The Meaning of Chains

by
Uli Sauerland
Submitted to the Department of Linguistics and Philosophy
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
August 1998
°
c Massachusetts Institute of Technology 1998. All rights reserved.

Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Department of Linguistics and Philosophy
August 31, 1998
Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Noam Chomsky
Institute Professor of Linguistics
Thesis Supervisor
Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Irene Heim
Professor of Linguistics
Thesis Supervisor
Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
David Pesetsky
Professor of Linguistics
Thesis Supervisor
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Michael Kenstowicz
Chair of the Linguistics Program
 The Meaning of Chains

by

Uli Sauerland

Submitted to the Department of Linguistics and Philosophy

on August 31, 1998, in partial fulfillment of the
requirements for the degree of Doctor of Philosophy

Abstract
This thesis investigates the mechanisms applying in the interpretation of syntactic
chains. The theoretical background includes a translation of syntactic forms into se-
mantic forms and a model theoretic explication of the meaning of semantic forms.
Simplicity considerations apply to all three stages of the interpretation process: syn-
tactic derivation, translation into semantic forms, interpretation of semantic forms.
Three main results are achieved. The first is that trace positions can have
semantic content beyond what is needed for the semantic dependency of trace and
binder. This extra content is some or all of the lexical material of the head of the
chain, as expected on the copy theory of movement. Two independent arguments
support this conclusion. One, discussed in chapter 2, is based on the distribution of
Condition C effects, where novel interactions between variable binding, antecedent
contained deletion and Condition C are observed. The second, developed in chapter
3, is based on conditions on the identity of traces observed in antecedent contained
deletion constructions. Both arguments lead to the same generalizations about what
lexical material of the head is interpreted in the trace position.
The second main result is that lambda calculus is superior to both standard
predicate logic and combinatorial logic as the mathematical model for the seman-
tic mechanism mediating the dependency of trace (or bound pronoun) and binder.
Chapter 4 argues this on the basis of the distribution of focus and destressing in
constructions with bound pronouns.
The third main result is that quantification must be allowed to range over
pointwise different choice functions. Chapter 5 shows that quantification over indi-
viduals is insufficient, and that pointwise different choice functions are required. The
result entails that the syntactic difference of A-chains and A-bar chains predicts a
semantic difference in the type of the variable involved, which is argued to explain
weak crossover phenomena.
Chapters 6 argues that the interpretation procedures developed in the pre-
ceeding chapters account for all cases. It is shown that only traces of the type of
individuals arise, and that scope reconstruction is a phonological phenomenon. The
latter result also supports the T-model of syntax.
Acknowledgments

I thank all the people who helped me on the way towards finishing this thesis. All three

members of my committee had a great impact on this thesis and on my thinking more

generally. David Pesetsky helped me clear up my thinking in countless meetings and

put it into writing. Irene Heim was always quick at finding interesting new challenges

for me and had usually more faith that I could meet them than I did. Noam Chomsky

expanded my vision of the bigger picture and helped me to sharpen many technical

details at the same time.

The intellectual environment this thesis grew in was provided by two, over-

lapping groups of people, the LF reading group at MIT and the inner hallway of

building E39. Danny Fox, especially, was enormously helpful. Also, Jon Nissenbaum,

Paul Hagstrom, Martin Hackl, Kai von Fintel, and Kazuko Yatsushiro caused many

improvements. I also thank my housemates Susi Wurmbrand, Jonathan Bobaljik,

and Meltem Kelepir for sharing many things with me.

When I came to MIT, I was a mathematician. Becoming a linguist was harder

than I had expected, and I thank the following people for their help. My class-

mates, Danny Fox, Paul Hagstrom, Norvin Richards, Martha McGinnis, Rob Pen-

salfini, Judy Yoon-Kyung Baek, Ingvar Løfsted, David Braun for being my friends.

My other teachers at MIT: Alec Marantz, Roger Schwarzschild, Ted Gibson, Mor-

ris Halle, Ken Hale, Michael Kenstowicz, Ken Wexler, and Robert Berwick. Hubert

Truckenbrodt and Colin Philipps for getting me through first year syntax. Those who

contributed to the liveliness of the LF-reading group over the past few years, espe-
cially Renate Musan, Diana Cresti, Lisa Matthewson, Michel DeGraff, Rajesh Bhatt,

Roumyana Izvorski, Sabine Iatridou, and Julie Legate. My former housemates Heidi

Harley, Hubert Truckenbrodt, Ayumi Ueyama, Danny Fox, Martha McGinnis, and

Marie Hélène Côté. And, the many other linguists I was in contact with; in the

MIT-community, Pilar Barbosa, Andrew Carnie, Ben Bruening, Vivian Lin, Cheryl

Zoll, Marie Claude Boivin, Wayne O’Neill, Maya Honda, Gina Rendon, Maire Noo-

nan, David Embick, Masa Koizumi, Yoonjung Kang, Bridget Copley, Karlos Arregui-

Urbina, Jeannette Schaeffer, Calixto Aguero-Bautista, Isabel Oltra-Massuet, Dylan

Tsai, Luciana Storto, Jay Rifkin, Cornelia Krause, Dag Wold, Carson Schuetze, Hooi-

Ling Soh, Takako Aikawa, Idan Landau, Eleni Anagnostopoulou, Taylor Roberts,

Philippe Schlenker, Sonny Vu, Shigeru Miyagawa, San Tunstall, Marlyse Baptista,

and Vaijayanthi Sarma; and elsewhere, especially Elizabeth Laurençot, Kyle Johnson,

Pauline Jacobson, Winfried Lechner, Sandiway Fong, Daniel Büring, Jason Merchant,

Gereon Müller, Sigrid Beck, Artemis Alexiadou, Chris Kennedy, Hotze Rullmann, Do-

minique Sportiche, Tim Stowell, Mamoru Saito, William Snyder, Angelika Kratzer,

Joachim Sabel, Hamida Demirdache, Akira Watanabe, Gennaro Chierchia, Viviane

Deprez, Gertjan Postma, John Frampton, Diane Jonas, Matthias Schlesewsky, Laurel

Laporte-Grimes, Takeo Kurafuji, Friederike Moltmann, Chris Wilder, Satoshi Oku,

Alan Munn, Christina Schmitt, Günther Grewendorf, Howard Lasnik, Yael Sharvit,

Satoshi Tomioka, Masao Ochi, Ayumi Matsuoka, Elena Herburger, Piroska Csuri,

Gisbert Fanselow, Josef Bayer, Markus Bader, and Michael Meng. I’m also grateful

to those who inspired me to pursue a career in linguistics: Peter Eyer, Arnim von Ste-

chow, Urs Egli, Manfred Kupffer, Ulf Friedrichsdorf, Shin-Sook Kim, Uta Schwertel,
and Wolfgang Sternefeld.

Most of section 6.2 was presented at the MIT LingLunch series, at the LF-

reading group at MIT, at the postersession of NELS 28 at the University of Toronto,

and at WCCFL 18 at the University of British Columbia at Vancouver. Some of

chapters 2, 3, and 5 was presented at the LF-reading group at MIT, at SALT 8 at

MIT and at the Colloquium Series at the University of California at Los Angeles. I

thank the audiences of these presentations for their comments.

During my five years at MIT, I received financial support from the German

Academic Exchange Service (DAAD), from the NSF-sponsored RTG grant to David

Pesetsky and from the Linguistics Department at MIT.

For being with me through all of this, I thank my family: Vater, Mutter,

Stefan, Angela, Kazuko, and Kai, who arrived when it was almost over.

[To match the page numbering of the filed version of the thesis the table of contents

starts on page 11.]

 Meinen Eltern,

Thankmar und Maria Luise Sauerland

Contents

1 Introduction 13

1.1 Background Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Binding into Traces 29

2.1 Scope, Condition C and Antecedent Contained Deletion . . . . . . . . 34

2.2 Variable Binding and Condition C . . . . . . . . . . . . . . . . . . . . 43

2.3 The A/A-bar Distinction . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.4 Relative Clause Internal Traces . . . . . . . . . . . . . . . . . . . . . 60

2.4.1 Two LF-Structures for Relative Clauses . . . . . . . . . . . . . 65

2.4.2 The Internal Head of Matching Relatives . . . . . . . . . . . . 75

2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

3 Identity of Traces 93

3.1 A Copy Identity Account of Kennedy’s Puzzle . . . . . . . . . . . . . 99

3.2 Semantic Content of the Trace . . . . . . . . . . . . . . . . . . . . . . 113

3.3 Wh-Traces and Focus in Chains . . . . . . . . . . . . . . . . . . . . . 137

11
 3.3.1 Pseudogapping and Traces . . . . . . . . . . . . . . . . . . . . 141

3.3.2 Focus and Wh-Traces . . . . . . . . . . . . . . . . . . . . . . . 144

3.3.3 Domain Expansion and Focus Index Sloppiness . . . . . . . . 161

3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

4 Linking Trace and Antecedent 185

4.1 Variables or Combinators . . . . . . . . . . . . . . . . . . . . . . . . . 192

4.1.1 Forcing Different Indices . . . . . . . . . . . . . . . . . . . . . 203

4.1.2 Forcing Index Identity . . . . . . . . . . . . . . . . . . . . . . 206

4.2 Predicates or Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . 227

4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

5 Interpreting Moved Quantifiers 237

5.1 A Choice Function Approach to All Quantifiers . . . . . . . . . . . . 240

5.2 Predictions of the Approach . . . . . . . . . . . . . . . . . . . . . . . 263

5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266

6 Conclusion/Outlook 269

6.1 The Type of Traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

6.2 Scope (or Total) Reconstruction . . . . . . . . . . . . . . . . . . . . . 276

6.2.1 A PF-movement Account of Scope Reconstruction . . . . . . . 280

6.2.2 The Scope Freezing Generalization . . . . . . . . . . . . . . . 284

12
Chapter 1

Introduction

Chains are dependencies between two or more positions of a syntactic structure which

are created by the syntactic computation, and the making of chains is an extensively

researched topic in syntax. Why is the meaning of chains interesting? For one, the

meaning of chains just like that of bound pronouns is, at least in an intuitive sense,

not compositional: the meanings of two disjoint pieces of structure, the head and the

tail of a chain, seem intimately related. For this reason, the study of the semantics

of chains promises to provide insight into the interpretation processes that apply to

syntactic structures. Little research has been done on the semantics of chains, and

the semantic dependency in a chain is standardly treated exactly parallel to pronoun

binding. This assumption, in particular, this thesis sets out to refute.

Some fundamental questions to ask about the semantics of chains are the

following: What aspects of the syntactic representation of a chain are relevant for

the semantics? What is the independent contribution of the head of a chain to the

meaning of the whole? What is the independent contribution of the tail of the chain?

13
What are the semantic processes applying to link the two (or more) positions of a

chain together? Are there differences between the semantic processes interpreting

bound variables and those interpreting chains? Are differences between syntactic

types of chains (A/A-bar) reflected in their semantics? This thesis tries to answer

the fundamental questions about the interpretation of chains using new diagnostics,

in particular focus semantics. While it in some cases provides new arguments for

classic assumptions like the use of λ-calculus, it argues in many cases for a substantial

revision of the semantics of chains.

The remainder of the introduction briefly clarifies the set of background as-

sumptions in which this thesis is embedded in section 1.1, and then provides an

overview of the thesis in section 1.2.

1.1 Background Assumptions

This thesis is intended as a contribution to a growing body of work that attempts

to account for the contribution of syntactic structure to sentence meaning by taking

both syntactic insights concerning structure forming and modifying processes seri-

ously, and at the same time, providing fully explicit model-theoretic statements of

the semantic rules involved. This research project with only minor difference is in-

troduced in two recent semantics textbooks by Larson and Segal (1995) and by Heim

and Kratzer (1998). The account of the meaning of structures within this project

can involve both (covert) syntactic operations that apply for purely semantic reasons

and semantic interpretation rules in whichever combination that leads to the simplest

14
overall account.

One insight of this research project, which I adopt, is that a clearer state-

ment of both the syntactic computation and the semantic mechanisms is achieved if

a third mechanism is hypothesized that translates the output of the syntactic com-

putation at the syntactic level of logical form into another representation to which

the semantic mechanisms apply. For example, von Stechow (1993:section 8) discusses

this assumption using the term transparent logical form for this intermediate level

of representation. I will sometimes use the term semantic form, but when it’s un-

ambiguous, I will often refer to this level of representation as logical form, as well.

Obviously, simplicity considerations also apply to the translation of logical forms into

semantic forms. I assume, in particular, that this translation procedure can delete

parts of phrases, but also insert new pieces of structure that are necessary to represent

dependencies semantically as discussed in chapter 4.

Within the three step interpretation procedure sketched in the previous para-

graph, the last step, the semantic mechanism, is the least tangible. Along with most

research on the topic, I content myself with stating semantic rules to define a model-

theoretic concept of truth for the semantic form structures. I assume that the notion

of truth in a model defined in this manner is related to semantic intuitions, in the

intuitive way that is commonly assumed in the field and has proved fruitful. The

semantic rules are split into lexical rules and a composition rule C interpreting com-

plex phrases. The composition procedure C is defined by recursion over the syntactic

structure. As for the basic operation that combines the interpretations of two phrases

into the meaning of one branching node, I assume with Heim and Kratzer (1998) that

15
there are two clauses to C, namely functional application and predicate intersection,

and that whichever of the two clauses that is compatible with the semantic types of

the two sub-phrases is the one that applies.

I start out with one assumption specific to the first step of the account of

chains, their syntactic derivation. Namely, I assume that the syntactic process that

creates chains is copying of the lexical material from the tail position of a chain to

the head position as endorsed for example by Chomsky (1995). This assumption is

strongly supported by reconstruction facts like those discussed in chapters 2 and 3

and in the references cited there. The other two stages of the interpretation of chains

are the main subject of the investigation.

The main tool used in this thesis to study the meanings of dependencies is

the semantics of focus and destressing/deletion. It is known that the explanation

of these phenomena involves sentence internal entailment relationships. Hence, they

can be used to test for the meaning of parts of a sentence, in particular a part that

contains a dependent element, but not its antecedent. Since the semantics of focus

and destressing is not so widely known, I introduce aspects of it as they become rele-

vant: the relationship between destressing and ellipsis in section 3.2, the concept of a

presuppositional skeleton and its relevance in section 3.3.2, focus indices and domains

in 3.3.3, the contrastiveness requirement in 4.1.1, and the relationship between pitch

accent and focus in 4.1.2. For readers who prefer a concise introduction, Kratzer

(1991) and Rooth (1996) provide a good overview of the basics of focus semantics.

The work of Rooth (1992b) is particularly relevant below for the account of sloppy

interpretations and the relationship between focus and ellipsis. Schwarzschild (1998)

16
presents some ideas that are used and modified in sections 4.1.1 and 4.1.2.

1.2 Overview

The thesis is structured in three parts. The first part, consisting of chapters 2 and

3, concerns the content of a trace position in a chain. The second part, chapters 4

and 5, studies the semantic mechanisms linking a trace and its antecedent. The third

part, chapter 6, provides evidence for the completeness of the solution given for the

semantics of chains in the other two parts.

The main point of chapters 2 and 3 is that a trace position in a chain can

contain lexical material of the head of the chain. Chapters 2 and 3 develop two

independent arguments in favor of this conclusion; chapter 2 looks at the distribution

of Condition C effects, chapter 3 looks at the identity requirement of two traces in

ellipsis constructions. Not only do these two chapters both argue for the presence

of lexical material in some cases; it’ll also be shown that both argue for the same

generalizations about when and what parts of the lexical material of the head of the

chain is represented in the trace position.

The discussion in chapter 2 is guided by the proposal of Chomsky (1993) that

whenever an R-expression in the head of a chain triggers a Condition C violation with

respect to a pronoun that c-commands only the tail position of the chain, this means

that the R-expression is lexically represented in the tail position. For example, the

R-expression John in (1a) is part of a wh-movement chain and triggers a Condition

C violation in the trace position of this chain. Therefore, I assume a semantic form

17
representation like (1b), where a part of the moved phrase including the R-expression

is lexically represented in the gap position. The correspondence between chapters 2

and 3 argues that lexical representation in the trace position of a chain is the right

approach to the distribution of Condition C phenomena.

(1) a. ∗ Which
W argument of Johni ’s father did hei defend.

b. [Which] did he defend [argument of John’s father]

Assuming representations like (1b), the distribution of Condition C effects shows

that DPs seem to split into independent parts that can be represented in different

positions of the chain at the level of semantic form, while the parts themselves cannot

be divided. I use the term segment for a part of a DP that seems to be always

represented in the same position. Segments are the NP-part, which I define as the

lowest NP-projection of the complement of D (excluding all adjoined modifiers), and

each modifier adjoined to the NP-part. The terminology is exemplified in (2), which

is a DP with two segments, the NP-part and one modifier.

(2) which argument of John’s that Mary had criticized

| {z } | {z }| {z }
Det. NP-part modifier
| {z }
segments

The following factors are shown to affect the presence of segments in the trace position

of a DP-chain in chapter 2: the surface position, the A/A-bar status of the chain, the

impossibility of self-contained reference in Antecedent Contained Deletion (ACD),

and the requirement that bound variables must be in the scope of their binder. In

18
particular the following two results are new: For one, while an ACD-relative must

be represented in a higher position of a chain (Fox 1995b), the NP-part must always

be represented in the trace position, as shown in section 2.1. Secondly, extending

arguments of Lebeaux (1992, 1995), section 2.2 presents evidence that a segment

containing bound variable must be represented in the scope of its binder, while other

segments of the same DP can be represented in a higher position. In section 2.3, the

first result together with other observations from the literature supports the claim

that the distinction of A-bar from A-chains can be reduced to the claim that the NP-

part must be represented in the tail position of an A-bar chain. Section 2.4 shows that

lexical material of the relative clause head is also present in the relative clause internal

trace position, but, depending on the semantic properties of the relative clause, the

representation is less direct than in chains. The result of chapter 2 is summarized in

section 2.5 as a set of ranked constraints.

The argument from the identity of traces in chapter 3 is based on paradigms

like (3), where the interpretation of elided material intended is indicated by a para-

phrase in angle brackets. Kennedy (1994) first observed that examples like (3a) are

ungrammatical. The observation that (3b) contrasts with (3a) is new, and not pre-

dicted by any existing account of (2a).

(3) a. ∗ Polly
P visited every town that’s near the lake Eric did hvisiti.

b. Polly visited every town that’s near the town Eric did hvisiti.

The examples in (3) involve ACD where the head of the ACD-relative is different from

19
the DP in the antecedent that corresponds to the trace in the elided VP. Contrasts

like (3) show that, for the acceptability of the construction, the NP-parts of the two

DPs involved—the head of the relative clause and the correspondent in the antecedent

of the trace in the elided VP—must be lexically identical. The account of (3) argued

for relies on the semantic form representations in (4). In (4), ACD is resolved by

quantifier raising of the DP corresponding to the trace, such that the elided VP and

its antecedent both contain a trace position. If the NP-part of the antecedent is

lexically represented in both trace positions as in (4), the paradigm (3) follows from

the identity requirement of VP-ellipsis. In (4a), the elided VP and its antecedent are

not identical and, therefore, (4a) is predicted to be bad. In (4b), however, the elided

VP and its antecedent are identical.

z elided
}| VP { z antecedent
}| {
∗
(4) a. [every town that’s near the lake Op E. visited [lake]] P. visited [town]

z elided
}| VP { z antecedent
}| {
b. [every town that’s near the town Op E. visited [town]] P. visited [town]

The account of (3) sketched here is developed in section 3.1. It’s also shown

there that the amount of lexical representation argued for by this account of Kennedy’s

observation is the same as that argued for by the distribution of Condition C. Section

3.2 slightly extends the paradigm in (3) and shows that the semantic properties of

the material lexically represented in the trace position affect the severity of the ill-

formedness in the case of mismatch. This is seen to argue that the lexical content of

a trace position at semantic form is indeed interpreted in the trace position. Section

20
3.3 shows why the effect of the identity requirement on traces is usually not observed

in cases of wh-movement other than ACD. Furthermore, the account there predicts

one exception where the effect is found in examples with wh-movement; namely, (5)

shows a contrast just like (3).

(5) a. I know which cities Mary visited, and now I would like to know the cities

Sue did hvisiti

b. ∗ I know which cities Mary visited, but I would like to know the lakes she

did hvisiti

Also in section 3.3, I present an argument that the lexical material in the trace position

is also represented in the head position of a chain, unless it contains a pronoun that

isn’t bound in the higher position. That is, I assume the semantic representation of

(6a) to be (6b) where the NP-part of the wh-phrase is represented in both positions

of the wh-chain.

(6) a. Which book does he like?

b. [which book] does he like [book]

The second part of the thesis considers the question how both positions of a

chain are interpreted together. As mentioned already, the interpretation of a chain,

and also that of a bound pronoun, is in a sense not compositional—the interpretation

of two positions, the binder and the bound phrase, is intimately connected. A number

of mathematical models has been proposed for the semantic mechanism that is at

21
work in such dependencies and the question is whether and how these views can

be empirically distinguished. In chapter 4, I present several arguments in favor of

λ-calculus as the model of the semantic mechanism underlying dependencies. The

result applies both to chains and to pronoun binding. But, the content of the trace

position, which is established in chapter 2 and 3, raises additional questions about

the interpretation of chains. Chapter 5 considers the case of a chain headed by a

quantificational DP, and develops a complete set of interpretive mechanism for this

case. The main claim there is that the quantification ranges not over individuals, but

over pointwise different choice functions.

The three mathematical models for the mechanism creating the semantic link

in a chain that chapter 4 compares are λ-calculus, combinatorial logic, and predicate

logic extended with restricted quantification. Since, as mentioned, pronoun binding

seems to involve the same concept of semantic link as chains, the simplest assumption

is that the same mechanism is involved in chains and pronoun binding. Because of

this assumption, some of the arguments of chapter 4 (which are based on examples

involving pronouns) carry over to chains as well.

Section 4.1 presents two arguments in favor of those models that involve vari-

ables (λ-calculus and extended predicate logic) and against combinatorial logic. Both

arguments revolve around the abstract configuration sketched in (7). Consider the

meanings of the two domains A and B, which both include a dependent element, but

not the binder of it. I show that the contribution of the dependent elements to the

meaning is identical for both domains on the combinatorial logic view. On the views

that use variables, however, the two dependents could contribute different variable

22
names (or indices).

z domain
}| A { z domain
}| B {
(7) binder . . . . . . dependent . . . . . . binder . . . . . . dependent . . .

How can the meaning of such domains that aren’t full sentences be investigated? Fo-

cus semantics has been argued to involve inference relationships between constituent

domains smaller than sentences. Hence, if it can be determined which domains are

considered in the licensing of focus and destressing, the meaning of domains like those

in (7) can be studied. Section 4.1.1 presents an argument along these lines based on

example (8) and on the fact that a focussed phrase must be contrastive to the cor-

responding phrase in the antecedent of its focus domain. In (8), the pronoun his in

the second conjunct can optionally be focussed and, therefore, must differ in meaning

from the pronoun his in the antecedent to be contrastive. If the focus structure of (8)

is as indicated, the two pronouns can be different in meaning on the variables view,

namely they can be interpreted as different variables. On the combinatorial view, on

the other hand, the contrastiveness requirement cannot be satisfied by (8). Since the

focus on the pronoun is optional in (8), other focus structures must be possible for

(8). Section 4.1.2 presents a second argument for the variables view, based on a case

where the focus structure is unambiguous, and therefore the focus on the dependent

required.

z antecedent
}| { z focus }|
domain {
(8) Every boy called his father and every TEAcher called HIS father

Section 4.2 attempts to draw a distinction between the λ-calculus view and the

23
extended predicate logic view. One argument for λ-calculus shows that the different

variables of the tails of two chains are not contrastive when the domains considered

are the sisters of the moved constituents. This result is expected if the variables are

bound within the sisters of the moved constituents by corresponding λ-operators, but

it’s unexpected if the two moved constituents themselves bind the variables as on the

predicate calculus view. A second argument presented in section 4.2 for λ-calculus

comes from the distribution of i-within-i reference. I conclude therefore that the

semantic form of (6a) is (9), where the translation from the syntactic logical form to

the semantic form might contribute the variable index and the λ-operator.

(9) [which book] λx that he likes [x, book]

Chapter 5 addresses the question how the lexical content of the trace position,

for example in (9), contributes to the interpretation of a chain. I consider exclusively

the case of a chain headed by a quantificational determiner, and argue later that this

might be the only case that arises. The approach developed is guided by the assump-

tion that the semantic mechanisms should apply in the same manner to all chains

headed by a DP. Specifically, no difference should be made between interrogative and

non-interrogative DPs. Because of this assumption, examples like (10a), where the

fronted DP contains a pronoun that’s bound in the trace position, are an important

case to account for. The semantic representation of (10a), that was argued for in

the preceding chapters, is given in (10b). For the semantics, I adopt and extend the

choice function approach of Engdahl (1980), which also relies on representations like

24
(10b).

(10) a. Which friend of heri ’s did every student invite?

b. which λx did every student invite [x, friend of heri ’s]

To extend the choice function proposal to non-interrogative, non-existential quanti-

fiers, it turns out that it must be modified. The modification that proves most fruitful

is a restriction of quantification to only pointwise different choice functions, instead

of all choice functions, where the definition of pointwise different is given in (11).

(11) f is pointwise different from g if and only if ∀x: f (x) 6= g(x)

One prediction of the approach that seems desirable is discussed in section 5.2. It

predicts that all DP-chains with lexical material in the trace position involve quan-

tification over choice functions, while chains with no lexical material in the trace

involve quantification over individuals. Since it was argued in 2.3 that the chains of

the former type are most A-bar chains, while all A-chains are of the latter type, a

type difference between A-bar chains and A-chains follows. If pronouns are of the

type of individuals, it follows that the head of an A-bar chain cannot bind a pronoun,

while the head of an A-chain can. In this way, the type difference is seen to predict

the distribution of weak crossover effects.

The third part of the thesis is the shortest and most tentative. Chapter 6

presents two results that go some way towards establishing the claim that the semantic

mechanisms developed in the previous part can account for the interpretation of all

25
chains that actually arise. Section 6.1 addresses a limitation of the mechanism of

chapter 5, namely its restriction to DP-chains. I present arguments from the literature

and one new argument based on facts from quantifier float in Japanese to show that

only chains where the type of the trace is the type of individuals occur at LF. This

implies that all occurring cases of chains are accounted for by the mechanisms already

developed, where the variable ranges over choice functions and the type of the entire

trace is that of individuals.

Section 6.2 addresses cases of so-called scope or total reconstruction, where

a moved quantificational phrase takes scope in its trace position. In such cases, all

lexical material of the chain is interpreted in the trace position, and none in the

head. This, however, seems to require me to partially withdraw the claim of section

4.2 that the sister of a moved phrase is interpreted as a predicate. If the head of a

chain is semantically empty, nothing would serve as the argument of this predicate.

Though the required modification is rather trivial, section 6.2 presents an argument

that the modification might not be needed at all. It is argued there that scope

reconstruction phenomena should instead be analyzed as cases of movement in the

phonological component of the grammar, which therefore doesn’t have any semantic

effect. Specifically, it’s shown that the PF-movement proposal together with the

assumption that movement must always target a c-commanding position makes a

correct prediction; namely, the generalization that scope reconstruction is blocked

in examples like (12) from Barss (1986) where the moved quantifier some politician

doesn’t c-command its trace.

26
(12) [How likely to tQP address every rally]wh is [some politician]QP twh ? (someÀlikely,

∗
likelyÀsome)

27
28
Chapter 2

Binding into Traces

At least since Ross (1967) and Lakoff (1968), it’s been known that dislocated phrases

can behave as if they were in their base position for the purposes of binding the-

ory as in (1) and (2) below. This phenomenon, Binding Reconstruction, still re-

mains only incompletely understood in many ways, but some significant proper-

ties of it have been discovered over the years: a correlation with the A/A-bar (or

NP-movement/wh-movement) distinction (Wasow 1972:66,142,147–57, 1979:157–75,

Riemsdijk and Williams 1981:204, Chomsky 1981),1 a distinction between arguments

and adjuncts of the moved phrase (Freidin 1986, Lebeaux 1988 with observations in

Riemsdijk and Williams 1981, Chomsky 1981:144), a difference between overt and

covert movement (Brody 1979, Chomsky 1981:196–197), a correlation between bind-

ing reconstruction for Condition C and for variable binding (Lebeaux 1992, 1995,

1
In the literature on scrambling (e.g. Tada 1993), differences with respect to binding reconstruc-
tion are often the only criterion for the A/A-bar distinction. Hence, it may seem strange to speak
only of a correlation. But, the A/A-bar distinction is needed independently for the statement of
locality conditions on movement, in particular for overlapping paths (see Chomsky 1977, Rizzi 1990,
Takano 1993, Müller 1993, 1998).

29
1998, Tada 1993:66-68, Chierchia 1995:129–170), and a correlation between narrow

scope and binding reconstruction (Heycock 1995, Romero 1996, 1997, Fox 1996, 1997,

1998b, Sportiche 1996).

(1) [Which friend of herj ’s]i did every studentj invite ti ?

∗
(2) [
[Which pictures of Johnj ]i did hej like ti ?

This chapter discusses facts that demonstrate that some parts of a moved phrase

show binding reconstruction effects, but other parts of the moved phrase don’t seem

to reconstruct. Two kinds of evidence are new: One novelty is evidence that shows

that the covert movement that resolves Antecedent Contained Deletion shows some

binding reconstruction effects. The other is evidence that, when variable binding

forces binding reconstruction of some part of the moved phrase, other parts of the

antecedent can still escape binding reconstruction.

In the discussion of binding reconstruction phenomena, I adopt the assumption

that binding reconstruction is syntactically represented by lexical material occupying

the trace-position at the level where binding theory applies. Throughout, I use the

notation exemplified in (3a) and (4): The relation between the lexical material that

is represented in the trace position and the lexical material in the top position of

the chain is expressed by a variable, x in the examples, which is part of the complex

trace and bound in the position marked by the λ-operator, that marks the sister of

the moved operator as a derived predicate. The use of λ-calculus as the mechanism

mediating the dependency of operator and trace is argued for in chapter 4. For the

30
moment though, it should be seen as a typographically more convenient version of

a notation like (3b) which is agnostic about the semantic mechanism mediating the

dependency.

(3) a. Which λx did every studentj invite [x, friend of herj ’s]?
| {z } |{z} | {z }
operator binder complex trace

b. Which did every studentj invite [friend of herj ’s]

The bound variable her’s in (3) must be in the scope of its binder, and therefore

I assume that in this case some of the lexical material is represented only in the

trace position, as shown in (3). For (2), it’s not clear whether to represent the

lexical information picture of John only in the tail of the chain as in (4a), or to

represent it doubly, in the head and the tail of the chain as in (4b) as suggested by

Danny Fox (p.c.). To block coreference between he an John in (2), however, either of

the representations in (4) suffices, since in both the pronoun he c-commands the R-

expression John. (4b) may seem redundant, but on the other hand, it’s more natural

on the assumption that the syntactic operation underlying movement phenomena is

copying, since it doesn’t require as much deletion as (4a). In section 3.3.3, I present

an empirical argument in favor of the latter view.

∗
(4) a. W
Which λx did hej like [x, pictures of Johnj ]?
| {z } | {z }
operator complex trace

b. ∗ [Which
[ picture of Johnj ] λx did hej like [x, pictures of Johnj ]?
| {z } | {z }
operator complex trace

31
 In (3b) and (4b), binding reconstruction is represented by syntactic material

that occupies the trace position. It is debated, though, whether the evidence neces-

sitates this syntactic view (cf. Lebeaux 1992, 1995, 1998 Chierchia 1995, Fox 1998b,

Romero 1997, Lechner 1998, Sharvit 1998 for discussion). An alternative view de-

veloped in detail by Barss (1986) assumes that a the head of a chain can be in the

semantic scope of a phrase that c-commands its trace under certain conditions. On

this view, the evidence in this section would be relevant towards stating restrictions

on this chain-binding strategy. For now, I use lexical material in the trace position

in the discussion of binding reconstruction as working hypothesis, though I suspect

that it’s less elegant to state the generalizations on the semantic reconstruction view.

I refer the reader to Fox (1998b) for an overview of arguments in favor of the syn-

tactic approach. The arguments in chapter 3 provide an additional argument for the

syntactic approach taken here, and in chapter 5 I make a proposal as to how the

lexical material in both the head and the tail position of a chain contributes to its

interpretation.

The goal of this chapter is to investigate what parts of the head of a chain

undergo binding reconstruction when the chain itself exists at the level of logical form.

In other words, since I assume that binding reconstruction is represented by lexical

material in the trace position, the goal of this chapter is to find out where what parts

of the lexical material of a chain are at the level where binding theory applies. In

particular, I present new arguments that the lexical material of a DP-chain can in

some cases be split between the top and the bottom position, in ways other then the

split between the quantificational D-head and its NP-complement postulated in (3)

32
and (4). This kind of situation is sketched in (5).

(5) [D ‘some lexical material’] λx . . . [x, ‘other lexical material’]

| {z } | {z }
operator trace

The derivation of LF-representations like (5) is straightforward assuming the copy

theory of movement (Wasow 1972, Chomsky 1981, Burzio 1986, Chomsky 1995). If

syntactic movement creates representations where all positions of a chain contain

all the lexical material of the moving phrase, the distribution of lexical material in

representations of the kind sketched in (5) can be derived by the application of deletion

in the positions of a chain. The goal of this chapter is then to find the conditions

under which this deletion operation applies.

Engdahl (1980:131–144) was the first to articulate the claim that binding re-

construction can cooccur with wide scope. The empirical evidence she gives for the

LF in (6a) (I discuss her semantics in Chapter 5) is of the type in (1): wh-questions

where the wh-phrase contains a bound variable pronoun. Her argument therefore re-

lies on two assumptions: that bound variable pronouns must be in the scope of their

antecedents and that wh-phrases take sentential scope. There’s hardly any alternative

to the first assumption (Though, see the discussion of Skolem-functions in Engdahl

(1986) and Chierchia (1993)). The second assumption is much harder to argue for,

and indeed not universally assumed—for example Hamblin (1973), Cresti 1997 and

Rullmann and Beck (1997) entertain LF-representations like (6b) for (1) where the

wh-chain of the overt form is not represented at all. Ultimately, I believe Engdahl’s

(1980) interpretation of (1) is correct, but it will take some effort to get there.

33
(6) a. which λx did every studenti invite [x, friend of heri ’s]?

b. did every studenti invite which friend of heri ’s?

| {z }
wh-phrase

To sharpen Engdahl’s argument, we need better tests for the location of the expression

heading the chain that don’t reconstruct. In addition to variable binding, the three

tests relevant for this chapter are Quantifier Scope, Antecedent Contained Deletion

(Sag 1976, May 1985, Larson and May 1990), and Condition C of the Binding Theory

(Ross 1967, Langacker 1969, Lasnik 1976, Chomsky 1981).

2.1 Scope, Condition C and Antecedent Contained Dele-

tion

Before applying Condition C to test for the LF-position of syntactic material, I’ll

summarize an argument from Fox (1995b) that Condition C applies at LF only. His

argument relies on the contrast in (7) (from Fiengo and May 1994:296 with some

modifications). (In (7) and in the following, I indicate the interpretation of a VP-

ellipsis site by a paraphrase given in angle brackets hi.)

(7) a. You introduced himi to everyone Johni wanted you to hintroduce himi toi

b. ∗ I introduced himi to everyone Johni wanted you to dance with.

In both examples in (7), the pronoun him c-commands the R-expression John

in the surface form. While Condition C rules out coreference between him and John

in (7b), Condition C doesn’t seem to apply in (7a) despite the surface c-command

34
of John by him. In (7a), coreference between the pronoun and the R-expression is

acceptable. Fox’s (1995b) as well as Fiengo and May’s (1994) account of (7) relies on

the account of Antecedent Contained Deletion (henceforth ACD) of Sag (1976:73),

May (1985) and Larson and May (1990). Sag and Larson and May, in particular,

argue, based on an interaction between scope and ACD, that quantifier movement is

required for the resolution of ACD. This is displayed by the LF-representation in (8).

h i
(8) everyone [λy Johni wanted you to introduce himi to [y]]

λx you introduced himi to [x].

In the LF-representation (8), the R-expression John is no longer in the c-domain of the

pronoun him, and therefore Condition C is not violated. If Condition C doesn’t apply

to the surface representation (7a), but to the LF-representation (8), it’s expected that

coreference between him and John is possible. I leave the question why quantifier

movement cannot obviate Condition C in (7b) for the next paragraph. Assuming

that there’s an answer to this question, the contrast in (7) is a strong argument that

Condition C applies only at LF (see Fox 1995b for arguments against the account of

(7) in Fiengo and May 1994).

Why quantifier raising doesn’t always bleed Condition C, for example not

in (7b), is the remaining question. It turns out that (7b) is actually not a good

example to raise the question for because (7b) doesn’t control for the scope of the

universal quantifier and, if Fox (1995a) is right, quantifier movement is blocked in (7b).

However, it is well known that even when a universal quantifier takes wide scope,

35
quantifier movement usually doesn’t obviate Condition C (Brody 1979, Chomsky

1981:196–7). This is shown by (9), where even on the ∀À∃ wide scope reading, the

pronoun him cannot be coreferent with the R-expression John. The contrast between

(9) and (7a) cannot be explained by conditions on whether quantifier movement

applies.2 Rather, the contrast must be explained by an interaction between ACD and

Binding Reconstruction, namely that ACD blocks Binding Reconstruction.

∗
(9) S
Someone introduced himi to everyone Johni wanted you to dance with.

As Fox (1995b) points out, the Copy Theory of Binding Reconstruction to-

gether with standard theories of ACD predict that ACD blocks Binding Reconstruc-

tion: The relevant standard assumption about ACD and VP-ellipsis in general is

that an elided VP must be (almost) identical to its antecedent (Sag 1976, Williams

1977, May 1985, Tancredi 1992, Rooth 1992b, Wold 1995, Fiengo and May 1994).

For now, I assume that exact lexical identity is required except for tense morphology

(see section 3.2 and 3.3 though). Consider (10) which is the LF-representation of

(7a) where the elided VP and its antecedent are indicated. In (10) without Binding

Reconstruction, the elided VP and its antecedent are identical.

2
To maintain the view that it is the application of quantifier movement that accounts for the
contrast between wide scope as shown by scope and wide scope for the resolution of ACD, the
following assumptions would need to be postulated: There is a second scope taking mechanism in
addition to quantifier movement which yields wide scope, but doesn’t resolve ACD. Furthermore,
quantifier movement applies only to resolve ACD, while the other mechanism applies yields wide
scope when there’s no ACD. A mechanism with exactly these properties has been formulated in the
literature, namely Quantifier Storage (Cooper 1983), though not for this particular problem. But,
consistent though it is, such an account is quite clearly not more than a restatement of the facts.

36
 operator
hz }| {i
(10) everyone [λy Johni wanted you to introduce himi to [y]]
| {z }
elided VP
trace
z}|{
λx you introduced himi to [x] .
| {z }
antecedent

But if there’s a copy of the lexical material of the relative clause in the bottom

position of the chain of quantifier movement as indicated in (11), the elided VP and

its antecedent do not satisfy the identity condition. Hence, the representation (11) is

not possible for the sentence (7a). This way, the identity requirement on an elided

VP and its antecedent blocks Binding Reconstruction into traces that are part of the

antecedent in ACD constructions.

operator
z }| {
∗
(11) e
everyone [λy Johni wanted you to introduce himi to [y]]
| {z }
elided VP
trace
z }| {
λx you introduced himi to [x, [λy Johni wanted you to introduce himi to [y]].
| {z }
elided VP (copy)
| {z }
antecedent

In (9), on the other hand, Binding Reconstruction is possible. The represen-

tation corresponding to Binding Reconstruction is given in (12), which then violates

Condition C. If the representation (12) is actually the only one available for (9), the

contrast between quantifier raising for ACD resolution and for wide scope alone is

accounted for.

h i
∗
(12) everyone [λy Johni wanted you to dance with [y]] λx someone introduced

himi to [x, [λy Johni wanted you to dance with [y]]]

37
To this end, Fox (1995b) adopts the assumption argued for in Chomsky (1993:36-37)

that there is a preference for A-bar chains to represent all lexical material except for

the D-head in the trace position—in effect, a preference for Binding Reconstruction.

After Chomsky (1993), this is often referred to as the Preference Principle; it states a

preference between two representations: the one with the lexical material interpreted

at the top of the chain, and the one with the lexical material interpreted at the

bottom of the chain. Fox (1995b) reformulates the Preference Principle as a economy

condition and adds that it can be overridden by the requirement of ACD that the

relative clause containing the elided VP must be represented outside of the antecedent

for deletion. Condition C is different from the requirement of ACD in that it doesn’t

motivate a violation of the Preference Principle. The interaction between ACD and

the preference principle is discussed further at the end of section 3.2.

In addition to the argument that Condition C applies at LF, Fox’s (1995b)

analysis has implications concerning what are possible LF-representations of A-bar

chains, and what preferences exist between them. At least two representations are

possible for an A-bar chain consisting of a quantificational operator and lexical mate-

rial restricting the operator: a representation where the entire restrictor occupies the

trace position of the A-bar chain, and a representation where at least a relative clause

modifying the restrictor doesn’t occupy the trace position, but only occupies the head

position. Of these, the former representation is preferred by the grammar and the

second one is only used in case ACD blocks the first one.3 As for the structure of the

3
As Danny Fox (p.c.) pointed out to me, this view predicts that quantifier movement should
always obviate Condition C if the R-expression triggering Condition C is part of the quantificational
determiner. It’s very difficult to construct relevant examples, but a weak preference in the predicted

38
representation in the case of ACD, Fox’s (1995b) analysis of (7a) gives us information

about the LF-position of the R-expression, which is the subject of the ACD relative

clause. Namely, it must be in a position higher than its surface position. This follows

from the assumption that the entire relative clause must be in a position higher than

its surface position, which is required in the ACD-cases for ACD resolution. The

evidence, however, leaves it open whether all lexical material of the DP that moves

for ACD resolution is represented in the top position of the A-bar chain or only as

much as is needed for ACD-resolution. In effect, Fox (1995b) develops both views.

Consider now the contrast in (13), which resolves the open issue. (13a) is

comparable to (7a): The R-expression that is in a Condition C configuration in the

surface form is the subject of an ACD relative clause just like in Fiengo and May’s

(1994) example above: Condition C is bled by ACD-resolving Quantifier Movement.

In (13b), on the other hand, the R-expression is part of the noun phrase that has to

move for ACD-resolution, but it’s not inside of the ACD relative clause. The new

discovery is that the Condition C effect remains in (13b).

(13) a. In the end, I did ask himi to teach the book of Irene’s that Davidi wanted

me to hask him to teachi.

direction seems to be detectable in (i). In (ia), it seems that Condition C is obviated on the reading
where John’s every takes scope over someone. In (ib), quantifier raising of John’s every to a position
above the subject is blocked because the subject is not a scope bearing element (Fox 1995a). As
expected, Condition C cannot be obviated in (ib).
(i) a. Someone must’ve fed himi Johni ’s every move over earphones.
b. ∗ K
Kasparov must’ve fed himi Johni ’s every move over earphones.

39
 b. ∗ In
I the end, I did ask himi to teach the book of Davidi ’s that Irene wanted

me to hask him to teachi.

Merchant (1998a) independently discovered similar facts and makes the same point.

I give his three examples in (14). In each of them, Condition C applies to block coref-

erence between the object pronoun and the R-expression that occurs in the external

head of the ACD-relative clause.

(14) a. ∗ I gave himi every report on Bobi ’s division you did hgive himi i.

b. ∗ I reported heri to every cop in Abbyi ’s neighborhood you did hreport heri

toi.

c. ∗ I showed heri every picture from Abbyi ’s mantlepiece you did hshow heri i

The examples in (15) show that coreference is possible if Merchant’s (1998a) examples

are modified such that the R-expression is inside of the ACD-relative clause. In (15b)

and (15c), the R-expression is part of the subject of the ACD-relative, like in (7a),

while, in (15a), the R-expression is part of the material pied-piped with the relative

operator. Note that, while in (13) the two sentences compared differ with respect

to the amount of material intervening between the pronoun and the R-expression

relevant for Condition C, the contrast among (14b) and (14b) could not be explained

in terms of such a difference.

40
(15) a. I gave himi every report whose section on Bobi ’s division you asked me to

hgive himi i.

b. I reported heri to every cop Abby’s neighbors allowed me to hreport heri

toi.

c. I showed heri every picture Abbyi ’s agent forgot to hshow heri i

Example (13b), and also Merchant’s (1998a) examples in (14), show that ACD

doesn’t block binding reconstruction of lexical material that is not part of the ACD rel-

ative clause. On the Copy Theory of Binding Reconstruction, this lexical material—

book of David’s in the example (13b)—must therefore occupy the trace position, and

thereby causes a Condition C violation. In fact, a structure with such a complex trace

must be forced in the the examples in (13), since the Condition C violating structure

is the only one available for (13b).

As (16), a tentative LF-representation for (13a), illustrates, if we assume that

only quantifier movement leaves such a complex trace, the elided VP is not identical to

its antecedent anymore, because the relative clause internal trace is a simple variable

while the QR-trace is complex. At least if VP-ellipsis requires identity of lexical

material (see chapter 3), (16) cannot be the LF representation of (13a).

h i
(16) the book of Irene’s λy Davidi wanted me to ask himi to teach [y]
| {z }
elided VP
λx I asked himi to teach [x, book of Irene’s]
| {z }
antecedent

41
Since (13a) allows VP-ellipsis, the relative clause internal trace must be lexically

complex, containing lexical material of the noun phrase it attaches to. I present

further arguments towards this conclusion in section 2.4. For (13a), this allows the

LF representation in (17), which satisfies the identity condition of VP-ellipsis.

h i
(17) the book of Irene’s λy Davidi wanted me to ask himi to teach [y, book of Irene’s]
| {z }
elided VP
λx I asked himi to teach [x, book of Irene’s]
| {z }
antecedent

For (13b), these consideration force the LF representation in (18). In (18), the quan-

tifier movement that resolves ACD left the noun phrase part book of David’s in the

trace position, where the pronoun him c-commands it. Therefore, Condition C is

violated in (18), and coreference between him and David is correctly ruled out.4

h i
∗
(18) the book of Davidi ’s λy Irene wanted me to ask himi to teach [y, book of Davidi ’s]
| {z }
elided VP
λx I asked himi to teach the [x, book of Davidi ’s]
| {z }
antecedent

In the contrast in (13), the offending R-expression was an argument of the head

noun the determiner takes as complement. The examples in (19) exhibit a similar

contrast to (13), but the offending R-expression is part of an adjunct in (19b). This

shows that there’s no difference between adjuncts and arguments in the construction

4
Actually, Condition C seems to actually be violated twice in (17): not only the instance of David
in the trace of quantifier raising, but also the instance of David in the relative clause internal trace
is c-commanded by a coreferent pronoun. While this extra violation causes no problem in (17), in
general no Condition C reconstruction effect is found in a relative clause, as (i) exemplifies. I address
this issue in section 2.4.
(i) The book of Billi ’s that hei was working on since 1971 finally appeared.

42
we’re considering here.

(19) a. In the end, we did advise himi to buy the computer compatible with Bev’s

that Noami hoped we would.

b. ∗ In
I the end, we did advise himi to buy the computer compatible with

Noami ’s that Bev hoped we would.

The two results of this section are worth repeating once again: One, an argu-

ment of Fox (1995b) was summarized which shows that Condition C of the Binding

Theory applies at LF only. Secondly, an argument was presented that even when

some of the material of the covertly A-bar moved phrase is missing from the trace

position, other lexical material still seems to occupy the bottom position of the A-bar

chain.

2.2 Variable Binding and Condition C

In this section, I’ll present more arguments that structures with part of the lexical

material of an A-bar chain interpreted in the bottom position and other parts inter-

preted only in the top position are possible LF-representations. The evidence here

will be based on interactions between variable binding and Condition C, which I from

now on assume to apply at LF only following Fox (1995b).

In fact, if Condition C applies at LF only, the distribution of Condition C

effects alone with overt A-bar movement is evidence for representations where the

lexical material is split between two positions. The relevant observation is the well

43
known contrast between (20a) and (20b) (Riemsdijk and Williams 1981, Freidin 1986,

Lebeaux 1988), which is usually seen as an argument-adjunct distinction. The fact is

that the R-expression inside an argument of the noun head causes a Condition C viola-

tion by Binding Reconstruction into the trace position, making (20a) ungrammatical.

The R-expression in (20b), however, which is contained in an adjunct, doesn’t cause

a Condition C effect in the trace position.

(20) a. ∗ [Which
[ argument that Johni was wrong]j did hei accept tj in the end?

b. [Which argument that Johni had criticized]j did hei accept tj in the end?

The contrast in (21) makes the same point as (20), but controls for structural differ-

ences between the wh-phrases.5 The only difference between (21a) and (21b) is that

in (21a) the R-expression John occupies an argument position of argument and Mary

is the subject of the adjoined clause, but in (21b) the two are switched around. Only

when the R-expression is in the argument position in (21a), does it cause a Condition

C violation via Binding Reconstruction into the trace position.

(21) a. ∗ [Which
[ argument of Johni ’s that Mary had criticized] did hei omit tj in

5
However, (20) controls for the depth of embedding better than (21). It’s known that the strength
of a Condition C violation correlates with the distance and depth of embedding (Chomsky 1981:196–
7). Differences in the severity of Condition C as illustrated in (i) are expected on the basis of
general processing conditions and in addition the fact that examples like (ia) might also constitute a
Condition B violation (Kuno 1997). Therefore, it’s important to control for the depth of embedding
whenever possible.
∗
(i) a. He
H i liked Johni .
∗?
Hei liked that Mary bought a picture of Johni .
b. H
c. ∗? H
Hei liked that Johni ’s grandfather’s stories were popular.

44
 the final version?

b. [Which argument of Maryi ’s that John had criticized] did hei omit tj in

the final version?

The LF-representations the Copy Theory would assign to (21a) and (21b) are given

in (22a) and (22b) respectively. Condition C is violated only in (22a), where the lower

trace contains the R-expression John.6

h i
∗
(22) a. Which argument of Johni ’s [λy that Mary had criticized [y]] λx hei omit

[x, argument of Johni ’s] in the final version

h i
b. Which argument of Mary’s [λy that Johni had criticized [y]] λx hei omit

[x, argument of Mary’s] in the final version

The representations in (22) show that whether a Condition C violation is incurred

or not is a function of where in the fronted DP the R-expression appears. If the R-

expression is inside the NP that’s the complement of which, it triggers Condition C in

the trace position. If the R-expression occurs inside a relative clause that modifies this

NP, it doesn’t trigger Condition C in the trace position. In stating this generalization,

it is useful to have a term for the part of a DP that is the complement of the determiner

excluding relative clauses and other adjuncts that are adjoined to it: Henceforth, I

call this the NP-part of a DP. In (23), I have marked the determiner, the NP-part

and modifiers to it of the wh-phrase of (21a). Using this terminology, the contrast in

6
I’m not representing the lexical content of the relative clause internal trace at this point because
it’s irrelevant here.

45
(21) argues that the NP-part of a wh-movement chain must reconstruct to the trace

position for Condition C. This is stated as a preliminary generalization in (24).

(23) which argument of John’s that Mary had criticized

| {z } | {z }| {z }
Det. NP-part modifier
(24) The NP-part must occupy the position of an A-bar trace. A modifier may,

occupy the position of an A-bar trace.7

The generalization in (24) assumes that only modifiers adjoined to the NP-part itself,

but not modifiers internal to the NP-part, can escape binding reconstruction in wh-

chain. Based on different assumptions, Tada (1993:65) arrives at this conclusion

and presents evidence for it from Japanese. In (25), though the R-expression John

occurs inside a relative clause and outside the c-domain of the pronoun kare in the

surface form, coreference is blocked by Condition C. (25) is predicted if only modifiers

adjoined to the NP-part of the moving phrase itself can escape Condition C.8

∗?
(25) [John
[ i -ni ki-ta tegami-o suteru-yooni]j karei -ga tsumani tj miji-ta
[JohnDAT came letterACC ]j throw-away heiNOM wifeDAT tj ordered.

‘To throw away the letter that came to John, he told his wife.’

7
Lebeaux (1988) proposes counter-cyclic adjunction of relative clauses as an explanation for the
Condition C obviation of overt movement. Recall though from the previous section, that covert
movement also displays obviation of Condition C with relative clauses, if the covert movement is
required for Condition C resolution. Hence, Lebeaux’s (1988) explanation is at least incomplete. I
come back to the question Lebeaux’s proposal at the end of this section.
8
Note that Tada’s account predicts that, for a relative clause inside a fronted predicate, overt
movement won’t obviate Condition C. That this prediction is correct is shown by the examples in (i)
from Takano (1995:(12)) (see also Heycock (1995)). Since Tada’s explanation of (i) isn’t dependent
on the VP-internal trace hypothesis, I conclude contrary to Takano (1995) and Heycock (1995) that
examples like (i) don’t bear on this hypothesis.
(i) a. Criticize a student that Johni taught, hei said Mary did.
b. How proud of a student that Johni taught did hei say Mary is?

46
However, Tada (1993:fn. 25) doubts the existence of a difference between modifiers

to the NP-part and modifiers internal to the NP-part for English because of (26a)

(attributed to Noam Chomsky, p.c.). In (26a), Condition is obviated even though

the relative clause containing the R-expression John is adjoined to the lower NP

book. This argument isn’t convincing because the lower NP itself could be part of

an modifier if we assume that the for-PP is an modifier to book. The obviation of

Condition C in (27a) shows that the for-PP is an adjunct, as does the separability in

the copular paraphrase in (27b) (see Schütze 1995).

(26) a. The award for the book that Johni wrote, hei never received.

b. The award for the book that Johni received, hei never cashed.

(27) a. Which award for Titanici did everybody agree iti deserved.

b. The award was for the book.

The contrast in (28) shows that only modifiers adjoined to the NP-part of the moved

phrase can escape binding reconstruction. The R-expression Bill occurs inside a

relative clause in both, (28a) and (28b). However, there’s a contrast depending on

whether this relative clause is part of an argument inside the NP-part of the fronted

phrase as in (28a), or a modifier to this NP-part.

rel. clause
z }| {
∗
(28) a. Which
W book of the woman Billi admires did hei give to hisi parents.
| {z }
NP-part
rel. clause
z}| {
b. Which |book about
{z } | the woman Bill i admires did hei give to hisi parents.
{z }
NP-part modifier

47
 Another way to enforce binding reconstruction is variable binding, which

brings us back to Engdahl’s paradigm mentioned at the beginning. As is well known,

overt wh-movement allows a binding of variable inside the moved material by a quan-

tifier that c-commands the trace position, as in (29a). The ungrammaticality of (29b)

shows that c-command of the trace position is indeed necessary.

(29) a. [Which paper of hisj ]i did every studentj plan to revise ti ?

b. ∗ [Which
[ paper of hisj ]i ti earned every studentj praise?

On the copy theory of binding reconstruction, the LF-representation this leads us to

postulate for (29a) is (30). The variable his is interpreted in the bottom position of

the A-bar chain where it’s c-commanded by the quantifier every student.

(30) Which λx every student λy [y] planned to revise [x, paper of hisy ]

That the representation (30) is correct is shown by the interaction between Binding

Reconstruction for Variable Binding and Binding Reconstruction for Condition C

Lebeaux (1992) observes.9 The contrast in (31), which is from Lebeaux (1992) with

minor changes, shows this interaction.

(31) a. [Which paper that hek gave to Maryj ]i did every studentk think t0i that shej

would like ti ?

9
Tada (1993:66-68) discusses an interaction between temporal dependencies and Condition C
that makes the same point as Lebeaux’s data. Chierchia (1995:129-170) shows data with fronted
conditionals that supports Lebeaux’s conclusion as well.

48
 b. ∗ [Which
[ paper that hek gave to Maryj ]i did shej think t0i that every studenti

would like ti ?

In (31a), variable binding can be satisfied via reconstruction in the position t0i , which is

c-commanded by the antecedent of he, namely every student, but not c-commanded by

the pronoun she. Therefore, she doesn’t trigger a Condition C effect in this position,

and she and the R-expression Mary can be coreferent. The interaction observed by

Lebeaux is predicted by the Copy Theory view of reconstruction, as is shown by the

LF-representation in (32). The relative clause which contains both the bound variable

pronoun and the the R-expression is interpreted in the intermediate trace position

that is not c-commanded by the R-expression Mary.

(32) [Which paper] λx every studenti think [x, paper, λz hei gave [z] to Maryj ] λy
| {z } | {z }
operator intermediate trace
shej would like [y, paper]
| {z }
lowest trace

In (31b), on the other hand, all reconstruction positions c-commanded by every stu-

dent are also c-commanded by the R-expression Mary. If Binding Reconstruction for

variable binding always forces Binding Reconstruction for Condition C to take place

as well, variable binding is correctly predicted to be blocked in (31b). The Copy

Theory captures the interaction observed; namely, that Binding Reconstruction for

variable binding forces Binding Reconstruction for Condition C as well. The repre-

sentation in (33) illustrates that e.g. interpreting the copy of the relative clause in

the lowest trace position leads to a Condition C violation.

49
 ∗
(33) [Which
[ paper] λx shej thinks [x, paper] λy every studenti would like
| {z } | {z }
operator intermediate trace
[y, paper [λz hei gave [z] to Maryj ]]
| {z }
lowest trace

As we also saw in the interaction of ACD-resolution and Condition C above, Lebeaux’s

(1992) data demonstrate that a relative clause cannot be split among different posi-

tions of a chain. Otherwise it should be possible to interpret the bound variable in a

position lower than the the R-expression Mary and thereby accomplish variable bind-

ing without violating Condition C. To find out whether variable binding can also be

satisfied by binding reconstruction of a part of a fronted constituent, we need to test

cases where we know the relevant parts of the fronted constituent can be interpreted

in different positions of the A-bar chains. Such cases have not been studied in the

previous literature.

The contrast in (34) shows that it is possible to accomplish variable binding

by reconstructing only parts of a fronted constituent. (34b) has the same structure

as Lebeaux example in (31b) and, as above, variable binding into the relative clause

brings about a Condition C violation. In (34a), on the other hand, the bound variable

is not part of the relative clause, and therefore reconstruction of the relative clause

isn’t forced. Therefore, (34a) doesn’t violate Condition C. Example (35) makes the

same point as (34).

(34) a. [Which paper of hisk that Maryj was given]i did shej tell every studentk to

revise ti ?

50
 b. ∗ [Which
[ paper that hek gave to Maryj ]i did shej tell every studentk to revise

ti ?

(35) a. [Which of hisk pictures that Maryj was shown]i did shej return to ti every

studentk .

b. ∗ [Which
[ picture that hek showed to Maryj ]i did shej return ti to every

studentk .

The LF-representation of (34a) is shown in (36). The bound variable his is

interpreted in the bottom position of the A-bar chain, while the R-expression Mary

with the relative clause is interpreted in the top position, such that Condition C isn’t

violated.10

h i
(36) Which [λz Maryj was given [z]] λx did shej tell every studenti to revise [x,

paper of hisi ]?

A second case of variable binding taking place in a lower position than the

interpretation of the relative clause is found when there are two relative clauses.

(37b) shows that it is possible to interpret one relative clause in a low position to

achieve variable binding, and at the same time represent the second relative clause

only in a higher position, such that Condition C isn’t violated.

10
Notice that here the NP-part paper of hisi cannot be represented inside the relative clause
because the bound variable his would not be bound. In section 2.4, I present an analysis of relative
clauses that predicts this.

51
 inner modifier outer modifier
z }| {z }| {
(37) [Which computer compatible with hisj that Maryi knew how to use]k did shei

tell every boyj to buy tk .

While (37) confirms the claim that two modifiers can be represented at LF in different

positions of a chain, (38a), where the position of bound variable and R-expression

is exchanged, points to a complication. Here, Condition C is violated even though

the bound variable and the R-expression relevant for Condition C occur in different

relative clauses. Given the contrast to (37), it seems as if reconstruction of the outer

relative clause forces reconstruction of the inner relative clause to take place as well.

inner modifier outer modifier

z }| {z }| {
(38) ∗ [Which
[ computer compatible with Maryi ’s that hej knew how to use]k did shei

tell every boyj to buy tk ?

It is hard to decide whether the presence of a bound variable in the outer modifier

in (38) is among the causes of the Condition C effect. Even (39) is not very good

though here the outer modifier doesn’t contain a bound variable. But, there seems

to be a slight contrast between (39) and (38).

inner modifier outer modifier

z }| { z }| {
??
(39) [
[Which computer compatible with Maryi ’s that I knew how to use]k did shei

tell Tom to buy tk ?

In (40), an example where the R-expression is part of the NP-part is used as an

additional item of comparison. It seems that (40b), where the R-expression is part

52
of an inner modifier and the outer modifier doesn’t contain a bound variable, allows

coreference more easily than (40a), where the R-expression occurs in the NP-part, or

(40c), where the the outer modifier contains a bound variable.

(40) a. ∗ Tell
T me which descriptions of Kanti ’s views that were published every

woman said hei agreed with?

b. ? Tell
T me which books describing Kanti ’s views that were published every

woman said hei agreed with?

c. ∗ Tell
T me which books describing Kanti ’s views that shej published every

womanj said hei agreed with?

Therefore, I conclude that the ordering effect between (37) and (38) is real,

though it isn’t predicted by anything said so far. I think the effect might shed light on

the question why relative clauses can escape binding reconstruction with overt move-

ment. Namely, the effect is predicted on a modification of Lebeaux’s (1988) proposal

Tada (1993:63-70) develops. Lebeaux’s (1988) proposal is that relative clauses can

adjoin to a wh-phrase after it has undergone wh-movement and for this reason need

not reconstruct for binding to the bottom position of an A-bar chain. Essentially,

Lebeaux proposes that adjunction need not obey the syntactic cycle at all. Tada

(1993), however, proposes rather than to abandon the cycle, to modify it to accom-

modate Lebeaux’s cases. In effect, Tada proposes that adjunction obeys the cycle,

but that adjunction to the specifier of the current cyclic domain is consistent with

the cycle. On Tada’s proposal, modifiers adjoining to a moved phrase must obey the

53
cycle with respect to the phrase they are adjoining to. Then, the ordering effect is

predicted: The cycle then makes sure that the inner relative clause must be adjoined

before the outer relative clause. If reconstruction to the position where a relative

clause was first adjoined is forced, the order of adjunction determines that the inner

relative clause must reconstruct at least as low as the outer relative clause. Note that

this account of (40) supports the central claim of Lebeaux’s (1988) account that the

reason Condition C can be obviated with overt movement is late adjunction. In the

previous section, we saw that also deletion of adjuncts at LF can cause obviation of

Condition C. Because of (40), I conclude that both mechanisms are needed.

The main point of this section, however, is that even in cases where variable

binding forces binding reconstruction of parts of a chain, others parts of the chain

don’t have to reconstruct. More specifically, example (34) showed that even the

NP-part of a wh-phrase can reconstruct while a relative clause adjoined to it can still

occupy a higher position. Cases like (34) will be important for the semantics of chains

in chapter 5, because in these cases the dependency between the two positions of the

chain must be more complex because the meaning of the complex trace depends on

the value of the bound variable.

2.3 The A/A-bar Distinction

The previous two sections were concerned with covert quantifier movement chains and

overt wh-movement chains. While there were differences between overt and covert

A-bar movement with respect to relative clause modifiers, the NP-part of the moved

54
phrase was always represented in the trace position. Recall that an R-expression

that is part of the NP-part always triggers a Condition C effect in the trace position,

as illustrated by (41a): The pronoun he that c-commands the trace position cannot

be coreferent with the R-expression Kai. It is well known that A-chains differ from

A-bar chains in this respect. Namely, an R-expression that is part of the NP-part of

an A-moved phrase doesn’t trigger a Condition C effect in the trace position. This

is illustrated in (41b), where the R-expression Kai and the pronoun him, which c-

commands the A-trace ti , can be coreferent. (I’m concerned with the interpretation of

(41b) where one takes scope over seem. In case seem takes scope over one—the case

of Scope Reconstruction—, a Condition C effect is found as Fox (1997) and Romero

(1997) show. See also section 6.2.)

(41) a. ∗ [Which
[ relative of Kaij ’s]i did hej say ti likes Kazuko.

b. [One relative of Kaij ’s]i seemed to himj to ti like Kazuko.

On the view that binding reconstruction is represented by lexical material in the trace

position, the fact (41b) indicates that in A-chains no lexical material of the head is

represented in the trace position. On the copy theory of movement, the behavior of

A-chains seems unexpected, since nothing seems to motivate deletion of the lexical

material in the trace position. Recall though from the previous section that, while the

NP-part was always represented in the trace position in A-bar chains, relative clause

modifiers generally weren’t required to be represented in the trace position in chains

created by overt wh-movement. Then, (41b) shows that the NP-part in an A-chain,

55
which is created by overt movement, behaves in the same way that modifiers behave

with overt A-bar chains.

At this point, there are various ways to state the difference between A-chains

and A-bar chains. It seems to me that the difference between the NP-part and

modifiers in an A-bar chain is unexpected because semantically both the NP-part

and the modifiers are alike: they contribute predicates that form the restrictor of

the quantificational determiner which is heading the moving DP. Hence, I propose to

capture the difference between A-chains and A-bar chains by means of the condition

in (42), which stipulates the unexpected behavior of the NP-part in A-bar chains.

(42) In A-bar chains, the NP-part of the moving DP must be represented in the

lowest trace position.

Obviously it’s desirable to derive (43) from something, but, at this point, I must

relegate the issue to future research. I hope to show, however, that the difference

between A and A-bar chains at the level of logical form can be reduced to the condition

in (42). In the remainder of this section, I present some tentative results that relate

to this project concerning the distribution of Condition C effects with modifiers in

A-chains. Obviously there are other differences between A-chains and A-bar chain.

In section 5.2, I show that differences with respect to weak crossover follow from

(42). For the different behavior with respect to the licensing of parasitic gaps, I

refer the reader to Nissenbaum (1998). Nissenbaum shows that this difference can

be derived from syntactic locality differences between different types of movement,

56
namely whether intermediate adjunction is required. For the differences with respect

to locality, I again refer the reader to the respective literature (Rizzi 1990, Chomsky

1995, Takano 1993, 1994, Müller 1993, 1996), which reduces main differences between

different movement types to . The open question remaining, is how the difference with

respect to intermediate adjunction sites are captured on this approach. However, this

problem doesn’t directly relate to the issue of the LF-representation of chains, and

hence is not crucial for the following.

The position I take above is that the NP-part of an A-chain is subject to the

same principles that determine the distribution of modifiers in all chains. These are

discussed in the previous two sections; namely, a preference for the surface position

which can be overridden by variable binding or ACD. The interaction with variable

binding, leads us to expect cases with A-chains where the determiner of the moving

DP is separated from the NP-part. It is difficult to determine whether this expectation

is fulfilled, as we see in (43) and (44).

In (43), I tried to force reconstruction of the NP-part of the A-moved phrase.

The question we’re interested in is whether (43) has the LF-representation in (44a),

where one takes scope in its surface position, but himself is interpreted as bound by

everybody. However, since (43) definitely allows the representation in (44b), where

one takes scope below seem and everybody, it is impossible to discern whether there

are also readings with wide scope for one. The kind of reading we might expect

(44a) to have—and it’s not so clear what this might be—could also be a specific or

wide-scope reading of (44b) (see Fodor and Sag 1982, Reinhart 1997, Kratzer 1995)

57
(43) [One picture of himselfj ]i seemed to everybodyj to ti be too small.

(44) a. [One] λx seemed to everybodyj to [x, picture of himselfj ] to be too small.

b. seemed to everybodyj to [one picture of himselfj ] to be too small.

A better test are interactions between variable binding and Condition C. The

paradigm in (45) resembles that in (34) and the judgment is similar, though the

contrast seems to be less sharp.11

(45) a. ∗ [A
[ picture that hek showed to Maryj ]i seemed to herj to have been given ti

to every studentk .

b. [A picture of hisk that Maryj was shown]i seemed to herj to have been ti

given to every studentk .

c. ∗ [A
[ picture of hisk meeting with Maryj ]i seemed to herj to ti have been given

to every studentk .

In (46), a slight contrast in the predicted direction is found, though again even the

better example (46b) is not perfect. Here the reason might be the complexity of the

construction, and the fact that it’s generally hard to reconstruct in an A-chain if an

overt full DP intervenes.

11
One problem with the examples in the text might be a minor violation of weak crossover.
However, in examples like (i), weak crossover is even weaker than it usually is (Burzio 1986:203,
Pesetsky 1994:221-223, Pica and Snyder 1994).
?
(i) A picture of hisj mother seemed to have been given to every studentj .

58
(46) a. ∗ [A
[ letter that hisk mother sent to Maryj ]i seemed to herj to appear to every

studentk to be ti interesting.

b. [A letter of hisk mother that Maryj had received]i seemed to herj to appear

to every studentk to be ti interesting.

Another prediction of the assumption that the A/A-bar difference reduces to

(42) is that covert A and A-bar chains should behave alike (except if ACD is in-

volved) because in covert A chains the preference to represent the NP-part in its

surface position also predicts it will be represented there. Most cases discussed as

covert A-movement in the older literature, namely movement to replace an exple-

tive, don’t exhibit any of the semantic effects associated with movement, therefore

aren’t regarded as covert A-movement at this point. However, there’s one case in

Modern Greek which seems to disconfirm my prediction. Namely, Alexiadou and

Anagnastopoulou (1997) argue that certain cases of clitic doubling in Greek involve

covert A-movement, and are hence similar to overt scrambling in languages like Ger-

man and Japanese. As we see in (47) (Alexiadou and Anagnastopoulou 1997:147),

Condition C is obviated by the covert A-movement in (47b), which indicates that

there the NP-part of the A-chain doesn’t occupy its surface position, but the top

position of the A-chain.12 This could be a problem for the approach taken here, and

definitely deserves further study. The fact alone that this construction in Modern

12
Alexiadou and Anagnastopoulou (1997) also present a contrast similar to (47), but using weak
crossover. Since it’s known thought that severity of weak crossover is affected by Pesetsky’s (1989)
D-linking and since clitic doubling seems to bring about a discourse effect similar to D-linking, I
consider Alexiadou and Anagnastopoulou’s (1997) weak crossover facts unconvincing.

59
Greek might be the a case of covert A-movement, the only one known to me, is in-

teresting. However, Sabine Iatridou (p.c.) finds the contrast in (47) less clear than

Alexiadou and Anagnastopoulou (1997) indicate, and therefore I ignore (47) for now.

(47) a. ∗ O Janis tisi epestrepse [to vivlio tis Mariasi ]j simiomeno

the John she gave back the book of Mary with notes

b. ? O Janis tisi toj epestrepse [to vivlio tis Mariasi ]j simiomeno

the John she it gave back the book of Mary with notes

In sum, despite the tentative nature of the evidence presented, it seems feasible

to fit A-chains into the picture developed for A-bar chains in the previous two sections.

I adopt the assumption that the difference between A and A-bar chain can be reduced

to (42). I come back to the A/A-bar distinction in section 5.2 with a discussion of

weak crossover.

2.4 Relative Clause Internal Traces

The relationship between the head of a relative clause and the relative clause internal

trace position is puzzling, as was first pointed out by Munn (1994): As shown in (48a)

and with more examples below, no Condition C effect is triggered in this position.

On the other hand, as shown by (48b) and more examples below a variable contained

in the head can be bound in the relative clause internal trace position.

(48) a. Which is the picture of Johni that hei likes?

b. Which is the picture of himselfi that everybodyi likes?

60
 This section is concerned with the absence of Condition C effects in relative

clauses. More specifically, only restrictive relative clauses are considered. Relative

clause formation obviously involves A-bar movement as the locality restrictions move-

ment show. But, as evidenced by (48) and further examples below, the relation be-

tween the relative clause head and the relative clause internal trace is different from

that between the head of a wh-chain and its trace in a question. The conclusion I

argue for in section 2.4.1 is that the proposal of Carlson (1977) is essentially correct:

There are two possible LF-structures for relative clauses, a matching structure and a

raising structure, and the two can be distinguished by means of their interpretation.

In section 2.4.2, I show that the two structures have many things in common and I

propose a derivation that can generate both the matching and the raising structure.

The result of this section not only solves the puzzle (48), but is also important for

chapters 3 and 5 where additional evidence for this analysis of relative clauses will be

achieved.

As already mentioned, the relation between the external head and the trace

inside the relative clause seems to be less direct than with wh-movement in questions

with respect to Condition C, as pointed out by (Munn 1994, Safir 1998)13 In examples

like (49a) (repeated from (48a)), (50a), and (50a) no Condition C effect if observed

even though the R-expression John occurs inside the NP-part of the relative clause

head. In the wh-questions, in the corresponding b)-examples attest coreference is

blocked by Condition C.

13
In some examples, though, a Condition C effect is observed with relative clauses, as I also show
below. In particular, I address examples of this kind that are due to Schachter (1973) in footnote
14 below.

61
(49) a. Which is the picture of Johni that hei likes?

b. ∗ Which
W picture of Johni does hei like?

(50) a. The pictures of Marsdeni which hei displays prominently are generally the

attractive ones.(Safir 1998:(38a))

b. ∗ Which
W pictures of Marsdeni does hei display prominently.

(51) a. I have a report on Bob’s division he won’t like.(Merchant 1998a:fn.1)

b. ∗ Which
W report on Bobi ’s division will hei not like.

The Condition C evidence seems to show that there is no material of the

external head in the relative clause internal position. In other respects though, the

relation between the external head and the relative clause internal trace position

seems to be just as tight as that in a wh-chain. One such case, first observed by

(Jackendoff 1968, Schachter 1973:32-33), are examples where the head of the relative

clause contains a variable that is bound by an expression inside the relative clause as

in (52). While an example like (52b) might not require c-command for the binding,

the fact that in (52c) her can be interpreted as a bound variable indicates that in this

case her must be able to occur in the scope of every professor.

(52) a. The interest in each otheri that John and Maryi showed was fleeting.

(Schachter 1973:43a)

62
 b. Une photo de luii que Jeani avait donnée à Marie a été
A photo of him that John has given to Mary has been

retrouvée hier. (Vergnaud 1974:256)

found again yesterday

c. The book on heri desk that every professori liked best concerned model

theory.

A second case are examples where the head of the relative clause forms an idiom

together with other lexical material inside the relative clause (Brame 1968). The

noun headway in (53a) cannot appear in any other environment than as part of the

idiom make headway. This suggest the position where headway is interpreted in (53a)

is the complement position of make. (53b) allows both an idiomatic interpretation (the

pictures John made with a camera) and a non-idiomatic interpretation (the pictures

John grabbed ). For the idiomatic interpretation, the same point could be made, as

for (53a).

(53) a. The headway John made proved insufficient.

b. All the pictures John took showed the baby.

Finally, Irene Heim (p.c.) mentions examples where a part of the head of the relative

clause seems to take scope below a relative clause internal scope taking element. The

preferred interpretation of example (54a) is one that can be paraphrased as “Gina

needs so many books for vet school such that no linguist would read that many

books”. In this paraphrase, many takes scope below need. Similarly, (54b) prefers an

interpretation paraphrasable as: there is a number such that Mary can take n-many

63
drinks, but she shouldn’t even have n-many drinks. In this paraphrase as well, the

quantifier n-many drinks takes scope below the relative clause internal modal can.

(54) a. No linguist would read the many books Gina will need for vet school. (need

À many)

b. Mary shouldn’t even have the few drinks that she can take. (can À few)

In the evidence so far, the relationship between the relative clause internal

trace position and the external head is alike to that between head and trace in an

A-chain: While there is usually no reconstruction and hence no Condition C effects,

binding and scope can force reconstruction. By contrast, the relationship between the

relative clause operator and the trace position is exactly like that in a wh-movement

chain. Not only the locality restrictions and weak crossover point in this direction,

but the A-bar nature of the relative clause internal movement can also be shown using

Condition C as a test. As Safir (1998) observes, lexical material that is pied-piped

by the movement of the relative clause operator behaves exactly like lexical material

in wh-chains with respect to Condition C. (55) shows the contrast between material

pied-piped with the relative clause operator and the external head. The pronoun he,

which c-commands the relative clause internal trace position, cannot be coreferent

with the R-expression John that is part of the material pied-piped with the operator

in (55a). In (55b), where the R-expression is part of the external head, on the other

hand, coreference is possible.

64
(55) a. ∗ I respect any writer whose depiction of Johni hei ’ll object to. (Safir

1998:34a)

b. I respect any depiction of Johni hei ’ll object to.

The pairs in (57) and (58) show that a familiar argument/adjunct contrast with

respect to the relative clause internal operator movement. No Condition C effect is

found in (56a) and (57a), where the the R-expression is part of an adjoined modifier

to the constituent moved in the relative clause. (56b) and (57b) show a Condition C

effect just like (55a).

(56) a. There’s a singer whose picture in Johni ’s office hei ’s very proud of. (Safir

1998:(34b))

b. ∗ There’s
T a singer whose picture of Johni ’s office hei ’s very proud of.

(57) a. Max is a prince Johni ’s description of whom hei varies when spies are

around. (Safir 1998:(34c))

b. ∗ Max
M is a prince whose description of John hei varies when spies are around.

The well-behaved nature of this chain internal to the relative clause, makes the rela-

tionship between the external head and the internal trace all the more interesting.

2.4.1 Two LF-Structures for Relative Clauses

What explains the difference between the examples with Condition C and those in-

volving binding, idioms and scope? The explanation, I pursue is based on the idea of

65
Carlson (1977) that relative clauses are structurally ambiguous at LF. I’ll first con-

sider only the LF-structures. Following Carlson (1977), I call the two LF-structures

for relative clauses the matching analysis (Lees 1960, 1961, Chomsky 1965) and the

raising analysis (Schachter 1973, Vergnaud 1974). The differences between the two

LF-structures and the main prediction of Carlson’s ambiguity view—that Condition

C reemerges when the raising analysis is forced—are spelled out in this section. In

the section 2.4.2, I then look at the matching analysis in more detail and show how

the derivation of the two LF-structures could be unified. The details of the semantic

procedures that interpret the structures proposed in this section are left to chapter 5.

The two structures are sketched in (58) for an example that forces the match-

ing analysis and in (59) for and example that forces the raising analysis. On the

matching structure in (58b) and (59b), the external head and the internal trace are

I assume not related via movement. Therefore, the external head is represented in

the relative clause external position at LF, but at least not literally in the relative

clause internal position. In (58b) and (59b), none of the lexical material of the ex-

ternal head is represented in the relative clause internal position. Notice that to

capture the Condition C evidence, the structures in (58b) and (59b) represent only

one possibility. In section 2.4.2, I present an argument that the relative clause ex-

ternal head is represented in some sense in the relative clause internal trace position

on the matching analysis, and revise the matching structures accordingly. As shown

in (58b), the matching structure assumed here is predicted to obviate Condition C.

On the other hand, the position of the external head at LF in (59b) rules out the

matching structure in case it contains a variable bound by a quantifier inside the

66
relative clause (unless we assume that the quantifier can move to a position outside

of the relative clause). Similarly, the matching analysis is ruled out in the examples

(53) with idioms and (54) with scope.

(58) a. the picture of Johni hei likes

head
z }| {
b. the picture of Johni λx hei likes [x] (matching)
6

head
z }| {
∗
t picture of John λx hei likes [x, picture of Johni ] (raising)
c. the
6

(59) a. the picture of himselfi everybodyi likes

head
z }| {
∗
t picture of himselfi λx everybody likes [x] (matching)
b. the
6

head
z }| {
c. the λx everybodyi likes [x, picture of himselfi ] (raising)
6

The raising analysis is sketched in (58c) and (59c). Here, I assume that the relation

between the internal trace position and the external head is one of movement. There-

fore, the R-expression John must be represented in the trace position in (58c), just

like in the case of wh-movement. Hence, (58c) violates Condition C. On the other

hand, it is possible to delete all but the lowest copy of the NP-part of this chain as

in (59c), and therefore it’s possible to completely delete any relative clause external

appearance of the relative clause head. This is in fact required for binding in (59c),

as well as for idiom interpretation in examples like (53) and for narrow scope as in

the examples in (54).

As shown by (58), Condition C can be used to enforce the matching analysis of

67
a relative clause, while (59) shows that variable binding enforces the raising analysis

of a relative clause. By the same logic as that of (59), idiom interpretations and

scope can also be used to ensure that the raising analysis is forced. The ambiguity

analysis immediately makes one prediction, namely that factors forcing one analysis

are incompatible with factors forcing the other, and raises one question, namely which

of the two analyses is chosen when none of the factors seen to choose one analysis is

at work. I address the question first, and then demonstrate the prediction.

The question is which analysis of a relative clause is chosen if none of the

factors mentioned determines the analysis. Part of the answer can be found in the

previous work on relatives (Carlson 1977, Heim 1987, Grosu and Landman 1998),

who argue that there is a difference in interpretation between the two analysis. For

the raising analysis at least four different interpretations should be entertained: an

amount reading (Carlson 1977, Heim 1987, Grosu and Landman 1998) as in (60a), a

multiple individual reading (Geach 1964, Sharvit 1996a, Sharvit 1996b) as in (60b),

a possibility modal reading (Hackl and Nissenbaum 1998) as in (60c), and maybe

also a kind reading in (60d) similar to the one Heim (1987:27–33) observes for what-

questions. The cases of raising relatives noted above can be subsumed under these

four types, namely the idiom cases seem to have either an amount reading as argued

by Carlson (1977) or a kind reading, the binding cases clearly have the multiple

individual reading, and the scope cases all have an amount reading.

(60) a. It will take us the rest of our lives to drink the champagne they spilled

that evening. (Heim 1987:(40))

68
 b. The woman every man invited is waiting in the lobby.

c. Sabine has come up with many problems for us to work on. (Hackl and

Nissenbaum 1998:(1))

d. The beer that there was for sale was too expensive for John.

The availability of the four different readings seems to be subject to a number of

different constraints: for example, the amount reading and the possibility reading are

only available with certain determiners, and the multiple individual reading is most

easily possible for the argument of a copular construction. However, in general the

restrictions on the four readings are only incompletely understood. Unfortunately, the

detailed investigation of the semantics of the different readings and the restrictions on

them are beyond the scope of the current investigation (see chapter 5). Despite this

lack of precise understanding, I think it’s safe to proceed with the assumption that

the raising structure is only chosen in cases with one the above four interpretations.

Specifically, I assume that the raising analysis is only chosen if the NP-part must be

deleted in all positions but the relative clause internal trace position. As shown in

section 2.4.2, this allows a fairly uniform derivation of both the matching and raising

structures. This is obviously required if it NP-part contains a variable that is only

bound in this position, or if it’s an idiom chunk that can only be interpreted in this

position, or if it takes scope below another relative clause internal quantifier. For the

cases in the category kind relative, I suggest that they involve some form of binding

as well, for example of an event argument.

69
 The remainder of this section demonstrates two predictions of the analysis of

relative clauses pursued here. First, consider the prediction mentioned above: The

analysis of relative clauses as structurally ambiguous pursued here makes the clear

prediction that the factors forcing one analysis are incompatible with those forcing

the other analysis. The one factor that, on this account, definitely forces the matching

analysis is obviation of Condition C. In the following examples we see that in all the

constructions that motivated the raising analysis, Condition C cannot be obviated.

First consider variable binding in (61) and (62). In both (61a) and (62a), the pronoun

her is interpreted as a variable bound by a quantifier in the relative clause.14 As

discussed above, this forces the raising analysis and therefore the Condition C effects

observed in (61a) and (62a) between the R-expression John in the external head and

the pronoun that c-commands the relative clause internal trace position confirm the

analysis.15

14
Schachter (1973:32) discusses the examples in (ia) and (iia), where a Condition C effect is
observed. These might fall into place here under the assumption that nouns like opinion and portrait
have an implicit subject argument that in (ia) and (iia) is bound from a relative clause internal
position (Jackendoff 1972).The examples in (ib) and (iib), which don’t show a Condition C effect,
don’t have this confound.
∗
(i) a. The
T (proj ) opinion of Johni that hei thinks that Maryj has is unfavorable. (Schachter
1973:(41b))
b. The opinion of Johni that hei thinks that Maryj has refute is described in hisi letter to
her.
∗
(ii) a. The (proi ) portrait of Johni that hei painted is extremely flattering. (Schachter
T
1973:(42b))
b. The (proj ) portrait of Johni that hei ordered two years ago was finally delivered.

15
The contrast between (62a) and (i) is unexpected so far. It indicates that even when the external
head must stand in a movement relationship with an intermediate position of the relative clause
internal chain, it can nevertheless stand in the more indirect matching relationship with the lowest
trace of the same chain. This might indicate that, in fact, not only the relationship of the lowest
trace to the external head, but in fact every link of the relative clause internal chain is ambiguous
between a raising and matching analysis.
?
(i) A review of Johni ’s debate with herj that every senatorj wanted himi to read landed in the
garbage instead.

70
(61) a. ∗ The
T letters by Johnj to heri that hej told every girli to burn were published.

b. The letter by himj to heri that Johnj told every girli to burn were published.

(62) a. ∗ A review of Johni ’s debate with herj that hei wanted every senatorj to read

landed in the garbage instead.

b. A review of hisi ’s debate with herj that Johni wanted every senatorj to read

landed in the garbage instead.

The use of idioms is another way to enforce the raising analysis. As Munn

(1994) already observes, the prediction that Condition C effects reemerge is confirmed

as shown by the pairs in (63) and (64).

(63) a. ∗ the
t picture of Billi that hei took (Munn 1994:(15c))

b. the picture of himselfi that Billi took

(64) a. ∗ The
T headway on Mary’s project she had made pleased the boss. (Nis-

senbaum, p.c.)

b. The headway on her project Mary had made pleased the boss.

Also, narrow scope of many in (65a) and few in (65b) seems to cause a Condition C

effect in the expected fashion.

(65) a. ∗ The
T many books for Ginai ’s vet school that shei needs will be expensive.

(need À many)

71
 b. ∗ The
T few coins from Billi ’s pocket hei could spare weren’t enough for all the

needy. (could À few)

In fact, a Condition C effect is found also with other amount readings, as expected. In

(66), the amount reading is forced because the relative clause internal trace occurs in

a there-existential construction (Carlson 1977). This, as proposed by Carlson (1977)

and above, forces the raising analysis, and therefore the Condition C effect in (66a)

is expected.

(66) a. ∗ It
I would have taken us all year to read the letters for Johnj hej expected

there would be.

b. It would have taken us all year to read the letters for himj Johnj expected

there would be.

The second prediction of the analysis of relative clauses is more intricate. It is

made by the position of the lexical material of the head at LF in the raising analysis.

What we saw just now, is that the lexical material of the head occupies a raising

relative clause internal position, and triggers a Condition C effect there. In section

2.1 above, I showed for examples like (13b), which is repeated in (67a), that the head

of the relative clause can occupy a position outside of the relative clause and trigger

a Condition C effect there. The LF-representation in (67b), which was argued for

above, can obviously only hold for matching relatives. I therefore predict that raising

relatives will not show the Condition C effect noticed in (67a).

72
(67) a. ∗ In
I the end, I did ask himi to teach the book of Davidi ’s that Irene wanted

me to hask him to teachi.

h
b. ∗ the book of Davidi ’s [λy Irene wanted me to ask himi to teach the [y, book
i
of Davidi ’s]] λx I asked himi to teach the [x, book of Davidi ’s]

To verify the prediction, we need to look at examples where covert movement of the

DP containing the relative clause is forced by ACD, as it is in (67a), but where the

head of the relative clause must occupy a relative clause internal position at LF.16

The examples in (68a) and (69a) demonstrate that the prediction is correct. Both

contrast with (68b) and (69b), where there is no ACD to block binding reconstruction

of the relative clause. They also contrast with (68c) and (69c), where there is ACD,

16
Wold (1995:26) shows that sometimes ACD is incompatible with binding reconstruction into the
relative clause, as for example in (ib) and in (iib), where the judgment is actually stronger as Danny
Fox (p.c.) observes.
(i) a. Sue likes every picture of himselfi that Johni painted.
b. ∗ S
Sue likes every picture of himselfi that Johni does.
(ii) a. Sue likes every picture of himselfi that every boyi painted.
b. ∗ S
Sue likes every picture of himselfi that every boyi does.
Wold’s (1995) effect can be explained by the lack of identity between the elided VP and its antecedent
in the LF-representation (iva) of (iib). For the test in the text, however, we can circumvent it, because
Danny Fox (p.c.) also shows that Wold’s (1995) effect isn’t found if there is a relative clause internal
trace position outside of the elided VP, where the variable binding can be satisfied as in (iii). The
LF-structure of (iii) is shown in (ivb). The examples in the text have an intermediate position just
like (iii). Notice, however, that this analysis conflicts with the main proposal of section 2.4.2 in an
interesting way.
(iii) Sue likes every picture of himselfi that every boyi hoped she would.
∗
(iv) a. [[every λx every boyi likes [x, picture of himselfi ]] λy Sue likes [y]
| {z } | {z }
£ elided VP antecedent ¤
b. every λx every boyi hoped [x, picture of himselfi ] λz Sue would like [z]
| {z }
elided VP
λy Sue liked [y]
| {z }
antecedent

73
but the relative clause isn’t forced to have a raising analysis.17

(68) a. John asked himi for the pictures of herj mother meeting Clintoni every girlj

wanted him to hask Clintoni fori.

b. ∗ John
J asked himi for the pictures of herj mother meeting Clintoni every girlj

had published.

c. ∗ John
J asked himi for the picture of the woman meeting Clintoni every girlj

wanted him to hask Clintoni fori.

(69) a. The host introduced himi to the writers of herj replies to Casanovai every

girlj refused to hintroduce himi toi.

b. ∗ The
T host introduced himi to the writers of herj replies to Casanovai every

girlj had hired.

c. ∗ The
T host introduced himi to the writers of the letters to Casanovai every

girlj refused to hintroduce himi toi.

The LF-representation I propose for the example (68a) is sketched in (70), where

irrelevant details about the lowest relative internal trace position are omitted. Since

the head of the raising relative clause occupies a relative clause internal position, it

escapes Condition C for the same reason that material inside a raising relative was

found to do so earlier. Namely, ACD forces deletion of the copy of the relative clause

17
The judgement in these cases is made easier, if they’re put in the context of a little story. For
(a), for example, the story might say that John is investigating girls whose mothers had affairs with
Clinton. It’s known that Clinton maintains photographic records of his affairs, and the girls each
would like to see some of the pictures of their mothers with Clinton from his archives, but are afraid
to ask him. Therefore, John asks Clinton for the pictures.

74
in the QR-trace position, and therefore the R-expression occurs only in a position in

the head of the QR-chain in (70).

h
(70) the [λx every girlj wanted him [x, pictures of herj mother meeting Clintoni ]
i
to ask Clintoni for [x] λy John asked himi for [y]
| {z } | {z }
elided VP antecedent

The case in (71) makes the same point as (68) and (69), but the raising analysis of

the relative clause is forced by enforcing an amount reading of the relative clause.

Again, ACD-resolution in (71a) obviates Condition C even when the R-expression

occurs outside of the relative clause on the surface.

(71) a. The company will send her to any fan clubs of Mary there are requesting

it hthat the company send Mary to themi18 .

b. ∗ The
T company will send her to any fan clubs of Mary there are.

2.4.2 The Internal Head of Matching Relatives

In this section, I look at the matching analysis of relative clauses in more detail. The

LF-representation of a matching relative clause assumed in (58b) above is repeated

in (72b). In this section, I argue that the representation is instead that in (72c),

where the internal position contains an elided NP the antecedent of which is the

external head. The argument in this section is based on data from Safir (1998);

18
Obviously the it in this case is not the usual VP-ellipsis of the textbook cases. However, as
David Pesetsky (p.c.) pointed out to me, such antecedent contained anaphora are expected to and
do indeed behave exactly like ACD.

75
an additional argument for (72c) is given in section 3.1. Recall that also that the

discussion of example (13) lead us to propose the structure (17), which is essentially

like (72c). The proposal (72c) takes the term matching seriously: At some point of

the derivation the internal head must be (almost) identical to the external head. This

raises the question at what point of the derivation matching must be satisfied. I argue

that the point of the derivation where this matching requirement must be satisfied is

LF. As I show below, this assumption also allows us to (almost) reduce the raising

analysis to a special case of the matching analysis.

(72) a. the picture of Johni hei likes

head
z }| {
b. the picture of Johni λx hei likes [x]
6

overt NP elided NP
z }| { z }| {
c. the picture of Johni λx hei likes [x, picture of himi ]
6

Consider first what the absence of Condition C effects tells us about the matching

analysis. As we saw already above (examples (55) to (57)), the trace position inside

the relative clause must at least contain a representation of the NP-part of the material

that is pied-piped with the relative clause operator. The question here is to what

extent the external head is represented in the trace position in a matching relative

clause. The fact that the external head triggers no Condition C effects inside a

matching relative could be explained by various degrees of indirect representation,

for example the relationship between the pronoun and its antecedent in (73a) is

such that no Condition C effect is obtained, or that between the elided VP and its

76
antecedent in (73b).

(73) a. John drew a picture of Maryi , but shei didn’t like it hthe picture of Maryi i.

b. Mary loves Johni and hei thinks that Sally does hlove Johni i, too. (Fiengo

and May 1994:220)

To explain the latter observation, Fiengo and May (1994) propose that the identity

relationship between the elided VP and its antecedent is satified, even when an R-

expression in the antecedent corresponds to a coreferent pronoun in the elided VP

(see also sections 3.2 and 3.3 on the identity relationship). Fiengo and May (1994)

introduce the term Vehicle Change for such cases where exact identity of syntactic

form is violated. The structure (74) is the LF-representation Fiengo and May (1994)

propose for (73b). I adopt Fiengo and May’s (1994) proposal, as is already indicated

in the structure in (72c).

(74) Mary love Johni and hei thinks that Sally loves himi
| {z } | {z }
antecedent elided VP

The argument I present for the claim that the internal trace contains an elided

representation of the external head is based on the observation of Safir (1998:(35)) in

(75). In (75a), it’s impossible for the quantifier anyone in the external head to bind

the pronoun he in the relative clause. In (75b), on the other hand, binding of him by

anyone is possible. As the similar contrast in (76) confirms, the relevant difference

is whether the pronoun in the relative clause is c-commanded by the relative clause

internal trace.

77
(75) a. ∗ Pictures
P of anyonei which hei /hisi mother displays prominently are likely

to be attractive ones.

b. Picture of anyonei that put himi /hisi mother in a good light are likely to

be attractive ones. (Safir 1998:(35))

(76) a. ∗ Mary
M exhibited the picture of every boyi that hei /hisi sister brought.

b. Mary exhibited the picture of every boyi that was brought by himi /hisi

sister.

Since the quantifier in all four examples in (75) and (76), doesn’t c-command the

pronoun it binds in the surface structure, all examples might be expected to be weak

crossover violations. But, at least since Gabbay and Moravscik (1974), Hintikka

(1974), Reinhart (1976), and May (1977:61-124), it’s known that DP-internal quanti-

fiers can bind a pronoun outside quite easily, as long as they take scope over it. This

is illustrated in (77). In fact, the status of (75b) and (76b) seems comparable to the

examples in (77). On an analysis, where the external head is not represented at all

in the relative clause internal position, however, (75a) and (76a) would incorrectly be

expected to be as good as the examples in (77), as well.

(77) a. One picture of everyonei is displayed by himi prominently.

b. Somebody from every cityi despises iti . (May 1985:68)

Safir (1998) proposes that the bad examples of the contrasts in (75) and (76)

should receive the same explanation as the badness of the examples in (78), which

78
display strong and weak crossover.

(78) a. ∗ He
H i is displaying a picture of everyonei .

b. ∗ Which
W picture of everyonei is hei displaying?

c. ?? His
H mother is displaying a picture of everyonei .

d.?? Which
W picture of everyonei is hisi mother displaying.

For a raising relative, of course, any account of (78) carries over to the cases

under discussion. However, there’s no evidence that Safir’s examples must receive a

raising analysis. Moreover, the examples in (79) where the raising analysis is ruled

out by Condition C shows the same contrast as Safir’s example.

(79) a. ∗ The
T Times will generally publish pictures of any womani visiting Clintonj

that hej told heri about.

b. The Times will generally publish picture of any womani visiting Clintonj

that hej thinks will offend heri .

I believe that any account of Safir’s discovery on the matching analysis has

to propose a representation of the external head in the internal position, but not

one related by movement to the external head. The particular version of this I

assume here is that the internal head contains a phonologically deleted version of the

external head. Implicit in this proposal is that the external head and the internal

head must match at the level of LF, and there only, since this is generally the case

for phonological deletion, for example in ACD. If we furthermore assume that the

79
quantifier in the external head undergoes quantifier raising to position outside of the

DP, the LF-representation of (75) can be sketched as in (80), where I assume that

the anyone leaves the NP-part one in its trace position. In (80), the copy of [x, one]

in the relative clause internal trace is c-commanded by the pronoun hex . Therefore,

(80) is predicted to be a case of strong crossover.

h
∗
(80) anyone
a λx pictures of [x, one] [which picture of [x, one]] λy hex displays
| {z } | {z }
external head internal head
i
prominently [y, pictures of [x, one]] are likely to be attractive ones.
| {z }
internal trace

One more revision of the structure in (80) is required: Though the analysis

of (75) in (80) successfully predicts a strong crossover violation, it’s not generally

the case that an elided correspondent of a trace in the antecedent shows strong

crossover effects, as shown by (81a) from Fiengo and May (1994:279). Assuming

exact identity of syntactic form between the elided VP and its antecedent, (81b) is

the LF-representation of (81a). To resolve ACD, the to-object of (81) must undergo

quantifier raising to a position outside of the VP. But, since the direct object, every

guy, binds a varialbe in the relative clause adjoined to the to-object, it must undergo

quantifier raising as well to a position where it c-commands the raised to-object.

But, then (81b) violates the strong crossover condition: The antecedent VP contains

a trace of quantifier raising in the direct object position, and therefore the elided VP

in the relative clause does as well if we assume identity of syntactic form. This trace

in the elided VP, however, is c-commanded by a coreferent pronoun hex . Therefore,

(81b) violates strong crossover.

80
(81) a. Mary introduced every guy to every woman he wanted her to hintroduce

him toi

b. ∗ [every
[ guy] λx [every woman λz hex wanted her to introduce [x, guy] to [z, woman]]
| {z }
elided VP
λy Mary introduced [x, guy] to [y, woman]
| {z }
antecedent

But, the obviation of strong crossover in (81a) is not surprising on a view where

strong crossover is reduced to Condition C, since Condition C violations disappear

under ellipsis as was shown by (73). Extending their notion of vehicle change, Fiengo

and May (1994) propose that a trace in the antecedent in the antecedent of VP-

ellipsis, just like an R-expression, can correspond to a pronoun in the elided material

(see also Merchant 1998b). Adopting this assumption, the LF representation of (81)

is given in (82), where the direct object in the ACD-relative clause is a pronoun. Since

pronouns are not subject to strong crossover, (82) doesn’t violate strong crossover.

(82) [every guy] λx [every woman λz hex wanted her to introduce himx to [z, woman]]
| {z }
elided VP
λy Mary introduced [x, guy] to [y, woman]
| {z }
antecedent

Fiengo and May (1994) point out the contrast between (83a) and (81a), which

lends strong support to their account of (81a). It seems that in (83a) a strong crossover

effect is maintained, though the potentially violating trace is also part of an elided

VP. As Fiengo and May (1994) argue, the apparent strong crossover effect in (83a)

should be analyzed as a Condition B violation. Assuming that the correspondent of

the trace in the elided material is a pronoun, the LF-representation of (83a) in (83b)

81
violates Condition B since the this pronoun himx is in the local domain of another

pronoun himx . In (82), on the other hand, the pronoun himx that corresponds to the

trace is far enough away from the other pronoun, such that Condition B is satisfied

in (82).19

(83) a. ∗ Mary
M introduced every guy to every woman she wanted him to hintroduce

him toi

b. ∗ [every
[ guy] λx [every woman she wanted himx to introduce himx to] λy

Mary introduced [x, guy] [y, woman]

Condition B suffices to rule out Safir’s example (75), which is repeated in (84a).

(84b) shows the LF-representation of (84a) assuming that the trace of quantifier

raising of anyone in the elided occurence of the head is changed to a pronoun. This

pronoun is expected to violate Condition B, just like the pronoun in (85) which is

part of the fronted wh-phrase.

(84) a. ∗ Pictures
P of anyonei which hei /hisi mother displays prominently are likely

to be attractive ones.

19
In other examples of strong crossover under ellipsis, like (ia) and (iia) it seems the effect remains
even when the distance of the trace and the c-commanding pronoun is big enough to satisfy Condition
B. The illformedness of these examples is explained by the parallel dependencies condition (45) on
122 (see also Fox (1998c)).
∗
(i) a. T mani whoi Mary said that she likes and whoi hei did hsay that hei likesi too.
The
(Ristad 1990:144)
b. ∗ T
The mani whoi Mary said that Sue likes and whoi hei did hsay that Sue likesi too.

82
 h
b. ∗ anyone
a λx pictures of [x, one] [which picture of himx ] λy hex displays
| {z } | {z }
external head internal head
i
prominently [y, pictures of himx ] are likely to be attractive ones.
| {z }
internal trace
∗
(85) W
Which picture of himi does everyonei display prominently.

The explanation of Safir’s observation (75) as a violation of Condition B,

makes new predictions concerning the locality of the the effect. The prediction is

that the effect should be obviated if more material intervenes between the trace in the

relative clause internal head and the pronoun that triggers the Condition B violation.

That this prediction is correct is evidenced by the contrasts in (86) and (87). (86a)

(repeated from (76a)) and (87a) display the same degree of illformedness as Safir’s

observation. (86b) and (87b), where the quantifier is embedded more deeply in the

relative clause head, however, are markedly better.

(86) a. ∗ Mary
M exhibited the picture of every boyi that hei bought.

b. Mary exhibited the picture of every boyi ’s mother that hei bought.

(87) a. ∗ John
J bought a picture of every girli that shei chose.

b. John bought a picture of every girli ’s father that shei chose.

The improvement exemplified by (86b) and (87b) is predicted by my account of

Safir’s observation. Because the quantifier is more deeply embedded, the pronoun

corresponding to the trace inside the relative clause internal head is not in the local

domain of its antecedent, and therefore doesn’t violate Condition B. The contrast

83
between (86a) and (86b) is hence analogous to that between (88a) and (88b).

(88) a. ∗ Which
W picture of himi did every boyi buy.

b. Which picture of hisi mother did every boyi buy.

As second way to increase the distance between the two positions that give

rise to a Condition B violation in Safir’s example, is by making the relative clause

longer in the way shown in (89b) and (90b). In (89b) and (90b), however, if any, only

a very small improvement is found as compared to (89a) and (90a).

(89) a. ∗ Mary
M exhibited the picture of every boyi that hei bought.

b.∗? Mary
M exhibited the picture of every boyi that hei thought John bought.

(90) a. ∗ John
J bought a picture of every girli that shei chose.

b.∗? John
J bought a picture of every girli that shei thought he would choose.

But, the status of (89b) and (90b) should be measured against an example like (91b),

where Condition B in an intermediate position of a chain is at issue. Since (91b) seems

to be not fully grammatical, Conditon B seems to apply in intermediate positions of

a chain. But, if this is the case, (89b) and (90b) are expected to ungrammatical as

well.

(91) a. Which picture of herselfi does every girli believe Bob likes?

b.∗? Which
W picture of heri does every girli believe Bob likes?

84
 The correlation between Condition B effects and Safir’s observation confirms

the account of the latter proposed above. Therefore the strong crossover case of (75)

supports a version of the matching proposal where the internal trace is identical to

the external head modulo vehicle change. Since the weak crossover case of (75) relates

to the unsolved problem of why no weak crossover is found in cases like (77), which

I also have no solution for, I leave this matter open. Together with the arguments in

section 3.1 and that surrounding the structure (17), I believe to have given conclusive

evidence for a representation of the external head internal to a matching relative

clause.

The next argument concerns the question where matching of the internal and

external head applies in matching relatives. The belief expressed above, that the rela-

tionship between the two NPs is that of an elided NP and its antecedent presupposes

that matching is verified at LF. And the evidence given above for the effects of vehi-

cle change in matching relatives already lend strong support to this conclusion. The

following argument provides further evidence that matching applies at LF. Consider

(92), which is repeated from (34) above. I argued that, at LF, the NP-part paper of

hisk of the wh-phrase is represented in the trace position ti , while the relative clause

that Maryj was given is represented at LF in its surface position. The LF-structure

of (92) is shown (93).

(92) [Which paper of hisk that Maryj was given]i did shej tell every studentk to

revise ti ?

85
 h i
(93) Which [λz Maryj was given [z]] λx did shej tell every studenti to revise [x,

paper of hisi ]?

Notice that the LF-structure in (93) satisfies matching since both, the internal and

the external head, are empty. If, however, matching was applying before the higher

copy of paper of hisk is deleted, the relative clause in (92) would be expected to

contain a copy of it, and specifically the bound variable pronoun hisk . This pronoun,

however, would not be in the scope of its binder in the LF-representation of (92).

Hence, matching must apply in (92) after the overt copy of paper of his has been

deleted.

At this point, the raising analysis proposed above can be analyzed as special

case of the matching analysis with one remaining phonological stipulation. Specifi-

cally, the argument that the matching applies at LF and that matching is satisfied in

case both NP-parts are empty as in (92) suggests that this is the case in the raising

analysis as well. Consider again the matching and raising structures in (94) (repeated

with modifications from (59)). Empty NPs are indicated in (94c) by empty brackets

[]. Both the external head position and the complement of the relative clause oper-

ator could be occupied by an empty NP, and therefore the structure (94c) satisfies

matching.

(94) a. the picture everybody likes

head elided NP
z }| { z }| {
b. the picture which picture λx everybody likes [x, picture] (matching)
6

86
 head
z }| {
c. the [] which [] λx everybody likes [x, picture] (raising)
6

Since the relative clause operator is related to the trace position by movement, the

structure of (94c) before LF-deletion applies must be that in (95). This structure, does

however, not directly reflect the facts of English pronounciation: in raising relatives

the head is pronounced in front of the relative clause operator, just like in matching

relatives. For matching relatives, I have already assumed a pronounciation rule that

bans pronounciation of the internal head. To get the pronounciation of (95) right,

I suggest that just in case the external head is empty, the internal head is actually

pronounced, namely in the position of the external head. This is clearly a stipulation,

but at this point it seems to be the best I can do.

(95) the [] which picture λx everybody likes [x, picture]

6

The assumption that raising relatives are a special case of matching relatives

provides a straightforward explanation for the restricted occurence of raising relatives

noted in the discussion of (60). Recall from section 2.2 that the lexical material of the

top copy of a wh-chain can only be deleted if this is required for the interpretation

of a bound variable, that is not bound in this high position. If this generalization

applies to relative clauses as well (and also encompasses the cases of narrow scope

in (54)), it predicts that the copy of the internal head in the position of the relative

clause operator can only be deleted if it contains a variable that is bound internal to

the relative clause. Furthermore, if the copy of the internal in the operator position

87
doesn’t delete, matching requires the external head to be non-empty as well. There-

fore, the external head can only be empty, if the internal head contains a variable

that’s bound internal to the relative clause. This explains that the raising analysis

of relative clauses is restricted to cases with a ‘special’ interpretation in (60)—only

the special interpretation requires that the internal head be deleted in the operator

position.

To conclude this section, let me summarize the main points concerning rel-

ative clauses. Based on the contrasting behavior with respect to Condition C and

other tests for Binding Reconstruction, I concluded that there are two possible LF-

structures for relative clauses: a raising and a matching structure. Of these, I claimed

the raising analysis to always be associated with a ‘special interpretation’ as exempli-

fied by (60), whereas the matching analysis I claimed to be the default. Furthermore,

I concluded that the relative clause internal trace of a raising relative forms a chain

with the external head, whereas on the matching analysis the internal head consists

of its own lexical material, but must be phonologically deleted under identity with

the external head.

2.5 Summary

In this chapter, I argued for four main generalizations about which material of a

moved DP seems to enter binding theory in the trace position. The discussion above

has shown that it’s useful to distinguish three types of parts of a moved DP, the

determiner D, the NP-part, which is the lowest NP-projection (excluding all adjuncts)

88
of the complement of D, and relative clauses and other modifiers adjoined to the NP-

part. For the concise statement of the generalizations, I use the term segment to refer

to either the NP-part or any modifier of a DP. The terminilogy is exemplified in (96).

(96) which argument of John’s that Mary had criticized

| {z } | {z }| {z }
Det. NP-part modifier
| {z }
segments

The generalizations can be stated as in (97), as conditions governing when deletion

applies to a copy of the NP-part or a Modifier in a chain. The way the generalizations

are stated in (97) reflects a hierarchy between them with (97a) being the highest

ranked. Generalizations lower in rank, are only fulfilled up to the an extent such that

the higher ranked generalizations are fulfilled.

(97) a. Recoverability: At least one copy of every segment of the restrictor must

remain represented.

b. Binding: Any occurence of a segment that contains a bound pronoun that

isn’t c-commanded by its antecedent must be deleted.

c. A-bar: The lowest position of an A-bar chain must contain a copy of the

NP-part.

d. ACD: If material inside a modifier is anaphorically related to the con-

stituent surrounding an occurence of this modifier, this occurence of this

modifier must be deleted.

e. Lebeaux’s Generalization: Copies of a segment in positions lower than

the copy that is pronounced may be deleted (in a particular order).

89
 f. Economy of Deletion: Segments must not be deleted.

Of the six generalizations, (97b) to (97e) have been argued for in detail above,

while (97a) and (97f) have been more or less presupposed as background assumptions.

Both (97a) and (97f) play an important role in the account. (97f) is, for example,

responsible for the fact that quantifier raising doesn’t obviate Condition C in exam-

ples like (98a) (repeated from (9)), where ACD isn’t involved. If it was possible to

delete the lower copy of the relative clause modifier in the LF-representation in (98b),

Condition C should be obviated by quantifier raising in (98a), contrary to fact.

(98) a. ∗ Someone
S introduced himi to everyone Johni wanted you to dance with.
h i
b. ∗ everyone [λy J.i wanted you to dance with [y]]

λx someone introduced himi to [x, [λy J.i wanted you to dance with [y]]]

The Recoverability constraint (97a) is required for examples like (99). Since the

every copy of the modifier who knows heri in (99) will contain a bound pronoun that

isn’t c-commanded by its antecedent, (97b) would force deletion of all copies of this

modifier. This would incorrectly predict that (99) should be grammatical, namely

with the same interpretation as the sentence The boy thinks that every girl is singing.

(97a) blocks deletion of all copies of the relative clause in the LF-representation of

(99) and therefore (99) is correctly predicted to be ungrammatical.

∗
(99) T boy who knows heri thinks that every girli is singing.
The

90
 Of the four main generalizations (99b) to (97e), I assume (99b) and (99c)

throughout in the form stated. Of course, it would be desirable to derive them from

other principles, but at this point such a step seems premature. (97e), I assume

involves seemingly countercyclic adjunction because of the discussion of example (37)

above. (99d), finally, has curious nature since it seems to involve look-ahead to

interpretation in form of the licensing of ACD. Alternatively, (99d) might be a deletion

rule that always applies in the case of an VP-deletion dependency where the elided VP

is contained in the antecedent it depends. Then, (99d) would require no lookahead,

but the dependency of the two VPs would need to be formally represented. At the

end of section 3.2, I present one argument for the latter view of (99d).

91
92
Chapter 3

Identity of Traces

Certain constructions impose an identity (or parallelism) requirement on two con-

stituents. If these constituents contain traces, we can ask the question under what

conditions two traces are identical. This chapter argues that two traces are identical

in the relevant sense if the lexical content represented in the trace positions is the

same. Therefore, this result provides independent support for the claim of the previ-

ous chapter that the lexical content of a moved phrase is partially represented at LF

in the bottom position of a chain. More precisely, I show a perfect correspondence;

namely, the same parts of the moved phrase are represented in the trace position for

the concerns of Binding Theory (previous chapter) and the concerns of the Identity

Condition (this chapter). In section 3.2, I argue for the stronger claim that the lexi-

cal material in the trace position is not only represented there, but interpreted in the

trace position.

A major part of this chapter concerns the analysis of a restriction on ACD

that was first studied in detail by Kennedy (1994). The restriction is demonstrated

93
in (1a), where ACD is blocked. The contrast between (1a) with ellipsis and (1b)

without ellipsis shows that the ungrammaticality of (1a) is due to a restriction on

VP-ellipsis.

(1) a. ∗ Polly
P visited every town in every country Eric did hvisiti.

b. Polly visited every town in every country Eric visited

c. Polly visited every town Eric did.

Kennedy’s puzzle is then to explain why ACD is possible in (1c), but not in (1a).

Descriptively, the difference between (1a) and (1c) is the following: In (1c), the ACD-

relative clause is attached directly to the NP that will undergo quantifier movement

for the resolution of ACD. In (1a), on the other hand, the ACD-relative clause is

attached to an argument of this NP. In fact, it’s marginally possible in (1a) (and

much easier with an overt complementizer that in the relative clause) to attach the

relative clause to the higher noun town, in which case ACD is grammatical.

Since I keep referring back to the same example for most of this chapter, it’s

more convenient to talk about the contrast in (2) instead of (1). For (1), the natural

reading is one where every country takes scope over every town. But, this scope shift

is irrelevant for the discussion, and would make the LF-representations more complex

than needed. Therefore, I talk about example (2a), where the scope shift is not

needed. In (2a) the judgment is more subtle than in (1a), since (2a) is grammatical

on a reading of the elided VP clause as visited every town in t, which is more marginal

in (1a). The fact to explain though is that both (1a) and (2a) don’t have the reading

94
of the elided VP as only visited.

(2) a. ∗ Polly
P visited every town in a country Eric did hvisiti.

b. Polly visited every town Eric did hvisiti.

Assuming that traces are interpreted as variables, (3) shows the LF representations for

(2a) in (3a) and for (2b) in (3b). In both cases, the quantifier every town has moved

to resolve ACD. As a result of this movement, the elided VP and the antecedent are

identical in both (3a) and (3b). As I argue now, assuming the representations in (3)

would make it impossible to account for (2) in a principled way—not surprisingly so,

since the VPs in (3a) and (3b) are the same.

elided
z }| VP{ antecedent
z }| {
∗
(3) a. [every town, in a country Opy Eric visited [y]] λx Polly visited [x]
6
6

elided
z }| VP{ antecedent
z }| {
b. [every town, Opy Eric visited [y]] λx Polly visited [x]
6
6

Kennedy (1994) showed that the explanation of (2) must be a constraint on ellipsis,

as mentioned above. The only constraint on ellipsis usually assumed is an identity

(or parallelism) condition that the elided VP and its antecedent must satisfy, where

I for now assume an intuitive concept of identity that is sharpened in sections 3.2

and 3.3 (In the end, the condition I assume is very similar to that of Rooth (1992b).)

The only place where a difference could be made between the two VPs in (3b) are

the traces. Therefore, I assume that (3) is evidence for a condition on the identity of

traces that distinguishes (2a) from (2b). This assumption underlies all approaches to

95
Kennedy’s puzzle I know of, namely those of Kennedy (1994) and Heim (1997a) and

the one developed.

The question where the approaches disagree is: What makes the traces in

(3a) different, whereas those in (3b) are identical? Sag (1976:66,103) first suggested

that traces are only identical for the purposes of VP-ellipsis if their binders are the

same. Sag also develops a particular way to implement this suggestion, namely via

two restrictions that apply to the indices conventionally used to mark relations of

dependence: First, different dependencies, even when they don’t overlap, must use

different indices, and, second, an elided VP is only identical to its antecedent if the

indices on all unbound traces (and other variables) are the same. If relative clause

internal traces are viewed as bound by the DP the relative clause is attached to,

these considerations yield the LF-representations sketched in (4), where crucially the

indices of the traces in (4b) are identical, but not in (4a). Both Kennedy (1994) and

Heim (1997a) develop Sag’s idea and apply it to cases like (2). I call Sag’s approach

as well as its descendants the index identity approach.

elided 6= antecedent
z }| VP{ z }| {
∗
(4) a. [every town, in a country Opy Eric visited [y]] λx Polly visited [x]
6
6

zelided
}| VP{ antecedent
z }| {
b. [every town, Opx Eric visited [x]] λx Polly visited [x]
6
6

The reason I think Sag’s index identity is not the right approach to Kennedy’s puzzle

(2) are contrasts like (5). Both examples in (5) have the same structure. The only

difference between (5a) and (5b) is the head of the relative clause. Since this differ-

96
ence isn’t expected to affect the indexation possibilities, the index identity approach

predicts (5a) and (5b) to have the same status; namely, both should be ungrammat-

ical. This prediction is wrong: (5a) is clearly better than (5b). Sections 3.1 and

3.2 contain numerous contrasts like (5) which make sure that (5) is representative

of a real generalization. The failure of the index identity approach to account for

this generalization leads me to reject it and to pursue an alternative approach to

Kennedy’s puzzle. It should say, though, that while I reject index identity as an

approach to Kennedy’s puzzle, this doesn’t justify provide an argument against the

index identity condition per se, but only against an account of Kennedy’s puzzle (2)

based on the index identity condition. In fact, I present empirical support for the

index identity requirement in section 4.1 and discuss in section 4.2 which assumption

of the index identity approach should be given up. Since I present Heim’s (1997a)

version of the index identity approach there in more detail and my approach is based

on quite different assumptions, I don’t discuss it any further in this chapter.

(5) a. Polly visited every town that’s near the one Eric did hvisiti.

b. ∗ Polly
P visited every town that’s near the lake Eric did hvisiti.

The contrast in (5) shows that lexical properties of the antecedents of the

traces affect the acceptability of examples with the structure of Kennedy’s puzzle.

My approach to Kennedy’s puzzle is inspired by the idea of Chomsky (1993) that the

trace positions contain copies of the lexical material of their antecedents, which was

also discussed in the previous chapter. Hence, I call this approach the Copy Identity

97
Approach. Consider the sketched representations for (2) in (6). In (6), I repeated the

head noun of the antecedent in the trace positions. In the sketch (6a) for the bad

example, (2a), the antecedent is different from the elided VP. In (6b), on the other

hand, the antecedent and the elided VP are identical.

6= antecedent
z elided
}| VP { z }| {
(6) a. ∗ [every town, in a country Op Eric visited country] Polly visited town
6
6

z elided
}| VP { zantecedent
}| {
b. [every town, Op Eric visited town] λx Polly visited town
6
6

I claim that the lexical material in the trace positions in the way captured by (6) is the

right explanation of Kennedy’s puzzle. This copy identity approach is developed in

section 3.1. It is shown, in particular, that the copy identity approach directly predicts

the contrast in (5) and similar such contrast. Another point, section 3.1 discusses

that relationship of the copy identity approach to the Condition C evidence discussed

in chapter 2. Notice that in (6) only parts of the moved phrases are represented in the

trace positions. Section 3.1 shows that the copy identity approach and Condition C

converge on the same conclusion as to which parts of a moved phrase are represented

in the trace position.

Section 3.2 makes a new argument concerning the lexical material represented

in the trace position, that goes beyond what could be tested using Condition C in

chapter 2. It argues that the lexical content of the trace position is not only formally

represented there, but contributes to interpretation in the trace position. I argue for

this based on the observation that the acceptability of examples that test the identity

98
of traces depends on the semantic relationship of the lexical content of the traces, as

well as on general grounds.

Section 3.3 considers facts like (7) where no effect of the copy identity is

observed, thought the elided VP and its antecedent contain traces with different

lexical content. I show that two mechanisms can circumvent the effect of the copy

identity requirement: focus percolation into the trace position and a kind of sloppy

reading. The former, I argue in section 3.3.2, applies to example (7a), while the latter

applies to (7b) as shown in section 3.3.3.

(7) a. I know which cities Mary visited, but I have no idea which lakes she did

hvisiti. (= (71a))

b. The cities Mary visited are near the lakes Bill did hvisiti. (= (71b))

3.1 A Copy Identity Account of Kennedy’s Puzzle

This section begins to develop an account of Kennedy’s observation (2) based on

the view that lexical material of the head of a chain is partially represented in the

bottom position of the chain. One of the conclusions of the previous chapter was

that A-bar traces of a DP always contain the lexical content of the NP-part of the

moved DP. (Recall, that I defined NP-part as the NP that is the sister of the D-head

of the DP minus all adjoined modifiers). In ACD-constructions, specifically, section

2.1 argued based on Condition C that exactly the NP-part is represented in the trace

position, whereas all the quantifier and the ACD-relative are represented only in

the top position of the QR-chain. Furthermore, the analysis of matching relatives

99
of section 2.4 argued that the relative clause contains an unpronounced copy of the

external head of the relative clause. If this is true of the ungrammatical example from

(2), repeated as (8a), the LF-representation must be (8b), which essentially the same

as (6a). The elided VP and its antecedent differ in (8b) with respect to the lexical

material that appears in the trace position. My proposal is that this difference blocks

VP-ellipsis in (8a).

(8) a. ∗ Polly
P visited every town in a [country Eric did hvisiti].

b. ∗ [every
[ town in a [country, Opy Eric visited [y, country]]]
| {z }
elided VP
λx [Polly visited [x, town]]
| {z }
6= antecedent

Compare (8) with the grammatical example of (2), which is repeated in (9a): The

lexical content of the two traces, the one in the elided VP and the one in the an-

tecedent, is identical, namely [x, town]. This is shown by the LF-representation in

(9b).

(9) a. Polly visited every town Eric did.

b. every [Opy Eric visited [y, town]] [λx Polly visited [x, town]]
| {z } | {z }
elided VP antecedent

In both (8b) and (9b) the names of the variables, x and y, inside the traces differ

between antecedent and elided VP. For now, assume that the names of variables are

ignored by the identity condition. In section 4.1, I argue contrary that this assumption

is in general wrong, but in section 4.2, I argue that in examples like (8b) and (9b)

the identity requirement for variable names can be circumvented. For the moment,

100
it’s easiest to assume that variable names generally don’t matter.

Now, consider one prediction of the copy identity approach already hinted at

in the introduction with (5). This prediction is that if the antecedents of two traces

have the same NP-parts, the traces should be considered identical even if the two

operators binding the traces are different. Consequently, ACD should be possible.

The contrast in (10) shows that this prediction is correct. (10b) is basically the same

as Kennedy’s example (8a). In (10a), however, the NP to which the relative clause

is attached to and the NP-part of the object quantifier are lexically identical. If the

second occurrence of town is destressed in (10a), the example is fully acceptable.1

(10) a. John visited every town near a town Mary did hvisiti.

b. ∗ John
J visited every town near a lake Mary did hvisiti.

The LF-representation for (10a) is shown in (11). The trace-positions in the elided

VP in the relative clause and the trace of quantifier raising in the antecedent both

have town as its lexical content, and therefore the elided VP and its antecedent mean

1
Some English speakers don’t find the improvement in (10a) very strong, but everybody I con-
sulted with found a strong contrast in the examples with one-anaphora below. I assume that speakers
who find (10a) unacceptable differ from those who do in whether they find it natural to destress the
second occurrence of town.
The destressing requirement is probably due to contribution stress would make to the meaning
in this construction. Consider (i), where also a repeated occurrence of the noun book is stressed:
The stress indicates a contrast between the book John read and the book Mary read, with respect
to their ‘bookness’. In effect, (i) entails that what John read wasn’t really a book. Therefore, I
assume the two nouns book in (i), despite having similar phonology, differ in the sense relevant for
the identity condition on traces. The destressing requirement argues therefore that the identity
required isn’t identity of lexical form, but identity of meaning. Section 3.2 presents more arguments
for this conclusion.
(i) John read a book and Mary a BOOK.

101
the same.2

h i
(11) every town near a [town, Opy Mary visited [y, town]] λx John visited [x, town]
| {z } | {z }
elided VP antecedent

It’s important to go through the argument that (10) makes to see that it’s

independent support for the result of the previous chapter: (10) shows that for the

well-formedness of ACD the head nouns of the antecedents of the two traces involved

must be identical, namely of the trace of QR and of the trace internal to the relative

clause. Why would there be such a requirement? As already mentioned in the

introduction, it’s established that an elided VP must be identical to its antecedent.

Therefore, I conclude that the head nouns are represented in the elided VP and its

antecedent, respectively. The only part of the two VPs related to the head nouns

are the traces. Hence, it’s natural to assume that, if anywhere, the head nouns are

represented in the trace positions. Therefore, (10) argues that the head noun of a

2
The copy identity approach shares the prediction (10) with—at least a benevolent interpretation
of—a proposal of Lappin (1984). Lappin proposes, in effect, that two traces or pronouns are identical
if they can be naturally interpreted as having the same intended range of possible values. (Lappin
1984:(10)) He, however, doesn’t discuss contrasts like (10) and his proposal is too vague to be sure
of this prediction. There are other differences between the copy identity approach I’m developing
and Lappin’s proposal. For one, Lappin doesn’t derive the identity condition from properties of the
semantic representation in the it’s done here, but suggests that the condition is pragmatic which,
as far as I can see, he presents no motivation for. Secondly, Lappin’s condition applies to all traces
and bound pronouns, which isn’t true of the copy identity approach pursued here as discussed in
section 3.3. The examples in (ib) and (iib) show that this aspect of Lappin’s (1984) proposal makes
wrong predictions (see also Fiengo and May 1994).
(i) a. Here is the man who Bill saw, and here is the man who he didn’t hseei. (Lappin
1984:(21b))
b. Here is the man who Bill saw, and here is the woman who he did hseei.
(ii) a. [Every friend of John’s]i wants Mary to kiss himi , but [none of the little fellows]j believes
that she will hkiss himj i (Lappin 1984:(10))
b. [Every friend of John’s]i wants Mary to kiss himi , while [every friends of Bill’s]j wants
Sue to hkiss himj i.

102
QR-chain is represented in the trace position, and that the head noun of the relative

clause external head is represented in the relative clause internal head position.

Notice that the argument is independent of the arguments given in chapter 2

in favor of the same conclusion. In 2.1, I argued with example (12a), repeated from

(13b) on page 40, that for Condition C the NP-part phrase that moves for ACD-

resolution remains represented in the trace position. The LF-representation proposed

for (12a) is repeated in (12b).

(12) a. ∗ In
I the end, I did ask himi to teach the book of Davidi ’s that Irene wanted

me to hask him to teachi.

h i
b. the book of David’s λy Irenei wanted me to ask him to teach [y, book of David’s]
| {z }
elided VP
λx I asked himi to teach [x, book of David’s]
| {z }
antecedent

Furthermore, I argued in section 2.4 based on Safir’s (1998) discovery in (13a), re-

peated from (75) on page 78, that also the relative clause internal trace position

contains lexical material. Namely, if the NP-part of the head of the relative clause is

represented there as in (13b), (13a) is predicted to violate strong crossover.

(13) a. ∗ Pictures
P of anyonei which hei displays prominently are likely to be attrac-

tive ones.
h
b. ∗ anyone
a λx pictures of [x, one] [which picture of [x, one]] λy hex displays
| {z } | {z }
external head internal head
i
prominently [y, pictures of [x, one]] are likely to be attractive ones.
| {z }
internal trace

103
The argument based on (10) provides independent confirmation of these two conclu-

sions of chapter 2. In the remainder of this section, I give further evidence for this

interpretation of (10) and the parallelism to the arguments of chapter 2. I start by

adding some more examples just like those in (10), then I show that it is not just

the head noun, but the NP-part that matters for the identity of traces, just like it

does for binding theory as argued in chapter 2. Finally, I show a difference between

A- and A-bar-movement that parallels the A/A-bar distinction found with respect to

binding.

Both examples in (10) marginally allow an interpretation of the elided VP as

visit every town near t. This is expected because extraction out of a reduced relative

clause is marginally possible as in (14a), and on this reading the operator binding

both traces is the same.

(14) a. ?? Which
W lake did you visit every town near?

b. ∗ Which
W lake did you visit every town that’s near?

As (14b) shows, extraction out of a full relative clause is impossible. The examples in

(15), repeated from (5) in the introduction, and (16) show a similar contrast to (10),

but the don’t allow a different reading of the elided VP than the indicated one.

(15) a. John visited a town that’s near the town Mary did hvisiti.

b. ∗ John
J visited a town that’s near the lake Mary did hvisiti.

(16) a. Jon ordered a drink that’s more expensive than the drink Sue did horderi

104
 b. ∗ Jon
J ordered a drink that’s more expensive than the dish Sue did horderi

The repetition of the same noun within one sentence is usually a little unnat-

ural and most speakers prefer to replace the second occurrence with a one-anaphor.

As the examples in (17) show, the good examples of (14), (15), and (16) are also

good with a one-anaphor in place of the repeated noun. We can ignore the ques-

tion whether one-anaphora are analyzed as NP-ellipsis (Lakoff 1968) or NP-pronouns

(Jackendoff 1977:58-60); on either assumption the facts in (17) are expected: Since

on either one the one anaphor is, semantically at least, not different from a full NP

that could be used to paraphrase it, the examples in (17) are expected to behave just

like (10a), (15a), and (16a).

(17) a. John visited every town near the one Mary did hvisiti.

b. John visited a town that’s near the one Mary did hvisiti.

c. Jon ordered a drink that’s more expensive than the one Martin did horderi

Consider the LF-representation of (17b) given in (18). In (18), the lexical content

of the trace position in the relative clause is indicated as town, though the external

head of the relative clause is one. However, the representation in (17b) is possible

if either of the following two assumptions is correct: one is a phonologically reduced

expression of town, or the content of the internal head of a matching relative must

have the meaning as that of the external head. I believe that, at the least the

latter assumption is correct for the reasons given in section 2.4. Hence, the LF-

105
representation in (18) is possible, and satisfies the identity requirement of the elided

VP straightforwardly.3

h i
(18) a town that’s near the one λy Mary did visit [y, town] λx John visited [x, town]
| {z } | {z }
elided VP antecedent

Jacobson (1998a) points out that the ungrammaticality of Kennedy’s examples

is also found in cases like (19a), where ACD is resolved by overt wh-movement rather

than by covert movement. In (19a), the antecedent of the trace of wh-movement is

town while the head of the relative clause is lake. As we see in (19b) and (19c), if the

two nouns are the same or one is anaphoric to the other, the example improves.

(19) a. ∗ Do
D you know which town near a lake Mary did hvisiti John visited?

b. Do you know which town near a town Mary did hvisiti John visited?

c. Do you know which town near the one Mary did hvisiti John visited?

Example (20) shows that the judgement doesn’t change if the other VP is elided—

the one that contains the trace of wh-movement. As Jacobson (1998a) already notes,

(20a), doesn’t allow deletion of the VP containing the trace of wh-movement. The

contrast with (20b) and (20c) shows that, again, the difference of lexical content of

the antecedents of the two trace in (20a) causes the ungrammaticality.

(20) a. ∗ Do
D you know which town near a lake Mary visited John did hvisiti?

3
In the next section, I present arguments that the kind of identity required between the elided
VP and its antecedent is identity of meaning. This implies that even if one occupies the relative
clause internal trace position, ACD should be possible as long as one means the same as town, the
content of the QR-trace in the antecedent.

106
 b. Do you know which town near a town Mary visited John did hvisiti?

c. Do you know which town near the one Mary visited John did hvisiti?

This argues for the LF-representation of (20a) in (21) (and it would be the one of (19a)

if the labels antecedent and elided VP were interchanged). In (21), the head noun of

the wh-phrase is represented in the position of the wh-trace, while the head noun of

the relative clause head is represented in the relative clause internal trace position.

Again, the identity requirement of the elided VP and its antecedent is clearly violated

by the content of the traces. That ACD is possible in the b) and c) examples of (19)

and (20), on the other hand, shows that the relative clause isn’t represented in the

position of the wh-trace.

h i
∗
(21) which town near a lake λy Mary visited [y, lake] λx John visited [x, town]
| {z } | {z }
6= antecedent elided VP

Note that, again, the conclusion just reached parallels the conclusion pointed out in

section 2.2, namely that overt wh-movement also represents the lexical content of the

NP-part in the trace position. There the argument that the NP-part of a wh-phrase

must be represented in the trace position, while a relative clause adjoined to it need

not be, was the argument/adjunct asymmetry with Condition C of Freidin (1986)

and Lebeaux (1988) illustrated in (22) (repeated from (20) on page 44).

(22) a. ∗ [Which
[ argument that Johni was wrong]j did hei accept tj in the end?

b. [Which argument that Johni had criticized]j did hei accept tj in the end?

107
 So far, there is one apparent difference between the conclusions reached here

based on identity conditions on traces and the conclusions reached in chapter 2 based

on the distribution of Condition C effects. Namely, the discussion in this section

has produced arguments that the head noun of an antecedent is represented in the

trace position. The distribution of Condition C effects argued that the NP-part,

which is the head noun plus its complement, is represented in the trace position.

The difference is only apparent: In the examples so far, the NP-parts of the relative

clause head and the DP moving for ACD-resolution consisted only out of the head

noun. The paradigm in (23) shows that similar contrasts are also found in a case

where the NP-parts have the same head-noun but the arguments of the head noun

is different. However, repeating a complex NP within the same sentence as in (23b)

is so unnatural that the contrast to (23c) is very weak. The contrast between (23a)

and (23c) is clear, though. (24) shows the relevant aspect of the LF-representation

proposed for (23c)—the elided VP is not identical to its antecedent.

(23) a. Bill gave a description of Mary that’s similar to the one John did hgivei

b.?? Bill
B gave a description of Mary that’s similar to the description of Mary

John did hgivei

c. ∗? Bill
B gave a description of Mary that’s similar to the description of Sue John

did hgivei

h
∗
(24) a description of Mary that’s similar to the description of Sue λy John did
i
give [y, a description of Sue] λx Bill gave a [x, a description of Mary]
| {z } | {z }
elided VP 6= antecedent

108
The lack of a contrast in (25) below shows that different adjuncts of the antecedents of

the two traces don’t block ACD. One possible explanation could be that the difference

between (23c) and (25b) mirrors the argument/adjunct distinction of binding theory.

A conclusive judgement on this issue, however, would need to take into account the

considerations brought up in the next section 3.2. (See footnote 6 below).

(25) a. John visited a town near Madrid that had signs for the one Bill did hvisiti.

b. John visited a town near Madrid that had signs for the one near Rome Bill

did hvisiti.

Example (26) is another place where the predictions of an identity of NP-parts

requirement differ from an identity of head noun requirement. In (26), the head of

the relative clause is an argument of the noun heading the NP-part of the DP that

moves for ACD-resolution: Even though the head-nouns of the two NP-parts involved

in (26) are identical, the examples are ungrammatical. The two NP-parts itself aren’t

identical in (26), because one contains the other. For example, in (26a), the NP-part

of the relative clause head is only picture, but the NP-part in the antecedent is picture

of a picture. This is captured in the representation in (27).4

(26) a. ∗ Susi
S produced a picture of a picture Meltem did hproducei.

4
The LF-representation I actually assume for (26a) must contain the relative clause in the trace
position as in (i), and therefore doesn’t allow ACD (See the discussion of example (28) on page 47).
With respect to question of whether the NP-part of the noun head is represented (i) leads to the
same conclusion.
(i) [a picture of a picture Meltem did produce] λx Susi produced [x, picture of a picture Meltem
did produce]

109
 b. ∗ Jonathan
J visited every relative of the relative Danny did hvisiti.

h i
∗
(27) a picture of a picture λy Meltem produce [y, picture]
| {z }
elided VP
λx Susi produced [x, picture of a picture]
| {z }
6= antecedent

The contrast in (28) brings out the difference between arguments and adjuncts as

a minimal pair. For the judgement, imagine that John’s art is painting pictures of

Dali’s pictures. One day, John meets Dali and Dali tells him about his plan for a new

great painting. John likes the plan a lot, and immediately makes his own plans based

on Dali’s plan. In this context, (28b) is an acceptable sentence, but (28a) remains

unacceptable.

(28) a. ∗ John
J is planning to paint many pictures of the one Dali is hplanning to

paint.i

b. John is planning to paint many pictures showing the one Dali is hplanning

to paint.i

Since the head of the relative clause one is an argument of the higher NP in (28a)

and therefore inside the NP-part of the DP that moves for ACD-resolution, (28a) is

expected to be bad. (28b), on the other hand, is expected to have the same status as

(10) because the head of the relative clause is contained in an adjunct to the higher

NP-part.

110
 Given the parallelism of chapter 2 and the conclusions here, it’s expected that

a difference between A- and A-bar-chains is also found with the trace identity require-

ment. Recall from section 2.3 that A-chains and A-bar-chains differ with respect to

Condition C as illustrated in (29) (repeated from (41) on page 55): While Kai in

the A-bar moved phrase behaves as if in the trace position with respect to Condition

C, the R-expression Kai in (29b) can be coreferent with the pronoun him. Hence,

section 2.3 concluded that the requirement that the NP-part must be represented in

the trace position of a chain, only applies to A-bar chains.

(29) a. ∗ [Which
[ relative of Kaij ’s]i did hej say ti likes Kazuko.

b. [One relative of Kaij ’s]i seemed to himj to ti like Kazuko.

The examples in (30) and (31) show a contrast between topicalization (A-

bar-movement) and passivization (A-movement) that argues that the requirement on

trace identity is sensitive to the A/A-bar-distinction as well. Namely, the passive

examples in (30a) and (31a) are acceptable, while the topicalization cases in (30b)

and (31b) are ungrammatical.

(30) a. The town near the lake that was hvisited by vandalsi seems to have been

visited by vandals, as well.

b. ∗ The
T town near the lake they did hvisiti, the vandals seem to have visited,

as well.

111
(31) a. The town near the lake that was visited by vandals seems to have been

hvisited by vandalsi, as well.

b. ∗ The
T town near the lake they visited, the vandals seem to have hvisitedi, as

well.

These contrasts argue, based on the identity criterion, that the trace in an A-chain

need not contain lexical material of the antecedent. Consider the LF-representation

of (30a) in (32). If both the elided VP and its antecedent contain only a variable

in the object position, but not the lexical material of the antecedent, the identity

condition of VP-ellipsis is satisfied.

h i
(32) The town near the lake λz [z, lake] λy was visited by vandals [y] λx seems
| {z }
elided VP
to have been visited [x] by vandals, as well.
| {z }
antecedent

Therefore, the contrasts in (30) and (31) provide independent support for the A/A-bar

distinction as stated in 2.3. There are, however, examples like (33) where deletion of a

VP containing an A-trace is blocked, even though the identity condition is predicted

to be satisfied. (34) is the LF-representation of (33a) assuming that the VP internal

subject hypothesis, which claims that the subject A-moves from a VP-internal posi-

tion to its surface position (see Webelhuth 1995:60-64 and references therein). In fact,

most of the examples discussed in the papers of Kennedy (1994) and Heim (1997a)

are examples with A-traces, and Kennedy and Heim both view examples like (33) as

support for the index identity view.

112
(33) a. ∗ A proof that God exists does hexisti (Wasow 1972:93)

b. ∗ Every
E man who said George would buy some salmon did hbuy some salmoni

(Kennedy 1994:(2b))

(34) A proof that God λy exists

| {z } λx does [x] exist
| {z }
antecedent elided VP

As Kennedy (1994:fn. 3) notes and Heim (1997a) discusses in detail, the grammati-

cality of examples like in (33) improves for many speakers with the addition of focus

particles like too, as well, or instead, which is not the case for examples with A-bar

movement like (2). Hence, I reject the conclusion that the examples in (33) should

receive a similar explanation as Kennedy’s A-bar movement cases. Example (33a)

is probably ill-formed because it requires scope reconstruction of the subject into a

VP-internal position (see for example Diesing 1992). In section 4.1, I provide an ac-

count that predicts that examples like (33b), while not ungrammatical, are difficult

to parse.

3.2 Semantic Content of the Trace

Two independent lines of argumentation established that parts of the antecedent of

trace are represented in the trace position at the LF-level. In chapter 2, the argument

was based on the distribution of Condition C effects. In the previous section 3.1, I

presented an argument based on the identity condition between traces imposed by

VP-ellipsis. While this correspondence is quite remarkable, it still leaves it open what

the contribution of the lexical material in the trace is to interpretation. Up to now,

113
it’s conceivable that the material in the trace doesn’t contribute to interpretation at

all, except in the cases of variable binding in section 2.2. In this section, I argue that

in the lexical material in the trace position is also interpreted there—it constitutes

the semantic content of the trace. In particular, I show that the range of entailments

drawn from a constituent containing a trace, but not its antecedent, is affected by

the semantic content of the trace.

The alternative position I’m arguing against here doesn’t, at least at this point,

look very attractive. The assumption that the lexical material in a trace position isn’t

interpreted there, but in the position of the antecedent, would necessitate the follow-

ing additional assumptions: To begin with, it requires the assumption that the lexical

material in the trace position is also represented in the antecedent, so that no infor-

mation is lost if the material in the trace position is ignored. This assumption is

unproblematic (In fact, I have been making this assumption throughout and give an

argument for it in section 3.3.3 below), except when the fronted material contains a

variable in (35a) (repeated from (34) on page 50). Here, I have been assuming a rep-

resentation like that in (35b), where the part of the fronted constituent that contains

the bound variable is only represented in the trace position of the wh-chain. On the

assumption that normally, lexical material in the trace position is ignored, either the

case of a bound variable must constitute an exception to this, or additional semantic

mechanisms that allow the interpretation of a bound variable in a position outside

the c-domain of its binder must be postulated. However, such mechanisms have been

postulated; for example Skolem-functions in Engdahl (1986) and Chierchia (1993)

and abstraction over assignments as in Sternefeld (1998) (technically a generalization

114
of Skolem-functions).

(35) a. [Which paper of hisk that Maryj was given]i did shej tell every studentk to

revise ti ?
h i
b. Which [λz Maryj was given [z]] λx did shej tell every studenti to revise

[x, paper of hisi ]?

A second consequence of the assumption that the lexical content of the trace is se-

mantically vacuous, is the existence of two kinds of deletion at the LF-level. This is

clear in examples like (36a) (repeated from (2)), and the same point could be made for

(13b) in section 2.1. In the QR-chain of the LF-representation of (36a), as repeated

in (36b), the ACD-relative clause is represented only in the operator position of the

chain, while the NP-part of the QR-chain is represented in both the operator and the

trace position of this chain. However, if the lexical material in the trace position is

also not entering interpretation in that position, (36) in effect involves two steps of

deletion: Before Condition C and the identity condition of ellipsis apply, the relative

clause is deleted in the position of the QR-trace. After the two conditions applied,

the NP-part is deleted in the trace position. While this position isn’t incoherent, it’s

also not particularly attractive from my point of view since the second step of deletion

operations seems unmotivated.

(36) a. ∗ Polly
P visited every town in a country Eric did hvisiti.

b. ∗ every
e [town, in a country Opy Eric visited [y, country]]
| {z }
elided VP

115
 λx Polly visited [x, town]
| {z }
6= antecedent

The argument I give now in favor of the trace actually having semantic content

is quite a bit stronger than the preceeding two arguments. The argument comes from

a closer look at Kennedy’s restriction on ACD. Consider the data in (37): (37a)

and (37b) are repeated from (15) above, however (37c) is new. Surprisingly, (37c)

is almost as good as (37a), though the lexical material of the QR-trace is predicted

to be town while that of the relative clause internal trace is city. As in (37a), the

judgement requires leaving city unstressed; if city is stressed, (37c) is unacceptable

(cf. footnote 1).5

(37) a. John visited a town that’s near the town Mary did hvisiti.

b. ∗ John
J visited a town that’s near the lake Mary did hvisiti.

c. ? John
J city Mary did hvisiti
| {z } that’s near the |{z}
visited a town
QRNP RCNP

For the following discussion it is convenient to have the following two terms at our

disposal: The NP-part of the DP that moves covertly for ACD-resolution I call QRNP,

and the NP-part of the head of the ACD-relative I call the RCNP. The empirical

generalization argued for in section 3.1 could then be stated as follows: ACD is

possible if and only if QRNP is equal to RCNP. (37c) shows that this generalization

is not exactly correct. While ACD is always possible when QRNP and RCNP are

identical, there seem to be more cases where ACD is possible. (38c), (38d), (39c),

5
In fact, for some people, it’s possible to leave lake in (37b) unstressed and then (37b) becomes
acceptable. The explanation given below for (37c) carries over to this case as well.

116
and (39d) show that (37c) isn’t the only exception while (38a) and (38b), as well as

(39a) and (39b) (repeated from (16)) display the contrast familiar from the previous

section.

(38) a. John lives in a city that’s close to a city Mary used to hlive ini.

b. ∗ John
J lives in a city that’s close to a castle Mary used to hlive ini.

c. ? John
J lives in a city that’s close to a town Mary used to hlive ini.

d. ? John
J lives in a city that’s close to where Mary used to hlivei.

(39) a. Jon ordered a drink that’s more expensive than the drink Sue did horderi

b. ∗ Jon
J ordered a drink that’s more expensive than the dish Sue did horderi

J ordered a cocktail that’s more expensive than the beer Sue did horderi
c. ?? Jon

J ordered a drink that’s more expensive than what Sue did horderi
d. ? Jon

It seems that the relationship between the RCNP and the QRNP that is required for

the ACD to be acceptable has a semantic character. While I found a great amount

of speaker variation with respect to examples like the above ones, every speaker I

consulted found that the semantic relationship of RCNP and QRNP was the deciding

factor in the judgements. This semantic character supports strongly that it’s the

semantic contribution of the trace content which determines the identity of traces.

The account I develop now makes the semantic character precise, and specifically

aims to accounts for the facts in (38) and (39). The generalization I end up with in

this section is that ACD is possible if and only if QRNP denotes a subset of RCNP as

in (38d) and (39d) or QRNP and RCNP are from the semantic field as in (38c) and

117
(39c). It remains to be seen whether this generalization is correct exactly as stated

here; I do believe, though, that the general point concerning the semantic character

is.6

This generalization I argue is implied by the conjunction of two things: the

assumption that the lexical content of the trace contributes to interpretation and the

right account of the identity requirement. As already argued, the identity requirement

imposed on RCNP and QRNP in ACD is part of the identity requirement of VP-

ellipsis in general. Therefore, I now present an account of the identity requirement

of VP-ellipsis. I, then, come back to ACD, and show that this account predicts the

generalization just mentioned.

It is a well known observation that VP-ellipsis requires some form of seman-

tic identity or sameness of meaning between the elided VP and its antecedent (see

Sag 1976:92-95 and references therein). For example, in (40a), the first conjunct is

ambiguous between a volitional reading and an idiomatic, non-volitional reading, but

the interpretation of the elided VP in the second conjunct has to correspond to that

of the first VP. Similarly, the first VP in (40b) can receive an interpretation like

put paint on the bike or one like made a picture of a bike, but the elided VP has to

correspond in meaning to the first.

6
In the present context, reconsider the example in (i), repeated from (25b). The well-formedness
of (i) could be due to the fact that the RCNP and the QRNP are from the same semantic field even
if the modifiers near Madrid and near Rome are represented in the trace positions. Therefore, it
seems at this point impossible to test using examples with the structure of Kennedy’s puzzle whether
modifiers are represented in the trace position.
(i) John visited a town near Madrid that had signs for the one near Rome Bill did hvisiti.

118
(40) a. John hit the wall and then Pete did hhit the walli

b. John painted a bike after Mary did hpaint a bikei

An important observation that plays a role in the following is that destressing of a

VP displays a semantic requirement very similar to that of VP-deletion (Tancredi

1992, Rooth 1992b, Wold 1995, Fox 1998a). For example, the examples in (41) show

the same disambiguation as those in (40). (I represent destressing in (41) and in the

following by italics and a reduced character size.)

(41) a. John hit the wall and then Pete hit the wall

b. John painted the bicycle after Mary paint the bicycle

The first to discuss in detail the claim that VP-ellipsis requires identity of

meaning is Sag (1976). Sag, however, rejects this claim, and opts instead for a

requirement that LF-representations must be identical. His only reason is the example

in (42), where a child is interpreted generically in the first VP, but existentially

in the elided VP. Sag’s conclusion isn’t forced by example (42), at least not on a

quantificational variability account of generic interpretations of indefinites (Wilkinson

1991). On such an account, the first conjunct contains a covert generic quantifier

usually that lends its quantificational force to the indefinite a child and could take

scope outside of VP. The indefinite a child can, on this account, receive an existential

interpretation in both the elided VP and its antecedent.

119
(42) They caned a child severely when I was a child, but not like Miss Grundy did

hcane a childi yesterday. (Sag 1976:(2.0.13))

Since there are examples like (40) where I know of no argument in favor of

an LF-difference between the two interpretations that VP-ellipsis draws a distinction

between, I assume that VP-ellipsis requires identity of meaning to an antecedent. This

is also assumed by Tancredi (1992), Rooth (1992b), Wold (1995), and Fox (1998a).

The nature of the identity condition affects the question what the nature of

the content of the trace is because the identity condition is sensitive to it. If there

is only an identity of meaning requirement on elided VPs, the lexical material in the

trace position must be interpreted there since it’s relevant to the identity condition.

However, exact semantic identity wouldn’t explain the examples like (37c). Hence,

the identity requirement must be a more complicated technical condition, not the

intuitive notion of identical I have made appeal to up to this point. Whether this

more complicated condition makes references to the form or to the meaning is what

needs to understood. Before coming back to (37c), I summarize the literature on

this question. In the literature, the main disagreement is whether there is also a

requirement of identity of form in addition to the identity of meaning requirement.

If there’s a requirement of identity of form, the requirement that traces must have

identical lexical content might be a purely formal requirement. In that case, there’s

no evidence that traces have semantic content other than being variables.

Rooth (1992b), in particular, argues that there is also a requirement of identity

120
of form. One of his argument has the following structure:7 As seen above, destress-

ing and deletion seem to share the semantic identity requirement. There are cases,

though, where destressing and also VP-ellipsis can be licensed by satisfying a weaker

requirement of indirect identity. Therefore, the semantic requirement allows indirect

identity. This, however, overgenerates possible interpretations for VP-ellipsis. There-

fore, there must be an additional requirement, identity of form, for elided material.

I now present Rooth’s (1992b) argument for identity of form in more detail. I then

summarize a different way to draw the distinction between destressing and deletion

argued for by Fox (1998a), which doesn’t require identity of form. Finally, I argue

that the facts from ACD above argue for Fox’s (1998a) statement of the condition

and also show that the trace has semantic content.

Strict semantic identity is too strong in cases of destressing like those in (43).

As Tancredi (1992) and Rooth (1992b) argue, such examples argue that destressing

can be licensed under identity of meaning with a sentence that is not part of the

discourse itself, but rather, entailed by the discourse. For example, the first conjunct

in (43b) entails that Mary is having a drink, and it’s semantic identity to the VP

of the second conjunct that can license destressing in (43a). I call this relationship,

where identity is satisfied by an entailment of the antecedent, indirect identity.

(43) a. John enjoyed one Russian novel, and even Bill read a book.

b. Mary ordered a beer and Sue is having a drink, too.

7
The other argument concerns the requirement that sloppy readings must have a same dependency
in the antecedent. See the dicussion of the parallel dependencies requirement at the end of section
4.1.2.

121
 While Tancredi (1992) claims that indirect identity is only found with de-

stressed VPs, Rooth (1992b) shows a case where an elided VP seems to be only

indirectly identical to its antecedent. The argument relies on an observation illus-

trated in (44a) and (44b). If her in the first conjunct refers to Mary, (44a) allows a so

called sloppy interpretation: the pronoun her in the elided VP need not refer to Mary,

but can also refer to Jane instead. However, as shown in (44b), the pronoun her in

the elided VP cannot be taken to refer to Sue. (44) argues for a constraint on sloppy

readings such as that given in (45) (see Ristad 1990, Fiengo and May 1994:96–117,

Rooth 1992b, Fox 1998c for further evidence for (45)). I assume (45) for the rest

of this chapter as an empirical generalization—I discuss briefly at the end of section

4.1.2 how Rooth (1992b) actually derives most cases of the requirement (44).

(44) a. First, John told Maryi I was bad-mouthing heri ,

and then Sue told Janej I was hbad-mouthing herj i

b. ∗ First,
F John told Maryi I was bad-mouthing heri ,

and then Suej told Jane I was hbad-mouthing herj i

(45) Parallel Dependencies: If a pronoun isn’t identical in reference to the cor-

responding pronoun in the antecedent, it must stand in the same structural

relationship to its binder as the corresponding pronoun in the antecedent.

Assuming (45), Rooth’s (1992b) argument is based on (46). In (46), as indicated

the pronoun her in the elided VP can refer to Sue, even when the the corresponding

pronoun in the antecedent refers to Mary. In (46), the condition (45) seems to be

122
violated, because the antecedent of the pronoun in the second conjunct is the subject,

whereas that of the corresponding pronoun in the first conjunct is the object.8 Rooth

proposes that the violation of (45) in (46) is only apparent; ellipsis in (46) isn’t

licensed by direct identity with the first conjunct, but by indirect identity where the

relevant entailment of the first conjunct is the sentence (47).

(46) First, John told Maryi I was bad-mouthing heri ,

and then Suej heard I was hbad-mouthing herj i (Rooth 1992b:(30))

(47) Maryi heard I was bad-mouthing heri .

Therefore, (46) argues that indirect identity can license ellipsis as well. At the end

of section 3.3.2, I summarize an additional argument from Jacobson (1998a) that

indirect identity can license ellipsis. However, there are many cases where indirect

identity is lenient: Rooth (1992b) and Tancredi (1992) show with examples likes those

in (48) that ellipsis must require more than indirect identity. In (48a), which contrasts

with destressing in (43a), the elided VP cannot receive the interpretation indicated

though it would satisfy indirect identity. Similarly, (48b) and (48c) clearly contrast

with (46), since the entailment argued to be involved in the licensing of deletion in

(46) cannot license deletion in (48a) and (48b).

8
Fiengo and May (1994:100) claim that the surface subject of the second conjunct of (45) is
in fact an object, and therefore (46) doesn’t violate the condition (45). But, as Danny Fox (p.c.)
points out, Fiengo and May’s (1994) account predicts that sloppy ellipsis should also be possible if
the order of the conjuncts is reversed, which isn’t the case as (i) shows. Rooth’s (1992b) analysis of
(46) makes the right prediction for (i).
∗
(i) First,
F Suej heard I was bad-mouthing herj , and then John told Maryi I was hbad-mouthing
heri i

123
(48) a. ∗ John
J enjoyed one Russian novel, and even Bill did hread a book.i

b. ∗ First
F someone told Mary about the budget cuts and then Sue did hhear

about the budget cutsi (Rooth 1992b:(15))

c. ∗ First
F John told Maryi I was bad-mouthing heri and then Suej did hhear I

was bad-mouthing herj i

The argument shows that VP-ellipsis has an additional requirement which distin-

guishes it from destressing. Rooth (1992b) proposes that this is an identity of form

requirement; specifically, he refers to the reconstruction relation of Fiengo and May

(1994). While this is a possible account of (48) it could also be the case that the

semantic identity requirement VP-ellipsis imposes is slightly stricter than that of de-

stressing, in a way similar to the proposal of Tancredi (1992). Fox (1998a:ch. 3)

develops such a proposal, which I adopt. The arguments of Fox (1998a) for his pro-

posal, and against that of Rooth (1992b), are too intricate to summarize here; instead

I point out some of the problems an identity of form proposal faces, before I sum-

marize Fox’s (1998a) proposal. The account of facts like (37c) I present then is an

additional argument for Fox’s (1998a) approach.

Consider the examples in (49) in the context of an identity of form require-

ment. In each of them, identity of form must be compromised because, if the elided

VP would have to be identical in the choice of lexical items to the antecedent VP,

all four sentences are predicted to be ungrammatical. Therefore, as Johnson (1996:7)

argues, examples like these are a significant challenge for any version of an identity

124
of form requirement, and the best developed proposal of this kind I’m aware of is

that of Fiengo and May (1994:220). According to Fiengo and May, the examples

require essentially of list expressions that satisfy identity of form despite being lexi-

cally different, for which Fiengo and May (1994) introduce the term vehicle change

as mentioned in section 2.4.

(49) a. John doesn’t see anyone, but Bill does. (Sag 1976:(2.3.39))

b. Jonathan didn’t have a red cent, but Susi did hhave moneyi.

c. John won’t leave until midnight, but Bill will hleave before midnighti

(Chomsky 1972a:(75))9

d. Because Sue didn’t want to buy Billi ’s dinner, hei had to hbuy hisi dinneri.

Based on other (stronger) arguments, Fox (1998a) proposes to replace the concept

of identity of form that Rooth (1992b) appealed to with a stricter condition on the

semantic relation between an elided VP and its overt antecedent. Specifically, he

proposes a restriction on indirect identity that amounts to the recursive condition in

(50):10

(50) The antecedent VPantecedent and the elided VPelided , which is part of a sentence

S, can satisfy indirect identity only if there’s no VPelided 0 such that

9
Grinder and Postal (1971) judge the sentence (49c) ungrammatical. However, Chomsky (1972a)
and my informants do find it acceptable with the appropriate contrastive foci on the subjects John
and Bill. See also Sag (1976:158-60) for discussion.
10
The difference is mainly that Fox states the condition for focus domains, rather than for elided
VPs specifically. This presupposes the semantics of focus which are only introduced in section 3.3.
Moreover, this aspect of Fox’s proposal would not be useful at this point.

125
 a. replacing VPelided with VPelided 0 in S yield a grammatical sentence S0

b. S0 is logically stronger than S

c. VPelided 0 is (directly or indirectly) identical with VPantecedent

The restriction (50) can also be seen as a constraint on the parsing (or recovery) of

elided material. Then it could be stated as follows: For an elided VP site, choose

a parse with an interpretation as strong as possible, but entailed by the antecedent,

and that’s compatible with the overt material surrounding the deletion site.

Fox’s condition accounts for Rooth’s problem that VP-ellipsis of indirectly

identical material was licensed in (46), but blocked in (48). Consider (48a), as re-

peated in (51), first. In this case, there is an interpretation of the elided VP site is

possible that results in a stronger statement and that is also indirectly (as well as

directly) identical to the first conjunct: Bill enjoyed one Russian novel. Therefore,

the interpretation indicated in (51) is blocked. In general, it will be the case that

the interpretation of the elided VP directly identical to the antecedent is the one

chosen, unless there is a requirement imposed by the material surrounding the elided

VP which blocks direct identity.

∗
(51) J
John enjoyed one Russian novel, and even Bill did hread a book.i

The difference between (51) and (46), repeated in (52), is that in (52) material outside

of the elided VP together with the constraint (45) forces indirect identity. The only

parse which satisfies (45) is the strict reading. But, the strict reading stands in no

126
entailment relation to the sloppy reading indicated in (52), and therefore doesn’t

block (52). Among the sloppy readings, condition (45) forces a parse of the elided VP

as a verb with an object pronoun. Of these, the one in (52) is the logically strongest,

and hence, is the one possible for the elided VP.

(52) First, John told Maryi I was bad-mouthing heri , and then Suej heard I was

hbad-mouthing herj i

In sum, Fox (1998a) claims that there’s no requirement of identity of form

between the elided VP and its antecedent, but only a slightly stronger semantic iden-

tity condition than the one for destressing. Going back to the question whether the

lexical material in the trace position is interpreted there, Fox’s (1998a) account is

only compatible with one answer. Namely if Fox is right, it predicts that traces must

have semantic content beyond being a variable, since it must be their semantic contri-

bution that’s blocking ACD in cases like (2). Additional support for this conclusion

and Fox’s (1998a) account comes from the facts in (37) to (39), as I show now.

Consider first the examples where RCNP denotes a subset of RCQR, like those

in (53) (repeated from (38d) and (39d)). The contrast in (54) shows that RCNP must

denote a subset of RCQR, and not the other way round.

(53) a. ? John
J lives in a city that’s close to where Mary used to hlivei.

J ordered a drink that’s more expensive than what Sue did horderi
b. ? Jon

(54) a. Last night, I talked to a bachelor who looked like the guy you did htalk toi

127
 b. ∗ Last
L night, I talked to a guy who looked like the bachelor you did htalk toi

I propose that in these cases indirect identity is involved. Consider the elided VP and

its antecedent in the LF-representation of (54a) in (55). The matching requirement

on relative clauses doesn’t allow a lexical content of the trace in the elided VP other

than guy (see (56) below). Therefore, if the antecedent in (55) has an entailment

where the lexical content of the trace is [x, guy], the elided VP satisfies the require-

ments for indirect identity that no logically stronger replacement of the elided VP is

grammatically possible.

h i
(55) a bachelor who looked like the guy λy you talked to [y, guy]
| {z }
elided VP
λx I talked to [x, bachelor]
| {z }
antecedent

Treating (53) and (54) as cases of indirect identity, therefore, requires that the con-

tribution of [x, bachelor] to the meaning of the antecedent is such that it allows an

entailment to the VP where it’s replaced by [x, guy]. For now, let us assume that

the meaning of the trace [x, bachelor] can be paraphrased as the indefinite a bach-

elor. Then, it’s indeed predicted that ellipsis is licensed in (55), since ‘I talked to a

bachelor’ entails ‘I talked to a guy’.

The possibility of indirect identity makes it necessary to briefly talk about the

analysis of relative clauses again. In section 2.4, I argued one class of relative clauses,

the matching relatives, involve both an internal and external head NP, and I suggested

that the internal head NP is obligatorily phonologically deleted and the antecedent

128
of it is the external head. If indirect identity could be satisfied in the relationship

of the internal and external head, this would have consequences on the account of

Kennedy’s puzzle. Consider Kennedy’s example in (55a), repeated from (2). Since,

being a country entails being something, an empty internal head is entailed by the

external head in (56a). But, as (56b) illustrates, (56a) is predicted to be grammatical

if a representation with an empty internal head was possible. However, entailment

isn’t the only requirement of indirect identity. If we assume that condition (50) carries

over to directly to the case of NPs, direct identity of the semantic content of internal

and external head is required in cases like (56a).

(56) a. ∗ Polly
P visited every town in a country Eric did hvisiti.

b. ∗ [every
[ town in a [country, Opy Eric visited [y, ]]]
| {z }
elided VP
λx [Polly visited [x, town]]
| {z }
antecedent

Now, consider the cases where NPRC and NPQR are from the same semantic

field, but no subset relation holds between them. In (57), (38c) and (39c) from above

are repeated. In contrast to the subset cases, in these cases NPRC and NPQR seem

interchangeable as (58) aims to show ((58a) repeated from (37c)).

(57) a. ? John
J lives in a city that’s close to a town Mary used to hlive ini.

J ordered a cocktail that’s more expensive than the beer Sue did horderi
b.?? Jon

(58) a. ? John
J visited a city that’s near a town Mary did hvisiti.

b. ? John
J visited a town that’s near a city Mary did hvisiti

129
Notice that for the examples to be acceptable, a particular intonation is required.

Even then, they still remain marginal for most speakers if compared to the cases

where NPQR and NPRC are identical. On the required intonation, NPQR is stressed,

while NPRC is unstressed. Though I haven’t been able to verify whether the pitch

on NPQR confirms this claim, I claim that NPQR bears a topic accent in the sense

of Büring (1996, 1998) since this helps in explaining the relative acceptability of (57)

and (58).

One of the functions of topic accents Büring discusses is that they signal the

presence of an alternative question to the assertion made. One of Büring’s (1995)

examples is (59a), where the topic accent on female is phonetically realized as a

falling pitch. With this intonation, (59a) indicates that the question What did the

male pop stars wear? is still open. (59b) answers this open question, and in (59b)

male cannot bear a topic accent. This is because, with the assertion of (59b), it is

known what all the pop stars wear. In (59b) male can though optionally bear focus.

(In (59) and the following, I indicate pitch accents with capital letters, semantic focus

with an F-subscript, and semantic topic with a T-subscript. With Büring (1995), I

assume that a pitch accent inside a semantic topic is phonetically realized as falling

pitch, whereas a pitch accent inside a semantic focus is realized by a rise in pitch.)

(59) a. The [FEmale]T pop stars wore [KAFtans]F .

b. The male pop stars wore [tuXEdos]F .

The examples in (60) and (61) indicate that the implicit question raised in-

130
dicated by a topic accent can license destressing and deletion, if it’s not ambiguous

which implicit question raised is the licensing one. Hence, (60) and (61) are acceptable

in a situation where only German and American beer or red and green gummibears

are under consideration.

(60) a. [John]F bought [German]T beer. [Mary]F bought the American one.

b. [John]F eats only [red]T gummibears. [Mary]F must’ve eaten the green one.

(61) a. [John]F bought [German]T beer. As for the American one, [Mary]F did.

b. [John]F eats only [red]T gummibears. As for the green ones, [Mary]F does.

What I claim is that the examples of (37) to (39) where RCNP and RCQR are from

the same semantic field satisfy the licensing condition on VP-ellipsis for the reason

the examples in (61) do. Assume that (37c) has the focus structure in (62). The

alternative needed to license VP-deletion is a question like Who visited a town?.

Alhough the matter is far from clear, the topic accent might make this antecedent

available. The requirement to be in the same semantic field, I believe, is a general

requirement on alternative questions raised by topic accents.

(62) [John]F visited a [city]T that’s near a town [Mary]F did hvisiti.

This concludes the discussion of the examples (37) to (39). I showed that a

class of such examples, where NPRC denotes a subset of NPQR, is actually predicted

by the assumption of Fox (1998a) that VP-deletion can be licensed indirectly if this

is forced by the overt material surrounding the deletion site. This is part of a theory

131
where all the licensing conditions for VP-ellipsis are only sensitive to meaning. As

for the second class of examples where NPRC and NPRC are from the same semantic

field, I offered a suggestion of how to incorporate these cases into the theory of

ellipsis. Both accounts relied on the assumption that the lexical material represented

in the trace position contributes to the interpretation of the trace. In particular, for

the examples above, the interpretation of an unbound trace [x, NP-part] could be

paraphrased as an indefinite ‘a NP-part’.

The remaining pages of this section contain a digression. The question it ad-

dresses is raised at the end of chapter 2 and concerns deletion of ACD-relatives in

a trace position. I state this as the requirement in (63) (repeated from (97d)). The

question raised above is whether this requirement involves look-ahead of the mech-

anism that deletes parts of a chain to the level where the licensing conditions for

ellipsis apply, or whether chain deletion applies in an ACD configuration indepen-

dently of whether this will license deletion or not. In chapter 2, it was impossible to

distinguish these possibilities empirically because it seemed that all examples with

an ACD configuration required deletion of the modifier in the trace position for the

licensing of VP-ellipsis. But, at this point, it becomes possible to draw a distinction,

and as I will show, the results lead me to conclude that no look-ahead is involved

with (63).

(63) ACD If material inside a modifier is anaphorically related to the constituent

surrounding an occurence of this modifier, this occurence of this modifier must

be deleted.

132
Consider the following prediction of the idea that VP-deletion can be licensed indi-

rectly via an entailment. In (64a), the head of the ACD-relative is introduced by an

upward entailing quantifier. In this case, VP-ellipsis in the relative clause is predicted

to be licensed even if the relative clause remains in its surface position. Consider the

LF-representation in (64b). For the antecedent and elided VP as indicated, indirect

identity is satisfied by the structure in (64b): Since a is an upward entailing quan-

tifier, the clause (64a) entails that John read a book. But, this provides a suitable

antecedent for the elided VP in (64b) if, as I argued above, an indefinite in the an-

tecedent is sufficient to license deletion of a VP where a trace with the same NP-part

corresponds to the indefinite.

(64) a. John read a book Mary did hreadi

elided VP
z }| {
b. John read [a book λx Mary did read [x, book]]
| {z }
antecedent

This prediction gives us way to test whether the condition in (63) is looking

ahead to see whether ACD can be licensed without deletion, or whether ACD applies

whenever the formal configuration of ACD arises. Namely, if (63) is looking ahead the

configuration in (64b) should arise in the licensing of ACD with indefinites. On the

other hand, if (63) doesn’t look ahead, it should apply in (64b), and therefore ACD

should require QR for its resolution even with upward entailing quantifiers. How can

we test this prediction? The first test that comes to mind is to use quantifier scope

as a diagnostic of whether QR applied. As we’ll see, the result seems to favor the no

look-ahead position, but is ultimately not decisive. A second diagnostic is Condition

133
C obviation effects discussed in section 2.1. Here again, the result argues for the no

look-ahead position and, in this case, I find the argument convincing.

First, the quantifier scope test. As Sag (1976:72-74) and Larson and May

(1990:112-15) observe, ACD forces the DP that the ACD-relative is part of to take

scope outside of the antecedent of the elided VP. This follows from—and to be more

precise, strongly supports—the assumption that covert movement to a position out-

side of the antecedent is required for the resolution of ACD. This is shown in (65),

where only the case where the elided VP is interpreted as indicated is relevant. The

fact observed by Sag is that then the quantifier every cannot take scope below want.

The explanation of this fact is that QR to a position outside of the antecedents is

required which forces an LF-representation like (65b). In (65b), however, every is

outside of the c-domain of want.

(65) a. Betsy’s father wants her to read everything her boss wants hher to readi

(everything À want, ∗ want À everything) (Sag 1976:(1.3.38))

h i
b. everything λy her boss wants her to read [y, thing] λx Betsy’s father

wants her to read

Sag’s example (65a), as well as the example of Larson and May (1990), demonstrate

the correlation between scope and ACD using the universal quantifier every. The ques-

tion at hand is whether upward entailing quantifier behave differently. The examples

in (66), based on suggestions by Irene Heim (p.c.) and Danny Fox (p.c.), show that

indefinites behave like universal quantifiers. Both involve the scope of the negative

134
polarity item anything which is upward entailing in its NPI-meaning. In this meaning

it must occur in the scope of an NPI-licensing expression, which in (66a) is negation

and in (66b) the verb refuse. In both examples, consider only an interpretation where

the NPI-licenser is part of the antecedent of the deletion in ACD, as indicated. If the

correlation between scope and ACD wouldn’t hold for upward entailing quantifiers,

the VP-deletions indicated could be licensed without moving the NPI to a position

outside the scope of its licenser. However, this prediction seems to be wrong—the

interpretations of the elided VPs indicated seem unavailable. This shows that with

upward entailing quantifier, Sag’s observation also holds: the ACD-containing DP

must take scope outside of the antecedent.

(66) a. ∗ John
J plans never to be anywhere Mary did hplan never to bei (∗ any À

not, ∗ not À any)

b.?? John
J is refusing to read anything Mary is hrefusing to readi (?? any À

refuse, ∗ refuse À any)

The result in (66) seems to favor the idea that the deletion in ACD involves no

look-ahead. In fact, though, the look-ahead view probably predicts the lack of a

narrow scope reading for the examples in (66) in the following way. Consider the

LF-representation for (66a) in (67). While it’s true that “John is in a place that

Mary is” entail that “John is in a place”, this entailment is not sufficient to license

indirect identity in (67). The entailment required for (67) would be from “John plans

never to be anywhere Mary plans never to be” to “John plans never to be anywhere”,

135
which obviously doesn’t hold. Therefore, (67) doesn’t satisfy indirect identity.

elided VP
z }| {
(67) John plans never to be anywhere λx Mary did plan never to be [x, place]
| {z }
antecedent

The failure of (67) to license indirect identity is not an accident, but inherent to

the logic of the scope argument: An example without this flaw would be one where

an upward entailing quantifier with an ACD-relative, entails the wide scope reading

while taking narrow scope. As Abusch (1994) argues, it is for pragmatic reasons

impossible to test whether a certain reading is present, if it’s entailed by another

reading whose existence is established. Hence, I believe the argument based on scope

argument doesn’t decide whether there’s look-ahead in ACD-resolution.

The second test for the look-ahead question is based the obviation of Condi-

tion C discussed in section 2.1. Recall that in an example like (68) (repeated from

(7a) on page 34) Condition C was obviated by ACD, because ACD requires dele-

tion of the ACD-relative clause in the position of the QR-trace, as shown by the

LF-representation in (68b).

(68) a. You introduced himi to everyone Johni wanted you to hintroduce himi toi
h i
b. everyone [λy Johni wanted you to introduce himi to [y]]
| {z }
elided VP
λx you introduced himi to [x].
| {z }
antecedent

Since indirect identity could license ACD without movement with upward entailing

quantifiers, the look-ahead view would predict that in these cases we shouldn’t find

Condition C obviation. The contrast in (69), which seems as strong as in the case

136
of the universal quantifier in (68a), falsifies this prediction. This argues that the

LF-representation in (70), though it would satisfy indirect identity, is ruled out by a

formal requirement that rules out any configuration where an elided VP occurs inside

of its antecedent. Therefore, I conclude that the no look-ahead position is correct.

(69) a. ? John
J introduced heri to a man Maryi wanted him to hintroduce her toi.

b. ∗ John
J introduced heri to a man Maryi wanted Bill to like.

elided VP
z }| {
(70) John introduced heri to a man λx Maryi wanted him to introduce her to [x, man]
| {z }
antecedent

3.3 Wh-Traces and Focus in Chains

This sections starts out with a number of apparent problems for the assumption that

traces have content that matters for the licensing of VP-ellipsis. The goal of the

section is to show that the right understanding of how focus works in chains, which is

actually mostly drawn from the literature, yields a natural solution to these problems,

and in fact provides new support for the main claim of this chapter.

The problems are examples like (71). Apparently, in constructions other than

ACD, the lexical content of a trace, if it’s there, doesn’t block VP-ellipsis (Evans

1988, Jacobson 1992)11

11
Sag (1976:63–67) and Williams (1977:130–31) claim based on examples like those in (i) that
wh-extraction from an elided VP is impossible. The examples in the text falsify this general claim
and the examples in (i) are probably ruled out for irrelevant reasons: In (ia), since the verb moves to
Comp, the elided VP is in the complement position of an empty head, which is generally impossible
(Lobeck 1992). (ib) and (ic), as Fiengo and May (1994:244) suggest, indicate a preference to delete
as much material as possible once material is deleted (See also 2 on Lappin 1984).
(i) a. What did Harry take a picture of?
∗
What
W did Bill? (Sag 1976:(1.3.18))

137
(71) a. I know which cities Mary visited, but I have no idea which lakes she did

hvisiti.

b. The cities Mary visited are near the lakes Bill did hvisiti.

Let me briefly sketch the solution before I spell it out in detail below. The first point

to note is that the examples in (71) require two different solutions. This is shown by

the surprising contrast in (72): (72a) shows that a VP containing a relative clause

internal trace can license deletion of a VP containing a trace of wh-movement with

different lexical content. (72b) shows that it’s impossible to license deletion in the

other direction—a VP containing the trace of wh-movement cannot be the antecedent

for an elided VP that contains a relative clause internal trace with different lexical

content.

(72) a. I know the cities Mary visited, but I would like to know which lakes she

did hvisiti.

b. ∗ I know which cities Mary visited, but I would like to know the lakes she

did hvisiti

In section 2.4, I discussed another difference between relative clauses and questions,

namely with respect to Condition C as illustrated in (73)(repeated from (49) on page

62). The account presented in section 2.4 essentially claimed that in a question, the

b. ∗ John
J who Bill saw and who Bob did, too. (Williams 1977:(93))
c. ?? We
W finally got in touch with John, who my brother Al tried to visit, but who he couldn’t
hvisiti (Sag 1976:(1.3.22))

138
fronted material is directly related to the trace position. The relationship between the

head of a relative clause and the relative clause internal trace position, on the other

hand, is less direct, and therefore allows the minor change in the lexical content of

the NP-part that’s represented in the trace position, needed to circumvent Condition

C.

(73) a. Which is the picture of Johni that hei likes?

b. ∗ Which
W picture of Johni does hei like?

My solution for (71a) makes use of this distinction between relative clauses and ques-

tions. I claim that material that’s directly related to the dislocated material, as in

a question, is essentially overt material, and therefore isn’t subject to the identity

requirement on elided material. Indirectly related material, on the other hand, is

subject to the requirement. This is the essence of my solution for (71a) presented in

more detail below.

The solution of (71a) just sketched doesn’t carry over to (71b), because here

the relative clause internal traces depend only indirectly on the external heads, which

are different. Because of the contrast in (72), this is a desirable result. The solution

for (71b) I propose draws an analogy between them and sloppy readings. Consider

the sloppy reading in example (74): While I have so far assumed that the semantic

contribution of a variable is ignored by the identity condition, this is usually assumed

to be incorrect—in fact, I argue that it’s incorrect in section 4.1. But, then the two

VPs are different in meaning, namely bribed John and bribed Bill. This problem looks

139
similar to the problem in (71b), since there as well the lexical content of the trace

depends on a phrase outside of the elided VP.

(74) Johni admitted that Mary had bribed himi .

Billj admitted that she had hbribed himj i, too. (Hardt 1992:(27))

I elaborate this analogy by extending Rooth’s (1992b) account for sloppy readings,

which I argue for in section 4.1, to the case of (71b). In particular, I show that

the explanation of the parallel dependencies requirement (45) also accounts for the

difference between the well-formed(71b) and the ill-formed (72b), as well as the ACD-

cases like (2): In (71b) the material the traces are related to is the external head of

the relative clause for both, the elided VP and the antecedent. In the ill-formed

examples, this isn’t the case, and therefore (71b) seems to be satisfactorily explained.

The remainder of this section is divided in three subsections, each of which

considers one possible way to circumvent the identity requirement of VP-ellipsis.

First, I consider pseudo-gapping but only to conclude that it doesn’t account for

cases like (71a) and (71b). Second, I consider Focus, and will argue that it provides

the solution for (71a), but not for (71b). Finally, I consider Domain Extension and

argue that it solves (71b).

3.3.1 Pseudogapping and Traces

Pseudogapping, illustrated in (75), as a potential way to circumvent the identity

requirement of VP-ellipsis, was brought to my attention by Fox and Nissenbaum

140
(1998). In pseudogapping, a VP is elided, except for the object which is pronounced

and can be different from the antecedent. In fact, the object must be different.

(75) a. While some visited cities, others did hvisiti lakes.

b. While some people advised Mary to visit cities, others did hadvise Mary to

visiti lakes.

It seems possible that most examples of apparent VP-ellipsis with a trace in ob-

ject position are really instances of pseudogapping, where the trace isn’t part of the

elided material (Cormack 1984, Jacobson 1992). Consider, for example, the LF-

representation for (71a) in (76). If pseudogapping only imposes an identity require-

ment on the elided material, and the trace in object position can be not part of

the elided material, (76) is predicted to satisfy the identity requirement. This is the

right prediction for (71a), but the account overgenerates massively: It predicts that

all examples of VP-deletion with an object trace should allow an analysis like (76),

unless the trace is too deeply embedded in the VP. This would incorrectly predict all

examples like Kennedy’s puzzle (2) to be good.

(76) I know which cities λx Mary |visited

{z } [x, cities], but I have no idea which
antecedent
lakes λy she did visit
| {z } [y, lakes].
elided VP-part

There is reason to doubt that pseudogapping involves deletion of parts of a VP. For

one, deletion on this account would be an unusual operation since it does target non-

constituents. For example, but the VP-parts hypothetically deleted in (75b) don’t

141
form a constituent. Based on this and additional arguments, all recent analyses of

pseudogapping (Jayaseelan 1990, Lasnik 1995, Johnson 1996) conclude that pseudo-

gapping is actually VP-ellipsis preceeded by movement of the object to a position

outside of VP. This means that, even in pseudo-gapping, the identity condition always

applies to a trace.

The remaining question is whether the lexical content of the antecedent is

represented in this trace or not. Or in other words, whether the object movement in

pseudogapping is A- or A-bar-movement. If it is A-movement, as shown in section

2.3 and at the end of section 3.1, the trace usually doesn’t contain material of the

antecedent, and pseudogapping would help to circumvent the identity requirement.

If it’s A-bar-movement, the content of the trace in pseudo-gapping would not be

different from content of the trace of wh-movement.

The recent analysis of pseudogapping by Jayaseelan, Lasnik and Johnson dis-

agree about the type of movement involved in pseudogapping. Jayaseelan (1990)

proposes heavy NP shift, Lasnik (1995) advocates object shift analogous to what is

found in Scandinavian languages, and Johnson (1996) opts for a movement analo-

gous to Dutch scrambling. At least Lasnik’s and Johnson’s point in the direction

of A-movement; object shift is always A-movement and scrambling in Dutch is A-

movement in many cases (Déprez 1990). However, the objective of all three papers

is to account for the locality restrictions of the movement and restrictions on the

type of object that can occur. Therefore the argument for A-movement is only in-

direct: the movement in pseudogapping shows restrictions reminiscent of restrictions

on A-movement some other languages exhibits.

142
 We can test for whether A- or A-bar-movement is involved in pseudo-gapping

by considering Condition C obviation. As seen in section 2.3, only A-movement obvi-

ates Condition C regardless of where in the moving phrase the R-expression occurs.

The examples in (77) show that pseudo-gapping must involve A-bar-movement by

the Condition C test. Neither in (77a) nor (77b) is it possible for her to corefer with

an R-expression that’s part of the NP-part of the moving DP.

(77) a. ∗ I gave heri a book and you did hgive heri i a picture of Maryi .

b. ∗ While
W some told heri to paint a portrait of John, others did htell heri to

painti a picture of Suei .

The contrast in (78) is parallel to the well-known argument/adjunct contrast of Frei-

din (1986) and Lebeaux (1988) discussed in section 2.2 above. If the R-expression

occurs in a modifier adjoined to the object-DP that moves in pseudogapping as in

(78b), it doesn’t cause a Condition C effect.

(78) a. ∗ While
W some believed himi everything, others did hbelieve himi i only the

story that Johni had met aliens.

b. While some believed himi everything, others did hbelieve himi i only the

story that Johni had evidence for.

Because of (78) and (79), I conclude that the object in pseudo-gapping undergoes

A-bar movement, and leaving a trace with the same lexical content as the trace of wh-

movement. Hence, pseudo-gapping can never obviate the identity requirement that’s

143
imposed on the lexical content of A-bar traces and isn’t involved in the explanation

of (71a).

3.3.2 Focus and Wh-Traces

Focus is another possibility to escape the semantic identity requirement. This is

illustrated in (79) using the identity requirement of destressing. Recall that (79a)

(repeated from (41b)) is ambiguous with respect to whether John produces a picture

of the bicycle or puts paint on the bicycle, but the interpretation of the destressed

VP in the second conjunct must correspond to that of the first conjunct. This was

used to show that the destressed VP has to satisfy the identity of meaning require-

ment, as well. Now consider (79b): it exhibits a similar requirement of sameness of

interpretation, but the object car cannot be subject to this requirement. Since car

must be focussed in (79b), this argues that focus is required for material in the scope

of an identity requirement that isn’t identical to the antecedent. In a sense, focus

must make material invisible to the identity of meaning requirement.

(79) a. John painted the bicycle and Mary painted the bicycle

b. John painted the bicycle and Mary painted the [CAR]F

Examples like (79b) show that the identity requirement is only imposed on the mean-

ing of the non-focussed parts of a destressed VP. Following Jackendoff (1972), Rooth

(1985), and Kratzer (1991), I refer to the meaning of the non-focussed parts of a

phrase as the presuppositional skeleton. It isn’t always intuitively obvious what the

144
meaning of the non-focussed parts is; for example, if the non-focussed parts aren’t

a constituent. To make the notion presuppositional skeleton more precise, I adopt

(with one minor difference) a formalization argued for by Kratzer (1991), which was

inspired by Jackendoff (1972) and Rooth (1985:12). Kratzer defines a function that

assigns to a phrase-marker its presuppositional skeleton, which I use the notation

[[[—]]] for. Informally, the value of [[[XP]]] is the meaning of an XP0 that is derived

from XP by replacing all focussed subconstituent of XP with designated variables of

the corresponding semantic type. Examples like (80a) with multiple foci, argue that

different focussed constituents must correspond to different variables in the presup-

positional skeleton. Otherwise, interpretation (80c), would be incorrectly predicted

to be possible for (80a).

(80) a. At the party, John only introduce [MAry]F 1 to [GRANDma]F 2 ,

b. ‘For any x1 different from ‘Mary’, and any x2 different from ‘Grandma’,

John didn’t introduce x1 to x2 .

c. ∗ ‘For
‘ any x1 different from ‘Mary’, and any x1 different from ‘Grandma’,

John didn’t introduce x1 to x1 .

I indicate the variables a focussed constituent translates as by a superscripting it to

the focussed constituent in the syntactic representation. This is exemplified by (80a).

Assuming a similar notation, Kratzer defines [[[—]]] by recursion over the syntactic

structure, where G is the assignment function for the focus variables:

145
(81) a. [[[[X]F n ]]]G = G(n)

b. [[[X0 ]]] = {[[X0 ]]}, if X0 is a terminal node

c. otherwise [[[[X Y]]]] = C([[[X]]]G ,[[[Y]]]G ), where C represents the function that

assigns to [[X0 ]] and [[Y0 ]] the semantic value [[X0 Y0 ]] for any X0 of the same

semantic type as X and any Y0 of the same semantic type of Y

Assuming Kratzer’s definition of presuppositional skeleton, the identity of

meaning requirement can be restated as a relationship of the meaning of the an-

tecedent to the presuppositional skeleton to take the role of focus into account.

Namely, the antecedent must be identical to the value of the presuppositional skeleton

under G for at least one choice of assignment function G.

(82) There is an assignment G such that [[antecedent]] = [[[VP]]]G

For (80), (82) requires that the meaning of the antecedent VP, painted the bicycle,

must be an element of the presuppositional skeleton given in (81). For this to be the

case, the interpretation of paint in the antecedent and the partially destressed VP in

(80) must be identical.

The example in (83a) from Fox (1995a) is another case where focus marks

material that escapes the identity requirement. This case involves VP-ellipsis. (83a)

allows an interpretation where the subject in both conjuncts takes scope below seem.

As we see in (83b), on this interpretation, the elided VP doesn’t seem to be identical

to its antecedent, unless we exempt focussed material from the identity requirement.

146
Notice that the example in (83) provides an additional argument against the identity

of form requirement discussed in the previous section: If there was an identity of form

requirement on elided VP, it would have to apply at the level of LF to allow ACD.

But, to allow (83a), it would need to be sensitive to focus. Therefore, an identity

of form requirement would have to redundantly replicate the identity of meaning

requirement ellipsis and destressing have in common.

(83) a. An American athlete seemed to Bill to have won a Gold Medal, and a

[RUSsian RUNner]F did to. (Fox 1995a)

b. |seemed to Bill to an American{zathlete have won a Gold Medal} and

antecedent
seemed to Bill to a [Russian runner]F have won a Gold medal
| {z }
elided VP

Focus is relevant for the cases we’re interested in, if traces or their lexical

content can be focussed. Clearly traces cannot bear the pitch accent that usually

indicates focus phonetically. But, it has been argued by Selkirk (1995) and references

therein that traces can inherit the F-marking of their antecedents: F-marking of a

constituent licenses the F-marking of its trace (Selkirk 1995:559). One of Selkirk’s

(1995) arguments based on the work of Bresnan (1971, 1972) starts with the obser-

vation that while usually, as we saw in (79b) above, material that is not identical

in meaning to preceding material must be focus marked and receive a pitch accent,

this doesn’t hold for the verb in case the object is also focussed. Reviewed in (84a) is

new information, but doesn’t need to bear a pitch accent. For comparison, when the

object in (84b) isn’t focussed, pitch accent on the verb is required.

147
(84) a. Bill read the article and Helen [reviewed]F [the BOOK]F .

b. Bill read the article and Helen [reVIEWed]F the article.

From (84), Selkirk (1995) concludes that an F-marked verb doesn’t need to receive

pitch accent if its complement is F-marked. Based on this generalization, she argues

that in (85b) the object trace must be F-marked. Consider (85b) in the context of

(85a). Again, both the object and the verb must be F-marked. However, a pitch

accent on the fronted object suffices to phonetically realize this F-marking. This is

explained by the same phonological principle that applied in (84a), if the trace of

the fronted phrase is F-marked. Otherwise though, (85b) not only requires a new

phonological principle for pitch placement, but one that refers to a syntactic notion

such as the antecedent of a trace. Therefore, (85b) argues that the trace in (85b) can

obtain F-marking from its antecedent.

(85) a. Bill read an article, but . . .

b. [Which [BOOK]F ]i did Helen [review]F [ti ]F ? (Selkirk 1995:(24))

Consider now example (71a) again, which is repeated in (86a). If Selkirk (1995)

is right, the trace of a wh-moved phrase is F-marked in case the moved phrase is F-

marked. On a copy theory of movement this can be restated as follows: Copies of F-

marked phrases are F-marked. Assuming this for (86a), yields the LF-representation

in (86b). To satisfy the focus-sensitive identity requirement in (82), there must be a

replacement of the focussed material inside the elided VP, such that the antecedent VP

148
and the elided VP mean the same. In (86b), the focus-sensitive identity requirement

is obviously satisfied because it’s possible to replace the lakes, the lexical content of

the trace in the elided VP, with cities to achieve identity.12

(86) a. I know which cities Mary visited, but I have no idea which [lakes]F she did

hvisiti.

b. I know which cities λx Mary visited [x, cities],

| {z }
antecedent
but I have no idea which [lakes]F λy she did visit [y, [lakes]F ]
| {z }
elided VP

This concludes the account of (71a). As shown the standard claim that focus in

the head of a chain is also represented in the trace position together with the equally

standard claim that focussed material isn’t seen by the identity requirement on elided

material, yields a straightforward explanation of this case. The next question is under

which circumstances a focus that is phonetically expressed on a one constituent is also

represented as F-marking in a dependent position. For the discussion, I refer to this

12
The account makes a prediction for examples like (ia) (repeated from (20a)). Since the elided
VP contains a trace of wh-movement in (ia), focus percolation should from the wh-phrase should be
possible, and the focus structure in (ib) should be result. But, then (ia) should be acceptable with
focus on town-phrase, since lake is a focus-alternative to town that would satisfy direct parallelism.
This prediction seems factually incorrect.
(i) a. ∗ Do
D you know which town near a lake Mary visited John did hvisiti?
b. which [town]F near a lake λy Mary visited [y, lake]] λx John did visit [x, [town]F ]
The structure of (iib) resembles that of the A-movement examples like (33) discussed at the end
of section 3.1 and in section 4.1, where the ellipsis site also contained a trace of overt movement.
However, in contrast to the A-movement cases, the addition of focus particles doesn’t seem to lead
to an improvement of (ia) as (ii) attests. While I still hope that a better understanding of what
are possible focus structures will provide an explanation for (ia), at this point, I have to leave the
matter open.
∗
(ii) Do you know which town near a lake that Mary visited John did instead.
D

149
situation as percolation of the F-marking from one position to another. I want to argue

that F-marking can only percolate if the two positions are related via movement.13

One piece of support for the claim that F-marking can only percolate to a

dependent within a chain comes from the sluicing paradigm in (87) and (88).14 I

assume an analysis of sluicing as IP-ellipsis (Ross 1969b, Chung et al. 1995). It’s

quite well known that the fronted wh-phrase in sluicing can be related to the trace

position in the elided IP either via movement, or via a different process that doesn’t

create a syntactic chain. For example, Chung et al. (1995:279) distinguish between

sprouting (involving a chain) and sluicing (not involving a chain). I cannot discuss

the process invoked by sluicing (in the narrow sense of Chung et al. 1995) in detail

(see also Reinhart 1994). Two properties of sluicing matter: that it isn’t sensitive to

syntactic islands and that it doesn’t involve formation of a syntactic chain. Therefore,

in the examples in (87), both sprouting and sluicing are possible, but in (88) only

sluicing is possible because an island intervenes between the antecedent and the trace

in the elided material, as can be seen from the paraphrases in both (88a) and (88b).

Then the facts in (87) and (88) yield the following conclusion: Sprouting, as in

(87a) is possible even if the NP-part of the wh-phrase is different from the NP-part

of the corresponding indefinite in the antecedent. Sluicing, however, as shown, in

(88), requires that the NP-part of the wh-phrase be identical to the NP-part of the

corresponding indefinite in the antecedent.

13
As is expected from the absence of lexical content in the trace position, Focus in A-chains
cannot percolate to the trace position. Hence, the only way for a F-marked DP in an A-chain
can contribute an F-mark in the trace position is by scope reconstruction (see section 6.2). Diesing
(1992) and Selkirk (1995) argue based on data from Berman and Szamosi (1972) that this prediction
is correct.
14
Examples like (87) came to my attention during a discussion with Chris Kennedy.

150
(87) a. An astronomer needs to find a lot of new supernovae for her Ph.D., but

I don’t know how many galaxies han astronomer needs to a find for her

Ph.D.i

b. An astronomer needs to find a lot of new supernovae for her Ph.D., but I

don’t know exactly how many hnew supernovae an astronomer need to a

find for her Ph.D.i

(88) a. ∗ An
A astronomer needs to find a quadrant that contains a lot of new super-

novae for her Ph.D., but I don’t know how many galaxies han astronomer

needs to find a quadrant that contains for her Ph.D.i

b. An astronomer needs to find a quadrant that contains a lot of new super-

novae for her Ph.D., but I don’t know exactly how many hnew supernovae

an astronomer needs to find a quadrant that contains for her Ph.D.i

For the account of sluicing, I adopt the assumption that an indefinite in the antecedent

can correspond to a trace with the same NP-part in the elided IP, which is in the

spirit of Reinhart (1994). Then the facts in (87) and (88) follow directly from the

assumption that focus can only percolate from the antecedent to its dependent if the

two are linked by a syntactic chain. Notice that in (87a), the NP-part of the wh-phrase

how many galaxies must be focussed. Since in (87a) the wh-phrase can be linked to

the trace position by a syntactic chain, the focus of the wh-phrase can percolate to

the trace position, as shown in (89a). Therefore, the elided IP in (89a) is identical

modulo its focussed parts to the antecedent. In (88a), on the other hand, no syntactic

151
chain can be formed between the antecedent and the trace position. I claim that, as

shown in (89b), F-marking cannot percolate to the NP-part of the trace because it’s

not linked to its focussed antecedent by a chain. But, if the NP-part of the trace

isn’t F-marked in (89b), it doesn’t satisfy the identity requirement. Therefore, (88a)

is ill-formed on the assumption that F-marking can only percolate in a chain.

(89) a. how many [galaxies]F λx an astronomer needs to find [x, [galaxies]F ]

b. how many [galaxies]F λx an astronomer needs to find a quadrant that

contains [x, galaxies]

In relative clauses, the question whether F-marking can percolate from the

external head to the relative clause internal trace is much harder to investigate. Re-

call from section 2.4 that there are two possible structures for a relative clause, the

matching and the raising structure, and that these are quite hard to distinguish based

on their interpretation. The main discussion of focus percolation in relative clauses

I’m aware of is found in (Bresnan 1971) and the replies to Bresnan’s paper (Lakoff

1972, Berman and Szamosi 1972, and Bresnan 1972). Especially, the discussion of

a correlation between a difference in interpretation and stress placement in Bresnan

(1972:337-40) is quite interesting for our current purposes. The discussion is based on

the notion of normal stress which unfortunately isn’t made very precise in the paper

itself. For the following, I assume that normal stress can be characterized as bear-

ing F-marking on the rightmost ‘most embedded’ constituent that can be F-marked

(Höhle 1979 1982, Cinque 1993, Schwarzschild 1998, Zubizaretta 1998). This assump-

152
tion predicts a difference in normal stress between the matching and raising analysis,

in examples like (90), where an object relative clause is attached to the object of the

main verb. On the raising analysis, normal stress should require pitch accent on the

relative clause head so that the trace in the object position of the relative clause is

F-marked. On the matching analysis, however, it should be impossible to F-mark the

trace in the relative clause, and therefore the verb in the relative clause should be

F-marked.

(90) a. I gave John the [BOOKs]F he wanted. (Bresnan 1972:(43))

‘I gave John the number of books that he wanted.’

‘Of books, I gave John the ones he wanted.’

b. I gave John the books he [WANted]F (Bresnan 1972:(44))

‘Of the books, I gave John the ones he wanted.’

∗
‘ gave the John the number of books that he wanted.’
‘I

In her discussion of (90), Bresnan rejects the claim of Lakoff (1972) that the normal

stress in an example with an object attached relative like in (90) can be freely assigned

to either the head of the relative clause as in (90a) or the verb inside the relative

clause as in (90b). Instead, she argues that the apparent optionality correlates with

a difference in interpretation. The two interpretations Bresnan characterizes are one

where where the relative clause applies to the entire head in (90a) and a concealed

partitive interpretation for (90b). The prediction seems therefore at least partially

borne out as (90b) doesn’t allow an amount interpretation. I find it impossible to

153
assess whether (90a) has only an amount or kind interpretation, or whether it also

has a restrictive interpretation.

Accent placement in (90) might also be affected by the implicature that if

John was given books, he probably wanted them. For the example (91), most of my

informants agree that (91b) prefers an interpretation where those is used to refer to

the same tokens of chips as those that used to be ours. (91a), on the other hand,

could be used when the chips are different tokens, but the amount of chips is the

amount of chips that we lost. Again, (91) confirms a part of the prediction, while it

leaves it open whether pitch accent on the head noun can be the ‘normal’ stress for

a matching analysis of the relative clause.

(91) a. Those are the [CHIPs]F we lost

b. Those are the chips we [LOST]F

One argument for the prediction concerning examples with pitch accent on the head

noun comes from the example in (92), where a raising analysis is ruled out by Con-

dition C. As predicted, the pitch placement in (92b) seems to be preferred in (92).

(92) a. ?? Those
T are the [AUNTs of Maryi ]F shei likes

b. Those are the aunts of Maryi shei [LIKes]F

With non-finite relatives in (93) the intuitions are sharper. In a neutral con-

text, Hackl and Nissenbaum (1998) argue that pitch accent on the head noun, as in

(93a), forces an interpretation where the relative clause has possibility modal force.

154
Pitch accent on the verb, on the other hand, forces an interpretation paraphrasable

only with a necessity modal. Hackl and Nissenbaum (1998) present arguments based

on Condition C that (93a) has a raising analysis, while (93b) has a matching analysis.

Therefore the prediction mentioned above is fully confirmed by (93).

(93) a. Sabine came up with many [PROblems]F for us to work on

‘Sabine came up with many problems we could work on.’

∗
‘Sabine
‘ came up with many problems we should work on.’

b. Sabine came up with many problems for us to [WORK]F on

∗
‘
‘Sabine came up with many problems we could work on.’

‘Sabine came up with many problems we should work on.’

Based on these arguments, I conclude that F-marking can only percolate to a

dependent within a chain. This predicts that focus on a fronted phrase can obviate

the effect of the identity condition on a trace position, only if the trace is related to

the antecedent directly via movement as in the case of wh-movement, but not in the

case of matching relatives. Though I regard the arguments based on normal stress

above as tentative, this conclusion must be correct: Otherwise, all the examples of

Kennedy (1994) like (2) would be predicted to be acceptable with the right pitch

placement, which isn’t the case as shown by (94a) and (95a).15

15
One person in the audience at the SALT 8 conference at MIT reported a general improvement
of Kennedy’s examples if the head of the relative clause is stressed. However, none of my informants
share this intuition, and the person in question was not a native speaker of English. In fact, as
discussed in 1 and also observed with the paradigm in (39), destressing the head of the relative
clause is required even in the good examples.

155
 In contrast to matching relative clauses, raising relatives are predicted to pat-

tern with wh-questions, because here the relationship of the external head to the

relative clause internal trace position was argued to be created by movement (section

2.4). The contrasts in (94) and (95) seem to confirm this prediction for the kind

reading of raising relatives. In (94a) (repeated from (10)) and (95a) (repeated from

(16)), placing pitch accent on the head of the ACD-relative lakes doesn’t improve the

example. But in (94b) and (95b), where a kind reading is possible, accenting the

head of the ACD-relative improves the example.

(94) a. ∗ John
J visited towns that are near the LAKes Mary did hvisiti.

b. John visits towns that are much nicer than the LAKes Mary does hvisiti.

(95) a. ∗ Jon
J ordered a drink that’s more expensive than the DISH Martin did

horderi.

b. John orders drinks that are more expensive than the DISHes Martin does

horderi

To see that the contrast in (94) is predicted, consider the LF-representations in (96).

In (96a), the lexical content of the trace in the elided VP is not F-marked, and

therefore blocks identity between the elided VP and the potential antecedent. In

(96b), on the other hand, the lexical content of the trace in the elided VP is F-marked,

and therefore irrelevant for the identity condition (82). Therefore, the antecedent and

the elided VP are considered identical in (96b).

156
 h i
∗
(96) a. towns that are near the [lakes]F λy Mary did visit [y, lakes] λx John
| {z }
elided VP
visited [x, towns]
| {z }
6= antecedent
h i
b. towns that are much nicer than the λy Mary did visit [y, [lakes]F ] λx
| {z }
elided VP
John visited [x]
| {z }
antecedent

In matching relative clauses, focus percolation is predicted to be possible for

overt material that is part of the internal head. This is pied-piped material surround-

ing the wh-word in a matching relative clause. Jacobson (1998a) points out examples

like those in (97) which confirm this prediction.

(97) a. Mary visited every country the [EMbassy]F of which [BILL]F did hvisiti.

b. John greeted every boy whose [MOther] [Sue]F did hlikei

c. Sue voted for every candidate the [FAther] of whom [BILL]F had hvoted

fori (Jacobson 1998a:(15))

Consider the LF-representation of (97a) in (98). The pied-piped material is repre-

sented in the relative clause internal trace position, but is focussed. Therefore, the

identity requirement is satisfied in (98).

h i
(98) every country λy Bill visited the [embassy]F of [y, country]
| {z }
elided VP
λx Mary visited [x, country]
| {z }
antecedent

157
Jacobson (1998a) makes a very interesting observation about (97) that is not relevant

to the current discussion, but strongly supports Rooth’s (1992b) view of the parallel

dependencies requirement and indirect identity. Notice that the parallel dependen-

cies requirement seems to be violated in (98) because the relationship of y in the

elided VP to its antecedent is different from that of x to its antecedent. Jacobson

(1998a) proposes therefore that (98) requires indirect identity, and the entailment of

the sentence that contains the antecedent in (98) that satisfies the parallel dependen-

cies requirement is one like that paraphrased in (99), where territory of is inserted to

make the dependency parallel to the y-dependency in the elided VP.

(99) λx Mary visited the territory of [x, country]

Jacobson points out that the contrasts in (100) and (101) corroborate the view

that (97) requires indirect identity. The examples (100a) and (101a) are very similar

to those in (97), except that the DP that moves covertly for ACD-resolution in (97) is

moved overtly by topicalization in (100) and (101). Because the elided VP precedes

its antecedent linearly, (100a) and (101a) are slightly harder to parse than (97), but

both acceptable. This is expected because the account just given for (97) carries over

to (100a) and (101a). The examples (100b) and (101b), however, are ill-formed.

(100) a. ? Every
E country the embassy of which Bill did hvisiti, Mary visited.

b. ∗ Every
E country the embassy of which Bill visited, Mary did hvisiti.

(101) a. ? Every
E candidate the father of whom Bill had hvoted fori, Sue voted for.

158
 b. ∗ Every
E candidate the father of whom Bill had voted for, Sue had hvoted fori

(Jacobson 1998a:(18))

Jacobson’s account correctly predicts (100b) and (101b) to be ill-formed. Consider

the LF-representation of (100b) in (102). The antecedent isn’t directly identical to

the elided VP in (102). The entailment that would be required to license indirect

identity would be one from “Bill visited the embassy of the country” to “Bill visited

the country”. Since this entailment isn’t valid, no antecedent for the elided VP is

available in (102).

h i
∗
(102) every country λy Bill visited the [embassy]F of [y, country]
| {z }
6= antecedent

λx Mary visited [x, country]

| {z }
elided VP

In this way, the asymmetry between deletion of the VP with the complex trace and

deletion of the VP with the simpler trace seen in (101) and (102) follows directly

from the fact that an entailment from a simple DP like country to a more complex

DP territory of the country seems always possible, whereas an entailment in the other

direction is usually impossible. Notice that (103), where the entailment from the

complex DP to the simpler one is licit, is better than (100b), as predicted. Therefore,

Jacobson (1998a) concludes that her paradigm involving pied-piping in the relative

clause lends strong support to the claim the VP-ellipsis can be licensed by indirect

identity.

159
(103) ? Every
E country the capital of which Bill visited, Mary did hvisiti, too.

In sum, Focus percolation in a chain explains the puzzle (71a) from the begin-

ning of this section, while still predicting the ACD examples like (2) to be bad. Since

focus percolation doesn’t apply to matching relative clauses, it also doesn’t predict

(71b) (repeated in (104a)) to be good, unless it could be argued to allow a raising

analysis. As (104b) shows, examples like (104a) are acceptable even when a raising

analysis of the relative clause is ruled out by Condition C. Hence, the explanation of

(104) cannot be focus percolation into the trace position.

(104) a. The cities Mary visited are near the lakes Bill did hvisiti.

b. The aunt of Maryi shei visited live near the uncle of Billi hei did hvisiti.

3.3.3 Domain Expansion and Focus Index Sloppiness

Example (71b) (just repeated in (104a)) still needs an account. What I suggest, is that

(71b) is similar to examples that have a sloppy reading. Consider (105a), repeated

from (74) above, on a sloppy reading of the second VP. The LF-representation of

(104a) is shown in (105), where the elided VP and a potential antecedent are indicated.

On the sloppy reading the name and value of the variable x in the antecedent and

that of y in the elided VP differ. So far I have been assuming that this difference

is irrelevant for the identity condition of VP-ellipsis. It is, however, widely assumed

that this assumption is incorrect and I present an argument against this assumption

in chapter 4. That means, though, that the elided VP and the antecedent in (105b)

160
are predicted to be not identical, and it looks surprising that ellipsis is possible in

(105a).

(105) a. Johni admitted that Mary had bribed himi .

Billj admitted that she had hbribed himj i, too.

b. John λx x admitted that Mary had |bribed

{z x}
6= antecedent

Bill λy y admitted that Mary had bribed y

| {z }
elided VP

Rooth (1992b), incorporating an idea of Ristad (1990) and Fiengo and May (1994) into

the focus semantics framework of Rooth (1992a), presents a solution for this problem.

He proposes that the identity requirement (82) need not be verified for the elided VPs,

but can instead be verified for any bigger constituent that contains the elided VP.

If the bigger domain includes the binder of the variable, the names and values of

the variables effectively aren’t visible to the identity condition since, once bound,

variables names have no semantic effect. For example, in the LF-representation in

(106b) the domain of identity indicated and the antecedent are semantically identical:

Both denote a function from individuals into truth values which yields true if and

only if the individual admitted that Mary had bribed it. Therefore, (106) satisfies

the identity condition in the expanded domain of identity.

161
(106) John λx x admitted that Mary had bribed x
| {z }
antecedent
Bill λy y admitted that Mary had bribed y
| {z }
| {z
elided VP}
domain of identity (∼P)

In (106), the expanded domains were exactly identical because the material between

the binder and the elided VP was identical. There are examples of sloppy readings

like (107) where this isn’t the case. But, the solution to this problem is the focus

sensitivity of the identity condition already argued for in (82) above: The material

that is different in (107) between the antecedent and ∼P must be focussed, as Hardt

(1992) notices, and therefore doesn’t block identity on Rooth’s (1992b) account.

(107) a. Johni admitted that Mary had bribed himi .

Billj didn’t admit that MAry had hbribed himj i

But hej admitted that SOMEbody had hbribed himj i (Hardt 1992:(31))

b. he λx x admitted that [somebody]F had bribed himx

| {z }
| {z
elided VP }
domain of identity (∼P)

I claim that (71b), repeated in (108a), is analogous to a sloppy reading. Recall

that on the matching analysis of relative clauses, which I argued for in section 2.4,

there are two heads, an internal head and an external head. Furthermore, the lexical

content of the internal head depends on that of the external head. One could say that

the external head binds the internal head, forgetting for the moment the argument in

2.4 that the internal head has an internal syntactic structure. Under this assumption,

the LF-representation in (108b) is possible, where N and M are variables over NP-

162
part meanings that express the dependency of the internal head on the external head.

In (108b), the antecedent is identical to ∼P, if Bill is focussed.

(108) a. The cities Mary visited are near the lakes Bill did hvisiti.

b. the cities λN λx Mary visited [x, N] are near

| {z }
antecedent
the lakes λM λy [Bill]F visited [y, M]
| {z }
| {z
elided VP }
domain of identity (∼P)

That’s essentially the solution of (71b): to capture the dependency of the non-focus

correspondent of a focussed phrase by a semantic mechanism. The one last difficulty

to consider is of a more technical nature: Based on Safir’s observation, I argued in

section 2.4 that the internal head of a matching relative is not just a variable, as in

(108b), but a copy of the lexical material of the external head. Whereas a variable can

be bound in the familiar way, for a copy of lexical material that’s unexpected. The

dependency of the internal head on the external head is not variable binding, but a

deletion-antecedent relationship since the internal head is obligatorily phonologically

deleted and therefore has to satisfy an identity requirement similar to that of VP-

ellipsis. In the following paragraphs, I first show that the difference between the

variable-binding relationship of the NP-parts assumed in (109b) and the deletion-

antecedent relationship of the NP-parts argued for by Safir’s fact causes a technical

problem for the account of (71b). I then show that the same property of a deletion-

antecedent relationship that’s needed to solve this technical problem is also needed in

examples of VP-deletion, one of them being Kratzer’s (1991) ‘Tanglewood-example’.

163
I then basically adopt the account of Kratzer (1991), but argue for one minor change.

After that, I discuss some new predictions of the solution of (71b).

To begin with, it’s necessary to distinguish the antecedents in examples with

more than phrase that imposes an identity requirement. Following Rooth (1992a),

I use numerical indices to mark the antecedents and the corresponding ∼Ps, the

phrases that need to satisfy the identity requirement. Using this notation, the focus

structure of (108a) is given in (109).

(109) the [cities]F λx Bill visited [x, cities

| {z }] are near
| {z }
antecedent2 ∼P2
| {z }
antecedent1

the [lakes]F λy Bill visited [y, lakes

| {z }]
| {z }
antecedent3 ∼P3
| {z }
∼P1

The question is why lakes in the trace position in the second relative clause is identical

to cities in the relative clause trace in the first antecedent. The properties of (109)

that are going to play a role in the a solution, are: one, that lakes stands in an

deletion-antecedent relationship to another copy of lakes; two, that the other copy of

lakes is focussed; and three, that the other copy is also part of the domain of identity

under consideration.

Notice that examples that raise the same issue as (109) can be given with

VP-destressing and deletion. Consider the VP in the while-clauses in (110). Except

for the focussed verb, the VP can be destressed under identity to the preceeding

material. But, part of the destressed material as well as the antecedent is a elided

164
VP that can be interpreted as indicated. The elided material is neither focussed nor

identical to the elided VP in the first clause. Hence, it seems to violate the identity

condition just like the lexical content of the the trace in (108a) does.

(110) a. John bought a book from the guy Bill did hbuy a book fromi, while Sue

[STOle]F a book from the guy Bill did hsteal a book fromi

b. Every girli writes like heri teacher does hwritei, while every boyj [TALks]F

like hisj teacher does htalki

The LF-representation of (110a) in (111) brings out the parallelism to (109) (To save

space, the lexical content of the traces is not represented in (111)). The bold-faced

instance of steal in ∼P3 isn’t identical to the corresponding verb buy in the antecedent,

but isn’t focussed. Therefore, it should block identity, on our assumptions so far.

h i
(111) John λx the guy λy Bill did buy a book from y λz x bought a book from z
| {z } | {z }
∼P2 antecedent2
| {z }
antecedent1
while
h i
Sue λx the guy λy Bill did steal a book from y λz x [steal]F a book from z
| {z } | {z }
∼P3 antecedent3
| {z }
∼P1

In both (109) and (111), the non-identical material of ∼P stood itself in an identity

relationship to another instance of the same material within ∼P. It seems that in

computing the identity requirement of ∼P such internal relationships need to be

taken into account. More generally the problem can be characterized in the following

165
way: If an unfocussed XP is related by an identity condition to a focussed YP and the

focus value of a domain that includes both XP and YP is computed, then the value

of XP should covary with the focus-alternatives of YP. The examples in (110) also

show that the solution given above, binding of the destressed XP by YP is insufficient

because in (111) the destressed XP isn’t c-commanded by YP.

A third place where the same problem comes up is (112a) from Kratzer (1991).

The interpretation of (112a) paraphrased in (112b) shows that the relationship be-

tween the focussed instance of Tanglewood and the instance of Tanglewood in the

elided VP must be visible for the interpretation of the focus particle only.

(112) a. I only went to [TANglewood]F because YOU did hgo to Tanglewoodi (Kratzer

1991:(15))

b. ‘The only place such that I went there because you went there is Tangle-

wood.’

Recall, at this point, the assumption of Kratzer (1991) introduced in (81) that

an F-marked constituent corresponds to a variable in the presuppositional skeleton.

The argument Kratzer gives for this assumption is based on (112a). Her point is for

the interpretation (112b) the elided instance of Tanglewood must also correspond to

a variable in the presuppositional skeleton and more precisely, it must correspond to

the variable as the focussed instance of Tanglewood.

Kratzer (1991) executes her proposal assuming on an identity of form require-

ment for VP-ellipsis, namely LF-copying of syntactic material. If it’s required that

166
the F-mark and the focus index of the overt instance of Tanglewood are also present on

the elided instance of Tanglewood, both instances are translated as the same variable

in the presuppositional skeleton by the procedure (81).

(113) a. I only went to [TANglewood]F 1 because YOU did hgo to TanglewoodF 1 i

(Kratzer 1991:(15))

As shown by (113), Kratzer’s (1991) proposal predicts that Tanglewood in

the elided VP is focussed. This seems counterintuitive on a deletion view of VP-

ellipsis and is in fact ruled out by most analysis of ellipsis. Furthermore, Kratzer’s

example can also be created with VP-destressing as in (114a) and does allow the same

reading, though most people prefer there in (114). In this case, it would be probably

contradictory to assume either LF-copying or that the destressed there or occurence

Tanglewood be focussed in (114a).

(114) I only went to [TANglewood]F because YOU went to there/?? Tanglewood16

Finally, the claim that F-marking is present on material in a ∼P that corresponds to

an F-marked constituent in the antecedent would interfere with my account Kennedy’s

example (2a) repeated in (115a). Under Kratzer’s (1991) assumption, the relative

clause internal copy of country is F-marked as shown in (115b). But, (115b) satisfies

the focus sensitive identity condition, and hence (115a) would incorrectly be predicted

to be acceptable.

16
While (114) is clearly acceptable with there, it seems to be considerably degraded when the
lexical item Tanglewood is reapeated. This is at present unexplained.

167
(115) a. ∗ Polly
P visited every town in a country Eric did hvisiti.
h i
b. ∗ every town, in a [country]F λy Eric visited [y, [country]F ]
| {z }
elided VP
λx Polly visited [x, town]
| {z }
antecedent

The problem with Kratzer’s account is that the correspondent of a focussed

phrase is always interpreted as a variable in the presuppositional skeleton—that’s

after all what it means ot be focussed. Rather, it seems that the correspondent of a

focussed phrase is only translated as a variable if the constituent the presuppositional

skeleton is computed for includes the focussed constituent itself. Therefore, I propose

to represent the correspondent of a focussed constituent not as focussed itself, but

as bearing focus index, specifically the same focus index as the focussed phrase it

corresponds to. Using this notation, consider the two representations in (116a) and

(117a). For the computation of the presuppositional skeleton of (116a), the F-index

on the instance of Tanglewood that corresponds to the focussed instance must be

ignored. The result is (116b). But, for the computation of the presuppositional

skeleton of (117b) the F-index on the same instance of Tanglewood must cause it to

be interpreted as a variable, such that (117b) is the result. Moreover, the identity

relationship internal to (117a) is probably the reason the focus index is present on

elided instance of Tanglewood.

(116) a. [[[went to [Tanglewood]1 ]]]

b. ‘went to Tanglewood’

168
  
 
(117) a. 
went to [Tanglewood]F 1 because [you]2 went to [Tanglewood]1 

| {z } | {z }
antecedent ∼P

b. ‘went to x1 because x2 went to x1 ’

Wold (1996, 1998) develops for independent purposes a formalism where a

focus can be sometimes interpreted as a variable and at other times as its lexical con-

tent, depending on the constituent the presuppositional content is computed for. He

calls this device a Switch Strategy. It’s an interesting question whether his motivation

for introducing this device and the use of it I make here are related in a more relevant

way. For the moment, I just apply Wold’s device to the case at hand.

Wold’s idea is to only use partial assignment functions for the computation of

the presuppositional skeleton and interpret a focus as a variable only if this variable is

defined. In difference to Wold, I assume a distinction between F-marked constituents

bearing a focus index and constituents that aren’t F-marked, but bear a focus index.

This difference is that the F-marked constituents must, in some sense, be able to

enforce an interpretation as variable not only of themselves, but also of the the non-

F-marked constituents bearing the same focus index.

Therefore, I postulate the interpretation rules of constituents bearing an F-

index as in (118). What F-marking adds to a focus index x is that it forces an

evaluation under an assignment G that assigns a value to x. Since, this effect influ-

ences the evaluation of the entire constituent under consideration, it can be used to

trigger that the non-F-marked constituents with the same focus index are interpreted

as variables.

169
(118) a. [[[[XP]F x ]]]G is only defined if x is in the Domain(G)

[[[[XP]F x ]]]G = G(x) where defined





 G(x) if x is in the domain of G
b. [[[[XP]x ]]]G =



 [[[XP]]]G otherwise

The identity requirement must now be revised to take advantage of the flexi-

bility the switch strategy provides. Moreover, it needs to ensure that the focus index

of the correspondent of a focussed phrase is that of the focussed phrase. Therefore,

the identity condition I propose has the two clauses in (119). The first clause is almost

the same as (82) except for the domain minimality requirement. This requirement

makes sure that a constituent bearing a focus index is only interpreted as a variable, if

either itself or another instance of the same focus index in ∼P also bears an F-mark.

The second clause of (119) makes sure that the correspondents in ∼P of a focussed

phrase in the antecedent bear the same focus index.

(119) [[antecedent]] = [[[∼P]]]G for a G with a minimal domain such that [[[∼P]]]G is

defined and for this G:

[[[antecedent]]]H = [[[∼P]]]H for any H that expands17 G

Coming back to (71b), repeated in (120a), consider the LF-representation in

(120b). As indicated by the raised focus indices in (120b), the identity condition

applying between the external and internal head of the relative clause, has ensured

17
An (assignment) function H expands an (assignment) function G if the domain of H is a superset
of the domain of G and for any x in the domain of G it’s the case that G(x) = H(x).

170
that the internal head bears the same focus index as the external head, though the

internal head isn’t F-marked. For the antecedent and ∼P indicated, the definition

of the identity condition just given is satisfied. The reason is that ∼P contains

an F-marked instance of the focus index 2, and therefore even the phrase [lakes]2

in the relative clause internal trace position is interpreted as a variable when the

presuppositional skeleton of ∼P is considered.

(120) a. The cities Mary visited are near the lakes Bill did hvisiti.

b. The [CIties]F 1 [MAry]F 3 visited [x, [cities]1 ] are

| {z }
antecedent
near the [LAkes]F 2 [BIll]F 4 did visit [y, [lakes]2 ]
| {z }
∼P

In this way, the modification of Kratzer’s (1991) approach developed above

accounts for (71b). The status of the focus indices probably needs to be thought

about more, if they occur independently of focus as I have claimed. Nevertheless, I

believe the essential idea of the account is right; that there is a dependency between

a focussed constituent in the antecedent and its correspondent in ∼P that is relevant

when the presuppositional skeleton of a phrase containing both of them is computed.

In particular, the account make one interesting prediction; namely, it’s expected

that the dependency it introduces is subject to the parallel dependencies requirement

(45).18 I argue now that two classes of examples show that the prediction is correct.

The first are the examples of Kennedy’s puzzle like (2a). The second one are cases
18
This might intuitively seem surprising, but it follows from Rooth’s (1992b) account of the parallel
dependencies requirement as I show at the end of section 4.1.2. Note in this context that ‘pseudo-
sloppy’ readings in (i) obey parallel dependencies, where the effect of a sloppy pronoun arises despite
that lack of c-command.

171
like (72b) from above, where the only potential antecedent for a relative clause is a

question with a different NP-part.

Consider first (2), repeated in (121a) with the LF-representation in (121b),

where the relevant focus indices are indicated. The dependencies that are at issue

in (121b) are: the dependency between country2 in the relative clause internal trace

position and the focussed instance countryF 2 in the external head; and the dependency

between the copy of townF 1 in the QR-trace and the copy of townF 1 in the phrase

that moved by QR for ACD-resolution.

(121) a. ∗ Polly
P visited every town in a country Eric did hvisiti.
h i
b. ∗ every [town]F 1 , in a [country]F 2 λy Eric visited [y, [country]2 ] λx Polly

visited [x, town]1 F

The two dependencies aren’t parallel: Compare the structural relationship of the

relative clause head to the relative clause in (122a), with that of the NP-part of the

moved phrase to its complement in (122b). The NP-part in (122a) c-commands its

dependent in the relative clause, while the NP-part in (122b) doesn’t.

(i) a. The policeman who arrested Johni read himi hisi rights and the policeman who arrested
Billj did hread himj hisj rightsi too.
b. ∗ T
The policeman who Johni talked to read himi hisi rights and the policeman who arrested
Billj did hread himj hisj rightsi, too.

172
 DP
H
  HH H

D NP
H
 HH
(122) a.

NP-part CP

H
 H

λy ...

IP
HH
 H
 HH

DP IP
b. H
 HH H
 H

D NP λx ...
H
 HH

NP-part Mod.

Now consider the second case: an example where the only potential antecedent

for a relative clause is a question with a different NP-part. The contrast in (123)

((123a) and (123b) repeated from (72)) illustrates that this case is also ill-formed.

(123a) is an example where a relative clause serves as the antecedent for a question,

while (123b), where the question is the potential antecedent for a relative clause, is

ill-formed. The controls in (123c) and (123d) show that the ill-formedness of (123b) is

due to the difference in lexical content of the antecedent of the relative clause internal

trace and the the wh-word in the question.

173
(123) a. I know the cities Mary visited, but I would like to know which lakes she

did hvisiti.

b. ∗ I know which cities Mary visited, but I would like to know the lakes she

did hvisiti

c. I know the cities Mary visited, but I would like to know which cities Bill

did hvisiti.

d. I know which cities Mary visited, but I would like to know the cities Bill

did hvisiti

The paradigm in (124) illustrates the same point. Again, the NP-parts of the trace

antecedents differ in (124a) and (124b). (124a), where the antecedent is a relative

clause, and the elided VP appears in a question, is acceptable just like (123a). (124b),

however, where the question is the antecedent and the relative clause contains the

elided VP is ungrammatical. In (124c) and (124c) the NP-parts of the two trace

antecedents are identical, and there is no contrast between the relative clause an-

tecedent, question with deletion case in (124c) and the question antecedent, relative

clause with deletion in (124d).

(124) a. We know which is the house Marlyse bought, but not which car Paul did

hbuyi

b. ∗ We
W know which house Marlyse bought, but not which car is the one Paul

did hbuyi

174
 c. We know which is the house Marlyse bought, but not which house Paul

did hbuyi

d. We know which house Marlyse bought, but not which is the house Paul

did hbuyi

The facts in (123) and (124) seem to be a remarkable discovery, since they essentially

recreate Kennedy’s puzzle without antecedent containment. It is, I believe, no small

achievement of the account developed here, that is predicts the entire paradigms in

(123) and (124) correctly. The examples with identical NP-parts ((123c), (123d),

(124c), and (124d)) are acceptable, because the content of the traces is identical, just

like the examples (10). Consider the LF-representation in for (123d) in (125). For

the ∼P indicated, the antecedent is suitable.

(125) I know which cities λx Mary visited [x, cities],

| {z }
antecedent
but I would like to know the cities λy [Bill]F did visit [x, cities]
| {z }
∼P

Next, consider the examples (123a) and (124a) where the NP-parts are different and

the antecedent of a question is a relative clause. In the question the focus of the head

of wh-chain can percolate to the trace position, as argued in the previous subsection.

This is indicated in the LF-representation in (126). Because the lexical content of the

trace in ∼P is focussed, it doesn’t block identity, and the antecedent and ∼P satisfy

the identity requirement.

175
(126) I know the cities λx Mary visited [x, cities],
| {z }
antecedent
but I would like to know [which [lakes]F ] λy she did visit [y, [lakes]F ]
| {z }
∼P

Finally, consider (123b) and (124b). The LF-representation of (123b) is given in

(127). Because focus cannot percolate to the trace position in a matching relative

clause, the ∼P indicated in (127) isn’t identical to the antecedent, even if the overt

occurrence of lakes is focussed.

(127) ∗ I know [which cities] λx Mary visited [x, cities],

| {z }
antecedent
but I would like to know the [lakes]F λy she did visit [y, lakes]
| {z }
6= antecedent

If the domain the identity condition is applied to is expanded as in (128), the focus

index dependency between the external and internal head of the matching relative

is part of ∼P, and now effectively counts as a focus on lakes. But, (128) violates

the parallel dependencies requirement: the dependency of the focus index 2 isn’t

parallel to that of focus index 1 in the antecedent. The difference between the two

dependencies is due to the fact that the relative clause is a sister of NP, while the

predicate created by wh-movement is a sister of DP. This is same structural difference

as that shown in (122). Therefore, the examples like (123b) are in all respects relevant

to the account pursued here alike to Kennedy’s puzzle (2).

176
(128) ∗ I know [which citiesF 1 ] λx Mary visited [x, citiesF 1 ],
| {z }
antecedent
but I would like to know the [lakes]F 2 λy she did visit [y, lakes2 ]
| {z }
6= antecedent

The account of (71b) is hence corroborated by these non-trivial prediction. In

the remainder of this section I show that, based on the account of (71b), an argument

can be made for the assumption that the NP-part of a chain is not only represented

in the trace position, but also in the higher positions of the chain unless binding

is involved. I introduced this assumption in the introduction of chapter 2 without

presenting any arguments in favor of it. At this point, an empirical argument for it

can be made.

Recall that the lexical material inside a relative clause trace can become in-

visible to the identity condition if the focus index dependency to the external head is

part of the domain considered. In addition the focus index dependency of the relative

clause trace material is subject to the parallel dependencies requirement as attested

by the ungrammaticality of (2) and (72) as just discussed. With this in mind, consider

the examples in (129).

(129) a. Which city that Mary did hvisiti is near the lake that John visited.

b. After I saw a lake that John visited, I was wondering which city that Mary

did hvisiti is more enjoyable.

The examples in (129) are like (71b), except that the relative clause that contains

the elided VP is part of a wh-chain. The previous discussion of wh-chains in sections

177
2.2 and 3.1, argued that the NP-part of the wh-chain and the relative clause can be

represented in different positions of the chain. However, the ellipsis in (129) requires

parallel focus index dependencies of the head of the relative clause and the relative

clause internal trace for the two relative clauses in both (130a) and (130b). If the

lexical content of the trace is only represented in the trace position, even if the relative

clause occurs in the top position of the chain the LF representation of (129a) is (130),

which doesn’t satisfy parallel dependencies for the focus indices 1 and 2.

h i
∗
(130) Which λy that Mary visited [y, city2 ] λx is [x, [city]F 2 ]

near the lakeF 1 λz that John visited [z, lake1 ]

Therefore, if the NP-part in a chain is represented in only one position, it follows that

the examples in (129) require reconstruction of the relative clause to the position that

the NP-part must reconstruct to. Reconstruction results in the LF-representation in

(131), where the focus index dependencies are parallel.

h i
(131) Which λx is [x, [city]F 2 λy that Mary visited [y, city2 ]]

near the lakeF 1 λz that John visited [z, lake1 ]

If, on the other hand, the NP-part of a wh-chain is represented not only in the trace

position, but in every position of a chain, reconstruction of the relative clause in

(129) is not required. Consider the LF-representation in (132), which is just like

(130) except for the additional, seemingly redundant, instance of cityF 2 in the head

position of the wh-chain. This extra instance of city satisfies focus index parallelism

178
in the domains indicated.

h i
(132) Which cityF 2 λy that Mary visited [y, city2 ] λx is [x, [city]F 2 ]
| {z }
∼P
near the lakeF 1 λz that John visited [z, lake1 ]
| {z }
antecedent

A test for reconstruction is Condition C. The examples in (133a) and (133b)

are structurally like the examples in (129), but reconstruction of the relative clause

attached to the wh-word is blocked by Condition C. It seems that coreference is

possible in (133a) and (133b) as indicated. Hence, I conclude that the examples in

(133) argue that the NP-part is represented in all positions of a wh-chain.

(133) a. After I saw the lake that John visited, I was wondering which city that

Maryi did hvisiti shei would prefer to that lake.

b. The person John met at the party knows which girl that Maryi did hmeet

at the partyi shei stayed in touch with afterwards.

3.4 Summary

In this chapter, I looked at ellipsis constructions where the elided constituent and

the antecedent contain a trace. The question I asked is when the two traces are

considered identical in the sense that is relevant for the licensing of ellipsis. It turns

out that the most intricate pattern of data is found in the examples with the structure

of Kennedy’s puzzle (2), which sections 3.1 and 3.2 are concerned with. Section 3.3

looks at other constructions with traces in ellipsis and shows why their properties

179
differ from those of Kennedy’s puzzle.

The main empirical discovery of section 3.1 are contrasts like (134) (repeated

from (5)). The only difference between (134a) and (134b) is the lexical content of the

NP the relative clause is attached to. This difference, however, affects the possibility

of ellipsis: If the head of the relative clause is identical in meaning to the NP-part of

object of the matrix verb visit, as in (134a) ellipsis is possible. Otherwise, ellipsis is

impossible.

(134) a. Polly visited every town that’s near the one Eric did hvisiti.

b. ∗ Polly
P visited every town that’s near the lake Eric did hvisiti.

Section 3.1 shows that contrasts like (134) argue that parts of the moved constituent

are represented in the trace position of movement. If traces have lexical content, it’s

expected that whether two traces are identical for the licensing of ellipsis is affected by

the lexical content of their antecedent. I argue that this is precisely the explanation

of (134). Furthermore, section 3.1 shows that the amount of lexical material in a

trace position that the identity criterion establishes is the same as that argued for

based on the distribution of Condition C in chapter 2. Specifically, I show three cases

where the two criteria lead to the same result: the effect of ACD, the integrity of the

NP-part, and the the A/A-bar distinction. The results of chapter 2 and section 3.1

together are therefore a much stronger argument for the lexical content of a trace,

then each result individually.

Section 3.2 establishes that the lexical content of a trace makes a semantic

180
contribution to the constituent containing the trace. I show that the semantic con-

tribution of the content of the trace to the elided constituent and its antecedent is

important for the licensing of ellipsis. One way I argue for this claim is to argue that

the licensing of ellipsis only looks at the semantic content of the elided constituent

and its antecedent—it requires identity of meaning, not identity of form. To make

this point I summarize the account of Fox (1998a) for the few cases that were thought

to require an identity of form requirement in addition to an identity of meaning re-

quirement in Rooth (1992b), and conclude that Fox’s (1998a) account renders the

identity of form requirement redundant. The second argument for the claim that the

semantic content of the traces determines the possibility of ellipsis comes from facts

like (135)(repeated from (39)). (135) shows that the acceptability of examples with

the structure of Kennedy’s example is affected by the semantic relationship of the

two NPs involved, the head of the relative clause, and the NP-part of the matrix

object. The relevance of this semantic relationship cannot be explained if the two

NPs, which constitute the content of the traces at logical form, don’t make a semantic

contribution in the trace positions. On the view that the content of a trace makes a

semantic contribution, on the other hand, the effect of the semantic relationship can

be predicted in the was section 3.2 discusses.

(135) a. Jon ordered a drink that’s more expensive than the drink Sue did horderi

b. ∗ Jon
J ordered a drink that’s more expensive than the dish Sue did horderi

J ordered a cocktail that’s more expensive than the beer Sue did horderi
c. ?? Jon

J ordered a drink that’s more expensive than what Sue did horderi
d. ? Jon

181
 While sections 3.1 and 3.2 look at the identity criterion in examples with

the structure of Kennedy’s puzzle, section 3.3 looks at other cases. If explain why

the effect of the identity requirement on traces is usually not observed in cases of

wh-movement other than ACD, as for example in (136) (repeated from (71)).

(136) a. I know which cities Mary visited, but I have no idea which lakes she did

hvisiti.

b. The cities Mary visited are near the lakes Bill did hvisiti.

Section 3.3.2 argues that (137a) should be explained based on the assumption a focus

on the head of a wh-chain is also represented in the trace position. Then, the general

observation that focussed material is not relevant for the identity condition of ellipsis

also explains why ellipsis is possible in (137a). This account, however, doesn’t carry

over to (136b), because I argue that the focus on the head of a relative clause is

not represented in the relative clause internal trace position in (136b). Section 3.3.2

argues that (136b) should be analyzed as a kind of sloppy reading of the dependency

of the internal head of the relative clause and the external head. In particular, I show

that the notion of sloppiness required for (136b) independently required following in

one case the argumentation of Kratzer (1991). Interestingly the account developed in

section 3.3.3 predicts that the identity criterion should apply in (137) (repeated from

(123)) in the same way as in examples that have the structure of Kennedy’s puzzle.

This prediction is seen to support the approach developed in section 3.3.3.

182
(137) a. I know which cities Mary visited, and now I would like to know the cities

Sue did hvisiti

b. ∗ I know which cities Mary visited, but I would like to know the lakes she

did hvisiti

183
184
Chapter 4

Linking Trace and Antecedent

In this chapter, I present several new results concerning the semantic mechanism that

links a trace to its antecedent. I call this mechanism the Dependency Mechanism. This

mechanism is a central piece of semantic competence, but it is unfortunately quite

difficult to study. The results I present compare three widely used mathematical

models of the dependency mechanism: one, the variable free model of combinatorial

logic; two, variable binding in the form of first order logic (with restrictors added);

and three, variable binding combined with λ-calculus. The three models are math-

ematically equivalent, as far as I know, and the choice between them is a matter

of convenience for mathematic purposes. But, I present arguments that for linguis-

tic purposes the third mechanism is the most appropriate. In the conclusions, I try

to isolate the factors that make model three more successful in accounting for the

data I present, in the hope that these properties are properties of the dependency

mechanism.

In the organization of this thesis, this chapter starts a new topic. The previous

185
two chapters have concerned the content of the trace position in the LF-representation

and its contribution to interpretation. Almost no attention was paid to the fact, that

ultimately the interpretation of the antecedent of the chain must be able to, in some

sense, involve the trace position. In the notation I used, a dependency was represented

by the λx next to the antecedent phrase and the x that’s part of the representation of

the trace. This chapter and the following concern the mechanism(s) that accomplishes

the semantic relationship of trace and antecedent.

(1) [Which book] λx did John read [x, book]

This chapter, in particular, concerns the aspects of the interpretation mechanism(s)

that chains have in common with the relationship of a bound pronominal and its

binder. For example the bound pronoun his in (2), is dependent on the interpretation

of its binder every boy in a similar way to a trace. Chapter 5 addresses what is

particular to chains, namely the semantic content of the trace position.

(2) Every boyi was riding hisi bicycle.

The notation used to represent the fact that which book and [x, book] in (1)

belong together semantically was chosen ad hoc, and I could have used instead any of

the notations in (3) or infinitely many other notations. In (3a) and (3b), the a variable

index is written in different positions. In (3c), the non-antecedents are marked instead

of the antecedent. And in (3d), the dependency is graphically indicated in a manner

similar to Higginbotham (1983). Probably, (3d) would have been the most appropriate

186
notation for the previous two chapters, since it doesn’t suggest anything about the

mechanism.

(3) a. [Which book]x did John read [x, book]

b. Whichx bookx did John read bookx

c. [Which book] .did .John .read [book]

d. which book did John read book

Obviously the notation isn’t interesting; the mechanism is. Of the three models

mentioned at the beginning, two use variables and assignment functions. Namely,

λ-calculus and an extended first order logic with restrictors have variable in common.

The third view, combinatorial logic, does without variables, but instead employs more

complex rules of combination. I call the former view the Variables View, the latter

the Combinatorial View. While I presuppose knowledge of the basics of the variables

view, I briefly introduce the combinatorial view below just before the beginning of

section 4.1. Before doing that, I sketch the kind of argument for the variables view I

detail in section 4.1.

The main difference between the two views is that on the variables view the

dependents in one dependency relation are different from those in any other depen-

dency relation, namely the variables involved are different.1 On the combinatorial

view, however, the dependents are more or less semantically vacuous, and there’s no

1
Technically, it’s only required to choose different variables for dependencies that overlap. In the
discussion surrounding (11), I present reasons to believe that non-overlapping dependencies might
also involve different indices.

187
semantic difference between the dependent of one dependency and those of another

one. (This is, as mentioned, introduced below in detail.) Using the identity condition

of focus semantics from the previous chapter, it’s possible to distinguish the two views

empirically: Consider a situation sketched in (4) where the domain of identity, ∼P,

contains a dependent element but not the phrase it depends on. Furthermore, the

antecedent in (4) contains a dependent element in a position corresponding to that

in the domain of identity ∼P, but the phrase it depends on is different.

antecedent domain of identity (∼P)

z }| { z }| {
(4) binder . . . . . . dependent . . . . . . binder . . . . . . dependent . . .

On the variables view, the contributions of the dependents to the meaning of the

antecedent and the meaning of ∼P could potentially be different in (4). Namely,

the variables chosen for the dependencies could differ. On the combinatorial view,

as explained below, the contribution the dependents make to the meaning of the

antecedent and ∼P in (4) are identical. It turns out that, on the variables view,

there’s some reason that the variables chosen for the two dependencies actually must

be different. Then, there’s a clear difference in prediction made for a structure like

(4). On the variables view, the dependent in ∼P must be focussed, since otherwise

the identity condition is violated. On the combinatorial view, no focus is required in

the same situation.

To test for the difference in predictions between the two views, a focus structure

and interpretation like that in (4) must be argued for in an actual example. The

dependent element of (4) is a pronoun that receives a sloppy reading, which is easy

188
to test for.2 The other part of the situation in (4) is that a domain is subject to the

identity condition, that includes the sloppy pronoun, but not the binder. Since the

question which domains are subject to the identity condition is getting important,

the following terminology is convenient: I call the domains that the identity condition

must apply to Focus Domains following Truckenbrodt (1995) and I’ll keep Rooth’s

notation to mark the domains that are subject to the identity condition with a ∼-mark

in the focus semantic representation of a sentence.

I present two arguments distinguishing the variables view from the combina-

torial view based on the difference in prediction for the situation in (4). The first

argument relies on additional restriction on the placement of focus that Schwarzschild

(1998) argues for. Informally, the requirement is that, if a phrase is focussed, it must

be different from the antecedent—Schwarzschild calls this the Avoid F(ocus) Prin-

ciple. Assuming this, the argument for the variables view comes from the fact that

the dependent pronoun can optionally be focussed as in (5). On the variables view,

because of the difference in variable name, Avoid F is satisfied if the focus domain in

(4) is considered. On the combinatorial view, since there is no difference between the

two pronouns, Avoid F cannot be satisfied for any choice of focus domain. Therefore,

the focus placement in (5) is predicted to violate Avoid F on the combinatorial view.

(5) Every boyi is riding hisi bike and every manj is riding [HISj ]F bike.

2
It seems impossible to create the configuration in (4) with the dependent element being a trace,
because the possibility of intermediate landing sites usually makes is possible that there is another
binder close enough to the dependent to be part of ∼P.

189
 While the first argument involved an optional focus domain, the second argu-

ment looks at a situation where a certain focus domain is forced. In that case, the

variables view predicts that a sloppy pronoun must be focussed, while the combina-

torial view predicts that it must be destressed. Below, I argue based on arguments

of Schwarzschild (1998) that essentially every branching F-marked constituent must

be a focus domain. Additional assumptions of Schwarzschild (1998) concerning the

relationship between the placement of F-marks and the placement of pitch accent

are introduced below and important for the argument. Together, they yield a very

precise picture of the focus structure of (6d) if (6d) is part of the discourse in (6),

and only left bears pitch accent in (6d). The focus structure argued for is indicated

in (6d). The prediction of the variables view is that (6d) is ill-formed without focus

on her, while the combinatorial view predicts (6d) to be well-formed. Since there is

a contrast between (6d) and both (7a) and (7b) when part of the same discourse, I

conclude that the prediction of the variables view is confirmed in (6d).

(6) a. A: Who cut the carrots?

b. B: John didn’t. Hei broke hisi right hand.

c. A: Did Mary cut the carrots?

d. B: No. ∗ Maryj cut [herj [LEFT]F hand]F

| {z }
| {z
∼P }
∼P

(7) a. B: No. Maryj cut [hisi [LEFT]F hand]F

| {z }
| {z
∼P }
∼P

190
 b. B: No. Maryj cut [[HERj ]F [LEFT]F hand]F
| {z }
| {z
∼P }
∼P

These two arguments for the variables view are presented in detail in sections

4.1.1 and 4.1.2. Together sections 4.1.2 and 4.1.1 argue not only that the variables

view is more appropriate than the combinatorial view of binding, but also that the

names of variables matter for the identity of meaning considerations relevant for focus

and destressing. This assumption, that the indices of variables matter for semantic

identity considerations, is underlying the index identity account of Kennedy’s puzzle.

This account is mentioned and argued against in the introduction of the previous

chapter 3. The apparent conflict between the argument against the index identity

view presented in chapter 3 and the arguments for the index identity requirement of

section 4.1 is addressed in section 4.2. Retracing a line of argumentation of Heim

(1997a), I show that the argument against the index identity view of Kennedy’s

puzzle argues only against one of the two popular incarnations of the variables view.

I follow Heim (1997a) in calling the two implementations the Formulas view and the

Predicates view. The difference between the two is indicated by the notation in (8).

The formulas view, indicated in (8a), uses as a model an extension of standard first

order logic to allow restricted quantification. The two arguments of a quantifier,

restrictor and scope, are formulas with an unbound variable, which the operator

binds. The predicates view uses λ-calculus as the model: the two arguments of a

quantifier are one-place predicates, and the quantifier itself is a function from tuples

of predicates to truth values. For index identity, the difference between the two views

191
is that on the latter the λ-predicate is a constituent where the index of the moved

phrase is bound, but that doesn’t include the moved phrase itself. As I show below,

this difference favors the predicates view.

(8) a. [Whichx book(x)] did John read [x, book]

b. [Which book] λx did John read [x, book]

4.1 Variables or Combinators

This section contrasts the view of binding based on the notion of a variable with that

of combinatorial logic by looking at the focus domains that include only a sloppy

pronoun, but not the binder of it. (See the discussion in footnote 2 about using traces

instead of pronouns.) It starts by introducing the two views, and in subsections 4.1.1

and 4.1.2 argues that the indices of the variables view matter for the focus structure

of a sentence. Namely, 4.1.1 shows that focus can force indices to be distinct, while

4.1.2 shows that absence of focus can under special circumstances force indices to be

identical.

Both views of dependency I discuss, the variables view and the combinatorial

view, are taken from mathematical logic. Since the variables view is older and more

popular, I explicate it first.

Within mathematical logic, the status of variables has been viewed differently.

In the first versions of first order logic, Frege (1884) and Whitehead and Russell

(1910), a variable has no status other than marking a dependency for the statement

of inference rules quantificational statements. An unbound variable, on this view, had

192
no well-defined meaning. With the advent of model theory, Tarski (1936), variables

did get a meaning, namely the refer to a value that’s provided by an assignment, a

kind of storage and retrieval mechanism. This later concept of a variable is what

has become to be the major model for dependent reference in linguistics. The use

of variables and assignments in semantics is well-known and very clearly presented

in recent textbooks (Larson and Segal 1995, Heim and Kratzer 1998). The essential

idea is that the meaning of a phrase XP is relative to an assignment functions that

must at least assign a value to the variables that occur free in XP. Secondly, the

meaning of a complex phrase XP = [X Y] is a combination of the meanings of its

parts X and Y relative to the same assignment function, except when one of the parts

is a binder. In case one of the parts of XP is a new binder, the assignment relative

to which the meaning of the sister is considered is modified, as shown in (9b) for

the empty operator λ. (See section 4.2 below for definitions for quantifiers on the

formulas view.)

(9) a. [[X Y]]g = C([[X]]g , [[Y]]g ) where C is the semantic composition function (see

section 1.1)

b. [[λζ Y]] = the function λx. [[Y ]]g[ζ7→x]

The variables view as presented so far leaves it open to which extent the in-

dices of unbound variables matter for the comparison of meanings needed for focus

semantics. The following three possibilities come to mind to state the identity re-

quirement for an antecedent XP and a focus domain YP: One, it could be that the

193
indices of variables don’t matter at all for the identity condition. This can be stated

as the requirement that there is an assignment g such that the meaning of XP under

g is identical to that of YP under g. The two other views have in common the as-

sumption that indices do matter, which can be expressed by an identity requirement

that for every assignment g the meaning of XP under g is identical to that of YP

under g. The difference between possibility two and three is whether reuse of indices

is possible In mathematical logic, the formula in (10a) is well-formed and has the

same meaning as (10b), because the two dependencies don’t overlap (as long as x and

y are unbound within the surrounding material indicated by dots in (10)). Possibility

two is to assume that similarly in semantic representations that choice of index for a

dependency is free except for the case of overlap.

(10) a. (∀x: . . . x . . .) . . . (∀x: . . . x . . .)

b. (∀x: . . . x . . .) . . . (∀y: . . . y . . .)

If reuse of an index for different dependencies was possible, the identity condition

could in most cases with variables be satisfied by reusing an index. The third possi-

bility is that the indices of variables do matter and that it’s reuse of an index is not

possible. The third possibility results in the strongest restriction and it’s the version

of the variables view advocated by Sag (1976) and Heim (1997a). Henceforth, when

I mention the variables view, I refer to this third possibility as the ‘official’ version of

the variables view. Heim (1997a) states the requirement that reuse of an index isn’t

possible as in (11) (see also Sag 1976, Chomsky 1986:75 and below)

194
(11) No Meaningless Coindexing: If an LF contains an occurrence of a variable

v that is bound by a node α, then all occurrences of v in this LF must be

bound by the same node α. (Heim 1997a:(24))

The results below argue in favor of not just the variables view, but more pre-

cisely, the third possibility of explicating it. One argument against the first possibility

considered above, that indices don’t matter at all, is the observation of McCawley

(1976:328) (also Bach, p.c.ṫo Williams 1977), that a deictic pronoun in an elided

VP must refer to the same individual as the corresponding deictic pronoun in the

antecedent VP does. This is shown for VP-deletion in (12a) and for destressing in

(12b). If deictic pronouns are interpreted as unbound variables, for which the dis-

course provides an appropriate assignment, McCawley’s observation can be stated as

a requirement that the index of the variable the pronoun him is interpreted as must

be the same in the focus domain and the antecedent. This requirement follows from

the second and the third possibility mentioned above, but not from the first.

(12) a. Betsy saw himi and Sandy did hsee himi i.

b. Betsy saw himi and Sandy saw himi .

Note, however, that the argument based on deictic pronouns depends very much on

the assumptions made for deictic pronouns. For example, if deictic pronouns are

phonetically reduced forms of proper names, the facts in (12) would also be expected

on the first view. As already mentioned the arguments I give below for the variables

view argue, in fact, for possibility three from above.

195
 Distinguishing the possibilities two and three empirically, I leave for below. It

seems, though, that possibility three is also conceptually simpler: If we assume that

the computational system of syntax doesn’t use variables, variables are introduced at

the point where the LF-structure of a sentence is translated into a semantic represen-

tation. As mentioned the reuse of an index must be prohibited on both possibilities,

if the dependency the index was first used for is overlapping with the one it’s being

reused for. For example, such a restriction is needed for (13a), where two chains are

overlapping. If in the translation of the syntactic representation containing chains

into a semantic representation containing variables the indices of variables could be

freely chosen, a semantic representation like (13b) must be blocked.

(13) a. ? What
W mani do you know what manj to talk to tj about ti ?

b. ∗ What
W manx do you know what manx to talk to x about x?

The easiest way to block (13b) is to postulate that different chains are always trans-

lated with a different variable index. This is possibility three. If, as possibility two

assumes, it’s sometimes possible to reuse an index the procedure translating syn-

tactic chains into operator-variable dependencies would need to verify whether the

no-overlap condition is satisfied. In particular, since this condition must be checked

globally, on the entire structure that is translated, this seems undesirable.3

Combinatorial Logic is the only alternative to variables in mathematical logic,

3
For pronouns, Condition C seems to be such a global condition on the translation of syntactic
representation into semantic representation (David Pesetsky, p.c.). But, even if the existence of such
global conditions is granted, this doesn’t yet justify an unrestricted proliferation of such devices.

196
as far as I know (Schönfinkel 1924, Curry 1930, Curry and Feys 1958, Hindley et al.

1972). Though far less popular than the variables view, a treatment of dependencies

modeled on categorial logic has been proposed by a number of people (Quine 1960,

Szabolcsi 1987, Hepple 1990, 1992, Dowty 1992, Jacobson 1992, 1993, 1994, 1998a,

1998b). Since the different adaptations vary in their terminology and range and no

standard has emerged, the exposition I give here uses the notation of Curry and Feys

(1958).

A constituent XP that contains a dependent element, but not its antecedent

is, on the categorial view, always interpreted as a function, that given an appro-

priate argument yields the interpretation that XP would have if the argument was

inserted in the position of the dependent element. The open argument position of

the dependent is kept open until the antecedent is encountered. To keep this position

open, the semantic composition mechanism of a combinatorial semantics are more

flexible. In the following, I annotate the composition rule that applies to determine

the interpretation of a complex phrase from its parts in the node dominating the

complex phrase. Most advocates of the combinatorial view use some convention like

this. As mentioned in the introduction, on the variables the semantic composition

rule applying for each phrase is probably predictable from the semantic types of its

parts (Klein and Sag 1985, Heim and Kratzer 1998), and therefore I haven’t indicated

the composition rules above. For the combinatorial view, I am not aware of any dis-

cussion of the predictability of the composition rule applying—because of the bigger

inventory of composition rules, the result of the variables view doesn’t carry over to

the combinatorial view.

197
 The definitions of the combinators, I assume, are those marking functional

application, function composition, and ‘duplication’. For the first two, in addition,

the direction must be indicated which I do with the signs . and / following for example

Steedman (1996). Functional application, which is usually not indicated by any sign,

therefore is indicated by just the direction mark as in (14).

 
 Z 
 
 
(14)   is defined as [[X]]([[Y]])
 XX 
 
 
X . Y

 
 Z 
 
 
  is defined as [[Y]]([[X]])
 XX 
 
 
X / Y

Function composition is indicated by the letter B and the direction mark, as defined

in (15).
 
 Z 
 
 
  is defined as λx[[X]]([[Y]](x))
(15)  XX 
 
 
X B. Y

 
 Z 
 
 
  is defined as λx[[Y]]([[X]](x))
 XX 
 
 
X B/ Y

Important for binding is the Duplicator. This simple version of the duplicator in (16)

when applied to a binary predicate X yields a unary predicate which is derived from

198
X by applying the same argument twice. In effect the duplicator enforces cobinding

of two argument positions. I indicate the points where the duplicator applies with

the letter W (Jacobson uses Z instead).

 
 Z 
 
 
(16)   is defined as λx.[[X]](x)(x)
 
XX 
 
 
W X

In the case of overlapping dependencies some version of the following generalized

Duplicator is needed. It’s however not needed for any of the cases below.
 

Z
 
 
(17)  
 XX  with n < m is defined as
 
 
Wn,m X

λxλy1 . . . λyn-1 λyn+1 . . . λym-1 .[[X]](x)(ym-1 ) . . . (yn+1 )(x)(yn-1 ) . . . (y1 )

Dependent positions can be treated as either semantically empty or be inter-

preted as the identity function. I follow Jacobson (1998b) in assuming the latter.

Consider then the example (18) of a bound pronoun. The semantic representation in

(20) with the combinatory rules indicated yields the bound interpretation.

(18) Every boyi called hisi mother.

199
 S
H
 H
  HH
 HH
HH
(19) NP
  HH
HH  H
 H
W VP
Every . boy H
 HH
  HH
  HH

called B. NP
HHH
 H

his B. mother

idDe

Compare (18) with an example where there’s no binding like (20). In the

semantic representation of (20) in (21), the combinatory rules are different from those

in (19).

(20) Every boy called Mary’s mother.

S
HHH
 H
 HH
 HH
  HH

(21) NP . VP
HH HH
 H  HH
 HH
Every . boy  H

called . NP
H
HH
 H

Mary’s / mother

200
 As the comparison between (19) and (21) indicates, the relationship between

a dependent and its antecedent is notated in the composition principles on the com-

binatorial view. Therefore, it must be assumed that in the translation of syntactic

chains into a semantic representation, the choice of combinator applying in each node

is at least partially determined, such that the dependency of a chain is correctly rep-

resented. At present it seems to me that this requires annotating the LF-structure

with the appropriate composition rule for each node, as I have done above.4

In contrast to the variables view, the combinatorial view seems to allow only

one possibility with respect to how the identity condition of focus semantics and

applies to a phrase that contains a dependent element, but not its antecedent. Since

there are no indices, a one dependent means the same another. Hence, there is a

difference between the combinatorial view and the official version of the variables

view. The prediction of the combinatorial view is that a sloppy pronoun, like hisj

in (22), is identical to its antecedent, no matter what domain of identity is under

consideration.

4
Pauline Jacobson (p.c.) notes that, if all restrictions on possible dependencies are part of inter-
pretation, the syntax-semantics mapping needs no restrictions on the composition principles. This
would be conceptually a simpler view of the syntax-semantic interface, and there are indeed clear
examples of dependencies ruled out for semantic reasons, for example a pronoun her presupposes
that its antecedent is of female gender. However, there seem to me to be equally clear cases of
semantically conceivable dependencies that are impossible because the corresponding syntactic rep-
resentation cannot be derived. A particularly well understood case is the difference between crossing
and nesting dependencies illustrated in (i) (from Pesetsky 1982:268 with minor modifications). As
Reinhart (1981), Rudin (1988), Koizumi (1994), and Richards (1997) show, the explanation of (i) is
actually an interaction of the Shortest Attract requirement of syntax with a morphological property
of English, namely how many Specifiers of CP are possible. Richards (1997) demonstrates that
languages with a different morphological property, show the opposite judgement pattern for (i).
(i) a. ? What
W mani do you know what manj to talk to tj about ti ?
∗
b. W What mani do you know what manj to talk to ti about tj ?
The arguments against the combinatorial view developed in sections 4.1.1 and 4.1.2, apply to the
view Jacobson suggests.

201
(22) Every boyi called hisi father and every teacherj called hisj father

For example the meaning of antecedent and ∼P indicated in (23), are exactly identical.

(23) Every boy / W [called B. [idDe B/ father]] and

| {z }
antecedent
every [teacher]F / W [called B. [idDe B/ father]]
| {z }
∼P

On the variables view, the antecedent and ∼P, as in (23) are not identical. Rather,

the extended focus domains in (24) must be considered to license destressing in (22).

(24) Every boy λx x called x’s father and

| {z }
antecedent
every [teacher]F λy y called y’s father
| {z }
∼P

Both views predict correctly that destressing is licensed in (22), though the

licensing focus domains are different. To be actually able to distinguish the variables

view from the combinatorial view empirically, it’s necessary to have a better under-

standing of the distribution of focus domains. Both section 4.1.1 and section 4.1.2

argue first for a certain generalization about the distribution of focus domains and

then consider the implication for the question of whether variables or combinators

are the correct view. The generalizations about the placement of focus domains are

based on the those of Schwarzschild (1998), but different.

202
4.1.1 Forcing Different Indices

Based on the old observation that focussed material must be ‘new’ in some sense,

Schwarzschild (1994, 1998) argues for a ban against superfluous F-marking, which

he calls the Avoid F condition. Obviously, this ban can be absolute since otherwise

there would be no F-marking at all. Therefore the ban against F-marking must

interact with the factors requiring focus, namely the identity requirements imposed

by focus domains ∼P. The nature of this interaction is not obvious. For the moment,

I assume the principle Avoid F as defined in (25), according to which the Avoid F

condition applies to a structure after the focus domains have been determined. I

discuss Schwarzschild’s version of defining the Avoid F condition in section 4.1.2 and

show below that (25) has empirical advantages.

(25) Avoid F: F-mark as little as possible without violating the identity require-

ments imposed by ∼Ps.

Direct evidence Schwarzschild (1998) gives for the Avoid F condition are

question-answer pairs like (26). The answer in (26b) is felicitous where exactly the

new material is focussed in the answer. The focus structure in (26c) is infelicitous

because there is no focus domain structure such that Mary would not have an an-

tecedent. This is expected if the entire answer constitutes a focus domain ∼P. Then

focus on the new information, John, is required since no antecedent with John in

the object position is available in the discourse in (26). The F-mark on Mary in

(26c), however, is avoidable since an antecedent is available that satisfies the identity

203
condition without focus on Mary, as (26b) attests.

(26) a. Who did Mary praise?

b. Mary praised [JOHN]F .

c. ∗ [Mary]
[ F praised [John]F

The argument for the variables in this section is based on the observation in

(27) that the example of a sloppy reading (22), the sloppy pronoun can optionally

be focussed. At first, an optional focus seems to be inconsistent with the idea of an

Avoid F condition, and it’s indeed inconsistent with Schwarzschild’s (1998) statement

of Avoid F as I show below. The way Avoid F is stated in (25), however, allows op-

tionality of F-marking in principle, if different choices of focus domains force different

amounts of F-marking. This is what I assume to be the case in (27). As was shown

in (43a) above, the absence of focus on the sloppy pronoun can be explained easily.

But, for the choice of focus domains considered there, since they tolerate the absence

of focus on the sloppy pronoun, Avoid F blocks focus on the sloppy pronoun.

(27) Every boyi called hisi father and every TEAcherj called HISj father.

To satisfy Avoid F, there must be a placement of focus domains such that the focus

on his is required in (27). At this point, the variables and the combinatorial view

diverge: on the variables view there is such a placement of focus domains, namely

that in (28). For ∼P1 in (28) to be identical to the antecedent, the variable y must

be focussed to be distinct from the antecedent.

204
(28) Every boy λx x called
| x’s
{z father} and every [teacher]F λy y called [y’s]F father
| {z }
antecedent1 ∼P1
| {z } | {z }
antecedent2 ∼P2

On the combinatorial view, on the other hand, there’s no distribution of focus domains

such that the focus marking on his is required. In particular, the choice of focus

domain indicated in (29) is identical to the antecedent. Therefore, Avoid F cannot

be satisfied for (27) on the combinatorial view.

(29) Every boy / W [called B. [idDe B/ father]] and

| {z }
antecedent
every [teacher]F / W [called B. [idDe B/ father]]
| {z }
∼P

Another example making the same point is (30), where in addition we see that

the strict reading in (30a) is indeed blocked by the focus on the pronoun. The lack

of the strict reading in (29a) is probably predicted by both the variables view ad the

combinatorial view, but not important at this point.

(30) a. ∗ John
J i called hisi mother and Billj called [HISi ]F mother.

b. Johni called hisi mother and Billj called [HISj ]F mother.

Irene Heim (p.c.) points out that examples like (31) where the ranges of

the quantifiers binding the two pronouns overlap don’t allow focussing of the sloppy

pronoun in the second conjunct. The difference between (31) and (27) is unexpected.

The observation might indicate that contrastiveness of a sloppy pronoun requires

more than a different index. An different way of thinking about (31) might be to ask

205
the question whether the semantic relationship of the two quantifiers affects whether

a focus domain that doesn’t include the quantifier can be considered. At this point,

I leave the issue brought up by (31) open.

∗
(31) I expected every student to call his father, but only every YOUNG student

called HIS father.

The first argument for the variables view, was based on the observation that

pronouns with different binders can contrast. This was shown to be unexpected on

the combinatorial view, while on the variables view the difference in indices provides

the necessary contrast.

4.1.2 Forcing Index Identity

The second argument for variables is an attempt to force a focus domain that in-

cludes a sloppy pronoun, but not its antecedent. As was argued above, the variables

approach predicts that in this case the sloppy pronoun must be focussed, while the

combinatorial view predicts that the sloppy pronoun need not be—in fact, because

of the Avoid F principle, must not be—focussed.

In the examples of sloppy readings considered so far, it was always possible to

extend the domain of focus such that it includes the antecedent of the pronoun. To

construct an example where this isn’t the case, I again rely on ideas of Schwarzschild

(1998). The first relevant observation of Schwarzschild, is that in cases like (32) the

answer to the question must obligatorily be a focus domain, since otherwise no focus

206
would be required in the answer. I propose for the moment to capture this observation

by the condition (32). If every sentence in a discourse must be a focus domain, the

identity requirement applies and forces new material to be focussed.

(32) Who praised who?

∗
M
Mary praised John.

∗
[MARY]
[ F praised John.

[Mary]F praised [John]F

(33) Every sentence in a discourse must be a focus domain.

The condition (33) is derived from other conditions below, but the empirical gener-

alization behind (33) seems correct, and the arguments in the following rely just on

(33).

The second important observation of Schwarzschild (1998) is that accent place-

ment within a focussed VP is sensitive to the same constraints as elsewhere. Consider

Schwarzschild’s (1998) contrast between (34) and (35). In both examples the VP must

be focussed since the question is asking for the VP-information. Nevertheless, the

pitch accent must be placed on John in (34b) and on praised in (35b).

(34) a. What did Mary do?

b. She [praised JOHN]F ?

(35) a. What did John’s mother do?

b. She [PRAISED John]F ?

207
Another class cases of Schwarzschild (1998) showing that the placement of the pitch

accent inside an F-marked constituent is affected by the discourse is illustrated by

the dialogue in (36). Only the pitch accent on Donca is required in (36c). But, the

the object of wreck must also be F-marked, because otherwise the entire sentence

(36c) isn’t a licit focus domain: If the object wasn’t F-marked, an antecedent of the

form Bill wrecked the convertible X would be necessary to license the entire sentence

as a focus domain. Since such a sentence is not part of the context but by (33) the

whole sentence (36c) must a focus domain, either both the subject and the verb, or

the object must be F-marked. Because only the object contains a pitch accent, I

conclude that the object is F-marked in (36c).

(36) a. John drove the convertible that Barry liked.

b. Aha. And Bill wrecked a boat?

c. No, Bill wrecked [the convertible that DONCA liked]F

| {z }
∼P

One conclusion, Schwarzschild (1998) draws from facts like (34), (35), and (36) con-

cerns the relationship between pitch accent and F-marking. Namely, he proposes

that it’s necessary and sufficient for the phonetic realization of F-marking that an

F-marked phrase contains a pitch accent, with the one exception stated in (37) for

examples like (34b), which is however irrelevant for the following.

(37) Phonological Realization of F-marking: Every F-marked phrase must

contain a pitch accent. (except for an F-marked verb whose complement is

208
 also F-marked and contains a pitch accent)

Condition (37) leaves it open on which word within a complex F-marked phrase the

pitch accent falls. But, the placement of pitch in a complex F-marked phrase is de-

termined by the preceeding discourse, in a similar way that determines the placement

of pitch within matrix sentences. For example, it’s impossible in the context of (36c)

to place the pitch accent on the verb liked.

∗
(38) N Bill wrecked [the convertible that Donca LIKed]F
No,

The focus domain mark on the entire sentence doesn’t make any prediction concerning

pitch placement within the F-marked phrase, because the effect of the F-marking on

the object is to make the information in its scope irrelevant to the focus domain it’s

part of. In the definition of the presuppositional skeleton (81) in section 3.3 the F-

marked constituents of a focus domain were replaced by variables for this reason. For

the same reason, any focus domain mark that includes the entire F-marked object

in (36c) will not distinguish between (36c) and (38). Therefore, there must be a

focus domain within the F-marked object to capture Schwarzschild’s observation that

the placement of pitch accent within an F-marked constituent is determined by the

same discourse considerations that determine pitch placement otherwise. There are a

number of possibilities to spell this insight out more precisely I present Schwarzschild’s

account of (36) first, but am ultimately going to draw slightly different conclusions

which are closer to Truckenbrodt’s (1995).

Schwarzschild (1998) proposes that all non-F-marked constituents are focus

209
domains. Furthermore, Schwarzschild states the Avoid-F principle as a global con-

dition, that requires minimization of F-marking by looking at the entire sentence up

to the requirement that non F-marked constituents must satisfy the identity condi-

tion. Together, the two assumption explain the paradigms in (34), (35) and (36),

as Schwarzschild shows in detail. Consider, for example (36) repeated in (39). As

already argued, the object in (39c) must be F-marked. However, it remains open

whether the subconstituents of the F-marked object are also F-marked. Because of

Avoid-F, F-marking is to be avoided here too. And since the antecedent the convert-

ible that Barry liked is part of the discourse, F-marking is only required on the noun

Donca, which must receive the pitch accent. In (39c), I indicated the two F-marked

constituents and all the focus domains that Schwarzschild’s proposal predicts. Notice

though that one attractive aspect of Schwarzschild’s proposal is that focus domains

need not be indicated, because the presence of a focus domain is indicated by the

absence of an F-mark.

(39) a. John drove the convertible that Barry liked.

b. Aha. And Bill wrecked a boat?

c. |{z}
Bill wrecked
| {z } [the
|{z} convertible
| {z } [DONca]F liked
| {z }]F
YP YP YP YP YP
| {z }
YP
| {z }
YP
| {z }
YP
| {z }
YP

Going back to sloppy readings, Schwarzschild’s (1998) proposal is only com-

patible with the combinatorial view of binding. Consider as an example of a sloppy

210
reading (40) (repeated from (22)). In the second conjunct of (40), only the noun

teacher must be focussed. Therefore, one of the many focus domains Schwarzschild’s

proposal predicts is the one indicated in (40). But, as argued above, ∼P in (40)

doesn’t have an antecedent on the variables view of binding, while the first conjunct

provides an antecedent on the combinatorial view of binding. I take this consequence

of Schwarzschild’s (1998) proposal to be undesirable because of the evidence presented

in the previous section against the combinatorial view of binding.

(40) Every boyi called hisi father and every [TEAcherj ]F called hisj father
| {z }
∼P

A second argument against Schwarzschild’s (1998) statement of Avoid F and

the distribution of focus domains is the optionality of focus that was observed in

the previous section. (41) repeats the example from (43a) and (27) where focus on

the sloppy pronoun his is optional. If Avoid F attempts to minimize the number of

F-marks for the entire sentence, F-marking of the sloppy in (41) is predicted to be

impossible, because the alternative Focus structure without this F-mark is possible.

(41) Every boyi called hisi father and every TEAcherj called hisj /HISj father.

For these two reasons, I adopt a different proposal concerning the distribution

of ∼Ps than Schwarzschild. Recall that the evidence in (34) to (36) shows that

non-focussed material within a complex phrase requires a discourse antecedent like

that of non-focussed material outside of an F-marked constituent. For destressed

material outside of any F-marked constituent this requirement was captured by the

211
generalization (33), which, however, was insensitive to the focus structure internal to

a complex F-marked phrase. It is therefore natural to consider the generalization (42)

analogous to (33), which forces all complex F-marked phrases to be focus domains.5

(42) Every complex F-marked phrase is a focus domain.

However, (42) makes the wrong prediction for (35) (repeated in (43)). Consider the

∼P indicated in (43b). It requires an antecedent of the form V John, which arguably

the discourse (43a) doesn’t provide an antecedent for.

(43) a. What did John’s mother do?

b. She [[PRAISED]F John]F

| {z }
|
∼P
{z }
∼P

It seems that in (43b), a focus domain is required inside the F-marked phrase, but

need not include more than the object John. Hence, I assume that the requirement

for a ∼P is related to the presence non-F-marked phrase, just like in Schwarzschild’s

(1998) proposal. But, in contrast to Schwarzschild’s proposal, I assume that it’s

sufficient for destressed material to occur in the scope (or domain) of a ∼P without

any F-marks intervening. To capture the fact that an intervening F-mark interrupts

the licensing between a ∼P and a destressed phrase, I define the notion of immediate

scope in (45). For the licensing of destressed (i.e. non-F-marked) material, I propose

5
The restriction to complex F-marked phrases, is needed because imposing (42) on F-marked
terminals would lead to circularity: If an F-marked terminal was a focus domain, this focus domain
would require domain F-marking of the terminal within this focus. This F-marking would create
another even smaller focus domain, which would bring about further requirements ad infinitum.

212
the condition in (44).

(44) Every non-F-marked phrase must be in the immediate scope of a ∼P.

(45) X is in the immediate scope of ∼P if there’s no F-mark dominating X, but

not dominating ∼P (and no other ∼-mark dominating X, but not dominating

∼P)6

The condition (44) accounts for the facts in (34) to (36), while allowing both

a variables and a combinatorial account of simple examples of sloppy readings like

(22). Consider first (35) (repeated in (43) and (46)). Because praised is F-marked

in (46b), it doesn’t need to be in the immediate scope of a ∼P. Therefore, the ∼P

inside of the complex F-marked constituent needs to only include John, as indicated

in (46b).

(46) a. What did John’s mother do?

b. She [[PRAISED]F John

| {z }]F
| {z
∼P }
∼P

Next, consider (36) (repeated in (47)). One possibility of licensing all destressed

constituents is the one indicated in (47). There are a number of other possible distri-

butions of focus domains that condition (44) permits, but in all of them there is at

least one focus domain within the complex F-marked constituent.

6
The requirement that the be no intervening ∼-mark is unnecessary at this point. I include it
though because then immediate scope expresses the intuition that the ∼P that X is in the immediate
scope of is the primary one where the discourse requirement of a destressed ∼P is verified. The
requirement does play a role below.

213
(47) a. John drove the convertible that Barry liked.

b. Aha. And Bill wrecked a boat?

c. No, Bill wrecked [the

| convertible that
{z DONCA liked}]F
| {z
∼P }
∼P

Thirdly, reconsider (22) (repeated in (48)) under the licensing condition (44). For

the second conjunct of (48), (44) allows the focus domain structure indicated; namely

only one focus domain that contains the sloppy pronoun and its antecedent. As shown

above, the focus structure in (48) is predicted to satisfy the identity requirement of

∼P on both the variables and the combinatorial view of binding.

(48) Every boyi called hisi father and every [TEAcherj ]F called hisj father
| {z }
∼P

I return now to the question of whether the combinatorial or the variables

view of binding is more accurate. In this section so far, it’s shown that Schwarzschild

(1998) conclusions about the distribution of focus domains are only compatible with

the combinatorial view, but a slightly different view of his facts allows us to maintain

either the variables or the combinatorial view. The other important result of the

discussion above, is that a destressed phrase that occurs in an F-marked constituent

must be in the scope of a ∼P that is smaller than the F-marked constituent. In

a sense, F-marked constituents are an upper boundary for the extension of focus

domains. I show now that this result together with the variables view makes a new

prediction about the availability of sloppy readings that is borne out.

214
 Recall that the variables approach requires that a destressed sloppy pronoun is

licensed in a focus domain that also includes the binder. In the examples considered so

far, it was always possible to choose a focus domain big enough to license a destressed

sloppy pronoun. The result of the discussion of Schwarzschild’s (1998) data, that F-

marking limits the extension of focus domains, can block licensing of a sloppy pronoun.

The prediction is that a sloppy pronoun that’s part of a F-marked constituent which

doesn’t include the binder requires F-marking.

To test the prediction, I use discourses similar to Schwarzschild’s (1998) ex-

ample (36), but with a pronominal dependency. Clear examples aren’t easy to create.

However, all my consultants agreed on the example (6), repeated in (49), from the

introduction. With pitch accent only on left, (49d) isn’t possible in the discourse (49).

(49) a. A: Who cut the carrots?

b. B: John didn’t. Hei broke hisi right hand.

c. A: Did Mary cut the carrots?

d. B: No. ∗ Maryj cut herj LEFT hand.

I show first that (49d) must have a focus structure like (50), in the discourse above.

The reasoning is analogous to that in (36) above: Because Mary cut is destressed, it

must be part of a focus domain with (49c) as its antecedent. But, then the object

her left hand must be F-marked. Condition (44) forces another focus domain internal

to the F-marked object to exist, because the destressed words her and hand must be

licensed by such a focus domain. Hence, there must be focus domain that contains

215
her, but not its antecedent Mary. If we assume that there is a preference to choose a

big focus domain, this forces the focus domain shown in (50).7

(50) B: No. ∗ Maryj cut [herj [LEFT]F hand]F

| {z }
| {z
∼P }
∼P

Then, (49d) is an example where the sloppy pronoun is in a focus domain that the

antecedent isn’t part of. As discussed above, the variables view predicts such an

example to be impossible, while the combinatorial view predicts it be acceptable. If

the judgement on (49d) is the one indicated, it therefore argues for the variables view.

As mentioned above, the judgement on (49d) is different when her is stressed.

This is also predicted by the variables view, because the focus makes the index of the

sloppy pronoun irrelevant for the identity condition on focus domains, as discussed

in the previous subsection.

(51) B: No. Maryj cut [[HERj ]F [LEFT]F hand]F

| {z }
| {z
∼P }
∼P

Another important control are the strict readings in (52). (52a) and (52b) are accept-

able with the same focus structure that was impossible for the sloppy reading in (50).

For the licensing of (52b), it either needs to assumed that the antecedent John cut

his hand has an alternative representation where his isn’t bound by the antecedent

7
In (49d), if her and LEFT hand form separate focus domains, with Mary the antecedent of the
focus domain of her, the example would be predicted to be acceptable even on the variables view.
At the end of this section I argue that this possibility is blocked by a condition of Truckenbrodt
(1995) that requires the maximalization of ∼Ps. The other examples I discuss in the following don’t
allow such a focus structure.

216
(Keenan 1971, Sag 1976:125, Reinhart 1981) or, as Rooth (1992b) suggests, that the

antecedent has an entailment of the right form to license (52b), e.g. Somebody cut his

hand, such that indirect identity in the sense of section 3.2 is satisfied.

(52) a. B: No. Maryj cut [hisi [LEFT]F hand]F

| {z }
| {z
∼P }
∼P

b. D: No. Maryj cut [Johni ’s [LEFT]F hand]F

| {z }
| {z
∼P }
∼P

The following two examples, illustrate the same point that (6) made. The

example in (53) is initially quite hard to imagine as a discourse. But, once this

difficulty is overcome most of my consultants agreed to the indicated judgment.

(53) a. A: John didn’t wash the dishes. John damaged the car his father was

leasing.

b. B: Aha. Did Mary wash the dishes?

c. A: No. ∗ Mary washed the car her father was SELLING.

Again, it’s instructive to compare (53c) with different pitch placements. The pitch

placements in (54) improve the example.

(54) a. A: No. Mary washed the car HER father was SELLING.

b. A: No. MARY WASHed the car her father was SELLING.

The contrast between (53c) and the alternatives in (54) is predicted by the variables

217
view of dependencies. Look at the focus structures of the three examples, as given

in (55). (55a) and (55b) are analogous to the previous example. (54b) as shown in

(55c) can be licensed with one focus domain that includes both the antecedent and

the sloppy pronoun. This ∼P can be licensed under identity to (53a).

(55) a. ∗ Mary
M washed [the car her father was [selling]F ]F
| {z }
| {z
∼P }
∼P

b. Mary washed [the car [her]F father was [selling]F ]F

| {z }
| {z
∼P }
∼P

c. [Mary]F [washed]F the car her father was [selling]F

| {z }
∼P

The example in (56), shows a preference in the predicted direction; namely that his

needs to be stressed. But, the judgment is even less clear than that in the previous

two examples. I suspect that to make (56) a more coherent discourse some people

assume that (56b) indicates that every American suspects something that his teacher

is something. If (56b) carries such an implicature with it, it could license the focus

structure in (57) for (56c). Hence, those people are expected to find (56c) acceptable.

(56) a. A: Every Canadian believes that his teacher is a genius.

b. B: Is that so? Well, every American suspects something.

(∗)
c. A: You’re right. Every American suspects that his teacher is an ALIEN

(57) Every American suspects that his teacher is an [ALIEN]F

| {z }
∼P

218
 This concludes the argument for the variables view, that is the main point

of this section. The remainder of this section contains two digressions. The first

digression is about how the idea of Truckenbrodt (1995) that the domain of focus

need to maximal could be incorporated into the version of Schwarzschild’s (1998)

system developed above. In particular, I argue that Truckenbrodt’s condition can

predict some cases where Kennedy’s puzzle seemed to arise with A-movement and

how it rules out the confound mentioned in footnote 7 above. The second digression

contains some remarks towards a potential third argument for the variables view.

Namely, it shows that it renders the parallel dependencies requirement (45) for page

(45) partially redundant. However, it shows also that, at this point, two subcases of

the parallel dependencies requirement remain.

Developing an idea of Rooth (1992a:114), Truckenbrodt (1995) argues that

focus domains are also relevant for the phonology of focus and phonological phrasing.

Furthermore, Truckenbrodt assumes that normally the focus domain surrounding a

focussed phrase is extended to include as much destressed material as possible—the

domain of a focus must be maximalized (Truckenbrodt 1995:126–30). Some of the

examples above are relevant to the question of how the domain maximalization idea

can be incorporated into the set of assumptions argued for above.

The first relevant point is that the domain maximalization condition cannot

compare all possible ways of placing focus domains. The reason is the same that led

me to abandon an Avoid F condition that compares all possible ways of placing Focus

domains and F-marks. Namely, examples like (58a) (repeated from (5)) where focus

is optional on a sloppy pronoun. As argued above, there must be two focus domains

219
in (58a), one surrounding the sloppy pronoun his, but not including the binder of it,

and one containing the entire clause. (58b), on the other hand, contains only one

focus domain—the one indicated. If the focus domain maximalization condition was

to force (58b) to only have the one focus domain of (58a), (58a) would be predicted

to violate Avoid F. Therefore, the two focus domains in (58a) must be permitted.

(58) a. Every boyi is riding hisi bike and every MANj is riding [HISj ]F bike
| {z }
| {z
∼P }
∼P

b. Every boyi is riding hisi bike and every MANj is riding hisj bike
| {z }
∼P

Nevertheless, Truckenbrodt’s intuition that focus domains can be too small

or rather too vacuous seems to right in other cases. I propose therefore that the

domain maximalization requirement only applies to focus domains that are trivial

in the sense of (59). At this point, the definition of immediate scope given in (45)

becomes important again: Recall that something is in the immediate scope of a focus

domain if no F-mark nor ∼-mark dominates it that is inside of the focus domain.

(59) A focus domain ∼P is trivial if either there is no F-mark in the immediate

scope of ∼P or there is no destressed material in the immediate scope of ∼P.

Notice that at least in (35) (repeated in (60)) a trivial focus domain was argued to

be possible. However, even increasing the scope of ∼P2 in (60b) wouldn’t lead to a

greater immediate scope of ∼P, because the sister of ∼P is focussed.

220
(60) a. What did John’s mother do?

b. She [[PRAISED]F John

| {z }]F
∼P2
| {z }
∼P1

Therefore, I assume that a trivial ∼P is blocked if an alternative focus structure is

possible where the ∼P has more in its immediate scope. In the two cases I talk about

now, the domain maximalization blocks a trivial ∼P to be the sister of another ∼P

or be immediately dominated by another ∼P. The first case is illustrated by (61)

(cf. footnote 7). In (61), ∼P2 is trivial, because it only contains destressed material.

Hence, I assume that (61) is blocked because of the possibility to replace ∼P2 and

∼P3 with one focus domain.

(61) B: No. ∗ Maryj cut [ herj [LEFT]F hand]F

|{z} | {z }
∼P2 ∼P3
| {z }
∼P1

The second case, are examples like (62) (repeated from (33b) on page 113). The

ill-formedness of examples like (62) was left unexplained in the earlier discussion.

Consider now the focus structure for (62) given: ∼P1 and ∼P2 are both trivial.

∗
(62) [
[Every man who said George would t buy some salmon]F did hbuy some salmoni
| {z } | {z }
antecedent1 ∼P1
| {z }
∼P2

Therefore the restriction on trivial focus domains proposed above requires instead the

focus structure in (63) where ∼P1 contains its antecedent. It is conceivable that this

configuration is either ungrammatical, or at least hard to parse.

221
 ∗
(63) [Every
[ man who said George would t buy some salmon]F did hbuy some salmoni
| {z }
antecedent1
| {z }
∼P1

This analysis of (62) lacks a lot of detail at the moment. Nevertheless, I believe that

it does look promising in the light of contrasts like those in (64) and similar ones in

Heim (1997a).

(64) a.∗? Every

E man who wants George to leave should hleavei.

b. ? Every
E man who wants George to leave did last time.

c. ? Every
E man who did wants George to leave.

The remaining paragraphs of this section point towards another potential ar-

gument for the variables view. Namely, I show that the variables view predicts some

cases of the parallel dependencies generalization in (65) (repeated with minor modifi-

cations from (45) on 122) as Rooth (1992b) points out in passing, while the combina-

torial view makes no prediction in this respect. The argument is very weak, though,

since the variables view doesn’t capture all cases that (65) account for, and therefore

the condition (65) is still needed. I mention it largely because I feel that the variables

view at least gives us a handle on the parallel depedencies requirement, and I hope

that the Rooth’s account can be extended to all cases of the parallel dependencies

condition. Another reason to mention it, is that to show that the instances of fo-

cus index sloppiness where the parallel dependencies requirement was seen to apply

belong to those cases that follow from the variables view.

222
(65) Parallel Dependencies: If a dependent isn’t identical in reference to the cor-

responding dependent in the antecedent, it must stand in the same structural

relationship to its binder as the corresponding dependent in the antecedent.

Consider the contrast in (66) (repeated from (44) on page (44)), which provides direct

evidence for (65). The sloppy interpretation in (66a), which satisfies (65), is possible,

while the sloppy reading indicated in (66b), which doesn’t satisfy (65) is blocked.

(66) a. First, John told Maryi I was bad-mouthing heri ,

and then Sue told Janej I was hbad-mouthing herj i

b. ∗ First,
F John told Maryi I was bad-mouthing heri ,

and then Suej told Jane I was hbad-mouthing herj i

The semantic representation of the second conjunct of (66b) on the variables view—to

be explicit, I assume λ-calculus in (67)—is given in (67). On the variables view, the

minimal focus domain that can be invoked for the licensing of deletion is one that

includes the binder of the variable x.

(67) Sue λx told Jane I was badmouthing x

| {z }
| {z
elided VP }
minimal ∼P

What is a possible antecedent for the minimal ∼P indicated in (67)? The fact this

it must be identical in meaning to the this ∼P modulo the focussed parts of ∼P

restricts the possible antecedents to predicates. Furthermore, I claim the structure

223
of the antecedent predicate needs to effectively correspond to the structure of (67).

The semantic contribution to ∼P of the parts (67) that aren’t focussed must be

exactly matched by a potential antecedent predicate, while for the focussed parts

of (67) the antecedent must contain material that makes an equivalent contribution

to its meaning. Since there are few cases where examples with different structures

have exactly the same meaning, the semantic identity requirement effectively limits

potential antecedents of (67) to predicates with the same internal structure. This

can be assumed in the account of the parallel dependencies condition without loss

of generality because the discussion of (46) on page (46) shows that if there are

cases where predicates with different structure are semantically identical, the parallel

dependencies condition is expected to be obviated. But, if the antecedent predicate

has the same structure, this means specifically that the variable in the same structural

position. In other words, the variables view predicts that any potential antecendent

of ∼P in (67), is a predicate denoting phrase that is structurally isomorphic to (67)

and where the variable predicated over appears in the same structural position as x

does in (67). This predicts that no antecedent is available in (66) for the ∼P indicated

in (67).

Before considering other potential choices of ∼P in (67), notice that while the

prediction of the variables view pointed out in the previous paragraph doesn’t block

all cases accounted for by the parallel dependencies requirement. The difference is

that the prediction just stated only requires that the antecedent predicate contain

a variable in the same position as ∼P, but it doesn’t require that there be a direct

dependency between the two positions. The parallel dependencies requirement, how-

224
ever, requires a direct dependency in this antecedent. As Fox (1998c) points out,

examples that show that the stronger requirement of the parallel dependencies con-

dition is necessary are those known in the literature as Dahl’s puzzle like (68a). The

absence of the interpretation paraphrased in (68b) is the crucial fact, which shows

that the representation sketched in (68c) impossible.

(68) a. Max said that he saw his mother and Oscar did hsay that he saw his

motheri, too.

b. Max said that Max saw Max’s mother and Oscar said that Max saw Oscar’s

mother, too.

c. Max said that he likes his mother

Now, return to the discussion of (66b) and consider the choice of focus domain

as in (69), where the argument of the λ-operator that binds the variable x is also

part of the focus domain. The considerations in the following carry over to any focus

domain which contains the argument of the relevant λ-operator, also in examples

where the focus domain contains other additional material than this argument. Again

the question is: What is a possible antecedent for the ∼P indicated in (69)?

(69) Sue λx told Jane I was badmouthing x

| {z }
| {z
elided VP }
∼P

The same considerations as above show that all the possible antecedents of ∼P that

need to be considered correspond to (69) in structure. Furthermore, it can be argued

225
that the structural positions occupied to x, which refers to Sue, and Sue must also

have be identical in reference in the antecedent, in the cases to consider for the

derivation of the parallel dependencies requirement. Since in the cases where the

parallel dependencies condition applies the reference of the correspondent of x must

be different from that of x, only this situation needs to be considered. But, this

difference in reference will block identity, unless it’s circumvented by focus in ∼P.

Since the pronoun corresponding to x in (66b) cannot be focussed, the only way

focus can affect the reference of x is to focus Sue in (69). Since in all elements of the

focus set of (69) the reference of the position of Sue and that of the position of x are

identical, this is required for the antecedent of ∼P as well. In other words, the two

positions of the dependency of (69) must have the same reference in any potential

antecedent of (69).

The prediction of the variables view just deduced again comes close to ren-

dering the parallel dependencies requirement redundant, but doesn’t fully succeed.

For the example (66b), the prediction explains that no antecedent is available for the

focus domain chosen in (69). There are, however, again examples that show that the

stronger requirement of the parallel dependencies condition is needed. Again, Dahl’s

puzzle represents one class of such examples. An additional class of cases that Rooth

(1992b) discusses, are examples like (70) where the two positions in the antecedent

have the same reference, but no dependency exists between the two positions.

∗
(70) 5 is (obviously) less than or equal to 5, and (of course) 7 is hless than or equal

to 7i, too.

226
 Summing up this last point, the variables view covers a substantial amount of

cases that are the empirical basis of the parallel dependencies requirement. At this

point though the prediction of the variables view doesn’t cover the cases (69) and

(70), and therefore the parallel dependencies requirement is still needed.

4.2 Predicates or Formulas

The previous section argued that the variables view of dependencies is correct, and

that the indices of unbound variables matter for semantic identity of phrases. As

mentioned in the introduction, the variables view itself can be spelled out along the

lines of two different mathematical models. One view, the formulas view, adopts the

assumption of first order logic that every quantifier can bind a variable. The other

view, the predicates view, follows λ-calculus (Church 1932, 1933) in assuming that

there is only one operator, λ, that can bind a variable.

Both positions are quite popular in linguistics: for example, Larson and Se-

gal (1995) assume and present in detail the formulas view, while Heim and Kratzer

(1998) explicate the predicates view. Heim (1997a) contrasts the two views, and ar-

gues that they differ in their predictions in the case of ACD constructions. In this

section, I summarize Heim’s argumentation, but then argue based on the new data

of the previous chapter 3 for the opposite conclusion of Heim’s paper; namely, for the

predicates view. I then give another argument for the predicates view, based on the

distribution of i-within-i reference.

For the example (71), the difference between the two views is represented by

227
the sketches of semantic representations in (72). In (72a), which exemplifies the

formulas view, the quantifier whichx takes two formulas with the unbound variable x

as its arguments. On the predicates view, exemplified by (72b), the two arguments

of the quantifier are predicates, the lexical predicate book and the derived predicate

λx did John read [x, book] .

(71) Which book did John read?

(72) a. Whichx [x book] [did John read [x, book]]

b. [Which book] λx did John read [x, book]

On both views, binding requires a new semantic composition principle. On the for-

mulas view, the rule has to apply to structures consisting of a quantifier and its two

arguments. If we assume, that which is essentially an existential quantifier (see the

following chapter), and example of such an composition rule is given in (73), and it

illustrates the general schema.

 
 CP 
 
 g
(73)   X XX  g[x7→a]
= 1 and [[N ]]g[x7→a] =
   = 1 iff an a exists such that [[R]]
 
 
whichx R N
1

On the predicates view, quantifiers themselves don’t require a special composition

rule; quantifiers can be understood as functions that take two predicates as an argu-

ment and yield a truth value. However, the λ-marking requires the special interpre-

tation rule in (74) that binds a variable and creates a predicate.

228
  
 YP 
 
 g
(74)   is interpreted as the function a 7→ [[Y ]]g[x7→a]
 
XX 
 
 
λx Y

The first argument for the predicates view has to do again with Kennedy’s

puzzle, as repeated in (75) from (2) on page 95.

(75) a. ∗ Polly
P visited every town in a country Eric did hvisiti.

b. Polly visited every town Eric did hvisiti.

Consider first the semantic representation in (76), which the formulas approach pre-

dicts for the examples. As shown in (75), the variable index of the trace in the elided

VP is different from that of the corresponding VP in the antecedent only in (76a),

which is the representation of the ungrammatical (75a).

(76) a. ∗ every
e x [x town in [ay [y country] [Eric visited [y, country]]]] [Polly visited
| {z }
elided VP
[x, town]]

b. everyx [x town Eric visited [x, town]] [Polly visited [x]town]

| {z }
elided VP

Since the indices of the two traces in (76a) are different, the parallel dependencies

requirement must be satisfied by the two. This is however not the case, as the

discussion of (122) on page 173 showed.

On the predicates approach, the semantic representations of (75) are those

given in (77). In both (77a) and (77b), the variables of the trace positions differ, and

in both cases there binders are in parallel positions.

229
(77) a. ∗ [every
[ town in a country λy Eric visited [y, country]] λx Polly visited [x,
| {z }
elided VP
town]

b. [every town λy Eric visited [y, town]] λx Polly visited [x, town]
| {z }
elided VP

Therefore, (75a) is predicted to violate the parallel dependencies requirement only

on the formulas approach. Heim (1997a) argues based on this observation for the

formulas approach. Since it predicts (75a) to be ill-formed, she concludes that it’s

right. But, in the light of the facts observed in the previous chapter, it turns out

that Heim’s observation actually can be used for an argument against the formulas

approach.

In the previous chapter I showed that (75a) is ruled out by the semantic content

of the trace position. Therefore the fact that the formulas approach also rules it out,

says little in favor of the formulas approach. In fact, it was shown with (5) on page

(5), repeated in (78), all that’s wrong with Kennedy’s example is the semantic content

of the trace.

(78) John visited a town that’s near the town Mary did hvisiti.

Namely, (78) has the same structure as Kennedy’s example above. The only difference

between the two examples is the lexical content of the trace position. Since (78)

is acceptable, it argues against the formulas approach which would predict it to

violate the parallel dependencies condition. As shown above, (78) is predicted to be

grammatical on the predicates approach.

230
 Heim (1997b) mentions two other cases where ACD is possible, but index

identity is not expected on the formulas view. Namely, comparatives as in (79a) and

partitives in (79b). The same point as for (78) can be made for the examples in (79).

(79) a. John can run faster than Mary can hrun fasti.

b. Bill visited the three oldest cities out of the ones that Mary had advised

him to hvisit ti.

The second argument for the formulas approach is based on the distribution

of i-within-i reference. I use the term i-within-i reference in the following way: A

pronoun exhibits i-within-i reference with a determiner D if the pronoun covaries

in reference with the quantification of the determiner D and occurs inside the DP

that determiner D projects. Chomsky (1981:212,229) observes the argument-adjunct

distinction with respect to i-within-i reference illustrated in (80). If the pronominal

anaphor itself occurs in the NP-part of the determiner a it cannot exhibit i-within-i

reference with this determiner as shown by (80a). If the pronoun occurs in a relative

clause adjoined to the NP-part it can refer i-within-i. The contrast in (80) shows that

an adjunct occurring inside the NP-part of a DP doesn’t allow i-within-i reference. In

example (81a) from Vergnaud (1974:31) the pronoun him occurs in a relative clause,

but one that is adjoined to an argument inside the NP-part of the relevant determiner.

i-within-i reference is impossible in (81a), while it’s possible in (81b) where the second

DP itself occurs inside a relative clause adjoined to the NP-part of the first DP.

231
(80) a. ∗ Kai
K drew [a picture of itselfi /iti ]i

b. Kai drew [a picture showing itselfi ]i

(81) a. ∗ the
t son of the woman who killed him was a Nazi (Vergnaud 1974:(62i))

b. the guy buried near the woman who killed him was a Nazi

The generalization illustrated is that i-within-i with a pronoun and a deter-

miner D is possible if and only if the pronoun occurs outside the NP-part of the the

DP projected by D. This is an argument for the predicates because the predicates

approach predicts precisely this generalization, while the formulas approach doesn’t.

First, witness the failure of the formulas approach which is already noted in Higgin-

botham (1983:416–18) and Jacobson (1994). Recall that, on the formulas approach,

both arguments of a quantifier must be open formulas containing a variable. For this

reason, the NP-complement on the formulas approach must contain a subject posi-

tion that contains a variable the quantificational determiner can bind. But, if this

subject position can be bound by the determiner, it’s predicted that the determiner

should also be able to bind variables elsewhere in the NP-part of its complement.

Hence, (82) is predicted to be a well-formed semantic representation on the formulas

approach, but the DP it corresponds to in (80a) is ill-formed.

∗
(82) ax [x picture of x]

The predicates approach predicts the distribution of i-within-i correctly: Recall that

the two arguments of a quantifier are predicates and that the quantifier itself doesn’t

232
bind a variable. Since the NP-part of a DP is a lexical predicate, it’s not necessary

to postulate a subject position in the NP-part. In fact, on the predicates approach,

it’s natural to postulate that there’s no subject position in the NP-part of a DP

that covaries with the determiners quantification. Then, the representation of illicit

i-within-i reference is that in (83) (for (80a)), which is ruled out because x isn’t

bound.

∗
(83) a [picture of x] λx . . .

Since relative clauses are derived predicates, they are predicted to allow i-within-i

reference on the predicates approach. Recall that derived predicates are created by

the λ-operator. Since the λ-operator can bind variables in its scope, representations

like (84) for (80b) are well-formed.

(84) a [[picture] [λx x showing x]]

I conclude that the distribution of i-within-i reference is only predicted by the predi-

cates approach, and therefore argues for it. Notice, by the way, that the combinatorial

view of binding also doesn’t predict the distribution of i-within-i reference, as Jacob-

son (1994) shows. On the combinatorial view, there is no difference between derived

predicates and lexical predicates. In addition to the two arguments for the predicates

view presented in this section, I know of two additional arguments for predicates:

Sauerland (1998) presents an argument based on the existence of polyadic quantifi-

cation, and Nissenbaum (1998) presents an argument based on the distribution of

233
parasitic gaps. Nissenbaum’s (1998) argument is the most ambitious; he claims the

existence of λ-operators as independent syntactic heads. While the other three ar-

guments only provide evidence that the complement of a moved phrase as well as

relative clauses are interpreted as predicates, they’re compatible with Nissenbaum’s

(1998) stronger claim.

It should be mentioned that the predicates view also predicts that an argument

of a lexical predicate cannot bind a pronoun in its scope. In (85a), Mary doesn’t

bind the pronoun her, because only the argument of a derived predicate can bind any

pronouns. Hence, the subject must have moved as in (85b) for it to bind the pronoun.

But, since it seems that many and maybe all DPs must move a short distance for

case reasons, this prediction is maybe not as bothersome as it looks at first.

(85) a. ∗ Mary
M likes herx bicycle.

b. Mary λx likes herx bicycle.

4.3 Summary

This sections investigates the contribution a dependent element to the meaning of a

constituent that doesn’t contain the binder of it. The tool that is employed to study

this question is the semantics of focus and destressing. Hence, the examples consid-

ered mainly head the abstract structure in (86). The question that focus semantics

can answer for a configuration like (86) is whether the semantic contribution of the

dependents to the antecedent and the focus domain are the same or not.

234
 antecedent focus domain (∼P)
z }| { z }| {
(86) binder . . . . . . dependent . . . . . . binder . . . . . . dependent . . .

The results show that the answer to the question depends on how much ma-

terial intervenes between the the binder and ∼P. One generalization that fits the

results is the following: If the focus domain ∼P is smaller the the sister of the binder,

the semantic contributions of the dependents differ between the two domains; if the

focus domain ∼P is the sister of the binder, the semantic contributions of the two

dependents are the same.

This generalization is predicted if λ-calculus is chosen as the mathematical

model for the semantics of dependencies. The two other models considered, combi-

natorial logic and extended predicate calculus, were shown to predict substantially

different generalizations which could are inconsistent with the data presented above.

Namely, the combinatorial logic model predicts that the contributions of the depen-

dents to the domains should always be identical, while the extended predicate calculus

model predicts that the contributions should always be identical.

235
236
Chapter 5

Interpreting Moved Quantifiers

The previous chapter argued, that the interpretation of chains involves a mechanism

that has the essential properties of variable binding. Still, for many of the structures

considered in the chapters 2 and 3, it’s not intuitively obvious how the interpretation

procedure applies to a chain to yield the correct meaning. For example, consider

(1) (repeated from (34) on page 50). The LF-representation of (1a) is given in (1b),

with the operator and the trace of the relevant chains marked. It is clear that the

variable x cannot refer to a single individual in (1b), because there need not be an

single individual paper such Mary told every student to revise it, for (1a) to be a

sensible question. But then, the question is what the variable x does refer to in the

interpretation of (1b).

(1) a. [Which paper of hisk that Maryj was given]i did shej tell every studentk to

revise ti ?

237
 b. [Which [λz Maryj was given [z]]] λx did shej tell every student
| {z }
operator

λw [w] to revise [x, paper of hisw ]

| {z }
trace

Another interesting observation about (1b) is that the semantic division of questioned

information and known information that the surface syntax of English suggests is not

transparent in the LF-syntax. For example in (2b), the answer matches the question

except for the wh-phrase. But, in an LF-representation like (1b) the wh-phrase and

the rest of the question don’t form separate constituents, as they seem to do on the

surface in (2a).

(2) a. Q: Which friend of her’s did Mary invite.

b. A: Mary invited Bill.

With non-interrogative DPs, an example of the kind of semantic representation

entertained is given in (3), repeated from (13) from page 39. The semantic represen-

tation of (3a) argued for is (3b). The main feature of (3b) that seems counterintuitive

are the three occurrences of book of Irene’s. It occurs in the trace position inside the

relative clause, in the trace position of quantifier raising and in the operator position

of quantifier. In fact the matching analysis of relative clauses predicts that it also

occurs in the operator position of the relative clause. To interpret the NP-part or

any other segment of the restrictor in more than one position seems redundant. But,

chapters 2 and 3 showed that the NP-part is often interpreted in the the trace position

and section 3.3.3 provided an argument that the NP-part was also represented in the

238
operator position is relevant at LF.

(3) a. In the end, I asked him to teach the book of Irene’s that David wanted me

to hask him to teachi

h i
b. the book of Irene’s λy that Davidi wanted me to teach [y, book of Irene’s]
| {z }
operator

λx I asked him to teach [x, book of Irene’s]

| {z }
trace

The most fundamental problem for interpretation seems to be the one posed

by (1), and in a more condensed way by (4), namely that the variable x in (4b)

cannot be understood as referring to an individual. One solution for this problem

was proposed by Engdahl (1980). She proposes that the variable in (4b) ranges over

choice functions. Since Engdahl’s (1980) solution for (4a) relies on representations like

(4b) I directly adopt it for the case of interrogative quantifiers. Therefore, Engdahl’s

proposal is presented in some detail in section 5.1.

(4) a. Which friend of heri ’s did every studenti invite?

b. Which λx did every studenti invite [x, friend of herselfi ]

I show then that Engdahl’s proposal doesn’t straightforwardly carry over to all non-

interrogative quantifiers. Rather than concluding that therefore the semantics of in-

terrogative and non-interrogative quantifiers is fundamentally, I show that it’s possible

to modify Engdahl’s proposal such that all quantifiers can be explained as involving

quantification over choice functions. At the end of section 5.1 I present an account for

239
the problem mentioned above that parts of the moved quantifiers must be interpreted

in other positions of the chain than the trace position.

Section 5.1 develops Engdahl’s proposal. I first present Engdahl’s proposal and

then go on to show that Engdahl’s choice function can be extended to cover all the

constructions considered in the previous chapters. The three main difficulties for this

extension are the following: First, the fact that interrogative DPs seem to have, as

we will see, always existential quantificational force, whereas non-interrogative DPs

can be headed by determiners with a different quantificational force. The second

difficulty is how to incorporate the contribution of material in the operator position.

And, finally I address the interpretation of intermediate traces.

Section 5.2 points out one important prediction of the choice function approach

developed in 5.1, namely that it predicts many weak crossover effects. The prediction

arises from the the type difference between pronoun and the variables involved in the

interpretation of chains, when choice functions are used.

5.1 A Choice Function Approach to All Quantifiers

The goal of this section is to develop a general interpretation procedure for all DP-

chains making use of the insights of the previous chapters. Since many of the DPs

considered in the previous chapters are wh-phrases, one task of the semantics is

to account for wh-words in questions in a similar way as for other quantificational

determiners. Hence, I start the section by summarizing Karttunen’s (1977) semantics

of questions which treats wh-words as existential quantifiers.

240
 It turns out that it’s easiest to talk about the meaning of a question when

it occurs as the complement of agree on—I owe this insight to Lahiri 1991:16–25

and Rullmann and Beck 1997. In other environments, the meaning of questions is

obscured either by the difficulty of understanding the semantic contribution of speech

acts (in the case of matrix questions) or by the factivity of the question-embedding

verb (in the case of other question-embedding verbs). In this section, I only consider

the contribution to meaning of a question that appears as the complement of agree

on, and refer to the specialized literature for the reduction of other cases to this one

(Groenendijk and Stokhof 1984, Berman 1991, Lahiri 1991, Dayal 1996, and Hagstrom

1998).

Consider now the example in (5). What is the contribution of the embedded

question which student Lisa invited to the meaning of (5)?

(5) Bill agrees with John on which student Lisa invited.

Assume that Bill and John both know the concept student fully and correctly. Then

the truth of (5) implies, that for any student x, (6a) and (6b) must have the same

truth value. And conversely, if for any student x, the sentences in (6) both have the

same truth value, (5) would be considered true.

(6) a. Bill believes that Lisa invited x

b. John believes that Lisa invited x

If this intuition is any guide, the semantics of agree on involves quantification. I adopt

241
the proposal of Lahiri (1991) that agree on involves quantification over propositions.

Then, the question must specify the range of propositions agree on quantifies over.

For (5), the semantics of agree on could be given as in (7). This meaning of agree on

leads to a certain view of the meaning of questions, which is due to Hamblin (1958,

1973). Namely, questions are essentially descriptions of a set of propositions—the set

of propositions agree on quantifies over.

(7) If Bill believes p, then John believes p and vice versa for all propositions p of

the form specified by which student Lisa invited

The remaining question is what set of propositions a specific questions is the

description of. In the above example, the propositions quantified over are of the form

Lisa invited x where x is a student.1 Hence, a proposition p is quantified over if there

is a student x such that p is the proposition “Lisa invited x”. In this paraphrase,

the contribution of the wh-word which seems to introduce existential quantification

over students. This is, in fact, one popular view of the meaning of wh-words since

Karttunen (1977) and is supported also by the morphological similarity amongst wh-

words and indefinite determiners in many languages (see for example Cheng 1991 and

Hagstrom 1998). It is now possible to isolate the contributions that the elements of a

question make towards its meaning on Karttunen’s (1977) approach, as given in the

tree. The three interpretation rules needed are given in (9).

1
The example also has a presupposition that Bill and John both believe that Lisa invited only
one student. This is not relevant for the point here and I’ll ignore it. It is though an interesting
aspect of the semantics of questions and I refer to Schwarz (1993) and Dayal (1996) for discussion.

242
 CP
HH
 H
 HH

λp C0
H
 HH

 HH

(8) which student C0

HH
∃[[student]]  H

λx C0

HH

C[+wh] IP

λq.p = q

(9) a. [[C[+wh] ]] = [λq.p = q]

b. [[which]] = ∃

c. λp is introduced at CP-level to bind p in C[+wh]

Note that while the interpretation rules (9a) and (9b) specify the meaning of lexical

entries, (9c) is unusual as a rule of the translation from syntactic logical form into

a more semantic form of representation for two reasons. For one, it’s specific to

questions but isn’t a rule specifying a lexical entry. Secondly, (9c) must introduce a

binder λp for the unbound proposition variable that (9a) introduced; hence, (9c) is

not a strictly local rule. However, as far as I know, there’s at present no satisfying

way around this undesirable feature of the semantics of questions.

With the semantics of wh-determiners in (9) in mind, look at Engdahl’s exam-

ple. In (10), it’s given as the complement of agree on. What is the set of propositions

243
agree on could quantify over in (10)?

(10) Bill agrees with John on which friend of heri ’s every studenti invited?

In a neutral context, the meaning of (10) can be elaborated in the following way:

If John and Bill both know which individuals are students and who is friends with

whom, the truth of (10) entails that for any x and y, if x is a student and y is a friend

of x, (11a) and (11b) have the same truth value.2 Conversely, if (11a) and (11b) have

the same truth value for any pair of x and y with x a student x and y a friend of x,

(10) would be considered true.

(11) a. Bill believes that x invited a y.

b. John believes that x invited a y.

This paraphrase of (10) using (11) seems to suggest that the subject universal every

student takes scope outside of the proposition p that’s quantified over in the inter-

pretation of (10). This, however, cannot generally be the explanation of examples

like (10).3 While there might be cases where a universal quantifier can take scope

outside of a question (cf. Higginbotham and May 1980, Groenendijk and Stokhof

1984, Chierchia 1993, Moltmann and Szabolcsi 1994), examples like Engdahl’s are

possible when the relevant quantifier cannot take scope outside of the question. In

2
In a marked context, for example when preceded by a discussion of three kinds of typical friend
relationships, boy-friend, oldest friend and grad-school buddy, (10) can be true even when the
entailment to (11) doesn’t hold.
3
The argument in the following is I believe due to Engdahl (1986). I haven’t been able to verify
this, however.

244
(12), the quantifier that binds a variable in the wh-phrase is separated from the ques-

tion’s complementizer position by a finite clause boundary. Furthermore, the higher

question-internal subject position in (12) is occupied by the quantifier a professor.

Since (12) requires that there must be a single professor such that John and Bill be-

lieve that he claimed that every student invited somebody for it to have any answer, a

professor takes obligatorily scope over every student. This is expected because of the

finite clause boundary. But, if every student cannot take scope over a professor, it can

also not take scope over the +wh-Comp that c-commands a professor. Nevertheless,

the binding of the variable in the fronted wh-phrase is possible in (12). Hence, this

kind of binding doesn’t require quantification to a position outside of the question.

(12) They agree on which friend of heri ’s a professor claimed that every studenti

invited. (a À every, ∗ every À a)

Since at least the interpretation example (12) involves a mechanism other scoping the

quantifier to a position outside of the questions, I assume that (10), as well, is inter-

pretable without scoping the subject quantifier to a position above the interrogative

quantifier. Then the propositions described by the question in (10) must be of the

form in (13), where y is a place-holder for the interpretation of the trace. Consider

the proposition of the form in (13) if y was restricted to individuals in its interpre-

tation. Then, (13) would entail that every student invited the same person, namely

y. But, that entailment is wrong since for the truth of (10) it’s not necessary that

every student invited the same person. Hence, y must be able to refer to different

245
individuals covarying with the quantification of the subject. In a way, p in (13) must

be equivalent to a conjunction of propositions of the form “x invites y” entertained

earlier.

(13) p = ‘every student invited y’

Since y cannot be the individual variable that intuition would favor, many less in-

tuitive possibilities are now open. But, the results of chapter 2, I believe, narrow

the options down significantly, leaving the proposal of Engdahl (1980) as perhaps the

most natural candidate. Recall at this point the conclusion of section 2.2; namely,

that the trace position in Engdahl’s example must contain the NP-part of the wh-

phrase. Hence, the propositions described by the questions in (10) must actually be

of the form in (14), where y is the variable bound by the wh-word.

(14) p = ‘every studenti invited [y, friend of heri ’s]’

The meaning of the NP-part in the trace position covaries with the quantifier that

binds her. For one given student that’s quantified over by the subject—let’s call her

Mary—the NP-part friend of her’s denotes the property friend of Mary’s that the

NP-part denotes. As shown above, the question meaning involves propositions of the

form “x invited z” with z being a friend of x. If y in (14) selects one of the individuals

that have the property the NP-part denotes, namely being a friend of Mary’s, the

result is a proposition of the form “Mary invited z”, with z a friend of Mary’s.

Engdahl’s (1980) choice function proposal captures the intuition just expressed,

246
that the variable bound by the wh-word selects an individual that satisfies the prop-

erty the NP-part expresses. A choice function is a function that assigns to properties

individuals which have this property. Formally, this is defined in (15). Sometimes,

I’ll use the abbreviation CF for either the set of choice functions or the term choice

function.

(15) f of type hhe, ti, ei is a Choice Function if x(f (x)) = 1 for all x ∈ domain(f )

For Engdahl’s example (repeated in (16a)) this results in the interpretation repre-

sented by (16b) and paraphrased in (16c).

(16) a. which friend of heri ’s every studenti invited?

b. λp∃λf (f ∈ Dhhe,ti,ti and f is a CF and ∀y ∈ {students}: y invited f (friends(y))

c. There is an f such that p means: for every student x, x invited the one

that f chooses for the property friends of x’s

The choice function f could, for example, always select the oldest of the people having

a certain property. For this f , the proposition p described by (16b) is “Every student

invited her oldest friend”. But, the selections by f made could also not correspond

to any natural definite description, for example f could choose the oldest friend for

one student, the youngest friend for another, and a friend that’s neither the oldest

nor the youngest for a third student.4

4
One problem of Engdahl’s approach is exemplified by the case where two students have exactly
the same friends. The properties of being a friend of these two students are extensionally identical,
and therefore the choice function should select the same individual for both the students. The

247
 Engdahl’s account would carry over straightforwardly to the interpretation of

other questions where the wh-phrase doesn’t contain a bound variable, if the NP-part

wasn’t interpreted in the head of the chain. Consider the representation (17b) for

the example (17a) (repeated from (5)). In (17b), the NP-part in the trace position

doesn’t vary with any quantifier in the sentence, hence the choice function f is only

applied to the property student for which it selects a student. Hence, the proposition

(17b) describes are exactly those of the form “Lisa invited x”, with x being a student.

But, given the argument in section 3.3.3 that the NP-part can appear in both the

operator and the trace position, the representation of (17a) is probably (17c) rather

than (17b). If this is correct, Engdahl account must be slightly modified in the way

shown below.

(17) a. Bill agrees with John on which student Lisa invited?

b. which λf Lisa invited [f , student]

c. [which student] λf Lisa invited [f , student]

In the area of non-interrogative quantifiers, Engdahl’s (1980) proposal has

been widely adopted for wide scope indefinites (Reinhart 1994, 1997, Kratzer 1995,

1998, Ruys 1993, Winter 1997 and Matthewson 1998). In this case, the syntactic

processes involved are different, namely the existential quantifier is not related to the

problem arises more sharply in examples like (i) where the property denoted by the NP-part, ancestor
of her’s, is necessarily the same for each person quantified over. I believe the problem indicates that
the formal notion of property isn’t fine-grained enough to reflect how the denotation of the NP-part
is conceptualized in such cases (cf. Kratzer 1998).
(i) Which ancestor of heri ’s did every common daughteri of John and Mary like best?

248
indefinite that restricts it by a chain as attested by island insensitivity in (18) and

further arguments in Ruys’s (1993). Hence, the question of whether the NP-part of

the indefinite also occurs in the operator position doesn’t arise.

(18) a. Mary will leave if we invite a philosopher.

b. ∃ [λf Mary will leave if we invite [f , philosopher]]

Both interrogative quantifiers and indefinites involve existential quantification,

as motivated above for questions. With universal quantifiers, quantification over

choice functions also works straightforwardly, at least if the occurrence of the NP-

part in the operator position, is ignored as von Stechow (1996) first pointed out.5

Consider the example (19a) assuming (19b) as its semantic representation. For (19b)

to be true, any way of selecting one of the cliffs must be one such that a girl is climbing

the selected cliff. This is sufficient to the intuitive meaning of (19a), namely that for

every cliff there’s a girl who is climbing it.

(19) a. A (different) girl is climbing every cliff.

b. every λf a girl is climbing [f , cliff]

In addition to the problem of material occurring in the operator position, a

generalization of the choice functions to the analysis of all quantifiers faces other

significant problems. Consider (20a) with the cardinal quantifier two taking wide

scope over the subject, assuming (20b) as its semantic representation. I show now

5
Kai von Fintel (p.c.) drew my attention to this part of von Stechow’s paper.

249
that (20a) is predicted to be true in a situation where there is only one cliff that a

boy is climbing.

(20) a. A (different) boy is climbing two cliffs.

b. two λf a boy is climbing [f , cliff]

Assume that College Rock is the only cliff that a boy is climbing. Then, f and g in

(21) are two different choice functions that make the predicate ‘λf a boy is climbing

[f , cliff]’ true. Hence, (20b) is predicted to be true in a situation where College Rock

is the only cliff climbed by a boy. But, this prediction is of course undesirable.

(21) f ([[‘cliffs’]]) = College Rock, f ([[mountains]]) = Everest

g([[‘cliffs’]]) = College Rock, g([[mountains]]) = Zugspitze

A similar problem arises also with the proportional quantifier most in example (22):

The predicate over functions in (22b) will be true of infinitely many functions f

even if only one girl is climbing a cliff. Hence, it’s not easily possible to determine

the proportion that would be required for (22a) to be true. More generally, all

quantifiers for the interpretation of which the cardinality of the domain is important

are problematic for the choice function proposal as developed so far because the

cardinality of the set of choice functions that satisfy a predicate is typically either

zero or infinite.6 For existential and also universal quantifiers this property of choice

6
For the truth of the kind of predicate arising in linguistic examples only the values of the choice
function for a finite set of properties is relevant. Under these circumstances, for example, the number
of choice functions satisfying the predicate will always be zero or infinite.

250
functions isn’t problematic, but for most other quantifiers it is.

(22) a. A girl is climbing most cliffs.

b. most λf a girl is climbing [f , cliff]

There are probably many more ways out of the problem posed by (20). In the

following, I discuss two of them. The first one is to restrict quantification to choice

function that vary over choice functions that are minimal in their domains. Imposing

this requirement on the choice functions considered ensures that two different ones of

the choice functions consider differ in their value for a property actually considered

in the evaluation of the sentence under consideration. To refute this approach, I

show then a second problem that arises when quantification over choice functions

is extended to all non-interrogative quantifiers. The second solution I present is to

assume that two choice functions are only considered different if they are pointwise

different—they differ in value for every property in their domain. This solution can

account for both (21) and the second problem discussed below.

The first way out of the problem of (20) is to restrict the domain of the choice

functions looked at to the properties that are ‘really relevant’. ‘Really relevant’ are

only the values of the choice function for those properties that it’s actually applied to

in the evaluation of the sentence in question. For example, in (22b), f is only evaluated

for the one property: cliff. I assume that in (22) only choice functions defined for this

one property are quantified over. The number of such choice functions is the same as

the number of cliffs, because the number of ways to select one element from one set

251
is exactly the number of elements the set has. Hence, by looking only at such choice

functions, it’s possible to define cardinal and proportional quantifiers in a way that

yields the right interpretation for (21) and (22).

Recall, though, that in example (10) above, the choice function must be defined

for more than one property: Because the argument of the choice function contains

a bound variable, the choice function must at least be defined for all the different

properties that arise given the values the bound variable ranges over. Since it is

desirable to postulate only one interpretation mechanism for all kinds of DPs, this

case must be allowed by the restriction imposed on the set of choice functions a

quantifier quantifies over. This is captured by the definition in (23).

(23) min(C) = { f ∈ C | ∀g ∈ C: domain(g) 6⊂ domain(f ) }

The general way to draw the restriction, hence, seems to be to restrict quantification to

choice functions that have a minimal domain such that the argument of the quantifier

are defined. The meaning of quantifiers can be given by the general schema in (25),

which is exemplified in (24) for two, most and which.

(24) a. [[two]](S) is true if and only if two different elements of min(domain(f ))

make S(f ) true.

b. [[most]](S) is true if and only if more than half of the elements of min(domain(f ))

make S(f ) true.

c. [[which]](S) is true if and only if one f ∈ min(domain(f )) makes S(f ) true

252
(25) [[Q]](S)=1 if and only if Q-many of min(domain(S)) are in { f | S(f ) = 1 }

Going back to the example (20), repeated in (26), the correct interpretations

are now predicted. The minimal choice functions that the λf predicate in (26b) is

defined for, are those that have only the predicate cliff in their domain. As argued

above, the existence of two such choice functions is equivalent to the existence of two

cliffs which a boy is climbing. Notice that whenever the content of the trace position

doesn’t contain a bound variable, the choice functions quantified over are predicted

to be ones with a singleton domain.

(26) a. A (different) boy is climbing two cliffs.

b. two λf a boy is climbing [f , cliff]

Next, consider again the interpretation of Engdahl’s example (27) (repeated

from (10)). The minimal choice functions that the λf -predicate in (27b) is defined

for are those that are defined for exactly all of the predicates friend(y) for the values

of y quantified over.

(27) a. . . . which friend of heri ’s every studenti invited?

b. λp∃λf (f ∈ Dhhe,ti,ti and f is a CF and ∀y ∈ {students}: y invited f (friends(y))

c. There is an f such that p means: for every student x, x invited the one

that f chooses from the set of friends of x’s

Because the semantics given for (27) is general to all DPs, the question arises

253
whether the possibility that a bound variable in the lexical content of the trace can

generally lead to quantification over choice functions with a domain of cardinality

greater than one. The question is what kind of interpretation would be predicted for

such a case. In the case of an indefinite quantifier, the resulting reading would be

equivalent to a narrow scope reading. But, consider a possible wide scope construal

of (28a) with the counting quantifier two taking scope over every. The semantic

representation of (28a) I’m entertaining is shown in (28b).

(28) a. Every student brought two relatives of hisi .

b. two λx every studenti brought [x, relatives of hisi ]

The reading (28b) is predicted to have according to the previous section is I claim not

available for (28a): Consider a situation with two students, one of which, brought two

relatives of his, Lynn and Eve, but the other, Bill, brought only one, Sue. Intuitively,

(28a) is false in such a situation. But, there are then two choice functions that

make the λx-predicate in (28b) true—namely, f , which selects Lynn for the property

relatives of John and Sue for the property relative of Bill, and g, which selects Eve

for the property relatives of John and Sue for the property relative of Bill. Hence,

(28b) is predicted to be true, which is incorrect.

The problem arises generally in the situation in (29), when Qb is a strong

quantifier, and Qa is a quantifier sensitive to the cardinality of its domain. It’s

generally difficult to obtain wide scope of one strong quantifier over another, but

seems to be marginally possible in examples like (30) (cf. Beghelli 1993, 1995, Sato-

254
Zhu 1996). Therefore, the incorrect reading is predicted in (28).

(29) Qa λx . . . [Qb NP]i . . . [x, . . . proi . . . ]

(30) Every student read exactly two books.

The problem brought to light by (28) and (30) is not a problem specific to

the approach I’m developing here. Rather it seems to arise by necessity from the

assumption that all DPs have a uniform semantics. These uniform semantics need to

provide an account for examples like Engdahl’s (10), where a bound variable occurs

inside a wh-quantifier. But, then the question whether this kind of binding is also

possible with examples involving other quantifiers is unavoidable, and probably some

representation equivalent to (29) must be allowed. Instead of giving up the idea of

a uniform DP-semantics as Engdahl (1980) and others do, I want to maintain that

representations like (29) are possible. But then, the interpretation of (28) cannot be

the one given above. I propose that counting quantifiers individuate choice functions

differently than assumed above. If we assume that the quantifier two requires that

there are two choice functions that are different in their value on every argument they

have in common, defined as the pointwise different relation in (31), (28b) is correctly

predicted to be false in the situation laid out above.

(31) f is pointwise different from g if and only if

∀x ∈ domain(f ) ∩ domain(g): f (x) 6= g(x)

More generally, it’s true that, for any set S of finite sets, the maximum number of

255
choice functions with domain S that are pointwise different from each other for each

possible pair is equal to the cardinality of the smallest set s ∈ S. Therefore, if counting

quantifiers require the choice functions that satisfy their domain to be pointwise

different, even representations like (29) are interpreted correctly. This approach can

also accommodate proportional quantifiers if it’s assumed that here the maximum

number of pointwise different choice functions that make the scope of the quantifier

true is compared to the maximum number of pointwise different choice functions that

make it false. Therefore, individuating choice functions with the pointwise different

relation makes it possible to maintain a uniform semantics for interrogative and non-

interrogative determiners.

Notice that the pointwise different requirement renders the restriction to min-

imal choice functions superfluous. Recall, that the observation that lead to the in-

troduction of the minimality requirement was the following. An example like (32a)

(repeated from (20)) with the representation in (32b) is predicted to true if the two

choice functions f and g defined by (33) are considered. However, the requirement

that the choice functions considered in the interpretation of a quantifier must be

pointwise different doesn’t permit the consideration of f and g as defined in (33).

Therefore, the pointwise different requirement can replace the minimality condition.

(32) a. A (different) boy is climbing two cliffs.

b. two λf a boy is climbing [f , cliff]

(33) f ([[‘cliffs’]]) = College Rock, f ([[mountains]]) = Everest

g([[‘cliffs’]]) = College Rock, g([[mountains]]) = Zugspitze

256
 The remainder of this section addresses the question of lexical content in po-

sitions other than the lowest trace position. At this time, what I can say about the

issue is mostly mapping out the issues that arise and to show that the account is not

incoherent because of these issues. To begin with, recall the problem: The schema

for the definition of quantifiers in (25) above assumes that only the operator is inter-

preted in the operator position of a chain. In chapter 2, I argued that this assumption

is incorrect: Both NP-parts and relative clauses can occur in the operator position

of a chain. The potential occurence of NP-parts in the operator position was argued

for in section 3.3.3. Recall that also relative clauses, according to chapters 2 and 3,

can occur in the operator position of a chain as well, and in contrast to the NP-part

don’t even need to be repeated in the trace position. One of the arguments given in

section 2.2 for this conclusion is Freidin’s (1986) observation that Condition C can

be obviated by over wh-movement in examples like (34) (repeated from (20) on page

44). The relative clause in (34) cannot occur in the trace position since its subject

doesn’t trigger a Condition C violation in this position.

(34) [Which argument that Johni had criticized]j did hei accept tj in the end?

It turns out that a relative clause in the operator position is actually easier

to interpret there, then the NP-part is. The reason is that, as argued in section 2.4

relatives clauses usually contain an internal head. Recall that the matching analysis

of relative clauses proposed in section 2.4 claims that the relative clause internal

trace position is occupied by the NP-part of the relative clause head. Hence, the

257
LF-representation of (34) proposed is the one in (35). If the relative clause internal

trace position also contains an NP-part, it seems natural to interpret relative clauses

as predicates of choice functions in the same way as it was proposed for the sister

of a moved quantifier above. The relative clause in (35) then denotes a predicate of

choice functions that is true if the choice function assigns to the property argument

an argument that John had criticized. Hence, if the choice functions that which

quantifies over in (35) is restricted to those that satisfy the predicate the relative

provides, the correct interpretation results in (35).

(35) [which argument λy Johni had criticized [y, argument]] λx did hei accept [x,

argument] in the end.

Generally, if a relative clauses has a non-empty internal head, it is interpreted as a

predicate of choice functions. And, the relative clause is true if the choice function

selects an individual that intuitively would make the relative clause true. Hence,

relative clauses that have an internal head can be combined with the operator they

share a position with as restrictors of the operator. The new definition schema for

quantifiers is given in next.

(36) [[Q]](S)=1 if and only if Q-many pointwise different choice functions of

min(domain(S)) ∩ {f | R(f ) = 1} are in {f | S(f ) = 1}

The NP-part is not as easily interpretable in the operator position as a rela-

tive clause is because the NP-part denotes a predicate of individuals, not of choice

258
functions. This mismatch in semantic types seems to arise naturally from the con-

siderations above. Namely, the discussion of Engdahl’s example (10) above showed

that derived predicates with lexical material in an internal position must be, at least

in some cases, be interpreted with a variable of type other than that of individuals.

Hence, they denote a predicate of something other then individuals. The NP-part,

on the other hand, must be a predicate of individuals since it can occur as an NP, for

example, in a copular construction like (37).

(37) John is a student.

At this point, there doesn’t seem to be an elegant way to let the NP-part in the

operator position contribute to the interpretation of the chain. One way of making

it interpretable is to assume that a predicate of individuals can be converted into a

(trivial) predicate of choice functions by the type shifting rule in (38). Then, the NP-

part in the operator position can be interpreted in the same way the relative clause

was interpreted.

(38) pet −→ λf hetie (f (p) = f (p))

The type-mismatch problem arises not only with the NP-part, which on the approach

taken here seems to be redundant in the operator position anyway, but also for those

relative clauses that don’t contain an internal head. Recall that for examples like (39a)

(repeated from (34) on page 50), the NP-part of the moved phrase has to reconstruct

because it contains a bound variable, but the relative clause has to be represented in

259
the operator position because otherwise (39a) would violate Condition C. As argued

in section 2.4, the fact that matching in a matching relative clause applies at LF

predicts for (39a) the LF-representation in (39b) where the NP-part isn’t represented

in the relative clause internal trace position. But then, the null hypothesis is that the

relative clause in (39b) is interpreted as a predicate of individuals. If this is right, the

type-shifting rule in (38) must apply for (39b) to be interpretable.

(39) a. [Which paper of hisk that Maryj was given]i did shej tell every studentk to

revise ti ?
h i
b. Which [λz Maryj was given [z]] λx did shej tell every studenti to revise

[x, paper of hisi ]?

A type-shifting rule like (38) is also required in the other direction, because

just like the NP-part isn’t interpretable as a predicate of individuals in the operator

position, a relative clause isn’t interpretable in the trace position as a predicate of

choice functions. One argument for this are examples like (40a) where the relative

clause contains a bound pronoun. The LF-representation predicted for (40a) is given

in (40b). The relative clause in (40b) is predicted to be interpreted as a predicate

of choice functions that is true if the value for the property paper is a paper that

relevant student hej read. In (40b), however, the interpretation of the relative clause

must be a property of individual, so that it can serve as the argument of the choice

function variable x. Hence, I assume that some kind of type-shifting can apply, like

the rule given in (41). For the relative clause in (40b) the rule in (41) results in a

260
predicate that’s true of a paper if hej read it, which is appropriate.

(40) a. [Which paper that hej read]i did every studentj like ti ?

b. [which paper] λx every studentj liked [x, paper λy hej read [y, paper]]

(41) If ∀f ∈ domain(P ) ∀p ∈ domain(f ): p(x) and domain(P ) 6= ∅,

P hhetieit −→ λx (P (fx ) = 1), where fx any f ∈ domain(P ) with ∃p: f (p) = x

With the type-shifting rule (41), the predicate that is the argument of which paper

in (40b) requires that the choice function x be defined for the properties paper that

he likes for each of the students. However, the NP-part, in the operator position is

interpreted as a predicate of choice functions that requires that they are defined for

the property paper. It’s easy to see the interpretation assigned to (41b) is nevertheless

correct. But, the fact that sometimes the content of traces that in some sense belong

to the same operator have different lexical content, does give rise to a problem in the

following.

Finally, material interpreted in intermediate traces requires the type-shifting

just mentioned, but also raises the other problem just hinted at. Consider the example

in (42a), repeated from (31) on page 48, and its LF-representation in (42b). The

intermediate trace of the chain in (42b) contains the relative clause and the NP-part

while the lowest trace and the operator position of the chain contain only the NP-part.

Notice that the higher part of the chain in (42b), the operator and the intermediate

trace, resembles the chain in (40), and the same interpretation procedure can apply.

But, how does the predicate created by λy that contains the lowest trace, contribute

261
to the interpretation?

(42) a. [Which paper that hek gave to Maryj ]i did every studentk think t0i that shej

would like ti ?

b. [Which paper] λx every studenti think [x, paper, λz hei gave [z, paper] to Maryj ]
| {z } | {z }
operator intermediate trace
λy shej would like [y, paper]
| {z }
lowest trace

For the moment, I pursue an approach in line with the structure given in (42b). Then,

the natural proposal seems to be to apply the λy-predicate to the choice function x.

But, this predicts the wrong interpretation for (42b): The denotation of the trace in

the λy-predicate is the value the choice function y assigns to the property paper. But,

the denotation of the trace in the intermediate position is the value x assigns to the

property paper hei gave to Mary. Now, consider a situation where there are papers

that every student gave to Mary, but no student thinks Mary would like the paper he

gave her. Rather, each student thinks Mary would like the paper P , which nobody

gave her. Intuitively, (41) should have no correct answer in such a situation. But, for

a choice function that assigns to the property paper the value P , the λx-predicate in

(42b) is true.

The problem with (42b), on the above account of how it’s interpreted, is

that there is no relationship as what the choice functions quantified over assign to the

lowest trace that contains just the NP-part and to the intermediate trace that contains

the NP-part and the relative clause. To remedy this problem, the λy-predicate in

(42b) must apply not to the choice function x, but to the result of applying x to the

262
content of the intermediate trace. Since, this is an individual, another type shifting

rule is needed. Namely, one like (43), which assigns to an individual x a choice

function that chooses x for all properties that of x.

(43) ζ e −→ xζ where xζ is the choice function with

domain(xζ ) = {pet | p(ζ) = 1} and x(p) = ζ for all p

5.2 Predictions of the Approach

The main prediction of the approach to the interpretation of quantifiers developed

in the previous section is, of course, that it assigns the right interpretation to all

the structures that were hypothesized in chapters 2 and 3. The approach makes

two further predictions which are worth mentioning. The first prediction concern

weak crossover effects: I show that the bijection principle of Koopman and Sportiche

(1982), which is one well-known generalization about weak crossover effects, follows

from the proposal of the previous section. The predictions stems from the fact that

chains with lexical content in the trace position were seen to involve a variable of

a higher type than that of individuals. This predicts effectively that A-bar chains,

which generally do have lexical content in the trace position, and A-chains, which

don’t, differ with respect to the type of the variable involved. It seems natural to

relate the possibility to bind a pronoun to this different in type. The other prediction

I point out below concerns the question what a chain with lexical content in the trace

position can be headed by.

The account of weak crossover effects is predicted in the following way. A

263
consequence of the system developed in the previous section is that all dependencies

where the trace position has lexical content involve binding of a variable of a type

other than the type e of individuals. It’s shown above that the higher type is required

in case the lexical content of the restrictor contains a pronoun that is bound only the

trace position, as in examples like (44) (repeated from (4)) for which Engdahl’s (1980)

choice function proposal was adopted. For all other chains with lexical material in

the trace position the motivation to use a higher type was also seen to provide an

account for the the appearance of lexical material in the trace position, and therefore

renders all other potential accounts using a lower type superfluous.

(44) Which friend of heri ’s did every studenti invite?

I believe that the proposed higher type is corroborated by the following analysis of

weak crossover effects. Weak crossover effects are cases where the a moved DP cannot

bind a pronoun from its derived position. Consider, for example, the contrasts in (45)

and (46). Only the a)-examples, where the trace position of the wh-word c-commands

the pronoun his allow binding.

(45) a. Whoi fed hisi dog? (Wasow 1972:135)

b.∗? Who
W i was hisi dog fed by?

(46) a. Which boyi received a postcard from hisi sister?

b.?? Which
W boyi did hisi sister send a postcard to?

I show now that the weak crossover effects in (45) and (46) is in fact predicted by

264
the approach outlined in the previous section. Recall from section 2.3 that A-bar

movement chains require that the NP-part is present in the trace position. Therefore,

the examples in (45) and (46) have LF-representations where the trace position has

lexical content. For example, (47) shows the LF-representation for (46b). In (47),

the variable f must range over choice functions for the chain to be interpretable. But

then, it seems plausible that the operator binding this choice function variable cannot

also bind a pronoun, since pronouns are plausibly of the type e of individuals.

??
(47) [
[Which boy] λf did hisf sister send a postcard to [f , boy]

The remaining question for the account of (45) and (46) is why binding of the pronoun

by the wh-quantifier seems possible in (45a) and (46a). Recall, though, from the

discussion of (85) on page (85) that a special mechanism must be postulated to

explain why a DP that doesn’t seem to have moved and therefore is not by virtue

of this movement the argument of a λ-predicate can nevertheless act as a binder, a

problem that Heim and Kratzer (1998) also point out for the predicates approach.

Whatever is the answer to this question, will allow the subject trace in (45a) and

(46a) to act as a binder. Specifically the mechanism allowing the binding could be a

short, string-vacuous A-movement step preceding the A-bar movement.

A-movement is predicted to obviate weak crossover because it doesn’t require

that the NP-part be represented in the trace position. That A-movement does obviate

weak crossover is, of course, well known, and illustrated here by (48).

(48) Which girli seemed to heri brother to be a good player.

265
 Therefore, this account of weak cross over predicts that a pronoun can only

be bound if it’s c-command by a trace in an A-position. This is, in effect, equivalent

to the bijection principle of Koopman and Sportiche (1982), since the lowest trace of

any NP occupies an A-position. It should be mentioned, though, that the account

also inherits all potential problems that of the bijection principle (see Safir 1984).

The second prediction of the approach developed in section 5.1 is a restriction

to essentially chains headed by a quantificational DP. This restriction seems to maybe

reflect the particular data considered in section there; namely, data involving DP-

chains headed by a quantificational determiner. While this is possible, I present one

argument in section 6.1 that the restriction is in some sense real.

5.3 Summary

In this chapter, I provided interpretation rules that assign the right interpretation to

all the examples of the previous chapters. The main tenet of the system laid out in this

section was that all determiner phrases involve the same interpretation principles. It

was shown that this assumption lead to the account of Engdahl (1980), which involves

a variable ranging over choice function to express the semantic dependency in a chain.

Engdahl’s proposal, which was originally only intended for interrogative quan-

tifiers, which all have existential force, is shown to raise problems when it’s carried

over to cardinal non-interrogative quantifiers. Both problems can be solved, however,

if it’s assumed that the lexical entries of quantifiers are such that two choice functions

quantified are only considered different it they are pointwise different.

266
 The choice function proposal developed in 5.1 was seen to predict the effect of

the weak crossover condition.

267
268
Chapter 6

Conclusion/Outlook

This conclusion doesn’t provide a summary of what was accomplished in the preceding

chapters. An overview of the thesis is given in section 1.2. Rather, it contains

two tentative remarks concerning the completeness of the account presented in the

previous chapters for the interpretation of chains.

My aim is here to make the claim plausible that the account of the syntax-

semantics interface developed in this thesis covers all cases of chains that arise. The

two sections look at the two cases that, at first, seem to show that this completeness

claim is wrong assuming. I present, in each case, an analysis that is compatible

with the completeness claim and then argue with new facts that this new analysis

is superior to the first analysis that isn’t compatible with completeness. This result

constitutes the strongest support for the exhaustiveness claim that seems possible. It

is, obviously, always possible that the system developed proves incomplete in other

respects, either ones I overlooked or ones that are discovered in the future.

The restriction of the account discussed in section 6.1 is was pointed out at

269
the end of section 5.2: The syntactic and semantic rules presented consider only the

case of a chain headed by a quantificational DP and at least the mechanism that

interprets chains with lexical content in the trace position, cannot straightforwardly

account for any other case. Section 6.1 summarizes one argument and presents a

second argument that only traces of type e (and maybe other non-functional types

like t) arise at the level of logical form. This result is actually even stronger than the

restriction just mentioned, since the interpretation of chains where the trace is of a

higher type than e but has no lexical content seems possible. Therefore, the result

implies that that the restriction of the account addressed is unproblematic.

Section 6.2 concerns the restriction of the account to cases of a DP-chain where

the quantificational determiner is interpreted in the head position of the chain. I show

first that this restriction does actually make the account of the syntax-semantics inter-

face properties of chains easier. I then develop a new account for scope reconstruction

phenomena that assumes that in scope reconstruction cases movement is actually not

seen by interpretation at all, but takes place in the PF-branch of grammar. This ac-

count is seen to predict a generalization about the availability of scope reconstruction

that is otherwise unexplained.

6.1 The Type of Traces

This section argues that the type of the trace position of a chain must be the type of

individuals e.1 If this claim is correct, it entails that only the two types of variables

1
With respect to type of truth values t, there is to my knowledge no evidence that t is different
from e, and the distinction drawn between the two types seems to be merely for expository purposes.

270
inside a trace that are proposed in chapter 5 arise. Namely, the type of individuals

if the trace doesn’t have any lexical content and the type of choice functions if the

trace has lexical content.

A restriction on the types of traces is first proposed by Heycock (1995), Beck

(1996) and Fox (1998b). While the proposals and the evidence differ, all three present

evidence only for the existence of traces of type e. Heycock (1995), in effect, proposes

a restriction to type e. The evidence Fox (1998b) uses to argue for the restriction,

for example, involves an interaction between scope reconstruction of A-moved quan-

tifiers and Condition C. He points out that in examples like (1) scope reconstruction

is blocked (see also 1997, and Sportiche 1996). This correlation is unexpected, if the

type of the variable corresponding to the A-trace could be the type of generalized

quantifiers he, eti because this type achieves the effect of narrow scope while syntac-

tically representing the A-moved quantifier in the higher position (von Stechow 1993,

Cresti 1995, Rullmann 1995, Chierchia 1995). In contrast, the interaction between

Condition C and scope reconstruction is predicted if the trace position can only cor-

respond to a variable of type e. Then, the moved quantifier must be syntactically

represented in the trace position for narrow scope. Therefore, the interaction between

Condition C and scope reconstruction argues for a restriction on the type of traces.

The data in section 6.2, where scope reconstruction is discussed, provide another ar-

gument to assume that to achieve scope reconstruction by means of a higher type

trace must be blocked.

The same seems to be true for degrees.

271
(1) A student of Davidi ’s seems to himi to be at the party. (∃ À seem, seem À

∃) (Fox 1998a:(46a))

In this section, I present further empirical evidence for a restriction of the

type of traces to the type e from quantifier float in Japanese. In this construction,

there’s a trace position associated with the moved nominal phrase in the complement

position of the numeral quantifier. I claim that the type of this complement position

can be either e or et resulting in two distinct interpretations, and show that the

interpretations associated with the higher type et require that the moved nominal

phrase be interpreted entirely in the complement position of the quantifier.

The argument relies on a new account of the partitive/cardinal ambiguity

found with cardinal floating quantifiers in Japanese (Kitagawa and Kuroda 1992,

Ishii 1997). Ishii (1997) observes that if the direct object that a floating quantifier is

associated with occupies a VP-adjoined position, only the partitive interpretation is

available: example (2) is infelicitous in a situation where only three books are salient.

(2) John-wa [urenokotta hon-o]i Mary-ni [t1 san-satu] ageta

JohnTOP left unsold booksACC MaryDAT three-CL gave

‘John gave Mary three (of the) unsold books.’ (partitive, ∗ cardinal)

Ishii (1997) also notes that examples like (3), where the nominal phrase associated

with the floated quantifier occupies an IP-adjoined position, allow both a partitive

and a cardinal interpretation. (Actually, it seems impossible to assess the presence of

the partitive interpretation in examples like (3) if a cardinal interpretation is available,

272
since the former entails the latter.)

(3) [Urenokotta hon-o]i John-wa Mary-ni [t1 san-satu] ageta

left unsold booksACC JohnTOP MaryDAT three-CL gave

‘John gave Mary three (of the) unsold books.’ (partitive, cardinal)

I claim that the cardinal interpretation of (3) requires reconstruction of the

nominal phrase urenokotta hon-o to a the complement position of the quantifier san-

satu. In support of this claim, I show that the availability of the cardinal reading

correlates with the availability of reconstruction in two cases. The first is (4). In

contrast to (3), (4) doesn’t allow a cardinal interpretation. Since reconstruction is

blocked by Condition C in (4), reconstruction is required for a cardinal interpretation.

(4) [Mary-gaj sukina hon-o]i John-wa kanozyo-nij [t1 san-satu] ageta

MaryNOM likes booksACC JohnTOP herDAT three-Cl gave

‘John gave Mary three of the books she liked.’ (partitive, ∗ cardinal)

The second correlation between the availability of reconstruction is involves a paral-

lelism of the data in (5) to the contrast between (2) and (3). Saito (1992) shows that

scrambling to a VP-adjoined position cannot reconstruct for anaphor binding, while

IP-adjoined scrambling can. The contrast in (5) shows that Saito’s observation also

holds for scrambling that strands a floated quantifier in the base position.

(5) a. John-ga [Hanako-to Mary-ni]i otagaii -no hon-o ni-satu ageta

JohnNOM Hanako-and Mary-to each otherGEN bookACC two-Cl gave

‘John gave Hanako and Mary two books of each other’s.’

273
 b. ∗ John-ga
J [otagaii -no hon-o]j [Hanako-to Mary-ni]i [t2 ni-satu] ageta
JohnNOM each otherGEN bookACC Hanako-and MaryDAT two-Cl gave

c. [otagaii -no hon-o]j John-ga [Hanako-to Mary-ni]i [t2 ni-satu] ageta

each otherGEN bookACC JohnNOM Hanako-and Mary-to two-Cl gave

Examples (2) and (3) showed that only scrambling to an IP-adjoined position allows

a cardinal interpretation. Hence, the availability of reconstruction (in (5)) again

correlates with the availability of the cardinal interpretation. I conclude that the

cardinal interpretation requires reconstruction.

At this point, the generalization is the following: If the sister of the floated

quantifier is a trace at LF, only the partitive reading is available. How does this

generalization relate to the type of the trace. I claim that the cardinal interpretation

of a numeral requires a complement of type he, ti, the type of first order properties.

The partitive interpretation of a numeral, on the other hand, takes a complement of

type e, the type of individuals. This is suggested by the English examples in (6):

(6) three |books

{z } vs. three of the
| books
{z }
he, ti e

If we assume that, in Japanese as well, the difference between the cardinal and the

partitive interpretation is represented by the type of the complement of the (floated)

quantifier, the generalization I arrived at above follows from the restriction of traces

to be of type e straightforwardly: If, at LF, the trace that’s the sister of the float-

ing quantifier is visible at LF, it must be interpreted as of type e and, therefore,

only the partitive interpretation is available. Hence, the Japanese facts support the

generalization that traces must be of type e.

274
 An alternative explanation of the above generalization would be the following.

Assume that floating quantifier constructions can either be generated by movement

or base-generated with a pronominal element pro occupying the complement position

of the quantifier (Kitagawa and Kuroda 1992). Furthermore, assume that when the

complement of the quantifier is pro, only the partitive interpretation is available, and

the if Q-float is generated by movement, this movement must reconstruct. Then the

facts above follow, without appeal to the condition on the type of traces. But, the

following facts argue that Q-float must always be generated by movement (see also

Miyagawa 1989). The argument is based on the ungrammaticality of (7), where the

floating quantifier is contained in a fronted VP, while the associated NP is stranded.

∗
(7) [
[Mary-ni [t1 san-satu] age-sae]j [Urenokotta hon-o]i John-wa t2 sita
MaryDAT three-CL give-even left unsold booksACC JohnTOP did

Notice that material stranded by VP-fronting can bind a variable in the fronted VP,

as shown in (8). Therefore, the ungrammaticality of (7) argues that the relation

of the associate of the floated quantifier and quantifier is not just one of binding,

but one derived by movement. Under this assumption, the ungrammaticality of (7)

follows from the proper binding condition (or recent proposals to derive proper binding

condition effects from shortest attract; Takano 1993, Kitahara 1994, Yatsushiro 1997).

(8) [Mary-ni zibuni -no hon-o san-satu age-sae]j daremoi -ga t2 sita
MaryDAT selfGEN bookACC three-CL give-even everybodyNOM did

‘Give three books of his to Mary, everybody did.’

I conclude that trace may only be of type e. Hence, there are only two types

275
possible for the variable in a chain: e if the trace has no lexical content and (et)e, the

type of choice functions if the trace has lexical content.

6.2 Scope (or Total) Reconstruction

The restriction of the interpretation procedure for chains considered in this section

is that the moved quantificational determiner must be interpreted in the operator

position of the chain. I have made this assumption throughout and will argue below

that it allows the account to be simpler than otherwise possible. The main case where

the assumption seems to be wrong are cases of scope reconstruction like (9) under

the interpretation where two takes scope below likely (see (11) below).

(9) [Two people from New York]i are likely to ti win the lottery next weekend.

That the quantification determiner of a moved DP must be interpreted in the higher

of position follows from an assumption argued for in section 4.2. There I argued that

the sister of a moved constituent is always interpreted as a λ-predicate. This implies

that at least parts of the moved constituent are interpreted in the derived position,

since otherwise this λ-predicate wouldn’t have any argument. But, if any part of

a moved chain is interpreted in the operator position the determiner head must be

interpreted there. Hence, it follows that at least the D-head of a moved DP is always

interpreted in the derived position.2

2
This argument is weakened, however, that the facts below argue that, for example for VP-
fronting, an interpretive mechanism must apply where all material is interpreted in the trace position.

276
 This chapter argues for a new analysis of scope reconstruction phenomena in

A-chains. Specifically, I mean by scope reconstruction cases where all material of

the moved phrase seems to contribute to interpretation only in the trace position.

Saito (1992), for example, uses the term total reconstruction for such cases. The type

of scope reconstruction seems to be mainly available with A-movement. I propose

that A-movement can take place in the PF-branch of the derivation and therefore

be not noticed at the LF-interface. The argument for this proposal is based on the

unavailability of reconstruction when the A-moved element doesn’t c-commands its

trace in the overt form as first observed by Barss (1986). As I show, only the PF-

movement analysis straightforwardly predicts this restriction on reconstruction.

In the remainder of the introduction, I summarize test for whether scope re-

constructions is possible in A-chains. May (1977) first notice scope reconstructions

in A-chains looking at quantifier scope. He observed that in raising constructions

the raised subject is scopally ambiguous with respect to a scope bearing element

that intervenes between the trace of the raised subject and its overt position. This

is illustrated by the examples (10a) and (11a), where the two different readings are

paraphrased in b. and c. In (10a), the wide scope reading (paraphrase (10b)) is

salient because our world knowledge about skiing competitions at the Olympic games

tells us that the possibility of there being two gold medal winners in one compe-

tition is vanishingly small. In (11a), on the other hand, the narrow scope reading

paraphrased in (11c) is the only one compatible with our world knowledge that, in

a lottery, it’s never the case that a particular individual has more than a very small

chance of winning.

277
(10) a. [Two Germans]i are likely to ti win the Gold Medal in this skiing race.

b. Two Germans have a good chance of winning. (two À likely)

c.#There
T is a good chance that two Germans will win. (likely À two)

(11) a. [Two people from New York]i are likely to ti win the lottery next weekend.

T New Yorkers have a good chance of winning. (two À likely)

b.#Two

c. There is a good chance that two New Yorkers will win. (likely À two)

These following three tests for scope reconstruction which rely on grammaticality

judgments are used below. The first of test uses negative polarity licensing in addi-

tion to Scope as a test for the scope reconstruction (Linebarger 1980, 1987). As is well

known, a negative polarity item (henceforth NPI) must be c-commanded by negation

or a downward entailing operator. What Linebarger shows is that the scope recon-

struction of an A-chain can feed NPI-licensing. This is illustrated in (12). Neither in

(12a) nor in (12b) does the negation c-command the NPI anything in the overt form.

Nevertheless, the NPI in (12a) can be licensed and the NPI-licensing seems to force a

scopal construal where negation takes scope over the subject. Given that there is an

A-trace of the subject below negation, it seems reasonable to assume that the scope

reconstruction of the subject A-chain feeds NPI-licensing in (12a).

(12) a. [A doctor who knows anything about acupuncture]i isn’t ti available.

b. ∗ [A
[ doctor who knows anything about acupuncture]i is ti available.

278
In raising constructions as well, the narrow scope interpretation can feed NPI-licensing.

This is illustrated in (13a), which contrast with example (13b), where there’s no nega-

tion c-commanding the A-trace, as well as with (13c), where negation is present, but

the A-trace is not c-commanded by it.

(13) a. [A doctor with any reputation]i is likely not to be ti available.

b. ∗ [A
[ doctor with any reputation]i is likely to be ti available.

c. ∗ [A
[ doctor with any reputation]i is ti anxious for John not to be available.

A second test for the availability of a narrow scope interpretation using gram-

maticality was discovered by Burzio (1986).3 It uses binomial each as test. The

contrast between (14a) and (14b) shows that normally binomial each must be c-

commanded by a distributive noun phrase in the overt form.

(14) a. The athletes demanded one translator each.

b. ∗ One
O translator each welcomed the athletes.

As Burzio notes, there’s one exception to this generalization which is illustrated in

(15): Binomial each attached to the direct object can be licensed by a distributive to-

phrase and, as Safir and Stowell (1987) point out certain other prepositional phrases.4

(15) The Olympic Committee assigned one translator each to the athletes.

3
Richard Kayne (p.c.) first drew my attention to Burzio’s work.
4
As David Pesetsky (p.c.) pointed out to me, licensing of direct object binomial each by the
following PP might itself involve a scope reconstruction of an A-chain, assuming the direct object
moved from a position below the goal-PP to its surface position. See Pesetsky (1994:221) and
footnote 11 on page 58 for corroborating data.

279
For our purposes, Burzio’s most important observation is that the scope reconstruc-

tion in an A-chain can feed each-licensing in the pre-PP position before a prepositional

phrase. This is shown in (16) for A-movement in passives, and in (17) for two-step

A-movement, one step being movement to the subject position of a passive and the

second step being movement to the subject position of a raising construction.

(16) a. [One translator each]i was assigned ti to the athletes.

b. ∗ [One
[ translator each] gave a speech to the athletes.

(17) a. [One translator each]i is likely to ti be assigned ti to the athletes.

b. ∗ [One
[ translator each]i is likely to ti give a speech to the athletes.

6.2.1 A PF-movement Account of Scope Reconstruction

The existence of a scope reconstruction in A-chains being established, consider the

proposals that have been made to derive the narrow scope interpretation. The three

proposals I know of are LF-lowering (May 1977, 1985, Chomsky 1995), the Copy The-

ory of movement (Wasow 1972:139, Burzio 1986, Chomsky 1993, Hornstein 1995) and

Semantic Reconstruction (von Stechow 1993, Cresti 1995, Rullmann 1995, Chierchia

1995). I don’t have room to summarize these proposals here in detail—in a nutshell,

LF-lowering assumes that covert movement doesn’t have to be to a c-commanding

position in the tree and thereby can undo the effect of overt raising. The Copy

Theory of movement assumes instead that a full copy of the moved phrase is left in

the trace position and the interpretive component of grammar can look at this lower

copy rather than the higher one. Semantic Reconstruction, finally, assumes that the

280
semantics of an A-chain dependency can optionally be of a higher semantic type,

which leads to a scope reconstruction. All three proposals have in common that they

assume that overt A-movement is followed by an invisible undoing operation as the

traditional term ‘scope reconstruction’ for the scope reconstruction suggests. I, as

already mentioned, believe that the term ‘reconstruction’ is misleading and propose

that no undoing of movement is necessary. Rather, I propose that A-movement in

the cases of a narrow interpretation, is not seen by the interpretive component of

grammar because it takes place in the PF-branch of grammar.

The question that needs to be answered by any account of narrow interpreta-

tion phenomena in A-chains is the following: What is the derivation of the PF-LF-pair

in (18)?5

(18) PF: Two people are likely to win the lottery

LF: are likely to two people win the lottery

My answer to this question relies on the T-model of grammar (also sometimes called

the Y-model or inverted Y-model) of Chomsky and Lasnik (1977), which I assume

here in the form given in Chomsky (1995). The T-model embodies three partially

interrelated assumptions: One, it assumes that complex representations are built up

and modified by simple operations, generalized transformations, which are inherently

ordered. Two, it assumes that operations can apply either having an effect on both

LF and PF, or their effect can be limited to only LF, or only to PF. Three, there is

link between the ordering of operations and where they have an effect: namely the

5
The PF-representations here and in the following are given in the form before real phonology
has applied.

281
operations that are visible only to one of LF or PF follow operations that are visible

to both. All three assumptions together have the consequence that LF-PF pairs are

derived by a partially ordered set of transformations that has the graphical shape

shown in (19).6

(19) T-model (Chomsky and Lasnik 1977)

Stem

Split
@
LF @
PF
@
R
@
LF-Interface PF-Interface

For the moment, assumption the of the T-model, that an operation can have an

effect at only one of the interfaces if it applies in one of the branches, is what we

need. This allows us to analyze LF-PF mismatches as operations that apply in one

of the branches. It seems natural to propose that (18) is derived by PF-movement

of two people from the embedded subject position. This is the derivation I propose

generally derives scope reconstructions in A-chains. In other words, I propose that

A-movement in general can optionally take place in the PF-branch of the derivation,

instead of taking place in the stem. For an illustration, consider (20).

6
The T-model incorporates an additional assumption, namely that operations which take more
than one simple representations as input (in Chomsky 1995 the only operation of this type is Merge)
must have an effect at both LF and PF. This assumption derives that there is exactly one Split
point in the derivation of one LF-PF pair and that the branch segments of any derivation are totally
ordered. This assumption, however, is not important for anything I’ll say in the following.

282
(20) [Two people]x are likely to tx win the lottery.

The proposal is that (20) has two possible derivations. In one derivation raising of

two people takes place in the stem and therefore the result of raising is visible to both

LF and PF. This derivation therefore gives rise to wide scope of two people over likely.

Crucially, I assume that there’s no way raising in the stem can be covertly undone.

So, this derivation yields only the wide scope interpretation. The second derivation

is one where raising of two people is takes place at PF, and its application is visible

only to PF, but not to LF. This derivation leads to a narrow scope interpretation of

two people below likely, because raising is not seen by the LF-interface.7

One immediate ramification the PF-movement proposal makes concerns the

level at which it is verified that obligatory overt movements have indeed taken place.

The PF-movement approach is incompatible with the view taken for example by

Chomsky (1995) that this verification only takes place at LF. Rather, PF must be

the level where the verification takes place for overt movement. At least, the mor-

phological requirement that triggers raising in (20)—the EPP-feature if current work

on the topic is to be believed—must be checked at PF. This consequence however, as

far as I can see, doesn’t cause any new problems; on the contrary, it now follows that

the EPP must universally be satisfied overtly (cf. Chomsky 1995).

7
It is technically conceivable, that in a derivation where raising is delayed until PF, quantifier
raising applies in the LF-branch to bring about the wide scope interpretation. I don’t have any
evidence bearing on this possibility.

283
6.2.2 The Scope Freezing Generalization

In this section, I show that the PF-movement approach predicts a generalization

Barss (1986) first hinted at regarding the availability of scope reconstructions, and

then argue that the generalization is indeed true. This generalization is the following

Scope Freezing Generalization (SFG).

(21) SFG: A moved quantifier QP cannot be interpreted in an A-trace position, if

the trace isn’t c-commanded by the overt position of QP.

The SFG blocks a scope reconstruction in cases where the trace left by A-movement

is inside a a constituent that subsequently undergoes movement itself. One such case

is example (22) from Barss (1986), who based solely on (22) suggests an analysis that

would account for the SFG. I address Barss’s account of the SFG at the end of this

section.

(22) [How likely to tQP address every rally]wh is [some politician]QP twh ? (someÀlikely,

∗
likelyÀsome)

Barss (1986) presents only the example (22) in support of the SFG. In the following, I

present some more. The kind of constructions that are relevant to testing the SFG are

ones where subsequent A-bar movement destroys the c-command relationship between

an A moved phrase and its trace, as it happened in (22). Example (23a) shows that

such a stranded A-moved phrase is capable of taking scope below a c-commanding

quantifier. (23b) shows that the stranded phrase can also take scope below a c-

284
commanding likely. Therefore, the lack of narrow scope of the stranded phrase in

(22) and the examples in the following must be due to the lack of c-command.

(23) a. [Every journalist]∀ asked [how likely to t∃ address every rally]wh [some

politician]∃ is twh . (∀À∃, ∃À∀)

b. John is likely1 to find out [how likely2 to t∃ address every rally]wh [some

politician]∃ is twh . (likely1 À ∃, ∃À likely1 )

The judgment in Barss’s (1986) example can be sharpened by using each-

licensing as introduced above as a test. As we see in (24), Barss’s judgment now

shows up as a grammaticality contrast: (24a) shows again that the scope reconstruc-

tion can feed each licensing. In (24b), where the SFG correctly blocks the narrow

interpretation, each cannot be licensed. (24c), on the other hand, without each is

grammatical, but it only has the reading with scope of one over likely.

(24) a. [One translator each]QP is likely to be assigned tQP to the athletes.

b. ∗ [How
[ likely to be assigned tQP to the athletes]wh is [one translator each]QP

twh ?

c. [How likely to be assigned tQP to the athletes]wh is [one translator]QP twh ?

In (22) and (24) it was wh-movement that destroyed the c-command relationship

between the A-moved QP and its trace. The contrasts in (25) and (26) show that

other types of A-bar movement, namely topicalization in (25) and though-raising in

285
(26), have the same effect.8

(25) a. ∗ . . . and [likely to be assigned tQP to the athletes]top [one translator each]QP

is ttop .

b. . . . and [likely to be assigned tQP to the athletes]top [one translator]QP is

ttop .

(26) a. ∗ [Likely
[ to be assigned tQP to the athletes]tr though [one translator each]QP

is ttr , there were still complaints.

b. [Likely to be assigned tQP to the athletes]tr though [one translator each]QP

is ttr , there were still complaints.

While in questions NPIs are independently licensed, with topicalization and

though raising we can also use NPI-licensing as a test for the availability of a scope

reconstruction. As the data in (27) and (28) show, the result from NPI-licensing

confirms the each-licensing data.

(27) a. ∗ . . . and [certain to be not tQP available]top , [a doctor with any reputation]QP

was ttop .

b. . . . and [certain to be not tQP available]top , [a doctor from cardiology]QP

was ttop .

8
Examples of VP-fronting like (25) are best if they are preceded by the same sentence with a
non-fronted VP as in (i). The dots preceding all examples of VP-topicalization serve as a reminder
to look at them in such a context.
(i) Martin said that one translator (each) is likely to be assigned to the athletes and, likely to
be assigned to the athletes, one translator (∗ each) is.

286
(28) a. ∗ [Certain
[ to be not tQP available]tr though [a doctor with any reputation]QP

is ttr , patients were waiting.

b. [Certain to be not tQP available]tr though [a doctor from cardiology]QP is

ttr , patients were waiting.

A-movement of subjects has been argued to take place not only in raising

constructions, but also with all other subjects from the VP-internal underlying subject

position to the EPP-position. Hornstein (1995) and Johnson and Tomioka (1997)

argue that inverse scope of the object over the subject in English transitive clauses

requires a scope reconstruction of the subject chain from the VP-internal subject

position to its overt position. Therefore, the SFG predicts that A-bar movement of

the VP will block inverse scope in transitive clauses. In fact, this prediction seems to

be a well-known fact (Fox, p.c. referring to Truckenbrodt, p.c.), though I don’t know

who first made this observation nor whether this has ever been made in print. The

contrasts in (29) and (30) show the prediction. While (29a) and (30a) allow inverse

scope, this interpretation is not available in (29b) and (30b).

(29) a. . . . and [a policeman]QP tQP stood in front of every bank. (∀À∃, ∃À∀)

b. . . . and [tQP stand in front of every bank]top [a policeman]QP did ttop . (∀À∃,

∗
∃À∀)

(30) a. Though [enough of us]QP were tQP defending every gate, the enemy broke

through. (enoughÀ∀, ∀Àenough)

287
 b. [tQP Defending every gate]tr , though [enough of us]QP were ttop , the enemy

broke through. (enoughÀ∀, ∗ ∀Àenough)

In sum, the SFG seems be corroborated by a number of tests. I show now that

the SFG is a consequence of the PF-movement analysis of narrow scope phenomena in

conjunction with the three assumptions in (31). Each of these additional assumptions

are independently motivated. I assume that assumption (31a), that wh-movement

and other types of A-bar movement take place in the stem, follows from the nature of

A-bar movement. The c-command condition on movement in (31b) could follow from

a better understanding of movement as discussed by Chomsky (1995). I also discuss

this assumption in section below in the context of a quantifier lowering analysis. Of

the T-architecture, I will make use of the order it imposes on the operations in a

derivation; specifically, that movement in the PF-branch takes place after movement

in the stem.

(31) a. Overt A-bar movement must take place in the stem.

b. c-command: Movement must target a position that c-commands the mov-

ing item. (Chomsky 1995)

c. T-architecture: PF-movement must take place later than stem movement.

(Chomsky and Lasnik 1977)

Consider now the derivations of a structure like (30) that would lead to narrow and

wide scope respectively. First, look at a potential derivation for narrow scope in (32).

288
For narrow scope, raising must be delayed until PF. But, wh-movement must take

place in the stem by assumption (31a). Assuming the T-model, the derivation in (32)

is forced. It violates either the EPP or the c-command condition on movement.

(32) Failing Derivation for Narrow Scope

HHH
 HH
 HH
  H H H   H H
 H
QP   HH PF *  H   H -
HH  t QP
H
twh
6  H
QP A EPP 6
wh-mvmt. LFAAU *

For wide scope, on the other hand, EPP-raising takes place in the stem, and can

therefore precede wh-movement, as shown in (33). As the derivation in (33) shows,

the EPP can be satisfied without incurring a violation of the c-command condition.

(33) Derivation for Wide Scope

HH
H
  HH
HH 
H H
 H
 HH  HH
-  H PF
 tQP  *

QP  HH 6 QP  HH 
H H
6 tQP tQP A
EPP wh-mvmt. LFAAU

I conclude that the SFG is a consequence of the PF-movement approach to

narrow scope phenomena. Before I begin to consider alternatives views of scope

reconstruction that actually explain the SFG, note that both the copy theory as

well a semantic approaches to scope reconstruction phenomena in A chains seem to

offer no perspective in accounting for SFG in an insightful way. On either account

the operation bringing about the scope reconstruction is different from movement,

and therefore a sensitivity of this operation to c-command would have to stipulated.

289
The strength of the PF-movement account, in this respect, is that the c-command

sensitivity of movement is an independently argued for property of movement which

carries over to PF-movement.

Now, consider two alternative explanations that might be given for the SFG.

Note here, that the evidence for the SFG came entirely from examples with the

structure in (34).

A mvmt.
(34) H
H ? H
H
6
A-bar mvmt.

The first potential explanation of the SFG was brought to my attention by David

Pesetsky and Želko Bošković, and is based on two assumptions of Lasnik and Saito

(1992): One, the generalized proper binding condition (GPBC), that traces must not

be unbound at any point of the derivation, and two, the assumption that a control

analysis is optionally possible in all raising constructions. These two assumptions

force a control analysis for all examples that have the structure in (34) as Lasnik

and Saito (1992:140-42) point out. control structures generally don’t allow scope

reconstruction, Lasnik and Saito’s (1992) analysis of structures of type (34) predicts

the SFG.

However, the assumptions underlying this account are at best controversial.

As Takano (1993), Kitahara (1994), and Müller (1996) show, the GPBC is not correct

in the form suggested by Lasnik and Saito (1992) and a empirically more accurate

condition accounting for all data attributed to the GPBC follows from the general

economy condition shortest attract. But, this condition allows a raising analysis for

290
structure such as (34). Moreover, the assumption that a control analysis is possible for

raising structures misses some distinctions between the two: As Wurmbrand (1998)

points out, real control can be ‘imperfect’ as in (35a): The PRO can refer to a plural

entity that the subject is a member of. Raising on the other hand doesn’t allow

‘imperfect’ readings, as (35b) shows.

(35) a. The mayor decided to PROthey gather in the lobby.

b. ∗ The
T mayor was likely to PROthey gather in the lobby.

The second alternative account of the SFG is the analysis Barss (1986) gives

for the example (22). He relies on a Q-lowering analysis of scope reconstruction

phenomena in A-chains and proposes that the c-command condition on movement

cannot only be satisfied by the landing site c-commanding the origin site, but is also

satisfied if the origin site c-commands the landing site. This modified, symmetric,

c-command condition allows lowering, but only to a position that is c-commanded by

the origin site. Barss (1986) claims that his account blocks lowering in a structure like

(34) because the landing site inside the fronted constituent here isn’t c-commanded

by the origin site.

It is first not clear that Barss’s (1986) account actually predict the SFG.

Consider a derivation, where the position inside the fronted constituent is reached

by two steps of movement: The first step raises the raised subject to a position

above the fronted constituent, and the second step lowers the subject into the fronted

constituent. This derivation doesn’t violate Barss’s weakened c-command condition.

291
 Secondly, Barss’s (1986) account inherits the problems that Q-lowering has.

In particular, the absence of overt lowering will need to be explained, which is not

trivial in many cases: Consider e.g. Japanese scrambling: Saito (1992) shows that

in Japanese a wh-phrase can be scrambled to a position outside of its scope domain

as in (36a). He therefore argues that Japanese scrambling can be freely undone.

Nevertheless, it is still impossible to scramble a phrase to a lower position in Japanese

as the ungrammaticality of (36b) attests.

(36) a. dono hon-oi Masao-ga Hanako-ga ti tosyokan-kara karidasita ka

which book MasaoNOM HanakoNOM library-from checked-out Q

siritagatteiru
want-to-know

‘Masao wants to know which book Hanako checked out from the library.’

b. ∗ Hanako-ga
H ti Masao-ga Taro-nii waratta-to omowa-seta
HanakoNOM MasaoNOM TaroDAT laughed-that believe-made

Finally, it seems to me quite likely that a strict c-command condition on

movement could easily be derived as a consequence of more general principles of

syntactic derivations—an issue that has received a lot of attention in recent work (cf.

Chomsky 1995). The symmetric c-command condition of Barss (1986), on the other

hand, seems to be a mere stipulation at this point, and it would be more natural

to assume no such restriction on lowering. Therefore, I conclude that c-command is

a general property of all movement. But then, the PF-movement account of scope

reconstruction in A chains is the only account of the SFG left.

292
References

Abusch, Dorit. 1994. The scope of indefinites. Natural Language Semantics 2.83–
135.
Alexiadou, Artemis, and Elena Anagnastopoulou. 1997. Toward a uniform
account of scrambling and clitic doubling. In German: Syntactic Problems –
Problematic Syntax , ed. by Werner Abraham and Elly van Gelderen, 143–161.
Tübingen: Niemeyer.
Barss, Andrew. 1986. Chains and Anaphoric Dependence. On Reconstruction and
its Implications. Cambridge: Massachusetts Institute of Technology dissertation.
Beck, Sigrid. 1996. Wh-Constructions and Transparent Logical Form. Universität
Tübingen dissertation.
Beghelli, Filippo. 1993. A minimalist approach to quantifier scope. In Proceedings
of NELS 23 , 65–80, Amherst. GLSA.
——. 1995. The Phrase Structure of Quantification Scope. Los Angeles: University
of California dissertation.
Berman, Arlene, and Michael Szamosi. 1972. Observations on sentential stress.
Language 48.304–325.
Berman, Steve. 1991. On the Semantics and Logical Form of Wh-Clauses.
Amherst: University of Massachusetts dissertation.
Brame, Michael K. 1968. A new analysis of the relative clause: Evidence for an
interpretive theory. Unpublished Manuscript, MIT.
Bresnan, Joan. 1971. Sentence stress and syntactic transformations. Language
47.257–281.
——. 1972. Stress and syntax: A reply. Language 48.326–342.
Brody, Michael. 1979. Infinitivals, relative clauses, and deletion. Mimeograph,
University of London.
Büring, Daniel. 1995. The great scope inversion conspiracy. In Proceedings of
SALT 5 , ed. by Mandy Simons and Teresa Galloway, 37–53, Cornell. CLC Pub-
lications.
——. 1996. The 59th-Street Bridge Accent. Universität Tübingen dissertation.
——. 1998. Focus and topic in a complex model of discourse. Manuscript, Cologne
University, Germany.
Burzio, Luigi. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht:
Kluwer.
Carlson, Greg N. 1977. Amount relatives. Language 58.520–542.

293
Cheng, Lisa Lai-Shen. 1991. On the Typology of Wh-Questions. Cambridge:
Massachusetts Institute of Technology dissertation.
Chierchia, Gennaro. 1993. Questions with quantifiers. Natural Language Seman-
tics 1.181–234.
——. 1995. Dynamics of Meaning. University of Chicago Press.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax . Cambridge, Mas-
sachusetts: MIT Press.
——. 1972a. Some empirical issues in the theory of transformational grammar. In
Goals of Linguistic Theory, ed. by Stanley Peters, Englewood Cliffs, New Jersey:
Prentice Hall. (reprinted in Chomsky 1972b).
——. 1972b. Studies on Semantics in Generative Grammar . The Hague, The Nether-
lands: Mouton.
——. 1977. On wh-movement. In Formal Syntax , ed. by Peter Culicover, Tom Wasow,
and Adrian Akmajian, 71–132. New York: Academic Press.
——. 1981. Lectures on Government and Binding: The Pisa Lectures. Berlin: Mouton
de Gruyter.
——. 1986. Barriers. Cambridge, Massachusetts: MIT Press.
——. 1993. A minimalist program for linguistic theory. In The View from Building
20, Essays in Linguistics in Honor of Sylvain Bromberger , ed. by Ken Hale and
Jay Keyser, 1–52. MIT Press.
——. 1995. The Minimalist Program. Cambridge, Massachusetts: MIT Press.
——, and Howard Lasnik. 1977. Filters and control. Linguistic Inquiry 8.425–504.
Chung, Sandra, James McCloskey, and Bill Ladusaw. 1995. Sluicing and
logical form. Natural Language Semantics 3.239–282.
Church, Alonzo. 1932. A set of postulates for the foundation of logic. Annals of
Mathematics (2) 33.346–366.
——. 1933. A set of postulates for the foundation of logic (second paper). Annals of
Mathematics (2) 34.839–864.
Cinque, Giugliemo. 1993. A null theory of phrase and compound stress. Linguistic
Inquiry 24.239–297.
Cooper, Robin. 1983. Quantification and Syntactic Theory. Dordrecht, The
Netherlands: Reidel.
Cormack, Anabel. 1984. VP anaphora: Variables and scope. In Varieties of
Formal Semantics, ed. by Fred Landman and F. Veltman, Dordrecht: Reidel.
Cresti, Diana. 1995. Extraction and reconstruction. Natural Language Semantics
3.79–122.
——. 1997. Some considerations on wh-decomposition and unselective binding. In
Proceedings of the Tübingen Workshop on Reconstruction, 147–176. Universität
Tübingen, Germany.
Curry, Haskell B. 1930. Grundlagen der kombinatorischen Logik. American
Journal of Mathematics 52.509–536.
——, and Robert Feys. 1958. Combinatory Logic, Volume I . Amsterdam: North-
Holland.
Dayal, Veneeta. 1996. Locality in Wh Quantification. Dordrecht, The Netherlands:
Kluwer.

294
Déprez, Viviane. 1990. On the Typology of Syntactic Positions and the Nature
of Chains: Move α to the Specifier of Functional Projections. Cambridge: Mas-
sachusetts Institute of Technology dissertation.
Diesing, Molly. 1992. Indefinites. Cambridge, Massachusetts: MIT Press.
Dowty, David R. 1992. ’Variable-free’ syntax, variable-binding syntax, the natu-
ral deduction Lambek calculus, and the crossover constraint. In Proceedings of
WCCFL 11 , 161–176, Stanford, California. Stanford University, Center for the
Study of Language and Information.
Engdahl, Elisabeth. 1980. The Syntax and Semantics of Questions in Swedish.
Amherst: University of Massachusetts dissertation.
——. 1986. Constituent Questions. Dordrecht, The Netherlands: Reidel.
Evans, Frederic. 1988. Binding into anaphoric verb phrases. In Proceedings of
ESCOL 5 , ed. by Joyce Powers and Kenneth de Jong, 122–129, Columbus. Ohio
State University, Working Papers in Linguistics.
Fiengo, Robert, and Robert May. 1994. Indices and Identity. Cambridge,
Massachusetts: MIT Press.
Fodor, Janet Dean, and Ivan Sag. 1982. Referential and quantificational indef-
inites. Linguistics and Philosophy 5.355–398.
Fox, Danny. 1995a. Economy and scope. Natural Language Semantics 3.283–341.
——. 1995b. Condition C effects in ACD. In Papers on Minimalist Syntax, MITWPL
27 , ed. by Rob Pensalfini and Hiroyuki Ura, 105–119. Cambridge, Massachusetts:
MIT Working Papers in Linguistics.
——. 1996. Where does binding theory really apply? Presentation at the LF reading
group, MIT, Cambridge, Massachusetts.
——. 1997. Reconstruction, variable binding and the interpretation of chains. In
Proceedings of the Tübingen Workshop on Reconstruction, 71–118. Universität
Tübingen, Germany.
——. 1998a. Economy and Semantic Interpretation. Cambridge: Massachusetts
Institute of Technology dissertation. (in progress).
——. 1998b. Reconstruction, variable binding and the interpretation of chains. Lin-
guistic Inquiry (to appear).
——. 1998c. Locality in variable binding. In Is the Best Good Enough? , ed. by Pilar
Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky,
129–144. Cambridge, Massachusetts: MIT Press and MIT Working Papers in
Linguistics.
——, and Jon Nissenbaum. 1998. A-bar traces and the copy theory of movement.
Notes, MIT, Cambridge, Massachusetts.
Frege, Gottlob. 1884. Grundlagen der Arithmetik . Breslau.
Freidin, Robert. 1986. Fundamental issues in the theory of binding. In Studies in
the Acquisition of Anaphora, Volume I , ed. by Barbara Lust, 151–188. Dordrecht:
Reidel.
Gabbay, Dov M., and J.M.E. Moravscik. 1974. Branching quantifiers, English
and Montague Grammar. Theoretical Linguistics 1.139–157.
Geach, P.T. 1964. Referring expressions again. Analysis 24. (reprinted as Geach
1972.

295
—— 1972. Theory of reference and syntax. In Logic Matters, chapter 3.4, 97–102.
Berkeley and Los Angeles, California: University of California Press.
Grinder, J., and Paul Postal. 1971. Missing antecedents. Linguistic Inquiry
2.269–312.
Groenendijk, Jeroen, and Martin Stokhof. 1984. Studies in the Semantics
of Questions and the Pragmatics of Answers. The Netherlands: University of
Amsterdam dissertation.
Grosu, Alexander, and Fred Landman. 1998. Strange relatives of the third
kind. Natural Language Semantics 6.125–170.
Hackl, Martin, and Jon Nissenbaum. 1998. Variable modal force in for-infinitival
relative clauses. Manuscript, Massachusetts Institute of Technology, Cambridge.
Hagstrom, Paul. 1998. Decomposing Questions. Cambridge, Massachusetts: Mas-
sachusetts Institute of Technology dissertation. (in progress).
Hamblin, C.L. 1958. Questions. Australasian Journal of Philosophy 36.159–168.
—— 1973. Questions in Montague English. Foundations of Language 10.41–53.
Reprinted in Partee (1976).
Hardt, Daniel. 1992. VP ellipsis and semantic identity. In Proceedings of SALT II ,
ed. by Chris Barker and David Dowty, 145–161, Columbus. Ohio State University,
Working Papers in Linguistics.
Heim, Irene. 1987. Where does the definiteness restriction apply? Evidence from
the definiteness of variables. In The Representation of (In)definiteness, ed. by
Eric Reuland and Alice ter Meulen, chapter 2, 21–42. Cambridge, Massachusetts:
MIT Press.
——. 1997a. Predicates or formulas? Evidence from ellipsis. In Proceedings of SALT
VII , ed. by Aaron Lawson and Eun Cho, 197–221. Ithaca, New York: CLC
Publications.
——. 1997b. Predicates or formulas? Evidence from ellipsis. Presentation at LF
reading group, MIT.
——, and Angelika Kratzer. 1998. Semantics in Generative Grammar . Oxford:
Blackwell.
Hepple, Mark. 1990. The Grammar and Processing of Order and Dependency: A
Categorial Approach. Scotland: University of Edinburgh dissertation.
——. 1992. Command and domain constraints in categorial theory of binding. In Pro-
ceedings of the Amsterdam Colloquium, University of Amsterdam, The Nether-
lands.
Heycock, Caroline. 1995. Asymmetries in reconstruction. Linguistic Inquiry
26.547–570.
Higginbotham, James. 1983. Logical form, binding, and nominals. Linguistic
Inquiry 14.679–708.
——, and Robert May. 1980. Questions, quantifiers, and crossing. The Linguistic
Review 1.41–80.
Hindley, J.R., B. Lercher, and J.P. Seldin. 1972. Introduction to Combinatory
Logic. Cambridge University Press.
Hintikka, Jaakko. 1974. Quantifiers vs. quantification theory. Linguistic Inquiry
5.153–177.

296
Höhle, Tilman. 1979. ‘Normalbetonung’ und ‘normale Wortstellung’: Eine prag-
matische Explikation. Leuvense Bijdragen 68.385–437.
——. 1982. Explikation für “normale Betonung” und “normale Wortstellung”. In
Satzglieder im Deutschen: Vorschläge zur syntaktischen, semantischen und prag-
matischen Fundierung, ed. by Werner Abraham, 75–153. Tübingen: Narr.
Hornstein, Norbert. 1995. The Grammar of Logical From: From GB to Mini-
malism. Cambridge: Blackwell.
Ishii, Yasuo. 1997. Scrambling and the weak-strong distinction in Japanese. In Is the
Logic Clear? Papers in Honor of Howard Lasnik, UConn WPL 8, ed. by Joeng-
Seok Kim, Satoshi Oku, and Sandra Stjepanović, 89–112. Storrs: University of
Connecticut.
Jackendoff, Ray. 1968. An interpretive theory of pronouns and reflexives. PEGS
paper number 27.
——. 1972. Semantic Interpretation in Generative Grammar . Cambridge: MIT Press.
——. 1977. X̄-bar Syntax: A Stude of Phrase Structure. Cambridge, Massachusetts:
MIT Press.
Jacobson, Pauline. 1992. Antecedent contained deletion in a variable free se-
mantics. In Proceedings of SALT II , 193–213, Columbus. Ohio State University,
Working Papers in Linguistics.
——. 1993. Bach-Peters sentences in a variable-free semantics. In Proceedings of the
Eigth Amsterdam Colloquium.
——. 1994. i-within-i effects in a variable free semantics and a categorial syntax. In
Proceedings of the Ninth Amsterdam Colloquium. University of Amsterdam, The
Netherlands.
——. 1998a. ACE and pied-piping: Evidence for a variable-free semantics. Presenta-
tion at SALT 8, MIT.
——. 1998b. Towards a variable-free semantics. Linguistics and Philosophy (to
appear).
Jayaseelan, K.A. 1990. Incomplete VP deletion and gapping. Linguistic Analysis
20.64–81.
Johnson, Kyle. 1996. When verb phrases go missing. GLOT 2.3–9.
——, and Satoshi Tomioka. 1997. Lowering and mid-size clauses. In Proceedings
of the Tübingen Workshop on Reconstruction, 177–198. Universität Tübingen,
Germany.
Karttunen, Lauri. 1977. The syntax and semantics of questions. Linguistics and
Philosophy 1.1–44.
Keenan, Edward. 1971. Names, quantifiers, and a solution to the sloppy-identity
problem. Papers in Linguistics 4.211–232.
Kennedy, Christopher. 1994. Argument contained ellipsis. Linguistics Research
Center Report LRC-94-03, University of California, Santa Cruz.
Kitagawa, Yoshihisa, and S.-Y. Kuroda. 1992. Passive in Japanese. Manuscript,
University of Rochester and UCSD.
Kitahara, Hisatsugu. 1994. Restricting ambiguous rule-application. In Formal
Approaches to Japanese Linguistics I , ed. by Masatoshi Koizumi and Hiroyuki
Ura, 179–209, Cambridge. MIT Working Papers in Linguistics.

297
Klein, Ewan, and Ivan A. Sag. 1985. Type-driven translation. Linguistics and
Philosophy 8.163–201.
Koizumi, Masatoshi. 1994. Layered specifiers. In Proceedings of NELS 24 , ed. by
Mercè Gonzàlez, 255–269, Amherst. GLSA.
Koopman, Hilda, and Dominique Sportiche. 1982. Variables and the bijection
principle. The Linguistic Review 2.139–160.
Kratzer, Angelika. 1991. The representation of focus. In von Stechow and
Wunderlich (1991), chapter 40, 825–834.
——. 1995. Scope or pseudoscope? Are there wide-scope indefinites? Manuscript,
University of Massachusetts, Amherst.
——. 1998. Scope or pseudoscope? Are there wide-scope indefinites? In Events in
Grammar , ed. by Susan Rothstein, Dordrecht: Kluwer.
Kuno, Susumo. 1997. Binding theory in the minimalist program. Manuscript,
Harvard University, Cambridge, Massachusetts.
Lahiri, Utpal. 1991. Embedded Interrogatives and Predicates that Embed them.
Cambridge: Massachusetts Institute of Technology dissertation.
Lakoff, George. 1968. Pronouns and reference. Mimeograph, Indiana University
Linguistics Club, University of Indiana, Bloomington (published as Lakoff 1976).
——. 1972. The global nature of the Nuclear Stress Rule. Language 48.285–303.
——. 1976. Pronouns and reference. In Notes from the Linguistic Underground, Syntax
and Semantics, Volume 7 , ed. by James D. McCawley, chapter 16, 275–335. New
York: Academic Press.
Langacker, Ronald W. 1969. Pronominalization and the chain of command.
In Modern Studies in English, ed. by David A. Reibel and Sanford A. Schane,
chapter 12, 160–186. Englewood Cliffs, New Jersey: Prentice Hall.
Lappin, Shalom. 1984. VP anaphora, quantifier scope, and logical form. Linguistic
Analysis 13.273–315.
Larson, Richard K., and Robert May. 1990. Antecedent containment or vacu-
ous movement: Reply to Baltin. Linguistic Inquiry 21.103–122.
Larson, Richard K., and Gabriel Segal. 1995. Knowledge of Meaning: An
Introduction to Semantic Theory. Cambridge, Massachusetts: MIT Press.
Lasnik, Howard. 1976. Remarks on coreference. Linguistic Analysis 2.1–22.
——. 1995. A note on pseudogapping. In Papers on Minimalist Syntax, MITWPL
27 , ed. by Rob Pensalfini and Hiroyuki Ura, 143–164. Cambridge: MIT Working
Papers in Linguistics.
——, and Mamoru Saito. 1992. Move α: Conditions on Its Application and Output,
volume 22 of Current Studies in Linguistics. Cambridge: MIT Press.
Lebeaux, David. 1988. Language Acquisition and the Form of Grammar . Amherst:
University of Massachusetts dissertation.
——. 1992. Relative clauses, licensing, and the nature of the derivation. In Per-
spectives on Phrase Structure: Heads and Licensing, ed. by Susan Rothstein
and Margaret Speas, volume 25 of Syntax and Semantics, 209–239. New York:
Academic Press.
——. 1995. Where does the binding theory apply? In University of Maryland Working
Papers in Linguistics 3 , 63–88. College Park: University of Maryland.

298
——. 1998. Where does the binding theory apply? (version 2). Technical Report
98-044, NEC Research Institute, Princeton, New Jersey.
Lechner, Winfried. 1998. On semantic and syntactic reconstruction. Studia
Linguistica.
Lees, Robert B. 1960. The Grammar of English Nominalizations. The Hague:
Mouton.
—— 1961. The constituent structure of noun phrases. American Speech 36.159–168.
Linebarger, Marcia. 1980. The Grammar of Negative Polarity. Cambridge:
Massachusetts Institute of Technology dissertation. Distributed by MITWPL.
——. 1987. Negative polarity and grammatical representation. Linguistics and Phi-
losophy 10.325–387.
Lobeck, Anne. 1992. Ellipsis. Oxford: Oxford University Press.
Matthewson, Lisa. 1998. On the interpretation of wide-scope indefinites.
Manuscript, Massachusetts Institute of Technology, Cambridge (revised form to
appear in Natural Language Semantics).
May, Robert. 1977. The Grammar of Quantification. Cambridge: Massachusetts
Institute of Technology dissertation.
——. 1985. Logical Form: Its Structure and Derivation. Cambridge: MIT Press.
McCawley, John. 1976. Notes on Jackendoff’s theory of anaphora. Linguistic
Inquiry 7.319–341.
Merchant, Jason. 1998a. On the extent of trace deletion in ACD. Manuscript,
University of Utrecht, The Netherlands and University of California, Santa Cruz.
——. 1998b. Guess what i found and where. In Proceedings of WCCFL 18 , Stanford,
California. Center for the Study of Language and Information.
Miyagawa, Shigeru. 1989. Structure and Case Marking in Japanese. Number 22
in Syntax and Semantics. New York: Academic Press.
Moltmann, Friederike, and Anna Szabolcsi. 1994. Scope interactions with
pair-list quantifiers. In Proceedings of NELS 24 , ed. by Mercé González, 381–395,
Amherst. University of Massachusetts, GLSA.
Müller, Gereon. 1993. On Deriving Movement Type Asymmetries. Tübingen:
Universität Tübingen dissertation.
——. 1996. Incomplete Category Fronting. Habilitationsschrift. Universität Tübingen.
——. 1998. Incomplete Category Fronting. Dordrecht, The Netherlands: Kluwer.
Munn, Alan. 1994. A minimalist account of reconstruction asymmetries. In Pro-
ceedings of NELS 24 , ed. by Mercè Gonzàlez, 397–410, Amherst. University of
Massachusetts, GLSA.
Nissenbaum, Jon. 1998. Movement and derived predicates: Evidence from parasitic
gaps. In The Interpretive Tract, MITWPL 25 , ed. by Uli Sauerland and Orin
Percus, 247–295. Cambridge: MIT Working Papers in Linguistics.
Partee, Barbara H. (ed.) 1976. Montague Grammar . New York: Academic Press.
Pesetsky, David. 1982. Paths and Categories. Cambridge: Massachusetts Institute
of Technology dissertation.
——. 1989. Wh-in-situ: Movement and unselective binding. In The Representation
of (In)definiteness, ed. by Eric Reuland and A. ter Meulen, 98–129. Cambridge:
MIT Press.

299
——. 1994. Zero Syntax: Experiencers and Cascades. Cambridge: MIT Press.
Pica, Pierre, and William Snyder. 1994. Weak crossover, quantifer scope and
minimalism. In Proceedings of WCCFL 13 , 334–349, Stanford, California. Center
for the Study of Language and Information, Stanford University.
Quine, Willard van Orman. 1960. Variables explained away. Proceedings of the
American Philosophical Society 104. (reprinted as Quine 1995).
——. 1995. Variables explained away. In Selected Logic Papers, Enlarged Edition,
227–235. Cambridge, Massachusetts: Harvard University Press.
Reinhart, Tanya. 1976. The Syntactic Domain of Anaphora. Cambridge: Mas-
sachusetts Institute of Technology dissertation.
——. 1981. A second COMP position. In The Theory of Markedness in Generative
Grammar: Proceedings of the 1979 GLOW Conference, ed. by Adriana Belletti,
Luciana Brandi, and Luigi Rizzi, 517–557. Pisa, Italy: Scuola Normale Superiore
di Pisa.
——. 1994. Wh-in-situ in the framework of the minimalist program. OTS Working
Papers 94/03, Utrecht University, Utrecht, The Netherlands.
——. 1997. Quantifier scope: How the labor is divided between QR and choice
functions. Linguistics and Philosophy 20.335–397.
Richards, Norvin. 1997. What moves where when in which language? . Cambridge:
Massachusetts Institute of Technology dissertation.
Riemsdijk, Henk van, and Edwin Williams. 1981. NP-structure. The Linguistic
Review 1.171–217.
Ristad, Sven Eric. 1990. Computational Structure of Human Language. Cam-
bridge: Massachusetts Institute of Technology dissertation.
Rizzi, Luigi. 1990. Relativized Minimality. Number 16 in Linguistic Inquiry Mono-
graphs. Cambridge: MIT Press.
Romero, Maribel. 1996. The (cor)relation of scope reconstruction and connectivity
effects. Manuscript, University of Massachusetts, Amherst.
——. 1997. Problems for a semantic account of scope reconstruction. In Proceedings
of the Tübingen Workshop on Reconstruction, 119–146. Universität Tübingen.
Rooth, Mats. 1985. Association with Focus. Amherst: University of Massachusetts
dissertation.
——. 1992a. A theory of focus interpretation. Natural Language Semantics 1.75–116.
——. 1992b. Ellipsis redundancy and reduction redundancy. In Proceedings of the
Stuttgart Ellipsis Workshop, ed. by Steve Berman and Arild Hestvik. Arbeitspa-
piere des Sonderforschungsbereichs 340, Bericht Nr. 29, IBM Germany, Heidel-
berg.
——. 1996. Focus. In The Handbook of Contemporary Semantic Theory, ed. by
Shalom Lappin, 271–297. Oxford: Blackwell.
Ross, John R. 1967. On the cyclic nature of pronominalization in English. In
To Honor Roman Jakobson, 1669–1682. The Hague, The Netherlands: Mouton.
(reprinted as Ross 1969a).
—— 1969a. On the cyclic nature of pronominalization in English. In Modern Studies
in English, ed. by David A. Reibel and Sanford A. Schane, chapter 13, 187–200.
Englewood Cliffs, New Jersey: Prentice Hall.

300
Ross, John Robert. 1969b. Guess who? In CLS 5 , ed. by Robert I. Binnick, Alice
Davison, Georgia M. Green, and Jerry L. Morgan, 252–278, Chicago. Chicago
Linguistics Society.
Rudin, Catherine. 1988. On multiple questions and multiple wh fronting. Natural
Language & Linguistic Theory 6.445–501.
Rullmann, Hotze. 1995. Maximality in the Semantics of Wh-Constructions.
Amherst: University of Massachusetts dissertation.
——, and Sigrid Beck. 1997. Reconstruction and the interpretation of which-
phrases. In Proceedings of the Tübingen Workshop on Reconstruction, 215–249.
Universität Tübingen, Germany.
Ruys, Eddie. 1993. The Scope of Indefinites. Utrecht University dissertation.
Safir, Ken. 1984. Multiple variable binding. Linguistic Inquiry 15.603–638.
——. 1998. Reconstruction and bound anaphora: Copy theory without deletion at
LF. Manuscript, Rutgers University, New Brunswick, New Jersey.
——, and Tim Stowell. 1987. Binominal each. In Proceedings of NELS 18 , 426–450.
Amherst: GLSA.
Sag, Ivan. 1976. Deletion and Logical Form. Cambridge: Massachusetts Institute
of Technology dissertation.
Saito, Mamoru. 1992. Long distance scrambling in Japanese. Journal of East
Asian Linguistics 1.69–118.
Sato-Zhu, Eriko. 1996. The Logical Intepretation of English and Japanese Sen-
tences. Stony Brook: State University of New York dissertation.
Sauerland, Uli. 1998. Plurals, derived predicates and reciprocals. In The In-
terpretive Tract, MITWPL 25 , ed. by Uli Sauerland and Orin Percus, 177–204.
Cambridge: MIT Working Papers in Linguistics.
Schachter, Paul. 1973. Focus and relativization. Language 49.19–46.
Schönfinkel, M. 1924. Über die Bausteine der mathematischen Logik. Mathema-
tische Annalen 92.305–316.
Schütze, Carson. 1995. Pp attachment and argumenthood. In Papers on Lan-
guage Processing and Acquisition, MITWPL 26 , ed. by Carson Schütze, Jennifer
Ganger, and Kevin Broihier, 95–151. Cambridge: MIT Working Papers in Lin-
guistics.
Schwarz, Bernhard. 1993. Gewisse Fälle eingebetteter Fragesätze. Master’s
thesis, Universität Tübingen, Germany.
Schwarzschild, Roger. 1994. The contrastiveness of associated foci. Manuscript,
Hebrew University of Jerusalem, Israel.
——. 1998. Givenness and optimal focus. Natural Language Semantics (to appear in
revised form).
Selkirk, Elizabeth. 1995. Sentence prosody: Intonation, stress, and phrasing.
In The Handbook of Phonological Theory, ed. by John Goldsmith, chapter 16,
550–569. London: Blackwell.
Sharvit, Yael. 1996a. Functional dependencies and indirect binding. In Semantics
and Linguistic Theory VI , ed. by Teresa Galloway and Justin Spence, 227–244,
Ithaca, New York. Cornell University, CLC Publications.

301
——. 1996b. The Syntax and Semantics of Functional Relative Clauses. New
Brunswick, New Jersey: Rutgers University dissertation.
——. 1998. Possessive wh-expressions and reconstruction. In Proceedings of NELS
28 , Amherst. University of Massachusetts, GLSA. (to appear).
Sportiche, Dominique. 1996. A-reconstruction and constituent structure. Hand-
out, University of California, Los Angeles.
Steedman, Mark. 1996. Surface Structure and Interpretation. Cambridge, Mas-
sachusetts: MIT Press.
Sternefeld, Wolfgang. 1998. The semantics of reconstruction and connectiv-
ity. Arbeitspapier 97, SFB 340, Universität Tübingen and Universität Stuttgart,
Germany.
Szabolcsi, Anna. 1987. Bound variables in syntax (are there any?). In Proceedings
of the Amsterdam Colloquium, 331–351. University of Amsterdam, The Nether-
lands.
Tada, Hiroaki. 1993. A/A-Bar Partition in Derivation. Cambridge: Massachusetts
Institute of Technology dissertation.
Takano, Yuji. 1993. Minimalism and proper binding. Manuscript, UC Irvine.
——. 1994. Unbound traces and indeterminacy of derivation. In Current Topics in
English and Japanese, ed. by Masura Nakamura, 229–253. Tokyo: Hituzi Syoboo.
——. 1995. Predicate fronting and internal subjects. Linguistic Inquiry 26.327–340.
Tancredi, Christopher. 1992. Deletion, Deaccenting and Presupposition. Cam-
bridge: Massachusetts Institute of Technology dissertation.
Tarski, A. 1936. Der Wahrheitsbegriff in den formalisierten Sprachen. Studia
Philosophica 1.261–405.
Truckenbrodt, Hubert. 1995. Phonological Phrases: Their Relation to Syn-
tax, Focus, and Prominence. Cambridge: Massachusetts Institute of Technology
dissertation.
Vergnaud, Jean Roger. 1974. French Relative Clauses. Cambridge: Mas-
sachusetts Institute of Technology dissertation.
von Stechow, Arnim. 1993. Die Aufgaben der Syntax (the objectives of syntax).
In Syntax: Ein internationales Handuch der zeitgenössischen Forschung (Syntax:
An International Handbook of Contemporary Research), ed. by Joachim Jacobs,
Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann, 1–88. Berlin:
de Gruyter.
——. 1996. Some remarks on choice functions and LF-movement. In Proceedings of the
Konstanz Workshop “Reference and Anaphorical Relations”, ed. by Klaus von
Heusinger and Urs Egli. Constance: Universität Konstanz.
——, and Dieter Wunderlich (eds.) 1991. Semantik: Ein internationales Han-
duch der zeitgenössischen Forschung (Semantics: An International Handbook of
Contemporary Research). Berlin: de Gruyter.
Wasow, Thomas. 1972. Anaphoric Relations in English. Cambridge: Massachusetts
Institute of Technology dissertation.
——. 1979. Anaphora in Generative Grammar . Ghent: Story Scientia.

302
Webelhuth, Gert. 1995. X-bar theory and case theory. In Government and
Binding Theory and the Minimalist Program, ed. by Gert Webelhuth, chapter 1,
15–96. Cambridge: Blackwell.
Whitehead, Alfred N., and Bertrand Russell. 1910. Principia Mathematica.
Volume I . Cambridge, Great Britain.
Wilkinson, Karina. 1991. Studies in the Semantics of Generic Noun Phrases.
Amherst: University of Massachusetts dissertation.
Williams, Edwin. 1977. Discourse and logical form. Linguistic Inquiry 8.101–139.
Winter, Yoad. 1997. Dependency and distributivity of plural definites. Manuscript,
Utrecht University.
Wold, Dag. 1995. Identity in ellipsis: Focal structure and phonetic deletion.
Manuscript, MIT.
——. 1996. Long distance selective binding: The case of focus. In Proceedings of
SALT 6 , ed. by Teresa Galloway and Justin Spence, 311–328, Ithaca, New York.
Cornell University, CLC Publications.
——. 1998. Cambridge: Massachusetts Institute of Technology dissertation. (in
progress).
Wurmbrand, Susi. 1998. Downsizing infinitives. In The Interpretive Tract,
MITWPL 25 , ed. by Uli Sauerland and Orin Percus, 141–175. MIT Working
Papers in Linguistics.
Yatsushiro, Kazuko. 1997. VP-scrambling in Japanese. In Is the Logic Clear?
Papers in Honor of Howard Lasnik. UConn WPL 8, ed. by Joeng-Seok Kim,
Satoshi Oku, and Sandra Stjepanović, 325–338. Storrs: University of Connecti-
cut, Working Papers in Linguistics.
Zubizaretta, Maria Luisa. 1998. Prosody, Focus and Word Order . Cambridge,
Massachusetts: MIT Press.

303