Академический Документы
Профессиональный Документы
Культура Документы
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
Meaning in Language
An Introduction to Semantics and Pragmatics
Second edition
by Alan Cruse
Pragmatics
by Yan Huang
Diachronic Syntax
by Ian Roberts
Linguistic Categorization
Third edition
by John R. Taylor
IN PREPARATION
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
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
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
Research questions 318
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
∗
Glossary 471
References 477
Index 485
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
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
– or as a string of words –
(3) biting 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
1
Parts of this section are adapted from Culicover and Jackendoff 2005:
Chapter 1.
4 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
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?
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.
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
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.
This page intentionally left blank
2
Syntactic Categories
The traditional lexical categories found in English are noun (1a), verb (1b),
preposition (1c), adjective (1d), and adverb (1e).
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:
2.1.1. Nouns
and
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
[N cat]
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
14 2. SYNTACTIC CATEGORIES
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
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
2.1. TRADITIONAL CATEGORIES 15
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
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.
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
18 2. SYNTACTIC CATEGORIES
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).
2.1. TRADITIONAL CATEGORIES 19
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
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.
20 2. SYNTACTIC CATEGORIES
(14)
B ARE FORM P RESENT PAST - ING FORM - EN FORM
∗
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.
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.
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
(24)
B ASE C OMPARATIVE S UPERLATIVE
These adjectives may also be modified by intensifiers like very and so.
tall
very angry
(25)
so
old
irritating
∗
(26) the more present King of France
∗
the more alleged assassin
∗
the more perfect idiot
4
This sentence is acceptable under another interpretation of perfect.
24 2. SYNTACTIC CATEGORIES
2.1.4. Prepositions
2.1.5. Adverbs
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.
26 2. SYNTACTIC CATEGORIES
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.
2.1. TRADITIONAL CATEGORIES 27
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.
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
28 2. SYNTACTIC CATEGORIES
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.
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
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.
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
syntactic properties
meaning
word formation
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.
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.
34 2. SYNTACTIC CATEGORIES
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.)
2.2. MORPHOSYNTAX 35
(56)
VERBAL root 3rd singular other -ing form past -en form
INFLECTION present present (progressive) (past
tense tense participle)
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.
36 2. SYNTACTIC CATEGORIES
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)
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
2.2.4.1. 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 )
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
(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.
38 2. SYNTACTIC CATEGORIES
(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.
2.2. MORPHOSYNTAX 39
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.
40 2. SYNTACTIC CATEGORIES
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
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
2.2. MORPHOSYNTAX 41
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.
42 2. SYNTACTIC CATEGORIES
(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.
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
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
44 2. SYNTACTIC CATEGORIES
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.
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
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.
46 2. SYNTACTIC CATEGORIES
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.
(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.
2.4. THE THEORY OF LINGUISTIC CATEGORIES 47
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.
11
This example is similar to utterances by Yoda (in the Star Wars movies), but is
not grammatical in English.
2.4. THE THEORY OF LINGUISTIC CATEGORIES 49
∗
(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).
50 2. SYNTACTIC CATEGORIES
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?
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
2.4. THE THEORY OF LINGUISTIC CATEGORIES 51
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.
52 2. SYNTACTIC CATEGORIES
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
[§2.1.1.]
3. The English morpheme -er has more uses than those discussed in
Exercise 2, as shown by the following examples.
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”.
54 2. SYNTACTIC CATEGORIES
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.
[§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
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”.)
[§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,
56 2. SYNTACTIC CATEGORIES
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
[§2.3.]
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
many
the
some
these
b. ∗ two books
my
any
Terry’s
each
(The notation ∗ (. . . ) here means that the string is ungrammatical when the
material in parentheses is omitted.)
[§2.4.]
Research questions
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.
but
What distance
(4) a. did the package fall?
How far
b. How much did it rain?
[§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
60 2. SYNTACTIC CATEGORIES
tight
tightly
b. You have to grab the handle ∗ .
firm
firmly
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.
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]
(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
(7) S
NP VP
QUAN N V NP
dogs
1
Evidence that VP is a constituent of the English sentence is given in
section 3.6.
64 3. BASIC SENTENTIAL STRUCTURE
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.
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
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).
(15) S: Proposition
V ...
3
This diagram and the discussion leading up to it is inspired by the treatment of
basic English structure in Huddleston and Pullum 2002.
68 3. BASIC SENTENTIAL STRUCTURE
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]
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.
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.
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
∗
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.
3.5.1. Case
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
‘Taro gave a/the flower to Hanako.’
c. Hanako-ni Taro-ga hana-o ageta
Hanako-DAT Taro-NOM flower-ACC gave
‘Taro gave a/the flower to Hanako.’
d. Hanako-ni hana-o Taro-ga ageta
Hanako-DAT flower-ACC Taro-NOM gave
‘Taro gave a/the flower to Hanako.’
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.
74 3. BASIC SENTENTIAL STRUCTURE
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”.)
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”.)
5
There are some contexts in German where the head noun also shows marking
for case.
3.5. MARKING GRAMMATICAL FUNCTIONS 75
(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
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.
3.5. MARKING GRAMMATICAL FUNCTIONS 77
(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.’
3.5.3. Agreement
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.’
3.6.1. Ellipsis
On the other hand, there are examples that suggest that an NP can be
omitted in subject position.
3.6. TESTS FOR CONSTITUENCY 81
(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.
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]
82 3. BASIC SENTENTIAL STRUCTURE
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?
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
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.
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]].
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.
84 3. BASIC SENTENTIAL STRUCTURE
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]]
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.
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.
3.6. TESTS FOR CONSTITUENCY 85
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).
86 3. BASIC SENTENTIAL STRUCTURE
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
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.
(8) a. President Smith called.
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.
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.
(88) a. President Smith called.
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.
3.6. TESTS FOR CONSTITUENCY 89
3.6.4.3. Pseudo-cleft
A test that is related to topicalization is the pseudo-cleft construction, illus-
trated in (91).
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 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 ___?
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.
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.
92 3. BASIC SENTENTIAL STRUCTURE
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]
Exercises
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?
94 3. BASIC SENTENTIAL STRUCTURE
[§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.]
[§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.]
96 3. BASIC SENTENTIAL STRUCTURE
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.]
98 3. BASIC SENTENTIAL STRUCTURE
Problems
(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.
[§3.6.4.]
100 3. BASIC SENTENTIAL STRUCTURE
Research questions
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.]
RESEARCH QUESTIONS 101
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
This page intentionally left blank
4
Phrasal Categories
4.1. X theory
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).
106 4. PHRASAL CATEGORIES
… X0 …
We will flesh out the details of the structure of each type of phrase as we
proceed.
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 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].
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∗ )
(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
– 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].
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
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.
these
d. furry poodles that I own
the
e. a poodle in the park
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.
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.
112 4. PHRASAL CATEGORIES
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]
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.
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.
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
4.5. THE ENGLISH VERBAL SEQUENCE 115
(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.
The structure is
(49) VP
V0 VP
[PROG. PART]
be
V0 …
[PROG. PART]
running
(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.
116 4. PHRASAL CATEGORIES
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.
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.
4.5. THE ENGLISH VERBAL SEQUENCE 117
(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.
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.
118 4. PHRASAL CATEGORIES
(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?
∗
(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 ]
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).
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).
4.5. THE ENGLISH VERBAL SEQUENCE 121
Exercise 4 asks you to derive a number of verbal sequences with and without
do.
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?
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
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).
122 4. PHRASAL CATEGORIES
(75) S
NP AUX AP
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.
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) XP
Spec X
X0 Comp
4.7.1. IP and CP
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.
124 4. PHRASAL CATEGORIES
(78) IP
NP I
I0 VP
tense (M)
(79) CP
Spec C
C0 IP
that NP I
I0 VP
tense (M)
(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.’
4.7. APPLICATIONS OF X THEORY 125
(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.
126 4. PHRASAL CATEGORIES
(82) DP
… D
D0 NP
Spec N
N0 …
(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
4.7. APPLICATIONS OF X THEORY 127
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?
128 4. PHRASAL CATEGORIES
∗
(85) Sandy will Terry read the book.
(86) IP
Spec I
I0 VP
DP V
Sandy V0 DP
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].
130 4. PHRASAL CATEGORIES
[§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.]
[§4.5.]
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
[§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) Niuean
e tau manu kula fulufuluola e:
A BS C PL bird red beautiful that
‘those beautiful red birds’
[Kahnemuyipour and Massam 2002]
132 4. PHRASAL CATEGORIES
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).
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.)
[§4.5.]
∗
(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.]
(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]
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.
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
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
RESEARCH QUESTIONS 137
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.]
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
This page intentionally left blank
5
Conceptual structure
and the lexicon
5.1. Overview
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
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
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.
142 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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
(3)
Linguistic expression Conceptual Structure Reference in the actual
world
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,
144 5. CONCEPTUAL STRUCTURE AND THE LEXICON
the notation that we use in this book for reference is superscript Greek
letters.
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.
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.
SYNTAX … [N book] …
CS … BOOK …
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
146 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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/
CS DIE (…)
5.3. CS relations
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
(11) STINK(CHEESE‰ )
(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.
(13) S
NP:WRITER VP
Mary V NP:WRITTEN
write a book
<WRITER,WRITTEN>
(16) SYNTAX [CATEGORY V]
CS WRITE(AGENT:X,PATIENT:Y)
(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
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)
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.
152 5. CONCEPTUAL STRUCTURE AND THE LEXICON
(22) SYNTAX S
NP VP
V NP
PHON write
GF Subject Object
CS WRITE(AGENT:X, THEME:Y)
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)
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.
5.5.1. Intransitives
(31) SYNTAX S
NP VP
V PP
go P NP
into
GF Subject
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
(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.
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.
158 5. CONCEPTUAL STRUCTURE AND THE LEXICON
(38) SYNTAX S
NP VP
V NP
seek
GF Subject Object
CS SEEK(AGENT:X, THEME:Y)
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.
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
(41) SYNTAX S
NP VP
Sandy V
runs
GF Subject
CS RUN(AGENT:X)
160 5. CONCEPTUAL STRUCTURE AND THE LEXICON
(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)
(44) Lexicon:
Sandy run
CS SANDY RUN(AGENT:X)
(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.
162 5. CONCEPTUAL STRUCTURE AND THE LEXICON
(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
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
CS CHASE(AGENT:TERRY, PATIENT:ROBIN)
NP VP
V NP
GF Subject Object
CS CHASE(AGENT:TERRY, PATIENT:ROBIN)
NP VP
Sandy V
runs
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.
5.6. COMPUTING CORRESPONDENCES 165
(54) SYNTAX S
NP VP
Sandy V
runs
GF Subject
CS RUN(AGENT:X)
(55) SYNTAX S
NP VP
Sandy V
PHON
runs
GF Subject
CS RUN(AGENT:SANDY)
N PP
friend P NP
of Sandy
CS FRIEND(X) SANDY
to
N PP
friend P NP
of Sandy
CS FRIEND(SANDY)
(58) Lexicon:
Sandy run friend
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
168 5. CONCEPTUAL STRUCTURE AND THE LEXICON
N PP V
friend P NP runs
of Sandy
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
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])]
170 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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.
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.
172 5. CONCEPTUAL STRUCTURE AND THE LEXICON
something, and so on. Thus, the CS representation has two arguments, but
there is only one syntactic argument, the Subject.
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.
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.
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.
174 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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) ⇓ ↔ ⇓
Theme/Patient Object
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
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]
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.
176 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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.’
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).
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).
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.
178 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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.
[§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.
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.]
Problems
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.]
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.
182 5. CONCEPTUAL STRUCTURE AND THE LEXICON
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
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.
This page intentionally left blank
6
Argument Correspondences
6.2. Passive
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.
NP VP
V PP
[PASSIVE]
P NP
seek
(=‘sought’) by
GF Subject Object
CS SEEK(AGENT:X, THEME:Y)
The Object GF is simply not associated with any argument and not realized
syntactically in the passive.
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.
188 6. ARGUMENT CORRESPONDENCES
NP VP
be VP
[PASSIVE]
V PP
[PASSIVE]
P NP
seek
(=‘sought’) by
GF Subject Object
CS SEEK(AGENT:X, THEME:Y)
! "
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
GF Subject Object
CS F(AGENT:X, THEME:Y)
190 6. ARGUMENT CORRESPONDENCES
GF Subject Object
CS F(AGENT:X, THEME:Y)
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
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.’
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]
CS F(h1:X,h2:Y,h3:Z)
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.
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]
(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]
GF Subject Object
CS F(AGENT:X, THEME:Y)
GF Subject Object
CS F(AGENT:X, THEME:Y)
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
(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
(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
(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.
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.
202 6. ARGUMENT CORRESPONDENCES
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.
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.
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.’
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.
GF Subject
CS F(h:X [FEATURES], …)
GF Subject
CS F(h:X [FEATURES], …)
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
CS F(h:X [FEATURES], …)
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),
206 6. ARGUMENT CORRESPONDENCES
… 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]
6.7.2. Clitics7
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.
208 6. ARGUMENT CORRESPONDENCES
b. Jean la préfère.
John her prefers
‘John prefers HER.’
c. ∗ Jean LA préfère.
(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
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.
(7)
SYNTAX S
NP VP
be VP
[PASSIVE]
V PP
[PASSIVE]
P NP
seek
(=‘sought’) by
GF Subject Object
CS SEEK(AGENT:X, THEME:Y)
210 6. ARGUMENT CORRESPONDENCES
(9) be (passive)
CATEGORY V
! "
SYNTAX VP
COMPS
PASSIVE
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.
212 6. ARGUMENT CORRESPONDENCES
(70) S
NP VP
Kim V NP
saw Sandy
NP VP
Sandy V VP
was V PP
see+en P NP
by Kim
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
6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE 213
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.
214 6. ARGUMENT CORRESPONDENCES
The simple answer is, if the NP does not move, the resulting sentence is
ungrammatical.
∗
(73) It was seen Sandy.
∗
Was seen Sandy.
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.
(74) IP
[ ] I
ca
se
I VP
V+en no case NP
case
216 6. ARGUMENT CORRESPONDENCES
(75) Recently released from prison, Sandy immediately looked for a job.
(76) IP
[ ] I
I0 VP
V VP
(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]
(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.’
218 6. ARGUMENT CORRESPONDENCES
(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]
6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE 219
[ ] I
cas
e
I0 VP
V0 VP
wurde V+en
getanzt
es VP
V0 VP
[PASSIVE]
werd- V0 …
[TENSE] [PASSIVE]
getanzt
GF Subject
CS DANCE(AGENT:X)
220 6. ARGUMENT CORRESPONDENCES
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.
13
See Haspelmath 1990 for discussion.
222 6. ARGUMENT CORRESPONDENCES
Theta criterion
i. Every θ-role is assigned to one and only one argument.
ii. Every argument is assigned one and only θ-role.
– 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 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.
224 6. ARGUMENT CORRESPONDENCES
(92) IP
NP I
[NP]
I0 …
[NP]
Exercises
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.]
(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.’
228 6. ARGUMENT CORRESPONDENCES
[§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
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)?
(1) Chichewa
a. fisi a-na-dul-a chingwe ndi mpeni.
hyena(1) SUBJ(1)-PAST-cut-ASP rope with knife
‘The hyena cut the rope with a 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.
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.
(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.]
Research questions
and which ones will not? Feel free to add more verbs to test your
hypothesis.
[§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.
(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.
RESEARCH QUESTIONS 233
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
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
234 6. ARGUMENT CORRESPONDENCES
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
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.]
(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.]
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
(3) V
to run
This rule will accommodate the patterns V-VPINF and V-NP-VPINF that are
seen in (2).
238 7. COMPLEX CLAUSES: RAISING AND CONTROL
try VPINF
appear VPINF
persuade NP VPINF
believe NP VPINF
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.
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]))
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.
(17) SYNTAX S
NP VP
V0 S
NP VP
try
PRO V0 ...
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.
7.1. INFINITIVAL COMPLEMENTS 243
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.
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.
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.
244 7. COMPLEX CLAUSES: RAISING AND CONTROL
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, . . .))
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.
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)
[ ] VP
V0 S
appear NP VP
V0 ...
CS APPEAR(F(θ:X, . . .))
246 7. COMPLEX CLAUSES: RAISING AND CONTROL
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.
7.1. INFINITIVAL COMPLEMENTS 247
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
(28) SYNTAX S
NP VP
V0 NP VP
persuade V0 ...
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.
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.
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
∗
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 ...
(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).
250 7. COMPLEX CLAUSES: RAISING AND CONTROL
(35) SYNTAX S
NP VP
V0 S
believe NP VP
V0 ...
Raising to object
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.
7.1. INFINITIVAL COMPLEMENTS 251
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.
A complication in the picture that we have presented here is the verb expect.
Consider the following sentences.
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.
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.
252 7. COMPLEX CLAUSES: RAISING AND CONTROL
7.1.6. Gerundives
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
(45) Terry considered (= thought about) Robin spending the night at a local hotel.
(47) I began to speak German to the children when they were young.
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 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
254 7. COMPLEX CLAUSES: RAISING AND CONTROL
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).
7.2. MORE CORRESPONDENCES 255
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:·,. . . ))
NP VP
V0 VP
expect V0 ...
to V
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 .
256 7. COMPLEX CLAUSES: RAISING AND CONTROL
(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.
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).
7.2. MORE CORRESPONDENCES 257
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
(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
TENSE INF
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.
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
(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
(68) S
[e] VP
V0 S
appear NP VP
Robin
260 7. COMPLEX CLAUSES: RAISING AND CONTROL
∗
(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.
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.
7.3. RAISING AS MOVEMENT 261
(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]
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.
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.
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
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
7.3. RAISING AS MOVEMENT 263
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).
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.
264 7. COMPLEX CLAUSES: RAISING AND CONTROL
∗
(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.
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.
7.3. RAISING AS MOVEMENT 265
(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
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
266 7. COMPLEX CLAUSES: RAISING AND CONTROL
(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 ]]]
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
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.
268 7. COMPLEX CLAUSES: RAISING AND CONTROL
(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
7.4. SYNTACTIC CONFIGURATION AND CONTROL 269
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.
ACC NOM
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.
7.4. SYNTACTIC CONFIGURATION AND CONTROL 271
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
(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
272 7. COMPLEX CLAUSES: RAISING AND CONTROL
(94) IP2
NP I
George I0 VP2
V0 IP1
expects NP I
case
Al I0 VP1
V0
to win
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.
7.4. SYNTACTIC CONFIGURATION AND CONTROL 273
(that) Al will win
(97) a. I expect very much ∗ Al to win .
for Al to win
(that) Al will 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
(that) Al will 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
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
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
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.
276 7. COMPLEX CLAUSES: RAISING AND CONTROL
(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.
7
The MDP was first proposed by Rosenbaum 1967.
7.4. SYNTACTIC CONFIGURATION AND CONTROL 277
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]
NP VP
George V0 S2
wants S1 VP
NP VP to be difficult
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.
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
7.4. SYNTACTIC CONFIGURATION AND CONTROL 279
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]].
280 7. COMPLEX CLAUSES: RAISING AND CONTROL
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).
(112)
SYNTAX S
NP VP2
V0 PP VP1
plead P0 NP V0 ...
with
GF Subject [Subject ]
9
See Culicover and Jackendoff 2005:Chapter 10 for extended discussion of the
semantic basis of control.
EXERCISES 281
Exercises
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.
[§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.)
[§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
NP VP
George V0 VP
expects V0
to win
GFs Subject [Subject ]
CS EXPECT(EXP:GEORGE·, THEME:WIN(EXP:·)
[§7.2.]
(1) A
B C
D
E F G
H I J K
[§7.4.3.]
[§7.4.3.]
284 7. COMPLEX CLAUSES: RAISING AND CONTROL
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
Problems
(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.’
2. Give the lexical entries for the English causative alternation melt as in
[§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.]
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
(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]
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.
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.)
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.]
RESEARCH QUESTIONS 291
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’
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.’
RESEARCH QUESTIONS 293
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]
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.
A. The following data show the behavior of alla “all”. What does this
suggest about the proper syntactic analysis of the infinitive?
Object
Subject
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
This page intentionally left blank
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).
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
(8) a. Sandy thinks that it is important [to check the oil level]arb .
b. ∗ Sandy thinks that it rained drunkarb .
(9) Sandy thinks that it is important [to check the oil level soberarb ]arb .
(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)]
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.
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.
AP
PP
(16) VP → V (NP)
NP
VPINF
8.1.4. Correspondences
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
NP VP
V NP AP
GF Subject Object
NP VP
V NP AP
GF Subject Object
(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
PRO flat
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.
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 ___.
8.2. SMALL CLAUSES 307
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
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,
8.2. SMALL CLAUSES 309
for example, why it can have a dummy subject only when the complement
selects a dummy subject, e.g.
(38) S
NP VP
V SC
appear NP AP
Sandy unpleasant
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
(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
(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’
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]
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
rather than
(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
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.]
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.]
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.
[§8.3.]
318 8. PREDICATION
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 .
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
RESEARCH QUESTIONS 319
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
8.1. 1 1
8.2. 2, 3, 4, 5
8.3. 6, 7 2
9
A constructions
Head Tail
CHAIN
322 9. A CONSTRUCTIONS
9.1. Questions
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?
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
buy
GF Subject Object
CS Q·(BUY(AGENT:SANDY,THEME:·))
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.
328 9. A CONSTRUCTIONS
NP …
[WH]
GF Subject
CS Q·(…, ·, …)
NP VP
Kim V S
ask NP NP VP
buy
GF Subject [ Subject Object ]
CS ASK(AGENT:KIM,THEME:Q·(BUY(AGENT:SANDY, THEME:·)))
9.2. TYPES OF WH-QUESTIONS 329
In this case, the phrase that appears in initial position is a PP that contains
a phrase whose specifier is [WH]:
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.
330 9. A CONSTRUCTIONS
(19) SYNTAX S
PP AUX NP VP
GF Subject Object
CS Q·(GET(AGENT:YOU,THEME:QUOTE,SOURCE:·) )
(20) SYNTAX S
NP AUX NP VP
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] ]?
9.2. TYPES OF WH-QUESTIONS 331
(25) Q· (. . . , ·, . . . )
(26) SYNTAX S
NP VP
Sandy ga NP V
nani o katta ka
‘what’ ‘bought’
GF Subject Object
CS Q·(BUY(AGENT:SANDY:THEME:·))
(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?’
9.2. TYPES OF WH-QUESTIONS 335
The word cossa also means “what”, but behaves like a complex wh-phrase
for historical reasons, being derived from che cossa “what thing”.
4
Curiously, only the last name can be replaced by who. So we cannot have
(i) ∗ You think you just saw who Einstein?
336 9. A CONSTRUCTIONS
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?
9.2. TYPES OF WH-QUESTIONS 337
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.
NP NP VP
[WH]
Sandy V [e] where
what
put
GF Subject Object
CS Q·‚(PUT(AGENT:SANDY,THEME:·,LOC:‚))
SYNTAX S
.
.
.
.
XP
[WH]
CS Q·(…, ·, …)
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:‚))
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
you V [ei ]
saw
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.
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.
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]
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
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
said P [e]
to
CS-link
(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
346 9. A CONSTRUCTIONS
[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.
that NP VP
I V PP sentential subject
voted P [e]
for
CS-link
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)
348 9. A CONSTRUCTIONS
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
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.
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.
9.5.1.1. 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.
9.5. THE THEORY OF WH-MOVEMENT 351
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).
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 ___
Spec C
C0 IP
[WH]
NP I
you I0 VP
V NPi
saw whoi
[WH]
9.5. THE THEORY OF WH-MOVEMENT 353
(88) IP
NP I
[·NUM]
I0 VP
[·NUM]
Spec C
NPi C0 IP
[WH]
whoi NP I
[WH]
you I0 VP
V NPi
saw whoi
[WH]
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
356 9. A CONSTRUCTIONS
(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).
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)
(92) D-structure
movement
S-structure
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.
358 9. A CONSTRUCTIONS
(93) CP
Spec C
PRO C0 IP
[WH]
… XP …
PRO X
[WH]
…
X0
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).
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
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
(103) SYNTAX S
XP … [e] …
CS F(…, ·, …)
IS TOPIC·
(104) SYNTAX S
NP NP VP
John I V [e]
like
CS LIKE(EXP:ME,THEME:·)
IS TOPIC·
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
(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.
(107)
CP
Spec C
C0 IP
(that) NP NP I
Sandy I I0 VP
V0 NP
like Sandy
364 9. A CONSTRUCTIONS
(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]
from C0 . This proposal maintains the strong view that all A positions are specifiers
of functional heads.
366 9. A CONSTRUCTIONS
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.]
(113) Whoi did you say [CP that you were talking to t i ]?
(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.
368 9. A CONSTRUCTIONS
(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 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.
9.7. MORE ON CONSTRAINTS 369
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]
(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 ]]
15
We omit examples of wh-questions and relative clauses that demonstrate the
same point.
9.7. MORE ON CONSTRAINTS 371
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.
(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
9.8.1. Questions
(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]
(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]
376 9. A CONSTRUCTIONS
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
(139) SYNTAX NP
N0 S
what [e]
GF GFi
CS THING·(PROPERTY:F(…, ·, …))
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.
378 9. A 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.
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
9.8. OTHER A CONSTRUCTIONS 379
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.
9.8.3.1. Clefts
The cleft construction has the following informal description.
is
was
(145) it FOCUSi [CP that . . . t i . . . ]
will be
etc.
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.
380 9. A CONSTRUCTIONS
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.
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.
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
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.
We might expect each of these to undergo piedpiping, since each one con-
tains a wh-phrase. Some definitely can –
EXERCISES 383
∗
(3) A picture of which man are you looking for?
[Cf. Which man are you looking for a picture of ?]
(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?’
384 9. A CONSTRUCTIONS
∗
(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.]
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
386 9. A CONSTRUCTIONS
c. free relative
d. cleft
e. pseudo-cleft
[§9.8.]
Problems
[§9.1.]
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.]
4. Consider the fact that in general it is not possible to have a zero relative
when the subject is relativized.
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).
[§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.]
(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
390 9. A CONSTRUCTIONS
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.]
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
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.]
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.
14. Dutch has a complementizer dat that functions like English that.
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.
394 9. A CONSTRUCTIONS
∗
(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?
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
S Ni
S Ni
… applei … -nun ei
[§9.8.]
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.]
Research questions
accounted for? That is, where in the grammar of the language is the p-
stranding characteristic located?
[§9.2.]
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.]
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.]
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
This page intentionally left blank
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
GF Subject Object
CS KISS(AGENT:[MOTHER(MARY·)]‚,PATIENT:·)
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
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:„)
10.2. Binding
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.
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 .
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.
406 10. COREFERENCE AND BINDING
(17) An anaphor and its antecedent must be local with respect to one another.
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
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.
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.
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.
408 10. COREFERENCE AND BINDING
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!
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.
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.
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(·))
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.
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
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.
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.
10.4. BINDING IN CS AND SYNTACTIC STRUCTURE 413
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
(49)
IP2
NP I
I0 VP2
V0 CP
expects Spec C
government
C0 IP
· I
no government by expect I0 VP
414 10. COREFERENCE AND BINDING
himi
(51) Johni saw a snake near · [locative PP]
himselfi
(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 (‚, ·) )
(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.
416 10. COREFERENCE AND BINDING
Let us review some key examples with these definitions and assumptions
in mind. For
(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 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:·)
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
10.4. BINDING IN CS AND SYNTACTIC STRUCTURE 419
∗
(69) Maryi saw heri .
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.
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)
(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]
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
422 10. COREFERENCE AND BINDING
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).
SYNTAX N
GF Subject
CS X· ·
10.5. ∗ Reconstruction
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.
NP NP VP
herself Mary V PP
talks P [e]
GF Subject
CS TALK(AGENT:MARY·, GOAL:·)
IS TOPIC·
424 10. COREFERENCE AND BINDING
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.
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 .
426 10. COREFERENCE AND BINDING
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.
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 .
(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.
428 10. COREFERENCE AND BINDING
introduced into the structure until after movement. Hence condition C does
not apply.
10.7. Summary
Exercises
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.]
[§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.
430 10. COREFERENCE AND BINDING
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.]
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
The condition will have to mention the CS representation and the distribu-
tion of the reflexive pronoun.
[§10.2.]
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.
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.]
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.
[§10.4.]
[§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.]
434 10. COREFERENCE AND BINDING
Research questions
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.
[§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.
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
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
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.
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)).
(6) a. A: Harriet has been drinking something.
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.’
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)
11.1. BARE ARGUMENT ELLIPSIS 441
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)
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.
11.1. BARE ARGUMENT ELLIPSIS 443
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.
(6) a. A: Harriet has been drinking something.
b. B: Yeah, 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
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.
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.
11.1. BARE ARGUMENT ELLIPSIS 447
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.
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).
(6) a. A: Harriet has been drinking something.
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.
11.1. BARE ARGUMENT ELLIPSIS 449
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?
b. Do X anaphora:
so
that
Robin ate a bagel (on Thursday), and Leslie did the same thing
something similar
...
(on Friday).
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
(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))
11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS 453
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
(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]]]
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
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
11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS 457
(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
SYNTAX S
NP AUX VP
Sandy can’t V NP
speak German
Fragment
SYNTAX S
NP AUX
Leslie can
(48)
Antecedent: NOT(ABLE(SPEAK(AGENT:SANDY,THEME:GERMAN))
Substitute: ABLE ROBIN
Result: ABLE(SPEAK(AGENT:ROBIN,THEME:GERMAN))
Here, Robin and Sandy form a match, and will never and WILL do. The
match might be analyzed as
(51)
Antecedent: FUTURE (NEVER(SPEAK(AGENT:SANDY,THEME:GERMAN))
Substitute: EMPH(FUTURE Ø ROBIN
Result: EMPH(FUTURE(SPEAK(AGENT:ROBIN,THEME:GERMAN))
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
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.
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.
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
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.
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.
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
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
Problems
∗
1. Consider the following exchange.
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
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: 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.
[§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
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
RESEARCH QUESTIONS 469
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.
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.]
11.1. 1, 2 1, 2, 3
11.2. 3 4, 5 1, 2
11.3. 4 6 3, 4
Glossary
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.
BD, 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
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
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