Coordination in Syntax CAMBRIDGE

This page intentionally left blank
C O O R D I NAT I O N I N S Y N TA X
Coordination in syntax is an important part of the analysis of sentence struc-

ture. Niina Ning Zhang addresses the issues raised by coordinate pairings
and the implications of these structures, looking in particular at examples
within English and Chinese. The volume covers the major questions regard-
ing coordinates in syntax, providing a fresh perspective to arguments raised
within previous literature. She explains how such coordinate complexes are
structured, how some coordinators can be combined with conjuncts of vari-
ous parts of speech, the fixed nature of some of these pairings and what
changes exist between the coordinate and non-coordinate constructions. The
theories raised are backed up by a rich variety of examples as well as provid-
ing a crosslinguistic perspective, contextualizing these ideas within current
syntactic research.
niina ning zhang is Associate Professor and Director in the Graduate

Institute of Linguistics at National Chung Cheng University. She previously
co-edited Ellipsis in Conjuction (2000) and has written many journal articles
on syntactic theory and practice.
In this series
78. henry smith : Restrictiveness in case theory
79. d. robert ladd: Intonational phonology
80. andrea moro : The raising of predicates: predicative noun phrases and the theory
of clause structure
81. roger lass: Historical linguistics and language change
82. j o h n m . a n d er s o n : A notional theory of syntactic categories
83. b e r n d h e i n e : Possession: cognitive sources, forces and grammaticalization
84. nomi erteschik-shir : The dynamics of focus structure
85. john coleman: Phonological representations: their names, forms and powers
86. c h r i s t i n a y . b e t h i n : Slavic prosody: language change and phonological theory
87. barbara dancygier: Conditionals and prediction
88. claire lefebvre: Creole genesis and the acquisition of grammar: the case of
Haitian creole
89. h e i n z g i e g e ri c h : Lexical strata in English
90. keren rice : Morpheme order and semantic scope
91. a p ri l m c m a h o n : Lexical phonology and the history of English
92. matthew y. chen : Tone Sandhi: patterns across Chinese dialects
93. gregory t. stump: Inflectional morphology: a theory of paradigm structure
94. joan bybee: Phonology and language use
95. l a u r i e b a u e r : Morphological productivity
96. t h o m a s e r n s t : The syntax of adjuncts
97. elizabeth closs traugott and r i c h a r d b . d a s h e r: Regularity in
semantic change
98. m a y a h i c k m a n n : Children’s discourse: person, space and time across languages
99. diane blakemore: Relevance and linguistic meaning: the semantics and
pragmatics of discourse markers
100. i a n ro berts and a n n a r o u s s o u: Syntactic change: a minimalist approach to
grammaticalization
101. d o n k a mi n k o v a : Alliteration and sound change in early English
102. ma rk c. ba k er : Lexical categories: verbs, nouns and adjectives
103. ca rlo ta s . s mi th : Modes of discourse: the local structure of texts
104. ro ch elle lieber: Morphology and lexical semantics
105. h o lg er d i es s el : The acquisition of complex sentences
106. s h a ro n i n k ela s and cheryl zoll: Reduplication: doubling in morphology
107. s u s a n ed w a rd s: Fluent aphasia
108. ba rba ra d a n cy g i er and eve sweetser: Mental spaces in grammar:
conditional constructions
109. h ew ba erma n , d u n s t a n b r o w n and greville g. corbett: The
syntax–morphology interface: a study of syncretism
110. ma rcu s to ma li n : Linguistics and the formal sciences: the origins of generative
grammar
111. s a mu el d . ep s tei n and t. daniel seely : Derivations in minimalism
112. p a u l d e la cy : Markedness: reduction and preservation in phonology
113. y eh u d a n . f a lk : Subjects and their properties
114. p . h . ma tth ew s : Syntactic relations: a critical survey
115. ma rk c. ba k er : The syntax of agreement and concord
116. g i lli a n ca tri o n a ra mch a n d : Verb meaning and the lexicon: a first phase
syntax
117. p i eter muysken : Functional categories
118. ju a n u ri a g ereka: Syntactic anchors: on semantic structuring
119. d . ro bert ladd: Intonational phonology, second edition
120. leo n a rd h . ba bby : The syntax of argument structure
121. b. ela n d res h er : The contrastive hierarchy in phonology
122. d a v i d a d g e r, d a n i e l h a r b o u r and laurel j. watkins : Mirrors and
microparameters: phrase structure beyond free word order
123. n i i n a n i n g z h a n g : Coordination in syntax
Earlier issues not listed are also available
CAMBRIDGE STUDIES IN LINGUISTICS
General Editors: p. austin, j. bresnan, b. comrie,
s. crain, w. dressler, c. j. ewen, r. lass,
d. lightfoot, k. rice, i. roberts,
s. romaine, n. v. smith
Coordination in Syntax
COORDINATION
IN SYNTAX
N I I NA N I N G Z H A N G
National Chung Cheng University
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
São Paulo, Delhi, Dubai, Tokyo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521767552
© Niina Ning Zhang 2009
This publication is in copyright. Subject to statutory exception and to the

provision of relevant collective licensing agreements, no reproduction of any part
may take place without the written permission of Cambridge University Press.
First published in print format 2009
ISBN-13 978-0-511-77000-5 eBook (NetLibrary)

ISBN-13 978-0-521-76755-2 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy

of urls for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
This book is dedicated to my late parents
Zhang Guangbi ddd and Fang Aiqi ddd.
Contents
Acknowledgments page xvi

Abbreviations xviii
1 Introduction 1
part i no special syntactic conf iguration 7
2 The complementation structure of coordinate complexes 9

2.1 Introduction 9
2.2 The binary-branching constituency of coordinate complexes 10
2.2.1 The asymmetry between conjuncts in binding 11
2.2.2 The asymmetry between conjuncts in possessee
pronominalization 12
2.2.3 The asymmetry between conjuncts in hosting coordinators 13
2.2.4 The asymmetry between conjuncts in coordinator floating 14
2.2.5 Conclusion and Dik’s challenges 16
2.3 The complementation structure of coordinate complexes 19
2.3.1 The dubious status of agreement in the syntax of
coordination 21
2.3.2 The impossibility for external conjuncts to be stranded 21
2.3.3 The possible interactions between coordinators and
internal conjuncts 26
2.3.4 Extraction from both internal and external conjuncts 30
2.3.5 The syntactic relation between conjuncts: conclusions 31
2.4 The possible modifier function of conjuncts 33
2.5 The issue of so-called bar-level sharing 35
2.6 Chapter summary 40
part ii no special syntactic category 41
3 The categorial makeup of coordinate complexes 43

3.1 Introduction 43
xi
xii Contents
3.2 The categories of coordinators and conjuncts 44

3.2.1 Coordinators without c-selection restrictions 45
3.2.2 Coordinators with c-selection restrictions 46
3.2.3 Representing the categorial dependency of coordinators
on conjuncts 49
3.3 The categorial makeup of coordinate complexes 50
3.3.1 Coordinate complexes headed by and-like coordinators 50
3.3.2 Categorial features of coordinators that have c-selection
restrictions 57
3.3.3 Categorial unification in Spec-Head and Head-Compl
relations 59
3.4 Against &P 60
3.4.1 The distributions of coordinate complexes are covered by
simplexes 61
3.4.2 Neither closed classes nor case inflection argue for &P 63
3.4.3 Retrospection 64
3.5 Against the Clausal Conjunct Hypothesis 65
3.6 The structure of coordinate complexes composed of more than
two conjuncts 69
3.6.1 The coordinator must be grouped with an edge conjunct 71
3.6.2 The category decisiveness of non-final conjuncts in
English 72
3.6.3 Borsley’s arguments against the layered complementation
in English 73
part iii no special syntactic constraint 77
4 The Conjunct Constraint and the lexical properties

of coordinators 79
4.1 Introduction 79
4.1.1 The CCi and CCe 79
4.1.2 Previous approaches to the CC 81
4.1.3 A new account of the CC 85
4.2 The CCi and the asymmetry in conjunct drop 88
4.2.1 Conjunct drop in right-branching coordinate complexes 88
4.2.2 Conjunct drop in left-branching coordinate complexes 89
4.2.3 Clause-final coordinator-like elements 90
4.3 The CCe and the Chinese de constructions 92
4.3.1 Two kinds of de constructions 93
4.3.2 The various categories of kernel-final constructions 95
4.3.3 De as the head of the whole complex 97
Contents xiii
4.3.4 The chameleon-like nature of de keeps the kernel elements

in situ 105
4.4 The CCe and the he/gen comitative constructions in Chinese 107
4.4.1 Introduction: he/gen constructions in Chinese 108
4.4.2 Coordinator properties of the comitative he/gen 109
4.4.3 Violation of the CCe in non-distributive coordination 114
5 The Element Constraint and the semantic relatedness

of conjuncts 124
5.1 Introduction 124
5.2 Asymmetrical coordination as a type of natural coordination 124
5.2.1 Natural coordination 124
5.2.2 Asymmetrical coordination 127
5.2.3 Some formal contrasts between natural and accidental
coordination 128
5.3 The EC violation in asymmetrical coordination 135
6 Three puzzles solved by rejecting the CSC 141

6.2 Deriving Split Argument Constructions by giving up
the CC 141
6.2.1 The Split Argument Construction (SAC) 141
6.2.2 The two DPs of a SAC form a coordinate complex 145
6.2.3 Deriving SACs by conjunct raising 145
6.2.4 Section summary 153
6.3 Deriving Modifier-Sharing Constructions by giving up
the CC 154
6.3.1 The Modifier-Sharing Construction (MSC) 154
6.3.2 MSCs have coordinate antecedents 155
6.3.3 Deriving MSCs by sideward conjunct raising 160
6.3.4 A comparison with the multiple dimensional analysis 166
6.4 Deriving Interwoven Dependency Constructions by giving up
the EC 168
6.4.1 The Interwoven Dependency Construction (IDC) 169
6.4.2 Previous analyses 171
6.4.3 IDCs exhibit parallel movement dependencies 172
6.4.4 Deriving IDCs by sideward extraction from conjuncts 173
xiv Contents
7 Relativized parallelism in syntactic complexes 177

7.2 The Relativized Parallelism Requirement (RPR) 178
7.2.1 The Coordination of Likes Constraint and other similar
constraints 178
7.2.2 The RPR: conjuncts must hold a coherence relation 181
7.3 The components of the RPR 183
7.3.1 Examples of the semantic relatedness between conjuncts 183
7.3.2 Examples of the resemblance between conjuncts in
semantic types 186
7.3.3 The CLC: two further issues 190
dependency chains 192
7.4 The RPR in language processing 195
7.4.1 The more tightly semantically connected, the easier
to process 195
7.4.2 The more parallel in merged structures, the easier to process 196
7.4.3 The more parallel in dependency chains, the easier to process 197
7.5 The nature of the RPR 198
7.5.1 The RPR is a filter on representations of syntactic complexes 198
7.5.2 The general economy motivation of the RPR 201
7.6 Chapter summary and conclusions for Part III 202
part iv no special syntactic operation 205
8 The derivation of coordinate clauses with identity

adjectives 207
8.2 The identity adjective same 208
8.2.1 The general plural-α licensing of identity adjectives 208
8.2.2 Major questions about the syntax of TLCs 213
8.3 Building well-formed conjuncts of TLCs 214
8.3.1 The existence of a silent nominal in the second conjunct 214
8.3.2 The interpretation of the silent argument in the second
conjunct 215
8.3.3 The syntactic category of the silent argument in the
second conjunct 216
8.4 Extraction of SEs out of their licensing coordinate complexes 216
8.4.1 The extraction of SEs out of first conjuncts 217
8.4.2 Carlson’s constraint and the motivation for the SE extraction 218
8.4.3 The silence of the pro-form in the second conjunct of a TLC 220
Contents xv
9 Forming Across-the-Board constructions without forking

movement 222
9.2 ATB constructions as TLCs 222
9.2.1 The identity readings of ATB constructions 223
9.2.2 The syntactic reality of a silent argument in the
second conjunct 225
9.2.3 The correspondence between extraction and identity readings 229
9.2.4 The compatibility between two types of wh-expressions 231
9.3 The respectively readings of certain ATB constructions 233
9.3.1 Munn’s respectively readings 233
9.3.2 The availability of respectively readings in modification
constructions 234
9.4 A comparison with other approaches 236
9.4.1 The characteristics of our approach 237
9.4.2 The null operator approach 237
9.4.3 The multiple dimensional approach 238
9.4.4 The deletion approach 240
9.4.5 The sideward movement approach 240
9.5 Chapter summary and conclusions of Part IV 241
10 Conclusions 242
References 247
Index 267
Acknowledgments
It is more than nine years since I attended a summer school course called The
Syntax of Coordination in Potsdam, 1999, taught by Chris Wilder. My interest
in the topic started then. My initial work on coordination received help from
Chris Wilder, of course, as well as other former colleagues at the Center for
General Linguistics (Zentrum für Allgemeine Sprachwissenschaft, ZAS) and
other linguists in Berlin, including Philippa Cook, Laura Downing, Werner
Frey, Dieter Gasde, Hans-Martin Gärtner, Ljudmilla Geist, Andreas Haida,
Silke Hamann, Daniel Hole, Manfred Krifka, Ewald Lang, André Meinunger,
Kerstin Schwabe, Ben Shaer, and Arthur Stepanov. After I moved to Taiwan
in the fall of 2003, the writing of this book has benefited from various kinds
of help from the local linguistics community. I thank Henry Yungli Chang,
Tzong-Hong Jonah Lin, Jowang Lin, Chen-Sheng Luther Liu, Miaoling Hsieh,
Shuying Shyu, Chih-Chen Jane Tang, Tingchi Tang, Sze-Wing Tang, Jen Ting,
and Wei-Tien Dylan Tsai. In the Linguistics Institute at National Chung Cheng
University, I have been blessed with the kind support of my colleagues Jung-
hsing Chang, Jim H. Y. Tai, Jane S. Tsay, and our administrative assistant
Shu-Fen Hsu.
Early versions of most of the chapters have been presented at many con-
ferences, including the 11th International Conference on Chinese Linguistics
(Nagoya, Aug. 20–22, 2002), the 1st International Workshop on East Asian
Linguistics (Kyoto, Aug. 23, 2002), the Conference on Null Subjects and Para-
metric Variation, Reykjavik (July 18–19, 2003), GLOW in Asia 4 (Seoul, Aug.
20–23, 2003), the 2nd Workshop on Formal Syntax and Semantics (Taipei, Sept.
27–29, 2003), the 16th North American Conference on Chinese Linguistics
(Iowa, May 21–23, 2004), the 2nd International Workshop on Theoretical East
Asian Linguistics (Hsinchu, June 12–13, 2004), GLOW in Asia 5 (New Delhi,
Oct. 5–8, 2005), NELS 36 (Amherst, Oct. 28–30, 2005), GLOW 29 (Barcelona,
April 5–8, 2006), the 4th Workshop on Formal Syntax and Semantics (Chiayi,
April 14–15, 2006), the Symposium on Chinese Syntax and Semantics (Hong
Kong, Aug. 18–20, 2007), and the 13th International Morphology Meeting
xvi
Acknowledgments xvii
(Vienna, Feb. 3–6, 2008). I am grateful to the audience members of these confer-
ences and readers of various parts of the book manuscript for their suggestions,
encouragement, and challenges, especially Mark Baltin, Lisa Lai Shen Cheng,
Norbert Corver, Wayne Cowart, David Adger, Jingqi Fu, Yang Gu, Anders
Holmberg, Cheng-teh James Huang, Richard Kayne, Paul Kiparsky, Paul Law,
Thomas Lee, Yen-hui Audrey Li, Danqing Liu, Feng-his Liu, Jianming Lu,
Norvin Richards, Mamoru Saito, Yang Shen, Dingxu Shi, Peter Svenonius,
Satoshi Tomioka, Juan Uriakereka, Henk van Riemsdijk, Akira Watanabe, Dan
Xu, Hang-Jin Yoon, James Yoon, and Bojiang Zhang. Special thanks go to
Neal Whitman, who wrote fifteen pages of comments on the manuscript, which
helped me with the final revision.
I also want to thank Elizabeth Cowper and Diane Massam for teaching me
formal syntax, and Jinguo Ding, Yucun Qi, Yili Xu, Dechun Wang, and Liejiong
Xu for inspiring me to work on the puzzles of language.
Previous versions of parts of the book have been published in Language
and Linguistics (2006, Vol. 7: 175–223), Lingua (2007, Vol. 117: 2134–2158),
Lingua et Linguistica (2007, Vol. 1: 7–46), Taiwan Journal of Linguistics
(2007, Vol. 5: 19–47), Canadian Journal of Linguistics (2009, Vol. 54), Studia
Linguistica (2009, Vol. 63), and Language Research (2008, Vol. 44: 121–163).
I am grateful to these journals for allowing me to include revised versions of
the papers in this book. The suggestions from the reviewers and editors of the
journals and from the reviewers and board members of Cambridge Studies in
Linguistics have all helped improve the strength of the argumentation presented
in this final version. I am indebted to these anonymous teachers. Needless to
say, all remaining errors are mine.
Since the end of 2003, the writing of this book has been supported by grants
from the National Science Council in Taiwan. In preparing for the publication
of this book, I have also received tremendous assistance from Andrew Winnard,
the Senior Commissioning Editor of Cambridge Studies in Linguistics, and the
editorial and production team of Cambridge University Press.
I owe extremely special thanks to my husband James Myers. Without his
insightful academic discussions and psychological support, as well as his edit-
ing of the English in the whole manuscript, this book would be uninterpretable.
Abbreviations
AC Asymmetrical Coordination
AP Adjective Phrase
ATB the Across-the-Board dependency
CC Conjunct Constraint (the first part of CSC)
CCe No movement of external conjuncts (part of CC)
CCi No movement of internal conjuncts (part of CC)
CCC Coordinate Constituent Constraint
CED Condition on Extraction Domains
CCH Clausal Conjunct Hypothesis
CHC the Coordinate-Head RC Construction
CLC Coordination of Likes Constraint
CLCfunc CLC with respect to grammatical functions
CLCsem CLC with respect to semantic types
CSC Coordinate Structure Constraint
DP Determiner Phrase
EC Element Constraint (the second part of CSC)
FM Focus Marker
IDC the Interwoven Dependency Construction
ID Identification feature
IP Infl Phrase
LF Logical Form
Mod Modifier
ModP Modifier Phrase
MSC the Modifier-Sharing Construction
NCC Null Conjunct Constraint
NP Noun Phrase
Num Number
PF Phonological Form
PP Preposition Phrase
PPC the Paired Pronoun same Construction
xviii
Abbreviations xix
PR Parallelism Requirement
QR Quantifier Raising
QT quotative morpheme
RC Relative Clause
RNR Right Node Raising
RPR Relativized PR
SAC the Split Argument Construction
SE Similarity Expression
TLC the Thematic Licensing same Construction
TP Tense Phrase
UTAH Uniformity of Theta-Assignment Hypothesis
VP Verb Phrase
&P Conjunction Phrase
In the Chinese examples:

CL classifier
EXP experiential aspect
INCH inchoative aspect
PRF perfect aspect
PRG progressive aspect
PRT sentence-final aspect particle
DE associative particle
Q sentence-final question particle
TOP topic
1 Introduction
All languages have coordinate constructions. Although generative grammar,

including recent work within the Minimalist framework, has made much
progress in reducing various types of construction-specific syntax to the mini-
mum, it has not answered the following four fundamental questions:
A. Does the derivation of coordinate constructions create any special syn-
tactic configuration, other than the general binary complementation
and adjunction configuration?
B. Does the derivation of coordinate constructions resort to any special
syntactic category, other than NP, VP, and so on?
C. Is the derivation of coordinate constructions subject to any special
constraint on syntactic operations, other than general conditions such
as the Minimal Link Condition?
D. Does the derivation of coordinate constructions require any special
type of syntactic operations, other than Merge and the step-by-step,
one-tail-one-head chains of Move?
In this monograph, my answer to all of these four questions is negative.
I argue against any special syntax of coordination. Consequently, no spe-
cial syntax is the real syntactic law of coordination, just as the Minimalist
program would lead us to expect.
However, the standard answer to Question A is affirmative. Coordination has
hitherto enjoyed the exclusive privilege of the flat multiple branching structure,
as shown in (1.1a), which can be found in nearly all linguistics textbooks.
(1.1) a. b.
conjunct
conjunct coordinator conjunct
coordinator conjunct
The binary-branching structure of coordination shown in (1.1b) has occa-
sionally been proposed (Yngve 1960: 456; Thiersch 1985; Munn 1987; Kayne
1994; Zoerner 1995; Johannessen 1996; among others). However, the key
1
2 Introduction
arguments against the binary-branching structure summarized by Dik (1968)

have never been refuted. For instance, if the combination of a coordinator and
one conjunct is a constituent, excluding the other conjunct, why is it never
able to undergo any regular movement? Since questions like this have not been
answered, the two opposing analyses still co-exist in the literature. However,
binary vs. not binary should be an issue of truth, rather than taste. In this mono-
graph, I not only answer Dik’s challenges to the binary structure analysis, but
also present a variety of new arguments to show that coordinate complexes have
a complementation structure, rather than any coordination-specific structure.
My arguments thus indicate that the relationship between two conjuncts is that
between a specifier (external conjunct) and a complement (internal conjunct),
with the head realized by the coordinator. Unlike previous binary approaches
(Munn 1993; Johannessen 1998; among others), I do not consider morphologi-
cal agreement in my argumentation. Since Koutsoudas (1968), it has been noted
that verbs may agree with the closest conjunct of a coordinate nominal, regard-
less of whether they precede or follow the nominal. This adjacency effect has
been accounted for from a processing perspective (e.g. Lorimor 2007; Steiner
2008). Moreover, it has long been observed that the denotation of coordina-
tion also plays a role in agreement (e.g. McNally 1993: 363; Huddleston and
Pullum 2002: 1283). Since in many cases morphological agreement in coordi-
nate construction is a processing or semantic issue, it cannot be used to argue
for any special structure of coordinate complexes. Instead, I use facts like the
following to argue for the structure in (1.1b): the asymmetry of conjuncts in
binding, in possessee pronominalization, in hosting coordinators, and in coor-
dinator floating.
Question B concerns the syntax of the categorial makeup of coordinate
complexes. In the literature, there are basically two approaches to this issue. In
the traditional approach, it has been assumed that the category of a coordinate
complex is simply that of the conjuncts. However, if the two conjuncts are of
different categories, like the nominal and clause in (1.2a) and the nominal and
PP in (1.2b), this approach does not tell us what the category of the whole
coordinate complex is and where it comes from.
(1.2) a. You can depend on [my assistance and that he will be on time]. (Sag
et al. 1985: 165)
b. John eats only pork and only at home. (Grosu 1985: 232)
The other approach is to claim that the category of all coordinate complexes
is &P (or CoP, ConjP, BooleanP), a special category exclusively for coordinate
constructions (Munn 1987; 1993; Zoerner 1995; Johannessen 1996; among
Introduction 3
others). I show that &P is both theoretically and empirically problematic. I

argue that in coordinate complexes headed by coordinators without any intrinsic
categorial features, like English and, the categorial features of a designated
conjunct move and provide the category features for the whole coordinate
complex. Thus, the category of a coordinate complex is always that of one of
the two conjuncts. The category feature movement argued for here not only
removes the problematic &P from the computation system, but also gives a
simple answer to Question C.
The main issue of Question C is how to account for the effects of the
Coordinate Structure Constraint (CSC, Ross 1967). This constraint is composed
of two parts: no conjunct may be moved (the Conjunct Condition, CC), and no
element may be extracted from conjuncts (the Element Condition, EC). The
CC and EC are illustrated in (1.3a) and (1.3b), respectively.
∗
(1.3) a. Which boy did John kiss [ and which girl]? (CC violation)
∗
b. What kind of herbs did you [[eat ] and [drink beer]]? (EC violation)
On the one hand, the CSC has been regarded as “the most problem-free
syntactic constraint ever discovered” (Postal 1998: 52), but on the other hand,
it has remained as the only construction-specific constraint in generative syn-
tactic theory. When Riemsdijk and Williams (1986: 28) introduce various con-
straints, they state “All the principles discussed here have since been modified,
generalized, or replaced. The fate of the CSC has been somewhat different,
however, because it has not interacted with the other constraints under these
revisions.” The CSC has survived for more than 40 years. It still challenges
generative linguistics, including the Minimalist program, which aims to abolish
all construction-specific constraints.
This monograph makes two contributions to syntactic theory with respect to
the CSC.
First, I review data showing that both the CC and the EC may be violated.
Representations that violate the CSC are fully acceptable if they satisfy a Rela-
tivized Parallelism Requirement, a processing filter. The Relativized Parallelism
Requirement is satisfied if conjuncts are semantically related to each other, or
if conjuncts show resemblance in semantic type and movement history.
Second, I propose a new account for the observed CSC effects. The observed
EC effects are explained by deviation from the Relativized Parallelism Require-
ment alone, and the observed CC effects are explained by the combination
of two factors: deviation from the Relativized Parallelism Requirement, and
the special lexical properties of and-like coordinators. These properties are
revealed when we compare such coordinators with those that have intrinsic
4 Introduction
categorial features, as in Chinese. Our proposed category feature movement,

which answers Question B above, accounts for the CC effects in coordinate
complexes headed by and-like coordinators. The categorial feature movement
from the external conjunct (i.e. first conjunct, in English) to the coordinator
brings about the effect that the conjunct may not move any more, since its
moving carrier, i.e., its categorial features (Chomsky 1995: 265), have gone.
Coordinators with c-selectional restrictions as well as null coordinators do not
need this categorial feature movement, and consequently the external conjunct
may move. By contrast, the internal conjuncts can never move away from
coordinators, since the latter, like many other types of head elements, may not
be stranded. This correctly predicts that internal conjuncts may move if the
coordinators are null.
Evidence for my new explanation of CSC effects comes from a wide
range of empirical studies, including detailed studies of comitative construc-
tions, a study of other types of head elements that also have no intrin-
sic categorial features, and asymmetrical coordination in both English and
Chinese.
Theoretically, this new account of the CSC is plausible. In the Minimal-
ist program, movement is driven by morphological considerations (Chomsky
1995: 262). Logically, it is also possible that the blocking of movement is
related to morphological properties of specific syntactic elements, in addition
to the generally recognized locality restrictions.
Empirically, one sees in this monograph that removing the CSC from the
computational system also enables us to understand the syntactic derivations of
three apparently puzzling constructions: Split Argument Constructions in both
English and Chinese, as in (1.4a) and (1.5a) (cf. (1.4b) and (1.5b)), Modifier-
Sharing Constructions in English, as in (1.6), and Interwoven Dependency
Constructions in both English and Chinese, as in (1.7).
(1.4) a. John married Jane.

b. [John and Jane] married.
(1.5) a. Tudou yijing shao-le niurou. [Chinese]

potato already cook-PRF beef
‘The potatoes have already been cooked with the beef.’
b. Baoyu shao-le [tudou gen niurou].
Baoyu cook-PRF potato and beef
‘Baoyu cooked the potatoes and the beef.’ (either separately or together)
(1.6) John met a man and Mary met a woman who knew each other well.
(1.7) How many frogs and how many toads did respectively Greg capture and Lucilli
train?
Introduction 5
These constructions seem to be in conflict with certain basic syntactic laws

observed elsewhere: theta-role licensing and the general semantics–syntax
mapping stated in Baker’s (1988; 1997) Uniformity of Theta-Assignment
Hypothesis, the identification of the syntactic relation between modifiers and
their split modified elements, and the identification of the launching site of
certain movement operations. As stated in the final sentence of Postal (1998),
the challenges brought by such constructions can be avoided only under pain
of maintaining a theory that denies that these constructions actually occur in
natural languages. In this monograph, I propose syntactic derivations for the
three constructions that require us to set aside the CSC. Specifically, both Split
Argument Constructions and Modifier-Sharing Constructions are derived by
conjunct raising, and Interwoven Dependency Constructions are derived by
element extraction from conjuncts. Thus, giving up the CSC may strengthen
the explanatory power of syntactic theory.
Finally, Question D asks how the Minimalist program explains the derivation
of Across-the-Board (ATB) constructions, as in (1.8a), which have motivated
so-called ATB movement, illustrated in (1.8b) (Ross 1967; Williams 1977).
(1.8) a. Who did Jim like and Jane hate?

b. Who did Jim like _ and Jane hate _? (ATB movement)
This alleged operation is specific to coordinate constructions. Unlike all other

movement chains, the chain of ATB movement is forking, with two tail links. By
contrast, I argue that ATB constructions are derived by the extraction of a rela-
tional expression from just one conjunct, with a binding dependency between
the relational expression and a silent pronominal element in the other conjunct.
Both the extraction operation and the pronoun binding dependency are moti-
vated independently of coordinate complexes. Thus this proposed derivation
does not require any ad hoc forking chains of movement.
In the course of addressing these four fundamental questions, many other
empirical issues are investigated more deeply in this monograph than ever
before, including certain conjunct-internal coordinators in Chinese and other
languages, coordinators that cannot be used in collective contexts, the syntactic
relationship between coordinate constructions and comitative constructions in
Chinese, the derivation of the identity relation construction such as The same
man Mary helped and Jane ruined, as well as the syntactic derivations of the
constructions represented by (1.4a), (1.5a), (1.6), and (1.7).
This monograph, however, does not cover all properties of coordination. As
noted above, morphological agreement is not discussed in this monograph, and
6 Introduction
neither are Right Node Raising, reduction forms such as gapping, VP ellipsis,
and sluicing, or diachronic aspects of coordination.
The monograph has four main parts, in addition to this introduction and
the final concluding chapter. Part I (Chapter 2) deals with question A; Part II
(Chapter 3) deals with question B; Part III (Chapters 4, 5, 6, and 7) deals with
question C; and Part IV (Chapters 8 and 9) deals with question D.
I No special syntactic
configuration
2 The complementation structure
of coordinate complexes
2.1 Introduction
A coordinate complex is a syntactic constituent consisting of two or more
units (called conjuncts), and its category is identical to that of at least one of
the conjuncts.1 Generally, there is an element (particle, clitic, affix) to link
the conjuncts. Such an element is called a coordinator, which can be further
classified as a conjunctive (e.g. and), disjunctive (e.g. or), and adversative
coordinator (e.g. but). How two (or more) conjuncts and a coordinator are
organized in a coordinate complex has been an open question: do they form a
flat multiple branching structure or any version of the basic binary-branching
structure? The goal of this chapter is to answer this question.
In this chapter, I examine the structure of coordinate complexes that are
composed of two conjuncts. This is the basic type of coordinate complex. I
will leave my discussion of coordinate complexes that are composed of more
conjuncts to Chapter 3. I will also leave discussion of the category of coordinate
complexes to that chapter.
The following claims have been seen in the previous literature, but have not
been generally accepted. This is why it is still necessary to argue for them in
this chapter:
(A) Coordinate complexes have a binary-branching structure, and thus one

conjunct is structurally closer to the coordinator than the other con-
junct. This constituency is not captured by the traditional flat multiple-
branching representations.
(B) The head of the structure is realized by a coordinator, and the conjunct
that is structurally closer to the coordinator is the complement of the
head, and the other conjunct is Spec of the head.
1 De Vries (2006: 239) states that the term conjunct is confusing, since it refers to one of the coordi-
nated elements, regardless of whether the coordination is conjunctive, disjunctive, or adversative.
However, the term conjunct is conventional in the syntactic literature on coordination.
9
10 The structure of coordinate complexes
(C) The semantic relation between conjuncts does not need to be sym-
metrical, and thus the asymmetrical syntactic relation between con-
juncts is compatible with the possibility of an asymmetrical relation
in semantics.
(D) Conjuncts, which are non-projecting elements in coordinate comple-
xes, can be of any constituency level (word-fragment, word, phrase),
and this freedom in conjuncts does not affect the complementation
structure of coordinate complexes.
The above four claims will be made one by one in Sections 2.2 to 2.5.
Section 2.6 is a brief summary.
2.2 The binary-branching constituency of coordinate complexes

This section discusses the constituency of coordinate complexes. I will advocate
a binary-branching structure for such complexes, making a syntactic distinction
between internal and external conjuncts.
The hypothesis that coordinate complexes are binary has been proposed in
De Groot (1949: 66, 112–113), Nida (1949: 42 fn. 25), Yngve (1960: 456),
Thiersch (1985), Schachter (1985: 46), and Munn (1987). It is in contrast to the
assumption that such complexes have a flat structure in which conjuncts are on
the same level, while the coordinator holds them together without being more
closely connected with any one of them. The two assumptions are represented
in (2.1a/a ) and (2.1b), respectively.
(2.1) a. or a′.
conjunct conjunct
coordinator conjunct conjunct coordinator
b.
conjunct coordinator conjunct
The structure in (2.1b) has been assumed in many works, including Blümel
(1914: 193, 205), Bloomfield (1933: 185), Bach (1964: 67–68), Chomsky
(1965: 12–13, 196 fn.7), Dik (1968), Dougherty (1969), Gazdar et al. (1985:
170), Goodall (1987), and Muadz (1991). It is still quite popular in the current
literature of coordination (e.g. Phillips 2003; Takano 2004; Peterson 2004;
Wurmbrand 2008; Johnson 2008). However, the cross-linguistic facts to be
presented in this section do not support this assumption. Theoretically, the
binary structures in (2.1a) and (2.1a ) are also superior to the multiple-branching
2.2 The binary-branching constituency 11
structure in (2.1b) under economy considerations (Yngve 1960: 453; Collins

1997: 77).
I will first present evidence for the binary-branching structure in 2.2.1 through
2.2.4, and then answer Dik’s (1968) challenges against the binary-branching
analysis in Section 2.2.5.
Various arguments for the binary-branching structure of coordinate com-
plexes have been seen in the literature. Among them, I choose a representative
one, namely the asymmetry between conjuncts in binding, and then add three
more myself: the asymmetry between conjuncts in possessee pronominaliza-
tion, in hosting regular coordinators, and in hosting floating coordinators. These
arguments show the structural closeness of the coordinator to one of the two
conjuncts, and thus indicate that the coordinator forms a constituent with the
conjunct, as in the two structures in (2.1a) and (2.1a ).
2.2.1 The asymmetry between conjuncts in binding

A well-cited argument for the binary-branching constituency of coordinate
complexes is Blümel’s (1914: 164) observation of a binding asymmetry in
coordination. The first conjunct as a whole can be the antecedent of a pronoun
in the second conjunct, but the second conjunct as a whole cannot be the
antecedent of a pronoun in the first conjunct. Relevant English data appear
in Dik (1968: 36, 57) and Moltmann (1992a: 28, 45, 50). For instance, in
the coordinate complex every man and his dog in (2.2a), the first conjunct
every man can be the antecedent of the pronoun his in the second conjunct.
However, in the coordinate complex his dog and every man in (2.2b), the second
conjunct every man cannot be the antecedent of the pronoun his in the first
conjunct.
(2.2) a. Every mani and hisi dog left.

b. ∗ Hisi dog and every mani left.
Likewise, in each of the following examples, the second conjunct contains a

pronoun which is co-referential with the first conjunct.
(2.3) a. [That Himmler appointed Heydrich] and [the implications

thereof] frightened many observers. (Bayer 1996: 580)
b. Pat is [a Republican] and [proud of it]. (Sag et al. 1985: 117)
In (2.3a), the word there is combined with the preposition of. In this usage,
there is co-referential with the first conjunct that Himmler appointed Heydrich.
In (2.3b) the pronoun it pronominalizes the whole first conjunct a Republican,
which is a predicate nominal here (see Déchaine and Wiltschko 2002: 410 for
a discussion of NP pronouns, in addition to the more familiar DP pronouns).
If we reversed the order of the two conjuncts in the above data, the resulting
forms would not allow the binding relation. We can see that the two conjuncts
are asymmetrical in binding: the first conjunct is structurally higher than the
second one, as in the structure in (2.1a). Accordingly, the structure of coordinate
complexes cannot be a flat multiple-branching one.2
2.2.2 The asymmetry between conjuncts in possessee pronominalization

The binary-branching constituency of coordinate complexes is also supported
by the asymmetry between conjuncts in possessee pronominalization. A pos-
sessee in the first conjunct can take part in pronominalization in the second
conjunct, but no possessee in the second conjunct may take part in pronom-
inalization in the first conjunct. In (2.4a), the semantics of the possessee NP
mother in the first conjunct takes part in the pronominalization of yours, which
is the second conjunct. However, in (2.4b), the semantics of the possessee NP
mother in the second conjunct may not take part in the pronominalization of
yours, which is the first conjunct.
(2.4) a. Sally’s mother and yours have turned vegetarian.

b. ∗ Yours and Sally’s mother have turned vegetarian.
The contrast indicates that the two conjuncts are asymmetrical. Such an
asymmetry is also observed between subjects and objects:
(2.5) a. Sally’s mother praised yours.

b. ∗ Yours praised Sally’s mother.
Subjects c-command objects. Data like (2.5) lead us to the generalization

that no c-commanded element may take part in possessee pronominalization.
The NP mother is c-commanded by yours in (2.5b), but not in (2.5a). If first
conjuncts c-command the second conjuncts, the contrast in (2.4) is accounted
for by the same generalization.
2 Another type of binding asymmetry between two conjuncts can be found between an element in
one conjunct and an element in another conjunct, for example the nominal one bottle of wine in
one conjunct and it in the other conjunct in the following data.
(i) John bought just [one bottle of wine]i and served iti with the dessert.
(ii) ∗ John bought (just) it and served [one bottle of wine] with the dessert.
i i
For relevant discussions, see Mittwoch (1979), Munn (1993), Collins (1988a; 1988b), Wilder
(1994; 1999: Section 4.4 and Section 3.4), Heim and Kratzer (1998: 280–297), Progovac (1998a;
1998b), Camacho (2003: 16).
Note that the possessee nominal that takes part in the pronominalization is
not referential. Thus the contrast seems not to be covered by binding Princi-
ple C, which considers pronominalization of referential nominals. The above
generalization captures both Principle C and the possessee pronominalization.
The asymmetry in possessee pronominalization suggests that the first con-
junct is structurally higher than the second one. Accordingly, the structure of
coordinate complexes cannot be a flat multiple-branching configuration.
2.2.3 The asymmetry between conjuncts in hosting coordinators

Coordinators are not hosted by every conjunct equally. Coordinators can be to
the left (prepositive), or right (postpositive) of a conjunct. Even in the same
language, one coordinator can be prepositive and another one can be postpos-
itive (Dik 1968: 47; Haspelmath 2004: 6). Ross (1967: 90–91) observes that
coordinators like Japanese -to or -si form a phonological unit with the preced-
ing conjunct whereas a coordinator like English and forms a phonological unit
with the following conjunct (see also Schachter 1985: 47 and McCawley 1988a:
523). The prepositive position of and is shown in (2.6a), and the postpositive
position of -si is shown in (2.6b).
(2.6) a. The son graduated // and the daughter got married.
b. musuko-ga sotugyoo sita-si //musume-ga yome-ni itta
son-nom graduation did-and daughter-nom bride-dat went
‘The son graduated and the daughter got married.’
If the two conjuncts of a coordinate complex were syntactically equal, it
would be possible for the coordinator to be merged with either conjunct and
thus be grouped with either conjunct phonologically, so long as its position
relative to the conjuncts is consistent. We would then predict the following
forms to be acceptable:
∗
(2.7) a. and the son graduated // the daughter got married.
∗
b. musuko-ga sotugyoo sita //musume-ga yome-ni itta-si
son-nom graduation did daughter-nom bride-dat went-and
The fact that such forms are not acceptable indicates that the two con-
juncts are not syntactically symmetrical, and only one of them is consistently
grouped with the coordinator. The acceptability contrast between (2.6) and
(2.7) clearly shows that coordinate complexes have a binary structure and and-
coordinate complexes are right-branching, whereas si-coordinate complexes
are left-branching.
One might use a certain iconicity principle to ascribe the unacceptability
of (2.7) to the edge positions of the coordinators, assuming coordinators must
occur between the linked elements. However, the Latin coordinator -que almost
always occurs after the first word of a conjunct, and if the conjunct is a single
word, -que does not occur in the middle of a coordinate complex. Data like
(2.8) falsify the alleged iconicity principle.
(2.8) Marcus Julius-que [Latin]
Marcus Julius-and
‘Marcus and Julius’
It is true that in some languages, a coordinator may occur with each conjunct.
However, in such cases, the deletable coordinator and the undeletable one are
grouped with different conjuncts. If we consider the undeletable one only, we
see the same asymmetry of conjuncts in hosting coordinators (see Zoerner
1995: 23 and Zhang 2006: 179). Deletable coordinators have been argued to
be focus particles parasitic on the real coordinators (Hendriks 2002; de Vries
2005; Zhang 2008a).
2.2.4 The asymmetry between conjuncts in coordinator floating

In this subsection, I report that in Mandarin Chinese and Hungarian, certain
coordinators can occur either between two conjuncts, or inside the second
conjunct, but they never occur to the left of or inside the first conjunct. I call
such coordinators floating coordinators.
In Chinese, the adversative coordinator ke(shi) ‘but’ and the conjunction
yushi ‘and thus’ conjoin clauses. They either precede the second conjunct or
immediately follow the subject (or topic) of the second conjunct (Shi 1986;
Lü et al. 1999). In (2.9), the subject of the second conjunct is wo ‘I.’ In (2.9a),
ke(shi) precedes the second conjunct, i.e., to the left of wo. In (2.9b), it follows
wo. (2.9c) and (2.9d) show that ke(shi) cannot occur in the first conjunct,
regardless of its position relative to the subject there.3
(2.9) a. Baoyu yao tiaowu, ke(shi) wo yao hui-jia.

Baoyu want dance but I want return-home
b. Baoyu yao tiaowu, wo ke(shi) yao hui-jia.
Baoyu want dance I but want return-home
Both a and b: ‘Baoyu wants to dance, but I want to go home.’
∗
c. Baoyu ke(shi) yao tiaowu, wo yao hui-jia.
Baoyu but want dance I want return-home
∗
d. Ke(shi) Baoyu yao tiaowu, wo yao hui-jia.
but Baoyu want dance I want return-home
3 The word ke(shi) is not an adverb. Adverbs may not precede a nominal in Chinese. See Zhang
(2006: 183) for more arguments against the adverb analysis of this word.
If the coordinator ke(shi) floats in the second conjunct only, we can conclude
that the coordinator has a closer relation to the second conjunct than to the first
conjunct.
The same restriction on the floating scope is seen in another coordinator yushi
‘and thus.’ This coordinator is used if the eventuality expressed by the second
conjunct is a consequence of the eventuality expressed by the first conjunct. It
can be regarded as the Chinese counterpart of and in (2.10) (or the Malagasy
coordinator dia ‘and then’; see Payne 1985: 24 for similar asymmetrical coor-
dinators in other languages). In (2.10a), for instance, the event denoted by the
second conjunct broke down in tears is a consequence of the event denoted by
the first conjunct heard the news.
(2.10) a. The child heard the news and broke down in tears.
b. John drank the poison and died.
The distributions of yushi are similar to that of keshi. It either precedes the
second conjunct, as in (2.11a) below, or follows the subject of the second con-
junct, as in (2.11b). Moreover, no other coordinator may precede the conjunct
in which yushi occurs. Importantly, yushi never precedes any element of the
first conjunct, as shown by (2.11c).
(2.11) a. Baoyu yi guli, yushi Daiyu huifu-le xinxin.

Baoyu once encourage and Daiyu recover-prf confidence
b. Baoyu yi guli, Daiyu yushi huifu-le xinxin.
Baoyu once encourage Daiyu and recover-prf confidence
Both a and b: ‘Baoyu encouraged her, and thus Daiyu recovered her
confidence.’
c. (∗ yushi) Baoyu (∗ yushi) yi guli, Daiyu huifu-le xinxin.
and Baoyu and one encourage Daiyu recover-prf confidence
My syntactic analysis of the floating of the two coordinators will be presented

in Section 2.3.3.
Parallel Hungarian data can be found in Bánréti (1994: 356ff.). In this lan-
guage, coordinators are divided into three groups, according to their distri-
butions. Coordinators such as és ‘and’ must occur between two conjuncts.
Coordinators such as azonban ‘however,’ called right-shifted coordinators,
occur either between two clausal conjuncts or within the second clausal con-
junct (following the topic or subject of the clause). Finally, coordinators
such as meg ‘and’ must occur within the second conjunct, specifically, they
must follow the topic or subject of the second conjunct. In (2.12a), the con-
junction meg follows the subject of the second conjunct, Péter. In (2.12b),
meg precedes the whole conjunct and the sentence is unacceptable (Bánréti
1994: 357).
(2.12) a. János a televiziót nézte, Péter meg a rádiót hallgatta.
John the TV-acc watched Peter and the radio-acc listened
‘John watched the TV, and Peter listened to the radio.’
b. ∗ János a televiziót nézte, meg Péter a rádiót hallgatta.
In this language, although coordinators such as meg must occur within the
second conjunct, no coordinator may occur within the first conjunct. The asym-
metry between the two conjuncts is obvious (conjunctions that must follow the
topic of the second conjunct are also observed in Nupe; see Kandybowics 2006,
Section 2).
The restricted floating scope of the coordinators suggests that in a coordinate
complex, the two conjuncts do not have an equidistant relation with the coordi-
nator. One of them is closer to the coordinator and is thus able to interact with
it. The scope of the interaction divides a coordinate complex into two parts:
(i) the first conjunct; (ii) the combination of the coordinator and the second
conjunct. The constituency of a coordinate complex in Mandarin Chinese is
thus [[conjunct [& conjunct]].4
2.2.5 Conclusion and Dik’s challenges

The above four arguments show that coordinate complexes do not have a flat or
multiple-branching structure. Instead, they show a binary-branching structure,
where the coordinator and one conjunct form a constituent, excluding the other
conjunct. I will call a conjunct that forms a constituent with a coordinator
an Internal Conjunct, and the one that does not an External Conjunct. A
binary structure can be either right-branching, as in (2.13a) (= (2.1a)), or left-
branching, as in (2.13b) (= (2.1a )). I will concentrate my study to the pattern
of (2.13a) only.
(2.13) a. (the right-branching type)

(e.g. English and-complexes)
internal conjunct
external conjunct
4 Ross (1967) uses the distribution of the German word aber ‘but’ to argue that coordinators must
be grouped with the second conjunct in this language. See Zhang (2006: fn 6) for a critical
comment on this aber argument, although my discussion here and Ross’s argument lead to the
same conclusion.
b.
(the left-branching type)
(e.g. Japanese to-complexes)
external conjunct
internal conjunct
Note that in these structures, what is shown is the constituency of sub-

components of a coordinate complex, rather than the surface position of
the elements within a constituent. Specifically, I care about which conjunct
belongs to the same constituent as the coordinator. Where the coordinator
surfaces with respect to this particular conjunct is a separate issue: it can
be to the left (prepositive), as seen in (2.6a), or right (postpositive) of the
conjunct, as seen in (2.6b), or align with a certain prosodic unit within the
conjunct (e.g. Latin -que). Even in the same language, one coordinator can be
prepositive and another one can be postpositive (Dik 1968: 47; Haspelmath
2004: 6).
Dik (1968: 53–55) argues against the binary-branching structure of coor-
dinate complexes, advocating a flat multiple-branching structure where all
conjuncts and coordinators are structurally on the same level.
I address his four challenges as follows.
(a) Assuming that there must be a parallelism between conjuncts, Dik
claims that the parallelism “is spoiled” by the binary-branching anal-
ysis (p. 53). However, we have seen in the arguments in 2.1 that
conjuncts are intrinsically asymmetrical. Therefore, I do not accept
Dik’s assumption that there must be a syntactic parallelism between
conjuncts. In addition to the binding asymmetry presented above, data
like (2.14a), which he also cites (p. 28), do not show a categorial paral-
lelism, and data like (2.14b) (Lakoff 1986) do not show any movement
chain parallelism.
(2.14) a. John walked slowly and with great care. (Adv and PP)
b. What kind of cancer can you eat herbs and not get ?
In general, not all coordinate complexes show syntactic parallelism.

Indeed, no binary-branching representation requires conjuncts to be
parallel to each other. Rather than “spoiling” anything, binary branch-
ing actually captures the facts.
Chapter 7 of this book will show that parallelism is a processing fil-
ter on representations of syntactic complexes rather than a constraint
on syntactic operations, and thus it has no effect on the structure-

building of syntactic complexes. In other words, under certain condi-
tions, parallelism may lead to the rejection of certain representations;
however, no structure-building operation needs to follow a parallelism
constraint. Data like (2.14) are well-formed exactly because they are
derived without the enforcement of parallelism.
(b) Dik claims that only the flat multiple branching structure is “able
to connect the difference between coordination and subordination
with different hierarchical descriptions” (p. 54). The difference, in my
analysis, is between the complementation structure of coordination
and the adjunction structure of subordination. In Section 2.3, I will
argue for the Spec-Complement relation between conjuncts and show
how conjuncts are syntactically different from adjuncts.
(c) Some early advocates of the binary-branching approach, such as van
der Lubbe (1958: 80), claimed that the closer connection between
coordinators and second conjuncts is evident from the following con-
trast: coordinators may be separated from first conjuncts by other
elements, whereas they may not be separated from second con-
juncts. Dik presents the data below to show that their claim is
wrong:
(2.15) a. John will come today but, as he said to me yesterday, he will not
be able to stay for the weekend.
b. I want you to know that he will come today and also that he will
not be able to stay for the weekend.
Dik is right in pointing out that insertion before the second conjunct
is possible. However, this only reveals sub-constituency inside the
constituent formed by a coordinator and the second conjunct. It does
not show that they themselves cannot form a constituent that excludes
the first conjunct. The possibility of insertion in examples like (2.15)
does not rule out alternatives to his flat structure.
(d) Dik’s strongest argument, in my judgment, is the following (p. 54):
If it were really true that the coordinator and the following member
constitute a unit with greater freedom of combination, we would
expect, e.g., that this unit could be shifted to the front of the whole
construction, as is indeed possible with subordinating coordinators
and certain adverbial modifiers . . . With coordinators, however, this
is always excluded; a fact which is at least partly accounted for if
the coordinator and the following member are not treated as a single
constituent.
2.3 The complementation structure 19
Dik’s challenge is based on a contrast observed by Sledd (1959:

101) and others. Sledd states that the cluster formed by a subordinator
like when and a clause may appear either initially or finally, as shown
in (2.16), whereas the cluster formed by and and a clause does not
have this freedom, as shown in (2.17):
(2.16) a. The Yankees and the Indians finished the second game of their
double-header, when the rain stopped.
b. When the rain stopped, the Yankees and the Indians finished the
second game of their double-header.
(2.17) a. The rain stopped and they finished the second game.
b. ∗ And they finished the second game the rain stopped.
Dik uses the constraint that coordinators cannot be moved together

with the second conjunct to argue that they do not form a constituent.
He concludes that coordinate complexes must have a flat structure,
as in (2.1b). Recently, Dik’s argumentation and conclusion reappear
in Bánréti (2003: Section 1). Unfortunately, advocates of the binary
structure of coordinate complexes make no comment on this challenge
of Dik. I will answer this challenge in Section 2.3.2. Briefly speaking,
the immobility of the constituent comes from its projecting level: it is
an intermediate projection. This property itself distinguishes coordi-
nate structure, which I will argue to be a complementation structure,
from adjunction structure (cf. (b) above).
I end this section with the following claims. Coordinate complexes have a
binary-branching structure. A coordinator forms a constituent with only one
of the conjuncts, the internal conjunct. The asymmetry between external and
internal conjuncts has an important impact on the computation of coordinate
complexes. Thus any multidaughter structure is problematic. Finally, all of
Dik’s challenges can be refuted.
In this section, I have focused my attention on the constituency issue
of coordinate complexes. I leave the issue of configuration to the next
section.
2.3 The complementation structure of coordinate complexes

I have not yet discussed the syntactic status of the sister of the external conjunct
in (2.13a), [&β]. There are three possibilities. The first two are adjunction
structures, as shown in (2.18a) and (2.18b).
(2.18)
a. (coordinate complex) b. XP (coordinate complex)
XP XP
(external conjunct) (external conjunct)
X X
(coordinator) (internal conjunct) (coordinator) (internal conjunct)
In (2.18a), [&β] is an adjunct to the external conjunct α. This structure is

proposed in Munn (1992; 1993) and adopted in Bošković and Franks (2000)
and Hartmann (2000). Munn’s (1992; 1993) structure for (2.19a) is shown in
(2.19b). In this structure, the projection headed by a coordinator is a right-
adjunct to the external conjunct. Consequently, it is the external conjunct that
is c-selected by the sister of the coordinate complex.
(2.19) a. Hobbs and Rhodes
b. DP
DP BP (Boolean Phrase, or &P)
Hobbs B DP
and Rhodes
In (2.18b), the external conjunct is an adjunct to the constituent formed by
the coordinator and the internal conjunct.
Both (2.18a)/(2.19b) and (2.18b) can be rejected if there is evidence to show
that the external conjunct and the internal conjunct are in a Spec-Complement
relationship. This is the third possibility. In this section, I advocate this possi-
bility. The complementation structure is shown in (2.20):
(2.20) XP
X′
(external conjunct)
X
(internal conjunct)
coordinator
This structure has been previously proposed by Munn (1987), Larson (1990),
Johannessen (1998), and Zoerner (1995). One of Zoerner’s arguments will be
introduced below in Section 2.3.3A. My arguments for (2.20) are the following:
No stranding of external conjuncts (Section 2.3.2);
The existence of interactions between coordinators and internal con-
juncts (Section 2.3.3);
The possibility of extraction from either internal or external conjuncts

(Section 2.3.4).
Unlike the works cited above, I do not consider agreement in my analysis.
I justify the exclusion of agreement in Section 2.3.1 (for critical comments on
other unreliable arguments for the structure in (2.20), see Chaves 2007: 23).
2.3.1 The dubious status of agreement in the syntax of coordination

Although conjuncts can be asymmetric in their nominal and verbal inflections
(e.g. agreement, case, and finiteness morphology; see Payne 1985: 27), inflec-
tion patterns are not directly related to the complementation structure. Since
Koutsoudas (1968), it has been noted that verbs may agree with the closest
conjunct of a coordinate nominal, regardless of whether they precede or follow
the nominal (also see Sadler 1999; 2003; Marušic et al. 2006). This adjacency
effect has been accounted for from a processing perspective (Lorimor 2007;
Steiner 2008). Moreover, it has long been observed that the denotation of coor-
dination also plays a role in agreement (e.g. McNally 1993: 363; Huddleston
and Pullum 2002: 1283; see also Section 5.2.3B of this book). Since in many
cases the contrast between singular and plural agreement is a semantic issue
in English (Humphreys and Bock 2005), using singular conjunct agreement to
argue for any special structure of coordinate complexes is misleading. Even
simplex plurals can trigger singular agreement, as seen in (2.21b) (Sag et al.
1985: 154) and (2.21c) (Bernard Comrie p.c. to Miao-Ling Hsieh):
(2.21) a. Ham and eggs is my favorite breakfast.

b. Flapjacks is my favorite breakfast.
c. Two bottles of wine {were/was} thrown into the soup.
Sag et al. (1985: 154 fn. 23) correctly point out that coordination is not
relevant to the verbal inflection in examples like (2.21a). Sobin (2004: 507)
also finds it problematic to use inflection as a diagnostic of the syntactic struc-
ture of coordinate complexes. Inconsistency in the manifestations of inflec-
tion has even led Peterson (2004: 650f.) to argue that coordinate complexes
have a headless structure. This conclusion seems unwarranted given that Lori-
mor’s (2007) experimental study concludes that “notional information, lexi-
cal plurality, adjacency, and linear (surface) word order play significant roles
in the computation and production of agreement” (cf. Cowart & McDaniel
2008).
2.3.2 The impossibility for external conjuncts to be stranded

My first argument for the complementation structure of coordinate complexes
is the immobility of the combination of a coordinator and the internal conjunct.
A coordinate complex can move as a whole, as seen in (2.22a), whereas

the constituents that are composed of a coordinator and an internal conjunct
cannot, as seen in (2.22b).
(2.22) a. [Tall and slim]i though Helen is i , . . .
b. ∗ [And slim]i though Helen is [tall i ], . . . (Postal 1998: 191)
The following data from McCawley (1988a: 267) show that clefting is possi-
ble for adverbials but not for the combination of the coordinator and the internal
conjunct. If clefting involves movement, the contrast in (2.23) is parallel to the
contrast presented above.
(2.23) a. It was before Jane arrived that Tom left. (McCawley 1988a: 267)
b. ∗ It was and Jane arrived that Tom left.
Recall that Dik (1968: 54) uses the immobility of the combination of the
coordinator and the internal conjunct to challenge the binary-branching analysis
of coordinate complexes (Section 2.2.5). Dik’s challenge has not been answered
in the literature. McCawley (1988a: 267) states that “and S is not a constituent
of a type that allows the mobility that time adverbs such as before S have,” but
he does not give any explanation. Kehler (2002: 61) also uses this constraint
to distinguish coordination constructions from adjunction (or “subordination”)
constructions.
The constraint is explained when one realizes that the combination of a
coordinator and its internal conjunct is an intermediate projection, which is
the sister of a Spec element, as seen in the structure in (2.20). As generally
recognized in generative syntax (Chomsky 1994; 1995: 253), no intermediate
projection may move. The illegal movement of an intermediate projection can
be illustrated by the following examples. In (2.24a), there is an I topicalization,
stranding the subject, which is at Spec of IP. In (2.24b), similarly, there is a
P wh-movement, assuming directly is in Spec of PP. In (2.24c), there is a P
wh-movement, adopting the general assumption that right is at a Spec position
if it precedes a spatial or temporal PP (see Emonds 1972; among others) (I
thank Chris Wilder for providing examples (2.24a) and (2.24b)).
(2.24) a. ∗ [Will see John]i , (she thinks that) [Peter i ].

b. ∗ [Under which book]i was it [directly i ]?
c. ∗ [To which town]i did you take the car [right i ]?
If an intermediate projection is not able to move, the constituent composed

of a coordinator and an internal conjunct can never move together, stranding
the external conjunct. The immobility of such constituents makes the structure
of coordinate complexes significantly different from that of adjunction con-

structions. As is well known, adjuncts introduced by words like after, before,
since, etc. enjoy the freedom of A-bar movement. We have seen such examples
in (2.23a). The mobility of the adjuncts indicates that they are not intermediate
projections.
The above analysis might be challenged by the hypothesis that (2.25a) is
derived from (2.25b) by the raising of [and he is an expert]:
(2.25) a. The professor [and he’s an expert], thinks the recession will continue.
b. The professor thinks the recession will continue [and he’s an expert].
We have, however, two doubts about this derivation relation between (2.25a)
and (2.25b). First, it seems that no phrasal movement in English lands to the
right of a subject. The syntactic status of this landing site in the language is not
clear. Second, the alleged movement cannot land to the left-peripheral position,
the default landing site for A-bar movement in English:
∗
(2.26) [And {he/the professor} is an expert], {he/the professor} thinks the
recession will continue.
I thus do not adopt this movement approach to data like (2.25a). Bors-
ley (1994: 240) states that [and he is an expert] in (2.25a) is a parenthetical
expression. I further assume that this parenthesis contains an implicit external
conjunct. Parentheses could be analyzed as adverbials, in the spirit of Potts
(2002). Importantly here, the complex does not undergo any movement. In
contrast, in (2.25b), the external conjunct is the clause The professor thinks the
recession will continue, and the internal one is [he is an expert].
I argue not only against any leftward movement of the sister of the exter-
nal conjunct, but also against any rightward movement of such a constituent.
Data like the following (2.27) have been cited to show that constituents that
are composed of a coordinator and a conjunct can be extraposed in English,
stranding the other conjunct (Collins 1988a; 1988b; Munn 1992: 19; 1993: 15;
Zoerner 1995: 10, 85; Progovac 1998a; 1998b; Cowper and Hall 2000: 33; see
also Huddleston and Pullum 2002: 1277; Fromkin et al. 2007: 149):
(2.27) a. John bought a book yesterday, and a newspaper.

b. Tomorrow he will come, I think, or the day after.
Such data are called Split Coordination in Höhle (1990: 141 fn.). Similar
examples can be found in German, as in (2.28), and in many other languages
(see Johannessen 1998: Section 6.2.1).
(2.28) Hans hat gestern ein Buch gekauft, und eine Zeitung.
Hans has yesterday a book bought and a newspaper
‘Hans bought a book yesterday, and a newspaper.’
The extraposition analysis is not conclusive. Cross-linguistically, we have

not seen any similar constructions in languages that unambiguously allow extra-
position. For instance, Amele, a Papuan language from New Guinea, allows
extraposition of adverbial clauses, but not of constituents that are composed of
a coordinator and a conjunct. This is shown in (2.29) and (2.30) (Roberts 1988:
55). (tod.p = today’s past tense)
(2.29) a. [Ija ja hud-ig-en fi] uqa sab man-igi-an.

1s fire open-1s-fut if 3s food roast-3s-fut
‘If I light the fire she will cook the food.’
b. Uqa sab man-igi-an [ija ja hud-ig-en fi].
3s food roast-3s-fut 1s fire open-1s-fut if
‘She will cook the food if I light the fire.’
(2.30) a. [Ija ja hud-ig-a qa] uqa sab man-i-a.
1s fire open-1s-tod.p but 3s food roast-3s-tod.p
‘I lit the fire but she cooked the food.’
b. ∗ Uqa sab man-i-a [ija ja hud-ig-a qa].
3s food roast-3s-tod.p 1s fire open-1s-tod.p but
In this language, the coordinator qa is attached to the end of the left-conjunct,

which should be the internal conjunct, and thus the internal structure of the
complex is left-branching. Roberts uses (2.29b) to show that the combination
of the fi ‘if’ and its complement can be extraposed, and uses (2.30b) to show
that the constituent formed by the coordinator qa ‘but’ and the first conjunct
cannot.
An alternative analysis of data like (2.27) is available. This is the Bare
Argument Ellipsis analysis (Hudson 1976; Neijt 1979; Moltmann 1992a: 22,
48, 228; Johnson 1996; Schwarz 1999: 354–355). Bare Argument Ellipsis
leaves only one non-verbal element in the elided sentence. In this analysis, the
second conjunct is regarded as a reduced sentence fragment (see Lang 1984:
21 for a review of early studies following this approach). (2.27a) thus can be
analyzed as follows:
(2.31) John bought a book yesterday, and he also bought a newspaper yesterday.
In Heim and Kratzer (1998: 249), it is assumed that in Bare Argument

Ellipsis, “the ‘remnant’ argument always has been topicalized (adjoined to
S) before the deletion.” The above sentence, for instance, should have the
following representation:
(2.32) John bought a book yesterday, and [[a newspaper]i he also bought ti yesterday].
Johnson (1996) argues against the extraposition analysis based on the seman-
tic type of the predicate involved. He observes that the analysis fails to predict
the pattern of grammaticality illustrated in (2.33) below (Johnson 1996: 69; see
Neijt 1979 for arguments against the extraposition analysis from an agreement
perspective):
(2.33) a. I introduced Carrie and Will to each other.

b. ∗ I introduced Carrie to each other and Will.
c. ∗ I introduced Carrie to each other and I introduced Will to each other.
In the extraposition analysis, the ungrammaticality of (2.33b) requires the

ad hoc restriction that a DP coordinate complex with a group interpretation
may not be discontinuous at S-structure. Later in Chapter 4, we will see that
the situation is just the opposite: a group interpretation allows conjuncts to be
separated, whereas a distributive one does not. On the other hand, if (2.33b)
is derived from the illegal (2.33c) through gapping, the ungrammaticality of
(2.33b) follows automatically. It is indeed the case that neither gapping nor
Bare Argument Ellipsis may have a collective predicate, as pointed out by
Moltmann (1992a: 49) (see also Section 5.2.3D of this book):
(2.34) a. ∗ John shared the coffee and Bill the cake.

b. ∗ John met and Bill (too).
Another restriction shared by Bare Argument Ellipsis and gapping is that

they cannot occur in embedded clauses (Carlson 2002: 11):
(2.35) a. Bill ordered beans and Sam rice. (Gapping)

b. ∗ Bill ordered beans and I think Sam rice. (Gapping)
(2.36) a. Bill ordered beans for supper and Sam too. (Bare Argument Ellipsis)
b. ∗ Bill ordered beans for supper and I think Sam too. (Bare Argument
Ellipsis)
The word too in data like (2.36a) has also been analyzed as an anaphoric
element, referring back to the VP in the first conjunct. If so, there is no ellipsis
(Goodall 1987: 28). However, as stated in Heim and Kratzer (1998: 259), not all
examples of ellipsis include such a particle. In the cases where no too occurs,
ellipsis is still possible.
I conclude that the plausibility of the deletion analysis casts doubt on the
extraposition analysis. In the deletion analysis, the constituent that is com-
posed of the coordinator and the internal conjunct is built independently
of the preceding clause, and therefore it is not moved from the preceding
clause.
It needs to be mentioned that Reinhart and Rooth (1991) propose an LF right-
adjunction approach to Bare Argument Ellipsis. Exactly the same arguments
are used in Cowper and Hall (2000) for an overt rightward movement of the
and-DP cluster. The approach is convincingly argued against by Moltmann
(1992a: 229).
I conclude that the immobility of the combination of a coordinator and an
internal conjunct suggests that the combination is an intermediate projection.
If so, the relation between an external conjunct and internal conjunct is a
Spec-Complement relation.
2.3.3 The possible interactions between coordinators and

internal conjuncts
The proposed complementation relation is also supported by the interactions
between coordinators and internal conjuncts. Such interactions are impossible
if one is an adjunct of the other as Moltmann (1992a) has claimed. I present two
types of syntactic interactions between coordinators and internal conjuncts, one
involving the launching site of head raising from internal conjuncts (used by
Zoerner 1995: 41 to argue, as I do, that coordinators and second conjuncts have
a head-complement relation), and the other involving the syntactic positions of
floating coordinators in languages such as Chinese.
Head raising from internal conjuncts to the positions of coordinators

One case showing an interaction between coordinators and internal conjuncts
is found in the Uto-Aztecan language Papago (O’odham). In this language, we
see an instance of head raising from Infl of the internal conjunct to the position
of coordinators. The canonical word order of the language is S-Aux-V, as seen
in (2.37a) and (2.37b). However, when two clauses conjoin, the Aux of the
second conjunct precedes the subject of the second conjunct, as seen in (2.37c)
(Zepeda 1983: 25–27) (the second auxiliary verb in (2.37c) has lost its initial
vowel through an independent phonological operation):
(2.37) a. ’Uwi ’o cipkan (Papago)

woman is working
‘The woman is working.’
b. ’A:ñi ’añ ko:s

I am sleeping
‘I am sleeping.’
c. ’Uwi ’o cipkan ñ ’a:ni ko:s
woman is working am I sleeping
‘The woman is working and I am sleeping.’
Zoerner (1995: 41) analyzes this phenomenon as head raising from Infl of the
second conjunct to the position of the coordinator. If this head raising analysis
is right, the second conjunct must be the complement of the coordinator (contra
Moltmann 1992a; Chaves 2007), since head movement may neither launch
from nor land in an adjunct.
The syntactic positions of floating coordinators

Another case showing an interaction between coordinators and internal con-
juncts is found in Mandarin Chinese.
Earlier I presented the fact that the coordinators ke(shi) ‘but’ and yushi ‘and
thus’ can either immediately precede internal conjuncts or immediately follow
the subjects (or topics) of internal conjuncts (Section 2.2.4). One pair of relevant
data is repeated here:
(2.38) a. Baoyu yao tiaowu, ke(shi) wo yao hui-jia.

Baoyu want dance but I want return-home
b. Baoyu yao tiaowu, wo ke(shi) yao hui-jia.
Baoyu want dance I but want return-home
Both: ‘Baoyu wants to dance, but I want to go home.’
The floating coordinators have two possible positions with respect to internal
conjuncts. The availability of the two positions for the coordinators indicates
that the coordinators and internal conjuncts interact. I have already showed
that a coordinator and the internal conjunct are merged as sisters (Section 2.2).
We also know that if there are interactions between sisters, they are in a head-
complement relation. Accordingly, internal conjuncts must be the complement
of the coordinators. If internal conjuncts were adjuncts, the dependency relation
between the two positions would be impossible. The floating phenomenon thus
supports the complementation configuration proposed in (2.20).
One property of floating coordinators is that they may not follow any non-
topic element. Adverbs, which are not able to function as topics, cannot precede
any coordinator. In (2.39), for instance, the manner adverbial buxiaoxin ‘care-
lessly’ is not a topic and it cannot precede keshi.
(2.39) a. Ta jijimangmang de qie cai, keshi buxiaoxin qie-po-le shouzhi.

he hurriedly de cut vegetable but carelessly cut-broken-prf finger
‘He cut the vegetable hurriedly, but he cut his finger carelessly.’
b. ∗ Ta jijimangmang de qie cai, buxiaoxin keshi qie-po-le shouzhi.
In order to capture this property, I assume that floating coordinators have an
unvalued [topic] feature, which can be valued by a topic of the internal conjunct
(the topic in (2.39) is null). The valuation can be accomplished by either of two
ways, namely by the Agree relation between the coordinator and the topic of
the internal conjunct, as in (2.38a), or by the movement of the topic to a Spec
position, as in (2.38b) (this point will be elaborated below).
Another property of floating coordinators is that when one follows the topic
of the internal conjunct, the external conjunct is a background topic, whereas
the topic of the internal conjunct is a contrastive topic. In (2.38b), for instance,
the first conjunct Baoyu yao tiaowu ‘Baoyu wants to dance’ is a background
topic, and wo ‘I’ is a contrastive topic.
Multiple topics, like multiple fronted wh-elements in certain Slavic lan-
guages, and multiple verbs in serial verb constructions, can be represented by
tucking-in structures, based on the Minimal Link and Local Move conditions
(see Collins 2002b: 10 for an implementation of Richards 1997). Accordingly,
I propose the following derivation for the multiple topic constructions. If two
CPs are coordinated, the head of the whole coordinate complex is also C
(further discussion of the categorial issue is given in Chapter 3). When the
coordinator ke(shi) or yushi is the realization of this C, the base-position of
the whole internal conjunct is the complement of this C, as shown in (2.40a).
In contrast to this, I claim, when the coordinator follows the topic of the
internal conjunct, the topic has tucked in to an inner Spec position, as shown
in (2.40b).
(2.40)
a. b. CP3
CP3 CP1 CP3
CP1 C′3 external conjunct topic C′3
external conjunct C3 CP2 C3 CP2
yushi internal conjunct yushi internal conjunct

ke(shi) ke(shi)
tucking-in
Note that in multiple-fronted wh-constructions and serial verb constructions,

all of the relevant elements move, whereas in (2.40b), the outer Spec element
is base-generated there. This difference does not affect the tucking-in analysis,
however. This is because the Local Move condition forces the mover to land
in the inner Spec position if the outer Spec is realized by an element of the
same type, regardless of whether the latter element is base-generated there
or not.
Topic movement across a C-element is also seen in clausal subordination in
Chinese, where complementizers such as yaoshi/ruguo ‘if,’ yinwei ‘because,’
suiran/suishuo ‘although,’ jiran ‘since’ can also follow the subject/topic of the
selected clause (Shi 1986: Section 3.7; McCawley 1988b: 181; Tsao 1996:
174; Zhou 2002: Chapter 5). In the following data, the two sentences are near
synonymous.5
(2.41) a. Ruguo qingkuang shushi, najiu jianjue geiyi zhicai.

if case true then resolutely enforce punishment
‘If the case is true, then (we) must enforce punishment resolutely.’
b. Qingkuang ruguo shushi, najiu jianjue geiyi zhicai.
The topics to the left of the complementizers are background topics rather
than contrastive topics. Presumably, such complementizers also have an unval-
ued topic feature, to be valued by a topic of the selected clause.
As in the coordinate complex (2.39a), adverbs, which cannot be topics,
cannot precede a complementizer. In (2.42), for instance, the adverb jingchang
‘often’ cannot precede ruguo ‘if.’
(2.42) a. Ruguo jingchang xiayu, najiu bu yao jiaoshui le.

if often rain then not should water prt
‘If it rains often, do not water (the plants) any more.’
b. ∗ Jingchang ruguo xiayu, najiu bu yao jiao shui le.
We can see that the behavior of coordinators patterns with that of typical
complementizers. This fact shows that coordinators are head elements. I have
assumed that the observed floating phenomenon is an indication of topic move-
ment. Since syntactic movement starts from complement rather than adjunct,
the floating phenomenon suggests that internal conjuncts are complements
rather than adjuncts.
5 In classical Chinese, the conjunction er could also occur between the subject/topic and the
predicate of the second conjunct of a coordinate complex. In Classical Latin poems and prose,
subordinators such as cum ‘when’ can also be “shifted” inside the relevant subordinate clause;
this inversion has been called topicalization (see Huttar 2004). In Nupe, conditional gá ‘if ’ also
follows the subject of the conditional clause, and like Hungarian meg, the conjunction ma and
ci must follow the topic of the second conjunct (Kandybowics 2005: Section 2.2).
In 2.2.4 I introduced the Hungarian fact that meg ‘and’ must follow the
topic of the second conjunct. If we extend the above analysis to the Hungarian
coordinate complexes headed by meg, we can claim that the topic movement
launching from the second conjunct to the left of meg is obligatory, assuming
that there are certain edge features involved (see Chomsky 2007; 2008).
2.3.4 Extraction from both internal and external conjuncts

The fact that phrasal extraction is possible from both internal and external
conjuncts distinguishes coordination from any adjunction structure.
This possibility is not expected given the Coordinate Structure Constraint
(CSC, Ross 1967: 89), which states: “In a coordination structure, no conjunct
may be moved, nor may any element contained in a conjunct be moved out
of that conjunct.” Since Grosu (1973), the CSC has been split into two parts.
The first part is that no conjunct may be moved, and the second part is that
no element may be extracted from conjuncts. Following Grosu (1973), I call
the first part of the CSC the Conjunct Constraint (CC), and the second part the
Element Constraint (EC). The CC and EC are illustrated in (2.43a) and (2.43b),
respectively.
(2.43) a. ∗ Which boy did John kiss [ and which girl]? (CC violation)
b. ∗ What kind of herbs did you [[eat ] and [drink beer]]? (EC violation)
Unlike the CC, the status of the EC is controversial. Grosu (1973), Gold-
smith (1985), and Lakoff (1986) present quite a lot of English data to show that
the EC can be violated. Postal (1998), however, claims that none of Lakoff’s
data can challenge the CSC, since some of the data do not have typical coor-
dination readings, or they are constrained in various ways. One weakness of
this argument is highlighted by the relation between the CC and the EC. Both
parts of the CSC have been supposed to apply only to coordinate construc-
tions. Therefore, if Lakoff’s data cannot be regarded as coordination data, they
should not be subject to the other part of the CSC either. Yet as correctly
pointed out by Postal, “The Conjunct Constraint is almost never questioned;
nothing in Lakoff 1986 is intended to challenge it” (p. 83). Postal then shows
how exactly the same data used by Lakoff to challenge the EC must obey the
CC. If Lakoff’s data strictly obey the CC, and if non-conjuncts are not con-
strained by either the CC or the EC, the conclusion must be that his data are
indeed coordination data, and therefore the EC can indeed be violated in certain
cases. I therefore accept Grosu (1973), Goldsmith (1985), and Lakoff’s (1986)
conclusion that the EC can be violated (see also Levine 2001: 161; Kehler
2002).
In (2.44), extraction from either conjunct is possible (Lakoff 1986):
(2.44) a. What kind of cancer can you eat herbs and not get ?
b. What kind of herbs can you eat and not get cancer?
The data in (2.44) represent so-called Asymmetrical Coordination (AC; Ross

1967; Schmerling 1975; Lakoff 1986; Levin and Prince 1986).6 The violation
of the EC is also found in other languages (see Chapter 5). Summing up, there
are good reasons to believe that the EC can be violated.
The extractability of conjunct-internal elements shows that AC does not have
an adjunction structure. As presented in Culicover and Jackendoff (1997: 209),
extraction from AC conjuncts is possible, whereas it is impossible from adjunct
subordinate clauses, which are islands. This contrast supports the structure
in (2.20), where neither conjunct is in an adjunct position. They cannot be
hosted in an adjunct, either, suggesting that neither (2.18a) nor (2.18b) is
likely.
In my proposal, external conjuncts are Spec elements. The fact that elements
can be extracted from external conjuncts might cast doubt on this proposal,
given the observations that subjects are Spec elements and that no element can
be extracted from subjects (the Condition on Extraction Domains, or CED;
Huang 1982). However, Takahashi (1994), Stepanov (2001), and Sabel (2002)
have argued that elements can be extracted from subjects when the subjects
remain in situ. In other words, the CED is effective only when subjects have
been raised. Returning to coordination, then, note that in all of the data above,
conjuncts are not in a derived position. Accordingly, extraction of elements out
of external conjuncts is as legal as extraction of elements out of in situ subjects.
The fact that elements may be extracted from conjuncts does not mean that
such an extraction is always possible. I will account for observed EC-like
effects, as seen in (2.43b), in Chapter 7. Nevertheless, the fact that extraction
is possible at all argues against an adjunction analysis of coordination.
2.3.5 The syntactic relation between conjuncts: conclusions

I have presented the following three arguments to show that external and internal
conjuncts are in a Spec–Complement relation, and coordinators are realizations
of the head of the projection, namely the immobility of the constituent com-
posed of a coordinator and internal conjunct (Section 2.3.2), the existence of
6 Note that the term asymmetrical coordination is used here in a semantic sense. Such coordination
is also called “fake coordination” (e.g. Szabolcsi and den Dikken 1999). Syntactically speaking,
all conjuncts are syntactically asymmetrical (see Section 2.2).
interactions between coordinators and internal conjuncts (Section 2.3.3), and

the possibility of extraction from either conjunct (Section 2.3.4).
My conclusion rules out both the structure in (2.45a) (= (2.18b)), where
the external conjunct is an adjunct, and Munn’s (1992; 1993) structure in
(2.45b) (≈ (2.18a), (2.19b)), where the sister of the external conjunct is an
adjunct.7
(2.45) a. * XP (coordinate complex)
XP
(external conjunct)
X
(coordinator) (internal conjunct)

b.
*
BP (Boolean Phrase, or &P)

(external conjunct) B
and (internal conjunct)
My conclusion does not support Moltmann’s (1992a: 52) claim that “coor-
dinators are formal adjuncts of (at least) one of the conjuncts.” This claim fails
to capture the interactions between coordinators and internal conjuncts (Sec-
tion 2.3.3). The interactions instead show a complementation relation between
coordinators and internal conjuncts.
Furthermore, in both Moltmann’s (1992a, especially p. 54 (87b)) and Munn’s
approaches, the constituents that are composed of coordinators and internal con-
juncts are regarded as maximal projections. The immobility of the constituents
(Section 2.3.2) cannot be accounted for by these approaches.8
I conclude that the relationship between external and internal conjuncts is
a Spec–Comp relationship, and that coordinators are the heads that integrate
them. Thus neither conjuncts nor coordinators are adjuncts. The Spec–Comp
relation is fundamentally different from an adjunction relation. In Chomsky’s
(2000) terminology, the former is built via Set-Merge, while the latter is built
via Pair-Merge.
7 Cowart and McDaniel’s (2008) experimental study also argues against an adjunction approach
to coordination in English.
8 Kandybowics (2005) claims that clausal coordination has an adjunction structure. See Zhang
(2006: 208) for critical comments on his arguments.
2.4 The possible modifier function of conjuncts 33
The complementation structure of coordinate complexes follows general

phrase structure. The structure is built by the regular operation Merge
(also Remerge; see Chapter 6). No special operation such as coordinate-α
(Johannessen 1998) or union of phrase markers (Goodall 1987: 36) is called
for, nor is any special mode of merger needed, such as the multiple-dimensional
phrase marker structure proposed in Muadz (1991), Moltmann (1992a), and de
Vries (2005). The syntax of coordination as second-order syntax (Lang’s 1984
term) uses no new tools that are not found in first-order syntax, in integrating
the sub-components of coordinate complexes.
One important implication of this conclusion is that coordination is not an
independent syntactic relation. Very often, coordination is contrasted with sub-
ordination (e.g. Dik 1968: 63; Huddleston and Pullum 2002: 1275). If the
definition of subordination is an adjunction relation, it is true that coordina-
tion is not subordination. Instead, it is a complementation relation (see Section
2.2.5 for our comment on challenge (b) of Dik (1968)), which is not exclu-
sive to coordination. If, however, the definition of subordination is the relation
between a projecting element and a non-projecting element, then adjuncts,
complements, and specifiers are all subordinate to their sisters. I have argued
that the two conjuncts are in Spec-Complement relation, so both conjuncts
are subordinate to the conjunction. In this sense, the asserted contrast between
subordination and coordination disappears. Again, the Spec-Complement rela-
tion is not exclusive to coordination. Furthermore, as I will show in the
next section, conjuncts are not always semantically distinct from modifiers.
In short, contrasting coordination to other syntactic relations is confusing
because it implies that it has a special structural status when it actually does
not.
2.4 The possible modifier function of conjuncts

I have argued against the adjunct status of conjuncts syntactically. Though
my focus in this book is not on semantics, in this section I briefly
state my understanding of the semantic relation between adjuncts and
conjuncts.
Consider first asymmetrical coordination, discussed in Section 2.3.4. In
(2.46a) below, the second conjunct expresses a purpose, like a purpose adver-
bial, and in (2.46b), the first conjunct expresses a “despite” meaning, again
like an adverbial (for more data like these in both English and other languages,
see Goldsmith 1985; Lakoff 1986; Kehler 2002; Blakemore and Carston 2005;
Amfo 2007):
(2.46) a. The screw whichi I’ve got to [try] and [find i ] holds the frammis to the
myolator. (Ross 1967)
b. What kind of music can you listen to and still get your work done?
It has been claimed in the literature that the apparent coordinators here are
not real, since and in (2.46a) can be replaced with the word to and and in
(2.46b) can be replaced with the adverb yet (Schachter 1977: 100; Zoerner
1995: 80; Postal 1998). However, this replacement approach is ad hoc. No
syntactic principle can tell us why the and in (2.46a) cannot be replaced by
yet and why the and in (2.46b) cannot be replaced by to. I have argued in
Section 2.3.4 that data like (2.46) are true coordination data. The “illogical”
relationship between modifiers and conjuncts is seen not only in asymmetrical
coordination constructions, but also in data like the following.9
(2.47) a. I need something like a book. (Lang 1984: 25)

b. I need a book or something.
(2.48) a. Tom left before Jane arrived. (McCawley 1988a: 262)
b. Tom left and then Jane arrived.
(2.49) a. Harry called up Bill, though he didn’t want to. (Bever et al. 1989: 335)
b. Though Harry called up Bill, he didn’t want to.
c. Harry called up Bill, but he didn’t want to.
In (2.47), the two examples are synonymous. In (2.47a), like a book is

syntactically a modifier of something, whereas in (2.47b), something and a
book are two conjuncts. The two examples in (2.48) are also synonymous,
although (2.48b) is a coordinate construction, whereas (2.48a) is not. In (2.49),
either of the two non-coordinate constructions in (2.49a) and (2.49b) can be
expressed by the coordinate construction in (2.49c) (also see Haspelmath 2007:
(74)).
Moreover, a wh-argument and a wh-adverb may also be remerged with a
coordinator to form a coordinate complex, as seen in data like (2.50) (see
Zhang 2007b for a syntactic analysis of such constructions):
(2.50) What and when does John (normally) eat?
9 It has also been claimed in the literature that aspectual come/go constructions such as (ia) are
derived from coordinate constructions such as (ib) below. However, Shopen (1971) argues against
this derivation relation (see Jaeggli and Hyams 1993: 319).
(i) a. They go visit the dentist every year.
b. They go and visit the dentist every year.
In all of these cases, the semantic differences between conjuncts and adver-
bials are blurred. The fact that conjuncts can semantically serve as modifiers
suggests that there is no sharp semantic contrast between conjuncts on the
one hand, and certain adverbials and attributives on the other. We thus should
not expect any construction-specific semantics to correlate with any (assumed)
construction-specific syntactic configuration in our analysis of coordination
complexes. This has been correctly pointed out by Dik (1968: 2):
There is no guarantee that there is a one-to-one relation between linguis-

tic structures on the one hand, and logical or psychological organization on
the other: items which are logically or psychologically “coordinated” may
be linguistically expressed in subordinative constructions, and vice versa.
Consequently, the explanation of the linguistic facts in terms of logical or
psychological phenomena almost automatically leads to a division of linguis-
tically similar items and to a grouping together of linguistically diverse ones,
on the basis of extra-linguistic considerations.
Ideas similar to Dik’s can be found in Levine (2001: 160), Carston (2002,
Chapter 3) and Amfo (2007: 682), among other places. I likewise conclude that
although conjuncts are syntactically different from adjuncts, they may function
semantically like adjuncts (modifiers). Note that this conclusion is in contrast
to Munn’s (1993: 63) claim that conjuncts are syntactically adjuncts but are
semantically different from adjuncts.10
2.5 The issue of so-called bar-level sharing

Borsley (1994: 233) claims that there are two types of sharing between con-
juncts and their hosting complexes: “a coordinate structure must not only share
categorial features with its conjuncts but must also share their bar-level.” I will
leave the issue of categorial sharing to Chapter 3. In this section, I address the
issue of so-called bar-level sharing.
The issue of constituent-level in coordinate complexes might seem to be
purely theoretical, but it does have empirical significance.
Pesetsky (1982: 440) states that conjuncts have the same bar-level as their
mother: “the constituent formed by the coordination of occurrences of Xn ,
where n is some number of bars, is itself Xn .” Since the mother and daughters
10 Not only the meaning of a conjunct can be encoded by an adjunct, the meaning of a predicate
can also be expressed by an adjunct:
(i) a. Oddly, Bill eats grass.
b. It is odd that Bill eats grass.
share the same bar-level in his analysis, the aunt of a conjunct (i.e. the sister
of the coordinate complex) should count as a sister of the conjunct for all
grammatical purposes. I illustrate this hypothesis in (2.51).
(2.51) The aunt-as-sister theory:

D C Since daughter A and mother C have the same bar-
level, and since C and D are sisters, the aunt–niece
A and B relation between D and A counts as a sister relation.
Although Pesetsky claims that “nothing important seems to follow from this
stipulation” (p. 440), some other authors do think the bar-level of conjuncts is
an important issue, and the claims made about this issue have been linked to
the analyses of the configuration of coordinate complexes.
A generally adopted assumption is that only phrases may appear as specifiers
and complements (e.g. Stowell 1981; Chomsky 1994). In a coordinate complex,
if the coordinator is the head, conjuncts must, accordingly, be phrasal, if they
are specifiers and complements. This is clearly seen in Kayne (1994). Borsley
(1994) argues that since conjuncts do not need to be phrasal, the hypothesis that
coordinate complexes have a complementation configuration is not convincing.
In this section I address Borsley’s challenge by considering word and word-part
conjuncts, which are smaller than phrases. I will show that the non-projecting
elements, specifically, specifiers or complements, do not in fact need to be
phrases. Once this theoretical issue is clarified, then, Borsley’s challenge to the
complementation hypothesis is diffused.
Let us first review Kayne’s (1994) hypothesis that conjuncts must be phrases.
Kayne (1994) claims that apparent word-level coordination involves con-
joined phrases to which a deletion process has applied (see also Wilder 1997:
63). Consider, however, examples like the following, which have been noted
since Abbott (1976), and are also discussed in Jackendoff (1977: 192). The
readings of the a-sentences are in contrast to those of the corresponding
b-sentences.
(2.52) a. Hobbs whistled and hummed a total of 16 tunes.

b. Hobbs whistled a total of 16 tunes and hummed a total of 16 tunes.
(2.53) a. Hobbs whistled and hummed the same tune.
b. Hobbs whistled the same tune and hummed the same tune.
(2.54) a. Hobbs criticized and insulted many people.
b. Hobbs criticized many people and insulted many people.
In (2.52a), for instance, there are 16 tunes involved, whereas in (2.52b), there
are 32 tunes involved. Obviously it is impossible for (2.52a) to be derived from
(2.52b) by deletion, since deletion must obey the recoverability condition on

interpretation (Chomsky 1965; 1968). Based on such data, Borsley (2005: Sec-
tion 4) correctly points out that Kayne’s deletion analysis of word coordination
is problematic. One has to admit that verbs can be conjuncts. Takano (2004)
uses similar arguments to show that verb conjuncts exist in both English and
Japanese. Bresnan and Thráinsson (1990) also argue for the existence of verb
coordination in Icelandic, independent of verb phrase coordination and ellipsis
constructions.
However, in Kayne’s (1994) theoretical framework as in ours, conjuncts
are Specifiers and Complements. Following the assumptions that Specifiers
and Complements must be phrasal, Borsley (2005) then uses the existence of
word-level conjuncts to argue against the Specifier and Complement status of
conjuncts.
My response to this argument is to distinguish constituent levels from pro-
jectivity. If we adopt Chomsky’s (1994) bare phrase structure theory, Specifiers
and Complements are simply non-projecting elements in a binary branching
representation; their sisters project, whereas they do not. The crucial property
shared by Specifiers and Complements is their non-projectability, rather than
their constituent level. We thus need to distinguish two types of dichotomy:
word vs. element larger than a word, and projecting vs. non-projecting. The
former is a morphological classification, whereas the latter is a purely syntactic
classification. The status defined by the former is stable in syntax, whereas the
status defined by the latter depends on the merger.
Accordingly, despite the fact that they are words rather than phrases, verb
conjuncts do not project within the coordinate complex, and thus they are still
regarded as non-head elements. On the other hand, the coordinate complex
formed by the verbs can be regarded as a head, if it selects a complement and
projects a VP. I list four possibilities in (2.55), and some illustrating examples
in (2.56).
(2.55)
Words or smaller elements Elements larger than a word

Projecting element e.g. coordinators in e.g. coordinate verb
coordinate complexes, such complexes, such as whistled
as and in (2.56b/c) and hummed in (2.56b)
Non-projecting e.g. word conjuncts, such e.g. phrasal conjuncts, such as
element as whistled in (2.56b) a Tibetan folk song in (2.56c)
(2.56) a. Hobbs [VP whistled [a total of 16 tunes]].

b. Hobbs [VP [V whistled and hummed] a total of 16 tunes].
c. Hobbs [VP whistled [DP Jingle Bells and a Tibetan folk song]].
In (2.56a), the verb whistled is a head, since it selects the DP [a total of

16 tunes], and projects the VP. In (2.56b), however, although it is a word,
the verb whistled alone is not a head in the coordinate complex [whistled and
hummed], since it does not project. Moreover, the whole complex, headed by
the coordinator, is also a head, since it selects the DP [a total of 16 tunes], and
projects the VP, although it is larger than a word. In (2.56c), the coordinator
projects a DP. Accordingly, the coordinator is a head in the coordinate DP
complex. The projected DP is not a projecting element when it is merged with
the verb whistled. Instead, the DP is selected by the verb, and the verb projects
a VP. Thus the verb is the head of the VP in this example.
In this approach, the notion of bar-level should be given up, and the rela-
tivity perspective of Chomsky’s (1994) bare-phrase structure must be adopted.
Whether an element is a head or not depends on whether it projects when it
merges with another element. The notions head and projection are crucial to
structure building of all levels, while the notion maximal projection (i.e. phrase)
is not (Williams 1989: 282). In this sense, I second Koopman and Szabolcsi’s
(2000) statements: “We are not assuming that the material that counts as a word
for the purpose of phonology is dominated by any particular kind of syntactic
label” (p. 223 fn. 8) and “Words do not correspond to syntactic atoms or heads”
(p. 39).11 If the notion of bar-level is not justified, the background for Pesetsky’s
(1982) theory that aunts count as sisters in coordination disappears, and so does
Borsley’s (1994) challenge to the complementation structure of coordination.
I have clarified that the phrase level of an element is not related to the syntactic
status of head or Spec/Complement. In fact, the issue of phrase level itself is far
from clear. The fuzziness can also be seen in so-called co-compounds (Wälchli
2003, see also Heycock and Zamparelli 2003: 451, fn.5), such as bow and
arrow, brother and sister, night and day. In such compounds, bare nouns are
conjoined by a conjunction, expressing a close lexico-semantic relationship
between them. Wälchli (2003: abstract) claims that “co-compounds in most
languages are intermediate between words and phrases.”
An issue related to the above discussion is the syntactic significance of
coordination of word-parts. Word-parts are not phrases but they can be
11 In Cowper and Hall (2000: 26), their approach to word-level element coordination is ascribed
to Marantz’s (1997) assumption that the same principles of composition govern phrasal, non-
phrasal, and morphological structures. Their approach is compatible with mine.
conjuncts. Such elements are again simply non-projecting elements in the

relevant coordinate complexes.
(2.57) a. to study psycho- and socio-linguistics

b. postmen and -women
c. inteligente y profunda-mente [Spanish]
‘intelligent and profoundly’
d. glücklicher-oder unglücklicherweise [German]
‘fortunately or unfortunately’
Coordinate complexes of word-parts might be considered to be Right Node
Raising (RNR) constructions, regardless of how the constructions are derived:
by movement (applying Postal 1998), special merger (Wilder 2008), deletion
(see Booij 1985; Munn 1993: 8), or incremental merger (applying Phillips
2003). Under such an approach, each conjunct in data like (2.57) is supposed
to be a word. If so, my above analysis of the word conjuncts applies here
and thus there is nothing new to say. However, Artstein (2005) argues that
the coordination is indeed at the word-part level, rather than at the word level
or higher. His conclusion suggests that the claim that bound forms are not
coordinatable is wrong (see textbooks such as Haspelmath 2002: 151 for such
a claim). His analysis seems to be intuitively natural, although he does not
mention the issue of projectivity at all.
Not only can conjuncts be part of a word, but so can entire coordinate
complexes. In the following Chamorro example, the incorporated object is a
coordinate complex (Chung and Ladusaw 2003: 87):
(2.58) Kao man-gäi-[fotgun kandit yan kahun ais] siha?

q agr-have-stove electricity and box ice they
‘Do they have an electric stove and a refrigerator?’
One also finds coordination of a word and a prefix (Huddleston and Pullum
2002: 1325):
(2.59) Please list all publications of which you were the sole or co-author.
I conclude that neither the existence of non-phrasal conjuncts nor the
existence of word-internal coordinate complexes poses any real challenge to
hypothesizing a complementation structure for coordinate complexes. Instead,
such facts may help us to get a deeper understanding of the relation between
the structure and sizes of syntactic elements.
In this section, I have examined an aspect of the projectivity issue of coor-
dinate complexes. I have showed that the bare phrase structure hypothesis,
which abandons the bar-levels of syntactic elements, provides us with a simple
analysis of word conjunct complexes. I therefore am able to maintain the con-

clusion reached in Section 2.3: coordinate complexes have a complementation
structure.
2.6 Chapter summary

The structure of coordinate complexes is not different from that of other syntac-
tic complexes. I have argued that a coordinate complex has a binary-branching
structure, and between the two conjuncts, one is internal and the other is exter-
nal to the coordinator. I also argued that the relation between external and
internal conjuncts is a Spec–Comp relation. The structure of coordination is
complementation, and coordinators are lexicalizations of the head of this com-
plementation structure. This is different from any adjunction analysis such
as Munn (1992). Moreover, the asymmetrical syntactic relation between con-
juncts is compatible with their possible asymmetrical relation in semantics.
Furthermore, I clarified that conjuncts, which are non-projecting elements in
coordinate complexes, can be of any constituency level (word-fragment, word,
phrase), and this freedom in conjunct size does not affect the complementation
structure of coordinate complexes.
My analysis shows that coordination does not introduce any new configura-
tion into the computational system. This is the first major conclusion of this
book:
No special configuration exists in coordination .
The theoretical implication of this conclusion is that unlike complemen-
tation and adjunction, the notion “coordination” is not primitive in syntactic
computation.
II No special syntactic
category
3 The categorial makeup
of coordinate complexes
3.1 Introduction
One clear fact about coordinate constructions is that the category of the whole
complex is identical to at least one of the conjuncts. The goal of this chapter is to
determine the syntactic computation behind this categorial feature unification.
I follow the traditional assumption that major parts of speech are composed
of categorial features, and the categorial features of a nominal are [+N, −V],
the categorial features of a verb are [−N, +V], the categorial features of an
adjective are [+N, +V], and so on.
In the theoretical syntax literature on coordination, it has long been proposed
that coordinators are head elements (De Groot 1949: 112, 222–223; Pesetsky
1982; Thiersch 1985; Munn 1987; 1992; 1993; Woolford 1987; Collins 1988a;
1988b; Kolb and Thiersch 1991; Anandan 1993: 38; Kayne 1994; Johannessen
1998; Zoerner 1995; see also Dik’s 1968: 53 review of De Groot’s proposal
and Progovac’s 1998a; 1998b review of later proposals). As we know, the
categorial features of a complex element are projected from the head of the
complex. However, unlike other head elements, coordinators such as and do not
have any categorial features. Thus the recognition that coordinators are head
elements does not account for the fact that the category of the whole coordinate
complex is identical to at least one of the conjuncts.
From the perspective of the structure of coordinate constructions, coordi-
nate complexes have a complementation structure in which one conjunct is in
the complement position (called the internal conjunct in Chapter 2) and the
other conjunct(s) is/are in the Specifier position(s) (called the external conjunct
in Chapter 2). By itself, however, such a structure does not account for the
categorial unification, since it does not force either the Spec of X or the com-
plement of X to share categorial features with the projected XP. As pointed
out by Borsley (1994: 227), claiming that a coordinate complex has a com-
plementation structure implies nothing about the categorial features of either
43
44 The categorial makeup of coordinate complexes
coordinate complexes or conjuncts. We thus need an account for the general-

ization that “the categorial makeup of coordinate structures must reflect that of
the conjuncts” (Borsley 1994: 226).
One common solution is to propose a new type of maximal projection, &P
(or CoP, ConjP) (e.g. Munn 1987; Zoerner 1995; Johannessen 1998; and even
textbooks such as Fromkin et al. 2007: 149). Can this &P be justified syntacti-
cally? For instance, is its distribution in contrast to that of other well-recognized
syntactic categories? I will present negative answers to these questions. Instead,
I will clarify that every coordinate complex is a regular N, or V, or A, or P
element.
Every coordinate complex must be equipped with categorial features, in
order to take part in syntactic computation (“All grammatical operations in
natural languages are category-based”; Radford 1997: 29). If a coordinator
is the head of a coordinate complex (Chapter 2), and if the coordinator does
not have intrinsic categorial features, the categorial features exhibited in the
coordinate complex as a whole must come from the conjuncts. This kind of
feature transference has been independently motivated. I will propose that the
categorial features of a coordinate complex come from just one of the conjuncts.
On the other hand, no element in a syntactic structure may have two sets
of categorial features. If a coordinator has its own intrinsic categorial features,
no transference of categorial features from a conjunct to the whole complex
occurs.
In Section 3.2, I demonstrate that in some languages, certain coordina-
tors have specific categorial requirements on conjuncts, in contrast to coordi-
nators in languages such as English, which can conjoin elements of almost
any category. In Section 3.3, I demonstrate that in the coordinate complexes
headed by the latter type of coordinators, the external conjuncts determine the
category of the whole coordinate complexes. In Section 3.4, I argue against
the construction-specific syntactic category &P. In Section 3.5, I argue against
the Clause Conjunct Hypothesis (e.g. Gleitman 1965; Tai 1969). Building on
the arguments in this and the previous chapter, I analyze the structure of coor-
dinate complexes that are composed of three conjuncts and one coordinator in
Section 3.6. Section 3.7 is a brief summary of this chapter.
3.2 The categories of coordinators and conjuncts

In this section, I address the categorial relation between coordinators and con-
juncts. I will show that cross-linguistically, coordinators may or may not c-select
conjuncts.
3.2.1 Coordinators without c-selection restrictions

The goal of this subsection is to examine the categorial relationship between
conjuncts and and-like coordinators in languages like English.
In English, conjuncts linked by and can be nearly any category and can
be both phrasal and word-level elements (the following examples are mostly
taken from Wilder 1999 and Cowper and Hall 2000), as well as word-parts (see
Chapter 2).
(3.1) Phrases:
a. NP We still need the [bat and ball].
b. NP my [friend and colleague] (Heycock and
Zamparelli 2005: 204)
c. DP [John and Mary] are coming.
d. φP (Pron) [your and her] letters; [yours and hers]
e. PP [In London and in Berlin], it is still cold.
f. AP the [red and blue] flag
g. VP Mary has [[left] and [gone to England]]
h. IP I don’t know if [[Mary left] and [Peter returned]]
i. CP [[What do you gain] and [what do you lose]]?
(3.2) Words:
a. N He is both the [father and employer] of my friend.
b. Num [[Two] and [three]] is five. (Dik 1968: 273)
c. P The events took place [in and around] Toronto.
d. A . . . both [[glad] and [sad]] about this . . .
e. V Judith [[washed] and [dried]] the towels.
f. I We both [[can] and [will]] visit her.
g. I Bill [[was] and [is]] the best tennis player in the
club.
h. C [[Can] and [will]] you do this?
i. C [[If] and [when]] she arrives, the party will begin.
Like English, many other European languages also permit conjuncts to be

words and phrases of almost any category (e.g. German und ‘and,’ discussed
in Welsche 1995: 64–65, 193; and Russian i ‘and,’ discussed in McNally 1993:
349). Outside the Indo-European family, one finds the same pattern with the
coordinators ja in Finnish, és in Hungarian, da in Georgian, at in Tagalog, and
so on (see Payne 1985: 28).
Meanwhile, it is also obvious that in all languages, any given coordinate
complex belongs to one of the well-recognized syntactic categories such as N
or V. If coordinators alone do not decide the category of a coordinate complex,
as shown by the above facts, what does? I answer this question in Section 3.3,
but first we need more background.
3.2.2 Coordinators with c-selection restrictions

We have seen that the English coordinator and coordinates conjuncts of various
categories. However, as pointed out by Payne (1985: 5), such a pattern “is by
no means universal.” For instance, coordinators in Japanese are sensitive to
the categories of the conjuncts. This sensitivity was noted by Yamada and
Igarashi (1967). Like many early generative syntacticians (Gleitman 1965; Tai
1969), they tried to derive all types of coordinate complexes from sentential or
proposition coordinate complexes, via deletion. In the “Remaining Problems”
part of their article (p. 154), they raise the question of how this Clausal Conjunct
Hypothesis explains the fact that coordinators vary according to the categories
of conjuncts. However, this “remaining problem” of theirs still remains today as
a challenge to the syntactic analysis of coordination, since the issue of category
is one of the fundamental issues of syntax (Radford 1997: 29). However, the
fact that coordinators, cross-linguistically, can be sensitive to the categories
of conjuncts has syntactic implications that have not been explored in any
framework.
Sensitivity to the categories of conjuncts is seen in both overt coordinators
and null coordinators, as I now show.
A. Overt coordinators that require conjuncts to be of specific categories

In many languages, different coordinators are used to coordinate conjuncts of
different categories. In Mandarin Chinese, for instance, the coordinators gen,
tong, yu, and ji coordinate nominals only (Chen et al. 1982: 238), whereas the
coordinators erqie and you cannot coordinate nominals. The contrast is shown
in (3.3). Moreover, the coordinator yushi ‘and thus’ coordinates clauses only
(see Chapter 2).
(3.3) a. Dai Jiaoshou xihuan he pijiu {gen/∗ you} lü-cha.

Dai Professor like drink beer and/and green-tea
‘Prof. Dai likes to drink beer and green-tea.’
b. Dai Jiaoshou shanliang {you/∗ gen} youmo.
Dai Professor kind and/and humorous
‘Prof. Dai is kind and humorous.’
Furthermore, the disjunction yaome ‘or’ cannot coordinate nominals,

whereas the disjunction huozhe ‘or’ can. The contrast is shown in (3.4).
(3.4) a. Lao Li {yaome/huozhe} zai du xiaoshuo, {yaome/huozhe}

Lao Li or/or prg read novel or/or
zai du baozhi.
prg read newspaper
‘Lao Li is reading a novel or is reading a newspaper.’
b. Lao Li zai du xiaoshuo {∗ yaome/huozhe} baozhi.

Lao Li prg read novel or/or newspaper
‘Lao Li is reading a novel or a newspaper.’
In the following table, we can see how the category compatibility between
coordinators and conjuncts is exhibited in various languages.
(3.5) Some patterns of category compatibility between coordinators and conjuncts
Language Coordinator Conjuncts Main references

Fujian kei nominals Payne (1985: 5)
ka non-nominal elements
Japanese to nominals
si finite clauses Kuno (1973)
te non-finite verbs, Adjs
Korean (k)wa nominals Martin and Lee (1986: 51)
ko clauses
Maasai oo non-clausal elements Caponigro (2003)
n clauses
Malagasy sy VPs Keenan (1976: 274)
ary clauses
Sissala ká clauses Blass (1989),
a VPs Johannessen (1998: 85)
rí (or arí) non-verbal elements
Somali iyo nominals
oo VPs Berchem (1991: 324–327)
na clauses
Turkish la nominals Haspelmath (2006: (58))
ıp VPs
Yapese ngea nominals Jensen (1977: 311–312)
(Austronesian) ma clauses
Coordinators are sensitive not only to the distinctions between major cate-
gories such as nominals and verbal elements, but also to finer syntactic classifi-
cations. For instance, in several languages, coordinators that conjoin pronouns,
which are DPs or φPs in the analysis of Déchaine and Wiltschko (2002), are
different from the coordinators that conjoin other types of nominals (Stassen
2000: 49, fn. 2). Moreover, in Gungbe, bò coordinates two IPs where the
subject of the first IP necessarily controls that of the second (Aboh 2009: Sec-
tion 2.1.1). Furthermore, haishi ‘or’ in Chinese serves to coordinate elements
related to a regardless-condition or interrogative operator, whereas huozhe ‘or’
coordinates elements elsewhere (cf. Li and Thompson 1981: 654); the Finnish
contrast between vai and tai is completely analogous (see Haspelmath 2007:
(69)). Similarly, the disjunction oo ‘or’ in Kanada can connect interrogative

clauses but not declarative clauses (Amritavalli 2003: 3). However, I will con-
centrate on the conjunct–coordinator interactions with respect to the major
syntactic categories, since interactions with respect to finer syntactic categories
do not raise any additional issues.
In this subsection, I have demonstrated that the choice of overt coordinators
can be sensitive to the categories of conjuncts.
B. Null coordinators that require conjuncts to be of specific categories

Not only overt but also null coordinators can impose categorial requirements
on conjuncts. For example, in Cayuga, a Northern Iroquoian language spoken
in Ontario, the coordinator hni’ conjoins nominals, as in (3.6a), whereas a null
coordinator conjoins verbs and clauses, as in (3.6b) (Mithun 1988: 342; also
see Johannessen 1998: 85).
(3.6) a. Ne:’ tsho: ne’ [onehe’ sahe’tá’ hni’] okwayethwe hne:’ (Cayuga)
it only the corn beans also we.planted contr
‘No, we only planted corn and beans.’
b. Tho tsho: nhe:yóht ake’tré’ atká:ta’
there only so.it.is I.drove it.stopped
‘I was just driving along and it stopped.’
Similarly, the Chadic language Háusá uses dà to conjoin DPs, whereas it
does not use any overt coordinator to conjoin verbal phrases or clauses (data
from Gazdar et al. 1985: 179 fn. 1 and Hartmann 2000: 22):
(3.7) a. Dà nı́ı́ dà kái dà shı́ı́, múú àbòòká nèè. (Háusá)
and I and you and he we friends are
‘I, you, and he, we are friends.’
b. Múúsá káá shá gı́yàà (∗ dà) káá gásà kı́ı́fı́ı́.
Musa inflPERF/3SG.MASC drink beer and infl roast fish
‘Musa drank beer and roasted fish.’
Zoerner (1995: 37) presents the following list (parataxis is another term for
null coordinator):
(3.8) Barasano: kede for coordinating NPs, parataxis elsewhere
Chemehuevi: wai for NPs, parataxis elsewhere
Tera: nde for NPs, parataxis elsewhere
Another example is Japanese, where a null coordinator cannot coordinate

conjuncts if the first one has tense-marking. In (3.9a), where the first conjunct
has a past-tense marker -ta, the coordinator sosite is obligatory (Takano 2004:
172 fn. 4; see Payne 1985: 27 for parallel data in other languages):
(3.9) a. John-ga sono ronbun-o kopiisi-ta ∗ (sosite) fairusi-ta. (Japanese)

John-nom that paper-acc copy-pst and file-pst
b. John-ga sono ronbun-o kopiisi (sosite) fairusi-ta.
John-nom that paper-acc copy and file-pst
Both: ‘John copied and filed that paper.’
Japanese also disallows the use of a null coordinator to conjoin an AP with a
relative clause. In (3.10a), the two modifiers of the nominal are both APs, and
the conjunctive -te is optional. In (3.10b), however, one modifier is an AP and
the other is a clause, and -te is obligatory (Saiki 1985: 371, 372).1
(3.10) a. akaku (-te) atsui hon (Japanese)

red -te thick book
‘the red and thick book’
b. minikuku ∗ (-te) Takashi-o nagutta otoko
ugly -te Takashi-acc hit man
‘a man who is ugly and who hit Takashi’
The contrasts between (3.9a) and (3.9b) and between (3.10a) and (3.10b)
indicate that the distributions of null coordinators do not always correspond
to whether the conjuncts are nominals or not. This fact does not support
Zoerner’s (1995) claim that the distributions are related to Case. Therefore
his conclusion that coordinators head a Case-related functional projection is
not convincing (see Bošković 2006 for a discussion of the Case-licensing of
conjuncts).
The two aspects listed in this subsection clearly indicate that coordinators,
like other types of syntactic head elements, can have restrictions on c-selection.
3.2.3 Representing the categorial dependency of coordinators on conjuncts

We have seen that coordinators such as and may occur in coordinate complexes
of any category. This fact means that such coordinators do not have any intrinsic
categorial features.
I have also presented cross-linguistic data showing that different coordina-
tors may have different categorial requirements on conjuncts. The fact that
coordinators can have such properties has generally been ignored. I mentioned
in Section 3.2.2 that it is a remaining problem in Yamada and Igarashi (1967).
In Sag et al. (1985: 133) and Gazdar et al. (1985: 179), this fact is mentioned
only in footnotes, and in Dik (1968) and Johannessen (1998), though this
aspect of coordination is discussed, it is not taken into consideration in their
1 According to Kuno (1973), -te is a gerundive marker. Saiki (1985: 371) calls -te coordination
morpheme. I will simply gloss the suffix TE.
syntactic analyses. If a coordinator has any special requirement for the category
of a conjunct, the relation between the coordinator and the conjunct must be
that between a head and its selected complement, since only head elements
exhibit c-selection restrictions on their sisters. We thus cannot adopt either
Goodall’s (1987: 32) hypothesis that coordinators are inserted at PF when the
structure of a coordinate complex is linearized, nor Moltmann’s (1992a: 52)
hypothesis that coordinators adjoin to conjuncts. The c-selection restriction
between coordinators and internal conjuncts presented here thus provides an
additional argument for the claim made in Chapter 2 that coordinators are head
elements.
3.3 The categorial makeup of coordinate complexes

Having examined the categorial relation between the components in a coordi-
nate complex, we now look at the categorial relation between these components
and the whole coordinate complex.
3.3.1 Coordinate complexes headed by and-like coordinators

A. The categorial decisiveness of external conjuncts
This section presents English data showing that the categorial features of exter-
nal conjuncts must be compatible with the c-selection of the heads that merge
with the coordinate complexes, whereas the categorical features of internal
conjuncts do not need to be.
Clausal conjuncts and prepositions The compatibility between the

category of external conjuncts with the whole coordinate complex is seen in the
complement of prepositions. (3.11b) shows that the preposition on may not take
tensed clauses as complements. However, if the complement of a preposition is
a coordinate complex, the internal conjunct can be a tensed clause. In (3.11a)
the internal conjunct is a tensed clause and the external one is a DP. (3.11c)
shows that if we switch the order of the two conjuncts in (3.11a), the result
is not acceptable. The contrast indicates that it is the external conjunct that
satisfies the c-selection of the preposition. The examples in (3.12) show the
same point.
(3.11) a. You can depend on my assistance and that he will be on time.

b. ∗ You can depend on that he will be on time.
c. ∗ You can depend on that he will be on time and my assistance.
(3.12) a. We talked about Mr. Golson’s many qualifications and that he had worked
at the White House.
b. ∗ We talked about that he had worked at the White House.
c. ∗ We talked about that he had worked at the White House and
his many qualifications.
The data in (3.11) and (3.12), first introduced by Sag et al. (1985: 165), have
previously been used in the literature to argue for the asymmetry between two
conjuncts (Munn 1996: 2; Johannessen 1998: 14).
Notice that not all prepositions allow the second conjunct of their object
to be a clause. The preposition despite, for instance, does not (Bayer
1996: 585):
∗
(3.13) Despite LaToya’s intransigence and that all the musicians quit, Michael
signed the contract.
Here we focus on the existence of the asymmetrical syntactic pattern exhib-

ited in data like (3.11a) and (3.12a), and its theoretical implications, rather than
possible constraints on the pattern implied by data like (3.13) (one may find
such constraints discussed in Whitman 2004). The crucial point here is that such
sloppiness in c-selection is never allowed for external conjuncts, as shown in
the (c) sentences above, where the external conjuncts cannot be clausal. Exter-
nal conjuncts thus must satisfy the category requirements that are imposed on
the whole coordinate complex.
Moreover, the acceptability contrast between data like (3.11a) and (3.11c)
cannot be semantic or pragmatic (contra Whitman 2004), since as Sag et al.
(1985: 166–167) noticed, the contrast disappears if the coordinate complexes
are not objects of prepositions. For instance, in a subject position, as in (3.14),
and in the position of the object of a verb, as in (3.15), the combination of a
clause and a nominal is fine:
(3.14) [[That our perspectives had changed over the years] and [the issue we
had worked on as students]] were the topics of discussion. [CP&DP]
(3.15) a. I didn’t remember until it was too late [[John’s inability to get along
with Pat], and [that he had no background in logic]]. [DP&CP]
b. I didn’t remember until it was too late [[that John had no background
in logic] and [his inability to get along with Pat]]. [CP&DP]
As we know, both clauses and nominals can be subjects, and both can be
the object of remember. Thus regardless of whether the external conjunct is
a clause or a nominal, they always satisfy the categorial requirement on the

coordinate complexes by the contexts in (3.14) and (3.15). The acceptability
contrast between (3.11c) and (3.12c) on the one hand, and (3.14) and (3.15b)
on the other hand, is explained if it is the external conjunct that must satisfy
the category requirements imposed on the coordinate complex.
Such a simple explanation cannot be achieved if we fail to see the asymmetry
between external and internal conjuncts. In order to account for the contrast
between (3.11a)/(3.12a) and (3.11b)/(3.12b), Sag et al. (1985: 166) formulate
a rule (their rule (127)) which “guarantees that in general the objects of prepo-
sitions are ‘ordinary’ NPs, rather than sentential ones. But if a prepositional
object is coordinated,” the rule “will permit a sentential NP to appear as a con-
junct.” As for the contrast between the acceptable (3.14) and the unacceptable
(3.11c)/(3.12c), another rule (their rule (132)) is formulated, which “says that
sentential NPs always follow their non-subject sisters.” Furthermore, as they
note in a footnote, ellipsis is also needed to account for (3.15b). None of these
ad hoc rules is needed in our approach.
In the discussion of (3.11) and (3.12), I view them from a categorial perspec-
tive. Alternatively, Goodall (1987: 46–49) views them from a Case-assignment
perspective. Specifically, it is generally assumed that prepositions assign Case,
and tensed clauses may not receive Case. If tensed clauses occur as the com-
plements of prepositions, as in (3.11c) and (3.12c), the result is not acceptable,
since the prepositions assign Case, but the Case may not be received by the
clause. He must therefore stipulate that in (3.11a) and (3.12a), the first conjunct,
which is a nominal, receives the Case assigned by the preposition, and then
the tensed clause does not receive Case from the preposition. I address this
stipulation at the end of Section 3.3.1B.
PP conjuncts and transitive verbs Another type of example showing

the categorial decisiveness of external conjuncts in English is illustrated below
((3.18d) is from Chris Wilder p.c. and the rest are from Grosu 1985):
(3.16) a. John devoured [DP only pork] and [PP only at home].
b. ?∗ John devoured [PP only at home] and [DP only pork].
(3.17) a. He read [DP only The Times] and [PP only on Sundays].
b. ?∗ He read [PP only on Sundays] and [DP only The Times].
(3.18) a. John writes only funny letters and only to funny people.
b. John eats the most unlikely things and at the most unlikely hours.
c. John has stolen more watches and from more unsuspecting victims than
anybody else ever will.
d. I eat [neither meat nor at restaurants].
In the acceptable sentence (3.16a), the first conjunct only pork is a nomi-
nal, satisfying the c-selection restrictions of the verb eat, whereas the second
conjunct only at home is a PP, not satisfying the c-selection restrictions of the
verb. If the two conjuncts switch their order, as in (3.16b), the sentence becomes
unacceptable. Similarly, in each of the other examples above, the two conjuncts
are of different categories, and the first conjunct must satisfy the c-selection
restrictions of the verb that selects the whole coordinate complex (note that
the focus marker only, or neither . . . nor, which also encodes a focus meaning
(Hendriks 2002, among others), is necessary in such examples; see Chapter 7
for my account).
The discussion of the data in (3.11) through (3.18) shows that in English, the
internal conjuncts do not always exhibit the expected category, although the
external ones always do.
Pollard and Sag (1994: 203) propose a Coordination Principle, which states
that the category of each conjunct is subsumed by (is an extension of) that of
the whole coordinate complex. They claim that this principle “guarantees that
whenever a syntactic environment imposes some condition on a phrase in a
given position X, that condition is respected by every conjunct of a coordinate
structure in position X.” Data like (3.19b) and (3.19d) have been assumed to
be ruled out by this principle. Specifically, the conjunct happy in (3.19b) does
not satisfy the condition on the complement of the verb remembered. Similarly,
neither of the two conjuncts in (3.19d), am walking to the store and that I left,
satisfies the conditions on the element to the right of Jessie believes Tracy.
(3.19) a. Pat remembered the appointment and that it was important to be on time.
∗
b. Pat remembered happy and that it was important to be on time.
∗
c. Pat remembered happy.
∗
d. Jessie believes Tracy [am walking to the store and that I left].
∗
e. Jessie believes Tracy [am walking to the store].
The unacceptability of (3.19b), according to our analysis, comes from the

fact that the external conjunct, happy, is not c-selected by the verb remember,
as shown in (3.19c). Similarly, the unacceptability of (3.19d), in our analysis,
comes from the fact that the external conjunct, am walking to the store, is
not c-selected by the relevant head of the small clause, as shown in (3.19e).
Therefore what Pollard and Sag try to account for is already covered without
the Coordination Principle.
Sag et al. (1985), Bayer (1996), and Whitman (2002) claim that in the
coordination of unlike categories, category contrasts between conjuncts are
neutralized. Like the approach taken in Pollard and Sag (1994), this neutral-
ization approach does not consider that the categories of the first and second
conjuncts have different syntactic effects. In our approach, no categorial fea-

tures of conjuncts are neutralized. Instead, in and-complexes the categorial
features of the external conjunct are decisive, whereas those of the internal
conjunct are syntactically invisible.
One might want to link this asymmetry in syntactic category effects to some
asymmetry of conjuncts in agreement. In the literature, it has been noted that in
certain cases the first conjunct (the external conjunct, in our terms) may deter-
mine the agreement pattern of the whole coordinate complex (Munn 1993;
Aoun et al. 1994; 1999; Johannessen 1996; 1998; among others). Consider-
ing the asymmetry of conjuncts in agreement, Borsley (1994: 242) states: “To
generate the correct structures, a linear precedence rule is necessary to ensure
that an unmarked conjunct precedes a marked conjunct, and some principle is
necessary to ensure the correct relation between the conjuncts and the coordi-
nate structure.” However, as I noted in Section 2.3.1, agreement is affected by
multiple factors, not all of which are syntactic. In particular, external conjuncts
do not actually play the decisive role in agreement patterns. Moreover, it is
not clear how agreement properties, even if they were consistent, could decide
category issues.
I have presented evidence for the decisiveness of external conjuncts in deter-
mining the category of coordinate complexes in English. This does not imply
that coordination can be legally combined anything with anything else; sen-
tences like ∗ John read books and in the park are not acceptable. Nevertheless,
with certain kinds of semantic parallelism, such as the focus parallelism sig-
naled by the reoccurrence of only in John eats only pork and only at home
(Grosu 1985: 232), a nominal can be conjoined with a PP (further discussion of
the relevant semantic constraints on coordination is given in Section 7.3.2B).
We conclude that in English, external conjuncts must satisfy the category
requirements that are imposed on the whole coordinate complexes.
B. The movement of categorial features from Specs to Heads

So far in our discussion of and-like coordinators, we have presented two facts.
First, the conjuncts linked by and can be of nearly any category, as can the
coordinate complexes headed by and (Section 3.2.1). Second, the categories
of such complexes are the same as their external conjuncts (Section 3.3.1).
The chameleon-like nature of coordinate complexes headed by and-like coor-
dinators is analyzed in different ways in the literature. Rothstein (1991a: 108)
suggests that although coordinators are heads, they do not project category
features. She claims “[t]he constituent they project has the same categorial
status as the complements, and satisfies subcategorisation frames accordingly”
(p. 103). Heycock and Zamparelli (2002: 14) also state: “we assume that CoordP
allows the transit of the categorial features of its conjuncts (in the trivial sense
that a coordination of DPs behaves as a DP, of VPs, as a VP and so forth); that
and is a head of CoordP.” Similarly, Yuasa and Sadock (2002: 89) claim that
“A coordinate constituent is one of two or more sister nodes whose categorial
information percolates to the mother node.” Moreover, as we introduced before
(see Section 3.3.1A), Pollard and Sag (1994: 203) propose a Coordination
Principle, which states that “In a coordinate structure, the CATEGORY and
NONLOCAL value of each conjunct daughter is subsumed by (is an extension
of) that of the mother.” However, none of these approaches covers the second
fact, i.e., only external conjuncts must have the same categorial features as the
coordinate complexes.
In Chapter 2, we reached the conclusion that coordinate complexes have
a complementation structure where the external conjunct and internal con-
junct are the Spec and Complement of the coordinator, respectively. Accord-
ingly, the two facts listed above can be captured by the syntactic structure
in (3.20):
(3.20) XP1 (coordinate complex)
XP2 X′1
(external conjunct)
X1 YP
(coordinator) (internal conjunct)
Since coordinators such as and may occur in coordinate complexes of any

category, such coordinators do not have any intrinsic categorial features. In
(3.20), the projection headed by a coordinator shares the category with the
external conjunct. The observed category-sharing between external conjuncts
and the coordinate complexes means that if an external conjunct is DP, the
coordinate complex must itself be DP, and the head of this latter DP is lexical-
ized by a coordinator. In (3.20) the two daughters of XP1 (XP2 and X1 ) are
identical in category; thus two elements of the same category are merged. We
call this structure the Upstairs-Twin-Structure. The Upstairs-Twin-Structure is
also seen in possessive DPs, e.g., John’s child, where the possessor DP (John)
is the Specifier of the hosting DP (see Landau 1999: 10, 13 for a recent defense
of this structure).
We now consider how the sharing of category between external conjuncts
and the coordinate complexes is syntactically implemented in (3.20).
A generally accepted assumption is that “Every word has some categorial

feature” (Collins 2002a: 43). In the previous sections, we have shown that
coordinators such as English and do not seem to have any intrinsic categorial
feature, and that their external conjuncts and coordinate complexes share cate-
gorial features. Since the heads of such complexes are realized by coordinators,
it must be that the categorial features of the external conjuncts are transferred to
the coordinators and then are projected to the whole complexes. Accordingly,
even though it is always and that shows up at the head position in (3.21), the
head of the coordinate complex is D in (3.21a), P in (3.21b), A in (3.21c), and
V in (3.21d).
(3.21) a. DP [[John] and [Mary]] are coming.

b. PP [[In London] and [in Berlin]], it is still cold.
c. AP the [[red] and [blue]] flag
d. VP Mary has [[left] and [gone to England]]
If a coordinator inherently has no categorial features (necessary for the

syntactic computation of the projected coordinate complex), it needs to get
them from somewhere. I propose that an and-like coordinator receives its
categorial features from the external conjunct. This is implemented as feature
percolation from external conjuncts to coordinators. After the transference, the
categorial features of the coordinators are projected to the coordinate complex.
Our claim that coordinators have no intrinsic categorial features and are thus
able to merge with conjuncts of various categories is parallel to the proposal
in Distributed Morphology that roots do not have categorial features, enabling
them to merge with functional morphemes of any categories (see Embick and
Noyer 2007; among others).
The proposed feature dependency of XP on the Spec of X is independently
seen in wh- and negation feature percolation. For instance, the negation feature
is percolated from Spec in (3.22a) and (3.22b), and the wh-feature is perco-
lated from Spec in (3.22c) (Grimshaw 1991; Webelhuth 1992; Koopman and
Szabolcsi 2000: 41):
(3.22) a. Nobody’s car would I borrow. [NEG feature percolation]

b. No one’s mother had baked anything. [NEG feature percolation]
c. Whose book did you read? [WH feature percolation]
If the negation in (3.22a) is sentential negation, which is able to trigger the

subject-modal inversion, the [NEG] feature must move out of the word nobody,
which is the Spec element of the possessive DP. Similarly, if the negation in
(3.22b) is the licensor of the negative polarity item any, the [NEG] feature
must move out of the word no one, which is the Spec element of the possessive
DP, in order to c-command the word any. Likewise, in (3.22c), in order to
check the [WH] feature of C, the relevant feature must move out of the word
whose, which is the Spec element of the possessive DP. In addition to these
features, Winter (2001: 177 fn. 29) mentions that some semantic features of a
possessive DP can be decided by the Specifier (possessor) in certain cases (for
more discussion of feature percolation, see Cole et al. 1993; López 2001).
The hypothesis that there is a syntactic relation between external conjuncts
and and-like coordinators with respect to categorial features links together
the two facts introduced in this subsection: first, conjuncts linked by and, as
well as coordinate complexes headed by and, can be of nearly any category,
and second, the categories of such complexes are the same as their external
conjuncts. In contrast to this categorial feature approach, Goodall’s (1987: 49)
Case-assignment approach to the data introduced in Section 3.3.1 does not
make such a link. Goodall (1987: 49) admits:
Although this analysis gives us an answer to our original question about (98)
[= (3.11a) and (3.12a) here – NZ], it also raises several new questions we had
not previously considered. The most fundamental of these is why it is that
Case-assignment can be blocked in coordination. Similarly, we would want to
know how this affects the operation of the Case Filter and the θ-criterion. I do
not believe that the model of coordination I am developing here shed any new
light on these questions, so I will simply note that they form an interesting
area of investigation that is worth pursuing.
3.3.2 Categorial features of coordinators that have c-selection restrictions

In this subsection, I show that Chinese coordinators such as he, gen, and erqie
not only have c-selection restrictions, but also have certain intrinsic categorial
features.
It is generally assumed that the c-selection of x must be satisfied by the
complement of x, rather than by the Spec of x. Moreover, c-selection must be
satisfied only once. Thus, it is impossible for all conjuncts to be subject to the
same c-selection.
We have now learned that two conjuncts may have different categorial fea-
tures. If a coordinator has c-selection restrictions, the category of its comple-
ment, i.e., the internal conjunct, is fixed. However, the category of its specifier,
i.e., the external argument, is not determined by any element. If a coordinator
has no intrinsic categorial feature, it may get any categorial features from the
external conjunct, and those features will decide the category of the whole
coordinate complex.
(3.23) a. Wo renwei {∗ yi ge tiancai / Baoyu yexinbobo}

I think one cl genius Baoyu ambitious
‘I think {∗ a genius / that Baoyu is ambitious}.’
b. ∗ Wo renwei [Baoyu yexinbobo] gen [yi ge tiancai]
I think Baoyu ambitious and one cl genius
The example in (3.23a) indicates that the verb renwei ‘think’ selects a clause
exclusively and disallows a nominal complement. The coordinator gen c-selects
a nominal exclusively and disallows a clause (see (3.3)). In (3.23b), the c-
selection of gen is satisfied. If gen had no intrinsic categorial features, the
non-nominal categorial feature of the first conjunct [Baoyu yexin bobo] would
be percolated to gen and then be projected to the whole coordinate complex,
satisfying the c-selection of renwei. However, (3.23b) is not acceptable.
Coordinate complexes headed by he/gen may be merged only with elements
that select nominals. Verbs such as renshi ‘know’ and jiao ‘teach’ select a
nominal exclusively and disallow a clausal complement, as shown in (3.24a).
The coordinate headed by he/gen may occur as the complement of renshi, as
shown in (3.24b).
(3.24) a. Wo renshi {yi ge xiaoshuojia / ∗ Baoyu yexinbobo}

I know one cl novelist Baoyu ambitious
‘I know {a novelist / that Baoyu is ambitious}.’
b. Wo renshi [yi ge xiaoshuojia gen yi ge yinyuejia].
I know one cl novelist and one cl musician
‘I know a novelist and a musician.’
The above discussion shows that he/gen has intrinsic nominal features.
The following examples show that the distributions of coordinate complexes
headed by erqie ‘and’ show just the opposite pattern:
(3.25) a. Wo renshi [yi ge xiaoshuojia {he/ gen/∗ erqie} yi ge yinyuejia].

I know one cl novelist and/ and/ and one cl musician
b. Wo renwei [Baoyu yexinbobo {erqie/∗ he/∗ gen} ta hen youqian].
I think Baoyou ambitious and / and/ and he very rich
‘I think that Baoyu is ambitious and that he is very rich.’
Such data show that coordinate complexes headed by erqie are not nominals,
and erqie must have intrinsic verbal categorial features.
The intrinsic categorial features of he/gen and erqie can also be seen in
(3.26).
(3.26) a. Baoyu xiande {gaoda/youqizhi/∗ qizhi}.

Baoyu seem tall/stylish/style
‘Baoyu seems to be {tall/stylish/∗ style}.’
b. Baoyu xiande [gaoda erqie youqizhi]. [AP & AP]

Baoyu seem tall and stylish
‘Baoyu seems to be tall and stylish.’
c. ∗ Baoyu xiande [jiankang-qingkuang he qizhi]. [NP & NP]
Baoyu seem health-condition and style
d. Baoyu guanxin [jiankang-qingkuang he qizhi]. [NP & NP]

Baoyu care.about health-condition and style
‘Baoyu cares about health condition and style.’
The example in (3.26a) shows that the verb xiande ‘seem, look like’
c-selects adjectives gaoda ‘tall’ and youqizhi ‘stylish’ but not the nominal
qizhi ‘style.’ The coordinate complex headed by erqie in (3.26b) satisfies the
c-selection of the verb, indicating that erqie does not have nominal features.
The coordinate complex headed by he in (3.26c) does not satisfy the c-selection
of the verb, although jiankang-qingkuang he qizhi ‘health condition and style’
is a well-formed coordinate complex itself, as seen in (3.26d). (3.26c) thus
indicates that he does not have adjectival features.
3.3.3 Categorial unification in Spec-Head and Head-Compl relations

We have argued that coordinate complexes headed by and have the Upstairs-
Twin-Structure in (3.20), repeated here as (3.27a). Another possible structure
that can be built by the computional system is the Downstairs-Twin-Structure in
(3.27b). In the latter structure, the category of a complex is identical to that of the
complement. For example, in Larson’s (1988) VP-shell structure, the head and
its complement are identical in category. Similarly, McCloskey (1999) proposes
a structure where C selects another CP, to represent the double that constructions
in Standard English (e.g. He thinks that if you are in a bilingual classroom that
you will not be encouraged to learn English). (3.27a) and (3.27b) are different in
that the former represents uniformity of category in the immediate constituents
of the top (or final) merger level of the syntactic complex, whereas the latter
exhibits uniformity of category in the immediate constituents of the next lower
(or earlier) merger level of the complex. A third type of structure that can be
built by the computational system is the well-recognized (3.27c).
(3.27) a. XP1 b. XP1 c. XP
XP2 X′1 cat-sharing YP X′1 YP X′
X1 XP2 cat-sharing X ZP
X1 YP
The Upstairs-Twin-Structure The Downstairs-Twin-Structure
e.g. possessive DP e.g. Larson’s VP-shell,
McCloskey’s double CP structure
This research thus shows that there is no construction-specific syntactic

category in coordination. The top projection in (3.27a), i.e., XP1 , is not an
&P, or any other new type of functional category. It is simply the same as
whatever the external conjunct (XP2 ) is. Some empirical problems of &P will
be presented in Section 3.4.
3.4 Against &P

According to our approach, the term coordinator is a convenient name for a
group of forms that realize the heads of coordinate complexes. Such words do
not have a (complete) set of categorical features intrinsically. By contrast, some
have argued for a special coordination-specific category like &P (or CoP for
Coordination Phrase, or BP for Boolean Phrase) in the literature (Munn 1987;
Zoerner 1995; Johannessen 1998; de Vries 2005; among others). Although the
notion of &P has been repeatedly criticized (Wilder 1999; Sag 2000: 8; and
Borsley 1994; 2005), the criticisms have been ineffective since the crucial issue
of the categorial makeup of coordinate complexes was not solved. The results
described in the previous sections now enable us to provide a more conclusive
argument against &P.
The c-selection restriction presented in this chapter, along with the interac-
tions between coordinators and internal conjuncts presented in the last chapter
(Section 2.3.3), provide substantial support for the claim that coordinators are
head elements, consistent with claims long advocated in the literature (see De
Groot 1949: 112, 222–223; Pesetsky 1982; Thiersch 1985; Munn 1987; 1992;
1993; Woolford 1987; Collins 1988a; 1988b; Kolb and Thiersch 1991; Anan-
dan 1993: 38; Kayne 1994; Johannessen 1996; 1998; Zoerner 1995). However,
the claim that coordinators are head elements is different from the claim that
coordinators form an independent functional category. In this section, we argue
against the category of &P.2
2 Camacho (1997) provides an alternative way of avoiding &P. In consideration of the free
categorial environment of and, Camacho (1997) assumes that and in the sentence Tom and Mary
saw the movie is simply in T (or I), with the following constituency:
3.4 Against &P 61
3.4.1 The distributions of coordinate complexes are covered by simplexes

It has been claimed that &P is the category of coordinate complexes, and
coordinators such as and and or are functional categories, independent of other
categories such as N, V, and A. In this subsection, I show that the distributions
of coordinate complexes are covered by simplexes, and that the categorial
contrasts of various coordinate complexes are covered by those of simplexes.
Therefore, coordinate complexes cannot be an independent category.
A. The syntactic positions of coordinate complexes are covered by
those of simplexes
No syntactic position is found to be taken by coordinate complexes exclusively.
For example, verbs such as compare select plural nominals, and the selected
element can be either a coordinate complex or a simplex plural nominal. Since
the classification of syntactic elements into syntactic categories is decided by
the syntactic distributions of the elements, the identification of any new category
must be based on the existence of a new distribution. Coordinate complexes do
not pass this test.
In fact, the distributions of coordinate complexes are not only covered by,
but are also more restricted than, simplex elements. The restriction is seen in
aspects like the following.
First, plurals can be subjects of quantity-denoting predicates and occur with
quantity-related modifiers such as a total of or its Chinese counterpart yigong
‘in total’, whereas coordinate complexes cannot ((3.28) is cited from Dougherty
1970b: 854).
(3.28) a. The men were {numerous/plentiful/interspersed}.

b. ∗ Jack, Bill, and Tom were {numerous/plentiful/interspersed}.
(3.29) a. a total of two persons

b. ∗ a total of John and Tom
(3.30) a. yigong liang ge ren

total two cl person
“a total of two persons”
b. ∗ yigong Baoyu gen Daiyu
total Baoyu and Daiyu
(i) [TP Tom [T’ and [TP Mary [T [VP saw the movie]]]]
However, Camacho realizes that in (i), the coordinated elements Tom and Mary do not form a
constituent, which as pointed out by Progovac (1998b: 5), counts against this analysis. In our
approach, it is also possible for and to be in T in certain contexts, but only if the external conjunct
is a T-element, as in John [was and is] the best tennis player in the club, or if the external conjunct
is a TP, as in I don’t know if [Mary left and Peter returned].
Second, plurals can occur in the partitive construction one of __, whereas
coordinate complexes cannot. The following data are cited from Büring (2002:
7):
(3.31) a. one of {us/the detectives}

b. ∗ one of {you and me/Schimansky and Tanner}
Third, plurals can occur in the partitive construction half of __, whereas
coordinate complexes cannot. The following data are cited from Dougherty
(1970b: 855).
(3.32) a. Half of the quartet wanted to quit.

b. ∗ Half of Mary, Sue, Jane, and Sally wanted to quit.
All of these examples show that if an element can be merged with a coordinate
nominal complex, it can also be merged with a plural simplex, but not vice
versa. The distributions of coordinate complexes are more restricted and thus
fully covered by those of simplexes. The coverage relation means that such
complexes do not represent a separate syntactic category in the computational
system.
B. The category contrasts of coordinate complexes are covered by

those of simplexes
If coordinate complexes could neutralize the category contrasts of other ele-
ments, the following four coordinate complexes would form a natural class in
merging with verbs, but they do not.
(3.33) a. [DP & DP]

b. [VP & VP]
c. [CP & CP]
d. [AP & AP]
A well-established fact about c-selection is that it is local. The complex of

(3.33a) can be selected by a transitive verb because transitive verbs c-select for
DPs and because, in our approach, the whole complex is a DP. By contrast, if a
coordinator projected an &P with its own categorial features, a transitive verb
could not merge with the complex of (3.33a). If we try to solve the problem
by positing that a transitive verb c-selects &P as well as DP, the verb should
also be able to merge with the complexes of (3.33b), (3.33c), and (3.33d), since
they are all &Ps. However, no such verbs have been found.
Since coordinate complexes do not show a distribution different from those
of the currently recognized categories, nor do they “bleach” contrasts in the
3.4 Against &P 63
currently recognized categories, they cannot represent an independent category.

Therefore, the assumed category &P does not exist in syntax.
3.4.2 Neither closed classes nor case inflection argue for &P
Here we examine two proposed arguments for the projection of &P: the closed-
class nature of coordinators and case inflection patterns of conjuncts.
A. Coordinators form a closed class

It has been claimed that because coordinators form a closed class and are
unstressed elements, they therefore can head an independent projection, &P
(Johannessen 1998). However, not all closed-class and unstressed elements
head independent projections. For instance, demonstratives, pronouns, and
Chinese locatives do not need stress, and they are all closed-class elements.
Nevertheless, they are hosted by DPs or NPs; there is no DemP, PronP, or LocP.
Similarly, pronouns have been argued to be proforms of DPs, ϕPs, and NPs
(Déchaine and Wiltschko 2002). Therefore, stress and class size cannot be used
as arguments for the existence of &P.
B. Coordinators and Case

It has been claimed that coordinators are Case-assignors, and thus &P is a
Case-related functional category (e.g. Zoerner 1995). For instance, in (3.34),
the external conjunct me is accusative, which is not expected to be true of the
whole coordinate complex in the subject position. The same point holds for the
internal conjunct me in (3.35) and both conjuncts in (3.36).3
(3.34) Me and Robin left.
(3.35) a. Robin and me left.

b. Huck and me ain’t cry-babies. (Mark Twain, Tom Sawyer, Chapter 16)
(3.36) a. Me and them left.

b. Him and me drinks nought but water. (Chapman 1995: 37)
The existence of such data does not mean that the coordinator and has any
intrinsic feature related to Case. We need to clarify that abstract Case encodes
a mutual checking (or licensing) relation between a nominal and a functional
head. For instance, the Nominative Case of a nominal and the Case feature of T
3 See Progovac (1998a) for an overview of the diversity of case-marking of conjuncts both within
and across languages, and Progovac (1998b) for a review of some approaches to this issue in the
literature.
check each other, and the Accusative Case of a nominal and the Case feature of
v (or AgrO, Tr) check each other.4 Non-nominal conjuncts do not have Case fea-
tures. If coordinators must have a Case-relation with conjuncts, the derivation
of coordinate non-nominal complexes will crash, since the Case-feature of the
coordinators cannot be checked. Zoerner (1995: 45 fn. 2) notices this dilemma,
but declares that the use of a coordinator in non-nominal coordinate complexes
“reflects the &0 ’s use as a discourse marker,” rather than a Case-assignor.
Since he does not explain what this “discourse marker” status of coordinators
means, this response does not help support the claimed Case-assignor status of
coordinators.
One also needs to distinguish morphological case from abstract Case (see
McFadden 2004 and Legate 2008 for recent discussions), which we under-
stand to be a formal licensing relationship between a nominal and a functional
head. Although I am not making any claim regarding the morphological cases
of pronominal conjuncts (see Schütze 2001: Section 5.2 for a default case
approach; also Bošković 2006), I claim that the coordinator and does not have
any intrinsic feature related to Case. Accordingly, the issue of Case cannot be
used as an argument for &P.
3.4.3 Retrospection
Based on the facts and arguments in this section, I conclude that coordinators
neither bleach the category features of the relevant coordinate complexes, nor
create any special category. Thus both of the following two assumptions are
problematic: that there is a category-neutral &P, and that &P is categorically
contrastive to other categories, such as N and V.
One needs to distinguish between two issues: whether coordinators can func-
tion as head elements, and whether they should be classified as an independent
functional category. One example to illustrate the difference is the word do
in English do-support. The word do can be a realization of T (or some other
functional projection, see Laka 1994), but it does not head an independent DoP,
in contrast to other functional and lexical categories. Similarly, we claim that
coordinators are realizations of lexical or functional elements, but they do not
constitute an independent syntactic category.
Grootveld (1994: 31) claims that coordinators head a functional projection
that is categorially non-distinct from the head’s complement, assuming all con-
juncts are such complements. Zoerner (1995: 19) claims that “An &P lacks
4 In Chomsky (2000), Case features of nominals are checked or erased by the phi-features of
functional categories under matching. This technical change does not affect the statement here.
Both types of features are still uninterpretable, and still must be checked or erased mutually.
inherent features such as [±V] or [±N], which would limit the number of sites
in which it could licitly surface; rather, it assumes the feature specification of
its conjuncts.” In Johannessen (1998), although CoP is claimed to be an inde-
pendent category, “CoP get[s] its major category feature from one conjunct”
(p. 112, also pp. 164, 168, 169). We can see that in these analyses, it makes
no empirical difference if one gives up the label of CoP or &P and uses the
category of a certain conjunct instead.
The &P proposal has been repeatedly criticized in the literature. Abney
(1991: 225) correctly points out that coordinators such as and do not show
any specific selectional properties. Based on the fact that coordinate complexes
have no fixed position and the so-called &P is never selected for, Wilder (1999:
Section 2) states that “analyzing the coordination constituent as a projection of
& is problematic.” Similar comments have been made by Sag (2000: 8) and
Borsley (2005).
Traditionally, coordinators such as and have been regarded as “empty words.”
In Hockett (1958: 153), coordinators “serve not directly as carriers of meaning,
but only as markers of the structural relationships between other forms.” In
Moltmann (1992a: 32), “‘coordinator’ does not denote a syntactic function.”
Blümel (1914: 52) even declares “Strictly speaking . . . such words should be
excluded from dictionaries” (see Lang 1984: 67 for a review). Following Chom-
sky’s (2007: 11) idea that a lexical item may have an edge-feature, which permits
it to be merged with another syntactic element, we assume that it is the edge-
features of coordinators that permit them to be merged with conjuncts. It is
plausible to suppose that coordinators have other features as well. For instance,
when a conjunction is merged with two individual-denoting conjuncts, it may
project a plural feature, even when each of the conjunct is singular (see Section
5.2.3B for further discussion of the issue); a disjunction such as or projects a
disjunctive feature, and adversative coordinators such as but projects an adver-
sative feature. However, and does not have intrinsic categorial features, and it
gets the features from the external conjuncts.
It is well recognized that not every part of speech corresponds to an inde-
pendent syntactic category. Pronouns, demonstratives, articles, and expletives
are all grouped into D-elements. Conjunction is also a part of speech, but this
does not mean that it must be an independent syntactic category.
3.5 Against the Clausal Conjunct Hypothesis

The fact that conjuncts can be any category and any category-level (see Section
2.5) is in conflict with Gleitman’s (1965) and Tai’s (1969) Clausal Conjunct
Hypothesis (CCH), which states that all conjuncts are clausal and that apparent
non-clausal conjuncts are the result of reduction. Since the CCH is still alive
(George 1980; Goodall 1987; Hoekstra 1994: 296; Johannessen 1998; Aoun et
al. 1994; Aoun and Benmamoun 1999; Camacho 2003; and Schein forthcom-
ing), in this subsection we provide further arguments against it.
The CCH requires two crucial operations. The first one is deletion, called
Tranformation of Coordination-Reduction, which is used to derive phrasal
coordination constructions. According to the CCH, (3.37a) is derived from
(3.37b).
(3.37) a. Louise and George rode bicycles.

b. Louise rode bicycles and George rode bicycles.
The second operation, as mentioned by Tai (1969: 144) and Goodall (1987:
18), is regrouping, which combines, for instance, Tom and Jane in (3.38) at
a certain step of the derivation, so that the verb is changed from the singular
form eats, as in the assumed source sentence (3.38b), into the plural form eat
in (3.38a):
(3.38) a. Tom and Jane eat bread and crackers respectively.

b. Tom eats bread and Jane eats crackers.
Lakoff and Peters (1966) argue against the CCH, observing that the CCH
cannot, for instance, account for the contrast between (3.39a) and (3.39b):
(3.39) a. John and Mary are erudite.

(Reading = John is erudite and Mary is erudite.)
b. John and Mary are alike.
(Reading = ∗ John is alike and Mary is alike.)
Lakoff and Peters note that if all coordinate complexes are derived from
clausal coordinate complexes, the source of (3.39b) must be an ungrammatical
one. Accordingly, the CCH must be wrong. Instead, they analyze (3.39a) as
clausal coordination and (3.39b) as phrasal coordination. Tai (1969) defends
the CCH, claiming that in data like (3.39b) the transformation of Coordination-
Reduction is obligatory. Conceptually speaking, the CCH makes the syntactic
derivation more complicated, since it requires both ungrammatical underlying
forms and a rule to make them grammatical. In any analysis without the CCH,
including ours, the coordination in (3.39a) is distributive and the one in (3.39b)
is collective. Both are DP coordinate complexes, and no clausal coordination
is involved.
The CCH has been extensively discussed in Dik (1968), Lang (1984: 86),
Goodall (1987: 18), McCawley (1968; 1988a: 277), Munn (1993), Wilder
(1994; 1997; 1999: 9), among many other places. There are still some recent
advocates of the CCH, as listed above, but McCawley (1968; see Dougherty
1970a: 532), Dougherty (1970b: 857), Wilder (1999: 13) and Eggert (2000)
present arguments against the CCH. Parallel to these works, Schwarzschild
(2001) also casts doubts on the CCH (t-only-conjunction in his terms) from a
semantic point of view.
However, although no clausal coordination is involved in data like (3.39), and
thus no conjunct reduction, I admit that some apparent nominal coordinate com-
plexes are indeed the result of reduction. The following example from Collins
(1988a; 1988b) truly looks like the result of RNR of two clausal conjuncts:
(3.40) Perhaps John, maybe Mary, and certainly Bill went to the store.
In (3.40), each conjunct is modified by an adverb. The adverbs modify

predicates, including nominal predicates, as in (3.41), and appositive nominals,
which are also predicates, as in (3.42). But they cannot modify individual-
denoting nominals, as shown in (3.43) (Munn 1993: 160; Schein forthcoming):
(3.41) a. John is perhaps a doctor.
b. It was perhaps John that I met.
(3.42) a. I met someone, possibly a priest, yesterday.

b. Someone, possibly Bill, came into the room.
(3.43) a. ∗ I met [possibly a priest] yesterday.

b. ∗ [Possibly someone] came into the room.
It thus seems that data like (3.40) have a clausal coordination source, semanti-
cally and syntactically (via RNR) (see Schein forthcoming). One new argument
for the reduction analysis of such examples is that if a conjunction cannot con-
join clauses, like he in Chinese, it cannot occur with any modal adverb in the
parallel coordinate complex:
(3.44) a. [Baoyu he (∗ yexu) Daiyu] qu-le Shanghai.

Baoyu and perhaps Daiyu go-prf Shanghai
‘Baoyu and Daiyu have gone to Shanghai.’
b. (Yexu) [Baoyu, Daiyu, he Yuanyang] dou qu-le Shanghai.
Perhaps Baoyu Daiyu and Yuanyang all go-prf Shanghai
‘(Perhaps) Baoyu, Daiyu, and Yuanyang all have gone to Shanghai.’
c. ∗ Yexu Baoyu, dagai Daiyu, he kending Yuanyang dou
Perhaps Baoyu probably Daiyu and certainly Yuanyang all
qu-le Shanghai.
go-prf Shanghai
In (3.44a), the modal adverb yexu ‘perhaps’ cannot occur between the two
conjuncts of the subject linked by he. This constraint suggests that real nominal
coordinate complexes exclude such adverbs, which in turn suggests that in
examples like (3.40), there is a clausal rather than nominal coordinate complex.
In (3.44b), the adverb occurs outside the coordinate complex, and thus it has
no interaction with the coordinate complex headed by he. In (3.44c), the modal
adverb dagai ‘probably’ and kending ‘certainly’ both occur inside the subject
coordinate complex headed by he, and as in (3.44a), this is not acceptable. The
interactions between the adverbs and the coordinate complexes headed by he
show that the occurrence of such adverbs in examples like (3.40) may indicate
that the coordination is a clausal one (see Section 5.2.3C for some other cases
of reduction of clausal conjuncts).
However, not all conjuncts are derived from clauses. No matter how appeal-
ing the CCH might be from a certain semantic point of view (e.g. Schein,
forthcoming), it cannot capture certain syntactic facts.5
First of all, in many languages, clausal and non-clausal coordination may use
different coordinators (see Section 3.2.2). In Mandarin Chinese, for instance,
ji and gen coordinate nominals only, and never coordinate clauses (Chen et al.
1982: 238), nor do the Japanese conjunction -to and the Shanghainese conjunc-
tion teq-. In contrast, Chinese erqie can coordinate verbal phrases, including
clauses, but it cannot coordinate any nominals. Moreover, the conjunction yushi
‘and thus’ coordinates only clauses (Chen et al. 1982: 515). If all conjuncts
are, or are derived from, clausal conjuncts, it is hard to explain the syntactic
distributions of these conjunctions.
Second, the CCH developed from the study of the distributive type of coor-
dination, such as (3.37a). According to the CCH, data like (3.37a) are the result
of a reduction of clausal conjuncts. In Chinese, however, the conjunction ji can
occur only in the distributive coordination of DPs (see Chapter 5) and is not able
to conjoin clauses. This is inconsistent with the CCH, which would wrongly
predict that all ji-coordinate complexes were derived from syntactically illegal
representations where ji conjoined clauses.
Third, if sentences like (3.45a) were derived from two clausal conjuncts, and
if the predicate in the bathroom and the predicate in the living room were in the
two clauses respectively, it would be impossible to identify the exact form of
5 Arguments against the CCH can also be seen in Vicente’s (to appear) study of but. He shows that
corrective but, as in (i), always requires clausal coordination, followed by PF deletion, whereas
contrastive but, as in (ii), may conjoin elements of other categories.
(i) Helen didn’t eat one apple but three bananas.
(ii) Helen ate one apple but three bananas.
3.6 The complexes of more than two conjuncts 69
the subject for each of the predicates ((3.45a) is from Heycock and Zamparelli
2002 and (3.45b) is from Krifka 1990: 168).
(3.45) a. The children are in the bathroom and in the living room.
b. The planes flew above and below the clouds.
Certainly the children cannot be the subject of either predicate in (3.45a).

Only some of the children are in the bathroom while others are in the living
room, and the sentence does not tell us how many are in each group. The chil-
dren thus does not have a syntactic position in the assumed two source clauses.
If we give up the CCH, however, the syntactic representation of (3.45) is clear.
The subject here is a simplex DP in its plural morphology. The predicate is a
coordinate PP complex. The subject is base-generated higher than the coordi-
nate PP complex. The copular are, which is required for non-verbal primary
predication in English, agrees with the plural subject. There is only one clause.
No clause-reduction occurs. (3.45b) presents the same challenge to the CCH.
Finally, as pointed out by Schmerling (1975: 220–221, cited by Sag et al.
1985: 151), AC constructions are not synonymous to the assumed unreduced
form. The AC in (3.46a) encodes a single event whereas (3.46b) may encode
two events.
(3.46) a. I went to the store and bought some whiskey.
b. I went to the store and I bought some whiskey.
AC data are not distributive (see Section 5.2.2). Data like (3.46a) are not
compatible with the CCH.
I conclude that the CCH cannot be syntactically right, in the generally
accepted syntactic framework.6,7
3.6 The structure of coordinate complexes composed of more

than two conjuncts
In this section, I present an analysis of the syntax of coordinate complexes that
are composed of three or more conjuncts and a single coordinator. Following
Winter (2006: 3), I call such complexes multiple coordinate complexes. I will
not discuss complexes that are composed of three or more conjuncts and two or
more coordinators, called repeated coordinate complexes in Winter (2006: 5).
6 In advocating the CCH, Schein (forthcoming) assumes that all theta roles are propositional. This
is not a generally accepted assumption.
7 Partee (2005: Section 2.2) states that the CCH also caused problems for the ordering of various
transformational rules proposed in the early seventies.
Repeated coordinate complexes are ambiguous and allow sub-grouping into

the basic types that we are covering. The differences between multiple and
repeated coordinate complexes presented by both Borsley (2005) and Winter
(2006) falsify Payne (1985: 19), Goodall (1987: 32, 33), and Moltmann’s
(1992b: 54) claim that the repetition of coordinators has no syntactic or semantic
effect.
Gazdar et al. (1985: 170) state that “no language can have, say, a three-place
coordinator morpheme. If there were one in English, pronounced triand, we
would find Kim, Sandy, triand Lee grammatical, but not ∗ Kim triand Sandy or
∗
Kim, Sandy, Lee, triand Tracy. We know of nothing like this in any attested
language.” Similarly, Zwart (1995: 12) says “I know of no language where
multiple coordination is actually obligatory.” The common implication of these
quotes is that unlike binary coordination, three-place coordination is not a
primary type of coordination.
Like binary complexes, multiple coordinate complexes allow both distribu-
tive and collective readings. For instance, (3.47a) is collective, (3.47b) is dis-
tributive, and (3.47c) is ambiguous between the two readings.
(3.47) a. John, Bill, and Tom collided. (Dougherty 1970b: 855 (39b))
b. John, Bill, and Tom knew the answer.
c. Hobbs, Rhodes and Barnes lifted the rock. (Borsley 2005)
One similarity between multiple coordinate complexes and the binary com-
plexes discussed so far is that extraction from one or more conjuncts is possible.
Examples like (3.48) are found in Lakoff (1986):
(3.48) What did he go to the store, buy, load in his car, drive home and unload?
The extractability indicates that none of the conjuncts of multiple coordinate
complexes is an adjunct. Thus we need to consider the following three analyses:
the flat multiple-branching structure (Ross 1967: 88–92; Jackendoff 1977: 51,
among many others) as in (3.49b), the multiple Spec structure as in (3.49c), and
a layered complementation Downstairs-Twin-Structure, as in (3.49d) (see the
&P-shell structure in Kayne 1994: 57; Zoerner 1995; and Johannessen 1998:
144).
(3.49) a. John Mary and Bill
b.
DP
DP DP and DP
John Mary Bill

c. DP1 d. DP5
DP4 DP4 D′5

DP3 D′1
John John D5 DP1
Mary D1 DP2
DP3 D′1
and Bill
Mary D1 DP2
and Bill
In this section, I argue against the flat structure analysis in (3.49b), and
adopt Borsley’s (2005) arguments against (3.49d). Consequently, (3.49c) is
the only choice. The configuration in (3.49c) is also seen in other construc-
tions cross-linguistically, and thus again does not create a coordination-specific
configuration.
3.6.1 The coordinator must be grouped with an edge conjunct

My first argument against the flat structure is that the solitary coordinator in
a multiple coordinate complex must be grouped with an edge conjunct. Dik
(1968: 41, 58) makes the generalization that “if only one coordinator is present,
its place is determined, generally, by the last member of the series,” and “an
almost universal rule puts the coordinator before or after the last member of
the series.”
(3.50) a. apples, oranges, and bananas

∗
b. apples, and oranges, bananas
∗
c. and apples, oranges, bananas
d. I’d like it to be (∗ or) black, (or) white, or grey.
e. Mary was fun to tease, easy to please, and known to have fleas.
(3.51) a. Yu, rou, gen shucai dou yao chi. (Chinese)

fish meat and vegetable all must eat
‘One must eat fish, meat and vegetables.’
b. ∗ Yu, gen rou, shucai dou yao chi.

fish and meat vegetable all must eat
c. ∗ gen yu, rou, shucai dou yao chi.
and fish meat vegetable all must eat
(3.52) (I) Marija, (i) Milan, i Petar studiraju lingvistiku. (Serbo-Croatian)

and Mary and Milan and Peter study linguistics
‘Mary, Milan, and Peter are students of linguistics.’
In (3.50a), which is cited from Dougherty (1970b: 853 (20)), the coordinator
and precedes the last conjunct. In (3.50b) and (3.50c), however, it does not,
and the two sentences are not acceptable. The same contrast is seen in other
examples in (3.50) and (3.51). The Serbo-Croatian example in (3.52) shows
that the first two occurrences of the coordinator i is optional, in the presence of
the last one. All of these data confirm that the last conjunct hosts the coordinator
in these languages.
3.6.2 The category decisiveness of non-final conjuncts in English

My second argument against the flat structure is that in English, all of the
non-final conjuncts determine the category of the whole coordinate com-
plex. We saw in Section 3.3.1 that only the category of the final conjunct
is allowed to be different from the category of the whole coordinate complex
in English. Now we present the fact that this is also true in multiple coordinate
complexes.
(3.53) a. You can depend on my assistance, John’s cooking skill, and that Mary will
be on time.
b. ∗ You can depend on my assistance, that Mary will be on time, and John’s
cooking skill.
c. ∗ You can depend on that Mary will be on time, my assistance, and John’s
cooking skill.
In (3.53a), the whole coordinate complex is selected by the preposition

on. In the complex, the first and second conjuncts are both DPs, which are
compatible with the c-selection of the preposition, whereas the last conjunct
is a clause, which is not compatible with the c-selection of the preposition.
However, the sentence is acceptable. If we change the order of the three con-
juncts, as seen in (3.53b) and (3.53c), the resulting sentences become unac-
ceptable. Therefore, in English, just as with two-conjunct complexes, the cate-
gory of the non-final conjuncts determines the category of multiple coordinate
complexes.
This and the previous subsection show the uniform behavior of all of the
non-final conjuncts in contrast to the final conjunct. The contrast falsifies the
multiple branching structure of (3.49b).
3.6.3 Borsley’s arguments against the layered complementation in English

Borsley (2005) presents three arguments against the layered complementation
structure of multiple coordinate complexes in English, i.e. (3.49d). All of his
generalizations indicate that no two conjuncts form a maximal projection inde-
pendent of the third conjunct. But grouping the last two conjuncts is represented
in (3.49c).
A. The both argument

Generally, the word both can occur to the left of X in a non-collective coordinate
construction [X and Y]. If any two conjuncts of [A, B and C] could form a
constituent independent of the third conjunct, we would therefore expect the
strings [both A, B and C] and [A, both B and C] to be acceptable. However,
neither is acceptable:
(3.54) a. ∗ Both John, Mary and Bill are asleep. (Lasersohn 1995: 151)
b. ∗ The chicken was both cold, sour and expensive. (Schwarzschild
1996: 143)
(3.55) a. ∗ John, both Mary and Bill are asleep.

b. ∗ The chicken was cold, both sour and expensive.
In Lasersohn (1995: 151), the unacceptability of (3.54a) is captured by the

claim that among the three conjuncts, “no two of them form a coordinate
structure to the exclusion of the third.” This is exactly the right description.
B. The gapping argument

Generally, if two clauses are conjoined and have parallel structures and identical
verbs, the verb in the second clause may undergo gapping. Borsley cites the fol-
lowing example from McCawley (1988a: 269) to show that this generalization
does not hold of B and C in [A, B and C].
∗
(3.56) Tom ate a hamburger, Alice drank a martini, and Jane a beer.
If [A, B and C] had a structure of [A & [B&C]], we would expect that

gapping can apply to C alone, taking the verb of B as the antecedent. Such a
gapping should be possible within the cycle of [B & C], independently of A,
which occurs outside the cycle. However, the above sentence is not acceptable,
posing a problem for the [A & [B&C]] analysis.8
8 Note that if the verbs in three conjuncts are identical, the last one can be gapped (Heycock and
Zamparelli 2000: (13b)):
C. The distributivity argument

In addition to the above two syntactic arguments, Borsley (2005) presents a
further semantic argument. Generally, [X and Y] may have a collective read-
ing if the context allows. This is also true for multiple coordinate complexes.
Thus (3.57) means either that each of the three persons lifted the rock indepen-
dently, a distributive reading, or that they lifted the rock together, a collective
reading.
(3.57) Hobbs, Rhodes and Barnes lifted the rock. (= (3.47c))
Note that if [H, R and B] had the structure [H & [R&B]], we would expect
Rhodes and Barnes to form a group in lifting the rock, leaving Hobbs to do so
independently. However, this sentence does not allow such a reading. Thus the
constituency of [H & [R&B]] is unlikely.
Similar observations are made in Dougherty (1970b: 858 (61)–(63), 860
(83)) and Winter (2001: 65 (99); 2006: 6).9
The lack of a group meaning for any conjunct pair within multiple-conjunct
complexes indicates that the multiple Spec structure in (3.49c) is right. In this
structure, the top Specifier cannot be stranded, nor can the two Specs form a
constituent, excluding the complement.
The same distributivity effect can also be seen in sentences with the distribu-
tive adverb respectively. Borsley describes the reading contrast between the
following two sentences:
(3.58) a. The two girls were seen by Rhodes and Barnes, respectively.
b. The two girls were seen by Hobbs, Rhodes and Barnes, respectively.
If [A, B and C] had the structure [A & [B&C]], we would expect the “matrix”
Specifier-Complement relation between A and [B and C] to be in construal with
the adverb respectively, as it is in (3.58a). If so, the coordination of [B and C] in
(i) John wrote the first chapter, Mary wrote the second chapter, and Bill wrote the
conclusion.
This fact might show the contrast between final and non-final conjuncts, but it does not show
that the last two conjuncts may form a maximal projection.
9 Note that if the rock in (3.57) is changed into the bare plural rocks, as in (i), a new reading
becomes possible: H and R lifted a rock (together) and R and B lifted another rock (together).
This is possible if R lifted a rock with one hand with H, and lifted another rock with the other
hand with B. See Winter (2001: 256) for discussion.
(i) Hobbs, Rhodes and Barnes lifted rocks.
Since this is still a symmetrical collective reading, the complexity is not incompatible with the
observation reported in the text.
(3.58b) could be collective. The intended meaning would be that one of the girls
was seen by Hobbs alone, and the other girl was seen by Rhodes and Barnes
together. However, Borsley points out that (3.58b) is semantically anomalous.
The fact that [Rhodes and Barnes] does not allow a collective reading here
again casts doubt on the assumed layered complementation structure in (3.49d).
In a multiple Spec structure, no single Spec may take part in computation
independently, and the cluster that excludes the outer Spec may not take part in
any computation either. For more semantic arguments in support of Borsley’s
conclusion, see Winter (2006).
All of these syntactic and semantic arguments support the multiple Spec
structure in (3.49c), as Borsley concludes at the end of his Section 3. Multiple
Spec structure is independently supported by other constructions (e.g. Pesetsky
2000; Ura 2000; Collins 2002b), and poses no theoretical problems. “[T]o say
that there are any number of specifiers is not an assumption, it’s just to say you
may continue to merge indefinitely: it merely states that language is a recursive
system” (Chomsky 2002: 133). We thus once more reach the same conclusion
as in Chapter 2:
No special configuration exists in coordination .
Multiple Spec structure is never required by any lexical item, and thus it is
not obligatory. This accounts for Gazdar et al.’s (1985: 170) observation that
no coordinator requires three or more conjuncts.
3.7 Chapter summary

We have presented cross-linguistic data to show that coordinators can c-select
conjuncts. We have also shown that in coordinate complexes headed by and-
like coordinators, the external conjunct determines the category of the whole
coordinate complex. In order to account for this fact, we have proposed that the
categorial feature is transferred from the external conjunct to the coordinator
in a coordinate complex. Moreover, since coordinate complexes neither show
a distribution other than that of the currently recognized categories, nor neu-
tralize contrasts in the currently recognized categories, they cannot represent
an independent category, thus arguing against the &P hypothesis. Furthermore,
by showing that coordinate complexes can be any category, we have argued
against the Clause Conjunct Hypothesis. Finally, based on the arguments in
this and the previous chapter, we have argued that the structure of coordinate
complexes composed of three conjuncts and one coordinator is a multiple Spec
complementation structure.
The theoretical contribution made in this chapter is that unlike categories such
as N and V, the notion “coordinator” is not primitive in syntactic computation.
If coordinators do not have intrinsic categorial features, it is the conjuncts
that provide the categorial features to them. We thus reach the second major
conclusion of this book:
No special category exists in coordination .
III No special syntactic
constraint
4 The Conjunct Constraint and the
lexical properties of coordinators
4.1 Introduction
4.1.1 The CCi and CCe
The Coordinate Structure Constraint (CSC, Ross 1967: 89) states: “In a coor-
dination structure, no conjunct may be moved, nor may any element contained
in a conjunct be moved out of that conjunct.” As emphasized by Postal (1998:
95), the CSC was intended as a linguistic universal. However, from a mini-
malist perspective it is implausible that the theory of syntax has, in addition
to operations such as Merge and Remerge, which apply in the computational
system in its constructive sense, construction-specific stipulations such as the
CSC, which give special instructions where the normal operations cannot apply.
Nevertheless, it seems that the CSC has remained relatively immune to reduc-
tion to other more general principles. The goal of this chapter is to reduce the
CSC to other general principles.
The CSC contains two parts (Grosu 1972; 1973). The first part is that no
conjunct may be moved, and the second part is that no element may be extracted
from conjuncts. Following Grosu (1972), we call the first part the Conjunct
Constraint (CC), and the second part the Element Constraint (EC). This chapter
discusses the CC, and the next discusses the EC. CC effects are shown in (4.1)
and (4.2).
∗
(4.1) a. Johni seems to be [ i and Mary] in the room.
∗
b. Whoi did John kiss [ i and a girl]?
∗
c. The speaker whoi I watched [ i and Bill] was vain.
∗
d. [Go to the club]i John might [ i and [have some fun]].
∗
e. Cani you [ i and will] stay at home?
∗
(4.2) a. Maryi seems to be [John and i ] in the room.
∗
b. Whoi did John kiss [a girl and i ]?
∗
c. The speaker whoi I watched [Bill and i ] was vain.
∗
d. [Have some fun]i John might [[go to the club] and i ].
∗
e. Willi you [can and i ] stay at home?
79
80 The CC and the lexical properties of coordinators
In each of the examples in (4.1), the first conjunct moves, and in each
of the examples in (4.2), the second conjunct moves. The CC is violated in
both groups of examples. The movement is an A-movement in (4.1/4.2a),
interrogative wh-movement in (4.1/4.2b), relativization wh-movement in
(4.1/4.2c), topicalization movement in (4.1/4.2d), and head movement in
(4.1/4.2e).1
CC effects are also observed in languages other than English. In the following
Norwegian examples, a conjunct of the coordinate object complex is raised in
(4.3a), and a conjunct of the post-verbal coordinate subject complex is raised
in (4.3b). Neither is acceptable (Johannessen 1998: 222). The unacceptability
of raising the subject out of the Spanish coordinate complex in (4.4a) shows
the same restriction (Zoerner 1995: 78 (25)).
∗
(4.3) a. Olai så jeg [Per og i ]. (Norwegian)
Ola saw I Per and
Intended: ‘I saw Peter and Ola.’
∗
b. Peri vasket klaer [Ola og i ].
Per washed clothes Ola and
Intended: ‘Ola and Peter washed clothes.’
∗
(4.4) a. Éli con cuidado [ i y ella] manejaron el coche. (Spanish)
he with care and she drove the car
b. [Él y ella] con cuidado manejaron el coche.
he and she with care drove the car
‘He and she carefully drove the car.’
Postal (1998: 83) states: “The Conjunct Constraint is almost never ques-
tioned.” Current research within the minimalist program has also kept silent
about the CC.
I will argue that the CC should be split into the CCe, i.e., no external con-
junct may move, and the CCi, i.e., no internal conjunct may move. The CCi
means that no coordinators may be stranded. I therefore group coordinators
with other types of elements that may not be stranded. As for the CCe, exter-
nal conjuncts in complexes headed with coordinators like and cannot move
because their categorial features have been transferred to the head, as argued
1 One needs to distinguish data like (4.1) from data like (i) and (ii), which are derived by PF
deletion (Section 2.3.2).
(i) John bought a book yesterday, and a newspaper.
(ii) Welchen Knaben hat Hans geküsst und welches Mädchen? [German]
which boy has Hans kissed and which girl
4.1 Introduction 81
in Chapter 3. In short, it is the lexical/morphological properties of the specific

type of coordinators that are responsible for the effects of the CC.
The organization of this chapter is as follows. I show the problems with
current approaches to the CC in Section 4.1.2 and present my own analysis
of the CCi and CCe in Section 4.1.3. Then in Section 4.2, I present the main
argument for my new account of the CCi. My two arguments for the new
account of the CCe are given in Sections 4.3 and 4.4. Section 4.5 is a brief
summary.
4.1.2 Previous approaches to the CC

The CC has previously been analyzed as a unified syntactic constraint (Sag
1982; Pesetsky 1982; Pollard and Sag 1994; Zoerner 1995), semantic con-
straint (Johannessen 1998), phonological constraint (Merchant 2001), or as a
consequence of the Parallelism Requirement (Napoli 1993).
A. The CC as unified syntactic constraint

The A-over-A principle account Sag (1982: 334) and Pesetsky (1982:
435) (see also Riemsdijk and Williams 1986: 20) claims that CC effects are
accounted for by the so-called A-over-A principle. This approach claims that
in (4.5), neither of the two lower NPs may be moved, since they are hosted by
the upper NP.
(4.5) NP
NP and NP
Let us examine the A-over-A principle itself. There are two versions of the
principle. One version says that if a rule ambiguously refers to A in a structure
of the form (4.6), the rule must apply to the higher, more inclusive, node A.
This principle is claimed to prevent extraction of the NP Africa out of the NP
my trip to Africa in which it is included in (4.7a) (Chomsky 1964; Ross 1967).
(4.6) . . . [A . . . [A . . .
(4.7) a. I won’t forget [NP my trip to [NP Africa]]

b. ∗ Africa, I won’t forget my trip to.
c. My trip to Africa, I won’t forget.
Chomsky (1964) points out several potential counterexamples to the A-over-

A principle:
(4.8) a. Who would you approve of my seeing?

b. What are you uncertain about giving to John?
c. What would you be surprised by his reading?
In (4.8a), for instance, the nominal who moves out of the nominal my seeing
who. Data like (4.9) also show that the principle is too strong. The extraction
of the DP that book is possible, although it is contained in the DP five reviews
of that book.
(4.9) That book, they published [DP five reviews of ]. (Gazdar et al. 1985: 148)
Since this version of the A-over-A principle is empirically inadequate, I do

not think its application to the CC can be justified. Some effects of the principle
can be covered by other constraints. The acceptability of (4.8) and (4.9), for
instance, can be captured by Kuno’s (1987) topichood analysis of extraction, or
by Davies and Dubinsky’s (2003) semantic condition for extraction from nom-
inals. Davies and Dubinsky claim that extraction from unambiguous process
nominals, such as (4.8), is generally possible, and extraction from representa-
tional nouns, such as (4.9), is possible when the verb is a verb of creation or a
verb of use (e.g. write, edit, publish), but not with a verb like destroy:
∗
(4.10) Who did they destroy a book about? (Davies and Dubinsky 2003: 3)
Another version of the principle says that if a rule ambiguously refers to A

in a structure of the form of (4.6), the rule must apply to the IMMEDIATELY
higher, more inclusive, node A. The following examples may show the principle
(Sag 1982: 334):
(4.11) a. Fido jumped [PP from [PP under [DP the table]]].
b. Fido ran [PP out [PP into [DP the meadow]]].
(4.12) a. From under which table did Fido jump ?

b. Out into which meadow did Fido run ?
(4.13) a. Which table did Fido jump from under ?

b. Which meadow did Fido run out into ?
(4.14) a. ∗ Under which table did Fido jump from ?

b. ∗ Into which meadow did Fido run out ?
The two sentences in (4.11) are canonical forms. In (4.12a,b), the matrix
PP has undergone a wh-movement. Since in (4.12) the wh-movement applies
to the matrix PP rather than the embedded PP, the A-over-A principle is not
4.1 Introduction 83
violated. In (4.13a), the moved wh-phrase is a DP, rather than PP. Similarly,
in (4.13b), the moved wh-phrase is a DP. The A-over-A principle does not
apply to (4.13), since the trace of the wh-DP is not immediately dominated by
another DP. Instead, it is immediately dominated by the PP headed by under or
into. In (4.14a), however, the trace of the PP under which table is immediately
dominated by another PP, which is headed by from. In this case, the A-over-A
principle is violated. Similarly, in (4.14b), the trace of into which meadow is
immediately dominated by another PP, which is headed by out, and again the
A-over-A principle is violated. Sag claims that the unacceptability of the two
sentences in (4.14) is captured by the A-over-A principle.
Like the former version, however, this alternative version of the A-over-
A principle is also empirically inadequate. First, it cannot explain quantifier
floating (Sportiche 1988). It is generally assumed, as shown in (4.15a), that the
quantifier all is a D-element, takes a DP complement, and projects another DP.
In quantifier floating constructions, the DP complement of all is raised out of
the hosting DP (see Adger 2003: 263 for a summary; also see Fitzpatrick 2006
for more discussion of quantifier floating).
(4.15) a. DP b. DP
D DP DP D′
(possessor)
all D NP
Second, the possibility of possessor DP raising out of possessive DP, as

illustrated in (4.15b) (Szabolcsi 1983; 1994; Landau 1999; among others),
indicates that the principle is too strong to rule in such acceptable movement
of Spec elements.
Since the A-over-A principle is not a valid principle with respect to movement
of either complements or Spec elements, it cannot be used to account for CC
effects.
The trace subcategorization account Pollard and Sag (1994: 201)

claim that under virtually any assumptions about the nature of coordinate struc-
tures, it is the mother of the coordinate structure that is (strictly) subcategorized,
not the individual conjuncts. Since a conjunct is never subcategorized for, it can
never be realized as a trace, following their Trace Principle, which states that
“Every trace must be subcategorized by a substantive head” (p. 172). However,
the movement illustrated in (4.15b) is wrongly ruled out in this account, since
it is the top DP rather than the possessor DP that is subcategorized by a head.
Moreover, the account is not able to explain the contrast between Can and will
you stay at home and ∗ Cani you [ti and will] stay at home, a contrast that has
nothing to do with subcategorization.
Sag (2000: 8) claims that the CC can be accounted for by the assumptions that
wh-traces are not syntactic constituents and that conjuncts must be syntactic
constituents. His claim might cover (4.1b) and (4.2b), but it cannot cover the
other data in (4.1) and (4.2), where no wh-conjunct appears. Thus the CC seems
to have nothing to do with the status of wh-traces.
The Minimal Link Condition account Zoerner (1995: 75ff.) proposes

that CC effects can be covered by the Minimal Link Condition, since he assumes
that an attracting feature external to a coordinate complex never looks as low
in the structure as the conjunct, but instead sees the feature on the &P node and
stops its search there. This proposal is too strong. If coordinate complexes have
a complementation structure, as he also argues, then conjuncts are Spec and
Complement elements. Empirically, raising of Spec elements and Complement
elements is possible in general, as seen in (4.15), so why would it be blocked
specifically in coordinate complexes? Theoretically, the assumed &P does not
c-command either conjunct, and thus the Minimal Link Condition is not relevant
here.2
B. A semantic approach
Johannessen (1998: 235) proposes a semantic account for the fact that the EC
can be violated (see Grosu 1973, and my Section 2.3.4 and Chapter 5) while
the CC cannot:
When something is extracted a special relationship is triggered between two

elements. However, when a whole conjunct is extracted, there are no longer
two available elements between which such a relationship could exist, and
extraction must be ruled out. Thus the semantic factors not only constrain
extraction of parts of conjuncts, but constrain extraction of whole conjuncts –
to such an extent that the latter is not possible.
However, according to this semantic approach, any movement at all will

cause an element to be away from its merger position, where it is integrated
into the structure semantically, and thus will give rise to the same semantic
problem claimed by Johannessen to be caused by CSC violations. Consider
possessor raising. If the possessor is moved away from the possessed nominal,
2 Goodall (1987) proposes an account for the CC in terms of Principle C of the binding theory.
Obviously, his effort does not consider non-nominal conjuncts, and thus it still needs another
account for such conjuncts. See Sag (2000: 6) for arguments against Goodall’s proposal.
4.1 Introduction 85
does that mean the semantic relation between the two is lost? Also, in data
like What kind of beer does John normally drink, the verb drink s-selects the
liquid-denoting what kind of beer. When the latter moves, does that mean the
semantic selection is lost? Since this semantic approach is not plausible, it
cannot be used to account for CC effects.
C. A phonological approach
Grosu (1981: 56) proposes a Null Conjunct Constraint (NCC), which states
that conjuncts may not be phonologically null. Merchant (2001) claims that
the CC can be covered by the NCC. In Merchant’s approach, the CC is related
to PF. Considering the cases of null external conjuncts (conjunct-drop) to be
presented in Section 4.2, we do not think that the NCC is applicable to external
conjuncts. Null internal conjuncts, however, are indeed not seen. Thus the NCC
can explain CCi effects, as will be discussed in Section 4.2.
D. A Parallelism Requirement approach

Napoli (1993: 409) speculates that the CC might be the result of a Parallelism
Requirement. However, in the presence of and, the CC cannot be the result of
the Parallelism Requirement, since this constraint can be satisfied by moving
both conjuncts, and yet the resulting coordinate complex is still unaccept-
able. Here are some unacceptable examples of the Across-the-Board type of
dependency:
(4.16) a. ∗ Which books did Bob read [ and ]?

b. ∗ I wonder who you saw [ and ]? (Gazdar et al. 1985: 178)
c. ∗ The Pre-Raphaelites, we found [ and ]. (Sag 1982: 332)
Similarly, Russian allows multiple wh-movement, but wh-movement of both

conjuncts is impossible:
∗
(4.17) [kokogo mal’čika]i [kakuju devočku]j ty l’ubiš’ [ i i j ]?
which boy which girl you love and
In the above discussion, I have shown that none of the unified approaches
to the CC, whether syntactic, semantic, phonological, or parallelism-based,
reveals the true nature of the CC. Since no satisfactory account for the CC is
available, I pursue my own account in the following sections.
4.1.3 A new account of the CC

In this section I give brief introductions to my accounts of the CCi and the CCe,
as background for the fuller arguments in Sections 4.2–4.4.
A. The CCi: certain head elements need an overt complement

CCi effects in the presence of and are well established, as seen in (4.2). As
an explanation of them, I propose that internal conjuncts may not move in the
presence of a coordinator because, like certain other head elements, coordi-
nators may not have silent complements. Thus, internal conjuncts may not be
silent, regardless of whether the silence is caused by movement or deletion.
The fact that certain head elements do not allow their complements to be silent
is seen from observations like the following. For instance, the complementizer
that cannot be stranded (data from Adger 2003: 290; see Abels 2003: 116 for
parallel examples in French and Icelandic):
(4.18) a. Everyone claimed [that the poison was neutralized].

b. ?[That the poison was neutralized] was claimed by everyone.
∗
c. [The poison was neutralized] was claimed that by everyone.
d. It was claimed by everyone [that the poison was neutralized].
∗
e. It was claimed that by everyone [the poison was neutralized].
The complementizer for also cannot be stranded (Law 2002: 84, among
others):
∗
(4.19) [John to leave]i is impossible [CP [for ti ]]
Although Chinese transitive verbs in general may be stranded, as seen in

(4.20a), causative verbs or particles such as shi, rang, jiao, ba ‘cause, make’
may never be stranded, as shown in (4.20b).
(4.20) a. Baoyu xiwang shang daxue, Daiyu ye xiwang.

Baoyu hope go university Daiyu also hope
‘Baoyu hopes to go to a university, so does Daiyu.’
b. ∗ Zhe jian shi rang wo hen nankan, na jian shi ye rang.

this cl matter make I very embarrassed that cl matter also make
Intended: ‘This matter embarrassed me, so did that matter.’
Similarly, articles such as the in English are never stranded, nor are degree
elements such as feichang ‘very’ in Chinese, nor are prepositions in many
languages including Chinese.3
3 Various analyses have been proposed to account for why complementizers may not be stranded
(e.g. Webelhuth 1989; Saito and Murasugi 1999; Wurmbrand 2001; Abels 2003). It is not true that
head elements may be stranded only when the Spec position is filled (Lobeck 1990). Preposition
stranding is possible in English, but not in languages such as Chinese, regardless of whether the
Spec of P is filled.
4.1 Introduction 87
It is thus possible that coordinators have the same property. They neither
license movement of their complements, nor allow their complements to be
deleted at PF. I will argue for this proposed account of the CCi in Section 4.2.
B. The CCe: categorial features are carriers of overt movement

Regarding the CCe, I claim that external conjuncts cannot move in the presence
of and because the categorial features of the former have been transferred to
the latter (see Chapter 3).
The theoretical background for this claim on the CCe is the following.
Any element that undergoes overt movement must have categorial features. In
Chomsky (1995: 265; see also Ochi 1999: 90), all overt movement chains are
composed of two sub-chains: a chain of formal features that take part in the
required checking (CH <FF>) and a chain of categorial features (CH <CAT>).
The latter chain does not involve feature checking, but it is simply the carrier
of the former chain. In Chomsky’s terms, overt formal feature movement is
always pied-piped with a categorial chain. This implies that in the absence of
categorial features, no overt movement is possible. If the categorial features of
the external conjunct have been transferred away, the conjunct cannot move
anymore, since the carrier is gone.
One might respond that in the copy theory of movement, a lower copy
remains after the movement, and thus if categorial features move from exter-
nal conjuncts, they still remain as a lower copy, before any deletion. How-
ever, as argued in Zhang (2004a) and asserted in Chomsky (2008: fn. 16),
movement does not create a new copy of the moved element. Instead, it is
simply an operation of Remerge. From a derivational perspective, I assume
that when x is remerged in a new syntactic position, it disappears in its old
position, syntactically (I make no claim about the PF representations). For
instance, in (4.21), the [NEG] feature disappears in the possessor nominal
after it is transferred to the sentential level (see Section 3.3.1A). This dele-
tion is attested by the fact that the sentence does not have a double negation
reading.
(4.21) Nobody’s car would I borrow. [NEG feature percolation]
I claim that movement itself can be defined as Remerge at the new position
and unmerge at the old position syntactically. Therefore, there is no syntactic
(as opposed to PF) operation of deletion or unmerge that is independent of
movement.
I will present two arguments for this analysis of the CCe in Sections 4.3 and
4.4.
4.2 The CCi and the asymmetry in conjunct drop

It is possible for one conjunct to be silent when its meaning is recoverable
from the linguistic or discourse context. I call this silence of conjuncts a
“conjunct-drop” effect. I claim that in the presence of an overt coordinator,
the internal conjunct, which is the complement of the coordinator, may not be
silent. Therefore internal conjuncts may not move, stranding the coordinator.
The main argument for my new approach to the CCi is the asymmetry between
external and internal conjuncts in conjunct drop.
4.2.1 Conjunct drop in right-branching coordinate complexes

In Chapter 2, we saw that coordinate complexes have a right-branching binary
structure in languages such as English and Chinese. In such complexes, external
conjuncts may be null, but internal conjuncts may not be. In (4.22), for instance,
the speaker directly starts the sentence with the coordinator and (Hankamer
and Sag 1976: 410).
(4.22) [Observing Ivan playing pretty good ragtime piano]

And he doesn’t have a left hand!
A coordinator used in such a way is called a discourse-initial or utterance-

initial coordinator (Hankamer and Sag 1976: 410; see also Huttar 2003). Han-
kamer and Sag (1976: 411) state: “It seems that such cases involve essentially
pragmatic omission of an understood left conjunct.” Similarly, in (4.23), the
last sentence starts with and, with the meaning of the first conjunct implied in
the previous sentence (one can find Chinese data parallel to (4.22) and (4.23)
in Shi 1986: Section 2.3; Shao and Rao 1985: 6 also discuss contextual or
discourse conjuncts in Chinese).
(4.23) “It’s time to put sentiment aside,” announced New York Times columnist
Nicholas Kristof one day last month. And who can disagree? (Michael
Kinsley, The Washington Post, Sept. 14, 2002)
By contrast, silent second (internal) conjuncts do not seem to be possible.

Although we use the elided expression “And you?” we do not say “You and?”
Moreover, the examples in (4.24b) and (4.25b) show that coordinator-final clus-
ters are totally unacceptable when used alone. Note that the negation encoded
by not scopes over the whole coordinate complex in each case, parallel to the
corresponding a-sentences. The a-sentences are cited from Postal (1998: 87)
and Lawler (1974: 370) (for a semantic description of the construction, see
Goldsmith 1985: 141).
(4.24) a. Can linguists [study negation]? Not e and stay sane they can’t.
b. ∗ Can linguists [stay sane]? Not study negation and e they can’t.
(4.25) a. Can I [go outside without any clothes on]? Not e and stay healthy, you can’t.
b. ∗ Can I [stay healthy]? Not go outside without any clothes on and e, you can’t.
The contrast in acceptability between the a-sentences and the

b-sentences in these examples indicates that the combination of the first
conjunct and the coordinator cannot function independently of the second
conjunct.
Similarly, as pointed out by Borsley (1994: 240), a parenthesis can be a
combination of a coordinator and a following conjunct, as in (4.26a). It is
obvious that a parenthesis may not be a combination of a coordinator and a
preceding conjunct, as in (4.26b).
(4.26) a. The professor, and he is an expert, thinks the recession will continue.
b. ∗ The professor, he thinks the recession will continue and, is an expert.
Similar patterns are found in other languages which have right-branching

coordinate complexes. In Irish, the external (i.e. first) conjunct may be a pro
while the internal (i.e. second) conjunct can never be a pro (McCloskey and Hale
1984; McCloskey 1986). In Chinese, as we have seen in (4.20a), a transitive
verb can end a clause, with the meaning of the implied object recoverable from
the context. Regardless of how such sentences are derived (pro object, deletion
of the object, or VP deletion after raising of the verb), the surface order of
(4.20a) is verb-final. Unlike verbs, however, coordinators in Chinese cannot
stand alone, that is, without an internal conjunct to their right.
The above facts from various languages all show that coordinators must be
adjacent to their internal conjuncts.
4.2.2 Conjunct drop in left-branching coordinate complexes

In contrast to and-like coordinators, which require an overt final conjunct, in
Japanese, the conjunctions -shi and -to and the disjunction -toka require an
overt initial conjunct. The structure of coordinate complexes headed by these
coordinators is left-branching (Munn 1987; Zoerner 1995: 11; also our 2.2.3),
making the initial conjunct the complement of the coordinator, and thus the
internal conjunct. Consistent with our analysis of the CCi, then, it is the final,
or external, conjunct that need not be overt. Hinds (1986: 93) describes the
relevant facts as follows. “For each of the types of coordination – and, but, and
or – there is a sentence type in which the coordinator ends the utterance. While
there may be a feeling of lack of closure, this feeling is not necessarily there at
all times.” He presents both types of examples. For the first type illustrated in
(4.27), he uses three dots at the end to show the lack of closure:
(4.27) a. Kyoo-wa atsui-shi, . . . (Hinds 1986: 86)
today-top hot-and
‘It’s hot today, and (moreover) . . . ’
b. Kyooto-e-wa ikanai-shi, . . .
Kyoto-to-top go.not-and
‘[I]’m not going to go to Kyoto, and (moreover) . . . ’
For the second type, Hinds notes that in (4.28a) (= his (318) on p. 93) “the
first sentence ends in falling intonation. The second sentence is reproduced
so the reader can see that it is not a continuation of the first sentence.” (QT
= quotative morpheme) Similarly, in (4.28b) (= his (319) on p. 93), “the
coordinator toka appears with falling intonation in answer to a question.”
(4.28) a. Nihon kaet-te ojisan ii tsut-tara unten suru kamoshirenai shi.
Japan return-when uncle OK qt-say-if drive do probably and
Nihon konde-ru kara wakannai.
Japan crowded since know-neg
‘When [I] return to Japan if my uncle says it’s OK [I]’ll probably drive, and.
Japan’s crowded, so [I] don’t know.’
b. Tatoeba ojiichan-ga shi- shinda toki toka.
e.g. grandfather-nom died time or
‘The time Grandfather died, or.’
The following data (Zoerner 1995: 33) further show that unlike English,
Japanese does not allow the first conjunct to be contextual or implied in the
discourse.
(4.29) a. A: Robin ate fish. B: And rice!
b. A: Robin-wa sakana-o tabeta. B: ∗ to gohan!
Robin-top fish-acc ate and rice
Since the -to and -shi coordinate complexes in Japanese are left-branching,
final conjuncts are external conjuncts. As in the last subsection, all of the data
in this subsection show that external conjuncts can be silent whereas internal
ones cannot.
4.2.3 Clause-final coordinator-like elements

In the above two subsections, I have shown that coordinators require the pres-
ence of internal conjuncts. In this subsection I show that although certain
coordinator-like elements can appear clause-finally in certain languages, they

have different formal properties from regular coordinators. Thus their existence
does not challenge the generalization that coordinators may not be stranded.
In German, the word oder ‘or’ can occur at the end of a root sentence,
building a yes–no question. The oder-final sentences are always uttered with
a rising intonation. Speakers of such sentences expect a confirmation. Thus
the function of this oder is similar to that of Isn’t it? in English (André Mein-
unger and Hans-Martin Gärtner, p.c.) or question markers in Chinese. In some
dialects, the counterpart of oder in this usage does not share its form with any
disjunction (André Meinunger, p.c.). The Norwegian sentence-final eller ‘or’
has the parallel function, as pointed out by an anonymous reviewer of Studia
Linguistica.
Another case in which questions are introduced by a sentence-final
coordinator-like element is the German expression Na und? ‘So what?’
Similarly, -oo in Malayalam occurs either as a disjunctive, as in (4.30a), or
as a sentence-final particle, as in (4.30b):
(4.30) a. John-oo Bill-oo Peter-oo (Jayaseelan 2001: 64)

John-or Bill-or Peter-or
‘John or Bill or Peter’
b. John wannu-(w)oo? (Jayaseelan 2001: 67)
John came-or
‘Did John come?’
Paralel Sinhala and Japanese data can be found in Jayaseelan (2008: 3)

and Hagstrom (1998). As correctly pointed out by Jayaseelan (2008: 3), the
disjunctive-like element at the clause-edge position “cannot be a disjunctive
connective, since it does not connect anything.” Following Hagstrom (1998)
and Jayaseelan (2001: 70), I treat the sentence-final coordinator-like elements
in these languages as clause-typing particles base-generated in the C-domain.
By contrast, regular coordinators do not introduce any clause-typing informa-
tion. Therefore, I do not think data like (4.30) affect the generalization that
coordinators require overt internal conjuncts.
In this section I argued that CCi effects are covered by a general surface con-
straint disallowing certain head elements from being stranded. In this approach,
CCi effects are not caused by any construction-specific syntactic constraint, but
instead they are syntax – phonological interface effects that also affect certain
other types of head elements. The phonological side of the constraint rules
out any representations in which the complement of the elements is a trace or
is affected by deletion, while the sensitivity of the requirement to the part of
speech of the head elements reveals a morphosyntactic side of CCi effects. For
instance, the complementizer that does not allow a null complement, whereas
prepositions such as with and of do (as in (4.9)). Consequently, we do not
consider the CCi to be a purely syntactic constraint specific to coordinate
constructions.
4.3 The CCe and the Chinese de constructions

We have seen that external conjuncts can be null in Section 4.2. Thus the
CCe cannot be a phonological constraint. I have proposed instead that external
conjuncts cannot move in the presence of and, because the categorial features
of the former have been transferred to the latter.
My arguments for this new account of the CCe come from two sources.
(A) When categorial transference occurs in other constructions, the cate-
gorial feature-providing elements may not move. In other words, CCe-like
effects are attested independently of coordinate constructions. (B) If coor-
dinators have their own intrinsic categorial features, the CCe may be vio-
lated, since external conjuncts do not transfer their categorial features away
and thus are able to move by pied-piping on the categorial feature chain. In
other words, CCe effects are predictably absent in certain types of coordinate
constructions. I will present argument A in this section, and argument B in
Section 4.4.
In Mandarin Chinese, de is a bound form (an enclitic, according to
C. R. Huang 1989). It is phonologically weak (it has an intrinsic neutral
tone, which means that the syllable is short and cannot bear any stress),
and is attached to an element to its left. It occurs between a modifier and
the modified element, as in (4.31a), or between a relational noun and its
semantic licensor, as in (4.31b), or between a noun and its complement, as in
(4.31c).
(4.31) a. [honghong] de pingguo

red de apple
‘red apple(s)’
b. [Lulu] de linju
Lulu de neighbor
‘Lulu’s neighbor’
c. [jintian dizhen] de yaoyan
today earthquake de rumor
‘the rumor that there’s an earthquake today’
In Li and Thompson (1981), de is called associative marker, since it occurs

between two elements that have a certain semantic relationship.4 The term
associative marker is abstract enough to cover the various semantic relations
between the two linked elements. The modified element in (4.31a), the relational
noun in (4.31b), and the Head noun in (4.31c) are the semantic kernel of the
complex nominal. I thus call them kernel elements. The non-kernel element of
a de construction functions as a major constituent of the kernel element.5 In the
examples discussed in this section, the underlined part is the kernel element
and the part in brackets is the non-kernel element.
In this section, I first distinguish two types of associative markers in Mandarin
Chinese (both pronounced de), one that occurs in kernel-final constructions and
one that occurs in kernel-initial constructions (Section 4.3.1). Then I present
data to show that the de of the former type occurs with kernel elements of various
categories (Section 4.3.2), argue that de heads the whole complex containing the
kernel and non-kernel elements (Section 4.3.3), and finally report that in such
constructions, kernel elements may not move (Section 4.3.4). If de is indeed
the head of the whole complex, the fact that it occurs with kernel elements of
different categories shows that it must get its categorial features from the kernel
element. Therefore, the immobility of the kernel elements supports my claim
that transferring the categorial features away from an element blocks it from
moving (Section 4.1.3B).
4.3.1 Two kinds of de constructions

De may follow a modifier, as in (4.32a), forming a kernel-final construc-
tion. It may also precede a modifier, as in (4.32b), forming a kernel-initial
construction.
4 I do not discuss other uses of de, such as in resultative constructions, in the shi . . . de ‘be
. . . de’ constructions, and in nominalization. I assume that the particle de in such constructions
is different from the associative marker discussed here.
5 The notion of major constituent is borrowed from Hankamer (1973: 18). A major constituent
is the modifier, complement, or subject of a certain element. Its existence has been attested at
least in gapping constructions: the remnants of a gapping operation must be major constituents
(Hankamer 1973: 18; Neijt 1979: 40, 111). Note that if both of the internal and the external
argument of a nominal precede the nominal, in addition to de, the preposition dui ‘to’ occurs to
the left of the internal argument (see Fu 1994: Chapter 4 and Saito et al. 2008: 257):
(i) Manzu de ∗ (dui) Luoma de huimie
Barbarian de to Rome de destruction
‘the barbarians’ destruction of Rome’
(4.32) a. Baoyu zai [hen kuai] de langdu. (kernel-final)

Baoyu prg very fast de read.aloud
‘Baoyu is reading (something) aloud fast.’
b. Baoyu langdu de [hen kuai]. (kernel-initial)
Baoyu read.aloud de very fast
‘Baoyu read (something) aloud fast.’
These two constructions are different in at least two respects. First, in kernel-
final constructions the phonological host of de may be a phrase, such as hen kuai
‘very fast’ in (4.32a), whereas in kernel-initial constructions it must be a head
element, such as langdu ‘read aloud’ in (4.32b) (Tang 1990: 431). Data like the
following show that de may not follow a phrase in kernel-initial constructions:
∗
(4.33) Baoyu langdu na pian wenzhang de [hen kuai]. (kernel-initial)
Baoyu read.aloud that cl paper de very fast
Intended: ‘Baoyu read that paper aloud fast.’
Since the categorial levels of the phonological hosts of the des in the two
constructions are different, I claim that they are different kinds of bound forms,
following Tang (1990: 431).
Second, the kernel element must be verbal in kernel-initial constructions,
whereas it can be other categories in kernel-final constructions. We have seen
the verbal kernel element langdu ‘read aloud’ in (4.32). The following examples
show that the kernel element can be a noun in kernel-final constructions (also
(4.31)), but not in kernel-initial constructions.6
(4.34) a. [hen shangxin] de xuesheng (kernel-final)
very sad de student
‘very sad students’
b. ∗ xuesheng de [hen shangxin] (kernel-initial)
student de very sad
We can see that de is similar to coordinators in that they both occur with
two syntactic constituents.7 Specifically, the de in kernel-final constructions is
6 In writing, many people use different characters to distinguish the associative marker that
precedes a nominal kernel element (d), as in (4.31), (4.34a), from the one that precedes a verbal
kernel element (d), as in (4.32a), and from the one that occurs in kernel-initial constructions
(d), as in (4.32b). However, the standard grammar books (e.g. Chao 1968; Zhu 1984) do not
make this distinction.
7 Unlike Rebuschi (2005) and Y. A. Li (2007), I do not claim that de and conjunctions are
semantically similar. A complex composed of a conjunction and two singular conjuncts can
similar to the English coordinator and in that they both occur in complexes
of different categories, whereas the de in kernel-initial constructions is similar
to the Chinese coordinator erqie, as in (4.35), in that neither may occur in
complexes of non-verbal categories.
(4.35) Wo renshi [yi ge xiaoshuojia {he/∗ erqie} yi ge yinyuejia].

I know one cl novelist and/and one cl musician
In the following subsections I focus on the and-like de of kernel-final con-

structions.
4.3.2 The various categories of kernel-final constructions

De in kernel-final constructions in Mandarin Chinese exhibits two major prop-
erties.
First, de may occur with non-kernel elements of various types.
(4.36) a. [lao] de jiaoshou b. [wu-li] de jiaoshou

old de professor room-in de professor
‘old professor(s)’ ‘the professor(s) in the room’
c. [yiqian] de jiaoshou d. [women] de jiaoshou

past de professor 1pl de professor
‘professor(s) of the old days’ ‘our professor(s)’
e. [yan malu] de shangdian f. [dui zongjiao] de taidu

along street de shop toward religion de attitude
‘the shops along the street’ ‘the attitude towards religions’
g. [zuotian lai] de jiaoshou h. [zuotian lai] de Li Jiaoshou

yesterday come de professor yesterday come de Li professor
‘the professor that came yesterday’ ‘Professor Li, who came yesterday’
i. [qita] de jiaoshou j. [suowei] de jiaoshou

rest de professor so-called de professor
‘other professors’ ‘so-called professor(s)’
satisfy the plurality requirement of collective verbs or predicates, but a complex composed of de
and two singular nominals may not, as seen in (i).
(i) Lulu hebing-le {wenjian jia he wenjia yi / ∗ wenjian jia de wejian yi}.
Lulu combine-prf file A and file B file A de file B
‘Lulu combined {file A and file B / ∗ file B of file A}.’
The non-kernel element is an adjective in (4.36a), a locative nominal in

(4.36b), a temporal nominal in (4.36c), a personal pronoun in (4.36d), a preposi-
tional phrase in (4.36e) and (4.36f), a relative clause in (4.36g), a non-restrictive
relative clause in (4.36h), and a non-predicative adjective in (4.36i) and (4.36j).
These various categories of the non-kernel elements have no impact on the
category of the whole complexes. All of the examples in (4.36) are nominals.
In these examples, de plays no role in deciding the category of the complexes,
either. Kitagawa and Ross (1982) use the following rule to capture this fact,
analyzing de as a general modifying marker (Mod):
(4.37) Mod Insertion: [NP . . . XP Nα ] → [NP . . . XP Mod Nα ], where Mod = de
Second, de may occur with kernel elements of various categories (Tang 1990:
421). We have seen that the kernel elements in (4.36) are nominal, whereas
those in (4.32) above are not. More data are shown in (4.38), with the kernel
elements underlined.
(4.38) a. Na ge [hen deyi] de xiaohai ling-le jiang. NP

that cl very proud de kid receive-prf award
‘That proud kid received the award.’
b. Na ge xiaohai [hen deyi] de ling-le jiang. VP
that cl kid very proud de receive-prf award
‘That kid received the award very proudly.’
c. Na tian wanshang ta [tebie] de xingfen. AP
that day evening he especially de excited
‘That evening, he was especially excited.’
d. [Diu-le qianbao] de ta hen zhaoji. Pronoun
lost-prf wallet de he very worried
‘He, who has lost his wallet, was very worried.’
e. [Diu-le qianbao] de Baoyu hen zhaoji. Proper name
lost-prf wallet de Baoyu very worried
‘Baoyu, who has lost his wallet, was very worried.’
The kernel element is the nominal xiaohai ‘kid’ in (4.38a), the eventive
predicate (verbal) ling-le jiang ‘received the award’ in (4.38b), the stative
predicate (adjectival) xingfen ‘excited’ in (4.38c), the pronoun ta ‘he’ in
(4.38d), and the proper name Baoyu in (4.38e). De occurs in all of these
kernel-final constructions.8
8 Proper names and pronouns can be modified by adjectives or relative clauses in Chinese, a
language that has no articles. Like the adjective charitable in the charitable Miss Murray, the
In the two kernel-final examples in (4.39), the non-kernel element is the same,
hen kuaile ‘very happy.’ However, the kernel element is the nominal xuesheng
‘student’ in (4.39a), but the verbal qian-le zi “signed the name” in (4.39b).
(4.39) a. Wo kanjian-le yi ge [hen kuaile] de xuesheng.
I see-prf one cl very happy de student
‘I saw a very happy student.’
b. Na ge xuesheng [hen kuaile] de qian-le zi.
that cl student very happy de sign-prf name
‘That student signed {his/her} name very happily.’
Considering a wide range of such data, Tang (1990: 424) states “all the
occurrences of de in question have the same behavior in that they do not carry
any specific semantic content but express a modifier/modifiee relation. It does
not appear to be the case that these des are of different sorts.” She proposes
that de is a functional head in all of these constructions.
We have mentioned that the de in kernel-final constructions is similar to the
English coordinator and. The two constituents linked by these elements can
be any category. However, for and-coordinate constructions, it is the left (i.e.
external) conjunct that determines the category of the whole complex, whereas
for de-constructions, it is the right (i.e. kernel) element that determines the
category of the whole complex.9
(4.40) a. [XP [ext.conjunct XP] and [int.conjunct YP]] b. [YP [non−kernel XP] de [kernel YP]]
4.3.3 De as the head of the whole complex

The goal of this subsection is to show that like coordinators, de is the head of
the whole complex. Many scholars have assumed that de is grouped with the
modifier (non-kernel), and the combination of the modifier and de functions
modifiers of proper names and pronouns are always non-restrictive in Chinese (J.-W. Lin 2003).
They can be either eventive, as in (4.35h), or stative, as in (i).
(i) Bu yuan fa-pang de ni yinggai duo yundong.
not want get-fat de you should more exercise
‘You, who do not want to get fat, should do more exercise.’
9 Other elements of the same form may also be distinguished with respect to whether they have
intrinsic categorial features. For instance, corrective but heads clausal coordinate complexes
only, and contrastive but heads coordinate complexes of various categories (see footnote 5,
Chapter 3). Also, the aspectual come and go are syntactically different from the lexical come and
go (Jaeggli and Hyams 1993). In Aboh (2009), it is argued that in many serializing languages,
the functional and lexical verbs have the same morphological form. Elements of the same form
may also be distinguished with respect to whether they have other intrinsic formal features (see
the two types of bound pronouns discussed by Kratzer (2009)).
as an adjunct of the kernel element, as shown in (4.41a). For these scholars,

then, de is not the head of the whole complex, and thus there is no structural
parallelism between (4.40a) and (4.40b) above. In this subsection, I argue
against this assumption and argue for the structure in (4.41b) for kernel-final
constructions. In this structure, de is the head, taking the non-kernel element
(XP) as its specifier and the kernel element (YP1 ) as its complement.
(4.41)
a. YP b. YP2
XP YP XP Y′2
non-kernel
non-kernel de kernel element element
Y2 YP1
element
de kernel element
I assume that if the kernel element is relational (e.g. taidu ‘attitude,’ linju
‘neighbor,’ or body part terms such as shou ‘hand’), it is possible that the non-
kernel element is merged with the kernel element first, satisfying the selection
of the latter (Castillo 2001; Ogawa 2001), and then is remerged as the Specifier
of de.
As for kernel-initial constructions, which are always verbal, I speculate that
they are derived from (4.41b) by the raising of the kernel element (YP1 ) (cf.
Karimi 2007 for Kurdish Ezafe constructions), when the head Y2 is realized by
another type of de, which is a suffix and takes a verb head as its morphological
base (Shen and Ting 2008). In other words, it is the kernel element, not the
non-kernel one, that is directly merged with de. I leave the issue of the exact
computation of kernel-initial constructions for future research. In any case, the
relation between de and the non-kernel element is not that between a head and
its complement in either kernel-initial or kernel-final constructions.
In kernel-final constructions, the phonological adjacency between the non-
kernel element and the enclitic de does not mean that they form a syntactic
constituent, as pointed out by C. R. Huang (1989: 30).
As is well known, phonological phrases are not necessarily isomorphic to
syntactic constituents. For instance, the syntactic constituency of (4.42a) is not
reflected in the phonological grouping in (4.42b) (Jackendoff 1997: 26).
(4.42) a. [DP a [NP [AP big] house]] b. [φ [ω a big] [ω house]]
Another example is seen in the position of the Latin coordinator -que. This
coordinator generally follows the first word of a conjunct, although it does not
form a syntactic constituent with the word. Compare the two coordinators, et
and -que, in (4.43).
(4.43) a. senatus et populus romanus [Latin]

senate and people Roman
‘the senate and the Roman people’
b. senatus populus-que romanus
senate people-and Roman
‘the senate and the Roman people’
Thus phonological constituency does not mean syntactic constituency. De

may undergo morphological merger with every major constituent of the whole
complex at PF, regardless of the constituent’s syntactic category, after the
complex has been built in syntax (see Matushansky 2006 for the notion of
morphological merger).
Despite these considerations, it has been sometimes assumed that de and
the non-kernel element to its left do form a syntactic constituent, excluding
the kernel element, as in (4.41a) (Cheng 1986; Tang 1990: 424; Ning 1993;
Rubin 2002; 2003; Aoun and Li 2003; among many other places). In Cheng
(1986); de is treated as a complementizer, taking the non-kernel element to its
left as its complement, although it “places no restrictions on the category of
its complement” (p. 319). In Tang (1990: 428), de is treated as a functional
category taking the non-kernel element as its complement, and surfaces to the
right of the element. In Rubin (2002; 2003), de is a Mod (for modification)
element, where “Mod is a functional category forming a shell around the content
of the modifier, XP” (Rubin 2002: 1) and ModP is an adjunct of the modified
element (Rubin 2002; 2003):
(4.44) YP (see Rubin 2002: 1)
ModP YP
(modified element)
Mod XP
de (modifier)
In contrast, in Zhang’s (1999) n analysis, Simpson’s (2003) D analysis, and

den Dikken’s (2006) Linker analysis of de, the complement of de is either the
kernel element or the element that contains the kernel element. In other words,
they do not group de with the non-kernel element, separate from the kernel
element. This is also the basic characteristic of (4.41b).
The arguments for the constituent status of the combination of de with a
non-kernel element, as represented in (4.44), are not convincing.
One argument given in Ning (1993) for a complementation relation between
de and the non-kernel element is that when the non-kernel element linked by
de is a clause, it must contain a gap, a requirement that can be treated as
a selectional property of a head on its complement. Contradicting this puta-

tive requirement, however, clauses to the left of de may actually be gapless
(Tang 1979: 243, 289; Aoun and Li 2003: 186; Chang 2006; Zhang 2008c).
Unlike English examples such as the time when I left, such constructions do
not have any syntactic position for a relative pronoun (there is no relative
pronoun in Chinese). An example was given earlier in (4.31c), and here is
another:
(4.45) Wo jisuan-chu-le [qiche xingshi] de sudu.
I calculate-out-prf car run de speed
‘I calculated the speed at which the car was running.’
Zhang (2008c) argues that the kernel element in (4.45) is a relational noun
and the relative-clause-like clause is the subject of the noun. Clausal subjects
contain no gap that is associated with their predicates (Rothstein 1991b: 145).
Therefore, nothing in data like (4.45) supports the constituency of a non-kernel
element and de.
Another argument for the assumed constituency of modifier plus de is that
every modifier can be followed by an instance of de (see Tang 1990), as in
(4.46a), and the combination may appear as a conjunct, as in (4.47b) (Aoun
and Li 2003: 150, 250 fn.12):
(4.46) a. jiaoshi-li (de) zhengzai jianghua de xuesheng

classroom-in de prog speak de student
‘the student who is speaking in the classroom’
b. zhuyao (de) erqie women yijing taolun-guo de shiqing
main de and we already discuss-exp de matter
‘the main matters that we have discussed’
There are various possible ways to explain the multiple occurrences of de in

(4.46). First, similar multiple occurrences of a functional word with multiple
parallel elements appear in other constructions where the functional word is not
grouped with the elements syntactically. For instance, determiners may occur
with each modifier in Greek (Androutsopoulou 1994), Albanian, and Hebrew,
but they do not take the modifiers as their complement.
(4.47) to megalo to kokkino to vivlio (Greek)

the big the red the book
‘the big red book’
If both de in (4.46) and to in (4.47) are treated as the head of the entire
nominal complexes, their property of being “spread” to each modifier may be
represented in a similar way. This analysis of multiple occurrences of de has

been proposed in Simpson (2003).
Second, based on the fact that only the final de is obligatory, we may claim
that other occurrences of de are not syntactic in (4.46). Instead, they might
simply demarcate major constituents of a complex (see footnote 5 of this
chapter), and such demarcation is not a syntactic operation.
Specifically, in (4.46a), the optional de marks each Spec in a multiple Spec
structure. As for (4.46b), the optional first de also behaves differently from
the final obligatory de. The contrasts in form and reading between (4.48a) (=
(4.46b)) and (4.48b) are telling. In (4.48a), the coordinator is erqie, the first
de is optional, and the reading is that of modifier coordination. In (4.48b),
however, the coordinator is gen, the first de is required, and the reading is that
of nominal coordination.
(4.48) a. zhuyao (de) erqie women yijing taolun-guo de shiqing (= (4.46b))
‘the main matters that we have discussed’
b. zhuyao ∗ (de) gen women yijing taolun-guo de shiqing
‘the main matters and the matters that we have discussed’
Recall that erqie may not conjoin nominals whereas gen conjoins nominals
only (see (4.35)). Thus the conjuncts in (4.48a) are not nominal whereas those
of (4.48b) are nominal. The structure of the first can be (4.49a), whereas that
of the second can be (4.49b), in which a PF deletion occurs.
(4.49)
optional spreading
a. [[AP zhuyao ] erqie [CP women yijing taolun-guo]] de shiqing
b. [DP zhuyao de shiqing] gen [DP women yijing taolun-guo de shiqing]

The mere availability of these alternative analyses indicates that the occur-
rence of de with each modifier in multiple-modifier constructions is not a
convincing argument for the assumed constituency of de plus modifier (An
ellipsis analysis of Greek multiple Det+Adj strings, as in (4.47), is proposed
in Panagiotidis 2007: Section 4.4).
Not only are previous arguments for (4.41a) unconvincing, but there are also
two additional crucial facts that support (4.41b) and falsify (4.41a), as I now
show. They both show the following contrast: in kernel-final constructions,
the relationship between de and the kernel element to its right is similar to
that between a head and its complement, as represented in (4.41b), whereas

in kernel-initial constructions, the relationship between de and the non-kernel
element to its right is not like that between a head and its complement.
The first fact concerns the possible silence of the element to the right of de. In
Chinese, ellipsis of a phrasal element is licensed by an overt head element that
takes the phrase as its complement. In (4.50), the null object is licensed by the
overt verb mai-le ‘buy-prf.’ Similarly, in (4.51a), the VP ellipsis is licensed by
the overt modal hui ‘will.’ In (4.51b), the VP ellipsis cannot be licensed by the
adverb ye ‘also,’ which does not take the VP as its complement. In (4.51c), the
VP ellipsis also cannot be licensed by the adverbial NP jin-nian ‘this year.’10
(4.50) Baoyu mai-le xigua, Daiyu ye mai-le xigua.
Baoyu buy-prf watermelon Daiyu also buy-prf watermelon
‘Baoyu bought a watermelon, so did Daiyu.’
(4.51) a. Baoyu hui mai baoxian, Daiyu ye hui mai baoxian.

Baoyu will buy insurance Daiyu also will buy insurance
‘Baoyu will buy insurance, and Daiyu will also buy insurance.’
b. ∗ Baoyu hui mai baoxian, Daiyu ye hui mai baoxian.
Baoyu will buy insurance Daiyu also will buy insurance
c. ∗ Baoyu qu-nian mai-le baoxian, Daiyu jin-nian ye
Baoyu last-year buy-prf insurance Daiyu this-year also
mai-le baoxian.
buy-prf insurance
The generalization to be drawn from the above data is that if XP is the

complement of Y, the presence of Y licenses the silence of XP, whereas if XP is
not the complement of Y, the presence of Y does not license the silence of XP.
In a kernel-final construction, the presence of de can license a null kernel
element, as in so-called Headless modification constructions like (4.52a) and
(4.53a). By contrast, if de is not overt, the kernel element cannot be null, as
seen in (4.52b) and (4.53b).
(4.52) a. [Dai yanjing de na ge ren] lai-le.
wear glasses de that cl person come-prf
‘The person who wears glasses came.’
b. ∗ [Dai yanjing de na ge ren] lai-le.
wear glasses de that cl person come-prf
10 In kernel-final constructions, the non-kernel (left) elements may not be null, whereas the kernel
(right) ones may be. De is thus different from and in coordinate complexes, which allows its
specifier (the external conjunct) to be null (see Section 4.2.1). The difference follows from the
fact that de is a bound form taking its preceding element as its phonological host, whereas and
requires its complement (the internal conjunct) to occur to its right.
(4.53) a. Zheli bi duodeshi, bie yong ni baba xie xin de bi.

here pen many not use 2sg dad write letter de pen
‘Here are many pens. Don’t use the one with which your dad writes letters.’
b. ∗ Zheli bi duodeshi, bie yong ni baba xie xin de bi.
here pen many not use 2sg dad write letter de pen
If de and the non-kernel element to its left form a constituent, excluding

the kernel element, it is not clear why the possibility of silence of the kernel
element depends on the occurrence of de. If de and the kernel element have
a head-complement relation, as I claim, the correlation is expected, since a
similar correlation is also observed in other head-complement constructions,
as in (4.50) and (4.51).
In kernel-initial constructions, however, de does not license a null modifier
to its right, as seen in (4.54). This is compatible with my claim that de never
takes a non-kernel element as its complement.
(4.54) a. ∗ Baoyu xie de hen zixi, Daiyu ze kan de hen zixi.

Baoyu write de very careful Daiyu however read de very careful
Intended: ‘Baoyu wrote (something) carefully, but Daiyu read (something)
carefully.’
b. ∗ Baoyu zou de hen man, Daiyu ze pao de hen man.
Baoyu walk de very slow Daiyu however run de very slow
Intended: ‘Baoyu walks slowly, but Daiyu runs slowly.’
Our second fact concerns the occurrence of the focus marker (FM) shi ‘be.’
Shi may occur in various positions, but not between a head and its complement.
This is shown in (4.55) (the _ positions are all possible positions for shi).
(4.55) a. Baoyu zuotian mai-le (∗ shi) na ben shu.

Baoyu yesterday buy-prf fm that cl book
‘Baoyu bought that book yesterday.’
b. Cong (∗ shi) Taipei lai-le (∗ shi) yi wei kexuejia.
from fm Taipei come-prf fm one cl scientist
‘A scientist has come from Taipei.’
The examples in (4.56) show that shi may not occur between de and the kernel
element to its right. This indicates that, as represented in (4.41b), the syntactic
relationship between de and the kernel element is like that between a head and its
complement.
(4.56) a. Na jian hongse de (∗ shi) qunzi zai chuang dixia.

that cl red de fm skirt at bed under
‘That red skirt is under the bed.’
b. Baoyu hen kuai de (∗ shi) pao-zou-le.

Baoyu very quick de fm run-away-prf
‘Baoyu ran away quickly.’
By contrast, in kernel-initial constructions, shi may occur between de and

the non-kernel element to its right, as shown in (4.57). This indicates that de
does not take a non-kernel element as its complement.
(4.57) a. Baoyu xie de shi hen zixi.

Baoyu write de fm very careful
‘Baoyu wrote indeed very carefully.’
b. Daiyu ku de shi hen shangxin.
Daiyu cry de fm very sad
‘Daiyu cried indeed very sadly.’
The focus marker shi has an interrogative version, shi-bu-shi ‘be-not-be,’

which introduces a yes-no question reading to the clause. The distribution
pattern of shi-bu-shi in de constructions is exactly the same as shi:
(4.58) a. Na jian hongse de (∗ shi-bu-shi) qunzi zai chuang dixia? (cf. (4.56a))
that cl red de fm skirt at bed under
‘Is that red skirt under the bed?’
b. Baoyu xie de shi-bu-shi hen zixi? (cf. (4.58a))
Baoyu write de fm very careful
‘Did Baoyu write very carefully?’
The above two arguments support our complementation structure in (4.41b),

repeated in (4.59).
(4.59) YP2
XP Y′2
non-kernel
element Y2 YP1
de kernel element
We elaborate the structure in (4.59) as follows. In this structure, the obligatory

occurrence of XP can be accounted for by either or both of the following views.
First, de as an enclitic needs some phonological material to its left. Second,
de has an edge feature (i.e. an EPP-like feature; see Chomsky 2007 for the
notion of edge feature), which requires the merger of XP at the Spec position.
Therefore, although a non-kernel element can be a modifier of the kernel and
modifiers are optional elements in general, in the presence of de the occurrence

of the non-kernel element becomes obligatory.
The structure proposed in (4.59) is like Zhang’s (1999) n analysis, Simp-
son’s (2003) D analysis (also see Saito et al. 2008), and den Dikken’s (2006)
Linker analysis of de in not grouping de and the non-kernel element together.
However, the current analysis differs from these in a derivational sense, since
here de does not have any intrinsic categorial features, its D or n category is
obtained from the kernel element, if the latter is a nominal, and de cannot be
D or n, if the latter is not a nominal, as in (4.38b) and (4.38c). This analy-
sis better captures the wide variety of de constructions. For example, in den
Dikken (2006), de is analyzed as a Linker, which is introduced in the syntactic
structure as a by-product of the application of Predicate Inversion, a move-
ment operation by which a predicate raises across its subject into a higher
position. However, de also occurs with non-predicative adjectives, such as qita
‘rest’ in (4.36i) and suowei ‘so-called’ in (4.36j). It is unlikely that the left
position of non-predicative adjectives is derived by predicate inversion. My
analysis thus gives a more accurate description of the facts than the Linker
analysis.
The category of a de complex is always identical to that of the kernel element.
As we know, the category of a complex should be projected from the head of
the complex. If de takes the kernel element as its complement, it is the head
of the complex. Since de, like the coordinator and, does not have categorial
features, it has to get categorial features from the kernel element, so that the
category of the projected complex can be specified. This implies that kernel
elements provide de with categorial features.
4.3.4 The chameleon-like nature of de keeps the kernel elements

in situ
In kernel-initial constructions, kernel elements must be verbal. Their position
on the left might be derived by movement, but I leave this speculation for future
research. What is important at this point in the argument is the cooccurrence
of two properties of kernel-final constructions: kernel elements can be any
category and they may not move.
In the following discussion, we set aside bare noun topic constructions like
(4.60a), since it is hard to rule out the possibility that the sentence-initial topic
is base-generated there and thus it is not moved from the gap position. Instead,
the gap position may be taken by a null element, replacable by an overt noun
as in (4.60b).
(4.60) a. Yinliao, ta mai-le san bei leng de .

beverage he buy-prf three cup cold de
‘As for a beverage, he bought three cups of a cold one / something cold.’
b. Yinliao, ta mai-le san bei leng de niunai.
beverage he buy-prf three cup cold de milk
‘As for a beverage, he bought three cups of cold milk.’
Instead, we focus on a set of elements that in constructions without a

category-less linking element may be separated from their modifiers, but in
the presence of such an element (e.g. as the kernel of a de construction) they
may not move. This set consists of proper names and pronouns.
Note first that proper names and pronouns may move in general:
(4.61) {Ta/Baoyu}, wo zao jiu renshi.

he/Baoyu I early then know
‘{He/Baoyu}, I got to know long time ago.’
Moreover, proper names and pronouns may be separated from their mod-
ifiers, such as non-restrictive relative clauses, cross-linguistically (de Vries
2002: 190):
(4.62) Ik heb Joop gezien, die twee zusters heeft. [Dutch]

I have Joop seen who two sisters has
‘I have seen Joop, who has two sisters.’
However, modified proper names and pronouns in Chinese (see footnote 8

of this chapter) require the presence of de, as shown in (4.63a) and (4.64a), and
they may not move, as shown in (4.63b) and (4.64b):
(4.63) a. Diu-le qianbao ∗ (de) ta/Baoyu hen zhaoji.

lose-prf wallet de he/Baoyu very worried
‘He/Baoyu, who lost his wallet, was very worried.’
b. ∗ Ta/Baoyu, diu-le qianbao de hen zhaoji.
he/Baoyu lose-prf wallet de very worried
∗
(4.64) a. Wo congbai boxue (de) Wang Jiaoshou.
I admire knowledgeable de Wang Prof.
‘I admire Prof. Wang, who is knowledgeable.’
b. ∗ Wang Jiaoshou, wo congbai boxue de .
Wang Prof. I admire knowledgeable de
It is a puzzle that de licenses a null or elided kernel element (see (4.52a) and
(4.53a)), but does not license the launching site of the movement of a kernel
4.4 The CCe and the Chinese he/gen constructions 107
element. Generally, it has been noted that the sites of ellipsis are all and only
those that meet the licensing conditions on the launching site of movement
(Zagona 1988a; Chapter 4; 1988b; Lobeck 1987a;1987b). In other words, if a
syntactic position is a possible ellipsis site, it is also a possible launching site
of movement. The fact that de licenses ellipsis but not movement thus needs
an explanation.
We correlate this fact of de constructions with the parallel fact observed in
coordinate constructions. Recall that and also allows the external conjunct to
be null (see Section 4.2), but does not allow movement of this same conjunct.
Importantly, it is the external conjuncts that provide the categorial features to
the whole coordinate complexes headed by and. We now see a parallel situation
in the de constructions. I use the following diagram to show the similarity of
the external conjunct of and (YP1 in (4.65a)) and the kernel element of the de
constructions (YP1 in (4.65b)):
(4.65) a. YP2 b. YP2
YP1 Y′2 XP Y′2
and XP de YP1
(not strandable) (enclitic)
This is the categorial-feature provider for

the whole complex. It may not move.
I concluded in the last subsection that the categorial features of a kernel

element have been transferred to de. The immobility of the kernel element in
de constructions supports the claim stated in Section 4.1.3B: Elements may not
move if they do not have categorial features, which serve as carriers of overt
movement.11
4.4 The CCe and the he/gen comitative constructions in Chinese

The CCe is obeyed in coordinate complexes headed by and-like coordinators,
which have no intrinsic categorial features. In this section, however, I show that
the CCe can be violated in coordinate complexes headed by coordinators that do
have intrinsic categorial features. Recall that the conjunction gen in Chinese has
intrinsic nominal categorial features (Section 3.3.2). We will see that the first
conjunct of a gen coordinate construction may move, although this movement
11 For a discussion of why CC effects are not found in covert wh-movement, see Zhang (2009).
occurs under a certain semantic condition, namely when coordination is non-

distributive.
4.4.1 Introduction: he/gen constructions in Chinese

The words he (han in some Mandarin varieties) and gen are usually translated
as ‘and’ or ‘with.’ Constructions containing he/gen always have two nominals,
DP1 and DP2, such as the underlined parts in the examples in (4.66) through
(4.68).
(4.66) Baoyu {he/gen} Daiyu ge mai-le yi liang che.
Baoyu he/gen Daiyu each buy-prf one cl car
‘Baoyu and Daiyu each bought a car.’
(4.67) a. Baoyu {he/gen} Daiyu he-mai-le yi liang che.

Baoyu he/gen Daiyu co-buy-prf one cl car
‘Baoyu and Daiyu bought a car together.’
b. Gongsi hebing-le disan bumen {he/gen} diliu bumen.
company combine-prf third branch he/gen sixth branch
‘The company combines the third branch and the sixth branch.’
(4.68) Weile gei Daiyu yixie xianjin, Baoyu {he/gen} Daiyu mai-le
to give Daiyu some cash Baoyu he/gen Daiyu buy-prf
yi jia gangqin.
one cl piano
‘In order to give Daiyu some cash, Baoyu bought a piano from Daiyu.’
Semantically, the construction represented by (4.66) is distributive. Distribu-

tive constructions are indicated by the occurrence of a distributive adverb, such
as ge ‘each,’ gezi ‘separately,’ and fenbie ‘respectively.’ In (4.66), the individ-
uals denoted by the two nominals Baoyu and Daiyu take part in car-buying
events separately.
The construction represented by the two examples in (4.67) is non-
distributive or comitative. Such constructions are indicated by the occurrence
of a collective predicate, such as xiangzhuang ‘collide’ and jiehun ‘marry,’
by a collective marker, such as the verbal prefix he- ‘co-,’ or the adverb yiqi
‘together,’ or by a collective verb, such as hebing ‘combine.’12 In (4.66), Baoyu
and Daiyu are both agents of their respective car-buying actions. In (4.67a),
however, Baoyu and Daiyu together are the agent of a single car-buying action.
Similarly, in (4.67b), disan bumen ‘third branch’ and diliu bumen ‘sixth branch’
together are the affected entity of the combining action.
12 See Teng (1970: 355) for a list of collective predicates in Chinese (he calls them “multiple-
reference verbs”).
DP1 and DP2 have the same thematic roles in both the distributive and
comitative constructions. The construction represented by (4.68) is neither
distributive nor comitative. In this sentence, Baoyu is the goal (also the agent)
while Daiyu is the source of the piano-buying action, and thus their semantic
roles are in contrast.
It is generally accepted that the string “DP1 he/gen DP2” is a nominal con-
stituent in distributive examples like (4.66) and post-verbal comitative exam-
ples like (4.67b), and the word he/gen is a coordinator inside this constituent.
Accordingly, the constituent is a coordinate nominal (see Zhu 1982: 176; among
others). If gen occurs to the right of a collective transitive verb such as hebing
‘combine,’ jiajie ‘graft,’ hunyao ‘mix,’ and bijiao ‘compare,’ as in (4.67b), it
cannot be a preposition. This is because in Chinese, no PP adjunct may occur
to the right of a verb. Collective verbs require a plural internal argument. In
(4.67b), neither of the two DPs to the right of the verb is plural. A plausible
analysis of the example is thus that the two DPs and gen form a coordinate
complex, which as a plural nominal satisfies the selection of the collective verb.
Now turn to examples like (4.68). It is not controversial that in such examples
the string “DP1 he/gen DP2” is not a syntactic constituent. Instead, DP1 is the
subject and “he/gen DP2” is an adjunct of the predicate, and the word he/gen
is a preposition or verb.
What is unclear is the syntactic constituency of preverbal comitative exam-
ples like (4.67a), and thus the status of the he/gen there. In 4.4.2 I argue that
in such constructions the string “DP1 he/gen DP2” is also a coordinate nomi-
nal. Then in Section 4.4.3 I present evidence showing the correlation between
collectivity and the possibility of separating the conjunct and coordinator. The
conclusion of this section is that the CCe can be violated in collective he/gen
coordination.
Note that in comitative constructions, the coordinators he and gen have
identical distributions. Thus in all of the comitative examples in this book, the
two words are interchangeable.
4.4.2 Coordinator properties of the comitative he/gen

Early discussion of the close relationship between comitative and coordi-
nate constructions can be found in Jespersen (1924: 90). Cross-linguistically,
the linking words in comitative constructions are often homonymous with
words that have other functions (Mithun 1988: 339, 349; among others). In
Chinese, the diachronic evolution of coordinators was verb > preposition
> coordinator (Liu and Peyraube 1994). In modern Chinese, the words he
and gen may introduce a comitative nominal, as in (4.67a) and (4.67b), a
source, as in (4.68), or a goal, as in (4.69), and they can also be verbs, as

in (4.70).
(4.69) Baoyu gen xuesheng jiang-qi-le guoqu de jingli.
Baoyu gen student tell-inch-prf past de experience
‘Baoyu started to tell his past experiences to the students.’
(4.70) Baoyu zong gen-zhe Daiyu.

Baoyu always follow-prg Daiyu
‘Baoyu always follows Daiyu.’
For the he/gen in comitative constructions, as in (4.67), I use the term com-
he/gen, and for the he/gen that functions as a preposition, as in (4.69) and
(4.68), I use the term prep-he/gen.
In this section, I present four arguments to show the coordinator status of he
and gen in comitative constructions. The first two argue for the constituency
of the string “DP1 com-he/gen DP2.” This constituent status supports in turn
the coordinator status of he and gen in preverbal comitative constructions. The
other two argue against the adjunct status of the string “he/gen DP2” in the
constructions. Among my arguments, only the third (the reversibility of the two
DPs) has previously been mentioned in the literature (Zhu 1982: 176).
A. The string “DP1 com-he/gen DP2” as a topic

My first argument for the constituency of the cluster “DP1 com-he/gen DP2”
is that the string may occur in a topic position.
(4.71) Baoyu gen Daiyu, wo tingshuo yiqi he-xie-le yi

Baoyu gen Daiyu I hear together co-write-prf one
bu xiaoshuo.
cl novel
‘Baoyu and Daiyu, I heard that (they) co-wrote a novel.’
Since only syntactic constituents may occur in topic positions, I conclude

that the cluster “DP1 com-he/gen DP2” must be a syntactic constituent.
B. The string “DP1 com-he/gen DP2” as an antecedent

My second argument for the constituency of the cluster “DP1 com-he/gen DP2”
is that the string may have an appositive, such as liang ge ren ‘two cl person’
in (4.72). Since only syntactic constituents may be antecedents of appositives,
I conclude that the cluster “DP1 com-he/gen DP2” is a syntactic constituent.
(4.72) Baoyu gen Daiyu (liang ge ren) yiqi qu-le Taipei.

Baoyu gen Daiyu two cl person together go-prf Taipei
‘Baoyu and Daiyu, the two persons, went to Taipei together.’
The above two arguments support my claim that “DP1 com-he/gen DP2” is a
syntactic constituent (an additional argument will be given in Section 4.4.3.F).
If so, the cluster “com-he/gen DP2” cannot be an adjunct of the predicate in this
construction. One might, however, still wonder whether the complex-internal
cluster he/gen-DP2 is a PP complement of DP1, like dui Daiyu in (4.73a), or a
PP adjunct of DP1, like yan malu ‘along the street’ in (4.73b).
(4.73) a. dui Daiyu ∗ (de) guanxin

to Daiyu de care
‘the care of Daiyu’
b. yan malu ∗ (de) shangdian
along street de shop
‘the shops along the street’
However, even if we ignore semantics (in (4.67a), for instance, gen Daiyu
is neither an argument nor a modifier of Baoyu), we still have two for-
mal considerations that argue against this preposition analysis. First, in Chi-
nese the complement and any adjunct of a nominal must occur to the left
of the nominal. In [DP1 com-he/gen DP2], the cluster he/gen-DP2 is to
the right of DP1. Second, if a PP modifies a nominal or functions as the
complement of a nominal, the functional word de must be present (Lü
et al. 1999: 19). In the constituent [DP1 com-he/gen DP2], however, de is
absent. It is thus unlikely that com-he/gen DP2 is an argument or modifier of
DP1.
C. The reversibility of DP1 and DP2

The third argument for the coordinator status of com-he/gen is that the two DPs
of the construction can exchange their positions, without affecting the basic
meaning of the construction, whereas the two DPs in prep-he/gen constructions
cannot do so (see Zhu 1982: 176). In (4.74) through (4.76), DP1 and DP2 can
switch their positions without affecting the basic reading.
(4.74) a. Baoyu he Daiyu dingqin-le.

Baoyu he Daiyu engage-prf
‘Baoyu and Daiyu are engaged.’
b. Daiyu he Baoyu dingqin-le.
(4.75) a. Qing he yang he-cheng shui.

hydrogen he oxygen combine-become water
‘Hydrogen and oxygen make water.’
b. Yang he qing he-cheng shui.
(4.76) a. Baoyu he Daiyu he-zhu-le yi zuo fangzi.

Baoyu he Daiyu co-rent-prf one cl house
‘Baoyu and Daiyu rented a house together.’
b. Daiyu he Baoyu he-zhu-le yi zuo fangzi.
The reversibility in the above data is parallel to the reversibility of the
following examples of distributive coordination.
(4.77) a. Baoyu he Daiyu ge zhu-le yi zuo fangzi.
Baoyu he Daiyu each rent-prf one cl house
‘Baoyu and Daiyu each rented a house.’
b. Daiyu he Baoyu ge zhu-le yi zuo fangzi.
In (4.78), however, the preposition gen introduces a source, and if the source
nominal Daiyu is exchanged with the agent nominal Baoyu, the basic meaning
is changed. The encoded transaction direction in (4.78a) is different from that
of (4.78b).
(4.78) a. Weile hua-diao yixie xianjin, Baoyu gen Daiyu mai-le yi

to spend-off some cash Baoyu gen Daiyu buy-prf one
jia gangqin.
cl piano
‘In order to spend some cash, Baoyu bought a piano from Daiyu.’
b. Weile hua-diao yixie xianjin, Daiyu gen Baoyu mai-le yi
to spend-off some cash Daiyu gen Baoyu buy-prf one
jia gangqin.
cl piano
‘In order to spend some cash, Daiyu bought a piano from Baoyu.’
Constituent reversibility is seen between conjuncts of a symmetrical coordi-

nation, but not between elements that have different theta roles. If the comitative
constructions in (4.74) through (4.76) are coordinate constructions, rather than
constructions with PP adjunctions, the constituent reversibility is accounted for.
Note that the reversibility of com-he/gen constructions is merely flexibility
in merger. This flexibility implies that it makes no difference which nominal is
merged with the com-he/gen first, but it does not mean that we have a syntactic
operation that can exchange the positions of two elements.
D. The obligatory occurrence of DP2

Removal of a PP adverbial does not affect the acceptability of a sentence, as
shown in (4.79a) and (4.79b).
(4.79) a. Baoyu gen Daiyu mai-le yi liang che.

Baoyu gen Daiyu buy-prf one cl car
‘Baoyu bought a car from Daiyu.’
b. Baoyu mai-le yi liang che.

Baoyu buy-prf one cl car
‘Baoyu bought a car.’
Like the collective transitive verbs in (4.67b), collective and relational pred-
icates select plural arguments. In a preverbal comitative construction, if DP1 is
singular, the presence of DP2 is obligatory. In examples like (4.67a), repeated
here as (4.80a), if we remove DP2 together with gen, the sentence becomes
unacceptable, as seen in (4.80b). The obligatory occurrence of DP2 together
with gen indicates that they are not a PP adjunct.13
(4.80) a. Baoyu gen Daiyu he-mai-le yi liang che. (= (4.67a))

Baoyu gen Daiyu co-buy-prf one cl car
‘Baoyu and Daiyu bought a car together.’
b. ∗ Baoyu he-mai-le yi liang che.
Baoyu co-buy-prf one cl car
Section summary
Recall that it is not controversial that post-verbal comitative he and gen are
coordinators. Our analysis has now demonstrated that they are coordinators,
not prepositions, in preverbal position as well. We are thus justified in glossing
comitative he and gen as ‘and’ in the remainder of this book.
If comitative constructions are coordinate constructions in Chinese, the
he/gen and the two DPs associated with them form a complex nominal in
their base positions. In this respect, comitative constructions behave the same
as other coordinate constructions, and as we have already shown that he and
gen, as coordinators, have intrinsic categorial features (3.3.2). Moreover, since
the conjuncts of comitative coordination are semantically related in a single
eventuality, the coordination here is collective coordination, in contrast to dis-
tributive coordination.
The conclusion achieved in this subsection predicts that in the context of a
comitative he/gen construction, if DP1 is not adjacent to he/gen, the surface
position must be derived by movement. Since DP1 is an external conjunct,
however, such movement would represent a violation of the CCe.
13 It is true that the English with-phrase is also obligatory with collective predicates taking a
singular nominal as subject:
(i) John is friends ∗ (with Bill).
Zhang (2007a: 146) uses facts like these to argue that comitative with is actually not a preposition,
and thus the behavior of English with does not support the PP adjunct analysis of the Chinese
com-he/gen constructions.
In the next subsection, I will demonstrate that separation of DP1 from comi-
tative he/gen does occur. Based on the existence of this separation and the
nature of its licensing, I will claim that the CCe can be relativized in collective
coordination.
4.4.3 Violation of the CCe in non-distributive coordination

In this subsection, I show that separation of the first conjunct from the rest
of a coordinate complex is possible in collective coordination, in contrast to
distributive coordination.
Since comitative complexes are coordinate complexes, their base positions
should be the same as those of any other coordinate complexes, depending
on the selecting elements. For example, if a comitative coordinate complex
is the subject of a transitive predicate, it is base-generated at Spec of VP. In
this subsection, I will show that the first conjunct can be raised from the base
position of subjects, thus violating the CCe. The raising is observed in six
distinct contexts, as shown in A through F below. In contrast to comitative
coordinate constructions, raising is not possible when distributive coordinate
constructions appear in these contexts. We can thus see a systematic contrast
between the two types of coordinate constructions with respect to CCe effects.
A. Raising verbs between first conjuncts and coordinators

The first indication that first conjuncts may move is seen in examples where
raising verbs appear between first conjuncts and coordinators.
In comitative coordinate constructions A CCe violation is seen in

(4.81), where the first conjunct and the coordinator are separated by the epis-
temic hui ‘might,’ which is a raising verb (C. J. Huang 1988b; Lin and Tang
1996).
(4.81) a. Huoche hui gen qiche xiangzhuang ma?
train might and bus collide Q
‘Might the train collide with the bus?’
It is generally assumed that if a subject precedes a raising verb, the word

order is derived by raising the subject from a position that is c-commanded
by the verb. If so, in (4.81), huoche ‘train’ is raised. In this case, the raising
launches from a coordinate complex, violating the CCe.
The readings of (4.81) can also be expressed as in (4.82). In (4.82), the whole
comitative coordinate complex occurs to the left of the modal. In this case, no
conjunct moves alone, and thus the example is irrelevant to the status of the
CCe.
(4.82) Huoche gen qiche hui xiangzhuang ma?

train and bus might collide q
‘Might the train collide with the bus?’
There is a consistent reading difference between examples like (4.81) and

their counterpart examples like (4.82) from the perspective of information struc-
ture: the two DPs in (4.81) are not symmetrical, whereas the two DPs in (4.82)
are. Specifically, in the absence of any contrastive stress, the DP preceding
the raising verb is foregrounded (emphasized), whereas the one following the
raising verb is backgrounded (not emphasized). In (4.81), huoche ‘train’ is fore-
grounded and qiche ‘bus’ is backgrounded; however, in (4.82), there is no such
difference between huoche and qiche. The reading difference can be captured
by Seiler’s (1974) generalization that comitative constructions leave the extent
of participation of the backgrounded partner in the action underspecified, from
mere “accompanying” to full-fledged “partnership.” The foregrounded partici-
pant of a comitative construction has the property of “Principality” (Teng
1970: 332) in contrast to the other participant. This reading contrast consis-
tently occurs in other comitative constructions in which the first conjunct is
separated from the rest of the coordinate complex.
In distributive coordinate constructions Unlike the case with comi-
tative constructions, in distributive coordinate constructions first conjuncts
cannot be separated from the coordinators by raising verbs. Fenbie in (4.83)
indicates that the coordination is a distributive one. In (4.83) the epistemic
hui ‘will’ cannot appear between the first conjunct Lao Li and the coordinator.
Since no conjunct may be separated from the coordinator in distributive
coordinate constructions, the CCe is obeyed.
(4.83) a. ∗ Lao Li hui gen Lao Wang fenbie qu-le Shanghai ma?
Lao Li will and Lao Wang separately go-prf Shanghai q
b. Lao Li gen Lao Wang hui fenbie qu-le Shanghai ma?
Lao Li and Lao Wang will separately go-prf Shanghai q
‘Might Lao Li and Lao Wang have gone to Shanghai separately?’
B. Adverbials between first conjuncts and coordinators

The second indication that first conjuncts may move is seen in examples where
adverbials appear between first conjuncts and coordinators.
In comitative coordinate constructions In comitative constructions,

first conjuncts can be separated from coordinators by adverbials such as yijing
‘already’ or by temporal or locative adverbials. In (4.84b) the first conjunct

and the coordinator are separated by the circumstantial zai Riben ‘in Japan.’
(4.84b) is nearly synonymous with (4.84a).
(4.84) a. [Akiu gen Baoyu] zai Riben jian-le mian.
Akiu and Baoyu at Japan meet-prf face
‘Akiu and Baoyu met in Japan.’
b. Akiui zai Riben [ i gen Baoyu] jian-le mian.
Akiu at Japan and Baoyu meet-prf face
The following example shows that no resumptive pronoun is allowed for

(4.84b):
∗
(4.85) Akiui zai Riben [tai gen Baoyu] jian-le mian.
Akiu at Japan he and Baoyu meet-prf face
The ban on resumptive pronouns in examples like (4.84b) indicates that the
sentence-initial nominal is not a gapless topic, and is thus not base-generated in
its surface position. Instead, it is moved out of a post-circumstantial position.
Since the conjuncts and coordinator of a coordinate DP complex must
be merged into a constituent which excludes any circumstantial, the occur-
rence of the circumstantial between the first conjunct and the coordinator
in (4.84b) suggests that the conjunct has been moved. If so, the CCe is
violated.
In distributive coordinate constructions Unlike the case with comi-

tative constructions, in distributive coordinate constructions first conjuncts can-
not be separated from the coordinators by adverbials. The distributive adverb
gezi in (4.86) indicates that the coordination is distributive. In such exam-
ples, the first conjunct and the coordinator must be adjacent, consistent with
the CCe.
(4.86) a. ∗ [Baoyu] zai Riben [gen Daiyu] gezi baifang-le pengyou.

Baoyu at Japan and Daiyu separately visit-prf friend
b. [Baoyu gen Daiyu] zai Riben gezi baifang-le pengyou.
Baoyu and Daiyu at Japan separately visit-prf friend
‘Baoyu and Daiyu visited their friends separately in Japan.’
C. Negation words between first conjuncts and coordinators

The third indication that first conjuncts may move is seen in examples where
negation words appear between first conjuncts and coordinators.
In comitative coordinate constructions Like the coordinator and in

English, coordinators in Chinese may not be negated. The coordinators he and
gen are not negated by the negation word bu or mei. However, in comitative
constructions it is possible for a negation word to occur between the first con-
junct and the coordinator. In (4.87), the first conjunct Baoyu and the coordinator
gen are separated by the negation word mei.
(4.87) Baoyu mei gen Daiyu he-mai yi liang che.

Baoyu not and Daiyu co-buy one cl car
‘Baoyu and Daiyu did not buy a car together.’
Since the conjuncts and coordinator of a coordinate DP complex must be

merged into a constituent that excludes any negation word, the occurrence of
the negation word between the first conjunct and the coordinator in (4.87)
suggests that the conjunct has moved. If so, the CCe is violated.
In distributive coordinate constructions It is impossible for a nega-

tion word to occur between the first conjunct and the coordinator in a dis-
tributive coordinate construction. The adverb fenbie in (4.88) indicates that
the coordination is a distributive one. In (4.88a), the negation word bu
may not occur between the first conjunct Baoyu and the coordinator gen.
Thus these distributive coordinate construction examples are consistent with
the CCe.
(4.88) a. ∗ Baoyu mei gen Daiyu fenbie mai yi liang che.
Baoyu not and Daiyu separately buy one cl car
b. Baoyu gen Daiyu mei fenbie mai yi liang che.
Baoyu and Daiyu not separately buy one cl car
‘Baoyu and Daiyu did not each buy a car.’
D. Relativization of first conjuncts

The fourth indication that first conjuncts may move is seen in examples where
first conjuncts are relativized.
In comitative coordinate constructions First conjuncts of coordinate

complexes can be relativized in Chinese, a fact noted in Tai (1969: 122). The
predicate he-zu fangzi ‘co-rent house’ in (4.89a) and xiangzhuang ‘collide’ in
(4.89b) are both collective. They each have a coordinate complex subject, but
the first conjunct of the complex is relativized.
(4.89) a. Gen Daiyu he-zu fangzi de xuesheng hen qiong.

and Daiyu co-rent house de student very poor
‘The student who co-rents a house with Daiyu is very poor.’
b. Wo kanjian-le gen huoche xiangzhuang de qiche.
I see-prf and train collide de bus
‘I saw the bus that collided with the train.’
Relativization is derived by either null operator movement (Chomsky 1977)

or the movement of the antecedent directly (Vergnaud 1974; Kayne 1994). In
either approach, some element is moved from the first conjunct position in
examples like (4.89).
In distributive coordinate constructions Unlike the case with

comitative constructions, first conjuncts cannot be relativized in distribu-
tive constructions. The following two sentences are a minimal pair. In
(4.90a), the collective adverb yiqi ‘together’ occurs, whereas in (4.90b),
the distributive adverb fenbie ‘separately’ occurs. The relativization of
the first conjunct of gen is fine in (4.90a), but not in (4.90b) (see Tai
1969: 134).
(4.90) a. Gen Daiyu yiqi zu fangzi de xuesheng hen qiong.

and Daiyu together rent house de student very poor
‘The student who co-rents a house with Daiyu is very poor.’
b. ∗ Gen Daiyu fenbie zu fangzi de xuesheng hen qiong.
and Daiyu separately rent house de student very poor
E. A-not-A forms of coordinators

The fifth indication that first conjuncts may move is seen in the A-not-A forms
of coordinators.
In comitative coordinate constructions In Chinese, A-not-A ques-

tions are yes-no questions. They are formed by the reduplication of the initial
syllable of a predicate (a verb or preposition) or by reduplication of a larger
prosodic unit of the predicate phrase, and an appropriate negation word (bu ‘not’
or mei ‘not’) between the reduplicant and the base. For instance, in (4.91a),
it is the first syllable of the verb xihuan ‘like’ that is reduplicated, whereas in
(4.91b), it is the whole verb xihuan that is reduplicated. The two sentences in
(4.91) are synonymous.
(4.91) a. Lao Li xi-bu-xihuan ni?

Lao Li like-not-like you
b. Lao Li xihuan-bu-xihuan ni?
Lao Li like-not-like you
‘Does Lao Li like you?’
The position of an A-not-A element marks the division between a VP and

the functional projections above VP, assuming that subjects move out of VP to
SpecIP in Chinese (Huang 1993). It is always the left-peripheral element of VP
that is in the A-not-A form. Usually the element is the verb or the first syllable
of the verb, but it need not be. For instance, if a VP starts with a directional PP,
it is the preposition of this PP that is in the A-not-A form, as seen in (4.92).
An adverb cannot be in A-not-A form (e.g. hai “still” in this example), but
I assume that this is a special property of adverbs that can be left for future
study.
(4.92) Zhe liang che hai wang-bu-wang nan kai?

this cl car still toward-not-toward south drive
‘Is this car still going to move to the south?’
Since the A-not-A formation can be applied to the first syllable of the pred-
icate, it is a syntax–phonology interface operation. Following the basic thesis
of C. J. Huang (1988a), I assume that the [Q] feature of an A-not-A form is
licensed by the relevant feature of a c-commanding functional head, presumably
Infl (or C; the choice is unimportant here).
Since an A-not-A form can mark the left edge of a VP, elements to the
left of the A-not-A form must either be base-generated outside of VP or
have moved out of VP. The data in (4.93) show that there are two possible
A-not-A constructions for a comitative subject. In the construction in (4.93a),
the coordinator gen is in the A-not-A form, whereas the nominal to the left
of the A-not-A word, Baoyu, is the surface subject. If a subject comitative
complex is base-generated at SpecVP, (4.93a) shows that the first conjunct
moves. The movement launches from the base-position of the subject (i.e.
inside the SpecVP) and lands at SpecIP. In this case, the CCe is violated. In
the other construction, the verb is in the A-not-A form, as seen in (4.93b). If
the whole comitative complex DP occurs to the left of the A-not-A word, as in
(4.93b), the whole complex DP is raised from SpecVP to SpecIP. In this case,
no conjunct moves alone, and thus such an example is irrelevant to the status of
the CCe.
(4.93) a. Baoyu gen-mei-gen Daiyu jiehun?

Baoyu and-not-and Daiyu marry
‘Did Baoyu and Daiyu marry?’
b. Baoyu gen Daiyu jie-mei-jiehun?
Baoyu and Daiyu marry-not-marry
‘Did Baoyu and Daiyu marry?’
In distributive coordinate constructions The coordinators of dis-

tributive coordinate constructions cannot appear in A-not-A form. In (4.94),
the presence of the distributive adverb fenbie ‘separately’ indicates that the
construction is a distributive coordinate construction. The intended meaning
of (4.94a) is expressed by (4.94b). In (4.94b), it is the verb, rather than the
coordinator, that is in the A-not-A form (as noted above, adverbs like fenbie
and gezi cannot be in an A-not-A form).
(4.94) a. ∗ Baoyu gen-mei-gen Daiyu fenbie jiehun?

Baoyu and-not-and Daiyu separately marry
b. Baoyu gen Daiyu fenbie jie-mei-jiehun?
Baoyu and Daiyu separately marry-not-marry
‘Did Baoyu and Daiyu get married, separately?’
In (4.94b), the whole coordinate complex precedes the A-not-A form. This
means that the whole complex is outside of VP, and no conjunct is raised by
itself out of the complex. (4.94a) shows that the coordinators of distributive
coordinate constructions cannot be in the A-not-A form. This fact indicates
that in distributive coordinate constructions no conjunct may be raised out of
VP alone, consistent with the CCe.
F. First conjunct-oriented resultatives

The sixth indication that first conjuncts may move is seen in the existence of
first conjunct-oriented resultatives.
In comitative coordinate constructions If a comitative coordinate

complex occurs with a resultative, the sentence is always ambiguous. The
secondary predicate can either predicate over the whole coordinate complex,
or just over the first conjunct:14
14 It is generally recognized that resultatives in Chinese can be subject-oriented, without any

reflexive support (Y. Li 1990; Zhang 2007c; among others).
(4.95) Baoyu gen Daiyu zhuang de toupoxueliu.

Baoyu and Daiyu collide de bleed
‘Baoyu and Daiyu collided so that Baoyu bled.’
‘Baoyu and Daiyu collided so that they both bled.’
The fact that the resultatives can take a single conjunct as subject would be
unexpected if the relevant conjunct remained inside the hosting DP. Accord-
ing to Williams (1980), subjects must c-command their predicates, including
secondary predicates. Thus in order to achieve the first reading for (4.95), the
first conjunct has to move out of the coordinate complex to c-command the
resultative. The empirical issue remains the same if one adopts a PRO approach
to secondary predication (Hornstein and Lightfoot 1987; Bowers 1993; 2001).
Specifically, if the subject of a secondary predicate is a local PRO, which c-
commands the secondary predicate, the controller of the PRO still needs to
be a constituent outside of the coordinate complex. In this PRO-approach, the
first conjunct in (4.95) is the controller of the PRO subject of the secondary
predicate. As the controller of the PRO, the first conjunct must be syntactically
outside of the coordinate complex.
On the other hand, the fact that resultatives can take whole coordinate com-
plexes as subjects, seen in the second reading for (4.95), indicates that the
raising of the first conjunct is not obligatory. This in situ possibility is parallel
to what we saw in the previous subsections. In all of these tests, there are
always two possibilities for the first conjuncts: they can either be raised or
remain inside the coordinate complexes.
Note that the ambiguity remains even if DP1 is separated from the string
gen-DP2:
(4.96) Baoyu zuotian gen Daiyu zhuang de toupoxueliu.

Baoyu yesterday and Daiyu collide mod bleed
‘Baoyu and Daiyu collided so that Baoyu bled yesterday.’
‘Baoyu and Daiyu collided so that they both bled yesterday.’
The existence of whole-coordinate-complex-oriented resultatives, seen in

the second reading of the above data, on the other hand, indicates that the
resultative may also take the whole coordinate complex in its base-position as
its subject. Since resultatives are generally analyzed as the complement of V,
it is c-commanded by the base-position of the coordinate subject, which is at
Spec of VP. Therefore, the predication relation is licensed.
The fact that resultatives can take whole coordinate complexes as subjects
also supports our claim that the two DPs form a constituent. The above examples
stand in contrast to the following. In the intended source reading of gen, the
gen-DP cluster is a PP, and the resultative is exclusively secondary predicate of
the DP to the left of gen, i.e., Baoyu in (4.97).
(4.97) Zhe ben shu, Baoyu gen Daiyu jie de dou fan-le.
this cl book Baoyu from Daiyu borrow de already tired.of-prf
‘This book, Baoyui borrowed from Daiyu so that hei already got tired.of it.’
Irrelevant reading: ‘This book, Baoyu and Daiyu borrowed from others so
that they both got tired of it.’ (distributive coordinate construction, gen
is a coordinator)
The lack of ambiguity in the reading of the prep-gen construction in (4.97) is

expected from the fact that the cluster Baoyu gen Daiyu is not a constituent, and
thus it cannot have a secondary predicate. Instead, only the DP to the left of gen,
which is the subject of the primary predication, can have a secondary predicate.
In distributive coordinate constructions In contrast to (4.95), the first

conjuncts of distributive coordinate constructions can never have any secondary
predicate. The following sentence is not ambiguous. Only the whole coordinate
complex can be the subject (or controller of the subject) of the resultative. We
thus see yet another contrast between comitative coordinate constructions and
distributive coordinate constructions regarding single conjunct raising.
(4.98) Baoyu he Daiyu qi de zhi fadou.

Baoyu and Daiyu angry de continuously shiver
‘Baoyu and Daiyu were so angry that they shivered continuously.’
G. Section summary
The six contrasts above between comitative and distributive coordinate con-
structions with respect to the mobility of first conjuncts show that CCe effects
are sensitive to the semantic type of the coordination: the first conjunct can
move in comitative coordination but not in distributive coordination.15
4.5 Chapter summary

In Chapter 3, I claimed that the categorial features of external conjuncts are
transferred to the coordinator and. In this chapter, I concluded that this transfer-
ence keeps external conjuncts in situ, since elements without category-features
15 One might wonder why external conjuncts may not be moved from a postverbal position,
regardless of the semantic type of the coordination. It is possible that there is some sort of
intervention effect for non-local movement like this, but I will leave this issue for future
research.
may not move overtly. Our analysis explains the effects that the CCe has been
intended to cover.
Our account for the CCe is supported by two facts. First, in Chinese de
constructions, kernel elements provide categorial features for de, which has
no intrinsic categorial features and is the head of the whole complex, and so
they may not move. Second, in comitative coordinate constructions in Chinese,
initial conjuncts do not provide categorial features for the coordinators, since
the coordinators have their own intrinsic categorial features (Section 3.3.2),
and so the conjuncts may move.
This account of the CCe can be falsified if we find coordination in which the
external conjunct may move and yet the coordinator has no categorial features.
The CCi, however, simply manifests a morphological property of coordina-
tors shared with many other types of elements, namely that they need to be
adjacent to their complement. Neither deletion nor movement of their comple-
ment (the internal conjunct) is possible.
Putting these points together, the CC is not a unified construction-specific
syntactic constraint. Instead, it is related to the lexical/morphological makeup
of coordinators such as and. In syntactic theory, the CCe, as part of the CSC,
has usually been treated as a constraint specific to the elements merged with
coordinators. However, as shown in Chapter 2, coordinate constructions do not
represent any special syntactic structure, and as shown in Chapter 3, coordina-
tors do not represent any special category. Thus, coordination is not a primitive
notion in syntax. Conjuncts, as regular Spec and Complement elements, may
undergo syntactic movement in general. Apparent CC effects actually derive
from idiosyncratic morphological properties of particular coordinators. In par-
ticular, English and is unspecified for categorial features, and the transference
of categorial features to it from the external conjunct blocks this conjunct from
moving. In the minimalist program, movement is driven by morphological con-
siderations (Chomsky 1995: 262). My study here shows that the blocking of
movement can also be related to morphological properties of specific syntactic
elements, in addition to the generally recognized locality restrictions (Chomsky
1975: 105).16
16 Given our claim that the CC depends on the nature of the overt coordinator, it follows that
conjuncts of a coordinate complex that does not have an overt coordinator (i.e. asyndetic
coordination) may move. In Chapter 6, we will show how this prediction explains the derivations
of certain constructions.
5 The Element Constraint and the
semantic relatedness of conjuncts
5.1 Introduction
Recall that Ross’s (1967) Coordinate Structure Constraint (CSC) has two parts,
the Element Constraint (EC) and the Conjunct Constraint (CC) (Grosu 1973).
The EC states that no element may be extracted from conjuncts, as seen in
(5.1a), and the CC states that no conjunct may move, as seen in (5.1b).
∗
(5.1) a. What kind of herbs did you [[eat ] and [drink beer]]? (EC violation)
∗
b. Which boy did John kiss [ and that girl]? (CC violation)
In the last chapter, I distinguished between two kinds of CC effects: those

caused by the morphological properties of and-like coordinators and those
caused by the distributiveness of coordination. The CC is absolute in coordinate
complexes headed by and-like coordinators, but can be violated in other types of
coordinate complexes, in non-distributive coordination. In this chapter, I present
a parallel correlation between violation of the EC and non-distributiveness.
Thus both parts of the CSC are relativized in non-distributive coordination.
Specifically, I will discuss the evidence showing that the EC can be violated in
so-called asymmetrical coordination (AC), but not in symmetrical coordination.
The goal of this chapter is to show that the EC depends solely on semantics
and thus is not a constraint on syntactic operations.
The chapter is organized as follows. In Section 5.2, I introduce the contrast
between natural and accidental coordination, and identify AC, as well as col-
lective coordination, as a type of natural coordination. In Section 5.3, I discuss
violation of the EC in natural coordination in both English and Chinese. Finally,
Section 5.4 is a brief summary of the chapter.
5.2 Asymmetrical coordination as a type of natural coordination

5.2.1 Natural coordination
Natural coordination is a semantic relation in which the entities expressed
by the conjuncts of a coordinate complex are closely related to each other
124
5.2 AC as natural coordination 125
(Haspelmath 2004; 2007; Wälchli 2005; Dalrymple and Nikolaeva 2006; see
also Mithun 1988: 332f.). A close relation is indicated if any element in the
context is semantically associated to the combined meaning of the conjuncts,
rather than to the meaning of each isolated conjunct. In contrast, accidental
coordination involves coordination of elements which do not have a close
semantic relationship with each other. For instance, if a coordinate complex
is associated with a relational adjective such as compatible or similar, the
conjuncts are semantically related and the coordination is natural, whereas if a
coordinate complex is associated with a non-relational adjective such as smooth,
the conjuncts are not semantically related to each other and the coordination is
accidental (see Dalrymple and Nikolaeva 2006: 825, 834).
Non-distributive coordination is natural coordination. Coordinate complexes
that are arguments of collective predicates or collective verbs, such as John
and Mary in (5.2), are non-distributive. In such coordinate complexes, neither
conjunct may function as an independent argument of the predicate or verb
(∗John collided in the street).
(5.2) [John and Mary] collided in the street.
The dichotomy between accidental and natural coordination covers various

dichotomies proposed in the literature: sentence conjunction vs. phrasal con-
junction (Lakoff and Peters 1966), non-joint vs. joint coordination (McCawley
1968), non-Boolean conjunction vs. Boolean conjunction in semantics (Massey
1976; Link 1983; Hoeksema 1983; Krifka 1990; among others), jia-er-bu-he
(adding-without-joining) and jia-erqie-he (adding-and-joining) (Lü 1979: Sec-
tion 75), and [+/− Separate] coordination (Payne 1985: 17). All of these
dichotomies depend on whether the conjuncts are semantically related to each
other or are essentially unrelated.
According to Haspelmath (2004; 2007), Wälchli (2005), and Dalrymple
and Nikolaeva (2006), the classification of natural and accidental coordination
is decided by both linguistic and non-linguistic contexts. Lexical semantics,
pragmatics, morphology, and syntactic forms all can play roles in distinguishing
the two types of coordination.
The semantic distinction between the grouping in natural (or collective)
coordination and the grouping in accidental (or distributive) coordination can
also be seen in plural nominals. We have shown that conjuncts of natural
coordination are related to each other and thus the whole coordinate complex
shows semantic integrity; by contrast, conjuncts of accidental coordination are
not related to each other and thus the whole coordinate complex does not show
semantic integrity. The accidental coordination reading of (5.3a) is that John
126 The EC and the semantic relatedness of conjuncts
can draw the cart and Peter can also draw the cart, while its natural coordination
reading is that they can draw the cart together. It has long been recognized that
this contrast also exists in plural nominals. (5.3b) is ambiguous in the same
way as (5.3a). In its natural grouping reading, it means that the men can draw
the cart together, whereas in its accidental grouping reading, it means that each
of the men can draw the cart individually.
(5.3) a. John and Peter can draw the cart. (Dik 1968: 87)
b. The men can draw the cart.
We can see the contrast between the two types of readings for plural nom-
inals not only in predication, but also in modification. In the natural grouping
reading of (5.4), the multiple individuals denoted by stories contradict each
other, whereas in the accidental grouping reading, each of them is internally
inconsistent.
(5.4) Jack told inconsistent stories. (Schwarzschild 2002: 16)
We highlight this parallelism between coordinate complexes and non-

coordinate elements in order to emphasize the fact that the contrast between
natural and accidental grouping does not depend on the construction being a
coordinate construction.
The natural and accidental grouping readings of plural nominals may also
co-exist in the same sentence (data like (5.5) are seen in Dowty 1987: 98 and
Lasersohn 1995: Chapter 5, among others):
(5.5) a. John and Mary are a happy couple and are well-adjusted individuals too.
b. John and Mary met at the bar and had a drink.
c. The students closed their notebooks, left the room, and then gathered in the
hall after the class.
In (5.5a), John and Mary is a natural coordination with respect to the first
conjunct are a happy couple, which is a collective predicate, but an accidental
coordination with respect to the second conjunct are well-adjusted individuals,
which is a distributive predicate. Similarly, in (5.5b) John and Mary is a nat-
ural coordination with respect to the first conjunct met at the bar, which is a
collective predicate, but an accidental coordination with respect to the second
conjunct had a drink, which is a distributive predicate. In (5.5c), the plural
nominal the students encodes an accidental grouping with respect to the first
and second conjunct, but a natural grouping with respect to the last conjunct
then gathered in the hall after the class.
5.2.2 Asymmetrical coordination

Natural coordination can be either nominal or verbal (Dalrymple and Nikolaeva
2006: 830). In this subsection I identify so-called asymmetrical coordination as
natural coordination of verbal elements. In examples like (5.6), the coordination
is not asymmetrical, since the conjuncts do not have to be semantically related
to each other, or encode a natural sequence of events. In (5.6a), for instance,
to play the piano has no semantic relation with to learn exotic languages. In
such accidental coordination examples, the conjuncts do not form an integrated
semantic whole. Goldsmith (1985: 134) calls the use of and in accidental
coordination “truth-conditional and.”
(5.6) a. Our first contestant likes to play the piano and (to) learn exotic languages.
b. He did some weeding and wrote a few pages of the paper.
Since Schmerling (1975), Goldsmith (1985) and Lakoff (1986), semantic

types of coordination other than the “truth-conditional” type have been explored
extensively in languages such as English (Culicover and Jackendoff 1997)
and German (Höhle 1990, among others). Goldsmith calls the conjunction
“temporal and” in examples like (5.7a), “causal and” in examples like (5.7b),
“despite” or “nonetheless and” in examples like (5.7c). One might like to add
the term “conditional and” to cover examples like (5.7d) (Bar-Lev and Palacas
1980; Culicover and Jackendoff 1997).
(5.7) a. Mary bought the newspaper after work and she read it on the train.
b. The light went off and I couldn’t see.
c. How many courses can we expect our graduate students to teach and (still)
finish a dissertation on time?
d. You drink another can of beer and I’m leaving. (= If you drink one more
can of beer, I’m leaving.)
Differences in the various uses of and in (5.7) are not syntactically signifi-
cant, and such examples have been covered by the general term asymmetrical
coordination (AC) or Fake Coordination in the literature. Like the accidental
coordination constructions in (5.6), AC constructions may express plural even-
tualities. What is special is that AC always encodes a certain semantic relation
between the conjuncts, in which the conjuncts form an integrated semantic
whole. Analyzing data like (5.7), Culicover and Jackendoff (1997: 213 fn. 13)
explicitly claim that “we understand the two events as being connected as parts
of a larger event; they did not occur independently, on different ‘occasions,’
so to speak.” In Huddleston and Pullum (2002: 1283), AC is identified as joint
coordination, yet another term for natural coordination.
In asymmetrical or natural coordination, conjuncts are semantically related to

each other in some dimensions; Goldsmith’s terms simply name the dimensions.
As pointed out by Dik (1968: 271), “the semantic aspect of and does not in itself
specify the kind of combination any further, but can give rise to a multitude
of different relations in the final interpretation.” Generally, it is the linguistic
(e.g. word order) and non-linguistic contexts that indicate the exact semantic
relation between the conjuncts in AC. For instance, by itself, and does not
indicate the differences in semantic relations between the conjuncts in the two
examples in (5.8).
(5.8) a. She died, and they buried her.

b. They buried her, and she died.
Nevertheless, the accidental vs. natural status of coordination can be signaled

by certain expressions. For instance, and similarly and and . . . too encode
accidental coordination, whereas and therefore, and as a result, and and then
encode natural coordination or AC. Although conjuncts in natural coordination
might function semantically like modifiers, they do not have an adjunction rela-
tion to another element syntactically. This has been argued for in Culicover and
Jackendoff (1997) and Kehler (2002: 61). I thus assume, as argued in Section
2.4, that AC is true coordination (contra Schachter 1977: 100 and Postal 1998).1
5.2.3 Some formal contrasts between natural and accidental coordination

The contrast between natural and accidental coordination can be formally
represented in various ways, beyond the well-recognized contrast between
collective adverbs such as together and distributive adverbs such as separately.
A. The coordinator ji in Chinese

Conjunctions can be exclusively distributive. The conjunction ji in Chinese is
such a conjunction, as demonstrated by the following evidence.
First, coordinate complexes conjoined by the conjunction ji, unlike those
conjoined by he and gen, cannot be the subject of collective predicates. In
(5.9a), for instance, shi pengyou ‘be friends’ is a collective predicate, and so
is shi yi dui fu-qi ‘be husband and wife.’ The subjects of the predicates can be
conjoined by he or gen, but not ji. (5.9b) shows the same point.
1 Note that although most of the examples above contain VP conjuncts, some of them have TP
conjuncts (going beyond the range of examples cited by Goldsmith 1985; Lakoff 1986; and
Culicover and Jackendoff 1997). Therefore, Postal’s (1998) claim that AC is restricted to VPs is
not accurate.
(5.9) a. Baoyu {he/gen/∗ ji} Daiyu shi pengyou/yi dui fu-qi.

Baoyu and/and/and Daiyu be friend /one pair husband-wife
‘Baoyu and Daiyu are {friends/a couple}.’
b. Baoyu {he/gen/∗ ji} Daiyu zai huayuan jianmian-le.
Baoyu and/and/and Daiyu at garden meet-prf
‘Baoyu and Daiyu met in the garden.’
Second, collective verbs are incompatible with objects conjoined by ji. In

(5.10), the verb hunyao “confuse” is collective, and the components of the
object cannot be conjoined by ji.
(5.10) Akiu hunyao-le lundian {he/gen/∗ ji} lunju.

Akiu confuse-prf claim and/and/and argument
‘Akiu confused the claims and the arguments.’
Third, unlike coordinate complexes conjoined by he and gen, those conjoined

by ji are never ambiguous between distributivity and collectivity. (5.11a) can
be ambiguous; although the default reading is collective (i.e. Baoyu has been
engaged to Daiyu), (5.11a) can also have a distributive reading, where Baoyu is
engaged to a person other than Daiyu, and Daiyu is engaged to a person other
than Baoyu. (5.11b), however, must be distributive, because of the presence
of the distributive adverb fenbie ‘separately.’ Like (5.11b), (5.11c) is also not
ambiguous, having only a distributive meaning, even though no distributive
adverb is present.
(5.11) a. Baoyu {he/gen} Daiyu dingqin-le. (default: collective)

Baoyu and/and Daiyu engage-prf
b. Baoyu {he/gen} Daiyu fenbie dingqin-le. (distributive)
Baoyu and/and Daiyu separately engage-prf
‘Baoyu and Daiyu are both engaged (separately).’
c. Baoyu ji Daiyu dingqin-le. (distributive)
Baoyu and Daiyu engage-prf
‘Baoyu and Daiyu are both engaged.’
In short, whereas he and gen can occur in either collective or distributive

contexts, the conjunction ji occurs in distributive contexts only.
Since ji is a distributive conjunction, the external conjunct of a ji-coordinate
construction may not be extracted. In (5.12a), the external conjunct Akiu is
adjacent to the conjunction gen, while in (5.12b), the external conjunct Akiu
and gen are separated by the circumstantial zai Riben ‘in Japan’ (see Section
4.4.3B). In the two examples in (5.12), the predicate is collective and thus
the coordination is a natural one. In (5.13a) the external conjunct Shizhang

‘mayor’ is adjacent to the distributive conjunction gen, while in (5.13b), the
external conjunct Shizhang and ji are separated by the circumstantial zai Riben
‘in Japan.’ However, unlike (5.12b), (5.13b) is not acceptable.
(5.12) a. [Akiu gen Baoyu] zai Riben jian-le mian. (= (4.85))
Akiu and Baoyu at Japan meet-prf face
b. Akiui zai Riben [ i gen Baoyu] jian-le mian.
Akiu at Japan and Baoyu meet-prf face
Both: ‘Akiu and Baoyu met in Japan.’
(5.13) a. [Shizhang ji qi furen] zai Riben canguan-le yi ge youeryuan.

Mayor and his wife at Japan visit-prf one cl kindergarten
‘The mayor and his wife visited a kindergarten in Japan.’
b. ∗ [Shizhang]i zai Riben [ i ji qi furen] canguan-le yi
Mayor at Japan and his wife visit-prf one
ge youeryuan.
cl kindergarten
Exclusively distributive conjunctions are also found in other languages, such
as i ‘and’ in Czech (Skrabalova 2005). In Udihe, the word zuŋe ‘and’ is used
for accidental coordination and the word mule ‘and, with’ is used for natural
coordination (Dalrymple and Nikolaeva 2006: 831). In Lenakel, the tight con-
junction m is for natual coordination and the loose conjunction məne is for
accidental coordination (Haspelmath 2004: 13).
B. Singular vs. plural agreement in English

One formal property of natural coordination in English is that it may trigger
singular agreement on verbs, auxiliaries, or pronouns (Moltmann 1997: 233ff.;
Huddleston and Pullum 2002: 1283).
As noted in earlier chapters, agreement in coordination is affected by multiple
factors, including notional information, lexical plurality, and linear (surface)
word order (Lorimor 2007; Steiner 2008). Here we focus on notional infor-
mation. Consider the variation in number agreement seen in the following
coordinate complexes (McNally 1993: 363; Moltmann 1997: 238; Sauerland
2003: (11)):
(5.14) a. Bread and butter is all we had to eat.
b. The bread and wine is on the table.
c. Bread and butter {are/∗ is} found in aisles 1 and 7, respectively.
d. Strawberries and cream is on the menu.
(5.14c) contains the distributive adverb respectively, making the coordination
accidental. Thus only the plural number agreement is possible, in contrast to
(5.14a), (5.14b), and (5.14d), where the singular number agreement is allowed.
The coordination in these three examples can be regarded as natural coordina-
tion.
Another relevant set of examples is the following:
(5.15) a. Three and five {is/are/makes/make} eight.

b. Three and five {are/∗ is} odd numbers.
(5.16) a. Three and five is even.

b. Three and five {are/∗ is} odd.
In (5.15a), neither conjunct alone can be the subject of the predicate eight.
The subject is therefore a natural coordination complex, which may take the
singular form of the copula. In (5.15b), however, each of the conjuncts in the
subject can be independently related to the predicate odd numbers, ruling out
natural coordination. Therefore, the plural form of the copula is used. Similarly,
the singular copula is is used in (5.16a) because even reflects the sum reading
of the coordinate subject. In contrast, the singular copula is cannot be used in
(5.16b) because odd is not the predicate of the sum reading of the coordinate
subject but of its distributive reading.
In the literature on the syntax of coordination, the fact that coordinate com-
plexes can trigger the singular number agreement in certain cases has been
studied by many scholars (Dik 1968: 210; McCloskey 1991: 594; Aoun et al.
1994; Aoun and Benmamoun 1999; Munn 1993: Section 2.5.7; 1999; Johan-
nessen 1996; 1998; among others). This quirky agreement is not restricted to
coordination, but is also found in nominals with a PP modifier (e.g. the gang
on the motorcycles and the gang near the motocycles; see Humphreys and
Bock 2005) (I thank Wayne Cowart for bringing my attention to the work by
Humphreys and Bock) and possessive nominals (den Dikken 2001: 22, 24). In
(5.17b) and (5.18b), the plural possessor participants triggers plural agreement
with the copula, regardless of its surface position. Similarly, the doors in (5.19b)
triggers plural agreement, although it is not the head of the complex nominal.
(5.17) a. The participants’ identity is to remain a secret.

b. The participants’ identity are to remain a secret.
(5.18) a. The identity of the participants is to remain a secret.

b. The identity of the participants are to remain a secret.
(5.19) a. The key to all the doors is missing.

b. The key to all the doors are missing.
I believe that McNally (1993: 363) is right in claiming that agreement resolu-
tion can depend on the denotation of the nominal complexes (see also Dougherty
1970b: 853 for an early discussion of the contrast between “semantic” and “syn-
tactic” agreement). Specifically, agreement depends on the contrast between
natural coordination and accidental coordination, where the former can trigger
singular agreement while the latter cannot. Since this contrast is not defined
(purely) syntactically, agreement cannot be used as a diagnostic in the syntac-
tic analysis of coordination (see Section 2.3.1). This conclusion fits with the
following generalization of McCloskey (1991: 565):
The semantic condition governing such agreement seems to be that plural

agreement is possible just in case the conjoined propositions are contradic-
tory or incompatible, or, more generally, when they specify a plurality of dis-
tinct states of affairs or situation-types. When the coordinated clauses denote
compatible propositions (that is, when they denote two or more propositions
that jointly specify a single complex state of affairs or situation-type), then
singular agreement is preferred or required.
McCloskey’s generalization is confirmed by Moltmann’s (1997: 238) obser-

vation that when both singular and plural agreement are possible, the latter
is used when the meanings of the two conjuncts are somehow “unrelated.” A
similar idea is seen in Sauerland (2003: Section 2.1).
Data like (5.14) above involve mass noun conjuncts. Coordinate complexes
composed of conjuncts of other semantic types can also trigger singular agree-
ment (Hegarty 2003; Moltmann 1997: 235). One such example is (5.20)
(McCloskey 1991: 565) (the Dutch counterpart of (5.20) shows the same accept-
ability contrast; Petra Hendriks, p.c.).
(5.20) That UNO will be elected and that sanctions will be lifted {is/??are} now
likely.
C. Other morphological representations

In the Finnish example in (5.21a), no single conjunct may be modified by
the relational adjective yhteensopivat ‘compatible’ and thus the coordination
yliopisto ja teatteri ‘university and theater’ is natural coordination. In this
language, natural coordinate complexes may be modified by plural adjectives,
as in (5.21a), whereas accidental coordinate complexes may not be modified
by plural adjectives, as shown in (5.21b). The intended meaning of (5.21b) is
expressed by (5.21c), in which each conjunct is modified by a singular adjective
(Dalrymple and Nikolaeva 2006: 826, 829).
(5.21) a. On vaikeaa suunnitella {yhteensopivat/∗ yhteensopiva} yliopisto

is difficult design compatible.pl/compatible.sg university
ja teatteri.
and theater
‘It’s difficult to design a compatible [university and theater].’
b. ∗ Tässä kaupungissa on hyvät yliopisto ja teatteri.
this.ines city.ines is good.pl university and theater
‘There is a good [university and theater] in this city.’
c. Tässä kaupungissa on hyvä yliopisto ja hyvä teatteri.
this.ines city.ines is good.sg university and good.sg theater
‘There is a good university and a good theater in this city.’
The contrast between natural and accidental coordination is also seen in

the following case patterns in Korean (Yoon and Lee 2005). In (5.22a), which
has a collective reading, the nominative case marker occurs only on the final
conjunct, whereas in (5.22b), which has a distributive reading, the nominative
case marker occurs on both conjuncts.
(5.22) a. John-kwa Mary-ka cip-ey ka-ss-ta
John-and Mary-nom home-loc go-pst-decl
‘John and Mary went home together.”
b. John-i kuliko Mary-ka cip-ey ka-ss-ta
John- nom and Mary-nom home-loc go-pst-decl
“John and Mary went home separately.”
Similar case patterns are found in Hebrew. In this language, the accusative
case marker et occurs only once, in front of the entire coordinate complex, if
the complex has a collective reading, as in (5.23a). By contrast, it precedes each
definite conjunct if the definite coordinate complex has a distributive reading,
as in (5.23b). Accordingly, if we use a verb which forces a collective reading
of the object, such as hifrid ‘separated’ (or hifgish ‘caused to meet’ or ixed
‘united’), only one et shows up, as seen in (5.24) (Winter 2001).
(5.23) a. Dan ar’a et Ruti ve Sara.
Dan saw acc Ruti and Sara
‘Dan saw Ruti and Sara together.’
b. Dan ar’a et Ruti ve et Sara.
Dan saw acc Ruti and acc Sara.
‘Dan saw Ruti and Dan saw Sara, separately.’
(5.24) a. Dan hifrid et Ruti ve Sara.
Dan separate acc Ruti and Sara
‘Dan separated Ruti and Sara.’
b. ??Dan hifrid et Ruti ve et Sara.
The genitive case marker ’s in English coordinate complexes also occurs once
if the reading is collective and on each conjunct if the reading is distributive.
(5.25a) means that John and Harry departed together and (5.25b) means that
they did so separately (McCawley 1968: 161):
(5.25) a. John and Harry’s departure for Cleveland [collective]

b. John’s and Harry’s departure for Cleveland [distributive]
Yoon and Lee (2005) argue that distributive examples like (5.22b) are derived
from clausal coordinate constructions by reduction, and the case of each case-
marked nominal is related to one case-assignor in the clause containing it. If
so, the possibility of reduction is again licensed by distributivity. Their analysis
may also be applicable to the Hebrew (accusative case) and English (genitive
case) constructions described above.
D. Other formal representations

The contrast between natural and accidental coordination is also reflected in
accent, gapping, and some special constructions.
If the semantics of the conjuncts in a coordinate complex are in contrast,
parallel elements in the conjuncts can bear contrastive accents. This is not
the case for AC. For instance, (5.26a) has two readings. It can mean either
that the events expressed by the two conjuncts are independent (distributive)
or that the first event is the cause of the second event (collective). Hendriks
(2004: 6) states that if all constituents but the finite verbs are pronounced with
contrastive accent, as in (5.26b), the AC reading seems to disappear, or at least
becomes highly marginal. In other words, distributive readings are compatible
with contrastive accent but collective readings are not.
(5.26) a. Sue became upset and Nan became downright angry. (AC possible)
b. SUE became UPSET and NAN became DOWNRIGHT ANGRY. (∗ AC)
Levin and Prince (1986) (also see Moltmann 1992a: 49; Hendriks 2004;
Chaves 2007: 30) observe that gapping is hardly ever found in AC. Compare
(5.26a) with its gapping counterpart in (5.27).
(5.27) Sue became upset and Nan downright angry. (∗ AC)
Unlike (5.26a), which has the two readings noted above, (5.27) has only the
distributive reading. In other words, distributive but not collective readings are
possible in gapping.
The contrast between distributivity and collectivity is also seen in so-called
SGF-coordination (Subject Gap in Fronted/Finite Clause Coordination, Höhle
5.3 The EC violation in AC 135
1990) in Germanic languages, as in (5.28). In this construction, the subject gap

in the second conjunct is referentially related to the subject of the first conjunct;
however, the subject of the first conjunct does not occur in the position parallel
to that of the gap in the second conjunct. In (5.28), the subject der Jäger ‘the
hunter’ follows the verb in the first conjunct, whereas the gap in the second
conjunct precedes the verb, assuming both conjuncts are in a V2 order.
(5.28) In den Wald ging der Jäger und fing einen Hasen. [German]
into the forest went the hunter and caught a rabbit
The construction is felicitous only with an AC reading, such as cause –

effect and contiguity readings (Frank 2002; Hendriks 2004: 15; see also John-
son 2002). This restriction means that the coordination in this construction is
natural, not accidental.
In this section, I have summarized various formal ways to represent the
contrast between natural and accidental coordination. The following sections
link this contrast with the EC.
5.3 The EC violation in asymmetrical coordination

In AC, extraction from either of the two conjuncts is generally possible, indi-
cating that there is no EC effect in natural coordination. Many such EC viola-
tions are reported and discussed in Ross (1967: Section 5.2.2), Grosu (1973),
Schachter (1977: 299), Goldsmith (1985), and Lakoff (1986), Heycock and
Kroch (1994: 272), Culicover and Jackendoff (1997: 201), Wilder (1999: 9),
Levine (2001: 156–161), Cormack and Smith 2005). (5.29) lists some exam-
ples.
(5.29) a. This is the thief that [you just point out the loot] and then we arrest [on the
spot]. (relativization)
b. How much wine can you drink and still stay sober? (wh-ques)
c. Swiftly John will run and end up falling down. (Adv-fronting)
(Benjamin Shaer, p.c.; see Shaer 2003: 243)
d. Off the boy went , and told his friends the news.
e. [This advice] the committee decided to follow and proceeded to set up a
new subcommittee. (nominal-topic)
f. Kiss her, I didn’t , and will probably regret it. (VP-topic)
In (5.29a), relativization occurs in the second conjunct. In (5.29b), the wh-

phrase moves from the first conjunct. In (5.29c) and (5.29d), manner and
direction adverbs, respectively, are fronted from the first conjunct. In (5.29e)
and (5.29f), the nominal topic and VP topic are fronted, respectively. (5.30)
shows that extraction is possible from either conjunct in AC (Lakoff 1986;
Kehler 2002: 128):
(5.30) a. Which knee did Terry run in these shoes and hurt ?
b. Which shoes did Terry run in and hurt her knee?
The examples in (5.29) and (5.30) show that the EC can be violated in
AC. The following examples in (5.31) show that when the EC is violated, the
coordination must be asymmetrical. The wh-element what moves from a single
conjunct in (5.31b), but nothing moves out of a single conjunct in (5.31a).
Carlson (1987: 539; see also Kehler 2002: 127, 139) observes that in (5.31b)
there must be a natural consequential relation between the two eventualities,
whereas (5.31a) allows a (less salient) reading in which the two eventualities
are not related.2
(5.31) a. John went to the store and bought some ice cream.
b. What did John go to the store and buy?
Cross-linguistically, the absence of EC effects in AC is reported in Johan-

nessen (1998: Chapter 6) and Peterson and VanBik (2004: 345–347). Violation
of the EC is also seen in Chinese. The conjunction erqie ‘and’ occurs in either
accidental coordination or the consecutive type of natural coordination. If erqie
is used in a consecutive sense, extraction from the first conjunct is possible,
as seen in (5.32): one has to read something before writing any notes about
it, so the two events have a consecutive relation. In (5.32a), the topic na fen
baozhi ‘that newspaper’ is associated with the object gap of the first conjunct.
In (5.32b), the relativized nominal baozhi ‘newspaper’ is associated with the
object gap of the first conjunct.
2 Kehler (2002: 139) reports that the repetition of the subject I in (ii) is not compatible with the
extraction from a single conjunct:
(i) I went to the store and I bought some whiskey.
(ii) This is the whiskey which I went to the store and (∗ I) bought.
Recall that (i) is ambiguous between a one-event reading and a two-event reading (see
Section 3.5). Only the former is an AC reading. Under the latter reading, the EC cannot be
violated, and thus (ii) is not acceptable if it shares the base-structure with (i), i.e., I is present
in the second conjunct. However, why can’t (ii) have an AC reading? I speculate that in such
examples, the repetition of the subject encodes a focus reading (I thank James Myers for pointing
out the information structure of such examples to me). Recall that both the accent in (5.26b)
and the gapping in (5.27) express focus, and they do not have an AC reading. If EC violations
are allowed only in AC and the presence of focus blocks the AC reading, the fact that I cannot
appear in the second conjunct of (ii) is accounted for.
5.3 The EC violation in AC 137
(5.32) a. Na fen baozhi, Baoyu kan-le , erqie hai xie-le biji.

that cl newspaper, Baoyu read-prf and also write-prf notes
‘That newspaper, Baoyu read it and wrote notes.’
b. na fen [Baoyu kan-le erqie hai xie-le biji] de baozhi
that cl Baoyu read-prf and also write-prf note de newspaper
‘the newspaper that Baoyu read and also wrote notes on it.’
As in English, if there is no semantic relation between the conjuncts (i.e. the

speaker reports two unrelated eventualities), extraction from a single conjunct
is impossible in Chinese. In (5.33b), the event of reading a newspaper is not
related to the event of taking a bath, so the coordinate complex conjoined by
erqie does not tolerate the object gap in the first conjunct, which is associated
with the relativized nominal baozhi ‘newspaper.’
(5.33) a. Akiu kan-le na fen baozhi, erqie xi-le zao.

Akiu read-prf that cl newspaper and wash-prf bath
‘Akiu read that newspaper and took a bath.’
b. ∗ na fen [Akiu kan-le erqie xi-le zao] de baozhi
that cl Akiu read-prf and wash-prf bath de newspaper
Since there is no semantic connection between the conjuncts in (5.33), the

coordination is accidental. The presence of the EC effect in (5.33b) and the
absence of the effect in (5.32) result from the contrast between accidental and
natural coordination.
The fact that natural coordination tolerates EC violations whereas accidental
one does not is also observed in Quantifier Raising, a covert movement opera-
tion (May 1985). Data like the following are discussed in Ruys (1992) and de
Vos (2005: 42):
(5.34) a. A policeman serenaded every widow.
‘Some policeman serenaded all the widows.’ [narrow scope]
‘For each of the widows, there was some policeman or other who serenaded
her.’ [wide scope]
b. A policeman went to town and (he) serenaded every widow as well.
‘Some policeman went to town and also serenaded all the widows.’
[narrow scope]
∗
‘Some policeman or other went to town and for each
of the widows, there was some policeman or other who serenaded
her.’ [wide scope]
The quantifier every in (5.34a) can have either narrow scope, as in the first
reading, or wide scope, as in the second reading. In the latter case, the quantifier
presumably undergoes covert movement and lands in a position higher than a
policeman. (5.34b) is a coordinate construction, and the expression as well

indicates its accidental coordination reading. In this example, the wide scope
reading of every disappears, suggesting that covert movement from the second
conjunct is impossible. Importantly, de Vos notes that in the contiguous reading
of (5.35), a natural coordination reading, both narrow and wide scope readings
of every are available. In other words, it is possible for the assumed covert
movement to launch from the second conjunct.
(5.35) A policeman went and serenaded every widow.
‘Some policeman serenaded all the widows.’ [narrow scope]
‘For each of the widows, there was some policeman or other who
serenaded her.’ [wide scope]
One might think that in examples like (5.29) above, the EC is not violated,
but instead the apparent gaps in the conjuncts are actually null resumptive pro-
nouns or conjunct-internal operator movement. The quantifier raising contrast
in (5.34b) and (5.35), however, does not permit this kind of analysis. There is
no possibility of either resumptive pronouns or operator movement in (5.35),
but the EC still seems to be violable.3
In addition to the semantic condition of natural vs. accidental coordination,
there is indeed a formal condition on certain EC violations: wh-adverbials
cannot be extracted from single conjuncts (Culicover 1990; Postal 1998: 66):
(5.36) a. Which student did Nora go to the drugstore, come home, and talk to
for an hour?
b. ∗ How long did Nora go there, come home, and talk to that student ?
As is well known, compared to wh-arguments, wh-adverbials show special

syntactic properties, and in particular they are usually less active. For instance,
no in situ wh-adverbial can be licensed in English (C. J. Huang 1982), no wh-
adverbial can undergo long-distance fronting in Chinese (Tsai 1994: 162), and
no multiple wh-adverbial fronting occurs cross-linguistically (Rudin 1988).
Therefore, the restriction of wh-adverbials implied by the above examples is
not related to the coordinate construction per se, and thus cannot be used to
argue for the EC as a formal syntactic constraint.
In short, in both overt and covert movement, when the EC is violated,
the coordinated conjuncts are semantically related. The situation is just as
Goldsmith (1985: 134) claims: “The Coordinate Structure Constraint fails here
not because of some formal failing of English or the grammar, but, it seems,
3 See May (1985: 59), Ruys (1993), Fox (2000), Szabolcsi (2001: 619), V. Lin (2001; 2002), Potts
(2002), and Kasai (2002) for more discussion on the relationship between QR and the CSC.
because of the meaning of the construction.” Specifically, as we have shown,

it is because of the semantic relatedness of the conjuncts found in natural
coordination.
In other words, extraction out of conjuncts is blocked only in certain special
cases (involving non-AC constructions and adverbials); there is no need to posit
a constraint against such movement in general.4,5
5.4 Chapter summary

In this chapter, I discussed the semantic contrast between natural and accidental
coordination, and identified both AC and collective coordination as natural
4 As shown in Chapter 4, the CC is not violable in and-complexes. Hence examples that combine
violations of the EC with violations of the CC in and-complexes are always unacceptable, and
shed no light on the EC itself. The following data are cited from Sag (1982: 332) and Gazdar
et al. (1985: 178).
(i) a. ∗I wonder who you saw [ and [a picture of ]].
b. ∗I wonder who you saw [[a picture of ] and ].
(ii) a. ∗ The Pre-Raphaelites, we found [ and [books about ]].
b. ∗ The Pre-Raphaelites, we found [[books about ]] and ].
5 Despite admitting that the traditional EC does not work in AC, Pesetsky (1982) presents an
alternative syntactic approach to EC-like effects. A crucial assumption of Pesetsky (1982) is that
in a symmetrical coordinate complex, the coordinator assigns a single theta role to a number
of conjuncts simultaneously, whereas in an asymmetrical coordinate complex, the coordinator
assigns different theta roles to the conjuncts. Pesetsky thus claims that the formal relations
between conjuncts and coordinators are special in AC, permitting EC violations in AC to be
licensed in the syntax. Unfortunately for this analysis, however, the theta-role approach to the
relation between coordinators and conjuncts is not compatible with the generally accepted theta
theory in at least the following ways.
By definition, in AC the semantic functions of the conjuncts are different (e.g. one conjunct
could be a cause and the other a result). However, we do not see any semantic relation between
conjuncts and coordinators. Thematic relations represent abstract semantic relations between
verbs and nominals. For instance, in John laughed, John is an agent of the event denoted
by laughed, whereas in John fell, John is a theme of the event denoted by fell. By contrast,
coordinators and conjuncts do not show any such semantic relation. In AC, the defining semantic
relation is expressed between the conjuncts, not between the conjuncts and the coordinator.
Moreover, with respect to the number of theta-role receivers, a theta-role approach to the EC
does not explain why a single coordinator in symmetrical coordination may assign a theta role
to a number of conjuncts simultaneously, whereas other theta-role assignors may only assign
a theta role to a single nominal. Similarly, with respect to the number of assigned theta roles,
a theta-role approach to the EC does not explain why a single coordinator in AC can assign
multiple theta roles to a number of conjuncts, whereas other theta-role assignors can assign only
one theta role to a nominal. The theta-role explanation for the absence of EC effects in AC is
thus ad hoc at best. It simply stipulates certain special mechanisms to formulate the peculiarities
of AC, without providing a general account for related phenomena.
coordination. I then examined violation of the EC in natural coordination in

English and Chinese. Finally, I showed that Pesetsky’s (1982) syntactic analysis
of the EC has serious problems.
In Chapter 4, I investigated how the availability of external conjunct move-
ment correlates with the non-distributiveness, showing that the CCe is rela-
tivized semantically. Parallel to this, in this chapter I showed how the availabil-
ity of element movement out of conjuncts correlates with non-distributiveness
as well, showing that the EC is also relativized semantically.
If, cross-linguistically, both the CC and the EC can be violated systematically
in non-distributive coordination (i.e., natural coordination), the CSC cannot be
a general syntactic constraint on movement.
In the summary in (5.37), Yes means the presence of the CSC effect, and No
means the absence of the CSC effect.
(5.37)
In natural coordination In accidental Explanation

coordination
EC No Yes RPR (see Chapter 7)
No, if the coordinator
has intrinsic categorial Yes RPR (see Chapter 7)
features (e.g. Chinese
he/gen)
Yes, if the coordinator has no intrinsic The categorial features of the
CCe categorial features and the external external conjunct are transferred
conjunct alone satisfies the c-selection to the coordinator, which is the
imposed on the whole coordinate head of the whole complex;
complex (e.g. English and) elements may not move if their
categorial features have been
removed (see Chapter 4)
Like many other types of head
CCi Yes elements, coordinators may not
be stranded (see Chapter 4)
6 Three puzzles solved
by rejecting the CSC
6.1 Introduction
In the last two chapters, I argued that neither the CC nor the EC is a constraint
on syntactic operations. The CSC is thus not part of the computational system.
Getting rid of an ad hoc stipulation is desirable by itself, but rejecting this
particular stipulation also allows us to improve our understanding of some
hitherto puzzling facts about natural language. In this chapter, I show what we
gain if we give up the CSC. Specifically, I examine three constructions that
seem to be in conflict with certain basic syntactic laws observed elsewhere.
The first construction seems to conflict with principles of selection, theta-role
licensing, and syntax–semantics mapping, the second with principles governing
the syntactic relation between elements and their shared modifier, and the third
with principles governing the launching site of certain movement operations.
Postal (1998) ends his book by stating that the third of these constructions raises
such serious challenges for syntactic theory that they can only be avoided by
denying that the construction actually exists. However, as I show in this chapter,
if we give up the CSC, all three of the constructions can be derived in accordance
with standard principles. Thus, abandoning the CSC may actually strengthen
the explanatory power of syntactic theory.
Each of the three remaining sections in this chapter is devoted to one of these
three constructions, which I will call Split Argument Constructions, Modifier-
Sharing Constructions, and Interwoven Dependency Constructions.
6.2 Deriving Split Argument Constructions by giving up the CC

6.2.1 The Split Argument Construction (SAC)
The Split Argument Construction (SAC) refers to sentences like (6.1), in which
preverbal and post-verbal arguments (underlined) share the same thematic rela-
tion with the verb, and which have a reading expressible as in (6.2), where a
coordinate complex containing the two nominals functions as a single argument
141
142 Three puzzles solved by rejecting the CSC
and theta-role bearer for the same verb. In English, the coordinate DP complex
in the non-SAC counterpart occurs in a subject position, as in (6.2).
(6.1) a. John married Jane.

b. Italy borders France.
(6.2) a. [John and Jane] married.

b. [Italy and France] border.
The example in (6.3) is a SAC in Chinese. In the corresponding non-SACs

in (6.4), the coordinate complex occurs in a subject position in (6.4a), and in
object position in (6.4b).1
(6.3) Tudou yijing shao-le niurou.

potato already cook-prf beef
‘The potatoes have already been cooked with the beef.’
(6.4) a. [Tudou gen niurou] shao-zai yiqi le.

potato and beef cook-at together prt
‘The potatoes and the beef have been cooked together.’
b. Baoyu shao-le [tudou gen niurou].
Baoyu cook-prf potato and beef
‘Baoyu cooked the potatoes and the beef.’ (either separately or together)
In English, in addition to marry and border, other collective verbs such as

date, divorce, hug, fight, meet may also occur in SACs (Levin 1993: 36–37).
Parallel examples can also be found in German:
(6.5) a. Die Prinzessin heiratet den Grafen.

the princess marries the earl
‘The princess marries the earl.’
b. Die Prinzessin und der Graf heiraten (sich/einander).
the princess and the earl marry one.another
‘The princess and the earl marry (one another).’
1 My definition of SACs excludes constructions in which the preverbal nominal and the postverbal
nominal do not appear as conjuncts in any alternative construction. Data like the following are not
SACs, although the two DPs seem to have a similar thematic relation to the verb. Such examples
might be derived by DP raising from complex DPs. However, this derivation has nothing to do
with coordination and thus is not our concern here. I thank Jianming Lu, Yang Shen, and Bojiang
Zhang for urging me to make this clarification.
(i) a. Na kuai rou chao-le rou-si. b. Juzi yijing bo-le pi.
that chunk meat fry-prf meat-shred orange already peel-prf peel
‘That chunk of meat has been cut into ‘The orange has been peeled.’
shreds and fried.’
6.2 Split Argument Constructions and the CC 143
In the Chinese SACs such as (6.3), the preverbal and post-verbal nominals
both seem to be patient(s) of the transitive verb. The verbs in such constructions
can be either ordinary transitive verbs such as shao ‘cook’ in (6.1), chao ‘fry,’
dun ‘stew,’ or collective transitive verbs such as lianxi ‘connect,’ ban ‘mix,’ and
jiehe ‘combine.’ The SACs in (6.6), in which collective verbs occur, alternate
with the non-SACs in (6.7).
(6.6) a. Lilun yinggai lianxi shiji.

theory should connect reality
‘Theories should be linked to reality.’
b. Na ge jidan zhenghao ban mifan.
that cl egg perfectly mix rice
‘That egg can be perfectly mixed with the rice.’
(6.7) a. [Lilun gen shiji] lianxi zai yiqi le.

theory and reality connect at together prt
‘The theory has been linked to reality.’
b. Na ge jidan gen mifan ban zai yiqi le.
that cl egg and rice mix at together prt
‘That egg has been mixed with the rice.’
A constraint on SACs shared by English and Chinese is that such a construc-

tion encodes only a single eventuality. In other words, the verb or predicate
may not be distributive. This can be seen from three sources of evidence. First,
distributive adverbs such as separately and both may not occur in SACs.
∗
(6.8) a. John married Jane separately. (cf. (6.1))
∗
b. Zhe ge jidan zhenghao fenbie pan fan. (cf. (6.6))
this cl egg perfectly separately mix rice
Second, intrinsically distributive verbs or predicates may not occur in

SACs:
∗
(6.9) a. John sneezed Mary.
Intended: ‘John and Mary sneezed.’
∗
b. Jidan xi-le doufu.
egg wash-prf tofu
Intended: ‘Eggs and tofu have been washed (together).’
Third, the contrast in interpretation between (6.3) and (6.4b) shows that
SACs exclude distributive readings.
Levin (1993: 36–37) calls the alternation between SACs and non-SACs in
English the Understood Reciprocal Object Alternation. The two DPs in English
SACs must have a reciprocal relation. For instance, even if John and Mary in
(6.10a) is interpreted as a couple that together receives a single gift, the reading
of this sentence still may not be expressed by (6.10b), since the relation between
the conjuncts in (6.10a) is not reciprocal. Note that the reciprocal relation is a
specific type of non-distributive relation.
(6.10) a. Someone gave John and Mary a gift.

b. John gave Mary a gift.
SACs challenge basic principles of generative grammar governing s-selection

of collective verbs, theta-role licensing theory, and the mapping between syntax
and semantics.
First, a collective verb requires a plural nominal. Yet when such a verb is in
a SAC, each of the two arguments may be singular; thus the selection of the
verb is not satisfied in the usual way. It seems that the selection of the verb is
not satisfied by any syntactic constituent in the surface representation.
Second, the verb in a SAC seems to assign two identical theta roles to two
nominals. In no other construction is this possible. The problem is also seen
from the perspective of the DPs. In a SAC the participant or thematic role of
one nominal is the same as that of the other. However, in no other construction
may two nominals receive identical theta roles from the same verb.
Third, if every SAC alternates with a construction in which the two DPs form
a coordinate complex, why do the two DPs in the SAC itself appear so far apart?
This separation represents an apparent mismatch between the positions of the
DPs and their interpretation. If the DPs in the two constructions have different
base positions, the alternation seems to violate Baker’s (1988; 1997) Uniformity
of Theta-Assignment Hypothesis (UTAH), which states that identical thematic
relationships between items are represented by identical structural relationships
between these items in their base positions.
Fourth, unlike the arguments of other transitive verbs, the subject and the
object of a SAC may exchange their positions, without affecting the basic
meaning of the construction:
(6.11) a. Italy borders France.

b. France borders Italy.
In this section, I will show that SACs can be derived by movement of the
external conjunct out of a coordinate DP headed by a null D. Since this analysis
requires denying that the CC is a genuine constraint on syntactic operations,
giving up the CC makes it possible to get a better understanding of the syntax
of SACs.
6.2.2 The two DPs of a SAC form a coordinate complex

I claim that the two DPs of a SAC form a coordinate DP in their base-positions.
Arguments for this claim come from the four challenges listed in the previous
subsection.
First, collective verbs such as marry and fight in English, and ban ‘mix’ and
lianxi ‘connect’ in Chinese, require their selected argument to be plural. In a
SAC, if neither of the two DPs is plural, neither alone can be the real argument
of the verb. If the two singular DPs form a coordinate complex, which is plural,
the collective verb is licensed in the construction.
Second, the theta-role licensing problem mentioned above does not occur
if the two nominals of a SAC form a complex in their base positions, and the
complex has a unique theta-role relation with the verb.
Third, the deviation from the mapping between syntactic positions and
semantic interpretations does not occur either, if the two nominals of a SAC
form a coordinate complex in their base positions. Then the two DPs in the
alternating constructions, such as (6.1) vs. (6.2), have the same base positions.
Fourth, the alternation in (6.11) is also found in coordinate constructions. If
the two nominals of a SAC form a coordinate complex in their base positions,
either one may be merged with the coordinator first, just like conjuncts.2
Note that like ordinary preverbal arguments, the preverbal DP of a SAC is
backgrounded, and like an ordinary postverbal argument, the postverbal DP of
a SAC is foregrounded, if the construction is uttered without any contrastive
stress. This difference in information structure is not relevant to the unification
of the theta role of the two nominals.
6.2.3 Deriving SACs by conjunct raising

A. DP Spec raising out of a complex DP
I have claimed that the two DPs of a SAC form a coordinate complex. Specifi-
cally, in a SAC, the first nominal (DP1 in (6.12)) is base-generated as the Spec
of the complex DP (DP3 in (6.12)), and the second nominal (DP2 in (6.12)) is
2 One might wonder whether comitative constructions like (ib) are derived from coordinate con-
structions like the SAC in (ic), since they pose the same four challenges. See Kayne (1994)
and Zhang (2007a) for arguments that English comitative constructions are indeed derived from
coordinate-like constructions.
(i) a. John and Jane met (each other) in the street.
b. John met with Jane in the street. (comitative construction)
c. John met Jane in the street. (SAC)
base-generated as the complement of the complex DP. The head of the complex
DP is a null D (D3 in (6.12)).
(6.12) DP3
DP1 D′3
D3 DP2
0
I propose that SACs are derived by the raising of the first conjunct of the
base-generated coordinate nominal, as shown in (6.12). The mechanism of this
movement is independently found in possessor-raising (see Szabolcsi 1983;
1994; Landau 1999, among others): in both cases, a DP Spec is raised out of a
complex DP.
Recall that I argued in Chapter 4 that in and-complexes, external conjuncts
may not move (the CCe effect) because their categorial features have been
transferred to the coordinator and. Recall also that external conjuncts may
move, however, if the head of a coordinate complex is not realized by and, but
by a coordinator that has categorial features, such as the conjunctions he and
gen in Chinese. Now in (6.12), the head of the complex DP3 is a null D, which
already has categorical features and so does not take any from DP1 . Therefore,
as in he/gen constructions, the movement of DP1 is computationally licensed
here.
In Chapter 4 (Section 4.4.3) I also argued that single conjuncts may move
only in non-distributive coordination. Since the verbs or predicates of SACs may
not be distributive, the proposed conjunct raising in (6.12) is also compatible
with this semantic condition on conjunct movement.
The raised DP1 in SACs parallels nominals appearing with “floating quan-
tifiers” like all, discussed by Sportiche (1988) and others. Like the position
of all, the surface position of the complex DP, which contains DP2 and the
launching site of DP1 , demarcates an argument position.
Before I move to address the syntactic positions of the two DPs of a SAC
in the next two subsections, I need to clarify that topicalization constructions
such as (6.13a) and ba constructions such as (6.13b) are semantically similar to
SACs. Although their existence does not challenge the analysis proposed here,
I put such constructions aside since it is controversial whether the preverbal
patient DP in them is derived by movement or not. If the DP is base-generated
at the surface position, it might simply be co-indexed with a null conjunct that
is conjoined with the postverbal DP.
(6.13) a. Tudou, ni keyi dun niurou.

potato you can stew beef
‘As for potatoes, you can stew them with beef.’
b. Baoyu ba tudou shao-le niurou.
Baoyu ba potato cook-prf beef
‘Baoyu cooked the potatoes and the beef together.’
B. The syntactic position of the preverbal DP of a SAC

The preverbal DPs of SACs exhibit the syntactic properties of nominals that
have undergone A or A-bar movement, supporting the conjunct raising hypoth-
esis illustrated in (6.12).
In English, it is not controversial that the preverbal DP of a SAC is in the
subject position, which is derived by A-movement. Like an ordinary subject,
the DP triggers agreement with the auxiliary, as seen in (6.14a). Moreover, like
an ordinary subject, the DP may precede a raising verb such as seem, as seen
in (6.14b). Accordingly, the proposed movement in (6.12) is A-movement in
English.
(6.14) a. John {has/∗ have} married Jane.

b. John seems to have married Jane.
Of course, the first DP (or its null version, i.e., a null operator) of an English
SAC may also undergo further syntactic operations, deriving various related
constructions:
(6.15) a. It was John who married Jane.
b. The man that married Jane was John.
I now discuss the syntactic positions of the preverbal nominals in Chinese

SACs. Consider the non-argument position first. If the subject position of a
SAC is filled by an agent, and no ba occurs, the preverbal patient DP, which is
associated with the postverbal DP, is not in an argument position. The preverbal
patient DP may be in a focus position, preceded by the focus marker lian ‘even,’
as shown by the underlined part in (6.16). This post-lian position of the focused
DP is derived by A-bar movement (Shyu 1995).
(6.16) Baoyu lian zuihou yi bao tudou dou dun-le niurou.

Baoyu even last one bag potato also stew-prf beef
‘Baoyu stewed even the last bag of potatoes with beef.’
Moreover, SACs may also have a relative clause version. If relativization

is derived by the A-bar movement of the modified element or a null operator
co-indexed with the modified element, as generally assumed, such SACs are
derived by the movement of single conjuncts.
(6.17) a. Baoyu xiang-qi-le na bao [RC dun-le niurou] de tudou.

Baoyu recall-inch-prf that bag stew-prf beef de potato
‘Baoyu recalled the bag of potatoes that has been stewed with beef.’
b. Baoyu xiang-qi-le na bao [RC dun-le tudou] de niurou.
Baoyu recall-inch-prf that bag stew-prf potato de beef
‘Baoyu recalled the bag of beef that has been stewed with potatoes.’
Like ordinary A-bar movement, the proposed A-bar movement of the external
conjunct in a SAC is not clause-bound, but is subject to island constraints. In
(6.18a), naxie niurou ‘that beef’ moves from the embedded clause to the matrix
clause, and in (6.18b), naxie niurou may not move out of a complex nominal.
(6.18) a. Lian naxie niurou, wo zhidao [Baoyu dou yijing dun-le tudou].
even those beef I know Baoyu also already stew-prf potato
‘Even that beef, I know Baoyu has stewed potatoes with.’
b. ∗ Lian naxie niurou, wo tingshuo [Baoyu dou yijing
even those beef I hear Baoyu also already
dun-le tudou] de xiaoxi.
stew-prf potato de news
We now examine the preverbal DPs of SACs that have neither an

agent nor ba, as in (6.3). If there is only one preverbal nominal in a
SAC in Chinese, the nominal, which is a patient, is in a subject posi-
tion. In Chinese, patient nominals of transitive verbs may occur in a
preverbal subject position (Cheng 1988; den Dikken and Sybesma 1998;
among others). Such constructions in Chinese can be either property-
denoting or episodic-eventuality-denoting, as shown in (6.19a) and (6.19b),
respectively.
(6.19) a. Tudou keyi zhu.

potato can boil
‘Potatoes can be boiled.’
b. Tudou yijing shao-hao-le.
potato already cook-ready-prf
‘The potatoes have been cooked already.’
Since the preverbal subject position in Chinese is an available position for

patients, it is plausible to assume that the preverbal DP of a Chinese SAC
is in the subject position. Here are three further pieces of evidence for this
assumption.
First, like an ordinary subject, the preverbal DP in such a construction pre-

cedes a modal, and a VP-adverbial. Both types of elements mark the left edge
of a predicate.
(6.20) a. Tudou keyi shao niurou.
potato may cook beef
‘One may cook potatoes with beef.’
b. Tudou zaojiu shao-le niurou.
potato early cook-prf beef
‘The potatoes were cooked with the beef a long time ago.’
Second, like an ordinary preverbal subject, the preverbal DP of a SAC in

Chinese may not be non-specific. The acceptability contrast in the SAC in
(6.21b) is parallel to that in (6.21a).
(6.21) a. {Na/∗ yi} bao tudou yijing huai le ma?
that/one bag potato already bad prt q
‘Has that bag of potatoes gone bad already?’
b. {Na/∗ yi} bao tudou yijing dun-le niurou le ma?
that/one bag potato already stew-prf beef prt q
‘Has that bag of potatoes been stewed with beef already?’
Third, like an ordinary subject with a quantity reading (Y. A. Li 1998), the
preverbal DP of a SAC may also have a quantity reading:
(6.22) Yi gongjin tudou shao-le liang gongjin niurou.
one kg potato cook-prf two kg beef
‘One kg of potatoes has been used to cook with two kg of beef.’
The landing site of the moved conjunct of such a SAC is not a topic position.
A topic may precede an overt or implicit agent of a transitive verb, and thus it
should license an agent-oriented adverb. Data like (6.23) tell us that no agent-
oriented adverbial is allowed in a SAC, although other types of adverbials are
allowed (e.g. yijing ‘already’ in (6.21b)). If there were an agent pro subject in
the construction, an agent-oriented adverb should be licensed.
(6.23) a. ∗ Tudou yukuai de shao-le niurou.
potato happy de cook-prf beef
b. ∗ Tudou guyi shao-le niurou.
potato deliberately cook-prf beef
The exclusion of agent-oriented adverbs also indicates that the preverbal

DP of a SAC is not in the position for an instrument adverbial, since such an
adverbial may occur with an agent-oriented adverb. In (6.24), yong guantou
‘with pan’ is an instrument and yukai ‘happily’ is an agent-oriented adverbial.
(6.24) Baoyu yong guantou yukai de zhu-le yi guan tang.

Baoyu with pan happy de cook-prf one can soup
‘Baoyu cooked a can of soup with a pan happily.’
Of course, like an ordinary subject, the preverbal DP of a SAC may undergo

further A-bar movement, to a topic position:
(6.25) Naxie tudoui , wo tingshuo i yijing dun-le niurou.

those potato I hear already stew-prf beef
‘Those potatoes, I heard that they have already been stewed with the beef.’
All of these indicate that the preverbal DP of a SAC is in a position that is

derived by A- or A-bar movement.
C. The syntactic position of the postverbal DP of a SAC

The unraised conjunct of the coordinate DP complex of a SAC remains inside
the complex, in a base position decided by the thematic relation between the
complex DP and the selecting verb.
Let us consider the thematic relation between the two DPs and the verb of
a SAC in Chinese. It is clear that both DPs have a patient reading. If the two
DPs form a complex DP, a plausible hypothesis is that the complex DP is base-
generated in the direct object position of the verb, i.e., inside VP (in Chinese at
least, postverbal patients must be in direct object position). When the external
conjunct is raised, the residue of the complex DP is phonologically realized by
the internal conjunct only, and this conjunct thus surfaces in the position of a
direct object.
One may wonder if the verb in a SAC may be passive (as suggested to me
by Yang Gu), but the object position of the postverbal DP rules this out. If
the verb were passive, the Case of the postverbal DP of a SAC would not be
licensed.
Now let us consider the thematic relation between the DPs and the verb of
a SAC in English. The default hypothesis is that the verb in a SAC and the
verb in the corresponding non-SAC are the same, and thus they have the same
argument structure. For instance, the verb met in (6.26a) and the verb met in
(6.26b) are the same.
(6.26) a. John met Mary.

b. John and Mary met each other.
(6.27) a. France borders Italy.

b. France and Italy border each other.
Levin (1993: 37) states that “[T]he action described by the verb in the
intransitive [i.e. non-SAC – NZ] variant of this alternation can be roughly
paraphrased by the transitive verb when it takes the reciprocal each other as
object.” If the reciprocal each other is in direct object position (also see Heim
et al. 1991), the base position of the coordinate complex John and Mary in
(6.26b) and France and Italy in (6.27b) should be that of external argument of
the verb, i.e., Spec of V.
One piece of evidence for the external argument position of the complex
nominal in examples like (6.26b) and (6.27b) is that the verb’s single argument
may not occur postverbally, unlike that of an unaccusative verb.
(6.28) a. There came the sound of gunfire from the street below.
b. From the street below came the sound of gunfire.
(6.29) a. ∗ There met John and Mary in the street.

b. ∗ In the street met John and Mary.
Another observation supporting the external argument position of the

complex nominal in examples like (6.26b) and (6.27b) is that the argu-
ment may not be passivized, unlike the internal arguments of transitive
verbs.
∗
(6.30) John and Mary were met in the street.
Intended: “John and Mary met each other in the street.”
The verbs in English SACs correspond to reciprocal verbs in other lan-

guages. Siloni (2008) reports a detailed study of such verbs and concludes that
“cross-linguistic evidence shows that the subject of reciprocals is an external
argument.”
If the complex nominals in examples like (6.26b) and (6.27b) are base-
generated in SpecVP, they undergo A-movement and land at SpecTP. In the
SACs in (6.26a) and (6.27a), however, it is the external conjunct of the coordi-
nate DP, rather than the whole coordinate DP, that is raised to SpecTP (see the
last subsection). Then the residue of the complex DP, which is phonologically
realized by the internal conjunct only, remains in Spec of V. Assuming the verb
is always moved out of its base position (see Johnson 1991; Koizumi 1995;
Runner 1995; Bowers 2002; and Baltin 2002; 2006 for various implementation
of this assumption), we capture the surface order in which the verb precedes
the element in Spec V. Moreover, since only the external conjunct is raised to
SpecTP, it alone bears nominative Case. Then the Case of the internal conjunct
may no longer be nominative, as seen in (6.31). Instead, it may check its Case
with V, as the pronoun other does in non-SAC constructions like (6.26b) and
(6.27b).
(6.31) Jane married {∗ he/him}.
This Case checking between V and the post-verbal DP in a SAC may also
explain why SACs do not permit reciprocals.
(6.32) a. ∗ John met Mary each other. (cf. (6.26b))

b. ∗ France borders Italy each other. (cf. (6.27b))
In (6.32a), both the pronoun other and the postverbal DP Mary check their
Case features with the single V, which is impossible in English. A similar
problem occurs in (6.32b).
Another explanation for the unacceptability of (6.32) comes from the anal-
ysis of each other in Heim et al. (1991). They claim that each in each other
undergoes an LF movement, adjoining to a group-denoting antecedent. Thus
in (6.26b), for instance, each adjoins to the group-denoting element John and
Mary. We can now see that in the SAC in (6.26a), after the overt raising of
the external conjunct, there is no longer a group-denoting element for each to
adjoin to. If LF movement occurs after the overt syntactic operations, it is too
late for each to find any group-denoting element. Consequently, representations
such as (6.32), which contain unlicensed each, are rejected.
This latter account for the absence of reciprocals in SACs may also explain
the absence of reciprocals in comitative constructions in English, a fact noted
by Baker (1992: 46):
(6.33) a. John and Mary kissed each other.

b. ∗ John kissed each other with Mary.
In Chinese, the reciprocal adverb huxiang ‘mutual’ is allowed in comitative

coordination if the two conjuncts are not separated, as in (6.34a), but not if the
first conjunct moves away, as in (6.34b) (see Chapter 4). In neither the English
comitative construction in (6.33b) nor the Chinese comitative construction in
(6.34b) is there any group-denoting nominal for the reciprocal to adjoin to at
LF, so both are rejected.
(6.34) a. Baoyu gen Daiyu yijing huxiang wo-shou-le.

Baoyu and Daiyu already mutual grasp-hand-prf
b. Baoyu yijing gen Daiyu (∗ huxiang) wo-shou-le.
Baoyu already and Daiyu mutual grasp-hand-prf
Both: ‘Baoyu and Daiyu have shaken their hands.’
I conclude that the preverbal DP of a SAC may occur wherever a language

permits an argument to move to. The postverbal DP of a SAC occurs in the
base position of the coordinate complex, where its thematic role is assigned by
the selecting verb. One consequence is that internal arguments split in SACs in
Chinese, whereas external arguments split in SACs in English.
6.2.4 Section summary

We have proposed that SACs are derived by the movement of the external
conjunct from a coordinate DP, a proposal that depends on rejecting the CCe.
If SACs can be derived syntactically in this way, we solve a variety of puzzles
posed by these constructions.
One detail remains neglected in this analysis, namely why, if the derivation
of SACs involves a null D in the head position of a coordinate complex, the
coordinator must appear overtly in certain other constructions, in both English
and Chinese.
In our analysis of SACs in English, the null D heading a coordinate complex
licenses single conjunct raising if the conjuncts have a reciprocal relation. But
why does the raising of the whole coordinate complex in the corresponding
non-SAC require the overt occurrence of the coordinator?
(6.35) John ∗ (and) Jane married.
Descriptively speaking, null coordinators in ordinary coordinate construc-

tions (i.e. not SACs) are marked in English (Payne 1985: 25), especially for
coordinate nominals, though as Kayne (1994: 144 fn.5) notes, null coordinators
are possible in coordinate verbal complexes, as in (6.36):
(6.36) They went, one to Paris, the other to London.
In Chinese, the raising of the external conjuncts is possible even when the
coordinator is overt (see Chapter 4), but nothing in the analysis requires it to
be overt. Nevertheless, the facts show that the raising of a single conjunct from
preverbal position (the cases discussed in Chapter 4) does indeed require the
overt appearance of the coordinator:
(6.37) a. Baoyu (gen) Daiyu he-mai-le yi zuo fangzi.

Baoyu and Daiyu co-buy-prf one cl house
‘Baoyu and Daiyu co-bought a house.’
b. Baoyu yijing ∗ (gen) Daiyu he-mai-le yi zuo fangzi.
Baoyu already and Daiyu co-buy-prf one cl house
‘Baoyu and Daiyu already co-bought a house.’
Despite this, the English constraint shown in (6.35) is not obeyed in Chinese:
(6.38) Baoyu (gen) Daiyu xiang-ai-le.
Baoyu and Daiyu mutual-love-prf
‘Baoyu and Daiyu fell in love with each other.’
In this book I focus on the generalization that conjuncts may move if the
coordinators have categorial features, regardless of whether the coordinators
are overt or covert. Null coordinators with nominal categorial features may
occur not only in SACs, but also in the construction to be discussed in next
section. It is obvious that not all overt coordinators have a null counterpart, and
not all null coordinators have an overt counterpart. At this point the principles
accounting for the overtness of coordinators in English and Chinese remain a
puzzle to me, and I leave this issue for future research.
6.3 Deriving Modifier-Sharing Constructions by giving up the CC

6.3.1 The Modifier-Sharing Construction
In this section, we examine constructions such as (6.39), where a relative clause
(RC) seems to be shared by two modified elements (Ross and Perlmutter 1970).
(6.39) a. Mary met a mani and John met a womanj [whoi&j knew each otheri&j well].
b. A mani came in and a womanj left [whoi&j were quite similar].
c. The house has a roomi and the shop has a cellarj [whichi&j are joined
by a small underground passageway].
In (6.39a), for example, the RC, as a modifier, seems to be shared by two
modified elements, antecedents, or Heads, namely a man and a woman.3 The
two DPs are distributed in the two matrix clauses respectively. Each of the
two nominals is singular, but the reciprocal each other, which requires a plural
binder, seems to be licensed by the combination of the two nominals (via the
relative pronoun who). I call the constructions Modifier-Sharing Constructions
(MSC, henceforth).
MSCs have been claimed to pose a serious problem for syntactic theory
(Alexiadou et al. 2000: 13). The RCs in MSCs are not adjacent to their
antecedents, but this separation of the modifier and its modified elements can-
not be generated either by extraposition or by deletion of the RC. This is shown
by the occurrence of relational expressions such as similar, each other, join,
which indicate that the RCs must be semantically related to the multiple Heads
3 I use “Head” to refer to the modified nominal of a RC, such as the man in the man who came.
Such a nominal is also called antecedent or “Head nominal.” I reserve the term “head” for the
projecting element of a syntactic constituent.
6.3 Modifier-Sharing Constructions and the CC 155
simultaneously. Syntactically, the RCs must be licensed by a plural nominal,

rather than either of the two singular Heads alone. For instance, (6.39a) cannot
be derived from either (6.40a) or (6.40b). The extraposition in (6.40a) is not
possible because the RC in the gap position _k is not well-formed, given that
each other is not licensed by any c-commanding plural element in the local
domain. The deletion in (6.40b) is not possible for a similar reason.
(6.40) a. ∗ Mary met a mani k and John met a womanj [who knew each other well]k .
b. ∗ Mary met a mani [who knew each other well] and John met a womanj
[who knew each other well].
As claimed by Gazdar (1981: 179), the RCs in MSCs “must be generated in
situ.” If so, the base position of the RCs and at least one of their antecedents
are not adjacent, unlike regular RC constructions. This raises the question of
whether MSCs are actually RC constructions at all. If they are not, they might
be derived in some other way, such as the way in which a depictive construction
is derived. In (6.41), the depictive young modifies the subject he, but the two
elements are not adjacent.
(6.41) He died young.
In short, we are confronted with an unclear syntactic relation between two
separate DPs and their shared modifier.
In this section, I propose that MSCs are derived from ordinary RC con-
structions. What is special about the constructions is that the two antecedents
of an MSC, as conjuncts of a coordinate complex, have undergone sideward
movement from the complex. My analysis will show how giving up the CC,
and accepting the operation of sideward movement, make it possible to under-
stand this special RC dependency construction, while avoiding problems with
an alternative approach.
The structure of this section is as follows. In Section 6.3.2, I argue that the
two apparent antecedents of an MSC originally form a coordinate complex,
the real antecedent of the RC. In Section 6.3.3, I propose that the observed
splitting effects of MSCs are syntactically derived by sideward movement of
the conjuncts out of this coordinate complex. In Section 6.3.4, I point out
certain disadvantages with an alternative analysis. Finally, Section 6.3.5 is a
brief summary.
6.3.2 MSCs have coordinate antecedents

I claim that the derivation of an MSC contains a step where the two apparent
antecedents are two conjuncts of a coordinate complex, and it is this complex
that is the real and unique antecedent of the relative pronoun.
In data like (6.42), RCs take a coordinate Head (Vergnaud 1974; Jackendoff
1977: 190; Link 1984). Such data are different from MSC data in that the
antecedents are not “split” into two matrix clauses.
(6.42) a. a mani and a womanj [whoi&j knew each otheri&j well]

b. a roomi and a cellarj [whichi&j are joined by a small underground]
passageway]
c. the boyi and the girlj [whoi&j were in love]
d. the cari and the truckj Opi&j that collided
Moltmann (1992a) calls both the RC in data such as (6.42) and the RC in
MSCs multiply-Headed RCs. We call the construction where a RC takes a
coordinate Head, such as those in (6.42), a Coordinate-Head RC Construction
(CHC).
I claim that the structure of (6.42a), for instance, is as follows:4
(6.43) DP
DPm [RC whom knew each other well]
DPi D′m
a man D
m DPj
and a woman
In this structure, DPm is a coordinate antecedent of the RC. I claim that

the antecedent of the RC of an MSC is also a coordinate DP. There are two
major differences between MSCs and CHCs. First, Dm in MSCs is realized by
a null conjunction, whereas it is realized by and in CHCs. Second, the two DP
conjuncts are separately re-selected by two verbs in MSCs, but not in CHCs. I
will present a full-fledged syntactic derivation of MSCs in Section 6.3.3. In the
present section, showing the syntactic similarities between CHCs and MSCs, I
present three arguments (A, B, and C) for the claim that in both constructions
the RC takes a single coordinate DP complex as its Head, and thus the RC is
not syntactically multiply headed.
4 In CHCs, the status of the RC as RC is not important. Other types of phrases can also be merged
with a coordinate DP. In (i), the phrase to the right of the coordinate DP is a PP modifier; and in
(ii), the phrase to the right of the coordinate DP is a complement (Jackendoff 1977: 191–192).
The frame of our structure in (6.43) is general enough to cover all of these types of examples.
(i) the boy and the girl [with a common background]
(ii) three members and two vice-chairmen [of interlocking committees]
A. Argument A: restrictive modification constructions

A well-established constraint on restrictive RCs in English is that they do not
take bare proper names (i.e. without an article) or pronouns as antecedents
(see Jackendoff 1977: 171), as shown in (6.44a) through (6.44d). The example
in (6.44e) further shows that a pronoun cannot be included in the Head of
the RC.
∗
(6.44) a. Pat {who/that} I like is a genius. (Sag 1997 (91))
b. The Pat {who/that} I like is a genius.
∗
c. He {who/that} I know is a genius.
∗
d. Mary met [him]i [whoi knew her well].
∗
e. I saw [him and the man]i whoi were late yesterday.
This constraint is not seen in non-RC modifiers such as depictives. In (6.45),

the depictive young may modify the pronoun he or the bare proper name Pat.
(6.45) {The man/He/Pat} died young.
Both CHCs and MSCs are restrictive modification constructions. In English,

this means that neither a CHC nor an MSC may allow proper name or pronoun
antecedents. Wilder (1999: 2.1) notes that neither of the apparent antecedents
of MSCs can be a pronoun, as shown in (6.46a). It is obvious that this is also
true of CHCs, as seen in (6.46b).
(6.46) a. ∗ Mary met [him]i and John met [a woman]j [whoi&j knew each other well].
b. ∗ [he]i and [a woman]j [whoi&j knew each other well]
Parallel to this pronoun constraint, the antecedents of both CHCs and MSCs
obey a proper name constraint:
(6.46) a. ∗ Mary met Billi and John met [a woman]j [whoi&j knew each other well].
b. ∗ [Bill]i and [a woman]j [whoi&j knew each other well].
The impossibility for pronouns and proper names to be antecedents of CHCs

and MSCs indicates that the RCs in both constructions are restrictive. If the two
types of constructions are derived in a similar way, the sharing of this property
is explained.
B. Argument B: D-features of the modified nominals

Another property shared by CHCs and MSCs is that the determiners of the
Head nominals must be identical. Vergnaud (1974) and Link (1984) note that
the determiners of the Head nominals must be identical in CHCs, and Moltmann
(1992a: 191, 195) further notes that this property is shared by MSCs:
∗
(6.48) a. a man and the woman who met last year (CHC)
∗
b. the father of John and a woman who know each other well (CHC)
∗
c. A man entered and the woman left who met last year. (MSC)
∗
d. John saw the man and Mary saw a woman who met last year. (MSC)
In (6.48a), a man and the woman is the Head of the RC who met last year.
Since the determiner a and the determiner the are not identical, the CHC is
not acceptable. Parallel to this, in (6.48c), the two antecedents are a man and
the woman, and again the difference in determiners causes the MSC to be
unacceptable. The other examples in (6.48) show the same pattern.
In fact, more precisely, the two modified nominals should share the same
definiteness or specificity, rather than the same form. Neal Whitman (p.c.)
points out the acceptability of “your piano student and my guitar student turned
out to know each other.” The two D-elements your and my are both definite.
Importantly, as pointed out by Moltmann, this restriction on determiners
does not hold for simple conjoined nominals, as seen in the following examples
(6.49).
(6.49) a. John met a man and the woman he saw yesterday.

b. John met that man and a woman.
In (6.49a), the coordinate object is composed of two conjuncts: a man, and

the woman he saw yesterday. The former is indefinite and the latter is definite.
The two conjuncts in (6.49b) are also different in their determiners, but like
(6.49a), the sentence is acceptable.
This restriction on the determiners of the Head nominals of both CHCs and
MSCs can be accounted for by our hypothesis that, as in CHCs, the actual
Heads of the RCs in MSCs are coordinate complexes. Semantically, RCs are
predicates of their Head nominals, and each predicate has only one subject.
This is also true of the Head nominal of the RC in a CHC or MSC; it must be a
unique nominal, and this unique nominal cannot have two different specificity
or quantificational features. In our analysis, the unique nominal is a coordinate
DP complex. Thus, the RCs of CHCs and MSCs do not have multiple Heads. In
an MSC, the conjuncts of this single complex Head later each undergo certain
syntactic operations, splitting the conjuncts into two clauses (see Section 6.3.3).
The fact that the determiners of CHCs and MSCs obey the same constraint
indicates that the two constructions may undergo certain identical steps in their
derivations. In one of these steps, I argue, both constructions have a single
coordinate complex as the antecedent of the relative pronoun, as shown in
(6.43).
Moltmann (1992a: 191) denies a semantic approach to the condition restrict-

ing the determiners of the Head nominals. She argues that (6.50a) could be
perfectly interpretable and (almost) equivalent to (6.50b), and similarly for
(6.51a) and (6.51b):
(6.50) a. ∗ a man and the woman who met last year

b. a mani and the woman that hei met last year
(6.51) a. ∗ A man entered and the woman left who met last year.
b. A mani entered and the woman left who met himi last year
Indeed, D-feature unification is a formal, not semantic, constraint. A crucial

difference between the a- and b-sentences above is that the RC in the latter is
the predicate of the simplex nominal, the woman. If the RC requires its Head,
the woman, to have a consistent set of specificity features, so should all other
RCs. Accordingly, the a-sentences are unacceptable because a man and the
woman do not provide a consistent set of specificity features for the RC, which
takes the combination of the two nominals as its Head (its semantic subject). If
the Head of an RC does not exhibit consistent specificity features, there cannot
be any predication relation between the Head and the RC.
The D-feature unification of the antecedents of CHCs and MSCs supports
our claim that the so-called multiply-headed RCs are actually RCs that take a
single-coordinate complex Head.
C. Argument C: licensing reciprocals

One more property shared by CHCs and MSCs is that they both license recip-
rocals, which need a plural antecedent (see Section 6.2.3C). In the CHC in
(6.52a), the reciprocal each other is licensed by the coordinate complex a man
and a woman, via the relative pronoun who. In the MSC in (6.52b), there is no
plural nominal to serve as the antecedent of the reciprocal, or no plural nominal
to ensure that the relative pronoun is plural so that it can license the reciprocal
locally, but the sentence is acceptable. If the derivation of (6.52b) involves a
representation similar to (6.52a), in which the two singular nominals, a man
and a woman form a coordinate complex, the licensing of the reciprocal is
accounted for.
(6.52) a. a man and a woman [who knew each other well] (= (6.42a))
b. Mary met a man and John met a woman [who knew
each other well]. (= (6.39a))
I thus claim that the antecedent of the relative pronoun in an MSC is a coor-
dinate complex, and the head of this complex is realized by a null conjunction
(the element e in (6.53b)). Thus regardless of how the relative clause is derived,
at a certain point of the derivation, we are forced to conclude that the structure
in (6.53b) is integrated into the derivation of (6.53a):
(6.53) a. Mary met a mani and John met a womanj [whoi&j knew each otheri&j well]
b. DPm
DPi D′m
a man Dm DPj
e a woman
6.3.3 Deriving MSCs by sideward conjunct raising

I have argued for the existence of a coordinate DP in MSCs (DPm in (6.53b)). I
now propose that MSCs are derived by the sideward movement of the conjuncts
of this DP.
A. Sideward movement
Before I spell out my proposal for the derivation of MSCs, I will briefly
introduce sideward movement. Generally speaking, the movement of α lands
in the same “tree,” that is, in a position that c-commands the launching site. The
movement from one tree into another tree is sideward movement, as illustrated
below.
(6.54) Step 1: α moves from Tree 1 to Tree 2
α
<α>
Tree 2 Tree 1
Step 2: Tree 1 is integrated into Tree 2 by a kind of remnant merger5

Since the landing site of sideward movement is in a different tree from
that of the launching site, it neither c-commands nor is c-commanded by the
5 Remnant merger here is parallel to remnant movement (den Besten and Webelhuth 1990; Müller
1998), in which part of a phrase is extracted, and then the rest of the phrase undergoes further
syntactic operation (Move or Merge).
launching site. Moreover, since the element that undergoes sideward movement
may be reselected by a verb in the new working site, it may land in a theta
position.
Sideward movement is discussed in Bobaljik and Brown (1997), Nunes
(2001), Hornstein (2001), and Nunes and Uriagereka (2000). The existence
of sideward movement is expected, if Remerge (Move), like Merge, simply
sets up new syntactic relations. As pointed out by Hornstein and Nunes (2002:
27), sideward movement does not add any new operation or condition to our
current computational system. Instead, it merely removes the stipulation that
movement must target the syntactic object that contains the trace. Hornstein
and Nunes (2002) claim that such a stipulation is a residue of D-structure, a
notion that is not compatible with the Minimalist Program. If no D-structure is
assumed and the computational system relies on generalized transformations to
build phrasal objects, the landing site of movement may lie beyond the domain
that contains the trace. “In other words, in a system that may operate with more
than one single-rooted syntactic object at once, as in Chomsky 1995, only brute
force would force movement to always target the same tree” (Hornstein and
Nunes 2002: 28). Sideward movement “is the result of taking the primitive
operations of the system seriously and exploring their full potential” (Boeckx
2007: 896).
Sideward movement has been argued to account for a number of unrelated
phenomena, such as adjunct control (Hornstein 2001), issues pertaining to
extraction domains (Nunes and Uriagereka 2000), PRO-gate effects (Hornstein
and Kiguchi 2003), donkey anaphora (Boeckx 2003), antecedent-resumptive
relations (Kayne 2002), binominal each (Boeckx and Hornstein 2005), and
(pseudo)gapping (Agbayani and Zoerner 2004).
B. Deriving MSCs by the evacuation of coordinate DP complexes

Rejecting the CC and assuming that conjuncts may move, I propose the fol-
lowing derivations for MSCs: the two antecedents of an MSC are originally
two conjuncts of a coordinate nominal, then each undergoes sideward move-
ment, landing in a new working site and being selected by a verb. After that,
a coordinate clausal complex is also constructed. In the old working site, a
complex nominal is constructed, in which the relative clause takes the remnant
coordinate nominal as its antecedent. Finally, the complex nominal adjoins to
the coordinate clausal complex. The whole process includes four major steps.
I illustrate the derivation of (6.55) as follows.
(6.55) Mary met a mani and John met a womanj [whoi&j knew each otheri&j well]
Step 1 : a man and a woman form a coordinate nominal, with a null D as its
head:
(6.56) DPm
DPi D′m
a man D DPj
m
e a woman
Step 2 : the DP a man undergoes sideward movement and is selected by

met. After certain additional syntactic operations, the TP Mary met a man is
constructed. Similarly, the DP a woman undergoes sideward movement and is
selected by met. After certain further additional syntactic operations, the TP
John met a woman is constructed. The two TPs and the conjunction and form a
coordinate clausal complex TPk (recall that the coordinator and itself does not
have any categorial features, and it obtains categorial features from conjuncts,
which is why it is under T in (6.57)).
(6.57) TPk
TP T k′ DPm
Mary met Tk TP DPi D′m

[a man]i
and John met t Dm DPj
i
[a woman]j
e tj
Step 3 : In the old working site, a complex nominal DPn is constructed, which
contains the RC and its antecedent, the remnant DPm :
(6.58) DPn
DPm [whom knew each other well]
DPi D′m
ta man Dm DPj
e ta woman
Step 4 : DPn , which was constructed in Step

3 , adjoins to TPk , which was
constructed in Step 2:
(6.59) TPk
final merger
TPk DPn
TP Tk′ DPm [whom knew each otherm well]
Mary met Tk TP DPi D′m

[a man]i
and John met ta man D DPj
m
[a woman] j
e ta woman
This proposed derivation of the MSC in (6.55) has the following character-
istics:
1. DPi and DPj form a coordinate complex headed by a null D in their
base-positions;
2. Both DPi and DPj are conjuncts and they move out of the coordinate
complex DPm ;
3. DPi and DPj are remerged with the verbs via separate sideward move-
ments;
4. DPn , which hosts the RC and the remnant of the DPm after the conjunct
movement, adjoins to TPk , as the last step of the derivation.
Consider 1. It is important that the Head of the coordinate nominal is null,
rather than an overt conjunction such as and. If it were not null, unacceptable
forms like the following would be derived:
∗
(6.60) Mary met a man and John met a woman and who knew each other well.
This null D is different from the conjunction and in two respects. First, it
has its own categorial features, and thus it does not get categorial features from
any conjunct. Following the arguments in Chapter 4, then, the conjuncts may
move. Second, as a null element, the null D does not need any phonological
host, and therefore the conjuncts linked by it may move. In other words, given
our analysis of the CC, we predict that both the CCe and the CCi may be
violated by the coordinate complex DPm .
Now we turn to 2. In this proposed derivation of (6.55), DPi and DPj are
conjuncts and they move. Recall that in Chapter 4 and 6.2 of this chapter,
I asserted that single conjuncts may move if the conjuncts are semantically
related to each other. Since in the derivation of MSCs both conjuncts move, the
semantic condition on single conjunct movement does not apply here.
Characteristic 3 can be described as follows. DPi and DPj are remerged with
the two transitive verbs, respectively. One might wonder whether the proposed
derivation violates the Complex NP Constraint (Ross 1967), since the two
conjuncts are extracted from a complex nominal. However, sideward movement
permits extraction from islands if the islands are not yet integrated into the
core structure (see Nunes and Uriagereka 2000; Hornstein 2001; Nunes 2004;
Taylor 2006). In my derivation, sideward movement occurs in Step 2 , and
the complex nominal is constructed in Step 3 . Since the sideward movement
occurs before any complex nominal exists, it does not violate the Complex NP
Constraint. Moreover, in a MSC it is the Head of the complex nominal, rather
than any element within the modifier or complement of a complex nominal,
that is moved. This kind of movement seems not to be subject to the Complex
NP Constraint.
Similarly, when the sideward movement occurs in Step 2 , DPn , which
becomes an adjunct in Step 4 , does not yet exist, so the sideward movement
does not violate any adjunct island constraint.
We now come to 4: DPn adjoins to TPk , as the final step of the derivation.
DPn is a complex nominal, containing an RC. There is no constraint on the
category of an adjunct, so if there is no illegal integration of the formal features,
syntactic merger is permitted.
We conclude this subsection with a description of one more property shared
by MSCs and CHCs, providing further support for the unified analysis. Recall
that in both CHCs and MSCs the determiners of the Head nominals must
be identical (Section 6.3.2B). Link (1984) observes that partitive nominals
apparently allow for RCs with multiple Heads, regardless of the differences in
the nature of the determiners of the conjunct nominals, as in (6.61a). Such data
are in contrast to non-partitive data like (6.61b). Moltmann (1992a: 202) notes
that this contrast in CHCs is also seen in MSCs, as shown in (6.62):
(6.61) a. all of the students and several of the professors who have met in secret
b. ∗ all students and several professors who have met in secret
(6.62) a. John saw all of the students and Mary met most of the professors who have
met in secret.
b. ∗ John saw all men i and Mary met most womenj whoi&j danced together.
With respect to (6.61a) and (6.62a), Moltmann (1992a: 202) explains that
“the reason why the agreement condition on the determiner is met in this case,
most plausibly, is that the students and the professors share definite determiners.
If this is true, then apparently the relative clause relates to the lower NPs, rather
than to the conjunct partitive NPs.” We adopt her account. In our analysis,
accordingly, both (6.61a) and (6.62a) have the following DP as the Head of the
RC whom have met in secret:
(6.63) DPm
DPi D′m
the students Dm DPj
e the professors
The derivation of (6.61a) is illustrated in (6.64). DPi the students undergoes

sideward movement to be remerged with of, and then of the students merges
with all, building the new nominal all of the students. Similarly, DPj the
professor undergoes sideward movement to be remerged with of, and then
of the professors is merged with several, building the new nominal several of
the professors. After that, the two newly built complex nominals merge with
the conjunction and, one after the other, forming the coordinate DP complex
all of the students and several of the professors. In the old working site, the
remnant DPm becomes the Head of the RC, forming DPn . Finally, DPn adjoins
to the coordinate complex.
(6.64) a. all of the students and several of the professors who have met in secret
b. DPk
final merger
DPk DPn
DP D′k DPm [whom have met in secret]
all of Dk DP DPi D′m

[the students]i
and several of ti Dm DPj
[the professors]j
e tj
The derivation of (6.62a) is illustrated in (6.65). DPi the students undergoes

sideward movement to be re-merged with of, and then of the students merges
with all. Thus a new nominal all of the students is built. Similarly, DPj the
professor undergoes sideward movement to be re-merged with of, and then of
the professors merges with most. Thus a new nominal most of the professors
is built. After that, the two newly built complex nominals are selected by the
transitive verbs saw and met, respectively. Other operations occur to build the
two clauses separately: John saw all of the students and Mary met most of the
professors. Then, a process occurs by which these two clauses merge with the
conjunction and, one after the other, and form the clausal coordinate complex
John saw all of the students and Mary met most of the professors. In the old
working site, the remnant DPm becomes the Head of the RC, forming DPn .
Finally, DPn adjoins to the coordinate clausal complex.
(6.65) a. John saw all of the students and Mary met most of the professors who
have met in secret.
b. TPk
final merger
TPk DPn
TP T′k DPm [whom have met in secret]
John saw all Tk TP DPi D′m

of [the students]i
and Mary met most ti Dm DPj
of [the professors]j
e tj
6.3.4 A comparison with the multiple dimensional analysis

We now compare our analysis with Moltmann’s (1992a; see also Wilder 2008:
253) multiple dimensional analysis of MSCs. Moltmann (1992a: 147) claims
that in the example A man came and a woman left who knew each other
well, “a man and a woman are implicitly coordinated.” The notion of implicit
coordination means that two elements function like a coordinate complex, but
do not form a coordinate complex syntactically. This differs from the present
proposal, where elements are conjoined by a null conjunction at a certain step
in the derivation of MSCs. Moltmann’s syntactic representation of (6.66a) is
(6.66b):
(6.66) a. A man came and a woman left who knew each other.
b. a man came
IP and NP V who knew each other
a woman left
In (6.66b), the two clauses are conjoined, but not the two NPs. Using the
same approach, the structure of (6.67a) is given by Wilder (1999) as (6.67b):
(6.67) a. John met a man and Mary met a woman who knew each other well.
b. IP
IP and IP
John I′ Mary I′
I VP I VP
V NP V NP
met a man CP
met a woman
who knew each other well
One major difference between our approach and the multiple dimensional
analysis is that in the multiple dimensional analysis, the apparent antecedents
do not move. Instead, they are linked to the RC by special tree branches.
One disadvantage of the multiple dimensional analysis is that the two overt
antecedents of the relative pronoun, a man and a woman in (6.67), never form
a constituent in the derivation, and thus additional constraints are required to
cover the facts presented in Section 6.3.2. For instance, if the two objects are
selected by two verbs respectively, they need not have the same D-features.
Then the unification of D-features discussed in Section 6.3.2B needs to be
accounted for. Moltmann (1992a) is indeed forced to stipulate certain con-
straints in order to restrict the derivations. In my approach, the restrictions
are captured naturally: the two nominals originally form a coordinate complex
serving as the unique antecedent of the relative pronoun, which does not tolerate
internal disagreement in D-features.
Another disadvantage of the multiple dimensional analysis is that it depends
on the linearization of multiple-dimensional structures, a more complex oper-
ation than linearalization from two-dimensional structures. Advocates of the
multiple-dimensional analysis have sought to formalize the necessary opera-
tion, but if their empirical goals can be met by the simpler two-dimensional
approach, I see no advantage to the more complex multiple-dimensional
approach.

In this section, I have proposed a syntactic derivation for MSCs such as Mary
met a man and John met a woman who knew each other well. I have claimed that
the two modified elements of such a construction are originally two conjuncts
of a coordinate nominal. Then each undergoes sideward movement from the

original working site to a new one, where it is selected by a verb. The two
nominals take part in the building of a coordinate clausal complex. In the old
working site, a complex nominal is constructed, in which the relative clause
takes the remnant coordinate nominal as its antecedent. Finally, the complex
nominal adjoins to the coordinate clausal complex. This analysis shows how the
rejection of the CC and the freedom in the selection of landing sites gained from
sideward movement make syntactic derivation of this special type of relative
clause construction possible, while at the same time avoiding the problems of
the alternative multiple-dimensional analysis.
This analysis also shows that the derivation of the relative clauses in MSCs
do not pose problems for either of the two major analyses of relative clause
constructions, namely the conventional Matching-Adjunction Analysis and the
Raising-Complement Analysis (Kayne 1994; Bianchi 1999; among others).
In the Matching-Adjunction Analysis, our proposed sideward movement may
occur before the adjunction of the relative clause to the Head nominal. In the
Raising-Complement Analysis, the relative pronoun and the antecedent DP
are initially merged together, and later the antecedent DP moves away from
the relative pronoun (Bianchi 1999; Zwart 2000). In this latter analysis, our
proposed sideward movement may occur before the antecedent DP is merged
with the relative pronoun. Thus just as MSCs do not require special multiple-
dimensional structures, RCs in MSCs can be analyzed with either of the two
major approaches to relative clause constructions.
6.4 Deriving Interwoven Dependency Constructions by

giving up the EC
This section examines the syntactic derivation of a construction that exhibits
dependencies between one coordinate complex and two syntactic gaps, with
each conjunct of the complex associated with one of the gaps, respectively. The
construction is shown in (6.68).
(6.68) [[Which nurse]i and [which hostess]j ]k did Fred date i and Bob marry j ,
respectively?
The construction illustrated in (6.68) is called the Interwoven Dependency

Construction (IDC) (Postal 1998). In this construction, conjuncti and conjunctj
in the coordinate complexk are associated respectively with gapi and gapj .
I claim that the construction is derived by sideward movement of ele-
ments from conjuncts. My analysis shows how rejecting the EC and accepting
sideward movement make it possible to give syntactic derivations for IDCs,
6.4 Interwoven Dependency Constructions and the EC 169
while avoiding the problems associated with alternative approaches. This effort
shows how the challenges posed by the construction can be faced without giving
up key assumptions of generative syntax (cf. Postal 1998).
I first introduce IDC data in Section 6.4.1, summarize previous analyses of
the construction in Section 6.4.2, motivate a movement approach in Section
6.4.3, and argue for a sideward movement analysis in Section 6.4.4. A brief
summary is given in Section 6.4.5.
6.4.1 The Interwoven Dependency Construction (IDC)

The IDC example introduced above is repeated here as (6.69):
(6.69) [[Which nurse]i and [which hostess]j ]k did Fred date i and Bob marry j ,
respectively?
The IDC contains a coordinate DP-complex, DPk in (6.69). The construction
poses a serious challenge to syntactic theory because each conjunct of the DP-
complex has a clear base position, but the complex as a whole does not. If
(6.69) is derived via wh-movement, the base position of which nurse should be
the object position of date, which is inside the first clausal conjunct, and that
of which hostess should be the object position of marry, which is inside the
second clausal conjunct. However, there doesn’t seem to be a launching site for
the complex which nurse and which hostess.
To the best of my knowledge, IDCs have previously been discussed most fully
in Tai (1969: 14), Dougherty (1970a; 1970b), McCawley (1988a), and Postal
(1998). IDCs need not involve wh-movement like in (6.69). The examples
in (6.70) illustrate IDCs involving subjects that have presumably been base-
generated below TP and raised to SpecTP; they too seem to have split launching
sites. Note the cross-category nature of the predicates here: they can be PP, AP,
and various projections extended from VP (such data are found in Dougherty
1969: 628; 1970a: 527, 544; Scha 1981; Link 1984: 246; McCawley 1988a:
536; Krifka 1990: 165; and Postal 1998: 108, 162):
(6.70) a. [Hans and Fritz]k are from Germany and from Switzerland.
b. [Logical and empirical truth]k are necessary and contingent, respectively.
c. [John and Bill]k were certain to leave and ready to leave respectively.
d. [John and Bill]k should go to N.Y. and will go to Chicago respectively.
e. [Kim and Sandy]k sang and danced, respectively.
f. [George and Martha]k respectively denounced and were denounced by
the governor.
IDCs are also found with topicalization. In each of the following examples,
if the two PP conjuncts have undergone A-bar fronting from the two predicate
conjuncts, the sentence can also be treated as an IDC ((6.71b) is from McCawley
1988a: 538).
(6.71) a. Under the pillow and in the drawer Lulu put her diary and hid her letters,
respectively.
b. On Fridays and (on) Saturdays, John respectively teaches and goes surfing.
IDCs are observed cross-linguistically (see Zhang 2007b). Although their

syntactic derivations have not been thoroughly studied, they have been claimed
to pose a serious challenge to any framework of generative syntax positing
movement operations (Postal 1998: 163).
Before I present an analysis of IDCs, I clarify the role of the adverb respec-
tively in them.
IDCs, by definition, have a distributive reading, thus permitting the adverb
respectively. However, it is not obligatory; respectively need not occur if the
distributive reading is made clear in some other way, as in the following.
(6.72) [The dogs and the roosters]k barked and crowed all night.
Nevertheless, in most of the IDC examples discussed here, if the adverb

respectively is absent, the sentences can be ambiguous between an IDC read-
ing (distributive) and an Across-the-Board dependency reading, if semantically
and pragmatically permitted. In the latter reading, the sentence-initial coordi-
nate complex is semantically related to the two clausal or predicate conjuncts
simultaneously (Postal 1998: 136). For instance, the Across-the-Board reading
of How many frogs and how many toads did Greg capture and Lucille train? is
a question asking for two numbers, x and y, such that Fred captured x frogs and
y toads and Lucille trained x frogs and y toads. Since IDCs are not Across-the-
Board constructions, I do not discuss this alternative reading here, but leave
Across-the-Board constructions to Chapter 9.
The possibility for respectively to occur in IDCs distinguishes them from
other constructions. The following coordinate complexes do not allow respec-
tively and thus are not IDCs.
(6.73) a. John and Mary are my two best friends and two oldest colleagues
(∗ respectively).
b. I consider John and Mary two excellent philosophers but two lousy parents
(∗ respectively). (see Heycock and Zamparelli 2005: 248)
Semantically, the adverb respectively is licensed by a plural element in the

context. In IDCs, it is licensed by the two coordinate complexes. Syntactically,
the adverb is an adjunct here that can be merged to either the left or the right
of VP or at any projection higher than VP. In (6.71b), for instance, it is merged
to the left of VP, and thus surfaces between the raised subject John and the
coordinate predicate teaches and goes surfing. In some other IDC examples,
the adverb occurs in sentence-final position. In this case, it may be merged to

the right of VP or some higher projection. I do not consider the adverb in my
proposed syntactic derivations of IDCs, since there is nothing special about the
syntax of this adverb.
6.4.2 Previous analyses

While Postal (1998) uses IDCs to challenge generative syntactic theory as
a whole, Tai (1969: 15) and Dougherty (1970a; 1970b) proposed syntactic
derivations for IDCs within generative frameworks. In Tai (1969: 15),
(6.74a) is derived from the structure in (6.74b) by the so-called Respectively
Transformation.
(6.74) a. John and James are eager to please and easy to please, respectively.
b. and S0
S1 S2
John is eager to please James is easy to please
Tai does not give any details about this special syntactic operation, but
consistent with much generative theorizing of the time, takes the Respectively
Transformation as a primitive operation requiring no further elaboration. Since
there seems to be no way to decompose this transformation into more basic
operations, I do not adopt this analysis.
In Dougherty’s approach, IDCs are derived by the so-called Respectively
Substitution Transformation (Dougherty 1970a: 544; 1970b: 887). In his
approach, the subject of the second clausal conjunct moves to the subject
of the first clausal conjunct to fill in a dummy element introduced by his phrase
structure rules, and the predicate of the second clausal conjunct also moves
to the predicate of the first clausal conjunct to fill in another base-generated
dummy element. Thus (6.75a) should be derived from the structure in (6.75b):
(6.75) a. Kim and Sandy sang and danced, respectively.
b. S0
S1 S2 respectively
NP1 VP1 NP2 VP2
NP3 NP4 VP3 VP4 Sandy V
Kim ∆ sang ∆ danced

Dougherty (1970b: 887) states that “[t]he Respectively Substitution Transfor-

mation operates to ‘fill in’ dummy elements introduced by the phrase-structure
rules of the base.” In (6.75b), there are two dummy elements, NP4 and VP4 ,
each marked by . They are replaced by NP2 and VP2 , respectively.
The techniques used in Dougherty’s derivation are problematic, from the
viewpoint of contemporary syntactic theory. In particular, the generation of the
dummy elements is not justified locally, but they are created merely in order to
be replaced later. We must thus pursue a new analysis of IDCs.
6.4.3 IDCs exhibit parallel movement dependencies

I argue that IDCs should be analyzed by movement originating from the clausal
conjuncts. I have four arguments for this movement approach, and against the
hypothesis that the coordinate nominal complexes in IDCs are base-generated
in their surface positions.
First, in varieties of English disallowing extraction of an indirect object
(including standard varieties) such extraction is also disallowed in IDCs (Postal
1998: 135).6 In (6.76a), the object of the preposition to is moved, whereas in
(6.76b), the indirect object is moved. The parallel acceptability pattern in
(6.77a,b) shows that the IDCs in (6.77) obey the same constraint on movement
seen in (6.76).
(6.76) a. [Which nurse] did Ernest sell cocaine to (and George sell heroin to )?
b. ∗ [Which nurse] did Ernest sell cocaine (and George sell heroin)?
(6.77) a. [[Which nurse]i and [which hostess]j ]k did Ernest sell cocaine to i and
George sell heroin to j , respectively?
b. ∗ [[Which nurse]i and [which hostess]j ]k did Ernest sell i cocaine and
George sell j heroin, respectively?
Second, if reflexive binding is a local structural relation, Postal’s (1998: 161)

analysis of examples like the following implies that the two wh-conjuncts of
the IDC show the same movement dependencies, since each wh-phrase can
license its reflexive before moving:
(6.78) [Which man]i and [which woman]j did respectively I talk to i about
himselfi , and you talk to j about herselfj ?
The reflexive binding here follows the strict gender and number agreement
restrictions between antecedents and reflexives.
6 See Bruening (2001: 236 fn. 5) for a discussion of dialect variation in whether indirect objects
can be moved.
Third, if s-selection is also a local structural relation, we infer that in the

following IDC, how many cakes is base-generated in the VP headed by bake,
since the latter s-selects the former. Similarly, we infer that how many letters
is base-generated in the VP headed by write, since the latter s-selects the
former.
(6.79) How many cakes and how many letters, respectively, did Mary
bake and John write this morning?
The s-selection relations between the predicates (barked, crowed) and the
subjects (the dogs, the roosters) in (6.72) demonstrate the same point.
Fourth, the gaps in IDCs cannot be bound variable pros, since such pros are
not available for PPs, in data like (6.71).
I conclude that the left-peripheral coordinate complexes in IDCs are not
base-generated in their surface positions.
6.4.4 Deriving IDCs by sideward extraction from conjuncts

Postal (1998: 160–162) has argued that it is impossible to move the whole coor-
dinate wh-complex in examples like (6.69) from the gap position of either or
both of the clausal conjuncts. We have seen that this is intuitively right: one gap
is semantically related to one conjunct of the wh-complex, and the other gap is
semantically related to the other conjunct of the wh-complex. There is no source
position for the whole coordinate wh-complex in the two clausal conjuncts.
This conclusion is consistent with the arguments given in the previous section
that IDCs involve multiple movement dependencies. Since these dependencies
seem to involve extraction from clausal conjuncts, accounting for them requires
rejecting the EC. I now propose a derivation for IDCs incorporating these
insights. The sentence in (6.80a), for instance, is derived from the steps in
(6.80b) and the structure in (6.80c). The important steps are marked by 1 ,
2 , and 3 (as stated in Section 6.4.1, we ignore the adverb respectively in our
representations).
(6.80) a. Kim and Sandy sang and danced, respectively.
b. Assemble: [VP Sandy danced];
The nominal Sandy undergoes sideward movement and is remerged with
and in the new working site, forming [and Sandy];
Assemble: [VP Kim sang];
The nominal Kim undergoes sideward movement and is remerged with
[and Sandy], building [DP Kim and Sandy];
The two VPs and the conjunction and form a coordinate VP complex;
The coordinate VP complex is merged with T and T is projected;
The complex [DP Kim and Sandy] is merged with T , appearing at SpecTP.
c. TP
DPk T′
DPi D′k T VP
Kim
Dk DPj VP V′
Sandy
and ti sang V VP
and t j danced
Keep in mind that when Sandy undergoes sideward movement, DPk does not

yet exist, and the merger of DPk with T occurs even later. Thus, from the local
perspective, the sideward movement of Sandy does not violate the Extension
Condition, the requirement that substitution operations in overt syntax always
extend their target (Chomsky 1993).
Similarly, (6.81a) is derived from the steps in (6.81b) and the structure in
(6.81c).
(6.81) a. [[Which nurse]i and [which hostess]j ]k did Fred date ti and Bob marry tj ,
respectively?
b. Assemble: [TP Bob marry which hostess];
The phrase which hostess undergoes sideward movement and is remerged
with and in the new working site, forming [and which hostess];
Assemble: [TP Fred date which nurse];
The phrase which nurse undergoes sideward movement and is remerged
with [and which hostess], building [DP which nurse and which hostess];
The two TPs and the conjunction and form a coordinate TP complex;
The coordinate TP complex is merged with did at C;
The complex [DP which nurse and which hostess] is merged with
C and appears at the SpecCP.
c. CP
DPk C′
DPi D′k C TPm

which nurse
Dk DPj did TPi T′m
which hostess
and Fred date t i Tm TPj
and Bob marry tj

In this proposed derivation, there are two parallel movement dependencies,

each with its own landing and launching sites. Specifically, each DP traveler
undergoes a sideward movement, landing at a position that neither c-commands
nor is c-commanded by the launching site. The landing site is in the working site
of the coordinating head of DPk , whereas the launching site is in the working
site of a VP. In the derivations illustrated above, the movement in step 2 is not
blocked by the movement in step 1 , because the landing site of the movement
in step 1 does not c-command the launching site of the movement in
step 2 . The two positions are in different working sites.
In Chapter 5, I concluded that elements may be extracted from single con-
juncts if the conjuncts are semantically related to each other, i.e., in AC. In
the derivation of IDCs, however, elements are extracted from both conjuncts.
As with MSCs, which also involve movement of both conjuncts, this semantic
condition does not apply here. I will say more about where such semantic
constraints do and do not apply in the next chapter.

IDCs have been claimed to pose a serious challenge to generative syntax. I
have responded to this challenge by arguing that the two conjuncts of the
left coordinate complex in an IDC independently undergo sideward movement
from the gap positions, forming a coordinate complex with a conjunction,
and later the newly built coordinate complex is integrated into the complex
clause.7
The analysis here shows how rejection of the EC and the landing site freedom
gained from sideward movement makes the syntactic derivation of certain
7 One might wonder whether IDCs display island effects. It has been claimed in the literature
that sideward movement permits extraction from islands if the islands are not yet integrated
into the core structure (see Nunes and Uriagereka 2000; Hornstein 2001; Nunes 2004; Taylor
2006). Although de Vos and Vicente (2005: 103 (21)) report some island effects in IDCs, my
own investigation does not give a consistent result. In October 2002, I asked six native English
speakers to judge the following adjunct island and tensed wh-island sentences:
(i) a. How many frogs and how many toads, respectively, did Greg get a prize after
capturing and Lucille get a prize after training?
b. How many frogs and how many toads, respectively, did Greg ask who has captured
and Lucille ask who has trained?
Three of the informants accepted both sentences and the remaining three rejected both. This
inconsistency in acceptability hints that IDCs may not display clear island effects, though a
larger judgment experiment would be needed to detect any subtle pattern that may exist (Myers
2009).
special constructions possible, while avoiding the problems associated with

alternative approaches proposed in the literature.8
6.5 Chapter summary

In the previous two chapters, I have argued that neither the CC nor the EC
is a syntactic constraint on the derivation of coordinate complexes. In this
chapter, I have presented derivations for three constructions that crucially rely
on rejecting the CSC. Indeed, if the CSC is truly a syntactic constraint, these
constructions seem to be difficult if not impossible to derive. This chapter has
thus demonstrated the empirical advantages of abandoning the CSC, already
shown to be problematic in earlier chapters.
In my proposed derivation for SACs, single conjuncts are moved from non-
distributive coordinate structures. In other words, the two conjuncts must be
semantically related. In my proposed derivations for MSCs and IDCs, however,
both conjuncts of coordinate constructions are affected by movement, and so
are not subject to this semantic constraint.
In the proposed derivations, the CC is ignored in SACs and MSCs and the EC
is ignored in IDCs. EC-like effects only arise in certain semantic, pragmatic,
or processing contexts, as explained in Chapter 7; there is thus no constraint
in syntax to block the derivations proposed here for IDCs. CC-like effects are
partially phonological and morphosyntactic, but since the coordinators assumed
in SACs and MSCs are null, the relevant lexical properties exhibited by and
do not exist. That is, no conjunct needs to serve as phonological host of a
coordinator, since the coordinator is null, and no conjunct needs to provide
categorial features to the coordinator, since the coordinator is assumed to be
a (null) D with its intrinsic D-features. Therefore, the conjunct movements
posited for SACs and MSCS are licensed.
8 The explanatory power of sideward movement might also benefit the analysis of other con-
structions, such as a split variable construction discussed in Zhang (2008b) and a Japanese
construction discussed in Takano (2002), which, consistent with this study, concludes that “UG
has a way of conjoining derived constituents, in addition to base-generated ones” (p. 275).
7 Relativized parallelism
in syntactic complexes
7.1 Introduction
In Chapters 4 and 5, I linked the effects of the CSC to the lexical–syntactic
properties of coordinators and the semantic relations between conjuncts. In this
chapter I examine the CSC together with some other constraints which have
also been claimed to apply exclusively in coordinate complexes.
Many authors consider Ross’s CSC to be related to or covered by the Par-
allelism Requirement (PR). The PR is a label for the generally recognized
requirement that the syntactic and semantic representations of the components
in syntactic complexes must not only be well-formed independently, but also
symmetrical or parallel to each other. In particular, it has long been realized that
the conjuncts in a coordinate complex obey something like the PR. However,
it is not clear what part of the cognitive system is responsible for the PR, nor
how far coordinate complexes actually obey it.
Moreover, although many authors have attempted to account for the CSC in
terms of some version of PR, or at least to link the two notions, the implementa-
tion of the PR itself remains a controversial issue. As a type of PR, the CSC has
been analyzed as a trigger of movement (Hornstein and Nunes 2002), a con-
straint on movement (Williams 1978: 37; Pesetsky 1982; and Postal 1998), and
as a constraint on merge (Johnson 2002a). However, it has also been claimed
in the literature that the CSC is not a constraint on syntactic operations at
all (e.g. Ruys 1992; Munn 1993; Rögnvaldssen 1993: Section 3.4; Heycock
and Kroch 1994: 271–273; Levine 2001; Lin 2001; Kehler 2002: 141; Potts
2002).
In this chapter I argue that (beyond the lexical–syntactic properties of coordi-
nators discussed in Chapter 4) all apparently coordination-specific constraints
(including, but not restricted to, the CSC) can be covered by a general relativized
parallelism filter. Clarifying the key issues in the PR literature, I show that this
filter is applied to linguistic representations rather than syntactic operations,
and that it is motivated by economy considerations in processing.
177
178 Relativized parallelism in syntactic complexes
The chapter is organized as follows. In Section 7.2, I first review four

coordination-specific constraints in addition to the CSC, and propose my own
generalization, the Relativized PR (RPR), to cover all of the relevant observa-
tions. Then in Section 7.3 I unpack the components of the RPR and illustrate
them with examples. In Section 7.4, I review some experimental results com-
patible with my conclusions. In Section 7.5, I show that the RPR does not affect
syntactic operations and that it is motivated by economy in processing. Finally,
based on a summary of this book’s argument so far, in Section 7.6 I conclude
that there is no coordination-specific constraint on syntactic operations.
7.2 The Relativized Parallelism Requirement (RPR)

7.2.1 The Coordination of Likes Constraint and other similar constraints
In this section, we review some problems with five constraints that have been
proposed exclusively for coordinate complexes, and propose our own unified
generalization for the relevant facts.
Ross’s (1967) CSC was designed to constrain the syntactic operation Move,
in coordinate constructions exclusively. Ten years before the CSC was pro-
posed, a different constraint had been proposed to restrict the syntactic opera-
tion Merge, also exclusively in coordinate constructions. Consider the following
data:
(7.1) a. the scene [PP of the movie] and [PP of the play]
∗
b. the scene [PP of the movie] and [that I wrote]
In (7.1a), both conjuncts are PPs. In (7.1b), however, the first conjunct is a
PP and the second one is a clause. In order to account for the unacceptability of
data like (7.1b), Chomsky (1957: 36; 1965: 212 fn. 9) claims that syntactically
different categories cannot be conjoined. This constraint has been referred to as
the Coordination of Likes Constraint (CLC) (see also Williams’s (1978) “Law
of Coordination of Likes,” and Gazdar 1981: 172 “only items of the same
syntactic category can be conjoined”).
Ten years after the CSC, Schachter (1977) proposed another constraint, the
Coordinate Constituent Constraint (CCC), which states that conjuncts must not
only belong to the same syntactic category but also have the same semantic
function. Unlike the CLC and CSC, the CCC is a constraint on surface repre-
sentations rather than on syntactic operations. The CCC is supposed to cover
the effects of both the CLC and the CSC. Let us see how.
On the one hand, Schachter points out that although the CLC can account
for (7.1b) above, it cannot explain data like (7.2) below. Both of the conjuncts
7.2 The Relativized Parallelism Requirement 179
are clauses in (7.2a), and adverbs in (7.2b), satisfying the CLC. However, the
sentences are still not acceptable (Gleitman 1965: 263; Schachter 1977: 89;
similar data can be found in Haspelmath 2007: (54)).
∗
(7.2) a. What are you doing and shut the door.
∗
b. John probably and unwillingly went to bed.
Schachter observes that the semantic functions of the two conjuncts in the
sentences in (7.2) are different. In (7.2a), one is a question and the other
is a command, and in (7.2b), one adverb is speaker-oriented (probably) and
the other is subject-oriented (unwillingly) (p. 89). Accordingly, he claims that
the CCC, which requires conjuncts to have the same semantic function, can
explain what the CLC cannot explain. Thus the CCC seems to be superior to
the CLC. The relationship between the CLC and “semantic categories” has also
been discussed by Sag et al. (1985: 143), Goodall (1987), and Munn (1993:
113–120; 1996: Section 3).
At the same time, Schachter claims that the CCC can also cover the CSC.
This is because extraction of a conjunct (the CC part of the CSC) or an element
from a conjunct (the EC part of the CSC) creates a syntactic representation in
which the two conjuncts are neither identical in syntactic category nor identical
in semantic function.
∗
(7.3) a. Who did John kiss and Mary? (CC)
∗
b. The lute which Henry [plays and sings madrigals] is warped. (EC)
In Schachter’s approach, after the movement of the first conjunct in (7.3a),

the two conjuncts of the coordinate complex become ∅ and Mary, which are
claimed to differ in both syntactic category and semantic function (Schachter
1977: 95). Similarly, after the extraction from the first conjunct in (7.3b), the
two conjuncts become [plays] and [sings madrigals]. “Plays is functioning
as a relation between Henry and lute; sings madrigals, on the other hand, is
functioning as a property of Henry. Relations and properties – or two-place
predicates and one-place predicates – are distinct semantic functions, and their
conjunction is prohibited by the CCC.” (See Gazdar 1981: 173 fn. 25 for a
similar claim about the CSC.) The CCC thus takes a representational approach
to the facts ascribed by the CSC to constraints on movement.
Since Schachter believes that the CCC entails the CSC, he assumes that the
CSC is descriptively valid. Thus despite acknowledging the acceptability of
extraction in AC, he claims that such examples are not instances of coordination
at all, since the word and in AC could be a kind of complementizer, rather than
conjunction (p. 94) (see arguments against this assumption in Section 2.4).
In previous chapters we have supported Culicover and Jackendoff’s (1997)

claim that AC is true coordination, and have shown that both the CC and the
EC can be violated in natural coordination. Two examples where the CSC is
violated are listed in (7.4). Since violations of the CSC create representations
that violate the CCC, the acceptability of examples like (7.4) indicates that like
the CSC, the CCC is too strong.
(7.4) a. What kind of herb can you [eat and not get cancer]? (EC violation)
b. Baoyu yijing [ gen Daiyu] he-mai-le yi zuo fangzi. (CC violation)
Baoyu already and Daiyu co-buy-prf one cl house
‘Baoyu and Daiyu already co-bought a house.’
Unlike Schachter (1977), who replaces the CLC (sensitive to syntax) with
the more restrictive CCC (sensitive to both syntax and semantics), Sag et
al. (1985: 143), Goodall (1987: 43), and Munn (1993) instead advocate a
less restrictive constraint requiring conjuncts to be identical only in semantic
types. For instance, two conjuncts of a coordinate complex must both denote
[+ manner], or both denote [+ temporal], and so on. This semantic constraint
successfully captures the acceptability of data like the following, which cannot
be accounted for by either the CLC or the CCC:
(7.5) Jermaine is boring and a fool. (Sag et al. 1985) (AP & NP)
In (7.5), the two conjuncts are not identical in their categories: one is an
AP and the other is an NP, violating the CLC and the CCC. But both denote
negative properties of a person. Let us call this semantic constraint CLCsem.
However, this constraint cannot account for examples like (7.6):
(7.6) Do that and I’ll give you $10. (Wilder 1999: 9)
In (7.6), the two conjuncts are not identical in their semantic type: one is a
command and the other is not.
In order to cover data like (7.5), Peterson (2004: 650) claims that conjuncts
should be identical in grammatical functions (such as OBJECT, ADJUNCT, and
PREDICATE). Let us call this constraint CLCfunc. However, this constraint
doesn’t work either, since it cannot account for data like (7.2b), where the
two conjuncts are both ADJUNCTS but the coordinate complex is still not
acceptable.
Various versions of the CLC have been also proposed in other works, such
as Dik (1968: 25), Hudson (1972: 28), Pullum and Zwicky (1986), and Gazdar
et al. (1985).
7.2 The Relativized Parallelism Requirement 181
We summarize the five constraints introduced so far in (7.7). The symbol “+”
marks the type of sentences for which judgments are predicted by the constraint,
and the symbol “−” marks the type of sentences for which judgments are not
predicted by the constraint.
(7.7) (7.1) (7.2a) (7.2b) (7.3) (7.4) (7.5) (7.6)

the CLC + -
the CLCsem + -
the CLCfunc - +
the CCC + + + - - -
the CSC + -
Since each of the five constraints fails to account for certain types of data,
we conclude that none of them is desirable.
7.2.2 The RPR: conjuncts must hold a coherence relation

I propose the following filter on the representations of syntactic complexes:
(7.8) Conjuncts of a coordinate complex must hold a coherence relation in

terms of:
a. Relatedness: they must be related to each other semantically, as in natural
coordination; or
b. Resemblance (Parallelism Requirement): they must hold a resemblance
relation in terms of both their semantic type and their dependency chains.
This filter states a symmetrically and inclusively disjunctive coherence rela-

tion between conjuncts. The inclusive disjunctive nature of this filter is that
conjuncts must satisfy either the relatedness stated in (7.8a) or the resemblance
stated in (7.8b), or both. The symmetrical nature of this filter is that the two
kinds of coherence relation are independent of each other and equally important.
The notion of coherence is adapted from Kehler (2002). Developing pro-
posals of David Hume, Kehler distinguishes three general classes of coherence
relations: Cause–Effect, Contiguity, and Resemblance. I group the first two
classes, together with other relations such as the conditional relation (as in
(7.6)), into my general Relatedness relation. This is what is exhibited in natural
coordination, in contrast to accidental coordination. One conjunct is related to
another conjunct in a context to the extent that its contextual effects in this
context are large (cf. Sperber & Wilson 1995: 125).
Kehler states that the EC cannot be violated when the Resemblance relation
is operative, whereas it can be violated when the other two classes of coherence
relations are operative (p. 116). Examples like (7.3b) show the former case and
examples like (7.4a) show the latter case. Empirically, both Kehler’s theory and
my (7.8) capture the acceptability pattern of the relevant data. In (7.8), I make
explicit the disjunctive nature of the relation between Resemblance and other
kinds of coherence relations. Precisely speaking, I propose it is the absence of
Relatedness as defined in (7.8) that makes the Resemblance relation become
obligatory. Therefore, EC effects correlate with the absence of Relatedness.
If the resemblance requirement is a Parallelism Requirement (PR), the coher-
ence filter in (7.8) should be understood as a relativized PR. I thus call this filter
the Relativized Parallelism Requirement (the RPR).
Let us now see how the RPR covers the representative data listed above:
(7.9) (7.1) (7.2a) (7.2b) (7.3) (7.4) (7.5) (7.6)

(7.8a): Relatedness + +
(7.8b): Resemblance in semantic type + + + + +
(7.8b): Resemblance in movement history + +
In the data in (7.1) through (7.3), no semantic relation between the conjuncts
is seen. Thus (7.8a) does not apply. However, (7.8b) does apply here. In the
acceptable example (7.1a), the conjuncts are of the same semantic type, whereas
in the unacceptable (7.1b), (7.2a), and (7.2b), the conjuncts are of different
semantic types. In (7.1b), the PP of the movie is the semantic licensor of the
relational noun scene, whereas the RC that I wrote is a modifier. In (7.2a),
the first conjunct is a question and the second one is an order. In (7.2b), the
speaking-oriented adverb probably and the subject-oriented adverb willingly
are of different semantic types. The first component of (7.8b) alone rules out
these examples. Moreover, in (7.2a), (7.3a) and (7.3b), movement occurs in
one conjunct, but not in the other. The second component of (7.8b) alone can
also account for their unacceptability.
In (7.4a) and (7.4b), a movement chain starts from a single con-
junct. The examples are acceptable, because the conjuncts in each coor-
dinate complex are related to each other. (7.4a) expresses a “neverthe-
less” relation between the two conjuncts, and (7.4b) expresses a togeth-
erness relation between the two conjuncts, and so their acceptability is
predicted by (7.8a). The conjuncts in (7.5) are of the same semantic
type (negative properties of persons), and so the acceptability of (7.5)
is predicted by (7.8b). The example in (7.6) is acceptable, because the
two conjuncts are semantically related to each other via a condition –
result relation, and so the acceptability of (7.6) is predicted by (7.8a).
One characteristic of the RPR is that syntactic category is not considered.

As shown in Chapter 3, generalizations relating to the syntactic categories of
conjuncts and coordinate complex have already been taken care of indepen-
dently. The RPR does not impose any new constraints in this regard.
Another characteristic of the RPR is that the coherence requirement is also
found in larger linguistic representations such as narratives (Kehler 2002). Thus
elements linked by overt coordinators and those that show conjunctive relations
but are not linked by overt coordinators are, plausibly, subject to the same filter.
7.3 The components of the RPR

The RPR states that if the conjuncts of a coordinate complex are neither related
to each other semantically, nor similar to each other in terms of semantic
type and dependency chains, the complex is not acceptable. In this section, I
elaborate on these two components of the RPR: relatedness and resemblance.
7.3.1 Examples of the semantic relatedness between conjuncts

A. AC
If the encoded eventualities of two conjuncts are related to each other, the coor-
dinate complexes are generally acceptable, regardless of whether the conjuncts
show resemblance in syntactic category, semantic type, or movement history.
This is the case with AC. In Section 5.2.2 I mentioned a few examples of AC,
and Huddleston and Pullum (2002: 1299–1304) provide a systematic classifi-
cation. In this subsection, I show how the RPR is superior to other proposed
constraints in handling various types of AC.
Interrogatives and declaratives In (7.10a), the first clausal conjunct

is interrogative and the second one is declarative. In (7.10b), the first clausal
conjunct is declarative and the second one is interrogative.
(7.10) a. How much wine can you drink and still stay sober?
b. Big Louie sees this mess and who’s going to be in trouble? (Culicover
and Jackendoff 1997: 210)
In each example above, there is a certain semantic relation between the

conjuncts. In (7.10b), for instance, the first conjunct encodes a condition for
the eventuality encoded by the second conjunct. In each case, the two conjuncts
are of different types semantically. They are also different in their movement
history: one conjunct contains a wh-movement chain, whereas the other does
not. It is Relatedness (7.8a), rather than Resemblance (7.8b), that accounts for
the acceptability of such examples.
In contrast to the above examples, the conjuncts of the following hold neither
a relatedness nor a resemblance relation.
(7.11) a. ∗ [What has Bill seen] and [he has heard the bad news]? (Culicover and
Jackendoff 1997: 211)
b. ∗ [Bill has seen the broken window] and [what has he heard]?
c. ∗ John asked me for a dollar and why the sky is blue.
In the examples in (7.11), one conjunct is a question and the other is not.
In none of these examples do the conjuncts belong to the same semantic type,
failing to satisfy (7.8b). At the same time, the two conjuncts in each coordinate
complex do not have any semantic relation, failing to satisfy (7.8a). Thus the
RPR is violated.
An EC-like effect is also observed in IDCs:
(7.12) a. What book and what magazine did John buy and Bill read respectively?
b. ∗ What book and what magazine did John buy, [Sue write the novel], and
Bill read respectively? (Kehler 2002: 125)
In (7.12a), both clausal conjuncts are interrogative. In (7.12b), however,

among the three clausal conjuncts, two are interrogative while the middle one,
Sue write the novel, is not. Thus in (7.12b), the conjuncts do not satisfy the
resemblance condition.
The coordination in IDCs is accidental, as signaled by the word respec-
tively, and so the conjuncts cannot satisfy Relatedness. IDCs are therefore only
acceptable if they satisfy Resemblance, as (7.12a) does by having conjuncts of
the same semantic type (interrogative).
Interrogatives and imperatives In examples like (7.13), an impera-

tive expression is coordinated with an interrogative one (Huddleston and Pul-
lum 2002: 1332). Though the two conjuncts in (7.13a) are of different semantic
types, they are nevertheless semantically related: the speaker uses the second
conjunct to ask the hearer to respond to the content expressed in the first con-
junct. However, in (7.13b), the two conjuncts are both of different semantic
types and unrelated to each other. Thus, it is not acceptable.
(7.13) a. Come around six, or is that too early?
b. ∗ What are you doing and shut the door.
Interrogatives and exclamatives In examples like (7.14), an inter-

rogative expression is conjoined with an exclamative one (Huddleston and
Pullum 2002: 1332). The two conjuncts both describe the same event, and thus
they are semantically related, making the sentence acceptable.
(7.14) I remember who was there and what a success it was.
Imperatives and declaratives In (7.15a) and (7.15b), the first clause

is imperative and the second one is declarative. Semantically, the first conjunct
encodes a condition for the second conjunct, and thus the two conjuncts are
semantically related to each other.
(7.15) a. Do that and I’ll give you $10. (= (7.6))

b. Be there or I’ll get angry.
Declaratives and exclamatives In examples like (7.16), an exclama-

tive expression is conjoined with a declarative one (Huddleston and Pullum
2002: 1332). The two conjuncts seem to have a “nevertheless” relation.1
(7.16) What a disaster it was and yet no one seemed to mind.
Passives and actives Passive and active constructions can also be

conjoined. In examples like (7.17), the first conjunct is a passive clause and the
second one is an active clause. The coordination is an AC. The two conjuncts
in (7.17) exhibit Cause-Effect and Contiguity relations (Kehler 2002).
(7.17) I was criticized yesterday and thus I criticized others today.
According to the RPR, the data in (7.10) through (7.17) are acceptable
because the two conjuncts are semantically related to each other.
B. Co-reference with whole conjuncts

Relatedness is also exhibited in co-reference. Coordinate complexes can be
acceptable without sharing semantic type across conjuncts, if one conjunct is
co-referential with an element inside the other conjunct.
(7.18) Let us examine [this alternative possibility] and [how it affects the
predictions of the LF Identity Condition]. (Heim and Kratzer 1998: 255)
In (7.18), the pronoun it in the second conjunct is co-referential with the first
conjunct this alternative possibility. This makes the two conjuncts semantically
related, and thus the coordinate complex satisfies (7.8a) and by extension, the
RPR. This is so even though the two conjuncts are not of the same semantic
1 This is a counterexample to the claim that the nevertheless-type of AC tends to have a modal or
to be non-episodic (see Kehler 2002: 140).
type (the second conjunct is a question whereas the first one is not) and hence
violates (7.8b).
Another type of example showing co-reference between one whole conjunct
and an element of the other conjunct is represented by the following (Zoerner
1995: 94; Progovac 1998b: 6).
(7.19) a. John read the book and quickly.

b. Robin knows Kim and intimately!
It is possible that they are derived from (7.20) by PF-deletion of the verb-
object string in the second conjunct (the underlined part). Since the content of
the deleted string can be recovered from the first conjunct, the deletion follows
Chomsky’s (1965: 144) recoverability condition on deletion.2
(7.20) a. John [read the book] and [read it quickly].

b. Robin [knows Kim] and [knows her intimately]!
In each example in (7.19), the whole first conjunct is co-referential with the
gap between the coordinator and the adverb in (7.20). Thus the two conjuncts
are semantically related.
7.3.2 Examples of the resemblance between conjuncts in semantic types

In this subsection we show that if conjuncts are of the same semantic type,
the coordinate complexes are generally acceptable, regardless of whether the
conjuncts are semantically related to each other or show any resemblance in
syntactic category.
A. Conjuncts of the same force

Question clauses and question words can conjoin since they share question
features, although their syntactic categories are different.
(7.21) a. [What and when] does John (normally) eat?

b. John asked [the time and where the bathroom was]. (Munn 1993: 119)
The two conjuncts in (7.21a) are both question words, and thus belong
to the same semantic type (see Zhang 2007b for a syntactic analysis of the
construction). In (7.21b), the two conjuncts of the coordinate complex belong
to different syntactic categories. One is a DP, and the other is a clause. However,
2 In (7.20), the pronoun it is structurally equivalent to the antecedent the book, and in (7.20), the
pronoun her is structurally equivalent to the antecedent Kim (see Elbourne 2005 and Kratzer
2009 for recent defenses of this analysis of pronouns).
they are of the same semantic type. The DP the time in these examples is an
implicit question, meaning “what time was it,” satisfying the s-selection of ask
in (7.21b) (see Grimshaw 1981; Pesetsky 1982).
Exclamative clauses and non-clausal exclamative expressions can also con-
join (Munn 1996: 4):
(7.22) It’s amazing how tall he is and the things he can do.
B. Focus feature-bearing conjuncts

If some focus-marking element is present in both conjuncts, such as the words
only and even and the degree words more and most, elements that are categori-
ally different can also be conjoined. We have seen examples like the following
in Chapter 3 (Grosu 1985, where parallel examples are given from Finnish,
Modern Hebrew, Kikuriya (Bantu), and Joluo (Nilotic)).
(7.23) a. John eats only pork and only at home.

b. ∗ John eats pork and at home.
(7.24) a. John eats the most unlikely things and at the most unlikely hours.
b. John has stolen more watches and from more unsuspecting victims than
anybody else ever will.
c. I eat [neither meat nor at restaurants]. (Chris Wilder, p.c.)
If both conjuncts are foci, they are of the same semantic type. Similarly,
if both conjuncts contain the same degree word, they can also be of the same
semantic type. The above examples thus satisfy (7.8b), and so their acceptability
is predicted by the RPR. Since the two conjuncts in (7.23b) are neither of the
same semantic type nor semantically related to each other, its unacceptability
is also predicted by the RPR.
C. Circumstantials (time and location)

Temporal expressions can be conjoined, as can locative expressions, regardless
of their syntactic categories, since they are of the same semantic type. In (7.25a)
and (7.25b), two temporal expressions are conjoined, although one conjunct is
a nominal while the other is a PP.
(7.25) a. John went to the library [yesterday] and [on Tuesday]. (NP & PP)
b. John plays at night and every Sunday. (Moltmann 1992a: 25) (PP & NP)
D. Other semantic types

Other types of semantic resemblance, not fitting into the above categories,
permit satisfaction of (7.8b) of the RPR. We have seen one such example in
(7.5), where the semantic type involves negative properties of persons. Further
examples illustrating the wide range of semantic types satisfying the RPR can
be found in Dik (1968: 28), Sag et al. (1985: 117f.), Gazdar et al. (1985: 174),
Zoerner (1995), Munn (1996: 4), Huddleston and Pullum (2002: 1335), among
other places. Consider the following.
(7.26) a. John is sick and in a foul mood. (AP & PP)

b. ∗ John is sick and in the park. (Munn 1993: 117)
(7.27) a. ∗ Dogs are mammals and are barking right now in front of my window.
b. ∗ A dog is a mammal and is barking right now in front of my window.
∗
(7.28) Liz made out Mason to be intelligent and Sarah to be angry.
In both (7.26a) and (7.26b), an AP is conjoined with a PP. The former

is acceptable, since both conjuncts denote properties of a person, the same
semantic type. The latter is not acceptable, since one conjunct denotes a property
and the other a location, different semantic types. Once more, we see that the
RPR captures the acceptability contrast while the CLC cannot. In (7.27) (taken
from Manfred Krifka’s talk at ZAS-Berlin, June 16, 2003), the two predicate
conjuncts are not of the same semantic type: one is generic and the other is
episodic, and they are not related to each other, either. In (7.28), the predicate in
the first conjunct, intelligent, denotes an individual-level property, whereas that
in the second conjunct, angry, denotes a stage-level property. In this example,
the conjuncts are not of the same semantic type, nor are they related to each
other.
The unacceptability of (7.26b) and (7.27) is not predicted by Sag et al.’s
(1985: 119) structural rule that coordinate complexes introduced by the verb
be allow conjuncts of any type. These examples also challenge CLCfunc, as
well as a predicate approach to the CLC (Sag et al. 1985; Chametzky 1987;
Morrill 1990; Zoerner 1995; cf. Munn 1993). It is not true that coordination is
always possible with elements that can be used as predicates. Such approaches
are inadequate because they neglect semantics.
The RPR is exhibited not only in the non-modifier type of coordinate com-
plexes, as in (7.26) through (7.28), but also in the modifier type of coordinate
complexes, including adverbials and modifiers of DPs (some of the examples
in (7.29) are from Peterson 2004: 650).
(7.29) a. John ran down the path, a marked man and desperately
afraid. (NP & AP)
b. In jeans and a T-shirt and sporting two days’ growth on his chin,
John presented a less than inspiring figure. (PP & VP)
c. Late for a meeting and growing impatient, the woman got out and
glowered down at the man as he struggled with a wrench. (AP & VP)
d. Anyone knowing the whereabouts of John Smith and afraid to
tell the police should contact the following emergency number. (VP & AP)
e. John walked [slowly] and [with great care]. (Adv & PP)
f. na ge landuo erqie jingchang chidao de xuesheng (AP & RC)
that cl lazy and often late.come de student
‘that student that is lazy and often comes late’
(7.30) a. ∗ I sat [on the couch] and [with fever]. (Progovac 1998b: 6)
b. ∗ Jessie believes Tracy [to be happy and walks] (Pollard and
Sag 1994: 204)
In (7.29a), the nominal conjunct a marked man and the adjectival conjunct
desperately afraid both express mental states. They can conjoin since they are
of the same semantic type. The other examples in (7.29) also contain conjuncts
of the same semantic type. However, the conjuncts in the examples in (7.30)
are neither semantically related to each other nor of the same semantic type.
In (7.30a), for instance, on the couch refers to a location, whereas with fever
refers to a state.
Note that the coordinate complexes in both (7.29) and (7.30) are syntactically
adjuncts. Hence their acceptability contrasts cannot be covered by Peterson’s
(2004: 650) CLCfunc.
The acceptable coordinate complexes in (7.23) through (7.29) all satisfy
(7.8b), regardless of whether they are modifiers or predicates, since the con-
juncts in each complex are of the same semantic type, and they do not contain
any unlike movement chains.
Not only are forms that do not satisfy the RPR judged unacceptable, but
the same holds for interpretations. For instance, the adverb clearly in (7.31a)
is ambiguous between a manner reading and an evidential reading (synony-
mous with obviously). In (7.31b), this adverb is coordinated with the adverb
loudly, which has only a manner reading. The evidential reading of clearly
in the conjoined structure is no longer available, as correctly predicted by the
RPR.
(7.31) a. Eve was clearly singing a lullaby.

b. Eve was clearly and loudly singing a lullaby.
The construction in (7.31b) is only possible because of the ambiguity of

clearly. With adverbs of unambiguously distinct types, conjunction is not pos-
sible at all, as in (7.2b), repeated here as (7.32):
∗
(7.32) John probably and unwillingly went to bed.
The effect of the PR on interpretation has been extensively discussed in Lang

(1984) and Fox (2000) (see also Potts’s 2002 review of Fox 2000). It thus will
not be addressed any more in this book.
7.3.3 The CLC: two further issues

A. Against a gapping account of CLC violations
The coordination of unlike categories is not consistent with a verb gapping
analysis.
It has often been assumed that sentences like (7.33a) (= (7.5)) are derived
from sentences like (7.33b), after deletion of the second copula.
(7.33) a. Jermaine is boring and a fool.

b. Jermaine is boring and is a fool.
If all cases of coordination of unlike categories were derived by copula-

deletion, the contrast in (7.34) would not be accounted for.
(7.34) a. The bouncer was muscular and was a guitarist. (Goodall 1987: 34)
b. ∗ The bouncer was muscular and a guitarist.
By contrast, in the RPR analysis, (7.34a) is fine because each conjunct

encodes a proposition about the same person, satisfying (7.8b). (7.34b) is
deviant because the two conjuncts are neither semantically related to each
other nor semantically of the same type: muscular denotes a physical property
but a guitarist denotes an occupation.
Another argument against the gapping approach is the fact that accept-
able unlike coordination constructions need not have any corresponding “full-
fledged” forms with syntactically like conjuncts (Munn 1996: 4). For instance,
(7.35a) cannot be derived from the unacceptable (7.35b) or (7.35c). Similarly,
(7.36a) cannot be derived from the unacceptable (7.36b) (see Munn 1993: 120;
Wilder 1999: 10; Peterson 2004: 648 for more discussion of problems with the
copula-gapping analysis).
(7.35) a. John knows neither the murderer, nor where the body is.
b. ∗ John knows neither the murderer, nor knows where the body is.
c. ∗ John knows neither the murderer, nor John knows where the body is.
(7.36) a. Slowly and with great care, was how John walked.
b. ∗ Walk slowly and walk with great care, was how John walked.
B. Coordinate complexes in predicate and subject positions

The present analysis also allows us to account for a puzzle posed by unlike
conjuncts that has not previously been solved in the literature of coordination.
Starting with Gazdar (1981: 172 fn. 24) it has been noted that although an
adjective and a noun can form a coordinate complex in a predicate position, as
in (7.37a), such a complex cannot occur in a subject position, as in (7.37b) and
(7.37c).
(7.37) a. Jermaine is boring and a fool.

b. ∗ Boring and a fool entered the restaurant.
c. ∗ A fool and boring entered the restaurant.
In fact, (7.37b) can be independently ruled out. In Chapter 3, I showed that

in English, the category of the external conjunct, here boring, determines the
category of the coordinate complex as a whole, here boring and a fool. Since
the complex in (7.37b) occurs as the subject of an eventive predicate, it must
be a DP, but it cannot be, because boring is not a DP.
∗
(7.38) Boring entered the restaurant.
As is well known, nominals are property-denoting in predicate positions, but

individual-denoting in the argument positions of eventive verbs. However, the
semantic types of adjectives are consistent. (7.37a) is fine because the predicate
nominal a fool denotes a property, and the word boring also denotes a property.
The two conjuncts are thus of the same semantic type. The two conjuncts of
(7.37c), however, are not of the same semantic type, since the first conjunct,
the nominal a fool, must be individual-denoting in the position of the subject of
the eventive verb enter, whereas the second conjunct, boring, is not individual-
denoting. Therefore, (7.8b) is not satisfied. Moreover, nothing indicates that
the two conjuncts are semantically related (a fool is not necessarily boring).
Thus (7.8a) is not satisfied, either. Therefore, the unacceptability of (7.37c) is
predicted by the RPR.
We can explain the following examples in a similar way (data from Gazdar
et al. 1985: 174; Borsley 1994: 231; Higginbotham 1987: 52):
(7.39) a. He is a sick man and suffering from fever.

b. ∗ [A sick man and suffering from fever] needs rest.
(7.40) a. His father was well known to the police and a devout Catholic.
b. ∗ The [well known and a Catholic] man was my father.
c. ∗ Soon [a Catholic and well known] started shouting again.
(7.41) a. John is a linguist and proud of it.

b. ∗ John met a linguist and proud of it.
c. ∗ He took offense and coffee.
If nominals such as a sick man, a Catholic, a linguist, and coffee are

individual- or entity-denoting in argument positions, they cannot be conjoined
with property-denoting elements, such as suffering from fever, well known, and
proud of it, or abstract notions such as offense, in these positions. Similarly, the
semantics of the adjective stupid and that of the proper noun John are different,
therefore they cannot be conjoined:
∗
(7.42) He is stupid and John. (Higginbotham 1987: 52)
The RPR explains these data (and similar examples in Sag et al. 1985: 141).

dependency chains
If conjuncts do not exhibit relatedness, they show resemblance in dependency
chains such as movement chains in terms of the semantic features and the
phonological realization of the chain links.
A. The semantic features of the links of dependency chains

If both conjuncts have dependency chains, they share a resemblance in the
thematic features of the chain links. Let us start with MSCs. In the two clausal
conjuncts of an MSC, the link of a dependency in one clausal conjunct must
have the same thematic features as the link of the parallel dependency in the
other clausal conjunct. For instance, they must both be themes, as in (7.43a), or
patients, as in (7.43b). But the pattern in (7.43c), in which one link in the first
conjunct is a patient but the parallel link in the second conjunct is an agent, is
not acceptable (more examples can be found in Moltmann 1992b: 175 (58)):
(7.43) a. [A man]i came and [a woman]j left [whoi&j knew each other well].
b. John met [a man]i and Mary met [a woman]j [whoi&j knew each other well].
c. ∗ John saw [a man]i and [a woman]j saw Mary [whoi&j were wanted by the
police].
The RPR can be satisfied even when theta roles are merely similar, rather
than identical. Thus the MSC in (7.44) is acceptable, even though a man is a
theme and a woman is a goal.
(7.44) John saw a man and Mary talked to a woman who were wanted by the police.
There are two ways to analyze the non-contrastiveness of a man and a woman
in (7.44). According to Reinhart (2002), only the features of agent and patient
are contrastive, whereas the features of other theta roles are not. Thus a man
and a woman in (7.44) have sufficiently similar thematic features to satisfy
the resemblence condition of the RPR. According to the typology in Platzack
(2008), the thematic roles of both a man and a woman in (7.44) belong to
family C (not inherently affected, e.g. path, theme), as opposed to family B
(inherently affected, e.g. undergoer, experiencer), as well as family A (e.g.
agent, instrument).
IDCs obey a similar constraint. In the two clausal conjuncts of an IDC, the
gap in one must be in a syntactic position similar to the gap of the parallel
dependency in the other. In other words, the gaps must be associated with
similar thematic roles. For instance, they must be either both agents or both
patients, as in (7.45a) and (7.45b), respectively. In (7.45c), the gap in one
conjunct is a patient whereas the gap in the other conjunct is an agent, and so
is not acceptable.
(7.45) a. [[Which nurse]i and [which hostess]j ]k i dated Fred and j married Bob
respectively?
b. [[Which nurse]i and [which hostess]j ]k did Fred date i and Bob marry j ,
respectively?
c. ∗ [[Which nurse]i and [which hostess]j ]k did Fred date i and j marry Bob,
respectively?
B. The phonological realization of the links of dependency chains

Conjuncts also show resemblances in how the links of dependency chains
are realized phonologically. Namely, they must be both silent or both overt
(e.g. expressed by a pronoun). Consider the wh-movement construction in
(7.46a) and the Chinese topicalization construction in (7.46b). In neither of
these examples does a gap in one conjunct occur with a pronoun in the other
conjunct.
(7.46) a. Which booki did Mary like i and John review (∗ iti )?
b. Na ge reni , Baoyu xihuan i , Daiyu taoyan (∗ tai ).
that cl person Baoyu like Daiyu dislike he
‘That person, Baoyu likes and Daiyu dislikes.’
Unlike the gaps in (7.46), parasitic gaps are not obligatory:

(7.47) Which paperi did Mary file i without reading (iti )?
The obligatory gapping in (7.46) is covered by the second part of the Resem-
blance condition of the RPR, relating to the resemblance of the phonological
realizations of dependency chain links. Specifically, no overt element internal
to the right conjunct may be co-indexed with a gap or silent element in the
left conjunct. The following examples are cited from Wilder (1997: 59, 60,
67, 94).
(7.48) a. ∗ [ i came in] and [{Maryi /shei } sat down].

b. ∗ [John i wine], but [Mary drinksi beer].
c. ∗ [John likes Beethoven] and [Mary likes Beethoven].
We can see that the constraint is also seen in ellipsis constructions, and the
offending overt forms in the second conjuncts are not restricted to pro-forms.
RNR constructions like (7.49) are fine because it is a gap rather than an overt
form in the second conjunct that is related to the gap in the first conjunct, and
the “shared” element (underlined in (7.49)) is outside the coordinate complex
(see Postal 1998; among others).
(7.49) John offered , and Mary actually gave , a gold Cadillac to Billy Schwartz.
This and the previous subsection show that conjuncts must show resemblance
in dependency chains. This can be covered by the following condition of Lang
(1984: 22):3
(7.50) Given some coordinate construction divided by a connector into two

sub-structures s1, s2, then for the specification of a pair cj1, cj2 within s1
and s2 respectively as conjuncts, it holds that
(a) as to their format, cj1, cj2 are delimited by the set of parallel-structured
matching constituents in s1, s2
(b) cj1 and cj2 must display parallel stress patterns
(c) cj1 and cj2 must be contrastable constituentwise
Item (a) of (7.50) explains the acceptability patterns of (7.43) and (7.45).
This is a PR effect. Moreover, from the perspectives of PF and information
structure, items (b) and (c) of (7.50) are valid as well. We can add another item:
between cj1 and cj2, it is impossible for one to be silent and the other to be
realized phonologically.
3 Zoerner (1995: 61) proposes a similar condition, the Condition on Index Association (CIA): In
a series of parallel index sequences, the nth term of one sequence associates syntactically and
semantically only with the nth of any other sequence.
7.4 The RPR in language processing

In this section, we review three types of processing research which show that
syntactically well-built representations are not processed equally. Rather, pro-
cessing is easier if the subcomponents of the representations have a tighter
semantic relation, or have parallel merged structures, or have parallel depen-
dency chains. The results of these processing studies are therefore compatible
with the RPR.
7.4.1 The more tightly semantically connected, the easier to process

It is generally recognized that elements that are more readily integrated into the
sentence are processed faster than elements that are not so readily integrated
into the sentence. For instance, the arguments of a verb are easier to process
than adjuncts of the verb. Thus Speer and Clifton (1998) found that readers
read the same prepositional phrases faster when they were arguments of a verb
than when they were adjuncts. A similar conclusion follows from the finding
that prepositional phrases that can function either as arguments or as adjuncts
tend to be understood as arguments (Schütze and Gibson 1999). Furthermore,
an experiment reported by Lin (2007) found that the possessors of inalienable
nouns (including kinship terms and body parts) were read significantly faster
than their alienable counterparts.
Conjuncts of natural coordination are more readily integrated into the sen-
tence than conjuncts of accidental coordination, so we expect that natural
coordination constructions should be processed more easily than accidental
coordination constructions. This is indeed the case. First, if a coordinate com-
plex is ambiguous between natural and accidental coordination readings, the
default is the former reading. We have seen the examples in (7.51) in Chapter
5. For (7.51a), the collective reading is the default one, whereas the distributive
reading is marked. For (7.51b), the AC reading is the default one, whereas the
non-AC reading is marked.
(7.51) a. Baoyu {he/gen} Daiyu dingqin-le.

Baoyu and/and Daiyu engage-prf
b. John went to the store and bought some ice cream.
As pointed out by Carston (1993: 29), the natural coordination reading “is
overwhelmingly more likely to be recovered by the hearer, and to have been
intended by the speaker,” than the accidental coordination reading. Second,
the experimental studies reported in Frazier et al. (1999) on English and Yu
(2008) on Chinese also show that natural coordination is processed faster than
accidental coordination.
The fact that natural coordination reading is the default reading of a coordi-
nate construction has been discussed from a pragmatic perspective since Grice
(1967), and accounted for by the pragmatic notion of relevance by Carston,
which he claims minimizes processing effort (p. 29; see also Sperber and
Wilson 1995: 260). Relevance is subsumed under the relatedness condition of
the RPR.
7.4.2 The more parallel in merged structures, the easier to process

Frazier et al. (2000) find that the coordination of syntactically like categories,
as in (7.52a), is processed faster than coordination of unlike categories, as in
(7.52b).
(7.52) a. John walked slowly and carefully, avoiding the broken glass.
b. John walked slowly and with great care, avoiding the broken glass.
Likewise, Frazier and Clifton (2001) and Carlson (2002) find that a conjunct
is read faster if it is structurally parallel to the preceding conjunct than if it is
not. (7.53a) was read more quickly than (7.53b), and the sentences in (7.54)
were read more quickly than those in (7.55).
(7.53) a. Hilda noticed a strange man and a tall woman when she entered the house.
b. Hilda noticed a man and a tall woman when she entered the house.
(7.54) a. Jim believed all Tom’s stories and Sue believed Jim’s stories.
b. Jim believed all Tom’s stories were literally true and Sue believed Jim’s
stories were fictitious.
(7.55) a. Jim believed all Tom’s stories and Sue believed Jim’s stories were fictitious.
b. Jim believed all Tom’s stories were literally true and Sue believed Jim’s
stories.
In (7.53a), both conjuncts have a [D A N] construction, whereas in (7.53b),

the first conjunct has a [D N] construction and the second conjunct has a [D A N]
construction. We can see that in (7.53a), the two conjuncts are identical in their
structure, and thus the sentence is read faster. In (7.53b), the two conjuncts are
not identical in their structure, and thus the sentence is read slower. Similarly,
in (7.54a) the object of each conjunct is a DP. The two conjuncts thus have
similar structures. In (7.54b), the object of each conjunct is a clause. The two
conjuncts thus also have similar structures. In contrast, in (7.55a) and (7.55b),
one of the two conjuncts has a nominal complement and the other conjunct has
a clause complement. Since these data are minimal pairs, the contrast in their
reading speed is directly related to the contrast in the degree of parallelism in

the conjuncts.
Parallelism makes processing easier and more efficient. However, the forms
that show more parallelism and those that show less parallelism are both syn-
tactically well-formed.
A similar experiment is reported in Luka and Barsalou (2005). It is observed
that grammaticality ratings are increased for sentences that share representa-
tional structures with those read earlier. Again, the more parallel, the easier to
process. This is covered by the resemblance condition of the RPR.
7.4.3 The more parallel in dependency chains, the easier to process

A final example showing the role of the RPR in processing is the following. If
each clausal conjunct contains a dependency chain, coordinate complexes with
non-parallel gaps are not acceptable.
(7.56) a. I know a man who [Bill likes and [Mary hates ].

b. ∗ I know a man who [Bill saw ] and [ likes Mary]. (Williams 1978: 34)
In (7.56a), the gaps in both conjuncts are object gaps, and thus they are
parallel. In the unacceptable (7.56b), however, the gap in the first conjunct is an
object gap, whereas the gap in the second conjunct is a subject gap. The gaps
in the conjuncts are not parallel and since they also show no semantic relation,
the sentence is unacceptable.
However, the following examples, which also have non-parallel gaps, are
fine, because they are saved by having semantically related conjuncts:
(7.57) a. This is the dress which Mary bought and cost $6,000. (Goodall
1987: 72)4
b. That’s the candidate who the unions endorsed and was the
overwhelming favorite of the Democrats. (Goodall 1987: 75)
Anderson (1983) reports from her experimental study that non-adjacent gaps
are more acceptable than adjacent ones. In (7.56b), (7.57a), and (7.57b), for
instance, the gaps are adjacent; in (7.56a), however, the gaps are not adjacent.
Generally speaking, the degree of route parallelism is lower when the two
gaps are adjacent than when the two gaps are not adjacent. The fact that the
two conjuncts in (7.57a) in (7.57b) are semantically related may reduce the
processing difficulty caused by the non-parallel dependency chains.
4 Zoerner (1995: 82) points out that (i) is natural only with heavy phonological stress on the
conjunction.
(i) Mary wore a dress that Ungaro designed and cost a fortune.
7.5 The nature of the RPR

7.5.1 The RPR is a filter on representations of syntactic complexes
Based on the relativized nature of the RPR, in this section we first examine
what the RPR cannot be. Then we will claim that the RPR is a filter on the
syntactic and semantic representations of syntactic complexes.
It has been generally recognized that all five of the constraints discussed in
Section 7.2.1 are related to the PR. In our approach, only the morphological
component of the CC, which is lexical-item-specific, is captured by mechanisms
other than the RPR (see Chapter 4). However, the RPR is responsible for all the
rest of the CSC, as well as all of the other constraints proposed for coordination.
A. Against a pure syntactic approach to the RPR

The RPR is not a syntactic operation In order to explain the pro-
cessing contrast between (7.53a) and (7.53b), Frazier and Clifton (2001) claim
that an operation of Copy α is present in (7.53a) but absent in (7.53b), and this
copy operation is cheaper than the regular step by step operation. If the PR is a
Copy α operation, what is copied is a syntactic category in (7.52a), a structure
in (7.53a), and a semantic feature in (7.31b).
However, the relativized nature of the RPR suggests that the PR is a global
evaluation: the RPR compares two conjuncts that have already been built. If
the two conjuncts are of semantically different types, they must be semanti-
cally related; if the two conjuncts are semantically related, they are allowed
to have non-parallel representations. If neither the relatedness nor the resem-
blance condition is satisfied, the coordinate complex is not acceptable. Such an
evaluation cannot be implemented at a point in the derivation where only one
conjunct has been built. Thus after we have built the internal conjunct, we can-
not know whether the proposed Copy α is operative or not. In other words, the
proposed Copy α is not motivated by any local formal requirement, unlike well-
recognized operations like Merge and Move. I therefore do not think that the PR
is a syntactic operation. Frazier and Clifton state in their footnote 11: “In fact
we suspect that Copy α is just the linguistic reflex of a more general cognitive
ability that we dub ‘ditto.’ ” It is possible that what they revealed in their study
is a PR filter, which reflects a general law of economy in language processing.
The RPR is not a trigger of syntactic operations Can the RPR be

a feature built inside coordinators to trigger syntactic operations? On the one
hand, Hornstein and Nunes (2002) claim that the PR is an interface require-
ment (p. 28) or a bare output condition (p. 37), while on the other hand, they
emphasize that “the Parallelism Requirement demands that movement apply

to all the conjuncts if it applies to any (Ross’s [1967] Coordinate Structure
Constraint)” (p. 37). They make the following claim (p. 41):
[A]fter a coordinating head merges with a given constituent X, it signals that

the computational system should proceed to build a constituent Y parallel
to X, with the lexical items available at the relevant derivational step. If the
available lexical material does not yield a (semantically) parallel structure,
then the Parallelism Requirement, locally enforced by the coordinating head
(or by the label of the syntactic object it heads), licenses the copying of
constituents of X in order to build Y.
In other words, they claim that if one of the conjuncts is syntactically legal,
parallelism alone can trigger an otherwise unlikely operation in building the
other conjunct. They further claim (p. 47) that in an ATB dependency, launching
movement from the late-merged conjunct can be driven by parallelism alone.
It is generally assumed that the properties or features of syntactic elements
trigger syntactic operations (following Collins 2002a we assume that Merge
is also triggered by checking selectional features). One might assume that
the PR is a feature of coordinators and can trigger syntactic operations, so
that the two conjuncts have similar representations (see the above claim of
Hornstein and Nunes 2002 and their claim [p. 40] that “the Copy operation must
be licensed either by Last Resort or the Parallelism Requirement”; similarly,
Nunes 2001: 336 claims that “the Parallelism Requirement functions as an
enabling condition that authorizes movement operations that otherwise would
not be licensed by Last Resort”). The observed relativity of the PR, however,
falsifies this assumption. The global nature of the RPR means that the PR cannot
structurally manipulate the building of the external conjunct. After we merge the
internal conjunct with a coordinator like and, there is no way for the coordinator
to signal whether or not the computational system should proceed to build
the external conjunct parallel to the internal conjunct already available in the
structure. I thus do not think that after the Merger of one conjunct, parallelism
alone can trigger an otherwise unlikely operation in building the other conjunct.
A more plausible possibility is that the two conjuncts and the whole coordinate
complex are built without any PR guidance, but the acceptability of the complex
is evaluated by the RPR, which is a filter on language processing.5
5 While some people have assumed the PR to be a trigger of syntactic operations, others have
posited an anti-PR as a trigger of syntactic operations. Thus Moro (1997; 2000) claims that
syntactic operations may be triggered in order to change a symmetrical relation into an asym-
metrical one.
The PR is not a constraint on syntactic operations Ross (1967) and

most advocates of the CSC, including Moltmann (1992b: 36, 41), believe that
the CSC is a constraint on movement operations. In Johnson (2002), the CSC,
assumed to reflect the PR, is taken as a constraint on structural building (Merge),
determining whether small or large conjuncts should be built. However, as we
have seen throughout this book, it has also been pointed out that the CSC is not
a constraint on movement (Ruys 1992; Munn 1993: 62, 97; and V. Lin 2001:
366). More generally, Potts (2002: 16) states that the CSC is “not a narrowly
grammatical restriction.” Our study supports the view adopted by this latter
group of authors.
More specifically, the arguments given above against the trigger status of
the PR can be used to argue against the syntactic constraint status of the PR,
since the RPR has no effect on the building of conjuncts. Instead, the evidence
suggests that the RPR evaluates representations rather than restricting syntactic
operations.
The CSC and CLC should not be used to make any claims about
syntactic structures Since the CSC and CLC are not syntactic constraints,
we cannot use them to make any claims about syntactic structures. In par-
ticular, we need to clarify two issues regarding the EC and the CLC,
respectively.
First, the EC part of the CSC has been one of the major arguments used
to support the VP-Internal Subject Hypothesis (Burton and Grimshaw 1992;
McNally 1992). Fortunately, the EC is not the sole argument for the hypothesis,
so giving up the EC need not have any effect on it.
Second, taking the CLC for granted, Bowers (1993; 2001) uses data like
(7.58a) to argue for the existence of PrP, which is a covering category for all
categories that can encode a predication relation.
(7.58) a. Jermaine is boring and a fool. [AP & NP] (= (7.5))

b. I want to emphasize {this point/this} and also that you should never
forget what your father told you. (Dik 1968: 28) [DP & CP]
c. You can depend on my assistance and that he will be on time.
d. [The death of his wife and that he is in prison now] worry me.
In this proposed PrP, the Spec of Pr is assumed to be a subject position,

and the complement is a predicate position (an earlier proposal of the category
PRED is seen in Jacobson 1987). However, if we follow this line of reasoning,
examples like (7.58b), (7.58c), and (7.58d) might be used to argue for V-ObjP,
Prep-ObjP, and SubjP, respectively. Although Subject and Object have indeed
been analyzed as syntactic features in Jackendoff (1977: 32) and Borsley (1983),
they are not regarded as syntactic categories independent of other categories
such as DP and CP. In fact, what Bowers calls PrP has been covered by VP
(or other extended projections, or den Dikken’s 2006 RelatorP or RP) (see
Munn 1993: 117 for arguments against the syntactic feature Predicate), and the
evidence for the existence of VP is independent of the CLC.
B. The RPR is a filter on representations of syntactic complexes

Since the PR simply evaluates the syntactic and semantic representations of
complexes, we can regard it as a representational filter, rather than a syntactic
operation, a trigger of syntactic operations, or a constraint on syntactic oper-
ations. One difference between a filter and the other choices is the timing of
the application: the former applies after the syntactic derivation is completed,
whereas the latter applies before or during syntactic operations.
7.5.2 The general economy motivation of the RPR

So far, what we have shown is that if the subcomponents of a syntactic complex
deviate from parallelism in their semantic types and movement links, without
any semantic relation to integrate them, the complex is too difficult to process,
and thus is rejected as unacceptable. The RPR is thus the result of a general
economy principle of processing.
Unlike this general principle, the five coordinate construction-specific con-
straints, the CLC, CLCsem, CLCfunc, CCC, and CSC, are ad hoc. Given the
general operations of Merge and Move, unlike categories can conjoin, and
conjuncts and their internal elements can move, as direct consequences of the
complementation structure of coordinate complexes argued for in Chapter 2.
There is no special categorial constraint on a complementation structure; exter-
nal and internal conjuncts are simply in a Spec-Complement relation. There
are also no special restrictions on movement for complementation structures;
specifiers and complements may move, and elements may be extracted from
them. As correctly pointed out by Anandan (1993: 33), “As there are no rules
for particular constructions such as interrogative, relative, passive and so on,
there are no rules for coordinate constructions also.”
I conclude that both parts of the RPR, Relatedness and Resemblance, reflect
economy in processing.
7.6 Chapter summary and conclusions for Part III

Empirically, I have shown that on the one hand, each of the five coordinate
construction-specific constraints proposed in the literature, namely the CLC,
CLCsem, CLCfunc, CCC, and CSC, can be violated. On the other hand, the
observed effects of these constraints can be accounted for in terms of two inde-
pendently motivated mechanisms: morphological constraints on the behavior
of (certain) conjunctions, and the RPR, which is a filter on linguistic represen-
tations in processing. The morphological constraints were addressed in Chapter
4, where I showed, first, that coordinators, like some other types of head ele-
ments, do not allow null complements, so if an internal conjunct moves, the
surface representation is not acceptable (accounting for the CCi), and second,
that if the coordinator does not have categorial features, it will take them from
the external conjunct, which will then be unable to move (accounting for the
CCe). As for the RPR, it is repeated as follows:
(7.60) Conjuncts of a coordinate complex must hold a coherence relation in terms of:
a. Relatedness: they must be either related to each other semantically, as in
natural coordination; or
b. Resemblance (Parallelism Requirement): they must hold a resemblance
relation in terms of both their semantic type and their dependency chains.
A theoretical consequence of my arguments is that we must remove these five

coordinate construction-specific constraints from the syntactic computational
system, since they are both ad hoc and empirically falsified. Moreover, they
should not be used as diagnostics when testing syntactic hypotheses of any
kind.
We now reach the third major conclusion of this book:
There is no special syntactic constraint on the deriva-

tions of coordinate complexes .
Before I end my discussion of the CSC, it seems fair to consider the ques-
tions of why the CSC was developed in the first place, and why it has been
so influential in syntactic research for more than forty years. As an answer,
recall the following facts. First, the EC part of the CSC is seen in accidental
rather than natural coordination. Second, the CCe part of the CSC is seen in
two situations: in accidental coordination and in coordinate complexes where
overt coordinators do not have any categorial features, as is typically found in
European languages. Third, the CCi part of the CSC follows from the fact that
coordinators are like various other head elements in taking their complement
7.6 Chapter summary and conclusions for Part III 203
as phonological host. Thus each piece of the putative CSC is due to a com-
pletely different factor. The CSC only looks like a unified construction-specific
constraint due to the coincidental confluence of these distinct factors in the
coordinate complex, a primitive notion in traditional grammar. However, when
one takes the broader view demanded by contemporary syntactic theory – con-
sidering other head elements, the semantic relations between conjuncts, and
coordinators in a wider range of languages – one sees the CSC melting away.
IV No special syntactic operation
8 The derivation of coordinate
clauses with identity adjectives
8.1 Introduction
Across-the-Board (ATB) constructions are coordinate constructions in which
each conjunct contains a gap, like the following example.
(8.1) Whoi did you say that Carrie likes i and Sarah hates i ?
The goal of this chapter and the next is to propose a new analysis of the
syntactic derivation of ATB constructions. It is generally assumed that the
constructions are derived by a kind of operation called ATB movement, which
“move[s] a constituent out of all the conjuncts of a coordinate structure” (Ross
1967: 107; see also Williams 1977; 1978). In other words, the single extracted
element in this construction has been assumed to move simultaneously from
multiple gaps. Such movement chains are called “forking chains,” and are
assumed to occur only in coordinate constructions.
In this book, however, I advocate the position, first stated in George (1980),
that ATB movement does not exist at all (see also Franks 1992; Munn 1992,
1993; Bošković and Franks 2000; and Hornstein and Nunes 2002). Moreover,
I support Munn’s (1992; 1993) claim that the extracted element in ATB con-
structions originates in the first conjunct only, and show that there is a binding
dependency between the extracted element and a silent pro-form in the second
conjunct. Both the extraction operation and the pro-form binding dependency
are motivated independently of coordinate constructions. The proposed deriva-
tion thus does not require any ad hoc forking chains.
The claims about ATB constructions that I make in this chapter are based
on an analysis of identity adjectives, such as English same, aiming to show
the parallelism between same constructions and ATB constructions. In my
proposed analysis, (8.2a) is derived from (8.2b):
(8.2) a. The same man Mary helped and Jane ruined.
b. [Mary helped [DP1 the same man]i] and [Jane ruined pro- φPi].
207
208 Coordinate clauses with identity adjectives
To anticipate, I will argue that in (8.2b), the pro-form in the second conjunct
takes the nominal the same man as its antecedent. The category of the pro-
form is pro-φP, which shares its gender, number, and person features with its
antecedent (see Déchaine and Wiltschko 2002). The pro-φP in same construc-
tions is silent at PF. Moreover, I will argue that a similarity expression such
as the same man in (8.2a) must be licensed by a plural element, which here
is the coordinate complex. This expression must move out of the coordinate
complex, simply because similarity expressions must be outside their licensing
elements (see G. Carlson 1987: 540).
The syntax of same gives us clues about the derivation of ATB constructions.
In the next chapter, I will demonstrate the similarities between ATB construc-
tions and same constructions, and extend my analysis of the latter to the former.
Thus, ATB constructions are derived without any forking movement.
In Section 8.2, I review certain syntactic properties of same constructions
and the important syntactic questions raised by sentences like (8.2a). These
questions form the research goals for the following two sections. In Section
8.3, I argue for the existence of a null pro-form in the second conjunct in
sentences like (8.2a). In Section 8.4, I argue that a similarity expression like
the same man in (8.2a) must move from the first conjunct and land outside the
coordinate complex. Section 8.5 gives a summary of the chapter.
8.2 The identity adjective same

Adjectives such as same, similar, equal, and different express similarity rela-
tions. Among them, different expresses a negative value of similarity and same
expresses a positive and superlative value of similarity, namely identity. I call
expressions that host a similarity adjective Similarity Expressions (SE). In
(8.2a), for instance, same is an identity adjective and the same man is an SE.
8.2.1 The general plural-α licensing of identity adjectives

Some elements need licensors in their context. For instance, a negative polarity
item such as any is licensed in a negative context. The licensor can be an
independent negation word such as not (e.g. He did not eat any candy) or a
negative feature expressed in a negative word such as refuse (e.g. He refused
to eat any candy). Similarly, verbs such as combine, predicates such as collide,
and adverbs such as respectively are all licensed in a plural context. A plural
context has two or more eventuality participants or two or more eventualities
(we will see examples later). Again, the forms of the licensors can vary (a plural
nominal or a coordinate complex). I call this general plural context a plural-α

context.
Like other similarity adjectives, identity adjectives must be licensed by a
plural-α context. Semantically, this is expected from the notion of similarity,
which presupposes the existence of multiple entities or multiple parts of wholes
so that a certain kind of “implicit comparison” (Carlson 1987: 531) can be
established. I divide plural-α contexts into three basic types. In the first type,
the SE is thematically related to the multiple verbal phrases, while in the latter
two types it is not.
A. The Thematic Licensing same Construction (TLC)

In the first type of plural-α contexts, the SE seems to get its theta role from the
multiple verbal phrases. I call identity adjective constructions of this type the
Thematic Licensing same Construction (TLC).
The examples in (8.3) represent the TLC. In each example, the coordi-
nate clausal complex provides a context of plural eventualities. In (8.3a), for
instance, it is the combination of the helping eventuality and the ruining even-
tuality that licenses the SE the same man. Moreover, in these examples each
conjunct of the coordinate complex seems to assign a theta role to the SE, if
we ignore the word same. In (8.3a) for instance, helped in the first conjunct
assigns a theta role to its object, and it seems that the (same) man is this object;
similarly, ruined in the second conjunct also assigns a theta role to its object,
which also seems to be the (same) man.
(8.3) a. The same man Mary helped and Jane ruined. (= (8.2a))
b. The same man got drunk and was arrested by the cops.
c. The same man praised you and seemed to hate you.
d. The same brush, John used for cleaning the toilet and Mary used for cleaning
the kitchen.
Putting aside the role of the identity adjective (same), this type of construction
intuitively seems to be defined in such a way that the relation between the whole
SE and the verb (or verb phrase) of each conjunct is a thematic licensing relation.
The TLC has another version in which the same-phrase, the SE, occurs in
the right-peripheral position of the sentence.
(8.4) John avoided and Bill ignored the same man. (Jackendoff 1977: 192)
I do not discuss such right-edge same-constructions (see Hartmann 2000:

78; Abels 2003: 127; Barker 2007; Johnson 2007 for some discussion). I
assume that the constructions belong to or should be dealt together with RNR
constructions, in the sense of Postal (1974) and Abbott (1976). It has been rec-
ognized that there are systematic syntactic differences between left-peripheral
argument-sharing constructions (ATB constructions) and RNR constructions
(e.g. the right-edge constraint on gap positions is seen in the latter but not in
the former) (Ross 1967; Wilder 1997; Sabbagh 2007; among others). For this
reason, I do not assume that the leftward dependency and the rightward depen-
dency can be derived in a unified way, and so I leave the “right” constructions
for future research.
Since theta-role relations are syntactic relations, the relation in a TLC
between the SE and its licensor (plural-α), the coordinate verbal phrase, is
a syntactic licensing relation. By contrast, in the two types of SE constructions
introduced in the following two subsections, the relation between the SE and
its licensor (plural-α) is not syntactic.
B. The Paired Pronoun same Construction (PPC)

One construction in which the relation between the SE and the plural-α is not
thematic is the Paired Pronoun same Construction (PPC). As in a TLC, in a PPC
the SE is licensed by a context of plural eventualities. In (8.5a), for instance,
it is the combination of the coming eventuality and the leaving eventuality
that licenses the SE the same man. In a PPC, the thematic licensing relation is
satisfied by a pronoun in each conjunct, and the two pronouns both take the SE
as their antecedent. In (8.5a), for example, the thematic licensing relation of
came and left are satisfied by the first he and by the second he, respectively, and
both pronouns take the same man as their antecedent. (8.5b) and the Chinese
example in (8.5c) show the same point.
(8.5) a. The same man, he came and he left.

b. The same brush, John used it to clean the toilet and Mary used it to
clean the kitchen.
c. Tong yi ge nühair, Baoyu gei ta xie-le
same one cl girl Baoyu to her write-prf
qing-shu, Fanjin gei ta song-le jiezhi.
love-letter Fanjin to her send-prf ring
‘The same girl, Baoyu wrote love letters to her and Fanjin sent a ring to her.’
This construction can be regarded as a gapless TLC. Semantically, in a TLC,

the missing argument in each conjunct must have the same interpretation as the
referent of the SE. Similarly, in a PPC, the pronoun in each conjunct must have
the same interpretation as the referent of the SE. Neither of the two pronouns
can refer to anyone else other than the referent of the SE. The relationship
between the SE and the two gaps in a TLC is similar to the bound pronoun
binding between the nominal some sheep and the two pronouns in (8.6) (Evans
1980: 339).
(8.6) Some sheep are such that John owns them and Harry vaccinates them in
the spring.
Similarly, the relationship between the SE and the two pronouns in a PPC is
like the bound pronoun binding between some sheep and the two pronouns in
(8.6). Therefore, in this semantic sense, PPCs are similar to TLCs.
Syntactically, however, the formal features of the verb in each conjunct are
saturated by the pronoun in a PPC. No thematic or Case relation exists between
the SE and the coordinate complex, unlike the situation in a TLC. The gapless
TLC data are similar either to the construction in (8.7a), where one nominal is
the antecedent of the two resumptive pronouns, or to the construction in (8.7b),
where one nominal is the antecedent of the two relative pronouns distributed
in the two conjuncts:
(8.7) a. Na ge nühair, Baoyu gei ta xie-le qing-shu, Fanjin gei ta
that cl girl Baoyu to her write-prf love-letter Fanjin to her
song-le jiezhi
send-prf ring
‘That girl, Baoyu wrote love-letters to her and Fanjin sent a ring to her.’
b. The manuscript [[the letter on the front of which and the scribbling
on the back of which] Harry deciphered] was in Gwambamamban.
(Postal 1972: 132)
C. The Non-Thematic Licensing same Construction (Non-TLC)

Another type of construction in which the relation between the SE and the
plural-α is not thematic is the Non-Thematic Licensing same Construction
(Non-TLC), which has four subtypes.1
Discourse plural licensing Same may have an indexical reading, as

in (8.8).
(8.8) John saw the same tree.
As described in G. Carlson (1987: 532), the most natural interpretation for

sentences like (8.8) involves a covert comparison between something referred to
in the sentence (a certain tree) and something that is understood by the listener
1 The adjective same can also appear in comparatives, as in John found the same solution as Mary.
Since the syntax of comparatives is not yet clear to me, I do not discuss this construction here.
as having already been contextually defined. The speaker is considering both

the current eventuality and a previous one in which a tree was mentioned. It is
this plurality of eventualities that licenses the identity adjective.
In (8.8), the SE the same tree has a thematic relation with the selecting verb
saw. The discourse licensing of the SE does not involve any thematic relation.
Quantifier presupposition plural licensing Same may be licensed by

a universal quantifier. In (8.9a), same is licensed by everyone, and in (8.9b),
same is licensed by all the furniture.
(8.9) a. Everyone saw the same tree. (Moltmann 1997: 136)

b. All the furniture is of the same color.
As is well known, when a universal quantifier is used with an individual-

denoting nominal in an episodic context, a group of individuals is presupposed
so that we can mention every item of the group. In (8.9a), the existence of
two or more persons is implied, and in (8.9b), the existence of several pieces
of furniture is implied. It is this implied plurality that licenses the identity
adjective.
Since a quantified nominal is not a verb or predicate, it does not have any
thematic relation with the SE.
Plural licensing with restrictive relative clauses Same may appear

in restrictive relative clause constructions, as in (8.10).
(8.10) The same man {that/who} we saw yesterday came today.
The semantic function of restrictive relative clauses is to restrict the referent

in a set of elements. Thus relative clauses also presuppose the existence of plural
elements. It is this presupposed plurality that licenses the identity adjective.
In (8.10), the SE the same man has a theta relation with the matrix predicate
came, rather than with its licenser, the relative clause. Within the relative clause,
we and the relative pronoun who (or a null operator) satisfy all the thematic
checking requirements.
Clause-internal plural licensing Same may occur in a construction

where an explicit plural-α occurs. This plural-α is realized by plural mor-
phemes, coordinate complexes, or plural numerals. The examples in (8.11)
represent this type of construction. The plural elements are underlined. These
can be either a plural nominal, like the students, or a coordinate complex, like
John and Mary, as seen in (8.11a). Keep in mind that unlike in TLCs, there is
no thematic relation between the SE and the plural-α in this construction (i.e.
the plural-α does not assign a theta role to the SE). In (8.11a), for instance, the
object the same picture is an SE and the plural subject the students licenses
the SE. However, unlike the situation in a TLC, the SE gets its theta role from
the verb saw, rather than its licensor, the subject. Likewise, the subject gets its
theta role from the whole predicate saw the same picture, rather than the SE.
Since the SE and its licensor are both arguments, neither assigns any theta role
to the other, so there is no thematic relation between them.
(8.11) a. {The students/John and Mary} saw the same picture.
b. The same salesman sold me these two magazine subscriptions. (G.
Carlson 1987: 532)
c. The same child slept in the bed and on the floor.
d. John played the same sonata slow and fast. (Moltmann 1992b: 231)
Since the plural-α is not a verbal element in these examples, it does not have
any thematic relation with the SE in the sentence. Note that the distinction
between this construction and TLCs is the latter property (thematic relations),
not the former (whether the plural-α is a verbal element). In the following
example, the SE adverbial is base-generated out of a coordinate verbal complex
and is directly licensed by it. Nevertheless, these examples are Non-TLCs
because the formal features of the verbs in the conjuncts are satisfied locally,
so there is no thematic relation between the SEs and their licensors.
(8.12) Zai tong yi tian, Baoyu wancheng-le boshi lunwen, Daiyu
at same one day Baoyu finish-prf PhD thesis Daiyu
sheng-le haizi.
bore-prf child
‘On the same day, Baoyu finished a PhD thesis and Daiyu gave birth
to a child.’
8.2.2 Major questions about the syntax of TLCs

After distinguishing TLCs from other SE constructions, we are now ready to
examine the syntactic properties of TLCs. TLCs have been regarded by Jack-
endoff (1977: 192–194), Abbott (1976: 642), Gazdar (1981: 180), Gazdar et al.
(1982: 664), and Postal (1998: 137) as a challenge to generative syntax. Such
data are special in that, if the SE is “reconstructed” into each clausal conjunct,
the sentence either becomes unacceptable or does not have the intended read-
ing. For instance, if no man has been mentioned previously in the discourse,
(8.13b) is not acceptable. One cannot use this sentence to initiate a description
of two events in which a single specific man had been involved. If someone has
been previously mentioned in the discourse, (8.13b) is acceptable, but then the
two occurrences of the same man both refer to the previously mentioned man
(i.e., the reading is changed into that of the discourse plural licensing type of
Non-TLCs).
b. (∗ )Mary helped the same man and Jane ruined the same man.
Why is it that the SE in TLCs cannot surface inside the coordinate complex?
I call this the Question of the Surface Position of SEs.
This question in fact requires us to solve another basic syntactic question, the
Question of the Base Position of SEs: where are the SEs in the constructions
base-generated? This second question entails questions like the following: Why
do we interpret SEs as having the same theta role expected for the missing part
in the conjuncts? How is the formal feature licensing of the verbal element
achieved if the argument that can implement the licensing is missing in each
conjunct of a TLC? For instance, in (8.13a), both helped and ruined are tran-
sitive. The agent of the former is Mary and the agent in the latter is John.
According to the Projection Principle (Chomsky 1981; among others), in the
presence of the agent, the transitive verb in each conjunct must have an object.
The verb and the object need to have formal feature relations (selection, theta
role, Case, and so on). But what is the object in each conjunct of the sentence?
In addition to the above two basic syntactic questions, one more fact is in
need of explanation, namely why the gap position in the second conjunct of a
TLC cannot be filled with an overt pronoun, as shown in (8.14). I call this the
Question of Overt Right Link of a Dependency.
(8.14) a. The same man Mary helped and Jane ruined (∗ him).
b. The same man Mary helped and (∗ him) Jane ruined.
All of these questions will be answered in sections 8.3 and 8.4.
8.3 Building well-formed conjuncts of TLCs

This section aims to answer the Question of the Base Position of SEs.
8.3.1 The existence of a silent nominal in the second conjunct

We can see that an SE functions as an argument of the predicate or verb in each
conjunct of a TLC, although it does not occur in either conjunct. However, no
single nominal may satisfy the formal requirements of two predicates or verbs
(e.g., selection, theta role, Case). Moreover, any (sideward) movement chain
relation between the two conjuncts is unlikely, since the tail and the head links
need not have identical forms. In (8.15), for instance, if the gap position in the
8.3 Building well-formed conjuncts of TLCs 215
second conjunct were the base position of the SE the same picture of herself,
the derivation would crash, since the feminine reflexive herself would not be
licensed in the conjunct.
(8.15) The same picture of herself, Mary painted yesterday and John bought today.
A plausible response to these problems is to hypothesize that the SE is

base-generated as an argument in the first conjunct only, and there is another
nominal, which is silent, in the second conjunct.
In each conjunct of a TLC, the existence of a nominal related to the argument
gap is first of all required by the general formal feature licensing, as stated above.
Second, it can also be seen in examples like (8.16), where the reflexive in each
finite clausal conjunct must be licensed by a local antecedent. Note that the
occurrence of the auxiliary has in each conjunct indicates that the construction
is a clause, not a VP, since auxiliaries surface at T and TP is a clause. If the
reflexive of the first clausal conjunct in (8.16) is licensed by the same guy, the
reflexive of the second clausal conjunct must be licensed by a different nominal
base-generated inside the clausal conjunct.
(8.16) The same guy has constrained himself in public and has indulged himself
in his home.
8.3.2 The interpretation of the silent argument in the second conjunct

After demonstrating the existence of an argument in the gap position of the
second conjunct of a TLC, we now determine its interpretation.
First of all, the φ-features of the assumed silent argument in the second
conjunct must be identical to that of the SE. The unacceptability of (8.17)
shows that the φ-features of the silent argument are not underspecified, since
they may not be different from those of the SE.
∗
(8.17) The same guy has constrained himself in public and has indulged herself
in her home.
Secondly, this assumed argument in the second conjunct of a TLC may

not have an independent referent. For instance, the TLC in (8.18) may not be
followed by the plural pronoun they, taking as its antecedent the combination
of the SE and the silent argument.
(8.18) The same man came today and will come tomorrow. {He/∗ They} had a
cold yesterday.
Semantically, the relationship between the silent argument and the SE is

like that between an anaphor or a resumptive pronoun and its antecedent. The
former has an unvalued [ID] (identification) feature, which is valued by its
antecedent (see Adger and Ramchand 2005: 173 and Harley 2005: Section 7
for formalization of the [ID] feature).
8.3.3 The syntactic category of the silent argument in the second conjunct
I now identify the syntactic category of the assumed silent argument in the
second conjunct of a TLC. I claim that the argument is a pro-form, rather than a
copy of the SE. I have shown in (8.17) that the φ-features of the silent argument
are identical to those of the SE, i.e., those of the head of the SE only. In (8.19),
for instance, the silent argument in the second conjunct must be a pro-form.
If the conjunct contained a copy of the SE, the same picture of herself, in any
step of the derivation, the feminine reflexive herself would not be licensed in
the conjunct.
(8.19) The same picture of herself, Mary painted yesterday and John bought
today. (= (8.15))
We can further identify the anaphor-like pro-form in the second conjunct of

a TLC as a pro-φP, defined in Déchaine and Wiltschko (2002) on the basis
of a cross-linguistic survey (including English) as never having independent
referents. This kind of (silent) pro-form for the lower link of A-bar dependencies
is also argued for in Adger and Ramchand (2005).
I thus conclude that the argument in the second conjunct of a TLC is a silent
pro-form, taking the SE as its antecedent. I leave the issue of why this pro-form
is silent to Section 8.4.3.
I can now answer the Question of the Base Position of SEs. An SE is base-
generated as an argument in the first conjunct, where it satisfies local formal
requirements like theta role and Case feature checking. In the second conjunct,
in the gap position there is a silent pro-φP satisfying its own local formal
requirements. Note that this conclusion depends on the assumption that the
verb or predicate of the second conjunct assigns its relevant theta role to a
nominal distinct from that of the first conjunct.
I have now established the existence of a silent argument in the second
conjunct of a TLC, as well as its interpretation and category. The SE and this
silent argument guarantee the syntactic well-formedness of each conjunct in
the TLC.
8.4 Extraction of SEs out of their licensing coordinate complexes

So far, I have demonstrated two dependencies in a TLC, namely the binding
relation between the SE and a silent pro-φP, and the licensing of the SE by a
8.4 Extraction of SEs out of coordinate complexes 217
plural-α. Each of these dependencies is independently motivated. Reflexives

encode the former dependency only, whereas relational nominals encode the
latter dependency only. Relational nominals such as combination, mixture,
marriage, comrade, or friend express relations between multiple individuals
and thus require a plural-α licensor, but they do not encode any binding relation.
The SEs that host the word same in TLCs are special in that they encode both
dependencies. For the binding relation, the SEs satisfy the formal conditions
of the pro-φP, whereas for the licensing relations, their own formal conditions
are satisfied by a plural context.
I have discussed the licensing of SEs of TLCs via the coordination of two
clausal conjuncts in Section 8.2.1A, and the binding relation in TLCs in Section
8.3. In this section, I show that it is the plural-α SE licensing relation that forces
the SE to move out of the coordinate complex. This obligatory raising also
correlates with the obligatory silence of the pro-form in the second conjunct.
We will thus be able to answer both the Question of the Surface Position of
SEs and the Question of Overt Right Link of a Dependency, raised in Section
8.2.2.
8.4.1 The extraction of SEs out of first conjuncts

I argued in Section 8.3 that the SE of a TLC is base-generated in the first conjunct
of the construction. Recall that the SE of a TLC cannot be “reconstructed” back
into both conjuncts (see (8.13)). I now argue that the association between the
SE and the gap in the first conjunct of a TLC is a movement chain. In other
words, the left-edge position of the SE of a TLC is derived by movement. My
first argument for the movement analysis comes from island effects.
(8.20) a. ∗ The same person, Bill lost business because he hired and Mary praised
a lot.
b. ∗ The same person, Bill lost business after he hired and Mary praised a lot.
c. ∗ The same student, the teachers who often praise are functionalists, and the
formalist teachers often criticize.
d. ∗ The same nurse, John is still wondering whether he should date and
Jack has dated.
In the above unacceptable examples, the SEs originate in islands. In (8.20a),

the SE the same person is related to the object of the verb hired, which is
inside a because-adverbial clause. In (8.20b), the SE the same person is related
to the object of the verb hired, which is inside an after-adverbial clause. The
impossibility of these dependencies can be covered by the adjunct island effect.
In (8.20c), the SE the same student is related to the object of the verb praise
inside the relative clause of the first conjunct, so the impossibility of this
dependency can be covered by the complex definite DP island effect. Finally,

in (8.20d), the SE the same nurse is related to the object of the verb date, which
is inside a question initiated with the word whether, so the impossibility of this
dependency can be covered by the wh-island effect.
I argued in Section 8.3 that the gap in the second conjunct of a TLC is a
silent pro-form, which is not formed by movement. Hence we do not expect it
to show any island effects, and this is indeed the case. Setting aside the marked
nature of topicalization in English, my English-speaking informants confirm
the relative acceptability of (8.21a,b), in contrast to (8.20a,d), respectively.
(8.21) a. The same person, Bill praised a lot and Mary lost business because
she hired.
b. The same nurse, John has dated and Jack is still wondering whether he
should date.
The unacceptable TLC examples in (8.20) also stand in contrast to the fol-
lowing acceptable PPC examples. Since there is no movement chain in deriving
the surface positions of the SE and the pronouns in a PPC, the acceptability of
the sentences is expected.
(8.22) a. The same person, Bill lost business because he hired him and Mary
praised him a lot.
b. The same person, Bill lost business after he hired him and Mary praised
him a lot.
c. The same student, the teachers who often praise him are functionalists,
and the formalist teachers often criticize him.
d. The same nurse, John is still wondering whether he should date her and
Jack has dated her.
My second argument for the movement approach is that an SE may contain

a reflexive bound by the subject of the first conjunct, as shown in (8.23a), but
may not contain a reflexive bound by the subject of the second conjunct, as
shown in (8.23b).
(8.23) a The same picture of herself, Mary painted yesterday and John bought
today. (= (8.19))
b. ∗ The same picture of himself, Mary painted and John bought.
The reconstruction effect of the reflexive indicates that movement has occurred
in the first conjunct.
8.4.2 Carlson’s constraint and the motivation for SE extraction

In this subsection I explain the motivation for the observed movement of SEs
in TLCs. For instance, what drives the SE the same man in (8.24a) to move out
of the coordinate complex?
8.4 Extraction of SEs out of coordinate complexes 219
b. (∗ )Mary helped the same man and Jane ruined the same man. (= (8.13b))
Recall that there are two dependencies in TLCs: the binding of the silent
pro-φP by the SE and the licensing of the SE by a plural-α. It is the latter
dependency that requires the extraction of SE out of the coordinate complex.
The requirement comes from a constraint noted by G. Carlson (1987: 540),
whereby the similarity adjectives same and different cannot surface inside their
licensing complexes. Examples like the following are unacceptable as TLCs
(they are only acceptable as the first type of Non-TLCs, with indexical SEs
licensed by the discourse context).
(8.25) a. ∗ John [spilled his milk and poached the same egg].
b. ∗ Brer Rabbit ran [into the briar patch and away from different enemies].
Carlson’s constraint is in fact not restricted to same and different construc-

tions. Consider the following:
(8.26) a. the meeting of {John and Mary/the boys/∗ the boy}

b. ∗ John and the meeting of Mary
(8.27) a. the mixture of the wine and oil

b. ∗ the wine and the mixture of oil
Collective expressions such as meeting, mixture, and combination are

licensed by non-singular arguments. In (8.26a), for instance, the word meeting
requires its argument to be plural. Either the coordinate complex John and Mary,
or the plural nominal the boys satisfies this requirement. The singular countable
DP the boy, however, does not. (8.26a) shows that collective expressions can
also be licensed by a coordinate complex. Why then are the coordinate contexts
in (8.26b) and (8.27b) unable to license the collective expressions? Carlson’s
constraint can provide the answer: such expressions cannot occur inside their
licensing coordinate complexes. For the same reason, the extraction of SEs out
of coordinate complexes is obligatory in TLCs.
I claim that SEs and collective expressions can be licensed in a derivational
way. After an SE or a collective expression moves out of a coordinate complex,
the complex is able to license the SE or the collective expression. In particular,
in TLCs, SEs are licensed by the coordinate verbal complexes that they move
out of. In (8.24a), for instance, the SE the same man is licensed after it moves
out of the coordinate complex.
If an SE or collective expression cannot be licensed at any step of the
derivation, the derivation will crash. (8.24b), (8.25), (8.26b) and (8.27b) all
illustrate this.
All of the examples that Carlson discusses are the fourth type of Non-TLCs.
In this type, an SE and its licensor are both in the same clause but there is
no thematic relation between them. Even though the SE in a TLC does have a
thematic relation with a conjunct of the licensing coordinate complex, Carlson’s
constraint can still be used to account for why the SE must be moved out of its
licensing coordinate complex in a TLC.
One may formalize this constraint in terms of feature-checking or some other
licensing relation. I will not commit myself to any specific formalization here,
since my approach depends only on the empirical observations motivating the
constraint.
This answers the Question of the Surface Position of SEs posed in Section
8.2.2. Why must the SE in a TLC be outside the coordinate complex? My
answer is that the coordinate complex is the licensor of the SE, and no SE can
be contained in its licensor, according to G. Carlson’s (1987: 540) constraint on
SEs. By contrast, in a PPC, the SE is already outside the licensing coordinate
complex, and so it does not move.
8.4.3 The silence of the pro-form in the second conjunct of a TLC

In Section 8.3, we determined that the null argument in the second conjunct
is a pro-form, rather than a full-fledged DP. We now need to explain why this
pro-form must be silent. For instance, why does the overt pronoun he in (8.28)
make the sentence unacceptable?
∗
(8.28) The same man came and he sat down.
If the gap in each conjunct of a TLC is replaced by a pronoun, the construction

becomes a PPC (see Section 8.2.1B).
(8.29) The same man, he came and he sat down.
The constraint seen in (8.28) indicates that when an SE is related to two gap
positions (as in a TLC) or two pronouns (as in a PPC), the sentence is fine;
however, if it is simultaneously related to a gap in one conjunct and a pronoun
in another (as in (8.28)), the sentence is not acceptable. In Section 7.3.4B,
we saw that conjuncts show resemblance in the phonological realization of
the links of dependency chains. If each conjunct contains a link of a certain
dependency chain, the links in the two conjuncts must be both silent or both
realized phonologically (e.g. by a pronoun). This PR therefore answers the
Question of Overt Right Link of a Dependency, posed in Section 8.2.2.
An independent issue is how the required silence of the pro-forms in TLCs
is achieved. On the one hand, the pro-forms may be inherently null pronouns,
like one version of pro in Japanese (Tomioka 2003). On the other hand, the
silence of the pro-forms may be the result of phonological deletion of an overt
pro-φP (Déchaine and Wiltschko 2002). Potentially, different languages may
choose different strategies to achieve silence of the pro-forms in TLCs required
by the PR, but I will leave this issue to future research.
Regarding the derivation of the TLC, I conclude that syntactically, an SE
is raised from the first conjunct and a silent pro-form is present in the second
conjunct, and semantically, as I mentioned in Section 8.2.1B, the relation
between the raised SE and the gap positions is similar to bound pronoun
binding (see Evans 1980: 339).2
8.5 Chapter summary

In this chapter, I have presented a new approach to coordinate clauses with
identity adjectives constructions. I have examined the syntactic properties of
same constructions and proposed a new analysis for the syntactic derivation of
constructions of type TLC, like The same man Mary helped and Jane ruined.
In the proposed derivation, the similarity nominal that contains the adjective
same is base-generated in the first conjunct, and moves out of the coordinate
complex. The raised relational nominal binds a silent pro-form, a pro-φP,
in the second conjunct. Thus the derivation of a TLC involves two major
dependencies: a movement chain and a binding dependency, and the raised SE
c-commands both gaps. These properties will then provide a new analysis for
ATB constructions, as shown in the following chapter.
2 The issue of pro-form binding with respect to the adjective same is discussed in Moltmann
(1992b), but she does not discuss TLCs. Comparing sentences like John and Mary bought the
same book (a Non-TLC, similar to my examples in (8.11)) with sentences like John and Mary
think they love each other, she analyzes the nominal containing the word same as a bound
element similar to a reciprocal. Her analysis thus covers certain types of Non-TLCs, which are
not the focus of this chapter.
9 Forming Across-the-Board
constructions without
forking movement
9.1 Introduction
Across-the-Board (ATB) constructions are coordinate constructions in which
each conjunct contains a gap, like the following wh-question in (9.1a) and
relative clause construction in (9.1b):
(9.1) a. Whoi did you say that Lulu likes i and Tubby hates i ?
b. I saw the person whoi Lulu likes i and Tubby hates i .
ATB constructions have been claimed to be universal (Goodall 1987: 77).

In this chapter, I present a new analysis of the constructions. In the previous
chapter, I discussed the syntax of the identity adjective same. In my proposed
analysis, (9.2a) is derived from (9.2b) by the movement of the SE the same
man from the external conjunct:
b. [[Mary helped [DP1 the same man]i] and [Jane ruined pro- φPi]].
In this chapter, I will demonstrate the properties shared by same constructions

and ATB constructions, and propose a syntactic derivation for ATB construc-
tions similar to (9.2b). This proposed derivation does not require any ad hoc
forking chains of movement.
In Section 9.2, I apply my analysis of same constructions to ATB construc-
tions. In Section 9.3, I address the issue of the respectively reading of ATB
constructions. In Section 9.4, I compare my new analysis of ATB constructions
with other analyses. Section 9.5 gives a summary of the chapter and Part IV of
the book.
9.2 ATB constructions as TLCs

So-called ATB constructions are in fact identity constructions, although there
is no explicit identity adjective. I claim that all argumental ATB constructions
222
are derived in the same way as proposed for TLCs in Chapter 8. This means
that the apparently shared arguments in ATB constructions are SEs with an
implicit morpheme meaning “same.” My derivation of (9.3a) is illustrated in
(9.3b).1
(9.3) a. Which picture of himself did Tom paint and Mary buy?
b. [[DP2 which 0 <same> picture of himself ] did Tom paint tDP2 and pro- φPi did Mary
buy t i ]
In (9.3b), the DP2 which ∅<same> picture of himself, which is base-generated

in the first conjunct, moves out of the coordinate complex, and binds a wh
pro-φP in the second conjunct.
In this section, I present arguments for this TLC approach to ATB construc-
tions. This approach is motivated by three observations about ATB construc-
tions, which will be presented in the three subsections of this section. First,
such constructions always have an identity reading (Section 9.2.1); second,
there is a silent argument in the gap position of the second conjunct (Section
9.2.2); and third, there is a correspondence between wh-extraction and iden-
tity reading in wh-in situ languages (Section 9.2.3). These observations, along
with Munn’s (1992; 1993) observation that movement in ATB constructions
launches from the first conjunct only, lead me to the new analysis of ATB
construction illustrated in (9.3b).
9.2.1 The identity readings of ATB constructions

The extension of the analysis of TLCs to ATB constructions is first of all
motivated by a well-recognized observation: ATB constructions are always
permitted to have an identity reading. As described by Moltmann (1992b: 126,
132), (9.4a) implies that the person that John married is also the person that
Bill proposed to, and (9.4b) can only concern a single man who both walked
down the street and was killed.
(9.4) a. Which woman did John marry and Bill propose to?
b. A man walked down the street and was killed.
The identity reading of ATB constructions is confirmed by the parallel seman-

tic anomalousness of (9.5a) (Heycock and Zamparelli 2000: 351) and (9.5b):
(9.5) a. # Tell me which documents John wrote today and Mary filed yesterday.
b. #The same documents, John wrote today and Mary filed yesterday.
1 In Section 9.3.2, I will claim that adjunct ATB constructions in English are not TLCs, but
Non-TLCs.
224 Forming ATB constructions without forking movement
The oddness of both (9.5a) and (9.5b) is accounted for by the common sense
that the same documents cannot be written today after being filed yesterday.
The left-peripheral nominal in ATB constructions can also be kind-denoting:
(9.6) Xiao mao, Baoyu xihuan, Daiyu taoyan.

small cat Baoyu like Daiyu dislike
‘Small cats, Baoyu likes and Daiyu dislikes.’
The Chinese sentence in (9.6) does not allow the reading that Baoyu likes
small cat A, while Daiyu dislikes small cat B. Instead, it means that Baoyu
likes all small cats and Daiyu dislikes all small cats.
Moreover, if the extracted argument of an ATB construction is definite,
adding (the) same to the argument does not change the meaning of the con-
struction. The two sentences of each pair below are synonymous.
(9.7) a. John robbed the bank and ran for president.

b. The same John robbed the bank and ran for president.
(9.8) a. The man robbed the bank and ran for president.
b. The same man robbed the bank and ran for president.
(9.7a) and (9.8a) are ATB constructions, whereas (9.7b) and (9.8b) are TLCs.
It seems plausible to posit that the ATB constructions are TLCs with an implicit
identity adjective.2
Identity readings are available for all types of ATB constructions. However,
Munn (1999) claims that some ATB constructions may also have respectively
readings, a possibility I will address in Section 9.3.
2 One difference between this assumed silent identity adjective and the overt form same is that the
latter generally needs to occur with the article the, a fact emphasized by Lasersohn (2000: 86).
Same may occur neither with the determiner a, nor in a determinerless nominal, as shown in (ia)
and (iia). This is different from the adjective different, which can occur with the determiner a,
as in (ib), and in a determinerless nominal, as in (iib).
(i) a. Ada and I have {the/∗ a} same hobby.
b. She is wearing a different dress every time I see her.
(ii) a. Meet friends with {the/∗ ∅} same hobbies.
b. All the men are from different towns.
Similarly, the German gleich ‘same’ must occur with a definite determiner. It cannot occur
with an indefinite determiner, nor in a determinerless nominal.
The silent identity adjective probably does not have this morphological property. For example,
the wh-phrases in (9.4a) and the indefinite nominals a man in (9.4b) are clearly not compatible
with the. Thus in examples like (9.4), the silence of the assumed identity adjective must be an
inherent property of this morpheme, rather than the result of phonological deletion of the word
same.
One intriguing case that needs to be clarified is represented by examples like

(9.9) (such data are reported in Moltmann 1992b: 137; Fox 2000).
(9.9) a. I would like to know how many books every student liked and every
professor disliked.
b. A guard is standing in front of every church and sitting at the side of every
mosque.
Among the several readings of (9.9a), we care about the one in which how
many books takes wide scope over every and an appropriate answer has to spec-
ify the quantity of books x such that every student liked x and every professor
disliked x (see Moltmann 1992a: 137 for a discussion of the other possible
readings). It is in this sense that the sentence is an ATB construction, where
the wh-phrase takes scope over the conjunction. Here, the identity reading is
associated with the quantity x.
In (9.9b), it is the quantity and property reading of a guard that is “shared”
by the two conjuncts. Our world knowledge rules out the individual reading of
a guard, since the same person cannot appear in different locations at the same
time. The sentence makes sense only in the quantity and property reading (see
Li 1998 for the claim that quantity-denoting indefinite nominals are NumPs
rather than DPs). In this kind of ATB construction, then, the identity reading is
associated with quantity and property.
9.2.2 The syntactic reality of a silent argument in the second conjunct

The extension of the analysis of TLCs to ATB constructions is also motivated by
the following observations: the left-peripheral extracted element is not moved
from the second conjunct, and there is a silent argument nominal related to the
gap position of the second conjunct.
Extracted reflexives in ATB constructions cannot be reconstructed into the
internal conjunct either semantically or syntactically. Citing Haı̈k (1985), Molt-
mann (1992a: 126) claims that (9.10) could mean “John likes himself and Bill
hates himself” if the reflexive were reconstructed into the second conjunct, but
most speakers can interpret (9.10) only as “John likes himself and Bill hates
John.”
(9.10) Himself, John likes and Bill hates .
The following examples of Munn (1992: 10) illustrate the same generaliza-
tion:
(9.11) a. Which picture of {himself/∗ herself} did John buy and Mary paint?
b. Which pictures of himselfi/∗ j did Johni buy and Billj paint?
In (9.11a), the extracted element, which picture of himself, cannot originate

in the second conjunct. If the derivation of this sentence contained the repre-
sentation of (9.12) at any step, it should have crashed immediately after this
step, thus making the assumed ATB movement impossible.
∗
(9.12) Mary painted which picture of himself
We conclude that the wh-phrase in (9.11) can only be base-generated in the

first conjunct, where the anaphor himself is licensed by the c-commanding and
local John.
Moreover, there are island effects for the gap in the first but not the second
conjunct of an ATB construction. In (9.13a), the gap in the first conjunct is in
an adjunct island, and the sentence is not acceptable, whereas in (9.13b), the
gap in the second conjunct is in the same adjunct island, but the sentence is
acceptable.
(9.13) a. ∗ Who did Bill lose business because he hired and Mary praise a lot?
b. Who did Bill praise a lot and Mary lose business because she hired?
Furthermore, if elements were extracted from two conjuncts at the same time,
one would expect it to be possible to extract a distinct wh-element from each
conjunct, at least in multiple wh-fronting languages. However, this is never the
case:
(9.14) a. ∗ Kogai staj on [vidi i ] i [jede j ]?
whom what he sees and eats
‘Whom what does he see and eat?’ (Russian, Kasai 2004: 169)
b. ∗ Kogoi kogoj Jan lubi i a Maria kocha j ?
whom whom Jan likes and Maria loves
‘Whom does Jan like and Maria love?’ (Polish, Citko 2003: (7))
In (9.14a), for instance, koga ‘whom’ is extracted from the first conjunct
and sta ‘what’ is extracted from the second conjunct. Though Russian per-
mits multiple wh-fronting in general, these multiple extraction operations are
impossible. In such operations, each wh-element is supposed to move to the
same C-domain in one fell swoop (cf. the derivation of IDCs).
Now I turn to the claim that there is a silent argument nominal in the gap
position of the second conjunct of ATB constructions. This claim is supported
by the following observations.
First, in examples like (9.15a), the predicate of the second conjunct requires
an external argument to check the relevant theta-features, and a silent argu-
ment nominal will satisfy this requirement. Entertaining requires an external
argument, a requirement satisfied by the assumed silent pro-φP. Second, in

examples like (9.15b), the reflexive himself in the second clausal conjunct
needs a local licensor in the tensed clause, and a silent argument nominal can
serve as this licensor.
(9.15) a. Who brought his guitar here and is entertaining himself now?
b. Maryi has invited Johnj and will ask about {herselfi /∗ himselfj }.
Cross-linguistically, the following Icelandic examples (Rögnvaldsson 1982;

1993: (17)) also suggest that the missing argument in the second conjunct exists
syntactically:
(9.16) a. Margir stúdentar náðu prófinu og var hrósað fyrirþað. [Icelandic]

many students passed test-the and were praised for it
‘Many students passed the test and were praised for it.’
b. Margir stúdentar náðu prófinu og þeim var hrósað fyrirþað.
many students passed test-the and they.d were praised for it
‘Many students passed the test and they were praised for it.’
The syntactic contrast between (9.16a) and (9.16b) is that the former has
no overt subject in the second conjunct, whereas the latter has the dative
subject þeim ‘they.’ Rögnvaldsson describes the above examples (= his (17)) as
follows.
Note that the argument which is missing from the second conjunct in (17a)
should have dative case, as shown in (17b) . . . [T]his means that the second
conjunct must have a separate subject position, even when the subject is
not phonologically realized, as in (17a). This is shown by the fact that the
verb and the participle in (17a) have disagreeing forms, var and hrósað,
instead of the forms voru and hrósaðir, which would be expected if the overt
nominative subject Margir stúdentar were the only available subject at all
levels of derivation.3
I thus claim that parallel to TLCs, in ATB constructions the second conjunct
has a silent argument, which is a silent pro-φP that takes the extracted DP as its
antecedent. The extracted DP is base-generated in the first conjunct, e.g. who
in (9.15a), and Mary in (9.15b).
3 In addition to demonstrating the syntactic reality of the null argument in the second conjunct
in ATB constructions, the Icelandic data in (9.16) also provide morphological evidence for the
claim that the conjuncts in sentences like (9.15b) cannot be intermediate projections. Instead,
each conjunct must be a full-fledged TP, with a subject in SpecT. This implies that there is no
conjunct of intermediate projection, consistent with the arguments in Chapter 2 (see Borsley
2005 for discussion of this issue).
In the literature on ATB constructions, it has been proposed by some that

there is a pro in the gap position of the second conjunct (see McNally 1992: 337
for a review). The main argument given by others against the pro approach is the
following: unlike the more familiar sort of pro, which can have an independent
referent, the gap in the second conjunct of ATB constructions cannot have a
referent independent of the extracted nominal. Godard (1989) and McNally
(1992: 338) correctly point out that filling the gap in an ATB construction with
an overt pronoun changes the meaning:
(9.17) a. Few politicians behave morally and are rewarded for doing so.
b. Few politicians behave morally and they are rewarded for doing so.
Note that (9.17a) means that there are few politicians who both behave
morally and are rewarded for doing so, but it is possible that there are also some
politicians who behave morally but are not rewarded for doing so. The implicit
theme of the second conjunct is bound by the quantifier phrase few politicians.
On the other hand, (9.17b) means that few politicians behave morally and all of
them are rewarded for doing so. In this sentence, the quantificational nominal is
not able to bind the pronominal subject of the second conjunct. In (9.17a), the
quantificational DP takes scope over both conjuncts, whereas in (9.17b), the
quantificational DP scopes over the first conjunct only (see Evans 1980: 339;
a similar contrast was first noted by Partee 1970). The pro analysis wrongly
predicts that (9.17a) will have a reading identical to (9.17b).
The contrast between (9.17a) and (9.17b) is accounted for in my TLC
approach to ATB constructions. I claim that (9.17a) patterns with a TLC in
the following ways. First, few politicians is moved out of the coordinate com-
plex, and that is why the quantifier scopes over the second conjunct. Second,
the silent pro-form in the second conjunct is a variable, not an independent
pro, that must be bound by the raised few politicians. This is similar to the
silent argument of a TLC identified in Section 8.3.2. Third, this pro-form is
co-indexed with the gap in the first conjunct, and thus as discussed in Section
8.4.3, its silence is not optional, unlike a pro.4 In (9.17b), by contrast, the overt
pronoun they gives this sentence a different structure from a TLC. Specifically,
in (9.17b), few politicians remains in the first conjunct, and that is why the
quantifier fails to scope over the second conjunct. If few politicians remains in
the first conjunct, the construction is not an ATB construction, and thus it does
not have to have an identity reading.
4 We can see that in addition to pro, PRO, and the lower occurrence of a movement chain, the
inventory of null elements also includes a null variable (see Tomioka 2003).
9.2.3 The correspondence between extraction and identity readings

The extension of the analysis of TLCs to ATB constructions is further motivated
by the hitherto unnoted correspondence between wh-extraction and identity
readings in wh-in situ languages.
I showed in Section 8.4 that an SE must be separated from its licensing coordi-
nate complex, regardless of whether it has a thematic relation with any conjunct
of the complex. In ATB constructions, it is recognized that the left-peripheral
element is outside of the coordinate clausal complex. In this subsection, I
present a correlation between the extraction of wh-elements in ATB construc-
tions and the presence of an identity reading. Thus parallel conjunct-internal
wh-phrases do not have an identity reading, while at the same time, the extrac-
tion of wh-phrases out of coordinate complexes is obligatory for an identity
reading.
Identical wh-expressions distributed in two conjuncts do not have an iden-
tity reading, cross-linguistically. Such constructions are either unacceptable,
as in (9.18a) (Bošković and Franks 2000: 110), or must have respectively
readings, as shown by (9.18b) and (9.18c) (Bošković and Franks 2000: 111–
112; Moltmann 1992b: 126). Exclusively respectively readings are also seen
in the Korean and Japanese examples in (9.19a) and (9.19b) (Cho and Zhou
2000).
(9.18) a. ∗ Who said that John bought what and that Peter sold what?
b. Which man said that John bought which house and that Peter sold
which house?
c. Which woman did John marry and which woman did Bill propose to?
(9.19) a. John-i enu salam-ul salangha-ko Mary-ka enu salam-ul

J-nom which person-acc like-and M-nom which person-acc
miweha-ni?
hate-q
‘Which person x, John loves x and which person y, Mary hates y?’
Not: ‘which person x, John loves x and Mary hates x’ (Korean)
b. John-ga dono hito-o aisitei-te Mary-ga dono hito-o
J.-nom which person-acc love-and M.-nom which person-acc
nikundeiru-no?
hate-q
‘Which person x, John loves x and which person y, Mary hates y?’
Not: ‘which person x, John loves x and Mary hates x’ (Japanese)
Similarly, in Chinese, in situ wh-phrases distributed in conjuncts do not have

an identity reading either, but instead have a respectively reading.
(9.20) a. Zhangsan xihuan shenme, Lisi bu xihuan shenme? (Wu 1999: 16)
Zhangsan like what Lisi not like what
‘What does Zhangsan like and what does Lisi like, respectively?’
b. Baoyu zenme qipian ni, Daiyu you zenme weixie ni?
Baoyu how cheat you Daiyu also how threaten you
‘How did Baoyu cheat you and how did Daiyu threaten you, respectively?’
c. Baoyu weishenme qipian ni, Daiyu weishenme weixie ni?
Baoyu why cheat you Daiyu why threaten you
‘Why did Baoyu cheat you and why did Daiyu threaten you, respectively?’
The ban on SE wh-phrases in conjuncts is parallel to what we see in (9.21)

(see Section 8.2.2):5
(9.21) (∗ )Mary helped the same man and Jane ruined the same man.
In contrast to the above type of data, the identity reading emerges when a
wh-phrase is moved out of the coordinate complex in wh-in situ languages (see
Hoh and Chiang 1990 for arguments for the focus-driven movement of clause-
initial wh-elements in wh-in situ languages). In this case, there is only one
overt wh-phrase. The construction is just like ATB constructions in English:
the wh-phrase is an SE, extracted from the first conjunct only.
(9.22) a. Enu salam-ul John-i salangha-ko Mary-ka

which person-acc J-nom like-and M-nom
miweha-ni? (Korean)
hate-q
‘Which person x, John loves x and Mary hates x?’
Not: ‘which person x, John loves x and which person y, Mary hates y’
b. Dono hito-o John-ga aisitei-te Mary-ga nikundeiru-no? (Japanese)
which person-acc J.-nom love-and M.-nom hate-q
‘Which person x, John loves x and Mary hates x?’
Not: ‘which person x, John loves x and which person y, Mary hates y’
5 A related fact is the following. Heim (1982: 150) observes that (ia) cannot mean that he likes
a cat and she hates the same cat. The Chinese version of the sentence in (ib) cannot have an
identity reading either. Similarly, as noted by van Oirsouw (1987: 32), the co-referential reading
of the two someones in (ic) is not preferred.
(i) a. He likes a cat and she hates a cat.
b. Baoyu xihuan yi zhi mao, Daiyu taoyan yi zhi mao.
Baoyu like one cl cat Daiyu dislike one cl cat
‘Baoyu likes a cat, and Daiyu dislikes a cat.’
c. Someone bought a box of cigars and someone bought a bottle of gin.
c. Shenme, Zhangsan xihuan Lisi bu xihuan? (Wu 1999: 17) (Chinese)

what Zhangsan like Lisi not like
‘What does Zhangsan like but Lisi does not like?’
d. Weishenme Baoyu qipian ni, Daiyu weixie ni?
why Baoyu cheat you Daiyu threaten you
‘Why did Baoyu cheat you and Daiyu threaten you?’
I conclude that wh-phrases with an implicit identity adjective, like other SEs,
must be raised out of the related coordinate complexes, cross-linguistically.
Recall that SEs may not be reconstructed into their licensing coordinate
complexes. If they did, the sentences would not be acceptable (see (9.21)
above). Now in ATB constructions, I have claimed that the extracted wh-
elements are SEs, and we have seen that if the SEs are not extracted out of the
coordinate complexes, no identity reading will be available. The correlation in
ATB constructions between the extraction of wh-elements and the presence of
an identity reading is expected under my TLC approach to ATB constructions.
I have presented three observations that led me to extend the analysis of
TLCs proposed in Chapter 8 to ATB constructions, which, I conclude, also
have an adjective (here implicit) meaning “same.”
9.2.4 The compatibility between two types of wh-expressions

In this subsection I present evidence that wh-expressions with an implicit
identity adjective and wh-expressions without an implicit identity adjective
can co-occur. Thus they are not in complementary distribution. Unlike the
latter type, the former type needs a plural-α licensor. All of these indicate that
they are different types of wh-expressions.
A. In multiple wh-fronting languages

The following data from Bošković and Franks (2000: 111) show that the wh-
element of the matrix clause and the raised wh-element from an embedded
coordinate complex can occur side by side.
(9.23) a. Koji kakvoj i kaza [ce Ivan e kupil j ] i [ce Petâr e prodal j ].
who what said that Ivan is bought and that Peter is sold
‘Which person x, which stuff y, x said that Ivan bought y and that
Peter sold y?’ (Bulgarian)
b. Koi štaj i tvrdi [da Jovan kupuje j ] i [da Petar projaje j ]?
who what asserts that John buys and that Peter sells
‘Which person x, which stuff y, x asserts that John buys y and that
Peter sells y?’ (Serbo-Croatian)
In these examples, the theta position of the first wh-word is in the matrix
clause. The second wh-word is an SE, introducing a binding dependency inside
the coordinate complex. This latter wh-word is extracted from the TLC, which
is selected by the matrix verb kaza ‘said’ in (9.23a) and tvrdi ‘asserts’ in
(9.23b).
B. In Chinese
The following Chinese examples show that the in situ wh-element of the matrix
clause can also occur with the wh-element raised from the embedded coordinate
complex.
(9.24) a. Shei zhidao shenme [Zhangsan xihuan Lisi bu xihuan]?

who know what Zhangsan like Lisi not like
‘Who knows what Zhangsan likes but Lisi does not like?’
b. Shei zhidao weishenme [zaochen taiyang zai dongbian], [xiawu
who know why morning sun at east afternoon
taiyang zai xibian]?
sun at west
‘Who knows why the sun is in the east in the morning and is in the west
in the afternoon?’
c. Ni zenme zhidao weishenme [Baoyu jie-le jiu] [Daiyu
you how know why Baoyu stop-prf alcohol Daiyu
jie-le yan]?
stop-prf smoking
‘How do you know why Baoyu stopped drinking and Daiyu stopped
smoking?’
In (9.24a), the theta position of the first wh-word shei ‘who’ is in the matrix
clause. This wh-word is not an SE. However, the second wh-word shenme
‘what’ is an SE, inquiring about an identical entity for the two states expressed
by the two conjuncts. In (9.24b), shei is also in the matrix clause, and is not an
SE, either. However, the second wh-word weishenme ‘why’ is an SE, inquiring
about an identical reason for the two states expressed by the two conjuncts.
Similarly, in (9.24c), the base-position of the first wh-word zenme ‘how’ is in
the matrix clause. This wh-word is not an SE. However, the second wh-word
weishenme ‘why’ is an SE, inquiring about an identical reason for the two
eventualities expressed by the two conjuncts.
The above observations about ATB constructions and the observations about
TLCs presented in Chapter 8 compel me to extend my analysis of TLCs to ATB
constructions. The lack of an overt adjective like same in ATB constructions
makes it superficially tempting to suppose that the extracted elements are moved
9.3 The possible respectively readings 233
from both conjuncts simultaneously. TLCs cannot be analyzed in this way,

however, and their deep syntactic and semantic parallels with ATB constructions
suggests that so-called ATB movement is merely an illusion.6
9.3 The respectively readings of certain ATB constructions

In this section, I divide ATB constructions into two types, modification con-
structions and formal-feature saturation constructions, and argue that since
the former are not TLCs, they allow both identity readings and respectively
readings.
9.3.1 Munn’s respectively readings

Munn (1999) claims that some ATB constructions in English may allow respec-
tively readings. For instance, (9.25a) can be answered by (9.25b), which pre-
sumes a respectively reading.
(9.25) a. Where did Mary vacation and Bill decide to live?

b. Mary vacationed in Paris and Bill decided to live in Toronto.
We should note at the outset that respectively readings for ATB constructions
are not available to all speakers, as observed by Moltmann (1992b: 131). In fact,
none of my English informants were able to get such readings for the examples
in Munn (1999). There may be some dialect variation here. Assuming, however,
that some speakers can get respectively readings for ATB constructions, how
do we account for them?
Gawron and Kehler (2004) point out that the respectively readings are
unavailable with unambiguously singular nominals. They show the contrast
between (9.26) and (9.27) below:
(9.26) a. In what city did Mary vacation and Bill decide to live?
b. #Mary vacationed in Paris and Bill decided to live in Toronto.
(9.27) a. In what cities did Mary vacation and Bill decide to live?
b. Mary vacationed in Paris and Bill decided to live in Toronto.
6 The putative ATB movement of head elements cannot be captured in terms of TLC, but there is
actually no compelling reason to believe that such movement exists. Consider gapping data like
(i), where the auxiliary will seems to be shared by the two conjuncts.
(i) I will buy a house and he ∅ sell his flat (∅ = will)
Although gapping has been seen as the result of head-ATB movement or VP-ATB movement
(Johnson 1996; 2009), it is however hard to exclude a null head analysis proposed by Hernández
(2007).
(9.26b) cannot be an answer to (9.26a), where the question expression what

city is singular. However, (9.27b) can be an answer to (9.27a), where the
question expression what cities is plural. Gawron and Kehler state that if a
plural reading is not excluded, an ATB construction might have a respectively
reading (they then present a semantic analysis of the readings, not relevant to
our purposes here; see Zhang 2008b for a syntactic analysis of the respectively
construction).
Munn (1999: 424; 2001: 383) reports that in examples like (9.28a) and
(9.29a), where the extracted element is a subject in the first conjunct, no
respectively reading is possible, and thus (9.28b) and (9.29b) cannot be their
answers.
(9.28) a. Which man murdered Sam and wounded Bill?

b. #Fred murdered Sam and Joe wounded Bill.
(9.29) a. Which restaurant was reviewed by Bill and criticized by Fred?

b. #His first restaurant was reviewed by Bill and his second restaurant was
criticized by Fred.
We can see that in (9.28a) and (9.29a), the extracted wh-nominal is also
singular. The obligatory identity reading is thus correctly predicted by Gawron
and Kehler (2004).
However, not all left-peripheral plural elements permit respectively readings.
In examples like (9.30), although the extracted nominal which documents is
plural, the respectively reading is still impossible.
(9.30) #Tell me which documents John wrote today and Mary filed
yesterday. (= (9.5a))
It seems that non-singularity of the left-peripheral element is not sufficient

to ensure the availability of a respectively reading for ATB constructions. I
present another condition in Section 9.3.2.
9.3.2 The availability of respectively readings in modification

constructions
In this subsection I contrast modification and formal-feature saturation con-
structions with respect to the availability of respectively readings of ATB con-
structions.
It seems that there are two cases in which respectively readings are possible.
9.3 The possible respectively readings 235
A. The left-peripheral phrase as a modifier

If the left-peripheral phrase is an adjunct, a respectively reading is possible.
The following examples are cited from Moltmann (1992b: 185, 233):
(9.31) a. In these two rooms, John died and Mary was born.
b. I can’t remember in which two rooms John died and Mary was born.
Moltmann reports that “most speakers get the reading of (81a) [= (9.31a)]
and of (81b) [= (9.31b)] in which John died in one of the two rooms and Mary
was born in the other one.” Munn (1999) also reports that respectively readings
are more likely in adverbial constructions.
B. The left-peripheral phrase as modified element

Respectively readings are also possible if the conjoined clauses are relative
clauses. The following examples are again from Moltmann (1992b: 233) (for
more examples, see Gawron and Kehler 2004 (47) and (48)).
(9.32) a. These are the masterworks that Bill painted and John composed.
b. The two masterworks that Bill painted and John drew are in this room.
c. These are the two women that Bill married and John proposed to.
The respectively readings remain available if the relative clauses are replaced
with PPs:
(9.33) a. the man and the woman with the two black dogs (Moltmann 1992b: 185)
b. the blue carpet and the red carpet in the bedroom and the living room
c. a man and a woman from two remote islands
C. Generalization
We thus see that if a plural expression is a modifier of a coordinate complex,
and if a plural expression is modified by a coordinate complex, a respectively
reading is possible. There is a contrast between modification and formal-feature
saturation constructions with respect to the availability of respectively readings
of ATB constructions.
The above data lead me to the following generalization: a respectively reading
is available in a modification relation between two plural elements. In other
words, Gawron and Kehler’s (2004) plurality condition and a modification
relation together may license a respectively reading of ATB constructions.
We now consider (9.27a) again. A respectively reading is possible because
the left-peripheral element in what cities is firstly plural and second has a
modification relation with the coordinate complex. In (9.30), however, although
which documents is plural, it has a thematic rather than modification relation

with the verb inside the coordinate complex. According to my generalization,
this sentence cannot have a respectively reading.
One account of the generalization could be that if there is no relation involv-
ing formal features (e.g. Case or theta-role) between the left peripheral element
and the coordinate complex, the construction is not a TLC. If it is not a TLC, the
two possible readings of the construction, an identity reading and a respectively
reading, can be captured as follows.
With the respectively reading, the left-peripheral element is not an SE.
Instead, the construction is interpreted as if an implicit adverb respectively
is present.
With the identity reading, the left-peripheral element may still be an SE,
which has an implicit same, as in other ATB constructions. However, in this
case, the construction is parallel to type IV of the non-TLCs (see Section
8.2.1C). The SE and the coordinate complex are base-generated independently
of each other. The relationship between the SE and the complex is just like that
between on the same day and three brothers in (9.34):
(9.34) Three brothers reached the summit of Everest on the same day.
In other words, in (9.27a), for instance, in what cities is base-generated

outside of the coordinate complex. An analysis like this has already been
proposed for adjunct ATB constructions by Woolford (1987: 168).
9.4 A comparison with other approaches

Our approach to ATB constructions in terms of null pronoun binding and SE
raising is different from all previous approaches. Munn (1993) and Zhang
(2004b), among others, argue against the forking movement chain approach,
but competing alternatives to this approach have been given in the liter-
ature, including the following: (a) Munn’s (1992) null operator approach,
where an element in the second conjunct moves locally and never keeps
company with any element in the first conjunct; (b) the multiple dimen-
sional analysis, presented by Goodall (1987), Muadz (1991), Moltmann
(1992b), Wilder (1999), Citko (2003), and Gracanin-Yuksek (2007); (c)
the deletion approach of George (1980); and (d) the sideward movement
approach, where a single element moves from one conjunct to the other,
and then out of the coordinate complex, advocated by Hornstein and Nunes
(2002).
9.4.1 The characteristics of our approach

A. No special operation in coordination
I have argued that the overt extracted element in ATB constructions has a
movement chain relation only with the gap in the external conjunct. This
means that ATB constructions do not require any special movement mode.
Thus coordinate constructions, the only contexts in which ATB patterns are
found, do not introduce any construction-specific type of movement to the
computational system.
Similar claims have formed part of other analyses of ATB constructions, and
though I will argue that my analysis is superior, I agree with their rejection of the
ad hoc notion of forking movement chains. I endorse the following statement
of Munn (2001: 370): “one nonetheless can (and in fact must) dispense with a
separate mechanism for ATB movement.”
B. Extraction from first conjuncts only

Like SEs in TLCs, the extracted elements of ATB constructions do not originate
in the two conjuncts at the same time. They are instead extracted from the
external conjuncts only. The silent pro-φP argument in the internal conjunct
never moves out of the conjunct.
C. ATB constructions are syntactically related to other constructions

(a) My analysis of ATB constructions is based on my solution to the problem
of TLCs. TLCs have been considered a serious challenge to generative syntax
since Jackendoff (1977: 192–194) (see Section 8.2.2), and as far as I know, this
book is the first effort to respond to the challenge. I have been able to show not
only that the well-established devices of syntactic theory are sufficient to derive
TLCs, but that the same devices can handle ATB constructions as well, without
the need for construction-specific operations. As a further benefit, the approach
advocated here also deepens our understanding of the syntactic and semantic
relations between these two constructions, hitherto considered independent.
(b) My analysis accounts for the raising of wh-phrases in coordinate con-
structions in wh-in situ languages, by showing that the raising is required for
the licensing of an identity reading. Thus, the approach advocated here also
deepens our understanding of the cross-linguistic syntactic properties of ATB
constructions.
9.4.2 The null operator approach

Munn (1992; 1993; see also Franks 1992; 1993) proposes a null operator
approach, also adopted by Bošković and Franks (2000). As shown in Munn’s
proposed structure in (9.35), there is no forking movement chain. Instead, a null

operator moves from the second conjunct and lands in the Specifier position of
the conjunction. The projection headed by the conjunction is a right-adjunct of
the first conjunct. The wh-word in the first conjunct alone moves to the CP.
(9.35) I wonder [ CP whoi [Jane detests ti ] [ Opi and [Harry adores ti]]]
This approach shares characteristics A and B of my analysis (see above),

but it does not have characteristic C. Thus it misses the deep parallels between
ATB constructions and TLCs. We also cannot accept Munn’s hypothesis that the
second conjunct is an adjunct of the first conjunct (see Chapter 2). Furthermore,
I have argued that the second conjunct of an ATB construction is part of a
licenser of an implicit relational word hosted in the extracted element, and is
therefore necessary for the identity reading of the construction.
9.4.3 The multiple-dimensional approach

The multiple-dimensional analysis, presented by Goodall (1987), Muadz
(1991), Moltmann (1992b), Grootveld (1994), Wilder (1999), Citko (2003),
and Gracanin-Yuksek (2007), tries to legalize the forking chain of the assumed
ATB movement. Moltmann (1992b: 121–122) claims that (9.36a) is derived
from the D-structure in (9.36b) and the S-structure in (9.36c). She claims that
the shared part, what in (9.36b), represents an implicit coordination, which,
according to my understanding, is not a syntactic coordination.
(9.36) a. What did John give to Bill and Sue show to Mary?
John give to Bill
b. SPEC(CP) IP-and NP-what (D-Structure)
Sue show to Mary
John give to Bill

c. NP-what C IP-and did (S-Structure)
Sue show to Mary
In addition to the problems with the multiple-dimensional approach noted
in Section 6.3.4, the main challenge to its application here is its assumption
that a single nominal can satisfy two sets of formal features of two verbs at the
same time (see my Section 9.2.2). Moltmann (1992a: 121) states that “In this
treatment of ATB extraction, for (241a) [= our (9.36a)] never more than one
occurrence of what has to be represented at any syntactic level. The reason is
that the NP node and hence every node it dominates are shared nodes; that is,
they are nodes that belong to both planes.” Yet it is not clear how the unique
what in the representation is related to both sets of thematic and Case relations
associated with the two clausal conjuncts. This is the same problem faced by
the earlier proposal of forking movement (also called factorization extraction)
(Ross 1967; Williams 1977; 1978).
The problem is also seen in so-called mixed-movement constructions (Molt-
mann 1992a: 124):
(9.37) a. Who laughed and seemed to be happy.

b. laughed
SpecCP-who C IP-t2
seemed t1 to be happy
In (9.37b), t2 is at SpecIP. It seems that there is only one subject position for
both laughed and seemed to be happy. The existence of t1 indicates that there is
an independent theta role bearing nominal originating in the second conjunct.
Yet it is a mystery how this nominal and the nominal that gets its theta role
from, laughed, are squeezed into the single position of t2 .
The multiple-dimensional analysis is costly not only in adding a new type of
structure (and dimension) to syntactic computation. It also invokes the notion
of implicit coordination, so-called m-planes (meaningful planes) and f-planes
(formal planes) (Moltmann 1992b: 36), and constraints such as CSR (Condi-
tion on Required Syntactic Relations in multiple-dimensional phrase markers,
Moltmann 1992b: 127). Moreover, f-planes are not always interpretable. For
instance, the sentence John compared the picture and the photograph has two f-
planes, where one is John compared the picture and the other is John compared
the photograph (Moltmann 1992b: 42–43). Neither is interpretable.
Everything else being equal, the analysis presented in this book should be
preferred because it is grounded on the well-established and simpler two-
dimensional structural configuration. Overall, the multiple dimensional anal-
ysis has none of the characteristics listed in 9.4.1. For instance, it cannot
explain the fact that extraction occurs only in the first conjunct in ATB
constructions.
The following comment of Dik (1968: 5) seems to be applicable to this
multiple-dimensional analysis:
Premature formalization . . . has certain dangers of its own, since formalized

systems tend to develop into self-contained organisms, independent of the
facts they were designed to account for. The simple fact that within such
systems a multitude of formal operations and manipulations is possible is
then easily taken as a proof of their scientific adequacy. But the only test
of the validity of a linguistic theory is whether it provides the means for a
satisfactory description and explanation of the facts of language.
9.4.4 The deletion approach

Another alternative approach is deletion of identical elements in conjuncts
(see George 1980; Hendriks 1991). However, as pointed out by Gazdar et al.
(1982: 675), the representations before the assumed deletion and the one after
the deletion do not have the same meaning. They demonstrate this with the
following two sentences.
(9.38) a. Who does everyone like and no one respect?

b. Who does everyone like and who does no one respect?
(9.38a) has an identity reading, whereas (9.38b) does not. If (9.38a) is derived
from (9.38b) by deletion of the second who, the deletion violates the recov-
erability condition on deletion (see Chomsky 1965: 144–145), which bans
deletions causing a meaning change. Therefore, the deletion approach is not
plausible here.
The contrast between (9.38a) and (9.38b) shows that ATB constructions have
an identity reading, or equivalently, such a reading is present if a wh-phrase
occurs outside the coordinate complex, and is absent if the wh-phrase occurs
conjunct-internally (see Section 9.2.3). This pattern is hard to explain in a
deletion approach.
Nevertheless, we still allow some kind of deletion in our proposed deriva-
tions of TLCs and ATB constructions, namely deletion of pro-φPs in second
conjuncts (see Section 8.4.3 for arguments relating to TLCs, which carry over
to ATB constructions). The pro-φPs that are affected by the deletion, however,
do not have the same forms with their binders (the SEs), nor does the deletion
affect the interpretations.
9.4.5 The sideward movement approach

The sideward movement approach of Hornstein and Nunes (2002; also Nunes
2001: 336–339) assumes that one element moves from one conjunct to the
other, and then moves out of the first conjunct. This approach has a major
inadequacy: it cannot cover the facts described in Section 9.2.2. For instance,
it does not account for the binding pattern in (9.39) noted by Munn (1992):
(9.39) Which picture of {himself/∗ herself} did John paint and Mary buy?
In this sentence, the merger of which picture of himself in the second conjunct,
where it is c-commanded by Mary, is illegal.
By contrast, in my approach the gap of the second conjunct in (9.39) is not a
trace, but a bound pronoun (syntactically, a pro-φP), and therefore it contains
no reflexive. This pronoun shares its φ-features with the head of its binder, and
9.5 Chapter summary and conclusions of Part IV 241
is either deleted at PF or is in a null phonological form. Thus the above problem

is avoided.
This sideward movement approach shares characteristic A with our approach
(see Section 9.4.1), but it lacks the other two characteristics B and C.
In fact, none of these alternative approaches to ATB constructions exhibits
characteristic C. That is, none of them considers the relationship between TLCs
(identity adjectives) and ATB constructions, either syntactically or semanti-
cally. Moreover, none of them considers the syntactic issues of ATB construc-
tions in wh-in situ languages.
9.5 Chapter summary and conclusions of Part IV

In this chapter, I have presented a new approach to ATB constructions without
resorting to ATB movement. I reached the following conclusions.
(a) ATB constructions are TLCs, cross-linguistically. The two constructions
share deep syntactic and semantic properties.
(b) There is no forking movement chain in human language, and thus there is
no syntactic operation that is exclusively used in coordination. I have argued that
the overt extracted element of ATB constructions has a movement chain relation
with only the gap in the external conjunct; like SEs, the extracted elements in
ATB constructions cannot originate in the two conjuncts at the same time. I
have argued that the silent pro-φP argument in the internal conjunct never
moves out of the conjunct.
This part of the book (this and the previous chapter) shows that forking
chains, which have been stipulated for coordinate constructions only, do not
actually exist. Thus coordinate constructions do not introduce any special type
of movement to the computational system. This is the fourth (and last) major
claim of this book:
No special operation exists in the computation of co-
ordinate complexes.
10 Conclusions
In this monograph, I have answered the four fundamental questions listed in

the introduction.
A. Does the derivation of coordinate constructions create any special syntac-
tic configuration, other than the general binary complementation and adjunction
configuration?
I argued that the structure of coordinate complexes is not different from
that of other syntactic constructions. First, I presented new arguments to sup-
port the claim that a coordinate complex has a binary-branching structure,
with one conjunct internal and the other external to the coordinator. Second,
I argued against any adjunction structure for coordination. Instead, I advo-
cated the complementation structure in (10.1), where the coordinator heads
a projection and the relation between the external and internal conjunct is
a Spec-Comp relation. In this structure, the combination of the coordinator
and the internal conjunct is an intermediate projection, which is not able to
move.
(10.1)
Spec
head complement
external conjunct coordinator internal conjunct
Third, I clarified that although coordination does not have an adjunction

structure, it still may express a semantic modification relation. Fourth, I exam-
ined an issue with respect to the projectivity of coordinate complexes, namely
the issue of word and word fragment conjuncts. Chomsky’s (1994) bare phrase
structure hypothesis, which abandons bar-levels for syntactic elements, pro-
vides us with a simple and thus desirable analysis of coordination of elements
of various constituency levels. Thus, one should not use the existence of non-
phrasal conjuncts to argue against the complementation structure of coordinate
242
Conclusions 243
complexes. Fifth, I claimed that when there is only one coordinator and three
or more conjuncts, only one conjunct is in the complement position and the
remaining conjuncts are in multiple Spec positions.
Overall, the structure of coordination is complementation. The theoretical
implication of this conclusion is that the notion “coordination” is not a primitive
syntactic relation, but rather instantiates the more general complementation
relation.
B. Does the derivation of coordinate constructions require any special syn-
tactic category, other than NP, VP, and so on?
I presented cross-linguistic data showing that coordinators can c-select con-
juncts, and may have ordinary categorial features. I concluded that if a coor-
dinator does not have any intrinsic categorial features, the coordinate complex
shares its category with the external conjunct; thus there is no special cate-
gory like &P. Coordinate complexes do not exhibit any distribution other than
that of the currently recognized categories, nor do they “bleach” the contrasts
among the currently recognized categories. Therefore, they cannot represent
an independent syntactic category.
C. Does the derivation of coordinate constructions require any special con-
straint on syntactic operations, other than general conditions such as the locality
condition?
The classic coordination-specific constraint is Ross’s (1967) Coordinate
Structure Constraint (CSC). This constraint disallows the movement of whole
conjuncts (the CC part of the CSC) and the extraction of any element from
conjuncts (the EC part of the CSC).
I concluded that the effects of the CC on external conjuncts can be accounted
for by the morphological properties of coordinators like and. I argued that
such elements have no intrinsic categorial features, although they are heads
of coordinate complexes. In order for a coordinate complex headed by such a
coordinator to take part in any syntactic computation, the categorial features of
the external conjunct must be transferred to the coordinator. External conjuncts
in and-coordinate complexes thus cannot move, having lost their categorial
features. This analysis is supported by my study of de constructions in Chinese.
As in and-constructions, the constituent that provides the whole complex with
categorial features in de constructions may not move. As for internal conjuncts,
I argued that they may not move because they are the phonological hosts of the
coordinators.
The observed CSC effects are accounted for not only by the morphological
properties of conjunctions, but also by the semantic relation between conjuncts.
The latter aspect was spelled out in the following two steps.
244 Conclusions
First, I demonstrated the distinction between accidental coordination and

natural coordination: in the former, conjuncts are not semantically related,
whereas in the latter they are. This distinction is encoded by language-specific
syntactic and morphological strategies. In particular, the bare and in English and
the conjunctions he, gen, and erqie in Chinese can occur in either accidental
coordination or natural coordination, whereas the conjunction ji in Chinese
occurs only in accidental coordination.
Second, I showed that the possibility of violating the CSC is correlated
with natural coordination. Cross-linguistically, the EC part of the CSC can be
violated in asymmetrical coordination constructions, but not in symmetrical or
accidental coordinate constructions. In Chinese, the CC part of the CSC can
be violated in comitative coordination constructions, but not in distributive or
accidental coordination constructions. In both asymmetrical coordination and
comitative coordination, conjuncts are semantically related to each other, thus
they both are natural coordination. The fact that the CSC can be violated in
natural coordination reveals a semantic aspect of the generalizations originally
motivating the constraint.
I thus investigated two factors related to the CSC: the morphological proper-
ties of coordinators, and the semantic distinctions between natural coordination
and accidental coordination. Neither factor requires the CSC as a constraint on
syntactic operations.
While I argued against the CSC as a syntactic constraint on the operation of
Move, I also argued against other constraints on the operation of Merge in build-
ing of coordinate constructions. I showed that each of the following constraints
can be violated under the proper conditions: the Coordination of Likes Con-
straint with respect to syntactic categories, the Coordination of Likes Constraint
with respect to grammatical functions, the Coordination of Likes Constraint
with respect to semantic types, and the Coordinate Constituent Constraint. The
observed effects of these constraints can be covered by my Relativized Paral-
lelism Requirement, which is a filter on syntactic representations in language
processing. The filter is repeated as follows:
(10.2) Conjuncts of a coordinate complex must hold a coherence

relation in terms of:
a. Relatedness: they must be related to each other semantically, as
in natural coordination;
or
b. Resemblance (Parallelism Requirement): they must hold a
resemblance relation in terms of both their semantic type and their
dependency chains.
Conclusions 245
The theoretical consequence is that, by accounting for the CSC and other
coordinate construction-specific constraints in terms of the morphological prop-
erties of coordinators and a processing filter, we remove all of these constraints
from the syntactic computational system. Empirically, rejecting the CSC makes
it possible to derive three hitherto recalcitrant constructions: the Split Argu-
ment Construction, the Modifier-Sharing Construction, and the Interwoven
Dependency Construction, as represented by (10.3a), (10.3b), and (10.3c),
respectively.
(10.3) a. Italy borders France.

b. A man came in and a woman left who were quite similar.
c. Which nurse and which hostess did Fred date and Bob marry, respectively?
I proposed that in Split Argument Constructions and Modifier-Sharing Con-

structions, conjuncts undergo movement, and in Interwoven Dependency Con-
structions, elements are extracted from conjuncts.
D. Finally, does the derivation of coordinate constructions require any special
type of syntactic operations, other than Merge and the step-by-step, one-tail-
one-head chains of Move?
In the generative syntactic literature, it has often been assumed that a special
forking movement is required to derive a certain type of coordinate construction,
namely Across-the-Board (ATB) constructions. Under this traditional assump-
tion, (10.4a) is derived by the movement chain in (10.4b), in which one head
link has two tail links.
(10.4) a. Who did Jim like and Jane hate?

b. Who did Jim like _ and Jane hate _? (ATB movement)
In order to find out whether the forking movement exists, I studied the
syntactic derivation of same constructions such as (10.5a). I found that the
relational nominal that contains the adjective same is base-generated in the first
conjunct, moves out of the coordinate complex, and binds a silent pro-form in
the second conjunct. My derivation of (10.5a) is illustrated in (10.5b).

b. [Mary helped [DP1 the same man]i] and [Jane ruined pro- φPi].
I argued that ATB constructions are derived in the same way as same con-
structions, by variable binding and an ordinary movement operation. Therefore,
246 Conclusions
there is no forking movement in the derivation of ATB constructions, and thus

no special syntactic operation in the derivation of coordinate complexes.
In short, I argued against any special syntax for coordination. Coordination
has no special status in syntax, but falls out automatically from structures,
categories, constraints, and operations that are already used in non-coordinate
constructions.
References
Abbott, Barbara 1976. “Right Node Raising as a test for constituenthood,” Linguistic
Inquiry 7: 639–642.
Abels, Klaus 2003. “Successive cyclicity, anti-locality, and adposition stranding,” PhD
diss., Storrs: University of Connecticut.
Abney, Steven 1991. “Syntactic affixation and performance structures,” in Katherine
Leffel and Denis Bouchard (eds.), Views on Phrase Structure, Dordrecht: Kluwer,
pp. 215–227.
Aboh, Enoch Oladé 2009. “Clause structure and verb series,” Linguistic Inquiry 40.
Adger, David 2003. Core Syntax: A Minimalist Approach. Oxford: Oxford University
Press.
Adger, David and Gillian Ramchand 2005. “Merge and move: wh-dependencies revis-
ited,” Linguistic Inquiry 36: 161–193.
Agbayani, Brian and Ed Zoerner 2004. “Gapping, pseudogapping, and sideward move-
ment.” Studia Linguistica 58: 185–211.
Alexiadou, Artemis, Paul Law, Andre Meinunger, and Chris Wilder 2000. “Introduc-
tion,” in Artemis Alexiadou, Paul Law, Andre Meinunger, and Chris Wilder (eds.),
The Syntax of Relative Clauses, Amsterdam: John Benjamins, pp. 1–51.
Amfo, Nana Abe Appiah 2007. “Clausal conjunction in Akan,” Lingua 117: 666–684.
Amritavalli, R. 2003. “Question and negative polarity in the disjunctive phrase,” Syntax
6: 1–18.
Anandan, K. N. 1993. “Constraints on extraction from coordinate structures in English
and Malalylam,” PhD diss., Central Institute of English and Foreign Languages,
Hyderabad, India.
Anderson, Carol 1983. “Generating coordinate structures with asymmetrical gaps,”
Chicago Linguistic Society (CLS) 19: 3–14.
Androutsopoulou, Antonia 1994. “The distribution of the definite determiner and the
syntax of Greek DPs,” in Proceedings of the 30th regional meeting of the Chicago
Linguistic Society, Chicago: University of Chicago Press, pp. 16–29.
Aoun, Joseph and Elabbas Benmamoun 1999. “Further remarks on first conjunct agree-
ment,” Linguistic Inquiry 30: 669–681.
Aoun, Joseph and Yen-hui Audrey Li 2003. Essays on the Representational and Deriva-
tional Nature of Grammar: The Diversity of Wh-constructions. Cambridge, MA:
MIT Press.
Aoun, Joseph, Elabbas Benmamoun, and Dominique Sportiche 1994. “Agreement and
conjunction in some varieties of Arabic,” Linguistic Inquiry 25: 195–220.
247
248 References
Artstein, Ron 2005. “Coordination of parts of words,” Lingua 115 (4): 359–393.
Bach, Emmon 1964. An Introduction to Transformational Grammars. New York: Holt,
Rinehart & Winston.
Baker, Mark 1988. Incorporation. Chicago and London: University of Chicago Press.
1992. “Unmatched chains and the representation of plural pronouns,” Natural Lan-
guage Semantics 1: 33–73.
1997. “Thematic roles and syntactic structure,” in L. Haegeman (ed.), Elements of
Grammar, Dordrecht: Kluwer Academic Publishers, pp. 73–137.
Baltin, Mark 2002. “Movement to the Higher V is Remnant Movement,” Linguistic
Inquiry 33: 653–658.
2006. “The nonunity of VP-preposing,” Language 82: 734–766.
Bánréti, Zoltán 1994. “Coordination,” in Ferenc Kiefer and Katalin É. Kiss (eds.), Syntax
and Semantics 27: The Syntactic Structure of Hungarian, San Diego: Academic
Press, pp. 355–414.
2003. “On the syntax of coordinate constructions,” Acta Linguistica Hungarica
50(3–4): 265–340.
Barker, Chris 2007. “Parasitic Scope.” Linguistics and Philosophy 30(4): 407–444.
Bar-Lev, Zev and Arthur Palacas 1980. “Semantic command over pragmatic priority,”
Lingua 51: 137–146.
Bayer, Samuel 1996. “The coordination of unlike categories,” Language 72: 579–616.
Berchem, Jörg 1991. Referenzgrammatik des Somali. Köln: Rüdiger Köppe.
Besten, Hans den and Gert Webelhuth 1990. “Stranding,” in Günther Grewendorf and
Wolfgang Sternefeld (eds.), Scrambling and Barriers, Amsterdam: Benjamins,
pp. 77–92.
Bever, Thomas G., C. Carrithers, W. Cowart, and D. J. Townsend 1989. “Language
processing and familial handedness,” in A. Galaburda (ed.), From Reading to
Neurons, Cambridge, MA: MIT Press, pp. 331–357.
Bianchi, Valentina 1999. Consequences of Antisymmetry: Headed Relative Clauses.
Berlin: Mouton de Gruyter.
Blakemore, Diane and Robyn Carston 2005. “The pragmatics of sentential coordination
with and,” Lingua 115(4): 569–589.
Blass, Regina 1989. “Pragmatic effects of coordination: the case of ‘and’ in Sissala,”
UCL Working Papers in Linguistics 1, University College, University of London,
pp. 32–52.
Bloomfield, Leonard 1933. Language. New York: Holt, Rinehart & Winston.
Blümel, Rudolf 1914. Einführung in die Syntax. Heidelberg: C. Winter.
Bobaljik, Jonathan David and Sam Brown 1997. “Inter-arboreal operations: head-
movement and the Extension Requirement,” Linguistic Inquiry 28: 345–356.
Boeckx, Cedric 2003. “(In) direct binding,” Syntax 6: 213–236.
2007. “Review of linearization of chains and sideward movement,” Language 83:
895–899.
Boeckx, Cedric and Norbert Hornstein 2005. “The status of D-structure: the case of
binominal each,” Syntax 8: 23–43.
Booij, Geert 1985. “Coordination reduction in complex words: a case for prosodic
phonology,” in Harry van der Hulst and Norval Smith (eds.), Advances in Non-
linear Phonology, Dordrecht: Foris, pp. 143–160.
References 249
Borsley, Robert D. 1983. “A note on preposition stranding,” Linguistic Inquiry 14:

338–343.
1994. “In defense of coordinate structures,” Linguistic Analysis 24: 218–246.
2005. “Against ConjP,” Lingua 115(4): 461–482.
Bošković, Željko 2006. “Case checking versus case assignment and the case of adverbial
NPs,” Linguistic Inquiry 37(3): 522–533.
Bošković, Željko and Steven Franks 2000. “Across-the-Board Movement and LF,”
Syntax 3: 107–128.
Bowers, John 1993. “The syntax of predication,” Linguistic Inquiry 24: 591–656.
2001. “Predication,” in M. Baltin and C. Collins (eds.), The Handbook of Contempo-
rary Syntactic Theory. Malden, MA: Blackwell. 299–333.
2002. “Transitivity,” Linguistic Inquiry 33: 183–224.
Bresnan, Joan and Höskuldur Thráinsson 1990. “A note on Icelandic coordination,” in
Joan Mailing and Annie Zaenen (eds.), Syntax and Semantics 24: Modern Icelandic
Syntax, San Diego: Academic Press, 355–365.
Bruening, Benjamin 2001. “QR obeys superiority: frozen scope and ACD.” Linguistic
Inquiry 32: 233–273.
Büring, Daniel 2002. “2 × Singular = plural,” Snippets 6: 6–7.
Burton, Strang and Jane Grimshaw 1992. “Coordination and VP-internal subjects,”
Linguistic Inquiry 23: 305–313.
Camacho, José 1997. “The syntax of NP coordination,” PhD diss., University of South-
ern California.
2003. The Syntax of Coordination. Dordrecht: Kluwer Academic Publishers.
Caponigro, Ivano 2003. “Unbalanced coordination in Maasai,” UCLA Working Papers
in Linguistics 9: 1–16.
Carlson, Greg 1987. “Same and different: some consequences for syntax and semantics,”
Linguistics and Philosophy 10: 531–565.
Carlson, Katy 2002. Parallelism and Prosody in the Processing of Ellipsis Sentences,
New York and London: Routledge.
Carston, Robyn 1993. “Conjunction, explanation and relevance,” Lingua 90: 27–48.
2002. Thoughts and Utterances: The Pragmatic of Explicit Communication. Oxford:
Blackwell.
Castillo, Juan Carlos 2001. “Thematic relations between nouns,” PhD diss., University
of Maryland at College Park.
Chametzky, Robert 1987. “Coordination and the organization of a grammar.” PhD diss.,
University of Chicago, Chicago.
Chang, Chia-Hao, Frank 2006. “Gapless Relative Clause Constructions in Mandarin
Chinese,” MA thesis, National Chung Cheng University.
Chao, Yuen Ren 1968. A Grammar of Spoken Chinese, Berkeley: University of Califor-
nia Press.
Chapman, Carol 1995. “A subject–verb agreement hierarchy,” Historical Linguistics 2:
35–44.
Chaves, Rui Pedro 2007. “Coordinate structures: constraint-based syntax–semantics
processing,” PhD diss., Universidade de Lisboa.
Chen, Qitong, Qingyan Chen, Xuanrong Chen, et al. (eds.) 1982. Xiandai Hanyu Xuci
Lishi [Examples of Functional Words in Modern Chinese], Beijing: Shangwu Press.
250 References
Cheng, Lisa Lai Shen 1986. “De in Mandarin,” Canadian Journal of Linguistics 31(4):
313–326.
1988. “Transitive alternations in Mandarin Chinese,” Talk presented at the Ohio State
University Conference on Chinese Linguistics, May 1988.
Cho, Sunggeun and Xuan Zhou 2000. “The interpretations of wh-elements in conjoined
wh-questions,” in Noriko Akatsuka, Susan Strauss, and Bernard Comrie (eds.),
Japanese/Korean Linguistics 10, CSLI publications, Stanford University, pp. 522–
531.
Chomsky, Noam 1957. Syntactic Structures. The Hague: Mouton and Co.
1964. Current Issues in Linguistic Theory, Mouton, The Hague.
1965. Aspects of the Theory of Syntax. MIT Press, Cambridge, MA.
1968. Language and Mind. Harcourt, Brace, and World, New York.
1975. “Questions of form and interpretation,” Linguistic Analysis 1: 75–109.
1977. “On wh-movement,” in Peter Culicover, Thomas Wasow and Adrian Akmajian
(eds.), Formal Syntax, New York: Academic Press, pp. 71–132.
1981. Lectures on Government and Binding. Dordrecht: Foris.
1993. “A Minimalist Program for Linguistic Theory,” in K. Hale and S. J. Keyser
(eds.), The View from Building 20, Cambridge, MA: MIT Press, pp. 1–52.
1994. “Bare phrase structure.” MIT Occasional Papers in Linguistics 5. (Also in
G. Webelhuth [ed.], [1995] Government and Binding Theory and the Minimalist
Program, Oxford: Blackwell.)
1995. The Minimalist Program. Cambridge, MA: MIT Press.
2000. “Minimalist inquiries: The framework,” in Roger Martin, David Michaels, and
Juan Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of
Howard Lasnik, Cambridge, MA: MIT Press, pp. 89–155.
2002. On Nature and Language. Cambridge: Cambridge University Press.
2007. “Approaching UG From Below,” in Uli Sauerland and Hans-Martin Gärtner
(eds.), Interfaces + Recursion = Language? Berlin: Mouton de Gruyter, pp. 1–
29.
2008. “On Phases,” in Robert Freidin, Carlos Otero, and Maria-Luisa Zubizarreta
(eds.), Foundational Issues in Linguistic Theory, Cambridge, MA: MIT Press,
pp. 133–166.
Chung, Sandra and William Ladusaw 2003. Restriction and Saturation, Cambridge,
MA: MIT Press.
Citko, Barbara 2003. “ATB wh-questions and the nature of Merge,” in Makoto Kadowaki
and Shigeto Kawahara (eds.), The Proceedings of NELS 33, Umass, Amherst:
GLSA Publications, pp. 87–102.
Cole, Peter, Gabriella Hermon, and Li-May Sung 1993. “Feature Percolation,” Journal
of East Asian Linguistics 2: 91–118.
Collins, Christopher 1988a. Part 1. “Conjunction adverbs.” ms., MIT.
1988b. “Part 2. Alternative analysis of conjunction.” ms., MIT.
1997. Local Economy. Cambridge, MA: MIT Press.
2002a. “Eliminating Labels,” in S. Epstein and D. Seely (eds.), Derivation and
Explanation in the Minimalist Program, MA: Blackwell, pp. 42–64.
2002b. “Multiple Verb Movement in = Hoan,” Linguistic Inquiry 33: 1–29.
Cormack, Annabel and Neil Smith 2005. “What is coordination,” Lingua 115: 385–418.
References 251
Cowart, Wayne and Dana McDaniel 2008. “What kind of thing is a coordination?:
qualitative comparisons of agreement relations in coordinate and subordinate sen-
tences,” ms. University of Southern Maine.
Cowper, Elizabeth and Daniel Hall 2000. “Intransitive and: locality, movement, and
interpretation,” in J. T. Jensen and G. Van Herk (eds.), Proceedings of the 2000
Annual conference of the Canadian Linguistics Association, Ottawa: Cahiers Lin-
guistiques d’Ottawa, pp. 25–36.
Culicover, Peter W. 1990. “Strange extractions,” ms., Center for Cognitive Science, the
Ohio State University, Columbus.
Culicover, Peter W. and Ray Jackendoff 1997. “Semantic subordination despite syntactic
coordination,” Linguistic Inquiry 28: 195–217.
Dalrymple, Mary and Irina Nikolaeva 2006. “Syntax of natural and accidental coordi-
nation: evidence from agreement,” Language 82: 824–849.
Davies, William D. and Stanley Dubinsky 2003. “On extraction from NPs,” Natural
Language and Linguistic Theory 21: 1–37.
de Vos, Mark Andrew 2005. “The syntax of verbal pseudo-coordination in English and
Afrikaans,” PhD diss., Leiden University.
de Vos, Mark and Luis Vicente 2005. “Coordination under Right Node Raising,” in
John Alderete, Chung-hye Han, and Alexei Kochetov (eds.), Proceedings of the
24th West Coast Conference on Formal Linguistics, Somerville, MA: Cascadilla
Proceedings Project, pp. 97–104.
Déchaine, Rose-Marie and Martina Wiltschko 2002. “Decomposing Pronouns,” Lin-
guistic Inquiry 33: 409–442.
Dik, Simon C. 1968. Coordination: Its Implications for the Theory of General Linguis-
tics. Amsterdam: North-Holland publishing company.
Dikken, Marcel den 2001. “Pluringulars, pronouns and quirky agreement,” The Lin-
guistic Review 18: 19–41.
2006. Relators and Linkers. Cambridge, MA: MIT Press.
Dikken, Marcel den and Rint Sybesma 1998. “Take serials light up the middle,” talk
presented at at GLOW 21, Tilburg University, April 15, 1998.
Dougherty, Ray C. 1969. “Review of Coordination: Its Implications for the Theory of
General Linguistics by Simon C. Dik,” Language 45: 624–636.
1970a. “Review of universals in linguistic theory,” Foundations of Language 6: 505–
561.
1970b. “A grammar of coordinate conjoined structures: I,” Language 46: 850–898.
Dowty, David 1987. “Collective predicates, distributive predicates, and all,” The Pro-
ceedings of Eastern States Conference on Linguistics (ESCOL) ’86: 97–115.
Eggert, Randall 2000. “Grammaticality and context with respect to and . . . and or . . .
respectively,” Chicago Linguistic Society 36: 93–107.
Elbourne, Paul 2005. Situations and Individuals. Cambridge, MA: MIT Press.
Embick, David and Rolf Noyer 2007. “Distributed Morphology and the syntax/
morphology interface,” in G. Ramchand and C. Reiss (eds.), The Oxford Handbook
of Linguistic Interfaces, Oxford: Oxford University Press, pp. 289–324.
Emonds, Joseph 1972. “Evidence that indirect object movement is a structure-preserving
rule,” Foundations of Language 8: 546–561.
Evans, Gareth 1980. “Pronouns,” Linguistic Inquiry 11(2): 337–362.
252 References
Fitzpatrick, Justin Michael 2006. “Two types of floating quantifiers and their A/A-bar
Properties,” C. Davis, A. Deal and Y. Zabbal (eds.), Proceedings of NELS 36,
pp. 253–265.
Fox, Danny 2000. Economy and Semantic Interpretation, Cambridge: MIT Press.
Frank, Anette 2002. “A (discourse) functional analysis of asymmetric coordination,” in
Miriam Butt and Tracy Holloway King (eds.), Proceedings of the LFG02 Confer-
ence, Athens: CSLI Publications, pp. 174–196.
Franks, Steven 1992. “A prominence constraint on null operator constructions,” Lingua
87: 35–54.
1993. “On Parallelism in Across-the-Board Dependencies,” Linguistic Inquiry 24(3):
509–529.
Frazier, Lyn and Charles Clifton 2001. “Parsing coordinates and ellipsis: copy α,”
Syntax 4: 1–22.
Frazier, Lyn, Jeremy M. Pacht, and Keith Rayner 1999. “Taking on semantic commit-
ments, II: collective versus distributive readings,” Cognition 70: 87–104.
Frazier, Lyn, Alan Munn and Charles Clifton 2000. “Processing coordinate structures,”
Journal of Psycholinguistic Research 29: 343–370.
Fromkin, Victoria, Robert Rodman, Nina Hyams 2007. An Introduction to Language.
8th edition, Boston, MA: Thomson Wadsworth.
Fu, Jingqi 1994. “On deriving Chinese derived nominals: evidence for V-to-N raising,”
PhD diss., University of Massachusetts.
Gawron, Jean Mark and Andrew Kehler 2004. “The semantics of respective readings,
conjunction, and filler-gap dependencies,” Linguistics and Philosophy 27: 169–
207.
Gazdar, Gerald 1981. “Unbounded dependencies and coordinate structure,” Linguistic
Inquiry 12: 155–183.
Gazdar, Gerald, Geoffrey K. Pullum, Ivan A. Sag, and Thomas Wasow 1982. “Coordi-
nation and transformational grammar,” Linguistic Inquiry 13: 663–676.
Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag 1985. Generalized Phrase
Structure Grammar. Oxford: Basil Blackwell.
George, Leland 1980. “Analogical generalization in natural language syntax,” PhD
diss., MIT.
Gleitman, Lila 1965. “Coordinating Conjunctions in English,” Language 41: 260–
293.
Godard, Danièle 1989. “Empty categories as subjects of tensed Ss in English or
French?,” Linguistic Inquiry 20: 497–506.
Goldsmith, John 1985. “A principled exception to the Coordinate Structure Constraint,”
in Papers from the Twenty-First Annual Regional Meeting of the Chicago Linguistic
Society, Chicago: Chicago Linguistic Society, pp. 133–143.
Goodall, Grant 1987. Parallel Structures in Syntax: Coordination, Causatives and
Restructuring. Cambridge: Cambridge University Press.
Gracanin-Yuksek, Martina 2007. “About sharing,” PhD diss., MIT.
Grice, Paul 1967. “Logic and conversation: the William James Lectures,” in H. P. Grice
(1989), Studies in the Way of Words, Cambridge, MA: Harvard University Press,
pp. 22–40.
References 253
Grimshaw, Jane 1981. “Form, function, and the Language Acquisition Device,” in C. L.
Baker and John J. McCarthy (eds.), The Logical Problem of Language Acquisition,
Cambridge, MA: MIT Press, pp. 165–182.
1991. “Extended Projection,” ms., Brandeis University.
Groot, A. W. De 1949. Structurele Syntaxis. The Hague: Servire.
Grootveld, Marjan 1994. “Parsing coordination generatively,” HIL diss., Leiden
University.
Grosu, Alexander 1972. “The strategic content of Island Constraints,” PhD diss., Ohio
State University.
1973. “On the nonunitary nature of the coordinate structure constraint,” Linguistic
Inquiry 4: 88–92.
1981. Approaches to Island Phenomena. Amsterdam: North-Holland.
1985. “Subcategorization and parallelism,” Theoretical Linguistics 12: 231–240.
Hagstrom, Paul 1998. “Decomposing questions,” PhD diss., MIT.
Haı̈k, Isabelle 1985. “The syntax of operators,” PhD diss., MIT.
Hankamer, Jorge 1973. “Unacceptable ambiguity,” Linguistic Inquiry 4: 17–68.
Hankamer, Jorge and Ivan. Sag 1976. “Deep and surface anaphora,” Linguistic Inquiry
7: 391–428.
Harley, Heidi 2005. “One-replacement, unaccusativity, acategorial roots, and bare
phrase structure,” in Slava Gorbachov and Andrew Nevins (eds.), Harvard Working
Papers in Linguistics 11, pp. 59–78.
Hartmann, Katharina 2000. Right Node Raising and Gapping. Amsterdam: John
Benjamins.
Haspelmath, Martin 2002. Understanding Morphology. New York: Oxford University
Press.
2004. “Coordinating constructions: an overview,” in M. Haspelmath (ed.) Coor-
dinating Constructions, Typological Studies in Language 58, Amsterdam: John
Benjamins, pp. 3–39.
2007. “Coordination,” in T. Shopen (ed.), Language Typology and Syntactic Descrip-
tion, vol. II: Complex Constructions, 2nd edn. Cambridge: Cambridge University
Press, pp. 1–51.
Hegarty, Michael 2003. “Semantic types of abstract entities,” Lingua 113: 891–927.
Heim, Irene 1982. “The semantics of definite and indefinite noun phrases,” PhD diss.,
University of Massachusetts, Amherst.
Heim, Irene and Angelika Kratzer 1998. Semantics in Generative Grammar. MA:
Blackwell Publishers.
Heim, Irene, Howard Lasnik and Robert May 1991. “Reciprocity and Plurality,” Lin-
Hendriks, Petra 1991. “Deletion in coordinate structures: the parallelism requirement,”
in Mark Kas, Eric Reuland and Co Vet (eds.), Language and Cognition 1, Yearbook
1991 of the research group for Linguistic Theory and Knowledge Representation
of the University of Groningen, pp. 99–110.
2002. “ ‘Either’ as a focus particle,” ms., University of Gronigen.
2004. “Coherence relations, ellipsis, and contrastive topics,” Journal of Semantics
21(2): 133–153.
254 References
Hernández, Ana Carrera 2007. “Gapping as a syntactic dependency,” Lingua 117:

2106–2133.
Heycock, Caroline and Anthony Kroch 1994. “Verb movement and coordination in a
dynamic theory of licensing,” The Linguistic Review 11: 257–283.
Heycock, Caroline and Roberto Zamparelli 2000. “Friends and colleagues: plurality and
NP-coordination,” NELS 30, pp. 341–352.
2002. “Conjunction and plurality: a case study in the integration of syntax and
semantics.” Project Proposal, University of Edinburgh.
2003. “Coordinated Bare Definites,” Linguistic Inquiry 34: 443–469.
2005. “Friends and colleagues: plurality, coordination, and the structure of DP,”
Natural Language Semantics 13: 201–270.
Higginbotham, James 1987. “Indefiniteness and predication,” in Eric Reuland and Alice
ter Meulen (eds.), The Representation of (In)definiteness, Cambridge, MA: MIT
Press, pp. 43–70.
Hinds, John 1986. Japanese. London and New York: Routledge.
Hockett, Charles Francis 1958. A Course in Modern Linguistics. New York: The Macmil-
lan Company.
Hoeksema, Jack 1983. “Plurality and conjunction,” in A. ter Meulen (ed.) Studies in
Modeltheoretic Semantics. Dordrecht: Foris.
Hoekstra, Eric 1994. “Expletive replacement, verb-second and coordination,” The Lin-
guistic Review 11: 285–297.
Hoh, Pau-San and Wen-yu Chiang 1990. “A focus account of moved wh-phrases at
S-structure in Chinese,” Lingua 81:47–73.
Höhle, Tilman 1990. “Assumptions about asymmetric coordination in German,” in Joan
Mascaró and Marina Nespor (eds.), Grammar in Progress: A Festschrift for Henk
van Riemsdijk. Dordrecht: Foris, pp. 221–235.
Hornstein, Norbert 2001. Move! A Minimalist Theory of Construal. Malden, MA: Black-
well Publishers.
Hornstein, Norbert and Hirohisa Kiguchi 2003. “PRO gate and movement.” Proceed-
ings of the 25th annual Penn Linguistics Colloquium. University of Pennsylvania
Working Papers in Linguistics 8(1): 33–46.
Hornstein, Norbert and David Lightfoot 1987. “Predication and PRO,” Language 63:
23–52.
Hornstein, Norbert and Jairo Nunes 2002. “On asymmetries between parasitic gap and
Across-The-Board constructions,” Syntax 5: 26–54.
Huang, Cheng-teh James 1982. “Logical relations in Chinese and the theory of gram-
mar,” PhD diss., MIT.
1988a. “Hanyu zhengfanwenju de mozuyifa” [Chinese A-not-A questions: a modular
approach], Zhongguo Yuwen 205: 247–264.
1988b. “Shuo shi he you” [On Be and Have in Chinese], The Bulletin of the Institute
of History and Philology 59: 43–64.
1993. “Reconstruction and the structure of VP: some theoretical consequences,”
Huang, Chu-Ren 1989. “Mandarin Chinese NP de – a comparative study of current
grammatical theories.” Nankang, Taipei (PhD diss., Cornell University 1987).
Huddleston, Rodney and Geoffrey Pullum 2002. The Cambridge Grammar of the
English Language. Cambridge: Cambridge University Press.
References 255
Hudson, Richard A. 1972. Systemic Generative Grammar. Mimeo, London: University

College.
1976. Arguments for a Non-transformational Grammar. Chicago: Chicago University
Press.
Humphreys, Karin R. and Kathryn Bock 2005. “Notional number agreement in English,”
Psychonomic Bulletin & Review 12(4): 689–695.
Huttar, George 2003. “Sum: discourse-initial ‘and’,” Linguist List 14–1163.
2004. “Summary on Latin preposition and conjunction order,” Linguist List 15–541.
1977. X Syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press.
Jackendoff, Ray 1997. The Architecture of the Language Faculty. Cambridge, MA: MIT
Press.
Jacobson, Pauline 1987. “Review of G. Gazdar, E. Klein, G. Pullum, and I. Sag,
Generalized Phrase Structure Grammar,” Linguistics and Philosophy 10: 389–
426.
Jaeggli, Osvaldo and Nina M. Hyams 1993. “On the independence and interdependence
of syntactic and morphological properties: English aspectual come and go,” Natural
Jayaseelan, Karattuparambil A. 2001. “Questions and question-word incorporating
quantifiers in Malayalam,” Syntax 4: 63–93.
2008. “Question particles and disjunction.” LingBuzz/000644.
Jensen, John Thayer 1977. Yapese Reference Grammar. Honolulu: University Press of
Hawaii.
Jespersen, Otto 1924. Philosophy of Grammar, New York: Norton (1965 edn.).
Johannessen, Janne Bondi 1996. “Partial agreement and coordination,” Linguistic
Inquiry 27: 661–676.
1998. Coordination. Oxford: Oxford University Press.
Johnson, Kyle 1991. “Object positions,” Natural Language and Linguistic Theory 9:
577–636.
1996. “Gapping. Ch. 2 of: In search of the English Middle Field,” ms., University of
Massachusetts–Amherst.
2002. “Restoring exotic coordinations to normalcy,” Linguistic Inquiry 33: 97–156.
2007. “LCA + alignment = RNR. Workshop on coordination, subordination and
ellipsis,” talk presented at University of Tübingen, June 2007.
2008. “The view of QR from ellipsis,” in Kyle Johnson (ed.), Topics in Ellipsis,
Cambridge: Cambridge University Press, pp. 69–94.
2009. “Gapping isn’t (VP) ellipsis,” Linguistic Inquiry 40: 289–328.
Kandybowics, Jason 2006. “Nupe coordinate structures: a syntactically heteroge-
neous class,” in Rebecca Cover and Yuni Kim (eds.), Proceedings of the
31st Annual Meeting of the Berkeley Linguistics Society (Volume 31S: Special
Session on Languages of West Africa), Berkeley: Berkeley Linguistics Society,
pp. 53–64.
Karimi, Yadgar 2007. “Kurdish Ezafe construction: implications for DP structure,”
Lingua 117: 2159–2177.
Kasai, Hironobu 2002. “Some remarks on the Coordinate Structure Constraint,” talk
presented at the GLOW in Asia (3), National Tsing Hua University, Tsinchu, Jan.
4–7, 2002.
2004. “Two notes on ATB movement,” Language and Linguistics 5(1): 167–188.
256 References
Kayne, Richard 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
2002. “Pronouns and their Antecedents,” in S. Epstein and D. Seely (eds.), Derivation
and Explanation in the Minimalist Program, Malden, MA: Blackwell Publishers,
pp. 133–166.
Keenan, Edward 1976. “Remarkable subjects in Malagasy,” in Charles Li (ed.), Subject
and Topic, New York: Academic Press, pp. 247–301.
Kehler, Andrew 2002. Coherence, Reference, and the Theory of Grammar. Stanford,
CA: CSLI Publications.
Kitagawa, Chistato and Claudia Ross 1982. “Prenominal modification in Chinese and
Japanese,” Linguistic Analysis 9: 19–53.
Koizumi, Masatishi 1995. “Phrase structure in minimalist syntax,” PhD diss., MIT.
Kolb, Hans-Peter and Craig Thiersch 1991. “Levels and empty categories in a principles
and parameters approach to parsing,” in Hubert Haider and Klaus Netter (eds.),
Representation and Derivation in the Theory of Grammar, Dordrecht: Kluwer,
pp. 251–302.
Koopman, Hilda and Anna Szabolcsi 2000. Verbal Complexes. Cambridge, MA: MIT
Press.
Koutsoudas, Andreas 1968. “The A over A convention,” Linguistics 46: 11–20.
Kratzer, Angelika 2009. “Building a pronoun: fake indexicals as windows into the
properties of bound variable pronouns,” Linguistic Inquiry, to appear.
Krifka, Manfred 1990. “Boolean and Non-Boolean ‘and’,” in Papers from the Second
Symposium on Logic and Language, Budapest: Akademiai Kiado, pp. 161–188.
Kuno, Susumu 1973. The Structure of the Japanese Language. Cambridge, MA: MIT
Press.
1987. Functional Syntax – Anaphora, Discourse and Empathy. Chicago and London:
The University of Chicago Press.
Laka, Itziar 1994. On the Syntax of Negation. New York: Garland Publishing Inc.
Lakoff, George 1986. “Frame semantic control of the coordinate structure constraint,”
in Anne M. Farley, Peter T. Farley, and Karl-Erik McCullough (eds.), Chicago
Linguistic Society 22, Part 2: Papers from the Parasession on Pragmatics and
Grammatical Theory, Chicago: CLS, pp. 152–167.
Lakoff, George and Stanley Peters 1966. “Phrasal conjunction and symmetric predi-
cates.” Mathematical Linguistics and Automatic Translation, Harvard Computation
Laboratory, Report No. NSF-17, VI, pp. 1–49 (reprinted in 1969).
1969. “Phrasal Conjunction and Symmetric Predicates,” in David A. Reibel and
Sanford A. Schane (eds.), Modern Studies in English, Englewood Cliffs, NJ:
Prentice-Hall, pp. 113–142.
Landau, Idan 1999. “Possessor raising and the structure of VP,” Lingua 107: 1–37.
Lang, Ewald 1984. The Semantics of Coordination. Amsterdam: John Benjamins.
Larson, Richard 1988. “On the double object construction,” Linguistic Inquiry 19:
335–391.
1990. “Double objects revisited: reply to Jackendoff,” Linguistic Inquiry 21(1): 589–
632.
Lasersohn, Peter 1995. Plurality, Conjunction and Events. Dordrecht: Kluwer Academic
Publishers.
References 257
2000. “Same, models and representation,” Proceedings of SALT X, Ithaca, NY: CLC
Publications, pp. 83–97.
Law, Paul 2002. “Adjunct relative clauses in Chinese,” in Zoe Wu (ed.) Proceedings of
the 13th Annual Meeting of the North American Conference on Chinese Linguistics.
Los Angeles: University of Southern California, pp. 80–97.
Lawler, John 1974. “Ample negatives,” Papers from the Tenth Regional Meeting (CLS).
Chicago: Chicago Linguistic Society, pp. 357–377.
Legate, Julie 2008. “Morphological and abstract case,” Linguistic Inquiry 39: 55–101.
Levin, Beth 1993. English Verb Classes and Alternations: A Preliminary Investigation,
Chicago: University of Chicago Press.
Levin, Nancy and Ellen Prince 1986. “Gapping and Causal Implicature,” Papers in
Linguistics 19(3): 351–364.
Levine, Robert D. 2001. “The extraction riddle: just what are we missing?,” Journal of
Linguistics 37: 145–174.
Li, Charles and Sandra Thompson 1981. Mandarin Chinese: A Functional Reference
Grammar. Berkeley: University of California Press.
Li, Ya-fei 1990. “On V–V compounds in Chinese,” Natural Language and Linguistics
Theory 8: 177–207.
Li, Yen-hui Audrey 1998. “Argument determiner phrases and number phrases,” Lin-
2007. “De: adjunction and conjunction,” talk presented at NACCL-19/IACL 15,
Columbia University, May 25–27, 2007.
Lin, Chien-Jer Charles 2007. “Processing (in)alienable possessions at the syntax–
semantics interface,” talk presented at the conference On Linguistic Interfaces,
University of Ulster, Northern Ireland, June 1–3, 2007.
Lin, Jo-Wang 2003. “On restrictive and non-restrictive relative clauses in Mandarin
Chinese,” Tsinghua Journal of Chinese Studies, New Series, 33(1): 199–240.
Lin, Jo-Wang and Chih-Chen Jane Tang 1996. “Modals as verbs in Chinese: a GB
perspective,” in The Bulletin of the Institute of History and Philology, Taipei:
Academia Sinica, vol. 66, pp. 53–105.
Lin, Vivian 2001. “A way to undo A-movement,” in K. Megerdoomian and L. A.
Barel (eds.), WCCFL 20 Proceedings, Somerville, MA: Cascadilla Press, pp. 358–
371.
2002. “Coordination and sharing at the interfaces.” PhD diss., MIT, Cambridge, MA.
Link, Godehard 1983. “The logical analysis of plurals and mass terms: a lattice-
theoretical approach,” in R. Bäuerler, Chr. Schwarze, A. von Stechow (eds.),
Meaning, Use and Interpretation of Language, Berlin and New York: De Gruyter,
pp. 303–323.
1984. “Hydras: on the logic of relative constructions with multiple heads,” in Fred
Landmann and Frank Veltmann (eds.), Varieties of Formal Semantics, Dordrecht:
Foris, pp. 245–257.
Liu, Jian and Alain Peyraube 1994. “History of some coordinative conjunctions in
Chinese,” Journal of Chinese Linguistics 22(2): 179–201.
Lobeck, Anne 1987a. “Syntactic constraints on ellipsis,” PhD diss., University of
Washington.
258 References
1987b. “VP ellipsis in infinitives: infl as a proper governor,” in J. McDonough and

B. Plunkett (eds.), Proceedings of NELS 17, GLSA, University of Massachusetts,
Amherst, pp. 425–441.
1990. “Functional heads as proper governors,” in Proceedings of the North
East Linguistic Society, vol. 20, Amherst: GLSA, University of Massachusetts,
pp. 348–362.
López, Luis 2001. “Head of a projection,” Linguistic Inquiry 32: 521–532.
Lorimor, Heidi 2007. “Conjunctions and grammatical agreement,” PhD diss., University
of Illinois at Urbana-Champaign.
Lü, Shuxiang 1979. Hanyu yufa fenxi wenti [Issues in the analysis of Chinese grammar].
Beijing: Shangwu Press.
Lü, Shuxiang, Linding Li, Jian Liu, et al. 1999. Xiandai Hanyu Babai Ci [800 Words
in Chinese]. Beijing: Shangwu Press (1st edition, 1980).
Lubbe, Henricus Franciscus Alphonsus Van Der 1958. “Woordvolgorde in het Neder-
lands; een synchrone structurele beschouwing,” diss. Utrecht. Assen [summary in
English].
Luka, Barbara and Lawrence Barsalou 2005. “Structural facilitation: mere exposure
effects for grammatical acceptability as evidence for syntactic priming in compre-
hension,” Journal of Memory and Language 52: 436–459.
Marantz, Alec 1997. “No escape from syntax: don’t try morphological analysis in the
privacy of your own lexicon,” U. Penn Working Papers in Linguistics 4(2): 201–
225.
Martin, Samuel Elmo and Yong-Sook C. Lee 1986. Beginning Korean. Tokyo:
Tuttle.
Marušic, Franc Lanko, Andrew Nevins, and Amanda Saksida 2007. “Last-conjunct
agreement in Slovenian,” in Richard Compton, Magdalena Goledzinowska, Ulyana
Savchenko (eds.), Formal Approaches to Slavic Linguistics 15: The Toronto Meet-
ing 2006. Ann Arbor: Michigan Slavic Publications, pp. 210–227.
Massey, Gerald J. 1976. “Tom, Dick, and Harry, and all the king’s men,’ American
Philosophical Quarterly 13: 89–107.
Matushansky, Ora 2006. “Head movement in linguistic theory,” Linguistic Inquiry 37:
69–109.
May, Robert 1985. Logical Form: Its Structure and Derivation. Cambridge, MA: MIT
Press.
McCawley, James 1968. “The role of semantics in grammar,” in Emmon Bach and
Robert Harms (eds.), Universals in Linguistic Theory. New York: Holt, Rinehart
and Winston, pp. 124–169.
1988a. The Syntactic Phenomena of English, Chicago: Chicago University Press.
1988b. “The comparative conditional constructions in English, German and Chinese,”
in Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society,
Berkeley: BLS, pp. 176–187.
McCloskey, James 1986. “Inflection and conjunction in modern Irish,” Natural Lan-
guage and Linguistic Theory 4: 245–281.
1991. “There, it, and agreement,” Linguistic Inquiry 22: 563–567.
1999. “Embedding the root,” talk presented at MIT, Nov. 12, 1999.
References 259
McCloskey, James and Ken Hale 1984. “On the syntax of person number marking in
modern Irish,” Natural Language and Linguistic Theory 1: 487–533.
McFadden, Thomas 2004. “The position of morphological case in the derivation:
a study on the syntax–morphology interface,” PhD diss., The University of
Pennsylvania.
McNally, Louise 1992. “VP-coordination and the VP-internal subject hypothesis,” Lin-
1993. “Comitative coordination: a case study in group formation,” Natural Language
and Linguistic Theory 11: 347–379.
Merchant, Jason 2001. The Syntax of Silence. New York: Oxford University Press.
Mithun, Marianne 1988. “The grammaticalization of coordination,” in J. Haiman and
S. A. Thomson (eds.), Clause Combining in Grammar and Discourse, Amsterdam:
John Benjamins, pp. 331–360.
Mittwoch, Anita 1979. “Backward anaphora in utterances conjoined with but,” ms.,
The Hebrew University.
Moltmann, Friederike 1992a. “Coordination and comparatives,” PhD diss., MIT.
1992b. “Reciprocals and same/different: towards a semantic analysis,” Linguistics
and Philosophy 16: 411–462.
1997. Parts and Wholes in Semantics. New York, Oxford: Oxford University Press.
Moro, Andrea 1997. The Raising of Predicates. Cambridge: Cambridge University
Press.
2000. Dynamic Antisymmetry. Cambridge, MA: MIT Press.
Morrill, Glyn 1990. “Grammar and logical types,” Proceedings of the Seventh Amster-
dam Colloqium, pp. 429–450.
Muadz, Husni 1991. “Coordinate structure: a planar representation,” PhD diss., Univer-
sity of Arizona.
Müller, Gereon 1998. Incomplete Category Fronting: A Derivational Approach to Rem-
nant Movement in German. Dordrecht: Kluwer.
Munn, Alan 1987. “Coordinate structure and X-bar theory,” McGill Working Papers in
Linguistics 4.1: 121–140.
1992. “A null operator analysis of ATB gaps,” The Linguistic Review 9: 1–26.
1993. “Topics in the syntax and semantics of coordinate structures,” PhD diss.,
University of Maryland, College Park.
1996. “Some problems of coordination,” talk presented at Max-Planck Gesellschaft
ASG, Berlin, June 5, 1996.
1999. “On the identity requirement of ATB extraction,” Natural Language Semantics
7: 421–425.
2001. “Explaining parasitic gap restrictions,” in P. Cullicover and P. Postal (eds.),
Parasitic Gaps, Cambridge, MA: MIT Press, pp. 369–392.
Myers, James 2009. “The design and analysis of small-scale syntactic judgment exper-
iments,” Lingua 119: 425–444.
Napoli, Donna Jo 1993. Syntax: Theory and Problems. New York: Oxford University
Press.
Neijt, Anneke 1979. Gapping: A Contribution to Sentence Grammar. Dordrecht: Foris
Publications.
260 References
Nida, Eugene A. 1949. Morphology: The Descriptive Analysis of Words. Ann Arbor:
University of Michigan Press.
Ning, Chunyan 1993. “The overt syntax of relativization and topicalization in Chinese,”
PhD diss., University of California, Irvine.
Nunes, Jairo 2001. “Sideward movement,” Linguistic Inquiry 32: 303–344.
2004. Linearization of Chains and Sideward Movement, Cambridge, MA: MIT Press.
Nunes, Jairo and Juan Uriagereka 2000. “Cyclicity and extraction domain,” Syntax 3(1):
20–43.
Ochi, Masao 1999. “Some consequences of Attract F,” Lingua 109: 81–107.
Ogawa, Yoshiki 2001. “The stage/individual distinction and (in)alienable possession,”
Language 77.1: 1–25.
Oirsouw, Robert van 1987. The Syntax of Coordination. New York: Croom Helm.
Panagiotidis, Phoevos 2007. “Determiner spreading as DP-predication.” Ms. University
Of Cyprus.
Partee, Barbara H. 1970. “Negation, conjunction, and quantifiers: syntax vs. semantics,”
Foundations of Language 6: 153–165.
2005, “Reflections of a formal semanticist as of Feb 2005,” ms., University of
Massachusetts, Amherst.
Payne, John 1985. “Complex Phrases and Complex Sentences,” in Timothy Shopen
(ed.), Language Typology and Syntactic Description, Cambridge: Cambridge Uni-
versity Press, pp. 3–41.
Pesetsky, David 1982. “Paths and categories,” PhD diss., MIT.
2000. Phrasal Movement and its Kin, Cambridge, MA: MIT Press.
Peterson, Peter G. 2004. “Coordination: consequences of a lexical–functional account,”
Natural Language and Linguistic Theory 22: 643–679.
Peterson, David A. and Kenneth VanBik 2004. “Coordination in Hakha Lai (Tibeto-
Burman),” in Martin Haspelmath (ed.), Coordinating Constructions, Typological
Studies in Language 58, Amsterdam: John Benjamins, pp. 333–356.
Phillips, Colin 2003. “Linear order and constituency,” Linguistic Inquiry 34: 37–90.
Platzack, Christer 2008. “Parametrized argument structure,” ms., Lund University.
Pollard, Carl and Ivan Sag 1994. Head-Driven Phrase Structure Grammar. Chicago:
University of Chicago Press.
Postal, Paul Martin 1972. “The two remarks on dragging,” Linguistic Inquiry 3: 130–
136.
1974. On Raising: One Rule of English Grammar and its Theoretical Implications,
Cambridge, MA: MIT Press.
1998. Three Investigations of Extraction, Cambridge, MA: MIT Press.
Potts, Christopher 2002. “The syntax and semantics of as-parentheticals,” Natural Lan-
Progovac, Ljiljana 1998a. “Structure of coordination, Part 1,” GLOT International 3(7):
3–6.
1998b. “Structure of coordination, Part 2,” GLOT International 3(8): 3–9.
Pullum, Geoffrey and Arnold Zwicky 1986. “Phonological resolution of syntactic feature
conflict,” Language 62: 751–773.
Radford, Andrew 1997. Syntactic Theory and the Structure of English: A Minimalist
Approach. Cambridge: Cambridge University Press.
References 261
Rebuschi, Georges 2005. “Generalizing the antisymmetric analysis of coordination to

nominal modification,” Lingua 115(4): 445–459.
Reinhart, Tanya 2002. “The theta system: an overview,” Theoretical Linguistics 28:
229–290.
Reinhart, Tanya and Mats Rooth 1991. “Bare argument ellipsis.” ms. Tel-Aviv University
and University of Stuttgart.
Richards, Norvin 1997. “What moves where when in which language?,” PhD diss.,
MIT.
Riemsdijk, Henk van and Edwin Williams 1986. Introduction to the Theory of Grammar,
Cambridge, MA: MIT Press.
Roberts, John R. 1988. “Amele switch-reference and the theory of grammar,” Linguistic
Inquiry 9: 45–63.
Rögnvaldsson, Eirı́kur 1982. “We need (some kind of) a rule of conjunction reduction,”
1993. “Coordination, ATB-extraction, and the identification of pro,” Harvard Working
Papers in Linguistics 3: 153–180.
Ross, John Robert 1967. “Constraints on variables in syntax,” PhD diss., MIT.
Ross, John Robert and David Perlmutter 1970. “Relative clauses with split antecedents,”
Linguistic Inquiry 1: 350.
Rothstein, Susan D. 1991a. “Heads, projections and category determination,” in
K. Leffel and D. Bouchard (eds.), Views on Phrase Structure, Dordrecht: Kluwer,
pp. 97–112.
1991b. “Syntactic licensing and subcategorization,” in S. Rothstein (ed.), Syntax and
Semantics, vol. XXV, Academic Press, Inc., pp. 139–157.
Rubin, Edward J. 2002. “The structure of modifiers,” ms., University of Utah.
2003. “Determining pair-merge,” Linguistic Inquiry 34: 660–668.
Rudin, Catherine 1988. “On multiple questions and multiple wh-fronting,” Natural
Runner, Jeffrey T. 1995. “Noun phrase licensing and interpretation,” PhD diss., Univer-
sity of Massachusetts at Amherst.
Ruys, Eddy 1992. “The scope of indefinites,” PhD diss., Utrecht: OTS dissertation
series.
Sabbagh, Joseph 2007. “Ordering and linearizing rightward movement,” Natural Lan-
Sabel, Joachim 2002. “A minimalist analysis of syntactic islands,” Linguistic Review
19: 271–315.
Sadler, Louisa 1999. “Non-distributive features and coordination in
Welsh”, in Miriam Butt and Tracy Holloway King (eds.), Proceed-
ings of the LFG ’99 Conference, Stanford, CA: CSLI Publications.
http://cslipublications.stanford.edu/LFG/4/lfg99-toc.html.
2003, “Coordination and asymmetric agreement in Welsh,” in Miriam Butt and Tracy
Holloway King (eds.), Nominals: Inside and Out, CSLI Publications, pp. 85–118.
Sag, Ivan 1982. “Coordination, extraction, and generalized phrase structure grammar,”
1997. “English relative clause constructions,” Journal of Linguistics 33(2): 421–484.
2000. “Another argument against wh-trace,” http://ling.ucsc.edu/Jorge/sag.html.
262 References
Sag, Ivan, Gerald Gazdar, Thomas Wasow, and Steven Weisler 1985. “Coordination
and how to distinguish categories,” Natural Language and Linguistic Theory 3:
117–171.
Saiki, Mariko 1985. “On the coordination of gapped constituents in Japanese,” CLS 21,
pp. 371–387.
Saito, Mamoru and Keiko Murasugi 1999. “Subject predication within IP and DP,” in
K. Johnson and I. Roberts (eds.), Beyond Principles and Parameters, Dordrecht:
Kluwer, pp. 167–188.
Saito, Mamoru, T.-H. Jonah Lin, Keiko Murasugi 2008. “N-ellipsis and the structure
of noun phrases in Chinese and Japanese.” Journal of East Asian Linguistics 17:
247–271.
Sauerland, Uli 2003. “A new semantics for number,” in R. Young and Y. Zhou
(eds.), Semantics and Linguistic Theory XIII, Ithaca: Cornell University, pp. 258–
275.
Scha, Remko 1981. “Distributive, collective and cumulative quantification,” in J. Groe-
nendijk, T. M. V. Janssen, and M. Stokhof (eds.), Formal Methods in the Study of
Language, Part 2, vol. 136 of Mathematical Centre Tracts, Amsterdam: Mathema-
tisch Centrum, pp. 483–512.
Schachter, Paul 1977. “Constraints on coordination,” Language 53: 86–103.
1985. “Parts-of-speech systems,” in Timothy Shopen (ed.), Language Typology and
Syntactic Description, vol. I, Cambridge: Cambridge University Press, pp. 3–61.
Schein, Barry (forthcoming). Conjunction Reduction Redux, Cambridge, MA: MIT
Press.
Schmerling, Susan 1975. “Asymmetric conjunction and rules of conversation,” in
P. Cole and J. L. Morgan, (eds.), Syntax and Semantics 3: Speech Acts, New
York: Academic Press, pp. 210–231.
Schütze, Carson 2001. “On the nature of default case,” Syntax 4(3): 205–238.
Schütze, Carson and Edward Gibson 1999. “Argumenthood and English prepositional
phrase attachment,” Journal of Memory and Language 40: 409–431.
Schwarz, Bernhard 1999. On the Syntax of Either . . . Or. Natural Language and Lin-
guistic Theory 17, 339–370.
Schwarzschild, Roger 1996. Pluralities. Dordrecht: Kluwer Academic Publishers.
2001. “Review of Winter (1998) Flexible Boolean Semantics: Coordination, Plurality
and Scope in Natural Language,” Glot International 5(4): 141–149.
2002. “The grammar of measurement,” ms., Rutgers University.
Seiler, Hansjakob 1974. “The principle of concomitance: instrumental, comitative, and
collective,” Foundations of Language 12(2): 215–247.
Shaer, Benjamin 2003. “‘Manner’ adverbs and the association theory: some problems
and solutions,” in Ewald Lang, Claudia Maienborn and Cathrin Fabricius-Hansen
(eds.), Modifying Adjuncts, Berlin: Mouton de Gruyter, pp. 211–259.
Shao, Jingmin and Chunhong Rao 1985. “On the conjunction you,” Yuyan Jiaoxue yu
Yanjiu [Language teaching and studies] 2: 4–16.
Shen, David Ta-Chun and Jen Ting 2008. “Morphological status of de in the V-de con-
structions in Mandarin Chinese: a study of interface between syntax and morphol-
ogy.” Paper presented at the 11th International Symposium on Chinese Language
References 263
and Linguistics, National Chia Tung University, Hsinchu, Taiwan, May 23–25,
2008.
Shi, Youwei 1986. “Hanyu lianci de gongneng, jiexian he weizhi” [the functions, criteria
and positions of conjunctions] Zhongyang Minzu Xueyuan Xuebao, 1986.3.
Shopen, Timothy 1971. “Caught in the act,” in Papers from the Seventh Regional
Meetings of the Chicago Linguistic Society, pp. 254–263.
Shyu, Shu-ing 1995. “The syntax of focus and topic in Mandarin Chinese,” PhD diss.,
USC Los Angeles, CA.
Siloni, Tal. 2008. “The syntax of reciprocal verbs: an overview,” in E. König and
V. Gast (eds.), Reciprocals and Reflexives: Theoretical and Typological Explo-
rations, Trends in Linguistics, Berlin: Mouton de Gruyter.
Simpson, Andrew 2003. “On the status of ‘modifying’ de and the structure of the
Chinese DP,” in Sze-Wing Tang and Chen-Sheng Liu (eds.), On the Formal Way
to Chinese Languages. Stanford: CSLI, pp. 74–101.
Skrabalova, Hana 2003. “Coordination: some evidence for DP and NumP in Czech,”
paper presented at the Fifth European Conference on Formal Description of Slavic
Languages (FDSL 5), University of Leipzig, Germany, November 26–28, 2003.
Sledd, James 1959. A Short Introduction to English Grammar. Chicago: Scott, Foresman
and Co.
Sobin, Nicholas 2004. “Expletive constructions are not ‘Lower Right Corner’ movement
constructions,” Linguistic Inquiry 35: 503–508.
Speer, S. R. and C. J. Clifton 1998. “Plausibility and argument structure in sentence
comprehension,” Memory and Cognition 26: 965–978.
Sperber, Dan and Deirdre Wilson 1995. Relevance: Communication and Cognition
(second edn). Oxford: Blackwell Publishers Ltd.
Sportiche, Dominique 1988. “A theory of floating quantifiers and its corrolaries for
constituent structure,” Linquistic Inquiry 19(3): 425–49.
Stassen, Leon 2000. “AND-languages and WITH-languages,” Linguistic Typology 4:
1–54.
Steiner, Ilona 2008. “Partial agreement in German: a processing issue?,” talk presented
at Linguistic Evidence, Tübingen, Feb. 1, 2008.
Stepanov, Arthur 2001. “Late adjunction and minimalist phrase structure,” Syntax 4:
94–125.
Stowell, Timothy 1981. “Origins of phrase structure,” PhD diss., MIT.
Szabolcsi, Anna 1983. “The possessor that ran away from home,” The Linguistic Review
3: 89–102.
1994. “The noun phrase,” in F. Kiefer and K. É. Kiss (eds.), Syntax and Semantics
27: The Syntactic Structure of Hungarian, San Diego: Academic Press, pp. 179–
274.
2001. “The syntax of scope,” in M. Baltin and C. Collins (eds.), The Handbook of
Contemporary Syntactic Theory, Malden and Oxford: Blackwell, pp. 607–633.
Szabolcsi, Anna and Marcel den Dikken 1999. “Islands,” Glot International 4(6): 3–8.
Tai, James H.-Y. 1969. “Coordination reduction,” PhD diss., Indiana University.
Takahashi, Daiko 1994. “Minimality of movement,” PhD diss., University of Connecti-
cut, Storrs.
264 References
Takano, Yuji 2002. “Surprising constituents,” Journal of East Asian Linguistics 11:
243–301.
2004. “Coordination of verbs and two types of verbal inflection,” Linguistic Inquiry
35: 168–178.
Tang, Chih-Chen Jane 1990. “Chinese Phrase Structure and the Extended X’-theory,”
PhD diss., Cornell University.
Tang, Ting-chi 1979. Studies in Chinese Syntax. Student Book Co., Ltd.: Taipei.
Taylor, Heather 2006. “Moving out of if-clauses: If an if-clause is sentence-initial . . . ,”
talk presented at GLOW 29, Barcelona, April 5–8, 2006.
Teng, Shou-hsin 1970. “Comitative versus phrasal conjunction,” Papers in Linguistics
2(2): 314–358.
Thiersch, Craig 1985. “VP and Scrambling in the German Mittelfeld,” ms., University
of Tilburg.
Tomioka, Satoshi 2003. “The semantics of Japanese null pronouns and its cross-
linguistic implications,” in Kerstin Schwabe and Susanne Winkler (eds.), The
Interfaces: Deriving and Interpreting Omitted Structures, Amsterdam: John
Benjamins, pp. 321–339.
Tsai, Wei-tien Dylan 1994. “On economizing the theory of A-bar dependencies,” PhD
diss. MIT.
Tsao, Feng-fu 1996. “hanyu de tisheng dongci” [Chinese raising verbs], Zhongguo
Yuwen 1996(3): 172–182.
Ura, Hiroyuki 2000. Checking Theory and Grammatical Functions in Universal Gram-
mar. New York: Oxford University Press.
Vergnaud, Jean Roger 1974. “French relative clauses,” PhD diss., MIT. (Revised version
in 1985, Dépendances et niveaux de représentation en syntaxe. Amsterdam: John
Benjamins.)
Vicente, Luis to appear. “On the syntax of adversative coordination,” Natural Language
and Linguistic Theory.
Vries, Mark de 2002. “The syntax of relativization,” PhD diss., Universiteit van
Amsterdam.
2005. “Coordination and syntactic hierarchy,” Studia Linguistica 59: 83–105.
2006. “The syntax of appositive relativization: on specifying coordination, false free
relatives, and promotion,” Linguistic Inquiry 37(2): 229–270.
Wälchli, Bernhard 2003. “Co-compounds and natural coordination,” PhD diss., Univer-
sity of Stockholm.
2005. Co-Compounds and Natural Coordination. Oxford: Oxford University Press.
Webelhuth, Gert 1989. “Syntactic saturation phenomena and the modern Germanic
languages,” PhD diss., University of Massachusetts, Amherst.
1992. Principles and Parameters of Syntactic Saturation. Oxford: Oxford University
Press.
Welsche, Birgit 1995. Symmetric Coordination. Tübingen: Max Niemeyer.
Whitman, Philip Neal 2002. “Category neutrality: a type-logical investigation,” PhD
diss., The Ohio State University.
2004. “Semantics and pragmatics of English verbal dependent coordination,” Lan-
guage 80: 403–434.
References 265
Wilder, Chris 1994. “Coordination, ATB and ellipsis,” in Cornelius Jan-Wouter Zwart
(ed.), Minimalism and Kayne’s Asymmetry Hypothesis, Groningen [Groninger
Arbeiten zur Germanischen Linguistik 37], pp. 291–331.
1997. “Some properties of ellipsis in coordination,” in Artemis Alexiadou and T. Alan
Hall (eds.), Studies on Universal Grammar and Typological Variation, Amsterdam:
John Benjamins, pp. 59–107.
1999. “The syntax of coordination,” Handout of Linguistic Summer School, Potsdam
University.
2008. “Shared constituents and linerarization,” in Kyle Johnson (ed.), Topics in Ellip-
sis, Cambridge: Cambridge University Press, pp. 229–258.
Williams, Edwin S. 1977. “Across-the-Board application of rules,” Linguistic Inquiry
8: 419–423.
1978. “Across-the-Board rule application,” Linguistic Inquiry 9: 31–43.
1980. “Predication,” Linguistic Inquiry 11: 203–238.
1989. “Maximal projections in words and phrases,” in Mark Baltin and Anthony
Kroch (eds.), Alternative Conceptions of Phrase Structure, Chicago: The Univer-
sity of Chicago Press, pp. 280–291.
Winter, Yoad 2001. Flexibility Principles in Boolean Semantics: The Interpretation of
Coordination, Plurality, and Scope in Natural Language. Cambridge, MA: MIT
Press.
2006. “Multiple coordination: meaning composition vs. the syntax–semantics inter-
face,” ms., Technion/NIAS.
Woolford, Ellen 1987. “An ECP account of constraints on Across-The-Board extrac-
tion,” Linguistic Inquiry 18: 166–171.
Wu, Jianxin 1999. “Syntax and semantics of quantification in Chinese,” PhD diss.,
University of Maryland at College Park.
Wurmbrand, Susi 2001. “Agree: the other VP-internal subject hypothesis,” in
K. Megerdoomian and L. A. Barel (eds.), Proceedings of WCCFL 20, Somerville,
MA: Cascadilla Press, pp. 635–648.
2008. “Nor: neither disjunction nor paradox,” Linguistic Inquiry 39: 511–522.
Yamada, S. and I. Igarashi 1967. “Co-ordination in transformational grammar,”
Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 20:
143–156.
Yngve, Victor H. 1960. “A model and a hypothesis for language structure,” Proceedings
of the American Philosophical Society 104: 444–466.
Yoon, James Hye and Wooseung Lee 2005. “Conjunction reduction and its consequences
for noun phrase morphosyntax in Korean,” in John Alderete, Chung-hye Han, and
Alexei Kochetov (eds.), Proceedings of the 24th West Coast Conference on Formal
Linguistics, Somerville, MA: Cascadilla Proceedings Project, pp. 379–387.
Yu, Chiung-Yi 2008. “The processing of natural and accidental coordination,” MA
thesis, National Chung Cheng University.
Yuasa, Etsuyo and Jerry M. Sadock 2002. “Pseudo-subordination: a mismatch between
syntax and semantics,” Journal of Linguistics 38: 87–111.
Zagona, Karen 1988a. “Proper government of antecedentless VPs in English and Span-
ish,” Natural Languages and Linguistic Theory 6: 95–128.
266 References
1988b. Verb Phrase Syntax: A Parametric Study of English and Spanish. Dordrecht:
Kluwer.
Zepeda, Ofelia 1983. A Papago Grammar. Tucson: University of Arizona Press.
Zhang, Niina Ning 1999. “Chinese de and the de-construction,” Syntaxis 2, 27–49.
2004a. “Move is remerge,” Language and Linguistics 5(1): 189–209.
2004b. “Against Across-the-Board Movement,” Concentric: Studies in Linguistics
30, 123–156.
2006. “On the configuration issue of coordination,” Language and Linguistics 7(1):
175–223.
2007a. “The syntax of English comitative constructions,” Folia Linguistica 41: 135–
169.
2007b. “The syntactic derivations of two paired dependency constructions,” Lingua
117: 2134–2158.
2007c. “A syntactic account of the direct object restriction in Chinese,” Language
Research 43 (1): 53–75.
2008a. “Repetitive and correlative coordinators as focus particles parasitic on coor-
dinators,” SKY Journal of Linguistics 21: 295–342.
2008b. “The syntactic derivations of split variable constructions,” ms. National Chung
Cheng University.
2008c. “Gapless relative clauses as clausal licensers of relational nouns,” Language
and Linguistics 9(4): 1003–1026.
2009. “Explaining the immobility of conjuncts,” Studia Linguistica 63(3).
Zhou, Gang 2002. Lianci yu xiangguan wenti [Coordinators and the relevant issues]
Hefei: Anhui Jiaoyu Press.
Zhu, Dexi 1982. Yufa Jiangyi [Lectures on grammar], Beijing: Shangwu Press.
Zoerner, Cyril Edward, III. 1995. “Coordination: the syntax of andP,” PhD diss., Uni-
versity of California, Irvine.
Zwart, C. Jan-Wouter 1995. “Review of Johannessen, J. B., Coordination: A Minimalist
Approach,” Glot International 1: 11–13.
2000. “A head raising analysis of relative clauses in Dutch,” in A. Alexiadou,
A. Meinunger, C. Wilder, and P. Law (eds.), The Syntax of Relative Clauses,
Linguistik Aktuell 32, Amsterdam: John Benjamins, pp. 349–385.
Index
Languages Latin 14, 17, 29n5, 98

Lenakel 130
Albanian 100
Amele 24 Maasai 47
Malagasy 15, 47
Barasano 48 Malayalam 91
Bulgarian 231
Norwegian 80, 91
Cayuga 48 Nupe 16, 29n5
Chamorro 39
Chemehuevi 48 Papago 26
Czech 130 Polish 226
Dutch 106, 132 Russian 45, 85, 226
Serbo-Croatian 72, 231

Finnish 45, 47, 132, 187
Sinhala 91
French 86
Sissala 47
Fujian 47
Somali 47
Spanish 39, 80
Georgian 45
German 16n4, 23, 39, 45, 80n1, 91, 127, 135,
Tagalog 45
142, 224n2
Tera 48
Greek 100, 101
Turkish 47
Gungbe 47
Udihe 130
Háusá 48
Hebrew 100, 133, 134, 187 Yapese (Austronesian) 47
Hungarian 14, 15, 29n5, 30, 45
Icelandic 37, 86, 227, 227n3

Persons
Abbott, Barbara 36, 210, 213
Japanese 13, 37, 46, 47, 48, 68, 89–90, 91, Abels, Klaus 86, 86n3, 209
176n8, 221, 229, 230 Abney, Steven 65
Joluo 187 Aboh, Enoch Oladé 47, 97n9
Adger, David 83, 86, 216
Kanada 48 Agbayani, Brian 161
Kikuriya 187 Alexiadou, Artemis 154
Kurdish 98 Amfo, Nana Abe Appiah 33, 35
Korean 47, 133, 229, 230 Amritavalli, R. 48
267
268 Index
Anandan, K. N. 43, 60, 201 Chiang, Wen-yu 230

Anderson, Carol 197 Cho, Sunggeun 229
Androutsopoulou, Antonia 100 Chomsky, Noam 4, 10, 22, 30, 32, 36–38,
Aoun, Joseph 54, 66, 99–100, 131 64n4, 65, 75, 81, 87, 104, 118, 123, 161,
Artstein, Ron 39 174, 178, 186, 214, 240, 242
Chung, Sandra 39
Bach, Emmon 10 Citko, Barbara 226, 236, 238
Baker, Mark 5, 144, 152 Clifton, Charles 195, 196, 198
Baltin, Mark 151 Cole, Peter 57
Bánréti, Zoltán 15, 16 Collins, Christopher 11, 12n2, 23, 28, 43, 56,
Bar-Lev, Zev 127 60, 67, 75, 199
Barker, Chris 209 Cormack, Annabel 135
Barsalou, Lawrence 197 Cowart, Wayne 21, 32n7, 131
Bayer, Samuel 11, 51, 53 Cowper, Elizabeth 23, 26, 38n11, 45
Berchem, Jörg 47 Culicover, Peter W. 31, 127–128, 128n1, 135,
Besten, Hans den 160n5 138, 180, 183, 184
Bever, Thomas G. 34
Bianchi, Valentina 168 Dalrymple, Mary 125, 127, 130, 132
Blakemore, Diane 33 Davies, William D. 82
Blass, Regina 47 De Vos, Mark Andrew 137–138, 175n7
Bloomfield, Leonard 10 Déchaine, Rose-Marie 12, 47, 63, 208, 216,
Blümel, Rudolf 10, 11, 65 221
Bobaljik, Jonathan David 161 Dik, Simon C. 2, 10, 11, 13, 16, 17–19, 22,
Bock, Kathryn 21, 131 33, 35, 43, 45, 49, 66, 71, 126, 128, 131,
Boeckx, Cedric 161 180, 188, 200, 239
Booij, Geert 39 Dikken, Marcel den 31n6, 99, 105, 131, 148,
Borsley, Robert D. 23, 35, 36–38, 43, 54, 60, 201
65, 70, 75, 89, 191, 201, 227n3 Dougherty, Ray C. 10, 61, 62, 67, 70, 72, 74,
Bošković, Željko 20, 49, 64, 207, 229, 231, 132, 169, 171–172
237 Dowty, David 126
Bowers, John 121, 151, 200, 201 Dubinsky, Stanley 82
Bresnan, Joan 37
Eggert, Randall 67
Brown, Sam 161
Elbourne, Paul 186n2
Bruening, Benjamin 172n6
Embick, David 56
Büring, Daniel 62
Emond, Joseph 22
Burton, Strang 200
Evans, Gareth 211, 221, 228
Camacho, José 12n2, 60n2, 66 Fitzpatrick, Justin Michael 83
Caponigro, Ivano 47 Fox, Danny 138n3, 190, 225
Carlson, Greg 136, 208, 209, 211, 213, Frank, Annette 135
218–220 Franks, Steven 20, 207, 229, 231,
Carlson, Katy 25, 196 237–238
Carston, Robyn 33, 35, 195–196 Frazier, Lyn 195, 196, 198
Castillo, Juan Carlos 98 Fromkin, Victoria 23, 44
Chang, Chia-Hao, Frank 100 Fu, Jingqi 93n5
Chao, Yuen Ren 94n6
Chapman, Carol 63 Gawron, Jean Mark 233–236
Chaves, Rui Pedro 21, 27, 134 Gazdar, Gerald 10, 48, 49, 70, 75, 82, 85,
Chen, Qitong 46, 68 139n4, 155, 178, 179, 180, 188, 191, 213,
Cheng, Lisa Lai Shen 99, 148 240
Index 269
Gärtner, Hans-Martin 91 Hyams, Nina 34n9, 97n9

George, Leland 66, 207, 236, 240
Gibson, Edward 195 Igarashi, I. 46, 49
Gleitman, Lila 44, 46, 65, 179
Godard, Danièle 228 Jackendoff, Ray 31, 36, 70, 98, 127–128,
Goldsmith, John 30, 33, 88, 127–128, 128n1, 128n1, 135, 156, 156n4, 157, 180, 183, 184,
135, 138 201, 209, 213, 237
Goodall, Grant 10, 25, 33, 52, 66, 84n2, 179, Jacobson, Pauline 200
180, 190, 197, 222, 236, 238 Jayaseelan, Karattuparambil A. 91
Gracanin-Yuksek, Martina 236, 238 Jaeggli, Osvaldo 34n9, 97n9
Grice, Paul 196 Jensen, John Thayer 47
Grimshaw, Jane 56, 187, 200 Jespersen, Otto 109
Groot, A. W. De 10, 43, 60 Johannessen, Janne Bondi 1, 2, 20, 23, 33, 43,
Grootveld, Marjan 64, 238 44, 47, 48, 49, 51, 54, 60, 63, 65, 66, 70, 80,
Grosu, Alexander 2, 30, 52, 54, 79, 84, 85, 81, 84–85, 131, 136
124, 135, 187 Johnson, Kyle 10, 24, 25, 135, 151, 177, 200,
209, 233n6
Hagstrom, Paul 91
Haı̈k, Isabelle 225 Kandybowics, Jason 16, 29n5, 32n8
Hale, Ken 89 Karimi, Yadgar 98
Hall, Daniel 23, 26, 38n11, 45 Kasai, Hironobu 138n3, 226
Hankamer, Jorge 88, 93n5 Kayne, Richard 1, 36–37, 43, 60, 70, 118,
Harley, Heidi 216 145n2, 153, 161, 168
Hartmann, Katharina 20, 48, 209 Keenan, Edward 47
Haspelmath, Martin 13, 17, 34, 39, 47, 125, Kehler, Andrew 22, 30, 33, 128, 136, 177,
130, 179 183–185, 233
Hegarty, Michael 132 Kiguchi, Hirohisa 161
Heim, Irene 12n2, 24, 25, 151, 152, 185, Kitagawa, Chistato 96
230n5 Koizumi, Masatishi 151
Hendriks, Petra 14, 53, 132, 134, 135, Kolb, Hans-Peter 43, 60
240 Koopman, Hilda 38, 56
Hernández, Ana Carrera 233n6 Koutsoudas, Andreas 2, 21
Heycock, Caroline 38, 45, 55, 69, 73n8, 135, Kratzer, Angelika 12n2, 24, 25, 97n9, 185,
170, 177, 223 186n2
Higginbotham, James 191, 192 Krifka, Manfred 69, 125, 169, 188
Hinds, John 89–90 Kroch, Anthony 135, 177
Hockett, Charles, Francis 65 Kuno, Susumu 47, 49n1, 82
Hoeksema, Jack 125
Hoekstra, Eric 66 Ladusaw, William 39
Hoh, Pau-San 230 Laka, Itziar 64
Höhle, Tilman 23, 127, 134 Lakoff, George 17, 30, 31, 33, 66, 70, 125,
Hornstein, Norbert 121, 161, 164, 175n7, 127, 128n1, 135, 136
177, 198–199, 207, 236, 240 Landau, Idan 55, 83, 146
Huang, Cheng-teh James 31, 114, 119 Lang, Ewald 24, 33, 34, 65, 66, 190, 194
Huang, Chu-Ren 92, 98, 138 Larson, Richard 20, 59
Huddleston, Rodney 2, 21, 23, 33, 39, 127, Lasersohn, Peter 73, 126, 224n2
130, 183, 184–185, 188 Law, Paul 86
Hudson, Richard A. 24, 180 Lawler, John 88
Humphreys, Karin R. 21, 131 Lee, Wooseung 133, 134
Huttar, George 29n5, 88 Lee, Yong-Sook C. 47
270 Index
Legate, Julie 64 Murasugi, Keiko 86n3

Levin, Beth 142, 143, 151 Myers, James 136n2, 175n7
Levin, Nancy 31, 134
Levine, Robert D. 30, 35, 135, 177 Napoli, Donna Jo 81, 85
Li, Charles 47, 93 Neijt, Anneke 24, 25, 93n5
Li, Ya-fei 120n14 Nida, Eugene A. 10
Li, Yen-hui Audrey 94n7, 99–100, 149, 225 Nikolaeva, Irina 125, 127, 130, 132
Lightfoot, David 121 Ning, Chunyan 99–100
Lin, Chien-Jer Charles 195 Noyer, Rolf 56
Lin, Jo-wang 97n8, 114 Nunes, Jairo 161, 164, 175n7, 177, 198–199,
Lin, Vivian 138n3, 177, 200 207, 236, 240
Link, Godehard 125, 156, 157, 164, 169
Liu, Jian 109 Ochi, Masao 87
Lobeck, Anne 86n3, 107 Ogawa, Yoshiki 98
López, Luis 57 Oirsouw, Robert van 230n5
Lorimor, Heidi 2, 21, 130
Lü, Shuxiang 14, 111, 125 Palacas, Arthur 127
Luka, Barbara 197 Panagiotidis, Phoevos 101
Partee, Barbara H. 69n7, 228
Marantz, Alec 38n11 Payne, John 15, 21, 45, 46, 47, 48, 70, 125,
Martin, Samuel Elmo 47 153
Marušic, Franc Lanko 21 Perlmutter, David 154
Massey, Gerald J. 125 Pesetsky, David 35, 38, 43, 60, 75, 81, 139n5,
Matushansky, Ora 99 140, 177, 187
May, Robert 137, 138n3 Peters, Stanley 66, 125
McCawley, James 13, 22, 29, 34, 66, 73, 125, Peterson, David A. 136
134, 169 Peterson, Peter G. 10, 21, 180, 188, 189, 190
McCloskey, James 59, 89, 131–132 Peyraube, Alain 109
McDaniel, Dana 21, 32n7 Phillips, Colin 10, 39
McFadden, Thomas 64 Platzack, Christer 193
McNally, Louise 2, 21, 45, 130, 132, 200, Pollard, Carl 53–54, 55, 81, 83, 189
228 Postal, Paul Martin 3, 5, 22, 30, 34, 39, 79,
Meinunger, Andre 91 80, 88, 128, 128n1, 138, 141, 168, 169–173,
Merchant, Jason 81, 85 177, 194, 210, 211, 213
Mithun, Marianne 48, 109, 125 Potts, Christopher 23, 138n3, 177, 190, 200
Mittwoch, Anita 12n2 Prince, Ellen 31, 134
Moltmann, Friederike 11, 24, 25, 26, 27, Progovac, Ljiljana 12n2, 23, 43, 61n2, 63n3,
32–33, 50, 65, 70, 130–132, 134, 156, 157, 186, 189
158, 164, 166–167, 187, 192, 200, 212, 213, Pullum, Geoffrey 2, 21, 23, 33, 39, 127, 130,
221n2, 223, 225, 229, 233, 235, 236, 180, 183, 184–185, 188
238–239
Moro, Andrea 199n5 Radford, Andrew 44, 46
Morrill, Glyn 188 Ramchand, Gillian 216
Muadz, Husni 10, 33, 236, 238 Rao, Chunhong 88
Müller, Gereon 160n5 Rebuschi, Georges 94n7
Munn, Alan 1, 2, 10, 12n2, 20, 23, 32, 35, 39, Reinhart, Tanya 26, 193
40, 43, 44, 51, 54, 60, 66, 67, 89, 131, 177, Richards, Norvin 28
179, 180, 186, 187, 188, 190, 200, 201, 207, Riemsdijk, Henk van 3, 81
223, 224, 225, 233–234, 235, 236, 237–238, Roberts, John R. 24
240 Rögnvaldsson, Eirı́kur 177, 227
Index 271
Rooth, Mats 26 Szabolcsi, Anna 31n6, 38, 56, 83, 138n3,

Ross, Claudia 96 146
Ross, John Robert 3, 5, 13, 16n4, 30, 31, 34,
70, 79, 81, 124, 135, 154, 164, 177, 178, Tai, James H.-Y. 44, 46, 65, 66, 117, 118,
199, 200, 207, 210, 239, 243 169, 171
Rothstein, Susan D. 54, 100 Takahashi, Daiko 31
Rubin, Edward J. 99 Takano, Yuji 10, 37, 48, 176n8
Rudin, Catherine 138 Tang, Chih-Chen Jane 94–101, 114
Runner, Jeffrey T. 151 Tang, Ting-chi 100
Ruys, Eddy 137, 138n3, 177, 200 Taylor, Heather 164, 175n7
Teng, Shou-hsin 108n12, 115
Sabbagh, Joseph 210 Thiersch, Craig 1, 10, 43, 60
Sabel, Joachim 31 Thompson, Sandra 47, 93
Sadler, Louisa 21 Thráinsson, Höskuldur 37
Sadock, Jerry M. 55 Ting, Jen 98
Sag, Ivan 2, 11, 21, 49, 51–52, 53–54, 55, 60, Tomioka, Satoshi 221, 228n4
65, 69, 81–84, 84n2, 85, 88, 139n4, 157, Tsai, Wei-tien Dylan 138
179, 180, 188, 189, 192 Tsao, Feng-fu 29
Saiki, Mariko 49, 49n1
Saito, Mamoru 86n3, 93n5, 105 Ura, Hiroyuki 75
Sauerland, Uli 130, 132 Uriagereka, Juan 161, 164, 175n7
Scha, Remko 169
Schachter, Paul 10, 13, 34, 128, 135, van der Lubbe, Henricus Franciscus
178–180 Alphonsus 18
Schein, Barry 66, 67, 68, 69n6 VanBik, Kenneth 136
Schmerling, Susan 31, 69, 127 Vergnaud, Jean Roger 118, 156, 157
Schütze, Carson 64, 195 Vicente, Luis 68n5, 175n7
Schwarz, Bernhard 24 Vries, Mark de 9n1, 14, 33, 60, 106
Schwarzschild, Roger 67, 73, 126
Seiler, Hansjakob 115 Wälchli, Bernhard 38, 125
Shaer, Benjamin 135 Webelhuth, Gert 56, 86n3, 160n5
Shao, Jingmin 88 Welsche, Birgit 45
Shen, David Ta-Chun 98 Whitman, Philip Neal 51, 53
Shi, Youwei 14, 29, 88 Wilder, Chris 12n2, 22, 36, 39, 45, 52, 60, 65,
Shopen, Timothy 34n9 66, 135, 157, 166–167, 180, 187, 190, 210,
Shyu, Shu-ing 147 236, 238
Siloni, Tal 151 Williams, Edwin S. 3, 5, 38, 81, 121, 177,
Simpson, Andrew 99, 101, 105 178, 197, 207, 239
Skrabalova, Hana 130 Wilson, Deirdre 196
Sledd, James 19 Wiltschko, Martina 12, 47, 63, 208, 216, 221
Smith, Neil 135 Winter, Yoad 57, 69, 74, 74n9, 75, 133
Sobin, Nicholas 21 Woolford, Ellen 43, 60, 236
Speer, S. R. 195 Wu, Jianxin 230, 231
Sperber, Dan 196 Wurmbrand, Susi 10, 86n3
Sportiche, Dominique 83, 146
Stassen, Leon 47 Yamada, S. 46, 49
Steiner, Ilona 2, 21, 130 Yngve, Victor H. 1, 10, 11
Stepanov, Arthur 31 Yoon, James Hye 133, 134
Stowell, Timothy 36 Yu, Qiung-Yi 195
Sybesma, Rint 148 Yuasa, Etsuyo 55
272 Index
Zagona, Karen 107 CED 31

Zamparelli, Roberto 38, 45, 55, 69, 73n8, CHC (the Coordinate-Head RC Construction)
170, 223 156n4, 156–160, 164
Zepeda, Ofelia 26 CLC (Coordination of Likes Constraint)
Zhang, Niina Ning 14, 14n3, 16n4, 32n8, 34, 178–181, 188, 190–192, 201, 202,
87, 99–100, 105, 107n11, 113n13, 120n14, 244
145n2, 170, 176n8, 186, 234, 236 CLCfunc (CLC with respect to grammatical
Zhou, Gang 29 functions) 180, 188, 189, 201, 202, 244
Zhou, Xuan 229 CLCsem (CLC with respect to semantic types)
Zhu, Dexi 94n6, 109, 110, 111 180, 201, 202, 244
Zoerner, Cyril Edward, III. 1, 2, 14, 20, 23, comitative 4, 5, 107–123, 113n13, 145n2,
26, 34, 43, 44, 48, 49, 60, 64, 70, 80, 81, 84, 152, 244
89, 90, 161, 186, 188, 194n3, 197n4 conjunct 9, External vs. Internal 2, 4, 10, 16,
Zwart, C. Jan-Wouter 70, 168 19–32, 40, 80, 88–90
Zwicky, Arnold 180 conjunct drop 85, 89, 88–90
coordinator 9
coordinator floating 2, 11, 14–16, 26–30
Subjects CSC (Coordinate Structure Constraint) 3–5,
AC (asymmetrical coordination) 31, 31n6, 30–31, 79, 138n3, 243–245, also see CC,
33–34, 128n1, 136n2, 139n5, 124–140, CCe, CCi, EC
179–180, 185n1, 244
accidental coordination 124–140, 181, 184, de construction 92–107, 123, 243
195–196, 202, 244 Downstairs-Twin-Structure 59, 70
Agree 28
agreement 2, 5, 21, 25, 54, 130–132, 147, EC (Element Constraint, the second part of
164, 172 CSC) 3, 30–31, 79, 124, 136n2, 139n4,
A-not-A 118–120 135–140, 141, 168, 173, 175–176, 179, 180,
A-over-A principle 81–83 181, 184, 200, 202, 243, 244
asyndetic coordination 123n16 ellipsis 37, 52, 101–103, 107, 194, Bare
ATB (the Across-the-Board dependency) 5, Argument Ellipsis 24–26, VP ellipsis 6
85, 199, 207–208, 210, 221, 222–241, external conjunct see conjunct
223n1, 227n3, 233n6, 245–246
fake coordination 31n6, 127
Bare Argument Ellipsis see ellipsis feature percolation 56–57, 58
Boolean coordination 125
gapping 6, 25, 73, 93n5, 134, 136n2, 161,
190, 194, 233n6
CC (Conjunct Constraint, the first part of
CSC) 3–4, 30, 79–123, 84n2, 107n11,
Head 154n3
123n16, 124, 139n4, 140, 141, 144, 155,
161, 163, 168, 176, 179, 180, 198, 243, 244
IDC (the Interwoven Dependency
CCC (Coordinate Constituent Constraint)
Construction) 168–176, 175n7, 184, 193,
178–181, 202, 244
CCe (no movement of external conjuncts, part 226
of CC) 80–81, 85, 87, 92, 107–123, 140, Internal conjunct see conjunct
146, 153, 163, 202 island 31, 148, 164, 175n7, 217–218, 226
CCH (Clausal Conjunct Hypothesis) 65–69,
68n5, 69n6, 69n7, 75 MSC (the Modifier-Sharing Construction)
CCi (no movement of internal conjuncts, part 154–168, 175, 176, 192–193
multiple-dimensional analysis 33, 166–167,
of CC) 80–81, 85–87, 88–92, 123, 140,
168, 236, 238
163, 202
Index 273
natural coordination 124–140, 181, 195–196, RNR (Right Node Raising) 6, 39, 67, 194,
202, 244 209
non-Boolean coordination 125 RPR (Relativized PR) 140, 178–203,
244
parataxis 48 SAC (the Split Argument Construction)
PPC (the Paired Pronoun same Construction) 141–154, 142n1, 145n2, 176
210–211, 218, 220 SE (Similarity Expression) 208
PR (Parallelism Requirement) 177–203, SGF-coordination 134
199n5, 220, 221 sideward movement 155, 160–166, 168–169,
173–176, 175n7, 176n8, 214, 236, 240–241
QR (Quantifier Raising) 137–138, 138n3 TLC (the Thematic Licensing same
Construction) 209, 221n2, 223n1, 233n6
reciprocal 144, 151–153, 154, 159, 221n2
reflexive 120n14, 172, 215, 216, 217, 218, Understood Reciprocal Object Alternation
225–226, 227, 240 143
relatedness 181–186, 244 Upstairs-Twin-Structure 55, 59
resemblance 181–182, 186–190, 244, also
see PR VP ellipsis see ellipsis

Coordination in Syntax CAMBRIDGE

Uploaded by

Copyright:

Available Formats

Coordination in Syntax CAMBRIDGE

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Coordination in Syntax CAMBRIDGE

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

Coordination in syntax is an important part of the analysis of sentence struc-

niina ning zhang is Associate Professor and Director in the Graduate

Cambridge University Press

© Niina Ning Zhang 2009

This publication is in copyright. Subject to statutory exception and to the

ISBN-13 978-0-511-77000-5 eBook (NetLibrary)

Cambridge University Press has no responsibility for the persistence or accuracy

Acknowledgments page xvi

part i no special syntactic conf iguration 7

2 The complementation structure of coordinate complexes 9

part ii no special syntactic category 41

3 The categorial makeup of coordinate complexes 43

3.2 The categories of coordinators and conjuncts 44

part iii no special syntactic constraint 77

4 The Conjunct Constraint and the lexical properties

4.3.4 The chameleon-like nature of de keeps the kernel elements

5 The Element Constraint and the semantic relatedness

6 Three puzzles solved by rejecting the CSC 141

7 Relativized parallelism in syntactic complexes 177

part iv no special syntactic operation 205

8 The derivation of coordinate clauses with identity

9 Forming Across-the-Board constructions without forking

In the Chinese examples:

All languages have coordinate constructions. Although generative grammar,

arguments against the binary-branching structure summarized by Dik (1968)

others). I show that &P is both theoretically and empirically problematic. I

categorial features, as in Chinese. Our proposed category feature movement,

(1.4) a. John married Jane.

(1.5) a. Tudou yijing shao-le niurou. [Chinese]

These constructions seem to be in conflict with certain basic syntactic laws

(1.8) a. Who did Jim like and Jane hate?

This alleged operation is specific to coordinate constructions. Unlike all other

(A) Coordinate complexes have a binary-branching structure, and thus one

2.2 The binary-branching constituency of coordinate complexes

conjunct coordinator conjunct

structure in (2.1b) under economy considerations (Yngve 1960: 453; Collins

2.2.1 The asymmetry between conjuncts in binding

(2.2) a. Every mani and hisi dog left.

Likewise, in each of the following examples, the second conjunct contains a

(2.3) a. [That Himmler appointed Heydrich] and [the implications

2.2.2 The asymmetry between conjuncts in possessee pronominalization

(2.4) a. Sally’s mother and yours have turned vegetarian.

(2.5) a. Sally’s mother praised yours.

Subjects c-command objects. Data like (2.5) lead us to the generalization

2.2.3 The asymmetry between conjuncts in hosting coordinators

2.2.4 The asymmetry between conjuncts in coordinator floating

(2.9) a. Baoyu yao tiaowu, ke(shi) wo yao hui-jia.

(2.11) a. Baoyu yi guli, yushi Daiyu huifu-le xinxin.

My syntactic analysis of the floating of the two coordinators will be presented

2.2.5 Conclusion and Dik’s challenges

(2.13) a. (the right-branching type)

Note that in these structures, what is shown is the constituency of sub-

In general, not all coordinate complexes show syntactic parallelism.

on syntactic operations, and thus it has no effect on the structure-

Dik’s challenge is based on a contrast observed by Sledd (1959:

Dik uses the constraint that coordinators cannot be moved together

2.3 The complementation structure of coordinate complexes

a. (coordinate complex) b. XP (coordinate complex)

(coordinator) (internal conjunct) (coordinator) (internal conjunct)