doesn’t move?
(iii) Where can it move to?
(iv) How far can it move?
Questions (i) and (ii) have already been answered: NP has to move because,
if it doesn’t, it does not have case properly assigned, and hence it violates
the requirement that case must be assigned to all referring NPs. (This
requirement is called the case filter.)
Why should there be such a requirement? The answer involves a some-
what complex chain of logic. First of all, we have observed that every NP
in a sentence has to have a θ-role assigned to it. Thus, for example, it is
impossible to have a direct object of an intransitive verb.
(74) a. Sandy fell hard.

b. ∗ Sandy fell the chair hard.
The requirement that all NPs have a θ-role is part of the theta criterion in
GB theory. If it is further assumed that an NP is not “visible” for θ-role
assignment unless it has case, it is possible to explain part of the case
filter in terms of the theta criterion. The solution takes advantage of the
fact that languages assign overt morphological case to thematic arguments.
Crucially, this reasoning is not based on overt morphological evidence in
a language like English or Chinese which lacks overt case morphology.
Nonetheless, the requirement that all NPs have case has been assumed in
much of MGG.
We state the case filter and the theta criterion informally as follows.
Case filter
An NP must have case assigned to it in order to be visible for θ-role
assignment.
Theta criterion
A referring NP must have a θ-role assigned to it.
Note that, on this analysis, the reason why the NP without case moves
is not necessarily the same as the reason why it cannot remain in place. It
may move because of the need to satisfy the EPP feature of the higher verb.
And, because it has moved, it is in a position where it gets case. If it does not
move, then the EPP feature of the higher verb can be satisfied by a dummy
NP. But the NP without case will still be in violation of the case filter. Hence
we have an account of the following.
∗
(75) It appears Robin to like yogurt.
It satisfies the EPP requirement of appears, but Robin lacks case owing to
the lexical properties of appears in this theory.
Question (iii) (where can the NP move to?) is answered in the same
way that it is answered in the case of the passive: the NP must move
to an available argument position. This much seems straightforward, if
we disallow in principle the creation of new structure through movement.
Question (iii) also has another component, though, which has to do with
the direction of movement. Can the NP move to the right? Can it move
down? There is no evidence to suggest that there are languages in which
downward movement is possible, and this fact has led to various constraints
on movement to rule out the nonexistent possibilities in principle. Most
notably, we see that the NP moves to a position that is higher than the
position that it moved from. Since this relation also plays a role in licens-
ing binding (see our discussion of control in this chapter and binding in
Chapter 10), the hypothesis has been explored that the relationship between
the moved NP and its original position can be reduced to other binding
relationships. 1
Question (iv) (how far can it move?) is more problematic. We can imagine
arbitrarily large structures in which there is an empty argument position
high up in the structure. For example, consider the following (be likely is a
raising predicate).
(76) [e] appears [S it to be likely [S George to win]]
In this example, it and George are not assigned case by appears and likely,
respectively. So, if one of them moves to [e], the other one can’t move. This
means that there will always be at least one NP that violates the case filter.
Since it does not bear a θ-role, it conceivably might be acceptable for it to
lack case, and therefore it would be acceptable for George to move to [e].
But it isn’t. The offending NP is in boldface in (77).
1
See Chomsky 1973.
∗
(77) George appears [S it to be likely [S ___ to win]]
This suggests that perhaps the case filter applies even to NPs that lack θ-
roles, and that the case filter cannot be reduced to the theta criterion. Or
there may be another reason why George cannot move to this position that
does not have to do with case.
In order to eliminate case as a possible account, let’s make sure that the
intervening dummy NP has case.
(78) [e] appears [CP that it is likely [IP George to win]]
There is nothing wrong with this configuration, and there is only one NP
that lacks case, George. So George should be able raise to [e]. But it cannot.
∗
(79) George appears that it is likely ___ to win.
Thus, it appears that raising, like control, must be local. Such an assumption
will rule out both of the illegitimate raisings seen here.
The most general way in which raising is guaranteed to be local in a
derivational theory is to require that all movements be local, which is a
central tenet of movement-based accounts. This line of development is
sketched out in Chapter 9, where we explore the development of locality
constraints on movement.
7.3.3. Interactions of raising, passive, and control
Raising, passive, and control interact in complex but entirely predictable

ways. It is helpful to visualize the interactions in terms of PRO and move-
ment, even if we do not want to assume PRO and movement as primitives
of our formal analysis. Consider the following structure.
(80) George tried [CP [e] to appear [IP [e] to be accepted PRO]]
Here, PRO becomes the subject of to be accepted, and then raises to become
the subject of to appear, where it is in a position to be controlled by
George.
(81) George tried [ PRO to appear [ to be accepted ]].
Now let’s try a complex one. To simplify the diagram, we use S instead of
IP/CP.
(82) S
NP VP
George V0 S
expects [e] VP
to be likely S
[e] VP
V0 VP
to be V0 NP S
persuaded Al [e] VP
V0 VP
to be V0 NP
shaved PRO
There are many movements that must apply here, as well as a control
relation. The NP PRO must move to the subject of to be shaved, where it
is controlled by Al. Al, in turn, must move to the subject of to be persuaded,
and then to the subject of to be likely. The result is
(83) George expects Al to be likely to be persuaded to be shaved.

S
NP VP
George V0 S
expects [e] VP
to be likely S
[e] VP
V0 VP
to be V0 NP S
persuaded Al [e] VP
V0 VP
to be V0 NP
shaved PRO
The sentence, while complex, is perfectly grammatical English.

Notice that in deriving this sentence, we have to apply the move-
ments and satisfy control from the bottom up. Alternatives are at
best more complex, and perhaps not even workable. The principle of
bottom-up application was originally called the cyclic principle; the rules
applied in order to the lowest S, then to the next higher S, and so on,
cycling up through the tree. In contemporary extensions of this prin-
ciple in the Minimalist Program (see Chomsky 1995), it is assumed
that the structure is put together word by word from the bottom-up
(or from the inside out), and that movements apply as soon as they
can; the result is essentially the same as bottom-up cycling through the
tree.
Complex constructions such as (83) can be formulated without movement
in terms of correspondences, as shown in (84). (We omit some details in the
CS representation to simplify the display.)
(84)
SYNTAX S
NP VP
George V1 NP VP
expects Al V AP
to be A VP
likely2 V VP
to be V VP
persuaded3 V VP
to be V
shaved4
GF Subject1 Object1 [Subject2 [Subject3 Object3 [Subject4 ]]]
CS EXPECT(EXP:GEORGE, LIKELY(PERSUADE(AGENT:X, PAT:AL·, SHAVE(AGENT:Z, PAT: · ))))
There are two passives, one “raising to subject” predicate and one “rais-
ing to object” predicate. Because shaved is a passive, ·, which is the
Patient of SHAVE, corresponds to the Subject GF of the lowest pred-
icate. · is bound by AL· , which is the Patient of PERSUADE. Since
7.4. SYNTACTIC CONFIGURATION AND CONTROL 267
persuaded is passive, AL· corresponds to the Subject of the persuaded

clause. Since persuaded is the infinitival complement of to be likely, which
is a “raising” predicate, its Subject corresponds to the Subject of to be
likely. But this clause is the infinitival complement of expects, which is
a “raising to object” verb. So this Subject is realized as the Object of
expects.
7.4. ∗ Syntactic configuration and control 2
7.4.1. Uniformity
The problem for syntactic theory is how to characterize the control rela-
tionship. It can be made part of the correspondence, as discussed in the
preceding sections, and tied to the lexical properties of the verbs. This is
the basic approach of HPSG (Head-driven Phrase Structure Grammar), on
which our treatment is based. Or it can be expressed in entirely configu-
rational terms, which is the traditional approach in mainstream generative
grammar (MGG). On this latter approach, the infinitival complement gets
the interpretation of a sentential complement because it is a sentential
complement.
The hypothesis that there is an invisible subject is compatible with a the-
ory that assumes UTAH (see Chapter 6, section 6.9). According to UTAH,
thematic relations always correspond to the same syntactic configurations.
If we require in addition that every thematic relation must be realized syn-
tactically (the theta criterion, Chapter 5, section 5.7.2), a thematic relation
that corresponds to overt subject must correspond to an invisible subject
when there is no visible constituent.
On this view, the role of the invisible subject is assigned to an NP
that is invisible, but is present in the syntactic structure. This NP is PRO.
Example (85) contains an illustration. The syntactic representation that we
use follows what is conventionally assumed in this approach, e.g. that the
sentential complement is a projection of I0 .
2
This section presupposes material discussed in Chapter 4, section 4.7.
(85) IP2
NP I
Georgei I0 VP2
V0 CP
tried Spec C
C0 IP1
NP I
PROi I0 VP1
V0
to run
The Agent role of run is assigned to PRO because it is the subject. The
control relation is not a binding relation between CS arguments but a
syntactic coindexing based on the configurational relation between the NP
George and PRO. We refer to this as syntactic control.
An immediate and important consequence of treating control as a syn-
tactic relation between a controller and an invisible PRO subject is that the
infinitival complement is sentential. So, if there is an overt subject, like Al, it
is reasonable to take it to be the subject of the infinitive, as in the following
example.
(86) George expects Al to win.
IP2
NP I
George I0 VP2
V0 CP
expects Spec C
C0 IP1
NP I
Al I0 VP1
V0
to win
In this case, Al gets the subject role of win, while George gets the subject role
of expects.
A syntactic theory of control raises a number of questions:
(i) What is the distribution of PRO?
(ii) What determines what the controller of PRO is in a given sentence?
(iii) What is the precise configurational nature of the relationship between PRO
and its controller?
(iv) How do we represent in the lexicon the difference between verbs that take
sentential complements with verbs that are inflected for tense and those that
take infinitival complements?
From the examples that we have seen, we know that PRO can be a subject,
but can it be anything other than a subject? Examples such as the following
show that it cannot. In each case, PRO is intended to be understood as
referring to the same individual as the coindexed controller.
(87) a. ∗ Georgei expects Al to defeat PROi . [meaning ‘George expects Al to defeat
him.’]
b. ∗ Georgei bought a book for PROi . [meaning ‘George bought a book for
himself.’]
c. ∗ Georgei claims that Susan insulted PROi . [meaning ‘George claims that
Susan insulted him.’]
Thus, it appears that PRO can only be a subject. But it cannot be the subject
of a verb that is inflected for tense.
(88) a. ∗ Georgei expects that PROi will win. [meaning ‘George expects that he
himself will win.’]
b. ∗ PROi forgives you. [meaning ‘He forgives you.’]
And PRO can start out as a non-subject as long as it moves into a subject
position.
(89) George expects PRO to be elected.
So, it appears that in English PRO (or less technically, the controlled
argument) can only correspond to the superficial subject of a non-finite
clause.
7.4.2. Case and PRO
To explain this distributional fact, GB theory sought an account of

PRO in terms of case. Recall from our earlier discussion of case and
the passive construction (Chapter 6, section 6.8) that an NP must have

case, and if it doesn’t, it must move to a position where it is assigned
case.
An early approach to PRO was to assume that it could only appear in
a position where case is not assigned. 3 Absence of case can follow in two
ways: (i) the phrasal head does not assign a case (as with an adjective or a
passive verb), or (ii) the position is one where the case assignment cannot
reach.
For example, a case-assigning head cannot normally assign case into the
phrasal projection of another head that also assigns case. So, for example,
a verb cannot assign accusative case to the subject of a tensed complement,
which is already assigned nominative within the complement.
(90) George expects [CP (that) [IP ∗him/he will win ] ]
ACC NOM
But we want expect to assign ACC to the subject of an infinitival comple-

ment.

him
(91) George expects ∗ to win.
he
In this case, the subject of the infinitive looks like the object of expect even
though (by assumption) it isn’t.
Considerations of this sort led to a series of theoretical proposals that
sought to reduce the distributional properties of case to a general structural
relationship between the case assigner and the assignee. In GB Theory and
the Barriers framework of Chomsky 1986, the central relationship is one of
government. Intuitively, a head X0 governs its arguments (and everything
else in its maximal projection XP), and is blocked from governing the argu-
ments of another governor. One governor establishes a barrier that keeps
out other governors. In (90), for example, the governor in the complement
that assigns nominative case to he blocks expects from assigning accusative
case to this position.
3
A theoretical argument that PRO must have this property was based on the
theory of binding, which we review in Chapter 10.

him
Now consider George expects ∗ to win, in which him has accusative
he
case. Since nominative case is not assigned in the infinitival, and accusative
case is, it is reasonable to hypothesize that the governor that creates the
barrier in (90) is in fact finite tense. Non-finite tense does not create such a
barrier, since it is not a case assigner.
If PRO does not have case assigned to it, then it must appear in a different
configuration from the overt subject of the infinitive that does get case.
Compare
(92) a. George expects [Al to win]

b. George expects [PRO to win]
Since both complements are infinitives, appeal to the presence or absence of

a case assigner will not suffice. If both subjects, Al and PRO, bear the same
syntactic relationship to expect, then either both will get case assigned by
expect, or neither will.
A hypothesis that proved to be very influential in this regard is that certain
projections, in particular CP and DP (see Chapter 4, section 4.7), also block
government from the outside. On this view, the infinitival complement that
has PRO is a CP, so that PRO is blocked from getting case, as illustrated
in (93). But when we have the overt NP, the complement is simply an IP, as
in (94).
(93) IP2
NP I
George I0 VP2
V0 CP
expects Spec C
case
C0 IP1
NP I
PRO I0 VP1
no case V
to win
(94) IP2
NP I
George I0 VP2
V0 IP1
expects NP I
case
Al I0 VP1
V0
to win
The configuration in (94), where the subject of the complement is assigned

case by the higher verb, is sometimes called exceptional case marking (ECM)
in MGG, because it involves assignment of case out of one clause into
another.
More recent analyses have demonstrated that in a theory that assumes
PRO, PRO must have case, just like any other NP. One reason is that direct
object PRO must undergo movement in the passive. To illustrate:
(95) George expects [PRO to be elected __ ]
The object position of to be elected is not a case position. If PRO did not
require case, then it would not have to move. But expects requires that
the subject of its complement be a full NP or PRO, so PRO has to move.
Treating the subject of the complement as a case position (as just discussed
in connection with ECM), regularizes the treatment of PRO to that of overt
NPs. An alternative motivation for PRO to have case is the assumption
in GB theory that an NP must have case in order to be assigned a θ-role
(Chomsky 1981).
As noted in Chapter 6, section 6.9, there have been more recent proposals
to the effect that the movement of an NP to subject position in English is
not triggered by case but by another feature, called the EPP feature. This
feature on the head of IP requires that there be an NP in its Spec. We will
not work through the logic that motivates this assumption, but simply note
it as a problem for the view that PRO lacks case. 4
4
For additional discussion, see Martin 2001.
As even this brief summary shows, the assumption that an apparently

simple structure, such as V-VP, has an abstract syntactic structure that is
essentially that of other sentential constructions produces very rich syntac-
tic consequences, and raises complex technical questions that are not easily
resolved.
To take another set of questions, if the complement of expects in (93) is
a CP, it must be explained why the head C0 does not appear, at least in the
standard dialect of English.
(96) George expects (∗ for) to win.
At the same time, if Al to win in (92a) is a constituent, we would expect

constituency tests to provide evidence that it is; however, constituency tests
seem to show that the string Al to win is not a constituent.
 
 (that) Al will win 
(97) a. I expect very much ∗ Al to win .
 for Al to win 
 
b. What I expect is ?Al to win .
 for Al to win 
 
 That Al will win 
c. ∗ , I very much expect.
Al to win
 For Al to win 
 
d. What do you expect? – ∗ Al to win .
 For Al to win 
 

 That Al will win 

∗
Al will win
e. ∗ is (un)expected.

 Al to win 

For Al to win
For each distinct example in which Al to win fails to function as a con-

stituent, it is necessary to construct an extension to the theory that will
account for the failure. For instance, for the examples in (97) we might
pursue a theory in which Al fails to get case because it is not immediately
adjacent to the case assigner expect. This might explain why the presence
of for usually (although not always) makes the infinitive grammatical –
for assigns case to Al. Such a theory must take into account the fact that
adjacency is not a problem for other NPs that are arguments of the case
assigner, as in
(98) a. Him, I really respect ___.

b. We were discussing ___ very enthusiastically [the things about the theory
that made us proudest].
7.4.3. Identifying the controller: c-command and MDP
What we have established in the preceding sections is when there is a control

relation, the infinitival VP must be linked to a CS representation in which
the controlled argument is represented. We have considered two formal
devices in which to express this relationship:
r a correspondence rule directly linking the infinitival VP syntax to a CS-
representation that expresses the argument corresponding to the subject;
r a syntactic representation in which there is an invisible PRO subject, which is
linked to the CS argument.
While these approaches are formally different, they share a crucial fea-
ture. What represents the controlled subject must be linked to some other
NP (or its corresponding CS-representation), the controller. 5 This is a
type of binding. Binding occurs when the interpretation of one expression
depends on the interpretation of another, its antecedent. For example, in
control, the controlled subject refers to the same thing as the controller
does. If the controller refers to a quantified set of individuals, then so does
the controlled argument.
(99) Every candidate expects to win.
In this example, each candidate holds the view that he or she will win, that
is
(100) every candidate x expects that x will win
One component of the syntactic theory of control, of which MGG is

an important exemplar, is an account of what the possible controllers are
in a sentence. 6 Consider a sentence like (101), using a representation in
5
A controller is a special case of an antecedent, to be discussed at greater length
in Chapter 11.
6
The other major syntactic treatment of control is that of Head-driven Phrase
Structure Grammar (HPSG); see Pollard and Sag 1994. The HPSG account does
not assume PRO, but a direct linking of the VP to the relevant portions of the CS
representation. A similar account is found in Lexical Functional Grammar (LFG);
see Asudeh 2005.
terms of PRO. The impossibility of a control relationship is indicated by

putting an asterisk of an impossible controller on the index attached to
PRO.

persuaded
(101) Georgei Alj [S PRO∗ i,j to step aside].
forced
We know that Georgei cannot be the controller of PRO; only Al j can be. But
how do we know this? Before we try to answer this question, let us consider
a few other examples that shed additional light on the syntactic relationship
between the controller and PRO.
(102) a. Georgei ’s motherj expected [PRO∗ i,j to win].
b. Al persuaded Georgei ’s motherj [PRO∗ i,j to bake some cookies].
∗
c. We visited Georgei and PROi was delighted to see us.
d. [PRO to err] is human.
Let us look first at the structure of (102a). (In order to simplify the tree
diagrams, we use S here instead of IP/CP.)
(103) S2
NPj VP2
NPi ’s N0j V0 S1
George mother expected NP NP1
PRO V0
to win
We see here that the controller, George’s motherj , is a sister of a node VP2
that dominates PRO, while the NP that cannot be the controller, Georgei ,is
not. The formal relationship between PRO and the NP George’s mother is
called c-command. We define it as follows.
C-command
A c-commands B if A is a sister of B or of a node that dominates B.
Only the controller c-commands PRO in this structure. The same difference
in c-command holds for (102b,c) – Exercise 8 asks you to show that this is
the case.
A syntactic theory of control assuming PRO would then have as a con-

dition that the controller of PRO must be an NP that c-commands it. The
structure of (101) shows two NPs that c-command PRO, George or Al.
(104) S2
NPi VP2
George V0 NPj S
persuaded
forced Al NP VP1
PRO V0 Adv
to step aside
But the interpretation of example (101) shows that only one of these
two NPs can be the controller. Continuing to assume an account of con-
trol based on syntactic configuration, we can distinguish between George
and Al by observing that while George c-commands Al and PRO, Al c-
commands PRO but not George. That is, although both c-command PRO,
Al is closer to PRO than George is. A reasonable hypothesis, then, is that
when there is more than one c-commanding potential antecedent, the clos-
est one is the actual antecedent. This hypothesis is known as the Mini-
mum Distance Principle (MDP), 7 and is widely (although not universally)
accepted.
Minimal Distance Principle

The controller of PRO is the closest potential antecedent that c-
commands it.
When PRO lacks a c-commanding antecedent there is no controller.

Example (102d) shows that in this case PRO is interpreted as referring to
an arbitrary individual. In this case PRO is called PROarb .
7
The MDP was first proposed by Rosenbaum 1967.
7.4.4. Problems with MDP
Additional examples suggest that the conditions on control are somewhat

more restrictive than the MDP, and also raise questions about whether a
strictly configurational account is correct. Consider first the following.
(105) George wants [S it to be difficult [S PROarb to vote twice]].
In this example there are two NPs that c-command PRO, George and it.
Since it is a dummy NP, we might take the view that it is not a possible
antecedent and that it is therefore ineligible as a controller. If so, then
there is a problem with MDP. George is the closest eligible controller, but
is not the controller of PRO, which has an arbitrary interpretation: the
sentence means “George wants it to be difficult for anyone to vote twice”
not “George wants it to be difficult for him to vote twice”.
So we might conclude that dummy it is a possible controller, and that it
produces arbitrary control because it has no reference. But a paraphrase of
this sentence without it also has arbitrary control.
(106) George wants [S [S PROarb to vote twice] to be difficult]
Here, [PROarb to vote twice] is the subject of to be difficult. The sentence is

somewhat awkward, but can be made more acceptable with the proper into-
nation – lengthening of the space between wants and to prevents the hearer
from associating to with wants. It can also be paraphrased as George wants
[[PROarb voting twice] to be difficult], which is also a case of arbitrary
control.
Let us look at the structure of (106).
(107) S3
NP VP
George V0 S2
wants S1 VP
NP VP to be difficult
PRO to vote twice
Compare this structure to that of subject control, as in (103). It appears

that we have a control relationship only when the controller is an argument
of the verb whose complement contains PRO as a subject. For example, in

(103), George’s mother is the subject of expects, and PRO is the subject of
the complement of expects. In (107), by contrast, George is an argument of
wants but PRO is not a subject of the complement of wants. MDP and c-
command do not correctly account for this case. By comparison, stating
control in terms of the lexical entry for want captures this relationship
directly, and permits no other possibilities.
Next, we note a well-known counterexample to MDP, involving promise.
(108) Ali promised Georgej [PROi,∗ j not to run again].
Here, contrary to expectations, the controller is not the closest NP, Georgej ,
but a more distant NP, Al i . This violates the MDP. Since this counterex-
ample is associated with a particular verb, it suggests again that the control
relation is represented as part of the lexical entry of the verb and not in
terms of c-command and the MDP.
Finally, there are cases of control where the controller does not appear to
c-command PRO at all.
(109) a. Georgei counted on Alj PRO∗ i,j to step aside.

b. Wei pleaded for several hours with Sandyj PRO∗ i,j to turn down the
stereo.
In cases such as these, the controller is the complement of a preposition.

Because of the branching structure of the PP, the controller does not actu-
ally c-command PRO. (Exercise 10 asks you to show this.)
What we see, then, is that the configurational theory of control is able
to represent the control relationship. But, if it is stated simply in terms of
c-command, it predicts that it will be more general than it actually turns
out to be in some respects, and more restricted in others. It can be further
restricted by the MDP, but the MDP is not sufficiently restrictive. The MDP
does not account for the fact that control is a local relation between a verb
and the subject of its complement. Counterexamples involving promise and
rely on/plead with show that at best the configurational account of control
requires additional modifications and stipulations.
7.4.5. The lexical representation of control
The fact that the complement containing PRO and the controller must be
locally related suggests that the relationship is in fact a lexical one and not a
configurational one. Recall the AVM representation of subject control. (48 )

is an elaboration of our earlier representation of expect.
(48 )
expect
  
CATEGORY V

  
  SUBJECT [3] 
 SYNTAX  COMPS VP2 [SYNTAX 
 
 
 SUBJECT [1] 
CS EXPECT(EXP:X1 · ,THEME:F2 (. . . , θ:·3 , . . . )
We are able to refer in this representation to the subject of expect X1 · and

to the coindexed ·3 subject of the complement. But we are not able to refer
to any other more distant arguments, because these are not available to us
in the lexical representation. In other words, the locality of control follows
directly from the fact that a lexical entry can only specify properties of its
arguments and complements, nothing else. 8
A lexical account of control is able to accommodate verbs like promise by
simply stating that it is the promiser that is committed to carrying out the
promised action, as part of the meaning of the verb. See the AVM in (110).
(110) promise
  
CATEGORY V

  
  SUBJECT [3] 
  COMPS (NP ) VP [SYNTAX 
 SYNTAX 

4 2
TENSE INF 

  

  SUBJECT [ 1] 
 
 OBJECT [4] 
CS PROMISE(AGENT:X1 · ,THEME:F2 (. . . , θ:·3 , . . . ),GOAL:Y4 )
This AVM says that promise is a subject control verb – the controller X1 ·
is coindexed with argument ·3 in the complement, which is linked to the
subject role.
It is possible to account for the control properties of many verbs in terms
of their conceptual structure in a very general way by recognizing that
8
The early literature on control referred to “long distance control”. An exam-
ple is
(i) George thinks [it would be ill-advised [PRO to run again]].
where the natural conclusion is that PRO is George. However, this is really a case of
arbitrary control in which there is an inference made that PRO refers to the same
individual as the distant argument. We can make this inference less plausible by
changing the context or the adjective.
(ii) I said I wanted to run again but George thinks [it would be ill-advised [PRO to
run again]].
implicit in verbs like want, force, persuade, intend, try, promise, and so on
are primitive semantic relations in which the intention to perform an action
or the obligation for performing it rests on a particular individual. 9 The
individual who has the intention or obligation is the controller, and the
individual who carries out the intention or obligation is expressed by the
controlled argument. For example, with promise the promiser undertakes
an obligation, hence there is subject control, while in persuade the persuader
places an obligation on someone else.
Finally, formulating matters in semantic terms allows us to sidestep the
problem raised by plead with, where the controller apparently does not
c-command PRO. The key here is that the preposition is not part of the CS
representation but is simply part of the syntactic realization of the oblique
argument of plead. We show this in the AVM in (111), and in the related
schematic in (112).
(111) plead with

  
CATEGORY V
      
  VP2 
   PP5    
 SYNTAX  COMPS     
  SUBJECT [3] 
  [with [4]] SYNTAX 
 
 
 SUBJECT [1] 
CS PLEAD(AGENT:X ,PATIENT:Y · , THEME:F (θ:· , . . .))
1 4 2 3
(112)
SYNTAX S
NP VP2
V0 PP VP1
plead P0 NP V0 ...
with
GF Subject [Subject ]
CS PLEAD(AGENT:X, PATIENT:Y·, THEME:F(θ:·, . . .)
9
See Culicover and Jackendoff 2005:Chapter 10 for extended discussion of the
semantic basis of control.
EXERCISES 281
To conclude this discussion of control, let’s look again at the questions

that we raised in section 7.4.1. We restate them more neutrally by not
presupposing an analysis in terms of PRO.
(i) What is the distribution of subjectless clauses?
(ii) What determines the controller in a given sentence?
(iii) What is the nature of the relationship between the unexpressed subject and its
controller?
(iv) How do we represent the difference in the lexicon between verbs that take
sentential complements with inflected verbs and those that take infinitival
complements?
Recall that in section 7.4.2 we worked out an answer to question (i)

in terms of case assignment and PRO. Problem 10 asks you to consider
how a lexical theory of control can account for this distribution. Regarding
question (ii), we have explored answers in terms of c-command and MDP
and in terms of the lexicon.
Question (iii) gets a different answer depending on what theory of control
we adopt. If it is a configurational theory of control, then the relationship
between PRO and its controller is a special case of binding. In MGG it has
been customary to describe binding relationships in syntactic terms and
in Chapter 10 we will take a closer look at how this has been done. In
the account given in terms of correspondences, the fundamental binding
relation holds between CS arguments, not NPs; this relation is reflected in
the infinitival structure and the absence of an overt NP corresponding to
subject.
The answer to question (iv) is straightforward if we take a lexical
approach to control. However, if we adopt a configurational approach to
control, then the special properties of verbs with respect to control become
more problematic. You are asked to look further into this question in
Problem 11.
Exercises
1. The following rule introduces infinitival VP into a larger VP in English.

NP
(1) VP→ V (NP)
VPINF
On the other hand, there is also a rule that accounts for the fact that PP,
Adv, and S may appear at the right edge of a VP.
(2) VP → V (NP) (NP) (PP∗ ) (Adv∗ ) (S)
First, how would you change rule (1) so that it incorporates the information
that is captured in rule (2)? Second, what claims does this new rule make
about the distribution of PP, Adv, and VP-final S in VPs that contain an
infinitival complement? Are these claims factually correct? Give examples
to support your answer.
[§7.1.]
2. Verify that the complement possibilities for the verbs promise, manage,
tell, and forget are what (6) in the text says they are.
[§7.1.]
3. For each of the following verbs, say whether it is an object control verb
or a raising to object verb. Justify your answer in terms of the thematic
structure and the possibility of having dummy NPs in the object position.
(1) a. force f. ask

b. believe g. declare
c. pressure h. judge
d. tell i. discover
e. instruct j. caution
[§7.1.]
4. Show that when it takes an NP and VP complement, expect is a raising
to object verb, not an object control verb. (Hint: Use the “Tests for raising”
summary.)
[§7.1.5.]
5. Categorize each of the following verbs as (a) subject control, (b) object
control, (c) no control, (d) raising to subject. (Hint: Use dummy subjects to
justify your answer.)
(1) a. make f. require

b. assume g. aspire
c. convince h. begin
d. rely on i. consider
e. compel j. wish
[§7.1.7.]
6. Using the AVMs in section 7.2 work out the correspondences for each
of the following sentences. Sentence (0) is worked out as an illustration.
EXERCISES 283
(0) George expects to win.

SYNTAX S
NP VP
George V0 VP
expects V0
to win
GFs Subject [Subject ]
CS EXPECT(EXP:GEORGE·, THEME:WIN(EXP:·)
(1) a. Sandy believes Kim to be a genius.

b. Kim persuaded Sandy to wash the floors.
c. Robin expected to be nominated.
d. Kim seems to have been attacked by the ducks.
e. Leslie tried to be nice to Robin.
[§7.2.]
7. List all of the c-command relations in the following tree.
(1) A
B C
D
E F G
H I J K
[§7.4.3.]
8. In the following examples, demonstrate that the controllers c-command

PRO and obey the Minimum Distance Principle, and the non-controllers do
not.
(1) a. Sandyi wants PROi to win.

b. Sandyi ’s motherj expected PRO∗ i,j to win.
c. Wei persuaded [some friends of Sandyj ]k PRO∗ i,∗ j,k to call.
[§7.4.3.]
9. Show that both George and Al c-command PRO in the following tree.
(1) IP2
NP2 I
George I0 VP2
V NPj CP1
persuaded Al Spec C
forced
C0 IP1
NP I
PRO I0 VP1
V Adv
to step aside
[§7.4.3.]
10. In the text it was claimed that in sentences like
(1) We pleaded for several hours with Sandy to turn down the stereo.
the controller Sandy does not c-command PRO. Draw the tree and show
that this is the case.
[§7.4.4.]
∗
11. Assuming a movement analysis of passive and raising, show the
derivation of each of the following sentences. Start with the lowest S, and
work your way up. Be sure to distinguish between those sentences in which
what is moving is an NP that controls PRO, and raising. Number and label
each movement and identify the control relations through coindexing.
(1) a. Al was made to suffer.
b. We were persuaded to eat the cookies.
c. The patient was expected to be examined by a specialist.
d. The patient was persuaded to be examined by a specialist.
e. The kids seem to have been persuaded to eat the cookies.
f. The kids seem to have been expected to eat the cookies.
(2) a. Sandy was persuaded to run for president.
b. Sandy was expected to run for president.
c. Sandy was expected to be persuaded to run for president.
PROBLEMS 285
(3) a. There was expected to be a problem.

b. There seems to have been expected to be a problem.
If you can, use bracketed strings, not trees.

[§7.4.]
∗
12. Show the correspondences for the sentences in Exercise 11 without
assuming movement. To do this, you must be explicit about the CS repre-
sentation, the correspondence between each CS argument and the GFs, and
the correspondence between the GFs and the syntactic structure.
[§7.4.]
Problems
1. In Chapter 6 we considered the Japanese causative. Here are some

examples.
(1) Japanese
a. Suzuki-san-wa musume-ni daigaku-e ik-ase-ta
Suzuki-Mr.-TOP daughter-DAT college-to go-CAUSE - PAST
‘Mr. Suzuki made his daughter go to college.’
b. Chichi-wa imooto-ni piano-o naraw-ase-ta
father-TOP younger sister-DAT piano-ACC learn to play-CAUSE - PAST
‘Father made younger sister learn to play the piano.’
Formulate a general correspondence to account for the correspondence

illustrated by these examples. Assume that the CS representation of
causation is CAUSE(AGENT/INSTR:X,THEME:F(. . . )), where Agent is
the role of an animate cause and Instrument the role of an inanimate
cause.
[§7.2.]
2. Give the lexical entries for the English causative alternation melt as in
(1) a. The ice melted.

b. The sun melted the ice.
Assume that the CS representation of causation is CAUSE(AGENT/

INSTR:X,THEME:F(. . . )), where Agent is the role of an animate cause
and Instrument the role of an inanimate cause.
[§7.2.]
3. In the text we focus on cases of control where the complement is

infinitival. The following examples show that a gerundive (a phrase headed
by V+ing) may be controlled.

into
(2) a. I talked Terry taking a picture of me.
out of
b. Terry considered spending the night at a local hotel.
c. I began speaking German to the children when they were young.
# $
Formulate the AVM for the lexical entries of talk intoout of , consider, and
begin that will account for the fact that they can appear in this construc-
tion. Be sure to specify the morphological form of the complement and to
represent the control relation.
[§7.2.]
4. Work out the MGG movement analysis of

(1) Robin was said to be a genius.
What do the following sentences suggest about this analysis?

∗
(2) a. Everyone said Robin to be a genius.
b. Everyone said Robin was a genius.
c. It was said that Robin was a genius.
(3) a. No one said anything.
b. Nothing was said by anyone.
[§7.3.3.]
5. How is the data in Problem 4 handled in an account that does not
assume movement? (Hint: To answer this question you will need to formu-
late a lexical entry for say that gets its behavior in the passive just right.)
[§7.3.3.]
6. Work out a MGG analysis of seems like/as if assuming that this
construction is a variant of the raising to subject analysis for seem to. What
problems/issues arise on this approach?

like
(1) a. It seems John really likes yogurt.
as if

like
b. Johni seems hei really likes yogurt.
as if

like
(2) a. It seems there would be a problem with this analysis.
as if

like
b. ∗ There seems there would be a problem with this analysis.
as if
PROBLEMS 287

like
(3) a. It seems everyone likes Maryi .
as if

like
b. Mary seems everyone likes heri .
as if
[§7.3.3.]
7. Formulate a non-movement alternative for the data in Problem 6. To

do this, you will have to specify the selectional properties of seems, the CS
representations, and the correspondences.
[§7.3.3.]
8. Suppose that we assume a syntactic account of control, in which PRO

is bound by a controller that c-commands it. It is necessary to guarantee
that verbs will have the particular control relations that we have observed.
State the c-selectional properties of expect, try, believe, and persuade
that will guarantee that they will have these control properties. Be clear
about any syntactic assumptions that you make regarding the distribution
of PRO.
[§7.4.3.]
9. In (97) in the text we gave a number of ungrammatical examples of NP-

VPINF that suggest that this sequence is not a constituent. We suggested that
it might be possible to maintain the assumption that it is a constituent and
at the same time explain these examples by appealing to the requirement
that a referring NP must have case assigned to it. What would the syntactic
conditions be for the assignment of case that would correctly predict that
these examples are ungrammatical, but that standard ECM and topicaliza-
tion examples are not problematic?
[§7.4.4.]
∗
10. We have seen that in an account of control where PRO is a syntactic
empty subject, PRO can only be the subject of a non-finite clause. But
in a lexical account of control there is no PRO. State as precisely as you
can the realization rule for the non-finite clause that will guarantee that it
lacks an overt subject on a lexical approach where there is no PRO. The
correspondence shown in (112) in the text is an example. (Hint: There are
two components to this realization that must be taken into account: the GF
of the argument and its CS representation.)
[§7.4.4.]
11. State as precisely as you can the lexical entries for the verbs try, believe,
forget, and expect in order to express the types of complements that each
may take.
[§7.4.5.]
Research questions
1. In a sentence with expletive there in subject position, the verb appears

to agree not with the subject but with the NP that follows it.

is
(1) a. There ∗ a lion in the closet.
are
∗
is
b. There lions in the closet.
are
We show in this chapter that there is a construction in English in which the
subject of an infinitival complement is realized as the subject of a higher
verb (“raising to subject”). In the case where the subject is there, the higher
verb (seem or seems in the following examples) appears to agree with the
NP following the form of be in the infinitive.
∗
seem
(2) a. There very clearly to be a lion in the closet.
seems

seem
b. There very clearly ∗ to be lions in the closet.
seems
How is this pattern to be accounted for? What assumptions do you have
to make about how agreement works in English in general and how it
works for there-sentences in particular. Does this phenomenon require a
movement analysis of raising, or can the data be accounted for without
movement?
[§7.3.]
2. Some languages use finite constructions to do the work that infinitivals
do in English, German, and Italian. Here is a Greek example.
(3) Greek
O Petros kseri na kolimbai.
the Peter-NOM knows-3. SG to swim-3. SG
‘Peter knows how to swim.’
[Alexiadou and Anagnostopoulou 1999]
The embedded verb in this case is in the subjunctive mood, and agrees in
person and number with the subject of the higher verb. Thus, it appears to
be a full S with a subject, although the subject is not visible. Since it is a

finite clause, it cannot contain PRO, which only occurs in infinitives. One
possibility that has been considered in the literature is that the subject in
this case is an invisible pronoun pro.
Greek also has a construction comparable to English ECM.
(4) Greek
O »eo‰orakis θeli ti Faranturi na tra„u‰ai
the Theodorakis-NOM want-3. SG the Faranturi-ACC that sing-3. SG
mono dika tu tra„u‰ja
only own-his songs-ACC
‘Theodorakis wants Faranturi to sing only his songs.’
(5) ‰en perimena ti Marina na „rapsi toso asxima sto
NEG expected-1. SG the Marina-ACC that write-3. SG that badly in
djagonisma tis fisikis.
exam the physics-GEN
‘I did not expect Marina to do so badly in the physics exam.’
[Kotzoglou 2002,6:40]
Notice that in these examples what we understand as bearing the subject

role of the complement is marked with accusative case, presumably by the
higher verb. At the same time, the complement is finite and agrees with this
NP, but has no overt subject. Again, it is natural to analyze this in terms of
a pro subject that is somehow linked to the accusative-marked NP.
Greek appears to have the kind of raising that English has, in the sense
that the raising verb agrees with the subject of the complement, e.g.
(6) The children seem/∗ seems to be friends.
However, in the Greek case, the verb in the complement also agrees with the
subject, even when the subject is part of an idiomatic expression.
(7) a. stamatisan/arxisan na mou benun psili

stopped.3 PL/started.3 PL that 1 SG. DAT enter.3 PL fleas.NOM . PL
st’aftia
in the ears
‘I stopped being/started becoming suspicious.’ (lit.: ‘Fleas stopped/started
entering my ears.’)
b. arxizoun na mou anavoun ta labakia
start.3 PL that 1SG. DAT light up.3 PL the lamps
‘I am beginning to get pissed off.’ (lit.: ‘My lamps start lighting up.’)
[Polinsky and Potsdam 2006]
Show that Greek raising is problematic both for an account of raising that
treats the subject of the higher verb as originating as the subject of the lower
verb, and for an account that relates the subject of the higher verb to the
complement through a correspondence.
[§7.3.]
3. Data from German suggests that there is a division among infinitival
VP complements, depending on the higher verb. Some infinitival comple-
ments (the coherent infinitives) act as though their verbs are part of the
higher sentence, while others (the non-coherent infinitives) act as though
they are distinct complements. Among the verbs that take the coherent
construction are versuchen (“try”), beginnen (“begin”), anfangen (“start”),
planen (“plan”), erlauben (“allow”), glauben (“believe”), gestatten (“allow”),
beabsichtigen (“intend”), erwägen (“consider”), and vergessen (“forget”).
The clearest difference between the two types of infinitive is that adverbs
and negation can appear between the verbs in a non-coherent construction
but not in a coherent construction. (In thinking about these examples,
keep in mind that German is V-final in subordinate clauses and that the
main verb appears in second position in a main clause. Pronominal objects
appear on the left edge of the VP.)
(1) a. (Er sagte) dass er sie zu kennen (nicht) bedauerte

(he said) that he her to know (not) regretted
‘. . . that he did (not) regret knowing her’
b. (Er sagte) dass er sie zu kennen (∗ nicht) schien
(he said) that he her to know (∗ not) seemed
‘. . . that he did (not) seem to know her’
[Wöllstein-Leisten and Heilmann 1997:5]
c. (Er sagte) dass er sie nicht zu kennen schien
(he said) that he her not to know seemed
‘that he didn’t seem to know her/that he seemed not to know her’
This difference itself suggests that in the coherent infinitive the complement
verb forms a unit with the higher verb but in the incoherent infinitive it does
not.
Show what the two syntactic structures are for the two constructions, and
show how the two structures correspond to the CS representations. What is
crucial is that if kennen schien is a unit, the object of kennen (in this case sie)
must still correspond to a thematic argument of the corresponding relation
in CS.
[§7.4.]
4. (This problem assumes that you have worked out an analysis of Ger-
man coherent and incoherent infinitives (Research question 3).)
A. In the coherent infinitive the entire verbal complex can be topicalized,
but not in the non-coherent infinitive.
(2) a. [Rasieren zu müssen geglaubt] hat sich Max noch nie
shave to must believed has himself Max still never
‘Max has never believed he had to shave.’
b. ∗ [Rasieren zu müssen bedauert] hat sich Max noch nie
shave to must regretted has himself Max still never
‘Max has never regretted having to shave.’
[Wöllstein-Leisten and Heilmann 1997:8, 9]
Explain how this difference follows from your analysis. Assume that topi-
calized material appears in the Subject position and that an untopicalized
logical subject is in VP.
B. The complement in a non-coherent infinitive construction can be
extraposed, but not in a coherent construction.
(3) a. ∗ dass Maria scheint, alle Verwandten zu kennen
that Maria seems, all relatives to know
‘that Maria seems to know all the relatives’
b. dass Maria prahlt, alle Verwandten zu kennen
that Maria boasts, all relatives to know
‘that Maria boasts of knowing all the relatives’
Explain how this difference follows from your analysis.

C. Only coherent verbs allow for the so-called Third Construction, in
which the two verbs are transposed.
(4) a. dass Hans das Auto zu reparieren versucht [normal order]
that Hans the car to repair tried
‘that Hans tried to repair the car’
b. dass Hans das Auto versucht zu reparieren [Third Construction]
that Hans the car tried to repair
‘that Hans tried to repair the car’
[Rambow 2003:5]
c. dass Hans das Auto zu reparieren bedauterte [normal order]
that Hans the car to repair regretted
‘that Hans regretted repairing the car’
d. ∗ dass Hans das Auto bedauerte zu reparieren
that Hans the car regretted to repair
Formulate the reordering rule in such a way that it accounts for this differ-
ence.
[§7.4.]
5. In Italian, pronominal arguments are expressed by clitics attached to

the verb. Normally a clitic attaches to the verb of which it is an argument.
But under certain circumstances, a clitic can attach to the higher verb (such
as a form of volere “want”), that is, to the verb that takes a complement that
contains the clitic. Here are some examples.
(1) Ci vengo con Maria
there go-1. SG with M.
‘I go there with Maria.’
(2) a. Vorrei andarci con Maria.
would.want-1. SG to.go-there with M.
‘I would like to go there with Maria.’
b. Ci vorrei andare con Maria.
there would.want-1. SG to.go with M.
‘I would like to go there with Maria.’
(3) a. Detesterei andarci con Maria.
would.detest-1. SG to.go-there with M.
‘I would hate to go there with Maria.’
∗
b. Ci detesterei andare con Maria.
there would.detest-1. SG to.go with M.
‘I would hate to go there with Maria.’
[Cardinaletti and Shlonsky 2004:521]
An interesting fact about this clitic climbing is the fact that, in Italian,
certain verbs, particularly verbs of motion, form the past construction using
the verb essere “to be” rather than avere “to have”, e.g.

Sono
(4) a. ∗ andato
Ho

be-1.SG
go- PAST. PRT.
have-1.SG
‘I went.’
∗
Sono
b. telefonato
Ho

be-1.SG
have-1.SG telephone-PAST. PRT.
‘I telephoned.’
Verbs that permit clitic climbing can form the past construction with essere
when the verb that heads the complement is a verb that takes essere.

Avrei
(5) a. voluto andarci con Maria.
?Sarei

would.have-1.SG
wanted to.go-there with Maria
would.be-1.SG
‘I would have liked to go there with Maria.’

Avrei
b. ∗ detestato andarci con Maria.
Sarei

would.have-1.SG
detested to.go-there with Maria
would.be-1.SG
‘I would have hated to go there with Maria.’
∗
Sono
(6) a. voluto questo.
Ho

be-1.SG
wanted this
have-1.SG
‘I wanted this.’
∗
sono
b. Non voluto!
ho

be-1.SG
not wanted
have-1.SG
‘I didn’t want to!’
[Cardinaletti and Shlonsky 2004:521]
A. Show how a restructuring analysis, in which a composite verb is

created from a control configuration, accounts for this data. State the
restructuring rule explicitly. Be precise about the relationship between the
syntactic properties of the lower verb and the syntactic properties of the
composite verb.
B. This data may also be taken as evidence that there are two ways in
which the CS predicate WANT (for the verb volere) and its complement
correspond to a syntactic structure (as an alternative to restructuring). State
the correspondence rules, as well as the correspondence rule for placing a
pronominal clitic.
[§7.4.]
6. Icelandic has “quirky” case, where subjects and objects of certain verbs
are marked with other than nominative or accusative case. Moreover, the
evidence shows clearly that the quirky case on an object remains with that
object when it is the subject of a passive.
(1) M’er var hjálpaD

me-DAT was helped
‘I was helped.’
[SigurDsson 2002c:694]
This question is concerned with what happens when these non-nominative

subjects are embedded in control and raising configurations.
Consider control first. The following examples show a few more verbs
that govern quirky case.
(2) a. Hana /∗ Hún vantaDi vinnu

her.ACC /∗ NOM lacked job
‘She lacked a job.’
b. Henni /∗ Hún leiddist
her.DAT /∗ NOM bored
‘She was bored.’
c. Hennar /∗ Hún var getiD
her.GEN /∗ NOM was mentioned
‘She was mentioned (by someone).’
Even when the controller and the infinitival subject position disagree in case,
control is possible. We mark the quirky case on PRO to help clarify what is
going on.
(3) Ég vonaDist til aD PRO-DAT verDa hjálpaD

I hoped for to be helped.
‘I hoped to be helped.’
[SigurDsson 2002c:694]
(4) a. Hún vonast til [aD PRO vanta ekki vinnu].
she hopes for to PRO.ACC lack not job
‘She hopes not to lack a job.’
b. Hana langar ekki til [aD PRO leiDast].
her.ACC wants not for to PRO.DAT bore
‘She does not want to be bored.’
c. ÞaD væri gaman [aD PRO verDa getiD].
It were nice to PRO.GEN be mentioned
‘It would be nice to be mentioned.’
[SigurDsson 1991a:328, 329]
(5) a. Hann lofar [aD PRO lesa bókina]
he promises to PRO.NOM read book
‘He promises to read the book.’
b. Hún vonast til [aD PRO finnast bókin skemmtileg]
he promises for to PRO. DAT find book amusing
‘He promises to find the book amusing.’
[BarDhal 1997]
(6) a. Haraldur vonast til aD lesa bókina
Harald.NOM hopes PRT to read the book.ACC
‘Harald hopes to read the book.’
b. Haraldur vonast til aD batna veikin.
Harald.NOM hopes PRT to recover-from the-disease.NOM
‘Harald hopes to recover from the disease.’
[Roehrs 2005]
c. Eg vonast til að vanta ekki efni í ritgerðina

I.NOM hope for to lack not material in thesis.DEF
‘I hope not to lack material for my thesis.’
[Vincent 2004]
A. The following data show the behavior of alla “all”. What does this
suggest about the proper syntactic analysis of the infinitive?
(7) a. Strákana vantaD i alla í skólann.

the boys.ACC lacked all.ACC. PL . MS in the school
‘The boys were all absent from school.’
b. Strákarnir vonast til [aD PRO vanta ekki alla í
the boys.NOM hope for to PRO.ACC lack not all.ACC in
skólann].
the school
‘The boys hope not to be all absent from school.’
(8) a. Strákunum leiddist öllum í skóla.
the boys.DAT bored all.DAT. PL . MS in the school
‘The boys were all bored in school.’
b. Strákarnir vonast til [aD PRO lei Dast ekki öllum í
the boys.NOM hope for to PRO.DAT bored not all.DAT in
skóla].
the school
‘The boys hope not to be all bored in school.’
[SigurDsson 1991a:331]
What would be required for a non-PRO account of alla agreement in

Icelandic that would capture the fact that it displays the case that would
appear on the subject if there was a subject? (Hint: The solution resembles
that for verbal agreement with pro. It is necessary to refer to the CS argu-
ment that corresponds to the GF that would be realized as a subject in a
tensed complement, but is suppressed in the infinitival. The quantifier alla
is be linked to this GF and derives its form from it.)
B. Consider next what happens when an infinitive is the complement of a
“raising” verb. There are two cases, one in which the NP is the object of the
higher verb and one in which it is the subject of the higher verb.
Object
(9) a. Ég tel [hanna hafa seD myndina]

I believe her. ACC have seen picture
‘I believe that she has seen the picture.’
b. Ég tel [henni hafa leiDst bókin]

I believe her.DAT have bored book
‘I believe that she has found the book boring.’
[SigurDsson 1992b]
Subject
(10) a. Ólafur las bókina

Olaf.NOM read book.the(ACC)
‘Olaf read the book.’
b. Ólafur byrjaDi aD lesa bókina
Olaf.NOM began to read book the(ACC)
‘Olaf began to read the book.’
(11) a. Ólafi leiddist
Olaf.DAT bored
‘Olaf was bored.’
b. Ólafi byrjaDi aD leiDast
Olaf.DAT began to bore
‘Olaf began to get bored.’
(12) a. Ólafi virtist hafa leiDst
Olaf.DAT seemed to.have bored
‘Olaf seemed to have been bored.’
b. Ólafi var taliD hafa leiDst
Olaf.DAT was believed to.have bored
‘Olaf was believed to have been bored.’
[SigurDsson 2002c:699, 698]
(13) a. Haraldur las bókina
Harald.NOM. read the-book.ACC
‘Harald read the book.’
b. Ég tel Harald hafa lesiD bókina
I believe Harald.ACC to-have read the-book.ACC
‘I believe Harald to have read the book.’
(14) a. Haraldi batnaDi veikin
Harald.DAT recovered-from the-disease.NOM
‘Harald recovered from the disease.’
b. Ég tel Haraldi hafa batnaD veikin
I believe Harald.DAT to-have recovered-from the-disease.NOM
‘I believe Harald to have recovered from the disease.’
[Roehrs 2005:2]
(15) a. Haraldur virDist hafa lesiD bókina.
Harald.NOM seems have read the-book.ACC
‘Harald seems to have read the book.’
b. Haraldi virDist hafa batnaD veikin.

Harald.DAT seems have recovered-from the-disease.NOM
‘Harald seems to have recovered from the disease.’
[Roehrs 2005:3]
(16) a. Drengina vantar mat
boys.DEF. ACC lacks food.ACC
‘The boys lack food.’
b. Drengina virðist vanta mat
boys.DEF. ACC seems lack.INF food.ACC
‘The boys seem to lack food.’
[Vincent 2004]
What is the generalization that is shown by this data? What conclusions

can you draw about the selectional properties of the verb with respect to
quirky case? How does this data support a movement account of the passive
and raising in Icelandic? What would be required to capture the same
generalizations under a non-movement analysis, e.g. a correspondence rule?
Can this analysis be extended to sentences that do not involve quirky case?
[§7.4.]
7.1. 1, 2, 3, 4, 5
7.2. 6 1, 2, 3
7.3. 4, 5, 6, 7 1, 2
7.4. 7, 8, 9, 10, 11, 12 8, 9, 10, 11 3, 4, 5, 6
8
Predication
Predication is the relation that holds between the subject NP of a sentence

and the main VP – the VP is called the predicate and is said to “predicate”
a property of the subject. A predication relation may also involve other
constituents of a sentence besides the subject and the VP. In these cases, the
question arises as to the syntactic relationship between the predicate and
the NP. We look at some of the main phenomena bearing on this question
in this chapter.
8.1. Secondary predication
8.1.1. Predicates and antecedents
Primary predication between subject and predicate is used to form a propo-

sition that is asserted, questioned, or an argument of some relation, among
other things. In the following examples, the subject is italicized and the
predicate is underlined.
(1) a. Kim went to bed.
b. Kim was hungry.
(2) a. Did Kim go to bed?
b. Is Kim hungry?
(3) a. I said that Kim went to bed.
b. I said that Kim was hungry.
In addition to this primary predication relation there is secondary predi-

cation, which is used to attribute additional properties to individuals and
states of affairs that are involved in primary predication. Consider the
following example.
(4) Kimi went to bed hungryi .
300 8. PREDICATION
Here we coindex Kim and hungry to show that hungry is predicated of Kim.
Hungry is not the main predicate, went to bed is. In this case, hungry is a
secondary predicate. We call Kim the antecedent of the secondary predicate.
Notice that Kim is both the subject of the main predicate and the antecedent
of the second predicate, and therefore bears two functions with respect to
the interpretation. A paraphrase of this sentence is
(5) Kim went to bed (at some time t) and was hungry (at time t).
The type of predication in which the secondary predicate holds of the

antecedent simultaneously with the primary predicate is called depictive
predication. Depictive predication may also be paraphrased with “while”:
“Kim went to bed while she was hungry”.
8.1.2. Predication and control
Secondary predication is similar to control (Chapter 7) in two respects:

r the antecedent of the secondary predicate is an argument that may appear some-
where in the sentence other than immediately adjacent to the predicate;
r the antecedent of the secondary predicate does not bear a grammatical relation to
the predicate.
Sentence (4) is therefore similar to a sentence like Kim expects to be hungry.

In this control example, Kim is not adjacent to the infinitive, and is the
subject of expects, not to be hungry.
Also, in some cases the predicate is selected by the verb, and in these cases
the choice of antecedent is restricted.
(6) a. Kimi made Sandyj angry∗ i,j .
b. Kimi kept Sandyj warm∗ i,j .
c. Ii consider Terryj intelligent∗ i,j .
This type of secondary predication resembles object control, e.g. Kim forced
Sandy to be polite, in which only the object can be interpreted as referring
to the subject of the infinitive.
However, there are some differences between the patterns shown by
predication and control. Examples such as the following show that, unlike
control, the antecedent of the secondary predicate is not always restricted
to a particular argument.
(7) a. Kimi showed Sandyj Robink drunki,j,k .
b. Kimi showed Sandyj the movie drunki,j .
8.1. SECONDARY PREDICATION 301
Example (7a) can be interpreted three ways, depending on which NP is

taken to be the antecedent of drunk. The interpretation in which the indirect
object Sandy is the antecedent is difficult to get when there is an animate
direct object, but example (7b) shows that it is in principle possible.
The secondary predicate in general cannot be understood as having an
arbitrary antecedent, in contrast with certain cases of control (Chapter 7,
section 7.3.3). That is, it is not possible to have a secondary predicate in a
clause that does not contain an argument that can serve as the antecedent of
the predicate. Compare (8a), which illustrates arbitrary control, with (8b),
which attempts to have a secondary predicate with an arbitrary antecedent.
(8) a. Sandy thinks that it is important [to check the oil level]arb .
b. ∗ Sandy thinks that it rained drunkarb .
Example (8b) cannot be understood as meaning that Sandy thinks that

someone was drunk while it rained.
However, a secondary predicate may have an arbitrary antecedent if its
antecedent itself is a case of arbitrary control. In example (9), the predicate
sober has the same arb antecedent as the infinitive.
(9) Sandy thinks that it is important [to check the oil level soberarb ]arb .
This example shows that the interpretation of the predicate is dependent

on the identification of an antecedent in the sentence. In contrast, the
antecedent of an infinitival can be arbitrary. These examples show that
the interpretation of the secondary predicate works very differently than
does the interpretation of the controlled infinitive. The interpretation of the
secondary predicate is dependent on the presence in the CS interpretation
of some argument that can serve as an antecedent, but this is not true for
control.
8.1.3. Resultative predicates
Secondary predication also allows for a resultative interpretation, besides

the depictive interpretation illustrated in (4). The predicate denotes a prop-
erty of the Theme resulting from the action taken by the Agent.
(10) a. Kimi hammered the metalj flatj . [resultative: “Kim hammered the metal
and as a result it became flat.”]
[cf. Kimi hammered the metalj nakedi . (depictive)]
302 8. PREDICATION
b. We painted the roomi purplei . [resultative: “We painted the room and as a
result it became purple.”]
[cf. Wej painted the roomi nakedj . (depictive)]
There are also intransitive resultatives.

(11) The pondi froze solidi . [“The pond froze and as a result it became solid.”]
Secondary predicates can be expressed by NPs and PPs as well as APs.

(12) a. Kim hammered the metal into a flat sheet.
b. I consider Terry a genius.
c. We painted the room a weird color.
d. Kim kept Sandy outside of the room.
e. Bill rolled out of the room.
f. Bill broke the bathtub into pieces.
g. The professor talked us into a stupor.
[Examples e–f from Goldberg and Jackendoff 2004:536]
Notice that some of these (a,c,e,f) can appear without the secondary predi-
cate, while the others (b,d,g) require a resultative.
(13) a. Kim hammered the metal.
∗
b. I consider Terry. [Ok with a different interpretation]
c. We painted the room.
∗
d. Kim kept Sandy. [Ok with a different interpretation]
e. Bill rolled.
f. Bill broke the bathtub.
∗
g. The professor talked us.
There is an unusual construction in English, the false reflexive, that uses

resultative secondary predication to make an intransitive verb into a transi-
tive reflective verb.
(14) Mary laughed herself silly.
This means “Mary laughed until she was silly” or “Mary laughed so much
that she became silly”. An intransitive verb cannot be used with a resultative
unless the verb denotes a change of state; compare (11) and (15).
∗
(15) Mary laughed silly.
Froze denotes a change of state, while laugh does not.

The phrase structure rule that introduces secondary predicates is straight-
forward. We state it so that it generalizes with the rule for infinitival com-
plementation.
8.1. SECONDARY PREDICATION 303
 

 AP 

 PP 
(16) VP → V (NP)  

 NP 

VPINF
8.1.4. Correspondences
In spite of their differences, there is a clear parallel between control and

secondary predication that we noted in section 8.1.2, in that both involve
predicates that require an external antecedent that supplies the “subject”
role. This role is often referred to as the external θ-role, and the argument to
which it is assigned is called the external argument. One difference between
the two constructions is that a controlled VP is a complement that itself
bears a semantic role assigned by the verb, while a secondary predicate is
not in many cases. 1 In English, at least, we get the following differences.
(17) a. Sandyi tried [VP to be nice]i

b. ∗ Sandyi tried nicei. .
We can capture this difference directly by recognizing that try selects an

infinitival VP complement but it does not select an AP complement. How-
ever, there are cases where an AP is a selected complement.
(18) I consider Sandyi intelligenti .
This is also a lexical phenomenon, as evidenced by the fact that not all
semantically related verbs take the same complement structure.
 

 ?think  

 

 thought  


 


 believe 

deem intelligent
(19) I Sandy .

 judge 
 guilty

 ∗ 
estimate 

 


 ∗
guess  

 ∗ say 

1
The verb keep assigns a Location role to its secondary predicate. The predicate
may denote a metaphorical rather than a literal location.
 
 in the house 
 
warm
(i) Sandy kept Kim .
 confused 
 
in the dark
304 8. PREDICATION
The correspondence between a secondary predicate and its CS represen-

tation is given in (20). We first illustrate the resultative interpretation.
(20)
SYNTAX S
NP VP
V NP AP
GF Subject Object
CS F(X, Y·; RESULT:PROPERTY (·))
For example, Sandy hammered the metal i flati .

The depictive secondary predicate has a similar correspondence, but the
CS representation is different.
(21)
SYNTAX S
NP VP
V NP AP
GF Subject Object
CS F(X·, Y; WHILE:PROPERTY (·))
For example, Sandyi reads the newspaper naked i .

In principle either of these two correspondences can apply when there is a
secondary predicate; the meaning of the predicate will determine whether it
can serve as a result, as a depiction, or both, as illustrated by the examples
in (7).
8.2. SMALL CLAUSES 305
8.2. ∗ Small clauses
In this section we consider the MGG approach to secondary predica-

tion. Since the predicate has an external argument and a sentential para-
phrase, the logic of uniformity suggests that it is a constituent of a
clause, and that the external θ-role is assigned to a subject NP in the
syntactic representation. In examples such as Sandy sleeps naked there
is no apparent subject of naked. As in the case of control, the formal
device that has often been proposed in MGG for this construction is
PRO, which is assumed in this case to be the subject of a clause that
lacks inflection. That is, naked is assumed to be clause of the form [PRO
naked].
This type of clause is called a small clause. Since the analysis assumes
a substantial amount of abstract structure, it is not entirely clear what the
category of the small clause is; we notate it for now as SC.
(22) S
NP VP
Sandy V SC
sleeps NP AP
PRO naked
A similar analysis is assumed for cases in which the predicate applies to the
object.
(23) S
NP VP
Sandy V NP SC
hammered the metal NP AP
PRO flat
As can be seen, analyzing secondary predication in this way shifts the

problem of identifying the antecedent of the predicate to the problem of
identifying the antecedent of PRO.
306 8. PREDICATION
On a small clause analysis, selection of the secondary predicate is trans-

lated into selection of SC, or whatever category it turns out to be. However,
the distribution of small clauses is substantially different from that of con-
trolled infinitives and gerunds, which suggests that conflating control and
secondary predication may not be straightforward. One difference, which
we have already seen, is that secondary predication, in contrast with con-
trol, lacks the possibility of an arbitrary interpretation. (See (8).) Another,
more subtle difference, is that secondary predicates can be questioned when
they are selected, as shown in (24), while selected and controlled infinitives
cannot be, particularly when they are controlled by objects, as in (25).
(24) a. How flat did Sandy hammer the metal?

How intelligent
b. do you consider Sandy?
What
c. What (color) did you paint the room?
(25) a. ∗ What did you force Sandy? (I forced Sandy to confess.)
b. ∗ What did you convince Terry? (I convinced Terry to leave.)
c. ∗ What did you beg Kim? (I begged Kim to call.)
For these reasons, the use of PRO to represent the relationship between
the secondary predicate and its antecedent does not appear to fall together
with control, and it imputes certain structure to the secondary predicate
that does not appear to have independent motivation.
However, there are two pieces of evidence that suggest that in certain cases
there can be small clauses of the form [SC NP AP]. But these small clauses
do not involve PRO. The first piece of evidence involves sentences such as
the following.
(26) [Sandyi angryi ] is a terrible thing.
Here we understand angry to be predicated of Sandy, and we understand

Sandy angry as the subject of is a terrible thing. In this case, Sandy angry
means “the state of affairs of Sandy being angry”. Not only can Sandy angry
be a subject but it can also be displaced.
(27) [Sandyi angryi ], I firmly believe ___ to be a terrible thing.
The distributional evidence thus supports the view that Sandy angry is a
constituent.
Sandy angry can also be a complement.
(28) a. We talked about [Sandyi angryi ].
b. [Sandyi angryi ], we talked about ___.
(29) a. We just couldn’t imagine [Sandyi angryi ].

b. [Sandyi angryi ], we just couldn’t imagine ___.
Here, the topicalization evidence again supports the constituent status of

the small clause. It is also possible to construct a pseudo-cleft with this
constituent in focus position.
(30) What I imagined was [Sandyi angryi ].
We discuss the structure of pseudo-clefts in some detail in Chapter 9,

section 9.8.3. The main point to note here is that what follows the form of be
in this construction must be a constituent. So we cannot have a pseudo-cleft
such as
∗
(31) What I put was [the groceries] [on the table].
based on
(32) I put [the groceries] [on the table]
Note that there are other verbs that appear to take a small clause com-
plement, but with these verbs the constituent structure is not confirmed by
topicalization or pseudo-cleft.
 
 made 
(33) a. We considered Sandyi angryi .
 found 
 
 made 
b. ∗ Sandyi angryi , we considered ___.
 found 
 
 made 
c. ∗ What we considered was Sandyi angryi . 2
 found 
The fact that there are true small clauses turns out to be an argument
against positing PRO as the subject of secondary predicates, which in turn
argues against positing PRO as the subject of non-finite complements.
Working out the logic of this argument is left for Problem 3.
Finally, let us consider the syntactic category of true small clauses. Given
that finite and non-finite sentences can function as the subject or comple-
ment of a verb, it is not unreasonable to suppose that a small clause is
an S.
2
This sentence What we found was Sandy, angry is grammatical when angry is
not interpreted as a predicate of Sandy, but as an appositive, meaning “who was
angry”. In this case there is an intonational break between Sandy and angry that we
do not find with secondary predication.
308 8. PREDICATION
 

 ∗Sandy angry 

that Sandy was angry
(34) a. We warned Kim about .

 Sandy being angry 

∗
for Sandy to be angry
 

 Sandy angry 

That Sandy is angry
b. is a terrible thing.

 Sandy being angry 

For Sandy to be angry
 

 Sandy angry 

that Sandy is angry
c. I just can’t imagine .

 ∗Sandy being angry 

(for) Sandy to be angry
The notable characteristic of this S is that it appears to lack a head, or,

if it has a head, it is not endocentric. It is of course possible to preserve
the generalization that all phrases are endocentric projections of heads by
simply stipulating that there is an abstract invisible head I0 of Sandy angry,
and perhaps also of Sandy being angry. Such a stipulation would need to be
supported by converging evidence that the head actually exists; otherwise it
serves only to maintain the generalization. Problem 4 asks you to explore
this question in more detail.
An alternative that does not assume PRO would be to say that the small
clause is a projection of the predicate, so that Sandy angry would be an AP,
Sandy a genius would be an NP, Sandy in the room a PP, and so on. It is not
clear that independent evidence can be found to support this categorization
beyond the assumption that the small clause must have the same category
as the predicate. Moreover, APs without overt subjects cannot appear as the
complement of verbs that take small clause complements.
∗
(35) a. Wei warned Kimi about PROi,j angry.
Sandy
∗
PROarb
b. angry is a terrible thing.
Sandy
∗
PROi
c. Ii just can’t imagine angry.
Sandy
These sentences would seem to be ungrammatical because they have APs in

contexts where NPs or Ss should appear.
Another case of small clauses involves sentences such as the following.
(36) Sandyi appears unpleasanti .
Recall that appear is a “raising” verb, meaning that the subject of this
verb is selected by the lower predicate only. This property of appear explains,
for example, why it can have a dummy subject only when the complement
selects a dummy subject, e.g.
(37) a. There appears to be a problem.

[cf. There is a problem.]
b. ∗ There appears to have eaten the peanut butter sandwich.
[cf. ∗ There ate the peanut butter sandwich.]
In the present case the lower predicate is not an infinitive. A natural

extension of the movement analysis of raising of the subject of the infinitival
complement is to assume that the complement is a small clause, and to raise
the subject.
(38) S
NP VP
V SC
appear NP AP
Sandy unpleasant
To make such an analysis plausible, it must be assumed, as in the case of

raising from an infinitive, that the NP lacks case, so that it has to raise.
∗
(39) It appears Sandy unpleasant.
An alternative is to assume that the complement of appear is simply AP,

and is one of the categories selected by appear. To the extent that this AP
has an external argument in CS, it is also straightforward to generalize
the correspondence rule so that the subject of appear is mapped into the
argument of the AP. (See Problem 5.)
Examples such as the following show that this non-movement alternative
is a plausible one. These examples show the AP complement is selected by
appear and other verbs, and does not alternate freely with an infinitival
complement. In (40a) we see first that these are all “raising” verbs. The
examples in (40b) show these verbs with infinitival complements, and those
in (40c) with AP complements.
310 8. PREDICATION
 

 appears  

 


 seems 

 happens 
 
(40) a. There turns out to be a problem.

 is likely 

 


 began 


 

continued
 

 appears  

 


 seems 

 happens 
 
b. Sandy turns out to be unpleasant.

 
 is likely 
 


 began 


 

continued
 

 appears 


 


 seems 


 happens 
∗ 
c. Sandy ?turned out unpleasant.

 
 ∗ is likely 
 


 ∗
began 


 

?continued
In order to make a movement analysis reflect these idiosyncrasies, we would

have to stipulate a selectional difference between an infinitival S comple-
ment and a small clause complement, thereby losing the generality that we
would gain by treating them both as cases of raising.
8.3. ∗ Secondary predication cross-linguistically
In this section we compare how English does secondary predication with

how other languages do it. Here are some examples from Chinese.
(41) Chinese
a. Wusong da si le laohu. (resultative, V-V)
Wusong beat die PERF tiger
‘Wusong beat the tiger so that it died.’
b. Wusong da de laohu liuxue le. (resultative, de)
Wusong beat DE tiger bleed PRT
‘Wusong beat the tiger so that it bled.’
(42) a. Wusong huo zhuo le yi zhi laohu. (depictive, V-V)
Wusong alive catch PERF one CL tiger
‘Wusong caught a tiger alive.’
8.3. SECONDARY PREDICATION CROSS-LINGUISTICALLY 311
b. Wusong rere de he le yi wan jiu. (depictive, de)

Wusong hot DE drink PERF one bowl wine
‘Wusong drank a bowl of wine hot.’
[Zhang 2001:192]
It can be seen that Chinese allows VPs to be interpreted as resultative

predicates; English requires so that in order to convey this relation. Also,
the syntactic location of the resultative predicate is different from that of
the depictive. Problem 6 asks you to try to work out the phrase structure
rules for the Chinese VP to accommodate these constructions.
We turn next to German secondary predicates. All verbs in German
except the inflected verb in a main clause appear in final position, essentially
the mirror image of the word order of English; the inflected verb appears
in second position in main clauses, and in final position in subordinate
clauses.
(43) German
a. Er ißt die Äpfel.
he eats the apples
‘He eats the apples.’
b. . . . daß er die Äpfel ißt.
that he the apples eats
‘. . . that he eats the apples’
(44) a. Ich habe das Buch gekauft.
I have the book bought
‘I bought the book.’
b. . . . dass ich das Buch gekauft habe.
that I the book bought have
‘. . . that I bought the book’
Given this mirror image property, we might expect German secondary

predicates to appear at the left edge of the VP. Here are some examples
to test this prediction.
(45) a. weil eri die Äpfelj ungewascheni/j ißt.

because he the apples unwashed eats
‘because he eats the apples unwashed.’
(He is unwashed or the apples are unwashed.)
b. weil eri ungewascheni/∗ j die Äpfelj ißt.
because he unwashed the apples eats
‘because he eats the apples unwashed.’
(He is unwashed.)
312 8. PREDICATION
c. ∗ weil ungewaschen∗ i/∗ j eri /der Manni die Äpfelj ißt.

because unwashed he/ the man the apples eats
[Müller 2004:8)
The examples show that German is not the mirror image of English with
respect to the syntax of secondary predication. In German, as in English,
the secondary predicate must follow the antecedent. In example (45a), it
follows both NPs, so either may be the antecedent. In example (45b) it
follows only er “he”, so there is only one possible interpretation. And in
example (45c) it precedes both NPs, so there is no interpretation of the
predicate.
Compare also the situation in Dutch, which, like German, is SOV in
subordinate clauses. In (46a,b) we see that the secondary predicate must
follow its antecedent, just as in English.
(46) Dutch
a. dat Jan [de peer]i [in stukken]i snijdt
that John the pear into pieces cuts
‘that John cuts the pear into pieces’
b. ∗ dat Jan [in stukken]i [de peer]i snijdt
that John into pieces the pear cuts
‘that John cuts the pear into pieces’
(47) a. ∗ John cuts [into pieces]i [the pear]i
b. John cuts [the pear]i [into pieces]2
[Neeleman 1994:176]
The comparison between English, German, and Dutch raises an interest-

ing problem regarding how to specify precisely when an NP can serve as the
antecedent for a secondary predicate. We have seen that in each language the
antecedent must precede the predicate regardless of the position of the V in
VP. But it is clear that simple linear precedence is not sufficient, because
an NP inside of another NP cannot be the antecedent of a secondary
predicate, e.g.
∗
(48) The picture of Robini turned redi .
So there must be a structural condition as well.

The natural candidate for such a structural condition is c-command. The
problem then becomes, how does the c-command relation work in the three
languages? There are two major analyses. The first, which we will call the flat
8.3. SECONDARY PREDICATION CROSS-LINGUISTICALLY 313
structure analysis, assumes that the verb, the direct object, and the predicate
are sisters. The following illustrates for English.
(49) VP
V0 NP AP
The comparable structure with V in final position will work for German
given that when there is more than one verb the verbs appear in the reverse
order to English.
(50) VP
NP AP V0
VP
VP V0
NP AP V0
In English and German we can require that the antecedent precedes and
c-commands the predicate. Since the subject c-commands everything in VP,
an AP in a transitive sentence will be c-commanded by both subject and
object.
The problem arises when we consider Dutch. In Dutch the order of verbs
is the same as that in English although they appear at the end of the VP.
That is, the auxiliary verbs precede the main verb. For example,
(51) dat Jan de deur nooit zal verven
that John the door never will paint.INF
‘that John will never paint the door’
Compare this with the German order, in which the auxiliary verb follows
the main verb.
(52) dass er uns nicht besuchen kann
that he us not visit.INF can
‘that he can’t visit us.’
For this reason, it has been proposed that the basic constituent order of
Dutch is the same as that of English, and that the direct object in Dutch
moves to the left (see Zwart 1993).
314 8. PREDICATION
(53) S
C NP VP
dat Jan VP
NP
that Jan
de deur V0
the door
verft
paints
If there is more than one verb, the NP would move to the left of the first
one, producing the order NP – V1 – V2 that we see in (51).
However, this movement does not account for the position of the AP,
which also is to the left of the Vs. If the initial order is the same as it is in
English, and the NP moves to the left, we get
(54) V1 V2 NPi APi → ∗ NPi V1 V2 ___ APi
rather than
(55) NPi APi V1 V2
While it might be possible to motivate movement of the NP (e.g. to get case

assigned to it), it is hard to see why the AP would move, unless NP and AP
form a constituent, a possibility that we return to below.
Another alternative is that V2 moves out and to the left, and then the
remaining VP, now missing its verb, moves to the left. Again, it is not clear
why this movement should occur. 3
(56) V1 [VP V2 NPi APi ] → V1 V2 [VP NPi APi ] → [VP NPi APi ] V1 V2
Similar questions arise if we assume that the order in Dutch (and perhaps
German) is the same as that of English, with movements of the verbs to the
right.
3
When there are movements that are not obviously motivated it is always possi-
ble to assume that there is a head H0 with a feature [F] that requires a constituent
of a certain category (in this case VP) to move to its specifier position. Such
movement analyses can always be formulated, but may not always be independently
justifiable.
PROBLEMS 315
One simple account is that the structures of English and German are as
we have shown them, and that the structure of Dutch is essentially that
of German, but with a reordering of the verbs. If this analysis is correct,
the reordered verbs must form a unit, since clusters of three verbs show
the same inverted order, with the lowest verb farthest to the right and the
highest verb farthest to the left.
(57) Dutch
(Ik denk) dat Jan het boek moet hebben gelezen
I think that John the book must have.INF read.PAST. PART
‘I think John must have read the book.’
[Zwart 1996:234]
When the highest verb is an auxiliary, other orders are possible, which
supports the idea that the verbal sequence is formed by reordering.
(58) a. (Ik denk) dat Jan het boek moet gelezen hebben
(I think) that John the book must read.PAST. PART have.INF
b. (Ik denk) dat Jan het boek gelezen moet hebben
(I think) that John the book read.PAST. PART must have.INF
[Zwart 1996:234]
Problem 7 asks you to explore and evaluate more complex alternative analy-
ses in terms of c-command.
Problems
1. Here is a construction in English in which the number of syntactic

arguments is greater than the number of CS arguments.
(1) a. Mary laughed herself silly.
∗
b. Mary laughed herself.
∗
c. Mary laughed John.
∗
d. Mary laughed John silly.
e. Mary laughed (∗ silly).
i. Describe the relationship between the argument structure of the lexical

entry of the verb laugh and the argument structure of the grammatical
sentence in (1a). Be explicit about the correspondences.
ii. State the relationship between the argument structure of the intransitive
verb and the grammatical sentence in (1a) so that the ungrammatical sen-
tences (1b–d) will be ruled out.
316 8. PREDICATION
iii. Is the relationship that you described in (i) a general one that holds for
all intransitive verbs, or is it restricted to a subclass of intransitive verbs?
Here are some intransitives to get you started: arrive, call, die, eat, hum,
sing, sleep, talk, work. If it is a general relationship, try to formulate it as
simply as possible.
[§8.1.]
2. In our syntactic analysis of secondary predication we have assumed

that the AP is a constituent of VP. Using constituency tests involv-
ing movement and deletion, determine if this is correct. Make sure you
consider secondary predicates with both subject antecedents and object
antecedents.
[§8.2.]
3. In the text we said that the fact that there are true small clauses turns
out to be a very strong argument against positing PRO as the subject of
secondary predicates, which in turn argues against positing PRO as the
subject of non-finite complements. Explain the logic of these statements.
(Hint: If there are true small clauses, and if PRO is a possible NP argument,
what predictions follow? Are they correct?)
[§8.2.]
4. As we noted in the text, it is possible to preserve the generalization

that all phrases are endocentric projections of heads by simply stipulating
that there is an abstract invisible head I0 of Sandy angry, which would be an
IP. Compare the distribution of this IP with tensed IPs; is there syntactic
evidence for or against this categorization?
[§8.2.]
5. Work out the lexical entries for one verb in the following example
that allows apparent raising from an AP complement, and one that does
not.
 

 seems 


 


 appears 

 happens 
 ∗ 
(1) Sandy ?turned out unpleasant.

 ∗
is likely 


 


 ∗
began 


 

?continued
PROBLEMS 317
(Hint: For those verbs that allow apparent raising, you must state explicitly
that the syntactic subject of seem is interpreted as the external argument of
the predicate.)
[§8.2.]
6. Write a phrase structure rule to account for the Chinese resultative and
depictive predicates in (41)–(42) in the text. Here are additional examples.
Assume the syntactic categories N, V, and A.
(1) a. (Akiu) ku lei le.

Akiu cry tired PERF
‘Akiu cried and as a result he felt tired.’
b. Akiu da shang le Fanjin.
Akiu beat wound PERF Fanjin
‘Akiu beat Fanjin so that Fanjin was wounded.’
c. Akiu ku shi le shoujuan.
Akiu cry wet PERF handkerchief
‘Akiu cried and as a result the handkerchief was wet.’
(2) a. Akiu wanr de wang le zuoye.
Akiu play DE forget PERF homework
‘Akiu played so much that he forgot the homework.’
b. Akiu kua de Fanjin buhaoyisi le.
Akiu praise DE Fanjin embarrassed PERF
‘Akiu praised Fanjin so that Fanjin felt embarrassed.’
c. Akiu ku de shoujuan ye shi le.
Akiu cry DE handkerchief also wet PERF
‘Akiu cried and as a result the handkerchief was wet.’
(3) a. Jia Zheng zai nu da Baoyu.
Jia Zheng PROG angry beat Baoyu
‘Jia Zheng is beating Baoyu angry.’
b. Akiu huo zhuo le Nanbatian.
Akiu alive catch PERF Nanbatian
‘Akiu caught Nanbatian alive.’
(4) a. Akiu hulihutu de mai le yi bao shipin.
Akiu confused DE buy PERF one package food
‘Akiu bought a package of food confused.’
b. Akiu lala de chi le yi wan Dandan-mian.
Akiu spicy DE eat PERF one bowl Dandan-noodle
‘Akiu ate a bowl of Dandan-noodle spicy.’
[Zhang 2001:193]
[§8.3.]
318 8. PREDICATION
7. In the text we outlined an analysis of secondary predication in German

and Dutch in which the VP structure is flat; that is, V, NP, and AP are
all sisters. This analysis requires that the antecedent both precede and c-
command the predicate. What changes in the structure would be required
in order to get rid of the “precede” requirement? (Hint: In order to make
such an analysis work, c-command must align entirely with precede, so that
if · c-commands ‚ then · precedes ‚. What would the structures have to
look like?)
[§8.3.]
Research questions
1. In the text we noted that when there is arbitrary control and secondary
predication, the two are understood as having the same reference. Hence
example (1) is understood to mean that whoever checks the oil should
be sober when doing so. It cannot be understood as meaning that it is
important in general for someone to check the oil level and that someone
else should be sober when this is done.
(1) Sandy thinks that it is important [to check the oil level soberarb ]arb .
Our notation arb is a sloppy way to represent this relationship, since we

have not said exactly what arb corresponds to in the CS representation.
Clearly, when we use arb, it means “for any arbitrary person x” and when
arb is used more than once in the same sentence as in (1), it is tied to
this same interpretation. (This relationship is a special case of binding; see
Chapter 7).
A similar case involves predication across the copular be.
(2) [To know her]arb is [to love her]arb .
This does not mean that for any person x to know her is equivalent to
some other person y loving her. The knower and the lover must be the same
person x.
Compare this case with
(3) [To err]arb is human, [to forgive]arb , divine.
In this example, the reference of the two arbs is not the same. The puzzle
presented by these examples has two parts. First, how do we represent
the arb interpretation properly in CS, so that we can show when two arbs
pick out the same (arbitrary) individual and when they do not? Second,
what are the syntactic conditions, if any, under which these two possibilities
occur? That is, are the interpretations seen here accidental, or are they the
consequence of the correspondences between syntactic structure and CS?
[§8.1.2.]
2. Formulate an account of the possible order of verbs in the Dutch
verbal cluster (see (57)–(58)). One obvious approach is to assume that the
VP is left-branching, and that various parts optionally reorder to the right,
as in (1).
(1) VP
NP AP V
VP
VP V
NP AP V1 V2
VP
VP V
VP V V3
NP AP V1 V2 moet
gelezen heben
Can the ordering possibilities be accounted for without assuming move-

ment but by allowing for alternative correspondences for the components
of the CS representations? Such an account would assume that there are
operators NECESSARY and PAST in CS corresponding to the auxiliary
verbs moet and heben.
(2) NECESSARY(PAST(READ(AGENT:X, . . . ))
320 8. PREDICATION
The trick is to formulate the individual correspondence rules in such a way

that just the possible linear orderings of the verbs are accounted for.
[§8.3.]
Section Problems Research questions
8.1. 1 1
8.2. 2, 3, 4, 5
8.3. 6, 7 2
9
A constructions
In this chapter we look at constructions in which there is a constituent in a

non-argument (that is, A ) position and a corresponding gap in an argument
position; the canonical example is the English-type wh-question such as
Who are you talking to ___? These are called A (Ā or “A bar”) constructions.
In contrast, the passive is an A construction, one in which two argument
positions are related.
Owing to the correspondence between the two positions, the type of
construction exemplified by English wh-questions is conventionally referred
to as A (Ā or “A bar”) movement, where the constituent in A position
appears to have been displaced from the position of the gap. The move-
ment terminology reflects the derivational perspective in mainstream gener-
ative grammar (MGG), according to which the S-structure of a sentence
is arrived at through successive application of formal operations on the
syntactic structure. Alternative accounts of the syntax of A constructions
have been developed in non-derivational terms, where the two positions (e.g.
who and the gap in Who are you talking to ___?) are related to one another
in virtue of the syntactic configuration, with no movement.
Regardless of how the relationship between the A constituent and the
gap is expressed, the two positions form an A chain (sometimes called an
A dependency). The constituent in A position is referred to as the head of
the chain while the gap is referred to as the tail. Such constructions are also
referred to as filler-gap constructions.
(1) Filler Gap
who are you talking to
Head Tail
CHAIN
322 9. A CONSTRUCTIONS
All A constructions have very similar properties. The distance between

the head and the tail of the chain can be as long as we like – that is, it is
unbounded, and the gaps cannot appear in certain syntactic configurations
(sections 9.4 and 9.7). The constituent in A position c-commands the gap.
This chapter focuses on three basic A constructions, questions, rela-
tive clauses, and topicalization. Chapter 11 considers a number of other
constructions that share some but not all properties with these basic A
constructions.
9.1. Questions
A wh-question in English is a sentence in which an interrogative constituent,

usually spelled “wh-”, appears in clause-initial position. We will call this
constituent a wh-phrase. The wh-phrase may consist of a single wh-word,
such as who, what, where, when, how, why. Or it may be complex, as in which
man, how many people, how far, etc.
There are two ways in which a wh-phrase can appear in clause-initial
position, and two types of clause-initial position. If the wh-phrase is a
subject, then it appears in clause-initial position in virtue of being a subject.
(2) a. Who just called?
b. What happened?
c. How many people believe that this theory is correct?
If the wh-phrase is not a subject, then it appears in clause-initial position

instead of where it would normally appear. We use the notation e here to
indicate what the normal position would be, and the coindexing to relate
this position to the constituent in initial position.
(3) a. Whati are you eating ei ?
[cf. You are eating pizza.]
b. Whoi were you just talking to ei so enthusiastically?
[cf. You were just talking to a participant.]
c. Wherei did you put the beer ei ?
[cf. You put the beer in the basement.]
d. Whyi is Sandy shrieking ei so loudly?
d . How comei Sandy is shrieking ei so loudly?
[cf. Sandy is shrieking because the TV doesn’t work.]
e. Wheni are you leaving for class ei ?
[cf. You are leaving for class after breakfast.]
9.1. QUESTIONS 323
f. Howi did you find the answer ei ?

[cf. You found the answer by surfing the Internet.]
g. How fasti can you swim ei ?
[cf. You can swim really fast.]
The presence of the wh-phrase in the initial position of the clause in main
clauses is illustrated in these examples. Between a clause-initial wh-phrase
and the subject there is an auxiliary verb. This phenomenon is called
(subject-Aux) inversion. Notice that example (3d ) with how come lacks
inversion.
∗
(4) How come is Sandy shrieking so loudly?
A question may also be embedded, in which case it is called an embedded

(or indirect) question. The following examples show that the wh-phrase
appears in initial position in the embedded question, but the embedded
question lacks auxiliary inversion in the standard dialect.
(5) a. I wonder [S who just called].
b. I wonder [S what happened].
c. I wonder [S how many people believe that this theory is correct].
(6) a. I want to know [S whati you’re eating ei ].
a . ∗ I want to know [S whati are you eating ei ]
b. We were wondering [S whoi you were just talking to ei ].
b . ∗ We were wondering [S whoi were you just talking to ei ].
c. Tell me [S wherei you put the beer ei ].
c . ∗ Tell me [S wherei did
you put the beer ei ].
why i
d. Can you explain [S Sandy is shrieking ei ]?
how comei

whyi
d . ∗ Can you explain [S is Sandy shrieking ei ]?
how comei
e. Do you know [S wheni you’re leaving for class ei ]?
e . ∗ Do you know [S wheni are you leaving for class ei ]?
f. It is unknown [S howi you found the answer ei ].
f . ∗ It is unknown [S howi did you find the answer ei ].
g. The clock will reveal [S how fasti you can swim ei ].
g . ∗ The clock will reveal [S how fasti can you swim ei ].
Wh-questions contrast with yes-no questions.

(7) a. Are you still there?
b. Can you fix the sink?
c. Do you speak English?
d. Don’t you want another drink of water?
When embedded, yes-no questions show a wh-phrase in initial position,

either whether or the superficially equivalent alternative if, and they lack
inversion.

whether
(8) a. Tell me [S you’re still there].
if

∗ whether
[Cf. Tell me [S are you still there.]
if

whether
b. I was wondering [S you can fix the sink].
if

∗ whether
[Cf. I was wondering [S can you fix the sink].]
if

whether
c. It is still unclear to me [S you speak Macedonian].
if

∗ whether
[Cf. It is still unclear to me [S do you speak Macedonian].]
if

whether
d. Do you know [S you want another drink of water].
if

∗ whether
[Cf. Do you know [S you do want another drink of water].]
if
(Problem 1 asks you to determine whether if and whether are completely

interchangeable, and if not, the restrictions on their distribution.)
Our description of the basic correspondence between the syntactic struc-
ture of a wh-question and its meaning takes into account the following: 1
(i) the wh-phrase appears in clause-initial position;
(ii) there is a gap in the position that defines the grammatical function of the wh-
phrase;
(iii) the sentence has an interrogative interpretation.
We represent the interrogative interpretation in CS by using an operator

Q that turns a propositional representation P into a question about that
proposition. So, if P is a proposition, Q(P) is a question corresponding to
P. We let Q have two variants, simple Q for the yes-no question and Q· for
the wh-question about some argument or adjunct variable · contained in
P. So the two types of CS representations for questions are Q(P) for yes-no
1
We use the term “correspondence” here the way it was introduced in Chapter 5.
There is a correspondence between the form of a sentence and its meaning, and
there are more general correspondences between parts of the form and parts of the
meaning.
9.1. QUESTIONS 325
questions, and Q· (P(. . . ,·,. . . )) for wh-questions, where · is a CS argument

or adjunct of P.
The question did Sandy buy the pizza will then have the CS representation
in (9), and what did Sandy buy will have the representation in (10).
(9) a. Sandy bought the pizza.
CS: BUY(AGENT:SANDY,THEME:PIZZA)
b. Did Sandy buy the pizza?
CS: Q(BUY(AGENT:SANDY,THEME:PIZZA))
(10) a. Sandy bought the cake
CS: BUY(AGENT:SANDY:THEME:CAKE)
b. What did Sandy buy?
CS: Q· (BUY(AGENT:SANDY:THEME:·))
The correspondence for what did Sandy buy is given in (11). [e] in this
structure is an empty NP in the direct object position.
(11) SYNTAX S
NP NP VP
what Sandy V [e]
buy
GF Subject Object
CS Q·(BUY(AGENT:SANDY,THEME:·))
As can be seen in this picture, there is a correspondence between the

syntactic position of what and the operator Q· . The CS proposition that
the interrogative is questioning is called the scope of the question. The
corresponding S to which the wh-phrase is attached is often referred to as
the (syntactic) scope of the wh-phrase. In English wh-questions, the wh-
phrase is in the initial position of the clause within its scope.
A particular correspondence between CS, GF, and syntax exists in a
language if the grammar of the language has correspondence rules that
license it (see Chapter 5). In the case of (11) there is one correspondence rule
that is specific to questions, which we call WH - QUESTION, and a number of
others that apply more generally in English, some of which we have already
seen in previous chapters.
(12) a. SUBJECT :
S
NP VP
Subject
b. OBJECT:
VP
V0 NP
Object
c. WH - QUESTION :
S
XP …
[WH]
Q·(…, ·, …)
d. buy: BUY(AGENT:X,THEME:Y)
e. LINKING : Agent ↔ Subject
Theme ↔ Object
The SUBJECT and OBJECT correspondence rules link Sandy and [e] to
the grammatical functions Subject and Object, respectively, in this example.
These are linked to AGENT and THEME, respectively, by the general
linking rule LINKING and the lexical entry for buy (see Chapter 5). Finally,
the wh-phrase in initial position is linked to the Q operator by the corre-
spondence rule WH - QUESTION.
The phrase structure rules of English are responsible for the syntactic
structure, while the rules of CS are responsible for where Q and BUY go
with respect to one another. What and the gap are connected to one another
9.1. QUESTIONS 327
through their correspondence with the CS operator Q and · and its link to
the argument ·.
The correspondence for the wh-question where the wh-phrase is in an A
position requires a wh-phrase in clause-initial position, the operator Q· in
CS, and a variable · in CS. It is important to note that the correspondence
rule (12c) is obligatory in an English wh-question. It is possible to have a
wh-phrase that is not clause-initial, but such cases are either not true wh-
questions (see section 9.2.3) or are multiple wh-questions, where one wh-
phrase is in initial position (see section 9.2.4).
Example (13) shows the correspondence between the variable and
a gap.
(13) S
XP … [e] …
[WH]
Q·(…, ·, …)
We call the set of links between the wh-phrase and the gap a CS-chain.
Notice that part of this chain is a link between Q· and · in CS. Here is a
preliminary definition.
CS-chain
Two syntactic constituents X and Y form a CS-chain when
(a) X c-commands Y and
(b) X and Y are both linked to the same element or a linked set of elements of a
CS representation.
As we will see, there are several types of chains. The kind that we are
concerned with here are those in which Y is empty, and X is in a non-
argument position, e.g. a wh-phrase at the front of the sentence. These are
linked to · and Q· , respectively, completing a CS-chain. A chain that links
a constituent in an A position with a gap is typically called an A chain.
When the wh-phrase is the subject of a main clause, it appears in clause-

initial position in English. There is no direct evidence that the wh-subject
is in an A position that is linked to a gap in subject position. The corre-
spondence is given below. (The question of whether subject questions are
A constructions is taken up further in Research question 2.)
(14) SYNTAX S
NP …
[WH]
GF Subject
CS Q·(…, ·, …)
Q· is linked to the sentence initial wh-phrase, and the SUBJECT correspon-

dence links · to the same phrase, through the Subject GF. There is a chain,
but no gap.
In an embedded question, such as those in (5)–(6), the scope of the
question is marked by the position of the wh-phrase. Compare the CS
representation for what did Sandy buy in (11) with that of Kim asked
what Sandy bought in (15). In both cases the scope is the proposition
BUY(AGENT:SANDY,THEME:·), which corresponds to the sentence
(what) Sandy bought e.
(15)
SYNTAX S
NP VP
Kim V S
ask NP NP VP
what Sandy V [e]
buy
GF Subject [ Subject Object ]
CS ASK(AGENT:KIM,THEME:Q·(BUY(AGENT:SANDY, THEME:·)))
9.2. TYPES OF WH-QUESTIONS 329
9.2. Types of wh-questions
9.2.1. Piedpiping and preposition stranding
In English and in other languages that have A wh-questions, phrases that

contain a wh-phrase may appear in clause-initial position. In (16a), from
whom is a PP that contains a [WH] NP complement, while in (16b), whose
story is an NP that contains a [WH] NP specifier.
(16) a. From whom did you get that fantastic quote?

a . [PP from [NP whom]]
b. Whose story would you rather believe?
b . [NP [NP whose] story]
The two possibilities can even be combined.
(17) On which page did you find that fantastic quote?
In this case, the phrase that appears in initial position is a PP that contains
a phrase whose specifier is [WH]:
(18) [PP on [NP which page]]
This phenomenon, where a phrase that contains a wh-phrase itself

behaves as though it is a wh-phrase with respect to the wh-question, is an
instance of what is called piedpiping. 2 It is important to note that piedpiping
is allowed only in certain configurations; Exercise 2 asks you to investigate
what these configurations are.
The correspondence for the wh-question (16a) with piedpiping of the PP
is given in (19).
2
The term “piedpiping” is due to Ross (1967). It derives from the story of
the Pied Piper of Hamelin. The idea (in a movement theory) is that the wh-
word brings the rest of its phrase along when it moves, just as the Pied Piper
brought the rats, and then the children, out of the town of Hamelin behind
him.
(19) SYNTAX S
PP AUX NP VP
did you V NP [e]

P NP
from whom get quote
GF Subject Object
CS Q·(GET(AGENT:YOU,THEME:QUOTE,SOURCE:·) )
The correspondence for the wh-question (16b) with piedpiping of the NP is

given in (20).
(20) SYNTAX S
NP AUX NP VP
NP N would you Adv V [e]
whose story rather believe
GF Subject Object
CS Q·(BELIEVE(EXP:YOU,THEME:STORY[POSS:·])
Notice in this case that the A chain does not directly link the NP whose
story with the gap. The chain involves the CS link Q· – ·, where · contained
in the representation of the Theme that corresponds to Object and thus to
the gap.
English is relatively rare among languages in allowing a wh-question in
which the gap appears in a PP without piedpiping.
(21) Who did you get that fantastic quote [PP from [e] ]?
This phenomenon, called preposition stranding, or p-stranding, does not

occur in French, and is very restricted in German. 3
(22) French
a. Tu parles à Jean.
you speak to Jean
b. A qui parles-tu
to whom speak-you
c. ∗ Qui parles-tu à
who speak-you to
(23) German
a. Er hat das Buch auf den Tisch gestellt.
he has the book on the table put
b. Auf welchen Tisch hat er das Buch gestellt?
on which table has he the book put
c. ∗ [Welchen Tisch]i hat er auf t i das Buch gestellt?
which table has he on the book put
Problem 2 explores how general p-stranding in English is.
9.2.2. In situ wh-questions
In some languages, e.g. Chinese and Japanese, questions do not involve

A chains. The wh-phrase remains in the position in the sentence where
it acquires its grammatical function. These languages are called wh-in-situ
languages – “in situ” is the Latin for “in place”.
Japanese is a strictly SOV language. We see that when the wh-phrase is
not the subject, it may follow the subject, as in (24a). Because Japanese
allows phrases to appear in various orders, the wh-phrase may also appear
initially, as in (24b). The scope of the question, rather than being marked
by the position of the wh-phrase, is marked by the suffix -ka on the verb.
(24) Japanese
a. Sandy-ga nani-o katta-ka
Sandy-SUBJ what-OBJ bought-Q
‘What did Sandy buy?’
b. Kore-wa nan desu-ka
what-TOP this is-Q
‘What is this?’
3
It appears that something resembling p-stranding occurs in some dialects of
German; see Fleischer 2000.
The correspondence between the structure of these types of questions and

their CS representations is straightforward. As we have seen, the CS of a wh-
question has an operator Q· that binds a variable · that corresponds to an
argument or adjunct in the syntactic structure.
(25) Q· (. . . , ·, . . . )
Even though the wh-question in Japanese is not an A construction, a wh-

question is Japanese has the same CS representation as a wh-question in
English. (After all, translations should have the same CS.) In this case, the
interrogative marker -ka corresponds to Q· , while the variable · corre-
sponds to the indefinite wh-form nani “what”. (We adopt a simple syntactic
configuration here in order to focus on the relevant details.)
(26) SYNTAX S
NP VP
Sandy ga NP V
nani o katta ka
‘what’ ‘bought’
GF Subject Object
CS Q·(BUY(AGENT:SANDY:THEME:·))
The scope of the wh-question is signaled by the position of the interrogative

-ka in the syntactic structure. The scope of the corresponding operator
Q· is the portion of CS that corresponds to the S that most immediately
dominates it. In an embedded question, -ka appears at the rightmost edge
of the embedded clause (27a), while, in a main question, it appears at the
right edge of the main clause (27b).
(27) a. [John-wa [Mary-ga nani-o mottekita-ka] itta-ndesu-ka].

John-TOP Mary-NOM what-ACC brought-Q said-POLITE-Q
‘Did John say what Mary brought?’
b. [John-wa [Mary-ga nani-o mottekita-to] itta-ndesu-ka].
John-TOP Mary-NOM what-ACC brought-that said-POLITE-Q
‘What did John say Mary brought?’
[based on examples in Ueno and Kluender 2003]
Another wh-in-situ language is Chinese.
(28) Chinese
a. Ni xihuan shei?
you like who
b. Zhangsan wen wo [shei mai-le shu].
Zhangsan ask me who bought books
‘Zhangsan asked me who bought books.’
c. Zhangsan wen wo [ni maile shenme]
Zhangsan ask me you bought what
‘Zhangsan asked me what you bought.’
d. Zhangsan xiangxin [shei mai-le shu].
Zhangsan believe who bought books
‘Who does Zhangsan believe bought books?’
e. Zhangsan renwei [ni maile shenme]
Zhangsan think you bought what
‘What does Zhangsan think you bought?’
f. Zhangsan zhidao [shei mai-le shu].
Zhangsan know who bought books
i. ‘Who does Zhangsan know bought books?’
ii. ‘Zhangsan knows who bought books.’
[Huang 1982]
In this case there is no interrogative marker, so the scope of the interrogative
has to be inferred. Thus, there is an ambiguity when the scope may either
be the embedded clause (narrow scope) or the main clause (wide scope), as
seen in the translations of (28f).
A superficially similar but ultimately different situation is found in
French, which has both A wh-questions and wh-in-situ. Consider the fol-
lowing examples from Rizzi 1991:75.
(29) French
a. Qui elle a rencontré e?
who she has met
‘Who did she meet?’
b. Qui a-t-elle rencontré e?
who has-t-she met
c. Elle a rencontré qui?

she has met who
d. ∗ A-t-elle rencontré qui?
has-t-she met who
What these examples show is that wh-questions in French may involve A

chains with or without inversion, or wh-in-situ without inversion. The latter
are true wh-questions, not echo or quiz questions.
Even more complex possibilities exist. For example, in Bellunese, a
Northern Veneto dialect of Italian, complex wh-phrases must appear in
sentence-initial position, while simple wh-words must appear in situ. Note
how Bellunese differs from French.
(30) Bellunese
a. Che libro à-tu ledest
what book have-you read
‘What book did you read?’
∗
b. À-tu ledest che libro?
have-you read what book
‘What book did you read?’
c. Quanti libri à-tu ledest
How.many books have-you read
‘How many books did you read?’
d. ∗ À-tu ledest quanti libri?
have-you read how.many books
‘How many books have you read?’
(31) a. À-tu incontrà chi?
have-you met who
‘Who did you meet?’
b. ∗ Chi à-tu incontrà?
who have-you met
‘Who did you meet?’
c. Sié-o stadi andé?
are-you been where
‘Where have you been?’
d. ∗ Andé sié-o stadi?
where are-you been
‘Where have you been?’
e. À-lo magnà che?
has-he eaten what
‘What did he eat?’
f. ∗ Che à-lo magnà?

what has-he eaten
[Obenauer 2004:345–6]
The word cossa also means “what”, but behaves like a complex wh-phrase
for historical reasons, being derived from che cossa “what thing”.
(32) a. Cossa à-lo magnà?

what has-he eaten
b. ∗ À-lo magnà cossa?
has-he eaten what
[Obenauer 2004:345–6]
In addition to the distinction between the simple and complex wh-phrases,

it is also striking that even with wh-in-situ, there appears to be inversion of
the tensed verb and the subject clitic. We leave the matter of accounting for
the Bellunese data for Problem 8.
9.2.3. English wh-in-situ
There are other constructions in English which appear to be genuine wh-in-

situ, where there is no initial wh-phrase. These are echo questions, exem-
plified in (33), and quiz questions, shown in (34). The echo question is
characterized by a rising intonation, which we mark here with , and
typically is appropriate only in a discourse where a prior assertion has been
made. The quiz question has a falling intonation.
(33) a. A: I just bought a Maserati.

B: You bought what?!
b. A: I just bought a Maserati.
B: You bought a what?!
c. A: I put the beer in the oven.
B: You put what where?!
d. A: I think I just saw Albert Einstein at the supermarket.
B: You think you just saw (Albert) who?!4
4
Curiously, only the last name can be replaced by who. So we cannot have
(i) ∗ You think you just saw who Einstein?
(34) a. For $100,000, Martha Washington was married to who.

b. For $100,000, Martha Washington was married to which famous
American.
c. For $100,000, English is spoken in how many of the world’s nations.
A skeptical echo question has a strong rising intonation on the inter-

rogative phrase, indicating that the speaker is surprised at what has been
asserted. In a repair echo question, the intonation does not rise so sharply,
and the speaker is expressing an inability to fully understand the corre-
sponding part of what was asserted.
The quiz question, on the other hand, is a true question, in that it asks for
information. However, unlike a direct question, a quiz question is typically
asked when the person asking the question (such as a quizmaster) is pre-
sumed to know the answer. (Quiz questions are often used by classroom
teachers for precisely this reason.) So someone who genuinely wants to
know what your name is would not say “Your name is what?” but rather
“What is your name?”.
It might appear at first sight that these questions have interpretations that
contain the interrogative operator Q. As we have seen Q can correspond
to a direct question, or it can be embedded. But an echo question or a
quiz question cannot be embedded, which suggests that there is some other
operator that encodes the meaning of these constructions.
For example, the following cannot be a report about John’s surprise; it is
only the speaker’s surprise about that thing that is communicated.
(35) John asked whether you bought a what?!
John said that you bought a what?!
Similarly, the quiz question cannot have narrow scope.

∗ said
(36) The teacher (that) Martha Washington was married to who.
asked
This sentence cannot report the quizmaster’s question about Martha
Washington. It can only be understood as a quiz question about the person
that John said Martha Washington was married to. That is,
(37) For $100,000, John said that Martha Washington was married to who.
The intonation suggests that the proper interpretation of the quiz question
is that of a statement with a part left blank for the respondent to fill in.
Finally, a phrase such as a what or Albert who cannot be used in a direct
wh-question.
(38) a. ∗ A what were you looking at.
b. ∗ Albert who did you talk to?
For these reasons, we do not treat these constructions as wh-questions,

and in particular we do not assign to them a CS representation using the
operator Q. Possibly there are special operators ECHO and QUIZ that
appear only with wide scope at CS, for reasons having to do with their
discourse function. Research question 3 asks you to analyze such questions
using the analogy with the interpretation of topicalization in section 9.6.
9.2.4. Multiple wh-questions
As we have seen, there are wh-in-situ languages, like Japanese and Chinese.
A language like English has special wh-in-situ constructions such as echo
questions and quiz questions. In addition, English also has multiple wh-
questions, where only one wh-phrase appears in the scope or clause-initial
position, and the others appear in situ. Here are some English examples.
(39) a. Who was talking to who(m)?

b. Where did you put what?
c. What did you put where?
d. Which camper sleeps in which bunkbed?
(40) a. I was wondering [who was talking to whom].
b. Can you tell me [where you put what]?
c. Tell me [what you put where].
d. I can’t figure out [which camper sleeps in which bunkbed].
Multiple wh-questions have some curious properties. In English, one of

the wh-phrases must appear in initial position, and only one.
∗
(41) a. Who to whom was talking?
∗
b. To whom who was talking?
∗
c. Where what did you put?
∗
d. What where did you put?
(42) a. What did Sandy put where?
b. Where did Sandy put what?
∗
c. (Did) Sandy put what where? [with non-echo intonation]
Representing the mapping between syntax and CS in a multiple wh-

question requires an extension of the schema in (11) that is licensed by the
correspondence rules in (12). In that schema, the interrogative operator Q·
binds the variable · that corresponds to the gap. In the case of a multiple
wh-question, however, there must be several superscripts on the Q operator,
one for each argument or adjunct that is questioned. Only one of them
corresponds to a gap. In the diagram in (43), the clause-initial wh-phrase

what forms a A chain with the gap in the VP. 5
(43) SYNTAX S
NP NP VP
[WH]
Sandy V [e] where
what
put
GF Subject Object
CS Q·‚(PUT(AGENT:SANDY,THEME:·,LOC:‚))
Since we already have a correspondence rule for a wh-phrase in an A

position, we need to add to our correspondence rules one that deals with a
wh-phrase that is not moved.
(44) WH - IN - SITU :
SYNTAX S
.
.
.
.
XP
[WH]
CS Q·(…, ·, …)
The wh-phrase in situ must correspond to a variable that is linked to the

interrogative operator. Other correspondence rules determine the function
5
What also forms a chain with where, which is mediated through Q·‚ and ‚.
This is a different kind of chain, since there are no gaps. It captures the fact that
the interpretation of where is dependent on that of what – the sentence is asking, for
each thing ·, for the place ‚ where Sandy put ·.
9.3. RELATIVE CLAUSES 339
of the wh-phrase in the CS representation, e.g. there is a correspondence

rule that says where an expression that corresponds to Location goes, as in
(43).
The correspondence shown in (44) combines with the one in (12c),
repeated here, to capture the generalization that a wh-in-situ question has a
wh-phrase in initial position and the other wh-phrases are in situ.
(12) c. WH - QUESTION :
S
XP …
[WH]
Q·(…, ·, …)
Japanese also allows multiple wh-questions. Since Japanese does not have
the English correspondence rule (12c), none of the wh-phrases is necessarily
fronted. Example (45) illustrates.
(45) Taroo-ga doko-ni nani-o okimashita-ka.
Taroo-NOM where-DAT what-ACC put.PAST-Q
‘What did Taroo put where?’
Again, the marker -ka indicates that this sentence is a question and indicates
the scope of the question. Both wh-phrases have the indicated scope and
the CS representation is the same as it would be for the corresponding
English question, where the interrogative operator Q·‚ is bound to both
of the variables.
(46) Q·‚ (PUT(AGENT:TAROO,THEME:·,LOCATION:‚))
9.3. Relative clauses
9.3.1. Relatives with gaps
English relative clauses, like wh-questions, contain gaps.

(47) a. This is the cat [which you saw e]
b. This is the cat [that you saw e]
c. This is the cat [you saw e]
We will refer to the type of relative in (47a) as a wh-relative, the type in (47b)
as a that-relative, and the type in (47c) as a zero-relative.
Our intuition tells us that a relative clause is composed of two sentences
that share an argument. In (47) the main clause is This is the cat and the
relative clause is based on You saw the cat. The relative clause in this case is
a restrictive relative clause, in that it expresses a restriction on the cat that
distinguishes from other cats (it is the one that you saw, not some other one).
A non-restrictive relative clause adds a description but does not restrict the
set of things referred to, as in
(48) This is the cat, “Tiger”, which you saw.
(This is the cat, “Tiger”, and you saw it.) Non-restrictive relatives are also
called appositive relative clauses. Appositive must have the wh-form and are
typically set off by intonation breaks (written as commas).
Our concern here is with the internal syntax of the relative clause. Clearly
it is part of the NP that it modifies, as shown by the fact that the relative
clause stays with the NP when the NP is topicalized.
(49) [The cat which you saw], I really like ___ a lot.
It is also clear that there is a gap in the English relative clause corresponding
to the argument that the relative clause modifies. The standard analysis of
relative clauses is one in which the relative clause is a sister of the head of
the modified NP. Here is a first approximation. Again, we set aside questions
about the internal structure of the clause and simply call it S.
(50) NPi
Det Ni S
the cat (that) NP VP
you V [ei ]
saw
We can formulate the correspondence between the observed forms and

their CS representation straightforwardly. First, we elaborate CS so that a
proposition can serve as a modifier. We use PROPERTY as the label for this
CS constituent. 6 (51) represents the interpretation of cat that you saw.
6
MOD is a feature used in some HPSG analyses of modification.
(51) CAT· [PROPERTY:SAW(EXPERIENCER:YOU,THEME:·)]
This represents the meaning “the cat that has the property that you saw it”.
The crucial feature of this representation is the binding between CAT
and the THEME of SAW, notated by the index ·. This binding relation is
reflected syntactically by the gap in the relative clause. The correspondence
for a that-relative is shown in (52).
(52) SYNTAX NP
Det N S
cat that NP VP
you V [e]
saw
GF Subject Object
CS CAT·(PROPERTY:SAW(EXP:YOU,THEME:·))
Note that there is a chain from cat to [e] that is mediated through CS: cat
is linked to the CS representation CAT· , which binds ·, and · is linked to
Object and then to [e]. What makes this chain possible is that in CS, the
modifier of CAT that corresponds to the relative clause contains a variable
· that is linked to CAT.
A similar correspondence works for the zero relative, with a slight wrinkle
due to the fact that in standard English the zero relative is ungrammatical
if the relativized argument in the relative clause is a subject.
(53) a. Show me the cat that scratched you.

b. ∗ Show me the cat scratched you.
Exploring why this might be so is left for Problem 4.

In the wh-relative, both the head noun and the relative pronoun corre-
spond to the variable in CS. The noun and the pronoun must agree, e.g.
 
 which 
 
(54) a. the cat ∗ who you saw . . .
 ∗ where 
 
 
 ?which 
 
b. the singer who I was listening to . . .
 ∗ where 
 
 
 ∗
 which  
c. the town ∗ who I was living . . .

 where  
In this construction, both head noun and the relative pronoun head the
chain. The correspondence in (55) illustrates for the cat which you saw.
(55) SYNTAX NP
Det N S
cat which NP VP
you V [e]
saw
GF Subject Object
CS CAT·(PROPERTY:SEE(EXP:YOU,THEME:·))
We will see in the next section how this double binding of the variable works
in more complex cases where there is piedpiping in the relative clause.
9.3.2. Piedpiping in relative clauses
Piedpiping occurs in English relative clauses, and it is somewhat freer than

it is in wh-questions. The examples in (56) show that both p-stranding and
piedpiping of a PP are possible in relative clauses. The examples in (57)
show that an NP will piedpipe if its specifier is a relative pronoun, as in
(57a), but the specifier itself cannot be in an A position linked to a gap in
specifier position of the NP, as shown in (57b). This is a general restriction
on chains that we discuss in more detail in section 9.4, called the left branch
constraint.

that
(56) a. the cati I was looking at ei
whichi
b. the cati [at whichi ]j I was looking ei
(57) a. the cati [whosei tail]j I was looking at e j
∗
b. the cati [whosei ] I was looking at [ei tail]
The correspondence for piedpiping relative clauses is more complex than

the simple cases discussed in section 9.3.1. Example (58) shows the corre-
spondence for (56b).
(58) SYNTAX NP
Det N S
cat PP NP VP
P NP I V VP
at which was
V [e]
looking
GF Subject
CS CAT·(PROPERTY:LOOK-AT(AGENT:I,THEME:·))
The head noun cat corresponds to CAT· in CS, which binds the variable
·, the Theme of the relation LOOK-AT. The PP at which satisfies the
requirement that in a relative clause with a relative pronoun, the relative
pronoun must appear in a clause-initial constituent. This PP is linked to
CAT· which, because it binds ·, allows the chain between at which and the
gap in the VP.
The examples in (59) and (60) show that larger phrases can piedpipe as
well, if they contain a relative pronoun.

that
(59) a. the cati I was looking at a picture of ei
whichi
b. the cati [apicture of
whichi ] I was looking at e j
that
(60) a. the tablei the cat was sitting on the corner of ei
whichi
b. the tablei , [sitting on the corner of whichi ] was the cat
Note that piedpiping is more restricted in the corresponding wh-questions.

(61) a. At which cat were you looking e?

b. Whose tail were you looking at e?
c. ?A picture of which cat were you looking at e?
d. ?Sitting on the corner of which table was the cat?
Problem 5 asks you to explore the full range of piedpiping possibilities in

English relative clauses.
9.4. Constraints on chains
The first comprehensive study of syntactic constraints was the landmark

work of Ross 1967. Ross formulated his constraints in terms of movement,
and much subsequent work has taken the same perspective. 7 In our terms,
the constraints can be seen as constraints on the syntactic configurations
that permit gaps to appear inside of them.
Many configurations do not permit a gap within them that is CS-linked to
something outside them. Such configurations are called islands or extraction
islands. Here is a typical case that illustrates the difference between an
extraction island for and the result of copying.
∗
(62) Gone with the Wind is a movie whichi I have to call my friend [every time I
look at ei ].
Example (62) is ungrammatical for most speakers.

“Extraction” from a particular syntactic domain ‰ is understood to mean
that there is a A chain whose head is outside of ‰ and whose tail is a gap
(sometimes called the trace) that is dominated by ‰. The most significant
of Ross’s constraints for present purposes are the following. We have para-
phrased them into more contemporary terminology.
i. The wh-island constraint: Nothing may be extracted out of an indirect question.
Suppose that we have an indirect question, such as

(63) I wonder what Sandy said to Robin.
If we form a wh-question where the extracted wh-phrase corresponds to the

complement of to, the result is ungrammatical. The examples in (64) show
a few cases in which something has been extracted from a wh-question.
(64) a. ∗ Who did you wonder [what Sandy said e to e]?
7
However, the constraints can be viewed as applying to chains, which need not
be derived by movement. See, for example, Koster 1978 and Pollard and Sag 1994.
9.4. CONSTRAINTS ON CHAINS 345
b. ∗ [To Terry], [I wonder what Sandy said e e]

c. ∗ I bought a book [that I couldn’t find out [how much e cost e]]
The configuration that blocks this extraction is illustrated in (65) for exam-
ple (64a).
(65)
wh-island
S
who did you wonder
NP NP VP
what Sandy V [e] PP
said P [e]
to
CS-link
ii. The complex NP constraint: No element contained in a sentence dominated by

an NP may be extracted from that NP.
Constructions that illustrate this constraint include relativization, wh-

questions, and topicalization from relative clauses. In (66a), which forms
an A chain with e in the relative clause. In (66b) which book forms such a
chain, and in (66c) this book does. The structure of (66a) is shown in (67).
(66) a. ∗ the book [which I know the man [who wrote e]]
b. ∗ Which book do you know the man [who wrote e]
c. ∗ This book, I know the man [who wrote e]
(67) the book [ which I know NP ]
complex NP
Det N S
the man NP VP
who V [e]
CS-link
While all of these cases of extraction are ungrammatical, note that the
extraction of a subject from a relative clause is worse than the extraction
of a non-subject. Compare the examples in (68) with those in (66). (We’ve

underlined the extracted phrase and its gap in subject position.)
(68) a. ∗ the man [who I read the book [which e wrote e]]
b. ∗ Who did you read the book [which e wrote e]
c. ∗ The man, I read the book [which e wrote e]
The complex NP constraint also applies to the sentential complements of

nouns like report.
(69) a. ??the dog [which I read a report [that you entered e] in a contest]]
b. ??Which dog did I read a report [that you entered e in a contest]
b. ??That dog, I read a report [that you entered e in a contest]
Extraction from a sentential complement is more acceptable than extraction

from a relative clause.
iii. The left branch constraint: The specifier of NP or AP cannot be extracted from
NP or AP, e.g.
The examples in (70) illustrate the consequences of extracting the specifier

from its phrase.
∗
(70) a. Whose were you looking at [e picture]?
∗
b. [How many] will you buy [e cars]?
∗
c. How will you buy [[e many] cars]?
∗
d. W hich did you see [e movie]?
∗
e. How is your son [e tall]?
As expected, these examples are saved by piedpiping.

(71) a. [Whose picture] were you looking at?
b. [How many cars] will you buy?
c. [How many cars] will you buy?
d. [Which movie] did you see?
e. [How tall] is your son?
The configuration that the left branch constraint applies to is exemplified in

(72) for (70b).
(72)
whose were you looking at NP
[e] N
picture
CS-link
9.4. CONSTRAINTS ON CHAINS 347
iv. The sentential subject constraint: No element can be extracted from an S that is
itself a subject.
This constraint is exemplified by the examples in (73)–(74).

∗
(73) a. Who does [that I voted for e] disturb you?
∗
b. Where is [that I put the money e] obvious?
∗
(74) a. Who would [for me to vote for e] be appalling?
∗
b. Where would [for me to put the money e] be safest?
When these sentential subjects are in an extraposed position, and no longer

in a syntactic subject configuration, extraction is not problematic.
(75) a. Who does it disturb you [that I voted for e]?
b. Where is it obvious [that I put the money e]?
(76) a. Who would it be appalling [for me to vote for e]?
b. Where would it be safest [for me to put the money e]?
The configuration that this constraint applies to is exemplified in (77) for

(73a).
(77)
who did S disturb you
that NP VP
I V PP sentential subject
voted P [e]
for
CS-link
v. The coordinate structure constraint (CSC): In a coordinate structure, (a) no

conjunct may be moved, (b) nor may any element contained in a conjunct be
moved out of that conjunct.
There are four distinct cases that fall under this constraint, depending on
whether the extracted phrase is a right or left conjunct, and whether it is the
conjunct itself that is extracted or a constituent of the conjunct. In (78a)
we see that it is impossible to extract the entire left conjunct, and in
(78b) we see that it is impossible to extract the entire right conjunct. (79a)
shows that it is impossible to extract from a left conjunct, and (79b) shows
that it is impossible to extract from a right conjunct.
∗
(78) a. Who did you see [e and Sandy]?
∗
b. Who did you see [Sandy and e]?
∗
(79) a. Who did you see [[a picture of e] and the latest movie]?
∗
b. Who did you see [the latest movie and [a picture of e]]
There are several different configurations that fall under this constraint. We
illustrate them in (80) using conjoined NPs.
(80)
who did you see NP
[e] and Sandy
who did you see NP
Sandy and [e]
who did you see NP
NP1 and NP2
a picture of [e] the latest movie
who did you see NP
NP1 and NP2
the latest movie a picture of [e]
Interestingly, extraction from conjuncts is not always ungrammatical. The

example in (81) shows that when there is parallel extraction out of both
9.5. THE THEORY OF WH-MOVEMENT 349
conjuncts, that is, across the board (ATB) extraction, there is no violation of
the CSC.
(81) Who did you see [[a picture of e] and [the latest move about e]]?
The original perspective on constraints such as these, and one that is still
quite prominent, is that they are universals of language. On this view, they
are not acquired by language learners through experience but form part
of the knowledge of language that language learners are born with. The
constraints are not associated with particular rules of grammar but apply
to all rules.
If the constraints are part of the human capacity for language, we expect
that learners would know them even in the absence of actual evidence
about them. In fact, there is little evidence that learners actually acquire
knowledge of the constraints over time. They do not appear to make errors
that would suggest that at some point they do not have this knowledge. See
Guasti 2002 for a recent survey.
While the constraints do in fact appear to be widely applicable, research
has shown that there are certain cases where they do not apply, raising
questions about their universality. We will review more recent work on
constraints in the next section, and then turn to the counterexamples and
apparent counterexamples.
9.5. ∗ The theory of wh-movement
The question of whether or not movement is the correct way to account

for A chains has been a persistent one in syntactic theory. The idea of
“movement” as a way of relating two positions in a sentence, one of which
is a gap, is an intuitively appealing one. On the other hand, “movement”
as an operation on syntactic structures involves particular computational
operations. If what is essential to an A construction is the A chain con-
figuration itself and not how it is derived, then movement does not play an
essential explanatory role. But if the movements per se and their interactions
contribute to an understanding of why some sentences are grammatical
and others are not, then there will be evidence for movement beyond its
intuitive appeal as a metaphor for the relationship between two positions in
a sentence.
In this section we look at how movement analyses account for the prop-
erty of A constructions. While the movement approach to A constructions
is by no means universally accepted or without its difficulties, it is so per-
vasive that it is important to recognize its influence on the development of
syntactic theory and its present form. In this and the remaining sections of
this chapter we sketch out the basic properties of the movement analysis
and consider how to apply it in those cases where it appears that there
is no plausible constituent that can be analyzed as moving. We look at
some particular problems that arise in the analysis of topicalization as
a movement construction, and consider a number of problems with the
standard view of constraints on extraction.
9.5.1. Basics of wh-movement
9.5.1.1. Chains
Mainstream syntactic treatments capture the properties of wh-questions
by positing movement. A simple movement analysis assumes a sequence of
syntactic structures such that the first structure has the wh-phrase in its
canonical position, and the next structure has the wh-phrase in a clause
initial position. Here is an illustration. The initial structure is D-structure,
which we introduced in Chapter 6, section 6.8.2.
(82) D-structure: [[ ][Sandy [buy what]]
Derived structure: [[what][Sandy [buy [e]]
A complex movement analysis envisions a more complex sequence of steps
through which the wh-phrase moves on its way to the clause initial position.
We will focus on the simple movement analysis here and take up complex
movement in section 9.7.1.
In standard derivational treatments, the A chain formed by movement is
defined in terms of the syntactic structure alone, without reference to CS.
In this case, coindexing of the constituents plays the role of linking the two
constituents. We call such a chain a movement chain.
Movement chain
Two constituents · and ‚ form a movement chain when
(a) · c-commands ‚ and
(b) · and ‚ are coindexed and
(c) ‚ is empty.
The gap in the wh-movement chain, ‚, is generally assumed to be a coin-

dexed trace of the moved wh-phrase. The trace is represented using the nota-
tion t, or by a full copy that is marked to indicate that it is phonologically
empty, as in (83). Indices are conventionally used to show the relationship
between a moved constituent and its trace.
(83) D-structure: [[ ][Sandy [buy whati ]]]
a. Derived structure: [[Whati ][Sandy [buy t–i ]]]
or
b. Derived structure: [[Whati ][Sandy [buy what–i ]]]
Compare this formulation of movement chain with our account of CS-

chain (see section 9.1). The two definitions are sketched in (84).
(84) a. CS-chain
SYNTAX … XP […, [e], …]
CS Q·(F(…, ·, …))
b. Movement chain
SYNTAX . . . XPi [. . . , ti , . . . ]
In the CS formulation, the two syntactic parts of the chain are linked to a
set of coindexed elements in the CS representation. One link is through the
bound variable · (corresponding to the gap) and the other is through the
operator, corresponding to the wh-phrase. This connection does the same
work as direct coindexing of the two syntactic parts of the movement chain,
which is a strictly syntactic representation that does not make use of the CS
representation.
The arguments for treating a trace as an invisible copy as in (83b) consist
primarily of evidence that moved wh-phrases behave as though they are in
situ with respect to a range of binding relations (see Chapter 10, section
10.5.3).
9.5.1.2. Structure preservation in A movements

Wh-questions in classical transformational grammar involve a single move-
ment to clause-initial position. It is normal practice on this approach to
identify the position to which something moves by an empty landing site.
The assumption that such a landing site exists is motivated by the gen-
eral restriction that movements must relate two independently motivated
positions in a structure, and cannot create structure. This assumption is
a version of the structure preserving hypothesis, which we introduced in
Chapter 6, section 6.8.3.
Consider in this light the movement analysis of the passive, which is an A
movement in derivational accounts. Recall that the direct object (and more
generally, an NP immediately following the V), moves from object position
into an empty subject position. Such a position is independently licensed by
the grammar of English; thus, the movement is structure preserving. It does
not build structure that is not already assumed to be there.
(85) [ ] . . . be see +en Sandy ⇒ [Sandy] . . . be see +en ___
X theory makes available a similar empty position for wh-movement,

given two important assumptions: (i) the full sentence is the maximal pro-
jection of C0 and (ii) every projection contains (or may contain) a Spec. In
this case, the projection of C0 is CP, and the specifier of CP is the initial
position in the clause. (This is the structure that we discussed in Chapter 4,
section 4.7.1.)
(86) CP
Spec C
C0 IP
[WH]
NP I
you I0 VP
V NPi
saw whoi
[WH]
This derivation is motivated in part by the assumption that syntactic struc-

ture should be sufficient to define the linear properties of sentences (see
Kayne 1994), and in part by the goal of making the structures used in
syntactic description as uniform as possible (see Culicover and Jackendoff
2005: Chapters 2 and 3).
In what follows, when we refer to the wh-movement analysis we will
assume for concreteness that in the English wh-question there is a wh-
phrase in Spec of CP and a trace, notated as t, in the canonical position
of the wh-phrase. A number of theoretical issues that have arisen around
this general approach, including whether the trace has internal structure,
are taken up elsewhere in this chapter and in Chapter 10.
9.5.2. Feature discharge
The question of why the wh-phrase moves to the clause-initial position,

and not to some other logically possible position, has received a range of
answers in MGG over the years. The contemporary view is that the com-
plementizer in a question, C0 , has a feature that must agree with a feature
of a constituent in specifier position. The analogy here is with agreement
between the inflectional head I0 and its specifier, which is the subject.
The technical device for representing agreement between constituents is
to assign the same feature values to them. For example, in the present tense
the subject in English agrees in number with the verb.

is
(87) a. The dog ∗ barking.
are

∗
is
b. The dogs barking.
are
(88) IP
NP I
[·NUM]
I0 VP
[·NUM]
The agreement relation is represented by requiring that the value of the

two features be the same. In the case of (88), · is either SG or PL, for
example. The feature on the head is said to be discharged in virtue of the

agreeing constituent in the Spec position. Discharging a feature satisfies
the agreement requirement. The configuration exemplified in (88) is one of
Spec-head agreement.
In the case of the wh-question, it is assumed that the complementizer has
a feature, call it [WH]. This feature on C0 identifies the CP as interrogative.
The feature is discharged if a wh-phrase is moved to its specifier.
(89) CP
Spec C
NPi C0 IP
[WH]
whoi NP I
[WH]
you I0 VP
V NPi
saw whoi
[WH]
C0 [WH] is said to be the trigger of the wh-movement.

Feature discharge has the following properties:
r the feature to be discharged is associated with a functional head, call it X0 ;
r in order to be discharged, there must be a phrase in the Spec position of the
projection XP of X0 ;
r the feature must be discharged at some point in the derivation, otherwise the
structure is ill-formed and the sentence is ungrammatical.
Some features are discharged through visible movement. An influential

theoretic proposal that has been explored in some detail over the years is
that other features are discharged through invisible or covert movement. We
discuss the motivations for covert movement in the next section.
Recall from our discussion of multiple wh-questions (section 9.2.2) that

in English exactly one wh-phrase may appear in clause-initial position in a
wh-question, and that one wh-phrase must appear in this position. These
facts appear to follow directly from the assumption that there is a C0 [WH]
whose feature is discharged by the presence of a wh-phrase in the specifier
position. The feature must be discharged, and it can be discharged only
once, through movement of a wh-phrase to Spec of CP. The requirement
that there be a wh-phrase in the Spec of C0 of a wh-question in a language
like English was originally called the wh-criterion, and is still informally
referred to by this term.
9.5.3. Covert movement
Let us consider now how the movement approach deals with wh-in-situ.
In early versions of movement theory, it was assumed that there were in
fact wh-in-situ constructions. In more recent versions of the theory (see
for example May 1985), it is assumed that wh-in-situ should have the
same basic properties as wh-movement. It is triggered by the interrogative
complementizer, represented as C0 [WH], and involves the movement of a
wh-phrase to Spec of CP. This view is a corollary of the widespread uni-
formity assumption in MGG that, to the extent possible, the same kinds of
constructions have shared syntactic properties (structural uniformity) and
derivations (derivational uniformity) within and across languages.
There is an obvious contradiction between the uniformity assumptions
and the fact that we do not see any overt displacement of wh-phrases in
wh-in-situ languages. One of these two possibilities must hold, given the
uniformity approach:
(i) C0 [WH] is “weak” in wh-in-situ languages, in that it does not have to be dis-
charged, or
(ii) C0 [WH] must be discharged in wh-in-situ languages, but somehow without overt
movement of a wh-phrase to the Spec of CP.
Alternative (i) was explored in early versions of this theory but not fully
developed, and we will not discuss it further here. Alternative (ii) is the
currently accepted view in MGG. On this view,
(a) there is movement of the wh-phrase to Spec of CP at a level of syntax that does
not correspond to word order, or
(b) there is a constituent of the wh-phrase with the feature [WH], or perhaps the
feature itself, that moves invisibly to Spec of CP, or
(c) there is a way in which C0 [WH] can be discharged by the wh-phrase in situ
without requiring any movement.
We discuss the first two possibilities in the remainder of this section, since
we are concerned here with analyses that assume covert movement. Option
(c) is a variant of (26), which assumes a direct correspondence between the
superficial structure of wh-in-situ and a CS in which there is an interrogative
operator with scope over the sentence. We leave discussion of the technical
details of this alternative to Problem 7.
The (a) approach utilizes movement in Logical Form (LF), a level of
syntactic representation that systematically represents those aspects of syn-
tactic structure that bear on the logical properties of sentences. Crucially,
movements that apply to LF representations do not have any affect on
the way a sentence is pronounced, only on its meaning. LF was originally
motivated to account in syntactic terms for scope ambiguities of the sort
illustrated in (90).
(90) Everyone in this room speaks two languages.
On one interpretation, for each person there are two languages that that
person speaks. On the other interpretation, there are two languages that
everyone speaks. Syntactic movement of the quantifier phrases in the LF
representation so that they are in different positions with respect to one
another in LF corresponds to the ambiguity.
(91) a. everyonei [[two languages]j [IP t i speaks t j ]] (that is, everyone is such that
there are two languages that he speaks)
b. [two languages]j [everyonei [IP t i speaks t j ]] (that is, there are two languages
such that every person is such that he speaks them)
These two syntactic representations correspond to the standard representa-

tions of quantifier scope in quantificational logic – they are in effect a way of
expressing particular logical properties of sentence (90) using devices from
syntactic theory, such as phrase markers and movement.
On this approach, a syntactic feature that does not have to be discharged
before phonological realization (currently called Spell Out) may remain
active and trigger movement in LF. This is the standard account of wh-
in-situ in MGG. On this view, covert wh-movement should be just like
overt wh-movement except that it has no phonetic consequences. We take
up in section 9.7.2 the question of whether covert movement is subject to
constraints on movement.
Schematically, the derivation of a sentence assuming LF movement fol-

lows the so-called T-model of Government Binding and Principles and
Parameters theory, where the movements that apply prior to Spell Out have
an effect on the Phonetic Form (PF) of the sentence, while those that apply
after Spell Out have an effect only on the representation of the meaning
(that is, at LF). 8
(92) D-structure
movement
S-structure
Spell Out movement
Phonetic Form (PF) Logical Form (LF)
Once it is assumed that there exists a level of syntactic representation in

which movement can take place after phonological form is derived but prior
to semantic interpretation, it is possible to use this level to account for other
aspects of meaning that do not explicitly correspond to superficial syntactic
structure. For example, the fact that a wh-in-situ has sentential scope (as in
who saw what) can be represented by moving the wh-phrase to the Spec of
C0 [WH] in LF. Such movement has consequences for the interpretation but
not for the observed order of constituents.
Alternative (b) assumes that there is an invisible interrogative specifier,
call it PRO[WH], that in some languages can undergo wh-movement in syn-
tax prior to Spell Out. Crucially, this specifier moves but does not piedpipe
the phrase of which it is a specifier.
8
A more contemporary view (the Minimalist Program), in the context of a
derivational syntactic theory (that is, one that makes crucial use of movement), is
that there are no distinct levels of D-structure and S-structure; rather, structures are
mapped to PF and LF as they are built up. The operation of forming phrases is
called Merge. On this view, the branching shown in (92) does not pertain to distinct
levels of representation but to individual representations as they are built up piece
by piece.
(93) CP
Spec C
PRO C0 IP
[WH]
… XP …
PRO X
[WH]
…
X0
It thereby discharges C0 [WH]. However, since PRO[WH] has no phono-

logical form, there is no phonological consequence of its movement, so it
appears that there is no movement. Since it does not cause piedpiping, the
entire phrase does not move.
9.5.4. Movement in relative clauses
Relative clauses with gaps lend themselves naturally to the movement analy-
sis as well. Such an approach must be concerned with what is in initial
position and its relationship to the gap.
A number of questions arise on such an analysis. Recall our observation
in section 9.3 that in English there are three types of relative clauses: those
with relative pronouns in clause-initial position (wh-relatives), those with
that in initial position (that-relatives), and those with neither in initial
position (zero-relatives).
r why does the relative pronoun move in the wh-relative?

r what happens to the complementizer that in the wh-relative and the zero-
relative?
r what produces the gap in a that-relative?
r what produces the gap in a zero-relative?
The first question may be answered by assuming that the complementizer

is C0 [REL], that the feature [REL] must be discharged through Spec-head
agreement, and that a relative pronoun has the feature [REL]. The derived
structure of the relative clause will then be (94).
(94) N [CP NPi [REL] [C0 [REL] [IP . . . t i . . . ]]]
It is then necessary to stipulate that the complementizer that does not

appear when there is a wh-relative pronoun in the specifier position of
C0 . Without this stipulation, we would be claiming, incorrectly, that the
following would be grammatical.
(95) a. ∗ the woman who that I saw

b. ∗ the cake which that I ate
It is of interest to note in this connection that in a number of languages it

is actually possible for a wh-form in clause-initial position to coexist with an
overt complementizer. (Examples b–g below are taken from Bayer 2002:2.)
(96) a. Middle English

whan that Aprile with his shoures soote, . . .
‘When April with his sweet showers, . . . ’
[Chaucer, Prologue to the Canterbury Tales]
b. Middle English
men shal wel knowe who that I am
‘One shall well know who I am’
c. Middle High German
nu hoert . . . wa daz er mir lougent niht aller mîner leide
now listen what that he me denies not all my pain
‘Now listen how much of my pain he denies’
d. Substandard French
Je ne sais pas quand que Marie arrivera
I NE know not when that Marie arrive-FUT
‘I don’t know when Marie will arrive’
e. Substandard Italian
Non so quando che Mario arriverà
not know.1 SG when that Mario arrive-FUT
‘I don’t know when Mario will arrive’
f. South-Thuringian (German)
West duu öpper, wi lang daß di walt beschtenna wörd?
know you perhaps how long that the world exist will
‘Do you know how long the world will last’
[Schleicher 1858]
g. Bavarian (German)
Frog’s doch, wia lang daß’s no dobleim woin!
ask-them PRT how long that-they still stay want
‘Ask them how long they still want to stay!’
Note that these are interrogative, not relative forms. In Bavarian similar
combinations may appear in relative clauses (Bayer 1984:216).
(97) Bavarian
a. Die Lampen (die) wo i g’seng hob wor greißlich
the lamp.NOM which.ACC that I seen have was ugly
‘The lamp that I saw was ugly.’
b. Der Mantl den wo i kafft hob
the coat which that I bought have
‘The coat which I bought’
c. Des Audo (des) wo i mecht is z’teia
the car.NOM which.ACC that I like is too-expensive
‘The car which I would like is too expensive’
This data supports the view that, at least in some languages, the clause-
initial position has a structure in which an initial constituent can appear
before an overt complementizer. (On a CP analysis, the structure would be
[CP Spec-C0 . . . ].) If this is the structure in English, it would be necessary to
stipulate that sequences like which that are excluded. Problem 11 asks you
to state this restriction as precisely as possible. 9
Somewhat more problematic is the analysis of that-relatives and zero-
relatives. Although there is a gap, nothing appears to have moved in these
constructions. The standard approach to such a state of affairs in MGG
is to posit the existence of a covert (that is, invisible) counterpart to the
overtly moved constituent, in this case a relative pronoun. The movement
of the covert constituent leaves a gap. 10
Most movement analyses of these relative clauses have assumed that there
is a special empty operator called OP. This operator is invisible, and has the
feature [REL]. The analysis would then be
(98) N [CP OPi [REL] [C0 [REL] [IP . . . t i . . . ]]]
There are some problems with this analysis that arise in connection with
piedpiping, which are taken up in Problem 2.
9
Shakespeare also uses the which: Our bodies are our gardens, to the which our
wills are gardeners. Othello, I(3)322.
10
Of course, since it is invisible if it didn’t move we would still see what appears
to be a gap in the argument position. However, in this case the feature [REL] would
not be discharged.
9.6. TOPICALIZATION 361
9.6. ∗ Topicalization
9.6.1. Basic structure
In informal English, a non-wh-constituent may appear in clause initial posi-

tion instead of its normal position in a construction called topicalization.
The fronted constituent is called the topic. Some examples are given in (99).
(99) a. John, I like e.
b. Things like that, I can never get used to e.
Constituents other than NPs can be topicalized in English. The following
examples show topicalization of VP, AP, and PP.
(100) a. They said that Sandy would win the election, and [VP win the election]i
Sandy did e.
b. They said that Sandy would be successful, and [AP successful] he has
been e.
c. [PP On the table] we put the groceries e, and [PP in the refrigerator] we put
the beer e.
In many cases, it is possible for the topic to originate in a lower clause.
(101) a. John, I would claim that many people dislike e.
b. Things like that, no one would admit that he had actually seen e.
c. ?They said that Sandy would win the election, and [VP win the election] I
suspect Sandy will e.
d. They said that Sandy would be successful, and [AP successful] I would say
he has been e.
And topics may precede a wh-phrase.
(102) To Sandy, what are you planning on giving e e?
One of the primary functions of topicalization in English is to mark a

constituent as “given” or “old” with respect to a discourse. As in the case
of wh-questions, our description of the correspondence between syntactic
structure and meaning shows the correspondence between the topicalized
constituent and its interpretation. For simplicity, we assume that in addition
to CS there is also an information structure (IS) representation that encodes
the discourse properties of the sentence. The topic corresponds to the
discourse function TOPIC, which is linked to a variable in CS, as illustrated
in (103). 11
11
An alternative notation would be one in which there is a TOPIC operator that
is analogous to Q. For expository purposes we have chosen to keep the CS and IS
parts of the interpretation apart.
(103) SYNTAX S
XP … [e] …
CS F(…, ·, …)
IS TOPIC·
When this correspondence applies to a full sentence, a CS-chain is formed.
(104) SYNTAX S
NP NP VP
John I V [e]
like
CS LIKE(EXP:ME,THEME:·)
IS TOPIC·
9.6.2. Topicalization as movement
Movement is the standard way in MGG of deriving the A chain (103) in the
case of topicalization. Just as wh-movement is triggered by C0 [WH] and the
movement of relative pronouns by C0 [REL] on this approach, a comparable
account of topicalization has to posit a triggering feature; let us call it [TOP].
9.6. TOPICALIZATION 363
A problem with topicalization as movement bears on the question of

where the topicalized constituent is attached. Assume that there is a com-
plementizer C0 [TOP] that triggers movement of the topic to its specifier. The
following would be the structure of Sandy, I like.
(105)
CP
Spec C
NP C0 IP
[TOP] [TOP]
NP I
Sandy
I I0 VP
V0 NP
like Sandy
One problem with this analysis, which is well documented in the syntax
literature, is that topicalization in embedded sentences places the topic after
the complementizer, not before it.
(106) a. Terry claims [that Sandyi , I like t i ].

b. ∗ Terry claims [Sandyi , that I like t i ].
This would not be problematic if the structure of topicalization was that of

(107), but (107) loses the parallelism with wh-movement, and the attach-
ment is not structure preserving.
(107)
CP
Spec C
C0 IP
(that) NP NP I
Sandy I I0 VP
V0 NP
like Sandy
In order to maintain the parallelism, it has been proposed that, in embedded

clause topicalization, there are two complementizers, the one realized as
that and a lower C0 [TOP].
(108) VP
V CP
Spec C
C0 CP
that Spec C
C0 IP
[TOP]
On the other hand, if the lower C0 is [WH] and the higher one is [TOP],
we will have one landing site for the wh-phrase and one for the topic for
examples like (102).
(109) CP
Spec C
C0 CP
[TOP]
Spec C
C0 IP
[WH]
Such an analysis is called iterated CP, and raises questions such as

why there can only be one that, one topicalized constituent, etc. See
Exercise 11. 12
12
A recent proposal due to Rizzi 1997, which we discuss in Chapter 11, assumes
that a topic goes into the specifier position of a functional head TOP that is distinct
9.7. MORE ON CONSTRAINTS 365
A somewhat more subtle problem is related to the idea that movement

is triggered by the need to discharge a feature. This idea is somewhat
plausible in the case of wh-questions, because of the possibility that the
overt realization of the wh-complementizer when there is no movement to
Spec of CP is whether. (This is not entirely without problems, however – see
Problem 12.) But there is no such complementizer in English for topics and
hence no overt morphological evidence to lend support to the analysis.
9.7. ∗ More on Constraints
In section 9.7.1 we review very briefly proposals to unify Ross’s constraints,

which were summarized in section 9.4. In section 9.7.2 we consider a range
of phenomena that suggest that these constraints, however they are formu-
lated, admit of exceptions under certain conditions.
9.7.1. Conditions and Barriers
As we saw in section 9.4, Ross’s constraints are descriptions of various

syntactic environments that do not permit extraction. Chomsky (1973), in
his paper “Conditions on transformations”, proposed what is known as
the Conditions Framework. Chomsky sought to explain Ross’s observations
by identifying a simple common basis for the various constraints. In the
1986 book Barriers, Chomsky proposed what is known as the Barriers
Framework, which is a further development of this approach. Related ideas
can be found in the current Minimalist Program (Chomsky 1995; Chomsky
2000).
The key idea of this approach is all movements are local. How one defines
locality is a technical matter that depends on what one assumes about
the syntactic details, and there have been various formulations over the
years.
One persistent idea is that local movement is movement within a sin-
gle phrasal projection, such as S(CP) or NP. The complex NP constraint
falls under this general idea of locality. For instance, to extract from a
from C0 . This proposal maintains the strong view that all A positions are specifiers
of functional heads.
relative clause, we would have to move a constituent out of a CP and

an NP.
(110) you have [NP a book [CP that describes whati ]] →

∗
whati you have [NP a book [CP that describes t i ]]
The effect is what was termed subjacency in the Conditions Framework: a

constituent cannot be moved out of two (or more) phrasal domains (that is,
S or NP) in one movement.
On the other hand, in MGG analyses there are cases in which it
appears to be necessary to move a constituent out of one sentence into
a higher sentence. For example, so-called raising to subject moves the
subject of an infinitival complement to the subject position of the higher
clause.
(111) Terry seems [S t to be angry]
For this reason, it was proposed that movement out of a phrase is possible
only from the left edge of the phrase (Chomsky 1973). The subject of the
complement is in fact at the left edge in (111), for example. But in a case
such as (112), it is not on the left edge, and it cannot be raised.
∗
(112) Terry seems [S that t is angry]
[cf. It seems that Terry is angry.]
There are, however, cases of apparent “long movement” out of a comple-

ment that seem to violate the condition that extraction from a sentence can
only be from the edge of the sentence. For example,
(113) Whoi did you say [CP that you were talking to t i ]?
If this condition is correct, it is necessary to analyze such cases as involv-

ing successive short movements. First, we move to the left edge of the
embedded sentence, and then out to the edge of the next sentence above.
Since X theory is extended to CP, it is possible to identify Spec of CP
as an “escape hatch” through which a constituent can move out of a
clause.
A schematic picture of how such extraction works for wh-questions and
other long distance A constructions is given in (114).
(114)
CP2
Spec C
C0 IP2
NP I
2 you I0 VP
V0 CP1
say Spec C
C0 IP1
1 that NP I
you I0 VP
V PP
talking P NP
to who
If something moves into the lower Spec, it must be stipulated that it can only
move from there to the higher Spec, and not into an argument position of
the higher clause. This type of analysis is often referred to as successive cyclic
movement, and is a central component of all contemporary derivational
treatments of A constructions. 13
In the Barriers Framework, Chomsky 1986 sought to derive the con-
straints from the basic notion of barrier. Intuitively, a barrier is a max-
imal phrasal node (also called a cyclic node) such as CP and NP. On
this approach, something can move across one barrier at a time but not
across more than one barrier, which produces the effect of the complex NP
constraint and allows successive cyclic movement.
However, it is not possible to extract from an adjunct, even when there
is only one maximal node. (This is sometimes called the adjunct island
constraint or the condition on extraction domains (CED), after Huang
1982.)
13
Because Spec of CP was called COMP in earlier syntactic analyses, it is
sometimes still referred to as COMP-to-COMP movement.
(115) a. ∗ Whoi did you call me [CP because I was planning to talk to t i ]?
b. ∗ They said I would buy that car, and [buy that car]i I went downtown [CP
because I wanted to t i ].
The Barriers Framework attempted to capture this fact by stipulating that

a node is not a barrier if it is a complement of a lexical head, such as V
or P. 14 For example, the CP complement of say in (114) is not a barrier,
while the CP headed by because in (115) is. (This is the basic difference
between the cases, but the actual computation of where the barriers are in
these structures and the formulation of the constraints in terms of barriers
in Chomsky 1986 is complex, and we do not try to reproduce it here.)
Another important constraint concerns the fact that a subject in English
cannot be extracted from its clause if it is immediately adjacent to its
complementizer. The classic examples, which gave rise to the name that-
trace or that-t effect, are as follows.
(116) a. Whoi did you say [t i called]?
b. ∗ Whoi did you say [that t i called]?
An intervening adverb ameliorates this effect –

(117) a. Whoi did you say [that most recently t i called]?
b. a person whoi I said [that under no circumstances t i should be allow to
participate]
This is called the adverb effect (Culicover 1993).

Gaps adjacent to other complementizers are as bad or worse than
those with that; these are said to illustrate the complementizer-trace
(complementizer-t) effect.
∗
(118) a. Whoi did you wonder [whether t i called]?
∗
b. Whoi did you wonder [if t i called]?
∗
c. Whoi did you wonder [whatj C0 [WH] ti said t j ]?
∗
d. Whoi did you ask [for t i to accompany you]?
The cases in (118a–c) not only show the complementizer-t effect but vio-
lations of the wh-island constraint. Example (118d) is included on the
assumption that for is a complementizer of infinitival embedded sentences.
The that-t effect does not apply at the highest node of a relative clause
when the complementizer is that. Consider the following example.
(119) the person that called
14
The actual formulation of the constraints in terms of barriers is somewhat
more complex than suggested here.
The movement analysis of section 9.5.4 assumes that the subject of the
relative clause is the empty operator OPi , and it moves to the Spec of C0 ,
which is that. So the relative clause has the following structure.
(120) (the person) [CP OPi that [IP t i called ]]
There is clearly a sequence that-t here, but it is not problematic. There have
been a number of proposals to deal with this apparent anomaly, but the
issue is still very much an open one in any analysis which posits a trace, or
the equivalent, in the subject position of the relative clause. Note that it is
not sufficient to say simply that that that-t effect does not arise when the
moved constituent is the empty operator, because it does occur in sentences
like the following.
(121) the person OPi that I believe [(∗ that) t i called]
So it appears that that-t is grammatical when that is the C0 of a relative

clause.
Chomsky 2000 proposes the notion of phase, which recapitulates many
of the earlier ideas of successive cyclic movement. Movement applies locally
within an NP or CP, which then may form part of a larger structure. A con-
stituent of a phase is not accessible to later operations unless it is on the left
edge – this is the phase impenetrability condition. Again we get the result that
successive movement to the left edge of a CP allows extraction (as in (114)),
but other extractions are blocked. It does not appear that this approach can
account for the CED or the that-t effect without additional stipulations.
9.7.2. Violability of constraints
As we noted in section 9.4, one of the most significant accomplish-

ments of contemporary syntactic theory is the observation that apparently
unbounded dependencies like wh-questions are subject to constraints. Hav-
ing observed that these constraints exist in natural language, questions
naturally arise about where they come from, and whether they are univer-
sal. While the constraints do appear to account for phenomena in many
languages, there are also exceptions.
9.7.2.1. Complex NP constraint

A few possible exceptions to the complex NP constraint have been found in
English and many have been found in Swedish. In the English examples in
(122), there appears to be extraction from relative clauses that are relatively
acceptable for some speakers.
(122) a. ?This is a dogi [OPi that I know a lot of people [who are afraid of t i ]
b. ?I just finished reading a booki [OPi that I once saw a movie [that was
very loosely based on t i ]]
In Swedish, extraction from a relative clause appears to be even freer. The

following examples are technically violations of the complex NP constraint,
but they are judged to be grammatical. The basic sentence is given in (123a)
– notice that there are two NPs in the relative clause, en pojke “a boy” and
en kyss “a kiss”, and one PP på Röda Torget “in Red Square”.
Example (123b) shows that the NP en flicka “a girl” can topicalize.
Swedish is a so-called V2 language, which means that the tensed verb
appears in second position in the main clause. As examples (123c–e) show,
the three constituents of the relative clause may also topicalize to the initial
position of the main clause, producing the V2 word order.
(123) Swedish [Allwood 1976:6] 15

a. Jag känner till en flicka [som gav en pojke en kyss på
I know of a girl who gave a boy a kiss in (the)
Röda Torget]
Red Square.
b. En flickai känner jag till t i [som gav en pojke en kyss på
A girl know I of who gave a boy a kiss in
Röda Torget]
Red Square
c. En pojkei känner jag till en flicka [som gav t i en kyss på
A boy know I of a girl who gave a kiss in
Röda Torget]
Red Square
d. En kyssi känner jag till en flicka [som gav en pojke t i på
A kiss know I of a girl who gave a boy in
Röda Torget]
Red Square]
e. På Röda Torgeti känner jag till en flicka [som gav en pojke en
In Red Square know I of a girl who gave a boy a
kyss t i ]
kiss
15
We omit examples of wh-questions and relative clauses that demonstrate the
same point.
The Swedish examples, while marginal for some speakers, are better than
the corresponding English examples, which are completely ungrammatical.
(124) a. ∗ The boy, I know of a girl who gave a kiss (to) in Red Square.
b. ∗ A kiss, I know of a girl who gave (to) a boy in Red Square.
c. ∗ In Red Square, I know of a girl who gave a kiss to a boy.
Another very different class of violations of the complex NP constraint

concerns languages in which there are no gaps in wh-questions and relative
clauses. As noted earlier in this chapter Chinese and Japanese have wh-in-
situ, which means that the wh-phrase is not in an A position. A Chinese
example of wh-in-situ is given in (125).
(125) Chinese
hufei mai-le shenme (ne)
Hufei buy-PERF what PRT
‘What did Hufei buy?’
[Cheng and Rooryck 2000:2]
If the constraints pertain only to A chains, then we would not expect

them to apply to wh-in-situ. However, derivational approaches typically
take wh-in-situ to be subject to invisible movement, so that it may be treated
uniformly with overt wh-questions. The question then arises as to whether
invisible movement is subject to the locality constraints.
On the face of it, covert movement, if it exists, is not subject to these
constraints. The relative clause in Chinese and Japanese precedes the head.
As seen in the Chinese examples in (126), shei “who” can be a constituent
of a relative clause and have a wide scope interrogative interpretation.
(126) Chinese
a. hufei xihuan nei-ben [NP [S shei xie de] shu]
Hufei likes that-CL who write DE book
‘∗ Whoi does Hufei like the book that t i wrote?’
(i.e. ‘Who is the person x such that Hufei likes the book the x wrote?’)
b. ni mai-le [NP [S shei xie de] shu]?
you buy-PERF who write DE book
‘∗ Whoi did you buy books that t i wrote?’
(i.e. ‘Who is the person x such that you bought [books [that x wrote]]?’)
[Tsai 1997]
To complicate matters somewhat further, note also that wh-in-situ in

French cannot occur in islands.
(127) a. Jean a acheté quoi

Jean has bought what
‘What has Jean bought?’
b. ∗ Jean aime le livre que qui a écrit?
Jean likes the book that who has written
‘Who is the person x such that Jean likes the book that x wrote.’
So, while this construction in French might be viewed as an “LF movement”

or an “invisible feature movement”, it must have different properties from
the Chinese/Japanese type of wh-in-situ that we have been looking at. This
conclusion is consistent with the observation that French wh-in-situ can
only have wide scope over the entire sentence (section 9.2.2).
9.7.2.2. Coordinate structure constraint

The CSC has two parts. One part rules out extraction of a conjunct, the
other part rules out extraction of a constituent of a conjunct. There is little
evidence if any that the first part of the constraint can be violated, but ample
evidence from English that the second part is too strong.
(128) a. How many coursesi can we expect our graduate students to teach t i and
(still) finish a dissertation on time? [Goldsmith 1985]
b. How many counterexamplesi can the coordinate structure constraint
sustain t i and still be assumed? [Lakoff 1986]
(129) a. They sat around all day in the kitchen and played with the cat.
b. This is the cati that they sat around all day in the kitchen and played
with t i .
c. ∗ This is the cat [with which]i they sat around all day in the kitchen and
played t i .
[Culicover and Jackendoff 1995]
Examples (129b) and (129c) show that the NP argument but not the PP can
be extracted from the right conjunct, showing that some islands selectively
allow extraction. Cinque 1990 calls contexts such as these weak islands;
Postal 1993 calls them selective islands. Roughly speaking, strong or non-
selective islands do not allow any extraction at all, while selective islands
allow the extraction of arguments but not adjuncts. Exercise 13 asks you to
test the other constraints that we have considered to see which, if any, create
selective islands.
9.8. OTHER A CONSTRUCTIONS 373
9.8. ∗ Other A constructions
There are many A constructions in English and other languages besides

those mentioned already in this chapter. We note their basic properties here.
Problems 19–24 ask you to work out analyses of some of these construc-
tions.
9.8.1. Questions
9.8.1.1. Infinitival questions

English has infinitival wh-questions, exemplified by (130a). Piedpiping and
p-stranding are both possible in such questions, as shown in (130b,c).
(130) a. I was wondering what to eat t.
b. I was wondering who to give the money to t.
c. I was wondering on which table to put the book t.
The infinitival question is a control configuration, since it has a full thematic

interpretation but lacks an overt subject. For example, (130a) is paraphrased
by I was wondering what I should eat. But there cannot be a for-NP phrase
that expresses the subject role.
(131) a. I was wondering what (∗ for you) to eat t.
b. I was wondering who (∗ for Sandy) to give the money to t.
c. I was wondering on which table (∗ for her) to put the book t.
Problem 19 asks you to state the correspondence between the syntactic

structure and the CS representation for this construction.
9.8.1.2. Partial wh-movement

Some languages have cases of partial wh-movement, in which a wh-phrase
appears in a position different from either the scope-marking position or
the in-situ position. The following are examples from German. 16
(132) German
a. Was glaubst du, weni wir t i einladen sollen?
what believe you who we invite should
Lit.: What do you believe who we should invite?
‘Who do you believe we should invite?’
16
Much of this data is controversial, and the judgments are not shared by all
speakers.
b. Was glaubst du, warumi wir ihn t i einladen sollen?

what believe you why we him invite should
Lit.: ‘What do you believe why we should invite him.’
‘Why do you believe we should invite him?’
c. Was glaubst du, wer kam?17
what believe you who came
Lit.: ‘What do you believe who came?’
‘Who do you believe came?’
[Sternefeld 2002]
This construction is characterized by the fact that there is a was “what”

in the position that marks the scope of the question, and the wh-phrase
appears clause-initially in the embedded clause.
It is possible to embed partial wh-movement, so that there are multiple
instances of was in each embedded sentence above the sentence that is the
scope of the question.
(133) German
Was glaubst du, was Peter meint, was Hans sagt, was Klaus
what believe you what Peter thinks what Hans says what Klaus
behauptet, mit wemi Maria t i gesprochen hat?
claims with whom Maria talked has
Lit: ‘What do you believe what Peter thinks what Hans says what Klaus
claims with whom Maria has talked?’
‘With whom do you believe that Peter thinks that Hans says that Klaus
claims Maria talked.’
[Riemsdijk 1983 cited in Sternefeld 2002]
Another language that has partial wh-movement is Hungarian. The follow-

ing examples are due to Horvath 1997:517.
(134) Hungarian
a. Mit kérdeztek, hogy kivel-talá alkoztam-e?
what-ACC asked-3 PL that who-with met-1 SG - Q - PRT
‘With whom did they ask whether I had met?’
b. Mit akartak tudni hogy kit
what-ACC wanted-3 PL know-INF that who-ACC
láttá’al-e?
saw-2 SG - Q - PRT
‘Who did they want to know whether you had seen?’
17
Was glaubst du, wer gekommen ist is better for some speakers. This construction
appears to be restricted to the verbs glauben “to believe”, and “sagen” to say, and
there is considerable disagreement about the grammaticality of the more complex
examples.
Comparable sentences in German are not grammatical.
(135) German
a. (i) ∗ Was fragst du, ob ich wen getroffen habe?
what ask you whether I who met has
‘∗ Who are you asking whether I met?’
(ii) ∗ Was fragst du, wen (ob) ich getroffen habe?
what ask you who whether I met have
‘∗ Who are you asking whether I met?’
b. (i) ∗ Was willst du wissen, ob ich wen gesehen habe?
what want you to-know whether I who seen has
‘?Who do you want to know whether I have seen?’
(ii) ∗ Was willst du wissen, wen (ob) ich gesehen habe?
what want you to-know who whether I seen have
‘?Who do you want to know whether I have seen?’
[Sternefeld 2002]
9.8.1.3. Multiple wh-movement

In section 9.5.2 we discussed the idea that a [ WH ] feature on C0 triggers the
movement of one wh-phrase to Spec of CP in multiple wh-questions, such
as Where did you put what? One problem with this line of analysis is that
in the Slavic languages, wh-questions have a wh-phrase in the clause-initial
position, but multiple wh-questions have all of the wh-phrases in the initial
position.
(136) Serbo-Croatian
a. Ko sta gdje kupuje?
who what where buys
‘Who buys what where?’
b. ∗ Ko kupuje sta gdje?
c. ∗ Ko sta kupuje gdje?
d. ∗ Ko gdje kupuje sta?
[Boškovic 1997]
(137) Russian
a. Kto čto kogda skazal?
who what when said
‘Who said what when?’
[Rudin 1988:446]
b. Bulgarian
Koj kogo vižda?
who whom sees
‘Who sees whom?’
c. Czech
Kdo koho videl?
who whom saw
‘Who saw whom?’
d. Polish
Kto co robił?
who what did
‘Who did what?’
[Rudin 1988:449]
Facts such as these clearly run up against the view that the appearance
of a wh-phrase in initial position of a wh-question is triggered by the
requirement that the [WH] feature of C0 has to be discharged. If one wh-
phrase will do this, why must the other wh-phrases appear in this position?
One answer that has been given is that although the feature is discharged,
it is “renewed”, but this is of course nothing more than an encoding of the
offending facts in terms of feature discharge.
9.8.2. Relatives
9.8.2.1. Free relatives

The free relative construction is interpreted as though it is a NP or PP that
contains a relative clause.

(138) a. What you said is funny.
The thing that you
said
b. I want to talk to who you talk to .
the person who you talk to
c. I will go where you go .
to the place where you go
For the case with what, it appears to be straightforward to formulate a direct

correspondence, as in (139), to capture the interpretation of the free relative.
GFi stands for an arbitrary grammatical function. Notice that the category
of the free relative is the category of what.
(139) SYNTAX NP
N0 S
what [e]
GF GFi
CS THING·(PROPERTY:F(…, ·, …))
Problem 20 asks you to formulate an analysis for where.
9.8.2.2. Internally headed relatives

Languages that do not have overt movement in questions also lack overt
movement in relative clauses. Here are examples from Korean illustrating
two ways in which this can happen. 18 (The prenominal marker -nun (glossed
as PNE) is attached to the right edge of the relative clause. The marker
kes (glossed as KES) is called a “dependent noun” in traditional Korean
grammar and means “thing”.)
(140) Korean
a. Tom-un [__ cayngpan-wi -ey iss-nun] sakwa]-ul mekessta.
Tom-TOP tray -TOP - LOC exist-PNE apple -ACC ate
‘Tom ate an apple that was on the tray.’
b. Tom-un [[sakwa-ka cayngpan-wi -ey iss -nun ] kes]-ul
Tom-TOP apple -NOM tray -TOP - LOC exist-PNE KES-ACC
mekessta
ate
‘Tom ate an apple, which was on the tray.’
[Chung and Kim 2003:43]
Korean is an SOV language. In (140a) the object of the verb mekessta “ate”
is sakwa “apple”, which is marked with the accusative case marker -ul. The
18
Japanese has similar constructions.
relative clause is cayngpan-wi-ey iss-nun “REL was on the tray”. Notice that
there is no overt relative pronoun. The relative clause precedes the head
noun.
In (140b) the relativized noun sakwa “apple” is in the relative clause
and there is no overt argument of the main verb mekessta “ate”. While
in Korean a relative clause precedes the head noun, this example also
shows that it is possible for the head noun to appear inside of the relative
clause. For this reason, the construction illustrated by (140a) is called
an externally headed relative clause, and that illustrated by (140b) is
called an internally headed relative clause.
Problem 21 asks you to formulate the correspondence rules for externally
and internally headed relative clauses.
9.8.2.3. Infinitival relatives

English has not only infinitival questions (see section 9.8.1.1) but infinitival
relatives.
(141) a. the person to do the job
∗
b. the person who to do the job
(142) a. the person (for you) to talk to
∗
b. the person who (for you ) to talk to
c. the person to whom (∗ for you ) to talk
Compare these with similar wh-infinitives.

∗
(143) a. I wondered to do the job.
∗
b. I wondered who to do the job.
(144) a. I asked who (∗ for you) to talk to
b. I asked to whom (∗ for you) to talk
The comparison highlights the fact that there are significant similarities
and differences between the two constructions. One similarity is that it
is impossible to overtly relativize or question the subject of an infinitival
(compare (141b) and (143b)). Another is that the infinitival relative may
have an overt complementizer and subject (with for) only when there is no
overt relative pronoun, and the wh-infinitival, since it must have an overt
wh-phrase, may not.
One difference is that it is possible to have an infinitival relative where
OPi is the subject of the relative, but there is no comparable wh-infinitival
(compare (141a) and (143a)). Another is that the infinitival relative may not
have an overt relative NP, but the wh-infinitive may. Problem 22 asks you to
work out a formal analysis of these constructions that captures these facts.
9.8.3. Clefts and pseudo-clefts
9.8.3.1. Clefts
The cleft construction has the following informal description.
 

 is 

was
(145) it FOCUSi [CP that . . . t i . . . ]
 will be 
 
etc.
FOCUS designates a constituent that has a contrastive or emphatic dis-

course function, as in It was the PRESIDENT that/who should have taken
responsibility for what happened (and not the CONGRESS). The focus
constituent is typically an NP, PP, or CP in Standard English, not an adverb
or an AP.
(146) a. It was the JELLO that I violently threw against the wall.
b. It was against the WALL that I violently threw the Jello.
c. ∗ It was VIOLENTLY that I threw the Jello against the wall.
(147) It is [that Sandy will WIN] that I believe.
(148) ∗ It was very ANGRY that Sandy was.
However, it is possible that the starred sentences here are better for some
speakers. In any case, the focus position of the cleft can be used as a test
for constituency. So NP-VP (as in expect there to be an explosion) and NP-
AP (as in cook the meat raw) sequences in general are not possible in focus
position, consistent with the view that these are not constituents. But a
true small clause, such as myself blond as the complement of imagine (see
Chapter 8) should appear in the focus position of a cleft, and does.
∗
(149) a. It was there to be an explosion that I expected.
that there would be an explosion
∗
b. It was the meat raw that I cooked .
myself blond that I imagined
When the focus of the cleft is an NP, it is possible to have a relative pro-
noun after the focus instead of that, and when the focus does not function
as the subject of the embedded clause, that may be omitted. Thus, the cleft
patterns very much like the relative clause.
(150) a. It was Sandy (that) I was talking to.

b. It was Sandy ∗ (that) called me.
c. It was Sandy who called me.
(151) a. It’s your book (that) I’m looking for.
b. It’s your book ∗ (that) bothers me.

c. It’s ?your book which I bought.
this
(152) a. It was on the table (that) I put the dog food.
b. It was on the table where I put the dog food.
(153) a. It’s next week that I’m leaving.

b. ?It’s next week when I’m leaving.
5 o’clock
Note the use of where as a relative pronoun when the focus is locative: cf.
the table where I put the dog food was very expensive, and the use of when if
the focus is a time.
9.8.3.2. Pseudo-cleft
The third construction that is related to relative clauses is the pseudo-cleft,
which looks like a free relative construction but is actually an embedded
question construction. The general form is
 

 is 

(154) [WH-QUESTION] was FOCUS
 will be 
 
etc.
Here are some pairs that illustrate the difference.
(155) a. What I bought is a new car.
b. What I bought is in the garage.
(156) a. What I believe is that the world is flat.
b. What I believe is false.
(157) a. Where Sandy put the beer is on the table.
b. Where Sandy put the beer is a dangerous place.
(158) What I found out is what Sandy forgot.
a. Meaning 1: I found out the answer to the question “what did Sandy
forget”.
b. Meaning 2: I found out something, e.g. that the world is round, which
Sandy had forgotten.
In each of these pairs, the focus in the a-examples supplies the answer
to the question. For example, in What I bought is a new car, the question
is “what did you buy” and the answer is “a new car”. In contrast, the b-
example provides a property of what is referred to by the free relative. So, in
9.9. SUMMARY 381
what I bought is in the garage, the phrase in the garage describes where the
car is. In the garage is not an answer to the question “what did you buy?”
A third type of sentence that is distinct from either of these two is an
indirect question where the question is a subject, e.g.
(159) a. What I bought is interesting.

b. Where Sandy put the beer bothers me.
c. What I found out is a mystery to everyone.
These cases are distinguished from the a- and b-examples above by the fact
that on their most natural interpretation they do not supply an answer to
the question, and they do not give a property of what the subject refers to.
Rather, they give a property of the question itself. It is interesting that these
embedded questions permit extraposition, while the pseudo-cleft and free
relative do not.
(160) a. What I bought is interesting. ∼ It is interesting what I bought.

b. What I bought is a new car. ∼ ∗ It is a new car what I bought.
c. What I bought is in the garage. ∼ ∗ It is in the garage what I bought.
The semantic differences noted here correlate with the syntactic analysis in
the following way. The subject of the free relative is an NP (or PP), and does
not participate in the extraposition alternation. The indirect questions are
sentential subjects whose predicates permit the extraposition alternation.
The pseudo-cleft is formed from an indirect question in subject position

is
and a predicate of the form was FOCUS, which does not permit an
will be
etc.
extraposition alternation.
9.9. Summary
In this chapter we looked at wh-questions, relative clauses, and topicaliza-

tion. These constructions are characterized by having a gap in an argu-
ment or adjunct position that matches up with constituents in the sentence
that supplies an interpretation associated with the gap. Using the corre-
spondence between syntactic structure and CS, it is possible to define A
chains, which link the gaps to these constituents through a binding relation
at CS.
In MGG, A chains are derived through movement. We summarized the

basic properties of movement analyses, which have been very influential in
contemporary syntactic theory. We also observed that A chains are subject
to extraction constraints. While movement is one way to derive A chains,
it is not clear that movement is a necessary formal device either from the
perspective of accounting for the distribution of A chains or the extraction
constraints.
Exercises
1. Say what the scope of the question is in each of the following sentences.
We will give you an example to get you started.
(0) a. What did Sandy buy.

The scope of the question is the entire sentence. It means “for which thing
x, Sandy bought x”.
b. I know what Sandy bought.
The scope of the question is the embedded sentence. The entire sentence
means, “I know the answer to the question ‘what did Sandy buy?’ ”.
(1) a. Where are you going?
b. Do you like Bare Naked Ladies?
c. Please tell me how much that car costs.
d. Sandy thinks that Leslie was wondering whether it is worth going out.
e. What did you say that you wanted to eat?
f. How many people did you tell that you were planning to cut your hair?
g. Whether Sandy is going to enter the race is completely unknown at this
point.
[§9.1.]
2. Here are some examples of English interrogative expressions.
(1) a. which man

b. how tall
c. how tall a man
d. to whom
e. to which man
f. a picture of which man
We might expect each of these to undergo piedpiping, since each one con-
tains a wh-phrase. Some definitely can –
EXERCISES 383
(2) How tall a man did you see?
and some definitely cannot –
∗
(3) A picture of which man are you looking for?
[Cf. Which man are you looking for a picture of ?]
Investigate the range of piedpiping in wh-questions in English by making

up your own examples. What kinds of phrases that contain a wh-word may
move, and what kinds of phrases may not? How deeply in the phrase can the
wh-phrase be and still permit piedpiping? Does it matter what the syntactic
function of the wh-phrase is in the phrase that contains it? (Don’t forget
about other types of phrases besides NPs and PPs.)
[§9.2.]
3. Construct the syntax–CS correspondences for the two sentences in (27)

in the text and verify that they have the scopes that are indicated by the
translations.
[§9.2.]
4. Construct the correspondence for the Japanese multiple wh-question
(45) in the text.
[§9.2.]
5. In (28) in the text we gave some examples of wh-questions in Chinese;

these are repeated here.
(1) Chinese
a. Ni xihuan shei?
you like who
b. Zhangsan wen wo [shei mai-le shu].
Zhangsan ask me who bought books
‘Zhangsan asked me who bought books.’
c. Zhangsan wen wo [ni maile shenme]
Zhangsan ask me you bought what
‘Zhangsan asked me what you bought.’
d. Zhangsan xiangxin [shei mai-le shu].
Zhangsan believe who bought books
‘Who does Zhangsan believe bought books?’
e. Zhangsan renwei [ni maile shenme]
Zhangsan think you bought what
‘What does Zhangsan think you bought?’
f. Zhangsan zhidao [shei mai-le shu].

Zhangsan know who bought books
i. ‘Who does Zhangsan know bought books?’
ii. ‘Zhangsan knows who bought books.’
[Huang 1982]
Illustrate in terms of correspondences the ambiguity of example (f). With

this in mind, how do you explain why (c) and (d) are not ambiguous?
[§9.2.]
6. How would you account for the facts of Exercise 5 in an LF-movement
analysis? (In order to answer this question, you should construct the syntac-
tic structures and show how the two different LF representations are derived
through movement.)
[§9.2.]
7. Draw a tree showing the syntactic structure of each of the following
NPs that contain a relative clause.
(1) a. the cat that scratched you

b. the town that I live in
c. the town where I live
d. every student who the university wants to learn about
e. any student who passes the quiz
[§9.3.]
8. Explain how the differences in the structures of the following sentences

correspond to the meaning differences associated with them.
(1) a. the child that gave the bone to the dog

b. the bone that the child gave to the dog
c. the dog that the child gave the bone to
[§9.3.]
9. For each of the following ungrammatical sentences, say which con-

straint or constraints it violates and why. (Hint: Figure out where the gap
would have to be.)
∗
(1) a. I bought a book that the fact that costs $50 really bothers me.
∗
b. Which of the books did they explain to you where you should shelve?
∗
c. This is the kind of peanut butter I really like and jelly.
∗
d. Who did the book case that fell on bruise her shoulder?
EXERCISES 385
e. ∗ Who did you warn Kim that was coming to stay for a while?
f. ∗ Whose did you really enjoy jokes?
g. ∗ How much did you buy that car, although you don’t really think it is worth?
[§9.4.]
10. Erteschik 1973 noted that there are long distance dependencies where
lexical semantics makes a difference for extraction possibilities from senten-
tial complements.

say
(1) a. What did Bill that Harry would like e for lunch?
??grumble

said
b. The man who Bill ∗ that Harry met e ordered a bagel.
grumbled

said
c. This book is longer than you ∗ that it was e.]
grumbled
Make a list of at least five verbs that allow long distance dependencies and
at least five that do not. Are there any characteristics that uniquely define
either set? What can we conclude about long distance dependencies from
these observations?
[§9.4.]
11. Suppose that a possible structure of CP in English is

(1) [CP Spec [C C0 CP]]
which is licensed by the rule

(2) C → C0 CP
Show how this analysis solves the problem posed by the following data.
(3) a. To Sandy, what did you say?
b. I believe that to Sandy, you should say nothing.
[§9.6.]
12. Work through the Swedish examples in (123) in the text and show that
they are indeed exceptions to the complex NP constraint.
[§9.7.]
∗
13. Chomsky 1977 argued that all A constructions behave identically
with respect to the extraction constraints (such as those in section 9.4 and
9.7). Test each of the following constructions to test this claim.
a. topicalization
b. infinitival relative
c. free relative
d. cleft
e. pseudo-cleft
[§9.8.]
Problems
1. If and whether are both used to signal embedded yes-no questions,

as seen in the examples in (8). Determine whether if and whether are
completely interchangeable, and, if not, what the restrictions on their dis-
tribution are. Here are some grammatical examples to get you started; you
will have to make up others that may or may not be grammatical.

if
(1) a. I was wondering you wanted to have something to eat.
whether
b. It is hard to know whether to stay or leave.
c. Whether or not you leave, we will stay.
d. Ask them whether you should stay or not.
e. Ask them whether or not you should stay.
[§9.1.]
2. Can every preposition in English be stranded? Is there an identifiable

syntactic or semantic factor that makes it difficult to strand a position
in certain configurations? Make up examples that will allow you to begin
to answer these questions. To get started, here are some cases that are
ungrammatical for some speakers, but perhaps not all.
(1) a. ?Which movie did you fall asleep during.

[cf. I fell asleep during Love Story.]
b. ∗ Which administration have you been living here since?
[cf. I have been living here since the first Bush administration.]
In these examples, the prepositional phrases are constituents of VP. [dur-

ing NP] and [since NP] are arguably adjuncts, and they are of course
temporal.
Construct similar examples to test the prepositions in the following lists.
(In order to keep the problem manageable, start with a selection of five
prepositions from each list.)
PROBLEMS 387
Locative/Directional
above from through
across in to
along inside toward (or
around into towards)
at near under
behind off underneath
below on upon
beneath onto up
beside out of up to
between outside within
beyond over
down past
Temporal
after during till
before since until
by throughout
Other
about despite of
against for with
among like without
[§9.2.]
3. English piedpiping applies to NP, AP, and PP but not to VP –
(1) a. [NP Whose horse]i were you riding t i ?

b. [AP How tall]i is Sandy t i ?
c. [PP To whom]i were you talking t i ?
d. [PP How far into the forest]i did Robin ride t i ?
∗
(2) a. [VP how carefully wrap the presents]i did the children t i ?
∗
b. [VP wrap the presents how carefully]i did the children t i ?
∗
c. [VP riding whose horse]i were you t i ?
– although VPs can undergo topicalization.
(3) . . . and [VP carefully wrap the presents] they did t i .

. . . and [VP riding the horse]i you were t i .
One way to account for piedpiping is to assume that there is a formal

operation that copies the [WH] feature of the Spec to the phrasal projec-
tion itself. Observing that whose and how are specifiers of their respective
phrases, state this operation as precisely as you can. What are the restric-
tions on this copying? (For example, how far up in a structure can the
copying go? Do the syntactic categories involved play any role?) Can such
an account be extended to account for piedpiping of the PP, and will it also
account for the fact that is is impossible to piedpipe VP?
[§9.2.]
4. Consider the fact that in general it is not possible to have a zero relative
when the subject is relativized.
(1) a. Show me the cat that scratched you.

b. ∗ Show me the cat scratched you.
State the condition that rules out English relative clauses such as the one
in (1b) as succinctly as you can. Your formulation should not rule out (1a),
nor should it rule out (2).
(2) a. The cat scratched you.

b. the cat you bought (is furry).
[§9.3.]
5. It is possible to have a sequence of relative clauses in an English NP.

What are the constraints on the form of these concatenated relative clauses?
Can any and all of them be overt, zero- or that-relatives? Here are a few
examples to get you started.
(1) a. the book [that I bought] [that I gave to Sandy]

b. the book [I bought] [that I gave to Sandy]
c. the book [that I bought] [which I gave to Sandy]
d. the book [which I bought] [which I gave to Sandy]
What generalizations if any emerge about the possible sequences of relative

clauses? (Hint: Try reordering the relative clauses in these examples and see
what happens.)
[§9.3.]
6. Starting from the examples given in section 9.3.2. in the text, determine
the range of possibilities for piedpiping in English relative clauses. Follow
PROBLEMS 389
the same approach that you took in Problem 2. Start with examples (56)–
(60) and add your own as needed.
[§9.3.]
7. It is possible to use feature discharge in a way that is equivalent to a

correspondence rule in not requiring covert movement. On such an analysis,
the position of C0 [WH] would mark the scope of the wh-question and
would be discharged without Spec head agreement if certain conditions are
satisfied. The superficial syntactic structure of a wh-question in a wh-in-situ
language would be essentially the following.
(1) CP
Spec C
IP C0
[WH]

NP I
VP I0
NP V
[WH]
i. State the conditions that must be satisfied that will guarantee that [WH]
is discharged only when the sentence is a legitimate wh-question. There are
three cases to consider: that shown in (1), the case where C0 is [WH] but
there is no wh-phrase c-commanded by C0 , and the case where there is a
wh-phrase but no C0 [WH].
ii. Compare your solution to one in which there is a correspondence rule
that relates the interrogative complementizer and the wh-phrase to a CS
representation in which the Q operator binds a variable.
[§9.5.]
8. The Bellunese data discussed in section 9.2.2. are unusual for two
reasons. First, a complex wh-phrase appears in clause-initial position but
a simple wh-word appears in situ. Second, even when there is a wh-phrase
in situ, there is inversion of the verb and the subject pronoun.
A. Using the formal devices of feature discharge and movement, work
out an analysis that derives the Bellunese data. Assume that the inflected
verb moves into C0 in order to discharge a feature of C0 , call it [ F ]. What

special assumptions do you have to make about the wh-in-situ cases?
B. As an alternative, try to account for the Bellunese data without making
use of movement, but simply formulating the conditions on interrogative
A chains. Compare the two analyses in terms of their complexity and
naturalness. Are there problems that are solved easily on one approach but
problematic for the other?
[§9.5.]
9. We have seen that piedpiping in relative clauses applies to NPs and PPs
quite freely (see examples (56)–(60)).
A. To explore piedpiping further, first test wh-questions, infinitival wh-
questions, and infinitival relatives to see to what extent they show the same
pattern. Are there any generalizations that succinctly capture any differ-
ences in piedpiping among these constructions?
B. Formulate an analysis of relative clauses in terms of an agreement
feature [REL] on the C0 of a relative clause that must be discharged. The
crucial part of the analysis will be stating under what circumstances a
constituent has the feature [REL] so that it can discharge C0 [REL].
[§9.5.]
10. Take the condition on distribution of [REL] that you arrived at in

Problem 9. Note now that there cannot be piedpiping in a zero-relative. For
example, if we analyze a zero-relative as containing the empty operator OP,
we have:
∗
(1) a. the cat, [PP at OPi ]j I was looking t j
∗
b. the cat, [NP OPi (’s) tail]j I was looking at t j
∗
c. the cat, [NP a picture of OPi ]j I was looking at t j
∗
d. the table, [VP sitting on the corner of OPi ]j was the cat
Formulate a condition that will block propagation of [REL] when the rel-
ative pronoun is OP, but permits it when the relative is an overt relative
pronoun.
[§9.5.]
11. State as precisely and as compactly as you can the condition that rules
out relative clauses and questions in English of the form
(1) a. ∗ the man who that you saw

b. ∗ I forgot when that you called.
PROBLEMS 391
but allows
(2) a. the man that you saw
b. I forgot when you called
A. Assume for the sake of doing this problem that the structure of the
clause is that of CP.
B. Then, compare your solution in A with the condition that you would
formulate if the structure was simply S, as in, for example, (11) in the text.
[§9.5.]
12. In our comparison of topicalization and wh-movement in the text we

offered the possibility that whether is the interrogative counterpart to that.
State explicitly the conditions on the distribution of whether, so that the
following pattern will be accounted for.
(1) a. I wonder whether Sandy loves Kim.
∗
b. I wonder Sandy loves Kim.
c. I wonder who Sandy loves.
∗
d. I wonder who whether Sandy loves.
∗
e. I wonder whether Sandy loves who.
∗
f. I wonder does Sandy love Kim.
[§9.6.]
13. English (and many other languages – see Merchant 2001) have a
construction called sluicing. It is exemplified in (1).
(1) a. I saw Terry talking to someone yesterday, but I don’t know who.
b. Sandy went to Florida, but I forgot with who.
One natural way of analyzing sluicing is to take it to be a type of ellipsis

construction, in which the Spec of CP contains a wh-phrase, and the IP-
complement of C0 is absent or empty.
(2) . . . I don’t know [CP who [C C0 [IP ]]
Hence sluicing can be seen as evidence for the CP/IP distinction.

i. Is this analysis of sluicing consistent with how sluicing behaves in main
questions?, e.g.
(3) A: I saw Terry talking to someone yesterday.
B: Who?
ii. Is there a counterpart to sluicing when what is fronted in the complement
is not a wh-phrase but a topic? As in
(4) I like Terry, but Sandy, I don’t like.
What conclusions follow?

[§9.6.]
14. Dutch has a complementizer dat that functions like English that.
(1) Ik weet dat Jan denkt dat Piet komt

I know that John thinks that Piet comes
However, Dutch shows some sequences in the complementizer position that

are not possible in English.
(2) a. Ik vraag me af [of [dat [ Ajax de volgende ronde haalt]]]

I ask me PRT if that Ajax the next round reaches
‘I wonder whether Ajax will make it to the next round.’
b. Dat is niet zo gek [als [of [dat [hij gedacht had]]]]
this is not so strange as if that he thought had
‘This is not as strange as he thought.’
c. Hij weet [hoe [of [je dat moet doen]]]
he knows how if you this must do
‘He knows how you must do this.’
d. Ze weet [wie [of [dat [hij had willen opbellen]]]]
she knows who if that he had wanted call
‘She knows who he wanted to call.’
[Hoekstra 1993]
(3) a. ∗ Ik geloof dat jouw boek ze waarderen
I believe that your book they appreciate
b. ∗ Ik geloof jouw boek dat ze waarderen
I believe your book that they appreciate
‘I believe they appreciate your book very much.’
∗
(4) a. Ik vergat dat jouw boek ze waarderen
I forgot that your book they appreciate
b. ∗ Ik vergat jouw boek dat ze waarderen
I forgot your book that they appreciate
‘I forgot they appreciate your book very much.’
(5) a. Ik geloof dat ze jouw boek waarderen.
b. Ik vergat dat ze jouw boek waarderen.
[van Gelderen 2000]
(6) a. Jan zal meedelen of (dat) Marie deze boeken leest.
Jan will announce if that Mary these books reads
‘Jan will announce whether Mary reads these books.’
PROBLEMS 393
b. ∗ Jan zal denken of (dat) Marie deze boeken leest.

Jan will think if that Mary these books reads
‘Jan will think that Mary reads these books’
[Barbiers 2002:50–1]
A. This data suggests that an iterated CP analysis might be correct for

Dutch. Work out such an analysis, being careful to specify the selectional
properties of each form als, of, and dat.
B. Now consider what an analysis of the possible sequences would look
like in strictly constructional terms. Compare the two analyses in terms of
which if either requires the less elaborate formal machinery.
[§9.6.]
15. In our discussion of Ross’s constraints we saw some examples that

show that when there is a violation of CNPC, putting a pronoun in place of
the gap improves the sentence, e.g.
(1) a. (?)Gone with the Wind is a movie whichi I cry [every time I see iti ].
b. ∗ Gone with the Wind is a movie whichi I cry [every time I see t i ].
This is called the resumptive pronoun strategy. Construct examples to see

if the introduction of a resumptive pronoun ameliorates violations of the
other constraints.
[§9.7.]
16. It is possible to extract a wh-phrase across another wh-phrase when

both originate in VP, e.g.
(1) I wonder whoj you gave whati to t j .
But when one of the wh-phrases is a subject, a violation occurs.

∗
(2) I wonder whatj whoi gave t j to Sandy.
In the syntax literature this is called a superiority violation, after the supe-
riority condition. Superiority condition: A wh-phrase cannot be extracted
across a wh-phrase that c-commands it.
A. Assuming that the definition of the Superiority condition in terms of
c-command is correct, what would be an appropriate syntactic structure
that would allow for such examples. (Hint: Will flat structure in VP work?)
B. Assuming that the structure is flat, how would you change the defini-
tion of the Superiority condition in order to allow for such examples? Your
solution should account for the following pattern.
∗
(1) a. What did who put there?
∗
b. Where did who put the beer?
∗
c. When did who leave?
∗
(2) a. What did you convince who that you said?
b. Who did you convince that you said what?
∗
(3) a. Who did you copy whose picture of ?
b. Whose picture of whom did you copy?
[§9.7.]
17. The adverb effect refers to cases in which the presence of an adverb
after the complementizer that ameliorates the that-t effect in cases of extrac-
tion of an embedded subject, as shown in (1).
(1) a. ∗ Whoi did you say that t i called?
b. Whoi did you say that most recently t i called?
Does the adverb effect apply in cases of extraction of embedded subjects

when the complementizer is not that? Construct examples to support your
answer. You should consider whether, if, and for, and wh-phrases.
[§9.7.]
18. Suppose that we account for multiple wh-questions by positing LF

movement of the wh-in-situ to a position that c-commands the part of
the sentence over which all of the wh-phrases take scope. For example, the
sentence
(1) What did you give t i to whom?
would have the LF

(2) whoj [whati [you give t i to t j ]]
If the movement constraints that we have discussed in this chapter apply to

LF movement, they should block multiple wh-questions in which one (or
more) wh-phrases are in an island. Discuss whether this is a correct predic-
tion for constructions that the following constraints have been applied to;
use examples to illustrate your answer.
a. complex NP constraint
b. that-t effect
c adjunct island condition
[§9.7.]
19. Consider the data in (130) in the text that illustrates infinitival ques-
tions in English. State the correspondence between the syntactic structure
PROBLEMS 395
and the CS representation as concisely as you can. For the syntax, it is

necessary to represent the fact that there is an initial wh-phrase, and an
infinitival VP. For the CS representation, there must be an interrogative
operator that binds a variable.
[§9.8.]
20. Formulate a non-movement analysis for where free relatives similar

to the analysis in the text for what free relatives in (139). Pay particular
attention to the question of what the category of the free relative is in this
case, and provide data to support your analysis.
[§9.8.]
21. Formulate the correspondence rules for externally and internally

headed relative clauses such as those given in (140) in the text. Assume that
the structures are the following where [ei ] marks an empty NP or N.
(1) Externally headed relative clause

NP
S Ni
… [ei] … -nun apple
(2) Internally headed relative clause

NP
S Ni
… applei … -nun ei
[§9.8.]
22. State a correspondence rule for the infinitival relative as illustrated in

(141)–(142) in the text. Show how it accounts for the differences between
infinitival relatives and infinitival questions.
[§9.8.]
23. Formulate an analysis of the English cleft construction in terms of

empty operator movement, and formulate the alternative analysis in terms
of a correspondence rule. In both analyses, you must make sure that there is
a chain that links the focus constituent with the gap in the embedded clause.
[§9.8.]
24. Formulate an analysis of the pseudo-cleft construction so that the

interpretation of the focus constituent appears in the correct position in
CS. State the correspondence as generally as you can, so that it does
not depend on the particular syntactic category or CS function of the
focus.
[§9.8.]
Research questions
1. Explaining why p-stranding occurs in some languages but not others is

a long-standing problem in syntactic theory. It is possible that the best we
can do is describe the difference without explaining it. But even doing this
is not entirely straightforward, as this exercise will demonstrate.
A. Using correspondences, provide a formal description of wh-questions
in a language like English that allows for p-stranding, and a formal descrip-
tion of wh-questions in a language that does not allow for p-stranding.
Both descriptions should consist of the essential links between the relevant
constituents in the syntactic structure, the GFs to the extent that they are
relevant, and the CS representation. What special statement has to be made
about English in order to allow it to have p-stranding?
B. Having described p-stranding in A, consider the fact that English
has p-stranding in all A constructions, including wh-questions and relative
clauses of all types. It even has p-stranding in the passive, which is an A
construction.
(1) a. Who did Sandy look at?

b. the picture that Sandy was looking at
c. this picture, which Sandy was looking at, . . .
d. This picture has rarely been looked at.
e. Sandy was taken advantage of.
So it appears that p-stranding is a general characteristic of English, not

of the specific constructions. How can this property of the language be
accounted for? That is, where in the grammar of the language is the p-
stranding characteristic located?
[§9.2.]
2. Example (81) in the text is an instance of across-the-board (ATB)

extraction. What are the limits of ATB extraction, and what constitutes
“parallel structure”?
A. Is parallel extraction always possible from conjoined NPs, even when
they are of different levels of complexity?, e.g.
(1) a. a picture of X and a movie about X
b. the destruction of X and the repair of X
c. the destruction of X and a movie about the repair of X
etc.
B. Is parallel extraction possible from conjoined VPs? What constitutes

parallelism in such a case? Consider, for example, conjoined VPs such as the
following.
(2) a. cook X and eat X
b. cook X and give X to Y
c. insult Y and give X to Y
d. give X to Y and reclaim X
e. give X to Y and insult Y
C. Is parallel extraction possible from conjoined APs?

D. Is parallel extraction possible from conjoined S’s, when the NPs (a)
have different θ-roles or (b) different GFs?
And so on.
In principle, these are very opened-ended questions, and a “yes/no”
answer is not particularly informative, even if it is possible. It is more
useful to narrow things down to a particular syntactic category, such as
VP, and then see whether there are syntactic and/or semantic (e.g. thematic)
conditions that determine when parallelism holds.
[§9.4.]
3. In section 9.1 of the text we raised the question of whether a subject

question such as
(1) Who saw Sandy?
is an A construction. On a movement analysis, who would be moved to

Spec of CP, leaving a trace in subject position.
(2) [CP [SPEC whoi ] C0 [ WH ] [IP t i . . . ]]
The question that needs to be addressed is whether there is syntactic evi-

dence for this structure. Discuss the kind of evidence that would bear on this
question and how you would go about determining if this evidence exists.
This turns out to be a difficult problem, because the sequence who saw . . .
is consistent with both the presence or the absence of a trace in subject
position.
[§9.5.]
4. Sketch out an analysis of the correspondences for echo and quiz ques-
tions on the analogy with that of topicalization that we gave in section 9.6.
How does this approach explain the fact that echo and quiz questions may
only have wide scope? Since topicalization may be embedded, how do you
explain this difference between the constructions?
[§9.5.]
5. While the complex NP constraint/subjacency is viewed as a syntactic

condition that blocks extraction from certain contexts, there is evidence that
some extractions from these contexts are not as ungrammatical as others.
For example,
(1) a. a book [that a critic [that I met] dislikes t]
b. ∗ a book [that I met a critic [who dislikes t]]
(=a book such that I met a critic who dislikes it)

that
c. ?a book [that there are only a few critics [ dislike t]]
(?)who
One difference between the NP that blocks extraction and the NP that
allows it is definiteness of reference in some informal sense. The first is of
the form I met a critic [who . . . ], which picks out a particular individual,
while the second is of the form there are only a few critics [who . . . ], which
simply states the existence of a set of individuals with a particular property.
Interestingly, the same type of difference appears to play a role in allowing
extractions from picture-NPs.
(2) Let me tell you about
a. a book [that I just read t].
b. ?a book [that I just burned [a review of t]].
(=a book such that I burned a review of it)
c. a book [that I just read [a description of t]].
d. ?a book [that I just read [Sandy’s description of t]]
e. ∗ a book [that I just burned [Sandy’s description of t]]
However, it does not appear that these examples can be made to follow from
CNPC.
Try to make the notion of “definiteness of reference” more precise. There
appear to be several factors involved, including (i) the specifier, (ii) the
θ-role assigned to the NP from which the extraction takes place, (iii) the
head noun of the NP. Isolate one of these and see if it is possible to say
in more precise terms under what conditions we get different levels of
acceptability judgments.
[§9.7.]
6. We noted in section 9.7.2.2 that certain violations of the coordinate

structure constraint are possible. In one subcase, the first conjunct contains
a verb like sit around and the second conjunct is a VP, as in
(1) a movie that we sat around all day and made fun of
A. What characterizes the class of predicates that can occur in the first
conjunct in this construction?
B. There is a similar construction in which the conjuncts are reversed. An
example based on (1) appears to be grammatical.
(2) a movie that we made fun of for a while and then sat around all day
Does the construction in (1) always permit such a variant, and if not, what
are the differences?
[§9.7.]
9.1. 1 1
9.2. 2, 3, 4, 5, 6 2, 3 1
9.3. 7, 8 4, 5, 6
9.4. 9, 10 2
9.5. 7, 8, 9, 10, 11 3, 4
9.6. 11 12, 13, 14
9.7. 12 15, 16, 17, 18 5, 6
9.8. 13 19, 20, 21, 22, 23, 24
10
Coreference and Binding
10.1. Coreference
The relation in which two NPs refer to the same thing or things is corefer-
ence. In this chapter we look at the syntactic factors that determine whether
two NPs may, must, or must not be coreferential.
Recall that we assign a referential index to every referring expression (see
Chapter 5). Suppose that we have two expressions in a sentence that are
intended to refer to the same thing. If we use the same name or description
in the sentence twice, then the representation of coreference is more or less
straightforward. As an example we use the sentence Mary’s mother kissed
Mary. We will take the CS representation of mother to be MOTHER(X). In
this case MARY fills the argument slot X, as shown in (1).
(1)
SYNTAX S
NP VP
NP ’s N V NP
Mary mother kissed Mary
GF Subject Object
CS KISS(AGENT:[MOTHER(MARY·)]‚,PATIENT:·)
As can be seen, the superscript · on the two arguments in CS indicates that

they refer to the same thing. The superscript ‚ on MOTHER(MARY· )‚
402 10. COREFERENCE AND BINDING
indicates that the reference of Mary’s mother is not the same as that of
Mary.
Using the same notation, we can represent the meaning of a sentence in
which the second instance of Mary is replaced by the pronoun her in more
or less the same way.
(2) Mary’s mother kiss her.
The only difference in this case is that her has no meaning independent of
its relationship to Mary, except that it denotes a female; hence its represen-
tation in CS is just the index, which indicates the reference, and the feature
[FEMALE].
(3)
SYNTAX S
NP VP
NP ’s N V NP
Mary mother kissed her
GF Subject Object
CS KISS(AGENT:[MOTHER(MARY·)]‚,PATIENT:·[FEMALE] )
As before, the identity of the indices indicates that the arguments Mary’s
mother and her are intended to refer to the same thing. The pronoun in her
(2) could of course be intended to refer to someone other than Mary, in
which case we would represent it in CS with a different index, e.g. „.
(4) CS KISS(AGENT:MOTHER(MARY· )‚ , THEME:„)
The coreferentiality of two NPs corresponds to a CS relation in which the

respective arguments denote the same thing. For example, if we use two
NPs that mean MARY· , there will be two arguments in CS that denote the
person Mary, as in (1).
We have already worked out a notation for representing coreference in
the case of control (Chapter 7). For example, Mary expects to win has the
CS representation
10.1. COREFERENCE 403
(5) EXPECT(EXP:MARY· , THEME:WIN(·))
Notice that only one of the CS arguments corresponds to an overt linguistic

expression.
Alternatively, if there are two instances of Mary in a sentence that are
intended to refer to the same individual, there are two instances of MARY
in CS with the same index.
(6) a. Mary swims and Mary runs.
b. SWIM(AGENT:MARY· ) & RUN(AGENT:MARY· )
If we use a pronoun to refer to Mary a second time, then the CS represen-

tation reflects this faithfully, with a variable · that matches MARY· . In the
following examples, MARY· corresponds to Mary and · corresponds to
she.
(7) a. Mary swims and she runs.
b. SWIM(AGENT:MARY· ) & RUN(AGENT:·)
(8) a. Mary expects she will win.
b. EXPECT(EXP:MARY· ,THEME:WIN(·))
And if we use a reflexive pronoun, it corresponds to a variable in CS that

has the same index as another argument.
(9) a. Mary loves herself.
b. LOVE(EXP:MARY· ,THEME:·)
This CS representation says that the EXPERIENCER and the THEME of

LOVE are the same person, namely Mary.
It is also possible to use an epithet, such as the idiot, or a descriptive noun
phrase, such as the poor girl, for coreference.

the idiot
(10) a. Mary’s mother kissed .
the poor girl

the idiot
b. Mary swims and also runs.
the poor girl
The question that we explore in this chapter is how the syntactic structure
of a sentence constrains the distribution of elements that bear these coref-
erence relationships to one another in CS. To give just a brief idea of what
is involved, note that Mary loves her cannot have the interpretation (9b) –
that is, it cannot mean “Mary loves herself”. It is grammatical to have a
pronoun in this position, as long as it is not coreferential with the subject.
That is,
(11) a. Mary loves her.
b. LOVE(EXP:MARY· ,THEME:‚)
10.2. Binding
We begin by assuming that the possibility of coreference depends at least

in part on the syntactic structure in which the expressions appear. This
assumption is useful for expository purposes; it allows us to lay out the basic
facts and the main components of the traditional analysis of the phenomena
in the mainstream literature. 1
We focus on the noun phrases themselves and not on their CS representa-
tions. From the perspective of reference there are basically two types of NPs,
those that have independent reference and those that depend on something
else for their reference. An NP that does not depend for its reference on
another NP is called a referring expression or R-expression. An example
of such an NP is Mary. NPs that are referentially dependent are typically
referred to as pronouns. Examples of pronouns are she and herself. The latter
is a reflexive pronoun.
10.2.1. Bound anaphors
In (12) there are two instances of Mary, both of which refer (or are
intended to refer) to the same individual. As an abbreviation of the full
structure, the reference of each NP can be indicated by an index – e.g. Maryi
and herj or herself i .Assignment of the same index to two NPs indicates
coreferentiality.
(12) Maryi was at the party. I think that Maryi is really terrific.
Keep in mind that the index on an NP is a way of avoiding mentioning both

the NP and what it denotes in CS, which is where the reference is actually
represented. This point will become important in section 10.4, where we
reformulate the analysis in terms of CS.
Referentially dependent expressions may in principle get their reference
from a variety of sources. Example (13a) is a case where the reference of she
is determined from the physical context. In example (13b) the reference of
she is gotten from the discourse. In example (13c) it is gotten from the same
1
This is a simplifying assumption. The possibility that conceptual structure also
plays a role is discussed in section 10.4.
10.2. BINDING 405
sentence. As in the case of control (see Chapter 7), the NP that the reference
depends on in (13b,c) is called the antecedent. 2
(13) a. [A woman appears on a TV news program that we are watching. One of
us says to the other:] She’s the new anchorperson.
b. A: Have you met the President of the University?
B: Yes, she’s very nice.
c. Maryi thinks that shei will win.
Reflexive pronouns are special, in that in general they must get their
reference from the same sentence (at least in English), and they must be
in a particular syntactic configuration with respect to their antecedent. The
following examples illustrate.
(14) a. Maryi likes herselfi .
b. [A woman appears on a TV news program.] ∗ I like herself!
∗
c. Maryi thinks that I like herselfi .
∗
d. Maryi thinks that herselfi will win.
Elements that behave in this way are called anaphors, to distinguish them
from ordinary pronouns. Another anaphor in English is the reciprocal each
other. A reciprocal must have a plural antecedent.
(15) a. The participantsi respected each otheri .
b. [All of the participants]i respected each otheri .
c. [John and Mary]i respect each otheri .
∗
d. Johni respects each otheri .
The binding theory of MGG seeks to account for the distribution of

reflexives and pronouns with respect to their antecedents in syntactic terms.
The examples in (14) might suggest at first that an anaphor (that is, a
reflexive or a reciprocal) and its antecedent must be in the same simple
sentence. Consider the structures:
(14 ) c. ∗ Maryi thinks [S that I like herselfi ].
d. ∗ Maryi thinks [S that herselfi will win].
(16) a. ∗ [John and Mary]i think [S that I like each otheri ].
b. ∗ [John and Mary]i think [S that each otheri will win].
Let us provisionally define local to mean “in the same simple sentence”.
Then a possible condition would be the following.
2
In the cases of referential dependency that we look at here the reference of the
dependent element (she and herself ) is the same as that of the antecedent. There are
other possibilities for referential dependency, one of which is explored in Research
question 2 at the end of this chapter.
(17) An anaphor and its antecedent must be local with respect to one another.
But the following examples show that locality, although it is necessary, is

not sufficient.
∗
(18) a. Herselfi likes Maryi .
b. Maryi likes herselfi .
∗
(19) a. Each otheri like [John and Mary]i .
b. [John and Mary]i like each otheri .
In both of these pairs the two coindexed NPs are in the same sentence, but
only (18b) and (19b) are grammatical.
For simplicity of exposition we concentrate on the examples with reflex-
ives. A look at the syntactic structure shows that in (18a) the reflexive c-
commands its antecedent, while in (18b), the antecedent c-commands the
reflexive.
(20) S S
NP VP NP VP
herself V NP Mary V NP
likes Mary likes herself
The central observation, then, is that an anaphor must be c-commanded by

its antecedent (that is, by an NP with the same index), in which case we say
that it is bound.
(21) · binds ‚ if and only if
a. · and ‚ are coindexed;
b. · c-commands ‚.
Adding the locality condition brings us to the following formulation.

(22) An anaphor must be locally bound. 3
Here “bound” is a syntactic relation involving coindexed NPs.

While there is general agreement about (22), there are data in English and
other languages that suggest that the binding of an anaphor is not always
strictly local; see section 10.4.3.
3
As formulated, this condition applies to all anaphors, not just reflexives. Exer-
cise 1 asks you to investigate whether other anaphors have the same properties as
reflexives with respect to this condition.
10.2. BINDING 407
10.2.2. Bound pronouns
On the other hand, a pronoun need not have an antecedent in the same
sentence. But if the antecedent and the pronoun are in the same sentence,
and if the antecedent c-commands the pronoun, the two cannot be “local”
with respect to one another. The example in (23) shows that the antecedent
cannot locally c-command the pronoun. Examples (24a,b) show that the
antecedent may non-locally c-command the pronoun. Examples (24c,d)
show that the antecedent may be in the same sentence as the pronoun even
if it does not c-command it.
(23) Maryi loves her∗ i .
[or ∗ Maryi loves heri .]
(24) a. Maryi thinks that shei is a genius.
b. Although shei is a genius, Maryi is very humble.
c. Maryi walked in and shei sat down.
d. Although Maryi is a genius, shei is very humble.
These examples suggest that a pronoun is not acceptable just when it is

locally c-commanded by its antecedent, assuming the same definition of
“local” as “in the same simple sentence”. There are in fact other con-
texts in which pronouns are not acceptable, but the following definitely
holds.
(25) A pronoun cannot be locally bound.
A synonym for “not bound” is free – thus, “A pronoun must be locally free”
is an equivalent way of stating this condition.
These observations together constitute the core of the binding theory of
Government Binding Theory (Chomsky 1981).
(26) Binding theory (general form)
A. An anaphor must be locally bound.
B. A pronoun cannot be locally bound.
We elaborate this formulation further in the next section.
10.2.3. Condition C
There are cases where the pronoun is not locally bound, yet cannot have
a given NP as antecedent. We know that a pronoun does not have to be
bound, so that cannot be the problem with the following.
(27) a. ∗ Shei loves Maryi .

b. ∗ Shei thinks that Maryi is a genius.
A plausible hypothesis might be that the pronoun simply cannot precede its
antecedent, but this is falsified by (28), as well as by (24c) in the preceding
section. In each case, she precedes Mary.
(28) a. When shei got home, Maryi sat down in front of the TV and ate some
pizza.
b. First shei wins the lottery and now Maryi gets elected to the Senate – how
lucky can you be!
The apparent configurational difference between (27) and (28) is that in

(27) the pronoun c-commands its antecedent while in (28) it does not. This
particular relationship has nothing to do with locality. It doesn’t matter how
far away the antecedent is from the pronoun – if the pronoun c-commands
it, there is a problem.
∗
(29) Shei thinks that everyone says that it is obvious that . . . that Maryi is a genius.
These facts suggest a third condition in the binding theory. The weak form
of this condition concerns the pronoun.
(30) A pronoun cannot c-command its antecedent.
A stronger form concerns the antecedent itself, and says that an R-

expression cannot be bound. For the cases that we have considered, the
results are the same, but the stronger form also rules out cases where two
coindexed NPs that are not pronouns are in a c-command relation.
(31) ?Maryi thinks that Maryi will win.
The strong form of this condition is probably too strong, since it rules out
the examples such as the following.
(32) a. Maryi behaves as though everyone who likes Maryi is somehow special.
b. Only Johni thinks that Johni is above suspicion.
c. Maryi is here because Maryi ’s friends made her come.
In each of these examples, the first NP Maryi c-commands the second NP

Maryi , hence the latter is bound. But these sentences are grammatical, hence
this strong form of the principle is too strong.
Summarizing, we have three conditions on binding that constitute the
binding theory. Classical binding theory adopts the stronger form of the
third condition that we discussed, and for completeness we note it here.
10.3. QUANTIFICATION 409
(33) Binding theory

A. An anaphor must be locally bound.
B. A pronoun cannot be locally bound.
C. A pronoun cannot c-command its antecedent. (Weak)
[C An R-expression must be free. (Strong)]
For present purposes we continue to take “local” to be “within the same

clause”. In section 10.4 we consider the related questions of what consti-
tutes “local” and whether the proper location of binding theory is syntax
or CS.
10.3. Quantification
Besides the binding of anaphors and pronouns, there is one other very
important case of referential dependency of pronouns that we must con-
sider, which involves quantification. Example (34) illustrates.
(34) Every horsei thinks that iti will win.
In this case there are a number of horses in some group, and every horse
in this group thinks “I will win”. In this case the pronoun it is bound by
every horse. This is a different sense of “bound” from that used earlier,
although the two are related. The meaning of (34) in which it is dependent
on every horse is called the bound variable interpretation of the pronoun.
We can represent it in terms of CS as follows, where here ∀HORSE· means
“every member · of the set of horses”.
(35) THINK(EXP:∀HORSE· ,THEME:WIN(·))
The binding relationship is shown in CS by assigning the same index to the

two arguments, where one is quantified.
There is also an interpretation of (34) in which it refers to something else,
in which case it is not bound in CS by HORSE· .
(36) THINK(EXP:∀HORSE· ,THEME:WIN(‚))
And there is another sentence –

(37) Every horsei thinks that every horsei will win.
− which has a different CS representation.

(38) THINK(EXP:∀HORSE· , THEME:WIN(∀HORSE· ))
This second interpretation attributes to each horse the belief that every
horse will win, i.e. that there will be a tie among all of the horses.
Usually the bound variable interpretation occurs only when the pronoun
is syntactically c-commanded by the quantified expression. However, there
are some cases that show that the scope of the quantifier must actually be
represented at CS.
∗
(39) a. Every horsei came in from the fields and iti was hungry.
∗
b. After you feed every horsei , give iti some water.
c. Every horse’si riderj thinks iti should be fed first.
d. The rider of every horsei naturally bets on iti to win.
e. Every personi that rides a horsej bets on itj to win.
These examples show that when the quantifier phrase is inside of an NP

that c-commands the pronoun, as in (39c,d,e), under certain circumstances
the “wide” bound variable interpretation is possible. This possibility occurs
perhaps because there is a one-to-one relation between the head noun,
which projects the NP, and the quantifier phrase within it. Effectively, as
we range over every horse in the set HORSE, we range over values of the
function RIDER(HORSE) as well. So one interpretation of rider of every
horse is
(40) ∀[RIDER(HORSE· )]‚
As we go from one horse · to another · , we go from rider ‚ to rider ‚ . Sim-

ilarly, every person that rides a horse establishes a one-to-one person/horse
relationship – for each horse there is a rider.
We can represent the quantificational status of these NPs syntactically
by literally raising the quantifier every to a position where it c-commands
the pronoun (invisibly, of course), and such an approach is fairly standard
in the literature. However, it is also worth developing a CS account of the
phenomenon; see Research question 1.
Some evidence that the basic binding relation is a semantic one that is
reflected in the syntax concerns the binding of implicit arguments. These
are arguments that are implicit in the meaning of a word but not overtly
expressed. Here are some examples. 4
4
Examples of this type were first pointed out by Partee 1989, following Mitchell
1986. The examples here are based on those in Storto and Carlson to appear.
10.4. BINDING IN CS AND SYNTACTIC STRUCTURE 411
(41) a. Sandy visited a local bar.

b. Every sports fan in the country was at a local bar watching the playoffs.
(42) a. Sandy faced an enemy.
b. Every participant had to confront and defeat an enemy.
Not only can local and enemy be interpreted with respect to Sandy but they
can also be interpreted relative to the quantifier every. So, for each sports
fan there is a bar local to that fan, and for each participant there is an
enemy of that participant. Problem 2 asks you to formulate representations
for these cases that account for the binding relationships.
10.4. ∗ Binding in CS and syntactic structure
10.4.1. The GB binding theory
Let us look again at the analysis of the “raising to object” construction in

MGG that we discussed in Chapter 7, as exemplified by (43b).
(43) a. George expects [S (that) Al will win]

b. George expects [S Al to win].
In mainstream syntactic analyses, as we saw in Chapter 7, the NP that

is interpreted as the subject of a non-finite complement is the syntactic
subject of the complement, as shown in (43b). This assumption is relevant
for binding theory, because it turns out that this NP acts as though it is
local with respect to the higher clause following the binding conditions
(33).

she
(44) a. Maryi expects [S that ∗ i will win].
herselfi
∗
heri
b. Maryi expects [S to win].
herselfi
As can be seen, the subject NP of an infinitival complement cannot be a

pronoun and must be a reflexive if it has an antecedent in the higher clause.
Thus, if both (33) and the structure shown in (44) are essentially correct,
the relation “local” has to be defined so that Mary and her/herself in (44b)
are local with respect to one another even though they are not in the same
clause.
Much of the history of contemporary syntactic theory turns on the res-

olution of this issue. 5 Mainstream generative grammar has adopted the
structure in (44b) because of its uniformity with that of (44a). On this
approach, the NP that receives the “subject” θ-role of the infinitival VP
is a syntactic subject of a sentential complement of a verb like expect. The
problem then is to account for the fact that her/herself behave as though
they are the direct object of expect. Other syntactic theories have in one way
or the other taken the view that her/herself in (44b) are arguments of expect
and not subjects of a different clause. On this latter approach, locality can
be interpreted in terms of “in the same clause”, what has come to be called
the clausemate relation.
The notion that the NP that follows expect is the direct object is also
supported by the observation that this NP becomes the subject in the
passive, e.g.
(45) Al is expected by George [to win].
but the subject of the tensed S does not

∗
(46) Al is expected by George (that) __ will win.
Thus, assuming that what appears to be the direct object (e.g. Al in (43b)
∗
heri
and in (44b)) is really a subject poses a serious problem.
herselfi
We review here briefly how the problem is resolved in MGG. As discussed
in Chapter 7, section 7.3.2, the subject of the infinitive is accessible to
the higher verb for case marking through the relation of government. A
cornerstone of Government Binding (GB) theory is the definition of locality
in terms of government. The verb expects in (44b) governs the subject of the
infinitive; locality can then be defined in terms of this verb and constituents
that bear a structural relation to this verb.
The following definition of governing category conveys the central defini-
tion of locality in GB theory.
(47) The governing category for · is the minimal category that contains
a. ·,
b. the governor „ of ·,
c. the subject of „.
5
See Chapters 2 and 3 of Culicover and Jackendoff 2005.
So, in the case of (44b), the governing category of the subject of the infinitive
is the higher IP, as shown in (48).
(48)
IP1
NP1 I
I0 VP
V0 IP2
· I
I0 VP
V is the governor of ·, NP1 is the subject of V, so IP1 is the governing

category of ·.
In (48) V0 is the governor of ·. But in a case where the complement of
the verb is a tensed S, the complement is a CP. By assumption, CP blocks
government – see Chapter 7, section 7.4.2. Hence NP1 and · are not in the
same governing category in this case, as illustrated in (49).
(49)
IP2
NP I
I0 VP2
V0 CP
expects Spec C
government
C0 IP
· I
no government by expect I0 VP
Given such a definition, and the terminology “free” instead of “not

bound”, the binding principles can be reformulated as follows. (We show
just the strong version of condition C, which is the one adopted in GB
theory.)
(50) GB binding theory

A. A reflexive must be bound within its governing category.
B. A pronoun must be free within its governing category.
C. An R-expression must be free.
This approach to binding is characterized by the fact that it takes binding

to be a syntactic phenomenon.
10.4.2. CS- and GF-binding
There are a number of phenomena in English and other languages that

challenge or appear to challenge this strictly configurational account of
binding. These phenomena suggest that it may be preferable to formulate
binding in terms of CS coindexing and the hierarchy of GFs introduced in
Chapter 5 as part of the account of linking between CS and syntax.
The first case that we consider is one in which a pronoun appears to be
locally bound, but no ungrammaticality arises. An anaphor is also possible
in this position.

himi
(51) Johni saw a snake near · [locative PP]
himselfi
On the assumption that there is a single syntactic structure involved here,

it does not seem possible to have a syntactic definition of binding that will
permit both him and himself. If him is permitted because it is not locally
bound, himself should be excluded. And if himself is permitted because it
is locally bound, him should be excluded.
Consideration of the meaning of this sentence suggests that there is a
relation between John· , a snake, and near · that is not reflected in the
syntax. We sketch out the correspondence in (52). It has the form of a
secondary predication construction, as discussed in Chapter 8.
(52)
SYNTAX S
NP VP
John V NP PP
saw a snake P NP
near him
GF Subject Object
·
CS SEE(EXPERIENCER:JOHN , THEME:SNAKE‚, LOCATION:NEAR (‚, ·) )
Crucially, the bound variable · is an argument of a relation distinct from

SEE. This correspondence suggests that him is acceptable because it is not
an argument of the relation SEE.
On the other hand, we have seen that an anaphor is also possible in this
context. We assume that the meaning of this case, with himself, is the same
as with the pronoun. Here is the correspondence.
(53)
SYNTAX S
NP VP
John V NP PP
saw a snake P NP
near himself
GF Subject Object
·
SEE(EXPERIENCER:JOHN , THEME:SNAKE‚, LOCATION:NEAR (‚, ·) )
CS
Note that the only difference between this correspondence and (52) is
that here the complement of near is himself, while in (52) it is him.
Notice that while · is not an argument of SEE, the NP himself is within
the same simple S as John. These examples are interesting because they
show that there are syntactic configurations that allow both pronouns and
anaphors.
This observation suggests that the anaphor is licensed because of a rela-

tionship between the GFs associated with saw. We define two closely related
binding relations, one for CS and one for GF, to try to capture what is
going on here. First, we define CS-binds, which holds between arguments
at CS.
(54) · CS-binds ‚ if and only if

a. · and ‚ are coindexed;
b. ·, ‚ are contained within the scope of a CS relation R.
The scope of R is the part of CS within the parentheses associated with

R(. . . ). “Locally bound” then becomes locally CS-bound.
(55) · locally CS-binds ‚ if and only if

a. · CS-binds ‚;
b. ·, ‚ are arguments of a CS relation R.
Next, we define GF-binds, which holds between arguments that correspond

to GFs.
(56) · GF-binds ‚ and only if

a. the GFs corresponding to · and ‚ are coindexed;
b. the GF of · is higher in the hierarchy than the GF of ‚;
c. the GFs correspond to the same clause.
For the definition of GF-binding, we assume the linking hierarchy of Chap-

ter 5, in which Subject is higher than Object, and Object is higher than
oblique argument. We assume that GFs are coindexed if they correspond to
coindexed CS arguments.
(57) L INKING (D EFAULT

  )  
 Agent   Subject 
↔
 ⇓   ⇓ 
Let us review some key examples with these definitions and assumptions
in mind. For
(58) Mary excused herself.
we have the correspondence

(59) SUBJECT S
NP VP
Mary V0 NP
excused herself
GF Subject· Object·
CS EXCUSE(AGENT:MARY·, PATIENT:·)
Here, herself is CS-bound because MARY· and · are coindexed, and they
are both contained in the scope of the relation EXCUSE. And herself is
GF-bound, because the Subject and Object are coindexed, and the GF
corresponding to MARY· , namely Subject, is higher on the hierarchy than
the GF corresponding to ·, namely Object.
If the anaphor and the antecedent are not arguments of the same CS-
relation or arguments of the same verb, ungrammaticality results.
(60) a. ∗ Mary expects that John will excuse herself.
b. ∗ Mary expects John to excuse herself.
c. EXPECT(EXP:MARY· ,THEME:EXCUSE(AGENT:JOHN,
THEME:·))
In these cases, the anaphor is not locally CS-bound. As shown in (60c), · is

an argument of EXCUSE but MARY· is not, so they are not arguments of
the same CS relation.
In (61a) himself is an oblique argument, and its antecedent is the Subject
of the same sentence. In the corresponding CS representation (61b), the
argument · is CS-bound but not an argument of the same CS relation as
JOHN· .
(61) a. John saw a snake near himself.
b. SEE(EXPERIENCER:JOHN· ,THEME:SNAKE‚ ,LOCATION:
NEAR(·,‚))
A preliminary hypothesis, then, is the following.

(62) An anaphor must be locally GF-bound. If it corresponds to a CS argument,
that argument must be CS-bound.
This condition is the counterpart of Condition A of the binding theory.

It rules out (60a,b) because the anaphor is not GF-bound, although it
corresponds to a bound CS argument. If the anaphor is the subject –
∗
(63) Himself loves John.
– it is not GF-bound, and hence is ruled out. An example such as

(64) a. John saw a snake near him.
b. SEE(EXPERIENCER:JOHN· ,THEME:SNAKE‚ ,LOCATION:
NEAR(·,‚))
shows that a pronoun cannot be locally CS-bound, but it can be locally

GF-bound and non-locally CS-bound. JOHN· does not locally CS-bind ·,
because · is not an argument of SEE, it is an argument of NEAR.
(65) A pronoun cannot be locally CS-bound.
This condition is the counterpart to Condition B of the binding theory.

To summarize to this point, see a snake near himi /himself i works as it does
because the variable is GF-bound but not locally CS-bound. The fact that
it is GF-bound allows for the reflexive, while the fact that it is not locally
CS-bound allows for the pronoun.
The CS representation of (64) differs crucially from that of a sentence
such as
∗
(66) Johni told Mary about himi .
In this case, him must be an argument of told, and is thus both CS-bound
and GF-bound.
(67) TELL(AGENT:JOHN· ,GOAL:MARY,THEME:·)
The fact that there is a GF condition for binding as well as a CS condition

also allows us to account for cases where an anaphor does not correspond
to a CS argument, e.g.
(68) a. Maryi was behaving herselfi .
b. Johni perjured himselfi .
Behave and perjure take only one CS argument, but they are syntactically
transitive and reflexive. In this case the reflexive does not correspond to
a CS argument, but it does correspond to a GF. Problem 8 asks you to
formulate the lexical entries for these verbs in order to account for their
syntactic argument structure.
The distribution of pronouns in terms of CS- and GF-binding is similarly
straightforward. An example such as
∗
(69) Maryi saw heri .
shows that the pronoun cannot be locally CS-bound.

In addition to John saw a snake near him, there are a number of other
examples that are problematic for a strictly configurational account of
binding in terms of the binding theory. Some English cases are exemplified
in (70).
∗
?himi
(70) a. Johni bought a nice picture of at a garage sale. [picture NP]
himselfi

b. Johni thinks that attractive pictures of himi are virtually priceless.
himselfi
[locality violation]

himi
c. Unpleasant stories about upset Johni . [“psych” verbs]
himselfi
Example (70a) shows a reflexive inside of an NP. On the assumption that

NPs have subjects just like Ss, following the DP structure of Chapter 4,
section 4.7.2., the governing category of the reflexive in this case would seem
to be the NP, since the governor is the N picture. This problem can be dealt
with by observing that there is no overt subject in this case that could be the
antecedent of the reflexive, and by redefining “governing category” slightly
so that it takes into account the availability of an antecedent. When there
is an antecedent in the NP, the governing category is the NP, as in John’s
picture of himself. 6
The same approach will work for (70b). Here pictures of himself is the
subject of the lower clause and cannot be its own antecedent, so the next
available antecedent is the subject of thinks. The problem, however, is that
him is also grammatical in this configuration. But the governing category
of him cannot be the higher sentence since the pronoun must be free in its
governing category and it would be bound in this larger governing category.
Again, the trick is to pick the smallest possible domain in which the con-
dition applies, which in this case would be the NP, in which the pronoun is
free.
Given the availability of a CS/GF account of binding, it is reasonable
to explore whether such an account might work for picture NPs. Such an
account would make crucial use of the semantics of the head noun. We do
not try to develop such an account here, but leave it for a Research question.
It is apparent that there is no c-command relation in (70c), which explains
why the pronoun is acceptable – it is free in its governing category. But then,
6
This is the analysis of Chomsky 1986.
so is the reflexive, yet the reflexive is also grammatical. This case provides
further evidence that there may be semantic as well as syntactic conditions
on the well-formedness of reflexives.
10.4.3. Long distance anaphora
What we are seeing here is not an isolated phenomenon. For instance,

in many languages there are anaphors that are bound to non-local
antecedents; this is called long distance anaphora. In most cases, the form
of the long distance anaphor is different from that of the local anaphor. We
give a few examples from Chinese, Italian, Dutch and Japanese.
(71) Chinese
a. Zhangsani renwei Lisii zhidao Wangwuk xihuan zijii/j/k
Zhangsan thinks Lisi knows Wangwu likes self
‘Zhangsani thinks Lisij knows Wangwuk likes himi/j /himselfk .’
b. Zhangsani renwei woj zhidao Wangwuk xihuan ziji∗ i/∗ j/k
Zhangsan thinks I know Wangwu likes self
‘Zhangsani thinks Ij know Wangwuk likes himselfk .’
[Cole and Hermon 1998:57, 62]
In Chinese the anaphor ziji may have a local or a long distance antecedent.
However, an intervening first person blocks the long distance interpretation
of the antecedent.
(72) Italian
a. Credo [ che Marioi sostenga [che tu abbia
I-believe that Mario claims.SUBJUNCT. that you have.SUBJUNCT.
parlato di sei e della sua famiglia in TV]].
spoken of self and of-the his family on TV
‘I believe that Mario claims that you spoke about him and his family on
TV.’ [Giorgi 1984:335]
b. ∗ Giannii pensava [che quella casa appartenesse ancora a
Gianni thought that that house belonged still to
se stessoi ].
himself
‘Gianni thought that that house still belonged to him.’ [Giorgi 1984:314)
In Italian the anaphor se may have a long distance antecedent if it is in

a subjunctive clause. The anaphor se stesso, on the other hand, is a local
anaphor and cannot have a long distance antecedent.
(73) Dutch
a. Max haat zichzelf.
Max hates self-self
‘Max hates himself.’
b. ∗ Max haat zich.
Max hates self
‘Max hates himself.’
c. Max hoorde mij [over zich praten].
Max heard me about self talk
‘Max heard me talk about him.’
∗
d. Max hoorde mij [over zichzelf praten].
Max heard me about self-self talk
‘Max heard me talk about him’
[Examples (71)–(73) from Cole et al. 2001:12]
In Dutch, zichzelf is a local anaphor, as shown in (73a), but not a long

distance anaphor (73d). Zich is not a local anaphor (73b) but a long distance
anaphor (73c).
(74) Japanese
Taro-wa Ziro-ga zibun-ga atama-ga ii to itta to
Taro-TOP Ziro-NOM self-NOM head-NOM good COMP say-PST COMP
itta
say-PST
‘Taroi said that Ziroj said that hei,j is smart.’
[McCready 2006]
(75) a. Taroi -ga [ Ziroj -ga zibuni/j -no kako-o Hanako-ni katatta
Taro-NOM Ziro-NOM SELF-GEN past-ACC Hanako-DAT told
to ] itta.
that said
‘Taroi said that Ziroj told hisi/j past to Hanako.’
b. [Hanakoi -wa [S Taro-ga [S Ziro-ga zibuni -o semeta ] to ] itta ]
Hanako-TOP Taro-NOM Ziro-NOM SELFi -ACC blamed that said
‘Hanakoi said that Taro said that Ziro blamed SELFi (=Hanako).’
(76) Mikei -ga zibuni -o semeta.
Mike-NOM self-ACC blamed.
[Kinoshita 2000]
The anaphor zibun in Japanese serves as both a long distance anaphor (74)–
(75) and a local anaphor (76).
The long distance anaphors have two particularly distinctive proper-
ties. First, they are fixed monomorphemic forms, e.g. Chinese ziji, Dutch
zich, Japanese zibun, Italian se. They do not show agreement, unlike the
inflected myself, himself, etc. Second, they must have a subject antecedent.
An account of long distance anaphors in terms of CS-/GF-binding is able
to implement the second requirement directly. The lexical entry for such an
anaphor, Japanese zibun, is given in (77).
(77) PHON /žibUn/
SYNTAX N
GF Subject
CS X· ·
As a consequence of the link between · and Subject, the antecedent of zibun

will have the Subject GF in any sentence in which it appears.
10.5. ∗ Reconstruction
10.5.1. A constructions and binding
We can now explore how binding and A constructions such as wh-

questions and topicalization interact. Generally a constituent in A position
typically behaves with respect to binding as though it was in the position of
the gap. A simple example that illustrates this point is the following.
(78) a. Maryi talks to herselfi frequently.

b. Herselfi , Maryi talks to t i frequently.
On a movement analysis, we would naturally consider applying the bind-

ing conditions before movement. On a non-movement analysis, we have to
consider what the relationship is between the A constituent and the gap, a
relationship that is expressed in terms of the chain that they form. In order
to simplify the discussion, we focus on the non-movement approach.
If the MGG binding theory of (33) is essentially correct, we must con-
clude that the reflexive is locally bound. That is, it has an antecedent that
10.5. RECONSTRUCTION 423
locally c-commands it. Maryi locally c-commands herself in (78a), but not
in (78b). So one of the following may be true.
(79) i. Binding condition A applies to the A chain, and (by definition) · c-
commands ‚ if · c-commands the trace of ‚ in argument position.
ii. Binding condition A applies to a representation in which the A
constituent is in the argument position.
On option (i), Mary actually c-commands herself in (78b) because Mary

c-commands t i in the chain of which herself is the head and t i is the tail.
On option (ii), Mary actually c-commands herself, which is the argument.
There are a number of variants of (ii); on one variant, the interpretation of
(78b) is constructed from the surface structure by putting the constituent
in A position into the position occupied by the trace. The result is an LF
representation, and the operation is called reconstruction.
Note that simple cases of reconstruction follow directly from our account
of binding in terms of CS/GF. Topicalization affects only the syntactic con-
figuration but not the CS representation or the assignment of GFs. In (78b)
herself is an oblique argument because it forms a chain with the gap, Mary
is a Subject, hence Mary GF-binds herself. The variable · corresponding to
herself in CS is locally CS-bound by MARY · .
(80)
SYNTAX S
NP NP VP
herself Mary V PP
talks P [e]
GF Subject
CS TALK(AGENT:MARY·, GOAL:·)
IS TOPIC·
Therefore, if (78a) is a valid case of anaphor binding, (78b) must be as well,

since the two are identical in terms of CS and GFs.
There are more complex cases that have been taken to argue for a move-
ment analysis of A constructions. Consider the following examples. 7
(81) a. Himselfj , Maryi claims Johnj talks to t j frequently.
b. It is himselfj that Maryi claims Johnj talks to t j frequently.
(82) a. Herselfi , Maryi claims Johnj talks to t i frequently.
b. It is herselfi that Maryi claims Johnj talks to t i frequently.
In (81) Johni locally c-commands the trace of himself. But in (82), Johni
locally c-commands the trace of herself, and the sentence should be ungram-
matical.
While the examples in (81) fall under either (i) or (ii), (82) suggests an
analysis in which there is a trace in the complementizer position of each
clause. This structure is consistent with the successive cyclic movement
analysis proposed in MGG, as discussed in Chapter 9, section 9.7.1.
(83) herselfi [Maryi claims [t i Johnj talks to t i frequently]]
On this view, Maryi locally c-commands the trace t i , which forms a chain
with herself i , and this is why a binding relation is possible.
However, if the trace in the initial position of the embedded sentence is
in fact local with respect to Maryi , so that the reflexive linked to it can be
bound, there is an apparent problem with sentences like the following.
∗
(84) Maryi says [that herselfi , Johnj talks to t i frequently].
[But cf. ?Maryi says [that it is herselfi that Johnj talks to t i frequently].]
Notice that in (84), Mary locally c-commands herself, but not the trace of
herself. This example thus poses a problem for the view that in (83), Mary
locally c-commands t i .
With this in mind, we could formulate condition A so that the reflexive
is locally bound only if (a) it has a local antecedent when it is not in an A
position or (b) a trace in its chain has a local antecedent. While complex,
this condition appears to take care of the three cases of reflexive binding
that we have encountered thus far:
(85) a. Maryi talks to herselfi frequently. [reflexive is locally bound]
b. Herselfi , Mary talks to t i frequently. [trace of reflexive is locally bound]
c. Herselfi , Maryi claims [t i John talks to t i frequently]. [intermediate trace
is locally bound]
7
These examples are not all perfect, but are judged acceptable by many speakers.
10.5. RECONSTRUCTION 425
In (85a) the reflexive is in situ, and it has a local antecedent. In (85b), the
reflexive has moved, and its trace has a local antecedent. And in (85c),
the reflexive has moved long distance and leaves a locally bound intermedi-
ate trace.
Stating the various conditions under which the reflexive can be bound
as we have done here covers the various observed cases, but it does not
constitute a particularly elegant account of the phenomenon. Example (84)
is particularly problematic. The reflexive is part of a chain that is bound
by John, which appears to rule out the possibility that the reflexive can be
bound by Mary. But in (81) and (82) the two possibilities appear to coexist
comfortably. The next two sections discuss various approaches to dealing
with this phenomenon.
10.5.2. The copy theory of movement
An alternative to reconstruction that has been proposed in recent years

within the context of A movement is that the gap is an invisible exact copy
of the A constituent. It is the copy itself that is subject to the binding theory.
This approach incorporates a syntactic account of binding, and avoids
reconstruction, but has very much the same effect. On this approach, (81a)
would have the following structure, where strikeout indicates phonetically
empty structure.
(86) Himselfj , Maryi claims Johnj talks to himselfj frequently.
While the reflexive is not bound, the copy is. We will see that this solution
does not generalize to the full range of cases where binding theory appears
to apply across syntactic binding domains.
Let us look briefly at how condition B of the binding theory interacts
with A constructions. According to this condition, a pronoun cannot be
locally bound. If a pronoun is locally bound in a given position, then if we
topicalize it we would expect it to continue to be locally bound, given the
corresponding pattern for reflexives. And if a pronoun is excluded for any
other reason, topicalizing it should not have any effect, assuming that the
original argument position determines the behavior of the pronoun with
respect to the binding theory. We illustrate using copies.
(87) a. ∗ Johni is proud of himi .
b. ∗ Himi , Johni is proud of himi .
(88) a. ∗ Johni would never write [a book about himi ]j .

b. ∗ [A book about himi ]j , Johni would never write [a book about himi ]i .
The data appears to support the view that from the perspective of con-
dition B, it is irrelevant whether the pronoun is in its argument position
or in an A position. The copy theory of movement captures this insight, as
does a non-movement account in which condition B applies not to syntactic
structure but to CS, as sketched in section 10.4.2. Nevertheless, there is an
important counterexample to this general perspective, regardless of how it
is implemented, which we take up in the next section.
10.6. ∗ Crossover and anti-reconstruction
It was noted very early in the syntactic literature 8 that there are com-
plex interactions in A constructions where the antecedents “cross over”
their dependents. (We have seen cases where the dependents cross over
the antecedents already, which motivate “reconstruction”.) When the
antecedent crosses over the dependent, it may end up in a position where it
c-commands the dependent. The question then arises as to whether binding
is possible. Here are some examples.
∗
(89) a. Shei is very proud of Maryi .
∗
b. Whoi is shei proudest of t i ?
(90) a. Heri father is really supportive of Maryi .
∗
b. Whoi is heri father really supportive of t i ?
∗
c. Heri father is really supportive of every studenti .
Example (89b) exemplifies strong crossover. In this case, the pronoun c-

commands the trace of the moved antecedent and the moved antecedent
c-commands the pronoun. (90b) exemplifies weak crossover. In this case,
the pronoun does not c-command the trace of the moved antecedent but the
moved antecedent c-commands the pronoun. Also of interest is (90c), where
there is no movement, yet the coindexing is problematic. Weak crossover
does not appear to occur in A constructions.
(91) Every studenti seems to heri father [CP t i to be destined for success].
This observation is consistent with the fact that A constructions do not

show reconstruction; see Problem 9.
8
Postal 1971.
10.6. CROSSOVER AND ANTI-RECONSTRUCTION 427
Strong crossover in (89) appears at first glance to be a special case of

condition C of the binding theory. On this view, she in (89) c-commands
its antecedent, and there is syntactic or CS reconstruction. This picture is
somewhat clouded by the fact that when what is in A position contains the
antecedent, condition C does not always apply. 9
(92) a. ∗ Hei later denied your claim that Johni was asleep.

Which
b. ∗ [ claim] that Johni was asleep did hei later deny?
Whose
(93) a. ∗ Hei later denied the claim that Johni made.
b. [Which claim that Johni made] did hei later deny?
(94) a. ∗ Hei printed the uglier picture of Tomi .
b. ∗ [Which picture of Tomi ] did hei print?
(95) a. ∗ Hei later printed the ugliest picture that Tomi took.
b. [The ugliest picture that Tomi took], hei later printed.
[Examples based on Munn 1994:399]
The (a) examples show that when the pronoun c-commands its antecedent,
condition C applies. But when the antecedent is in a relative clause, and the
relative clause is moved, as in (93b) and (95b), condition C does not apply. It
is as though reconstruction of the relative clause is somehow blocked. These
cases are therefore referred to as anti-reconstruction. Syntactic accounts
of anti-reconstruction formulate the interaction between movement and
the binding theory so that, at the point at which condition C applies, the
pronoun does not c-command its antecedent. This can be done in a number
of ways, but all of them have the flavor of a stipulation.
One particularly influential suggestion is that of Lebeaux 1990. Lebeaux’s
proposal is that, as syntactic structure is built up, first arguments are
introduced, then there is movement, and then adjuncts are introduced into
syntactic structure in a way that reflects their surface position. Condition
C applies whenever structure is built. On this approach, condition C will
rule out all of the (a) examples, since no movement applies and the pronoun
c-commands the antecedent. In examples (92a) and (94a), the antecedent
is contained in an argument. Then condition C applies, and there is a
violation. When there is subsequent movement, as in (92b) and (94b),
condition C has already applied, and the examples are ungrammatical. But
in examples (93b) and (95b) the antecedent is in an adjunct, which is not
9
The literature suggests that for some speakers of English, the b examples are all
acceptable.
introduced into the structure until after movement. Hence condition C does
not apply.
10.7. Summary
This chapter has looked at reference relations between expressions, in

particular, coreference, anaphora, and quantification. We have seen that
the coreference and binding possibilities are constrained by some simple
principles. The original formulation of these principles strictly in terms of
syntactic structure in GB theory is very elegant, but fails to deal with a
number of problematic cases.
We showed that it is possible to reformulate these principles in terms
of the levels of representation that enter into syntax/CS correspondences,
that is, GF-binding and CS-binding. These principles cover the phenomena
that fall under the classical MGG binding theory, and go beyond them. For
example, given GF-/CS-binding, we do not have to assign complex syntactic
representations to sentences like John saw a snake near him in order to make
the binding facts work out.
Exercises
1. Review the distributional properties of reflexives illustrated in (14) in

the text (as well as any that you happen to have noticed on your own). The
most salient is that the reflexive in argument position be locally bound by
its antecedent, but there are other possibilities. For example,
(1) a. John tried to sell Mary a picture of himself.
b. As for herself, Mary says that she is going to spend the winter in Acapulco.
c. It is herself that Mary thinks should be elected president.
d. The person that John is most proud of is himself.
e. The person that John thinks Mary is really angry at is himself.
Do all of these have the property that the reflexive is bound by its
antecedent? Explain your answer.
2. Compare the reflexives in Exercise 1 with the reciprocal (each other).
Are the distributions of the two exactly the same? If not, are the differences
in distribution systematic? Can the these cases be accommodated within the
binding conditions of (26)?
[§10.2.]
EXERCISES 429
3. Show how each of the following sentences is consistent with the binding
conditions in (26) in the text.
(1) a. The dog in the corner is scratching itself.
b. Maryi loves heri mother.
c. Although shei is a genius, Maryi is very humble.
d. Maryi walked in and shei sat down.
e. Although Maryi is a genius, shei is very humble.

hei
f. Johni ’s motherj thinks that should have been asked to drive.
shej
[§10.2.]
4. Explain why each of the following sentences is ungrammatical in terms

of the binding conditions in (26) in the text. Make special note of any cases
where you have to assume a particular definition of “local” in order to have
the principles apply correctly.
∗
(1) a. The childreni were fighting with themi .
∗
b. Johni says it was obvious to himselfi that the theory was correct.
∗
c. I saw Maryi and gave herselfi the book.
∗
d. Johni called, but himselfi couldn’t talk too long.
∗
e. It bothers herselfi that Maryi can’t speak Chinese.
[§10.2.]
5. Explain why the following examples are problems for the binding
theory in (26) in the text. The cases marked as grammatical are acceptable
for some speakers of English.
(1) a. The studentsi were all wondering [CP what each otheri would say].
[cf. ∗ The horsei was wondering [CP what itselfi would eat]]
b. ?The studentsi all believed [that each otheri would solve the problem].
[cf. ∗ The horsei was convinced [that itselfi would win].]
c. The studentsi all were wondering [what would happen to each other].
[cf. ∗ The horsei was wondering [what would happen to itselfi ].]
d. ∗ The studentsi all believed [that nothing would happen to each otheri ].

each otheri
e. ∗ The studentsi all believed [that I would be nice to ].
themselvesi
Is there a generalization that distinguishes the grammatical examples from
the ungrammatical examples?
[§10.2.]
∗
6. Consider the following cases of the bound quantifier interpretation.
Pay attention to the commas, which indicate intonational breaks.
 
 before 
(1) a. No studenti handed in heri exam after shei heard the bell.
 when 
∗
b. No studenti handed in heri exam, although shei hadn’t heard the bell.
c. No studenti handed in heri exam in spite of having heard the bell.
∗
d. No studenti handed in heri exam, in spite of having heard the bell.
e. No studenti handed in heri exam because shei hadn’t heard the bell.
∗
f. No studenti handed in heri exam, because shei hadn’t heard the bell.
Evaluate the extent to which the binding facts shown here are consistent
with the constituency of the adjuncts, as revealed in the standard con-
stituency tests. (Hint: What configurations do the binding facts argue for,
in view of the binding theory? Are these compatible with the constituency
tests?)
[§10.3.]
7. Explain how the following examples are accounted for in a syntactic

approach to binding theory if we assume the syntactic account of raising
and control.
∗
heri
(1) a. Maryi tried to exonerate .
herselfi
∗
heri
b. I persuaded Maryi to throw on the mercy of the court.
herselfi
∗
heri
c. Maryi seems to be proud of .
herselfi
(Hint: First, figure out what the syntactic structures are, assuming PRO and
raising. Then consider how PRO and the trace of the raised NP interact with
the pronoun and the anaphor.)
[§10.4.]
8. Formulate an account of the facts in Exercise 7 in terms of correspon-

dences, without assuming movement. Rather, apply the CS-/GF-binding
conditions (54) and (56) in the text. Compare your analysis to that of
Exercise 7.
[§10.4.]
Problems
∗
1. The following sentences show that it is necessary to formulate the
Binding theory in terms of CS relations, if control does not involve an empty
subject NP.
PROBLEMS 431
(1) a. To respect oneself is absolutely necessary.

b. It is very important not to take oneself too seriously.
First, show that this conclusion follows. Next, formulate Condition A as a

condition on the correspondence between CS and syntax. Assume that the
CS representation of an anaphoric relation is one in which two arguments
of the same relation have the same index, i.e.
(2) F(X· ,·)
The condition will have to mention the CS representation and the distribu-
tion of the reflexive pronoun.
[§10.2.]
2. A pronoun cannot be locally c-commanded by its antecedent. However,

the following sentence is grammatical.
(1) Maryi loves heri mother.
How must the definition of “local” be changed so that this sentence is not
ruled out by condition B of the binding theory?
[§10.2.2.]
3. Some quantifiers reside in the meanings of words like always and often,
e.g.
(1) a. Someone always comes late to the party.

b. At a party like that I often drink a beer.
The interpretation of (1a) can be that there is a particular person who

always comes late to every party, or that for every party there is someone
that comes late to it. The interpretation of (1b) can be that there is a par-
ticular beer that I often drink, or that it is often the case that I drink a beer
(a different one). The scope ambiguity is signaled by the adverb, but it also
involves the simple present tense in English, which is interpreted as habitual
or repeated action. How would you account for this scope ambiguity in
terms of syntactic movement? In terms of a correspondence with CS?
[§10.3.]
∗
4. The pronoun it can be bound by her exam in example (1a), but not in
example (1b).
(1) a. Every student handed her exam in before it was completed.

b. ∗ Every student handed her exam in, although it was clearly incomplete.
Discuss these facts from the perspective of an analysis of the bound quan-
tifier interpretation in terms of syntactic configuration and in particular c-
command. In particular, what has to be added to the representation so that
her exam binds it in (1a)?
[§10.3.]
5. Consider the following pair of sentences discussed in the text.

(1) a. Every horse thinks it will win.
b. Every horse thinks every horse will win.
State explicitly the correspondence rules between the syntactic structures

and the CS representations that will capture the meaning difference. Then,
extend your account to the difference between the following sentences.
(2) a. Every horse loves itself.
b. Every horse loves every horse.
Finally, does your account explain the following difference?

(3) a. Every horse is fast and every horse is a potential winner.
b. Every horsei is fast but it∗ i is not necessarily a potential winner.
(Hint: The problem here is how to distinguish between an · at CS that is

simply coindexed with another NP and one that is bound by it.)
[§10.3.]
6. Consider the following sentence.

himselfi
(1) Maryj bought from Johni a picture of .
herselfj
Assume that reflexives in picture NPs are covered by the same theory that
deals with reflexives in verbal argument positions. Why does this sentence
pose a problem for the syntactic binding theory?
[§10.4.]
7. Keeping in mind your answer to Problem 6 consider the examples
in (1).

himselfi
(1) a. Johni talked to Maryj about .
herselfj

himselfi
b. Johni talked about ∗ to Maryj .
herselfj

himselfi
c. Johni talked about Maryj to ∗ .
herselfj

himselfi
d. John talked to ∗ about Maryj .
herselfj
PROBLEMS 433
Show that these sentences pose an additional problem for the binding the-
ory (beyond that identified in Problem 6), whether the theory is formulated
in syntactic or semantic (that is, CS) terms.
[§10.4.]
8. Formulate the lexical entries for the verbs behave and perjure in order
to account for the following data.
(1) a. Maryi was behaving herselfi .

Ottoi was behaving himselfi .
To behave oneself is good.
∗
To behave Susan is good.
b. Johni perjured himselfi .
Susani perjured herselfi .
To perjure oneself is bad.
∗
To perjure Otto is bad.
(2) a. ∗ Maryi was behaving Susan.
b. ∗ John perjured Otto.
(3) a. Maryi was behaving.
b. ∗ John perjured.
[§10.4.]
9. A persistent question in derivational approaches to syntax is whether

there is reconstruction in the case of A movement. What do the following
examples have to say about this question?
(1) Johni seems to himselfi to be the most likely winner.

(2) a. It seemed to Johni that hei would win.
b. ∗ Hei seemed to Johni to have won.
(3) a. Every horsei was fed by itsi rider.
b. ∗ Heri horse was fed by every rideri .
[§10.6.]
10. Work out the derivation of the examples in (92)–(95) in the text to
test Lebeaux’s account of binding. The key elements are: (a) condition C
applies whenever structure is built; (b) arguments must be introduced before
movement; (c) adjuncts are introduced late into the structure, even after
movement, in their surface position.
[§10.6.]
Research questions
1. Consider the sentence

(1) Every horse thinks that every horse will win.
It is possible but not necessary that the two sets of horses are the same, just
as it is possible but not necessary that an R-expression like Mary refers to
the same person when it is used twice in the same sentence. When corefer-
ence is intended, the two arguments in CS that correspond to the two NPs
should have the same index. For some arbitrary relation F we would have
(2) F(. . . ∀HORSE· , . . . , ∀HORSE· , . . . )
How do we make sure that this CS is not equivalent to one in which the
second argument is just the variable · that represents the bound variable
interpretation?
(3) F(. . . ∀HORSE· ,. . . , ·,. . . )
[§10.3.]
2. Examples (41)–(42) in the text, repeated here, show some cases in which
a binding relationship does not involve a pronoun but implicit arguments.
(41) a. Sandy visited a local bar.
b. Every sports fan in the country was at a local bar watching the playoffs.
(42) a. Sandy faced an enemy.
b. Every participant had to confront and defeat an enemy.
A. Is the binding of an implicit argument subject to exactly the same

conditions as the binding of a pronoun with respect to the binding theory?
B. What is the lexical representation of a word that has a bound implicit
argument such that the binding properties follow when it is inserted in a
sentence? Does this representation account directly for the observations in
A, or are additional stipulations necessary?
C. What do your answers to these questions suggest about how binding
theory should be formulated more generally, in terms of the roles of syntax
and CS?
A reasonable approach to these questions would be to construct the
CS representation of a sentence like (41b) so that it expresses the desired
binding properties, and proceed from there.
[§10.3.]
3. Formulate the binding conditions on the distribution of R-expressions

in terms of CS- and GF-binding.
[§10.4.]
∗
4. Formulate an account of the conceptual structure of picture NPs
such that binding conditions in terms of CS- and GF-binding will yield
the following judgments.
∗
?himi
(1) a. Johni bought a nice picture of at a garage sale.
himselfi

himi
b. Johni thinks that attractive pictures of are virtually priceless.
himselfi
[§10.4.]
∗
5. Extend the binding theory account developed in the previous question
to cases involving “psych” verbs, e.g.
 
 upset 
 
 bother 
 
himi
(1) Unpleasant stories about offend Johni .
himselfi  

 irritate 
 
please
[§10.4.]
6. The interactions between A constructions and binding discussed in
section 10.5 turn out to be special cases of a more general phenomenon.
Consider the following sentences.

herselfi
(1) a. What Maryi bought from Johnj was [a picture of ].
himselfj
b. The person that Maryi likes the best is herselfi .
c. Herselfi is the one who Maryi should be talking seriously to right now.
A. Since the reflexives in these examples are not locally c-commanded by

their antecedent, they are not accounted for by condition A of the binding
theory. Develop a CS version of condition A that will account for these
cases.
B. Evaluate your answer to A with respect to the standard examples that
constitute the original motivation for condition A. Ideally your account in
part A should correctly predict the ungrammaticality of sentences like those
in (14), for example.
[§10.5.]
7. Consider the following example.

(1) a. Heri father is very supportive of the efforts that Maryi is making.
b. Which efforts that Maryi is making is heri father most supportive of ?
Why is there no crossover violation in (1b)?

[§10.6.]
∗
8. Work out an account of weak crossover in terms of the CS-/GF-
binding analysis of section 10.4.2.
[§10.6.]
∗
9. Work out an account of anti-reconstruction in terms of the CS-/GF-
binding analysis of section 10.4.2.
[§10.6.]
10.1.
10.2. 1, 2, 3, 4, 5 1, 2
10.3. 6 3, 4, 5 1, 2
10.4. 7, 8 6, 7, 8 3, 4, 5
10.5. 6
10.6. 9, 10 7, 8, 9
11
Fragments
In this chapter 1 we look at the syntactic properties of sentential fragments.

A fragment can be used in place of a full sentence when its meaning can be
inferred from the context. Example (1B) is a typical example of a construc-
tion called VP ellipsis.
(1) A: Who will eat that pizza?

B: I will.
The English expression I will in B’s response in (1) is not a full sentence; it
lacks a VP. The interpretation of I will in this case depends on a previous
sentence. B’s response means “I will eat the pizza” because the context is
“Who will eat that pizza?”. If the context was, for example, “No one will
smell quite that bad”, or “Who will pay for the gas?” the response “I will”
would have a very different interpretation.
The general problem is how the fragment receives the interpretation of
a full sentence. The answer must take into account at least the form of the
fragment and the overall linguistic context. The syntactic question is to what
extent the expression containing the fragment acquires its interpretation in
virtue of its syntactic structure and the syntactic structure of the antecedent.
A frequently encountered view in contemporary generative grammar is
that the fragment is the visible portion of a complete sentence, part of which
is invisible. This invisible portion of the sentence also has an interpretation.
We refer to this as the deletion approach to fragments, because of a long
tradition in syntactic analysis in which the invisible material is a deleted or
at least phonetically empty variant of a full overt structure.
In (1), for example, the syntactic structure of B’s response would contain
a phonetically null VP with the interpretation “eat the pizza”. We represent
1
This chapter draws from material from Chapter 7 of Simpler Syntax (Culicover
and Jackendoff 2005).
438 11. FRAGMENTS
this as an empty VP with a structure that is identical to that of the visible

VP eat the pizza, except that the empty phrases have no overt phonetic form.
This is indicated by striking out the relevant words. 2
(2) [S [NP I ][INFL will] [VP [V eat ][NP [DET the][N pizza]]]]
The alternative view is that there is no invisible syntactic structure

when there is a fragment. On such a view, the structure of B’s response is
simply (3).
(3) [S [NP I] [INFL will]]
On either approach to the analysis of fragments there are substantial tech-

nical challenges. For example, if we posit that there are invisible constituents
with full interpretations, we must specify under what syntactic conditions
an invisible constituent may occur, and what determines its interpretation.
On the other hand, if we assume that there is no invisible structure in a
fragment construction, then we have to say what can and cannot be a frag-
ment, and explain exactly how the fragment receives a full interpretation.
In this chapter we look here at some representative fragment constructions,
focusing primarily on English. We consider the following two constructions,
in addition to VP ellipsis.
Bare Argument Ellipsis (BAE)
(4) a. A: Harriet has been drinking something.
B: Yeah, scotch.
b. A: When is Robin coming?
B: On Tuesday.
Gapping
(5) Sam plays saxophone, and Susan sarrusophone.
In BAE, the fragment consists of a single constituent that does not resemble
a sentence. In gapping there appear to be two fragments, each of which
matches up with a constituent of the antecedent. And, as we have seen, in
VP ellipsis, the fragment is a sentence that is missing an overt VP.
The literature on ellipsis and other fragment constructions is vast, and
there are many specific varieties and puzzling facts that we do not have space
2
In order to simplify the presentation we do not consider here a number of
alternatives that have been proposed in the literature on ellipsis. One is that the
VP has no internal structure but is a proform (analogous to a pronoun). Another
is that the VP has a full overt structure and that parts of it are deleted by a formal
operation.
11.1. BARE ARGUMENT ELLIPSIS 439
to discuss here. Our concern will be to describe some basic phenomena and
to elaborate the two general approaches just noted. Our discussion focuses
on the general question of how to account for the fact that a fragment
typically functions as though it is the visible portion of a complete syntactic
structure under identity with some antecedent structure. A number of addi-
tional phenomena and issues are addressed in the Problems and Research
questions at the end of this chapter.
11.1. Bare argument ellipsis
11.1.1. Two approaches
We begin with bare argument ellipsis (BAE) because it is the most basic type
of fragment construction. Yet it illustrates most of the properties of more
complex constructions. Consider the following examples. B’s responses in
(6) are interpreted as though they are abbreviations of the full sentences as
indicated below B’s response. B’s response may be understood as an elab-
oration (as in (6a)), a correction (as in (6b)), or as supplying information
relevant to A’s utterance (as in (6c)).
B: Yeah, scotch.
‘Yeah, Harriet has been drinking scotch.’
b. A: Has Harriet been drinking bourbon again?
B: No, scotch.
‘No, Harriet’s been drinking scotch.’
c. A: What has Harriet been drinking?
B: Scotch.
‘Harriet has been drinking scotch.’
What is responsible for these interpretations? There are a number of

possible solutions, depending on what specific assumptions we make about
the syntactic structure of the fragment. Let us consider our two basic
possibilities, one in which the fragment has a full syntactic structure, and
the other in which it does not.
(7) a. Syntactic account, with interpretation of empty structure:
The responses in (6) have a syntactic structure rather like the full para-
phrases of B’s responses, except that all the parts that correspond to rep-
etitions of A’s sentence are represented as phonetically empty structure. The
mechanism for interpreting syntactic structure supplies the interpretations
440 11. FRAGMENTS
of the empty categories through their correspondence with A’s sentences,

e.g.
Syntax: [[NP Harriet ] [has been [VP [V drinking ] [NP scotch]]]]
Semantics: ‘Harriet’s been drinking scotch.’
b. Semantic/pragmatic account:
The responses in (6) have just the syntactic structure present at the surface.
The mechanism for interpreting syntactic structure supplies the rest of the
details of the interpretation, matching B’s statement with the structure of
A’s sentences, e.g.
Syntax: [NP scotch]
Semantics: ‘Harriet’s been drinking scotch.’
We visualize these two alternatives schematically, in order to see what is

involved in relating the superficial form of the fragment to the correspond-
ing interpretation in each case.
(8) Syntactic account
Antecedent
SYNTAX S
NP VP
Harriet V NP
drink something
CS DRINK(AGENT:HARRIET, PATIENT:Y)
BAE
SYNTAX S
NP VP
Harriet V NP
drink scotch
CS DRINK(AGENT:HARRIET, PATIENT:SCOTCH)
As the diagram shows, in the syntactic account we have to relate the

empty NP in the sentence that contains the fragment to the overt NP
Harriet in the antecedent so that the meaning HARRIET can be included
in the CS representation of the fragment sentence. Similarly for drink.
While this seems straightforward for the simple case of (6), we will show in
section 11.1.2 that the general statement of the relationship can be
complex.
(9) Semantic/pragmatic account

Antecedent
SYNTAX S
NP VP
Harriet V0 NP
drink something
GF Subject Object
CS DRINK(AGENT:HARRIET, PATIENT:Y)
BAE
SYNTAX NP
scotch
CS DRINK(AGENT:HARRIET, PATIENT:SCOTCH)
In the semantic account, the CS representation DRINK(AGENT:

HARRIET, PATIENT:X) has to be constructed on the basis of the con-
text, by matching scotch with something and identifying the part of the
CS that corresponds to something. The meaning of the fragment, in this
case SCOTCH, must then be fitted into the appropriate place in the CS
representation to get DRINK(AGENT:HARRIET,PATIENT:SCOTCH).
442 11. FRAGMENTS
11.1.2. Problems for a syntactic account of Bare

Argument Ellipsis
The syntactic account encounters problems when the fragment has a mean-
ing that does not correspond perfectly to the antecedent. Consider the
following exchange.
(10) A: I hear that Harriet’s been drinking again.
B: Yeah, scotch.
In this case, B’s response “Yeah, scotch” cannot be interpreted by simply
plugging the CS representation SCOTCH into the CS representation of
A’s statement. Doing that would produce a meaning along the lines of “I
(that is, A) hear that Harriet’s been drinking scotch”. But this is not what B
means. What B is saying is “Harriet has been drinking scotch”.
Here is another case that is similar but in the end quite different.
(11) A: Ozzie mistakenly believes that Harriet’s been drinking again.
B: Yeah, scotch.
In this case, what B is saying is “Ozzie mistakenly believes that Harriet has
been drinking scotch”, not “Harriet has been drinking scotch”.
Here is another example, in which B’s response has no coherent interpre-
tation.
(12) A: Ozzie doubts that Harriet has been drinking again.
B: ∗ Yeah, scotch.
A’s statements in (10), (11), and (12) have the same syntactic structure,
so there is no overt syntactic difference that tells us how to interpret
B’s response. What matters is the content of A’s sentence. B’s response is
interpreted with respect to the meaning of what A is saying, not with
respect to the syntactic structure of what A is saying. In other words,
semantic/pragmatic considerations – what “makes sense” – determines the
appropriate interpretation of B’s response.
Let us suppose that BAE has full syntactic structure, and that the invisible
structure is identical to that of the antecedent. Given (10), we would have
to say that the invisible structure containing yeah scotch cannot contain an
invisible copy of I hear (that). . . . Given (11) we would have to say that
the invisible structure must contain an invisible mistakenly believes (that)
and cannot simply be drinking. And given (12) we would have to say the
invisible structure containing yeah scotch cannot contain an invisible copy
of the verb doubts or Harriet has been drinking again.
Thus, the relationship between the BAE sentence and the antecedent is
not simply one of syntactic identity. The semantic/pragmatic conditions
must be invoked in order to explain precisely which of the many possible
identity relationships is the correct one. It is in fact not clear that syntactic
identity is relevant in these cases, since our reconstruction of the meaning of
the fragment must be formulated in terms of the meaning of the antecedent.
Further cases arise in which the form of the antecedent and the form of
the response are syntactically incompatible:
(13) a. What did you do to Susan?
– Kiss her. [cf. ∗ I kissed her to Susan. / ∗ I kissed Susan to her.]
b. What’s that frog doing in my tomato sauce?
– Swimming. [cf. ∗ That frog’s doing swimming in my tomato sauce.]
In these cases, there is no natural syntactic analysis that would derive the
fragment by simply having phonetically empty but meaningful structure
that is licensed on the basis of identity with a grammatical antecedent, as
the paraphrases show.
Semantics is even more deeply involved in instances of BAE where the
syntactic relation between the antecedent and response is less direct.
(14) a. A: Why don’t you fix me a drink?
B: In a minute, OK?
[cf. the infelicity of Why don’t I fix you a drink in a minute as a response:
the response is understood as I’ll fix you a drink in a minute]
b. A: How about fixing me a drink?
B: In a minute, OK?
[the response is understood as I’ll fix you a drink in a minute, OK?]
c. A: Let’s get a pizza.
B: OK – pepperoni?
[the response is not interpreted as OK, let’s get pepperoni pizza?, which is
odd at best; it is understood as something like OK, should we get pepperoni
pizza?]
d. A: Would you like a drink?
B: (i) Yeah, how about scotch?
(ii) No, but how about some lunch?
[cf. ∗ How about I would like a scotch/some lunch? as well as other
improbable variants]
e. A: Harriet’s been drinking again.
B: How stupid! [= ‘How stupid of Harriet to drink again.’]
f. A: I hear there’s been some serious drinking going on around here.
B: i. Not Sam, I hope. [= ‘I hope it’s not SAM who’s been drinking.’]
444 11. FRAGMENTS
ii. Not my favorite bottle of scotch, I hope. [= ‘I hope they haven’t been
drinking my favorite bottle of scotch.’]
g. A: Would you like a cookie?
B: What kind? [= ‘What kind of cookie have you got/are you offering?’; =/
‘What kind of cookie would I like?’]
h. A: Are you hungry?
B: How about a cookie? [What’s the paraphrase?]
i. A: Hey, look! There’s John over there, reading Moby Dick.
B: Are you blind? It’s Sam and Harry Potter.
Exercise 1 asks you to look closely at these examples in order to verify that
the meanings of the fragments are not present in the syntactic structure of
the antecedent.
Let’s push the syntactic approach a bit further. Consider again the
exchange in (6a), which we repeat here.
b. B: Yeah, scotch.
What is visible in the fragment is scotch, and what is invisible is Harriet

has been drinking. A standard assumption on the syntactic approach is that
the invisible material in the fragment must correspond to a single complete
constituent of the antecedent, and not simply to a string of words or to an
incomplete structure. In this case, the invisible material is Harriet has been
drinking, which is not a constituent. One way to have Harriet was drinking
be a constituent is to extract the direct object, e.g. by topicalization.
(15) NPi [S Harriet has been drinking t i ]
If we assume that there is a representation of both A’s statement and B’s

response that has exactly this syntactic structure, then we can see how the
two can correspond.
(16) A: Harriet has been drinking something
Related structure: somethingi [S Harriet has been drinking t i ]
B: Yeah, scotchi [S Harriet has been drinking t i ]
Since the two structures contain identical constituents, namely [S Harriet

has been drinking ti ], we can allow the invisible constituent in B’s response,
producing
(17) B: Yeah, scotchi [S Harriet has been drinking t i ]
This is the general strategy for deriving BAE syntactically. It rests on

the identity of constituents between two sentences, and movement such as
topicalization in order to produce the identical constituents. With this in

mind, we consider an example in which the required topicalization appears
to be very problematic.
(18) A: What kind of scotch does Harriet drink?
B: Expensive. [= ‘Harriet drinks expensive scotch.’]
[cf. ∗ Expensive, Harriet drinks scotch.]
In this case, the identical constituent must have the form

(19) [S Harriet drinks [t i scotch]]
But in order to derive this constituent, we would have to extract the adjec-
tive from the NP, in violation of the left branch constraint (see Chapter 9,
section 9.4).
Here are some additional examples that have similarly impossible extrac-
tions. (Exercise 2 asks you to discuss which constraints are being violated in
these examples.)
(20) a. A: Let’s get a pizza.
B: Pepperoni?
[cf. ∗ Pepperoni, let’s get a pizza; ∗ It is pepperoni that let’s get a
should we
pizza]
b. A: Did Susan say that she saw PAT Smith?
B: No, KIM.
[cf. ∗ Kim, Susan said that she saw [t Smith].]
c. A: Is that a Navy flight suit?
B: No, ARMY.
[cf. ∗ Army, that is a [t flight suit].]
d. A: How many pounds does that pumpkin weigh?
B: Over a thousand.
[cf. ∗ Over a thousand, that pumpkin weighs [t pounds].]
e. A: Is Sviatoslav pro-communist or anti-communist these days?
B: Pro.
[cf. ∗ Pro, Sviatoslav is [t-communist] these days.]
(21) a. A: Harriet drinks scotch that comes from a very special part of Scotland.
B: Where?
[cf. ∗ Where does Harriet drink scotch that is from?]
b. A: John met a guy who speaks a very unusual language.
B: i. Which language?
[cf. ∗ Which language did John meet a guy who speaks t?]
ii. Yes, Albanian.
[cf. ∗ Albanian, John met a guy who speaks t.]
446 11. FRAGMENTS
c. A: John met a woman who speaks French.

B: i. With an English accent?
[cf. ∗ With an English accent, John met a woman who speaks French t?]
ii. And Bengali?
[cf. ∗ And Bengali, did John meet a woman who speaks French t?]
d. A: The administration has issued a statement that it is willing to meet with
one of the student groups.
B: Yeah, right – the Green Rifle Club.
[cf. ∗ The Green Rifle Club, the administration has issued a statement that
it is willing to meet with t.]
e. A: They persuaded Kennedy and some other senator to jointly sponsor
the legislation.
B: Yeah, Hatch.
[cf. ∗ Hatch, they persuaded Kennedy and t to jointly sponsor the
legislation.]
f. A: For John to flirt at the party would be scandalous.
B: Even with his wife?
[cf. ∗ Even with his wife, would for John to flirt t at the party be
scandalous?]
There are two possible conclusions to draw from facts such as these.
One is that extraction constraints apply only to visible structure. While this
approach solves the problem that these examples pose, it does not explain
them. In fact, it raises another puzzle, which is why extraction constraints
apply only to visible structure.
In contrast, we may say that there are no apparent violations of con-
straints on movement in these cases because there is no movement, and,
in fact, no offending structure. But we still have to explain how the full
interpretation of BAE is arrived at. Before we do that, let us consider
some important evidence that shows that the syntax of the antecedent does
nevertheless play a role in BAE and must be taken into account in the
mechanism that interprets BAE.
11.1.3. Reasons to believe syntax is involved in BAE
While there are strong arguments against deriving BAE strictly syntacti-
cally, as we have just seen, there is also a strong argument that shows
that syntax is involved in licensing BAE. The fragment in general has
syntactic features appropriate to its being a part of the antecedent sentence.
For instance, in German, bare arguments such as those in (22) have case
morphology appropriate to the verb in the antecedent sentence.
(22) a. A: Wem folgt Hans?
who.DAT follows Hans
‘Who is Hans following?’
B: Dem Lehrer.
the.DAT teacher
‘The teacher.’
b. A: Wen sucht Hans?
who.ACC seeks Hans?
‘Who is Hans looking for.’
B: Den Lehrer
the.ACC teacher
‘The teacher.’
[Hankamer 1979:394]
The verb folgen “follow” assigns dative case to its object, while suchen
“seek” assigns accusative case. The bare argument in (22a) that corresponds
to the object of folgen must have dative case, while the bare argument in
(22b) that corresponds to the object of suchen must have accusative case.
Merchant 2003 adduces similar examples in Korean, Hebrew, Greek,
Russian, and Urdu. A parallel phenomenon in English appears in (23).
(23) a. A: I hear Harriet has been flirting again.
B: i. Yeah, with Ozzie.
ii. ∗ Yeah, Ozzie.
b. A: John is very proud.
B: Yeah, of/∗ in his stamp collection. [cf. proud of/∗ in NP]
c. A: John has a lot of pride.
B: Yeah, in/∗ of his stamp collection. [cf. pride in/∗ of NP]
The verb flirt requires the “flirtee” to be marked by the preposition with.
Only this requirement can explain the need for the preposition in the frag-
ment. The very close paraphrases (23b,c) push the point home further: they
differ only in that proud requires its complement to use the preposition of,
while pride idiosyncratically requires in. The replies, using BAE, conform to
these syntactic requirements, just as if the sentence was complete.
The conclusion from this and a wide range of similar evidence is that BAE
cannot be based purely on a semantic relation to the CS of the antecedent.
The syntactic properties of the antecedent are relevant, and the bare argu-
ment acts as though it is in a complete syntactic structure. Yet the previous
448 11. FRAGMENTS
section has shown that BAE cannot be based purely on a relation to syntax,
since the interpretation of BAE in the general case involves not a literal
copy of the antecedent but rather a pragmatic discourse relation to the
antecedent.
How are these two conflicting results to be resolved? The general idea is
that the syntax of the fragment guides the match with the antecedent, the
CS of the antecedent provides the missing meaning, and the syntax of the
fragment determines the meaning contribution of the fragment to the total
meaning. We look at this in the next section.
11.1.4. A resolution: indirect licensing
The semantic and syntactic evidence cited in the preceding section shows
that syntactic properties of a fragment must satisfy syntactic conditions
imposed by the antecedent. What this means is that the fragment is not
part of a larger syntactic structure, but it behaves as though it is. And
it is interpreted as though it occupies a particular position in a syntactic
structure, even though it doesn’t. We call this indirect licensing. 3 Here is
how it works in BAE.
Consider how to construct the interpretation of BAE. Again, we begin
with the very simple example (6a).
b. B: Yeah, scotch.
What we have to do is (i) identify the part of the antecedent that scotch
matches, called the target, (ii) find the part of the CS representation of the
antecedent that corresponds to the target, and (iii) construct a representa-
tion for the BAE example that substitutes the interpretation of scotch for
the interpretation of the target in the CS representation of the antecedent.
Here is a sketch of how this procedure would work. We assign the feature
[BEVERAGE] to scotch in the lexicon so that it satisfies the selectional
requirements of drink. We assume that an utterance is not restricted to the
category S but may be any category.
(24) fragment
S YNTAX: [NP scotch]
CS: SCOTCH[BEVERAGE]
3
This term is introduced in Culicover and Jackendoff 2005.
antecedent
S YNTAX: [S Harriet [VP drink [NP something]]
CS: DRINK(AGENT:HARRIET,THEME:[BEVERAGE])
substitution procedure:
1. scotch syntactically matches the direct object something of the antecedent.
2. something corresponds to THEME:[BEVERAGE] in the CS.
3. scotch is a [BEVERAGE] so it matches semantically.
4. Substitute CS of scotch for interpretation of complement of drink in
antecedent, yielding
SCOTCH[BEVERAGE] ≡ THEME:[BEVERAGE]
5. result: DRINK(AGENT:HARRIET,THEME: SCOTCH [BEVERAGE])
The critical part of this substitution rule is step 1, the syntactic match
between the fragment scotch and something in the antecedent. A minimum
requirement is that the target should be of the same syntactic category as
the BA. Moreover, we have already seen that this substitution cannot work
unless syntactic selection is taken into account. That is, in German it would
not be sufficient for the fragment to be an NP; it would have to have the right
case properties as determined by the verb in the antecedent. And in English,
as we saw with the verb flirt, it would not be sufficient for the fragment to be
a PP; it would have to have the form [with NP]. A theory of interpretation
of BAE must make the notion of match explicit.
While syntactic compatibility is necessary for there to be a match, dis-
course conditions also constrain the possible matches. For example, if the
antecedent contains a focus constituent (that is, new or contrastive infor-
mation), then the fragment (typically) has to match the focus and therefore
must share the selected syntactic properties of the focus. One example
of focus is emphatic or contrastive stress. In the following cases, the BA
matches the stressed constituent.
(25) A: Does Ozzie love M ARILYN?
B: No, H ARRIET.
(26) A: Does O ZZIE love Marilyn?
B: No, J OE.
In a wh-question, the wh-phrase is the focus because it asks for the hearer
to supply new information. For example, the answer to a who question must
be an NP that denotes a person. The wh-phrase defines the focus and the
BAE matches it.
(27) A: Who does Ozzie love?
B: Harriet.
450 11. FRAGMENTS
We have considered just the simplest cases of BAE here. A number of more
complex examples are offered at the end of this chapter.
A construction that resembles BAE very closely is sluicing, illustrated
by the examples in (28). In each case there is a fragment in the form of a
wh-phrase. The wh-phrase questions some explicit or implicit part of the
antecedent. For example, who in (28a) asks “who (was Sandy talking to)”,
and similarly for (28b,c).
(28) a. Sandy was talking to someone, but I don’t remember who.
b. Sandy went somewhere, but I forgot where.
c. A: Sandy left early.
B: Why?
B : When?
In examples where the wh-phrase is not selected, as in (28c), sluicing looks

like an interrogative variant of BAE.
(29) A: Sandy left early.
B: Yeah, because it was snowing. [fragment denoting reason]
B : Yeah, right after lunch. [fragment denoting time]
Interestingly, sluicing fragments appear wherever a full wh-question may

appear. For example, remember in (28a) takes an interrogative complement.
Sluicing is thus typically analyzed as a syntactic deletion construction,
because the wh-phrase is assumed (in mainstream approaches) to have
moved to initial position in the clause. But our discussion of BAE shows
that there is an alternative. We have seen that BAE cannot be analyzed
in terms of movement and deletion. It must have a more general solution
in which the fragment is syntactically licensed and interpreted through a
match with the antecedent. Such a solution for BAE generalizes to sluicing,
taking into account that in some cases (e.g. (28a,b)), the syntactic context
requires the fragment to be interrogative.
11.2. VP ellipsis and related constructions
We now turn to a second major class of fragment constructions, those that

stand in freely for canonical VPs. Within this class there are two major types,
VP ellipsis (30a) and do X anaphora (30b).
(30) a. VP ellipsis:
Robin ate a bagel for breakfast, and Leslie did too.
11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS 451
b. Do X anaphora:  

 so 


 that 

Robin ate a bagel (on Thursday), and Leslie did the same thing

 

 something similar 
 
...
(on Friday).
11.2.1. The syntax of VP ellipsis
VP ellipsis in English is characterized by the absence of an overt VP. One

question is under what syntactic conditions a VP may be omitted. It has
been observed that English ellipsis occurs adjacent to an auxiliary verb. (We
indicate ellipsis by showing the omitted VP with single quotes.)
(31) Sandy can’t speak German, but Robin can [‘speak German’].
A VP that is not the complement of an auxiliary cannot undergo ellipsis.

The clearest case of this is the English finite verb phrase, which occurs only
when there is no apparent auxiliary. The following is impossible.
∗
(32) Sandy doesn’t speak German, but Leslie [‘speaks German’].
It is then noteworthy that VP ellipsis is also possible after to, and an

infinitival VP including to cannot undergo ellipsis.
(33) a. Sandy wants to go to Paris, but I don’t want to [‘go to Paris’].
b. ∗ Sandy wants to go to Paris, but I don’t want [‘to go to Paris’].
To generalize this case with (31)–(32), the simplest hypothesis is that to is

a non-finite auxiliary that alternates with finite tense, with the structure in
(34).
(34) S
AUX
to
Since the modals are necessarily finite, this will explain why they cannot
appear in infinitives: ∗ to will go, ∗ to can remember, ∗ to must leave. There
are a number of ways to formulate this treatment of to, depending on
what assumptions we make about the precise syntactic details. This is left
to Exercise 3. Note that the structure in (34) is not compatible with the
452 11. FRAGMENTS
assumption made in Chapter 7 that infinitival to is an inflection of V under

VP, e.g.
(35) [VP [V to go] [PP to Paris]]
Research question 2 focuses on resolving this conflict.

Summarizing to this point, VP ellipsis is the absence of an overt untensed
VP. Because ellipsis involves just the omission of the VP, a tensed sentence
with VP ellipsis has many of the characteristics of a full sentence, i.e. a
subject, tense, and an auxiliary, but no VP. Hence it is unlike BAE, which is
just an orphan constituent.
As indicated earlier, there are two basic ways to analyze VP ellipsis
syntactically. Either the VP is present but invisible, or it is simply not
present. These two alternatives are illustrated in (36), for Robin can [‘speak
German’]. For concreteness we show the CS representation of the modal as
an operator that takes as its argument the entire proposition.
(36) a. Empty VP
SYNTAX S
NP AUX VP
Robin can V NP
speak German
CS ABLE(SPEAK(AGENT:ROBIN, THEME:GERMAN))
b. No VP
SYNTAX S
NP AUX
Robin can
GF Subject
CS ABLE(SPEAK(AGENT:ROBIN, THEME:GERMAN))
Note that we are assuming here that the syntactic rules of English permit
an S that contains a subject and inflected auxiliary, but no VP.
On the syntactic approach to ellipsis the empty VP gets the interpretation
“speak German” because a constituent of the antecedent has this meaning
and matches the empty VP. We have indicated this in the diagram by provid-
ing the empty VP structure, and by assigning to the empty verb the identity
of the verb speak and to the empty NP the identity of the NP German. The
meanings of these constituents are plugged into the interpretation of the
elliptical sentence to give it a complete interpretation.
The key question on this approach is, under what circumstances does a
VP have a particular structure with particular empty constituents? Clearly,
the answer has something to do with a match between the empty VP and a
VP in the antecedent. For instance, if the empty VP is [VP [V eat] [NP sushi]]
and the antecedent contains [VP [V speak] [NP German]], there could be
no match. We know this because the sentence Kim can’t speak German but
Robin can does not mean “Kim can’t speak German but Robin can eat
sushi”. The empty VP must be identical in syntax and meaning with the
antecedent. The technical problem is to say precisely what the antecedent is
and what in the antecedent the empty VP must be identical to.
On the other hand, if there is no empty VP, we need to explain how the
interpretation is constructed. By analogy with BAE, we might envision a
procedure whereby the fragment is matched with a target in the antecedent.
The interpretation of the part of the antecedent that is not part of the match
is what is supplied for the interpretation of the fragment.
At this point it might appear that the two approaches are more or less
equivalent. In both cases we must match the visible part of the sentence to
the antecedent to determine which VP in the antecedent corresponds to the
empty VP. Is the interpretation of the missing VP supplied by finding an
actual VP in the antecedent and using its interpretation, as in (36a), or is it
determined on the basis of the CS of the antecedent, as in (36b)? There are
constructions in English for which the second approach is the only possible
one, as we discuss in the next section. These are VPs like do so, which we call
VP anaphora. As we will see, in general the interpretation cannot be reduced
to the identification of some syntactic constituent of the antecedent – the
interpretation has to be constructed. Given that this approach is indepen-
dently required, there is no reason to assume that VP ellipsis by contrast
involves an empty VP, in the absence of strong independent evidence that
requires that we make this assumption. This situation is parallel to what we
454 11. FRAGMENTS
saw in the case of BAE and sluicing (section 11.1.4), where we saw that the
general matching-and-interpretation solution covers cases where movement
and deletion have been proposed in the mainstream literature.
11.2.2. VP anaphora
VP anaphora is typically of the form do X. We take do so as representative;

other possibilities are do the same, do the same thing, do that, do otherwise,
do something else, do likewise, and do it. Do so, and more generally do X,
requires its antecedent to be an action, as seen in (37), while VP ellipsis
does not, as seen in (38). (We use negative examples in order to distinguish
the do of do so ellipsis from do-support in VP ellipsis.)
(37) Do so anaphora:
a. ∗ Robin dislikes Ozzie, but Leslie doesn’t do so. [State]
b. ?∗ Robin fell out of the window, but Leslie didn’t do so.
[Non-action event]
c. Robin read the newspaper today, but Leslie didn’t do so. [Action]
(38) VP ellipsis:
a. Robin dislikes Ozzie, but Leslie doesn’t. [State]
b. Robin fell out of the window, but Leslie didn’t. [Non-action event]
c. Robin read the newspaper today, but Leslie didn’t. [Action]
Consider (37c). It is logically possible that do so gets its interpretation
by somehow being linked to the VP read the newspaper today. The analysis
would have the following form.
(39) [i]Robin read the newspaper today, but
[ii]Leslie didn’t do so.
1. do so is a pro-VP and needs to get an interpretation.
2. Identify a VP in the antecedent that may supply the interpretation
of do so.
a. read the newspaper today is a VP in the antecedent.
b. Link do so to read the newspaper today by coindexing them.
3. On the basis of 2b, assign the interpretation of read the newspaper
today to do so.
The critical part of this procedure is 2a, where a target VP in the
antecedent is identified. The case against a syntactic analysis of do so
anaphora involves sentences where the meaning of do so does not corre-
spond to any VP in the antecedent. Note first that the elliptical clause may
include one or more fragments. Consider first (40).
(40) a. Robin smokes a pipe after dinner, and Leslie does so during breakfast.
[do so = smokes a pipe]
b. Robin flipped the hamburgers with a spatula, and Leslie did so with a
chef’s knife. [do so = flip the hamburgers]
The fragments do so during breakfast and do so with a chef’s knife show that
do so may have the interpretation of part of the antecedent VP. If we want
to maintain a syntactic account, we will have to say that the structure of the
antecedent VP is more or less the following, with a small VP inside of the
larger VP.
(41) a. [VP [VP smoke a pipe] [PP after dinner]]]
b. [VP [VP flip the hamburgers] [PP with a spatula]]]
However, in the examples in (42) we see that the interpretation

of do so does not need to correspond to a contiguous part of
the antecedent VP. (These sentences are most felicitous if the into-
nation highlights the contrasting constituents, e.g. twelve and eight
in (42a).)
(42) a. Robin slept for twelve hours in the bunkbed, and Leslie did so for eight
hours. [do so = sleep . . . in the bunkbed]
b. Robin cooked Peking duck on Thursday in order to impress Ozzie, and
Leslie did so on Friday.
[do so = cook Peking duck . . . in order to impress Ozzie]
In order for do so in (42a) to mean sleep in the bunkbed, there would have
to be a VP of the form sleep in the bunkbed in the antecedent. But the
antecedent has the VP sleep for twelve hours in the bunkbed. The identity
condition cannot be satisfied unless we assume that constituents are moving
around in the VP and that the interpretation of do so can be established
before they move.
While this explanation cannot be ruled out, it does not appear to be par-
ticularly compelling. Moreover, there are examples that cannot be worked
around in this way. Note that the fragment associated with do so may be
an adjunct, as in all the above examples, but it may not be an argument, as
shown by the following examples.
(43) a. Robin read a book on the train, while Leslie was doing so on the bus.
[on the bus = adjunct]
b. ∗ Robin put a book on the couch, while Leslie did so on the table.
[on the table = argument]
c. ∗ Robin ate a hot dog, while Leslie did so a pickle. [a pickle = argument]
456 11. FRAGMENTS
d. ∗ Robin said that syntax is wonderful, and Leslie did so that phonetics is
even better. [that phonetics is even better = argument]
However, there are two special adjuncts to do X that can be matched with
arguments of the antecedent: to NP can be used to match a Patient, and
with NP can be used to match a Theme.

so
(44) a. Robin broke the window with a hammer and Mary did to the
the same
tabletop.
∗
. . . and Mary broke (the window) to the tabletop with a hammer.

so
b. John turned the hotdog down flat, but he wouldn’t have done with
that
filet mignon.
∗
. . . but he wouldn’t have turned (the hotdog) down flat with filet mignon.
∗
. . . but he wouldn’t have turned (the hotdog) with filet mignon down flat.

do the same
In these examples the anaphoric does not correspond to any VP
do so
do that
of the antecedent.
Furthermore, the phenomenon of “vehicle change” illustrated in (45)
(Fiengo and May 1994) also can be taken as evidence against a purely
syntactic match for do X. The different form of the antecedent does not
appear to inhibit interpretation of the pro-VP, as the following example
shows.
 
∅





 do so 
(45) Robin is eating frogs’ legs, but I never could do it .

 

 do that
 

do the same thing
Based on the form of the antecedent, the literal reconstruction of do X is

eating frogs’ legs, but ∗ I could never eating frogs’ legs is impossible. The same
problem arises even more dramatically when the sentence is . . . , but I never
could, with VP ellipsis instead of do X. In this case, the empty VP following
I never could would have to be identical to the target eating frogs’ legs in the
antecedent.
11.2.3. The interpretation of ellipsis
Using our analysis of BAE as a model, let us work out how to interpret
VP ellipsis and related constructions. As we noted already, we treat the
ellipsis construction as a fragment that must find a match in the antecedent.

Consider again the example in (31), repeated here.
(31) Sandy can’t speak German, but Robin can [‘speak German’].
To find the match, we must determine that Robin matches Sandy and can
matches can’t. The basis for determining such a match is that the syn-
tactic structure of the two sentences matches, in the technical sense that
the structure is the same and the phrases in each position contrast. In
the diagram in (46), we illustrate this matching by linking the matching
phrases.
(46) Antecedent
CS NOT (ABLE (SPEAK (AGENT:SANDY, THEME:GERMAN))
SYNTAX S
NP AUX VP
Sandy can’t V NP
speak German
Fragment
SYNTAX S
NP AUX
Leslie can
We identify the matching constituents and their interpretations as follows.

In the left column we put the constituents of the fragment that need to find
targets, and in the right column we put the corresponding targets.
458 11. FRAGMENTS
(47) Fragment Target

[NP Robin] ↔ [NP Sandy]
ROBIN SANDY
............................................................
[AUX can] ↔ [AUX can’t]
ABLE() NOT(ABLE())
In order to construct an interpretation for the fragment, we take the

interpretation of the antecedent and substitute the interpretation of the
corresponding constituents of the fragment for the interpretation of their
targets. Since Robin corresponds to Sandy, ROBIN is substituted for
SANDY in the CS. Since can corresponds to can’t, ABLE() is substituted
for NOT(ABLE()) in the CS.
(48)
Antecedent: NOT(ABLE(SPEAK(AGENT:SANDY,THEME:GERMAN))

Substitute: ABLE ROBIN
Result: ABLE(SPEAK(AGENT:ROBIN,THEME:GERMAN))
This is a very simple example, of course, but even so it raises impor-

tant questions. Most importantly, on what basis is it determined that two
constituents match? It may be a point-for-point syntactic match, but this
requires a very particular syntactic analysis. For example, we represented
can’t and can as AUX in order to make the match go through, but on many
analyses these are both complex, and may have different syntactic analyses.
In addition to the syntactic complexities, there are other factors that play
a role in determining what matches. Contrastive stress is one.
(49) Sandy will never learn German, but Robin WILL.
Here, Robin and Sandy form a match, and will never and WILL do. The
match might be analyzed as
(50) Fragment Target

[NP Robin] ↔ [NP Sandy]
ROBIN SANDY
..........................................................................
[AUX WILL ] ↔ [AUX will]
EMPH(FUTURE()) FUTURE
..........................................................................
Ø ↔ [ADV never]
Ø NEVER
To construct the interpretation of the fragment, for each matching pair

of constituents, we substitute the interpretation of the constituent in the
fragment for that of the target constituent in the CS of the antecedent.
(51)
Antecedent: FUTURE (NEVER(SPEAK(AGENT:SANDY,THEME:GERMAN))

Substitute: EMPH(FUTURE Ø ROBIN
Result: EMPH(FUTURE(SPEAK(AGENT:ROBIN,THEME:GERMAN))
This approach to interpretation of the fragment generalizes to do X, and

generalizes with the analysis of BAE. Consider the following examples.
(52) a. Robin is eating frogs’ legs, but I would never do it.

b. A: Robin is doing something.
B: Yeah, eating frogs’ legs.
The match in (52a) is as shown in (53). The clause with do it identifies

ROBIN as the Agent of some act.
(53) Pro-VP Target

I ↔ Robin
would never ↔ is
do it ↔ eating frogs’ legs
. . . ACT(AGENT:I) ↔ EAT(AGENT:ROBIN, PAT:FROGLEG)
Result: NEVER(FUTURE(EAT(AGENT:I, PAT:FROGLEG)))
Do it is thus given the interpretation of frogs’ legs. The match in (52b) is as

shown in (54). Here, doing something identifies ROBIN as the Agent.
(54) BAE Target

eating frogs’ legs ↔
doing something
Robin
is
EAT(AGENT:X, PAT:FROGLEG) ↔ ACT(AGENT:ROBIN)
Result: EAT(AGENT:ROBIN, PAT:FROGLEG)
In this case, the interpretation of eating frogs’ legs is substituted into the
interpretation of Robin is doing something.
In the Problems a number of more complex examples are given in order
to bring out additional aspects of this interpretation scheme.
460 11. FRAGMENTS
11.3. Gapping
11.3.1. Basic gapping facts
In gapping, everything in a second conjunct is absent except for (usu-

ally) two constituents with contrastive focus intonation (indicated by small
caps). 4
(55) a. ROBIN speaks F RENCH, and L ESLIE, G ERMAN.
b. ROBIN speaks F RENCH to Bill on Sundays, and L ESLIE , G ERMAN.
c. ROBIN wants to speak F RENCH, and L ESLIE, G ERMAN.
d. ROBIN has been speaking F RENCH , and L ESLIE , G ERMAN.
Notice that all of the material in the second conjunct is absent, including
the verb or verbs, and the tense. We show this by putting the “gapped” (that
is, missing) material in quotes. Moreover, what is gapped is not necessarily
a constituent.
(56) a. ROBIN speaks F RENCH, and L ESLIE [‘speaks’] G ERMAN.
b. ROBIN speaks F RENCH to Bill on Sundays, and L ESLIE [‘speaks’]
G ERMAN [‘to Bill on Sundays’].
c. ROBIN wants to speak F RENCH, and L ESLIE [‘wants to speak’]
G ERMAN.
d. ROBIN has been speaking F RENCH , and L ESLIE [‘has been speaking’]
G ERMAN.
In all these cases, the two phrases in the second conjunct are understood as
subject and object. But in general this need not be the case.
(57) a. On S UNDAYS, Robin speaks F RENCH, and on T UESDAYS, [‘Robin
speaks’] G ERMAN.
b. On S UNDAYS, Robin sleeps in the AFTERNOON, and on T UESDAYS,
[‘Robin sleeps’] until DINNERTIME.
Intuitively, the function of gapping is to contrast pairs. For example, in

(55) what is contrasted are the pairs <Speaker, Language>. Gapping allows
4
It is marginally possible to have three focus constituents, but as the number of
foci goes up the acceptability goes down, e.g.
(i) a. Robin speaks French on Tuesdays, and Leslie, German on Thursdays.
b. ∗ With Yves, Robin speaks French on Tuesdays, and with Otto, Leslie,
German on Thursdays.
We assume that the difficulty is due to processing the various constituents in the
absence of an overt verb, and is not a deep syntactic fact.
11.3. GAPPING 461
for as many conjuncts as we want, each of which contains the values for the
variables being contrasted.
(58) ROBIN speaks F RENCH , LESLIE [‘speaks’] G ERMAN, O TTO [‘speaks’]
C HUKCHI , and S VIATOSLAV [‘speaks’] O SSETIC.
However, the verb must appear in the first conjunct.

∗
(59) ROBIN [‘speaks’] F RENCH , LESLIE speaks G ERMAN, O TTO [‘speaks’]
∗
ROBIN [‘speaks’] F RENCH , LESLIE [‘speaks’] G ERMAN, O TTO [‘speaks’]
Following our approach to BAE, we may take each constituent of a

gapped conjunct to be a fragment, to be interpreted through a match with
an antecedent. The following shows how this would go for the simple case
of (56a).
(60) fragments
S YNTAX: [NP Leslie] [NP German]
CS: LESLIE[HUMAN] GERMAN[LANGUAGE]
antecedent
S YNTAX: [S Robin [VP speaks [NP French]]
CS: SPEAK(AGENT:ROBIN[HUMAN],THEME:
FRENCH[LANGUAGE])
substitution rule:
1a. Leslie matches the subject Robin of the antecedent in syntactic
properties.
1b. German matches the direct object French of the antecedent in
syntactic properties.
Therefore,
2a. Leslie corresponds to AGENT:[HUMAN] in the CS.
2b. German corresponds to THEME:[LANGUAGE] in the CS.
3a. LESLIE is [HUMAN].
3b. GERMAN is [LANGUAGE].
Therefore,
4a. Substitute CS of Leslie for interpretation of subject of speak in
antecedent.
4b. Substitute CS of German for interpretation of complement of speak
in antecedent.
LESLIE[HUMAN] ≡ AGENT:ROBIN[HUMAN]
GERMAN[LANGUAGE] ≡ THEME:FRENCH[LANGUAGE]
5. result:
SPEAK(AGENT:LESLIE[HUMAN],THEME:
GERMAN[LANGUAGE])
462 11. FRAGMENTS
As in the case of BAE, it is necessary to state precisely under what

circumstances there is a match. The matching constituents in the gapped
conjunct must have all of the syntactic properties that they would have if
they were actually substituted into the antecedent. For example, syntactic
category, case, and preposition selection must be the same.

she, him
(61) a. John loves Mary, and ∗ .
her, he

with
b. Ozzie was flirting with Harriet, and Sandy ∗ Leslie.
from
Moreover, the relative order of the constituents must be the same as it would
be if they were substituted into the antecedent. So, for example, (62) does
not have the interpretation that Leslie loves Robin.
(62) Sandy loves Kim, and Robin, Leslie.
In order for Leslie to be understood as substituting for Sandy, it would have

to precede Robin.
Data of the sort just noted is of course consistent with the view that gap-
ping actually involves deletion of phonological material from a syntactically
complete structure. If the second conjunct of (62) was actually Robin loves
Leslie and we simply deleted loves, the order Robin, Leslie would follow
immediately. The next section discusses the difficulties in maintaining such
an intuitively appealing syntactic approach to gapping.
11.3.2. Why gapping cannot be syntactic deletion
Just as in the case of VP ellipsis, a syntactic approach to gapping is to try

to analyze it in terms of an empty constituent, in fact, to try to reduce it
to VP ellipsis. On such an approach, both of the overt fragments would
have been moved out of the VP, and the VP would then be deleted (or
the empty structure would be licensed) on the basis of identity with an
identical antecedent VP. In (63) we show a derivational analysis for the
simple example (62).
(63) Sandy loves Kim, and Robin loves Leslie ⇒ [movement of NPs]
Sandyi Kimj [t i loves t j ] and Robink Lesliem [t k loves t m ] ⇒ [deletion of
identical structure]
Sandyi Kimj [t i loves t j ] and Robink Lesliem t k loves t m ⇒ [movement of V]
Sandyi lovesn Kimj [t i t n t j ] and Robink Lesliem t k loves t m
11.3. GAPPING 463
The movements would of course have to be independently motivated in

order for the analysis to be most convincing.
On this analysis, every constituent of the original structure of Sandy loves
Kim has been moved. While a single topicalization is certainly possible in
English, multiple topicalization and movement of the verb to the left is
problematic, especially if the primary (or only) motivation is to derive the
gapping construction. Nonetheless, an approach along these lines has some
appeal because, at first glance, it appears to be possible to delete a non-
constituent – and a discontinuous non-constituent at that (e.g. speaks . . . to
Bill on Sundays in (56b)).
(56) b. ROBIN speaks F RENCH to Bill on Sundays, and L ESLIE [‘speaks’]
G ERMAN [‘to Bill on Sundays’].
This in turn provides motivation for a movement analysis that consolidates

the deleted material into a single constituent (albeit one containing traces).
Suppose, by way of further illustration, that we want to derive (56b).
We topicalize F RENCH and G ERMAN in the two conjuncts, producing the
intermediate structure (64).
(64) F RENCHi , ROBIN speaks t i to Bill on Sundays, and G ERMANj L ESLIE
speaks t j to Bill on Sundays.
Now, let’s topicalize Robin and Leslie. (Note that these structures do not
actually correspond to grammatical sentences of English.)
(65) ROBINk F RENCHi , t k speaks t i to Bill on Sundays, and L ESLIEm G ERMANj
t m speaks t j to Bill on Sundays.
This multiple topicalization produces an identical constituent of the form

(66) t speaks t to Bill on Sundays
in the two conjuncts, if we ignore the difference in indices. If we assume that

the derivation of these sentences involves this multiple topicalization, then
we can carry out the deletion in the second conjunct at the point at which
the identity appears.
(67) ROBINk F RENCHi , t k speaks t i to Bill on Sundays, and Lesliem G ERMANj
t m speaks t –j to Bill on Sundays.
Thus, we are able to derive Leslie German through the deletion of a con-
stituent.
The problem that this derivation raises, of course, is that the superficial
order in the main conjunct in (56b) is not Robin French. Moreover, English
464 11. FRAGMENTS
does not appear to allow multiple topicalization of arguments, as shown

by the fact that sentences such as (67) are not grammatical. In order to
make this derivation go through, it must therefore be stipulated that the
topicalization is only an intermediate stage of the derivation, and that at
some point speaks ends up moving further to the left, to produce (68) (or
Robin and Leslie move back into their original positions).
(68) ROBINk speaksn F RENCHi , t k t n t i to Bill on Sundays, and L ESLIEm
G ERMANj t–m speaks t j to Bill on Sundays.
Under what circumstances such a derivation is technically feasible is a

complex question that we will not go into here. It is important to recognize
that multiple movements along these lines can be made to produce surface
orders with very different structures than what we might first hypothesize. A
reasonable goal would be to see if it is possible to explain the facts without
resorting to such devices, as we suggest in section 11.3.1.
It is important to note that there are cases in which the putative move-
ments needed to account for gapping will violate movement constraints.
This is similar to the situation that we encountered in the case of BAE.
Consider the following examples.
(69) a. ROBIN thinks that the New York T IMES will endorse George W. Bush,
and L ESLIE, the Washington P OST.
b. ROBIN is reading a book written by J OHN U PDIKE, and L ESLIE, A NN
T YLER.
c. ROBIN knows a lot of reasons why DOGS are good pets, and L ESLIE,
CATS .
Example (69) shows that gapping can involve constituents of adjacent

clauses. If there was movement in (69a), it would be illegitimate, because
the Washington Post is the subject of a that-clause and movement would
violate the that-t condition.
∗
(70) . . . and the Washington Post, Lesliei , thinks that t j would endorse George
W. Bush.
Similarly, movement in (69b,c) would have to involve violation of con-

straints. (Exercise 4 asks you to explain what movement constraints would
be violated in these examples.)
One can of course stipulate that such impossible movements are allowed
only if they are subsequently eliminated by deletion (perhaps along the
lines of Merchant 2001), or that they occur only at LF. But such a move
EXERCISES 465
is less preferred as an explanation of these examples if there is a suitable

interpretive procedure that captures the facts directly.
11.4. Summary
This chapter has been concerned with how fragments get their interpre-
tation. A widespread approach to this problem has been to assume that
fragments are the visible portions of complete structures. The evidence that
we have given suggests that this cannot be correct in every case, although
there are some cases for which it does work. Our conclusion is that there
must be an interpretive mechanism that reconstructs the interpretation of a
fragment based on a match between the fragment and an antecedent, and
the interpretations of the fragment and the antecedent.
We showed that this interpretation mechanism must make use of the
syntactic form of the fragment in order to determine what it matches
in the antecedent. The basic constraint is that a fragment cannot match
a given constituent in the antecedent if it does not possess the syntac-
tic properties that would allow it to function like its target does in the
antecedent.
Exercises
1. The examples given as (14) in the text are intended to show that the
interpretation of a fragment in BAE must involve semantic and pragmatic
information, and cannot be explained simply in terms of the syntactic
structure of the antecedent. Discuss each of the examples d–i and show how
it makes this point.
[§11.1.2.]
2. The examples given as (21) in the text are intended to show that a
syntactic account of BAE requires that violations of syntactic constraints
must be suspended if the offending structure is invisible. Discuss each of the
examples and show how it makes this point.
[§11.1.2.]
3. Discuss how the following facts constitute evidence that to is an
untensed auxiliary.
466 11. FRAGMENTS

will
∗
(1) a. I expect Robin to can leave.
must

have called
b. I expect Robin to .
be sleeping
Why is the following example a problem for this view?
(2) I expect Robin not to call.

[§11.2.1.]
4. Show how a movement derivation of the examples in (11.4) violates

the movement constraints. For each example, work out the derivation, say
which constraint is violated and show how.
[§11.3.2.]
Problems
∗
1. Consider the following exchange.
(1) A: Harriet’s been drinking again.

B: Yeah, scotch.
Here, the fragment extends the meaning of the antecedent. It is interpreted

as though it was added into the antecedent in the appropriate syntactic
position, and licensed as though it was in that position. As a result, the
fragment acquires all the semantic features and semantic relations that
follow from its role in the interpretation. Discuss the specific information
that must be extracted from the antecedent and the lexical representation of
drink in order to properly carry out this interpretation.
[§11.1.4.]
2. Consider the following exchange.
(1) A: Who wants to go to the movies?

B: Me!
Note that me has the case form of a direct object, although it is interpreted
as though it is the subject of want(s) to go to the movies.

I
(2) ∗ want to go to the movies.
Me
And the form I is very unnatural at best as a bare argument.

PROBLEMS 467
(3) A: Who wants to go to the movies?

B: ∗ ?I!
Discuss these facts from the perspective of the proposal that BAE is the
visible part of a complete syntactic structure. What would have to be added
to the analysis to account for the facts noted here?
[§11.1.4.]
∗
3. The following examples are instances of BAE that are more complex
than those discussed in the text. For each one, discuss what kinds of infor-
mation has to be retrieved in order to properly interpret the bare argument,
and where this information comes from (e.g. the antecedent, the lexicon,
common sense, world knowledge, etc.).
(1) a. A: Ozzie was flirting.

B: Yeah, with Harriet.
B: ∗ Yeah, Harriet.
b. A: Has Harriet been drinking scotch again?
B: No, bourbon.
c. A: I hear Harriet’s been drinking again.
B: Yeah, scotch.
d. A: I hear Harriet’s been drinking again.
B: Yeah, every night in her room.
[§11.1.4.]
4. The following dialogue poses a problem for the syntactic analysis of
VP ellipsis. Explain what the problem is.
(1) A: Who ate the spaghetti and who drank the wine?
B: I did!
[§11.2.1.]
∗
5. Following the model of (49)–(51), state explicitly what the corre-
spondences are for the following cases of VP ellipsis and show how the
interpretation of the ellipsis sentence is constructed. Some of these examples
constitute very difficult puzzles, and in those cases indicate what the puzzles
are.
(1) a. People who eat meat shouldn’t.

b. Not many people bought anything, but Kim did.
c. Kim talks to the same people that Robin does.
d. John likes to listen to stories about himself, but Mary doesn’t.
468 11. FRAGMENTS
e. Robin gave out the answers without intending to.

f. Robin runs faster than Kim does.
[§11.2.3.]
6. State the syntactic conditions on gapping suggested by the following
sentences.
(1) a. Robin speaks French, and Leslie [‘speaks’] German.
b. Robin wants to speak French, and Leslie [‘wants to speak’] German.
c. Robin speaks French, but Leslie [‘speaks’] only German.
d. Robin speaks French, and not Leslie [‘speaks’] German.
∗
e. Robin speaks French, whenever Leslie [‘speaks’] German.
∗
f. Robin [‘speaks’] French, and Leslie speaks German.
∗
g. Whenever Robin speaks French, Leslie [‘speaks’] German.
[§11.3.]
Research questions
1. In many respects VP ellipsis is similar to pronominal coreference

(Chapter 10). For example, an elliptical VP may appear in a subordinate
clause with a full VP antecedent to the left or to the right of it. Similarly, a
pronoun in a subordinate clause may have an antecedent to the left or right
of it.
(1) a. Sandy likes to eat raw oysters, although I don’t [VP like to eat raw oysters].
b. Although I don’t [VP like to eat raw oysters], Sandy likes to eat raw oysters.
(2) a. Johni likes to eat raw oysters, although hei is usually very cautious.
b. Although hei is usually very cautious, Johni likes to eat raw oysters.
Explore the extent of this similarity by determining whether VP ellipsis

is subject to condition B of the binding theory. The challenge will be to
construct examples in which the elliptical VP c-commands the overt VP.
[§11.2]
2. In the text we noted that when the verb is infinitival, to behaves as
though it is a tenseless AUX with respect to ellipsis.
(1) a. . . . but I won’t.
b. . . . but I didn’t want to.
This suggests that the infinitive is an S that lacks a subject. On the other
hand, in Chapter 7 we treated to as an inflection on the verb, and the
infinitival phrase itself as a VP, not an S. Work out an analysis of infinitival
complements that resolves this apparent conflict. (Hint: Does the category
of an infinitival complement have to be VP?)
[§11.2]
3. As noted in the text, sluicing is a construction in which a wh-question
is expressed by a wh-phrase fragment.
(1) Sandy was talking to someone, but I don’t remember who.
This construction is usually derived in MGG by deleting the absent material

under identity with the antecedent after movement. We show the trace of
movement as a copy.
(2) Sandy was talking to someone, but I don’t remember who [Sandy was talking
to who].
One motivation for such an analysis is that it provides a uniform account

of the contexts that license sluicing based on the contexts that license wh-
questions. So, in a context where a wh-question is impossible, sluicing will
also be impossible. For example, while who Sandy was talking to is a possible
argument of talk to in (3a), it is not a question. And we do not get sluicing,
as shown in (3b).
(3) a. Sandy was talking to someone, but I would never talk to who [Sandy was
talking to who].
‘Sandy was talking to someone, but I would never talk to who Sandy was
talking to.’
b. ∗ Sandy was talking to someone, but I would never talk to who [Sandy was
talking to who].
‘∗ Sandy was talking to someone, but I would never talk to who.’
A. Here are some further examples of the same construction.

(4) a. Sandy was angry, but I don’t know why [Sandy was angry why]
b. Sandy left the room, but I don’t know exactly when [Sandy left the room
when]
(5) a. Sandy was talking to someone, but I don’t remember to who [Sandy was
talking to who].
b. Sandy was talking to someone, but I don’t remember who [Sandy was talk-
ing to who].
(6) a. Sandy was talking, but I don’t remember to who [Sandy was talking to
who].
b. ∗ Sandy was talking, but I don’t remember who [Sandy was talking to who].
Formulate a definition of “identity” in syntactic terms that will license these

and similar cases of sluicing.
470 11. FRAGMENTS
B. The syntactic account of sluicing is relatively straightforward for the

types of cases given in Part A. Data such as the following raise some
complications.
(7) a. ∗ Sandy was talking to Leslie, but I don’t remember who.
b. Sandy was talking to Leslie, but I didn’t remember who until later.
(8) a. Sandy likes to drink beer, but I don’t know what brand.
b. Sandy likes to drink cold beer, but I don’t know exactly how cold.
(9) One of my friends lost some money playing poker last weekend, but I can’t tell
you who or how much.
Explain why these cases are problems for the simple syntactic account of
sluicing based on the data in Part A.
C. Formulate the syntax/CS correspondence for wh-questions in such a
way that it licenses sluicing as well as full wh-questions. Then show how to
extend the interpretive account of BAE given in section 11.1.4 to sluicing.
[§11.3.]
4. The following examples illustrate ones anaphora.

(1) a. Sandy drives a red car, and Leslie drives a green one.
b. Sandy drives a red car. Mine is green.
Show that it is possible to formulate interpretation mechanisms for these

cases that parallels that for do X and VP ellipsis, respectively. State the
interpretation mechanisms explicitly.
[§11.3]
11.1. 1, 2 1, 2, 3
11.2. 3 4, 5 1, 2
11.3. 4 6 3, 4
Glossary
A position: an argument position in a syntactic structure. (See also A

position.)
A chain: a syntactic dependency between two parts of a sentence, in
which one part (the head of the chain) is in a non-argument (A ) position.
(Sometimes called an A dependency).
A position: a non-argument position in a syntactic structure.
adjunct: expresses a refinement of the meaning of a phrase, not an essen-
tial component of the meaning. (See also argument.)
agreement: the situation in which the form of one word or phrase reflects
syntactic or semantic properties of another word or phrase.
allomorphy: a situation in which a morpheme takes various phonetic
forms.
antecedent (of a predicate): the phrase that is either the subject of a
predicate, or interpreted as though it is the subject of a predicate.
antecedent (of a referential dependency): the phrase that determines the
reference of another phrase, such as a pronoun.
antipassive: a construction in which the thematic role that would cor-
respond to the direct object in the default case is either suppressed or
expressed as an oblique object.
applicative: a construction in which the thematic role that would corre-
spond to an oblique object in the default case is expressed as the direct
object.
argument: a phrase in the sentence that refers to anything that is neces-
sarily involved in the relation expressed by the verb.
attribute value matrix (AVM): a display that shows each feature of a word
and its corresponding value.
binding: a situation where the interpretation of an expression depends on
the interpretation of its antecedent in virtue of the syntactic relationship
between them.
canonical structure: the structure that is most typical in a given language.
472 GLOSSARY
case licensing: the situation where a particular morphological case is

compatible with a given syntactic configuration.
c(ategorial)-selection: the specification of the categories of the syntactic
arguments of a head.
causative: a construction that has the interpretation that an Agent causes
a Patient to do something.
clitic: a word that cannot stand alone but must be attached to another
word.
coercion: an interpretation that is added to the normal interpretation
of a word as a consequence of the syntactic configuration in which it
appears.
competence: the language user’s knowledge of his or her language. (See
also performance.)
complement: a non-subject argument.
construction: a syntactically complex expression whose meaning is not
entirely predictable from the meanings of its parts and the way that they are
combined in the structure.
control: the dependency relation between an overt argument of a verb and
the missing subject argument of a non-finite complement.
coreference: the relation in which two NPs refer to the same thing or
things.
correspondence rules: explicit statements that relate the sound, syntactic
structure, and meaning of words or phrases.
CS-chain: a syntactic chain that is mediated by a binding relation in
conceptual structure.
derivational morphology: morphological structure that systematically
relates words of different categories.
detransitivization: suppression of the direct object of a verb. (See also
antipassive.)
displacement: a situation in which a part of a sentence is not in its canon-
ical position.
dummy or expletive subject: a subject of a sentence that has only a gram-
matical function but no meaning.
ellipsis: a construction in which part of a sentence is omitted, and its
meaning can be determined from context.
endocentric (phrase): a phrase whose category is determined by its head.
(See also exocentric.)
GLOSSARY 473
exceptional case marking (ECM): the construction in which the NP that

is understood as subject of an infinitival has the grammatical properties of
the object of the higher verb.
exocentric (phrase): a phrase whose category is not determined by its
head. (See also endocentric.)
external Ë-role: the thematic role that is by default assigned to the subject
of a predicate.
extraposition or it-extraposition: a construction in which the logical sub-
ject of a sentence is adjoined to the end of the VP, and the grammatical
subject is a dummy.
feature composition: a process by which the features of the parts of a word
or phrase become the features of the entire word or phrase.
fragment: an expression that is not a complete sentence, but can be given
a complete sentence interpretation in context.
GF: a grammatical function, such as Subject and Object.
grammar: a description of a language that specifies the knowledge that
speakers of a language have about what constitutes an expression in the
language.
grammatical: in conformity with the rules of the grammar of a
language.
head (of a chain): a phrase that is linked to a gap to form a chain.
head (of a phrase): the word that determines the category of a phrase. (See
also endocentric.)
implicit argument: the CS argument of a word that does not correspond
to an actual linguistic expression.
inflectional morphology: morphological structure that determines a
word’s grammatical properties.
indirect licensing: a situation in which the form of a fragment is deter-
mined by matching the fragment with a constituent in a complete sentence
in the context.
intransitive (sentence): a sentence that lacks a direct object.
island: a syntactic configuration that does not allow a gap within a phrase
to be linked to a constituent outside of the phrase.
lexicon: the totality of unpredictable knowledge about the form, struc-
ture, and meanings of words and expressions in a language.
logical object: the CS argument that by default corresponds to a syntactic
direct object.
474 GLOSSARY
logical subject: the CS argument that by default corresponds to a syntactic

subject.
long distance anaphor: a pronoun that must have an antecedent in the
sentence, where the antecedent may not or need not be in the same clause.
maximal projection: a projection of a head that is not a constituent of a
larger phrase of the same category.
monostratal (or non-derivational) theory: a syntactic theory that assumes
only one level of syntactic representation. Contrasts with derivational (or
transformational) theories. (See also transformation.)
morphosyntax: the relationship between the form of a word and its func-
tion and distribution in a phrase or sentence.
negative polarity item: a word or phrase that can be used only if it is in a
particular relation to another word or phrase that signifies negation.
paradigm: the set of inflectional forms that a word may take in different
syntactic contexts.
passive: a construction in which the thematic role that would correspond
to the subject in the default case is either suppressed, or expressed as an
oblique object.
performance: the processing strategies by which knowledge of language is
put to use by a speaker.
phrase structure rule (PSR): a statement that describes the possible form
of a phrase of a particular category.
piedpiping: a construction in which a phrase that contains a wh-phrase is
displaced as though it is a wh-phrase itself.
p(reposition)-stranding: a construction in which the complement of a
preposition is displaced, leaving a gap in the place of the complement of
the preposition.
pro (“little pro”): a hypothesized invisible pronoun.
PRO: a hypothesized invisible subject of a non-finite clause.
pro-drop: languages that have pro subjects are called pro-drop languages.
proform: a generic element, such as a pronoun, that can be used to express
the same meaning as an arbitrarily complex expression of the same category.
projection: a phrase that contains a head of the same category.
quirky case: case marking that does not fit the canonical pattern in a
language.
raising: a construction in which an argument of a verb (subject or object)
is interpreted as though it is actually the subject of the complement VP of
that verb.
GLOSSARY 475
recursion: a phrase of a certain category may contain a phrase of the same

category.
referring expression or R-expression: an NP that does not depend for its
reference on another NP.
scrambling: the phenomenon by which arguments and adjuncts of a sen-
tence may appear in more than one linear order. Sometimes called free word
order.
s(emantic)-selection: the specification of the CS arguments associated
with a head.
semantic anomaly: ill-formedness that is due to a violation of a semantic
restriction.
structure preserving: a transformation is structure preserving when the
result of its application is a structure that is independently described by the
phrase structure rules of the grammar.
subcategorization: c(ategorial)-selection.
thematic roles (or Ë-roles): the distinguishing property of a CS argument,
such as Agent or Patient.
thematic structure: the set of thematic roles associated with an expression.
topicalization: a construction in which a displaced non-interrogative con-
stituent appears in initial position in a clause.
transformation: a formal operation (or rule) that systematically relates
two syntactic trees as part of the description of the syntactic structure of
a sentence or phrase.
transitive (sentence): a sentence that contains a direct object.
ungrammaticality: a situation in which the arrangement of the words and
phrases of a sentence in terms of their categories and/or morphological
form is incompatible with one or more rules of the grammar.
Universal Grammar (UG): the capacity or capacities that humans have
that is specific to the task of acquiring a language.
wh-in-situ: a non-displaced wh-phrase.
References
A, S (1987). The Noun Phrase in its Sentential Aspect. Cambridge, MA:
MIT Press.
A, A, and C W (eds.) (1998). Possessors, Predicates
and Movement in the Determiner Phrase. Amsterdam, Philadelphia: John
Benjamins.
and E A (1999). “EPP without Spec,IP”. Specifiers:
Minimalist Approaches, ed. by David Adger, Susan Pintzuk, Bernadette Plun-
kette, and George Tsoulas, 93–109. Oxford: Oxford University Press.
A, J S. (1976). “The Complex NP-Constraint as a Non-universal Rule”.
University of Massachusetts Occasional Reports, University of Massachusetts,
Dept of Linguistics.
A, I. W, and J. K (2005). “Passive Without Passive Morphology? Evi-
dence from Manggarai”. The Many Faces of Austronesian Voice Systems: Some
New Empirical Studies, ed. by I. W. Arka and M. D. Ross, 87–117. Canberra:
Pacific Linguistics.
A, A (2005). “Control and Semantic Resource Sensitivity”. Journal of
Linguistics 41:465–511.
B, M (1988). Incorporation: A Theory of Grammatical Function Changing.
Chicago: University of Chicago Press.
B, S (2002). “Remnant Stranding and the Theory of Movement”.
Dimensions of Movement, ed. by Artemis Alexiadou, Elena Anagnostopoulou,
Sjef Barbiers, and Hans-Martin Gärtner, 47–67. Dordrecht: John Benjamins.
BD, J (1997). “Oblique Subjects in Old Scandinavia”. Working Papers
in Scandinavian Syntax 60:25–50.
B, J (1984). “Towards an Explanation of Certain That-t Phenomena: the
COMP Node in Bavarian”. Sentential Complementation, ed. by W. de Geest and
Y. Putseys, 23–32. Dordrecht: Foris Publications.
(2002). “Decomposing the Left Periphery: Dialectal and Cross-linguistic
Evidence”. Proceedings of the Israeli Association of Theoretical Linguistics
(IATL) 18.
B, B J. (1994). Case. Cambridge: Cambridge University Press.
B, E (1997). “On Certain Violations of the Superiority Condition,
AgrO, and the Economy of Derivation”. Journal of Linguistics 33:227–54.
C, A, and U S (2004). “Clitic Positions and Restructur-
ing in Italian”. Linguistic Inquiry 35(4):519–57.
478 REFERENCES
C, L L-S, and J R (2000). “Licensing Wh-in-situ”. Syn-
tax 3(1):1–19.
C, N (1957). Syntactic Structures. The Hague: Mouton.
(1970). “Remarks on Nominalization”. Readings in English Transformational
Grammar, ed. by Roderick Jacobs and Peter Rosenbaum, 184–221. Waltham,
MA: Ginn and Co.
(1973). “Conditions on Transformations”. Festschrift for Morris Halle, ed. by
Steven Anderson and Paul Kiparsky, 232–86. New York: Holt, Rinehart and
Winston.
(1977). “On Wh-Movement”. Formal Syntax, ed. by Peter W. Culicover,
Thomas Wasow, and Adrian Akmajian, 71–132. New York: Academic Press.
(1981). Lectures on Government and Binding. Dordrecht: Foris Publications.
(1986a). Barriers. Cambridge, MA: MIT Press.
(1986b). Knowledge of Language: Its Nature, Origin, and Use. New York:
Praeger.
(1995). The Minimalist Program. Cambridge, MA: MIT Press.
(2000). “Minimalist Inquiries: the Framework”. Step by Step: Essays on
Minimalist Syntax in Honor of Howard Lasnik, ed. by Roger Martin, David
Michaels, and Juan Uriagereka, 89–155. Cambridge, MA: MIT Press.
C, C, and J-B K (2003). “Differences between Externally and
Internally Headed Relative Clause Constructions”. Proceedings of the 9th
International Conference on HPSG, ed. by Jong-Bok Kim and Stephen Wechsler,
43–64. Stanford University.
C, G (1990). Types of A-Bar Dependencies. Cambridge, MA: MIT
Press.
C, P, and G H (1998). “Long Distance Reflexives in
Singapore Malay: An Apparent Typological Anomaly”. Linguistic Typology
2.
and C.-T. J H (2001). Long Distance Reflexives: The State of
the Art. New York: Academic Press.
C, W (2003). Typology and Universals, Second edition. Cambridge:
Cambridge University Press.
C, P W. (1993). “Evidence against ECP Accounts of the That-t
Effect”. Linguistic Inquiry 24:557–61.
and R J (1995). “Something Else for the Binding Theory”.
Linguistic Inquiry 26:249–75.
(2005). Simpler Syntax. Oxford: Oxford University Press.
D, W D. (1986). Choctaw Verb Agreement and Universal Grammar.
Boston: D. Reidel Publishing Company.
D, M (1999). A Grammar of Tukang Besi (Mouton Grammar Library,
20). Berlin: Mouton de Gruyter.
REFERENCES 479
(2002). “The Ingredients of Grammatical Functions”. http://courses.nus.edu.

sg/course/ellmd/TkBpivotsubj.pdf.
E, J (1970). Root and Structure Preserving Transformations. Bloom-
ington, IN: Indiana University Linguistics Club.
E, T (2002). The Syntax of Adjuncts. Cambridge: Cambridge University
Press.
E, N (1973). On the Nature of Island Constraints. Unpublished doctoral
dissertation. MIT.
F, R, and R M (1994). Indices and Identity. Cambridge, MA:
MIT Press.
F, J (2000). “Preposition Stranding in German Dialects”. Syntactic
Microvariation, ed. by Sjef Barbiers, Leonie Cornips, and Susanne van der Kleij.
Amsterdam: Meertens Institute.
F, B (1991). “Ergativity, Focus and Verb Morphology in Several
South Sulawesi Languages, Western Austronesian and Contact Languages”.
Papers from the Fifth International Conference on Austronesian Linguistics, ed.
by R. Harlow. Auckland, New Zealand: Linguistic Society of New Zealand.
G, A. (1984). “Toward a Theory of Long Distance Anaphora: A GB
Approach”. The Linguistic Review 3:307–59.
G, T (2001). Syntax: An Introduction. Amsterdam, Philadelphia: John
Benjamins.
G, J B (1958). “The Child’s Learning of English Morphology”.
Word 14:150–77.
G, A E, and R J (2004). “The English Resultative as a
Family of Constructions”. Language 80:532–68.
G, J (1985). “A Principled Exception to the Coordinate Structure
Constraint”. Papers from the 21st Regional Meeting of the Chicago Linguistics
Society, ed. by William Eilfort, Paul Kroeber, and Karen Peterson, 133–43.
Chicago: Chicago Linguistics Society.
G, J (1963). Universals of Language. Cambridge, MA: MIT
Press.
G, M T (2002). Language Acquisition: The Growth of Grammar.
Cambridge, MA: MIT Press.
H, J (1979). Deletion in Coordinate Structures. New York: Garland
Publishing.
H, H, and E R (2002). “Person and Number in Pronouns:
A Feature-Geometric Analysis”. Language 78:482–526.
H, M (1990). “The Grammaticization of Passive Morphology”.
Studies in Language 14(1):25–72.
H, J A. (1994). A Performance Theory of Order and Constituency.
Cambridge: Cambridge University Press.
480 REFERENCES
H, E (1993). “Dialectal Variation Inside CP as Parametric Variation”.

Dialektsyntax, ed. by W. Abraham and J. Bayer. West-deutscher Verlag. Opladen,
Special issue 5, Linguistische Berichte.
H, J (1997). “The Status of ‘Wh-expletives’ and the Partial Wh-
movement Construction of Hungarian”. Natural Language and Linguistic Theory
15:509–72.
H, C.-T. J (1982). Logical Relations in Chinese and the Theory of
Grammar. Unpublished doctoral dissertation, MIT.
H, R A., and G K. P (2002). The Cambridge
Grammar of the English Language. Cambridge: Cambridge University Press.
I, O A. (2005). Case-Asymmetry. Munich: Linccom Europa.
J, R (1977). X-Bar Syntax: A Study of Phrase Structure. Cambridge,
MA: MIT Press.
(1983). Semantics and Cognition. Cambridge, MA: MIT Press.
(1990). Semantic Structures. Cambridge, MA: MIT Press.
(2002). Foundations of Language. Oxford: Oxford University Press.
J, A (1987). Transitivity and Grammatical Relations in Inuktitut.
Université d’Ottawa.
K, A, and D M (2002). “Inversely Ordered DPs
in Niuean”. UQAM Working Papers in Linguistics, Proceedings of the Canadian
Linguistic Association Conference 2002, ed. by Sophie and Stance Somesfalean.
K, R M., and J B (1982). “Lexical-functional Grammar: A
Formal System for Grammatical Representation”. The Mental Representation of
Grammatical Relations, ed. by Joan Bresnan. Cambridge, MA: MIT Press.
K, J J., and P M. P (1964). Toward an Integrated Theory of
Linguistic Descriptions. Cambridge, MA: MIT Press.
K, Y (2003). “L1 Acquisition of Japanese Zero Pronouns: The Effect
of Discourse Factors”. CLA Annual Conference Proceedings, 109–20. Montréal:
Université du Québec à Montréal.
K, R S. (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT
Press.
K, T (2000). “HONNIN: A Critical Application of Katada” (1991).
http://www.lang.nagoya-u.ac.jp/proj/genbunronshu/21-2/kinoshita.pdf. (Unpub-
lished paper).
K, J (1978). Locality Principles in Syntax. Dordrecht: Foris Publications.
K, G (2002). “Greek ‘ECM’ and how to control it”. Reading
Working Papers in Linguistics, 6, ed. by Michalis Georgiafentis and Spyridoula
Varlokosta, 39–56.
K, I. S., P. N V, and M S. P (1988).
“Antipassive in Chukchee: Oblique Object, Object Incorporation, Zero Object”.
Passive and Voice, ed. by Masayoshi Shibatani, 651–706. Amsterdam: John
Benjamins.
REFERENCES 481
K, P R. (2004). Analyzing Syntax: a Lexical-functional Approach.

Cambridge: Cambridge University Press.
L, G (1986). “Frame Semantic Control of the Coordinate Structure
Constraint”. Papers from the Parasession on Pragmatics and Grammatical The-
ory, 22nd Regional Meeting, ed. by A. Farley, P. Farley, and K.-E. McCullogh.
Chicago: Chicago Linguistic Society.
L, B, and L R. G (1985). Language and Experience:
Evidence from the Blind Child. Cambridge, MA: Harvard University Press.
L, H (1999). Verbal Morphology. Minimalist Analysis. Oxford:
Blackwell Publishers.
L, D (1990). “The Grammatical Nature of the Acquisition Sequence:
Adjoin-A and the Formation of Relative Clauses”. Language Processing and
Language Acquisition, ed. by Lyn Frazier and Jill De Villiers. Dordrecht: Kluwer
Academic Publishers.
L, C (1983). “Directions for Interlinear Morphemic Translations”.
Folia Linguistica 16:193–224.
M, S, and M T (1986). A Dictionary of Basic Japanese
Grammar. Tokyo: The Japan Times Ltd.
M, C D. (1996). Ergativity. Stanford, CA: CSLI Publications.
M, R (2001). “Null Case and the Distribution of PRO”. Linguistic
Inquiry 32:141–66.
M, R (1985). Logical Form. Cambridge, MA: MIT Press.
MC, E (2006). “Discourse Subordination and Logophoric Binding”.
Research on Language and Computation 4:1–14.
M, J (2001). The Syntax of Silence. Oxford: Oxford University Press.
(2003). “Subject-auxiliary Inversion in Comparatives and PF Output Con-
straints”. The Interfaces: Deriving and Interpreting Omitted Structures, ed. by
Kerstin Schwabe and Susanne Winker, 55–77. Amsterdam: John Benjamins.
M, J (1986). The Formal Semantics of Point of View. Amherst, MA:
GLSA Publications, University of Massachusetts.
M, S (2004). “An Analysis of Depictive Secondary Predicates in
German without Discontinuous Constituents”. Proceedings of the HPSG04 Con-
ference, ed. by Stefan Müller. Center for Computational Linguistics, Katholieke
Universiteit Leuven. CSLI Publications.
M, A. (1994). “A Minimalist Account of Reconstruction Asymmetries”. NELS
24, ed. by Merce Gonzalez, 397–410. University of Massachusetts, Amherst.
N, A (1994). Complex Predicates. Unpublished doctoral dissertation.
Utrecht University, http://www.phon.ucl.ac.uk/home/ad/ pubs.html.
O, H-G (2004). “Nonstandard Wh-questions and Alternative
Checkers in Pagotto”. The Syntax and Semantics of the Left Periphery, ed.
by Horst Lohnstein and Susanne Trissler. Berlin, New York: Mouton de
Gruyter.
482 REFERENCES
P, B (1989). “Binding Implicit Variables in Quantified Contexts”. CLS

25, 342–65. Chicago: Chicago Linguistics Society.
P, D (1978). “Impersonal Passives and the Unaccusative
Hypothesis”. Proceedings of the Fourth Annual Meeting of the Berkeley
Linguistics Society, ed. by Jeri J. Jaeger et al., 157–89.
P, S (1984). Language Learnability and Language Development.
Cambridge, MA: Harvard University Press.
and R J (2005). “The Faculty of Language: What’s Special About
It?” Cognition 95:201–36.
P, M, and E P (2006). “Expanding the Scope of Control and
Raising”. Syntax 9:171–92.
P, C, and I S (1994). Head-driven Phrase Structure Grammar.
Chicago: University of Chicago Press.
P, P M. (1971). Crossover Phenomena. New York: Holt, Rinehart and
Winston.
(1993). “Remarks on Weak Crossover Effects”. Linguistic Inquiry 24:539–56.
P, J (1995). The Generative Lexicon. Cambridge, MA: MIT
Press.
R, O (2003). “Coherent Constructions in German: Lexicon or Syntax?”
Verb Constructions in German and Dutch, ed. by Pieter A. M. Seuren and Gerard
Kempen. Amsterdam: John Benjamins.
R, H  (1983). “Correspondence Effects and the Empty Category
Principle”. Studies in Generative Grammar and Language Acquisition, ed. by Yoku
Otsu, Henk van Riemsdijk, Kazuko Inoue, Akio Kamio, and Noriko Kawasaki,
5–16. Tokyo: International Christian University.
R, L (1991). “Residual Verb Second and the Wh-Criterion”. Technical
Reports in Formal and Computational Linguistics, 2. Faculté des Lettres,
Université de Genève.
(1997). “The Fine Structure of the Left Periphery”. Handbook of Generative
Syntax, ed. by Liliane Haegeman, 281–338. Dordrecht: Kluwer Academic
Publishers.
R, D (2005). “Icelandic Case Fluctuation and Movement into Theta-
Positions”. University of Connecticut Working Papers in Linguistics 13, ed. by
Masashi Nomura, Fumikazu Niinuma, and Lara Reglero, 195–229. Cambridge,
MA: MIT Working Papers in Linguistics.
R, P (1967). The Grammar of English Predicate Complement
Constructions. Cambridge, MA: MIT Press.
R, J R. (1967). Constraints on Variables in Syntax. Unpublished doctoral
dissertation. MIT.
R, C (1988). “On Multiple Questions and Multiple wh-Fronting”.
Natural Language and Linguistic Theory 6:445–501.
REFERENCES 483
S, J M. (1980). “Noun Incorporation in Greenlandic Eskimo”.

Language 56:300–19.
S, A (1858). Volkstümliches aus Sonneberg im Meininger
Oberlande. Weimar: Böhlau.
SD, H Á (1991). “Icelandic Case-marked PRO and the
Licensing of Lexical Arguments”. Natural Language and Linguistic Theory 9:
327–63.
(1992). “The Case of Quirky Subjects”. Working Papers in Scandinavian Syntax
49:1–26.
(2002). “To Be an Oblique Subject: Russian vs. Icelandic”. Natural Language
and Linguistic Theory 20:691–724.
S, R. C. (2003). Ways a World Might Be: Metaphysical and
Anti-Metaphysical Essays. Oxford: Clarendon Press.
S, W (2002). “Wh-Expletives and Partial Wh-Movement: Two
Non-Existing Concepts?” German in Typological Context, ed. by Werner
Abraham and Jan-Wouter Zwart, 285–306. Amsterdam: John Benjamins.
S, G, and G C (2006). “Sherlock Holmes Was in No
Danger”. Drawing the Boundaries of Meaning: Neo-Gricean Studies in Pragmatics
and Semantics in Honor of Laurence R. Horn, ed. by Betty Birner and Gregory
Ward. Amsterdam: John Benjamins.
T, W.-T. D (1997). “On the Absence of Island Effects”. Tsing Hua Journal
of Chinese Studies, New Series 27:125–49.
T, N (1996). An Introduction to Japanese Linguistics. Oxford:
Blackwell Publishers.
U, M, and R K (2003). “Event-related Brain Indices of
Japanese Scrambling”. Brain and Language 86:243–71.
 G, E (2000). “The Absence of Verb-Movement and the Role
of C: Some Negative Constructions in Shakespeare”. Studia Linguistica 54:
412–23.
 R, H (1999). “Clitics: a State-of-the-art Report”. Clitics of
the Languages of Europe, ed. by Henk van Riemsdijk. Berlin: Mouton de
Gruyter.
V, N (2004). “Control & Complementation Revisited”. http://www.essex.
ac.uk/linguistics/LFG/www-lfg.stanford.edu/lfg2004/school/material/seminars/
vincent.ppt. (PowerPoint presentation). University of Manchester.
W, W. D. (1979). “The Syntax and Semantics of Causatives in Nepali”. South
Asian Languages Analysis, 1, ed. by B. Kachru, H. H. Hock, and Y. Kachru,
145–56. University of Illinois, Urbana-Champaign.
W, T (2002). Postverbal Behavior. Stanford: CSLI.
W-L, A, and A H (1997). “The Syntax-
Semantics Interface”. Arbeitspapiere des Sondersforschungsbereichs, Tuebingen.
484 REFERENCES
W, E (2000). “Advances in African Linguistics”. Trends in African

Linguistics, 4, ed. by Vicki Carstens and Frederick Parkinson, 103–17. Trenton,
NJ: Africa World Press.
Z, A, J M, and H T (1990). “Case and
Grammatical Functions: the Icelandic Passive”. Syntax and Semantics, 24:
Modern Icelandic Syntax, ed. by Joan Maling and Annie Zaenen, 95–136. New
York: Academic Press.
Z, N (2001). “The Structures of Depictive and Resultative Constructions
in Chinese”. ZAS Papers in Linguistics 22:191–221.
Z, C. J-W (1993). Dutch Syntax: A Minimalist Approach. Groningen:
University of Groningen.
(1996). “Verb Clusters in Continental West Germanic Dialects”.
Microparametric Syntax and Dialect Variation, ed. by James R. Black and
Virginia Motapanyane, 229–58. Amsterdam, Philadelphia: John Benjamins.
Index
A position 471 of PRO in small clause 305

A position 471 of VP anaphora 454
A chain, see chain of VP ellipsis 437, 453–5
absolutive 76–7, 191, 197 argument 20, 69, 146–7, 471
across the board extraction 349 bound 255, 257
accusative case 42, 44, 73–4, 77, external 303
153, 173–5, 189, 191, 246, 260, implicit 171, 410, 473
270 oblique §5.5.2
adjective §2.1.3, 30, 32, 44, 45–7, article 27
49–50, 104 n.1, 110, 176, 187, 270, ATB extraction, see across the board
attributive 22 extraction
predicative 22 attribute value matrix 11, 42–3, 144,
adjunct 69–70, 122, 127, 176, 222, 252, 471
427, 455, 471 auxiliary verb 21, 112–22
adjunct island constraint 369–70 AVM, see attribute value matrix
adverb effect 368
adverbial 26, 82 BAE, see bare argument ellipsis
Affix hopping 113–14, 119–20 bare argument ellipsis 438, §11.1, 452,
agreement 15, 47, 49, 72, §3.5.3, 80, 459, 461, 464
85–6, 116, 197 n.6, 421, 471 barrier 270, 367–8
between specifier and head 215, Bellunese 334–5
223, 353–5, 358 Biak 37
in existential construction 201, binding 255, 274, 281, 341, §10.2,
232–3 §10.4–5, 426–7, 471
in Italian 203–6 CS- 416, 418, 428
allomorph 14 GF- 416, 418, 423, 428
allomorphy 14, 471 binding theory 405–9, §10.4.1, 418,
anaphor 405, 414–18, 424; see also 422–4, 425, 427
binding theory Bulgarian 376
long distance 420–2, 474 Burushaski 76
antecedent
of bare argument ellipsis 447–9 c-command 275–8, 312–13, 315, 322,
of binding 405, 407, 408, 471 327, 350, 406–10, 419, 423–4,
of predicate 300, 471 426–7
486 INDEX
c-selection 169, 171–2, 472 constraints 263, 344, §9.4, §9.7, 445–6,
canonical structure 84–5, 471 464; see also individual constraints
case construction 33, 472
abstract 215 control 472
and grammatical function 72–7 and predication §8.1.2, 303, 305–10
licensing 215, 472 arbitrary 252, 276–7
morphological 36, 39–40 in gerundives 252
quirky 155, 175, 474 in infinitival question 373
see also absolutive, accusative, dative, object 246–8, 250–2, 300–1,
ergative, instrumental, nominative obligatory 239
case filter 215 subject 241, 251–3, 277
case licensing 215 syntactic 268
category, see lexical category; syntactic see also PRO
category controller 241, §7.4.3
chain 321–2, 327–8, 330, 331, 334, 338, coordinate structure constraint 347–9,
341–3, 344, 349–51, §9.5.1.1, 362, 372
371, 422–5 coordination §3.6.3, 90
Chichewa 192, 194, 229 coreference §10.1, 472
Chinese 331, 333, 371–2, 420–1 correspondence 2–4, 8, 139, §5.2,
Choctaw 77–8 150–4, §5.6, 211, 324–5
Chukchee 197, 227 correspondence rule 152–3, 159–60,
cleft §9.8.3.1 211, 274, 324–7, 338–9, 472
coercion 20 n.3, 171, 472 CS, see conceptual structure
competence 4, 472 cyclic principle 266
complement 8, 21–2, 46–7, 68–9, Czech 376
103–5, 109, 111, 114, 116–17, 121,
124–6, 172, 174, 187–9, 201, 222, D-structure 210–11, 213–14, 350–1,
229, 279, 306–10, 329, 346, 379, 357 n.8
415, 450, 451, 472 dative alternation 193–5, 230–1
gerundive 252 dative case 42, 73–4, 155 n.5, 174, 447
infinitival §7.1, 253–60, 267–73, deep structure, see D-structure
278–81, 302–3, 309, 366, 368, degree 111
411–13 deletion 437–8, 462–4
complementizer 124, 353–5, 358–60, demonstrative 27
363–5, 368 derivation 9, 114, 205, 210–11, 216,
complex NP constraint 345–6, 365, 220, 223, 321, 350, 352–3, 357, 367,
367, 369–71 371
concept 141–3 determiner 3, 27, 110
conceptual structure 140–2 detransitivization 197, 472
conjunction 28, 83 do 88–9
constraint on extraction domains do support 120–1
(CED) 367–8 do X anaphora 451, 454–6
INDEX 487
Dutch 218, 220, 421 head

Djirbal 224–5 functional vs. lexical 126
of a chain 321, 342, 473
ECM, see exceptional case marking of a phrase 44, 63, 68, 103, 473
endocentric 44, 103, 106, 123, 308, head feature principle 115
472 HPSG 8 n.3, 115 n.6, 267, 274 n.6,
epithet 403 340 n.6
EPP, see extended projection principle
EPP feature 262–3, 272 Icelandic 175, 293–7
ergative 75–6, 77, 79, 96–7, 155 n.5, idiom 33, 145, 158, 243
191, 197 idiomaticity 33
exceptional case marking 272, 473 impersonal passive 218–19, 228, 230
existential 201 indirect licensing 448, 473
exocentric 44, 103, 473 indices 401–3
extended projection principle 223 information structure 361
extraposition 201, 222, 381, 473 instrumental 42, 74, 174, 176, 190, 197
Inuit 96
false reflexive 302 inversion, see subject-AUX inversion
feature discharge §9.5.2, 356, 358, island 344, 372, 473; see also
360 n.10, 365, 376 constraints
focus 89, 379–81, 449, 460 Italian 202–3, 205, 218, 233–4, 292–3,
for complementizer 273, 368, 378 359, 420–1
French 41, 104, 124–5, 207–8, 234, 331, iterated CP 364
333–4, 359, 371–2
Japanese 73–4, 103–4, 191, 195–6, 206,
generative grammar 4 285, 331–3, 339, 371–2, 421
gapping 438, §11.3 Jiwarli 76–77
German 55, 74, 124, 218–20, 290, 291,
331, 359, 373–5, 447 Kalkatungu 197
GF, see grammatical function Konjo 79
GF correspondence rules 153–4, 173–4
gloss 39, 42 labeled bracketing 13, 62–3, 65
government 270–1, 412–14 landing site 351–2
Government Binding (GB) theory 173, language acquisition 5,
221, 250, 262, 269–70, 272, 357, and constraints 349
412–14 and syntactic categories 50–2
grammar 3, 5, 473 Latin 191
grammatical 3, 473 left branch constraint 342, 346, 445
grammatical function 150–1, 414, 473; lexical category 11
see also object, subject lexical entry 29, 145–6, 148, 151–2,
Greek 288–9 155–6, 169, 172, 195, 205
Greenlandic Eskimo 227–8 LF, see logical form
488 INDEX
LFG 8 n.3, 274 n. 6 logical 186–7, 254, 473

linking rules 150–1, 154, 185, 209, 254, oblique §5.5.2, 172, 174, 187, 191–3,
416 197, 199
locality operator 244, 324, 326–7, 331, 337,
of binding 405–7, 411–14, 416–18 361 n.11,
of movement 264, 369 empty 360, 369
Logical form 356–7, 423, 464
p-stranding 331, 342–3, 373, 474
Manggarai 216–17 paradigm 29, 46, 474
meaning 2, 141; see also conceptual adjective 23, 55
structure inflectional 30
merge 213 n.11, 357 n.8 irregular 34
Middle English 359 noun 15–16, 18, 36–40
Middle High German 359 regular 34
Minimalist Program 213–14, 266, 357, verb 19–20, 34–5
365 particle 92
Minimum Distance Principle 276–8 passive §6.8
morphology 15, 28, 49, 51 n.15, and infinitival complement 250–1,
derivational 30–2, 53, 472 254, 256–7, 260–1, 263–4, 266–7,
inflectional 30, 473 270, 272, 284
movement 84, 344 performance 4, 474
in LF, see Logical form PF see Phonetic form, phonetics
in passive 215–21 phonetic form 1, 357, 437
in raising 259–60 phonetics 139
in topicalization §9.6.2 phonology 139
in wh-question §9.5 pied-piping 329–30, §9.3.1, 357–8, 373,
successive cyclic 367, 424 474
Polish 376
negation 90, 117, 125 predicate 67
Nepali 97, 191, 196 depictive 300–2, 304, 310–11
Niuean 131 resultative §8.1.3, 304
nominative 41–2, 73–4, 153, 173–5, Principles and Parameters theory
189, 220 357
number 2, 15–16, §2.4.2.1, 55, 71–2, 77, pro 205, 474
81, 86, 204–5 PRO 241–2, 264–72, 274–8, 474
and predication 305, 308
object in questions 357–8
and canonical structure 84 see also control
direct 7, 9, 70–1, 73, 77, 156, 191–2, pro-drop 205–6, 474
197–8, 215–16 proform 81, 474
indirect 70–1, 73 projection principle 223
INDEX 489
pronoun 404–5, 419; see also binding Russian 39–40, 41–3, 74, 96–7, 98,
theory 153–4, 174–6, 190, 375
proposition 67, 299
pseudo-cleft §3.6.4.3, 307, §9.8.3.2 s-selection 169–71, 475
pseudo-passive 235 scope 325, 331–3, 339, 356–7, 410
selection §5.7
quantifier 27, 57, 110, 409–11 in bare argument ellipsis 448–9
question 6–7, 325–7 in passive 186, 215
echo 335–7 in raising to object 248–9
embedded 323, 328, 332, 380 in raising to subject 243
infinitival 373 in subject control 239–40
multiple §9.2.4, 375 in verbal sequence 114–15
quiz 335–7 see also c-selection, s-selection
partial wh- 373 selection restriction 170, 186
wh- 90–1, 124, 322–8, §9.2, 343–4, semantic anomaly 17–18, 20, 475
449–50 semantic role, see thematic role
yes-no 323–4 sentential subject constraint 347
Serbo-croatian 375
raising 475 Slovene 36
subject §7.1.2, 253 sluicing 450
to object §7.1.4, 253, 266–7, 411 Spanish 217, 234–5
reciprocal 405 Spell Out 356–7
recursion 108, 475 structure preserving hypothesis §6.8.3,
reference 141–2, 202, 206, 240; see also 352, 363
coreference subcategory 21
referential index 144–5, 240, 401 subcategorization 169, 238, 475; see
referring (R-) expression 404, 408–9, also c-selection
475; see also binding theory Subjacency 366
reflexive 175, 403, 404–6, 411, 414–18, subject 67, 70–2, 98, 100
422, 424–5 and canonical structure 84
relative clause 110, §9.3, §9.5.4 default linking 151–2
appositive 340 expletive 200, 223, 227, 242–4, 472
free 376–7, 380 in VP 128–9
infinitival 378 logical 186–7, 254, 474
internally headed 377–8 subject-AUX inversion 118–19, 121,
non-restrictive 340 125, 323, 334–5
restrictive 340 substitution 11, 13, 17–19, 87
that- 340–1, 358, 360 S-structure 210, 321, 357
wh- 340–2, 358–9 Swahili 206, 226
zero- 340–1, 358, 360 Swedish 370–1
root 34, 39 synonymy 142
490 INDEX
tag question 72, 77, 96, 100 uniformity 9, 104–5, 125, §7.4, 305,
tail 321 355, 412; see also UTAH
that Universal Grammar 5–6, 475
complementizer 124, 358–9 UTAH 224, 267
relative, see relative clause
that-t effect 368–9, 464 V-raising 125
thematic role 66, 146, 173, 475 V2 language 220, 370
external 303, 305, 473 verb phrase (VP) 68–70, 87–9
thematic structure 148, 475 anaphora 453, §11.2.2
theta criterion 172–3, 180, §6.9, 264, ellipsis 80–1, 437, §11.2, 462
267 in English verbal sequence 115–17
topicalization 84, 86–9, 307, §9.6,
370, 422–3, 425, 444–6, 463–4, West Greenlandic 76
475 wh-criterion 355
trace 351, 369 wh-in-situ 331, §9.2.3, 337, 355–6,
transformation 113–14, §6.8, 245, 475 371–2, 475
tree 13, 63–5 wh-island constraint 344–5, 368
Tukang Besi 78, 226–7 word 29
Turkish 226
zero-allomorph 14, 35
UG, see Universal Grammar
ungrammaticality 3, 17–18, 171, 475 Ë-role, see thematic role

Culicover - Natural Language Syntax PDF

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Culicover - Natural Language Syntax PDF

Diunggah oleh

Hak Cipta:

Format Tersedia

   

Natural Language Syntax

The Grammar of Words

A Practical Introduction to Phonetics

Natural Language Syntax

Principles and Parameters

A Semantic Approach to English Grammar

Semantic Analysis: A Practical Introduction

Cognitive Grammar: An Introduction

Translation: Theory and Practice

ISBN 978–0–19–923017–4 (Hbk.)

3. Basic sentential structure 61

4. Phrasal Categories 103

5. Conceptual structure and the lexicon 139

6. Argument Correspondences 185

6.8.3. Structure preserving movement 212

7. Complex clauses: raising and control 237

9.8.1. Questions 373

10. Coreference and Binding 401

10.3. Quantiﬁcation 409

11. Fragments 437

The aim of this book is to provide an introduction to the study of natural

I am deeply indebted to many people who have contributed to the design,

helpful comments on all of the chapters, and Melanie Henschke, Nora

INSTR instrumental (case)

TOP topic (feature)

1.1. What is syntax?

The conventional answer to the question “What is syntax?” in theoretical

This form is often called phonetic form, or PF.

– or as a phrase of the category noun phrase (NP) consisting of a sequence

Owing to the fact that the number of possible correspondences in a

(6) a. the biting dogs (are vicious)

The symbol “∗ ” before a string of words as in (6b) indicates that it is

1.2. The goals of linguistic theory1

In our study of syntax we are not simply interested in the description of a

of particular phenomena are of fundamental importance, but are in

of native speaker competence corresponds in some interesting way to the

1.3. Where does syntactic theory ﬁt in?

Syntactic theory sits squarely in the middle of this general perspective. A

A theory of syntax must tell us that it is possible to form a question as

1.4. Simpler Syntax

Interacting with the problem of how language is acquired is the question

(9) a. Mary expects that she will win.

A related difference between Simpler Syntax (and a number of other

This chapter is concerned with syntactic categories. Section 2.1 introduces

2.1. Traditional categories

(1) a. dog, unicorn, truth, Mary, encouragement, plumber, . . . [noun]

Substitution: The result of substituting a word of a category C for another word of

We develop the general idea of substitution and its relation to grammatical-

In general, if a sentence contains the word cat, it is possible to replace cat

And we can have

(3) a. The cat is sitting on the mat.

(5) a. one very fat furry cat

Notation: Representing categories

Naturally, in a complete description of a string of words, all of the words

dog dog- SG /dOg/

What these examples show is that where there is a morphological mark-

of alternate forms whose distribution is governed by meaning differences (in

Notation: Representing paradigm properties

In an AVM we list each feature and its corresponding value. A non-

This discussion of categories thus far shows three things:

(8) a. Unicorns meow.

arrangement of the words and phrases of a sentence according to their