X Syntax:

A Study of
Phrase Structure
Linguistic Inquiry Monographs
Samuel Jay Keyser, general editor

1. Word Formation in Generative Grammar

Mark Aronoff
2. X Syntax: A Study of Phrase Structure
Ray Jackendoff
3. Recent Transformational Studies in European Language
S. Jay Keyser, editor
4. Studies in Abstract Phonology
Edmund Gussmann
X Syntax:
A Study of
Phrase Structure

Ray Jackendoff

The MIT Press, Cambridge, Massachusetts,

and London, England
 \
,<z^ S)

Fifth printing, 1991

© 1977 by The Massachusetts Institute of Technology

All rights reserved. No part of this book may be reproduced

in any form or by any means, electronic or mechanical, in¬
cluding photocopying, recording, or by any information stor¬
age and retrieval system, without permission in writing from
the publisher.

Printed and bound in the United States of America

Library of Congress catalog card number 81-80193

ISBN 0-262-60009-9
 Foreword

We are pleased to present this monograph as the second in the series Linguistic Inquiry
Monographs. These monographs will present new and original research beyond the
scope of the article. Because of their originality it is hoped they will benefit our field by
bringing to it perspectives that will stimulate further research and insight.

Samuel Jay Keyser

For Elise
 Contents

Preface. xi

1. Preliminaries . 1
1.1. Goals. 1
1.2. Assumptions about the Structure of Linguistic Theory. 2
1.3. Lexicalism and Interpretivism . 5

2. Motivation for the X Convention . 7

2.1. Chomsky's Arguments. 7
2.2. Adjectives and Nouns . 18
2.3. Adjectives and Adverbs. 23
2.4. Coming Attractions. 26

3. A Theory of Phrase Structure . 29

3.1. Reformulation of the X Convention . 29
3.2. The Syntactic Distinctive Features. 31
3.3. The Phrase Structure Rule Schema. 33
3.4. The Generalized Subject Relation. 37
3.4.1. Modifying Chomsky’s Analysis. 38
3.4.2. Arguments for the Uniform-Level Subject Relation. 42
3.4.3. Treatment of the Complementizer. 45
3.5. The Auxiliary. 47
3.6. Two Exceptions to the Uniform Three-Level Hypothesis. 50
3.7. Summary and Generalization of Rules. 53
3.8. Glossary of Lexical and Syntactic Categories. 56

4. Complements. 57
4.1. The Three Levels of Complements. 57
4.2. The V' and N' Complements . 64
4.2.1. Easy Parts of V'. 64
4.2.2. Particles and Predicates. 66
4.2.3. The N' Complement. 69
4.2.4. Generalization of V' and N' and the Definition of Object. 71
4.3. The V" and N" Complements . 72
4.4. Complements of [ — Subj] Categories. 76
4.4.1. Adjectives. 76
 CONTENTS

4.4.2. Adverbs. 78
4.4.3. Prepositions. 78
4.4.4. Particles. 79
4.4.5. Case Markers and Prepositions. 80
4.5. Generalization of Phrase Structure Rules. 81
4.6. Summary. 85
4.7. Appendix: The Passive and Related Rules. 86
4.7.1. NP-Preposing . 86
4.7.2. Agent-Postposing. 88
4.7.3. NP-Postposing . 90
4.7.4. The Determiner . 92
4.8. Appendix: Arguments for Intraposition. 95
4.9. List of Transformations. 101

5. NP Specifiers. 103
5.1. Introduction . 103
5.2. Articles and Quantifiers. 103
5.2.1. Semantic Roles of NP Specifiers . 103
5.2.2. Phrase Structure Rules. 104
5.2.3. We and You. 106
5.3. The Partitive. 106
5.3.1. Partitives as N' Complements. 107
5.3.2. Position of the Quantifier in Partitive Constructions . 110
5.3.3. The Partitive Constraint. Ill
5.3.4. Substantivization. 114
5.4. Pseudopartitives. 119
5.4.1. Arguments for the Construction. 119
5.4.2. Pseudopartitive Specifier Constraints . 123
5.5. Numerals. 126
5.5.1. Comparison to Quantifiers and Group Nouns. 126
5.5.2. Numerals as Nouns. 128
5.5.3 Three Adjustment Rules . 130
5.6. Summary. 133

6. Specifiers of X"' . 137

6.1. Introduction . 137
6.2. Measure Phrases. 137
6.3. Quantifier Phrases in X".141
6.4. Degree Phrases . 143
6.4.1. Attachment of Deg'" to Q"' . 144
6.4.2. The Degree System of Adjectives and Adverbs . 146
 CONTENTS

6.4.3. Comparative Specifiers. 147

6.4.4. Enough. 150
6.4.5. Summary of Degree Words. 156
6.5. Recursion in Degree Phrases . 157
6.6. Summary of Rules. 165

7. Relative Clauses. 169

7.1. Introduction . 169
7.2. Appositives. 171
7.3. Interpretation of the Appositive Relative Pronoun. 173
7.4. Constraints on the Determiner and the Head in Restrictives. 177
7.5. Restrictive Relatives on Quantified NPs . 182
7.6. Stacking of Restrictive Relatives. 185
7.7. Conjoined Heads. 190
7.8. Interpretation of the Restrictive Relative Pronoun. 194
7.9. Appendix: Against the Conjoined Source for Appositives. 197

8. Degree Clauses. 201

8.1. Introduction . 201
8.2. The Extraposition Theory and its Evidence for the X Convention .... 201
8.3. Arguments against Extraposition of Degree Clauses. 207
8.4. A Generalization in the X"-Complement Theory . 209
8.5. The Location of the Binding Degree Word. 211
8.6. Why an Interpretive Theory Is Needed. 214
8.7. Interpreting PRO. 218
8.8. Recapitulation . 219

9. Deverbalizing Rules. 221

9.1. The Deverbalizing Rule Schema . 221
9.2. Gerundive Nominals. 222
9.3. Tensed Complements and Free Relatives . 224
9.4. Gerundive PPs. 226
9.5. Passive VPs . 232
9.6. Final Remarks . 234

10. More General Implications. 239

10.1 Explanatory Adequacy . 239
10.2. The Challenge for Semantics . 241

References 245
 Preface

In the fall of 1967, Noam Chomsky gave the series of lectures at MIT in which he
introduced the Lexicalist Hypothesis, one of the first suggestions that the expressive
power of a generative grammar is not concentrated primarily in its transformations.
Many of us who heard those lectures found the new theory tremendously exciting, and
we looked forward to its impact on the study of syntax.
Its immediate effect was more one of spirit than of substance. "Remarks on
Nominalization”, the written form of the lectures, was the first authoritative stand
against the growing school of Generative Semantics; "lexicalist" came to be a term
describing anyone who wanted to retain deep structure in transformational grammar, or
who opposed transformational derivation of lexical items, or who proposed surface
structure rules of semantic interpretation. "Remarks" and Chomsky’s next major
paper, "Deep Structure, Surface Structure, and Semantic Interpretation", became
rallying points for a lexicalist-interpretivist school, mostly consisting of students of
Chomsky’s from 1965 and later.
But although "Remarks” had a great effect on the approach to syntactic descrip¬
tion, the most important issues it raised received surprisingly little attention for several
years. Chomsky’s arguments against the transformational derivation of lexical items
and for the cross-categorial application of transformations were widely appreciated and
extended. But his proposals on the organization of the lexicon and the theory of phrase
structure, although intriguing, were sketchy and programmatic, and this seems to have
discouraged further investigation. Many of us, of course, had inherited a misplaced bias
against phrase structure from the early work in generative grammar, which had little of
interest to say about phrase structure grammar, stressing only its insufficiency for
linguistic description; this prejudice too stood in the way of serious research. More
glamorous issues were demanding attention as well: in the late sixties, the relation of
syntax to semantics was the main arena; in the seventies, constraints on the transfor¬
mational component.
In the past few years, though, the structure of the lexicon has come to be of
growing concern, as phonology, syntax, and semantics have all put pressure on it to
solve their problems. And the gradual reduction in the power of transformations has led
syntactic theory, like other aspects of our lives these days, towards simpler ways—in
this case back to phrase structure. Thus the scope of Chomsky’s original proposals is
only just now beginning to be appreciated.
My original conception of this monograph was as a general overview of the
XII PREFACE

Lexicalist Hypothesis and of its implications for all components of the grammar. But as
I got involved in working it out, the importance of phrase structure stood out more and
more strongly, until it became clear to me that I would have to devote the entire book
to it. And still it kept growing, until the original four or five chapters became the
present ten, taxing my patience to a degree I hope I have spared the reader.
In its present form, the first two chapters of the book introduce the motivations for
a richer account of the base. Chapter 3 is the conceptual core of the book: it elaborates
Chomsky’s sketches in “Remarks” into a highly constrained and explicit theory of
constituent structure. The remainder of the book justifies the theory and works out its
consequences through a detailed examination of English phrase structure, giving a
much more wide-ranging and systematic survey of this aspect of the language than has
to my knowledge appeared in recent years. Thus the book is conceived of as a
contribution both to the study of English syntax and, because of the implications of
English for universal grammar, to linguistic theory as a whole.
As is usual for a book that has been so long in preparation, parts of it have
appeared in various forms elsewhere. The most rudimentary version was a paper called
“Speculations on Presentences and Determiners”, published by Indiana University
Linguistics Club and now mercifully out of print. “Introduction to the X Convention”,
also published by IULC, was the real germ of the book; readers familiar with that paper
will find passages from it scattered throughout chapters 2, 3, and 4, though they will
notice that I have since adopted three levels of bars or primes in major phrasal
categories rather than two. A paper published in Language, “Morphological and
Semantic Regularities in the Lexicon”, was originally intended as a chapter of this
book, but it was crowded out by the insidiously encroaching demands of the phrase
structure component. Finally, a paper entitled “Constraints on Phrase Structure
Rules”, consisting mostly of excerpts from chapters 3, 4, and 9, is appearing in Studies
in Formal Syntax, edited by Akmajian, Culicover, and Wasow (1977).
This monograph has benefited from the advice of numerous colleagues and
students over the years. My discussions on this material with Adrian Akmajian, John
Bowers, Joan Bresnan, Noam Chomsky, Peter Culicover, Joe Emonds, Jay Keyser,
Joan Maling, Ann Reed, Lisa Selkirk, and Henk van Riemsdijk have been particularly
fruitful, and I am grateful for their suggestions and encouragement. I especially want to
thank Akmajian and Culicover, in whose homes I precipitously arrived as a houseguest
in the summer of 1975 when chapters 6 and 7 were at white heat, for their patience and
forbearance in the face of my obsessive enthusiasm.
I also need to express my gratitude to my wife Elise, who, though she arrived on
the scene after the book was already well along, quickly came to an understanding of
what I found fascinating in this stuff and found herself beginning to care that I cared
about it. I have an overwhelming amount to thank her for, perhaps not least of which is
her sobering comment on the whole enterprise: “Impressive, but I'd rather read
Newsweek. ...”

Brandeis University, August 1977

X Syntax:
A Study of
Phrase Structure
 1: Preliminaries

1.1. Goals

Like all work in generative grammar, this study is concerned with describing what it is
that a human being knows when he knows how to speak a language. In particular, we
take it to be essential to separate those aspects of a speaker’s knowledge that are
universal from those that are language-particular, under the hypothesis that the
language-particular features must be learned by the speaker, but the universal parts
may be innate, i.e. determined by the structure of the human organism itself. Success
in separating universal from language-particular components enables us to make
interesting claims about the nature of the mind.
Given the fact that children learn languages in a relatively short period of time and
from rather fragmentary data, we would prefer a theory of linguistic structure in which
the job of the language learner is as easy as possible, that is, in which the universal
component is maximized. The extreme case, of course, would be if the entire language
were innate; but this hypothesis is clearly false, since children learn different languages
depending on the linguistic environment in which they are raised. Therefore, the theory
must strike the proper empirical balance between the diversity of human languages and
the need for them to be leamable.
This vision of the goals of linguistic theory originates in the work of Chomsky; it is
explicated in great detail in many of his writings, particularly Aspects of the Theory of
Syntax. What distinguishes the school of generative-transformational grammar from
other extant linguistic theories is the emphasis given to these goals at every level of
linguistic description, from the high-level formulation of general theory to the most
intricate low-level details of particular languages. The present study, it is hoped, is no
exception.
The particular area of inquiry we will address here is the theory of phrase
structure. Traditional grammar is concerned with defining parts of speech and constitu¬
ent structure, with such concepts as the grammatical relations subject and object, and
with the relations between modifiers and heads. The present study is an attempt to deal
with these traditional issues in the framework provided by the goals and methodology
of generative-transformational grammar.
7 RAY JACKENDOFF

The major questions we will be attempting to answer are these: What are the
lexical categories (parts of speech) available to linguistic theory? How are they
associated with syntactic structures, and what is the relationship between lexical
categories and the syntactic constructions in which they take part? What relationships
exist among lexical categories and how can they be expressed? What relationships exist
among larger syntactic categories and how can they be expressed?
These questions will be dealt with in the context of an attempt to describe the
grammar of English. Of course, in some respects, this choice of data severely limits the
range of possible linguistic phenomena to be used as input in formulating the theory. In
taking this approach rather than a relatively superficial survey of the grammars of many
languages, I am presupposing two methodological biases: first, that it is only by
investigating a system very thoroughly that one can begin to understand it; second, that
evidence from a single language can in fact provide substantial evidence for linguistic
universal. These biases are of course not empirical claims, but only strategies for
arriving at empirical claims. No matter how a claim is arrived at, it must of course be
tested against a great many details of a great many languages.

1.2. Assumptions about the Structure of Linguistic Theory

By the time of Aspects of the Theory of Syntax (1965), it was generally agreed that
linguistic theory must provide five distinct components, each with its own characteris¬
tic types of rules and characteristic interactions with the other components. In other
words, the organization of the grammar into components, the rule types, and their
interactions are not something that the child must learn in learning a particular
language, but rather are part of his hypothesis about what he is to learn. The field has
seen many disputes as to exactly the nature of these components, what aspects of the
language each accounts for, how the rules are formalized and constrained, and how the
rules interact within components and across components; the settling of these questions
is an empirical matter. But the general separation of the theory into components will be
taken for granted here.
The five components of the grammar are the lexicon, the categorial or phrase
structure component, the transformational component, the phonological component,
and the semantic component. The first two are often considered subcomponents of a
larger component called the base component; the first three, or the second and the
third, are often lumped together as the syntactic component. Because of the concern of
this study with the nature of phrase structure, we will be dealing primarily with the
form of the categorial component. We will make no attempt to discuss the other
components at all exhaustively. But since any decision about the form of the base is
nearly always intimately tied up with the other components, particularly the transfor¬
mations and the semantics, we will feel no obligation to ignore them.
The lexicon is the repository of idiosyncratic information about individual forma-
tives of the language. Each formative is assigned a lexical entry which describes its
 PRELIMINARIES

phonological, semantic, and syntactic properties. The lexicon also contains a class of
rules called lexical redundancy rules, which may include word formation rules,
morpheme structure rules, and rules of allomorphy.
The categorial component includes a set of context-free phrase structure rules,
which define the hierarchical syntactic categories of the language and simultaneously
determine the left-to-right order of the immediate constituents of each category. The
phrase structure rules generate a set of phrase markers whose initial symbol is S and
whose terminal symbols are the lexical categories of the language.
The base also includes a lexical insertion rule which substitutes lexical formatives
for terminal symbols in the phrase markers generated by the categorial component. In
the Aspects theory, lexical insertion is constrained by the strict subcategorization and
selectional restrictions of the heads of phrases. We will assume here a less restricted
lexical insertion rule which checks only the lexical category and the strict subcategori¬
zation restrictions; selectional restrictions are regarded as well-formedness conditions
on semantic interpretations, as argued in Jackendoff (1972, chapter 1). We will also
assume that terminal symbols can be left unfilled by lexical items, resulting in an
“empty node” in the sense of Emonds (1970); and that a terminal symbol may
dominate a phonologically null item PRO, used as an anaphoric element in elliptical
constructions.1 An empty node must be filled at some point in the derivation in order
for the derivation to be well-formed; PRO must receive a reading by the rules of
anaphora or other specific rules of interpretation, in order for the semantic representa¬
tion to be well-formed. The phrase markers with lexical formatives generated by the
base component of a language L constitute the possible deep structures of sentences of
L.2
The deep structures are taken as input to the transformational component, which
relates them to the surface structures of L, and thence via the phonological component
to the phonetic representations of L.3 We will assume here that the transformations of
L are extrinsically ordered and that their application is governed by the principle of the
transformational cycle, as follows: certain syntactic categories are designated by
linguistic theory as cyclic nodes. In the Standard Theory of Aspects, the cyclic nodes
are the S nodes; it is now generally acknowledged that at least NPs are also cyclic
nodes. In applying the transformations to a sentence, apply in order Tj, . . . , Tn first to
the structures dominated by the most deeply embedded cyclic nodes, that is, to those

1 PRO takes over the function of the empty node A used in Jackendoff (1972), which did not have the
same meaning as Emonds’s empty node. However, Wasow (1975) argues that they need not be distinct
entities.
2 It should be clear from these assumptions that I will not be concerned with theories of transforma¬
tional grammar which use unordered base strings or posttransformational lexical insertion. Nor will I concern
myself here with providing explicit arguments against such theories. However, the analysis arrived at here
constitutes an implicit challenge to their proponents, in that it deals with a range of facts and issues generally
ignored in discussion of the significance of deep structure in generative grammar.
3 There has been considerable evidence that earlier levels of syntactic derivation also serve as input to
the phonology. See Bresnan (1971), for example.
4 RAY JACKENDOFF

which do not dominate any other cyclic node. After T,, . . . , T„ have applied to these
domains, they apply again in order to the structures exhaustively dominated by cyclic
nodes, all cyclic nodes dominated by which have already governed an application of the
cycle. The cycle is repeated under the same conditions until the highest S node is
reached.
We will assume furthermore the principle of the strict cycle (Chomsky (1973, 243)):
“No rule can apply to a domain dominated by a cyclic node A in such a way as to
affect solely a proper subdomain of A dominated by a node B which is also a cyclic
node.’’ Without this convention, it would be easy to circumvent the ordering of
transformations within embedded contexts.
We will take it for granted that there is a class of transformations which take place
on the final (highest) cycle only. In the best possible case, the last-cyclic transforma¬
tions have the same conventions on application as the cyclic transformations, i.e. they
apply in the highest clause only and cannot affect only a subordinate cyclic domain.
However, last-cyclic transformations which violate this condition have appeared in the
literature, and I will leave the issue open.
It has often been suggested that there is also a class of precyclic transformations,
which naturally must have different conventions of application than the cyclic rules.
However, no convincing arguments for such rules have appeared within the particular
variant of transformational grammar being assumed here; clearly the theory would be
more highly constrained if they were unnecessary.
In the Standard Theory of Aspects, the semantic component is taken to be an
interpretive component which derives semantic representations from the deep struc¬
tures of L. However, if taken literally, this assumption turns out to lead to the
Generative Semantics position that deep structures and semantic representations are
isomorphic, a position advocated by Lakoff (1971a), McCawley (1971), and others.
Since in this theory the form of underlying phrase markers must reflect purely semantic
considerations, all the aspects of syntactic generality in a language expressible by the
phrase-structure rules of the Standard Theory are lost.
Alternatively, one can give up the assumption that deep structure is the only
syntactic source of semantic information, adopting instead what Chomsky has called
the “Extended Standard Theory”. Arguments for this modification to the Standard
Theory appear in Chomsky (1970b) and Jackendoff (1972), among others, and I will not
go into those arguments here. In the Extended Standard Theory, various aspects of the
interpretation of a sentence are derived from its deep structure, from its surface
structure, and possibly from its end-of-cycle derived structures. Although this theory,
with its variety of rule types and rather complex organization, is less immediately
attractive than the Standard Theory or Generative Semantics, it appears to be rather
tightly constrained, in that the formal universals organizing the grammar are considera¬
ble.
A recent alternative to the Extended Standard Theory has been provided by the
 PRELIMINARIES

"trace theory" of movement rules, developed in Chomsky (1973; 1975), Wasow (1971),
and Fiengo (1974). In this theory, all movements of constituents leave behind anaphoric
"traces" which serve to mark the previous position of the moved constituent. Certain
very general constraints on movement rules follow immediately from independently
motivated conditions on anaphora which are imposed on the traces. While the
consequences of this theory for the transformational component are still not well
understood, it has been pointed out by John Goldsmith that the trace theory permits all
semantic interpretation to take place at the level of surface structure. Jackendoff
(1975b) shows that this property permits a number of disparities in the semantic
component of the Extended Standard Theory to be resolved. Furthermore, it restores
to the semantic component the conceptual simplicity of the Standard Theory. We will
leave open which version of the Extended Standard Theory is to be preferred, as it is
largely irrelevant to our concern with phrase structure.

1.3. Lexicalism and Interpretivism

Chomsky’s paper "Remarks on Nominalization” (1970a, dating originally from 1967)

contrasts the (at that time) usual transformational derivation of nominals such as the
derivation of nominals from sentences such as Someone derives nominals with the
"Lexicalist Hypothesis”, in which nominals are generated directly by the base
component. Among the consequences of the Lexicalist Hypothesis is the claim that the
theory of grammar must include a way to refer to more than one syntactic category,
using a single term of the structural description of a grammatical rule. Thus the
Lexicalist Hypothesis not only changes the nature of the relations among formatives
such as derive and derivation, but also has profound consequences for the role of
syntactic categories in linguistic theory. Chapter 2 reviews Chomsky’s evidence for the
Lexicalist Hypothesis and adds further evidence; subsequent chapters explore its
consequences for the categorial component of the grammar.
It is perhaps important to emphasize the logical independence of the Extended
Standard Theory and the Lexicalist Hypothesis, since these positions have been
identified by and large with the same individuals and since both have been set in
opposition to Generative Semantics. There is no a priori reason why the existence of
surface structure semantic interpretation has any connection with the choice of a
lexical or transformational source for derived nominals. One could well imagine a
lexicalist version of the Standard Theory or a transformationalist version of the
Extended Standard Theory. In practice, however, the two hypotheses turn out to
supplement each other, for what I think are two separate reasons.
The first reason is methodological. The proponents of Generative Semantics have
attempted to constrain the theory of grammar by restricting the number of rule types,
categories, and derivational levels; and they are willing to accept as the price of this
constraint a high degree of abstractness and complexity in syntactic derivations. Those
6 RAY JACKENDOFF

who have adopted the lexicalist and surface structure interpretivist positions, however,
are willing to accept a wider range of rule types, categories, and rule interactions within
the theory of grammar, provided each component can be sufficiently constrained. They
often are rather skeptical about the ability of traditional logic to provide a model of any
part of semantic representation. This methodology turns out to be congenial both to the
Lexicalist Hypothesis and to the Extended Standard Theory of semantic interpretation.
But there are empirically based ties between the two hypotheses as well. For
instance, if one wants to maintain that the interpretation of sentences containing
negation is determined solely from their deep structure, one must permit transforma¬
tions which drastically alter the shape of lexical formatives; Klima (1964) posits such
changes as neg + any —> no, neg 4- often —* seldom, and neg + many —> few. Though
one could maintain a narrow Lexicalist Hypothesis for nominalizations and still admit
such derivations, they certainly violate the spirit of the Lexicalist Hypothesis, in that
they account for what are at best semiproductive lexical processes by means of a
transformation. Only by adopting a surface structure interpretive theory of negation
(e.g. Jackendoff 1969a; 1972, chapter 8) can we maintain a lexicalist theory of negative
morphemes. Hence surface structure interpretation is a necessary condition for a
desirable extension of the Lexicalist Hypothesis; the two theories thus complement
each other empirically as well as methodologically.
 2: Motivation for the X
Convention

This chapter provides the background against which the rest of this study is set.
Section 2.1 is concerned with Chomsky’s treatment of the relation between verbs and
their nominalizations; section 2.2 with adjectives and their nominalizations; and section
2.3 with adjectives and adverbs. Our approach is to show that none of these
relationships among lexical items can be accounted for transformationally. The task of
expressing these relationships must therefore fall on other components of the grammar.
One kind of relationship which it becomes necessary for the grammar to express is
cross-category generalization. We recapitulate Chomsky’s development of the X
Convention as a way to incorporate cross-category generalization, and we show how
this leads to a theory of the base.

2.1. Chomsky’s Arguments

Chomsky’s paper “Remarks on Nominalization’’ has as its most immediate concern the
syntax of derived nominals such as John’s criticism of the book and John’s refusal of
the offer. The point of departure is the considerable number of differences between
derived nominals and “gerundive nominals” such as John’s criticizing the book and
John’s refusing the offer. Basically, gerundive nominals behave like regular transforms
of sentences, but derived nominals do not.
Here are some of the differences. Gerundive nominals occur with aspectual verbs,
but derived nominals do not.

(2.1) John’s having criticized the book

Byrne’s having been refusing the offer just when Nixon arrived
(2.2) *John’s have(ing) criticism of the book
*John’s have(ing) been refusal of the offer

Gerundive nominals use adverbs before the gerund; derived nominals use adjectives
before the nominalization.

(2.3) John’s sarcastically criticizing the book

John’s emphatically refusing the offer
8 RAY JACKENDOFF

(2.4) John’s sarcastic criticism of the book

John’s emphatic refusal of the offer

Derived nominals can take part in partitive expressions and can be pluralized;
gerundive nominals cannot.

(2.5) *some of John’s criticizing the book

*John’s many refusings the offer
(2.6) some of John’s criticisms of the book
John’s many refusals of the offer

But gerundive nominals can take a range of adverbials that are prohibited in derived
nominals.

(2.7) John’s criticizing the book too often

John’s refusing the offer in a suspicious manner
(2.8) *John’s criticism of the book too often
*John’s refusal of the offer in a suspicious manner

Gerundive nominals, but not derived nominals, can undergo Subject Raising, Dative
Shift, and Particle Movement.

Subject Raising

(2.9) its being certain that John will win

John’s being certain to win
Dative Shift

John’s giving the book to Bill =>

John’s giving Bill a book
Particle Movement

John’s looking up the information

John’s looking the information up
(2.10) the certainty that John will win
*John’s certainty to win

John’s gift of a book to Bill

*John’s gift of Bill a book

John’s looking up of the information d£>

*John’s looking of the information up

Derived nominals may use the in the absence of a subject, while gerundives may not;
compare (2.10) with *the being certain that John will win.
The properties of gerundive nominals are exactly what one would expect if they
were transforms of sentences. But transformations which predict this behavior in
 2: MOTIVATION FOR THE X CONVENTION

producing derived nominals from sentences will necessarily be rather difficult to state
and order.1
A second kind of data distinguishing gerundive from derived nominals concerns
syntactic productivity and regularity. While all sentences (except those whose subjects
are too complex to have a possessive form) have gerundive nominals, many, for
example those in (2.11), have no derived nominals.

(2.11) It is easy to please John.

John amused the children with his stories.
It seems that John is gone.
(2.12) its being easy to please John
John’s amusing the children with his stories
its seeming that John is gone
(2.13) *the easiness to please John
*John’s amusement of the children with his stories
*its seem that John is gone

Conversely, many derived nominals do not correspond precisely to well-formed

sentences.

(2.14) *John doubted about their proposal.

*It is probable John’s leaving.
*John advised to Bill.
(2.15) ^John’s doubting about their proposal
*its being probable John’s leaving
*John’s advising to Bill
(2.16) John’s doubts about their proposal
the probability of John’s leaving
John’s advice to Bill

Transformations which perform all the minor syntactic adjustments necessary to

produce (2.16) from well-formed sentences will be numerous and highly idiosyncratic, if
they can be stated at all.
A third argument concerns semantic regularity. Derived nominals typically have
meanings rather idiosyncratically related to the parallel verb, whereas gerundive
nominals have absolutely predictable semantic relationships with the verb. Chomsky
mentions examples such as laughter, marriage, construction, actions, activities,
revolution, belief, doubt, conversion, permutation, trial, residence, qualifications,
specifications (p. 19/189),2 intelligence, deed, and writings (note 10, p. 19/note 11, p.

1 In particular, Ross (1967) argues from the existence of a transformation producing derived nominals to
the conclusion that Dative Shift and Particle Movement are last cyclic: if they were cyclic, they would have
no way of knowing that the sentence they operate in was to be turned into a derived nominal on the next
cycle. Jackendoff (1972, chapter 9) argues that this conclusion leads to an ordering paradox, for Dative Shift
must take place in cyclic derivations such as John is believed by Harry to have been given a book by Bob.
21 will give the page references on Chomsky’s Studies on Semantics in Generative Grammar followed
by those in Jacobs and Rosenbaum (1970).
10 RAY JACKENDOFF

217). Other examples are easy to find. The point of these examples is that a
transformational derivation of at least some readings of these words, from sentences
involving the verbs laugh, marry, etc., must involve the presence of semantic material
idiosyncratic to the particular nominalization. For example, John's deeds does not
mean ‘things which John did’ but rather at least ‘fairly significant things which John
did’. The derivation of the political sense of revolution from revolve is fairly obscure;
similarly for the sense of construction in the verb-particle construction in English. In
short, a transformational derivation of these nominals entails an incredible proliferation
of idiosyncratic transformations.
Finally, there are a great number of nouns which occur in structures similar to
those observed in derived nominals, but for which there is no corresponding verb at all.
Chomsky cites constructions like the weather in England, the story of Bill's exploits,
the message from Bill to Tom about the meeting, a war of aggression against France,
John's attitude of defiance towards Bill, his habit of interrupting (p. 30/196), the author
of the book, the assistant vice-chancellor of the University (p. 32/196), and John's
poems/sonnets/Alexandrines (about Homer) (note 10, p. 20/note 11, p. 217). A
consistent transformationalist approach must derive these nouns from verbs too.
Following such an approach, Lakoff (1971b) comes to derive king from a verb by
analogy with ruler; he then invokes the notion of “positive absolute exception” to the
nominalization rule, in order to guarantee that the verb underlying king never appears
at the surface. Such a solution is indicative of the unnaturalness of the transformational
theory.
To describe the differences between gerundive and derived nominals in a system¬
atic way, Chomsky claims that gerundives have an underlying structure like (2.17),
containing an S,3 but that derived nominals have an underlying structure like (2.18), in
which the head is a noun and no S node appears.

(2.17)

3 We will propose a somewhat different structure in section 3.6. but for the moment we adopt
Chomsky’s formulation.
 2: MOTIVATION FOR THE X CONVENTION

(2.18) NP

In order to propose a structure like (2.18), in which criticism is a deep-structure

noun, we must provide a mechanism in the lexicon to show the relation between
criticism and the verb criticize. Chomsky proposes that criticize and criticism form a
single lexical entry, unmarked for the syntactic feature differentiating nouns from
verbs. When inserted under a verb node, this entry will direct the use of the
phonological form criticize; when inserted under a noun node, it will be realized as
criticism. In an alternative formalism argued for in Jackendoff (1975b), criticize and
criticism are separate lexical entries whose relationship is specified by a lexical
redundancy rule.
Whichever formalism one chooses, there is an important innovation in this theory:
the relations among surface lexical formatives can be expressed by other than
transformational means. As Chomsky points out (p. 17/188), early work on transforma¬
tional grammar had no other means but transformations to relate lexical items; hence
the transformational position on nominalizations was the only possible theory. But with
the introduction of a lexicon and syntactic features into linguistic theory, as in Aspects,
a lexicalist position could be formulated, in which these regularities were expressed in
the lexicon. As the evidence seems to make the transformationalist position appear
problematic, it behooves us to explore the consequences of the lexicalist position.
As a first consequence, we arrive at immediate explanations for the syntactic
differences between gerundive and derived nominals pointed out above. Gerundives,
being sentences in deep structure, are expected to contain aspect; since -ing substitutes
for Tense and Modal, their absence is accounted for. Derived nominals, being NPs, will
not contain auxiliary elements. Likewise, it is typical for sentences to exhibit the
preverbal adverbs which appear in gerundive nominals, and typical for NPs to exhibit
the prenominal adjectives which appear in derived nominals. The verb of a sentence
cannot be pluralized or quantified, but the noun of an NP may; hence the contrast
between gerundive and derived nominals of (2.5)-(2.6). The time and manner adverbi-
als of gerundives are typical of sentences but not of NPs, so derived nominals do not
12 RAY JACKENDOFF

permit them. We expect Subject Raising, Dative Shift, and Particle Movement in Ss but
not in NPs, hence the differences in (2.9)—(2.10). Only NPs are expected to use the
definite article the, so only derived nominals can use it.
The differences in productivity and regularity also make more sense within the
Lexicalist Hypothesis. If gerundives are simple transforms of sentences, we correctly
expect unfailing productivity of forms and a systematic semantic relationship to
sentential forms. On the other hand, lexical relationships are notoriously irregular and
idiosyncratic, and it is therefore not at all surprising that derived nominals should be as
erratic as they are. Furthermore, if derived nominals are nouns, we would expect (as
observed) that there are NPs of the same structure as derived nominals, not related in
any way to sentences, simply because there happen to be no verbs lexically related to
their head nouns.
Thus the Lexicalist Hypothesis makes a principled differentiation between the two
types of nominalizations, explaining their differences as an automatic consequence of
their structures. Given the proposed structures, the properties of the two types of
nominalizations could not be too much different than they are without complicating the
grammar. On the other hand, the Transformationalist Hypothesis must resort to a great
deal of descriptive complexity to yield the observed properties of derived nominals.
These properties are thus taken to be accidental, in that there is no logical connection
between them, and in that the grammar would be much simpler if in fact the differences
between gerundive and derived nominals did not exist. Stated differently, in the
Transformationalist Hypothesis, the language learner must learn a large number of
related transformations to produce derived nominals; but in the Lexicalist Hypothesis,
the language learner needs to learn only the differences in deep structures (2.17) and
(2.18) in order to produce all the rest of the differences from independently learned
rules.
To make this argument fully convincing, we must of course show that all the rules
applying to derived nominals in the lexicalist account are independently motivated. To
begin with, examine the alleged base form (2.18). It has often been claimed that the
possessive form in John's house, for example, is derived by reduction from a deep
structure relative clause construction such as the house which John has. Under this
theory, possessive NPs do not appear in the base. However, (2.18) claims that the NP
John occupies prenominal position in the base form, so the base must apparently be
enriched to accommodate this structure. Chomsky shows, however, (pp. 37-38/200-
201) that this enrichment of the base is independently necessary: the “inalienable”
sense of John's leg, in which the leg is part of John’s body rather than a wooden leg he
happens to own, cannot be paraphrased by the leg that John has. For example, we
cannot say The leg that John has hurts meaning John's leg hurts. Further, expressions
such as John’s forte have only an inalienable sense; *the forte which John has is
nonsensical. This suggests that at least the inalienable sense of possession is not
derived from a reduced relative clause, but rather is generated in the base.
Further evidence comes from phrases like John’s book and John’s proof of the
 2: MOTIVATION FOR THE X CONVENTION

theorem. Here appropriate relative clause paraphrases are the book which John wrote
and the proof of the theorem that John discovered. Insofar as the relative clause
describing John’s relationship to the object is dependent on the nature of the object and
is often still open to variation, a transformation reducing the relative clause to a
possessive will violate the recoverability requirement on deletion transformations.
Suppose, however, that the possessives are generated in the base. One projection rule
for the possessive position can specify a rather loose notion of “intrinsic connection’’
between the possessive NP and the object denoted by the larger NP. This notion would
be sharpened by the semantic nature of the larger NP: if it is written material, intrinsic
connection denotes the writer; if it is an idea, intrinsic connection denotes the
discoverer; and so forth. Placing the burden of specifying intrinsic connection on the
semantic component (or preferably on real-world knowledge) instead of on the syntax
eliminates the violation of recoverability. It simultaneously explains the creativity in
the use of intrinsic connection: for example, John's chair may denote the chair that
John owns (alienable possession), or, by intrinsic connection, the chair that John built,
designed, or habitually sat in.4 Of course, a complete account must be much clearer
about the nature of intrinsic connection that I can be here. My conjecture, however, is
that it is a rather fundamental cognitive function, of which the examples above are
simply special cases.
Thus there is some justification for including the possessive position in the base,
independently of its use in derived nominals. What about the prepositional phrase
following the head noun in (2.18)? Again, it is often claimed that PPs in this position are
reduced relative clauses, and so this position is unnecessary in the base. For example,
a book about Harry and a picture of the lizard are easily derived from a book which is
about Harry and a picture which is of the lizard. Furthermore, the reduction rule has
good independent motivation, since it provides a straightforward source for such
phrases as the child sleeping in the alley and a book yellow with age. Chomsky points
out, however, that at least some PPs in NPs cannot be derived from plausible relative
clauses, for example the weather in England, the weather in 1965, the author of the
book, his attitude of defiance, his advantage over his rivals, the reason for his refusal,
and his habit of interrupting. Since there are no constructions such as *the weather
which is in England, for example, it seems plausible simply to expand the base rules to
permit certain PPs to be generated under NP. But this extension is precisely what we
need to generate (2.18) as a base form.
As further justification for this expansion of the base rules, we observe that many
adjective phrases can include prepositional phrases or complement clauses which
cannot be reduced relative clauses, for example afraid of Bill, eager to please, pleased
with his work, apprehensive about the job, and sorry to be of trouble. This suggests that

4 Cf. “Remarks”, pp. 45/205-206. One might with good reason try to interpret possession as a special
type of intrinsic connection: there seems little justification to syntactically distinguish own or have from build
or write purely for the purposes of stating a reduction transformation, when a suitable semantic generaliza¬
tion might eliminate the reduction rule altogether.
14 RAY JACKENDOFF

all three categories N, V, and A permit a range of following phrases. We will show in
chapter 4 that prepositions likewise exhibit a range of associated structures.
To express this generalization, Chomsky introduces the variable X to stand for any
lexical category symbol, and uses the notation X (henceforth here X') to denote the
node immediately dominating X. The general base rule schema which expresses the
existence of material to the right of the head of a phrase is (2.19).

(2.19) X' -> X - Comp

Comp (complement) is an abbreviation for some sequence of nodes: since it never

seems to be referred to as a constituent, it does not stand for a node. (2.19) is an
abbreviation for the four rules (2.20).5

(2.20) V' -> V - Compv

N' —» N - CompN
A' —» A - CompA
P' —» P - Compp

Chomsky further suggests that the material preceding the head of a major category
phrase is introduced by a base rule of the form (2.21). X (henceforth here X')
designates the node which is two nodes above a major category node, and [Spec, X']
(the specifier of X', henceforth noted as Specx) is an abbreviation for the material
preceding the head.

(2.21) X" —» Specx - X'

(2.21) is an abbreviation of the rules shown in (2.22):

(2.22) V" —» Specv - V'

N"-* SpecN - N'
A"^ SpecA - A'
P" Specp - P'

Again, the material in the specifier varies from category to category. Chomsky
associates Specv with the traditional auxiliary node, SpecN with the traditional
determiner, and SpecA with traditional degree phrases. We will discuss SpecP in
chapters 4 and 6. Chomsky does not say explicitly whether he considers the specifier to
be a constituent or, like Comp, an abbreviation for a sequence of constituents. His
diagrams show it as a constituent; however, we will argue here that it should not be.
To complete the base rules, Chomsky uses the initial rule (2.23).

(2.23) S —> N" - V"

Thus the trees (2.17) and (2.18) can be rewritten as (2.24) and (2.25). 1 enclose the
traditional names of the nodes in parentheses.

5 Chomsky does not subscript the various Comps by category. However, since the the Comp varies
from one category to another, I have included the subscript to indicate the possibility of variation.
(2.24) N" (NP)

the play
16 RAY JACKENDOFF

Chapter 3 suggests some modifications to Chomsky’s translation of the traditional

nodes into the X notation. However, for the moment we will assume his analysis.
Return to the base structure (2.18), now further analyzed as (2.25), for the derived
nominal John's criticism of the play. One of the stronger arguments for deriving this
nominal from a sentence is that the selectional restrictions parallel those of the
sentence John criticized the play, the range of noun phrases that can occupy possessive
position in the derived nominal is identical with the range of possible subjects in the
sentence, and the range of noun phrases following of is identical with the range of
direct objects in the sentence. In order to capture this generalization for a wide range of
nominals, we do not want to state twice in the grammar the selectional restrictions
shared by criticize and criticism; nor do we want to state twice the set of rules which
enforce these selectional restrictions, once for NP and once for S. One way to avoid
stating the selectional restrictions and rules twice is to derive the nominal from the
sentence, for then the selectional restrictions will only be stated on the verb, and the
rules for enforcing the restrictions will be stated only for the domain S.
However, Chomsky’s use of a more complex lexicon, including related lexical
items, eliminates the problem of stating selectional restrictions twice. The lexicon will
specify, as part of the relation between criticize and criticism, the similarity in
selectional restrictions. So even in the lexicalist approach the selectional restrictions
need only be stated once.
As for the rules which enforce selectional restrictions, we first observe that such
rules must exist in noun phrases as well as sentences, since selectional restrictions for
weather and attitude must prohibit, for example, *the weather in our idea and *my
shirt’s attitude of hunger. Furthermore, Chomsky suggests that along with the
generalization of base rules as in (2.19), there is a corresponding generalization in the
projection rules. For example, the notion “subject of’ can, where semantically
appropriate, be generalized to possessive NPs, and “object of’ can be generalized to
postnominal NPs. This generalization ensures that the rules enforcing selectional
restrictions on criticize and criticism are in fact the same rules in Ss and NPs, and they
need not be stated twice. Thus, if this generalization can be expressed, and if the
lexical relations can be stated successfully, the Lexicalist Hypothesis can relate the
selectional restrictions of derived nominals to those of sentences without recourse to a
transformational derivation.
A second possible difficulty for the Lexicalist Hypothesis concerns the Passive
transformation. We observe that the relation of the derived nominals (2.26a) and (2.26b)
seems to be the same as that between the active sentence (2.27a) and its passive
(2.27b).

(2.26) a. the enemy's destruction of the city

b. the city's destruction by the enemy
(2.27) a. The enemy destroyed the city.
b. The city was destroyed by the enemy.
 2: MOTIVATION FOR THE X CONVENTION

In order to capture the similarity of the two relations, the grammar should not have to
state, in addition to the standard Passive, a transformation having very similar effects
to the Passive but in the domain NP. Within the Transformational Hypothesis this
generalization is quite simple; all four phrases above are derived from the same
underlying form, but (2.26a) undergoes Nominalization, (2.27b) undergoes Passive, and
(2.26b) undergoes Passive followed by Nominalization. (2.26b) thus is the nominaliza¬
tion of a passive sentence.
In the Lexicalist Hypothesis this solution is not available, since there is no
Nominalization transformation. Instead Chomsky suggests that, like the base rules and
the projection rules, some transformations, in particular the Passive, may apply over
the domain NP as well as S. If this is the case, (2.26b) is derived from the same
underlying form as (2.26a), but it undergoes the generalized form of the Passive. Hence
its relation to (2.26a) is the same as the relation of (2.27b) to (2.27a), precisely as
required, and no extra rules need be added to the grammar. Section 4.7 will discuss the
generalized Passive in more detail.
The general nature of the claims made by the X Convention are now clear. The
structural schema (2.28), in which X represents any lexical category, is claimed to
constitute a linguistically significant generalization of the structures associated with
major categories.

(2.28) X"

That is, we expect there to exist rules whose structural descriptions refer to a
range of structures including more than one value of X. For example, we expect to find
rules whose domains include V" and N" or perhaps A' and N': but we do not expect to
find rules whose domains include nodes at different levels, for example P' and A".
The fact that some rules generalize to more than one major category of course does
not mean that all rules generalize over more than one major category. If that were the
case, there would be little reason to distinguish among major categories. Rather, we use
nongenerality of rules to distinguish one category from another. For example, the rule
of Cleft Sentence Formation can be used as a test to distinguish N" and P" from A" and
V".
The existence of cross-category generalizations is therefore not taken to be
evidence that one category is a subclass of some other category, as suggested by some,
for example, Lakoff (1971b), Ross (1969), and Postal (1971, chapter 18). Such a tactic
may merely shift around the names assigned to categories, since the distinctions must
18 RAY JACKENDOFF

still be made. For example, if we claim that all adjectives are deep structure verbs, we
must still assign a syntactic feature ±F which distinguishes surface structure verbs
from surface structure adjectives, and which is referred to by all rules in which
adjectives and verbs differ. In effect, then, we may have simply changed the traditional
name “adjective” into “ + F” and the traditional name “verb” into “ —F”, not a very
insightful proposal.
Of course, in a theory of grammar which does not include the X notation or some
similar device, there is no way to express cross-category generalization directly. Thus
one is forced to resort to a supercategory, with all the syntactic difficulties such a
proposal entails. Chomsky’s claim, then, is that cross-category generalizations play an
important role in grammar, and that the theory of grammar thus must contain a way of
expressing them.
The traditional method of expressing a classificatory system which includes cross-
classification is with a set of distinctive features. For the purpose of stating generalized
rules in the X notation, chapter 3 will introduce a set of syntactic distinctive features.
But first, let us examine other evidence for the Lexicalist Hypothesis, of a similar
nature to Chomsky’s but involving different categories.

2.2. Adjectives and Nouns

Bowers (1968a), from which much of the material in this section is derived, discusses
the relationship between adjectives such as high, long, and deep, and their nominals
height, length, and depth, comparing the transformationalist and lexicalist positions.
The basic fact to be accounted for is that the genitive phrase in nominals like the
building’s height, the railroad’s length, the lake’s depth, the slope’s steepness, the
table’s width, the paper’s thickness, and the plane’s lateness is restricted in precisely
the same way as the subject of the corresponding sentences The building is high. The
railroad is long, The lake is deep. The slope is steep, The table is wide, The paper is
thick, and The plane is late. Likewise for nonmeasure adjectives and their nominals:
the genitive NPs in Mary’s beauty, John's perversity, the rock’s whiteness, the table’s
flatness, the desert's dryness, the sky’s cloudiness, that proposal's absurdity, and
Sarah’s diffidence are restricted in precisely the same way as the subjects of the
sentences Mary is beautiful, John is perverse, The rock is white, The table is flat, The
desert is dry, The sky is cloudy, That proposal is absurd, and Sarah is diffident. Under
a transformationalist theory these relations are expressed by a transformation which
converts embedded predicative Ss into the corresponding NPs, the predicate adjective
becoming a noun. Under the Lexicalist Hypothesis the adjectives and nouns are given
separate but related lexical entries; the noun phrases are generated by the base rather
than by the transformational component.
For purposes of comparison, we will again contrast the derived nominals with
gerundive nominals. Gerundive nominals occur with degree adverbs, but derived
 2: MOTIVATION FOR THE X CONVENTION

nominals occur with degree adjectives:

(2.29) the building’s being extremely high

the table’s being relatively flat
(2.30) the building’s extreme height
the table’s relative flatness

There is a range of degree phrases possible in gerundive nominals that is impossible in

derived nominals.

(2.31) the building’s being very high

the table’s being too flat
(2.32) *the building’s very height (wrong reading)
*the table’s too flatness

The characteristic equative construction for gerundive nominals is as ... as-, for
derived nominals the same ... as is sometimes applicable.

(2.33) the building’s being as high as that tree

the table’s being as flat as the desert
(2.34) The building has the same height as the tree.
?The table has the same flatness as the desert.

Subjectless gerundive nominals cannot use the definite article; subjectless derived
nominals may.

(2.35) *the being high

(2.36) the height

Thus the adjective phrases in adjectival gerundive nominals have all the properties one
would expect of adjective phrases; the corresponding derived nominals have the
internal structure of noun phrases, taking all the characteristic noun modifiers such as
the definite article, the same, and prenominal adjectives. Such behavior is consistent
with the Transformational Hypothesis, in that one can always state transformations to
make all the necessary adjustments; but the existence of these transformations is taken
to be an accident of English syntax—things could just as well be otherwise. Under the
Lexicalist Hypothesis, however, only the gerundive nominals contain an underlying
AP; the derived nominals are deep structure NPs. This theory thus predicts the
difference between the two types of nominals and describes the behavior of derived
nominals as an automatic consequence of the behavior of ordinary NPs. Since ordinary
lexical properties determine syntactic behavior, the Lexicalist Hypothesis says that
things would be less general if they were otherwise, for instance if (2.31) instead of
(2.32) were ungrammatical and everything else were the same.
20 RAY IACKENDOFF

As with the deverbal nominals, there are irregularities in the relationship between
adjectives and nouns. Height does double duty for tall and high, there being no
*tallness. In addition, the nouns elevation and altitude cover part of the same semantic
area but are unrelated to adjectives. *Fastness is supplanted by speed (which is in turn
related to an adjective speedy, which is in turn related to a noun speediness), *oldness
is supplanted by age, and *afraidness is supplanted by fear (which is related to an
adjective fearful). Ability covers a much wider semantic range than its adjective root
able', the semantic relation between clean and cleanliness is irregular. Adjectives which
take infinitive complements (She is pretty to look at; It is easy to please Bill) do not
have direct nominal counterparts *her prettiness to look at, *the ease/easiness to
please Bill, though the latter type is related to a gerundive complement construction the
ease/easiness of pleasing Bill. In a transformational theory, each of these irregularities
must be derived by a lexically specific transformation or constraint. In the Lexicalist
Hypothesis, they simply represent deviations from the most ideal relationship; irregular
pairs contain more independent information than regular ones, but do not otherwise
complicate the grammar.
Under the Lexicalist Hypothesis, gerundive nominals and derived nominals will
have structures (2.37) and (2.38), respectively.

(2.37) N"

perverse
 MOTIVATION FOR THE X CONVENTION

As with the nominals of section 2.1, we must account for the generality of selectional
restrictions, e.g. for the fact that perverse restricts the subject in (2.37) in precisely the
same way that perversity restricts the genitive phrase in (2.38). The transformationalist
theory of course explains this relationship as an automatic consequence of the deep
structure ascribed to (2.38), which, like (2.37), will contain the S John be perverse.
In the lexicalist framework, we must again appeal to the use of redundancy rules in
the lexicon to express the similarity of the selectional restrictions of perverse and
perversity. As for the rules which enforce selectional restrictions, we can appeal to the
extension to NPs of the notion “subject-of ’ proposed in section 2.1. Since the genitive
NP in (2.38) is to be considered the subject of (2.38), the rules which impose selectional
restrictions on grammatical subjects will apply equally in (2.37) and (2.38). Thus, as in
section 2.1, the Lexicalist Hypothesis requires a cross-category generalization to be
expressed, but this time in terms of adjectives and nouns.
Bowers points out that there is a similarity in the semantic functions of various
subordinate clauses in APs and NPs.

(2.39) a. The table is so wide that the cloth won’t cover it.
b. The table is as wide as I thought it was.
c. The table is wider than I thought it was.
d. The table is too wide to be useful.
 RAY JACKENDOFF

I has
(2.40) a. The table such width that the cloth won’t cover it.
lis
b. The table has the width 1 thought it had.
c. The table has more width than I thought it had.
• ''T'l . li (doesn’t have the
d. The table 1. _ width to be useful.
fisn t ot a

While there is not a clear one-to-one correspondence between these phrases, there is
enough similarity to require explanation. In particular the relation between the equative
(2.39b), a comparative clause, and (2.40b), a relative clause, is striking. Bowers
suggests a similarity in the structures of comparative and relatives roughly like (2.41)
and (2.42).

An extension of the grammatical relation obtaining between a noun and its relative
clause to that between an adjective and its comparative clause is called for. The
Lexicalist Hypothesis by now leads us to expect such extensions. We will deal with
this particular relation in chapter 8.
Throughout this section we have ignored an alternative form for the derived
nominals: the height of the building, the width of the table, etc. Bowers argues that this
form is transformationally derived, and that the rule deriving it from the form with a
genitive NP accounts for some of the difficulties in Chomsky’s treatment of the passive.
We will deal with this rule in connection with the passive, in section 4.7.
 MOTIVATION FOR THE X CONVENTION

In sum, then, adjectives and their nominalizations present a picture quite similar to
verbs and their nominalizations. The same difficulties accrue to the Transformationalist
Hypothesis, and the same problems of lexical relations and cross-category generaliza¬
tions must be solved by the Lexicalist Hypothesis.

2.3. Adjectives and Adverbs

The similarities between adjectives and adverbs in English are far more striking than
those between other pairs of categories we have considered, and many people have
considered it self-evident that adverbs should be derived transformationally from
adjectives. Nevertheless, we will show here that a theory which does not distinguish
between the two categories in deep structure leaves something to be desired.
First observe why a transformational theory is so tempting. Nearly all adverbs
consist morphologically of an adjective plus the -ly suffix, and the meanings of
adjective-adverb pairs are usually quite close. Furthermore, the modifier systems of
the two categories are identical, as pointed out by Bowers (1968b), among others.

rather
(2.43) a. John’s car is incredibly ' fast.
very
b. Its acceleration is so rapid that you’ll never catch him.
c. That quartet is more beautiful than you can imagine.
d. The stop was too sudden to react to.
e. The sun was not as bright as the sunlamp.

b. It can accelerate so rapidly that you'll never catch him.

c. They played the quartet more beautifully than you can imagine.
d. It stopped too suddenly for us to react.
e. The sun didn’t shine as brightly as the sunlamp.

But deriving adverbs from adjectives in any semantically enlightening way turns
out to be a process conspicuously lacking in generality: observe the variety of sources
needed (examples from Jackendoff (1972, 52-55)).

(2.45) John is careless at driving his car. —» John drives his car carelessly.
The manner in which John disappeared was elegant. —> John disap¬
peared elegantly.
It was easy for Stanley to win the race. —» Stanley won the race easily.
I am frank in saying there is no reason for it. —> Frankly, there is no
reason for it.
The time at which Harry was known as "The Red Death” was a former
time. —> Harry was formerly known as "The Red Death”.
24 RAY JACKENDOFF

The extent to which this new development complicated matters is

double the extent to which matters were complicated before. —* This
new development doubly complicates matters.

There are many cases where the related adjective exists but provides no convincing
paraphrase for the adverb.

(2.46) a. The men were individually asked to leave.

*It was individual that the men were asked to leave.
*The manner in which the men were asked to leave was individual.
b. Ira readily accepted the offer.
Ira was ready to accept the offer, (wrong meaning)
*The manner in which Ira accepted the offer was ready.
c. Stanley completely ate his Wheaties.
?*Stanley’s eating of his Wheaties was complete.
*The degree to which Stanley ate his Wheaties was complete.
d. Irving finally ran away.
*It was final that Irving ran away.
*The event in which Irving ran away was final.
e. Tom absolutely refuses to go.
*The degree to which Tom refuses to go is absolute.
*Tom is absolute in refusing to go.
f. Actually, John can’t lose.
*It is actual that John can’t lose.
g. I was merely trying to help.

*The |j^gl|1ggr| 'n which I was trying to help was mere.

h. This data virtually shatters the transformational theory.

*The degree to which this data shatters the transformational theory is
virtual.

Thus the transformational theory is again faced with the necessity for a large number of
more or less lexically specific transformations. For cases like (2.46), a transformational
derivation cannot even succeed in properly relating the adjective and the adverb forms,
so the motivation for the derivation is vitiated. These are just the kinds of semantic and
syntactic irregularities we found in the previous sections.
Because of the great similarity in the syntax of adjectives and adverbs, it is difficult
to find arguments against the transformational analysis which are based on the contrast
in structure between APs and AdvPs (like the arguments based on (2.1)—(2.10) and
(2.29)—(2.36)). However, there is one major syntactic difference between the two kinds
of phrases: many adjectives strictly subcategorize a PP or an S in their complements,
but the related adverbs do not.
 MOTIVATION FOR THE X CONVENTION

(2.47) a. Tired (of the noise), John left the room.

Tiredly (*of the noise), John left the room.
b. Fearful (of a revolt), the king ordered a purge.
Fearfully (*of a revolt), the king ordered a purge.
c. The manner in which John grimaced was expressive (of his needs).
John grimaced expressively (*of his needs).

d. Happy
I at their departure
[that they were leaving
j, John waved goodbye.

Happily
*at their departure
*that they were leaving
j, John waved goodbye.

I with anticipation
e. Eager ( John chewed his nails.
[to leave

*with anticipation
Eagerly ( John chewed his nails.
*to leave

In the transformational theory of adverbs, all the various adverb-creating transforma¬

tions could be restricted so as to apply only to adjectives without complements;
alternatively, in order to capture the generality of the phenomenon, a surface structure
constraint could be imposed forbidding complements in adverb phrases. But this latter
ploy, the more general of the two, is rather clearly a lame attempt to enforce at the
surface a restriction which is easily stated in deep structure under the Lexicalist
Hypothesis: the phrase structure rules do not generate complements in adverb phrases.
Thus, as in the other two cases we have discussed, there are structural reasons as well
as semantic arguments and considerations of irregularity which favor the Lexicalist
Hypothesis.
This last consideration also argues against a mixed transformationalist-lexicalist
account of adverbs, first proposed (to my knowledge) in Emonds (1970). Emonds
observes that there are no deep-structure positions which allow both adjectives and
adverbs: adjectives are dominated by NP and Predicate (i.e. the complement of be,
become, etc.—cf. Chomsky (1965, 107)), and adverbs are dominated by VP, S, and
Degree. Emonds suggests that there is a single category A which can appear in all of
these positions but receives the -ly suffix transformationally or during lexical insertion
when it is in the adverb positions.
This account successfully avoids the difficulties of stating adverb formation
transformations, since adverbs are not derived from adjective paraphrases but simply
from adjectives inserted in adverb positions. Presumably the irregularities in meaning
are accounted for in the lexicon, as in the Lexicalist Hypothesis. But there is no way to
account for the difference in complement structure; an AP is predicted to have the
same structure no matter what position of the sentence it is generated in. Treating
adjectives and adverbs instead as separate categories permits the base rules to assign
them different complement structures.
26 RAY JACKENDOFF

In fact, Emonds’s theory does not differ from Chomsky’s theory of derived
nominals, except in one crucial respect: Chomsky distinguishes the two categories verb
and noun, for obvious reasons, but Emonds collapses adjectives and adverbs into a
single category, their differences being less apparent. What Emonds’s theory and the
original transformationalist theory account for is the almost complete similarity of
modifier structure and the morphological similarity.
The Lexicalist Hypothesis, on the other hand, is based on expressing the
differences between categories. It thus must find a new way to express the similarities,
since the use of a common category label is not possible. Again, we need a way to
express cross-category generalizations, just as in the previous two cases.

2.4. Coming Attractions

This chapter has shown that the Transformationalist Hypothesis runs up against serious
difficulties in dealing with a number of different problems, and that there is a potential
solution to these difficulties called the Lexicalist Hypothesis. Chomsky’s article
“Remarks” sketches a great number of suggestions concerning the formulation of the
Lexicalist Hypothesis and its consequences, but does not deal with any of the issues in
much detail.
A great deal of subsequent work has verified the need for the theory to express
cross-category generalizations, as predicted by the X Convention. For instance.
Bowers (1968b), Selkirk (1970), Bresnan (1973), and Milner (1973) explore the generali¬
zations among the determiner systems of NPs, APs, and AdvPs that we will address in
chapter 6. Dougherty (1970-1971) shows that the generalizations about surface con¬
junction of various constituents are not to be accounted for by Conjunction Reduction
from conjoined sentences, but must be treated as cross-category generalizations in the
base. Van Riemsdijk (1973; 1976) and Jackendoff (1973) show that prepositions form a
lexical category that participates in the X Convention.
Outside the categorial component, Selkirk (1974) shows that the X Convention
plays a role in determining liaison in certain dialects of French. Akmajian (1975) argues
that NP is a cyclic node, as part of an account of an extraposition rule that applies in
both Ss and NPs. Jackendoff (1971) shows that the rules of Gapping and VP-Deletion
have close NP analogues. Horvath (1976) demonstrates how a topicalization process in
Hungarian appears within several different syntactic categories. Chomsky (1973) uses
the cross-categorial notion of subject in defining his Specified Subject Constraint on
rule application. Perhaps most striking, Bresnan (1976a) revises Chomsky’s A-over-A
Condition into a cross-categorial “X-over-X” Condition, showing that constraints on
rule application depend in part on how cross-categorially the rules are stated.
The Lexicalist Hypothesis has also sparked an interest in the structure of the
lexicon. Such works as Halle (1973), Aronoff (1974; 1976), Siegel (1974), Rardin (1975),
and Jackendoff (1975a) have begun to open up the lexicon as a rich area of inquiry
independent of syntax and phonology.
 2: MOTIVATION FOR THE X CONVENTION

What has remained relatively neglected is the theory of phrase structure. Though
all the works just cited make reference to the X Convention, none attempts to explore
systematically its implications for questions such as these: How is the X Convention to
be formulated so as to provide the best account of the syntactic structures of the
language? What is the system of syntactic features relating lexical and syntactic
categories? Can the notions “possible syntactic category” and “possible phrase
structure rule” be characterized?
The rest of this study will be concerned with these questions, and how the
grammar of English can be brought to bear on them. Chapter 3 restates the X
Convention, proposes a theory of syntactic features and a highly restrictive phrase
structure schema, and uses them to reanalyze the grammar of the subject and the
auxiliary. Chapters 4 through 8 apply the theory to substantial fragments of English
phrase structure, defending detailed solutions to problems raised by complements, NP
specifiers, the degree system, relative clauses, and degree clauses. Chapter 9 investi¬
gates a second phrase structure rule schema, one which generates a previously
unremarked class of “deverbalizing rules”; chapter 10 restates a general overview, lest
it have been lost in the intervening furor.
 3: A Theory of Phrase
Structure

Chapter 2 presented arguments that the X Convention plays a role in the the theory of
syntax. This chapter will formalize what I take to be a quite conservative version of the
X Convention, the formally simplest and most restrictive statement of the theory that
appears adequate for the description of English. In adopting this approach and thus
requiring reformulation of many well-accepted analyses which would weaken the
theory, I am consciously adhering to the research strategy recommended by Chomsky
in the preface of Syntactic Structures:

Precisely constructed models for linguistic structure can play an important role, both
negative and positive, in the process of discovery itself. By pushing a precise but inadequate
formulation to an unacceptable conclusion, we can often expose the exact source of this
inadequacy and, consequently, gain a deep understanding of the linguistic data. More
positively, a formalized theory may automatically provide solutions for many problems
other than those for which it was explicitly designed.

In attempting to flesh out the X Convention into a full theory, I will separate
Chomsky’s proposal into a number of parts, some of which will be supported here,
others of which will not. Essentially, I will retain what I take to be the conceptual core
of Chomsky’s proposal, but I will modify most of the details.

3.1. Reformulation of the X Convention

The X Convention can be taken as a theory of syntactic categories in universal

grammar, making three principal claims. First, universal grammar includes a set of
syntactic distinctive features which defines the possible lexical categories of human
languages. A particular language chooses its repertoire of lexical categories from among
those provided by universal grammar, just as it chooses a phonological repertoire from
the possibilities provided by universal grammar. Presumably the choice must be made
in a constrained way—one would hardly expect a language with adverbs but no nouns.
But this too is parallel to phonological theory, where one does not expect a language
with the sound u but no a.
The second claim of the X Convention is that each lexical category X defines a set
30 RAY JACKENDOFF

of syntactic categories X', X", . . . , X*, the supercategories of X, related by phrase

structure rules of the form (3.1).

(3.1) Xn^ . . .X"-1. . .

(3.1) is a phrase structure rule schema provided by universal grammar. It results in

phrase structure configurations of the form (3.2).1

(3.2) Xfc

The head of a phrase of category X” can be defined in two different ways, either as
the X"-1 that it dominates or as the lexical category X at the bottom of the entire
configuration. We make use of both senses in this study, making it clear when
necessary which sense is meant. Both reflect traditional usages of the term. Refine¬
ments of (3.1) in section 3.3 will guarantee that the head is unique, that is, each X" will
dominate one*and only one X”-1. Section 3.6 will take up a number of exceptions to this
generalization.
The third claim of the X Convention is that rules of grammar are stated in terms of
syntactic feature complexes and the prime notation. To be more precise, each term of
any rule of grammar must be either a specified lexical or grammatical formative or of
aFiT
the form . This third claim gives substance to the other two claims, since it is by

attempting to write rules of grammar in these terms that we test hypotheses about the
feature system and the hierarchical organization of categories.
To flesh out these claims, we must answer at least these questions: What are the
syntactic distinctive features? Is there a maximum value for n in (3.1), and is it the same
for each category? Are there any constraints on what may appear in place of the
ellipses in (3.1)? Does (3.1) provide the only phrase structure rules of the language or

1 Ivan Sag has pointed out (personal communication) that this particular aspect of the X Convention, the
notion that syntactic categories are projections of lexical categories, is a fundamental assumption of Harris’s
approach to syntax (Harris (1946; 1951)). On the other hand, Harris does not (at least explicitly) use this
notion to get at cross-category generalization, nor does he impute any significance to the number of
supercategories associated with a lexical category.
 3: A THEORY OF PHRASE STRUCTURE

are there other types? The rest of this chapter is devoted to formulating one possible
set of answers to these questions; subsequent chapters defend these solutions on the
basis of the grammar of English.

3.2. The Syntactic Distinctive Features

The choice among competing systems of distinctive features should be made on the
basis of how easy it is to state actual rules of the language in terms of the proposed
systems. One presumes that rules are more likely to generalize to “natural classes”,
those that take fewer features to pick out. This criterion is essentially the same as the
one used in justifying phonological feature systems, and should therefore be altogether
familiar.
In “Remarks on Nominalization” Chomsky analyzes the major lexical categories
N, A, and V into the features ±N and ±V in the following way:

(3.3) +N -N

+V A V

-V N

A fourth major syntactic category is P(reposition), about which we will have much to
say in this study. Presumably Chomsky’s system would analyze it as [—N, — V],
completing the matrix in (3.3).
In distributing two distinctive features over four major categories, there are
actually only two systems possible other than (3.3):

(3.4) a. N V b. N V

A P P A

The choice among the three must be made on the basis of what categories go together
in the rules of English (and other languages, of course). Anticipating the generalizations
that appear in the rest of this study, we will adopt (3.4a) as our feature system. It will
emerge that there are many rules which generalize across supercategories of N and V,
and this is not expected in a feature system like (3.3). Similarly, there are many rules
which generalize across supercategories of N and A, and many to V and P, and this is
not expected in (3.4b). About the only rule that makes (3.3) and (3.4b) look more
natural than (3.4a) is the formation of cleft sentences, in which the clefted item can be
only an NP or a PP. Since the combination of N and P is so rare, and the combination
of V and A at least equally rare, we will feel justified in provisionally accepting (3.4a)
as the major division of lexical categories. As it would be too laborious to test the three
systems continually throughout this study, I leave it to the interested reader to check
the adequacy of this choice.
The classification of categories made by feature system (3.4a) corresponds to some
32 RAY JACKENDOFF

traditional divisions of categories: N and A are often called “substantives”, and A and
P are often thought of as “modifiers”. (3.4a) also has the interesting property of
corresponding to two significant divisions in the phrase structure rules of English.
Though these divisions may not strictly obtain in other languages, we will for
convenience name the features after the corresponding properties of English phrase
structure.
The first division is between those categories which have a subject, N and V, and
those for which no syntactic subject relation exists, A and P. We will accordingly call
the feature ±Subj, and designate N and V as +Subj and A and P as -Subj.2
To emphasize the heuristic, nontheoretical significance of the names for the
features, it is worth pointing out that in French, for example, nouns cannot take NP
subjects (other than possessive pronouns, which may be articles). Nevertheless, we
assume that French nouns and English nouns have the same syntactic features, and
that it is only the way these features appear in the grammar that differs from one
language to another.
The second major division in the phrase structure of English is between those
categories whose complements may include a surface NP direct object after the head,
i.e. V and P, and those categories whose complement cannot contain a surface NP, i.e.
N and A. The feature will be named ±Obj; V and P are +Obj and N and A are -Obj.
Again, the use of the term “direct object” is strictly syntactic: since there is no NP *his
consideration the offer and no AP *afraid Bill, but only his consideration of the offer
and afraid of Bill, these categories are not considered to have syntactic direct objects.
Next consider the “minor” lexical categories of English, starting with adverbs.
They clearly must be related to adjectives, and we will designate them both as [-Subj,
—Obj], Since their major difference with respect to associated phrase structure is in the
ability to take a complement, we will invent a feature to distinguish them and call it
±Comp: adjectives are +Comp and adverbs are -Comp.
We will also use the feature Comp to describe other minor lexical categories in the
language. The clearest case is the category Particle: particles such as up, on, and away
are morphologically identical to prepositions but permit no complements. To express
this relationship, we designate both particles and prepositions as [-Subj, -t-Obj];
prepositions are +Comp and particles are —Comp.
The difference between modal verbs and ordinary verbs must be stated somewhere
in the grammar, and ±Comp is a good place to localize it, calling verbs [ + Subj, +Obj,
+ Comp], and modals [ + Subj, +Obj, —Comp]. This analysis makes them separate

2 One might question the claim that adjectives do not have a subject, since it is often assumed that John
in John is tall is the subject of tall. This assumption is incorrect. Although tall imposes a selectional
restriction on the NP in NP is tall, the NP bears the grammatical relation "subject-of’ to the verb he, not to
the adjective. This becomes clearer if other verbs are substituted for he, e.g. John became tall. John made it
tall, etc. An NP bearing the “subject-of’ relation to an adjective would have to be contained in the AP, as
the subject of a noun is contained in the NP. Since there is no AP *John('s) fearful corresponding to the NP
John's fear, for example, we conclude that adjectives do not have syntactic subjects. For a discussion of how
the selectional restriction is imposed without a grammatical relation, see Jackendoff (1974b), especially
section 5.
 3: A THEORY OF PHRASE STRUCTURE

categories, yet their close feature relationship makes plausible the historical creation of
the category of modals in English (cf. Lightfoot (1974)): a particular class of verbs
underwent change in one syntactic feature, moving into a category available in
universal grammar but previously unrealized in English. We will deal with the English
auxiliary in somewhat more detail in section 3.5.
Finally, there is a clear candidate for a -Comp category associated with nouns:
the rather heterogeneous system of articles and quantifiers. Chapter 5 will further
divide these into two classes, called Articles (Art) and Quantifiers (Q). Although the
actual division will not be determined until chapter 5, we introduce the relevant feature
±Det here for completeness: articles are [ + Subj, — Obj, -Comp, +Det]; quantifiers are
[ + Subj, —Obj, —Comp, -Det],
We will use the feature ±Det also to pick out a special class of adverbs, the
"degree words” so, too, as, etc. We will call this category “Degree” (Deg) and
designate it by the features [-Subj, -Obj, —Comp, + Det]; ordinary adverbs will be
[-Det], This choice will be justified in chapter 6; it depends primarily on parallelisms
between Art and Deg, which we therefore wish to analyze as differing in only one
feature.
There is no place in this system for coordinating conjunctions and complementiz¬
ers. However, they do not participate in the grammar in the same way as the other
categories, in that they do not strictly subcategorize complements and specifiers, so
their exclusion is motivated. Section 3.6 will suggest places for them in the general
theory. Subordinating conjunctions, on the other hand, can be described as preposi¬
tions with sentential complements, so they do participate in the feature system.
The feature system thus looks like this:

3.3. The Phrase Structure Rule Schema

As stated in section 3.1, the second claim of the X Convention is this: every lexical
Category X must be dominated in phrase structure by a hierarchy of categories X', X",
34 RAY JACKENDOFF

. . . , Xfc; furthermore, the set of syntactic categories in a language is completely

determined by the lexical categories plus the hierarchical categories projected from
them. In formal terms, the claim is that all possible phrase structure rules are of the
form (3.1), repeated here:

(3.1) Xn -> . . .X"-1. . .

We will weaken this claim slightly in section 3.6. But aside from minor exceptions, we
will claim that (3.1) represents the canonical form for all phrase structure rules.
Notice what sorts of rules (3.1) excludes. One kind of phrase structure rule that
has been widely accepted in the literature is so-called Chomsky-adjunction,3 which
appears in a popular source for relative clauses (3.6a), one account of the verb-particle
construction (3.6b), one theory of manner adverbs (3.6c), and many other places.

(3.6) a. NP —* NP - S
b. V -* V _ Prt
c. VP —» VP - Adv

Since these rules generate no category on the righthand side that is one level lower
than the category on the lefthand side, they are impossible within schema (3.1).
In Vergnaud’s (1974) account of relative clauses, there is a phrase structure rule of
the following form:

(3.7) N" —» ±Def - N'" - S

Such a rule is impossible within schema (3.1), since (3.1) requires a rule expanding N"
to contain an N' somewhere on its right-hand side.
Berman (1974, 109) proposes a number of structures for comparative constructions
which illustrate violations of (3.1) and which seem in general counterintuitive. There is
nothing in the standard theory of phrase structure to rule them out, though:

(3.8) (= Berman’s (127))

a. AP

QP

3 This term has nothing to do with Noam Chomsky, and he regrets its existence (personal communica¬
tion). However, since it seems firmly established in common linguistic usage, I will retain it, sparingly, with
apologies.
 3: A THEORY OF PHRASE STRUCTURE

In (3.8a), AP dominates QP, but immediately dominates no category A'. In (3.8b), in

addition to the Chomsky-adjunction, the lower AP dominates N rather than an A'; in
(3.8c), the lower AP dominates only PP and no A’.
Thus (3.1) is a relatively restrictive theory of phrase structure rules, ruling out
many counterintuitive structures as well as a number that have been taken rather
seriously in the literature. In particular, the Chomsky-adjoined source of relative
clauses (3.6a) enjoys a wide following. Chapters 4 through 8 will be concerned with
showing that (3.1) permits well-motivated descriptions for a wide range of these
constructions, and thus that (3.1) is not too restrictive a theory of phrase structure.
Schema (3.1) leaves open two issues: first, what is the maximum value of n and
does it differ from category to category? Second, what is permitted on the righthand
side besides Xn_1, filling in the ellipses in (3.1)?
Various answers to the first question appear in the literature. In Chomsky’s
original formulation, n equals 2 for nouns and 3 for verbs (assuming the verb is the head
of the sentence). Vergnaud (1974) and Siegel (1974) have n equal to 4, at least for
nouns; Dougherty (1968) has n equal to 3 for nouns and 6 for verbs; Jackendoff (1971;
1974a) has n equal to 2 for all categories. The best theory, of course, provides just
enough structure to make the relevant structural differences and no more. It now
appears to me that n must equal 3 for verbs and nouns. Chapter 6 will show that for
reasons of structural parallelism with the syntax of NPs, n must equal 3 for several
36 RAY JACKENDOFF

other categories as well, including the minor lexical categories Q and Adv. Thus, at the
expense of some otherwise superfluous structure, the most uniform hypothesis is that n
equals 3 for all categories. We will call an X'" a major phrasal category.
This hypothesis makes possible an answer to the second question above, close to
the proposal of Emonds (1976, chapter 1). We will claim that every category to the left
or right of X"-1 in (3.1) is either a major phrasal category or a specified grammatical
formative such as have, number, case, or tense, and probably that it is optional. Thus
the canonical form (3.1) for phrase structure rules can be refined to (3.9).

(3.9) Xn (Cj). . .(Q) - X"-1 - (Cj+1). • .(Cfc), where 1 < n < 3, and for all C,,
either C{ = Y'" for some lexical category Y, or Cj is a specified
grammatical formative.

We will refer to rule schema (3.9) as the Uniform Three-Level Hypothesis.

(3.9) rules out many further phrase structure combinations, for example the
situation in (3.8a,b) in which QP dominates two lexical categories, Q and A. More
generally, (3.9) guarantees that the category of a phrase and its head are unambiguously
determined by the head: no situation such as (3.10a,b) can arise, in which two different
categories dominate precisely the same constituents.

A" Q" A" Q"

We combine the feature notation of section 3.2 with the prime notation to arrive at
distinctive feature analyses of all syntactic categories. For example, NP (now N'") is
designated as [ + Subj, -Obj, +Comp]'"; a rule applying either to P' or V' will mention
the feature complex [ — Obj, +Comp]'. The combination of the two notations makes the
desired claim about possible cross-category syntactic generalizations: generalizations
must be across categories of the same level, so it is impossible, for example, to pick out
both N" and Q' with a single term of a structural description. This hypothesis, as will be
seen, is often crucial in deciding which of a number of possible structures to assign to a
construction, in that parallelisms with the structures associated with other lexical
categories constrain the choices in interesting ways.
Before going on, let us clarify our terminology. We will continue to use the
traditional terms S, NP, AP, AdvP, PP, and QP informally for the major phrasal
categories associated with V, N, A, Adv, P, and Q, respectively; the Three-Level
Hypothesis claims that these traditional symbols for the major phrasal categories are
equivalent to what rule schema (3.9) defines as V'", N"', A'", Adv'", P'", and Q'". A
glossary of all category names and their feature analyses appears at the end of this
chapter.
 3: A THEORY OF PHRASE STRUCTURE

Chomsky uses the term specifier to refer to the material in a phrase to the left of
the head, and complement to refer to the material to the right of the head. According to
the tree structures in “Remarks”, Chomsky considers specifier to represent a syntactic
category, but complement is simply an abbreviatory term for some concatenation of
ordinary syntactic categories. However, there is to my knowledge no evidence that
either complements or specifiers function as constituents—they do not move or delete
as units, and unlike normal constituents, no part can be designated as a head.
Consequently, I will use the terms specifier and complement for expository conveni¬
ence only, with no theoretical significance implied.4 In (3.9), the terms Clf . . . , C} will
be referred to as X" specifiers, and Cj+1, . . . , Cfc as X" complements.
This terminology is suitable for discussion of English, though not perhaps for other
languages, since it calls the direct object in an SOV language part of a specifier, not of a
complement. In fact, we will see even in English that there are a few cases where a
particular grammatical relation is defined as part of an X’ specifier for some categories
and part of an X‘ complement for others. It must be understood, therefore, that the
distinction between specifier and complement is to be regarded here as of no theoretical
significance, but only as a convenience.
The next two sections work out two major aspects of English syntax, the subject
and the auxiliary, within the Uniform Three-Level Hypothesis. Besides exemplifying
the hypothesis in the most immediate way possible, they set the stage for the detailed
analyses of complements, specifiers, and relative and degree clauses in subsequent
chapters. After the discussion of the subject and the auxiliary, section 3.6 discusses
certain phrase structure expansions not covered by the Uniform Three-Level Hypothe¬
sis.

3.4. The Generalized Subject Relation

The grammatical relation “subject-of (an S)” is defined by Chomsky (1965, 71) as
[NP,S]; that is, in a string exhaustively dominated by S, the subject is a substring
exhaustively dominated by NP, and that occurrence of NP is directly dominated by the
S. We would like to define the relation “subject of an NP” for derived nominals in such
a way that it generalizes structurally with the relation “subject of an S”, so that the
various rules involving subjects need not be stated separately for NPs and Ss.
This is the first of many analyses to be presented here in which structural
parallelism across categories is a crucial consideration. The general principle entailed
by the X Convention is that if parallel grammatical relations exist in two different
categories, the categories must be syntactically parallel with respect to that grammati-

4 Hornstein (1975) claims that calling the specifier not a category but a concatenation of nodes is
unjustified, and that a treatment in which specifier is a category makes a stronger claim. The latter point is
correct, in that my view of the specifier ascribes less highly differentiated structure to sentences. However,
since neither he nor anyone else has given arguments that this extra structure is necessary (e.g. in terms of
movement or deletion), the weaker position seems to be the more supportable.
38 RAY JACKENDOFF

cal relation. In this way rules involving that grammatical relation can be stated so as to
apply to both categories, by appropriate use of syntactic distinctive features.5

3.4.1. Modifying Chomsky’s Analysis

Let us compare Chomsky’s proposed deep structures for John has proved the theorem
and several of John's proofs of the theorem.

(3.11) a. S

b. N"

John

5 Note that the X Convention says nothing about what to do with nonparallel structures. Hornstein
(1975) objects to the generalization of the subject relation on the grounds that many other aspects of Ss and
 3: A THEORY OF PHRASE STRUCTURE

(The details of SpecN are from Chomsky’s (30), p. 37/200.) We would like to say that
John is subject in both cases. Observe, however, that the subjects are not in
structurally parallel positions. In particular, the subject of the S (3.11a) is dominated by
the third node above the head verb, but the subject of the NP (3.11b) is dominated by
the second node above the head noun. Furthermore, the subject of the NP is embedded
in the Specifier node, whereas the subject of the S is directly dominated by S.
Let us first see what can be done about the structure of the derived nominals. We
can make some improvement by taking several of not as part of the determiner of this
NP but as part of a higher NP, thus:

(3.12) N"

N'

N" ’s proofs of the theorem

N'

John

Chomsky suggests this possibility in note 26, p. 38/note 27, p. 219. Essentially this
structure for quantifier constructions is argued for in Jackendoff (1968); we will develop
it further in chapter 5.
Next, observe that the possessive affix's occurs with all NPs in subject position in
nominals, including those that arrive in that position through a transformation (e.g. the
NPs are not parallel—for example, Ss have auxiliaries and complementizers, and NPs have determiners. But
these differences are irrelevant: the X Convention says simply that when parallelisms exist, they must be
expressed.
40 RAY JACKENDOFF

city’s destruction by the enemy). This suggests that a relatively late obligatory
transformation adds the affix to whatever NP occupies surface subject position. (In
section 3.6 gerundive nominals will be analyzed in such a way that they too receive
their ’5 by this transformation.) With such a rule, we can eliminate the node Poss in
(3.12) and place the NP dominating John directly under SpecN, yielding (3.13) instead of
the lower N" in (3.12).

(3.13) N"

John

Next suppose we take the view of specifiers proposed in the last section: there is
no category Specifier; rather “specifier” is simply an abbreviation for a concatenation
of nodes. Then SpecN and Specv can be eliminated in (3.1 la) and (3.13), and John is in
nearly parallel positions in the two structures. It remains to modify one or the other of
the structures so that John is attached the same number of nodes up from the head. By
dropping a node in the S (3.11a) we get a parallelism of the form (3.14); by adding an
extra node in the NP (3.13) we get the pair (3.15).
 3: A THEORY OF PHRASE STRUCTURE

b. N"

proof of the theorem

prove the theorem

b. N'"

proof of the theorem

(3.14) is the two-level hypothesis of Jackendoff (1974a); (3.15) is the three-level

hypothesis proposed in section 3.3. In either pair, John is in precisely parallel positions
in NP and S; we can therefore define the generalized grammatical relation “subject-of”
as [N", [ + Subj]"] in (3.14) and as [N'", [ + Subj]'"] in (3.15). The choice of (3.15) over
42 RAY JACKENDOFF

(3.14) will be dictated by the analyses of auxiliaries in section 3.5, of complements in

section 4.1, and of NP specifiers in sections 5.2 and 6.2.

3.4.2. Arguments for the Uniform-Level Subject Relation

Here are some arguments which favor the uniform-level theories (3.14) or (3.15) over
Chomsky’s mixed-level theory. The arguments all show that where rules generalize
across categories, it is invariably Ss, not VPs, that behave like NPs, APs, and PPs.
Hence, by the assumptions of the X Convention, Ss must have the same number of
primes as the other major phrasal categories.
First, observe that the rule of Topicalization preposes the major syntactic
categories NP, AP, and PP:

(3.16) a. My brother),) everyone expects Bill to like.

b. Taller than Marvin),) no one ever expected you to be. (in “Yiddish”
English)
c. Into the bucket(,) we asked you to put the bananas.

Topicalization also applies to sentences, not to VPs:

(3.17) a. That you were comingC) no one ever expected Bill to find out.
b. *Coming tomorrow(,) no one ever expected Bill to find out that you
were.

If the rule of Topicalization is to generalize across categories, the appropriate

generalization thus is to sentence, arguing that it is of the same level as the other major
syntactic categories.
The pronoun it can have as its antecedent an NP, an AP, or an S:

(3.18) a. A car drove up and 1 looked at it.

b. She is heavy, but she doesn’t look it.
c. Bill came, but 1 didn’t know it.

The anaphoric expression for VP is do so or do it, which does not generalize directly
with any other category. Hence the generalization calls for S rather than VP being
similar to NP and AP.
Similarly, NPs, APs, PPs, and Ss serve as antecedents of appositives.

(3.19) a. Charlie talked to Wendy, who was carrying the groceries.

b. Karl is famous, which you’ll never be.
c. The tree was in the clock tower, which was certainly an odd place for
it to be.
d. Harold left abruptly, which surprised no one.

If the rule which determines antecedents of appositives is to be stated in the simplest

fashion, it should generalize over all four categories. In the X Convention this is
 3: A THEORY OF PHRASE STRUCTURE

possible only if all are of the same level. (See sections 4.1 and 7.2-7.3 for further
discussion of appositives.)
Certain predicates permit a PP to replace the subject it.

(3.20) a. It’s a long way to Tipperary. => To Tipperary is a long way.

b. It’s not too long from Groundhog Day to Purim. => From Groundhog
Day to Purim is not too long.

This PP movement appears to generalize with the movement of S into subject, called
variously Subject Replacement or Intraposition.6

(3.21) a. It’s no wonder that Harry’s so brilliant. That Harry’s so brilliant is

no wonder.
b. It’s a marvel that they ever get along. => That they ever get along is a
marvel.

Again there is no generalization of this sort between PP and VP. Thus the phrase
structure rule schema demands that PP and S be of the same level if this generalization
is to be captured.
PP and S generalize again in a rule which extraposes certain PPs from within the
subject and also extraposes relative clauses.7

(3.22) a. A review of that book appeared yesterday. A review appeared

yesterday of that book.
b. A man who was from Philadelphia came in. A man came in who
was from Philadelphia.

The rule of Gapping applies in conjoined Ss and NPs, with precisely parallel
conditions (see Jackendoff (1971) for details):

(3.23) a. Max plays saxophone and Medusa(,) sarrussophone.

b. Max’s recording of Klemperer and Medusa’s of Bernstein

There is no corresponding generalization with VP. Again, the way in which the rule
generalizes argues that S, not VP, should be of the same level as NP.
Thus there are numerous reasons to consider S structurally parallel to the other
major syntactic categories, arguing for the uniform-level theory.
One difference between Ss and NPs, pointed out by Emonds (1976), is that
subjects are obligatory in Ss but not in NPs. This fact is an apparent counterexample to
the claim in section 3.3 that only heads are obligatory constituents. Since Emonds too
accepts this claim, he concludes that the subjects of Ss are outside of the constraints of

6 I assume here the account of Emonds (1970), not that of Emonds (1976). Some discussion appears in
sections 4.3 and 4.8.
7 Or, if there is not an extraposition transformation, the generalization is in the projection rule. See
Akmajian (1975) for evidence that this rule applies also in NP, a further generalization of NP and S'.
44 RAY JACKENDOFF

the X Convention, being generated by Chomsky’s special phrase structure rule S —> N"
- V".
However, let us examine the constraints on S subjects a little more closely. It is
generally considered to be the case that those sentences which lack surface subjects,
namely /o-infinitives, do so because the underlying subject has been deleted or is the
phonologically null form PRO. However, at least two recent articles, Lasnik and
Fiengo (1974) and Brame (1975), have argued that certain of these constructions contain
surface infinitives with no deep subject. Their claim is that these infinitives are VP
complements in deep structure, not Ss. Such a solution violates the phrase structure
rule schema (3.9), since it involves generating a V" to the right of a head. But an
alternative is to generate them as subjectless Ss, i.e. as Ss which do not even contain a
PRO in subject position. This alternative is equally consistent with Lasnik and Fiengo’s
and Brame’s arguments. Then the phrase structure rules could generate an optional
subject in S, its obligatoriness under most conditions being due to conditions extrinsic
to the phrase structure rules. Whether such conditions can be independently motivated
must be left for future research; but such a solution simultaneously meets Emonds’s
objection and provides a way of generating so-called VP complements within the
present framework. Should this solution not prove viable, we can, of course, always
accept a weakening of the optionality condition to allow obligatory subjects.
A second difference between S subjects and NP subjects pointed out by Emonds
concerns a general constraint in English against embedding sentences in specifiers,
including subjects of NPs.

(3.24) a. *[a man that Bill knowsfs brother

b. *[six more than Bill saw] boys '—in SpecN
c. *a [proud to be an American] man

d. *[more than Mildred is] beautiful

|—in SpecA
e. *[six hours that we regretted] late
f. *[two bags of flour that we needed] more—in SpecQ
g. *[the man that you hatedfs destruction of the records—in SpecN (NP
subject)

Note that the bracketed phrases in (3.24) are grammatical in other contexts, and
that if the subordinate Ss are eliminated from the bracketed phrases, (3.24a-g) become
grammatical. Apparently the only exception in English to this restriction is subject
position in Ss, which obviously may contain a wide range of embedded sentences.
Emonds argues therefore that this embedding constraint would be more general if the
subject were not considered a part of the specifier of V, but were rather outside V", as
in Chomsky’s system. In a uniform level theory, on the other hand, the sentence must
be an exception to the embedding constraint.
 3: A THEORY OF PHRASE STRUCTURE

However, there are languages in which even subjects may not contain sentential
complements. One is Walbiri, as pointed out by Hale (1975). Interestingly enough,
another is Old English, as pointed out by O’Neil (1976), who studies how relative
clauses in subjects came to be possible in English through a structural reanalysis of
certain topicalized constructions. In order to maintain Emonds’s generalization, one
would have to make the rather dubious claim that the historical change in English
involved altering phrase structure in a radical way, moving the subject from V" into
V'". A much more plausible change is that the embedding constraint itself was relaxed
to permit the observed exceptions; this accords with O’Neil’s analysis. Thus, accepting
the nongenerality of Emonds’s embedding constraint does not seem at all to be a
consequence damaging to the uniform-level theory.

3.4.3. Treatment of the Complementizer

One problem that remains in justifying the Uniform-Level Hypothesis is the treatment
of the complementizer. The question is where the complementizer is attached to the
sentence. All the movement rules cited in section 3.4.2 move the sentence with its
complementizer, and no rule ever leaves the complementizer behind. Thus, the
simplest hypothesis is that the complementizer is a left sister of the subject, attached to
V" in (3.14a) or to V'" in (3.15a). Unfortunately, this hypothesis appears to run afoul of
the analysis of Gapping, for Gapping applies only if the complementizer is absent from
the second clause:

(3.25) a. It is hard to believe that Jack hates swimming and (*that) Fred
fishing.
b. For Jack to hate swimming and (*for) Fred fishing would be a
tremendous surprise.

One explanation of this is that the difference in structures is as shown in (3.26a) versus
(3.26b).

(3.26) a. S

Jack hates swimming Fred fishing

46 RAY JACKENDOFF

b. S

If Gapping were defined over S conjunction but not over S conjunction, the difference
would be immediately accounted for.
Other evidence against the theory that the complementizer is the sister of the
subject is provided by Bresnan (1974), who shows that Right Node Raising applies only
to single constituents, but that it can apply to sentences with or without their
complementizers:

(3.27) a. Mike wouldn’t tell us, but Randy readily volunteered, that Jenny was
drinking again.
b. I’ve been wondering whether, but wouldn’t positively want to state
that, your theory is correct.

Again this argues that the complementizer is a sister of the entire S, not of the subject.
This raises a difficulty for the generality of movement rules under the X
Convention: since S is one level higher than NP, i.e. apparently V4, it cannot be moved
by the same rule that moves an N'", a P"\ or an A'". There are at least two possible ways
to deal with this problem. One would be to claim that S is indeed V4 (assuming a three-
level S and NP), but that the A-over-A Convention actually generalizes over category
levels: the structural description of a rule would be met by the highest-level constituent
of the category mentioned by the rule. For instance, by mentioning V in a rule, the
structural description would be met by V4; by mentioning N, the structural description
would be met by N'". In order to pick out the lexical category verb, a rule would have
to mention specifically V°. This is a not unattractive solution, although some care
would be necessary in checking its feasibility. It has the interesting property of vitiating
all the above arguments for parallelism based on movement rules, since a rule
mentioning, say, +Subj will apply to either N'" or V4.8

8 Emonds has pointed out (personal communication) that this solution is in some sense a return to one
attractive aspect of Harrisian transformations: one would state, say, an NP movement rule as ‘‘move N”; all
the modifiers would be moved along automatically. This emphasizes the centrality of the notion “head” to
syntax in a highly suggestive fashion.
 3: A THEORY OF PHRASE STRUCTURE

A less drastic solution is to claim that both S and S are instances of V'":

(3.28)

Comp

N'" Aux V''

Under this proposal, the A-over-A Convention guarantees that any movement rule
moves S rather than S, while preserving the generalization with the movement of other
X'". Such a structure represents a weakening of the theory of phrase structure, since it
contains the forbidden configuration of a category dominating itself. Section 3.6 and
chapter 9 will show, however, that this exception can be accommodated within a highly
constrained class of rules, generated by a second phrase structure rule schema of
considerable interest.
Which of these two solutions for the complementizer turns out to be correct,
however, does not seem to be of major importance to the main hypothesis. For
simplicity, we will assume the second solution, in which the complementizer is
Chomsky-adjoined to S. This problem disposed of, the generalization of subjects
provides considerable evidence for the Uniform-Level Hypothesis.

3.5. The Auxiliary

The problem of this section is where the various parts of the auxiliary are attached and
whether they provide evidence for the general structure of the S. The position adopted
here is a modification of the analysis in Jackendoff (1972, section 3.8), based in part on
1966 lectures by Klima and on Emonds (1970); other revisions have been proposed by
Akmajian and Wasow (1975). In turn, these are all variants of the original analysis of
Chomsky (1957).
The verb and its complements apparently form a constituent that does not include
the modal, since there is a rule (pointed out by Ross (1967)) which fronts a VP after
modals under certain conditions:

(3.29) They said she may attempt to leave, and attempt to leave she will.

The rule of VP-Deletion provides evidence that the verb and its complements form a
constituent which does not include the aspectual verbs have and be, since a VP may
delete, leaving aspect intact.9

9 Hornstein (1975) presents this as evidence against the two-level theory of S, in which aspect must be a
daughter of V'. As will be seen, a three-level theory accommodates it.
48 RAY JACKENDOFF

(3.30) a. We asked them to stay, and they could have.

b. Michael said he had been practicing, but I don’t really believe he had
been.

Jackendoff (1972) gives evidence that the first auxiliary is a daughter of S, but that
subsequent auxiliaries are not daughters of S. The evidence is that sentence adverbs
such as frankly, probably, and evidently occur in all possible positions as daughter of
S—initial, final with comma intonation, and before the auxiliary. They also occur after
the first auxiliary, but not after subsequent ones.

frankly
I have amused
(3.31) a. George will probably ' the children by the time
[be amusing
evidently
we get there.
frankly
have .amused
b. ?*George will ' probably them when we get there.
be amusing
evidently

A simple way to account for this difference is to claim that the first auxiliary is a
daughter of S but that the second auxiliary forms a constituent with the verb and its
complements; then an adverb following the first auxiliary can be a daughter of S, but an
adverb following the second auxiliary cannot be.
These three arguments for constituency lead to a three-level theory of the S:

(3.32) V'"

One of the problems in Chomsky’s original analysis of the English auxiliary is the

M
recurrence of the configuration have in transformational rules as the first verbal
be
element, including all uses of be and the aspectual use (and in some dialects the
possessional use) of have. The solution of Klima, Emonds (1970), and Jackendoff
(1972) is an obligatory transformation Have-Be Raising which moves the appropriate
uses of have and be to a position under Aux, just in case there is no modal present.
 3: A THEORY OF PHRASE STRUCTURE

After this transformation, any transformation can refer to the first auxiliary by
mentioning the node Aux in its structural description, and the need for the repetitive
braces is eliminated.
McCawley (1975) objects to the rule of Have-Be Raising on the grounds that it is
entirely arbitrary to treat have and be as unlike other verbs in undergoing this rule; and
that if modals are a category separate from verbs, there is no reason for modals, have,
and be to behave alike. The theory of categories of sections 3.2 and 3.3, however,
makes possible a refinement of the analysis which meets McCawley’s objections in a
rather interesting way.
Recall that the syntactic distinctive features of Modal are [+Subj, +Obj, —Comp],
contrasting with verbs only in the last feature. Now suppose that the lexical entries for
the appropriate uses of have and be are assigned the syntactic features [+Subj, +Obj],
being unmarked for Comp. Then, by the usual conventions for applying rules, any rule
which applies to the feature complex for modals, [+Subj, +Obj, -Comp] will also
apply to have and be but to no other verbs. In other words, [ + Subj, +Obj] corresponds
to Choms cy’s (1957) category v; [ + Subj, +Obj, -1-Comp] to V; [ + Subj, +Obj, -Comp]
M
to have The exceptionality of have and be and their falling together with modals is
be
expressed naturally in terms of such a lexical feature analysis; and the description of
the auxiliary thus looks considerably less arbitrary than before.10
For a further refinement, observe that the Uniform Three-Level Hypothesis claims
that Modal is the progenitor of a series of larger categories M', M", and M'". The
category Aux, which so far has not played any role in the X Convention, can therefore
be analyzed as M"\ Although there is no empirical evidence that as many as three
levels are necessary, the superfluous structure is harmless and makes a more highly
constrained theory possible; furthermore, the extra rules are part of universal grammar,
and thus add no cost to particular grammars. We will arbitrarily attach Tense as a
daughter of M'" rather than at any of the lower levels, yielding the following phrase
structure rules:11

10 Wendy Wilkins has pointed out (personal communication) that this analysis brings to mind the
arguments in Chomsky and Halle (1968, 382-385) which show a possible illegitimacy in the use of partially
specified phonological feature matrices. However, examination of those arguments reveals that they depend
on the fact that phonological rules may alter distinctive features. Since, under the assumptions of the
Lexicalist Hypothesis, syntactic category features are never changed by transformations, the use of partially
specified syntactic features is not susceptible to Chomsky and Halle’s criticisms.
11 Akmajian (personal communication) has objected to this approach to the auxiliary on the grounds that
it is a distortion of the actual facts of English, and that there are auxiliary constituents in other languages
which show no evidence of a modal. The approach taken here, however, is forced on me by the highly
restrictive phrase structure rule schema (3.9). The quotation from Chomsky in the introduction to this
chapter is relevant here: either this analysis will prove that the theory must be weakened in a highly specific
way, or else it will provide a new analysis of what is otherwise an unprincipled exception to the general
theory.
50 RAY JACKENDOFF

Now since the phrase structure rule schema also predicts that a head is obligatory
in M'", the only way a modal can be omitted from a tensed sentence is to generate the
empty node A under M. Thus both Have-Be Raising and Do Support can be stated in
such a way as to fill this empty node with a category nondistinct from M, making one
further peculiarity of the English auxiliary somewhat less peculiar.12
One further minor point: the phrase structure rule schema (3.9) claims that all
complements and specifiers are optional, at least from the point of view of phrase
structure rules. This is certainly true of M'" in rule (3.33a), since in infinitive clauses
there is no evidence that there has ever been a tense or modal present. We can thus feel
justified in regarding the obligatoriness of Aux in main clauses as a semantic condition,
or at least as a condition extrinsic to the phrase structure rules proper.
It is a little more difficult to justify making Tense optional in (3.33b), since the
accepted syntactic treatment has always required it. Its obligatory presence seems in
any event to be a minor exception to the overall generalization that only heads are
obligatory in phrase structure expansions.13

3.6. Two Exceptions to the Uniform Three-Level Hypothesis

One obvious exception to the theory of phrase structure presented so far is coordina¬
tion. It is generally agreed that the node dominating conjoined Ss is an S, that the node
dominating conjoined NPs is an NP, and so forth.

12 If there are no auxiliaries, and Affix Hopping attaches Tense to the main verb, presumably the entire
M'" is deleted, so as to meet Emonds’s (1976) condition that no empty nodes appear in surface structure.
13 Emonds (1976) suggests that only phrasal categories are always optional; grammatical formatives are
optional or obligatory. This weakening accounts for at least the present cases properly.
 3: A THEORY OF PHRASE STRUCTURE

(3.34) a. S

mushrooms toadstools

This violates even the hierarchic arrangement of categories predicted by (3.1), the
primitive version of the phrase structure rule schema. Furthermore, no constituent of a
coordinate construction can be identified as its head. Clearly a separate phrase
structure rule schema is necessary. (3.35) is one possible form.

(3.35) X* X1 - (conj - X')*

This permits coordination of any syntactic category. Whether it can be refined to a

more restrictive form is left for future research.
A different sort of exception to the phrase structure rule schema emerges from
considering the structure of gerundive nominals such as Noam's inventing a new
theory. The arguments of chapter 2 showed that they are closely related to sentences,
having the same possibilities for modification as sentences and having interpretations
with direct and productive relationships to the corresponding sentences. Emonds (1970)
shows, furthermore, that (unlike that and for-to complements) they have precisely the
distribution of NPs. Hence a first approximation to their structure might be (3.36).

(3.36) N'"

Schachter (1976) and Horn (1975) independently suggest an alternative structure, in our
terms represented as (3.37).
52 RAY JACKENDOFF

(3.37) N'"

N'" N"

ANoam ing V"

invent a new theory

(3.37) has at least three immediate advantages over (3.36). First, it enables the rule
inserting ’s (POSS) on the subject of an N'" to be generalized to the subject of gerunds
without further ado. Second, it makes possible a certain limited class of gerunds such
as (3.38) (quoted from Schachter).

(3.38) a. There is no enjoying this world without thee.

b. This telling tales out of school has got to stop.

The replacement of the subject by a normal NP determiner is inexplicable in structure

(3.36) , but is an expected possibility in (3.37). Third, it explains why gerunds allow
aspect (a V" specifier) but not tense and modal (V"' specifiers), since gerunds contain a
V" but not a V"'. Horn presents further arguments that the subject of gerundive
nominals behaves like the subject of an NP, not an S, where these differ.
The differences between gerundive and derived nominals observed in chapter 2 can
still be explained in terms of (3.37), since they all depend on the difference between the
internal structure of V" and N", a difference which is still present under the theory of
(3.37) . Since the Lexicalist Hypothesis generalizes the subject relation to NP, Noam
can still be interpreted as the subject of the verb invent, at worst by ignoring the feature
+Obj at some point in the derivation. Hence (3.37) appears to be a viable alternative
theory of gerundive nominals, with a certain amount of evidence in its favor.
Observe, however, that neither (3.36) nor (3.37) conforms to the phrase structure
rule schema, since the main N'" has no noun head. Hence it is necessary in either case
to treat gerundives as an exception to the schema, by adding phrase structure rule
(3.39a) to generate (3.36) or (3.39b) to generate (3.37).

(3.39) a. N'" _h, v"'

b. N" -* ing - V

Either of these rules provides a way to use a verb phrase as a noun phrase, i.e. not
as a verb phrase with respect to the context in which it is embedded. I believe that
these rules are not an unprincipled exception to the phrase structure rule schema of
section 3.3, but are part of a class of “deverbalizing” phrase structure rules described
by the schema (3.40).
 3: A THEORY OF PHRASE STRUCTURE

(3.40) X! —> af - V*

That is, a deverbalizing rule must expand a category as the verbal supercategory of the
same level, and there may be no complements or specifiers other than an affix or other
grammatical formative indicating the change of category.
This class of rules has up to now not been distinguished, perhaps because there has
been no theory of phrase structure sufficiently constrained that rules such as (3.39)
emerged as exceptional. Chapter 9 presents a number of candidates for deverbalizing
rules, some of which are not at all obscure.
One particularly relevant example should be mentioned now. Recall that in section
3.4.3 we proposed that the complementizer be generated with rule (3.41).

(3.41) V'" -» Comp - V'"

As Emonds has pointed out (personal communication), this rule is an instance of

schema (3.40), where X happens to be V, and the complementizer is a particular kind of
affix or grammatical formative. Thus the apparently exceptional treatment of the
complementizer takes its place as one of a principled set of violations of the main
phrase structure rule schema.

3.7. Summary and Generalization of Rules

The X Convention as formulated in section 3.1 makes three claims: the class of possible
lexical categories is determined by a set of distinctive features; the class of syntactic
categories is determined by elaborating the lexical categories in terms of the prime
notation; rules of grammar are to be stated in terms of these features and primes. We
have argued that this theory can be strengthened into the Uniform Three-Level
Hypothesis: for every lexical category X, there are syntactic categories X', X", and
X'", and no more, and the major phrase structure rules elaborating these categories are
of the form given by rule schema (3.9), repeated here:

(3.9) Xn -* (Cj). . .(Cj) - X”-1 - (Cj+1). . .(Cfe), where 1 < n < 3, and for all C„
either C, = Y'" for some lexical category Y, or C, is a specified
grammatical formative.

One argument has appeared which bears on the need for three levels: the
constituent structure of the auxiliary requires a level for aspectual verbs between the S
level and the V' level (section 3.5). Further arguments appear in subsequent chapters.
Rule schema (3.9) provides the bulk of the phrase structure rules of the language,
but there are at least two other schemata which generate possible configurations ruled
out by (3.9): the schemata for coordination (3.35) and for deverbalizing rules (3.40).

(3.35) X' -> X' - (conj - X')*

(3.40) X1 —» af - V*

These provide places in universal grammar for the parts of speech which do not
54 RAY JACKENDOFF

participate in the X Convention, namely coordinating conjunctions, which appear in

(3.35), and complementizers, which appear as one realization of af in (3.40).
The theory leaves a number of questions in my mind. The first group concerns the
feature system. It is clear that a theory of markedness for syntactic categories is
necessary, which will predict the relative probability of parts of speech across
languages: nouns and verbs are presumably universal, adjectives and prepositions less
common, adverbs rarer, modals rarer still. Also, one of the drawbacks of the present
feature system is that it incorrectly predicts that generalizations across -Comp
categories should be as common as those across +Comp categories. Is this a question
of markedness? Or are there substantive universals outside the feature system (such as
the presence of Tense in sentences and deictic elements in NPs) which simply override
possible generalizations?
A second group of questions concerns the adequacy of the constraint on the form
of C, in schema (3.9). Is it necessary to weaken the claim that all C, are optional, and if
so, how? Is the restriction to Y'" viable? If not, what is a properly restrictive way of
weakening it?
Continuing with the summary of the chapter, here is the collection of phrase
structure rules generated by (3.9) that have been mentioned so far.

(3.42) X'" Rules

Traditional Notation X Notation
a. S —> (NP) - (Aux) - PredP V'" —*■ (N'") -JM"')-V"
(N'" - N"
bNP-(fc«pj> "Nom
N" C[Art )
c. Aux —» T - M" M' T - M"
(3.43) X" Rules
a. PredP —»(have - en) (be - ing) VP V" —» (have - en) (be - ing) - V'
b. Nom —> N' N" —> N'
c. M" -> M' M"—> M'
(3.44) X' Rules
a. VP^ V - (NP) V' —»• V - (N"')
b. N' —> N - (PP) N' —> N - (P'")
c. M' -» M M' —> M

The main generalization worth noting at this point is that between (3.42a) and
(3.42b), involving the position of the subject. These two rules can be generalized via
feature notation into (3.45).
nt
(3.45) X + Subj
+ Subj -Obj
- <(M"')> - X"
<+Obj> ( + Comp)
_+Comp _ +Det

As in phonology, the angle brackets indicate simultaneous occurrence: if +Obj is

 3: A THEORY OF PHRASE STRUCTURE

chosen on the left, +Comp must be and M"' may be chosen on the right; if —Obj is
chosen on the left. Comp is unspecified on the right, so either N'" or Art"' may be
chosen, and M"' may not be chosen.
Unfortunately, the clumsiness of this notation will turn out to be proportional to
the complexity of the rules being collapsed. (3.45) is a relatively simple case, and far
more unsightly cases will appear in the next chapter. We will eventually conclude that
the traditional abbreviatory conventions are not sufficient to express the linguistically
significant generalizations of English phrase structure. Insofar as the choice of
abbreviatory conventions is directly linked to the evaluation measure, this conclusion
raises deeper questions for linguistic theory than we will be able to answer here.
On the following page is a glossary of lexical and syntactic categories.
56 RAY JACKENDOFF

3.8. Glossary of Lexical and Syntactic Categories

Complete Traditional Category Name

Major Category Feature
Grouping X Analysis for X' for X" for X

+ Subj
V + Obj VP PredP S, S
+ Subj _ + Comp_
L+Obj J + Subj
M + Obj Aux
. — Comp.

+ Subj
N -Obj Nom Nom NP
_ + Comp_
~+Subj
+ Subj -Obj
Art Det
L-Obj J -Comp
_+Det
+ Subj
-Obj
Q QP
-Comp
-Det

-Subj
p +Obj PP
-Subj _+Comp_
L+Obj J -Subj
Prt + Obj
_-Comp_

-Subj
A -Obj AP
_ + Comp_
-Subj
— Subj -Obj
Deg Det (?)
.-Obj . -Comp
_+ Det
"-Subj
-Obj
Adv AdvP
-Comp
— Det
 4: Complements

The Uniform Three-Level Hypothesis of chapter 3 predicts that complements to a

phrase can be attached in three possible places: to X', X", or X'". Since any constituent
following the head of a phrase could in principle be attached in any of the three
complement positions, we must be concerned with how to pick out the correct position
for various sorts of complement phrases. This chapter will explore the complement
systems of each of the major categories V, N, A, and P. We will show that there are
principled distinctions among the three levels of complements, and that the comple¬
ment system of English reveals substantial cross-category generalization.

4.1. The Three Levels of Complements

If we classify complements on semantic grounds, we find that there are three distinct
ways in which a complement may be integrated into a semantic interpretation: as a
functional argument, as a restrictive modifier, and as a nonrestrictive modifier. We will
attempt to identify these respectively with X', X", and X'" complements. Thus we will
be able to claim that there is a strong correlation between syntax and semantics in the
complement system.1
Let us begin with functional arguments. Those lexical items which strictly
subcategorize phrases in their environment can be thought of as semantic functions
which take as their arguments the interpretations of the strictly subcategorized phrases.
For example, the verb give strictly subcategorizes a subject, an object, and an indirect
object, and can be thought of as a semantic function f(x,y,z) which maps ordered triples
of terms into propositions. Such an approach is developed in Katz (1966; 1972) and
Jackendoff (1972; 1976), among other works, and has been used implicitly in almost
every approach to semantic interpretation in generative grammar. Similarly, the noun
part (of) strictly subcategorizes an NP, and can be treated semantically as a function
g(x) which maps terms into terms; the adjective proud (of) maps terms into properties

1 Williams (1975) comes to similar conclusions about the complement system, on the basis of somewhat
different evidence. Readers familiar with his work will notice that I have only three complement categories
instead of his four. His types I, II, and IV correspond to X', X", and X'" complements, respectively. His type
III, the most problematic for him, appears to divide into X" and X"' complements in a way not fully clear to
me.
58 RAY JACKENDOFF

(or predicates, or whatever kind of object the intension of adjectives is); the preposition
to maps terms into (intensions which pick out) directions. Aside from subjects, all
functional arguments in English come immediately after the head, typically preceding
all other modifiers, and many grammarians have had the intuition that they are the most
“tightly bound” to the head of all the complements. This intuition can be expressed by
assigning all and only the strictly subcategorized phrases to the X' complement in deep
structure.
In many cases a phrase can be identified as strictly subcategorized by the fact that
it cannot be omitted from the sentence without incurring ungrammaticality. For
example, the PP in Joe put the book on the table must be strictly subcategorized, since
*Joe put the book is ungrammatical. However, this is only a sufficient condition for
strict subcategorization, not a necessary condition, since many words optionally
subcategorize phrases. For example, tell occurs in the contexts John told Bill a lie and
John told Bill, but a lie appears to operate semantically as a functional argument in the
former sentence and hence must be strictly subcategorized. Similarly, when adjectives
and nouns strictly subcategorize, their arguments are usually optional.
A second criterion for X' complements in Ss and NPs is provided by certain
anaphoric processes. The phrase do so appears to be a pro-V', and may be followed
only by material which is outside the V' complement (usually part of the V"
complement).2 Consider (4.1).

(4.1) a. Joe bought a book on Tuesday, but Sim did so on Friday,

b. *Joe put a book on the table, but Sim did so on the chair.

The ability of on Tuesday to follow did so indicates that it is a V" or V"' complement in
this sentence; the inability of on the chair to follow do so indicates that it must be
inside of V' in the antecedent sentence.
A parallel anaphoric process in NPs is the use of the pro-N' one. Lakoff (1970a)
observes that there is a contrast in the applicability of one which depends on the nature
of the complement.

(4.2) a. Jack met the king from England, and I met the one from France,
b. *Jack met the king of England, and I met the one of France.

Notice that the PPs of England and from England play different semantic roles, since
there may be a king of England from France. Of England appears to be a functional
argument, since it specifies part of the function of the king; from England, on the other
hand, specifies a somewhat inessential part of kinghood. Notice further that the two
PPs may appear together only in one of the two possible orders: we cannot get *the
king from France of England. These facts can be described simultaneously if we
suppose that of England is an N' complement and from England is an N" complement.

2 This distinction in the use of do so was first observed by Lakofif and Ross (1966). 1 assume that the
interpretation of do so will be carried out by focus-dependent anaphoric processes such as Akmajian (1973)
describes.
 4: COMPLEMENTS

An N" complement would automatically follow an N' complement; the different

positions would correspond to different semantic roles.

Based on this reasoning, the proper formulation of the difference between (4.2a,b)
appears to be this: the pronoun one cannot be followed by the phrase of NP within the
N' complement. It can, however, be followed by other N' complements, as will be seen
shortly.3 Thus, the inability of a particular of NP phrase to follow the pronoun one
(where the antecedent is count) is a sufficient test for its being an N' complement. (4.4)
is the forbidden configuration.

(4.4) N'"

This test requires some care. At first glance it would appear that of water must be
an N" complement in (4.5), since it follows one.

(4.5) Bill has two quarts of wine and one of water.

However, we notice that one can be pluralized in (4.2a) but not in (4.5).

(4.6) a. I met the ones from France.

b. *The quarts of wine and the ones of water were left behind.

31 am grateful to Noam Chomsky for pointing this out to me. See footnote 4.
60 RAY JACKENDOFF

The difference is explicated if we realize that there are two morphemes one that can
function anaphorically in an NP, a numeral (a kind of Q, alternating with two, three,
etc.—see section 5.5) and a pronoun. Only the latter, which takes the place of the head
of the NP, can be pluralized to ones. (4.6b) shows that the one in (4.5) is the numeral
oneQ, and that in fact the pro-N' oneN cannot be followed by the complement of water
here. We conclude therefore that of water is indeed an N' complement in (4.5), and so,
by parallelism, is of wine.
There are certain cases that appear to be ambiguous between N' and N"
complements without appreciable difference in meaning. For example, because of the
parallel between the direct object in (4.7a) and the PP in (4.7b), one would want to
claim that the PP is an N' complement.

(4.7) a. Bill pictured Fred.

b. Bill’s picture of Fred

Yet the pictures of Fred and the ones of Harry is also acceptable, arguing that of Fred
must be an N" complement, similar to the pictures which are of Fred. The simplest
solution is to accept both sources for such a case.
Another test to distinguish N' from N" complements is based on a distinction
noticed by Lakoff (1970b):

(4.8) a. Fathers of few children have any fun.

b. *Fathers with few children have any fun.

Notice that of few children is, by our previous criterion, an N' complement, since
*ones of few children is ungrammatical. But with few children can be paraphrased by
the relative clause who have few children, suggesting that it is an N" complement. This
conjecture correctly predicts that only one order of the two complements is possible,
i.e. fathers of few sons with many daughters but not *fathers with many daughters of
few sons. To describe the difference in grammaticality between (4.8a,b), then, one is
led to the hypothesis that a quantifier may extend its scope out of an NP dominating it
if it is in the N' complement but not if it is in the N" complement. Thus any could be in
the scope of few in (4.8a) but not in (4.8b); since any must be within the scope of
negation, (4.8b) is unacceptable. I can offer no reason why this distinction should exist,
but it is confirmed by considering the possibility of vr/7-questions and relatives:
according to the previous criterion, only N' complements can be wh-ed.

(4.9) a. Fathers of which children had fun?

I met some children the fathers of whom like to drink,
b. *Fathers with which children had fun?
*1 met some children the fathers with whom like to drink.

Another pair with this contrast is (4.10).

 4: COMPLEMENTS

(4.10) a. Arguments with few people yield any satisfaction.

Arguments with which people satisfy you?
He is a person arguments with whom are fruitless.
b. ?*Arguments with few premises yield any satisfaction.
?*Arguments with which premises satisfy you?
?*This is a premise arguments with which are useless.

(4.10b) is all right if it is read arguments against which premises, parallel to arguments
with which people, but impossible if read arguments employing which premises. It is
plausible to assume that a person or premise against which an argument is directed is
strictly subcategorized, i.e. an N' complement, but that the instrument of argument, the
premise employed, is an N" complement like other instrument phrases. This is
supported by the grammaticality of arguments with Bill with few premises and the
ungrammaticality of *arguments with few premises with Bill. Thus, this distinction
between complement types supports the proposed interaction between level of comple¬
ment and scope of quantifiers and wh; we see that inability of quantifiers or wh to
extend scope outside the NP seems to be a sufficient condition for N" rather than N'
complements.
Since there has been little study of a possible distinction among different kinds of
NP complements, I am not familiar with enough strong cases to completely verify the
validity of the two tests described above, much less to explain why these particular
distinctions should obtain. However, the proposed distinction between N' and N"
complements does seem to be fairly well borne out by the few examples cited here.4
The distinction between X" complements and X'" complements is somewhat more
straightforward. In sentences, the V" complements are the expressions of manner,
means, accompaniment, instrument, purpose, and other so-called VP adverbials.
Semantically, they map predicates into predicates of the same number of arguments,
and they contribute to the main assertion of the sentence. As such, they can be
focused, clefted, and affected by sentence negation:

(4.11) a. John hit the nail softly.

b. It was with a hammer that John hit the nail.
c. We didn’t buy this for your benefit.5

Because they add extra truth conditions to the assertion of the sentence, restricting the
extension of the sentence, V" complements can be called restrictive modifiers.
V'" complements, by contrast, add no conditions to the assertion of the sentence,

4 Notice, by the way, that (4.10) provides evidence that the ones test is valid only for of-NP
complements, not for other PPs in N', since both of the following examples are acceptable.

(i) Arguments with Bill are less fruitful than ones with Harry. (N' complement)
(ii) Arguments with many premises are less impressive than ones with few premises. (N" complement)
5 Just because they are in V" and focused does not mean they are automatically affected by negation.
Much depends on intonation. Cf. Jackendoff (1972, sections 6.7 and 8.6).
62 RAY JACKENDOFF

but rather add some sort of auxiliary assertion (one of whose arguments is usually the
main assertion). They include sentence adverbials of all sorts, sentential appositives,
parentheticals (e.g. John is a fink, i think), and various other subordinate clauses.
When they occur at the end of a sentence, they are set off by comma intonation. Since
they are not part of the main assertion of the sentence, they cannot be focused, clefted,
or affected by sentence negation:

(4.12) a. *John hit the nail, of course.

b. *It was J Pr°t)ably I j0jin the najj
(in my opinion J
c. *John didn’t hit the nail, I think, (cf. John didn’t hit the nail, I don’t
think.)

The geometry of the sentence predicts that V'" complements must follow V"
complements, and this prediction is obviously borne out (except for some stylistic
inversions).

(4.13) John hit the nail softly, of course.

?*John hit the nail, of course, softly.

?John ran away quickly, probably.

*John ran away, probably, quickly.

John hit the nail with a hammer, which surprised no one.

*John hit the nail, which surprised no one, with a hammer.

John hit the nail with a hammer, I think.

?John hit the nail, I think, with a hammer.6

Similar contrasts can be found in NPs. The clearest example is the difference
between restrictive and nonrestrictive (appositive) relative clauses; the former follow
previous modifiers without a break, may contain foci, and may be affected by sentence
negation; the latter are separated by comma intonation, may not contain foci, and may
not be affected by sentence negation.

(4.14) a. I didn’t see the man who brought the strawberries.

b. *1 didn’t see the man, who brought the strawberries.

We would like to claim, therefore, that restrictive relative clauses are deep structure N"
complements and that appositives are deep structure N'" complements. But since many
proposals exist for deriving relative clauses from other sources, we will delay the
defense of this claim to chapter 7, where more space can be devoted to it.

s The insertion of parentheticals into V" may be what Banfield (1973) calls a stylistic transformation, one
of a class of rules with somewhat different properties from ordinary transformations. Alternatively, Emonds
(1976) gives an extensive treatment of parentheticals in which the complement is moved to the right of the
parenthetical.
 4: COMPLEMENTS

However, other constructions besides relative clauses can appear in N" and N'"
complements. For example, the PPs in the king from England, the weather at 6:00, the
man with a big nose, etc., are not in the N' complement, follow the head without a
break, and can be focused and affected by sentence negation.

the king from England

(4.15) We didn’t mention the weather at 6:00
the man with a big nose

They are thus candidates for N" complements. Descriptive adjectives, despite their
prenominal position, have similar semantic properties to these PPs, arguing that they
are attached to N" (see section 7.4 for more discussion). This of course makes them
parallel in structure and function to preverbal VP adverbs, which are in V", consistent
with the predictions of the X Convention.
By contrast, there are certain nonsentential appositives which can be characterized
as N"' complements:

(4.16) a. I will sell you these bagpipes, the finest in all Poland, for only 4000
zloty.
b. Perhaps you have heard of my brother, known the world over as a
notorious womanizer.
c. She presented Picasso, then in his blue period, with a blueberry pie.

One might want to argue that all these are reduced nonrestrictive relative clauses, and
this is certainly conceivable. In any event, they have the characteristic comma
intonation and inability to be affected by sentence negation which we identify with an
X'" complement.
Again, the N'" complements must follow the N" complements, as predicted by the
geometry of the NP:

(4.17) the man that brought the strawberries, who was dangerous
*the man, who was dangerous, that brought the strawberries

these bagpipes from Poland, the finest known

*these bagpipes, the finest known, from Poland

the man with a big nose, then in his blue period

*the man, then in his blue period, with a big nose (only ok as two
appositives)

We see therefore that there are three distinct kinds of complements in both
sentences and NPs, corresponding to the three levels predicted by the phrase structure
rule schema. Furthermore, the parallels in structure and function across the two
categories S and NP are clear and confirm the view of grammatical parallelism
advocated in chapter 3.
In the remaining major categories, AP and PP, the complements are less produc-
64 RAY JACKENDOFF

tive, but the three types are distinguishable. Both adjectives and prepositions strictly
subcategorize arguments, as was pointed out above. Most of the modifiers at the A"
and P" level are part of the degree system, which precedes the head; they will be
treated separately in chapter 8. At the A'" and P " level, there are appositive clauses
again, as usual marked by comma intonation.

(4.18) a. Martha is proud of her height, which you’ll never be.

b. We went from Aspen to Denver, which seems like a long way, in less
than four weeks.

As usual, the order of the appositive and the strictly subcategorized arguments cannot
be reversed:

(4.19) a. *proud, which you’ll never be, of her height

b. *from, which seems like a long way. Aspen to Denver
*from, Aspen to, which seems like a long way, Denver

Thus the facts of APs and PPs are consistent with the Three-Level Hypothesis,
although they do not push it to its limits as do Ss and NPs.
This section has shown that the three levels of complements made possible by the
theory of phrase structure of chapter 3 can be rather clearly distinguished, both
syntactically, through ordering and anaphoric behavior, and semantically, through the
means of integration into the interpretation. We have seen that there is strong evidence
for all three levels in S and NP, and for at least two levels in AP and PP.
The rest of this chapter will be devoted to the details of X' and X" complements,
concerning itself in particular with what cross-category generalizations appear in their
phrase structure. We begin with V' and N', the most elaborate and best-known
structures, then move to V" and N" complements in section 4.3. Section 4.4 deals with
the complements of A, Adv, and P, and section 4.5 discusses how the rules generalize.
The appendices deal with further problems of NP and S complements.

4.2. The V' and N' Complements

This section will be concerned with showing how a number of important syntactic
differences between Ss and NPs stem from differences in the phrase structure rules for
V' and N'.

4.2.1. Easy Parts of V'

A first approximation to the phrase structure rule for V' is (4.20).

(4.20) V' —» V - (NP) - (Prt) - ({^p}) - ({AQpP}) “ (pp) ~ (pp) - (S)

The verb and direct object are obvious. Skipping the next three positions for a moment.
 4: COMPLEMENTS

the two PPs are to provide for the strictly subcategorized arguments in constructions
like (4.21a). To see the need for both PPs, note that neither PP can follow do so
(4.2lb,c), although (4.2Id) shows that the verb talk can in other cases serve as the
antecedent of do so. We conclude that both PPs in (4.21a) are in V'.

(4.21) a. John talked to Bill about Harry.

b. *John talked to Bill about Harry, but he didn’t do so about Fred.
c. *John talked about Harry to Bill, but he didn’t do so to Fred.
d. John talked to Bill about Harry on Sunday, but he didn’t do so on
Thursday.

[AdvP]
Return now to the position. It is needed for the strictly subcategorized
LQP J
phrases in constructions like these:

. . ., , . I handsomely I
(4.22) a. Thejobpa,d(us)|too|.n|e j.

b. Bill worded the letter carefully.

Note that the AdvPs and QP in (4.22) cannot be omitted without incurring ungrammati¬
cally, and that, unlike most such phrases (cf. (4.23c)), they cannot follow do so.

(4.23) a. ?*The job on Tuesday paid us jj^ijtti™6^ j ’ ^Ut one °n

Wednesday did so jvery Poorly j
(enough J
b. ?*John worded his speech carefully, but Fred did so carelessly.

c. John talked about Martha J thoughtfully I but precj did

(quite enough J
I inconsiderately I
so
[far too much J'

Thus the adverbs in (4.22) pass the tests for V' complements.
The S in (4.20) represents the subordinate clause position in sentences like No one
would claim that Ss are NPs. There is no doubt that such an S is strictly subcatego¬
rized; it cannot be omitted, and it cannot follow do so. Furthermore, it is little
remarked that although the S must follow all other strictly subcategorized phrases, it
may be followed by manner and degree phrases.

(4.24) a. He claimed that Ss are NPs in a loud strident voice,

b. He believes that Ss are NPs quite fervently.

We will suppose, therefore, that these Ss are daughters of V' in deep structure, and
66 RAY JACKENDOFF

that their more usual surface position at the end of V" is the result of an extraposition
transformation, perhaps the same one that extraposes heavy NP direct objects.7

4.2.2. Particles and Predicates

Returning to the two positions after the direct object, we first find Particle.8 Emonds
(1972a) argues that in verb-particle constructions such as (4.25), the separated form
(4.25a) is underlying and the form with the particle next to the verb (4.25b) is derived.

(4.25) a. Howard looked the answer up.

b. Howard looked up the answer.

The principal objection to Emonds’s position is that look up is a lexical item and thus,
in the standard theory, must be inserted under a single constituent in deep structure;
(4.25b) but not (4.25a) can fulfill this requirement on deep structure. However, Emonds
points out idioms like take NP to task and take NP for a ride ('dupe NP’), which (like
(4.25a)) are discontinuous and which therefore provide independent evidence against
monocategorial lexical insertion. This vitiates the argument against (4.25a) being an
underlying form.
Following the particle is what Emonds has called the Predicate position. This
constituent appears in sentences like (4.26).

(4.26) We elected Harrison chairman.

We named the baby Hortensia.
We painted the bam red.
You make me sick.

It has been argued from time to time that these sentences have a deleted complement to
be. However, such sources as *We named the baby to be Hortensia, *You make me
(to) be sick, the alleged deep structures, are implausible, since elect, name, and paint
(at least) never appear at all with an infinitive complement. Thus, given the choice of
adding a transformation or extending the base to account for these verbs, the latter
solution appears better motivated. In fact, we can then claim that the copula involves
the Predicate position in John is a boy and John is sick, capturing the semantic
generalization between (4.26) and the copula directly in the formulation of the base and

7 As ever so slightly suggestive evidence for this extraposition, notice that no intonation breaks are
possible in (4.24), but pauses may be inserted in the extraposed forms:
(i) He claimed(,) in a loud strident voice(,) that Ss are NPs.
(ii) He believes(,) quite fervently(,) that Ss are NPs.
And even when the pauses are not inserted, the intonation in these forms is somewhat more “leisurely”.
Since such pauses are characteristic of deletion sites or traces (ct. Selkirk (1972)), (i)-(ii) must be the derived
forms, not (4.24).
8 According to the phrase structure rule schema, this should probably be Prt'", with no complements or
specifiers in the expansion of Prt'" down to Prt. Since universal grammar provides this much of the phrase
structure rules, using Prt'" instead of Prt in (4.20) incurs no extra cost to the particular grammar of English.
We will make the revision explicit in section 4.4.
 4: COMPLEMENTS

projection rules, without recourse to a transformation. This makes be a verb like any
other in its predicative uses, avoiding the extra phrase structure rule of Syntactic
Structures and Aspects.
Emonds (1970) claims that the Predicate position is dominated by a node Pred
which is in turn dominated by VP. Such an analysis would violate our theory of
categories, of course. I believe, however, that postulating this node is as much a
mistake as postulating a node Agent. Rather, Predicate, like Agent, is a semantic
relation which may be assigned by the projection rule for the NP/AP position in the
VP.9 As syntactic evidence that the second NP node is not distinguished by a
dominating node Pred, we offer a few expressions which use the second NP position
for other than a predicate, and which do not appear to be derived by any sort of dative
shift (these examples were suggested to me by Bowers (1971)):

(4.27) a. Margaret struck

{Joe a heavy blow
*a heavy blow to Joe

b. I envy
{you your beauty
*your beauty to you

c. He called
(her names
names to her (wrong meaning)

We need some justification for placing the Prt position before the Predicate. One
piece of evidence comes from sentences like (4.28), with a predicate AP.

(4.28) a. ?*They painted up the barn red.

b. They painted the barn up red.
c. *They painted the barn red up.

Clearly the best position for up is between the direct object and the predicate. I know
of no cases which contain a direct object, a particle, and an NP predicate; but John
grew up a Catholic is suggestive, in that the particle cannot be interchanged with the
predicate as it can with direct objects.10

9Jackendoff (1976), following Gruber (1965), characterizes this semantic relation as Identificational
Location, showing how it fits into a general account of functional semantic interpretation.
10 In Jackendoff (1974a) it was claimed that a Catholic is not strictly subcategorized in this example.
However, it cannot follow do so as can the similar construction to which it was compared there:

I
(i) *John grew up a Catholic and Bill did so a Moslem.

drunk
a saint
having forgotten who he was J
f, but Bill did

sober
so ja sinner
[in full knowledge of the hazards
Though the NP case in (ii) is somewhat worse than the others, it seems better than (i). The evidence that Prt
precedes predicate NP is admittedly inconclusive; but there is no evidence at all against it. This is a case
where we can let the theory decide.
68 RAY JACKENDOFF

Another bit of evidence for the base configuration V-NP-Prt-NP is that it permits
a simple structure-preserving analysis of Particle Movement and Dative Shift. Notice
first that if there is a single object, it can be generated in either NP position, i.e. on
either side of the particle. Hence the difference between (4.25a) and (4.25b) need not be
the application of a transformation at all: it may be due to the deep structure option as
to object position.
Next, observe that when there is a direct object, a particle, and a PP in the
complement, the object may as usual be on either side of the particle, as in (4.29). But,
as Emonds (1972a) points out, in a double-object complement, the particle must fall
between the two objects, as seen in (4.30).

(4.29) a. The secretary sent a schedule out to the stockholders.

b. The secretary sent out a schedule to the stockholders.
(4.30) a. The secretary sent the stockholders out a schedule.
b. ?*The secretary sent out the stockholders a schedule.
c. *The secretary sent the stockholders a schedule out.

This constraint can be accounted for if Dative Shift is a structure-preserving rule which
moves the object of to into the NP position directly after the verb (lower case denotes
an empty node):

(4.31) Dative Shift11

X - Va - np - (Prt) - NP - [PPto - NP] - Y r>
1 — 2 — 7 — 4 — 5 — </)—</>—8
OPTIONAL

According to this statement of Dative Shift, the structural description of the rule
can be met only if the first NP position in V' is empty, that is, if the direct object is
generated in the position following Prt. Thus, the configuration (4.30a) is the only
possible ordering of the two objects and the particle under this analysis. In turn, the
analysis justifies the base configuration V - NP - Prt - NP in (4.20).
Finally, there are some possible improvements in the statement of (4.20), making

11 In this rule Va denotes the class of verbs that govern Dative Shift.
Emonds (1972a) describes these facts with a Dative Shift that actually interchanges the two objects. The
present analysis is better in that it does not violate the generalization that transformations move only one
constituent at a time.
(4.31) is however sufficient to describe only Emonds’s Dialect B, which happens to be mine. For the
other two more complex dialects he describes, a movement of Prt to postverbal position will still be
necessary. Since Prt is not a phrase node, this rule does not fall under the structure-preserving constraint, but
rather under Emonds’s notion of a “local transformation”.
An interesting feature, perhaps an advantage, of this analysis is that it permits Dative Shift to be
governed by deep structure strict subcategorization rather than the traditional rule feature. Verbs which
permit Dative Shift strictly subcategorize an optional empty second NP, permitting the structural description
of (4.31) to be met. Verbs which do not allow Dative Shift will omit the Predicate NP in their strict
subcategorization. Thus, in fact, Vacan be replaced simply by V in (4.31), if it is the case that all verbs which
allow Dative Shift also allow Particle Movement in the presence of a particle.
 4: COMPLEMENTS

use of feature notation. The predicate position, can be rewritten as [-Obj,

(AdvP
Qp
[
| position can be rewritten as [-Obj, -Comp, -Det]'". Since it

appears that strictly subcategorized adverbs and QPs as in The job paid handsomely/
enough never cooccur with the predicate position, these two positions can be collapsed
into one (syntactically, though not semantically) with the feature complex [-Obj,
-Det]'". Similarly, there never seems to be a need for two PPs and an S all at once. If
this is the case, the last two terms in (4.20) can be collapsed into [ + Obj, +Comp]"'.
Thus, the final form of the base rule, using feature notation where necessary, is (4.32).

-Obj + Obj
(4.32) V' V - (NP) - (Prt) - ( ) - (PP) - ( )
-Det +Comp

4.2.3. The N' Complement

This subsection will show that certain differences between the phrase structure of V'
and N' account for a number of further divergences in their syntax. The expansion of
N' is at most (4.33).

(4.33) N' -> N - (Prt) - (NP) - (PP) - (

Again skipping the Prt and NP for just a moment, we see that two PP positions are
necessary for phrases like our talk about the war with Bill, paralleling the two PPs in
the V' of We talked about the war with Bill. The S in (4.33) appears in such
constructions as Frank’s claim [to Tony) that Sim was the culprit, where the S
corresponds to the one in the parallel V' of Frank claimed (to Tony) that Sim was the
culprit.12 There never seems to be a need for two PPs and an S at once, so I have
collapsed the second PP with the S; in feature notation the combination would be
identical to the one at the end of (4.32). Returning to the Prt, in the rare cases when
there is a particle in N', its position is always immediately after the head, e.g. Bill’s
looking up of the information, not *Bill’s looking of the information up.
There remains the NP position. In surface structure, of course, N cannot be
followed by an NP object or predicate. An N' corresponding to a V' with a direct
object generally contains of (although there are exceptions such as doubt).

12 Note, by the way, that the S at the end of N' confirms Emonds’s (1970) hypothesis that that and for-
to complements may be generated directly under V', rather than being extraposed there from under NP, as
proposed by Rosenbaum (1967). The Lexicalist Hypothesis claims that Frank claimed that Sim was the
culprit and Frank’s claim that Sim was the culprit must have parallel structures. In Rosenbaum’s
complement system, the deep structures are [sFrank past [VPclaim [NPit [gthat Sim was the culprit]]]] and
[NPFrank claim [gthat Sim was the culprit]], clearly nonparallel structures. To solve this difficulty, one must
assume either that both Ss are in [NPit S] constructions or that they are both attached to X'. As Emonds’s
arguments motivate the latter configuration, it is the structure of V', not of N', that must be changed from
Rosenbaum’s system.
70 RAY JACKENDOFF

(4.34) destroy the city/destruction of the city

explain the result/explanation of the result
establish a claim/establishment of a claim
etc.

Verbs which take PP complements generally preserve the preposition in the derived
nominal:

(4.35) persist in being stubborn/persistence in being stubborn

pray for war/prayer for war
escape from despair/escape from despair
remind him of his appointment/reminder of his appointment
lust for power/lust for power
approve of your behavior/approval of your behavior

Since in these cases the form of the complement is preserved in derived nominals,
Chomsky (1970a) suggests (p. 41-42/204 and note 27, p. 42/note 28, p. 219) that the
same is true of direct objects in underlying form: they occur in both V' and N' without
the preposition, and an obligatory transformation inserts of to produce the surface
form.13 Note that this of cannot be the preposition of, since there is no PP structure into
which it is inserted. Rather, it must be a specified grammatical formative like poss.
Another instance of the specified grammatical formative of will appear in section 5.4.
The transformation inserting of is therefore to be stated as (4.36).

(4.36) 0/-Insertion
[n.N - (Prt) -NP-X] ^>l-2-o/ + 3- 4
OBLIGATORY

13 Note that of is the only formative inserted. In his doubts about the proposal, about will be present in
deep structure, and about the proposal is a PP. The fact that the verb and noun do not agree in their
complement structure must be entered in the lexicon as a complication of their normal verb-noun pairing.
Chomsky’s footnote cited above defends the o/-insertion as opposed to an o/-deletion in VP. A further
argument appears in the discussion of the passive, section 4.7.
Jackendoff (1969c) argues that possessive constructions like a picture of John's are a kind of partitive,
analogous to one of John's pictures. If this is the case, the of is inserted by (4.36). Since 0/-Insertion applies
only to the NP immediately after the head noun, the genitive phrase must be the leftmost element. This
explains Chomsky’s judgment (p. 46/206) of the “quite clumsy phrase the portrait of Aristotle by Rembrandt
of the Metropolitan Museum's”, since there is no way for the grammar to generate it. If by Rembrandt is
omitted, the ungrammaticality is even clearer. Also, note the contrast of a portrait of the Met's by
Rembrandt and ?*« portrait by Rembrandt of the Met's.
Jackendoff (1974a) takes a somewhat different view of the inserted of: it is not a specified grammatical
formative, but a real preposition which fills an empty head of a PP complement. This analysis was necessary
in part because the notion of specified grammatical formative was not included in the theory of phrase
structure assumed there. The present analysis considerably simplifies the generalization of the grammatical
relation object, and is much closer to Chomsky’s account.
To distinguish between these two theories, one would ideally like to use extraction to test the
constituency of o/-NP. But the results are inconsistent. W/i-Preposing cannot carry the of along: *Of which
boys did you meet the fathers? But, as will be seen in section 5.3.1, Extraposition from NP does move o/-NP.
arguing against the present view. On the other hand, Akmajian and Lehrer (1976) show that certain apparent
extractions of o/-NP must be base-generated rather than extracted, so perhaps Extraposition from NP is such
a case too. I leave the question open.
 4: COMPLEMENTS

Thus, the underlying form of the nominals in (4.34) will be destruction the city,
explanation the result, establishment the claim.
The absence of an NP position preceding Prt in (4.33) explains why Dative Shift
and Particle Movement do not occur in N'. We have analyzed Particle Movement in V'
as the choice between two underlying phrase structure positions for the direct object,
one on either side of the particle. Since N' contains only the NP position after the
particle, the other ordering is impossible and the object can only follow Prt. Similarly,
Dative Shift in V' requires a structure-preserving movement of the indirect object into
the NP position before Prt. Since N' contains no such position, Dative Shift cannot
apply.
Furthermore, the absence of Predicate complements like *the election of John (of)
President follows from the fact that there is only one bare NP position in N': a
construction parallel to the V' elect John President would require two NP positions in
N' and there is only one. The closest parallel uses a PP with as: the election of John as
President. Thus, our analysis of the phrase structure of N' gives a principled
explanation of three apparently unrelated differences between V' and N' on the basis of
the N' lacking the first NP position.
Having justified base rule (4.33), we now address ourselves to how it generalizes
with V'.

4.2.4. Generalization ofV' and N' and the Definition of Object

Here are the two phrase structure rules we wish to relate.

-Obj +Obj
(4.32) V' V - (NP) - (Prt) - (
+ Comp_ )
(PP) - (
_ —DetJ ’
+ Obj
(4.33) N' N - (Prt) - (NP) (PP) - ( )
-l-Comp

All positions are parallel, except for two: N' lacks the first NP position, and allows
only an NP where V' allows NP, AP, QP, or AdvP. Using subscripted angle brackets,
the two rules can be collapsed as (4.37), which takes only three features more than the
rule for V' alone.14

X
+ Subj \
4-Subj
4-Comp/2 +Obj
(4.37) (+Obj )x X- ((NP))1-(Prt"')- - (PP) - ( )
-Obj 4-Comp _
(-Obj >2
-Det
_ + Comp_

We would also like to develop a definition of the grammatical relation “object-of”.

In terms of the Aspects notation for grammatical relations, the generalized object

14 We change Prt to Prt'" as suggested in note 8, page 66; this will be relevant in section 4.4.
72 RAY JACKENDOFF

relationship should be stated as [N'", [ + Subj]'], that is, in a string exhaustively

dominated by [ + Subj]', the object is a substring exhaustively dominated by N"', and
that occurrence of Nis directly dominated by the [ + Subj]'. While this would give the
correct results in N', it would incorrectly pick out both the object and the Predicate in
the V' elect John President.
It is not clear to me whether this is a serious problem. But if it is, the definition of
object can be sharpened to include order as well as domination relations:

X
(4.38) In , the object is the N'" immediately dominated by X' in
_+Subj
the configuration [VX - (Prt) - N'" - Y].

Note that this definition allows an NP on either side of Prt to be designated as object,
as long as there is no nonempty NP between it and the head.
For completeness, we should also define the grammatical relation Predicate.

(4.39) In V', the Predicate is the italicized phrase immediately dominated by V'

in the configuration [v-V - (N'") - (Prt) - f- Y].

Some Predicates will of course be structurally indistinguishable from some direct

objects, for example, Bill became a butterfly (predicate) vs. Bill saw a butterfly
(object). But since the choice is determined by the verb, no structural difference is
needed. The same two interpretations are possible within a PP, as in such a minimal
pair as The coach turned into a driveway vs. The coach turned into a pumpkin. Again
the possible interpretations of turn can determine which way to read the NP without a
structural difference. Hence, it should not be disturbing that (4.39) defines a set of
positions which overlaps partially with the set of positions defined by (4.38). In a
particular reading of a particular sentence, only one grammatical relation will be
applicable to an NP in one of the ambiguous positions, and that grammatical relation
will be established by the verb.
This completes our discussion of the phrase structure of V' and N'. We have seen
that they are parallel, except for one crucial position which N' lacks. The missing NP
position has been used to explain a number of further differences. Section 4.7 is an
appendix which discusses some movement rules in and out of V' and N'. We now go
up one level to V" and N".

4.3. The V" and N" Complements

As pointed out in section 4.1, the X" complement contains restrictive modifiers. In V"
these are AdvPs and PPs of manner, means, time, instrument, and so forth, as well as
various adverbial clauses. In N", the complements include PPs of time, place,
accompaniment, and so forth, relative clauses, and (for semantic reasons, their prehead
 4: COMPLEMENTS

position notwithstanding) APs. The phrase structure of X" is on the whole rather
simple; this section will discuss three intricacies involved in the proper formulation of
X" phrase structure rules and in the use of X" positions.
First let us deal with the differences between AdvPs and PPs in V". The most
salient difference is that AdvPs may appear preverbally as well as postverbally,
whereas PPs may only be postverbal.

Furthermore, although PPs can appear one after another, there is a stylistic
preference for avoiding two consecutive adverbs, wherever they are attached (see
Jackendoff (1972, chapter 3) for discussion of adverb attachment to V" vs. V'")-

(4.41) a. ?Bill frankly cleverly left, (two S adverbs)

b. ?Bill probably suddenly left. (S and VP adverbs)
c. ?Bill quickly completely finished, (two VP adverbs)
d. ?Frankly, probably Bill left, (two S adverbs)
e. ?Bill finished completely, probably. (VP and S adverbs)
f. ?Bill finished completely quickly, (two VP adverbs)

Note that if the adverbs are separated the examples in (4.41) become acceptable.

(4.42) a. Frankly, Bill cleverly left.

b. Probably, Bill suddenly left.
Bill suddenly left, probably.
Bill probably left suddenly.
c. Bill quickly finished completely.
d. Frankly, Bill probably left.
Frankly, Bill left, probably.
e. Probably, Bill finished completely.
Bill completely finished, probably.
f. Bill quickly finished completely.

Since this condition can involve any type of adverb, and in particular can cross
constituent boundaries, it cannot be stated as part of a set of context-free phrase
structure rules, but must rather be stated as some sort of an output string condition.
Thus the phrase structure component may be left free to generate an indefinite number
of adverbs as well as an indefinite number of PPs, and the number actually possible will
be constrained by the output condition and the necessity that they all receive
interpretations which can be integrated into the interpretation of the sentence.
These considerations suggest, for a first approximation, the phrase structure rule
(4.43), where X* denotes a concatenation of an indefinite number of Xs.
74 RAY JACKENDOFF

[PP
(4.43) V"—> (have - en) - (be - ing) - (AdvP) - V' - (
[AdvP

There are two things wrong with (4.43). First, the final term, expressed in distinctive
features, is [-Subj, aObj, aComp], not a wonderfully natural combination. Further¬
more, in the N" complement there is no AdvP, and no juggling of features will be able
to combine [aObj, aComp] with [ + Obj, -1-Comp] in a particularly felicitous fashion.
[PP
This suggests that the combination i A1 _ is incorrect, though it appears necessary
lAdvP
because PPs and AdvPs may be freely interspersed in V".
The second difficulty is the repetition of AdvP in two positions in V", though it has
essentially the same semantic functions in either position. Within the usual framework
of phrase structure grammar, this cannot be avoided, since the V' gets in the way of
collapsing the two occurrences. Moreover, the manner adverb may occur before an
aspect that has not been moved up into the Aux, as in John will completely have
finished. Thus in fact three AdvP positions appear necessary.
The usual solution to this problem is to generate the desired constituent in a single
position in underlying structure, then to move it to its various surface positions by
means of a transformation. Here, however, the nature of the generalization lends itself
to a different treatment: the generalization is that AdvP may appear in any position
where it is a daughter of V". To express this generalization, we will follow the
suggestion of Keyser (1968) and allow the phrase structure component to designate
certain constituents as "transportable”: a transportable node will have free word order
with respect to all of its sisters.15 We will indicate transportability with a feature
[+Trans], and assume that [—Trans] is unmarked, that is, that if the feature is not
mentioned in a rule, [-Trans] is intended (or filled in by marking conventions). We can
now revise (4.43) to (4.44).

Adv
(4.44) V" —> (have - en) - (be - ing) )* - V' - (PP)* - (S)
-<[ .+Trans

The position of the Adv in (4.44) allows collapsing with the rule for N", (4.45), to
the rule (4.46).

(4.45) N"—> (A'")* - N' - (PP)* - (S)

ft
X -Subj
+ Subj -Obj
(+Obj)j —> ((have - en) - (be - ing) )j - ( /(—Comp [\
<-Obj>2 \l+Trans l/i
+ Comp _< + Comp>2
(PP)* - (S)

15 Jackendoff (1972, chapter 3) argues that free adverb order in VP is impossible because strictly
subcategorized adverbs must be postverbal. In the present treatment, the placement of strictly subcatego¬
rized adverbs in V' separates them syntactically from other adverbs, which can now be free to wander
throughout V".
 4: COMPLEMENTS

Note that the feature +Trans appears only in V", so that adjectives will appear only
prenominally and will not be free within N".
A second problem in the structure of V" is the ability of a strictly subcategorized
PP (i.e. a PP in V') to appear after a manner adverb, as in John gave the beans quickly
to Bill. To account for this, either the adverb must be lowered into the AdvP position in
V', or the PP must be raised into the first PP position in V". Both operations are
structure preserving and in that respect harmless.
One piece of evidence that might be thought relevant to the choice is that a strictly
subcategorized PP cannot follow do so even when it follows a manner adverb:

(4.47) *John gave the beans quickly to Bill, and Fred did so (slowly) to Susan.

This apparently argues that the adverb is moved down into V'. But in fact such an
argument is incorrect; it is also the case that strictly subcategorized clauses cannot
follow do so even when extraposed around various PPs in V".

in a loud voice
(4.48) *John said suddenly that smoking was fun, but Susan
at 6:00

did so (' ) that it was bad for you.

at 5:00

Thus there must be an independent mechanism to rule out (4.48), possibly having to do
with the trace left behind in V' by the extraposed constituent; whatever this mechanism
is, it will also rule out (4.47) under the theory that the PP to Susan has been
extraposed. Thus there are no strong considerations to decide which constituent moves
in inverting a manner adverb with a strictly subcategorized PP. The spirit of the
Extended Standard Theory does include a general uneasiness with lowering rules, so
perhaps the raising of the PP is a better solution.
Finally, there is an S at the end of both V" and'N". This is the position of
restrictive relative clauses in N"and of comparative clauses in both V"and N"; chapters
7 and 8 will discuss these uses at some length. It is also the position for various
restrictive subordinate clauses in V" such as in It only hurts when I laugh, and the
position to which extraposed V' sentential complements move by a structure-preserv¬
ing rule, for example in (4.48). A more controversial use is as the position for sentence-
final subject complements, as in It bothers Bill that she came.
The controversy arises over whether the V" complement is only a derived position
for these complements, as claimed by Rosenbaum (1967), or whether it is an underlying
position, as claimed by Emonds (1970). Emonds shows that there are a vast number of
differences between that and for-to complements on the one hand and gerundive (poss-
ing) complements on the other, concluding that only gerundives behave like NPs.
Rosenbaum's theory, in which all three types of complements are dominated by NP in
76 RAY JACKENDOFF

underlying structure, is at a loss to explain this fact without a great number of unusual
conditions on the Extraposition transformation, which derives (4.49b) from (4.49a).

(4.49) a. [NPit [sthat she came]] bothers Bill

b. It bothers Bill that she came.
c. That she came bothers Bill.

Emonds, on the other hand, shows that the syntactic distribution of complement types
is much better accounted for if (4.49b) is underlying and (4.49c), which occurs only in a
rather restricted class of contexts, is derived from it by a simply stated root
transformation of Subject Replacement (or Intraposition).
On general syntactic grounds, the Intraposition theory has much to recommend it.
However, a number of difficulties in this theory are pointed out in Higgins (1973), and
in response to them Emonds (1976) abandons the Intraposition theory in favor of a
variant of Rosenbaum’s analysis. The advantageous properties of Emonds’s earlier
analysis are accounted for in the new analysis, but only in a somewhat artificial and
suspect fashion. The issue is not crucial to the present theory, but Emonds’s original
view of these complements does fit attractively into our analysis here. Section 4.8
presents some new considerations which favor the Intraposition analysis, and which
show that V" is the correct position for subject complements to be attached.
The V" position of subject complements has a parallel in N". Chomsky (1970a)
points out (p. 44/203) that sentences such as (4.50a) are paralleled by nominals like
(4.50b).

(4.50) a. It is necessary for John to leave,

b. the necessity for John to leave

Because Chomsky assumes the complement clause in the sentence is an extraposed

subject, the X Convention forces him to generate the complement of the nominal before
the head too. He then suggests that Agent Postposing (the first part of Passive)
accomplishes the extraposition of the clause in (4.50b) to its surface position. The
assumption that (4.50a) is in its underlying order, as Emonds (1970) suggests, eliminates
the dubious extraposition in the nominal, since the X Convention then requires (4.50b)
to be in its underlying order as well. Thus if the S in V" is a subject complement
position, so is the S in N".

4.4. Complements of [-Subj] Categories

4.4.1. Adjectives

The syntax of A' is similar to that of N', but more limited. Adjectives can take PP
complements, as in good at chess, yellow with age, eager for help. They can also take
S complements, as in afraid that Bill is fierce, eager to please, proud to be a frog, and
happy that he won. There are a few examples such as dependent on Bill for help which
 4: COMPLEMENTS

have two PPs, and dependent on Bill to help him with PP followed by S. Thus the
\ PP]
phrase structure rule for A' includes at least (PP) - n =, f).

A significant number of adjectives related to transitive verbs take o/-NP in their

complements: fearful/considerate/desirous/solicitous of NP, for example, are related to
fear /consider I desire /solicit NP. To simplify the statement of lexical relations, we can
consider these particular o/-NP complements to be simple direct objects in deep
structure, treating the of as a specified grammatical formative which is inserted
transformationally. Thus, we can generalize the definition (4.38) of the grammatical
relation object to adjectives; and rule (4.36), 0/-Insertion, generalizes to the domain
[—Obj]', saving a feature in its statement and simultaneously simplifying lexical
relations. Thus the expansion of A' must be something like (4.51).

+Obj
(4.51) A'-> A - (NP) - (PP) - ( )
+Comp_

In the A" complement there is a class of preadjectival modifiers, for example

completely/utterly/partially corrupt. These are structurally parallel to the adverbial
degree modifiers in Ss, as in John has been completely /utterly /partially corrupting the
government, and to prenominal adjectives as well, so base rule (4.52) will be a special
case of the more general rule for this position. Since the adverb can in turn take a few
modifiers (particularly not, as in a not utterly divine concert), the adverb position must
be Adv'" and not just Adv. Further, A" contains an S for degree clauses, to be
discussed in chapter 8.

(4.52) A" —» (Adv'") - A' - (S)

(4.53) is an AP which illustrates some details of the structures just proposed.

(4.53)
78 RAY JACKENDOFF

4.4.2. Adverbs

On the whole, adverbs take no complement: there is no *fearfully of Bill parallel to

fearful of Bill, for example. However, there are a few PP complements to adverbs, for
example unfortunately for our hero. The question arises as to whether they are
attached to Adv' or to Adv". We would like to maintain the generalization that
[-Comp] categories do not strictly subcategorize anything, i.e. that they have no X'
complements. This would require that for our hero is an Adv" complement.
Evidence within the adverb phrase itself is hard to come by, but the parallel AP
constructions are suggestive. Notice that in (4.54a), the PP for our hero can prepose (or
alternate with a V" complement which preposes) to form (4.54b). But a PP strictly
subcategorized by AP, as in (4.54c), cannot prepose.

(4.54) a. It was unfortunate for our hero that Rome burned.

b. For our hero, it was unfortunate that Rome burned.
c. Bill is dependent on Fred.
d. ?*On Fred(,) Bill is dependent, (ok only with “Yiddish” topicaliza-
tion)

This difference may be ascribed to attachment of the two PPs in different places,16
arguing that for our hero in (4.54a) and in the parallel AdvP is an X" complement.
As for S complements, we will take the position that subordinating conjunctions
are prepositions rather than adverbs, so “adverbial clauses” introduced by because,
although, etc. are not Adv' complements. There appear to be no -ly adverbs with S
complements. Thus, the phrase structure rule for Adv' is the minimal (4.55), and we
can maintain the claim that [—Comp] categories take no X' complements.

(4.55) Adv' —> Adv

Fike adjectives, adverbs may be preceded by utterly, completely, etc.; if such

degree adverbs are dominated by A", they must also be within Adv". Again the
presence of a few modifiers such as not and quite (as in The state is governed quite
utterly corruptly, infelicitous primarily because of the repetition of -ly) shows that the
degree modifier is an Adv'" rather than just an Adv. Adv" also contains degree clauses,
as well as the PPs just mentioned. Hence the base rule for Adv" is (4.56); (4.52) can also
be revised to include the PP.

(4.56) Adv" (Adv'") - Adv' - (PP) - (S)

4.4.3. Prepositions

Although the usual uninformed view is that the complement of a preposition is

invariably NP, it is shown in Jackendoff (1973), using in part evidence from Klima
(1965) and Emonds (1972a), that the base rule for P' is at least (4.57).

16 Williams (1975) gives this sort of difference as criterial for preposing in Ss.
 4: COMPLEMENTS

(4.57) P' -> p - (NP) - (PP)

If both optional constituents are omitted, we get a class of "adverbs” without -ly
such as here, there, outside, downstairs, beforehand, and afterward. If NP is present,
we get the familiar forms in the house, down the hatch, after the ball, etc. If only PP is
present, we get phrases like out of the barn, down into the darkness, from inside the
barrel, and up to your bedroom. If both NP and PP are present, we get phrases like
across the street from Bill’s house, down the street toward Bill, and to Bill in New
York. The expansion P - PP may use an intransitive preposition in the lower PP, as in
over here, down there, and from within. It can also be expanded with a P - PP in the
lower PP, as in from out of the darkness, down from above the altar, etc. Thus (4.57)
seems to be perfectly productive and free of restrictions; I find it surprising that it has
gone unnoticed for so long. Note that the definition of the grammatical relation
"object”, (4.38), generalizes directly to P', so (4.38) can now be left altogether free of
category features on X.
Klima (1965) claims that subordinating conjunctions can also be analyzed as
prepositions which take an S complement (from which the complementizer that is
deleted obligatorily in modern English, optionally in Middle English). This provides the

simplest description of the relation between the prepositions in j the ball and
the "conjunctions” in
(before I
the ball is over: before and after, like many verbs,
after J
nouns, and adjectives, allow either an object or a subordinate clause. The S does not
cooccur in the complement with either NP or PP, to my knowledge, so it must be an
alternative to them, yielding this phrase structure rule:

(4.58) p' p- {(g)P) ” (PP)|

PPs also permit the degree Adv"', as in not completely up the tree, quite utterly out of
his mind, partly into the room, etc. As will be seen in chapter 8, they also allow an S
for degree clauses. Thus the base rule for P" is in part (4.59):

(4.59) P"-* (Adv"') - P' - (S)

4.4.4. Particles

According to the feature system of chapter 3, the [-Comp] counterpart of preposition

is the category Particle. Particles take no complement structure at all; but if not
followed by an NP or AP in V', they can occur with a few of the prehead modifiers that
are characteristic of prepositions, as we see from (4.60a). Before NP or AP, these
modifiers are impossible, though, as shown by (4.60b).

(4.60) a. I’ll look the answer right up, sir.

He ate his lunch all up.
You can turn the faucet completely off.
80 RAY JACKENDOFF

b. We painted the house (^completely) up red.

I’ll look (*right) up the answer, sir.
He ate (*all) up his lunch.

Expressing this restriction is a problem. One’s first impulse is to generate PrtP as a

PP to the right of all NPs in V', then to move PrtP leftward, optionally, if it consists
only of Prt. However, such an account does not explain why, in double object
constructions in V', Prt can appear only between the two NPs but not after them, nor
why Prt always precedes the object in N'; these were the advantages of the alternative
analysis in section 4.2. I leave the resolution of this dilemma for future research.
In any event, the phrase structure expansions for Prt are something like (4.61).

(4.61) a. Prt' Prt

b. Prt" —> (Adv'") - Prt'

These rules do not account for the specifiers all and right in (4.60a); the rules for PP
will have to be emended in parallel fashion.

4.4.5. Case Markers and Prepositions

Before discussing generalization of the phrase structure rules for complements, we

digress briefly to make a small point about case marking. It is often claimed that
prepositions are not a lexical category, but rather that they are simply case markers on
noun phrases, possibly even inserted by transformations. This approach is institutional¬
ized in Fillmore’s case grammar (Fillmore (1968)); Postal (1971) trivializes the role of
prepositions still further. These analyses are based on the mistaken assumption that the
only possible complement to a preposition is NP; if prepositions enforce strict
subcategorization restrictions and occur with such bizarre complements as PP and even
NP - PP, the analysis is obviously untenable. Nevertheless, there is some relationship
between prepositions and cases which we would do well to clarify at this point.
The reason for considering prepositions to be case markers is semantic: preposi¬
tions in one language often translate as case markers in another language, and vice
versa. And even within a single language, prepositions and case markers often perform
very similar semantic functions, as alternative realizations of indirect objects, for
example.
One might therefore suggest that case markers are transformationally reduced
prepositions. Though diachronically the reduction of prepositions to case markers is
undeniable, it is not clear that such a process plays a role in synchronic grammar.
Alternatively, it is possible to consider case markers and prepositions as distinct
syntactic entities, the ranges of whose interpretations overlap to some extent. Such is
the case with adverb phrases and prepositional phrases, for example (cf. Jackendoff
(1972, chapter 3)). Prepositions then can be considered a lexical category on a par with
nouns, verbs, and adjectives, while case markers are affixes that are attached to NPs.
 4: COMPLEMENTS

Taking this view allows a more unified theory of case marking in languages such as
German and Russian, which have a mixture of prepositional phrases and cases:
prepositions, like every other lexical category, determine the cases of their comple¬
ments. Thus the lexical and transformational machinery necessary to enforce case
marking is very general, applying to the complements of all lexical categories. By
contrast, in the prepositions-as-case-markers view, the case markings due to preposi¬
tions must be accounted for separately from those due to other categories. Thus,
treating prepositions as a genuine lexical category, besides accounting for obvious facts
about their distribution, permits a more unified treatment of case marking.

4.5. Generalization of Phrase Structure Rules

Having described a substantial number of phrase structure rules for individual

categories, let us see what the prospects are for generalizing them.
Here are the rules for nontrivial expansions of X', in the forms developed in
preceding sections, without collapsing.

+ Obj
(4.32) V' V-(NP)-(Prt"')-([J^J ) - (PP) - ( +Comp )
+Obj
(4.33) N' N - (Prt'") - (NP) -(PP)-( )
+ Comp
+Obj
(4.51) A' A (NP) - (PP) - ( _+Comp_ )
(NP) -(PP)j
(4.58) P' P
(S)

Certain generalizations about word order are apparent from inspection of these
rules. In X', the head precedes all other phrases. There is a strong recurrence of N'",
P'", and S in the rules for X', and their relative order is the same in all cases where they
cooccur: N"' always precedes P'" and S, and P"' always precedes S. If these are
linguistically significant generalizations (as I believe they are), the evaluation measure
for the theory of grammar ought to be constructed in such a way as to favor this
grammar over one (for example) in which V' expands as V - N'" - P'" but P' expands
as P - P'" - N'". The usual way to construct an evaluation measure is to develop
notational conventions in terms of which the more general case takes fewer symbols to
state. This is the purpose of parentheses, braces, and distinctive feature notations (cf.
Chomsky and Halle (1968, chapter 8)).
In this case, though, a number of instances arise in which the ordinary abbrevia-
tory conventions are of no avail. First, suppose that we try to collapse just the V' and
P' rules so that the appropriate parts of the complement coincide. In simplified form,
showing only the relevant parts, we get rule (4.63) as an abbreviation for (4.62a,b).
82 RAY JACKENDOFF

(4.62) a. V' —> V - (S)

(4.63) X
(+Subj)!
(N "') - (P'")
<-Subj)2 -* X - - <(S)>
. <(S))*
+Obj
+ Comp _

Capturing the generalization about the relative order of N'" and P'" misses the
generalization about the presence of S in the complement. In (4.63) the two occur¬
rences of S have nothing to do with each other, and either could be changed to
something else without affecting the complexity of (4.63). This is clearly an incorrect
result. The only way to improve on this is with a complex Boolean condition such as in
(4.64), simulating the effect of the braces in (4.62b) within the constraint of a linear
string.

(4.64) X
+ Obj -*X-(N,w)1-(P'")2-(S)3
_+Comp_
Condition: If X= P, not (3 and (1 or 2)).

Apparently, then, all the X' rules can be collapsed, while preserving the generali¬
zations about order, only by means of negative Boolean conditions. The problem
becomes worse if we try to refine all the X' rules to prevent nonexistent expansions.
For example. A' appears not to allow NP to cooccur with any of the other
complements; thus, a more refined rule would contain a set of braces different from
that in P', requiring more Boolean conditions still.
It is hard to judge how good such a solution is. The introduction of more or less
arbitrary Boolean conditions increases the power of a phrase structure grammar
considerably, and one would have to ask very seriously how they can be constrained.
Furthermore, it is not clear how one would define an evaluation measure over such a
set of conditions, so as to know exactly what is a generalization and what is not.
Nonetheless, the numerous partial generalizations of English seem to require some
such rather complex solution. I can offer no interesting suggestions for alternatives to
Boolean conditions. A serious study of the phrase structure rules of a number of other
languages would no doubt elucidate what problems must be solved by an explanatory
theory of cross-category phrase structure rules.
To make it clear that this is not a pseudoproblem, let us just mention a few other
cases in the phrase structure component of English where the traditional abbreviatory
conventions are inadequate. First reconsider the rules for X'" developed in chapter 3:
 4: COMPLEMENTS

(3.42) a. V—> (N'") - (M'") - V"

h n -n"

Anticipating the results of chapter 6, we will also mention these rules for the A'" and
Adv"' specifier for comparison.

(4.65) a. A'" —> (Deg"') - A"

b. Adv"' —» (Deg'") - Adv"

There is a clear relationship between the Degree phrase in A'" and Adv'" and the
Article in Nand in fact a few words such as this and that appear in both categories.
What Deg'" and Art'" have in common is that they are the [—Comp, +Det] counterparts
of their respective heads. Now observe that M'" is likewise the [-Comp] counterpart of
its head, a generalization of possible interest, not pointed out in chapter 3. Again,
however, we are faced with a dilemma: Deg'" in (4.65) can be collapsed either with M'"
in (3.42a) or with Art'" in (3.42b), but not with both; there is no way to collapse all four
rules at all without a Boolean condition that restates the braces.

aSubj
X
/30bj
(4.66) aSubj (N'"), - ( )2 - X"
-Comp
L/30bj
+ Det
Conditions: If X = [—Subj], not 1
If X = N, not (1 and 2)

This case is thus formally parallel to the V' - P' generalization pointed out before.
A different sort of problem arises in trying to collapse the rules for X".

Adv
(4.44) V" (have - en) - (be - ing) - ( )* - V' - (P'T - (S)
+ Trans_
(4.45) N" (A'")* - N' _ (p'")* -(S)
(4.52)' A" (Adv'") - A' - (P"') -(S)
(4.56) Adv" (Adv'") - Adv' _(P"') -(S)
(4.59) (Adv'") - P' -(S)
(4.61b) (Adv "j - Prt'

The problem lies in capturing the generalization that all categories have a prehead
modifier that is [-Subj, —Obj]. N" and V" have special wrinkles in the modifier: both
allow more than one; in V" it is transportable; and in N", unlike all other categories, it is
[+Comp], Expressing these deviances with any of the usual notations would be
extremely clumsy, if at all possible. Of course, unconstrained Boolean conditions can
84 RAY JACKENDOFF

handle the problem:

-Subj
(4.67) X" —* (have - en), - (be - ing)j - ( Obj )2 - X' - .
Det J
Conditions: If X 4- V, not 1
If X = [+Subj], 2 is *
If X = V, 2 is [+Trans]
If X = N, 2 is [ + Comp]
If X =* N, 2 is [-Comp]

These conditions are still less constrained than those in (4.66), since they specify actual
features and not just presence or absence of constituents. The problem is to find some
device less constrained than traditional abbreviatory conventions but more constrained
than such arbitrary conditions.
Perhaps the most striking case is one that will be motivated in more detail in
chapter 6. All [ + Comp] categories except verbs permit a measure phrase (a kind of NP)
or a quantifier phrase in the X" specifier:

a. two parts steel (NP - N')

much steel (QP - N')
b. two miles long (NP - A')
little interested (QP - A')
c. two miles down the road (NP - P')
far down the road (QP - P )

In the verb phrase there is also an alternation of a measure phrase and a quantifier
phrase in V", but this time it is to the right of the head:

(4.69) a. He jumped into the air two times. (V' - NP)

b. He talked about sex too much. (V' - QP)

To make matters worse, the measure phrase and (under certain conditions) the QP can
follow the head in PP:

(4.70) a. down the road two miles (P' - NP)

b. down the road quite far (P' - QP)

There is thus a generalization that [ + Subj, —Obj, -Det]"' can be generated as a

daughter of X". But traditional phrase structure rules cannot express this generalization
at all, because of the difference in word order. In this case even Boolean conditions do
not provide much help. The best one could do is something like (4.71) (relying on
 4: COMPLEMENTS

semantic conditions to prohibit both at once in PP):

+ Subj w
+ Subj
-Obj )l-. . - X' - . • -( -Obj
_-Det . Det _
Conditions: If X = V, not I
If X = [—Obj], not 2

But this does not capture the generalization that the same expression appears on both
sides of X'; the grammar would be no more complex (in terms of number of symbols) if
the subscripted terms in (4.71) were of entirely different categories but used the same
number of features. One might suggest instead'that all these modifiers are generated
before the head, and a transformation permutes them obligatorily around a V' and
optionally around a P'. Such a rule, however, would be non-structure preserving, and
besides would be supported by no independent syntactic evidence, at least in the case
of V'.
Though this is the only such case I know of in English, there is reason to believe
that it is not such a rare situation. It would arise in its most drastic form in a language
with a verb-final V' but a noun-initial N'; German, if it is SOV, as has been rather
persuasively argued by many people, is such a language. Languages in which adjectives
follow nouns but adverbs precede adjectives would be another such case; so would
SOV languages with prepositions rather than postpositions. Such languages are cited by
Greenberg (1963).
Greenberg points out, however, that such languages are relatively rare; his
“universals” say that the opposite correlations occur with “overwhelmingly greater
than chance frequency”. This suggests that the evaluation measure for phrase structure
rules counts parallelisms of word order across X" as generalizations, as we have come
to expect in the present study. But it appears also that parallel grammatical relations
with differing orders across X" also count as generalizations, though not as strongly;
present notations have no way at all to express this. Again, detailed study of the phrase
structure of languages other than English is necessary, in order to establish exactly
what generalizations there are and how the formalism for phrase structure rules should
be constrained so as to explain why these and no other generalizations appear.

4.6. Summary

Section 4.1 showed that complements divide semantically into three types: functional
arguments, restrictive modifiers, and nonrestrictive modifiers, and that these can be
identified with X', X", and X'" complements, respectively. Thus, there is a systematic
relationship between the syntax and the semantics of the complement system.
The rest of the chapter developed phrase structure rules for the X' and X"
complements of most syntactic categories in English, showing the extent of cross¬
category generalization in the complement system. A close examination of the existing
86 RAY JACKENDOFF

generalizations revealed, however, that many of them cannot be expressed with the
traditional notational conventions.
There follow two appendices. The first deals with the generalization of the Passive
to NPs; the second presents some new arguments about Intraposition. Besides the
phrase structure rules listed in the previous section, this chapter and its appendices
discuss a substantial number of transformations; they are listed after the appendices.

4.7. Appendix: The Passive and Related Rules

This appendix describes how Chomsky (1970a) generalizes the Passive transformation
to NPs, and proposes a number of improvements on his analysis.

4.7.1. NP-Preposing

Consider the relationship between the phrases in (4.72a) and the synonymous readings
of the phrases in (4.72b).

(4.72) a. the destruction of the enemy

the murder of John
the picture of John
the proof of the theorem
b. the enemy’s destruction
John’s murder
John’s picture
the theorem’s proof

If we relate them with a transformation, taking (4.72a) as the underlying order, we can
explain two things: the syntactic relationship of these phrases to the sentences X
destroyed the city, X murdered John, etc., in which the NP is a deep object; and the
ambiguity of (4.72b), where the genitive phrase can originate either as deep subject or
deep object. Chomsky calls the transformation deriving (4.72b) from (4.72a) NP-
Preposing (p. 41/203).
This rule applies only to NPs with a direct object (in the sense of section 4.2): there
are no NPs *war's prayer, *my despair’s escape, *power's lust corresponding to the
prayer for war, the escape from my despair, the lust for power. This provides some
interesting evidence that the of of direct objects in NPs is not present at some earlier
stage of derivation. The argument goes as follows: the sense of approve which takes a
direct object, e.g. approve the plan, has the nominals the approval of the plan and the
plan's approval. But the sense of approve which takes an o/-phrase, e.g. approve of
your behavior, has only the nominal the approval of your behavior and not *your
behavior's approval. We can explain this difference by claiming that the former sense
of approval, like the verb, lacks of in deep structure, but the latter sense requires it.
Then only the former sense will permit NP-Preposing, although the deep structure
difference of the two senses is obscured by the later rule C/-Insertion.
 4: COMPLEMENTS

Some apparent irregularities in the application of NP-Preposing are explained by

Wasow (1975) as a consequence of the structure-preserving hypothesis. Wasow points
out that for NP-Preposing to be structure preserving, the preposed NP must be moved
into an independently generable empty NP position. If an empty subject NP were for
some reason impossible, NP-Preposing could not take place. He then observes the
contrasts (4.73) versus (4.74).

(4.73) a. The report of the invasion troubled John.

b. The reading of the honor roll embarrassed John.
c. The singing of the aria bored John.
d. John enjoyed the leveling of the city.
(4.74) a. The authorization of the invasion troubled John.
b. The publication of the honor roll embarrassed John.
c. The composition of the aria bored John.
d. John enjoyed the destruction of the city.

In (4.73) John may not be understood as having subject function in the nominal, but in
(4.74) he may. Wasow claims that the difference is due to the ability of the nominals in
(4.74) to take an empty (or PRO) NP specifier and the inability of the nominals in (4.73)
to do so; the difference is justified independently in Wasow and Roeper (1972). Wasow
then points out that this difference correctly predicts that only the nominals in (4.74)
undergo NP-Preposing, since only they allow an empty NP into which the object can
prepose.

(4.75) a. *The invasion’s report troubled John.

b. *The honor roll’s reading embarrassed John.
c. *The aria’s singing bored John.
d. *John enjoyed the city’s leveling.
(4.76) a. The invasion’s authorization troubled John.
b. The honor roll’s publication embarrassed John.
c. The aria’s composition bored John.
d. John enjoyed the city’s destruction.

Wasow’s analysis does not exhaust all the situations where NP-Preposing fails, but it
accounts for an interesting class of cases.
For the moment we will assume that the is inserted to fill an empty NP in the
specifier position; we will return to this in section 4.7.4. Under this assumption, NP-
Preposing can be stated as (4.77), where lower case denotes an empty node.

(4.77) NP-Preposing
np-Y-N-NP-Z
=>4-2-3-</>-5
OPTIONAL
88 RAY JACKENDOFF

4.7.2. Agent-Postposing

Next consider the relationship of the phrases in (4.78a) to those in (4.78b).

(4.78) a. the enemy’s destruction of the city

Lange’s portrait of Mozart
Euler’s proof of that theorem
b. the destruction of the city by the enemy
the portrait of Mozart by Lange
the proof of that theorem by Euler

Again, the synonymy of the pairs can be captured by a transformational relationship.

The parallelism of (4.78a) to The enemy destroyed the city, Lange portrayed Mozart,
and Euler proved that theorem suggests that (4.78a) is the underlying order and (4.78b)
is derived. To state the transformation in a structure-preserving framework, we need to
provide the complements in (4.78a) with a PP node whose NP is empty. We could also
leave the preposition empty; but by filling it with by in underlying structure, we have
some semantic material to associate with the agentive function evident in (4.78b).

(4.79) Agent-Postposing
[NP NP - X - N - Y - [PP by np] - Z]
=> np - 2- 3 -4 - 5-1-7
OPTIONAL

The underlying structures of (4.78b) will be the enemy destruction the city by np, etc.;
(4.78a) will simply lack the final PP.
If we order NP-Preposing after Agent-Postposing, we can apply them in sequence:

(4.79)

(4.80) the enemy destruction the city by np

(4.77)

np destruction the city by the enemy =>

(Poss—Insertion)
the city destruction ct> by the enemy
the city’s destruction 4> by the enemy

But the final form in (4.80) is precisely what we observed in chapter 2 as a parallel to
the passive sentence The city was destroyed by the enemy.
The Passive transformation, although traditionally written as a single rule which
moves two NPs, can clearly be broken into two components, the movement of the
subject into the /ry-phrase and the movement of the object into subject position. The
first component obviously generalizes with Agent-Postposing: the bv-phrase is used in
precisely the same way, and the NP is moved out of parallel positions. The rules
combine as (4.81).
 4: COMPLEMENTS

(4.81) Agent-Postposing—Passive, Part I

X
[X"'NP - Y - + Subj - Z - [ppby np]-W]
_+Comp
z>np-2-3-4-5-l-7

The second component of the sentential Passive does not immediately generalize
so clearly with NP-Preposing. This part of the Passive, unlike NP-Preposing, is
obligatory, since there are no sentential forms like (4.78b) (e.g. *it destroyed the city by
the enemy). Furthermore, under certain conditions the passive construction permits a
preposition to intervene between the NP to be preposed and the verb, as in This bed
has been slept in by an infamous violinist. The second half of the Passive must also
introduce the auxiliary be-en.

(4.82) Passive, Part 2

np - Y - V - (P) - NP - Z
=> 5 - 2 - be-en - 3- 4- </> - 6
OBLIGATORY

(4.82) could clearly be combined with (4.77) by the usual notations. However, the
difference between optional and obligatory use casts a certain amount of suspicion on
the generalization. Fiengo’s (1974) analysis of the passive provides some hope of
improvement. In Fiengo’s theory, be-en is present in the underlying structure of
passives, and semantic conditions on traces guarantee that only passive surface
structures are associated with it. Furthermore, he generalizes the rule so as to use it to
derive examples like These books sell easily from sources like np sell these books
easily; thus be-en plays a role neither in the structural description nor in the structural
change of the Passive. Because of the presence of various independent conditions,
then, Fiengo’s analysis eliminates the be-en and the obligatory application from (4.82).
The only remaining discrepancy is then the optional preposition, so the generalization
of NP-Preposing looks somewhat more plausible than in the analysis of “Remarks”,
interpreted strictly. Conceivably the generalized form could be at least as clean as
(4.83) .

(4.83) Generalized NP-Preposing

+ Subj
np-Y- <+Obj) - <(P)> - NP - Z
+ Comp_
5 _ 2 - 3 - 4 - <j> - 6
OPTIONAL

Besides the generalization with NP passives and the elimination of a rule moving
two constituents simultaneously, there is some independent reason to separate the two
components of the Passive. As pointed out by Emonds (1970), sentences like (4.84)
90 RAY JACKENDOFF

could be generated as deep structures with empty subjects and derived by the use of
(4.83) alone, avoiding the suspicious rule of By-Someone Deletion used traditionally in
deriving examples of this sort.

(4.84) Frege was born in 1485.

John is | sa*^ [ to like tamales.

(rumored I

4.7.3. NP-Postposing

Continuing with the syntax of N'", Chomsky appeals (p. 44/205) to

a minor transformational rule which will replace by by of under certain conditions,
permitting the refusal to leave of those men (or the refusal of those men to leave).
Presumably it is this rule that applies in the case of the nominals the growling of the lion.
etc.

However, Bowers (1968a) argues against the by —> of rule, proposing instead that there
is an independent rule of NP-Postposing with the following effect (stated in the present
framework):

(4.85) NP-Postposing
NP-Y-N-np-Z
=> np - 2 - 3 - 1-5
OPTIONAL

Bowers’s argument concerns the forms of nominals related to nonpassivizable

sentences. Compare (4.86a) with (4.86b).

(4.86) a. the building’s height

the railroad’s length
the lake’s depth
the desert’s dryness
that proposal’s absurdity
the rock’s whiteness
b. the height of the building
the length of the railroad
the depth of the lake
the dryness of the desert
the absurdity of that proposal
the whiteness of the rock

In order to avoid duplicating the strict subcategorization restrictions, selectional

restrictions, and projection rules for these two forms, we can relate them transforma¬
tionally. If we choose (4.86b) as the underlying order, we can derive (4.86a) by NP-
Preposing. But this choice has unfortunate consequences for the statement of lexical
 4: COMPLEMENTS

relations. For these nouns and the adjectives high, long, etc., would be related in the
simplest way if they took syntactically parallel arguments. Since there are no forms
*high of the building, *long of the railroad, etc., but rather only The building is high,
The railroad is long, etc., the optimal relation between nouns and adjectives under the
X Convention favors (4.86a) as the underlying order. This of course requires the
existence of a rule like (4.85).
Similar considerations apply in the following nominals.

(4.87) a. John’s resemblance to Mary

the parcel’s weight
the book’s cost
b. the resemblance of John to Mary
the weight of the parcel
the cost of the book

Again a transformational relationship is called for. If we choose (4.87b) as the

underlying order, NP-Preposing can derive (4.87a), but we lose the parallelism of
grammatical relations with John resembles Mary, The parcel weighs a lot, and The
book costs a lot. If we take (4.87a) as the underlying order, so as to have parallelism
with the verbal forms, we could conceivably use Agent-Postposing and Chomsky’s by
—» of rule. But in these cases there is evidence against the use of Agent-Postposing,
since the corresponding verbs do not undergo it: *Mary is resembled by John, *A lot is
weighed by the parcel, *A lot is cost by the book. Thus if we want to keep lexical
relations as simple as possible, we need NP-Postposing to derive (4.87b) directly from
(4.87a), with no intermediate step involving by. Bowers argues further that choosing
(4.86a) as the underlying form of the adjectival nominalizations correctly predicts that
(4.86b) does not alternate with an indefinite form *a height of the building, etc., since
NP-Postposing always leaves behind a definite article. We incorporate this argument
into our account of determiners below.
NP-Postposing derives various examples of Chomsky’s: it applies to “intransitive”
nouns such as those in (4.86a) and (4.87a), yielding also the growth of the corn from the
corn's growth (cf. Chomsky, p. 25/192), the fall of Rome from Rome’s fall, the
growling of the lion from the lion’s growling, and the refusal of those men to leave from
those men's refusal to leave (I happily find Chomsky’s example the refusal to leave of
those men ungrammatical). NP-Postposing thus has an effect exactly opposite to that of
NP-Preposing. Since the underlying structures involved are generally well-motivated,
the existence of two transformations with opposite effects should not be disturbing. In
certain cases, though, it may as a result be difficult to determine which of two forms is
underlying and which derived, for example the friend of the princess vs. the princess’s
friend. Some evidence will appear shortly.
The system of rules at this point thus includes Agent-Postposing in NPs and Ss,
NP-Preposing in NPs and Ss, and NP-Postposing in NPs.
92 RAY JACKENDOFF

4.7.4. The Determiner

A major difficulty with the analysis so far, essentially a variant of Chomsky’s

analysis,17 is that it deals only with NPs whose determiner is either a possessive or the.
It therefore gives no account of NPs such as (4.88).

(4.88) a. a victory by the Vietnamese

b. three books by Harry
c. every effort by Bill to win
d. some proofs of that theorem by well-known mathematicians

The putative source of the by-phrases in these examples, Agent-Postposing, would

leave behind an empty NP in the determiner, which by assumption would have to be
realized as the, not as any of the other determiners above. Transformational insertion
of an arbitrary determiner in place of the empty NP makes no sense; but the determiner
and the agent cannot both be generated in the same position in the base either (unless
we resort to Emonds’s dubious doubly-filled nodes). Apparently the only possible
alternative remaining is to allow the by-phrase to be generated in the base, adding a
projection rule that can read it as an Agent. But this throws doubt on the existence of
the Agent-Postposing transformation: why derive something by transformation that is
already generable by the base?
Now if Agent-Postposing is eliminated in NPs and replaced by a projection rule for
a complement by-phrase, it would be missing a generalization to retain the transforma¬
tion in sentences. Rather, we must conclude that passive by-phrases in Ss are also
generated by the base and interpreted by a generalized form of the projection rule for
the NP by-phrase. According to this analysis, the underlying form of The city was
destroyed by the enemy is np past (be-en) destroy the city by the enemy, and the only
transformation that takes place is (4.83), the preposing of the object.
Alternatively, one might consider retaining Agent-Postposing in Ss as a second
source of by NP in addition to base generation. The application of this rule could be
used to account for some of the discrepancies between NP passives and S passives,
such as *the sense of danger by John, *the fear of Harry by John, and *the respect for
Mary by John vs. Danger was sensed by John, Harry was feared by John, and Mary

17 Our account differs from Chomsky’s analysis not only in the treatment of NP-Postposing in NPs,
discussed above, but also in the treatment of the determiner. Instead of an empty NP determiner in the
destruction of the city by the enemy, filled transformationally by the, Chomsky proposes that the determiner
position have the syntactic feature analysis [+NP, +Def], with the latter feature spelled out as the if NP is
moved away. Since at the time of “Remarks” empty nodes were not an available device, Chomsky’s solution
appeared much more necessary. Jackendoff (1974a) makes an attempt to work it out in some detail; the
complexity of the resulting theory has been criticized with some justification by Homstein (1975). Besides its
complexity, there are two arguments against Chomsky’s stratagem of combining category and definiteness
features at a single node in the base. First, no one has found another use for this device since 1967, the time
at which “Remarks” first appeared. Second, this device strikes me as utterly inadequate to cope with
examples like (4.88). I thus feel little regret in abandoning it.
 4: COMPLEMENTS

was respected by John (these cases pointed out by Hornstein (1975)). Note that these
all are cases in which the by-phrase does not designate an agent, so they could not be
interpreted correctly by the proposed projection rule. They are thus good candidates
for a transformational derivation of the by-phrase from subjects. On the other hand,
this solution might make it more difficult to give a semantic account of those verbs
which do not undergo Passive at all, e.g. resemble, weigh, etc. I leave the issue open.
Note that even if Agent-Postposing is abandoned altogether, we are not led to
something like case grammar, in which all agents are generated as by-phrases in
underlying structure. Rather, NPs can be interpreted as agents either in subject position
or in by-NP complements, and there is no transformational relationship between the
two positions at all.18
Since NP-Preposing requires an empty NP in subject position, and since the
determiner of an NP may be either an article or an empty NP, the following derivations
may take place.

(every
that
(4.89) that ' proof of famous theorem
(the
(no transformations apply)
( NP-Preposing
r famous theorem —>

1^1 famous theorem’s proof

I The—Insertion
r famous theorem- =>

I that
the proof of famous theorem

We are thus positing two sources of the proof of that theorem, one in which the is base¬
generated as an article, and one in which the is inserted transformationally to fill an
empty NP.
One might think that the arguments for eliminating Agent-Postposing would apply
equally against (4.85), NP-Postposing. But interestingly enough, a substantial number
of the nominals which we proposed to derive by NP-Postposing appear only with the
article the, so counterexamples like (4.88) do not arise.

18 Emonds (1976) proposes that all agents originate in a by-phrase and are optionally preposed to subject
position. Though I have not investigated the consequences of this alternative thoroughly, it strikes me that it
will require at least equal complexity in the grammar, and that it is somewhat less plausible than the present
theory, which simply invests an optional phrase structure position with its own interpretation, similar to
instrument and accompaniment phrases.
94 RAY JACKENDOFF

the

(4.90) cost(s) of the book

"some
"every
f the
*a
*some
1f resemblance(s) of John to Mary (compare la Yhe resem-
some
v-*every
blance(s) between John and Mary)
the

' height of the building

"some
"every''

Within the proposed analysis, a very simple account of this determiner restriction
emerges: these nouns subcategorize only NPs in the determiner, not articles and
quantifiers. Thus a configuration like (4.89a) cannot come about. The is introduced
transformationally when the NP position either is empty in the base or becomes empty
through the application of NP-Postposing.
A similar case is the action reading of destruction, which subcategorizes an object.
Though some destruction is all right, *some destruction of the city is not. We can
describe this difference by claiming that the action reading subcategorizes not only an
object but also an NP subject. Thus, the only article that can appear with it is the,
inserted if the subject happens to be empty at the surface.
Suppose we find nominals which are like (4.90) in that what is the subject of the
corresponding sentence follows the head of the nominal, but which differ from (4.90) in
that other determiners besides the are possible. Our solution must be to generate these
forms in the base instead as a result of NP-Postposing. In this case the verb and its
nominalization will not be optimally related, since the verb’s subject will correspond to
the noun’s object instead of to the noun's subject as in (4.90).
Some speakers may consider growth and growling relevant examples of this sort,
finding grammatical all growth of corn (related nonoptimally to Corn grows) and all
growling of lions (related to Lions growl). I find these examples marginal at best, and
all is the only determiner other than the that sounds even faintly plausible. In any
event, the existence of such nonoptimal examples is not a problem: I would guess that
this is an area where individuals’ lexicons will differ, since the evidence to decide
between a subject which postposes and an object which preposes is relatively scanty in
the corpus available to a language learner, and the choice has no serious consequences
elsewhere in the grammar.
This completes our account of rules moving NPs in and out of the complement.
There are still no doubt quite a number of anomalies and irregularities which we have
not attempted to deal with, but we have added much precision and detail to Chomsky's
 4: COMPLEMENTS

rather sketchy analysis, making several improvements along the way. In the end, we
claim that NP-Preposing takes place in Ss and NPs, that Agent-Postposing may take
place in Ss, and that NP-Postposing takes place in NPs. An empty subject position in
an NP is filled by the\ if articles and quantifiers other than the appear in an NP, there
can have been no NP position in the determiner, and hence no NP to the right of the
head can have been an underlying subject.

4.8. Appendix: Arguments for Intraposition

Section 4.3 claimed that the S at the end of V" and N" is an underlying position for so-
called “subject complements” such as the S in It is obvious that Bill left, It disturbs me
that Bill left, and the necessity for Bill to leave. This is the view of Emonds (1970)
(though not Emonds (1976)), who claims that the synonymous sentences That Bill left is
obvious and That Bill left disturbs me are derived by moving the complement leftward
from underlying final position by a root transformation called Intraposition. Emonds
opposes the traditional Rosenbaum (1967) Extraposition theory, which claims the
derivations go in the opposite direction.
This section presents a number of arguments for the Intraposition theory, to meet
certain objections that have arisen since the publication of Emonds’s work. I do not,
however, claim to meet all extant arguments against Intraposition. I presuppose the
reader’s familiarity with Emonds’s arguments.
The most objectionable feature of the Intraposition theory was its account of
double complement verbs such as prove. Since Emonds provided only one S position at
the end of the sentence, he was forced to claim that (4.91a), for example, was derived
from an underlying structure (4.91b) in which both complements occupied the same
position, a situation Emonds called a “doubly filled node”, which would have to be
eliminated before surface structure.

(4.91) a. That Charley arrived so soon proves that he isn’t guilty,

b. S

Since the device of doubly filled nodes seems morally repugnant to most linguists,
Emonds’s account of (4.91a) seemed rather implausible, and many considered this
96 RAY JACKENDOFF

alone sufficient to dismiss the entire Intraposition analysis, or even the whole structure¬
preserving hypothesis.
However, the problem of prove turns out to have a better solution within the
Intraposition theory. There are some speakers who find (4.92a) and particularly
(4.92b,c) acceptable.

(4.92) a. It proves that Charley isn’t guilty that he arrived so soon.

b. It proves Charley not to be guilty that he arrived so soon.
c. It would prove that Charley isn’t guilty for him to arrive so soon.

For these speakers, the doubly filled node solution is impossible, since the surface
structure is not at all ill-formed. These examples show that Emonds was wrong in
positing only one final S position.
In the present theory there are two distinct positions available for sentential
complements, one in V' and one in V". It is possible to claim that the first clause in
(4.92) is the V' complement and the second the V" complement, interpreted as a
subject:

(4.93)

N'" M

that he arrived so soon

for him to arrive so soon

prove I that he isn’t guilty

[c/> him to be guilty

This would explain in particular why the clauses in (4.92) cannot be interchanged
without changing meaning, a fact unexplained by a structure like (4.91b). It then
remains to explain why (4.92) is unacceptable to many speakers (and notice that this
problem exists in the Extraposition hypothesis as well). The most plausible solution in
view of the dialect split is an output constraint similar to that for consecutive adverbs
(see section 4.3), ruling out consecutive Ss within V". Thus, doubly filled nodes can be
eliminated in favor of a theoretical device of an independently necessary sort, itself
 4: COMPLEMENTS

required in the competing theory. Once we realize this, the problem of prove simply
ceases to decide between the two theories.
Another objection to the Intraposition theory concerns the relationship of the
clause to the subject it. How does the grammar determine that the clause has all the
semantic properties of a subject, without the defining syntactic property of preceding
the auxiliary? Rosenbaum’s solution is to claim that in the underlying form the clause is
in subject position, accounting for its interpretation in the simplest possible way.
Emonds, on the other hand, is forced to claim that there is an anaphoric relationship
between it and the clause, established by an independent projection rule.19 Thus, the
Intraposition theory needs a rule not needed by the Extraposition theory.
However, there is some evidence that the interpretation rule is independently
necessary. First, there are languages (e.g. French and Hungarian) in which a clause can
never occupy subject position. There is thus no syntactic evidence for generating it
there, and for these languages the anaphoric theory looks rather more attractive. Thus
the rule of anaphora must be a possibility in linguistic theory. Second, there are
sentences in English which cannot in good conscience be generated by an Extraposition
theory but which have final clauses with the function of subject: compare (4.94a) and
(4.94b).

(4.94) a. It bothers me that you’re tickling me.

It doesn’t surprise me that she came,
b. It bothers me when you tickle me.
It wouldn’t surprise me if she came.

The (b) sentences have the preposed form (4.95a), not (4.95b).

(4.95) a. When you tickle me, it bothers me.

If she came, it wouldn’t surprise me.
b. *When you tickle me bothers me.
*If she came wouldn’t surprise me.

But since the clauses in (4.94b) have the same semantic function as those in (4.94a),
their interpretation requires precisely the rule of anaphora proposed by Emonds to
account for (4.94a), and in this case the alternative of an Extraposition analysis is far
less plausible.

19 Emonds says the relevant anaphoric relationship is coreference, which strikes me as somewhat
incoherent, in that it and the complement S do not constitute two separate references to the same event. A
more appropriate kind of anaphora might be termed “grammatical anaphora”, by contrast with the usual
“referential anaphora”. When a relationship of grammatical anaphora is established, any projection rules
which would normally apply to the proform by virtue of its syntactic position are directed to apply instead to
the antecedent. Thus the antecedent receives an “indirect” grammatical relation. This kind of anaphora is
probably the type involved in interpretation of traces, and perhaps with reflexives of certain types; it has an
effect rather like a bound variable in logical notation.
98 RAY JACKENDOFF

Section 3.4.2 pointed out a different class of cases, in which a PP appears to take
subject function:

(4.96) a. It’s a long way to Tipperary.

b. It's a long time from Groundhog Day to Purim.

These PPs can be moved into the subject in the same environments as that and for-to
complements can, and there is no problem in generating them under V". To interpret
(4.96), however, a rule of anaphora is needed which specifies the PP as what is being
measured, i.e. the subject. This is the same rule needed for the Intraposition analysis.
Thus, since the anaphora rule is needed anyway, it does not weigh against Emonds’s
theory, as originally thought.
The fact that the necessary rule of anaphora may be a surface structure rule of
interpretation provides the key to a problem of derived structure raised by Higgins
(1973). Higgins is concerned with the derivation of sentences like (4.97) (his (58)):

(4.97) a. It is thought to be unwise that you didn’t try harder,

b. That you didn’t try harder is thought to be unwise.

In the Extraposition theory, these are both derived from (4.98a), by Passive and
Raising in the main clause. It is not clear on which cycle Extraposition takes place in
(4.97a), so its derived structure could be either of the options in (4.98b). The fact that
Extraposition is generally obligatory in subordinate clauses argues that the complement
is probably attached to the lower VP.

(4.98) a. S

think S
 4: COMPLEMENTS

b. S

\
\

be unwise

Higgins assumes that in the Intraposition theory the deep structure of (4.97) is (4.99).
Hence (4.97a) is derived by Passive and Raising of it; the derived structure is (4.98b),
with the complement attached to the lower VP.
(4.99)

PRO Pres V

think

To produce (4.97b), Intraposition now has to raise the complement over a clause
boundary, and Higgins correctly shows that this raising produces difficulties.
100 RAY JACKENDOFF

However, closer consideration of (4.97a) shows that the complement clause is

attached in derived structure to the higher VP, not the lower one. For if the by-phrase
of the passive is realized lexically, it precedes the complement rather than following it:

(4.100) a. It is considered to be unwise by many people not to try harder,

b. ?*It is considered to be unwise not to try harder by many people.

In the Extraposition theory this fact is difficult to account for, since the obligatoriness
of Extraposition in subordinate clauses would force the that-clause to attach to the
lower VP, on the earlier cycle. However, if the rule of anaphora in the Intraposition
theory is like all other anaphora rules in applying to surface structures, it can be
restricted to the environment (4.101), making only the higher attachment of the
complement possible as a derived structure.

(4.101)

A movement over a clause boundary is thus unnecessary in deriving (4.97b) by

Intraposition, and the problem of properly locating the it relative to the complement S
does not arise.
This analysis of course implies that the deep structure of (4.97) also has the
complement in the upper clause, thus:

(4.102)
 4: COMPLEMENTS

This seems to violate our linguist’s intuitions about grammatical relations, since the
clause that you didn’t try harder is entirely outside of the clause it to be unwise of
which it is understood as the subject. This apparent anomaly disappears when we
realize that this subject relation is not a normal grammatical relation, but is established
only indirectly by a surface structure rule of anaphora. If it were not raised into the
upper clause, the anaphora rule could not apply, and that you didn't try harder would
be left without a semantic function.
This section has dealt with three objections to the Intraposition theory. First, we
showed that the double-complement verbs such as prove should not be accounted for
with doubly filled nodes, as Emonds suggested; rather, one complement is in V' and
one in V". The constraint against concatenated complements of the same type applies
equally in the Intraposition and Extraposition theories. Second, the rule of anaphora
relating it to the V" complement was shown to have independent motivation in both
English and other languages. Third, the fact that the anaphora rule applies to surface
structure enables the theory to meet Higgins’s objection about derived structure.

4.9. List of Transformations

Dative Shift (section 4.2)

(4.31) X - Va - np - (Prt) - NP - [PPto - NP] - Y

^>l_2-7-4-5 - </>-</>-8
OPTIONAL

Agent-Postposing in Ss (possibly) (section 4.7)

NP-Y-V-Z - [ppby np] - W
=>np-2-3-4-5 - 1-7
OPTIONAL

Generalized NP-Preposing (section 4.7)

-l-Subj
(4.83) np - Y - <+Obj> - <(P)> -NP-Zz>5-2-3-4-0-6
. +Comp
OPTIONAL

NP-Postposing (section 4.7)

(4.85) NP -Y-N-np-Z^np-2-3- 1-5

OPTIONAL

Intraposition (sections 4.3, 4.8)

[v„it - Y - [V»Z -S]-W]=>4-2-3-§-5
OPTIONAL—ROOT (may generalize to movement of [ + Obj]'")
102 RAY JACKENDOFF

Extraposition of S (section 4.2)

[X,Z - S] - Y - s]x. => 1 - s - 3 - 2

OPTIONAL

Extraposition of PP (section 4.3)—generalizes with Extraposition of S

[ x-Z - PP - W] - Y - pp - U]x. z>l-pp-3-4-2-6

OPTIONAL

Oi-Insertion (section 4.2, generalized in section 4.4)

[X'
X - (Prt) -NP-Y] =>l-2-o/+ 3- 4
L-obj J
OBLIGATORY

Poss-Insertion (section 3.4)

[n„N'" - Y - NT => I + poss - 2 - 3

OBLIGATORY

The-Insertion (section 4.7)

[N*np - Y - NT => the - 2 - 3

OBLIGATORY

That -Deletion (section 4.4)

[P-P - that - S] => I - <f> - 3

OBLIGATORY

Output Constraints

? Y - Adv - Adv - Z (section 4.3)

7Y-S-S-Z (section 4.8)
 5: NP Specifiers

5.1. Introduction

Chapter 3 dealt with aspects of the specifier that are central to the syntax of sentences:
the subject and the auxiliary. This chapter and the next extend the study of specifiers to
systems which play a minor role in sentences but which are important in NPs, APs, and
PPs: determiners and expressions of quantification and degree.
There are problems in studying specifier systems that do not arise in studying
complements. First, specifier systems involve very small numbers of lexical items and
are riddled with idiosyncrasies. Thus general phrase structure rules must be supported
on the basis of impoverished and skewed surface distributions. Only with judicious use
of lexically determined transformations and interpretation rules can any order at all be
brought into the system. One way to bring more data to bear on the problems is to
exploit the hypothesis of cross-category generalization, and that will be a major source
of evidence here.
A second problem with specifier systems, at least in English, is that it appears
much less possible to correlate semantic regularities with syntactic positions, as we did
in complements. This may of course be a function of our ignorance about the semantics
of deixis, quantification, and measuring; but it may also be a brute fact about English or
about specifiers in general. In the absence of a coherent semantic theory of specifiers, I
will make the latter assumption, hoping that a better understanding of the syntax may
make possible a more disciplined approach to the semantics.
This chapter deals with the syntax of the NP specifier. Chapter 6 extends this
account to AP and PP, proposing a reanalysis of Bresnan’s (1973) theory of the degree
system.

5.2. Articles and Quantifiers

5.2.1. Semantic Roles of NP Specifiers

There are three different semantic roles that the elements of the NP specifier can play.
Without being able to define these semantic roles precisely, we can still make the
distinction among the three roles intuitively, at least for the clearest cases, and let the
104 RAY JACKENDOFF

most general analysis decide in the unclear ones, hopefully to be corroborated by

evidence yet to be found. Among the demonstratives are the traditional definite articles
the, this, that, these, those, the interrogatives which and what, and possibly the
indefinite article a and the singular some (as in Some man is at the door). Among the
quantifiers are each, every, any, all, no, many, few, much, little, and other uses of
some. The third class is numerals, including all the usual cardinals plus a dozen, a
couple, a few, and a little. Notice already the confusion that is cropping up, for some
belongs to different categories depending on its stress, and negative few and little
belong to a different category from nonnegative a few and a little. These irregularities
will be made clearer in the course of the analysis.
A highly plausible semantic constraint can be based on distinguishing these three
specifier roles:

(5.1) Specif er Constraint

An NP specifier may contain at most one demonstrative, one quantifier,
and one numeral.

We will show that (5.1) interacts with the syntactic constraints on specifiers to produce
some of the otherwise curious aspects of specifier distribution.

5.2.2. Phrase Structure Rules

Let us turn now to the syntax itself. If N' is reserved for strictly subcategorized
arguments, we are left with two NP specifier positions to account for: the N" and the
N'" specifier. The N'" specifier is the position where genitive NPs occur in complemen¬
tary distribution with demonstratives. We have proposed two possible ways to generate
this alternation, in sections 3.7 and 4.5. Assuming the former for simplicity, we have
rule (5.2) generating demonstratives in the N'" specifier.

(5.2) N'"-^({Art '"})- N"

Genitive NPs and demonstratives may be followed by certain quantifiers, but not
by others:
'some '
Fred’s' Fred’s
each many
the the
(5.3) a. < , > - all dwarf(s) few dwarfs
those those
no several
\
which /
which
any /

There is no apparent semantic reason for this division: why should *Fred’s some apples
mean anything less innocuous than Fred's many apples? In order to account for this
distribution, then, we will resort to dividing the class of semantic quantifiers into two
syntactic categories. Those in (5.3a) will be assigned the category Art, which is
generated only by phrase structure rule (5.2). Their inability to cooccur with demon¬
stratives is a syntactic constraint, namely the availability of only one Art position in the
 5: NP SPECIFIERS

NP specifier. The quantifiers in (5.3b), however, will be assigned to the syntactic

category Q, which we will generate in the N" specifier:

(5.4) N"-» (Q'") - (A'")* - N' - . . .

We choose the N" specifier rather than the N'" specifier for this QP for reasons to be
discussed in section 6.3. Our analysis thus assigns (5.3b) this structure:

(5.5) N"'

Since some quantifiers are now Arts and some are Qs, the phrase structure
component will generate structures in which two quantifiers appear, one in each
position, e.g. *no many men, *all several men, *any much wine. These are ruled out
semantically, however, by the Specifier Constraint (5.1), which forbids two (semantic)
quantifiers in the same NP specifier. If the second quantifier is replaced by a numeral,
so that the Specifier Constraint is satisfied, acceptable constructions like no three men,
all six men, and any dozen men are generated.
For convenience in exposition, we will when necessary refer to the quantifiers in
(5.3a) as N"' quantifiers and those in (5.3b) as N" quantifiers. The former have the
syntactic features [+Subj, -Obj, -Comp, +Det] and the latter [ + Subj, -Obj, -Comp,
-Det].1

1 A few doubtful cases should be mentioned here. Little behaves like an N" quantifier but its positive
much behaves like an N'" quantifier:
John’s [ little j
(0 the Ji*muchJ
food

In most other relevant contexts, though, much patterns with many, few, and little, as will be seen throughout
this chapter. It is difficult to know whether to call much an Art in this use or to call it a Q and introduce
arbitrary restrictions that prohibit it from occurring with Art. More research into the semantics of English
quantifiers may help decide which is the appropriate solution. I assume it is a Q.
A second apparent anomaly is that every (but not its near relative each) can occur after genitives but not
after articles in the semi-idiomatic construction (ii).
,... , I Fred’s! fmovel
(..) Wehungon(*the j every {word j.

It is not clear whether every is functioning in its normal syntactic position here (indicating that it is actually a
Q) or whether this use as a Q is simply an idiomatic exception. I assume the latter.
106 RAY JACKENDOFF

5.2.3. We and You

We should mention two little-remarked articles in English which provide evidence for
the correctness of the feature system. Postal (1966) is concerned with the connection
between the normal pronominal use of we and you and their use in expressions like we
three guys, you five rascals. Plural but not singular vow has this use productively: I saw
you guys/*guy yesterday; though (inexplicably) singular you can be used this way in
vocatives: Come here, you bastard! There are dialects in which them guys is also
acceptable.
Postal effectively discounts the theory that the pronoun is head of this construction
and the other elements are appositives (we, who are three guys). Thus the pronoun is
apparently in article position. Since Postal has no syntactic features to work with, he
resorts to describing the relationship between the two uses by deriving the pronominal
use from an underlying we ones, so that we is always an underlying article. He
proposes this mechanism as a general derivation for all definite pronouns.
In the present framework there is an alternative. We and youplut can be assigned
the feature matrix [ + Subj, — Obj, + Det], leaving the feature Comp unspecified. Then
they can be placed in either N or Art positions. No transformations or abstract
underlying forms are necessary. Note that we and you can then be substituted into the
determiner of an NP in two distinct ways:

(5.6) a. N

N'" N" Art'" N"

Jwe 1
lyouj

(5.6b) will remain in its underlying form, giving we/you three guys. (5.6a), however, will
undergo Povs-Insertion to become ourlyour three guys.
To recapitulate, this section has shown that there are three semantic sorts of NP
specifiers: demonstratives, quantifiers, and numerals. They are divided up among two
NP specifier positions, in N'" and in N", and are of the categories Art and Q,
respectively. Demonstratives always appear in N'", but some quantifiers appear in N"'
and some in N". In order to rule out structures with two quantifiers, one in each
position, we appealed to the Specifier Constraint, a semantic condition. Finally, the use
of we and you as articles has provided evidence for the feature system.

5.3. The Partitive

All of the quantifiers mentioned in the previous section occur not only with a simple
noun (e.g. many men, all horses) but also with a definite of-phrase called a partitive
 5: NP SPECIFIERS

(e.g. many of the men, all of the horses). Some of them alter their form when used with
a partitive (e.g. every man, no men but every one of the men, none of the men)', this
alternation will be discussed in section 5.3.4. All numerals allow a partitive phrase; so
do the demonstratives those and which, and the adjective (?) other and superlatives
(e.g. the oldest of the men). This section will deal with the structure of the partitive
construction.2

5.3.1. Partitives as N' Complements

Many sources (including Chomsky in “Remarks”) account for the partitive construc¬
tion by treating the noun following of as the head of the construction, and the quantifiei
preceding of as a “prearticle”. Of is obligatorily inserted if the article is definite:

several the

However, Jackendoff (1968) proposes a different structure, in which of the N" is

part of the N' complement. For justification, consider definite of-phrases after group
nouns, e.g. a group of the men, a gallon of the wine. In these phrases, of the N"
behaves like an ordinary object of the NP, for it always precedes N" modifiers, as seen
in (5.8), and the pronoun ones cannot substitute for the head alone, as seen in (5.9).

(5.8) a. a gallon of the wine

(in thekitchen I
from Sicily J

, * „ Jin the kitchen I r .

h. *a gallon (from sicily j of the wme
(5.9) a. groups of the men from Siberia and ones from Japan (= groups of the
men from Japan)
b. *groups of the men and ones of the women

Selkirk (1977) further points out that these of the N" can undergo Extraposition
from NP, which regularly moves N' complements.

2 As will be seen, the arguments of Selkirk (1977) have contributed greatly to this section and the next.
These are in turn based on Akmajian and Lehrer (1976). I understand that Jean-Claude Milner’s recent
doctoral dissertation reaches similar conclusions, but I have not had the opportunity to see it.
108 RAY JACKENDOFF

(5.10) (= Selkirk’s (65))

a. A lot of the leftover turkey has been eaten.
A lot has been eaten of the leftover turkey.
b. Only a handful of those questions concerning electromagnetism were
asked.
Only a handful were asked of those questions concerning electro¬
magnetism.
c. He gave a rather large number of his books by famous authors to
Mary.
He gave a rather large number to Mary of his books by famous
authors.

Thus a structure (5.11), which treats of the N" as a constituent in the N' complement, is
most plausible for group nouns.3

(5.11)

The partitive o/-phrase serves the same function after quantifiers as after group
nouns: it designates a set out of which certain individuals (or a certain subset) is
selected. For count nouns, the o/-phrase must be plural.4
Quantifiers and group nouns with partitives share three other properties. First,
both prohibit quantification in the o/-phrase (see section 5.3.3 for more discussion):

3 Our account of NP objects says that of N'" is not a PP, since of is a transformationally inserted
grammatical formative. The fact that Extraposition moves it like any other PP may be evidence for the
position of Jackendoff (1974a), in which NP objects are really PPs with an empty P filled by of. I abandoned
this theory in chapter 4 because the evidence for it was not as strong as I previously thought, in light of other
revisions, and the complication of the object relation was so severe. Henceforth I present trees in a way
neutral between the two theories. See also chapter 4, note 13, p. 70.
4 This needs appropriate modification to deal with the semantics of mass nouns, but in any event the
semantic parallelism holds: compare a gallon of the wine and much of the wine.
 5: NP SPECIFIERS

* ( a group

(5.12) <
a number
> of
{some men
many all men
few

Second, if an N" quantifier or a group noun is preceded by the and has a definite of-
phrase, a restrictive relative clause is necessary.

(5.13) a. *the group of the men

*the many of the men5
b. the group of the men that you met
the many of the men that you met

This relative clause is not attached to the o/-phrase, as can be seen if the o/-phrase is
preposed:

(5.14) a. *Of the men that you met, mTn^

^ l e many j
f aren
t ^ere anymore-

the group
b. Of the men, that you met aren’t here anymore.
(the many

Third, as Selkirk (1977) points out, the partitives after quantifiers, like those after group
nouns, can undergo Extraposition from NP. The examples in (5.15) are comparable in
acceptability to those in (5.10).

(5.15) a. (Not) much has been eaten of the leftover turkey.

b. Only a few were asked of those questions concerning electromagne¬
tism.
c. He gave several to Mary of his books by famous authors.

Each of these arguments raises vexing problems for a theory like (5.7). First, if
quantifiers can take relative clauses of their own, “prearticle” must really be some sort
of NP. But (5.7) then incorrectly predicts that a relative clause on the prearticle should
precede the head noun rather than follow it. Second, the preposing rule which derives
(5.14b) would have to extract the head of the phrase, leaving the modifiers behind, an
otherwise unprecedented operation. Third, Extraposition from NP would have to move
a nonconstituent including the head, instead of a PP in the N' complement.
By contrast, the theory that the partitive is an N' complement explains the
syntactic parallelisms exhibited in these examples. Parallel (though still unknown)
constraints can be stated for the restrictions demonstrated in (5.12)-(5.13). Since
relative clauses always follow N' complements, the proper order of partitive and
relative clause in (5.13) follows from the base rules. And since transformations can in
general remove complements from their heads, the preposing in (5.14) is within the

5 I have no explanation for why the group of them is all right but *the many of them is not.
1 10 RAY JACKENDOFF

province of traditional transformations, and Extraposition from NP can apply in normal

fashion'.

5.3.2. Position of the Quantifier in Partitive Constructions

This theory of the partitive leads to two possible deep structures for many of the men.

(5.16a) is essentially the structure which Jackendoff (1968) chose, in part because
empty nodes were not an available device at the time. Ideally it requires some
explanation of why quantifiers cannot be preceded by adjectives; (5.16b) requires no
such explanation. We will develop solution (5.16b) here; section 5.5 will show that
numerals (but not quantifiers) do occur in construction (5.16a) and in fact can be
preceded by adjectives in that use.
To develop the description of (5.16b), we need a projection rule to assign the
empty head PRO an interpretation something like "unit(s)” when it is immediately to
the right of any of the articles, quantifiers, or adjectives that govern partitives. All of
these words must be lexically marked to govern the projection rule.6

(5.17) Partitive Projection Rule

X
pron UNIT/
_+partitive .

(5.17) interprets PRO in configurations such as (5.18a) and (5.18b), as well as in

(5.16b).

6 The interpretation here given roughly as UNIT must be general enough to apply to mass phrases too,
e.g. some of the wine, where perhaps the interpretation AMOUNT is more appropriate than UNIT. Wald
(1976) develops a theory of "slices”, based on nonatomic mereologies, in which mass nouns and plurals can
be dealt with homogeneously. The term common to UNIT for count nouns and AMOUNT for mass nouns is
Wald’s term SLICE.
 5: NP SPECIFIERS

Art'" N" Art'" N"

PRO the men PRO the men

But in (5.19a) PRO cannot be interpreted because it is not adjacent to the partitive word
many, and in (5.19b) PRO cannot be interpreted because there is no partitive word to
govern the rule. Since an uninterpreted PRO produces anomaly, *many old of the men
and *your of the men are eliminated.

(5.19) a. N"

N"

you(r)

PRO the men

This subsection has shown, then, that (5.16b) is a practicable structure for many of
the men, given the Partitive Projection Rule. Since we are saving (5.16a) for another
construction, (5.16b) is our only choice.7

5.3.3. The Partitive Constraint

Given structure (5.16b) for many of the men, we must now ask what happens to this
structure if the lower NP is indefinite, as in for example (5.20).

7 It is worth mentioning that the ungrammatically of *many old of the men is a stumbling block to many
earlier analyses of the partitive. After trying many alternatives, I have convinced myself that there is no
purely syntactic solution. See further discussion in section 5.3.4, particularly footnote 8.
1 12 RAY JACKEN DOFF

(5.20) N

Jackendoff (1968) claims that the of in this construction obligatorily deletes, giving
another possible source for many men besides the simple one of section 5.2, (5.21).

(5.21) N'"

N"

However, Selkirk (1977) argues that this claim makes a number of incorrect
predictions. Hence only (5.21) is a correct structure, and some way must be found to
eliminate the underlying form (5.20).
One of Selkirk’s arguments involves cases where a definite N' complement is
acceptable, but the indefinite without of predicted by (5.20) is impossible:

(5.22) a. much of that story/*much story

b. as little of his speech/*as little speech
c. some of the book/*some book (as [—count])
(5.23) a. one of her brothers and sisters/*one brother and sister
b. one of the cattle/*one cattle
c. one of her successive failures/*one successive failure
 5: NP SPECIFIERS

Conversely, (5.20) predicts that one single thing has the impossible definite counterpart
*one of the single things. (5.21), on the other hand, deals with these facts easily, as part
of the rule enforcing selectional restrictions and number agreement between quantifiers
and their heads.
Second, Jackendoff (1968) claims that Of-Deletion corrects the number of the head
noun when the quantifier is singular, so that underlying each of men becomes each man
rather than *each men. But Selkirk points out that it could not correct the number
within a relative clause on the head noun. Thus we would incorrectly expect paradigms
like (5.24).

(5.24) each of the politicians who were elected/*each politician who were
elected

(5.21) would avoid this problem, since politician in each politician would be underly-
ingly singular, not plural.
Third, Extraposition from NP would be expected to apply in (5.20) just as it does
in (5.16b), yielding sentences like (5.25), parallel to (5.15).

(5.25) a. *Not much has been eaten (of) leftover turkey.

b. *Only a few were asked (of) questions concerning electromagnetism.
c. *He gave several to Mary (of) books by famous authors.

On the other hand, (5.21) claims that turkey, questions, and books are heads of the
subject NPs in (5.25), and that Extraposition from NP should move constituents that
are one level further down into the NP. This prediction is correct:

(5.26) a. Only a few questions were asked concerning electromagnetism,

b. He gave several books to Mary by famous authors.

Thus the simpler (5.21), generated by the rules of section 5.2, is the only possible
underlying structure for many men. Selkirk proposes therefore to filter out (5.20) by a
generalization of the constraint illustrated in (5.12) that rules out *many of some men,
*many of all men, and various other specifiers in the partitive NP.
The nature of the constrainst that filters out (5.12) and (5.20) is not (to my
knowledge) yet understood. In observational terms, it seems to be approximately
(5.27).

(5.27) Partitive Constraint

In an of-N'" construction interpreted as a partitive, the N"' must have a
demonstrative or genitive specifier.

Thus, (5.27) rules out *many of all men, *many of some men, and *few of many men,
as well as *many of men, since they all lack demonstrative or genitive specifiers; but it
permits many of the men, few of the many men, and many of his friends.
Notice that (5.27) refers to semantic functions in the specifier, not to syntactic
positions. This argues that (5.27) is part of the semantic component, and further
1 14 RAY JACKENDOFF

motivation for it should be pursued in the semantics of specifiers and partitives. I am

not prepared to do that here. However, the observation that the Partitive Constraint
deals with the semantics of the construction removes any burden from the syntactic
component of filtering out the offending structures.
This subsection has thus established that many men is generated by the syntax as
(5.21); structure (5.20) does not undergo 0/-Deletion, and though it is syntactically
well-formed, it is ruled out by the Partitive Constraint, a part of the semantic
component.

5.3.4. Substantivization

To complete the account of the partitive, we need to deal with the phonological change
that takes place in certain articles that govern partitives:

(5.28) no men / none of the men

every man / every one of the men
each man / each (one) of the men

It is often thought that the change in form is a phonological coalescence of the article
with one, so that none of the men, for instance, is underlyingly no one of the men. But
since none can refer to mass nouns, as in none of the rice, this view is mistaken. We
will treat the alternation rather as the addition of an inflectional marker subst, added by
the following rule:

(5.29) Substantivization I
Art - [nPRO] 1 +subst - 2
OBLIGATORY

No+subst will be spelled as none; every +subst will be spelled as every one, etc. For
those articles such as all which do not change form, the addition of subst will not
change phonological shape.
This sort of rule changing phonological shape might be thought to violate the
Lexicalist Hypothesis. However, since it occurs with only a limited number of
morphemes and, unlike nominalization, has no semantic effect, we can regard it as a
rule of inflectional morphology, like Poss-Insertion and Affix Hopping, which the
Lexicalist Hypothesis permits.
In order to show that (5.29) is independently necessary, we will show that its
application is not restricted to partitive constructions; other rules which interpret a
PRO also create the appropriate configuration. One such rule can be called N'-
Anaphora (N-Deletion in Jackendoff (1971)). This is the rule operative in the final NPof
(5.30).
 5: NP SPECIFIERS

(5.30) John wanted to read the dossiers of famous linguists. he

/ Morris’s
yours
these
one
every one
succeeded in reading S none
some
many
three
one
*ones

The anaphoric element in this construction is a pro-N' PRO which occupies the head of
the NP, referring in (5.30) to the N' dossiers of famous linguists. Any specifier is
possible before PRO (as long as there is no intervening adjective); the impossibility of
ones shows that one in this construction is the specifier one, not the noun.
In the N'-Anaphora construction, the same quantifiers change phonological shape
as did in the partitive construction above. In addition, the possessive pronouns change
shape: my, your, etc. are replaced by mine, yours, etc. Again, the change cannot be a
result of coalescence with an underlying head one(s), since N'-Anaphora, unlike one(s)-
Anaphora, can apply to antecedents with mass nouns as heads.8

yours
I and
(5.31) John wanted some rice. ■ none ' was moldy.
[but
some

The proper analysis appears to be an extension of (5.29) to include possessive

pronouns.

(5.32) Substantivization II
f Art )
NP
+Poss - [nPRO] => 1 +subst - 2
.Pro.
OBLIGATORY

8 The fact that this construction appears with mass nouns goes unnoticed by Postal (1966), who claims
to derive all pronouns from underlying articles plus one. The solutions of Perlmutter (1970) and Jackendoff
(1969c) suffer from the same objection, though they at least note it, with some puzzlement. Sommerstein
(1972) has the opposite problem: he inserts one if the head has been deleted and there is no determiner or
noun left for the number morpheme to attach to. Thus his rules incorrectly derive *My coarse sand is next to
your fine one. In the present theory these objections (raised also by Siegel (1974)) do not arise, because one
and PRO are distinct anaphoric elements in NPs, not transformationally related.
116 RAY JACKENDOFF

Substantivization also applies when PRO is interpreted by the rule of N'- Gapping
(see Jackendoff (1971)). This rule differs from N'-Anaphora in that it can apply only if
the antecedent is directly conjoined to the gapped NP, and in that it permits N'
complements to be left behind.

(5.33) a. *Joe’s destruction of the pictures was appalling, and Fred’s of the
records was even worse.
(N'-Anaphora: NPs not conjoined, but N' complement left behind)
b. Joe’s destruction of the pictures and Fred’s of the records appalled
everyone.
(N'-Gapping: NPs conjoined)

Again, the missing head can be regarded as a PRO; so, for example, the subject of
(5.33b) is Joe’s destruction of the pictures and Fred's PRO of the records.
N'-Gapping is of considerable interest to the theory of syntactic distinctive
features, for it has properties very similar to the well-known rule of Gapping in
sentences. These similarities are discussed in Jackendoff (1971), and 1 have nothing to
add here. What is germane to the present discussion is that Substantivization applies in
N'-gapped constructions as well, yielding yours and none in (5.34), for example.

(5.34) a. Joe’s destruction of the pictures and yours of the records

b. few stories about Fred and none about Bill

A fourth application of Substantivization is in “postposed genitive” constructions

such as a friend of ours. This o/-phrase occurs after certain N'-complements but
appears to precede N"-complements:

(5.35) a. a picture of Bill of John’s

?*a picture of John’s of Bill
b. a picture of John’s with a gold frame9
?*a picture with a gold frame of John’s

The source for this construction proposed in Jackendoff (1969c), which I see no reason
to abandon, comes out in the present framework as (5.36).

9 (5.35b) shows that this o/-phrase cannot be produced by extraposing the genitive phrase from the
determiner to the end of the NP, as some have claimed, including Siegel (1974).
 5: NP SPECIFIERS

(5.36) N'

\
\ \

>

PRO

We in (5.36) undergoes Poss-Insertion and Substantivization to become ours.

This source preserves the generalization that genitive NPs arise only in determiner
position. However, the underlying position of the postposed genitive is not in the
determiner of the main NP, from which it is then extraposed. Instead, it is in the
determiner of a complement NP rather like a partitive. (5.36) is thus similar to the
structure of one of our friends, except that friend is in the main NP rather than in the
partitive. PRO must be interpreted by a special projection rule, devised for this
construction. Again, an underlying one(s) instead of PRO is impossible, because of
mass constructions such as some information of theirs.
We have thus shown that Substantivization is a well-motivated rule of English,
applying to N'" specifiers followed by PRO, where PRO can be interpreted by the
Partitive Projection Rule, N'-Anaphora, N'-Gapping, or the Postposed Genitive Projec¬
tion Rule. Thus, the alteration of the form of no, every, etc., and the use of a PRO head
are not peculiar to partitives, but rather are independently necessary. These results
thus confirm the analysis of the partitive given above.
To sum up the results of this entire section on the partitive, we have shown that
quantifiers and certain demonstratives occur in the following grammatical structures:
1 18 RAY JACKENDOFF

(5.37) Demonstratives
b. N'"

tree(s)
(5.38) N'" Quantifiers
a.

r
each >V each
all . • all
[some C.
some J

tree(s)
(5.39) N" Quantifiers
a. N b. N'"

trees PRO the trees

 5: NP SPECIFIERS

Each of these specifiers is marked [+Partitive] and thus in the (b) cases PRO can be
interpreted by the Partitive Projection Rule; the proper specifiers in the subordinate
NPs are guaranteed by the Partitive Constraint. Certain of the N'" specifiers in this
construction change their form in case (b) through the application of the rule
Substantivization.
This completes our discussion of the partitive. We now turn to two other
constructions in the NP specifier, pseudopartitives and numerals.

5.4. Pseudopartitives

5.4.1. Arguments for the Construction

Section 5.3.1 argued that group nouns followed by partitives such as a bunch of the
men have the following structure:

(5.40) N'"

ackendoff (1968) assumes that when the lower NP is indefinite, as in a bunch of men,
the structure is essentially the same:

(5.41)
120 RAY JACKENDOFF

However, Selkirk (1977) argues that a bunch of men has a different structure, (5.42),
which she calls the pseudopartitive. Her arguments are quite similar to those for
quantifiers given in section 5.3.3.

(5.42) N'"

In this structure, the grammatical formative of is a daughter of N". Other than the of
(5.42) is parallel to the structure of many men, (5.21).

(5.21) N"'

N"

Before presenting three of Selkirk’s arguments, let us see how a solution like (5.42)
affects the grammar. Ignoring the of for a moment, we see that there is now a choice
between Q'" and N'" in the N" specifier. In a theory without syntactic distinctive
features, this would involve a complication of the grammar. But since, in the present
theory, Q'" and N'" differ only in the feature Comp, the choice between Q"' and N'"
makes the grammar actually one feature simpler than if Q'" alone were possible. The
rule for N" thus reads like this:
 5: NP SPECIFIERS

+ Subj
(5.43) N" ( -Obj ) - (A'")* - N' - . . .
Det

Chapter 6 will elaborate on this rule considerably.

Returning to the of, we have the choice of introducing it in underlying structure
(with suitable use of angle brackets in (5.43) to make sure it appears only in the N'"
case), or by a simple local transformation. Section 6.2 will give one argument that it
should appear as a specified grammatical formative in the phrase structure rule; we will
assume this position in advance, though little depends on it.
Selkirk’s first argument for (5.42) concerns the Partitive Constraint of section
5.3.3. She points out that group nouns with partitives are apparently subject to the
Partitive Constraint, since the specifier of the NP after of must contain a demonstrative
or genitive:

(5.44) *a bunch of some men

*a gallon of much wine
*a number of all objections
(5.45) a bunch of the men
a gallon of our wine
a number of those objections

Since one of the things the Partitive Constraint rules out is a null determiner in the
partitive phrase, it rules out (5.41), unless (5.41) is an otherwise unmotivated exception.
On the other hand, (5.42) contains no NP interpreted as a partitive, so the Partitive
Constraint does not apply to it.10
Selkirk’s second argument concerns the application of Extraposition from NP.
Recall that in a definite partitive with structure (5.40), Extraposition from NP moves
the o/-phrase:

(5.10) a. A lot of the leftover turkey has been eaten.

A lot has been eaten of the leftover turkey,
b. Only a handful of those questions concerning electromagnetism were
asked.
=> Only a handful were asked of those questions concerning electro¬
magnetism.

10 Note that there is another construction, exemplified by a group of three men, in which the o/-phrase
does not contain a demonstrative specifier. But this o/-phrase does not have a partitive interpretation, since a
group of three men means not a group taken out of three men, but a group consisting of three men. Hence the
Partitive Constraint does not apply, and a nondemonstrative specifier is permitted. The possibility of such a
“consistive” interpretation is what makes a pot of steel ambiguous: one reading is like a group of men, has
structure (5.42), and denotes a certain amount of steel; the other reading is like a group of three men, has
structure (5.41), and denotes a pot made out of steel. In fact, the consistive reading seems to require a
nondemonstrative specifier and thus is mutually exclusive with the partitive reading. The Partitive Constraint
may thus in fact be part of the structural description of the semantic rule that differentiates between these
two readings of the N' complement.
122 RAY JACKENDOFF

c. He gave a rather large number of his books by famous authors to

Mary.
=> He gave( a rather large number to Mary of his books by famous
authors.

If (5.41) were the structure for a bunch of men, we would expect Extraposition from
NP to apply in precisely parallel fashion. However, it does not:

(5.46) a. ?*A lot has been eaten of leftover turkey.

b. ?*Only a handful were asked of questions concerning electromagne¬
tism.
c. ?*He gave a rather large number to Mary of books by famous
authors.

On the other hand, structure (5.42) correctly predicts the unacceptability of (5.46),
since it claims that, for example, leftover turkey is not an N'" in the N' complement.
Furthermore, structure (5.42) predicts a different result. It requires a rather large
number of books by famous authors to have this structure:

(5.47) N'"

N"

Thus, if Extraposition from NP applies to anything in this structure, it should apply to

by famous authors. This prediction is correct:

(5.48) a. Only a handful of questions were asked concerning electromagnetism,

b. He gave a rather large number of books to Mary by famous authors.

Hence (5.42), not (5.41), predicts the correct application of Extraposition from NP.
A third argument concerns the interpretation of relative clauses. Selkirk points out
that in (5.49a) the relative clause can apply to either a number or those daffodils, and
the interpretations are quite distinct, whereas in (5.49b) there is only one interpretation.

(5.49) a. She bought him a number of those daffodils, only two of which were
faded.
b. She bought him a number of daffodils, only two of which were faded.
 5: NP SPECIFIERS

This difference is easily accounted for if a number of those daffodils includes two N'"
to which relative clauses can be attached, but a number of daffodils contains only one
such N'". This is predicted by the theory that a number of those daffodils has structure
(5.40) and a number of daffodils has structure (5.42). It is not predicted by the theory
that a number of daffodils has structure (5.41).
We conclude therefore that a group of men is not a true partitive. The syntax
does in fact generate the string ambiguously, with partitive structure (5.41) and
pseudopartitive structure (5.42), but (5.41) is ruled out semantically by the Partitive
Constraint.

5.4.2. Pseudopartitive Specifier Constraints

Since pseudopartitives contain an N'" in the N" specifier, there is nothing to stop this
subordinate Nfrom having its own specifiers. Thus the phrase structure rules cannot
be prevented from generating such ungrammatical combinations of specifiers as these:
* John’s John’s
all a
(5.50) i > < > group of trees
some every
_ N'" those Nw _N"' that -N' -
' y ^ y -
It is clear that only one Nspecifier may be allowed to precede the group noun. The
problem is whether the permitted specifier is always that of the matrix N'", always that
of the subordinate N'", or sometimes one and sometimes the other.
First consider articles. It appears that when a group noun in a pseudopartitive is
preceded by an Art'", it is always the Art'" of the group noun, not that of the matrix
N'". To see this, notice that when the group noun is singular and the head noun is
plural, an Art before the group noun must be one that permits a singular head.

a
every
(5.51) aJ > group of people
this
that

all
we
you group of people
these
those

On the other hand, if the group noun is plural, the Art is one that permits a plural head.

all
we
(5.52) i you groups of people
these
Those
124 RAY JACKENDOFF

Since the number of the group noun controls the choice of articles, the assumption that
the article is a specifier of the group noun keeps the statement of these particular
agreement constraints the simplest.
Similarly, where quantifiers precede a group noun in a pseudopartitive, they
appear to quantify over instances of the group noun rather than over the head noun.
For example, all groups of people quantifies over instances of groups, not of people.
Compare this to all three people, which quantifies over instances of people (of whom
there are three), and which we have assumed has all in the N'" specifier of people. If
this semantic difference is to be represented syntactically, all in all groups of people
must be a specifier of groups.
Thus the evidence seems to indicate that an NP containing a group noun in its N"
specifier cannot have an Art'" in its N'" specifier. Can it have an N'" instead? The
answer seems to be yes. To see this, consider group nouns such as number and
quantity which place restrictions on their specifiers. When used in genuine partitives,
they cannot be used with a possessive:

(5.53) *John’s rather large number of those objections

*Fred’s substantial quantity of that wine

But possessives can precede them in pseudopartitives:

(5.54) John’s rather large number of objections

Fred's substantial quantity of wine11

This difference can be explained if number and quantity exclude possessives from their
N"' specifiers: in the true partitives (5.53), the possessives will have to appear
improperly in the specifiers of number and quantity, as in (5.55a), but in (5.54) they
may be in the specifier of objections and wine, as in (5.55b).

(5.55) a.

ft
N"' N

John’s

11 I have no explanation for the fact that (5.54) becomes much worse if the adjectives are omitted. Note
that this paradigm is parallel to the one with N" quantifiers, since we have *John's many of those objections,
like (5.53), but John's many objections, like (5.54).
 5: NP SPECIFIERS

b. N'"

Thus it is possible to have a genitive in the matrix N'" specifier. There also seem to
be cases where a genitive in the group-noun specifier is necessary. The noun blame
does not permit a genitive specifier, since, for example, only the blame for the accident
is possible, not *Fred’s blame for the accident. Nonetheless, the pseudopartitive
construction with blame does permit a genitive, since Fred’s share of blame for the
accident is possible. The simplest account of this seems to be to attribute the genitive
to the specifier of the group noun:

(5.56) N'"

Fred’s N'

share

Thus, in pseudopartitives, only the group noun may contain an Art"', but
apparently either the group noun or the head may contain a genitive. Other than one
126 RAY JACKENDOFF

combination which will be useful in the next section, I leave precise statements of these
curious restrictions for future research.
The one combination of interest is this: just in case the matrix NP has an N'"
specifier, as in (5.54), the group noun is left with no surface specifier at all. This is a
little strange, since singular count nouns like number and quantity normally require
some specifier. One way to make the situation a little better is to claim that the
underlying form of (5.54) is John's [N»a rather large number] of objections, Fred's [N™«
substantial quantity] of wine, with the indefinite article as the specifier of the group
noun. Then the following local transformation would delete the indefinite article:
(5.57) Pseudopartitive a-Deletion
[n„,N"' - a - X] =>1-0-3
OBLIGATORY

Thus we can claim that the group nouns in (5.54) do have the specifier they require in
underlying form, but it is deleted.
This solution does not deal with the possibility of other specifiers of the group
noun cooccurring with a matrix genitive. I assume they are eliminated by a surface
constraint which is necessary anyway to eliminate such combinations as two genitives
in a row. Thus a is the only group noun specifier cooccurring in underlying structure
with matrix genitives, because it is deleted and hence the surface constraint is satisfied.

To sum up our discussion of pseudopartitives, we have argued that the N" specifier
contains an N'" position which alternates with the Q'" position there. It may be
occupied by group nouns which are followed by the specified grammatical formative of.
Furthermore, the group noun may contain its own specifiers, and there is a complicated
set of constraints coordinating the group-noun specifier and the matrix N'" specifier.

5.5. Numerals

We have so far neglected the syntax of numerals. They come in two varieties, which I
will call cardinals and seminumerals. Cardinals are words like three and seven, which
need not be preceded by an article; seminumerals are words like dozen and hundred,
which require an article before them.
The grammar of numerals is somewhere in between that of quantifiers and that of
group nouns. This section will argue that numerals are nouns which trigger certain local
transformations whose effect is to make the superficial behavior of numerals closer to
that of quantifiers.

5.5.1. Comparison to Quantifiers and Group Nouns

This subsection will demonstrate the similarities of numerals to group nouns and
quantifiers, by showing how numerals behave in paradigms discussed in previous
sections.
First, like quantifiers and pseudopartitive group nouns, numerals appear between
N"' specifiers and A'"s.
 5: NP SPECIFIERS

John’s
|three
(5.58) these old trees
[dozen
all

This suggests that they are in the N" specifier, either as quantifiers (5.59a) or as group
nouns (5.59b).

(5.59) a. N'" b. N'"

Numerals also occur in partitive constructions, with structure (5.60a) if they are
quantifiers and (5.60b) if they are nouns.

The Partitive Constraint applies to these constructions,

(5.61)
(three I ^ I some men
a dozen J [all men

as does the constraint on quantifiers and group-noun partitives that if they are definite,
a relative clause is required.
128 RAY JACKENDOFF

three
(5.62) a. *the of the trees
(dozen

I three
b. the of the trees that weren’t cut down
[dozen

Extraposition from NP can be applied to the partitive:

(5.63)
[Three j
weren’t cut down of the trees you planted.
[a dozen J

By contrast, in the simple construction (5.58) (as in the parallel cases with quantifiers
and group nouns), Extraposition from NP applies not to the noun being counted, as in
the partitive, but to a PP one level further down:

f Three 1
(5.64) a. * A ^ r weren’t cut down (of) trees you planted.

Three j
b. pieces were eaten of the leftover turkey.
A dozen J

Thus numerals parallel the behavior of quantifiers and group nouns. If they are
quantifiers, the simple use (5.59a) is straightforward, and in the partitive use (5.60a)
they govern the Partitive Projection Rule. If they are nouns, the partitive use (5.60b) is
straightforward, and the simple use (5.59b) is like a pseudopartitive (though lacking the
of). The next question we will address is which of these analyses is correct. So far the
analysis of numerals as quantifiers looks more promising, since the paradigm (5.59a)-
(5.60a) is just like that of many and few. However, the next subsection will argue that
they are nouns.

5.5.2. Numerals as Nouns

The argument that numerals are nouns is based on their specifier system. The
quantifiers many, few, and little can be preceded by degree words such as so, as, too,
and how, chapter 6 will explore this specifier system in considerable detail. Numerals,
of course, can never be preceded by degree words:

as I three
(5.65)
too [dozen
how

On the other hand, nouns such as group, bunch, and number are preceded by typical
nominal modifiers such as articles and adjectives; if numerals are nouns, they should
have such specifiers too.
The fact that the seminumerals dozen and hundred always require a nominal
specifier (most often a) is a hint that we should look for other evidence that numerals
are nouns. The clearest evidence comes from a construction little noted in the literature
 5: NP SPECIFIERS

(although it appears in Jespersen), in which even cardinals are preceded by adjectives

and by the indefinite singular article:

(5.66) a beautiful two weeks

a dusty four miles of road
another whole seventeen pages

The phrases are close syntactic parallels to group noun pseudopartitives like (5.67).

(5.67) a tremendous group of people

a useless couple of days
another whole bunch of bananas

A similar parallel obtains with true partitives:12

(5.68) a. the most beautiful two of those weeks

a dusty four of those many miles of road
another whole seventeen of SPE’s pages
b. a tremendous number of the senators
a useless couple of the remaining days
another whole bunch of those bananas

Since quantifiers cannot be preceded by adjectives, but nouns certainly can, it

appears most expedient to treat the numerals in these constructions as nouns, and
therefore to assign (5.66) and (5.68a) the following structures:

(5.69) a. N'"

weeks

121 have no explanation for why good examples of partitives like (5.68a) are so hard to find.
130 RAY JACKENDOFF

If numerals are indeed nouns, the next step in the analysis must be to account for
the differences between numerals and group nouns; the next subsection will deal with
this problem. But first it is worth mentioning two special cases, one and few.
We have repeatedly alluded to two different morphemes spelled one, a numeral
and a pronoun. The numeral can govern a partitive and cannot be pluralized; the
pronoun allows no N' o/-complement and has the plural ones. If the numeral one were
like other numerals in being a noun, we would expect, parallel to (5.66), the
ungrammatical *a beautiful one day. In order to prevent this, we will claim that the
numeral one, unlike two, three, etc., is a quantifier and hence takes no nominal
specifiers. The pronoun one, on the other hand, takes nominal specifiers but is not a
group noun, so it cannot occur in structures (5.69a,b) either.
Section 5.2.1 hinted at the distinction between few and a few. We can now spell
it out more clearly. A few, because of its indefinite article, can be identified as a
seminumeral, hence a noun. As confirmation, we note that it can be preceded by
adjectives, e.g. a miserable few people showed up. On the other hand, few is a
quantifier, as can be seen by its ability to take degree words {so few, as few, etc.) and
adverbs (e.g. We met miserably few people). Similarly, a little is a seminumeral and
little is a quantifier.13
We now turn to the adjustment rules which distinguish cardinals, seminumerals,
and group nouns from each other.

5.5.3. Three Adjustment Rules

First let us distinguish cardinals from seminumerals. Seminumerals are like group
nouns in that they are always preceded by a specifier, in the unmarked case the article
a. (5.66) and (5.68) showed that cardinals also can occur with articles if an adjective
intervenes. To assimilate cardinals to the normal case of group nouns and seminumer¬
als, we will assume that cardinals too always have an article in underlying structure,
but that it is deleted by the following lexically governed local transformation:

(5.70) Cardinal a-Deletion

a
-r n
_+Card
2

OBLIGATORY

According to this analysis, the underlying structure of two weeks is a two weeks,
parallel to a dozen weeks and a bunch of weeks (we deal with of in a moment); (5.70)
deletes a before two. A beautiful two weeks, parallel to a beautiful dozen weeks and a
beautiful bunch of weeks, will not be affected by (5.70) because of the adjective
intervening between the article and the numeral. Similarly, in That two weeks we spent
in Bermuda was awful, the singularity of the article shows it is a specifier of the

13 One exception to this clear distinction is a very few, which has both a nominal and a quantifier
specifier. 1 have no explanation.
 5: NP SPECIFIERS

numeral; since the article is not a, it does not delete. (5.70) thus accounts for the
difference between cardinals and seminumerals.
Next we must distinguish numerals from group nouns. As pointed out in section
5.5.1, they differ in that group nouns but not numerals are followed by of when they
occupy the N" specifier. Under the assumption that the pseudopartitive of is an
underlying specified grammatical formative (a point that will be somewhat justified in
section 6.2), the simplest way to account for this difference is to posit another local
transformation, governed by all numerals:14

(5.71) Numeral of-Deletion

N
[N»X - _0/_Y-N'-Z]z>1-2-</>-4-5-6
,+ Num .
OBLIGATORY

These two rules yield the following derivations:

six
a bunch hundred
(5.72) of weeks a +Num of weeks
of weeks . +Num
_+Card.
apply 1 1 I
(5.71) n.a. a hundred weeks a six weeks
apply 1 1 i
(5.70) n.a. n.a. six weeks

There are a few lexical doublets that show how superficial the presence of the
feature [+Num] is. In my dialect, the word couple alternates freely between group
noun and seminumeral status, since I can say either a couple of trees or a couple
trees.15 In the majority dialect, dozen is a seminumeral; but there are dialects which
allow a dozen of eggs, where dozen is a group noun. Though hundred, thousand, and
million are seminumerals, hundreds, thousands, and millions are group nouns; thus we
have the contrast of six million men (numeral) vs. the somewhat fusty six millions of
men (group noun). In the present analysis, these differences follow solely from the
presence or absence of the exception feature [ + Num],
These two local transformations, Cardinal o-Deletion and Numeral o/-Deletion, are
sufficient to account for the differences between numerals and group nouns observed in

14 Under any other assumptions about how the pseudopartitive of is introduced, the exception feature
[ + Num] will play a similar role in preventing of from occurring in the surface. I assume there is a separate
idiosyncratic 0/-Deletion rule applying in the partitives all of the men, both of the men to form all the men,
both the men.
15 There is also the intermediate reduction a coupla trees, occurring with other group nouns as well, e.g.
a buncha trees. This reduction depends on of being a daughter of N". Notice that in the following example,
the reduction is impossible:

(i) We bought a couple (of) pictures of Fred, and a {7*coup/a °^}

The reason is that the final NP in this S is the result of N'-Gapping and hence has the structure [N4N»a couple
[n,PRO of Bill]]). The presence of PRO inhibits the reduction of of.
132 RAY JACKENDOFF

section 5.5.1. To complete our account of the local rules applying to numerals, we
observe that Pseudopartitive a-Deletion, developed in section 5.4.2, applies also to
seminumerals, since the derivations of John’s bunch of cows and John’s dozen cows
are precisely parallel in this respect:
(5.73) N'"

Art'" N" N

a N' cows

{bunch
dozen
(a—Deletion)
I bunch of]
> John’s
(o/-Deletion)
1!dozen ofj
bunch of I
cows

> John’s cows

dozen J

However, there is a difference between group nouns and numerals. Section 5.4.2
showed that when a group noun in a pseudopartitive is preceded by an article, the article
must be in the specifier of the group noun, not in the matrix NP. For example, that
group of men has singular that in the specifier of group; but *those group of men, with
plural those in the specifier of men, is ungrammatical. Numerals apparently lack this
constraint, since we have both that three weeks, with that in the specifier of three, and
those three weeks, with those in the specifier of weeks.16

10 As further justification for these structures, notice the distinction between every three weeks and all
three weeks. Every requires a singular head, and thus cannot be in the specifier of weeks. Moreover, the
meaning of every three weeks is not ‘every one of three weeks’ but something like ‘every group of three
weeks’: every quantifies instances of three. Thus the evidence points to every being in the specifier of three,
like that in (5.74a). By contrast, all requires a plural head, and in all three weeks it quantifies instances of
weeks. This means that it must be in the specifier of weeks, and hence that all three weeks has structure
(5.74b).
 5: NP SPECIFIERS

N N

three three
Now since seminumerals normally appear at the surface with their own determiners,
we would expect, parallel to (5.74b), *those a dozen weeks. Instead, we find those
dozen weeks. Apparently Pseudopartitive a-Deletion must be extended so as to delete a
in this configuration as well.
In the revision of Pseudopartitive a-Deletion, the feature system collapses Nand
Art'" in the first term of the rule:
(5.75) Pseudopartitive ^-Deletion, revised
+ Subj
-Obj -a - X =>1-0-3
_ + Det _ -
LN'"

OBLIGATORY

Thus the underlying structure of those dozen weeks is those a dozen of weeks, which
undergoes Pseudopartitive a-Deletion and Numeral o/-Deletion to reach the surface
form.
This completes our discussion of numerals. We have shown that, despite their
similarities to quantifiers, their specifier structure proves them to be nouns very much
like group nouns. Most differences between them and group nouns, and between
cardinals and seminumerals, are accounted for by simple local transformations.

5.6, Summary

Here is a list of NP specifiers classified in terms of the syntactic and semantic

properties discussed in this chapter.
134 RAY JACKENDOFF

Articles
those which
Semantically demonstratives: a, the, this, that, these, what, we.
+ Part. ’ .+ Part _
yOUpiur
Semantically quantifiers (all [ + Partitive]): any, all, no, some each, every, either,
neither, both

Quantifiers (all [ + Partitive]): many, much, several, few (negative), little (negative), one
(nonpluralizable)

Nouns
Group nouns: group, gallon, bunch, number, lot, score, hundreds, thousands, millions,
couple (some dialects), dozen (some dialects)
Seminumerals ([ + Num, -Card]): hundred, thousand, million, few (nonnegative), little
(nonnegative), couple (some dialects), dozen (most dialects)
Cardinals ([ + Num, + Card]): two, three, etc.
Pronouns: we, you, one (pluralizable), PRO

The following two phrase structure rules are responsible for the NP specifier
system:

+ Subj
(5.76) N'" —M -Obj ) - N"
_+Det
+ Subj
(5.77) N" ( -Obj ) _ (A'")* - N' - . . .
Det _

In (5.76), the feature matrix can be expanded as N"', which becomes a genitive, or as
Art'", which is a demonstrative or N"' quantifier. In (5.77), the feature matrix can be
expanded as N'", which is a group noun or numeral in a pseudopartitive construction,
or as Q"', an N" quantifier.

(5.78) N'"

Certain configurations generated by applying these phrase structure rules freely are
filtered out by constraints we have mentioned. The combination of an N'" quantifier
and an N" quantifier is ruled out by the Specifier Constraint. The combination of an
Art'" in N'" and a group noun in N" is ruled out by a constraint justified in section 5.4.2;
 5: NP SPECIFIERS

but an Art'" in N'" followed by a numeral in N" is not ruled out, as we showed in
section 5.5.3. Further constraints are necessary and have been left for future research.
Partitive constructions take one of these three forms:

(5.79) a. N'" b. N'"

PRO PRO

The Partitive Constraint restricts the possible specifiers of the complement N'" to those
which contain a genitive or demonstrative. In (5.79a,b), PRO is interpreted by the
Partitive Projection Rule, which gives PRO the reading UNIT or AMOUNT.
We have developed four local transformations. The first, Substantivization,
changes the form of various articles and possessive pronouns when they are directly
followed by PRO. For example, no becomes none, every becomes every one, and my
becomes mine. This rule applies regardless of the interpretation of PRO; thus the
changes can be observed not only in partitive constructions such as (5.79a), but also in
instances of N'-Anaphora, N'-Gapping, and postposed genitives.
The second local transformation. Pseudopartitive a-Deletion, deals with situations
in which a group noun or seminumeral occupies an N" specifier, and the N'" specifier is
136 RAY JACKEN DOFF

nonempty. This rule deletes the indefinite article of the group noun or seminumeral,
giving derivations such as these:

(5.80) John’s [a rather large number] of objections

=> John’s rather large number of objections
John's [a dozen] trees => John’s dozen trees

hundred days

The other two local transformations deal with the differences between group
nouns, seminumerals, and cardinals. We assume all three of these have underlying
articles, and that all are followed by of when in the N" specifier (pseudopartitive
position). Numeral o/'-Deletion eliminates the of after all numerals, giving contrasts
like a pair of men (group noun) vs. a few men (seminumeral). Cardinal a-Deletion
deletes a immediately before cardinals, giving contrasts like a couple weeks (seminu¬
meral) vs. two weeks (cardinal); the rule does not apply if there is an intervening
adjective, so that the article appears both with seminumerals (a beautiful few weeks)
and cardinals {an ugly two weeks).
The relevance of all this to the general theory of phrase structure presented here is
this: first, the NP specifier system has long been an area subject to wild unsystematic
proposals about phrase structure. We have shown that it can be analyzed quite
consistently within the highly restrictive theory of phrase structure presented in chapter
3, and that its irregularities are accounted for either by local transformations or by
nonsyntactic means.
Second, we have seen that the specifier system is much less consistent in its
semantics than the complement system. Furthermore, a reader of this chapter who is
concerned with semantics will have seen how little is really known about the semantics
of specifiers, despite nearly a century of study of the formal logic of quantifiers and
demonstratives. It is hoped that our systematic survey of the syntactic possibilities of
specifiers will serve as a source of irritation to those in linguistic semantics who think
they know something about quantification.
 6: Specifiers of X

6.1. Introduction

This chapter presents an analysis of the specifiers of degree, developing two major
cross-categorial phrase structure rules for specifier systems which introduce measure
phrases and quantifier phrases in X" and degree phrases in X'". We will show that the
minor lexical categories Q and Deg enter into cross-category generalizations with major
categories, justifying the use of Q'" and Deg'" predicted by the Uniform Three-Level
Hypothesis of phrase structure.
We draw heavily on Bresnan’s (1973) lengthy and insightful discussion of specifier
systems, which makes copious use of cross-category generalization and the X Conven¬
tion. We will propose a number of improvements on Bresnan’s analysis which in turn
confirm the general theory of phrase structure adopted here.
The general procedure behind the analysis is the same one we followed in chapter
4. A particular construction found in one category is checked for its occurrence in other
categories. An analysis is then developed consistent with the facts in all categories and
with the phrase structure schema. In particular, several crucial hypotheses are verified
here by reference to PP and its specifiers, to which Bresnan does not appeal, and which
turn out to be a surprisingly rich source of evidence.
We will build up the specifier system piecemeal, starting with measure phrases,
then going on to quantifier phrases. There follows a detailed discussion of the
occurrence of degree phrases and how recursion takes place in their specifiers.

6.2. Measure Phrases

Both adjective phrases and prepositional phrases can be modified by a prehead noun
phrase often called a measure phrase.
(6.1) a. two feet long
seven miles wide
b. five miles down the road
three inches along the seam
By means of conjunction and various preposing transformations, it can be shown that
the phrases in (6.1) are constituents.
138 RAY JACKENDOFF

(6.2) a. The pool is sixty feet long and thirty feet wide.
b. Six feet tall you’ll never be!
c. His house is to the left and six miles down the road.
d. Four hundred yards up the street they encountered a panther.

Apparent counterparts of these construction in NPs are such phrases as an inch of

rope and five miles of road, especially since the parallel between, for example, two feet
long and two feet of length is semantically appealing. As mentioned in section 5.4,
Selkirk (1977) delineates the cases in which these measure phrases are in fact
specifiers, followed by of: those in which no article follows the of. Thus two feet of
rope has the desired structure, with two feet as a measure phrase and rope as the head;
but two feet of the rope has the partitive structure in which of the rope is in the N'
complement.
The of in the cases we are interested in here is therefore not a preposition but a
specified grammatical formative like Poss, introduced either by the base or a local
transformation. To decide which source is correct, observe that there is another NP
construction, not often discussed, which fills the syntactic paradigm (6.1) even better
than an inch of rope does.

(6.3) a. This mixture is two parts alcohol and three parts water.
b. The book was (one) half garbage.

Expressions like two parts alcohol could be little but NPs, since they are made up of a
numeral and two nouns; but they may be used only in predicate positions, e.g. after be,
seem, stay, and so forth. Thus there are two uses of measure phrases in the specifier of
NP, with different interpretations. If the of in the use first mentioned is base-generated,
we can provide a deep structure cue for the difference in interpretation.
Where should the measure phrase be attached within NP? It precedes adjectives,
which are in N": two gallons of ordinary water, three parts clear alcohol, not *ordinary
two gallons of water, *clear three parts alcohol. Thus it is either in N" or N"\ If we
attach it to N", we automatically account for the fact that measure phrases do not
receive the Poss invariably attached to N'" in the N'" specifier. Thus we arrive at the
structures (6.4a,b).
 6: SPECIFIERS OF X

Given (6.4) as the structure for measure phrases in NP, we are constrained by the
X Convention to propose structures (6.5a,b) for (6.1a,b). We then have the partial
phrase structure rules (6.6) and a grammatical relation (6.7).

(6.5) a. A'"

A"

N'" A'

two feet A

long

(6.6) a. N" —» (N'" - (of)) - N'

b. A" —» (N'") - A'
c. P" —» (N"') - P'

(6.7) A measure phrase is the italicized constituent in the configuration

[x* N'" . . . X' . . .].

The projection rules for measure phrase will require that it consist of a quantified count
noun, with further selectional restrictions on the noun depending on the nature of the
X' to which it is a sister. The difference between the two interpretations of measure
phrase in NP will depend on the presence or absence of the specified grammatical
formative of.
The major category which is still missing in the paradigm above is V. There is
nothing preceding the verb which looks like the constructions in (6.1)—(6.4), but
measure phrases do appear postverbally.

(6.8) a. Jill ran around the track three times,

b. Fran stayed in Africa three years.1

Note that the measure phrase appears after a PP in both of these examples, a new
position for NP. In fact the measure phrase must appear after all strictly subcategorized
phrases (barring Heavy NP Shift), as shown by (6.9).

1 The measure phrase in (6.8b) is intuitively quite close to the PP for three years, and it is quite natural
to assume that it is the result of a For-Deletion. On the other hand, since three times does not appear to be
the result of a reduction, there is a base position available for three years, and we might as well make use of
it, sparing ourselves a rather idiosyncratic transformation.
140 RAY JACKENDOFF

(6.9) a. Charlie put the book on the table three times.

b. ?Charlie put the book three times on the table.
c. *Charlie put three times the book on the table.
d. Charlie told Edna the story three times.
e. *Charlie told Edna three times the story.
f. *Charlie told three times Edna the story.

The position of the measure phrase is thus the same as that of postverbal manner
adverb, so it must be a daughter of V", generated by the partial phrase structure rule
(6.11).

(6.10)

N'"

Charlie 1ST"

three times

(6.11) V" —» V' - (N'")

The posthead position of the measure phrase appears also as a second option in
PP, as in down the road three miles, up the river a way. By the X Convention, this
must be attached to P", parallel to V", and there must be a partial phrase structure rule
(6.12).

(6.12) P" —» P' - (N "')

There is a curious partial generalization here. All major categories permit a

measure phrase as a daughter of X", but it is a left sister of N' and A', a right sister of
V', and either a left or right sister of P'. Section 4.5 discussed the problems involved in
expressing this partial generalization in terms of conventionally stated phrase structure
rules, and 1 have nothing to add here.
In order to generalize the grammatical relation measure phrase to all categories we
must apparently resort to a definition of the Aspects sort, which involves not order but
only domination.
 6: SPECIFIERS OF X

(6.13) A measure phrase is an N'" immediately dominated by X". (In the Aspects
notation, [N"\ X"].)

This section has shown, then, that NPs with the function of measure phrases occur
in the X" of all major lexical categories, providing another significant cross-category
generalization. We next show that these NPs alternate with QPs in the same position.

6.3. Quantifier Phrases in X"

Section 5.2 claimed that the quantifiers many, few, much, little, and several are
daughters of N". They differ from the articles all, each, every, some, etc., which are
daughters of N'" and cannot be preceded by genitives or other articles. The operative
phrase structure rule was (6.14).

(6.14) N" -* (Q'") - (A'")* - N' - . . .

Since the N" quantifiers do not cooccur with measure phrases, we can immediately
generalize (6.14) with (6.6a) to form (6.15).

+ Subj
-Obj
(6.15) N"->( - ((of))) - (A'")* - N' - . . .
— Det
( + Comp)

In (6.15), the realization of the constituent broken down into features has two
possibilities. If it is [+Comp], the features spell out N'" and the of is possible. If it is
[-Comp], the features spell out Q'" and the of cannot occur.
This generalization of measure phrase and Q'" carries over to the other major
categories. In sentences, we find QPs postverbally, as in You talk too much. Since
these QPs are not strictly subcategorized, they belong in V", so we can generalize (6.11)
to (6.16).

+ Subj
(6.16) V" V'-( -Obj )
Det

In PPs there are quantifiers far and long (as in far down the road and long after the
accident) which alternate with measure phrases, and which, as we will see, take the
same modifiers as many and much. This justifies generalizing (6.6c) to (6.17).

+ Subj
(6.17) P" ( -Obj ) - P'
Det

The QP, like the measure phrase, may follow the P' as well, as in down the road too
far. This use is subject to the restriction that the QP must contain a specifier of its own,
142 RAY JACKENDOFF

since *down the road far is out. I have no explanation for this restriction; for present
purposes I will conveniently ignore it, generalizing (6.12) to (6.18).

+ Subj
(6.18) P" P' - ( -Obj )
Det

Having dealt with the generalization of measure phrase and QP in N", V", and P",
we turn to A". The situation in A" is more complex than the other three cases, and
presents a difficult choice. Bresnan (1973) chooses one possibility, which, as we will
see, has repercussions for her entire analysis of the degree system. We will show that a
different choice provides a more satisfactory analysis throughout. The rest of this
section will outline the problem, and its thread will be repeatedly picked up in the rest
of the chapter.
The problem is this: on the whole adjectives do not take simple quantifiers; for
example, *much long and *little long are ungrammatical. However, in the comparatives
more beautiful and less beautiful, more and less appear to be the comparative forms of
the quantifiers much and little. The question is how to account for this curious
alternation.
Bresnan describes it by claiming that much deletes obligatorily before adjectives,
leaving only its modifiers. For example, underlying as much beautiful becomes as
beautiful', underlying much beautiful becomes beautiful. In more beautiful, much
becomes more by affixation of the comparative -er\ since it is no longer the
phonological much, it does not delete.
One immediate difficulty with Bresnan’s account is the nonexistence of *little
beautiful, since a deletion of the quantifier little here could hardly result in anything
with the expected meaning. Yet since less beautiful is to be derived from underlying -er
little beautiful, some way must be found to rule out the noncomparative form.
There seem to be two ways out of this difficulty. First, Bresnan could be right, and
all adjectives could take quantifiers, but various deletions and semantic constraints
could prevent all forms but the proper comparative forms. Under this theory, (6.6b)
would immediately generalize to (6.19), parallel to (6.15) and (6.17).

+ Subj
(6.19) A ( -Obj A'
.-Det

As an alternative to Bresnan’s theory, I propose that usually adjectives do not take

quantifiers, and that the exceptions more and less are not quantifiers, but arise from a
different source, to be discussed in section 6.4.3. Under this theory, (6.6b) does not at
first glance appear to generalize to (6.19).
However, Bresnan points out some evidence that there is a generalization anyway
(note 4, p. 278): there are at least two adjectives which appear with surface quantifiers.
 6: SPECIFIERS OF X

much
(6.20) a. You and I are little 'different from Bill.
far

much
b. You and I are alike.
little

In Bresnan’s theory, these adjectives must be lexically marked to take optional rather
than obligatory A/wc/i-Deletion. However, this fails to explain why little and far can
occur with these adjectives and not with others.
Under the alternative view I have proposed, there is another account: these
adjectives are exceptions to the generalization that adjectives do not take quantifiers.
Such exceptionality is not too surprising, in view of the fact that adjectives that permit
measure phrases must also be lexically marked and are almost equally rare (high, wide,
long, tall, old, thick, broad, and perhaps a few others). In order for these exceptional
adjectives with quantifiers to exist, the phrase structure rules must allow a quantifier,
and thus the generalization of measure phrase and QP in (6.19) is motivated.
This view of QP in AP eliminates all need for Bresnan's Mwc/i-Deletion transfor¬
mation, along with the difficulties it poses for the occurrence of little and far. Rather,
the few cases where there is surface evidence for a QP are taken as exceptions in
underlying structure, provided for by general phrase structure rules. One would hope
that the exceptionality ultimately proves to be semantically motivated, leaving the
syntax and lexicon maximally simple.
Whether we take Bresnan’s view or the present one on the presence of quantifiers
in A", this section has shown the need for phrase structure rules (6.15), (6.16), (6.17),
(6.18), and (6.19). We thus see that the base rules for measure phrases and quantifiers
are fully general among the major lexical categories, aside from the problem of order in
V" and P". The parallelism of the X" specifiers justifies us in considering all these rules
to be special cases of a single phrase structure rule. Furthermore, the recurring
parallelism of NP and QP in the X" specifier is evidence for the feature analysis that
relates the lexical categories N and Q.

6.4. Degree Phrases

There are a number of words we will classify as Deg which, among other things,
precede the quantifiers of the preceding section. (6.21a) illustrates them in QP within
N"\ (6.21b) in QP within P"\ and (6.21c) in QP within V"'.
144 RAY JACKENDOFF

how
this
that I far along the road I
b.<
so [ long after the accidentj
too
as

(this 1
that I much I
so [little J
c. He talks too

as {iT.tlei as Bi"
\

How does he talk?

This section will develop the phrase structure rule generating them, showing that it
too is subject to an interesting cross-category generalization.

6.4.1. Attachment of Deg'" to Q"'

There is little strong evidence for where the degree words in (6.21) are attached to Q"\
However, the X Convention suggests the hypothesis that they are Q'" specifiers. This
accounts for the parallelism in the use of this and that, which as specifiers in N'" are of
the category Art but with similar demonstrative meaning. Under this theory, the N'"
specifier will then contain a [ + Subj, — Obj, -Comp, +Det]'" and the Q"' specifier will
contain a [-Subj, -Obj, -Comp, +Det]"'. The lexical items this and that will either
occur in both categories, differing by the feature Subj, or, preferably, they will simply
be unmarked for this feature. This parallelism between Deg and Art is the reason for
assigning Deg the feature analysis proposed in chapter 3. A further bit of evidence for
the parallelism arises in section 6.4.4.
Bresnan suggests a different attachment of degree words, to what in our terminol¬
ogy would be Q". Section 6.5 will call into question the validity of her argument for this
structure. Anticipating the refutation of her argument, I will adopt the solution
suggested by the X Convention and generate degree phrases in QP with the partial
phrase structure rule (6.22). Its relationship to the partial phrase structure rule (6.23)
for N'" is obvious.

(6.22) Q"' -> (Deg'") - Q

(6.23) N'" —► (Art'") - N'
 6: SPECIFIERS OF X

Example structures are given in (6.24).

N'"

the road

many far

much
146 RAY JACKENDOFF

There are three questions that must be dealt with in implementing this structure.
The most crucial is how the degree phrase is applied to A'" and Adv'". The next
subsection deals with this. There are also consequences for the treatment of the words
more, less, and enough, which we turn to in sections 6.4.3 and 6.4.4.

6.4.2. The Degree System of Adjectives and Adverbs

Section 6.3 claimed that adjectives (with a few rare exceptions) do not occur with
quantifiers. Therefore the degree words as, so, too, this, that, and how in A'" and
Adv'", as in (6.25), must be generated by a phrase structure rule that introduces degree
phrases directly in the specifier of A'" and Adv'".
as
so

that
how
The X Convention suggests that the degree phrase be generated in the A'" and
Adv'" specifier, as in (6.26), so that the rules generalize with (6.22) and to a lesser
extent with (6.23).
(6.26) a. A'" -> (Deg'") - A"
b. Adv'" (Deg'") - Adv"
This analysis thus claims that adjectives appear with the same degree words as much
because of the similarity of the phrase structure rules (6.22) and (6.26).
Bresnan expresses this generalization a different way: by quantifying adjectives
with a much which is modified by a Deg'". This much is deleted transformationally,
leaving only the degree phrase at the surface. Section 6.3 pointed out the difficulty in
this view: it requires all adjectives to allow a QP, thus leaving unexplained the
distribution of the quantifier little. Thus Bresnan’s means of expressing the generaliza¬
tion is questionable, whereas in the present analysis, such a distribution is an automatic
consequence of the proposed base structure
A complication Bresnan points out turns out to follow immediately from the
present theory. She observes (note 4, p. 278) that those adjectives such as alike and
different that occur at the surface with much have two possible forms with degree
words, for instance as different and as much different. She uses this as evidence that
Much-Deletion is optional with these adjectives. The present theory permits a more
direct account, using phrase structure rules alone: the two base forms in (6.27).r

2 A third possibility would attach Deg'" to A"' and Q"' to A". I have no arguments against such a
structure, except for the possibility that combinations of quantifiers and degree words as independent
specifiers may be excluded by some generalization of the Specifier Constraint of section 5.2. I know of no
cases where both are needed; but I understand the semantics of degree phrases and quantifiers sufficiently
poorly that I am loath to make a strong commitment ds to how the unnecessary combinations are to be
eliminated.
 6: SPECIFIERS OF X

(6.27) a. A'" b. A

much
' little ’
v
far -/

Hence the alternation follows not from a transformational exception, but from the
unusual subcategorization possibilities of these adjectives. This provides further
evidence that Deg'" is generated in A"' and Adv'" as well as in Q'", yielding a further
cross-category generalization in the specifier system.

6.4.3. Comparative Specifiers

Section 6.3 pointed out that the hypothesis that most adjectives do not take quantifiers
has as a consequence that more in more beautiful cannot be the comparative of the
quantifier much. Let us suppose instead that it is a Deg in the A'" specifier, parallel to
as, too, so, and how in (6.27a).
If more is a Deg, though, we expect it also to occur in the Deg"' within a Q"\
giving the incorrect forms *more many, *more much, etc. We correct this problem by
providing the idiosyncratic spelling rules (6.28) to yield the existing forms.3

3 Note that these rules are not a disadvantage with respect to Bresnan’s theory, since she needs similar
rules to attach her comparative morpheme -er to quantifiers:
-er - much => more
-er - many more
etc.
148 RAY JACKENDOFF

(6.28) more - much more

more - many more
more - little => less
more - few fewer
more - far => farther

The theory that more is a Deg thus produces structures such as these

(6.29) a. A'"

beautiful

b N'"

many
v-

more
 6: SPECIFIERS OF X

P"

far
-

farther

As a consequence of the theory that more in more beautiful is not the comparative
of much, less in less beautiful cannot be the comparative of the quantifier little either.
Rather, less must also be able to be a Deg, and less beautiful is then identical in
structure to (6.29a).
We further predict that less should be applicable to quantifiers as well as to
adjectives, substituting for more in constructions like (6.29b,c). One such possibility
exists—less far, as in less far down the road; but there are no *less much, *less many,
*less little, or *less few. Thus, while Bresnan’s theory must find a way to generate less
far and eliminate *little beautiful, the present theory must find a way to rule out the
other four less-Q combinations. The four bad forms, if they existed, would presumably
be synonymous with the existing forms more little > less, more few > fewer, more
much > more, and more many > more, respectively, which somehow seem “less
negative”. This fact may be useful someday in giving an interesting account for what
must remain for now an unprincipled filter.
Now consider the rule that forms comparative adjectives such as taller, presum¬
ably from the source more tall. As is well known, this rule applies only to a particular
class of lexically marked adjectives, mostly monosyllabic. In the present theory it
applies to this structure:
150 RAY JACKENDOFF

(6.30) A

tall

taller

Observe that this configuration is precisely parallel to that in the QPs of (6.29b,c). Thus
the rule forming comparative adjectives generalizes via the X Convention with the rule
forming comparative quantifiers.
In Bresnan’s system, there is of course also a rule that turns more - A into A-er.
However, in her theory the more - A rule attaches the quantifier more to the head and
thus has little to do with the rule forming comparative quantifiers, which attaches the
Deg -er to the head (see footnote 3, p. 277). Thus the present theory expresses a further
generalization missed by Bresnan’s analysis, justifying the assignment of more to the
category Deg, and further justifying the absence of QP in most APs.
As a final note, observe that all the arguments given here for the comparative Deg
more apply equally to the superlative Deg most, which also occurs in the specifiers of
A'", Adv'", and Q'". Like more, most undergoes spelling rules that attach it to
quantifiers and lexically designated adjectives; the environment is exactly the same as
with more, so the rules generalize. The word least, like less, occurs in the specifier of
A'" and Adv'" and is therefore also a Deg. Since least does not appear comfortably with
any quantifiers at all (even *least far is unacceptable), we can assume either a
generalization of the filter for less or some sort of semantic constraint. Thus the
superlatives behave almost precisely like the comparatives.

6.4.4. Enough

We now turn to the most syntactically exceptional word in the specifier system.
enough. It provides a great deal of evidence for the analysis of preceding sections.
 6: SPECIFIERS OF X

As Bresnan points out, enough is peculiar in that it obligatorily follows adjectives,

adverbs, and quantifiers that it modifies, and it optionally follows nouns. This behavior
is unlike that of all other modifiers it alternates with, which invariably precede their
heads.

r ~\

so
(6.31) a. beautiful enough / ' as ' beautiful
too

so
b. quickly enough / 'jas ' quickly
too

so
I far
as
C' {fewJen0Ugh/ 1 Ifew
too

. J enough pudding ! . J much I ...

d [pudding enough j [little } puddlng

Bresnan proposes a local transformation to account for this, moving enough around its
head. In present terms, this rule can be stated as (6.32), where X° denotes a lexical
category.

(6.32) Enough-Shift
enough - X° => <t> - 2 + I
OBLIGATORY if X° = A, Adv, Q
OPTIONAL if X° = N

I much
Bresnan treats enough as a Q, on the strength of similarities like
[enough

pudding and )Ugh J men• Small enough is taken to have the underlying structure

enough small, undergoing Enough-Shift; the parallel *much small becomes small by
obligatory Much-Deletion.
Consider, however, the alternative view that enough is a Deg. This would explain
its cooccurrence with quantifiers in expressions like far enough down the road, little
enough pudding, and few enough men, which would then have the following struc¬
tures.
152 RAY JACKENDOFF

(6.33) a. P'" b. N'"

P" N"

enough Q'

Q Q

far little]
few j

Since Bresnan points out (note 3, p. 277) that her analysis is unable to generate these
forms, we have already found a significant advantage to our analysis.
An apparent difficulty for our view that enough is a Deg is that it predicts the
existence of the forms *many enough men and *much enough pudding, parallel to
(6.33b). To correct this discrepancy, we introduce a rule to delete many and much after
enough ,4 changing these incorrect forms to enough men and enough pudding. This rule
is much more restricted than Bresnan’s very general rule of Much-Deletion, since it
localizes the irregularity in a particular combination of lexical items.

(6.34) Many-Much Deletion

enough - } => I - <t>

OBLIGATORY

Thus enough men and enough pudding have the structure (6.35).

4 This assumes the rule precedes Enough-Shitl. Alternatively, it could delete many and much before
enough, if it were ordered after £>iowg/!-Shift. The choice does not seem crucial.
 6: SPECIFIERS OF X

(6.35) N'

N"

| many] ^
[ muchj

If the Q in (6.35) is changed to few or little, we get the forms few enough men and little
enough pudding by Enough-Shift. If the Deg is changed to as, we get as many men and
as much pudding, since neither Many-Much Deletion nor Enough-Shift applies.
Bresnan’s arguments that enough is a Q are based on a large number of parallels
between it and more, which, as we mentioned, she takes also to be a Q, the
comparative of much and many. However, our analysis also accounts for these
parallels. There are two relevant cases. In the specifiers of adjectives and adverbs,
enough and more are both Deg'" attached to X'", accounting for one set of parallelisms.
In the specifiers of nouns, prepositions, and verbs, however, surface enough and more

are dominated by Q' and are derived from underlying enough

(much I and more
many J
{much ]
many J ’ resPechvely> by the local transformations we have proposed. Hence the

parallelism of enough and more is a consequence of our analysis as well as Bresnan’s.

In addition to accounting for the combinations little enough, few enough, and far
enough, the theory that enough is a Deg has other advantages. A minor dividend
concerns the complement it binds (as in enough pudding to sink a ship). In note 11,
p. 289, Bresnan claims that although enough, like the degree words so, as, and too,
governs a complement clause, this is not an argument that it is a Deg. As justification,
she points out that sufficiently, an adverb, also governs a complement (e.g. It’s
154 RAY JACKEN DOFF

sufficiently long for us to use it); so governing a degree clause is not a sufficient
condition for something to be a Deg. But in fact, there are no other quantifiers that
govern complements, so enough would be in any event an unusual quantifier, though
not an atypical Deg. In our analysis, then, the fact that enough governs a complement
is not unexpected.
The present analysis also has a much more striking advantage. If enough is a Deg,
it can be made to function also as an Art by eliminating the feature [ — Subj] from its
lexical representation. Under this assumption, it would resemble the bicategorial this
and that. It could then enter into two possible structures within NP, (6.35) and (6.36).

(6.36) N'"

pudding

Now suppose that Enough-Shift, like other transformations, is subject to the condition
of the strict cycle, i.e. it can apply only on the lowest possible X'" domain in which
enough appears. Then it will be unable to move enough around pudding in (6.35),
because of the Q'" dominating enough. On the other hand, in (6.36) the rule can apply
to yield pudding enough. Hence the two possible NP structures, stemming from the
categorial ambiguity of enough, interact with a well-motivated constraint on transfor¬
mations to produce the variation between enough pudding and pudding enough. Thus
Enough-Shift can be made obligatory in all contexts, an improvement on Bresnan’s rule
(6.32), which singles out NP for special treatment.
Now consider what happens in A'". Under the present analysis of A'", tall enough
has the underlying structure (6.37).

(6.37) A'"

tall
 6: SPECIFIERS OF X

In (6.37), enough obligatorily permutes around tall, just as in pudding enough and far
enough.
But observe how enough behaves with the exceptional adjectives alike and
different. Superficially it looks as if enough permutes around them optionally rather
than obligatorily, since we have both alike enough and enough alike. But there is a
more interesting account: since alike and different, unlike most adjectives, allow a Q"\
they can occur in two distinct underlying forms with enough:

(6.38) a. A''' b. A"

A"

enough A'

alike I | alike j
different J [different/

far
• little ’
much

In (6.38a), enough permutes around the head to give alike enough and different enough,
parallel to (6.37). In (6.38b), enough permutes around the Qs far and little to give little
j
alike f alike 1
enough [different }' fQr enoug^ [different J muc^ *n (6.38b), Many-Much
Deletion takes place, and because of the strict cycle, enough does not permute around
the adjective; the forms derived are thus enough alike and enough different. This is just
like what happens in the nominal paradigm (6.35)—(6.36). What at the surface looks like
an exceptional optionality in Enough-Shift with these adjectives turns out to be a
consequence of the fact that they allow a QP in their specifier.
Thus the analysis of enough as a Deg confirms many details of the specifier system
argued for in previous sections. It makes possible the use of enough with the
quantifiers little, few, and far4, it makes possible the generalization that degree words,
156 RAY JACKENDOFF

but not quantifiers, govern complements; and, most striking, it makes the apparent
optionality of Enough-Shift in certain cases emerge as an automatic result of the base
rules and the principle of the strict cycle. Furthermore, the exceptionality of alike and
different, rather than being spread out over two transformations, Mwc/i-Deletion and
Enough-Shift, is localized in a single phrase structure difference: unlike other adjec¬
tives, they allow a QP.

6.4.5. Summary of Degree Words

To recapitulate the argument of this section, we have claimed that the category Deg
includes the words as, so, too, how, more, less, most, and least, plus the bicategorials
this, that, and enough, which function also as articles.
The constituent Deg'" is generated as an X'" specifier in A'", Adv'", and Q'",
partially generalizing with the Art'" position in the N'" specifier.

/ >
as many some
1 so
> <
the V
how which f
\ this ,this .
The relevant phrase structure rules are these:

(6.26) a. A"' —» (Deg'") - A"

b. Adv'" —*■ (Deg'") - Adv"
(6.22) Q'"-► (Deg'") - Q"
(6.23) N'"-> (Art'") - N"
 6: SPECIFIERS OF X

Using feature notation, these can be collapsed as (6.40).

in
( + Subj)
+ Subj \
-Obj
(6.40) ,+ Comp/ ) - X''
-Comp
-Obj
_ +Det
-Det

The primary argument for the feature relationship of Deg and Art and for the X'"
attachment of Deg'" has been the categorial ambiguity of this, that, and enough. This is
admittedly not overwhelming evidence, but in the absence of any other basis for the
decisions, we have followed the preferences dictated by the X Convention.
The irregularities of the degree system have been accounted for with various local
rules. More and most trigger spelling rules which attach them to quantifiers and certain
adjectives. Combinations of less and least with quantifiers (other than the combination
less far) are ungrammatical and must be ruled out by some as yet unprincipled filter.
Enough triggers two local rules, £>zowg/i-Shift and Many-Much Deletion.
Having determined the occurrence of Deg'", we will now explore its internal
structure.

6.5. Recursion in Degree Phrases

Quantifier phrases with certain Degs can be further modified by measure phrases and
quantifier phrases.

many
(6.41) a. five times much
far j
(many enough
much => five times 'enough
[far far enough
five times
many
five miles
little
many
c. too much
much
few
far
far
a bit

five times many five times more

five miles little five miles less
much * more - much ■ => « much ► more [
far few far fewer
. a bit Jar J a bit farther
158 RAY JACKENDOFF

The presence of the measure phrases and quantifier phrases depends on the
presence of the degree words, as can be seen by eliminating the degree words in (6.41).
Furthermore, which measure phrases and quantifier phrases are possible depends on
the choice of degree word, as can be seen by substituting other degree words into
(6.41). This suggests that the measure phrases and quantifier phrases are in the
specifiers of the degree words. Bowers (1968b) and Selkirk (1970) therefore argue that
this recursion is through the Deg'", as in (6.42).

(6.42) Q'"

many

In (6.42) I have attached the new specifier to the Deg" rather than to Deg' or Deg'".
This choice enables us to generate these specifiers with the phrase structure rule (6.43),
which generalizes with rules (6.15)—(6.19) of section 6.3. Furthermore, the definition
(6.13) of the grammatical relation measure phrase can apply with complete generality.
These generalizations argue that Deg is a full-fledged participant in the X system, as
predicted by the Uniform Three-Level Hypothesis of phrase structure.

+ Subj '»
(6.43) Deg" -Obj ) - Deg'
Det _

Given this measure phrase or QP in the Deg" specifier, we are in a position to

generate such recursive wonders as (6.44), for which it is left to the reader to find a
context, if (s)he cares.
 6: SPECIFIERS OF X'"

(6.44) . as many times more feet too tall

Deg' Q

Deg many

as
160 RAY JACKENDOFF

b. five times as much too far away from us

five N

times

These structures are generated by the recursion of the two basic phrase structure rules
developed in this chapter, the one generating a Deg'" in various X'" specifiers, and the
other generating NP or QP in a wide range of X" specifiers.
 6: SPECIFIERS OF X

Bresnan’s analysis of these constructions is different. She claims that the recursion
of measure phrase and QP goes (in our terms) through the Q"\ Thus, instead of (6.42)
she proposes the structure (6.45).

(6.45) Q'

many

However, if we adopt the structure of A'" argued for in sections 6.3 and 6.4.2, the
X Convention compels us to adopt (6.42): since most adjectives do not subcategorize a
quantifier, the only structure available for much too tall is (6.46), parallel to (6.42). No
structure can be constructed for much too tall that parallels (6.45).

(6.46) A'

tall

Since Bresnan purports to eliminate (6.42) as a possible structure, it is incumbent on us

to deal with her argument against it.
Bresnan points out that (6.45) describes too many as a constituent but (6.42) does
162 RAY JACKENDOFF

not. She then claims that there is a transformation that moves too many, hence it is a
constituent, and (6.45) must be correct. The paradigm in question is (6.47).

(6.47) a. I have as many too many marbles as you.

b. I have as many marbles too many as you.
c. I have six too many marbles.
d. I have six marbles too many. (= Bresnan’s (132)—(133))

(6.47a,c) contain recursive degree phrases like (6.41), i.e. either structure (6.42) or
(6.45). Bresnan claims that (6.47b,d) are derived from (6.47a,c) by an optional
reordering transformation which moves too many to the right of marbles, a rule called
QP-Shift. Since only under hypothesis (6.45) is too many a constituent in (6.47a,c), the
requirement that only constituents be moved favors (6.45) over (6.42).
I claim, however, that there is no such transformation as QP-Shift, but rather that
(6.47b,d) are base-generated forms with the following structure:

(6.48) N'"

N"

Q'" N'

Deg Q PRO

N" Deg many

marbles
 6: SPECIFIERS OF X'"

J six I
According to this claim, i f marbles in (6.47b,d) is functioning as a measure
las many J
phrase in Deg"', just as Jive times is in (6.42). By contrast, the QP-Shift theory of
(6.47b,d) claims that marbles is the head and does not form a constituent with

fas many J
].
To differentiate between the theories, observe first that the string six marbles too
many is paralleled by five days too long (after the accident), five miles too far (down
the road), and five pounds too much, in which measure phrases precede degree
modifiers. The only question in establishing the parallelism is the unusual use of
marbles as a unit of measure. But it is not implausible to say of a tray for marbles or an
array of marbles that it is six marbles wide; marbles is a reasonable unit of measure if
one happens to be counting marbles. Such is the case in (6.48): presumably the
interpretation of PRO must be fixed as marbles. The PRO posited as head in (6.48) can
furthermore be replaced by a lexical item in certain (albeit limited) cases where the unit
of measure differs from what is being measured, as in (6.49).

(6.49) a. This book took five reams too many pages,

b. This cake took five pounds too much sugar.

Thus the relevant features of the structure (6.48) are independently motivated, and the
appropriate interpretation is assigned to it by regular rules. Hence QP-Shift appears
unnecessary in order to generate the paradigm (6.47).
Furthermore, the data which motivate constraints on QP-Shift can be explained
immediately by structure (6.48). Bresnan claims that QP-Shift must follow 0/-Insertion
in order to prevent (6.50) (Bresnan’s (136)—(137)).

(6.50) a. I have six more of them.

b. *1 have six of them more.
c. I have half a dozen too many of these marbles.
d. *1 have half a dozen of these marbles too many.

However, contra Bresnan, an of can appear in this construction:

(6.51) a. I have hundreds more marbles.

b. I have hundreds of marbles more.
c. I have five pounds too much sugar.
d. I have five pounds of sugar too much.

The real constraint seems to be that the noun cannot be definite. But this is a general
constraint on measure phrases:

(6.52) a. *five of them too far

b. *half a dozen of those miles too long
164 RAY JACKENDOFF

’six 1
Hence the constraint follows from the claim that marbles is a measure
as many J
phrase in (6.47b,d).
Second, Bresnan claims that QP-Shift must be restricted to ‘‘count” Qs:

(6.53) a. much too much bread

b. *much bread too much

But this restriction too is not quite accurate, as much appears after sugar in (6.5Id).
Again, the real restriction is a general one on measure phrases, namely that they must
be count. Since by the present hypothesis much bread would have to be a measure
phrase in (6.53b), it is ungrammatical. But in (6.5 Id) five pounds of sugar is count, so
the construction is acceptable.
Thus the QP-Shift theory of (6.47b,d) requires special constraints on an extra
transformation. The base theory uses an independently motivated base construction to
describe (6.47b,d); the constraints on the paradigm follow automatically from independ¬
ent properties of measure phrases.
Returning to the issue of (6.42) vs. (6.45), we see that Bresnan has not made a case

for the constituency of j many in (6.41); hence we are free to choose (6.42) rather

than (6.45) as the proper structure. (6.46) is therefore a viable structure for much too
tall, precisely parallel to (6.42), and it is in turn possible for us to continue to maintain
that most adjectives cannot take quantifiers.
Having defended the recursion of measure phrase and QP through the Deg", we
have completed our analysis of the degree system. Though we have not dealt with the
complete range of data discussed by Bresnan, in particular the relation of so and such,
our analysis here has been successful in dealing with the major points of Bresnan’s
theory, eliminating such irregularities as the nonoccurrence of less before most
adjectives, the optional application of A/wc/z-Deletion and Enough-Shift under certain
conditions, and the suspect transformation QP-Shift. The reader will have noticed that
at each step of the analysis, the issue of whether most adjectives take quantifier
phrases has played a crucial role, for it led to differences in the treatment of the
exceptional adjectives alike and different, the position of degree phrases in A'", the
categorial status of more, less, and enough, and finally the recursion of degree
phrases.5

5 Lest the part of Bresnan’s theory we have not dealt with should be thought unassailable, it is worth
mentioning that there are problems with her treatment of the such/so alternation, for which I have been
unable to find a more satisfactory solution. The crux of the issue is her transformation AP-Shift, which
transforms taller a man (parallel in the base to as tall a man) into a taller man, and so tall a man into such a
tall man. She claims (p. 308) that AP-Shift is obligatory when the NP lacks a determiner, so that underlying
*so tall 4> men becomes such tall men. However, this alternation is not general, since none of the other
constructions of the form Deg A a N permit a plural alternant:

a man
(i) men
 6: SPECIFIERS OF X

The analysis here has been carried out within the strong hypothesis of phrase
structure proposed in chapter 3. We have shown that the entire range of specifier
possibilities we have discussed is the product of two major phrase structure expansions
which apply to a large number of categories. In arriving at these rules, we made the
fullest possible use of the X Convention to make up for the scarcity of data within any
single category, under the assumption that cross-category generalization is an impor¬
tant part of the evaluation measure in syntactic theory.
As a final note, we should briefly mention a sizable group of specifiers we have not
touched on, which have a wide range of cross-category application, such as not,
almost, just, quite, nearly, practically, hardly, scarcely, and others. Some of these are
patently adverbs, and in NPs are replaced by their adjectival parallels; but the rest are
of some as yet undetermined category, perhaps Deg. Almost is a good example of
cross-category applicability—it can modify even the articles every and all, though up to
now we have had no need for internal structure in Art'".
(6.54) a. John almost fell. (in V'")
b. The pudding has become almost glue. (in N'")
c. The men look almost alike. (in A'")
d. We pushed John almost out the window. (in P'")
e. Almost every groundhog saw his shadow. (in Art'")
f. John left almost happily. (in Adv'")
g. You talk almost too much. (in Deg'")
Only the categories Q'", M'", and Prt'" are missing. Such behavior is not atypical
among this class of modifiers. Study of this class within the lexicalist framework would
no doubt yield a rich source of evidence for feature analyses and phrase structure rules.

6.6. Summary of Rules

The following phrase structure rules were developed in this chapter:

X'" rules

(6.23) N'" -> (Art'") - N"

(6.22) Q'" -» (Deg'") - Q"
(6.26) a. A'"' -> (Deg'") - A"
b. Adv'" (Deg'") - Adv"

This suggests that such tall men may not be a transform of underlying [so tall] men, but has some other
source, parallel to what tall men.
On the other hand, enough, like the comparative, does not share this restriction to singular heads, since
tall enough men is grammatical. Thus AP-Shift must be obligatory for comparatives (*taller a man) and
optional for A enough a man, and there must be a constraint ruling out the plural cases in (i). In all, the
system is highly idiosyncratic.
There is one semantic factor worth mentioning. Many of the constructions which require this indefinite
article—(i) plus more of a man, so much of a man, and perhaps others—can be used only predicatively. A
similar restriction appeared on NPs with measure phrase specifiers such as two parts water. One might
speculate whether the analysis of (i) involves some bizarre expansion of the measure phrase in the N"
specifier. On the other hand, hardly a man and scarcely a man can be used as subjects. Since they count
men, and the phrases in (i) do not, they are semantically more like QP in the N" specifier.
166 RAY JACKENDOFF

The categories expanded here are the class [-Obj, — Det]"'. The generalization is
evident. As in the rules collected in section 4.5, NP is different by one feature. In that
case many categories expanded X" with a prehead Adv'", but N'" used an A'", a
difference in the feature Comp. Here the difference between Deg and Art is in the
feature Subj. The same problems of felicitous collapsing arise here as did in that case.

X" rules

+ Subj
(6.16) V" -> V' - ( -Obj
Det
+ Subj
-Obj
(6.15) N" -> ( - ((of))) - (A'")* - N' - . .
— Det
_( + Comp)_
+ Subj
(6.19) A" —> ( -Obj )-A'

-Det
+ Subj
(6.17) P"-> ( -Obj )- P'
-Det
+ Subj
(6.18) P" -> P' - ( -Obj )
-Det
+ Subj
(6.43) Deg" —» ( -Obj ) - Deg'
_—Det

To this we can add (6.55), since much differently is an Adv'", and none of the adjectives
that allow measure phrases have corresponding adverbs.

(6.55) Adv" —> (Q'") - Adv'

(or, perhaips with use of the N'" being vacuous for semantic reasons)
+ Subj
Adv" —» ( -Obj ‘) - Adv'
-Det .

Thus the rule creating this X" specifier expansion applies to all [ + Comp] categories and
all [-Subj, -Obj] categories.
One serious problem in collapsing these rules, of course, is that of the order of the
N"7Q " term in V" and P". We discussed this in section 4.5, and I have nothing to add
here.6

6 We have not, however, touched on the problem of combining these X" rules with the rules of chapter
4. The major difficulty is with the prehead Adv’" in A", P", and Adv". Consider A", for example. (6.19) must
 6: SPECIFIERS OF X

We have developed the following transformations.

(6.28) More /most - Q spelling rules

generalizable
More/most - A spelling rules
(6.32) Enough-5/n/r (revised)
enough - X° </> - 2 + 1
OBLIGATORY
(6.34) Many-Much Deletion

— tzh'
OBLIGATORY
- f

The results of this chapter should be by now obvious enough that I need make no
further polemic about how the specifier system demonstrates the virtues of the X
Convention and the proposed set of syntactic distinctive features. It is worth pointing
out again, though, that the inclusion of Q'" and Deg'' in the generalized phrase structure
rules is a strong argument that the X Convention and the Three-Level Hypothesis apply
not only to the major categories N, V, A, and P, but to minor categories as well.
Finally, for convenience, here is a summary of the major differences between our
theory and Bresnan’s.

Bresnan X Syntax
1. Adjectives subcategorize quantifiers Most adjectives do not subcategorize
quantifiers
2. A few adjectives such as alike and A few adjectives such as alike and
different govern optional rather than different subcategorize quantifiers
obligatory Much-Deletion
3. Enough is a Q Enough is a Deg and an Art
4. Much deletes before adjectives Much and many delete after enough

be combined with (4.52).

(4.52) A" —» (Adv"') - A' -. . .
The Adv’" and the N"7Q'" cannot cooccur; for example *five miles incredibly wide and *much surprisingly
alike are out. This would initially suggest that the rule collapses N"\ Q'", and Adv'" into a single term. Yet

the feature configuration that picks out these three terms is the improbable [-Obj, —Det, }]■
Consideration of the N" rule suggests that this is the wrong way to collapse (6.19) and (4.86) anyhow.
For in N" the parallel N"7Q"' and A'" terms are separate for good syntactic reasons, for example, many tail
men. This suggests that for the sake of generality the A" rule must be (i), in principle allowing the N"7Q"' and
the Adv'" to cooccur.
+ Subj
(i) A" —» ( -Obj ) - (Adv"') - A'
-Det
But since both modifiers have the semantic effect of designating a degree, there are semantic reasons for their
noncooccurrence. We may thus invoke a principle similar to the Specifier Constraint of section 5.2 in
justifying (i). Similar arguments apply in dealing with the rules for P" and Adv".
168 RAY JACKENDOFF

5. -er is the comparative Deg More and less are comparative Degs
6. Less far cannot be generated Less much, etc., must be ruled out
7. Adjectives are compared by means of Adjectives are compared by means of
a Q'" a Deg'"
8. Degree phrase recursion is through Q'" Degree phrase recursion is through
Deg'"

I would like to emphasize again here my debt to Bresnan’s work. Despite the
substantive difference between our analyses, the spirit of the work is in total
agreement. Though I feel the analysis here is superior in several respects, it could not
have been begun were it not for Bresnan’s insight in organizing the mass of data in
what I feel is so close to final form.
 7: Relative Clauses

7.1. Introduction

This chapter is concerned with where relative clauses are attached in deep structure. It
is concerned only secondarily with the internal structure of relative clauses and with
the processes of relative pronoun formation and wh-preposing, since these have little
bearing on the form of the base.
Chapter 4 claimed that restrictive relative clauses are daughters of N" and that
appositives are daughters of N'", as in (7.1).

(7.1) the man who came to dinner, who Bill dislikes

N'"

man who came to dinner

restrictive apposidve

Though this claim fit in nicely with the general framework proposed in chapter 4, we
postponed defending it in any detail until the present chapter, where we will show how
it accounts for many well-known facts about relatives and compare it with other widely
accepted analyses of relative clauses. We will show that our theory, the “NP-
complement” theory, accommodates the data used as evidence for other theories at
least as well as the other theories themselves do.
The major theories with which the NP-complement theory is in contention are the
“determiner” theory, illustrated in (7.2), and the “Chomsky-adjoined” theory, illus¬
trated in (7.3).
170 RAY JACKENDOFF

(7.2) the man who came to dinner

on
\
\
\
\
\
N"' N'" \
\
\
\
\
\
\
\

who came to dinner

The determiner theory, advocated by Smith (1964) and Stockwell, Schachter, and
Partee (1973), among others, is consistent with the constraints on phrase structure rules
adopted in chapter 3. But since this theory requires an obligatory extraposition, the
NP-complement theory is prima facie less complex. Hence any evidence that the NP-
complement theory is equally adequate must be taken very seriously.
The popular Chomsky-adjoined theory, advocated by Ross (1967), is appealing
because of the simplicity of stating the antecedent of the relative pronoun: it is just the
entire lower NP. However, (7.3) is not a possible structure within the phrase structure
rule schema of chapter 3, which requires every X'to immediately dominate an X1-1. If
at all possible, we would like not to have to weaken the schema for the sake of relative
clauses.
We will not consider the hypothesis in which restrictive relative clauses originate
as sentences conjoined to some clause dominating the NP, as advocated by Thompson
(1971), for example. Section 7.9 will advance arguments against Ross’s (1967) well-
known theory that appositives are so derived, and all these arguments apply with equal
or greater force to restrictive relatives.
The next two sections deal with the relatively simple problems presented by
 7: RELATIVE CLAUSES

appositives, after which five sections are devoted to the much more complex issues
involved in restrictives. In many cases, we will show that the theory of complements
laid out in chapter 4 uncovers otherwise unexpected generalizations which competing
theories cannot express—in other words, that what the X Convention leads us to
believe is the simplest solution is in fact simplest.

7.2. Appositives
To start with, an adequate theory of appositives must account for numerous differences
between restrictive relatives and appositives. Appositives must appear to the right of
restrictives:
(7.4) a. The man that came to dinner, who was drunk, fainted,
b. *The man, who was drunk(,) that came to dinner fainted.
An NP may contain two or more concatenated (“stacked”) restrictive relatives; but an
NP may take multiple appositives only if they are conjoined:
(7.5) a. the man who came to dinner who hated lox
b. The man, who came to dinner, who hated lox
c. the man, who came to dinner and who hated lox

Appositives, unlike restrictives, cannot be introduced by the complementizer that or </>

instead of a w/t-word:

f*(that) Bill saw! ,

(7.6) The man, 1 , f, sneezed.
( who Bill saw J

Appositives, unlike restrictives, must be able to modify proper nouns:

(7.7) a. John, who came to dinner, sneezed,

b. *John that came to dinner sneezed.

X"'
As pointed out in chapter 4, appositives can occur in any whereas
_+Comp_ ’
restrictives can occur only in NP:

which I
(7.8) a. Relative clause formation is obligatory in NPs,
That J aC'
counts for the difference in surface shape, (in V"')

not
b. That Sheila was beautiful,
That i she was, was

realized until later, (in A'")1

c. Solving this problem will take from now until doomsday,

{-*ch) is more time than we’ve got. (in P'")

1 (7.8b) is taken from Ross (1969), where it is used as an argument that APs must be dominated by NP.
The X Convention, of course, allows a different account, namely cross-category generalization.
172 RAY JACKENDOFF

The differences in intonation, complementizer, and distribution between restric-

tives and appositives argue rather strongly that the two kinds of clauses have different
syntactic sources. The fact that appositives always follow restrictives is explained by
the NP-complement theory as a consequence of restrictives being N" complements and
appositives being N'" complements. Furthermore, the comma intonation before apposi¬
tives is characteristic of all X'" complements, since sentence-final sentence adverbs,
parentheticals, and the like have similar intonation:

of course'
probably
(7.9) This ice cream tastes like spaghetti.
I bet
no doubt

Since the two kinds of relatives arise from different base positions, the difference
illustrated in (7.5) can be a consequence of the base rules. We will argue in section 7.6
that (7.5a) is generated by the phrase structure rule (7.10), which permits multiple
clauses.

(7.10) N"-> . . . N' . . . (S)*

The rule for appositives, on the other hand, lacks the star, so multiple appositives are
possible only through conjunction under the single S.
(7.11) N'" -» . . . N" . . . (S)

Note, by the way, that appositives do not concatenate in S, AP, or PP either; again
only conjunction is possible:
(7.12) a. Relative clause formation is obligatory in NPs, which accounts for the
f *</> ^
difference in surface shape, which only your theory predicts.
_ and _
*<t>
b. That Sheila was beautiful, which she was. which
and
she hadn’t been two years previously, was not realized until later,
c. Solving this problem will take from now until doomsday, which is longer
r*4> i
than most problems take. which is more time than we’ve got.
and

I intend to provide no account of the differences in complementizer treatment

between restrictives and appositives. However, the fact that the clauses arise in
different base positions makes it relatively simple to distinguish between them in
whatever rules insert or delete relative pronouns and complementizers. Thus the NP-
complement theory is not inconsistent with the observed differences.
The difference in base position of the two kinds of relatives also implies that they
have different grammatical relations in the sentence, that is, they are interpreted by
different projection rules. Two different types of projection rules are involved: the
general rules for N" or N'" complements, and the rules for interpreting relative
pronouns in N" relatives and in X'" relatives. One consequence of the former was
 7: RELATIVE CLAUSES

mentioned in section 4.1: restrictives, like other X" complements, can be focused and
negated, but appositives, like other X'" complements, cannot:
(7.13) a. We didn’t talk to the man who married SUSAN, (we talked to the man
who married JAYNE)
b. *We didn’t talk to the man, who married SUSAN.
The second kind of projection rule deals with the relationship between the relative
pronoun and the head. The rules for interpreting the relative pronoun in appositives
(section 7.3) and restrictives (section 7.8) will account for the differences observed in
(7.7) and (7.8), as well as a number of other differences to be pointed out in later
sections. Section 7.9 will show that a competing hypothesis of appositives, deriving
them from underlying conjunction, encounters serious difficulties.

7.3. Interpretation of the Appositive Relative Pronoun

Before the development of interpretive theories of anaphora, the interpretation of the
relative pronoun was a potential obstacle to the NP-complement theory. For example,
in the analysis of Chomsky (1965), the deep structure of a relative pronoun is taken to
be a literal copy of its antecedent; most work of that period makes a similar
assumption. But if the relative clause is inside the NP, the relative pronoun cannot
originate as a copy of the head, for, with either N" or N'" attachment of the head, an
infinite regress results:
(7.14) N"'

Ai

N from Siberia N V"

man

(wh)(the)

N from N V"
Siberia

man wh- . . .
174 RAY JACKENDOFF

This problem for the NP-complement theory can be eliminated by adopting an

interpretive theory of the relative pronoun, parallel to the by now familiar interpretive
theory of definite pronouns of Jackendoff (1972), Wasow (1972), and others. In the
interpretive theory, the relative pronoun is a relative pronoun in deep structure, not
some more fully specified NP form. The only semantic features it needs to carry are
those necessary for agreement with the head and with the verb of the relative clause.
Its semantic structure, if it were composed of syntactically separable elements, would
look like this:

(7.15) a. who
N'"

wh [PERSON] ([PLURAL])

b. which
N

wh [INANIMATE] ([PLURAL])

c. where

p"

P N""

wh [PLACE] ([PLURAL])
 7: RELATIVE CLAUSES

The bracketed expressions in capitals represent semantic markers of the usual sort. The
wh is a placeholder which triggers rules of semantic interpretation.
The simplest possible rule of anaphora is one which marks the relative pronoun an¬
aphoric to the entire dominating NP. The infinite regress of (7.14) does not arise, since
such a semantic rule does not provide the relative pronoun with internal syntactic
structure. Rather, two NPs are simply designated as coreferential, and one happens to
dominate the other. Such a rule seems appropriate for appositive relative pronouns,
though not for restrictive relative pronouns, for which we will propose a different sort
of rule in section 7.8.
The appositive rule can be stated like this:
(7.16) Appositive Wh-Interpretation
X
is anaphoric to Y"', in the configuration
_ + wh_
[Y.Y"[g[comp. • - X'" . . . ] S]]
(7.16) has been stated in its most general form, so as to account for appositives in V'",
A"', and P'" as well as in N'". The proper notion of anaphora for N'" appositives is
obviously coreference; but in other Y'" the proper interpretation of the notion
“anaphoric” is less clear. Whatever that notion may be, it is in any event independ¬
ently motivated, since the demonstratives it and/or that exhibit exactly the same kind
of anaphora as which. Compare the following pairs:

(7.17) a. John likes your idea, but it’s crazy.

(in 1ST)
John likes your idea, which is crazy.
b. Bill came late, and that bothered Susan.
(in V'")
Bill came late, which bothered Susan.
c. Bill is drunk all the time—is that how you’d like to be?
Bill is drunk all the time, which is probably how you’d like (in A'")
to be.
d. Bill went into the tree, and that’s where I’d like to go too.
(in P'")
Bill went into the tree, which is where I’d like to go too.
That is, relative pronouns in appositives can be anaphoric to the same constituents as
ordinary demonstrative pronouns can. However such anaphora is defined for demon¬
stratives, the definition will be equally appropriate for appositive relative pronouns as
well.
A rule of anaphora like (7.16) makes it very natural that appositives can be applied
to proper nouns; the burden is thus on the rule for restrictives to explain why
restrictives cannot be applied to proper nouns. However, there is also a case in which
restrictives can occur but appositives cannot, with certain quantifiers in the head:

(7.18) a. | Every| man w^° drives a Cadillac is insane.

b. Everyj man’ w*1° drives a Cadillac, is insane.

176 RAY JACKEN DOFF

Ross (1967) cites this construction as evidence for the conjunction source, since the
putative conjoined source of (7.18b) is also ungrammatical.

Any
| man is insane, and he drives a Cadillac.
Every
The present theory provides an explanation as well. Observe that everyone and no
one can be coreferential with a pronoun which is in a restrictive relative clause or in
other subordinate clauses within the scope of the quantifier (understood anaphora
indicated by italics):
(7.20) a. Everyone there had a wife who loved him.
b. No one wanted Sue to waken him.
c. Everyone got from Boston to a place he had been to before.
But such anaphora is impossible if the him is within an appositive.
(7.21) a. *Everyone there had a wife, who loved him.
b. *No one wanted Sue to leave, which suited him.
c. *Everyone got from Boston to Cincinnati, which seemed to him like a
long way.
The ungrammaticality of (7.21a,b,c) follows from the more general fact that appositives
are immune to the scope of quantifiers and negation. (7.22) confirms this fact by
demonstrating that a negation outside an appositive cannot condition some-any type
alternations within the appositive, though any can appear in restrictive modifiers in
parallel positions.
(7.22) a. I didn’t see a man who had had any drinks.
some
I didn’t see Bill, who had had drinks.
"any
b. That Sheila isn’t nervous to any appreciable extent surprises me.
sometimes
That Sheila isn’t nervous, which she is, surprises me.
"ever
c. We couldn’t get from Boston to a place any of us had been.
some
We couldn’t get from Boston to Cincinnati, which seemed to
'any
of us like a short way.
Since two NPs can be coreferential only if they are subject to the same logical
operators, and since only the pronouns in (7.20) and not those in (7.21) can be
quantified by the subject, coreference is possible only in (7.20).2 But the anaphora
established between an appositive relative pronoun and its head by (7.16) is claimed to
be of the same nature as anaphora with ordinary pronouns. Therefore, in (7.18b), the
relative pronoun, which is within the appositive and hence obligatorily unquantified, is
marked anaphoric to a quantified head. Such a situation is semantically anomalous, just
as in (7.21); hence (7.18b) is ungrammatical.

2 Cf. the Modal Coreference Condition of Jackendoff (1972, chapter 7), for elaboration of this point.
 7: RELATIVE CLAUSES

These two sections have shown that the NP-complement theory of relatives
provides a principled distinction between restrictive and appositive relatives, and that
an interpretive theory of the appositive relative pronoun is not only statable but also
revealing. We now turn in some detail to restrictive relatives.

7.4. Constraints on the Determiner and the Head in Restrictives

This section and the next three will deal with four different kinds of data which provide
apparent counterevidence to the claim that restrictive relatives are attached to N" in
underlying structure.
The first case concerns interaction between restrictive relative clauses and the
determiners of their heads. These interactions have often been taken as evidence that
the underlying position of restrictives is in the determiner (see (7.2)), since in such a
theory the necessary constraints can be stated as constraints on the form of a single
constituent; in the NP-complement theory, on the other hand, they must be stated in
some sort of discontinuous fashion. This allegedly favors the determiner theory. We
will show, however, that the determiner theory must be stretched to absurdity and the
discontinuous environments are independently necessary.3
Let us start with some cases in which the - X - Rel can appear but the - X alone is
ungrammatical.
(7.23) a. *the Paris
the Paris that I love
b. *the manner/way
the manner/way in which he did it
c. *the four of the boys
the four of the boys that came to dinner
There are also some cases (pointed out by Perlmutter (1970)) in which the proper article
is dependent on the content of the relative clause:

(7.24) He greeted me with

warmth I had not expected

Consider first (7.23a). According to the determiner theory, the application of an

Art'" to a proper noun can be restricted to a particular kind of Art"', namely the + 5. In
the NP-complement theory, the restriction involves the cooccurrence of two distinct
constituents, the before the proper noun and S after it; such a discontinuous constraint
is more complex to state.
Vergnaud (1974), however, observes that there are phrases like the Paris of the 30s
and the Paris of my youth, in which PPs satisfy the constraint on use of the definite
article. These particular PPs, moreover, are not reduced relative clauses, since *the

31 am particularly indebted in this section to the arguments of Reed (1974) and Vergnaud (1974).
178 RAY JACKENDOFF

\of the 30s

Paris that is ungrammatical. In order to prevent discontinuity in
[of my youth
the statement of this constraint, the determiner analysis must propose that the PP also
originates in the determiner and is extraposed:

(7.25) N N'"

4>

of the 30s N N P"'

Paris Paris of the 30s

The deep structure in (7.25) of course entails adding an otherwise unmotivated option
in the base rule for the determiner. The NP-complement theory, which countenances a
discontinuous constraint anyway, can generate the PP in its surface position and simply
generalize the constraint to a further class of modifiers.
Exploring further, we discover phrases like the old Paris and the unknown Paris,
with prenominal adjectives satisfying the determiner constraint. Since the former of
these means ‘the part of Paris which is old’ or ‘the Paris of the old days’, neither of its

readings can be derived simply from the phrase the Paris that j old by relative

clause reduction. Thus a base-generated prenominal adjective is semantically as

adequate as any other source. The determiner theory, in order to preserve the
generality of the constraint for (7.23a), is forced to put the adjective in the determiner,
thus:

old
 7: RELATIVE CLAUSES

There is of course no need for an extraposition rule this time, but the proposed
structure looks even more suspicious than the one in (7.25).
Actually, a different generalization emerges from these examples: a certain class of
restrictive modifiers (of any syntactic category) permits the use of the definite article
with proper nouns. APs, PPs, and restrictive relative clauses can all be interpreted as
restrictive modifiers, by whatever projection rules happen to apply to them. It follows
then that there is no privilege in being attached to the determiner; hence (7.23a) is not
evidence for the determiner theory of restrictive relatives.
Either of the theories, of course, can provide projection rules that interpret APs,
PPs, and restrictive relatives all as restrictive modifiers. However, only the NP-
complement theory permits the generalization, pointed out in chapter 4, that all
restrictive modifiers are daughters of N". This permits a single projection rule with a
single generalized structural description to apply to all three categories. Such a
generalization is impossible within the determiner analysis, since relative clauses
originate in an entirely different position than prenominal adjectives or postnominal PPs
do.4
As further evidence that relative clauses and PPs can perform the same semantic
function, consider pairs such as these, pointed out by Vergnaud (1974):

(7.27) a. the reason

of his arrival
c. the manner
in which he arrived

Since in each case the PP and the relative are nearly synonymous, a grammar in which
they are related to the NP by the same projection rule is to be preferred to one in which
they are not.5
Summing up this argument, the supposed advantage of the determiner theory can
be maintained only at the expense of positing deep structures like (7.25) and (7.26). The
NP-complement theory permits a different sort of generalization, one that positions
restrictive relatives so that the projection rule for restrictive modification can be of the
simplest form.
A parallel argument can be constructed for (7.23b).

4 The generalization is likewise impossible in Ross’s theory of restrictives, though Ross allows such a
variety of Chomsky-adjoined constituents and unrestricted transformations that it is difficult to tell what
claims he could make here.
5 Does the fact that these phrases cannot be one-pronominalized mean that both the PP and the relative
are under N' here?
of his departure of her arrival
(i) *the time and the one
at which he left at which she arrived }
180 RAY JACKENDOFF

(7.23) b. *the manner/way

the manner/way in which he did it

Here the head nouns cannot occur at all without the relative clause. The alleged
advantage of the determiner analysis is that the constraint on the occurrence of these
nouns is confined to a single constituent, the determiner. But again, the existence of
PPs and adjectives that suit the same purpose calls into question the generality of such
a solution:

(7.28) the way of the wise

the manner of his arrival
a new way
a pompous manner

The same argument applies: the generalization concerns the need for a restrictive
modifier, of whatever category (or a possessive this time, as in his manner). Thus the
only possible syntactic consideration is the generality of the configurations that can be
interpreted as restrictive modifiers, and the advantage of the determiner analysis
vanishes.

Similarly with (7.23c), where the is made acceptable not only by a relative clause,
but by an appropriate PP or AP, as in (7.29).

(7.23) c. *the four of the boys

the four of the boys that came to dinner
(7.29) the four of the boys with green eyes
the usual four of the boys

And with (7.24):

(7.24) He greeted me with

warmth of an old friend

(7.30) a. He greeted me with
warmth of great intensity

usual warmth
b. He greeted me with
unusual warmth
 7: RELATIVE CLAUSES

Notice that at least some of these examples cannot be derived by reduction of relative
clauses: * the four of the boys that was/were usual; *the four of the boys that were with
green eyes; *the warmth that was of an old friend. So again the constraints concern all
restrictive modifiers, not just relative clauses.
Next consider the inability of proper nouns to take restrictive relative clauses:

(7.31) *John that came to dinner

The determiner analysis claims that this constraint is related to the inability of proper
nouns to take determiners. But in fact, no restrictive modifiers at all are possible with
proper nouns, so the generalization of the NP-complement theory accounts for this
constraint adequately.
One might claim that a phrase like poor John is a counterexample to the
generalization of the NP-complement theory. But poor here is not a restrictive
modifier; it does not constrain the choice of possible referents of the phrase. Rather,
something more akin to the projection rule for appositives must integrate poor into the
interpretation of poor John. I leave open whether this entails a different structural
position for such an adjective.
There remains one sort of argument for the determiner analysis, the inability of
restrictive relatives to cooccur with a genitive NP in the determiner:

(7.32) *John’s book that you stole

Here other restrictive modifiers are possible (e.g. his old book, his book with a faded
cover), so the previous argument for the NP-complement theory does not apply. The
determiner hypothesis predicts this ungrammaticality by claiming that when the
determiner is filled by NP the expansion Art - S is unavailable.6 Section 7.8 will show
how this constraint can be explained as a consequence of the relative pronoun’s
interpretation, making the determiner theory’s explanation unnecessary.
This section has thus shown that the motivations for the determiner analysis,
namely the dependencies between the relative clause, the article, and the head, are
(with one exception) consequences of more general constraints on all restrictive
modifiers. Since restrictive relatives must be interpreted as restrictive modifiers no
matter what syntactic analysis is assigned to them, there is in fact little ground for

6 Many proponents of the determiner analysis, however, simply require the extraposition of the
possessive to the form the book of his that you stole, a much more mechanical explanation, and one available
to all three analyses of relatives. But a syntactic relation between the determiner genitive and this postposed
genitive is very difficult to support, as argued in Jackendoff (1969b). For a different analysis of the postposed
genitive, see section 5.3.4.
182 RAY JACKENDOFF

preferring the determiner analysis. Rather, the NP-complement analysis appears to

allow the most general statement of the projection rule for restrictives, since the
constraints show that restrictives in nearly all respects behave just like other daughters
of N".

7.5. Restrictive Relatives on Quantified NPs

Another problem in the analysis of restrictive relatives (to my knowledge, first pointed
out by Dean (1966)) concerns the interpretation of relative clauses in NPs with
quantifiers.

a. many men who like knitting

(7.33) Mary knows b. many of the men who like knitting
c. the many of the men who like knitting

In (7.33a) the antecedent of the relative pronoun appears to be men or many men; in
(7.33b) the men; in (7.33c), the many. The problem is to specify quantifier and relative
clause structures which interact properly to create these interpretations.
I shall not be concerned here with describing various attempts to account for these
facts within other frameworks, for most writers seem to bog down hopelessly on this
issue, mainly in accounting for the partitive structures. Rather I will be content to show
that the analysis of partitives in section 5.3 interacts with the NP-complement theory to
provide an adequate account of (7.33).
Here are the possible structures. When there is no partitive N' complement, as in
(7.34), the relative clause can be attached only in one position; when there is a
partitive, as in (7.35), the relative can be attached to either the higher or the lower NP.

(7.34) (the) many men who like knitting

N'"

men
 7: RELATIVE CLAUSES

(7.35) a. many of the men who like knitting

N'"

PRO the men who like knitting

b. many of the men who like knitting

N'"

N"

c. the many of the men who like knitting

N'"
184 RAY JACKENDOFF

d. the many of the men who like knitting

In (7.34) the relative clause modifies many men-, there is no way to separate its
modification into that of many vs. that of men. Such a prediction seems correct. Many
of the men who like knitting is, however, ambiguous. Most typically it seems to have
the structure (7.35a), with the relative attached to the men. But a phrase such as many
of them who like knitting, in which the relative clause cannot be attached to the
complement (*them who like knitting), shows that structure (7.35b) is possible too.
Further, if evidence from number agreement is at all reliable, (7.36a) must have
structure (7.35a), and (7.36b) must have structure (7.35b).

I one I
(7.36) a. of the men who like knitting
[each J

I one I
b. of the men who likes knitting
[each J

Finally, of the pair (7.35c,d), only the latter is a correct structure for the many of
the men who like knitting, since, as was shown in section 5.3, there is a constraint
requiring a relative clause attached to the upper NP in this construction. This can be
shown either by preposing, as in (7.37), or by number agreement in the relative clause,
as in (7.38).

(7.37) a. Of the men, the many who like knitting are the happiest,
b. *Of the men who like knitting, the many are the happiest.

likes
(7.38) the (only) one of the men who knitting
*like

Of course, if a second relative clause is added, it may modify the lower NP. Compare
the preposing and number agreement in (7.39) to (7.37)—(7.38) to see that one restrictive
in (7.39) is attached to the upper NP, and one to the lower.
 7: RELATIVE CLAUSES

(7.39) a. Of the men who like bowling, the many who like knitting are the
happiest.
b. the (only) one of the men who like bowling who likes knitting

We have shown, then, that the NP-complement theory of restrictives combined

with the N'-complement theory of partitives provides the right structures and interpre¬
tations for constructions involving both quantifiers and relatives. Thus the evidence
from quantification supports the NP-complement theory of relatives.

7.6. Stacking of Restrictive Relatives

Another kind of apparent counterevidence to the NP-complement theory concerns the

phenomenon of “stacked” restrictives. To see what problems it raises, observe that
(7.40a) and (7.40b) appear to differ in meaning.

(7.40) a. the men who hated lox who came to dinner

b. the men who came to dinner who hated lox

It is commonly claimed that the difference is due to the fact that the second relative
clause modifies or restricts both the head and the first relative clause. Thus the
antecedent of the first relative pronoun in (7.40a) is the men\ of the second, the men
who hated lox. The theory par excellence for expressing this difference is Ross’s
Chomsky-adjoined theory, since (7.40a), for example, can receive the deep structure
(7.41).

Within both the circled and boxed NPs, the subordinate clause can relativize, so each
NP becomes the men who hated lox. Then the boxed NP is identical with the circled
NP and can itself become a relative pronoun to yield the desired surface form (7.40a).
Thus the intuitions about (7.40) are accounted for by “stacking” the relative clauses.
In the NP-complement theory, however, it appears difficult to account for the
186 RAY JACKENDOFF

“stacked” interpretation, since the most plausible structure with two restrictive
relative clauses simply concatenates them:

men

The fact that the relative clauses are not stacked in (7.42) appears to be a serious
drawback. We will show, however, that this is not a drawback but a virtue of the NP-
complement theory.
Consider the kind of context in which the “stacked” interpretation emerges most
clearly:

(7.43) a. The men who hated lox who came to dinner left early, but that wasn’t
I the men who hated lox ! . . ~
true ot |tjlose j who arrived alter 8:00.

b. ??The men who came to dinner who hated lox left early, but that wasn’t
true of the men who arrived after 8:00 who hated lox.

(7.43a) sets up a situation in which the men who hated lox who came to dinner
contrasts with the men who hated lox who arrived after 8:00. According to the rules of
focus and presupposition of Chomsky (1970b), Akmajian (1973), and Jackendoff (1972,
chapter 6), the contrasting constituents must be foci and thus must receive prominent
stress. The contrasting parts of the two subject NPs in (7.43a) are the final relative
clauses, where highest stress normally appears. In (7.43b), though, the contrasting
relative clauses are medial in the NP, and the contrast can be obtained felicitously only
by placing a strong stress on the contrasting relatives and effacing stress on the final
relatives:

(7.44) The men who came to DINNER who hated lox left early, but that wasn't
true of the men who arrived after eight O’CLOCK who hated lox.

This suggests that the phenomenon of “stacking” is not to be accounted for in the
syntax, but rather in the system of presupposition and focus. Two questions are raised
by such a conjecture: first, are the rules of presupposition and focus capable of deriving
interpretations which account for the intuitions of stacking? Second, if stacking has
 7: RELATIVE CLAUSES

nothing in particular to do with how relative clauses are attached, can other things
produce stacked interpretations as well?
The rules for focus and presupposition derive (7.45a) as the interpretation of the
first clause of (7.43a), (7.45b) for (7.43b), with normal stress, and (7.45c) for (7.44), with
special stress.7

(7.45) a. Presupposition: some men who hated lox left early

Focus: who came to dinner
b. Presupposition: some men who came to dinner left early
Focus: who hated lox
c. Presupposition: some men who hated lox left early
Focus: who came to dinner

In (7.43a) and (7.44) the presuppositions can be matched in the two clauses, and the
foci contrast properly. But in (7.43b) the presupposition of the second clause involves
men who arrived after 8:00 and the foci do not contrast, hence the observed infelicity.
Notice also that in (7.45a) the focused relative clause specifies members of the set
of men who hated lox; in (7.45b) the focused relative clause specifies members of the
set of men who came to dinner. These are exactly the properties required to satisfy the
intuition of a “stacked” interpretation, and they are achieved without any syntactic
counterpart of the semantic stacking.
“Stacked” interpretations can also be produced with pairs of PPs in the NP
complement. As with relatives, the stacking can be reversed by changing stress.

(7.46) a. The weather in Kansas in 1875 was better than

| that
in 1936.
[the weather in KansasJ
b. The weather in Kansas in 1875 was better than the weather in Australia
in 1875.

(7.46a) is felicitous only with stress on 1875 and 1936\ (7.46b) is felicitous only with
stress on Kansas and Australia. It is clear that the paradigm is parallel to (7.43)—(7.44).
Furthermore, the same pattern appears in the case of a PP followed by a restrictive
relative:

(7.47) a. The man with a scar who came to dinner hated

jthe one j wfo0 COoked lunch.
(the man with a scar J
b. The man with a scar who came to dinner was fond of the man with green
eyes who came to dinner.

Again, (7.47a) must be read with stress on dinner and lunch, and (7.47b) with stress on
scar and eyes.

7 I am oversimplifying here, assuming only one of a number of possible foci linked to the given stress
pattern, namely the relative clause. The other possibilities are irrelevant to the context of (7.43) and (7.44).
See the references cited above for details of the treatment of focus and presupposition assumed here.
188 RAY JACKENDOFF

Since the syntactic subordination has now been shown unnecessary, stacking is no
longer a ground for preferring the Chomsky-adjoined analysis of restrictive relatives
over the NP-complement theory. The latter assigns the relevant NPs in (7.43)—(7.47)
the following structures:

(7.48) a. N'"

men

b. N'"

man with a scar who came to dinner

 7: RELATIVE CLAUSES

This analysis treats (7.43)-(7.47) as altogether comparable, since both PPs and relatives
can be attached to N".
Only one simple change must be made in the phrase structure rule for N" to
account for multiple relatives: the S at the end must be replaced with S*.

(7.49) N"^ ... - N' - (P'")* - (S)*

A moment’s consideration shows in fact that our analysis of stacking in terms of

focus vitiates the alleged advantage of the Chomsky-adjoined theory of restrictives—its
claim that syntactic stacking reflects semantic stacking. A sentence with reversed
stacking such as (7.44) will require a deep structure in which the order of the relative
clauses is reversed, and an extraposition rule will be necessary to create the proper
order. For parallelism, a sentence with stacked PPs such as (7.46a) will require a
structure such as (7.50), for which there is little evidence, and the reversed stacking in
(7.46b) creates the same problems as (7.44).

(7.50) N'"

the weather in Kansas

(7.47a) will not be a problem, but the reversed stacking in (7.47b) requires the deep
structure (7.51), which appears at the surface only with great awkwardness, unless the
S is extraposed.

81 must emphasize again that PPs like with a scar cannot in general be derived by reducing relative
clauses.
190 RAY JACKENDOFF

Furthermore, the rules dealing with the interpretation of focus and presupposition on
the basis of surface stress are independently necessary, and make it possible to produce
“stacked” interpretations with normal stress patterns and no syntactic subordination of
the “inside” relative clause.
The syntactic stacking of the Chomsky-adjoined theory is thus not only superflu¬
ous but even disadvantageous. Hence the phenomenon of stacking is not only not
counterevidence to the NP-complement theory, but also provides important evidence in
its favor. The generalization underlying this result is the same as in section 7.4:
restrictive relatives work exactly like all other restrictive modifiers.

7.7. Conjoined Heads

The fourth and final class of apparent counterexamples to the NP-complement theory
involves data first pointed out (to my knowledge) by Ross; the argument in the form
given here is due to Vergnaud (1974). Consider the NPs in (7.52).

(7.52) the car and the truck that collided

the boy and the girl who met in Vienna

These NPs cannot be derived by conjunction reduction, since the sources would have
to be the ungrammatical *the car that collided and the truck that collided, *the boy who
met in Vienna and the girl who met in Vienna. The verbs of these relative clauses
require plural or conjoined subjects; hence the relative pronoun’s antecedent cannot be
either of the nouns alone.9
Ross’s analysis deals with such examples straightforwardly:10

(7.53)

that collided

9 Dougherty’s (1970-1971) discussion of conjunction is, I believe, sufficient to eliminate such distant
sources as the car that collided with the truck and the truck that collided with the car. A source once
suggested by Ross (in lectures at MIT), the car wh- [car and truck] collided and the truck wh- [car and truck]
collided, is semantically incomprehensible.
10 Vergnaud is actually proposing a slightly more sophisticated theory than Ross’s; however, with
respect to this particular argument the two theories are equivalent. I will therefore couch the argument in
terms of Ross’s theory, which will probably be more familiar to the reader.
 7: RELATIVE CLAUSES

In this structure, the relative clause is sister of a constituent that dominates both NPs,
and the antecedent of the relative pronoun can be the conjoined NP. On the other hand,
the NP-complement theory provides no convincing structure, since the Art'" is superior
to the relative clause:

that collided

Since only a theory like Ross’s can come even remotely close to a suitable structure for
(7.52), and it does so with great ease, Vergnaud considers this construction overwhelm¬
ing evidence for such a theory.
However, if Vergnaud’s argument is correct, then PP restrictive modifiers must be
outside the NP too, since they occur in the same sorts of constructions:

with the same birthday

with mutual interests
(7.55) the boy and the girl
with different-colored eyes
with a common background

These examples can be derived neither by Relative Clause Reduction nor by Conjunc¬
tion Reduction, since both (7.56a) and (7.56b) are ungrammatical.

(7.56) a. *the boy and the girl who were with the same birthday (Relative Clause
Reduction source)
b. *the boy with the same birthday and the girl with the same birthday
(Conjunction Reduction source)

If these PPs are in the NP complement, (7.55) raises the same problem in any theory as
(7.52) does in the NP-complement theory of relatives: where is the PP attached?
192 RAY JACKENDOFF

Lest one should immediately yield to the temptation to attach these PPs outside
the NP, we observe that N' complements create this problem too.
(7.58) three students and two teachers of the same language
three members and two vice-chairmen of interlocking committees
Vergnaud’s argument forces us to conclude that the proper bracketing is [[three
students] of French] and [[three members] of the committee], not the otherwise highly
motivated [three [students of French]] and [three [members of the committee]]. Things
are beginning to look fishy.
To make life even worse, the X Convention as usual leads us to discover similar
examples in sentences:

similar issues
(7.59) a. John avoided and Bill ignored the same man
men with the same birthday

the same tune

b. John whistled and Mary hummed ' at equal volumes
together

None of these sentences can arise from Conjunction Reduction or Right Node Raising,
since the presumed sources would either have the wrong reading or be ungrammatical:

similar issues
(7.60) a. John avoided • the same man • and Bill ignored
men with the same birthday

similar issues
the same man
men with the same birthday
 7: RELATIVE CLAUSES

the same tune

b. John whistled ' *at equal volumes ' and Mary hummed
^together

{the same tune

*at equal volumes
together

By Vergnaud’s argument, then, the phrases in braces must be right sisters of conjoined
subject-verb pairs:

(7.61)

John whistled Mary hummed

But this flies in the face of the presumably well-established fact that the phrases in
braces are direct objects or other V' and V" complements.
It is really unfair to saddle Vergnaud with this argument, since such arguments
based on conjunction have been considered strong evidence for structure throughout
the history of transformational grammar. We have seen, however, that this type of
argument leads ultimately to conclusions which contradict the most fundamental and
best accepted notions of phrase structure.
The question is, which set of arguments should be given up? If we wish to retain
the arguments based on conjunction, we give up all distinctions between various types
of complements, since they must all reduce to the Chomsky-adjoined variety. All the
arguments of chapter 4 distinguishing complement types must go by the wayside and all
evidence for internal structure of NPs and Ss must be accounted for in some other way.
I prefer to believe that the error lies rather in the theory of conjunction, always a
troubled area in transformational grammar. There are to my knowledge no remotely
coherent formalizations of Right Node Raising; Gapping is ever problematical; and
even Dougherty’s (1970-1971) highly restricted version of Conjunction Reduction
comes to grief in examples like these:

got drunk and was arrested by the cops

(7.62) The same man ' is rarely easy to please and eager to please
praised you and seemed to hate you

Dougherty invokes a Conjunction Reduction transformation specifically in order to deal

with conjoined derived verb phrases. But the underlying source he needs is ungram-
194 RAY JACKENDOFF

matical here:

(7.63) *The same man got drunk and the cops arrested the same man.
*It is rarely easy to please the same man and the same man is rarely eager
to please.
*The same man praised you and it seemed [(for) the same man to hate
you].

If I am correct, and the anomaly lies in the theory of conjunction, we can retain all
the usual arguments for phrase structure; only the arguments from conjunction have to
be rejected. I have no positive proposals to offer on the analysis of (7.52)—(7.62). But
this choice defuses Vergnaud’s otherwise powerful argument for the Chomsky-adjoined
theory of relatives; thus the NP-complement theory of relatives, like the theory that
there are direct objects in sentences, can be maintained. What is really needed is a
thoroughly revised theory of conjunction.

7.8. Interpretation of the Restrictive Relative Pronoun

Having dealt with potential difficulties for the NP-complement theory of restrictives in
the areas of determiner restrictions, quantification, stacking, and conjunction, we now
turn to the problem of interpreting the restrictive relative pronoun. As in the case of
appositive relative pronouns, we assume an interpretive theory, in which the relative
pronoun is inserted into underlying structure and \vh-preposed; a semantic rule applies
to the surface structure to establish the interpretation of the pronoun. The problem is to
find out exactly what this interpretive rule does. If it is properly defined, it can account
for otherwise puzzling facts about relative clauses.
Section 7.3 showed that the appropriate interpretive rule for the appositive relative
pronoun was one that established anaphora between the relative pronoun and the head.
A similar rule of anaphora could be developed for restrictive relative pronouns, but
there is reason to consider a different interpretation. To see this, let us look a bit more
closely at the semantics of NPs.
The semantic functions of various parts of an NP down to the N" level are of at
least four types: demonstrative or deictic, classificatory, measuring, and commenting.
The demonstrative function is taken by articles; they show whether the individual in
question is old or new in the discourse, unique, pragmatically designated, generic, and
so forth. Elements with the classificatory function place conditions on what sort of
object the individual in question is; the N' and the restrictive modifiers have this
function. Measuring and commenting functions are taken by numerals (and perhaps N”
quantifiers) and by appositives, respectively.
Traditional logic has not had much to say about the latter two functions, but it
makes a sharp distinction between the former two. All the classificatory functions have
been represented as predicates; those demonstrative functions that have been treated at
all emerge as operators such as quantifiers or the definite description operator. Thus,
 7: RELATIVE CLAUSES

for example, (7.64) might be represented as (7.65) in one version of logical notation,
restricted quantification. For illustrative purposes, I will use the rather free translation
characteristic of traditional logic.

(7.64) a. Some brown beaver died.

b. The picture of Bill from London fell down.
c. Every duck in Australia is pink.
(7.65) a. 3xx iS brown, x is a beaver (die(*))
b. foil down {lXx is a picture of Bill, x is from London)

C- Vxx is a dUck, x is from Australia (pink(x))

In these translations, the article has become an operator; each classificatory constituent
has been supplied with an “x is” and placed in the restriction on the variable controlled
by the operator.
Restrictive relative clauses require a slightly different treatment. In this notation,
the most appropriate translation of The man who Bill saw died is (7.66).

(7.66) die(?xx a ma)!; gm saw x)

Here the relative clause has become a restriction on the variable. The translation rules
have not added the fixed material “x is”, as in (7.64); rather, the relative pronoun has
itself been replaced by the bound variable x. In other words, ordinary classificatory
elements can be made to restrict the operator by adding fixed external material, but
restrictive relative clauses must be tampered with internally, in the position corre¬
sponding to the relative pronoun. The tampering involves replacing the relative
pronoun by the bound variable controlled by the operator.
I suggest that something parallel to these logical translations is going on in the
natural language semantics of restrictive relative clauses. The relative pronoun is the
syntactic marker of the position which must be bound by the demonstrative operator of
the head. Since the syntactic correlate of the demonstrative is the Article position, the
rule for interpreting the relative pronoun must establish a “binding” relationship
between the relative pronoun and the article:

(7.67) Restrictive Wh-Interpretation

X
is bound to Art'", in the configuration
+ wh_
Art'"[N». . . [§[ Comp . . . X' . ] S] . . . ]

The exact characterization of the “binding” relationship depends on the ultimate form
of semantic notation chosen; in the rough logical translation rules above, “is bound to”
would mean “is replaced by the variable of’.
The crucial mention of the Art'" in (7.67) may be used to account for one of the
differences between appositives and restrictives. Since N'" is the only category which
dominates Art'", restrictives can be related only to NPs. But since appositive relative
196 RAY JACKENDOFF

pronouns are interpreted as anaphoric to the entire head, the internal structure of the
head is not crucial, and appositives may thus be generalized to all major syntactic
categories.
The dependence on Art'" in (7.67) plays a role in accounting for certain ungram-
maticalities as well. Section 7.4 mentioned one such example:

(7.32) *John’s book that you stole

John's is not an Art'", so the relative pronoun cannot be bound; since the example does
not receive a complete interpretation, it is anomalous. Next consider the ungrammati¬
cally of (7.68a), and compare it to (7.68b,c).

(7.68) a. *the destruction of the city that you observed

b. the destruction that you observed
c. the destruction of the city, which you observed

Recall the discussion of section 4.7.4. There it was argued that when the is the only
specifier possible in a derived nominal, it arises not through expansion of Art'", but
through insertion into an empty N'". The destruction of the city is such a case, and the
destruction is not, since we have, for example, *much destruction of the city but much
destruction. Thus in (7.68a), the is dominated by N'" and the structural description of
(7.67) cannot be met, hence the phrase is ungrammatical; in (7.68b), the is dominated
by Art'", so the phrase is grammatical. Since the interpretation of appositive relative
pronouns does not depend on the presence of an Art'", (7.68c) too is grammatical,
though its determiner is dominated by N'".
This analysis of the relative pronoun eliminates the usual puzzles (cf. for example
Kuroda (1969)) about whether relative pronouns are definite or indefinite: they are
neither. In fact, the issue hardly makes sense in terms of the interpretation given by
(7.67) , since the bound variable is not referential at all, but simply has the effect of
converting the proposition represented by the relative clause into a property. So, for
example, a man who you know denotes an individual not completely identifiable from
previous discourse (i.e. indefinite), who is male, adult, and known by the hearer. The
man who came to dinner denotes an individual identifiable by this description plus
previous discourse or general knowledge (i.e. definite), who is male and adult and who
came to dinner. A relative clause may be “definitizing”, i.e. render an otherwise
indefinite NP definite, just in case it provides grounds for unique identification which
were not present in the NP without the relative. Thus there is no transformation of the
form that Perlmutter (1970) and others adopt, changing the article of an NP to the in the
presence of a restrictive relative clause.
In cases where the demonstrative element is interpreted as picking out a sort or
degree, it does so in the relative clause as well. For example, the object of We drank
the wine we always drink denotes not a particular physical body of wine (barring
 7: RELATIVE CLAUSES

regurgitation), but a sort or amount of wine. If this interpretation is localized in the

semantic markers of the determiner, it will correctly apply both to the head and to the
relative clause.
In an effort to give a plausible underlying determiner to the relative pronoun in
a generic sentence like (7.69a), Stockwell, Schachter, and Partee (1973) are forced to
propose a source like (7.69b), with an attendant lowering transformation to derive the
surface form.

(7.69) a. A beaver that eats worms builds leaky dams,

b. If a beaver eats worms, it builds leaky dams.

In the present analysis such a strategem is unnecessary. The generic character of the
head NP is determined by the indefinite article a, and this binds both the head N beaver
and the relative pronoun. Thus the conditional character of the relative clause in (7.69a)
is a result not of some unusual underlying form, but of the binding by (7.67) of the
relative pronoun to a generic determiner.
Similarly for NPs with quantifiers in the Art'" such as every man who owns a
Cadillac. There is no need to resort to an underlying form such as if a man owns a
Cadillac, he . . . in order to capture the semantic connection between the relative
clause and the main clause. Rather, binding takes place in the usual fashion, as shown
in (7.65a,c), and the quantifier automatically affects the relative clause.
A more subtle analysis than we have given here is required to account for the fact
that any can show up in quantified relative constructions such as every man who has
any sense. However, the failure of the present analysis to predict this is not serious,
since these peculiar anys turn up not only in relative clauses but also in restrictive PPs,
for example in every man with any sense. Thus the possibility of any here is not to be
explained in the semantics of restrictive relative clauses, but in a more general account
of restrictive modification, which we have not attempted to provide.
This completes our account of restrictive clauses. We have shown that the NP-
complement theory of restrictive relatives is capable of dealing with a wide range of
apparent counterevidence, and that in fact it permits a more general account of
restrictive modification than either the determiner source or the Chomsky-adjoined
source. Moreover, it requires neither an extraposition transformation, like the determi¬
ner theory, nor a weakening of the fundamental base rule schema, like the Chomsky-
adjoined theory. This last consideration is most important to the general goals of this
study, and it is the reason we have devoted so much space to the problems of relative
clauses.

7.9. Appendix: Against the Conjoined Source for Appositives

In the Chomsky-adjoined theory of restrictives, restrictives are attached as high in the

NP as they can possibly be. Thus there is no remaining position in NP to attach an
198 RAY JACKENDOFF

appositive that distinguishes it from a “stacked” restrictive:

(7.70) NP

Thus there can be no surface structural significance associated with the intonation
break characteristic of appositives, already a serious problem.
To provide a deep structure difference between restrictives and appositives, Ross
(1967) proposes deriving appositives from underlying conjoined clauses. For example,
(7.71a) is to be derived from something more or less like (7.71b).

(7.71) a. That man, who came to dinner, sneezed,

b. That man sneezed, and he came to dinner.

Thus appositives have a different underlying source from restrictives, and many of the
differences between them can be captured in terms of the underlying forms or the
derivations of the two kinds of clauses.
Ross himself (section 6.2.4.1) points out a serious difficulty with the conjunction
source: appositives can occur within imperatives and questions, though the sentences
(7.73a,b), putatively more like the underlying form, are ungrammatical.

(7.72) a. Go to Cincinnati, which is on the Ohio River.

b. Are we landing in Washington, which is on the Potomac?
(7.73) a. *Go to Cincinnati, and it is on the Ohio River.
*Cincinnati is on the Ohio River, and go there,
b. *Are we landing in Washington, and it is on the Potomac?
^Washington is on the Potomac, and are we landing there?

Ross therefore halfheartedly suggests that appositives be derived from pairs of

independent sentences, in effect by a Syntactic Structures-style “generalized transfor¬
mation”. Thompson (1971) rightly objects to this. But the only alternative she offers,
without argument, is that structures like (7.73) indeed do underlie (7.72), but if
appositive formation does not take place, the conjunction is expediently deleted to form
two independent sentences. In effect she circumvents the counterexamples by brute
force, explaining nothing.
Were this not difficulty enough, the derivation of the surface form requires a
transformation lowering the appositive into the NP. Even if lowering rules are
countenanced (as they are not in most lexicalist theories), one should have serious
 7: RELATIVE CLAUSES

reservations about lowering appositives into such strong islands as conjoined NPs
(7.74a) and complex NPs (7.74b).

(7.74) a. Haj, who likes beer, and Sandy, who hates it, couldn’t agree on one
restaurant.
b. Paul’s book about the transgalactic lexicon, which is the hottest topic of
the decade, is eagerly awaited.

In defense of the conjunction source, Ross points out a sentence type which he
claims is an intermediate step in the derivation from normal conjunction to appositives.

(7.75) Bill, and I’ve known him for many years, isn’t as smart as he makes
himself out to be.

Here a conjoined sentence has been inserted after the subject, supposedly halfway to
becoming the appositive who I've known for many years. Unfortunately for this
argument, though, the presumed intermediate step ranges from dubiously grammatical
to impossible when substituted into examples such as (7.72) and (7.74):

(7.76) a. *Go to Cincinnati, and it’s on the Ohio River.

b. *Are we landing in Washington, and it’s on the Potomac?
c. *Haj, and he likes beer, and Sandy, and she likes pizza, couldn’t agree
on one restaurant.
d. ?*Paul’s book about the transgalactic lexicon, and it’s the hottest topic
of the decade, is eagerly awaited.

The transformation producing (7.75), in other words, has all the constraints expected of
a rule combining two conjoined sentences; the rule producing appositives does not.
Hence the construction in (7.75) cannot be an intermediate step on the way to
appositives; in fact its constraints constitute evidence against Ross’s theory of
appositives.
We see then that Ross’s theory still lacks a viable underlying form for appositives.
Of course, one way to avoid the problems of the conjunction source is to generate
appositives in their surface position, as part of NP. But since restrictives already
occupy the highest possible position within NP, there is no room in NP for a higher
base position for appositives. Thus the disposition of appositives in Ross’s theory is
left enigmatic.
 8: Degree Clauses

8.1. Introduction

Section 2.2 mentioned a parallelism, pointed out by Bowers (1968b), between NPs with
restrictive relatives and APs with degree clauses associated with the degree words so,
too, enough, as, more, and less, such as those in (8.1).

(8.1) a. He’ s so tired that he’ll never get up in time.

b. He’ s too tired to get up.
c. He’s tired enough to sleep 14 hours.
d. He’s as tired as he’s ever been.

e. He’s tired than you are.

Bowers was the first (to my knowledge) to express this parallelism in terms of the X
Convention. Selkirk (1970), Milner (1973), and Reed (1974) develop the parallelisms
further. Postal (1974) has an extensive discussion of their semantic similarities, giving a
nonlexicalist formalization.
If this parallelism is genuine, and if restrictive relatives are attached to N" in deep
and surface structure, as argued in chapter 7, the X Convention compels us to adopt a
theory of degree clauses in which they too are attached to X" in deep and surface
structure. The present chapter compares such a view to the prevalent view that degree
clauses are extraposed from degree phrases. It will be shown that the X"-complement
theory is in fact quite plausible, though not yet entirely compelling. Thus the
framework proposed here is at least consistent with the facts of degree clauses, even if
it cannot yet be shown that it predicts them.

8.2. The Extraposition Theory and its Evidence for the X Convention

Degree clauses can appear at the surface in any of the major syntactic categories A'",
N'", V'", and P"\ Bowers points out that they always follow strictly subcategorized
complements in surface structure. Furthermore, since they always precede appositives,
202 RAY JACKENDOFF

the most likely surface position for them is in the X" complement. No rearrangement of
the complements in (8.2)-(8.5) is possible.1

(8.2) In A'":
a. he’s [so [[afraid to lose A«] that he’s freaking out, A»] which you’ll never
be Am]
b. he’s [too [[proud that he’s won A.] to admit it wasn’t such a big
deal A"]A'»]
he’s [too [[sickA,] to admit it wasn’t such a big deal,A„] which you’ll never
beA», ]
fof Bill
c. he’s [[[afraid enough to freak out,A»] which you'll
[that he’ll lose
never beA-»]

d. that she was [ j [[eager to winA,] than Fred was,A„] which she

hadn’t been expected to be,A,»] surprised me.

(8.3) InN'":
a. we heard [[so many [rumors that Dewey had wonN,] that we started
believing themN,»]N,»]
b. we heard [[too many [rumors that Dewey had wonN.] to question the
papersN„]N,„]
c. we heard [[as many [rumors that Wallace had wonN,] as we could
stomach,N»] which wasn’t manyN„,]
(8.4) In V'":
a. [he [[gave so little to the fundv-] that he looked stingy,v»] which surprised
no onev™]
b. [he didn’t [[say enough to us about how he did itv-] to justify himself,v»]
which was a shamev»]

c. [this county [has [elected j Democrats to the Senatev<] than

any other in Missouri,v»] which is hard to explainv»]

(8.5) In P'":
a. we waited [[long enough [after the earth stopped shakingp-] for things to
settle downP»]P,»]
b. [[as much [out of placeP<] as he isp.]P»], you're worse
c. we pushed him [[farther [into the roomP,] than Harry had been,p«] which
still wasn’t very farP»]

It is rather obvious that there is a connection between the degree words and the

1 The status of appositives in some of these examples is rather curious. The appositives in (8.3c) and
(8.5c) seem to comment on the quantifier rather than the head, and no other type of appositive seems
possible in N"' or P'" of this sort. I have no explanation.
 8: DEGREE CLAUSES

associated degree clauses. If the degree words are omitted in (8.2)—(8.5), the sentences
become ungrammatical unless the clauses in X" are omitted too. Furthermore, each
degree word governs characteristic complementizers:2

(8.6) so . . . that S
too . . . for-to S

as ... as S

This strongly suggests that degree clauses are strictly subcategorized complements of
Deg, despite their surface position. Bowers (1968b) and Selkirk (1970) take this view,
generating deep structures like (8.7) and extraposing the degree clause to its surface
position.

(8.7) a. A

b. A

2 For the complementizer status of as and than, see Bresnan (1970; 1972; 1974).
204 RAY JACKENDOFF

N'"

N"

d. N

The similarity of (8.7) to the determiner theory of restrictive relatives is evident:

 8: DEGREE CLAUSES

(8.8) N'"

Art"' N"

As Bowers and Selkirk point out, the extraposition operations in (8.7a,b) and (8.8) are
structurally parallel; only by use of the X Convention can this parallelism be expressed.
Furthermore, the deep structures are parallel, in that Deg and Art are both X'"
specifiers; in the present theory of syntactic features, they differ by the same feature,
±Subj, as their respective heads, A and N.
(8.7c,d) are more complex, but essentially the same as (8.7a,b), particularly if one
assumes the S to be extraposed in two stages, first to the end of Q"\ then to the end of
N'". Under this assumption, the extraposition transformation used in (8.7a,b) and (8.8)
needs only to be generalized to Q'" via the X Convention; the double extraposition will
be carried out cyclically, and no additional rules are necessary.
Similar double extrapositions take place in (8.9), where the categories dominating
Deg'" are different.

(8.9) a. as extraordinarily lively as he used to be

A'"

A"

Adv'" A'

as / S Adv' lively
206 RAY JACKENDOFF

b. too far along the road for us to see him

)1,1

D"

With more complex degree specifiers, the number of extrapositions increases:

(8.10) as much too much bread as I could stand

N'"

N"
 8: DEGREE CLAUSES

It is clear that the rule accounting for surface position of degree clauses will have to
make liberal use of the X Convention, whether or not it generalizes with the rules for
restrictive relatives.

8.3. Arguments against Extraposition of Degree Clauses

In chapter 7 we rejected the view that restrictive relatives are generated under Art'", in
favor of generating them under N". Does this mean that they are in fact not parallel in
deep structure to degree clauses? Or should we seek arguments similar to those of
section 7.4, to show that degree clauses too should be generated in their surface
position under X", restoring the parallelism?
Reed (1974) gives arguments relevant to this question. First, she points out that the
expression the same governs a degree clause with the complementizer as, in which the
permissible deletions are comparable to those in other comparatives.

you
(8.11) a. He looks the same as you do • . (in A'")
I think he used to (look)

you
b. He’s reading the same book as you are (reading) (in N'")
I think you used to

you
c. He spoke the same as ‘ you do (in Adv'")
I think you used to

But in N'" it also occurs with the complementizer that and the deletion pattern
characteristic of restrictive relative clauses:
you’re reading
*you (are)
(8.12) He’s reading the same book that
I think you used to read
?I think you used to

Thus the same must be allowed to govern either a degree clause or a relative clause;
since the latter is permissible only in N'", it will of course not crop up in other
categories. This option is stated most simply if degree clauses and restrictive relatives
are derived from the same underlying position, either in the specifier or in the X"
complement.3
One of the arguments adduced against the determiner theory of restrictives was
that there are PPs that serve the same semantic function as restrictives. Reed observes
that there are similar examples for some degree clauses;

J One possible drawback of this argument is that which cannot be substituted for that in (8.12), as would
be expected in an ordinary relative clause. This may indicate that the same does not govern a genuine
relative clause. On the other hand, it may indicate that we have been mistaken in regarding relative clauses
with that and those with which as deriving from precisely the same source. If there were an underlying
difference, the same could be constrained to govern only the former type.
208 RAY JACKENDOFF

I to wear shorts I
(8.13) a. He’s too fat
[for shorts J

b. He’s tall enough

(to be a butler
for a butler

I as you are
c. He’s as sly
las a fox

d. He’s taller
{than you are
than six feet

(Notice that in the latter two, the PP case is unlikely to be a reduction from the
questionable las sly as a fox is, *taller than six feet is.4 Rather, as and than, like for,
must double as prepositions and complementizers.) Thus if degree clauses are underly¬
ing complements of Deg, extraposed to their surface position, these PPs must be too.
Reed next points out that the degree clauses and PPs of (8.13a,b) occur with a
number of adverbial specifiers too, with similar meaning:

I to be a President I
(8.14) a. He’s quite young
[for a President J'

to be in school
b. He’s awfully young
for school

for us to use (it)

c. This book is sufficiently boring
for a textbook

I to have made that mistake I

d. He’s a pretty careful scholar
[for a mistake like that [
And in fact they can occur without any specifier:

(8.15) He’s young

{to be in school
for school

Since at least these particular degree clauses and PPs seem to have an existence
independent of the degree words, Reed argues that they must be generated in deep
structure to the right of the adjective, parallel to restrictive relatives, and that their

4 Bresnan (1973) says the underlying clause in taller than six feet is than six feet = x much, invoking an
underlying verb “ = ” which never appears on the surface and which has a reading suspiciously similar to the
copula.
 8: DEGREE CLAUSES

interaction with the degree words is purely semantic. The base rule for A", therefore,
must be of the form (8.16).

(8.16)

I have no further empirical arguments as to whether other degree clauses are

generated in their surface position or whether they are extraposed to their surface
position from the Deg'" complement. However, note that the extraposition theory
certainly goes against the spirit of the Extended Standard Theory, in that it claims an
underlying structure for which there is no purely syntactic evidence. Only cooccur¬
rence restrictions suggest the relationship of degree clauses to Deg'", and furthermore
there are no well-formed surface structures which contain a true complement of Deg. If
cooccurrence restrictions could be stated as a function of the surface configuration, the
grammar would be spared both an otherwise unmotivated phrase structure rule and the
transformation needed to eliminate from the surface that phrase structure rule’s effects.
We therefore will explore the alternative theory that degree clauses are generated
directly in their surface position in X", attempting to state an interpretive rule that
accounts for the semantic dependency between degree clauses and degree phrases.

8.4. A Generalization in the X"-Complement Theory

We first observe that the X"-complement theory of degree clauses restores the
parallelism between degree clauses and restrictive relatives pointed out by Bowers,
Selkirk, and others. Not only is there a structural parallel, since both types of clauses
are generated as X" complements, but there is a parallelism in their interpretation rules
as well. Recall that the rule for interpreting restrictive relative pronouns developed in
section 7.8 involves establishing a binding relation between the Article and an element
of a clause in the following configuration:

(8.17) N'"
210 RAY JACKENDOFF

Similarly, the simplest cases of degree clauses, such as (8.7a,b), are interpreted by
establishing a semantic relationship between an X'" specifier and part or all of a clause
in the X” complement:

More complex cases have a less direct syntactic relation between the Deg'" and the S;
it is interesting, however, that at least the simplest cases so closely parallel restrictive
relatives.
As some indication that this parallelism is not just an accident, recall the
discussion in section 4.8 of sentences with “subject complements”, for example It
bothers me that Tom isn't here. It was argued that such a complement clause is a V"
complement, interpreted by establishing an anaphoric relation between it and the
dummy subject it. The configuration is precisely parallel to (8.17) and (8.18):

Thus it seems possible to claim that there is a general rule of anaphora between X'"
specifiers and X" complement clauses, with three special cases. If the X'" specifier is an
NP, the X" complement is interpreted as a subject clause; if the X'" specifier is an Art'",
the X" complement is interpreted as a restrictive relative clause; if the X'" specifier is a
Deg'", the X" complement is interpreted as a degree clause. Notice that this generaliza¬
tion is more or less exhaustive, since N'", Art'", and Deg'" are the only major syntactic
categories other than M'" that occur in any X'" specifier.5 Furthermore, both constitu-

5 Since sentence adverbs, parentheticals, etc., are interpreted appositively, they are grouped semanti¬
cally with complements, not with specifiers, even when they precede the head.
 8: DEGREE CLAUSES

ents that take N'" in the X'" specifier, namely V'" and N"\ take subject clauses; the
only constituent with an Art'", namely N"', is the only one with restrictive relatives; all
constituents that take Deg'", namely A'", Adv'", and Q'", also allow degree clauses.
Thus an interpretive rule for degree clauses generated in their surface position appears
to fit into a general cross-categorial class of anaphoric rules for which there is some
independent motivation. Furthermore, a grammar incorporating such a rule would
express the observed similarities between restrictive relatives and degree clauses, while
maintaining the NP-complement theory of relatives argued for in chapter 7.

8.5. The Location of the Binding Degree Word

The configuration (8.18) is, however, only the simplest in which degree clauses can be
interpreted. This section will deal with more complex situations.
Besides cases like (8.18) in which the degree clause is bound to the X"' specifier of
the immediately dominating phrase, it can be bound to an X" specifier:

(8.20) a. sufficient food to survive for a week

N'"

b. sufficiently insensitive to survive Brandeis

A"

A"

This configuration is also the one that appears with same and different:
212 RAY JACKENDOFF

(8.21) the same books as you read

different books than you read
N"'

Also, as seen in section 8.2, the relationship can pass through a (probably
indefinite) number of embedded X'" and X" specifiers:

(8.22) a. too many stories about Bill for us to bear (too is Q'" specifier of N"
specifier)
b. as extraordinarily lively as Bill is (as is Adv'" specifier of A" specifier)
c. sufficiently many men for us to stop counting (sufficiently is Q" specifier
of N" specifier)
d. as much too tall (for the job) as Bill is (as is Q'" specifier of Deg"
specifier of A'" specifier)
e. as much too much bread as I could stand (as is Q'" specifier of Deg"
specifier of Q'" specifier of N" specifier)

The most general condition I have been able to discover for the position of the
specifier is this:

(8.23) The specifier binding a degree clause must be the head of a phrase on the
leftmost extremity of that X'" which most directly dominates the degree
clause.

This condition certainly holds in all the examples given so far, except for those in V'"
(8.4).

(8.4) a. He gave so little to the fund that he looked stingy, which surprised no
one.
b. He didn’t say enough to us about how he did it to justify himself, which
was a shame.

c. This county has elected j Democrats to the Senate than any

other in Missouri, which is hard to explain.

 8: DEGREE CLAUSES

As evidence of the essential correctness of (8.23), consider (8.24). The cannot cooccur
with a degree clause. (8.23) predicts this: since the is the leftmost specifier, the degree
clause is bound to the instead of to the degree word, producing semantic anomaly.

(8.24) a. the too many stories about Bill (*for uc to bear)

b. ?the sufficiently many men (*for us to stop counting)
Since there seem to be very few other cases at all in which a degree specifier can be
preceded by another independent specifier phrase, the generalization stated in (8.23)
appears to be correct.
Note that the condition speaks of the degree word being not leftmost, but the head
of a leftmost phrase. This permits it to take its own specifiers:

(8.25) a. [five times too] many men for us to catch

b. [twice as] much soup as we wanted
c. [three pounds more] sugar than was necessary

In fact, this statement of the condition permits two different degree words to bind
clauses in the same phrase:

(8.26) too much more bigoted than Vera is for us to like him
A"'

too much

Since the lower Deg'", of which too is the head, and the upper Deg'", of which more is
the head, are both at the leftmost extremity of the A'", both too and more can bind
degree clauses. The fact that the clauses are in the order they are in will be accounted
for in section 8.7.
It thus appears that (8.23) is reasonably close to the correct relationship between
degree clauses and their binding specifiers, though the cases in V'" still remain to be
accounted for.
214 RAY JACKEN DOFF

8.6. Why an Interpretive Theory Is Needed

Once the degree clause is related properly to its associated specifier, the interpretations
of these two elements must be related to the rest of the sentence. In the cases of the
clauses governed by enough, so, and too, this presents no further syntactic problem,
just one of understanding the semantics. Since the semantic problems must be
surmounted no matter where the degree clause is attached in underlying structure, their
solution does not affect the decision between the extraposition and the X"-complement
theories.
In comparative clauses, though, there is a missing element whose reading bears
some relation to the phrase dominating the clause. For example, in John read more
books than Fred read, the object of the comparative clause is missing. Defining the
interpretation of this missing element involves syntactic problems as well as semantic
ones, as it turns out.
Selkirk (1970) and Bresnan (1973) argue that this deleted (or PRO) string begins
with a w/z-like dummy degree word x and is otherwise identical to (or anaphoric to) the
string between the governing degree word and the clause.6 In (8.27) the bracketed string
represents the material missing from the surface.

(8.27) a. John is as tall as Bill is [x tall]

b. John has as many books as Fred has [jc many books]
c. Susan is more incredibly smart than Moe is reputed to be [x incredibly
smart]
d. we bought as many too many books as you bought [x many too many
books]

In Bresnan’s and Reed's (1974) analyses, the comparative clause is attached

outside the constituent dominating the degree word:

(8.28) a. A'"

x tall (f>
(or underlying PRO)

In Lees (1961), x is identified with the Deg that, just as wh at the time was identified with the Art the.
 8: DEGREE CLAUSES

b. 1ST

N"' S

more books than Bill has N

x many books => $

(or underlying PRO)

Under this analysis of the surface structure, the deletion or anaphora can be based on a
relationship between full constituents. In this it is structurally similar to the Chomsky-
adjoined theory of relative clauses, in which the relative pronoun’s antecedent is taken
to be the entire head NP.
But this analysis is also parallel to the Chomsky-adjoined theory in another way: it
violates the theory of phrase structure of chapter 3, having an X'" whose head is an X'"
instead of an X". Besides threatening the general theory of phrase structure, it
contradicts the conclusion of section 8.2 that degree clauses are attached to X". That
conclusion, in turn, was based on the theory that appositives are attached to X'", one of
the strong points of the present theory of phrase structure.
In the X"-complement theory of degree clauses, though, the constituent relation¬
ship of (8.28) is impossible. The upper and lower X'"s cannot be identical, since one
contains an S and the other does not. The most one can say is that a substring of the
upper X'" is identical with the whole lower X'" aside from the binding Deg'".

(8.29) a. A

tall as Bill is A
216 RAY JACKENDOFF

b. N"

N"

Thus the X"-complement theory of degree clauses has the same apparent difficulty of
interpretation as the NP-complement theory of relatives: nonconstituent identity.
The problem is exacerbated by stacked degree clauses such as the ones in
Bresnan’s example (8.30).

(8.30) Mary swam as many more laps than Joan swam as Linda swam.

The degree words and clauses are related from the inside out: the inner degree clause
goes with the second degree word and the outer clause goes with the first degree word.
The outer degree clause, furthermore, measures not simply a number of laps, but a
number of laps more than Joan swam. In Bresnan's outside-X'" theory, a stacked
structure can be constructed which accounts for this, preserving constituent identity in
the anaphoric process:
 8: DEGREE CLAUSES

This is precisely parallel to the Chomsky-adjoined theory’s account of the anaphoric

relations in stacked restrictive relatives (cf. example (7.41), p. 185).
In the X"-complement theory of degree clauses, the anaphoric relations in this
example can be established only stringwise, not as structural identity between single
full constituents:
(8.32) Mary swam N'"

N"
218 RAY JACKENDOFF

In a deletion theory of comparatives this structure would be untenable, since (at least
under the usual assumptions) constituent identity, not string identity, is necessary for
deletion.
Thus, if we are to adopt the X"-complement theory of comparative clauses, in
keeping with the strong version of the X Convention, we are forced to develop an
interpretive theory for the missing element of comparatives, a theory that is based on
identity of strings, not structure. The next section proposes such a theory.

8.7. Interpreting PRO

In an interpretive theory of comparatives, the righthand bracketed elements in (8.29)

and (8.32) are replaced by deep structure PRO:

(8.33) a. as tall as Bill is [A-»PRO]

b. more many books than Bill has [N»PRO]
c. as many more many laps than Joan swam [N»PRO] as Linda swam
[ N« PRO]

There are three steps involved in interpreting such structures. The first step is to
locate the degree word which binds it, i.e. which satisfies condition (8.23). This part of
the process is common to the deletion and interpretive theories. For concreteness,
suppose that the degree word is more, that the complementizer is than, and that the
clause is dominated by N'", as in (8.33b).
The next step of interpretation is to give content to PRO within the clause. In the
cases we have been considering, PRO must be of the same category as the X'" most
immediately dominating the clause.7 Assuming Bresnan’s account of the semantics,
PRO should be interpreted in the following way. Designate the string between more and
than as Z. Replace the Deg'" dominating more by a Deg'" dominating *, and assign PRO
the interpretation of the string* - Z. This string is equivalent to Bresnan’s underlying
structure.
It must be shown that * - Z has a coherent interpretation. Z contains some string
of specifiers (for example many in (8.33b)), followed by the N' and possibly some
daughters of N" to the right of the head. Since more is the head of a Deg'" at the
leftmost extremity of N'", replacing this Deg'" by * yields a complete specifier for* -

7 We are not considering any cases of what Bresnan (1976a) calls Sub-Deletion, for example (i) (elements
missing in the surface are bracketed).
(i) She is less beautiful than she is [x] funny.
We bought more books than we sold [x many] cars.
He has as many too many books as we have [x many] too few pencils.
Here the deleted elements are identical only to part or all of the specifier of the phrase dominating the degree
clause. These cases do not constitute a potential threat to the X"-complement theory even with deletion,
since the parallel strings are full single constituents, not partial constituents as in (8.29) and (8.32). 1 leave for
future research how Sub-Deletion should be stated interpretively, though 1 assume it will retain many of the
features of Bresnan’s elegant deletion analysis.
 8: DEGREE CLAUSES

Z. The N' of course has a coherent interpretation. Any N" complement within Z is
either an ordinary restrictive modifier, which can be integrated into the interpretation
independent of the rest of the contents of the N" complement, or else it is of the type
which must be bound by a specifier. If it is of the latter type, the binding specifier must
also be contained within x - Z for the interpretation of .v - Z to be coherent. But this is
precisely what is found in the case of stacked degree clauses: any degree clause
between a degree-word-degree-clause pair is matched by its own corresponding degree
word. For example, in (8.33c), the first PRO is in a clause governed by more, so x - Z
is x many laps. The second PRO is in a clause governed by as, so x - Z is x many more
many laps than Joan swam PRO, exactly the desired interpretation.
The final step in interpreting the degree clause is to assign an interpretation to x
itself. Presumably this involves logically binding it in some way to the governing degree
words, by a process similar to that for restrictive relative pronouns. But since an
appropriate logical formalization of degree words is not yet extant, this step must be
left for future research. Again, this problem is common to the deletion and interpretive
theories.
Thus, of the three steps involved in an interpretive theory of degree clauses, the
first and last are independently necessary in the deletion theory. The middle step, the
interpretation of PRO, has been shown to provide coherent interpretations of the
desired sort; this step can be carried out with the comparative clause generated as a
daughter of X" instead of as a daughter of a higher X'", despite the nonmatching of
constituent structure. Thus, like relative clauses, comparative clauses need not be
exceptions to the general phrase structure rule schema.
In turn, appositives must be attached higher than degree clauses, as shown in
(8.2)—(8.5). If we had needed to generate comparative clauses at a higher node than X",
this would have forced appositives into a position which was also an exception to the
phrase structure rule schema. Generating comparative clauses in X", however, permits
appositives to remain in X'", in conformity with the arguments of sections 4.1 and 7.2.

8.8. Recapitulation

The argument has gone as follows. First, whatever the specific details of one’s analysis
of degree clauses, a theory of grammar is necessary in which cross-category generaliza¬
tions can be made, since degree clauses occur in such a wide variety of positions.
Second, the theory that degree clauses are generated in their surface position has
certain syntactic advantages over the theory that they are extraposed from Deg'". But
this requires the more complex semantic rule (8.23) to establish the binding of the
degree clause to the Deg'". Finally, in order to preserve the X"-complement theory of
comparative clauses, thus avoiding a weakening of the general theory of phrase
structure, we had to face the problem of string rather than constituent identity. An
interpretive rule was developed which was shown to provide coherent interpretations
even to the complex stacked comparatives.
220 RAY JACKENDOFF

Along the way, we have observed somewhat distant but still striking generaliza¬
tions among degree clauses, restrictive relatives, and possibly even subject comple¬
ments. It has been particularly instructive to see how the determiner, Chomsky-
adjoined, and NP-complement theories of relatives discussed in chapter 7 are paralleled
by corresponding theories of degree clauses, how much alike the arguments for and
against the three theories are in the two domains, and how in both cases an interpretive
theory of the missing constituent in the clause is necessary to save the X"-complement
theory. If nothing else, these similarities constitute an argument on the metalevel in
favor of a close, though abstract, relationship between the two constructions.
 9: Deverbalizing Rules

9.1. The Deverbalizing Rule Schema

Section 3.6 pointed out that gerundive nominals are an exception to the normal
hierarchical arrangement of syntactic categories predicted by the X Convention: they
are dominated by NP but contain no head N. It was suggested that the phrase structure
rule that describes them is one of a class of systematic exceptions to the Uniform
Three-Level Hypothesis, characterized by the following phrase structure rule schema:

(9.1) Deverbalizing Rule Schema

X‘ af - V*

(9.1) expands a supercategory of the lexical category X as a supercategory of verb of the

same level, plus a grammatical formative or affix. This chapter will motivate several
rules of this type in English. The overall effect of applying a deverbalizing rule will be
an X'" whose head is not an X but a V; such an X'" will display some properties of X"'s
and some properties of sentences. (9.2) shows the three possible skeletal structures,
ignoring complements and specifiers.

The structures grow internally less X-like and more sentencelike, as i goes from 1 to 3
and as the complements and specifiers are determined by more supercategories of V
 RAY JACKENDOFF

and fewer supercategories of X. Externally, however, they are all still X"'s and occur in
positions characteristic of X'", not S.
Note that (9.1) strongly restricts the possible exceptions to the major rule schema
in several important respects. First, the only category that can appear on the right is V:
this seems to be true of all category-switching rules I have encountered. Second, even
when the category of the head of a phrase is switched to V by the application of one of
these rules, the hierarchy of primes is undisturbed: there is no instance, say, of A'
expanded as an affix plus V" or V'". Third, at the level where the category is switched
by a deverbalizing rule, there are no further complements or specifiers other than the
grammatical formative that marks the change of category. In each of these three
respects, it is clear what kind of evidence would entail weakening the restrictions; for
the moment, though, (9.1) seems like a plausible initial hypothesis.

9.2. Gerundive Nominals

Gerundive nominals are the most NP-like of the three complement types studied by
Rosenbaum (1967). As Emonds (1970) shows, they differ from that and for-to
complements in that they occur in all NP positions, including subject position in
questions, relative clauses, and for-to complements.

John’s leaving
(9.3) a. What would ‘ *that John left ■ reveal about him?
*for John to leave

John’s leaving
b. a man who ' *that John left ' would irritate
*for John to leave

{John’s leaving
*that John left r to bother us.
*for John to leave J

Thus gerundive nominals are apparently NPs, though they lack a noun head. To find
the level at which they change from N' to V\ we investigate what complements and
specifiers they take.
The level which explains the most about the structure of gerundive nominals is the
X" level. To recapitulate the discussion of section 2.1, gerundive nominals permit
within them all the transformations such as Dative Shift and Particle Movement
characteristic of V' but not N'. They are preceded by preverbal adverbs, characteristic
of V", not by adjectives, characteristic of N". They may contain perfect and progressive
aspect, which we assigned to V" in section 3.5. They may contain because and result
clauses characteristic of V" and not of N".
On the other hand, gerundive nominals do not contain tense and modal, character-
 9: DEVERBALIZING RULES

istic of V'" and absent in N'". They contain a possessive NP in the subject,
characteristic of N'", not of V'". As Schachter (1976) points out, they occur with a
limited range of articles, as in There is no enjoying this world without thee, thus
behaving like N'", not V'". Like N'" and unlike V'", they do not permit sentence
adverbial PPs; (9.4) compares gerundive nominals to t/iat-clauses to show that the
impossibility of these particular sentence adverbials is not a restriction of all subordi¬
nate clauses:

to our delight
(9.4) *John’s ' in his haste ' leaving so early didn’t distress Sue.
for some reason

to our delight
That John(,) ' in his haste (,) left so early didn't distress Sue.
for some reason

Further arguments of this sort appear in Schachter (1976) and Horn (1975), and,
though advocating a different theory, in Williams (1975). The point is clear. Gerundives
have the constituent structure of sentences up to the X" level and that of NPs above
that. Thus the appropriate skeletal structure is (9.2b), and the phrase structure rule is
(9.5).

(9.5) Gerundive Nominals

N" -> ing - V"

Ing, like other affixes, undergoes Affix Hopping to attach to the first verbal form. Thus
the underlying form of John's having left is (9.6), exactly its surface form except for the
insertion of poss and the attachment of verbal affixes.

(9.6) N

leave
224 RAY JACKENDOFF

9.3. Tensed Complements and Free Relatives

Let us turn now to tensed subordinate clauses. As Bresnan (1970) shows, these include
not only r/n/t-clauses but also comparative clauses with the complementizers as and
than, and indirect questions with the complementizer wh. All of these are indubitably
V"'s, since they contain, in addition to all the V" structure, tenses, modals, and
nongenitive subjects; what sentence adverbials they lack can plausibly be excluded for
semantic reasons. The discussion of chapter 3 showed that that-clauses, because their
movement generalizes with that of other X"', must be X'" of some sort. Furthermore, it
was shown that the complementizer is not within the Vthat dominates the subject and
the auxiliary. Thus a structure like (9.2c) appears most appropriate for tensed
complement clauses. The only question is what category the dominating X'" should be.
Chapter 4 alluded to the dispute over whether t/?at-clauses are dominated by NP in
underlying structure. We have taken the position that they are not, that their position at
the end of X' and X"is base-generated, and that their appearance in the subject position
of certain clauses results from the application of the root transformation Intraposition.
Thus we claim that the X'" in (9.2c) is V'" for tensed complement clauses, and that
there is no separate category S as Bresnan (1970) and Chomsky (1973) assume. Under
this theory, the phrase structure rule generating tensed complements is (9.7), and
typical structures are given in (9.8).1

Of course, if it were to turn out that that-clauses are dominated by NP, a possible
realization of (9.1) to describe this would be (9.9).

(9.9) N'" —» that - V'"

1 In principle this rule could incorrectly apply to its own output to yield sequences of complementizers.
I assume, perhaps unjustly, that either semantic considerations or a surface constraint prevent such an
eventuality.
 9: DEVERBALIZING RU LES

But there is another kind of tensed complement for which a dominating N"' looks a
great deal more plausible: free relatives.
Free relatives occur in all NP positions, including those in (9.3) from which that
and for-to complements are excluded:

(9.10) a. What would what the FBI found out reveal about John?
b. a man to whom what you found out would be a nuisance
c. It would be disgraceful for what you found out to be revealed.

Furthermore, free relatives, like gerundive nominals and unlike that and for-to
complements, do not have an alternate extraposed position:

What John saw

John's seeing a mouse
(9.11) a. > would surprise no one.
That John saw a mouse
For John to see a mouse
*what John saw

b. It would surprise no one

Thus free relatives behave externally much more like NPs than that and for-to
complements do.
It is generally assumed that a free relative is derived from a noun phrase with a
relative clause and some sort of dummy head, such as the thing which John saw or
PRO which John saw. However, as Jespersen (1909-1949, Vol. Ill) points out, such a
derivation would not provide an immediately plausible source for free relatives with
wh-ever, as in Whatever John saw interested me, We were interviewed by whoever we
met. Still worse are free relatives of the form what money I have, what help you can
offer him, etc., in which the wh-word is followed by a noun also fronted from the
relative clause. Underlying sources like the thing [s/ have wh-money] or PRO [syo//
can offer him wh-help] do not make sense as relative constructions—particularly if we
adopt the interpretive view of relative pronouns argued for in chapter 7.
Jespersen argues that the clause itself should be considered as the entire NP. In the
present theory, this can be expressed as the claim that free relatives are generated with
phrase structure rule (9.12), another deverbalizing rule; then what money / have is
derived from (9.13).

(9.12) Free Relatives

nt
N’" -> wh - V
226 RAY JACKENDOFF

(9.13) N'"

This hypothesis enables us to make a structural distinction between indirect

questions and free relatives, while still claiming that their great similarity is no
accident. Quite simply, free relatives are N'" dominating wh - V'", and indirect
questions are V'" dominating wh - V'". This would be the structural difference, for
example, in the well-known old minimal pair What lay on the table was the tissue (free
relative) and What lay on the table was the issue (indirect question).2
Thus if there is any construction in which an N'" immediately dominates an S, it is
much more likely to be free relatives than /7ta/-complements, since free relatives occur
in all NP positions and are most easily derived by a rule like (9.12). Hence we now
have three different applications of rule schema (9.1). (See also note 6, p. 238.)

9.4. Gerundive PPs

There are some -ing complements which, unlike most gerundives, alternate with PPs
rather than NPs:

(9.14)

b. Bill remained

There is no possibility of a subject in these -ing complements; further, like the

corresponding PPs and unlike most gerunds, they cannot cleft: *It was running that
John kept Bill. The subcategorization of the verbs in (9.14) would be simplified if in fact
these complements were not NPs but PPs. Jackendoff (1976) shows that they behave
semantically as expressions of location, so PP constituency is not semantically
implausible.

2 The latter of these examples thus has a V'" in surface subject position. 1 believe a stylistic inversion
has taken place—notice that the highest stress in this sentence is on table, in the middle of the sentence, not
on issue, the expected position. By contrast, in the free relative sentence, the highest stress is at the end, as
expected.
 9: DEVERBALIZING RULES

This class of gerunds allows certain V" complements (e.g. Moe went on doggedly
leaving for work at 6 a.m.), though others are excluded, probably for semantic reasons.
Thus if they are generated by a deverbalizing rule, they must be of structure (9.2b) or
(9.2c). If they are dominated by PP, choosing (9.2b) would automatically explain the
absence of subject, tense, and modals, since P'" contains none of these. The phrase
structure rule would be (9.15), and the structure of (9.14a) would be (9.16).

(9.15) Gerundive PPs

P" -> ing - V"
(9.16)

P"

run
Emonds (1970) suggests that the progressive be belongs to the class of verbs that
takes this sort of -ing complement. The main attraction of Chomsky’s (1957) analysis of
the progressive auxiliary was that it explained the inability of progressive to be
followed by modals, perfect, or another progressive. But Emonds shows that this may
be part of a more general set of semantic restrictions, since the verbs of (9.14) cannot
be followed by perfect or progressive either.

(9.17) a. *John kept Bill [havin8 run 1

(being running J

b. *John remained
{having worked
being working
I having sung
c. *Moe went on
[being singing
Notice also that be takes a PP semantically parallel to the progressive: compare John is
working to John is at work. Hence, however these restrictions are accounted for, they
can be generalized to progressive be if John is singing is assigned the structure (9.18).
228 RAY JACKENDOFF

(9.18)

N M V"

John pres M V'

A V

be

ing V"

Asing

As further support for this analysis, Jackendoff (1976) shows that the verb be fills a gap
in the semantic paradigm given by the verbs in (9.14), and that the interpretation of the
progressive predicted by this semantic analysis is not implausible, at least in these
cases. Hence we have some reason to consider progressive be as a main verb which
takes a gerundive PP complement.
Other than the absence of subjects and the alternation with PPs, 1 have given little
evidence that these gerunds are dominated by PP. One further argument concerns the
transformation which inverts a simple verb with the subject in the presence of a
preposed locational PP (see Emonds (1976, 37-40) for a formulation of this rule):

(9.19) Into the opera house raced Harpo.

Up the stairs ran the criminals.
On Cambridge Common stands a statue.

When preposed, certain gerunds cause such an inversion too:

(9.20) Buried here lies the producer of Gone with the Wind.
Bouncing out of the room came the garbagemen.
Screaming down the hall ran two famous linguists.

Note that we cannot appeal to a cross-categorial formulation of the inversion rule in

this case, since no preposed verbal constructions other than these particular subjectless
gerunds cause the inversion. If, however, they are dominated by PP, as in (9.16), the
inversion rule can be stated simply in terms of PP.
 9: DEVERBALIZING RULES

Another use for rule (9.15) may be to account for the gerundive constructions in
(9.21), discussed by Akmajian (1977).

(9.21) The sun rising in the east is a beautiful sight.

Akmajian argues that the sun rising in the east is not a clause, but rather that the sun is
the head of the construction. He proposes the following structure:

(9.22) NP

Unfortunately, this structure contains two configurations ruled out by our theory of
phrase structure: NP over NP, and VP dominated by some category other than S or
af-X".
In order to settle the first problem, we observe that these complements occur after
restrictive relatives (9.23a), appear with proper nouns (9.23b), and exclude appositives
(9.23c,d).

(9.23) a. The man you met falling on his face was a sad sight.
b. Bill falling on his face was a sad sight.
c. ?*Bill, who we all love, falling on his face(,) was a sad sight.
d. *Bill falling on his face, who we all love, was a sad sight.

This suggests that the gerundive may be treated as a new type of N'" complement, in
alternation with appositives.3

;l This proposal is substantively different from Akmajian’s analysis in two ways. First, in object
position, these gerundive constructions permit appositives: We saw Bill, who ire all loved, driving away.
Akmajian shows that in object position there are two distinct structures for the same string: one with the
object dominating the gerund, yielding passives like Bill driving away could not be seen by anyone, and one
with the main VP dominating the gerund, yielding passives like Bill was seen driving away. Akmajian claims
the latter is derived from the former by an extraposition rule. However, this would not predict the difference
in the possibility of appositives in subject vs. object position. As an alternative, we may claim that if the
gerundive is generated in the NP, it always excludes an appositive, but that these gerundives can also be
base-generated in VP, in which case the object is permitted an appositive. We thus reject Akmajian’s
extraposition rule in favor of use of the base. In fact, if these gerunds are PPs, as I have claimed, the
necessary phrase structure rule already exists, since it generates Bill was seen on TV, with a simple PP,
parallel to Bill was seen driving away, with a gerundive PP.
A more serious problem for the present analysis is the existence of passive gerunds of this sort, for
instance Bill being struck by lightning was a sad sight. In Akmajian’s structure the Passive can take place
under the domination of the upper NP; but in the present analysis, there is no complete NP to the left of the
gerund that could be moved there by the Passive transformation. In view of current doubts about the
existence of the Passive transformation (cf, Freidin (1975), Schachter (1976), Bresnan (1976b). for example),
this argument does not seem anywhere near as damning as it would have a few years ago. One can at least
leave the issue in abeyance instead of giving up altogether.
230 RAY JACKENDOFF

As for the second problem with (9.22), the bare VP, we observe that these
gerundives alternate with PPs of approximately the same semantic import and with the
same syntactic distribution:

(9.24) a. Gretchen at the spinning wheel is lovely.

b. A man we like so much in such a mess is a disheartening sight.
c. ?*Bill, who we all love, in such a mess(,) was a sad sight.
d. *Bill in such a mess, who we all love, was a sad sight.

This suggests that the lack of subject, tense, and modal is explained not, as Akmajian
suggests, by generating a bare VP, but rather by generating V" under P", again using
phrase structure rule (9.15). Thus our alternative to (9.22) is (9.25).4

(9.25) N'"

As a final example of phrase structure rule (9.15), consider what happens if we

move the P'" in (9.25) down one level to N". Since in N", P'" precedes relative clauses,
we should get a class of subjectless gerunds which follow N' complements and precede
restrictive relative clauses and appositives. And indeed there is such a class, the so-
called gerundive relatives:

(9.26) a. A man owning a Cadillac is not to be envied these days.

b. A father of three children begging for free time should not be sneered at.
c. The man paddling the canoe(,) who is wearing a red hat(,) is your
cousin.

Gerundive relatives are often thought to be derived from ordinary relatives by deletion

4 A more traditional transformational analysis, of course, would probably suggest the reverse: that the
PP cases are derived from the gerundives by a being deletion. Though I suspect there is such a
transformation involved in cases like John angry is a scary sight, it is not so plausible for John with a scar
like that is a sad sight, since with in this use does not occur after be.
 9: DEVERB ALIZING RULES

of the relative pronoun plus be, so-called Whiz-Deletion. (9.26a) shows that this is
impossible, since *a man who is owning a Cadillac is ungrammatical. Williams (1975)
gives a number of other arguments against Whiz-Deletion, which I will not reproduce
here.
(9.26b) shows that gerundive relatives follow N' complements; (9.26c) shows that
they precede other restrictive and nonrestrictive relatives. Hence they are in the
position of PPs under N". Since they are without subjects, tenses, and modals, and
since there is no NP position under N" after the head, it seems most plausible on all
grounds to invoke (9.15) again.
This section has shown that P" quite plausibly expands as V'' in at least three
different circumstances in the grammar. In each case the verb appears as a present
participle; in each case a subject is impossible; in each case PPs appear in identical
positions with parallel interpretations. We thus have evidence for a heretofore
unsuspected phrase structure rule, another instance of phrase structure schema (9.1).
I have said nothing about infinitival complements so far in this chapter, and
perhaps a word is in order at this point. The problems posed by infinitival complements
are in many ways like those of gerundives; they share the obligatory absence of tense
and modal, and the sometimes optional, sometimes obligatory absence of a subject. But
infinitival complements are far more widespread and varied than gerundives, and thus
less easy to categorize.
The standard approach to infinitival complements, of course, is that they are all Ss
with for-to or (following Bresnan (1972) and Chomsky (1973)) <$>-to complementizers. It
has occasionally been suggested that the “obligatory Equi” complements, such as
those with try, are generated with bare VP complements. The approach to complemen¬
tation proposed in this chapter suggests that a more enlightened use of phrase structure
rules might help sort out some of the complications of infinitivals. For example, the for-
to clauses which alternate with r/zat-clauses, such as the complements of prefer, wish,
etc., are probably best analyzed as V'", as usual. But the many infinitivals which
alternate with PPs of parallel interpretation and not with f/iaf-clauses, for example
those in (9.27), may well be expansions of PP similar to gerundive PPs.5

(9.27)

5 Faraci (1974) presents a wide range of examples of this type, suggesting a phrase structure rule PP
P - S
232 RAY JACKENDOFF

It is not clear to me, however, what sort of phrase structure rule should be used to
generate infinitivals, particularly for-to infinitivals. The fact that parallels in other
languages to full for-to complements are relatively rare, by comparison with parallels to
that-clauses, simple to-complements, and gerundives, may indicate that they are some
particularly exotic variation on the phrase structure rule schema for which there is no
other evidence in English. In any event, I leave the issue for future research.

9.5. Passive VPs

A final tentative example of a deverbalizing rule comes from Fiengo (1974), who
revives a proposal of Chomsky (1955), in which the passive verb phrase is an AP. The
passive auxiliary is present in deep structure, and the necessity of preposing the direct
object is guaranteed by semantic constraints (which, however, Fiengo does not spell
out in much detail). The deep structure of a passive is (9.28a) or (9.28b); its surface
structure is (9.28c) in the present framework.

(9.28) a.

love Evan
 9: DEVERBALIZING RULES

b.

be A"

en V"

be A"

love t
234 RAY JACKENDOFF

The t in direct object position in (9.28c) is the “trace” left behind by movement rules in
Fiengo’s theory, a pronoun which is anaphorically bound to the NP originally
occupying that position. We will not go into Fiengo’s motivation for this theory, as it
goes far beyond the scope of the present study. However, we observe that the analysis
requires a deverbalizing phrase structure rule, here represented as (9.29), in that there
is an AP without an adjectival head.

(9.29) Passive VP
A" -> en - V"

One piece of evidence for identifying the passive verb phrase as AP is that passive
verb phrases occur, with the correct meaning, after certain verbs other than be, for
example (9.30).

(9.30) a. Fred got arrested by the cops.

b. Mary got Kathy arrested by the cops.
c. George had Jerry studied by a team of shrinks.

Since these environments also permit APs (Fred got sick, Mary got Kathy sick, George
had Jerry angrier than a swarm of bees), an A'' structure like the one in (9.28) would
make the generation of these otherwise troublesome examples rather simple. Such a
theory would, of course, have to develop a semantic explanation of why passive VPs
do not occur in all adjectival environments, a problem which 1 must leave for future
research.

9.6. Final Remarks

We have shown that there is evidence for at least Five instances of rule schema (9.1) in
English; we repeat these here.

(9.5) Gerundive Nominals

N" ing - V"
(9.7) Tensed Complements
(that)
V'" as I V'"
Ithan|
\
wh /

(9.12) Free Relatives

N'" -> wh - V'"
(9.15) Gerundive PPs
P" -> ing - V"
(9.29) Passive VPs
A" -» en - V"
 9: DEVERB ALIZING RULES

Though not all of these rules have been equally well motivated, their similarity justifies
the claim that they constitute a principled class of exceptions to the Uniform Three-
Level Hypothesis.
It is interesting to notice what happens if we let / = 0 in schema (9.1), a possibility
we have not considered so far. The resulting rules are of the form X —> af - V and are
congruent with a subset of the word formation rules in the lexicon, for example those
relating derived nominals to verbs. And in fact, the historical roots of (9.5), (9.15), and
(9.29) are undoubtedly in word formation rules which still exist in the lexicon of
English. Corresponding to gerundive nominals are derived nominals of Germanic verbs,
such as building and writing. It is not difficult to imagine a historical restructuring in
which the affixing is taken to be at a higher level, thus enabling the verb transparently
contained in the noun to take some of its verbal complements. The reanalysis is from
(9.31a) to (9.31b).

(9.31) a. Deverbal Noun b. Gerundive Nominal

N'" N

build

Similarly, the past participles used in passive sentences are paralleled by past
participles used as passive adjectives such as amused, disturbed, shut, and so forth.
Again, one can imagine a reanalysis of the affix at a higher level, so that alongside the
lexical process (9.32a) arose the syntactic process (9.32b). Wasow (1977) has an
interesting discussion of syntactic tests that distinguish the two constructions from one
another.
236 RAY JACKENDOFF

Gerundive PPs are historically derived from the deverbal noun preceded by the
preposition on or at, a possible relic of which is the proform at it used to refer to these
constructions. The preposition degenerated to the prefix a-, which in turn dropped off
altogether. The transition is represented as (9.33); the structure of the intermediate step
is conjectural.
(9.33) a.

work
 9: DEVERBALIZING RULES

Now consider the implications of these changes for universal grammar. There are
word formation rules for deriving almost every lexical category from almost every
other lexical category. Is it a coincidence that the only types that have developed into
phrase structure rules are those which involve deverbal constructions? Rule schema
(9.1) claims that it is not, since it restricts the class of category-switching expansions in
syntax to those in which V is on the righthand side. For example, (9.1) predicts that we
would never find a historical change like the one from (9.34a) to (9.34b).

If there were a syntactic structure (9.34b), we would expect it to have N" modifiers
such as adjectives and quantifiers, e.g. *great fearful or *much beautiful. No such
category-switching constructions have come to my attention in which the head is other
than V. By contrast, changes in the opposite direction, changing syntactic structure
into lexical structure, do occur; for example, the adjective full developed into an affix
that converts nouns into adjectives. Thus rule schema (9.1) explains why certain
historical changes could have taken place and others could not have.
This completes our brief survey of deverbalizing rules. I consider them an
extremely interesting class of rules, putting on a principled basis a sizable number of
what previously had to be treated as altogether arbitrary phrase structure expansions.
In particular, they seem to me to provide a framework for a richer syntax of
complementation than has previously been considered, one that can stimulate a fresh
approach to the crucial problem of constraints on extraction. Moreover, as we have
just mentioned, they provide a previously unsuspected locus for historical change in
syntax.
238 RAY J ACKEN DOFF

In turn, it is only because we adopted such a highly constrained theory of phrase

structure as the Uniform Three-Level Hypothesis that the class of deverbalizing rules
has emerged as a unified whole. The fact that the exceptions to the main phrase
structure rule schema can be so clearly and systematically delimited is evidence that we
chose the correct set of constraints in the first place. Thus the discussion in this chapter
serves indirectly as an argument for the main thesis of this book, the formalization of
the X Convention in terms of syntactic features and primes.6

6 After this book went to press, it came to my attention that there are also free relatives dominated by
PP, for example the subordinate clauses in She put it where l wanted it and I’ll go whenever you do. This
implies a further deverbalizing rule:
(i) P'" wh - V'"
 10: More General Implications

10.1. Explanatory Adequacy

Chapter 3 formulated the X Convention as a theory of grammatical categories. This

theory claims that the possible lexical categories are defined by a set of syntactic
distinctive features, and that the syntactic categories are defined in turn as the
supercategories of the lexical categories by way of the bar (or prime) notation. The
theory was made more explicit by proposing the syntactic features Subj, Obj, Comp,
and Det to describe the lexical categories of English; finally, the theory of syntactic
categories was elaborated into a theory of phrase structure rules, the Uniform Three-
Level Hypothesis.
The Uniform Three-Level Hypothesis claims that every lexical category X has
exactly three supercategories X', X", and the major phrasal category X'", and that all
modifiers are either major phrasal categories or grammatical formatives. This claim was
formalized as the rule schema (3.9),

(3.9) Uniform Three-Level Hypothesis

X” —* (Cj) . . . (C,) - Xn_1 - (Cj+1) . . . (C*), where I < « < 3, and for all Q,
either Q = Y'" for some lexical category Y, or Q is a specified grammatical
formative.

We further claimed that there are only two kinds of exceptions to (3.9), coordinate
conjunction and deverbalized constructions, each constrained by a characteristic
phrase structure rule schema.

(3.32) Coordinate Conjunction Schema

X* -> Xj - (Conj - Xfi*
(9.1) Deverbalizing Rule Schema
X1' -> af - V

We claim, then, that all possible phrase structure rules of natural language are
described by these three schemata.
Starting in section 3.4 we embarked on a tour of English, concerned with showing
that its phrase structure is described by this theory. We took particular care to show
that many previous analyses of phrase structure which are not instances of the three
240 RAY JAC'KENDOFF

schemata are either empirically inadequate or replaceable by descriptively equivalent

rules which do conform to the present theory. It is now worth stepping back to try to
assess the value of this mass of particular analyses as justification of the theory.
There are three separate issues that must be dealt with in evaluating a theory.
First, does it generate enough grammars that the correct grammars are among them?
Second, does it constrain the class of grammars sufficiently to exclude various kinds of
conceivable but nonexistent rules? Third, does it characterize the notion “linguistically
significant generalization” in such a way that the language learner can arrive at the
correct grammar on the basis of fragmentary evidence?
The first issue, that of the descriptive adequacy of the theory, has been dealt with
at length in this study, through detailed investigation of the grammar of English. At
those points where there was some question as to the need to expand the class of
grammars described by the theory, we have tried to suggest the most feasible ways of
doing this. In order to evaluate the descriptive adequacy of the theory more fully, one
would have to look more seriously at other languages. A notable lack in the theory, for
example, is a mechanism to deal with free word order languages; the feature
“transportable” we introduced for the free order of English adverbs in V" and V'" is
doubtless only a makeshift.
As for the issue of constraining the class of grammars, this theory is to my
knowledge the most highly constrained theory of phrase structure rules to date (other
than the trivialized generative semantics base, which contains only the two rules S —> V
- NP - (NP) and NP —» N - (S)). Inasmuch as it excludes many well-known analyses in
the literature, forcing us to develop and defend alternatives, the constraints it imposes
are hardly self-evident or gratuitous. Of course, it is entirely possible that ways can be
found to constrain the theory more. Having been up to now preoccupied with showing
that it is not too constrained, I have not given much consideration to this problem.
The third issue, that of characterizing the notion “linguistically significant generali¬
zation”, is a knottier one. In formal terms, what is desired is an evaluation measure
within the theory, which determines which of two given grammars is more highly
valued, i.e. more “general”. The type of linguistic generalization we have been
concerned with in this study is the cross-category generalization: a term of a rule is
simpler, i.e. more general, if it is satisfied by a set of syntactic categories expressible
by fewer syntactic features. For example, a term satisfied by any of the categories N",
Q". A", Adv" would be more general than a term satisfied by any of the categories V'",
V", N", and P'", because the former could be expressed with two features, whereas the
latter would take four. It would also be simpler than a term satisfied by N" alone, which
would take three features. This sort of feature-counting is precisely parallel to that in
phonology.
Though the notion of embodying the evaluation measure in a feature-counting
mechanism is perfectly straightforward when dealing with individual terms of rules, as
demonstrated by our success in stating generalized grammatical relations, it has proven
a great deal more problematic in dealing with entire rules. As shown in the discussion
 10: MORE GENERAL IMPLICATIONS

of section 4.5, the generalizations among phrase structure rules are characteristically
only partial, giving rise to several sorts of situations in which the notational conven¬
tions traditionally incorporated in the evaluation measure fail to appropriately reduce
the number of features in the rule. Since the evaluation measure as a simple feature¬
counting device depends on the existence of appropriate abbreviatory conventions,
there are only two solutions to the problem: either develop new abbreviatory
conventions, or develop an evaluation measure that does something more than just
count features. We have shown that one of these alternatives is necessary in the theory
of the phrase structure component.
It may not be out of place here to point out that Jackendoff (1975a), discussing the
nature of lexical generalizations, shows that they too resist analysis in terms of a
feature-counting evaluation measure. A different sort of evaluation measure is pro¬
posed there, better adapted for the partial generalization endemic to the lexicon:
instead of attempting to collapse related lexical entries into a single more general entry,
it directly measures the redundancy between fully specified entries. A parallel approach
in the phrase structure component would use the feature and prime system not to
collapse rules but to enable the evaluation measure to judge, given two or more rules
with all categories fully specified, how much alike they are, i.e. how easy it is to learn
one given the other. Whether such an approach would be feasible I leave for future
research: it has been shown, in any event, that the usual approach is inadequate.
Of course, the evaluation measure (at least as presently conceived of) would value
most highly a grammar in which all categories of the same level had exactly the same
phrase structure expansion, aside from the head. Such a grammar seems hardly likely,
and there are probably many substantive universals which prevent this ideal from being
achieved. For example, we have good reason to expect M'" only under V'" and Art'"
only under N'", and these very specific restrictions are undoubtedly part of universal
grammar. For a more controversial case, observe that the grammar of English would be
simplified if nouns and adjectives came to be followed directly by NP, as verbs and
prepositions are. Yet this change seems highly unlikely. One possible explanation is
that universal grammar places such general restrictions on phrase structure expansions
as prohibiting surface objects in N' and A', prohibiting subjects in A'" and P'", and
generally excluding all X' modifiers from the minor lexical categories. (This would
amount to the claim that my choice of feature names has theoretical significance.)
Whether or not these specific restrictions are correct, they are indicative of the way in
which substantive universals could in principle prevent the phrase structure component
from achieving maximal generality.
Having discussed the implications of this study for the syntactic component, we
will conclude by turning to its impact on semantics.

10.2. The Challenge for Semantics

Throughout this study I have made strong assumptions about the capabilities of the
semantic component of the grammar. Though many readers will no doubt see these
242 RAY JACKENDOFF

assumptions as “sweeping difficulties under the rug”, my sense of the situation is that,
given an apparently arbitrary restriction on syntax, the account of it that puts the least
burden on the language learner is that it follows from independently motivated rules
and restrictions on semantic interpretation; it can then be claimed that knowing the
meaning of the construction in question automatically results in the application of the
observed constraint. If instead the constraint has to be added to the syntax, either
through universal grammar, or worse yet, in the language-particular syntax, no real
explanation has been provided. It seems to me that the success of this methodology has
been amply proven in Jackendoff (1972; 1974b; 1975c; 1976) and need not be further
defended here.
In the present study, however, I have been somewhat lax about actually proposing
semantic solutions to syntactic constraints. In a few cases, such as the binding of
restrictive relative clauses to their articles and the interpretation of relative pronouns
and the PRO in degree clauses, I have given explicit solutions. But the cases left open,
such as the Partitive Constraint, pseudopartitive determiner constraints, the binding of
degree clauses to their degree words, and the nonrecursion of the complementizer
node, have been far more numerous. Thus the present study leaves unsolved a great
number of particular problems, some of which are demonstrably semantic, some of
which have been called semantic admittedly on the basis of little more than hope.
More generally, we have explored a wide range of syntactic constructions whose
semantics have never been seriously studied. And though for the most part we have
mentioned the semantics of these many constructions only peripherally, it is clear that
they provide a rich field of inquiry for natural logic. For example, consider the range of
modifiers within the NP. At the N' level there are functional arguments, which are well
understood by logic. At the N" level there are restrictive modifiers (generally treated in
logic as conjoined predications or restrictions on quantifiers), the quantifiers many and
few (not among the quantifiers traditionally treated by logic), and numerals, measure
phrases, and pseudopartitive group nouns (outside the province of traditional logic). In
N'" there are subjects (again functional arguments), genitive NPs of intrinsic connection
(not treated by logic), demonstratives and N'" quantifiers (only partially within
traditional logic), appositives (again treated as conjoined predication), and gerundive
PPs (semantics unknown). The PP, which has in the past been treated with equal
disdain by syntax and semantics, has proved to have an extremely rich syntax,
including a characteristic system of quantification and degree modification; this syntax
implies the existence of a corresponding semantics, up to now totally unexplored. In
order for the semantic component of language to achieve even observational adequacy,
it is clear that there is a vast range of phenomena yet to account for.
But of course the syntactic investigation here was not directed toward merely the
complexity of phrase structure but toward uncovering the richness of cross-categorial
generalization in phrase structure and in grammatical relations. Since grammatical
relations, as the environments over which selectional restrictions are defined, are in
fact the structural descriptions of rules of semantic interpretation, cross-categorial
 10: MORE GENERAL IMPLICATIONS

syntactic generalization automatically implies cross-categorial semantic generalization

as well. Each of the major lexical categories gives rise to functional arguments, to
measure and degree modification, to quantification, and to restrictive and nonrestric-
tive modification. Many of the minor lexical categories take some of these modifiers
too. In one particularly extreme example pointed out in chapter 6, it was observed that
the degree expression almost can serve as a modifier to seven different categories.
Such cross-categorial semantic generalizations have never to my knowledge played a
role in any semantic theory.
As I see it, such generalizations have profound consequences for semantic theory.
Under the usual assumption that the aspects of meaning represented in deep structure
are compositionally determined from the structure, each X'" in a sentence must be
provided with an interpretation into which all of the modifiers of that X'" have been
integrated. The fact that the same grammatical relations appear in different categories
suggests that the interpretations of the categories have internal structure that is parallel
in important ways. On the other hand, the differences in possibilities among the
categories argue that the similarities are not to be captured by reducing all categories
semantically to predicates, as the traditional logical program would have it.
The argument I have just given parallels the syntactic arguments of chapter 2,
aimed at showing that various categories are syntactically distinct and not derived one
from the other. And I think the conclusions are parallel as well: alongside a lexicalist
syntax there must be a lexicalist semantics, which takes seriously the categorial
pluralism of language. Such a semantic theory would take as a basic assumption that
language has such a variety of syntactic devices at its disposal in part because of the
variety of semantic entities and relationships it has to express. And where syntactic
relationships are parallel or nonparallel across categories, that may often be a reflection
of parallelisms or nonparallelisms in semantic structure.
To be sure, we should not expect point-by-point parallelism between syntax and
semantics: we have seen cases both where a single syntactic position served two
semantic purposes and where the same thing was expressed with two different
syntactic constructions. But to me it is inconceivable that something as syntactically
rich as natural language serves the purpose of expressing something as ontologically
impoverished as any present-day form of logic.
Given the relationships between syntax and semantics that have been discussed in
this study, a plausible initial hypothesis emerges as to the proper strength of their
parallelism in universal grammar. A primary claim is that each major phrasal category
(X'") corresponds to a semantic (or ontological) category. In the simplest cases, NPs
correspond to a semantic category which essentially picks out things; Ss to a category
which picks out events and states of affairs; APs to properties (not predicates); and PPs
to places (i.e. spatial relationships to things). Each of these major semantic categories
has an internal structure whose possibilities are provided by universal grammar; the
constituents of these internal structures are the interpretations of the grammatical
relations possible within the major phrasal categories.
244 RAY JACKENDOFF

What one might guess not to be universal about the correspondence is exactly how
the phrases expressing internal constituents of the major semantic categories are
distributed syntactically within the major phrasal categories. For instance, we might
expect to find modifiers of the same semantic type occurring within X" in one case and
within X'" in another. We have found at least two such examples in English alone: the
occurrence of semantic quantifiers at both the N'" and N" levels, and the alternation of
nonrestrictive relative clauses in N'" with nonrestrictive adjective phrases in N". It
would not be surprising to find much greater variability across languages in this
respect.
On the other hand, one might expect the correspondences within each X'" not to
be totally arbitrary. For example, nonrestrictive modifiers probably cannot be more
deeply embedded in X"' than restrictive modifiers, and functional arguments other than
the subject quite possibly must occur in X'. Among the tasks of cross-linguistic
investigation would be to discover exactly how constrained these relationships are, and
to specify how extensive cross-categorial generalizations are both in universal grammar
and, where universal grammar leaves the question open, within particular languages.
Such a program of lexicalist semantics, then, would conceive of the hierarchical
aspect of rules of phrase structure as being to a significant extent dictated by semantic
considerations. However great or small the relationship actually turns out to be, the
result is of interest. For to whatever extent the form of the syntax is not determined by
semantics, this is evidence for Chomsky’s (1975) contention that the language faculty is
an autonomous mental organ whose structure is as different from other mental organs
as the heart is different from the liver. On the other hand, the closer that syntactic form
is related to semantic form in particular or universal grammar, the more it is possible to
use purely syntactic arguments as a basis for hypotheses about semantic structure.
As may be gathered from the preceding remarks, I believe that the relationship
between syntax and semantics is a strong one, and that it can and should be exploited
in order to counteract the irresponsibility toward syntactic form that is characteristic of
most present-day work in logic. Nor should this relationship be used, as many linguists
have used it, to bring evidence from a mathematically conceived formal semantics to
bear on syntax. Rather, the syntax itself provides precise and highly differentiated data;
a lexicalist semantics would use this evidence to shed light on the much murkier
question of how natural logic is structured. I consider Jackendoff (1975c; 1976) as
examples of such an approach, using syntactic paradigms to motivate novel semantic
analyses of some generality.
Since semantic representation is presumably closely related to cognitive structure,
such a program of research has a further consequence. To the extent that a lexicalist
semantics proves successful, the study of language provides an unparalleled tool for
investigating not just a single organ of the human mind, but perhaps some of the most
fundamental principles of mental organization. The potential implications for psychol¬
ogy are vast. It is hoped that the present study provides some impetus toward their
discovery.
 References

Akmajian, A. (1973) "The Role of Focus in the Interpretation of Anaphoric Expressions,” in S.

Anderson and P. Kiparsky, eds. (1973), 215-225.
Akmajian, A. (1975) "More Evidence for the NP Cycle,” Linguistic Inquiry 6.1, 1 15-129.
Akmajian, A. (1977) "The Complement Structure of Perception Verbs in an Autonomous
Syntactic Framework,” in A. Akmajian, P. Culicover, and T. Wasow, eds. (1977).
Akmajian, A., P. Culicover, and T. Wasow, eds. (1977) Studies in Formal Syntax, Academic
Press, New York.
Akmajian, A. and A. Lehrer (1976) "NP-like Quantifiers and the Problem of Determining the
Head of an NP,” Linguistic Analysis 2.4. 395-413.
Akmajian, A. and T. Wasow (1975) "The Constituent Structure of VP and AUX and the Position
of the Verb BE,” Linguistic Analysis 1.3, 205-245.
Anderson. S. and P. Kiparsky, eds. (1973) A Festschrift for Morris Halle, Holt. Rinehart, and
Winston, New York.
Aronoff, M. (1974) Word-Structure, unpublished Doctoral dissertation, MIT, Cambridge, Massa¬
chusetts.
Aronoff, M. (1976) Word Formation in Generative Grammar, Linguistic Inquiry Monograph No.
1, MIT Press, Cambridge, Massachusetts.
Bach, E. and R. T. Harms, eds. (1968) Universals of Linguistic Theory, Holt, Rinehart, and
Winston, New York.
Banfield, A. (1973) "Narrative Style and the Grammar of Direct and Indirect Speech,”
Foundations of Language 10, 1-39.
Berman, A. (1974) Adjectives and Adjective Complement Constructions, Doctoral dissertation.
Harvard University, Cambridge, Massachusetts; published as Formal Linguistics Report
NSF-29, Susumu Kuno, principal investigator, Harvard University Department of Lin¬
guistics.
Bierwisch. M. and K. E. Heidolph, eds. (1970) Progress in Linguistics, Mouton, The Hague.
Bowers, J. S. (1968a) "Some Adjectival Nominalizations in English,” unpublished mimeograph,
MIT. Cambridge, Massachusetts; Lingua 37, 341-361 (1975).
Bowers. J. S. (1968b) "Adjectives and Adverbs in English,” Indiana University Linguistics Club.
Bloomington; Foundations of Language 13, 529-562 (1975).
Brame, M. (1968) "A New Analysis of the Relative Clause: Evidence for an Interpretive
Theory,” unpublished paper, MIT, Cambridge, Massachusetts.
Brame, M. (1975) "On the Abstractness of Syntactic Structure: The VP Controversy,”
Linguistic Analysis 1.2, 191-203.
Bresnan, J. W. (1970) "On Complementizers: Towards a Syntactic Theory of Complement
Types,” Foundations of Language 6, 297-321.
Bresnan, J. W. (1971) "Sentence Stress and Syntactic Transformations,” Language 47, 257-281.
Bresnan, J. W. (1972) The Theory of Complementation in English Syntax, unpublished Doctoral
dissertation, MIT, Cambridge, Massachusetts.
246 RAY JACKENDOFF

Bresnan, J. W. (1973) “Syntax of the Comparative Clause Construction in English,” Linguistic

Inquiry 4.3. 275-344.
Bresnan, J. W. (1974) “The Position of Certain Clause-Particles in Phrase Structure,” Linguistic
Inquiry 5.4, 614-619
Bresnan, J. W (1976a) “On the Form and Interpretation of Syntactic Transformations,”
Linguistic Inquiry 7.1, 3-40
Bresnan, J. W. (1976b) “Toward a Realistic Theory of Transformational Grammar,” paper
presented at AT&T conference on language at MfT, March 1976.
Chomsky, N. (1955) The Logical Structure of Linguistic Theory, Plenum, New York (1975).
Chomsky, N. (1957) Syntactic Structures, Mouton, The Hague.
Chomsky, N. (1965) Aspects of the Theory of Syntax, MIT Press, Cambridge, Massachusetts.
Chomsky, N. (1970a) “Remarks on Nominalization,” in R. Jacobs and P Rosenbaum, eds.
(1970), 184-221; in N. Chomsky (1972), 11-61.
Chomsky, N. (1970b) “Deep Structure, Surface Structure, and Semantic Interpretation,” in D.
Steinberg and L. Jakobovits, eds. (1971), 183-216; in N. Chomsky (1972), 62-119.
Chomsky, N. (1972) Studies on Semantics in Generative Grammar. Mouton, The Hague.
Chomsky, N. (1973) “Conditions on Transformations,” in S. Anderson and P. Kiparsky, eds.
(1973), 232-286.
Chomsky. N. (1975) Reflections on Language, Pantheon. New York.
Chomsky, N. and M. Halle (1968) The Sound Pattern of English, Harper & Row, New York.
Dean, J. (1966) “Determiners and Relative Clauses,” unpublished mimeograph, MIT, Cam¬
bridge, Massachusetts.
Dineen, F., ed. (1966) 19th Monograph on Languages and Linguistics, Georgetown University
Press, Washington, D.C.
Dixon, R. M. W., ed. (1975) Grammatical Categories, Australian Institute of Aboriginal Studies.
Dougherty, R. C. (1968) A Transformational Grammar of Coordinate Conjoined Structures,
unpublished Doctoral dissertation, MIT, Cambridge, Massachusetts.
Dougherty, R. C. (1970-1971) “A Grammar of Coordinate Conjoined Structures,” Language
46.4. 850-898, and 47.2, 298-339.
Emonds, J. E. (1970) Root and Structure-Preserving Transformations, Doctoral dissertation,
MIT, Cambridge, Massachusetts; published by Indiana University Linguistics Club,
Bloomington, Indiana
Emonds, J. E. (1972a) “Evidence that Indirect Object Movement Is a Structure-Preserving
Rule,” Foundations of Language 8, 546-561.
Emonds, J. F. (1972b) “A Reformulation of Certain Syntactic Transformations,” in P. S. Peters,
ed. (1972).
Emonds, J. E. (1976) A Transformational Approach to English Syntax, Academic Press, New
York.
Faraci, R. (1974) Aspects of the Grammar of Infinitives and For-Phrases, unpublished Doctoral
dissertation, MIT, Cambridge, Massachusetts.
Fiengo, R. (1974) Semantic Conditions on Surface Structure, unpublished Doctoral dissertation,
MIT, Cambridge, Massachusetts.
Fillmore, C. (1968) “A Case for Case,” in E. Bach and R. T. Harms, eds. (1968), 1-88.
Fillmore, C. and D. T. Langendoen. eds. (1971) Studies in Linguistic Semantics Holt, Rinehart,
and Winston, New York.
Fodor, J. A. and J. J. Katz, eds. (1964) The Structure of Language, Prentice-Hall, Englewood
Cliffs, New Jersey.
Freidin, R (1975) “The Analysis of Passives.” Language 51.2, 384-405.
 REFERENCES

Greenberg, J. (1963) “Some Universals of Grammar with Particular Reference to the Order of
Meaningful Elements,” in J. Greenberg, ed. (1963).
Greenberg, J., ed (1963) Universals of Language, MIT Press, Cambridge, Massachusetts.
Gruber, J. S. (1965) Studies in Lexical Relations, Doctoral dissertation, MIT, Cambridge,
Massachusetts; published by Indiana University Linguistics Club, Bloomington, Indiana.
Hale, K. (1975) “The Adjoined Relative Clause in Australia,” in R. M. W. Dixon, ed. (1975).
Halle, M. (1973) “Prolegomena to a Theory of Word Formation,” Linguistic Inquiry 4.1, 3-16.
Harris, Z. (1946) “From Morpheme to Utterance,” Language 22, 161-183.
Harris, Z. (1951) Methods in Structural Linguistics, University of Chicago Press, Chicago,
Illinois.
Higgins, F. R. (1973) “On J. Emonds’s Analysis of Extraposition,” in J. Kimball, ed. (1973).
Horn. G. (1975) “On the Nonsentential Nature of the Poss-Ing Construction,” Linguistic
Analysis 1.4, 333-388.
Hornstein, N. (1975) “S and the X Convention.” Montreal Working Papers in Linguistics, Vol.
4, McGill University Department of Linguistics, Montreal.
Horvath, J. (1976) “Focus in Hungarian and the X Notation,” Linguistic Analysis 2.2, 175-197.
Jackendoff, R. (1968) “Quantifiers in English,” Foundations of Language 4.4, 422-442.
Jackendoff. R. (1969a) “An Interpretive Theory of Negation,” Foundations of Language 5.2,
218-241.
Jackendoff, R. (1969b) “Les constructions possessives en anglais,” Langages 14, 7-27.
Jackendoff, R. (1971) “Gapping and Related Rules,” Linguistic Inquiry 2.1, 21-36.
Jackendoff, R. (1972) Semantic Interpretation in Generative Grammar, MIT Press, Cambridge,
Massachusetts.
Jackendoff, R. (1973) “The Base Rules for Prepositional Phrases,” in S. Anderson and P.
Kiparsky, eds.(1973), 345-356.
Jackendoff, R. (1974a) “Introduction to the X Convention,” published by Indiana University
Linguistics Club, Bloomington, Indiana.
Jackendoff, R. (1974b) “A Deep Structure Projection Rule,” Linguistic Inquiry 5.4, 481-506.
Jackendoff, R. (1975a) “Morphological and Semantic Regularities in the Lexicon,” Language
51.3, 639-671.
Jackendoff, R. (1975b) “Tough and the Trace Theory of Movement Rules,” Linguistic Inquiry
6.3, 437-446.
Jackendoff, R. (1975c) “On Belief-Contexts,” Linguistic Inquiry 6.1, 53-93.
Jackendoff, R. (1976) “Toward an Explanatory Semantic Representation,” Linguistic Inquiry
7.1, 89-150.
Jackendoff, R. and P. Culicover (1971) “A Reanalysis of Dative Movements,” Foundations of
Language 7.3, 397-412.
Jacobs, R. and P. Rosenbaum, eds. (1970) Readings in English Transformational Grammar,
Ginn, Waltham, Massachusetts.
Jespersen, O. (1909-1949) A Modern English Grammar, George Allen & Unwin, London.
Katz, J. J. (1966) The Philosophy of Language, Harper & Row, New York.
Katz, J. J. (1972) Semantic Theory, Harper & Row, New York.
Keyser, S. J. (1968) Review of S. Jacobson, Adverbial Positions in English. Language 44.2, 357—
373.
Kimball, J., ed. (1973) Syntax and Semantics, Vol. 2, Academic Press, New York.
Kimball, J., ed. (1975) Syntax and Semantics, Vol. 4, Academic Press, New York.
Klima, E. S. (1964) “Negation in English,” in J. A. Fodor and J. J. Katz, eds. (1964), 246-323.
Klima, E. S. (1965) Studies in Diachronic Syntax, unpublished Doctoral dissertation, Harvard
University, Cambridge, Massachusetts.
248 RAY JACKEN DOFF

Kuroda, S.-Y. (1969) “English Relativization and Certain Related Problems,” in D. Reibel and S.
Schane, eds. (1969), 264-287.
Lakoff, G. (1970a) "Global Rules,” Language 46.3, 627-639.
Lakoff, G. (1970b) “Repartee, or a Reply to ‘Negation, Conjunction, and Quantifiers’,”
Foundations of Language 6.3, 389-422.
Lakoff, G. (1971a) “On Generative Semantics,” in D. Steinberg and L. Jakobovits, eds. (1971),
232-296.
Lakoff, G. (1971b) On Syntactic Irregularity, Holt, Rinehart and Winston, New York.
Lakoff, G. and J. R. Ross (1966) “A Criterion for Verb Phrase Constituency,” in Mathematical
Linguistics and Automatic Translation, Report NSF-17, Harvard Computation Labora¬
tory, Harvard University, Cambridge, Massachusetts.
Lasnik, H. and R. Fiengo (1974) “Complement Object Deletion,” Linguistic Inquiry 5.4, 535—
572.
Lees, R. B. (1961) "Grammatical Analysis of the English Comparative Construction,” Word 17,
171-185; in D. Reibel and S. Schane, eds. (1969).
Lightfoot, D. (1974) “The Diachronic Analysis of English Modals,” Montreal Working Papers in
Linguistics Vol. 3, McGill University Department of Linguistics, Montreal, 115-145.
McCawley, J. D. (1971) “Where do Noun Phrases Come From?” in D. Steinberg and L.
Jakobovits (1971), 217-231.
McCawley, J. D. (1975) “The Category Status of English Modals,” Foundations of Language 12,
597-601.
Milner, J.-C. (1973) “Comparatives et relatives,” in J.-C. Milner, ed. (1973).
Milner, J.-C., ed. (1973) Arguments Linguistiques, Maison Marne, Paris.
O’Neil, W. (1976) “Clause Adjunction in Old English,” ms., to appear in General Linguistics.
Perlmutter, D. (1970) “On the Article in English,” in M. Bierwisch and K. E. Heidolph, eds.
(1970).
Peters, P. S., ed. (1972) Goals of Linguistic Theory, Prentice-Hall, Englewood Cliffs, New
Jersey.
Postal, P. M. (1966) “On So-Called ‘Pronouns’ in English,” in F. Dinneen, ed. (1966); in D.
Reibel and S. Schane, eds. (1969), 201-224.
Postal, P. M. (1971) Crossover Phenomena, Holt, Rinehart and Winston, New York.
Postal, P. M. (1974) “On Certain Ambiguities,” Linguistic Inquiry 5.3, 367-424.
Rardin, R. (1975) Studies in Derivational Morphology, unpublished Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Reed, A. (1974) The Structure of English Relative Clauses, unpublished dissertation, Brandeis
University, Waltham, Massachusetts.
Reibel, D. and S. Schane, eds. (1969) Modern Studies in English, Prentice-Hall, Englewood
Cliffs, New Jersey.
Rosenbaum, P. (1967) The Grammar of English Predicate Complement Constructions, MIT
Press, Cambridge, Massachusetts.
Ross, J. R. (1967) Constraints on Variables in Syntax, Doctoral dissertation, MIT, Cambridge,
Massachusetts; published by Indiana University Linguistics Club, Bloomington, Indiana.
Ross, J. R. (1969) “Adjectives as Noun Phrases,” in D. Reibel and S. Schane, eds. (1969), 352—
360.
Schachter, P. (1976) “A Nontransformational Account of Gerundive Nominals in English,”
Linguistic Inquiry 7.2, 205-241.
Selkirk, E. O. (1970) “On the Determiner Systems of Noun Phrase and Adjective Phrase,”
unpublished paper, MIT, Cambridge, Massachusetts.
 REFERENCES

Selkirk, E. O. (1972) The Phrase Phonology of English and French, unpublished Doctoral
dissertation, MIT, Cambridge, Massachusetts.
Selkirk, E. O. (1974) “French Liaison and the X Convention,” Linguistic Inquiry 5.4, 573-590.
Selkirk, E. O. (1977) “Some Remarks on Noun Phrase Structure,” in A. Akmajian, P. Culicover,
and T. Wasow, eds. (1977).
Siegel, D. (1974) Topics in English Morphology, unpublished Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Smith, C. (1964) “Determiners and Relative Clauses in a Generative Grammar of English,”
Language 40.1, 37-52; in D. Reibel and S. Schane, eds. (1969), 247-263.
Sommerstein, A. (1972) “On the So-Called Definite Article in English,” Linguistic Inquiry 3.2,
197-210.
Steinberg, D. and L. Jakobovits, eds. (1971) Semantics: An Interdisciplinary Reader, Cambridge
University Press, New York.
Stockwell, R., P. Schachter, and B. Partee (1973) The Major Syntactic Structures of English,
Holt, Rinehart, and Winston, New York.
Thompson, S. A. (1971) “The Deep Structure of Relative Clauses,” in C. Fillmore and D. T.
Langendoen, eds. (1971), 76-96.
Van Riemsdijk, H. (1973) “The Dutch P-Soup,” unpublished mimeograph, University of
Amsterdam.
Van Riemsdijk, H. (1976) “Extraction from Prepositional Phrases and the Head Constraint,”
unpublished mimeograph. University of Amsterdam.
Vergnaud, J.-R. (1974) French Relative Clauses, unpublished Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Wald, J. D. (1976) Stuff and Words: A Semantic and Linguistic Analysis of Non-Singular
Reference, unpublished Doctoral dissertation, Brandeis University, Waltham, Massachu¬
setts.
Wasow, T. (1971) Anaphoric Relations in English, unpublished Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Wasow, T. (1975) “A Note on Empty Nodes,” Foundations of Language 12, 603-607.
Wasow, T. (1976) Anaphora in Generative Grammar, Story-Scientia, Ghent.
Wasow, T. (1977) “Transformations and the Lexicon,” in A. Akmajian, P. Culicover, and T.
Wasow, eds. (1977).
Wasow, T. and T. Roeper “On the Subject of Gerunds,” Foundations of Language 8.1,44-61.
Williams, E. (1975) “Small Clauses in English,” in J. Kimball, ed. (1975), 249-273.
 1*jU? iXV* -

-br¬
other recent titles in the series
Linguistic Inquiry Monographs

-->**•* . _ v V
» ‘ £»*~ •» * »■* >“♦. -w ♦v ^

v*

Recent Transformational Studies in practically none of it is in English. The results

European Languages obtained in the monograph, especially with re¬
edited by Samuel Jay Keyser gard to the vowel-zero alternations, hold not
This is the third publication in the Linguistic only for Polish but with some straightforward
Inquiry Monograph Series, which presents modifications can be shown to hold for all
new and original research beyond the scope of other Slavic languages as well.
the article format.
Contents: "On a Morphologically Governed An Encyclopedia of AUX:
Vowel Alternation in French” by Francois C. A Study in Cross-Linguistic Equivalence
Dell, CNRS, Paris, and Elisabeth O. Selkirk, by Susan Steele,
University of Massachusetts at Amherst: with Adrian Akmajian, Richard Demers,
Why Subject Sentences Don’t Exist” by Jan Eloise Jelinek, Chisato Kitagawa, Richard
Koster, University of Amsterdam; “ Trace Oehrle, and Thomas Wasow
Theory and French Syntax” by Jean-Yves This significant contribution to the generative
Pollock, Universite de Paris XII et VII; “A theory of syntax confronts a fundamental
Restructuring Rule in Italian Syntax” by Luigi question inherent in the study of univers&ls of
Rizzi, Scuola Normale Superiore, Pisa; "Re¬ grammar: How can claims of universality be
sult Clauses and Conditions on Rules” by empirically verified? This book puts the ques¬
Alain Rouveret, Universite de Paris VIII; "On tion of the similarity of categories across the
the Diagnosis ofWh Movement” by Henk van grammars of particular languages on a substan¬
Riemsdijk, University of Amsterdam. tive empirical basis by providing a principled
method of cross-linguistic comparison, a
Studies in Abstract Phonology method based on equivalence classes of lan¬
by Edmund Gussmann guage-particular properties relative to a univer¬
The issue of abstractness is currently one of sally applicable definition..While the critical is¬
the central questions in the ongoing research sue is general to a consideration of any
into the theoretical foundations of modem syntactic category, the framework proposed is
phonology. Linguists are divided on the degree applied specifically to a category called AUX.
of abstractness admissible in phonological rep¬ A considerable body of evidence is introduced
resentation. While some would impose far- for both the framework and the category
reaching restrictions on admissible abstract¬ through the detailed examination of a number
ness, others have felt that such restrictions are of languages—Lummi, Egyptian Arabic, Japa¬
unrealistic. This monograph offers extensive nese, Luiseiio, German, and English.
arguments for admitting a high degree of ab¬
stractness in phonology. The MIT Press
The empirical material on which the mono¬ Massachusetts Institute of Technology
graph is based is the phonology of modem Pol¬ Cambridge, Massachusetts 02142
ish. A substantial portion of the phonological
rules of the language are studied and their the¬
oretical implications are investigated in detail.
Although there exists a rich literature on Polish
phonology, very little of it has been written
with clear theoretical issues in mind, and
JACXP