BOSTON UNIVERSITY GRADUATE SCHOOL OF ARTS AND SCIENCES

Dissertation

ELLIPSIS, RIGHT NODE RAISING, AND ACROSS-THE-BOARD CONSTRUCTIONS

by

SEUNGWAN HA B.A., Korea University, 1999 M.A., Korea University, 2001

Submitted in partial fulfillment of the requirement for the degree of Doctor of Philosophy 2008

© Copyright by SEUNGWAN HA 2007

Approved by

First Reader

_______________________________________________________ Paul Hagstrom, Ph.D. Associate Professor of Linguistics Boston University

Second Reader _______________________________________________________ Shigeru Miyagawa, Ph.D. Professor of Linguistics Massachusetts Institute of Technology

Third Reader

_______________________________________________________ Kyle Johnson, Ph.D. Professor of Linguistics University of Massachusetts, Amherst

iv

ACKNOWLEDGEMENTS

One summer night in 2005, I read a bed time story for my 2 year-old son, written in Korean, the title of which was Mwues-i mwues-i ttok kasul ka? ‘Which of those are identical?’ While I was reading the book to him, I wondered what the syntactic structure of this title would look like and supposed that it should be a Right Node Raising (RNR) construction. My interest in RNR has begun with the title of this old Korean children’s book, so I wish to thank the author for inspiring me with the initial input for my dissertation topic. I should admit, however, that I still have not figured out for sure whether or how I can analyze the sentence as RNR, even after having completed this dissertation. Other than the author of that book, there are several people to whom I am deeply grateful for. First, I would like to thank for my advisor, Paul Hagstrom. I have bothered him in every possible way a graduate student can. Paul has been always available for meetings, including late in the evening and during his sabbatical. He has sincerely answered all kinds of linguistics questions that I have had during my graduate school years – syntax, semantics, language acquisition, and even information structure. His insightful answers to, and comments on, my questions have led to the discussion of my linguistic work in the past five years. Since he was so good at organizing ideas, I always had a hope that he could convert my unorganized thoughts into linguistic generalizations. Of course, he has never disappointed me. He has been thoughtful and patient to read and comment on many drafts of this dissertation, and kind enough to correct English grammar for my conference abstracts, term papers, manuscripts, and this dissertation, of course. Without his efforts, I could not have managed to present papers in many conferences in recent years, and submitted this dissertation as an accomplishment of my graduate school years at Boston University.

v There are also two external readers I have exploited for this dissertation: Shigeru Miyagawa and Kyle Johnson. I first met Miyagawa sensei ‘teacher’ in one of the MIT colloquium talks five years ago. Since then, he has always been interested in what I am doing and provided me with kind and encouraging words whenever I came across him. I was especially lucky that I was able to audit his syntax seminar course at MIT in fall 2005 where he presented his new ideas on Japanese focus movement in terms of satisfying the EPP. My ideas concerning the importance of contrastive focus for ellipsis in RNR was inspired by his lecture notes from that class. I owe an enormous debt to Kyle Johnson. Without the benefit of countless hours spent with him discussing the nature of the ellipsis feature of Right Node Raising and the right edge effects, this dissertation could not have come out as it is now. He has always clarified my vague ideas and pointed out what predictions I should be able to make—and then whether the prediction is borne out. Although I had to cut some part of my idea of this dissertation topic, “thanks” to him for his suggestions and comments, it did not take a long time for me to realize what he suggested was correct. I would also like to express my profound gratitude to the other committee members: Victor Manfredi and Cathy O’Connor. I was very lucky that I could have Victor in my thesis committee at the last minute. Even within a relatively short time, the influence on this dissertation was huge. Victor always pointed out the importance of the focus and prosody in Right Node Raising and convinced me to insert pitch contours for some of the examples. He also kindly agreed to be the subject for the voice recordings to be analyzed. Cathy also plays an important role in this dissertation. She made me realize how important context is in the examples of my dissertation. She provides me with a possible way to develop a corpus for RNR examples with using prompt phrases, such as if not, or let alone. I was not able to include the corpus data in this dissertation, due to the time limitation, but I am very thankful that

vi she gave me the next project to pursue. I also wish to show my sincere gratitude to Cathy, as the director of the Applied Linguistics program at Boston University, for her encouragement, financial and moral support. I want to thank other professors in the Applied Linguistics program at Boston University for providing me the wonderful environment for studying linguistics and psychology, and for undertaking research: Shanley Allen, Jonathan Barnes, Cathy Harris, Michela Ippolito, Jacqueline Leiderman, and Marnie Reed. I would also like to thank professors in the department of Linguistics and Philosophy at MIT, who generously let me audit their classes and have an occasional meeting for the past five years. Specifically, I would like to thank Danny Fox, Irene Heim, Shigeru Miyagawa, David Pesetsky, Norvin Richards, and Ken Wexler. Portions of the material in this dissertation were presented at the 42nd annual Chicago Linguistics Society conference (CLS 42), the 3rd Workshop on Altaic Formal Linguistics (WAFL 3), the 37th North East Linguistics Society conference (NELS 37), the 81st annual Linguistic Society of America conference, the 15th annual Conference of the Student Organization of Linguistics in Europe (ConSOLE 15), and the 4th Workshop on Altaic Formal Linguistics (WAFL 4). I thank the audiences at those conferences for their questions and comments. I am indebted to the following people for judging English sentences for me: Jonathan Barnett, Eric Goldstein, Paul Hagstrom, Kyle Johnson, Victor Manfredi, and Cheryl Small. The following people also deserve special mention. I benefited from discussions with Klaus Abels, Duk-Ho An, Hee-Don An, Asaf Bachrach, Seth Cable, Dae-Ho Chung, Shin Fukuda, Steven Franks, Michela Ippolito, Roni Katzir, Shin-Sook Kim, Jason Merchant, Jairo Nunes, Jaehee Park, Jongun Park, Shoichi Takahashi, Mark de Vries, Hideaki Yamashita, and James Yoon.

vii Finally, I want to express my sincere gratitude to my wife, Bowon, and my son, Joshua, for their love, support and patience. I also wish to thank my parents and my brother who have given me the warm support and encouragement. This dissertation is dedicated to my family.

viii

ELLIPSIS, RIGHT NODE RAISING, AND ACROSS-THE-BOARD CONSTRUCTIONS (Order No.

)

SEUNGWAN HA Boston University Graduate School of Arts and Sciences, 2008 Major Professor: Paul Hagstrom, Associate Professor of Linguistics

ABSTRACT This dissertation investigates two unique syntactic structures in natural language, known as Right Node Raising (RNR) and Across-The-Board (ATB) constructions. It establishes a connection between these constructions and the more familiar ellipsis phenomenon, reducing RNR and ATB to special cases of backwards ellipsis under coordination. First, I review previous accounts of RNR – movement, phonological deletion, and multiple dominance – and provide arguments against them. I propose that RNR is best understood as an ellipsis phenomenon, on the grounds that RNR shares many properties with ellipsis such as sloppy identity and Vehicle Change (Fiengo and May 1994). Second, I discuss licensing conditions for RNR. I argue that the licensing conditions can be reduced to semantic identity; namely, the conjuncts must stand in a mutual entailment relation, modulo existential closure over focused material, at the point of semantic interpretation. Following Merchant’s (2001) feature analysis of sluicing, I propose that ellipsis in RNR is triggered by an ellipsis feature (the ERNR feature). An ERNR feature is only valued if three requirements are met: i) it enters the derivation with a contrastively focused lexical item, ii) at PF, the complement of the

ix feature is forced to be silent, and iii) a mutual entailment relationship between the conjuncts must be established. When all the conditions are satisfied, RNR is licensed. Third, I turn to the striking similarities between RNR and ATB constructions and argue that RNR constructions are the underlying structures for ATB constructions. A single occurrence of the RNR target undergoes movement to the topmost position, stopping by every intermediate phase. With this analysis, we can solve several long-standing mysteries about ATB constructions. The conceptual problem of two independent movement operations targeting a single landing site does not arise. Furthermore, the proposed analysis sheds new light on the origin of the “single-identity reading” of ATB constructions.

x

TABLE OF CONTENTS

Chapter 1 Right node raising and ellipsis……………………………………………….1 1.1 Right node raising………...………….………………………………………………………..3 1.1.1 Backwards anaphora constraint..………………..……………………………………….3 1.1.2 Backwards ellipsis is RNR.………………………..…………………………………….5 1.2 Properties of RNR………………………………………….………………………………….9 1.2.1 Syntactic properties…………………………………..………………………………….9 1.2.2 Semantic properties.……………………………………….…………………………...14 1.2.3 Prosodic properties……………………………………………………………………..17 1.2.4 Pragmatic properties……………………………………………………………………22 1.3 Overview………………….………………………………………………………………….23

Chapter 2 Syntax of right node raising……………………..……………………………27 2.1 Movement analyses…….…………………………………………………………………….31 2.1.1 Arguments for movement…………..…………………………………………………..32 2.1.2 Arguments against movement………………………..………………………………...36 2.1.2.1 Preposition stranding……………………………………………..………………...36 2.1.2.2 Island insensitivity………………………………………………………..……...38 2.1.2.3 Right roof constraint………..……………………...………………………………39 2.1.2.4 Non-constituents………….………………………………………………………..40 2.1.3 Sabbagh’s (2007) cyclic linearization…………………………….……………………41 2.1.3.1 Fox and Pesetsky’s (2005) cyclic linearization………………..…………………...42 2.1.3.2 Cyclic linearization of RNR……………..…………………………………………43

xi 2.1.4 Challenges for Sabbagh’s analysis………..……………………………………………47 2.2 Non-movement accounts………….………………………………………………………….55 2.2.1 Strict phonological deletion account…………..…………………………………….…56 2.2.2 Multiple dominance account..………………………………….………………………59

Chapter 3 Right node raising is ellipsis………………………………………..………...74 3.1 Problems of the previous accounts..………………….………………………………………74 3.1.1 Vehicle change effects in RNR……………………………………..………………….75 3.1.2 Morphological mismatches in RNR…………….…………………..………………….83 3.1.3 Sloppy identity in RNR……………………………….………………..………………86 3.1.4 Parallelism in RNR………………………………………..………………………….90 3.1.5 Certain dialects of British do…………………………..…………………………….…98 3.1.6 Ellipsis vs. the strict phonological deletion……………..…………………………….100 3.2 Potential problems for the ellipsis account and solutions…………………………………103 3.2.1 Relational modifiers……………………….………………………………………….103 3.2.2 Korean relational modifier kakkak………………..…………………………………110 3.2.3 Scope differences………………………………………………..…………………….116

Chapter 4 Licensing right node raising…………………………………………..……..118 4.1 Structural isomorphism………………………………………………………….………….120 4.2 Semantic identity…………………………………………………………………………..122 4.3 The Hartmann-Féry analysis………………………………………………………………130 4.3.1 Focus structure and RNR.…………………………………………………………….131 4.3.2 Prosodic structure of RNR..…………………………………………………………..135

xii 4.4 Problems for the Hartmann-Féry analysis..………………………………………………...139

Chapter 5 The ERNR analysis……………..………………………………………..…….150 5.1 Ellipsis features…………………………………….……………………………………….150 5.1.1 Merchant (2001)………………………………………………………………………150 5.1.2 The ERNR feature…………………………….…………………………………………152 5.2 Consequences of ERNR………………………………….……………………………………159 5.2.1 RNR and deaccenting…………………………..……………………………………..160 5.2.2 Ellipsis properties………………………………..……………..……………………..162 5.2.2.1 Sloppy identity……………………………………….…………………………..162 5.2.2.2 Vehicle change……………………………………………………………………163 5.2.2.3 Morphological mismatches……………………………………………………….165 5.2.3 Double object-dative asymmetries………………..………………..…………………169 5.2.4 Korean dummy plural markers………………………………………………………..174 5.2.5 Structural mismatches………………………………………………………………..176 5.3 ERNR in verum focus…………………………………………………………………………177 5.4 Non-constituency in RNR revisited………………………………………………………...184 5.5 Right edge restriction under the ERNR analysis………………………………………………192 5.6 Summary of the ERNR analysis……………………………………………………………….193

Chapter 6 Across-The-Board constructions…………………………………...………..195 6.1 Coordinate structure constraint………………………………………………….…………196 6.2 Challenges for the CSC…………………………………………………………….…….…197 6.2.1 Multiple copies from movement………………………………………………..…….198

xiii 6.2.2 Single-identity reading…………………………………………………………..……198 6.2.3 Extractions out of only one conjunct………………………………………………....200 6.2.4 Lack of LF ATB movement………………………………………………………..…201 6.3 Alternative accounts………………………………………………………………………...203 6.3.1 Sideward movement…………………………………………………………………..203 6.3.2 Parasitic gap analysis…………………………………………………………………212 6.3.3 Multiple dominance…………………………………………………………………..215 6.3.3.1 Citko’s (2003, 2005) parallel merge……………………………………………..215 6.3.3.2 Bachrach and Katzir’s (2006a) delayed spellout…………………………………217 6.3.4 Conceptual problems for the multiple dominance accounts……………….…………224 6.3.4.1 Problems for the delayed spellout account………………………….……………224 6.3.4.2 Problems for the parallel merge account………………………………….……...226

Chapter 7 A RNR account of ATB constructions………….………….………..………229 7.1 Williams’ (1978) insight……………………………………………………………………230 7.2 Ellipsis and the apparent ATB movement…………………………………………….…….232 7.2.1 Similarity between RNR and ATB…………………………………………………...232 7.2.2 ERNR and successive cyclic wh-movement…………………………………………….235 7.2.3 The CSC revisited: Multidimensional analysis……………………………………….237 7.2.4 ERNR and the multidimensional view of the CSC……………………………………..240 7.2.5 ERNR and weak crossover effects………………………………………………………246 7.2.6 ERNR and ATB extractions from double object positions……………………………...247 7.3 Consequences…………………………………………………………………….…..……..250 7.3.1 Single-identity vs. paired answers…………………………………………..………250

xiv 7.3.1.1 Deriving paired answers…………………………………………………….……251 7.3.1.2 Deriving single-identity answers……………………………………………..…..252 7.3.2 Functional readings of bound pronominals…………………………………………...253 7.3.3 Movement out of the second conjunct………………………………………………..255 7.3.3.1 Selectional properties………………...…………………………………………..255 7.3.3.2 Crossover effects……….………………………………………………………...256 7.3.3.3 Island effects……………..……………………………………………………….258 7.3.3.4 Binding facts………..…………………………………………………………….261 7.4 Comparing analyses………....……………………………………………………………...263 7.4.1 The source of movement: Evidence against parasitic gap analyses..……………….263 7.4.2 The elided copy: Evidence against multiple dominance analyses……………..……..269 7.4.2.1 The paired interpretations…………..……………………………………..……...269 7.4.2.2 ATB left branch extraction and the right edge condition……….………………..270 7.4.3 Interim summary……………………………………..……………………………….274 7.5 ERNR and LF ATB constructions………………………….………………………………….275 7.5.1 ATB quantifier raising………………………………..………………………………275 7.5.2 Relational modifiers in RNR revisited……………………….……………………….279 7.6 ATB constructions irrelevant to RNR……………………………….……………………...283 7.6.1 DP coordination and the CSC…………………………………………………..…….283 7.6.2 Subject ATB constructions……………………………………………………………284 7.7 Gapping: ATB movement vs. vP-ellipsis…………………………………………………..286 7.8 Exceptional cases and information structure………………………………………………..287 7.8.1 Williams’ example revisited…………………………………………………………..287 7.8.2 Information structure………………………………………………………………….289

xv 7.8.3 ATB out of an ECM subject position…………………………………………………292

8 Concluding remarks..………………………………………………………………...295 Bibliography……………………………..…………………………………………….301

xvi

LIST OF TABLES

Table 1. Comparison between non-movement accounts…………………………………………98 Table 2. Resemblance vs. non-resemblance…………………………………………………….291

xvii

LIST OF FIGURES

Figure 1. Pitch contour of (30a)……..……………………………………………………………18 Figure 2. Pitch contour of right node raising (30b)…………………………………………….18 Figure 3. The upstep-downstep pitch contour in RNR…………………………………………..19 Figure 4. The upstep-downstep pitch contour in RNR………………………………………….135 Figure 5. Upstep in a RNR sentence……………………………………………………………137 Figure 6. Upstep and phrasing in gapping…...………………….………………………………144 Figure 7. Gapping without downstep……………………………………………………………145

xviii

LIST OF ABBREVIATIONS

A

Antecedent

ACC

Accusative case

ACD

Antecedent Contained Deletion

ATB

Across-The-Board

BAC

Backwards Anaphora Constraint

CONJ

Conjunction

CP

Complementizer Phrase

CSC

Coordinate Structure Constraint

DEC

Declarative marker

DM

Distributive Marker

DP

Determiner Phrase

DPM

Dummy Plural Marker

E

Ellipsis or Ellipsis feature

ECM

Exceptional Complement Marker

F

Focus marker

F-clo

Focus-closure

FocP

Focus Phrase

FUT

Future tense

H

High tone

IP

Intonation Phrase

L

Low tone

LBE

Left Branch Extraction

xix LCA

Linear Correspondence Axiom

LF

Logical Form

NPI

Negative Polarity Item

NOM

Nominative case

Op

Operator

PF

Phonological Form

PhP

Phonological Phrase

PL

Plural marker

PPI

Positive Polarity Item

PRES

Present tense

RER

Right Edge Restriction

RM

Relation Modifier

RNR

Right Node Raising

RRC

Right Roof Constraint

QR

Quantifier Raising

SG

Singular marker

TOP

Topicalization marker

TP

Tense Phrase

U

Utterance

VP

Verb Phrase

&P

Coordination Phrase

1

CHAPTER 1 RIGHT NODE RAISING AND ELLIPSIS

Right Node Raising (RNR) is a construction formed normally under coordination, where a gap appears in the first conjunct, as shown in (1a-c). It appears as if a single constituent has simultaneously moved out of each conjunct to the right edge of the sentence. This was the usual analysis of RNR and, in fact, the source of the name for the construction.

(1)

a. Mary LIKES, but Jane HATES – baseball. b. Mary CAN, but Bill CAN’T– swim across the river. c. John gave TO MARY, and Mary gave TO SUE – an expensive present.

RNR was first described and analyzed within generative grammar by Ross (1967) (under the name Backward Conjunction Reduction). The commonly used name Right Node Raising was invented by Postal (1974). Those traditional analyses treat RNR as being derived by rightward movement. There have been a couple of other approaches to the examples in (1) that do not appeal to movement; in some, the RNR target is deleted at PF under identity with its antecedent in the second conjunct (phonological deletion accounts), and in others the RNR target is shared by the conjuncts (multiple dominance accounts). However, the name “RNR” is still used in those analyses, despite its suggestion that movement is involved. In this dissertation, I will also use the

2 term “RNR” with no implication of movement. Rather, RNR refers to a variant of ellipsis in the first conjunct here.1 The aim of this dissertation is to dissolve the uniqueness of the RNR construction and provide a way to accommodate RNR within the more common phenomenon of ellipsis. To do so, I will discuss various properties of RNR at length, find commonalities between ellipsis and RNR, and account for some apparent differences between them. In the end, I will propose an ellipsis analysis of RNR. Another goal of this dissertation is to shed a new light on analyses of AcrossThe-Board (ATB) constructions. I will propose that the underlying structure of ATB constructions is RNR.2 The structure of this introductory chapter is as follows. In section 1.1, I will define what IS, and what is NOT – a RNR sentence.3 In particular, we will see evidence that what has been referred to as backwards ellipsis is, in fact, RNR. I will provide evidence that the distribution of backwards sluicing is closer to RNR than it is to forward sluicing, and that backward VP-ellipsis does not fit in licensing conditions of forward VP-ellipsis. 4 In section 1.2, I will outline the general properties of RNR in terms of syntax, semantics, prosody, and pragmatics, and address how they are similar to or different from forward ellipsis. This section will also provide background to the discussions in the subsequent chapters of this dissertation. In section 1.3, I will lay out the overview of the rest of the dissertation.

1

I will indicate RNRed materials with strikethrough inside an angled bracket. The properties of ATB constructions will be discussed in chapter 7. 3 Capital letters refer to accented materials, and what follows a hyphen (–) indicates a RNR target. 4 Forward ellipsis is defined as omitting material that follows the antecedent. Backwards ellipsis is just the opposite in the sense that the omitted material precedes its antecedent. 2

3

1.1 Right Node Raising This section is concerned with what counts as RNR. In this dissertation, RNR is defined as any sentence in coordination where a gap occurs in the first conjunct and its antecedent in the second. Therefore, all the examples in (1) are considered as RNR. RNR can even target expressions below the word-level in the first conjunct. A part of the word under-generation is elided in the first conjunct in (2a), and a similar case in German is observed in (2b), due to Hartmann (2000).

(2)

a. This analysis suffers from both UNDER- and OVER-generation. b. Frühlingsblumen

und Herbstblumen

Springtime flowers and autumn flowers (Hartmann 2000: 57)

An interesting question arises as to whether the gap in the first conjunct counts as VP-ellipsis or RNR in (1b). Apart from the fact that VP is elided in the first conjunct, rather than in the second, it appears be reasonable to assume that VP-ellipsis occurs here. Or we cannot tell at best whether it is RNR or VP ellipsis. In the next subsection, I will argue that (1b) is not an example of VP-ellipsis since the sentence does not follow licensing conditions of forward VP-ellipsis, and that it bears the same properties with RNR, which I will discuss in chapter 5 in more detail.

1.1.1 Backwards Anaphora Constraint In this subsection, I will examine whether backwards ellipsis is RNR. The first proposal about constraints on backwards ellipsis found in literature is the Backwards Anaphora Constraint (BAC), proposed by Langacker (1969), as given in (3).

4 (3)

Backwards Anaphora Constraint An anaphora preceding its antecedent needs to be contained in a subordinate clause.

Let us consider the examples in (4). In the case where backwards anaphora is contained in an adjunct clause, as in (4a), the BAC is satisfied because the adjunct is a subordinate clause of the antecedent. On the other hand, in a coordination of two matrix clauses as in (4b), the clause containing a cataphoric pronoun is not subordinated to the matrix clause; the first conjunct ccommands the second conjunct, assuming Boolean Phrase of the coordinate structure (Munn 1993, Zoerner 1995). Thus, the BAC would be disobeyed and the sentence is ungrammatical in (4b).

(4)

a. After shei bought a lot of apples from the farmer, Maryi sold them to her neighbors. b. *Shei bought an apple, and Maryi sold it to her neighbors.

Assuming that a null anaphora occupies the ellipsis site, Langacker claims the constraint applies to backwards ellipsis, too, with examples in (5a-b). If the clause containing ellipsis exists in the adjunct clause, as in (5a), the backwards anaphora is correctly predicted to be acceptable. On the other hand, when the ellipsis site is in the first conjunct of the coordinate structure, the backwards anaphora is predicted to be unacceptable, as shown in (5b).5

(5)

a. Because Billie didn’t pro, John drank beer. b. *Billie didn’t pro, but John drank beer.

5

The judgment is Langacker’s.

5 However, according to my informants, sentences, like (5b), have been judged grammatical with the focus in didn’t and beer as in (6). Winkler (2000) also shares this judgment. She claims that there may be a context where (5b) would be acceptable. For example, in the question-answer pair in (7), the answer becomes acceptable. This type of sentence will be revisited in chapter 5.

(6)

BILLIE DIDN’T , but JOHN drank beer.

(7)

Q: Who did and who didn't watch Casablanca? (Talking of Anne and Manny) A: ANNE DID but MANNY DIDN'T – watch Casablanca. (Winkler 2000: 6, fn 5)

If this is correct, it leads us to two important conclusions: First, the BAC is not applicable to backwards ellipsis, on the grounds that backwards anaphora in (4) CAN – but backwards ellipsis in (6) CANNOT – be explained by the BAC. Second, this implies that the site for backwards ellipsis is something other than a null anaphor. We will see more evidence in chapter 2.

1.1.2 Backwards ellipsis is RNR Assuming that the BAC is not the correct analysis for (5), we need to consider what backwards ellipsis is. As far as I know, there has been no other specific proposal for the nature of backwards ellipsis since Langacker (1969). Researchers generally assume that backwards ellipsis is just the reverse of forward ellipsis; for example, the VP gap is found in the first conjunct in backwards VP-ellipsis, rather than in the second conjunct. Now let us consider (8). (8a) is an example of forward VP-ellipsis. It is striking that (8b) is ungrammatical. (8b) is just minimally different from (8a) with respect to where VP-ellipsis

6 occurs; that is, the ellipsis site is the first conjunct in (8b). Therefore, if (8b) were a type of VPellipsis, then whatever licensing conditions allow VP-ellipsis in (8a) should also be able to license ellipsis in (8b). The ungrammaticality of (8b) shows that this is not the case.

(8)

a. John likes the show, and Mary does, too. b. *John does, and Mary likes the show, too.

Interestingly, if the dummy do is replaced by didn’t, so that the two conjuncts are in contrast, grammaticality dramatically improves, as shown in (9).

(9)

JOHN DIDN’T, but MARY liked the show.

This indicates that there is some difference between backwards ellipsis and forward ellipsis. The gap in the first conjunct seems to be neither a regular ellipsis nor a backwards anaphora. What is rather important for the elision to occur in first conjunct is to establish contrast between the conjuncts. Crucially, the same appears to hold for RNR. For example, (10a) is acceptable, but (10b) is not. Notice that a contrastive focus is assigned on the verb in (10a), but not in (10b). I will argue that what allows RNR in (10a) is the existence of the contrastive focus in the first conjunct (Hartmann 2000).

(10) a. JOHN LOVED, but BILL HATED – going fishing. b. ?*JOHN loved, and BILL loved – going fishing.

7 The detail of licensing conditions for ellipsis in relations to contrast will be discussed in chapter 5. Here I will just assume that backwards ellipsis is RNR, as stated in (11).

(11) Backwards ellipsis Backwards ellipsis is RNR and follows the same licensing conditions as RNR.

We have already observed in (8) that backwards VP-ellipsis does not have the same distribution as forward VP-ellipsis. Let us now consider backwards sluicing, a phenomenon discussed by Giannakidou and Merchant (1998). (12a-b) are examples where sluicing occurs in the first conjunct, which Giannakidou and Merchant dub reverse sluicing.

(12) a. It is not clear WHEN and WHERE – that poor guy borrowed $10,000. b. No one was sure of IF but John proved to us WHEN – Mary won a Pulitzer Prize.

It is puzzling by the fact that (12b) is grammatical, given that its counterpart forward sluicing is not acceptable in (13).6

(13) *John told us that Mary won a Nobel Prize, but nobody was sure of if.

6

Giannakidou and Merchant (1998) distinguish reverse sluicing (i) from RNR (ii). I argue that for both of the cases, contrastive focus must be assigned on wh-phrases and if, and that with the contrastive focus, it is not necessary to have if SO between the conjuncts in this particular case. (i) Frank wondered if and when – the guests would arrive. (ii) Frank wondered if, and if SO, WHEN – the guests would arrive. (Giannakidou and Merchant 1998: 248-249)

8 Let us consider (13), first. Merchant (2001) argues that an E(lliptical) feature enters the derivation in T. He assumes that the E feature is a strong feature bundle [wh*, Q*] and needs to be checked by [+wh, +Q] in the C head. The impossibility of sluicing in (13) can be explained by a failure to check the E feature. Since the complementizer if lacks [+wh, +Q], it fails to check the E feature. Thus, sluicing cannot occur. The same would be predicted to hold in (12b) if backwards sluicing were constrained by the same conditions as forward sluicing. The prediction is not borne out. Backwards sluicing in (12b) is possible, although here too the C head lacks [+wh, +Q], which is necessary for the E feature to be checked. Therefore, this also indicates that backwards sluicing is not a regular ellipsis. If so, (12b) would be incorrectly predicted to be ungrammatical. If we assume that examples in (12) are RNR, the mystery would be easily solved. The condition for RNR we assume at this point is the assignment of a contrastive focus on any element just prior to the RNR target in the first conjunct. Since IF bears a contrastive focus with respect to WHEN in the second conjunct, it satisfies the RNR licensing condition. I will elaborate more in detail in chapter 5. To summarize this section, I presented evidence that backwards ellipsis is different from forward ellipsis; backwards VP-ellipsis and sluicing do not seem to be the same as forward counterpart ellipsis phenomena. Rather, backwards ellipsis shares a crucial property with RNR. Without contrastive focus, no backwards ellipsis is licensed, just like RNR. Therefore, the predictions made by the hypothesis in (11), that all backwards ellipsis is actually RNR, appear to be borne out; RNR refers to any gap in the first conjunct where its antecedent can be found in the second conjunct.

9

1.2 Properties of RNR 1.2.1 Syntactic properties Similar to ellipsis, RNR sentences are missing some material. However, compared with ellipsis, RNR bears the elided materials not in the second conjunct, but in the first. Apart from this difference of directionality, RNR exhibits several similarities to ellipsis. VP-ellipsis can occur in the embedded clause in (14a), and so can RNR in (14b).

(14) a. John thought Mary was going to donate his car to the charity, and Mary thought John was going to . b. John thought Mary was trying to SELL , and Mary thought John was trying to DONATE – his car to the charity.

The possibility of RNR in (14b) indicates that RNR is different from gapping. Gapping can occur in the matrix clause (15a), but it cannot in the embedded clause (15b). Therefore, RNR should not be treated as a type of gapping.

(15) a. Mary ate Sushi, and Bill Teriyaki. b. *Mary said that she ate Sushi, and Bill said that he Teriyaki.

Furthermore, backwards gapping is not possible in English in (16), so if RNR were a type of gapping, we need to answer why gapping is only available in the second conjunct.

(16) *Bill the car, but Mary bought the bike.

10 RNR also shows some other hallmarks of ellipsis. I will briefly lay out three points, which will be discussed at length in chapter 3. First, Fiengo and May’s (1994) Vehicle Change effects appear in both forward ellipsis and in RNR. Fiengo and May propose that proper names and pronouns share a binary feature [±pronoun], but the reconstruction of the elided material is not sensitive to the value of this feature, so proper names can be reconstructed as pronouns. Let us consider Vehicle Change effects with examples in (17). In forward VP-ellipsis (17a), had John been faithfully reconstructed in the ellipsis site, it would have been improperly c-commanded by a coreferential noun phrase. However, Vehicle Change allows John to be recast as him, so no Principle C violation occurs. The same holds in RNR (17b) as well; the proper name in the RNR target can be shifted into a pronoun and Principle C violation does not arise.

(17) a. Sue said Bill wrote a mean joke about Johni on the blackboard, and hei told us that Mary did , too. b. Hei hopes Susan WON’T , but the secretary knows that she WILL – fire Johni at the end of this year.

Second, strict morphological identity is not required in RNR, just as in forward VP-ellipsis. Warner (1986), Lasnik (1999), and Lightfoot (1999) show that inflections need not match under VP-ellipsis in (18a). In (18a), the past tense morpheme is attached to the verb in the first conjunct, while the root form needs to be in the second. (18b) shows that tense can be mismatched in RNR, too.

11 (18) a. John slept, and soon Mary will , too. b. John WOULDN’T , but Mary HAS – negotiated her salary with the company for next year.

Finally, sloppy identity is available in RNR as it is in forward ellipsis. Let us first consider ambiguities in forward VP-ellipsis in (19). Sag and Williams (1976), among others, point out that the pronoun in the VP ellipsis can serve as a bound variable. Therefore, the pronoun can be bound by either the subject in the first conjunct (19a), or the subject of its own (19b). It can also refer to the third-party (19c).

(19) John likes his father, and Bill does , too. a. John likes John’s father and Bill likes John’s father, too.

(Strict reading)

b. John likes John’s father and Bill likes Bill’s father, too.

(Sloppy reading)

c. John likes Chris’s father and Bill likes Chris’s father, too.

(Third-party reading)

The same holds for RNR. The pronoun his in (20) is also multiply ambiguous: sloppy (20a), strict (20b), and a third-party reading (20c).

(20) Johni likes, but Billj hates – hisi/j father. a. John likes John’s father, but Bill hates Bill’s father.

(Sloppy reading)

b. John likes Bill’s father, but Bill hates Bill’s father.

(Strict reading)

c. John likes Chris’ father, but Bill hates Chris’ father.

(Third-party reading)

12 The phenomena discussed above have been argued to be properties of forward ellipsis in literature. Thus, there is a good reason to believe that RNR is also an ellipsis phenomenon. On the other hand, RNR also has some unique properties, which distinguishes it from forward ellipsis, so let us examine the differences. Forward ellipsis is known to be able to draw its antecedent from outside its own sentence. For example, in (21), the antecedent of the elided VP go to the ball game is found in another sentence.

(21) Jill and her friends decide to go to the ball game this Saturday. Well, I’m not particularly interested in baseball, but I will , too.

On the other hand, (22) indicates that a VP outside the sentence boundary cannot be an antecedent for the RNR target in RNR.

(22) a. *After a long consideration, Jill and her friends decided to . Well, I’m not a huge fan of the baseball, so I’m not gonna – go to the ball game. b. *Mary used to LOVE . Well I’m sure that she’s changed her mind now, but Bill still LIKES – David Letterman show.

It has been assumed in ellipsis literature that only constituents can be elided, but it appears that the RNR target does not have to form constituents (Yoon and Lee 2005). 7 In a Korean 7

The typical sentences given to show that non-constituent ellipsis is impossible include examples like (i) (p.c. Kyle Johnson). Given that the counterpart RNR version is also ruled out in (ii), (i-ii) are not conclusive examples to show that constituency has to be obeyed in ellipsis, but not necessarily in RNR. In fact, I was not able to find any pair of examples that could tease apart RNR and forward ellipsis with respect to constituency. (i) a.*John considers Mary smart, but Sally doesn’t happy. b.*John didn’t stand near Sally, but Mary did Sue.

13 example (23), the embedded and the matrix VPs are RNRed. Since they do not form constituents, (23) would be predicted to be ungrammatical, contrary to fact, if we simply assume that RNR is the same as ellipsis.

(23) [Bill-un B.-TOP

[Mary-ka ] kuliko M.-NOM math-ACC well do-C

[John-un

[casin-i

J.-TOP

self-NOM math-ACC well do-C

think-PRES-DEC

CONJ

swuhak-ul cal hanta-ko] sayngkak-han-ta]. think-PRES-DEC

‘Bill thinks that Mary does math well, and Johni thinks hei does.’

This is one of the puzzles that I will try to solve in this dissertation. Constituency in this kind of example will be re-evaluated in chapter 5, where I will propose that RNR observe constituency as forward ellipsis does. In a nutshell, I will argue that there are multiple deletion processes in (23) so that the embedded and the matrix clause are deleted independently. RNR affects the entire right edge of the first conjunct whereas there is no such restriction in forward ellipsis. The constraint is known as the Right Edge Restriction in literature (Postal 1974, McCawley 1982, Swingle 1993, Wilder 1997, 1999, Hartmann 2000, Sabbagh 2007). Once RNR starts in the first conjunct, pronunciation cannot resume until the coordinator is reached (at least, in English).8 (24) is ungrammatical, since the VP adjunct in the first conjunct remains pronounced.

(ii) a. * Sally doesn’t happy, but John considers Mary smart. b. *Mary did Sue, but John didn’t stand near Sally. 8

In Korean, on the other hand, RNR targeting only a middle expression seems to be possible. Notice that in the Korean example in (i), only the object DP is elided in the first conjunct, and the sentence is acceptable. It is unlikely that (i) arises from the use of a null pronoun in object position, given that null pronouns are prohibited in other backwards contexts, such as (ii). (i) Bill-i sse-ess-ko, Mary-ka ku chayk-ul caymiisske ilk-ess-ta. B.-NOM the book-ACC write-past-CONJ M.-NOM the book-ACC joyfully read-PAST-DEC.

14 (24) *John BORROWED from the bank, and Bill STOLE a large amount of money from the bank.

To summarize, RNR exhibits both similarities to and differences from ellipsis. RNR is similar to ellipsis with respect to sloppy identity, Vehicle Change, and morphological mismatches, but RNR is different from ellipsis because it needs some peculiar constraint, such as the Right Edge Restriction. Continuing to argue that RNR is ellipsis, I will raise questions of whether RNR and ellipsis differ in the cases above. My conclusion will be the differences are just apparent.

1.2.2 Semantic properties For licensing RNR, there is some semantic requirement that needs to be satisfied; that is, the first conjunct containing the RNR site must be semantically identical with the second conjunct containing the antecedent. I will discuss semantic identity conditions for ellipsis and RNR in chapters 4 and 5. For the purpose of a brief introduction to semantic parallelism requirements, here I will just adopt Merchant’s (2001) elliptical-GIVENness, stating the two conjuncts must mutually entail each other by existential closure of focused constituents (i.e. F-closure). Let us take (25). In (25a), the subject and the verb are focused in each conjunct. By hypothesis, focused constituents are existentially closed (i.e. Focus-closure), and F-closure of each conjunct results in the same semantic formula, ∃x∃R [x R-ed the baseball]. The antecedent entails a F-closure of RNR, and RNR entails a F-closure of the antecedent, which satisfies the e-GIVENness condition. In (25b), the e-GIVENness condition is not satisfied. Let us see why. The F-closure of the antecedent is ∃x [x left], which the RNR clause does not entail. And the F-closure of RNR is ‘Bill wrote the book, and Mary enjoyed reading the book.’ (ii) *pro pro

chayk-ul ilk-ess-ko, Mary-ka sakwa-lul mek-ess-ta. book-acc read-past-conj M.-nom apple-acc eat-past-dec

15

∃x∃R [x R-ed Mary], which the antecedent does not entail. Therefore, the two conjuncts are not semantically identical, so RNR in (25b) is not licensed.

(25) a. JOHN CAUGHT , but BILL MISSED – the baseball. b. *JOHN LOVED , and MARY left.

Another crucial semantic requirement is that there must be a contrastive focus just prior to the RNR target (Hartmann 2000). In (26a), the verb likes as the pre-RNR element is contrastively focused, contrasted with dislikes in the second conjunct. If there is no contrast, the sentence is degraded in (26b).

(26) a. BILL LIKES , but MARY DISLIKES – the TV show. b. ?*BILL likes , and MARY likes – the TV show.

Note that Forward ellipsis differs from RNR in this respect. It does not require that the material preceding the ellipsis site bears contrastive focus, as shown in (27a). And its presence does not suffice to license ellipsis, as shown in (27b).

(27) a. Bill shouted loudly, and Chris did, too. b. *MARY HATES black coffee, but JOHN LIKES .

(Forward Ellipsis)

c. MARY HATES , but JOHN LIKES – black coffee.

(RNR)

What makes RNR possible appears to be due to contrastive focus. A crucial pair of examples demonstrating this premise is given in (28). The VP is elided in both (28a) and (28b), but it is

16 important to notice that only (28a) is possible. Meanwhile, the corresponding forward VP-ellipsis examples in (29a-b) are acceptable. Thus, the ungrammaticality of (28b) indicates that the licensing conditions of RNR do not follow from those of forward ellipsis. The difference will be discussed in chapter 5, and I will argue that contrastive focus is the licensor for RNR, whereas syntactic features in a functional head license forward ellipsis.

(28) a. JOHN COULDN’T , but MARY COULD – shout loudly. b. *JOHN COULD , and MARY shouted loudly, too.

(29) a. Mary could shout loudly, but John couldn’t b. Mary shouted loudly, and John could , too.

The importance of those semantic requirements leads us to adopt semantic licensing conditions for RNR in this dissertation, as many of the accounts for forward ellipsis do (Sauerland 1998, Merchant 2001, among others). There are, of course, complications to address under the hypothesis that RNR follows the same kind of semantic licensing conditions as forward ellipsis, since some considerable differences exist between them. Those are also the reasons why previous researchers have proposed that these are two different phenomena. In chapters 4 and 5, following Merchant’s (2001) e-GIVENness, I will propose various conditions under which an ellipsis feature for RNR can appear, which makes the deletion of the RNR target possible at PF, just like forward ellipsis.

17

1.2.3 Prosodic properties RNR has a unique prosodic structure. It is different from its overtly pronounced counterpart with respect to aspects of its prosodic structure, such as pitch contour, pauses, and pitch accents. Let us first compare the same sentence with and without pronouncing the RNR target and their pitch tracks in (30).

(30) a. Mariana drove the red car, and Jenny dented the red car. b. Mariana drove, and Jenny dented – the red car.

As shown in figures 1 and 2, there are marked differences between (30a) and (30b) in terms of phonological phrasing. First, the nuclear pitch accent is assigned on the verb prior to the RNR target, and the verb in the second conjunct is also accented in (30b). On the other hand, in (30a), the nuclear accent is on the object DP in the first conjunct and on the verb in the second conjunct because it is contrasted. Second, in both sentences, the boundary tone at the end of the first conjunct is rising. However, we observe that the high boundary tone is more prominent in RNR than in the non-RNR sentence. The boundary tone is so prominent that the first conjunct counts as an Intonational Phrase (IP) in (30b), but not necessarily for the non-RNR sentence.

18

L*

H

Mariana

L*

H

Mariana

L*H H* LHL*H L* H LL% Mariana drove the red car, and Jenny dented the red car. drove the red car and Jenny dented the red car Figure 1. Pitch contour of (30a)

L*H L*HHL*H L* H H* LL% Mariana drove, and Jenny dented the red car. drove and Jenny dented the red car Figure 2. Pitch contour of Right Node Raising (30b)

Third, as Féry and Hartmann (2005) generalize, the upstep-downstep pitch contour is observed in the RNR sentence (see figure 3). That is, the second IP is downstepped, and the phonological phrases inside are also downstepped. Féry and Hartmann (2005) argue that this specific phonological phrasing signals the hearer that there is a missing material in the right edge of the first conjunct. The phonological phrasing discussed here will be revisited in chapter 4.

19

First IP Second IP ` First PhP Second PhP Figure 3. The upstep-downstep pitch contour in RNR (Féry and Hartmann 2005: 102)

Two other empirical facts related to the prosodic structure for RNR are worth mentioning here. Abbott (1976) and Swingle (1993) observe that the RNR target needs to form its own IP (cf. Féry and Hartmann 2005). If the target is too weak to form an independent IP, RNR cannot be constructed. In (31), although the RNR constructions satisfy syntactic or semantic requirements, they are not acceptable RNR examples for an independent reason. Since the pronoun and the demonstrative cannot stand alone as an IP, RNR is not licensed.

(31) a.*ALICE COMPOSED , and JOHN PERFORMED – it b. *MARY BOUGHT , and FRED STOLE – that. (Swingle 1993: 87)

The prosodic structure of (31a-b) is drawn in (32a-b). Notice that the RNR construction is acceptable when the content words are used for the RNR target, so that it can form its own IP in (32c).

20 (32)

IP qp Φ Φ C C W

W

IP ei Φ Φ C C 2 W W W performed

IP

W

a. *Alice

composed

and

John

it.

b. *Mary

bought

and

Fred

stole

that

c. John

bought

and

Fred

stole

the car (Swingle 1993: 94)

Following McCawley (1988), Swingle (1993) also notes that the pre-RNR constituent cannot be a function word.9 (33a-b) are ruled out, on the grounds that the pre-RNR word in the first conjunct is an indefinite article and the one in the second conjunct is a possessive, and that modals, as a contracted form, cannot be stranded in the pre-RNR position.

(33) a. *Ted has always wanted A , so I’ve given him MY – coffee grinder. b. *I think I’D , and I know that John’LL – buy one of those portraits of Elvis.

Although my consultants agree with the judgment of (33), it does not seem to turn on a prohibition of function words in the pre-RNR position. When the indefinite article is replaced by a possessive contrasting with the one in the second conjunct, grammaticality is improved, as shown in (34a). Similarly, if the modals are not contracted, the result is fully grammatical, as shown in (34b). The grammaticality of (34a-b) indicates that the pre-RNR constituent must be

9

‘Pre-RNR element’ is used to refer to the constituent prior to the RNR target in this dissertation.

21 able to be contrastively focused, and that neither an indefinite article nor the contracted form of modals can be focused.

(34) a. ?Ted has always wanted MY , and I’ve always wanted HIS – coffee grinder, so we will exchange it. b. I think I COULD , and I know John WILL – buy one of those portraits of Elvis.

It is interesting to compare (33) with ellipsis. Contracted auxiliaries and modals cannot be stranded in VP-ellipsis (35a-b), but their full-fledged forms can be followed by VP-ellipsis (35cd). This suggests that contraction prior to deleted materials is generally prohibited as a prosodic restriction. However, unlike in RNR, the element preceding the ellipsis site needs not be focused in forward VP-ellipsis. Thus, it is not focus that makes forward VP-ellipsis possible, while focus does matter for RNR.

(35) a. *I think I’d buy one of those portraits of Elvis, and I know that John’ll, too. b. *Kathy is going to need a new car, and Bill’s, too. c. I think I’d buy one of those portraits of Elvis, and I know that John will, too. d. Kathy is going to need a new car, and Bill is, too.

Another difference is that a possessive pronoun does not license the elision of its complement NP in forward ellipsis. Let us consider (34a) and (36). Compared with (34a), the possessive my in (36) cannot license NP-ellipsis (Lobeck 1995). This indicates that there is a difference in the importance of prosodic cues between forward ellipsis and RNR. The contrastive focus is enough

22 to license and signal RNR, but this is not the case for forward ellipsis. This difference will be discussed in chapter 5.

(36) *I’ve always wanted Ted’s coffee grinder, and he has always wanted my.

1.2.4 Pragmatic properties As mentioned earlier, RNR is a relatively rare construction in language. One reason could be that RNR is a processing burden for both speakers and hearers. Natural language is produced and comprehended in a left-to-right fashion, so the previous utterance in discourse serves as background information for the following sentences.10 Forward ellipsis is more natural in this respect, since the previous discourse provides the antecedent for the ellipsis site. However, RNR is different. RNR is a burden on both interlocutors because the speaker needs to bear the whole second conjunct in mind even as s/he produces the first conjunct, and because the hearer needs to wait until the end of the second conjunct to be able to assign an interpretation to the first conjunct. A natural question to ask is what kinds of information a speaker conveys with RNR that justifies this processing burden. I suggest that RNR plays two important roles in discourse. First, RNR is used to optimize contrast in coordination. What the speaker can accomplish with an RNR construction is pronunciation of only (contrastively) focused materials in conjuncts and a minimization of repeated material, which is the RNR target. With an RNR construction, the hearer can better receive the speaker’s intention of contrast between the conjuncts. Second, RNR is sometimes adopted in order to maintain discourse coherence. Let us consider a sentence in (37). (37) is ill-

10

I use “left” here to refer to earlier and “right” to refer to later, without regarding the orthographic conventions of the language (which could be different, as in Hebrew).

23 formed, but the oddity is not due to its syntactic structure. Rather, it is repetition of identical material in both conjuncts that significantly reduces discourse coherence.

(37) #JOHN WATCHED the new movie, but MARY LIKED the new movie.

Normally, pronominalization of one of the arguments is utilized as in (38a). I argue that RNR can be another option to enhance discourse coherence as in (38b).

(38) a. John watched the new movie but Mary liked it. b. JOHN WATCHED, but MARY LIKED – the new movie.

Unlike pronominalization, RNR can accomplish a parallel focus structure. Notice that the subject and the verb in the parallel position of each conjunct needs to be contrastively focused for RNR to be constructed, which in turn allows the speaker to achieve maximizing contrast. To summarize, I suggested that RNR is used to maximize the contrast effects in the coordinate structure at the expense of incurring some processing burden and RNR also helps the discourse be more coherent.

1.3 Overview The dissertation is structured in three parts with six core chapters. The first part, consisting of chapters 2 and 3, concerns the syntactic representation of RNR. The second part, chapters 4 and 5, is a study of the semantic licensing conditions for RNR. The third part, chapters 6 and 7, provides extensions of the ellipsis analysis of RNR to ATB constructions.

24 The main concern of chapter 2 is to review the previous accounts of RNR. The discussion in chapter 2 begins with the question of whether the RNR position in the first conjunct is a null pro, or whether it is a full-fledged unpronounced syntactic representation. I will review the debates on the question as they were applied to VP-ellipsis and examine how each account fits in with RNR. Then, in sections 2.1-2.2, we move on to the derivational approaches of RNR. I will lay out three contemporary accounts of RNR and evaluate each of them: movement, strict phonological deletion, and multiple dominance. In chapter 3, I discuss non-movement analyses in more detail and argue that they fail to capture the empirical data. As an alternative, I propose a new non-movement analysis of RNR as a variant of ellipsis. I argue that the previously problematic empirical evidence can be best explained by the ellipsis account. In section 3.2, I will also discuss challenges (potentially) faced by the ellipsis account, and attempt to resolve them. Chapter 4 concerns the semantics of RNR – the licensing conditions for RNR. Since I will have established an ellipsis account for RNR in the previous chapter, here I discuss what licenses RNR, and whether RNR observes the same licensing conditions for ellipsis. The chapter begins with licensing conditions for forward ellipsis. In sections 4.1-4.2, two competing analyses for forward ellipsis will be introduced. We focus on Rooth’s (1992b) account, which is based on structural isomorphism, and Merchant’s (2001) e-GIVEN account, which is based on semantic identity. I adopt Merchant’s account which convincingly shows that structural isomorphism is incorrect. In section 4.3, I will review the previous analysis which makes a proposal for licensing conditions for RNR (Hartmann 2000, Féry and Hartmann 2005). This previous account is based on syntactic isomorphism, and I will provide evidence that it encounters both under- and overgeneration problems for RNR as it does for VP-ellipsis.

25 Chapter 5 is the most important chapter in this dissertation, in which I will propose a new analysis of RNR in terms of an ellipsis feature. In section 5.1, I will provide background for Merchant’s (2001, 2004) E feature analysis for sluicing and fragment answers. And then I propose an E-feature for RNR, which I dub the ERNR feature. ERNR has syntactic, semantic, and phonological requirements, all of which need to be satisfied for the deletion of the RNR target to be licensed. In syntax, the lexical item containing contrastive focus can bear the ERNR feature in the first conjunct, and the sister of the feature is determined as the RNR target. In semantics, a mutual entailment relationship must be established between the conjuncts. In phonology, the deletion must initiate immediately following the ERNR feature to the right edge of the first conjunct. In section 5.2, I will discuss various consequences of the ERNR analysis: i) RNR and deaccenting would follow from the different strength of e-givenness conditions, ii) the ellipsis properties, such as sloppy identity, Vehicle Change, and morphological mismatches, can be explained without additional assumptions. In section 5.3, I will discuss cases where the RNR target can be different from the string that is actually deleted. In section 5.4, we examine whether RNR is constrained by constituency, and I argue that it is. I propose that there can be multiple ERNR features in a sentence, and the highest feature determines the constituent that will serve as the RNR target. In chapter 6, I will review accounts of ATB constructions, which have been taken to be an exception to the Coordinate Structure Constraint (CSC, Ross 1967). In section 6.2, I will review evidence that poses challenges for Ross’ assumption about ATB constructions. To resolve these problems, there have been three alternative accounts: sideward movement, parasitic gap, and multiple dominance, each of which will be reviewed in section 6.3. In chapter 7, I will propose a new analysis of ATB constructions. In section 7.2, I argue that the underlying structure of ATB constructions is RNR, and the RNR target in the second conjunct moves out of the coordinate structure successive cyclically. This analysis is called the RNR

26 analysis of ATB constructions. Under this analysis, an alternative view of the CSC needs to be exploited. I adopt Fox’s (2000) proposal that the CSC is not a movement constraint, but a condition on the LF representation. The consequences of the RNR analysis will be discussed in section 7.3. The RNR analysis will capture both paired and single-identity readings in answers to ATB wh-questions, and I will show evidence that movement comes from the second conjunct, from selectional properties, crossover effects, island effects, and binding effects. In section 7.4, the RNR analysis will be compared with two previous accounts which make directly opposite proposals: (i) parasitic gap analyses, under which the movement comes out of the first conjunct (Munn 2001), and (ii) multiple dominance accounts, which assume a single copy is shared by both conjuncts, whereas the ellipsis account supposes that there are two copies of the ATB constituent. In section 7.5, we will revisit LF ATB constructions and observe that ATB QR is possible. I take a new perspective on the nature of ATB constructions, arguing that ATB is a type of strategy by the computational system to obtain a single copy of ATB constituents for a singleidentity interpretation. We will observe some cases where ATB constructions are derived without any overt movement. Section 7.6 is concerned with some exceptional ATB constructions that RNR is not the underlying structure of. I will argue that subject ATB constructions are among them. Specifically, I propose that the subject is base-generated beyond the coordinate structure. In section 7.7, I suggest that gapping should be analyzed as ellipsis, contra Johnson’s (1996) ATB analysis. In section 7.8, I discuss exceptions allowed by some specific information structures.

27

CHAPTER 2 SYNTAX OF RIGHT NODE RAISING

In this chapter, we are concerned with how RNR is represented in syntax. Ultimately, I will argue that RNR is an ellipsis phenomenon, so we will first turn to look at accounts of ellipsis. There have been at least two kinds of accounts proposed in the forward ellipsis literature: a derivational approach and a pro-form approach. After providing background on this representational issue in forward ellipsis, I will provide evidence arguing for a derivational approach in RNR. The derivational approach assumes that full-fledged syntactic structure is represented in the elided position at least in a certain syntactic level (either at PF or at LF). On the other hand, the pro-form approach supposes that there is no internal syntactic structure present in the elided site, but instead a null pronoun there (Wasow 1972, Williams 1977, Chao 1987, Hardt 1992, 1993, Lobeck 1995). Currently most literature leans toward the derivational approach, criticizing the pro-form analysis for the following reasons: i) missing antecedent phenomena, and ii) different distributions between ellipsis and anaphora. I present each argument in a brief manner in this chapter. For more detailed arguments for the derivational approach, see Johnson (2001) for VP ellipsis and Merchant (2001) for sluicing. The pro-form approach treats the ellipsis site as a pronoun, of which the phonological content happens to be empty. Thus, this analysis predicts that the ellipsis site would behave the same as an overt pronoun. Let us first consider the do it anaphora in (1a). The antecedent for the anaphora is buy anything for Christmas, which includes a Negative Polarity Item (NPI). The pronoun it in

28 the last conjunct needs an antecedent, but the NPI cannot provide one for the pronoun, so a missing antecedent problem arises. Thus, (1a) is ungrammatical. The pro-form analysis predicts that the null anaphora would show the missing antecedent phenomenon in (1b), too, given the assumption that the null pro is a kind of anaphora. However, (1b) is grammatical, so the prediction is not borne out.

(1)

a. *My uncle didn’t buy anything for Christmas, but my aunt did it for him and it was bright red. b. My uncle didn’t buy anything for Christmas, but my aunt did pro and it was bright red. (Bresnan 1971: 591)

On the other hand, if we assume that there is a derived vP structure unpronounced in the ellipsis site as the sentence in (2), the missing antecedent problem does not arise. The elided portion may be reconstructed with a Positive Polarity Item (PPI) something (Klima 1964), which in turns allows the pronoun it to find its antecedent something in the ellipsis site. How this reconstruction is possible in VP-ellipsis will be elaborated in section 4.2. In a nutshell, an antecedent VP containing an NPI licenses the elision of the VP containing a PPI, because an NPI and a PPI have the same semantic denotation. We will assume that elided material must establish semantic identity with its antecedent; that is, the VPs of both conjuncts need to mutually entail each other, modulo ∃-type shifting (Merchant’s (2001) e-GIVENness condition). (2a-b) are the formulas of the first two conjuncts after existential closure takes place. Since both an NPI and a PPI are existential quantifiers, they are not distinct. (2a) and (2b) are identical and mutually entail each other. Therefore, VP-ellipsis is possible and the pronoun can find its antecedent in (2).

29 (2)

My uncle didn’t [vP1 buy anything for Christmas], but my aunt did [vP2 ] and it was bright red. a. VP1: ∃x∃y[x buy y for Christmas] b. VP2: ∃x∃y[x buy y for Christmas]

The missing antecedent problem arises in RNR, under the pro-form analysis. Let us consider (3). (3a) is ungrammatical since the pronoun it in the last conjunct does not have an antecedent. The NPI fails to provide one for the pronoun. The analysis also predicts (3b) to be ungrammatical, contrary to fact, since the pro occupying the ellipsis site does not have an antecedent.

(3)

a. *My uncle DID it, but my aunt DID’T buy anything for Christmas, and it was bright red b. My uncle DID pro, but my aunt DID’T buy anything for Christmas, and it was bright red.

The grammaticality of (3b) is explained under the derivational account. The NPI in the ellipsis site may be reconstructed with a different morphological form (i.e. PPI), as in (4). And the elision of the RNR target in the first conjunct is licensed because the NPI and PPI mutually entail each other. Now the pronoun it finds an antecedent in something. Therefore, the ellipsis site plays an important role for bridging the overt antecedent and the pronoun.

(4)

My uncle DID , but my aunt DIDN’T buy anything for Christmas, and it was bright red.

Besides the missing antecedent phenomenon, the contrast in (5) also demonstrates different distributions between RNR and anaphora. Notice that (5a) is acceptable but (5b) is not, a fact that

30 would be hard to capture under the pro-form analysis. Therefore, the contrast tells us that the RNR position in (5a) has different properties from overt anaphora.1 Under the derivational view, the unacceptability of (5b) is attributed to vacuous quantification of the wh-movement in the first conjunct since the moved wh-phrase does not leave a variable to bind at LF. Given that the same problem does not occur in (5a), it indicates that there is an internal movement of wh-phrase in the elided position.

(5)

a. I remember which articlej Chomsky DID , and which articlei he DIDN’T – submit ti to LI. b. *I remember which article Chomsky DID IT/SO, and which article he DIDN’T submit to LI.

Also, notice that the article that the first and the second conjunct refer to might differ from each other. This is only possible if there exists a variable bound by each wh-phrase. Under the derivational approach, this empirical fact is easily captured in that the ellipsis site contains internal syntactic structure, and wh-movement must occur in each conjunct. Since wh-movement occurs independently in both conjuncts, the variable bound to each wh-word has a different index (6a).

(6)

a. [which articlej Chomsky did ], and [which articlei he didn’t submit ti to LI] b. *[which articlej Chomsky did pro], and [which articlei he didn’t submit ti to LI]

1

I assume backwards (VP-)ellipsis is RNR. Refer to section 1.1 for the detail.

31 On the other hand, if there is no internal structure in the ellipsis site, as the pro-form analysis asserts, the configuration would be (6b). Notice that no variable has been left in the first conjunct, so that there are two wh-operators and one variable. Each variable must be bound by a single operator (according to the Bijection Principle, Koopman and Sportiche 1983). The wh-operator in the first conjunct has no variable to bind at LF, which incurs a vacuous quantification violation. To sum up, the same evidence that supports the derivational approach over the pro-form approach in ellipsis literature can also be found for RNR examples. The contents of the remainder of this chapter are the following. In sections 2.1-2.2, I will review the derivational approaches for RNR proposed in the previous literature – movement, strict phonological deletion, and multiple dominance – and argue that those previous analyses encounter either conceptual or empirical problems. In the next chapter, I will propose an alternative analysis, which I call the ellipsis account, and attempt to resolve the problems.

2.1 The movement analysis There have been largely two types of derivational accounts of RNR in current literature, differing with respect to the role played by syntactic movement: movement and non-movement analyses. Traditionally, RNR has been treated as a situation in which the identical constituent in the first and the second conjuncts undergoes rightward Across-The-Board (ATB) movement (Ross 1967, Hankamer 1971, Marling 1972, Bresnan 1974, Postal 1974, 1998, Abbott 1976, Grosu 1976, Hudson 1976, Williams 1978, Goodall 1983, Sabbagh 2003, 2007), as in (7). Let us call this the movement analysis. In (7), the RNR constituent every student moves out of the coordinate structure and lands in a higher position.

32 (7) Kim bought an apple for, and Jane gave it to – every student. &P qp &P DP qp 6 CP1 & CP2 every student 6 6 Kim bought Jane gave it to [t] an apple for [t]

2.1.1 Arguments for movement Arguments for the movement analysis are related to the rightward moved constituent scoping over both conjuncts. Therefore, the arguments show that the RNR target containing a quantifier can take wide scope over quantificational material inside the coordinate structure or the target bearing a relational modifier makes distributive interpretation available in RNR. Let us first consider when the RNR constituent contains a relational modifier. Examples of relational modifiers are different, same, similar, or respectively, all of which can yield distributive readings when they take scope over the coordinate structure. Let us consider (8). It is interesting to note that (8b) allows an interpretation that the underlying sentence (8a) does not, namely a distributive reading where Peter sang a song and Mary recorded a song, and the two songs are the same (Abbott 1976, Jackendoff 1977, Gazdar 1981). Rightward movement enables the modifier to take scope over the two predicates in coordination and permits a distributive reading.

(8)

a. Peter sang the same tune, and Mary recorded the same tune. b. Peter sang, and Mary recorded – the same tune.

33 Similarly, we observe that the sentence in (9) allows a distributive reading where John sold gold rings from South Africa and Mark bought raw diamonds from South Africa. This interpretation is straightforwardly captured when we assume that the relational adverb respectively scopes over the moved target complex NPs and distributes over each conjunct.

(9)

John sold, and Mark bought – gold rings and raw diamonds from South Africa, respectively.

These two examples in (8-9) indicate that to obtain a distributive reading, the relational modifiers must move out of the conjuncts and take scope over the coordination, and that this can be done via rightward movement. Another argument for the movement account comes from different scope-taking possibilities for quantifiers between RNR cases and their underlying structures (Sabbagh 2003, 2007). Let us compare (10a) with (10b). Sabbagh claims that inverse scope (i.e. different nurses for different patients) is possible in (10a), but not in (10b).

(10) a. Some nurse gave a flu shot to, and administered a blood test for – every patient who was (∃>∀, ∀>∃)

admitted last night.

b. Some nurse gave a flu shot to every patient who was admitted last night, and administered a blood test for every patient who was admitted last night.

(Only ∃>∀)

(Sabbagh 2007: 366)

Under the movement analysis, inverse scope in (10a) can be easily explained since the universal quantifier moves outside the conjunct where it can take wide scope over the existential quantifier. On the other hand, without movement in (10b), inverse scope is predicted to be unavailable,

34 assuming that covert ATB movement is prevented (Bošković and Franks 2000). 2 Thus, only surface scope is possible in (10b). Sabbagh argues that non-movement analyses, where the RNR constituent does not undergo overt movement, cannot account for the scope ambiguity of (10a) since these accounts must assume (10b) is an underlying structure of (10a). If the target object is deleted, rather than moved, we expect (10a) to have only surface scope, contrary to fact. Sabbagh (2003, 2007) makes a similar argument using Antecedent Contained Deletion (ACD). Let us examine (11). Assuming a rightward ATB movement account for RNR, he claims that (11) can have two antecedent constructions, where the embedded VP is an antecedent in (11a), or the matrix VP including try is an antecedent in (11b). To resolve the ACD, he assumes that the RNR target containing the ellipsis site must raise to a position where it can scope over its antecedent, meaning adjoining to the matrix CP (cf. Baltin 1987, Larson and May 1990, May 1985).

(11) The doctor tried to give a flu shot to, and administer a blood test for – every patient that the nurse did e. a. e = [[give a flu shot to t], and [administer a blood test for t]]. b. e = [tried to [give a flu shot to t], and [administer a blood test for t]]. (Sabbagh 2007: 367)

The important case here is the availability of the interpretation (11b). According to Sabbagh, non-movement analyses can account for the interpretation (11a), but not (11b). Let us examine 2

Bošković and Franks (2000) claim that ATB movement is not possible at LF by showing the lack of widescope reading of the example (i). (i) Every scientist reviewed some article, and every journalist critiqued some article. ≠ There is some article x, such that every scientist reviewed x and every journalist critiqued x. (*∃>∀)

35 why this is so in (12), which is the underlying structure of (11) before movement. In (12), the covert movement in the ACD construction occurs in each conjunct, so each ellipsis site (e1 and e2) is part of an independent covert movement. Ellipsis is known to be subject to parallelism constraints. Since the only possibility that e2 can take for its antecedent is administer a blood test for t, e1 must take the parallel antecedent, give a flu shot to t. Thus, the interpretations (12a-b) are obtained, which is the same as (11a).

(12) The doctor tried to give a flu shot to every patient that the nurse did e1, and administer a blood test for every patient that the nurse did e2. a. e1 = give a flu shot to t, (*and administer a blood test for t.) b. e2 = administer a blood test for t, (*and give a flu shot to t.)

Let us now turn to whether the in-situ analyses can capture the interpretation (11b). Notice that the coordination only includes the embedded clauses in (11b). To scope over the matrix verb, the ACD constituent must undergo covert ATB movement adjoining to the matrix CP, which we have already noted to be impossible.3 Therefore, the only possible interpretation for (11) under non-movement analyses is (11a). Then, the non-movement analyses need to account for why the elided version (13) allows scope ambiguity, which is not possible in its underlying (12). In chapter 7, I will propose that an ellipsis account does make scope ambiguity possible, and contra Bošković and Franks (2000), I will show that ATB movement at LF is possible.4

(13) The doctor tried to give a flu shot to , and administer a blood test for – every patient that the nurse did e2. 3 4

See footnote 2 in this chapter. See sections 7.5.1-7.5.2 for the detailed discussion.

36

2.1.2 Argument against movement Despite the advantages of the movement analysis, there has been a fair amount of criticism of this account. The major problems discovered in the previous literature can be characterized as involving four issues; a) preposition stranding, b) island insensitivity, c) the Right Roof Constraint, and d) constituency.

2.1.2.1 Preposition stranding Bošković (1997/2004) argues that RNR needs to be distinguished from Heavy NP shift (HNPS). He points out that HNPS cannot strand a preposition in English. Let us compare (14a-b). While the target NP can be dislocated in (14a), it cannot strand the preposition as in (14b).

(14) a. Mary criticized t last week the paper you presented at the LSA. b. *John will talk about t next weekend the paper you presented at the LSA.

With this information, let us first consider HNPS cases under coordination in (15). (15a) is the underlying structure before the target NP is shifted. Bošković (1997/2004) introduces two possibilities to derive HNPS under coordination. The first possibility is that HNPS occurs in an ATB fashion; the target NP is dislocated out of both conjuncts simultaneously, as shown in (15b). The second possibility is that the target NP moves only out of the second conjunct. The one in the first conjunct is deleted by the second conjunct under identity, as shown in (15c).5 HNPS would leave the preposition in the second conjunct stranded, which makes (15b-c) ungrammatical.

5

Bošković (1997/2004) chooses the second option on the grounds that (i) is improved, compared to (15b). Since there is no movement involved in the first conjunct, the preposition is not stranded in (i). Therefore, the differences in acceptability can be explained.

37 (15) a. Mary criticized the paper you presented at the LSA last year and John will talk about the paper you presented at the LSA tomorrow. b. *Mary criticized t last year and John will talk about t tomorrow – the paper you presented at the LSA. c. *Mary criticized last year, and John will talk about t tomorrow –the paper you presented at the LSA.

Let us now turn to RNR. If movement were involved in RNR, we would predict that (16) is not possible. This is so because rightward movement out of the first (16a) or the second (16b) conjunct would strand the preposition. Contra the prediction, the sentences in (16) are grammatical. These facts crucially indicate that movement is not involved in (16a-b).

(16) a. Mary criticized t and John talked about t – the paper you presented at the LSA last year. b. Mary talked about t and John criticized t – the paper you presented at the LSA last year.

Furthermore, if there were movement in the second conjunct (16), it would be string vacuous movement, as shown in (17a).6 To avoid vacuous movement, the DP should move out of the whole conjunct as rightward HNPS.

(i) ?Mary talked about , and John will discuss t tomorrow – the paper you presented at the LSA. These interesting differences will be recast in chapter 7. I will propose that ATB constructions are derived in a similar manner. 6 The unavailability of string vacuous movement can be tested with a quantifier scope. Let us consider (i). (i) John [wrote a paper that criticizes t] every idea I’ve ever come up with.

(*∀>∃) (p.c. Danny Fox)

If string vacuous movement were possible, the universal quantifier would be able to take wide scope in (i), contrary to fact.

38 (17) Mary criticized t, and John talked about t [the paper you presented at the LSA] yesterday. a. *[… t … TP], and [… t …

TP]...[the

paper you presented at the LSA] yesterday TP].

b. [… t … TP], and [… t … tomorrow TP] [the paper you presented at the LSA]&P].

A movement analysis of RNR requires that preposition stranding be allowed in this construction even in languages that otherwise disallow it. According to McCloskey (1986), Irish is not a preposition stranding language, so extraposition cannot strand a preposition, as in (18). However, RNR allows preposition stranding, as shown in (19), which is puzzling for the movement analysis.

(18) *Bhí mé a éisteacht le

inné

Was I listen(prog) with yesterday

[DP clár

mór fada ar an ráidió faoin toghachán]

program great long on the radio about-the election

‘I was listening yesterday to a great long program on the radio about the election.’

(19) Nil sé in aghaidh an dlí a thuilleadh a bheith ag éisteacht LE Is-not it against the law anymore

be(-fin) listen(prog) with

nó ag breathnu AR – [DP ráidió agus teilifís an Iarthair] or look(prog)

on

radio and television the West(gen)

‘It is no longer against the law to listen, or to watch, Western radio and television.’ (McCloskey 1986: 184-185)

2.1.2.2 Island insensitivity The examples in (20) show that RNR is insensitive to various island constraints. (20a) is a whisland, (20b) is a complex NP island, and (20c) is an adjunct island. If RNR arises from a

39 movement operation, it is not ordinary movement, given that ordinary movement is subject to island constraints.7

(20) a. Susan wonders when John ordered t, and Bill wants to know when he returned t – the tickets for the opera.

(Wh-island)

b. John likes a professor who lectured on t, and Mary likes a graduate student who debunked t – a recent theory of Right Node Raising.

(Complex NP island)

c. Josh was happy after he heard of t, but Willy got angry after finding out about t – the news that the food chain will no longer carry live lobsters.

(Adjunct island)

2.1.2.3 Right Roof Constraint Rightward movement has been argued to obey a bounding restriction, which is called the Right Roof Constraint (Ross 1967, Akmajian 1975, Johnson 1986, Rochemont 1992, McCloskey 1999). The Right Roof Constraint states that rightward movement is allowed only as far as the edge of the first cyclic node (= vP or PP). For example, the sentences in (21) are ungrammatical due to a violation of the Right Roof Constraint. Assuming that the adjunct drunk adjoins to TP, the rightward movement of the DP is out of the cyclic node boundary in (21a). And the DP moves out of PP in (21b), so ungrammaticality results.

7

Sabbagh (2007) notes that RNR, at least, observes the Coordinate Structure Constraint (CSC).

(i) *John left cookies t, and Mary left chocolates t – and candies. Whether the CSC should be analyzed as a constraint on movement is under debate (Ross 1967, Fox 2000, among others). This issue will be discussed in chapter 7. In brief, following Fox (2000), I will argue that the CSC is not a constraint on movement per se.

40 (21) a. *Max [vP described t for Bill] drunk, a popular Broadway musical. b. *Jamie walked [PP into t] suddenly, the dean’s office.

(vP) (PP) (Sabbagh 2007: 350)

Let us consider RNR in terms of the Right Roof Constraint. If the Right Roof Constraint is a constraint on all rightward movement, it should also constrain RNR. Any movement analysis of RNR would necessarily violate the Right Roof Constraint since the RNR target in the first conjunct, at least, moves across bounding nodes. In (22), for example, the RNR target moves beyond the vP in the first conjunct, which should constitute a violation of the Right Roof Constraint. Thus, the acceptability of an example like (22) indicates that there is no rightward movement involved in RNR.

(22) Billie [vP reserved a plane ticket for t], and Mary [vP fetched a porter for t] – their boss

2.1.2.4 Non-constituents Non-constituents also seem to be allowed in RNR, which poses a challenge to the movement analyses (Abbott 1976). Let us consider a Korean example in (23). Here part of the embedded clause oa-se ‘come-because’ and part of the matrix clause cip-e ka-ess-ta ‘went home’ would need to undergo movement, under a movement analysis. And they do not form a constituent. If the movement analysis were correct, we would be forced to say that those non-constituents can move.

41 (23) [AJ-nun [TP Mary-ka ], AJ-TOP

M.-NOM come home-to go-PAST-DEC

kuliko [Sue-nun [TP John-i oa-seo] cip-e kas-ess-ta.] CONJ

S.-TOP

J.-nom come

home-to go-PAST-DEC.

‘AJ went home because Mary came, and Sue went home because John came.’

To sum up, the examples above do not show some properties that normal movement has been argued to observe, and these are major reasons that movement analyses have been considered incorrect in literature. In the next subsection, I will introduce Sabbagh’s (2007) solutions for the problems. In particular, we will see Sabbagh’s treatments of the Right Roof Constraint effects and island insensitivity, which are based on Fox and Pesetsky’s (2005) cyclic linearization hypothesis.

2.1.3 Sabbagh’s (2007) cyclic linearization Sabbagh (2007) claims that rightward movement is in fact unbounded, and unconstrained by the Right Roof Constraint. Adopting the successive cyclic spellout mechanism proposed by Fox and Pesetsky (2005), he claims that as long as rightward movement does not contradict the linear order established in each spellout domain, the traditional movement constraints, such as islands and the Right Roof Constraint, can be violated.8 Let us first consider how Fox and Pesetsky’s system works.

8

Following Chomsky’s original definition, Fox and Pesetsky use the term “spellout domain” to refer to the constituents that are sent to PF interface and assume that spellout domains include CP, VP and DP. However, while Chomsky defines spellout domain as the complement of the phase head, they do this in a different way: When the derivation reaches a spellout domain D, D is linearized. Under Chomsky’s definition of spellout, what is linearized would be a complement of the head of D. This difference will play an important role in the critique of Sabbagh’s account later in this chapter.

42

2.1.3.1 Fox and Pesetsky’s cyclic linearization Fox and Pesetsky (2005) propose that linearization applies by spell-out domain, and crucially that the linear ordering cannot contradict orderings established in previous spell-out domains. Fox and Pesetsky’s cyclic linearization proposal is summarized in (24).

(24) Cyclic Linearization (Fox and Pesetsky 2005) a. Certain syntactic domains (e.g. vP and CP) are spell-out domains, where linearization applies (which is relatively compatible with Chomsky’s phases). b. The linear ordering of syntactic units is affected by Merge and Move within a spellout domain, but is fixed once and for all at the end of each spell-out.

Let us consider what (24) entails with a linearization of (25). Given the assumption that vP and CP are spellout domains, the first spellout domain to be considered is vP. In (25a), X and Y are merged within vP, and linearization of the vP spellout domain fixes X as preceding Y. In (25b), Z is merged beyond vP. When CP is spelled out, Z is linearized preceding vP, so Z precedes X and Y. If Y then moves to precede Z, an ordering contradiction results, as shown in (25c). The fact that Y precedes vP in the CP spell-out domain contradicts the ordering statement that X precedes Y in the vP spell-out domain. What if an element at the left edge of vP moves out? If X moves from the edge of vP to CP domain, as shown in (25d), the next spellout domain (= CP) will have the linearization ‘X precedes Z and Z precedes vP’. Here this ordering is not contradictory with the spellout of vP since X
43 (25) a. [vP X Y]:

X
b. [CP Z [vP X Y]]:

Z
c. *[CP Y1 Z [vP X t1]]:

Y
(Ordering contradicted)

d. [CP X1 Z [vP t1 Y]]:

X
(Ordering preserved)

Fox and Pesetsky demonstrate how their analysis can capture the fact that object shift in Scandinavian is possible only when elements that had preceded the object in vP still precede the object after it has shifted (cf. Holmberg 1999). This means that the verb must move and thus precede the object in CP spell-out domain, as shown in (26a). If the verb does not move, then object shift would be blocked since an ordering contradiction would result in (26b).

(26) Object Shift blocked by (unmoved) verb intervener

a. [CP Jag kysste henne inte [vP tV tO]]. kissed her

not

b. *[CP Jag har henne inte [vP kysst tO]]. I have her

(V
not

(*V
kissed (Fox and Pesetsky 2005: 19)

2.1.3.2 Cyclic linearization of RNR Sabbagh (2007) applies Fox and Pesetsky’s analysis to RNR. He defines a locality condition (for rightward movement) in terms of the linear order preservation: that is, locality is observed as long as the ordering is preserved amongst spellout domains. Bearing this assumption in mind, let us consider how island insensitivity and the Right Roof Constraint effects can be explained.

44 As discussed previously, a sentence, like (27), would constitute a typical complex NP violation if the RNR target is raised out of the original position, and such a case has been a problem for movement analyses of RNR. Under Sabbagh’s proposal, however, the island insensitivity can be explained because the rightward movement of the RNR target would not bring out an ordering contradiction between spellout domains. Let us follow his proposal in more detail. The problem for moving out of complex NP islands is that a DP as a spellout domain is linearized without providing an escape hatch for movement. Let us consider (27). The wh-phrase moves successive cyclically from the object position of vP1 to the matrix SpecCP and linearized by spellout domain, as shown in (27a-e). Notice that the wh-phrase cannot move to the specifier position of DP (i.e. islands), so the linearization of DP spellout domain would be (27c). In the next spellout domain (27d), when the wh-phrase moves to the edge of vP2 and is linearized, the preservation of linear order would be violated, because here the wh-phrase precedes the rumor.

(27) *[CP2 Who did John [vP2 _ hear [DP the rumor [CP1 _ that Mary will [vP1 _ get married to _]? a. vP1 = who>get>married>to b. CP1 = who>that>Mary>will>get>married>to c. DP = the>rumor>who>that>Mary>will>get>married>to Ordering contradiction d. vP2 = who>hear>the>rumor>that>Mary>will>get>married>to e. CP1 = who>did>John>hear>the>rumor>that>Mary>will>get>married>to

However, Sabbagh claims that moving something rightward from the right edge of an island is different because it would not influence the linear order. Let us consider how this works with an example in (28), where a wh-island exists in each conjunct. He assumes that each conjunct is

45 linearized first, and then the RNR target would move rightward out of the conjuncts in an ATB fashion.

(28) Josh will find out who needed t, and Bill will find out who sold t – the camera. a. Josh will [find out [who [needed t vP] b. Bill will [find out [who [sold t vP]

CP]

CP]

t’’ vP] the camera

t’’ vP] the camera

The first conjunct is spelled out in the following manner. The RNR target moves to the right edge of the embedded vP, and the vP is spelled out. The linearization of vP would be sold>the>camera. Since SpecCP in the embedded clause has already been occupied, it is an island. Therefore, the RNR target cannot move to the right edge of the embedded CP.9 The embedded CP is spelled out, and the linear order would be who>sold>the>camera. In the matrix vP spellout domain, the RNR target moves to the right edge of vP, and the spellout of the vP follows. The linearization would be find>out>who>sold>the>camera. It is important to notice that unlike the leftward movement case in (27), the linear order here is still preserved from the embedded CP domain. There is no constituent to move across at the right edge, so there is no effect on linear order. Finally, the whole first conjunct is linearized as Josh>will>find>out>who>needed>the>camera, and the linear order is preserved at all times. The second conjunct is linearized in the same way: Bill>will>find>out>who>sold>the>camera. Therefore, Sabbagh’s cyclic rightward movement anlaysis appears to capture the insensitivity to island effects. 9

Sabbagh (2007) assumes that the RNR target stops by the embedded SpecCP position, as in (i).

(i) a. Josh will [find out [who [needed t vP] t’ CP] t’’ vP] the camera. b. Bill will [find out [who [sold t vP] t’ CP] t’’ vP] the camera

(Sabbagh 2007:381)

I think this is a mistake, because if it were possible, we would lose the explanations for the existence of island effects in leftward movement.

46 The conjuncts are linked by a coordinator in such a way that the first conjunct is merged to the specifier position of &P and the second conjunct. &P1 is not a spellout domain, so the linear order between the conjuncts is not determined yet. Sabbagh claims that the RNR target in each conjunct undergoes ATB movement, as in (29).10 Now &P2 is spelled out and linearized.11 The linear order would be Josh>will>find>out>who>needed>and>Bill>will>find>out>who>sold> the>camera. Thus, the linear order is preserved. Under Sabbagh’s analysis, the RNR constructions are possible because rightward movement does not cause any ordering conflicts.

(29)

&P2 qp &P1 DP qp the camera CP1 and CP2 3 3 CP DP CP DP 6 t 6 t

Sabbagh claims that the Right Roof Constraint effect can also be captured under the linear order preservation analysis. Let us consider the sentence in (30).12 First, the RNR target moves to the right edge of the vP. Then, vP is spelled out and linearized. Such movement is allowed without violation of linear order preservation, since it occurs within the vP spellout domain. It further moves to the next spellout domain (CP) in each conjunct in (30a-b). The linear order is preserved, so this sequence of rightward movement is considered to obey the Right Roof Constraint.13 10

However, recall that we argued that &P should be a spellout domain in section 2.1.4. See section 2.1.4 why &P1 is not a spellout domain but &P2 can be. 12 Sentences in (30-31) are from Sabbagh (2007). 13 RNR is known to observe the Right Edge Restriction (Postal 1974, McCawley 1982, Swingle 1993, Wilder 1997, 1999, Hartmann 2000), which requires that the RNR target occupies the right edge of the first 11

47 (30) Josh will [vP donate t to the library], and Maria will [vP donate t to the museum] – each of these old novels. a. Josh will [[donate t to the library vP] t’ vP] t’’ CP] each of these old novels. b. Maria will [[donate t to the museum vP]

t’ vP] t’’ CP] each of these old novels.

Rightward movement is not always able to cross overt material. Sabbagh presents an example where the string looks similar to (30), but movement is blocked due to violation of linear order preservation. Let us consider (31). The adjuncts drunk and sober are assumed to adjoin to TP. When the RNR target moves to the right edge of the vP, he claims that further movement would cause a violation of the linear order preservation. The detail will be discussed in the next section where I will argue that this movement is licit under Fox and Pesetsky’s system, contra Sabbagh’s account.

(31) *Josh [vP described t for Jamie] drunk, and Maria [vP reenacted t for Sally] sober – a popular Broadway musical.

2.1.4 Challenges for Sabbagh’s analysis There are two concerns for Sabbagh’s movement analysis. The first challenge relates to when the coordinator can be linearized. Let us consider (28) again, where an independent linearization process takes place in each conjunct. The next spellout domain needs to be linearized after the coordinator is merged and the RNR target moves out of the coordinate structure in (29). That is the only way that the the whole sentence could be linearized with no contradiction in linear order. conjunct. Sabbagh assumes that the Right Edge Restriction is not violated in (30), since the RNR target moves to the right edge inside each conjunct.

48 On top of the spelled out second conjunct, let us assume that a coordinate structure is merged, as in (32a). Then, the first conjunct is merged to the specifier position of &P, as in (32b).

(32) a.

&’ 3 and CP 3 CP DP 6 the camera Bill will find out who sold t

b.

&P1 wp CP &’ 3 3 CP DP and CP 6 the camera 3 Josh will find out CP DP who needed t 6 the camera Bill will find out who sold t

The necessary assumption would be that the coordinate structure &P1 is not a spellout domain. If &P1 were spelled out, the linearization would be Josh>will>find>out>who>needed>the>camera >and>Bill>will>find>out>who>sold>the>camera. Since linearization cannot be altered in the subsequent spellout domains, no further rightward ATB movement would be allowed. When the RNR target moves out of the conjuncts, the coordinator and precedes the RNR target the camera, as shown in (33a-b). This is not a welcome result because the linear order is not preserved; in the previous spellout domain, the RNR target in the first conjunct preceded the coordinator. Thus, this suggests that &P1 should not be a spellout domain. &P2 could be a candidate for a spellout

49 domain, so that the linear order between the RNR target and the coordinator can be newly determined.14

(33) a.

&P2 qp &P1 DP wp the camera CP &’ 3 3 CP DP and CP 6 t 3 Josh will find out CP DP who needed t 6 t Bill will find out who sold t

b. Josh>will>find>out>who>needed>and>Bill>will>find>out>who>sold>the>camera.

However, this assumption would overgenerate. Let us consider (34). Only the object DP of the second conjunct moves out of the coordinate structure in (34), which is a violation of the Coordinate Structure Constraint (Ross 1967).15 However, Sabbagh’s analysis predicts (34) to be grammatical since the linear order is preserved. Suppose that only &P2 is a spellout domain and butternut squash moves out of the conjuncts successive-cyclically. As shown in (34b-c), the linear order ate>butternut>squash in CP2 is preserved in the next spellout domain &P2. To explain the ungrammaticality of (34), Sabbagh’s account should assume that the Coordinate Structure Constraint must be observed whether or not rightward movement obeys linear order preservation. Or the linear order preservation analysis could only apply to RNR. Either way, further explanation would be necessary.

14 Alternatively, one could assume that there is an additional phrase at the top, a RootP, which serves as a spellout domain containing the whole sentence. 15 See also chapter 7 for discussions about Coordinate Structure Constraint.

50

(34) *[&P2 [&P1 Sally ate green beans, and Bill ate t], respectively, butternut squash]. a. CP1 = Sally>ate>green>beans b. CP2 = Bill>ate>butternut>Squash c. &P2 = Bill>ate>green>beans>and>Bill>ate>respectively>butternut>squash

A more critical empirical problem is that Sabbagh’s analysis fails to capture the ungrammaticality of (31), repeated in (35); that is, his analysis should predict rightward movement of the RNR target to be grammatical since it observes linear order preservation.

(35) *Josh [vP described t for Jamie] drunk, and Maria [vP reenacted t for Sally] sober – a popular Broadway musical.

To account for (35), Sabbagh adopts Chomsky’ (2001) definition of spellout, rather than Fox and Pesetsky’s (2005). There is a subtle difference between the concepts of spellout in Chomsky (2001) and Fox and Pesetsky (2005). Specifically, in Chomsky’s system, the complement of the phase head is spelled out at the point of finishing the phase. For example, upon finishing vP, VP is spelled out; upon finishing CP, TP is spelled out. On the other hand, in Fox and Pesetsky’s system, upon finishing a spellout domain, the whole spellout domain is spelled out (not just an internal part of it). Fox and Pesetsky assume that the spellout domains are vP and CP. Let us see how Chomsky’s system works with an example in (36).

51 (36) Which man should John hire? a. [vP John v [VP hire which man] ↓ (Move which man to the edge of vP) [vP Which mani [vP John v [VP hire ti] ↓ Spell-Out(VP): hire ↓ (Move which man to the edge of CP) b. [CP Which mani should [TP John [vP ti’ v [VP hire ti] ↓ Spell-Out(TP): John>hire ↓ c. Spell-Out (Matrix): Which man>should>John>hire

It is important to notice that VP and TP are spelled out in (36a-b) at the completion of the vP and CP phase, which means that spellout occurs after movement, if any, is completed within each phase. That is why the wh-phrase is not spelled out in (36a-b). If it were spelled out before movement in each phase, an ordering contradiction would occur between spellout domains. Let us consider (37). If the spellout of VP occurred prior to wh-movement to the edge of vP, the whphrase would be linearized in the vP phase (37a) and hire is followed by which man. Afterwards, let us assume that the wh-phrase moves to the edge of vP. If TP were spelled out in the CP phase before wh-movement to SpecCP, the linearization would be John>which man>hire as in (37b). The linear order is not preserved from the previous VP spellout domain since the wh-phrase precedes the verb. Therefore, the derivation of (37) does not work. The fact that the derivation of (37) is not right plays an important role in revealing that Sabbagh’s example in (35) cannot be ruled out under the linear order preservation analysis.

52 (37) Which man should John hire? a. [vP John v [VP hire which man] ↓ [vP [vP John v [VP hire which mani] ↓ Spell-Out(VP): hire>which man ↓ b. [CP should [TP John [vP which mani v [VP hire ti] ↓ Spell-Out(TP): John>which man>hire ↓ c. [CP Which mani should [TP John [vP ti’ v [VP hire ti] ↓ Spell-Out (Matrix): Which man>should>John>hire

Ordering Contradiction

Bearing this in mind, let us first follow how Sabbagh’s analysis derives the ordering contradiction in (38). Take a closer look at the analysis with the first conjunct of the example, as in (38). In the vP phase (38a), the RNR target moves to the right edge of vP and VP is spelled out, which results in the linear order, described>for Sam. Then, the subject depictive modifier (DepP) drunk is adjoin to TP. In the CP phase, TP is spelled out as in (38b). Sabbagh seems to assume that TP is spelled out prior to the rightward movement of the RNR target. Thus, the RNR target precedes the depictive modifier. In the root phase, the DP moves further and the whole sentence is spelled out as in (38c). The ordering contradiction results between DP and DepP. DP precedes DepP in (38b), but this is contradicted in (38c). Thus, this rightward movement is not linearizable, which in turn is illicit.

53 (38) [Max [described t for Sam vP] drunk TP] a Broadway musical a. [vP v [VP described DPi PPfor] ↓ [vP v [VP described ti PPfor] DPi] ↓ Spell-Out(VP): described > PPfor ↓ b. [C [TP Max [vP v [VP described ti PPfor] DPi]] DepP]] ↓ Spell-Out(TP): Max > described > PPfor > DPi > DepP ↓ c. [[CP C [TP Max [vP v [VP described ti PPfor] ti]] DepP]] DPi] ↓ Spell-Out(Matrix): Max > described > PPfor > DepP > DPi

(Sabbagh 2007: 384)

However, this is not a correct way of applying Chomsky’s spellout. Movement must occur first in a given phase and then spellout follows. We have observed in (36-37) that the timing of spellout makes different predictions in terms of ordering preservation between spellout domains; the linear order is preserved only when spellout occurs only after the phase is complete. Now let us reanalyze (38) in such a way that TP is spelled out when all the movement to the CP phase is completed in (39). The DP undergoes further movement to adjoin to CP and then the spellout of the CP phase would be (39b). The DP is outside the TP domain, thus not linearized in the CP spellout. Instead, it would be linearized when the matrix clause is spelled out, as shown in (39c). Notice that ordering confliction does not arise between DP and DepP in all spellout domains. This indicates the current analysis of Sabbagh (2007) does not capture the unacceptability of (35) with ordering preservation.

54 (39) [Max [described t for Sam vP] drunk TP] a Broadway musical a. [vP v [VP described DPi PPfor] ↓ [vP v [VP described ti PPfor] DPi] ↓ Spell-Out(VP): described > PPfor ↓ b. [[CP C [TP Max [vP v [VP described ti PPfor] ti]] DepP]] DPi] ↓ Spell-Out(TP): Max > described > PPfor > DepP ↓ c. [[CP C [TP Max [vP v [VP described ti PPfor] ti]] DepP]] DPi] ↓ Spell-Out(Matrix): Max > described > PPfor > DepP > DPi

The linear order of a sentence like (39) is also preserved under Fox and Pesetsky’s analysis. Recall that on their analysis the spellout of CP linearizes all the components in CP, not just the complement of C (i.e. TP) in the same circumstance. Let us linearize (39) under Fox and Pesetsky’s analysis in (40). In the vP spellout domain (40a), the DP moves to the right edge of vP and the vP is spelled out. DepP adjoins to TP. The DP further moves adjoining to the right edge of CP, and then spelled out as in (40b). The linear order is preserved, so it can be linearized and is predicted to be grammatical.16

16

One possible way to solve this puzzle under Sabbagh’s system is to assume that the DegP is an island, so that movement across it is prohibited. However, this does not sound convincing either, because leftward movement has no problem crossing over the DegP in (i). (i) Which Broadway musical did John describe for Sam drunk?

55 (40) a. [vP Max v [VP described DPi PPfor] ↓ (Move DP) [vP Max v [VP described ti PPfor] DPi] ↓ Spell-Out(vP): Max>described > PPfor> DPi ↓ b. [[CP C [TP Max [vP v [VP described ti PPfor] ti]] DepP]] DPi] ↓ Spell-Out(CP): Max > described > PPfor > DepP > DPi

This appears to indicate that the locality of rightward movement cannot be derived solely from the linear order preservation. Therefore, I conclude that Sabbagh’s analysis of the locality of rightward movement with linear order preservation suffers from both a conceptual problem and a critical empirical failure.

2.2 Non-movement accounts In response to the problems with movement analyses of RNR, non-movement analyses have been proposed, such as i) Strict Phonological Deletion (Wexler and Culicover 1980, Booij 1985, Levine 1985, 2001, Kayne 1994, Bošković 1997, Wilder 1997, Hartmann 2000, Mukai 2003, Abels 2004, An 2006) according to which the RNR elements are deleted at PF but remain at LF, as illustrated in (41), and ii) Multiple Dominance (McCawley 1982, Goodall 1983, 1987, Erteschik-Shir 1987, Wilder 1999, Yoon and Lee 2005, de Vos and Vicente 2005, Bachrach and Katzir 2006a) according to which the RNR element, the DP in (42), is dominated by both the verb in the first conjunct and the verb in the second.

(41) a. PF: Chris wrote , and Kate reviewed – an article. b. LF: Chris wrote an article, and Kate reviewed an article.

(by PF-deletion)

56 (42) Chris wrote, and Kate reviewed – an article. &P qp TP1 &’ 3 3 DP VP & TP2 Chris t 3 V DP VP Wrote Kate ry V DP reviewed 5 an article

2.2.1 The strict phonological deletion account Strict phonological deletion accounts consider RNR as a strict PF-phenomenon. RNR is licensed when there exists an identical string of phonemes in the second conjunct. In (41), since the phonological string, an article, exists in the antecedent, the identical string in the first conjunct can be deleted at PF. The deletion is assumed not to affect the other interfaces such as overt syntactic derivation or Logical Form. An important consequence under this account is that RNR does not have to observe syntactic constraints, such as constituency or any movement constraints. Therefore, non-constituents would be eligible to be elided, solving one of the problems for the movement analysis. The possibility that RNR can apply below the word level supports this account as well (43-44). In (43), the subpart of the word, generation in undergeneration, can be elided since there is its antecedent in the second conjunct. The same holds for the Dutch example in (44), due to Hartmann (2000).

(43) This analysis suffers from both UNDER- and OVER-generation.

57 (44) Frühlingsblumen

und Herbstblumen

Springtime flowers and autumn flowers (Hartmann 2000: 57)

Insensitivity to islands is predicted under the strict phonological deletion analysis since no movement is assumed to be involved. The prediction is naturally captured in (45a-b).

(45) a. Susan wonders when John ORDERED , and Bill wants to know when he RETURNED – the tickets for the opera.

(Wh-island)

b. John likes a professor who LECTURED ON
, and Mary likes a graduate student who DEBUNKED – a recent theory of Right Node Raising.

(Complex NP island)

Likewise, the other movement-related problems, such as preposition stranding and the Right Roof Constraint, do not arise in the deletion accounts. The possibility of preposition stranding in languages where movement cannot strand a preposition is straightforwardly explained under the deletion accounts, since the complement of the preposition (i.e. the RNR target) does not undergo any movement. Also, the Right Roof Constraint is simply not relevant because the RNR target does not move, but is deleted in-situ in the first conjunct. The deletion accounts attempt to explain why the RNR target needs to be in the right edge of the conjuncts, the constraint called the Right Edge Restriction (Postal 1974, McCawley 1982, Swingle 1993, Wilder 1997, 1999, Hartmann 2000, Sabbagh 2007). Let us examine Hartmann (2000) where the Right Edge Restriction is captured by the edge alignment of prosodic phrases. Hartmann assumes that prosodic constituents are organized hierarchically, and that a constituent carrying nuclear accent always occupies the rightmost edge position within prosodic phrases

58 because grid-marks representing prominence relations are assigned in the rightmost position. Let us consider this with an example (46).17 At each prosodic level, there is a prominent head of a prosodic constituent which is represented by grid-marks, and the head at the utterance level is identified with the nuclear accent (Pierrehumbert 1980, Hartmann 2000). Thus, the nuclear accent is assigned to the object DP in (46), which occupies the rightmost edge of the utterance.

(46)

x (

x)U

(x

x)IP

(x)φ (x ( )w (

x )w

John plays

( )w ( the

x)φ )w trombone. (Hartmann 2000:99)

On the other hand, the nuclear stress is assigned on an element bearing a contrastive focus in RNR sentences, and this element is immediately followed by the RNR target. Let us consider (47). Here the preposition is contrastive between the conjuncts, so the head of the prosodic constituents at the utterance level is the preposition in each conjunct.

17

U = Phonological Utterance, IP = Intonational Phrase, φ = Phonological Phrase, and w = Phonological Words.

59 (47) (

x

( x

x

)IP

(x

x

(x

( x )φ

(x

Mary

sleeps ON and

x

x

)U

x

)IP

x

Bill sleeps BESIDES



(x

x)φ

the sofa. (Hartmann 2000: 100)

Hartmann claims that the contrastive focus carrying a nuclear accent in the first conjunct needs to be aligned in the rightmost edge position, and that if the RNR constituent follows the element bearing the contrastive focus, it is incorporated into the preceding phonological phrase (Target Incorporation, Hartmann 2000: 98). The Target Incorporation can be satisfied by the deletion or deaccenting of the RNR target. Therefore, under Hartmann’s analysis, RNR is a kind of strategy to make sure that the alignment constraint in (48) is satisfied. And this is basically how she derives the Right Edge Restriction as consequence of alignment constraint.

(48) Alignment constraint in RNR Delete the RNR target so that the rightmost edge in the first conjunct can be aligned with the head of the prosodic constituent at the utterance level.

2.2.2 The multiple dominance account Another non-movement analysis proposed in previous literature is the multiple dominance account. McCawley (1982) is usually credited as the first to propose representations of the sort employed in the multiple dominance account, but the formulation of the multiple dominance account of RNR explored here is that, formalized by Wilder (1999).

60 The main concerns of the multiple dominance hypotheses are to linearize a RNR structure, and to derive the Right Edge effect in RNR. (49) shows, for example, that only the right edge of the first conjunct can be RNRed. The example in (49b) is ungrammatical, by contrast, since the RNR target in the first conjunct is not at the right edge of the first conjunct.18

(49) a. Bill recorded t, and Jason watched t in December – every Red Sox game. b. *Bill sent t a present, and Mary congratulated t – all the winners.

The multiple dominance accounts attempt to derive the right-edge condition in RNR with linearization. The crucial point of the accounts is that the only possible way to linearize the RNR target is at the right edge of each conjunct. They assume that the target of RNR has just a single copy, and this copy is merged to the two conjuncts simultaneously.19 In a simple RNR example, like Mary wrote and Bill read the article, the target DP, the article, is simultaneously merged with the verb in the first conjunct, wrote, and that in the second conjunct, read. The structure of the example is shown in (50).

18

We need to assume that the rightward movement of the DP all the winners to the right edge of the first conjunct in (i) is prohibited. This will be discussed in chapter 5, following Baltin (2001). (i) * [vP [vP Bill sent t a present] all the winners] 19

Citko (2005) dubbed this Parallel Merge.

61 (50) Chris wrote, and Kate reviewed – an article. &P qp TP1 &’ 3 3 DP VP & TP2 Chris t 3 V DP VP Wrote Kate ry V DP reviewed 5 an article

Let us examine Wilder’s (1999) multiple dominance account in more detail. Wilder adopts a slight reinterpretation of Kayne’s (1994) Linear Correspondence Axiom (LCA) to account for the linearization of RNR. Kayne’s LCA states that a terminal node A precedes a terminal node B if A asymmetrically c-commands B. Let us consider how the word order of (50) would be derived. In the first conjunct, Chris asymmetrically c-commands wrote, so Chris precedes wrote in linear order. And wrote asymmetrically c-commands its object DP, so wrote precedes an article. Inside the DP, the indefinite article asymmetrically c-commands the NP article, hence an precedes article. The second conjunct is linearized in the same way. Kayne (1994) assumes that if a phrase XP asymmetrically c-commands a phrase YP, every terminal node in XP precedes every element in YP. In (50), the first conjunct also asymmetrically c-commands the second conjunct, which results that every element in the first conjunct precedes everything in the second. The result of linearization under Kayne’s LCA as originally formulated for (50) would be (51).

62 (51) Asymmetric c-command and linearization Chris > wrote Chris>an article Chris>& Chris>Kate Chris>reviewed

Chris>an article

wrote>an article wrote>& wrote>Kate wrote>reviewed wrote>an article an article>& an article>Kate an article>reviewed an article>an article &>Kate

&>reviewed

&>an article

Kate>reviewed

Kate>an article

reviewed>an article

However, an ordering conflict necessarily arises in this linearization. For example, the verb in the second conjunct reviewed precedes the RNR target an article in one place, and the RNR target precedes the verb in another place. In addition, reflexivity violation also arises in this linearization because an article precedes itself. Under Kayne’s LCA, therefore, the RNR sentence in (50) is predicted not to be linearizable. Wilder revises Kayne’s LCA in a way that only the terminals which are fully dominated can be linearized and the terminals of constituents that are shared are not in the image of either conjunct. Full dominance defined by Wilder is in the following (52).

(52) Full dominance X fully dominates α iff X dominates α and X does not share α. (Wilder 1999: 591)

With this new definition, let us reconsider (50). Wilder assumes that the asymmetrical ccommand relationship determines the linearization of the terminal nodes inside TP1 in (53a) and the linearization inside &’ in (53b).

63 (53) a. TP1: Chris>wrote>an>article b. &’: and>Kate>reviewed>an>article

TP1 asymmetrically c-commands &’, so by hypothesis every terminal in TP1 precedes every terminal in &’. Wilder supposes that the RNR target is not in the image of both conjuncts since the shared material is not fully dominated. Thus, linearization would not contain the shared constituent, as shown in (54a), and the only possible word order from the linearization of each terminal node is (54b).

(54) a. Chris > wrote Chris>&

Chris>Kate

Chris>reviewed

wrote>& wrote>Kate

wrote>reviewed

&>Kate

&>reviewed Kate>reviewed

b. TP1>&’: {Chirs>wrote}>{and>Kate>reviewed}

When we compare those three sets of linearization in (53-54), the only place that the RNR target, an article, can be linearized without any ordering contradiction would be the right edge of the second conjunct, as in (55).

(55) Chirs>wrote>and>Kate> reviewed>an>article

Let us examine if the shared material is linearized between wrote and and in (56). The linear order in (56) is preserved with respect to TP1 in (53a). However, an ordering contradiction occurs

64 when we compare the linearization between (56) and (53b); an article precedes and, Kate, and reviewed in (56), but an article follows them in (53b). Thus, (56) is not linearizable.

(56) Chirs>wrote>an>article>and>Kate> reviewed

Wilder claims that the multiple dominance account can explain why SVO languages do not allow backwards gapping, but SOV languages do (57-58). Let us first consider backwards gapping in English (57). The linearization inside TP1 and &’ is shown in (57a-b). TP1 asymmetircally c-commands &’, hence terminals in TP1 precedes terminals in &’, as in (57c). Note that the shared material is not in the image of the linearization. The verb needs to be linearized either in the first conjunct (57d) or in the second (57e). However, ordering contradiction occurs in both cases; for (57d), a red wine precedes drank, which contradicts the ordering (57a), and for (57e), drank precedes and, and Mary, which contradicts the ordering (57b).20 Thus, the backwards gapping sentence in (57) cannot be linearizable.

(57) *Jane [e] a red wine, and Mary drank an orange juice. &P qp TP1 &’ 3 3 DP VP & TP2 Jane t 3 DP DP VP 5 Mary ry a red wine V DP drank 6 an orange juice 20

It is interesting to notice that this analysis also rules out the linearization of regular gapping in English (57e). Thus, we can conclude that gapping in general cannot be captured by the multiple dominance account.

65 a. TP1: Jane>drank>a>red>wine b. &’: and>Mary>drank>an>orange>juice c. TP1>&’: {Jane>a>red>wine}>{and>Mary>an>orange>juice} d. *Jane>a>red>wine>and>Mary>drank>an>orange>juice e. *Jane>drank>a>red>wine>and>Mary>an>orange>juice

On the other hand, a Korean backwards gapping example in (58) matches the assigned linearization. Basically, the linearization is the same with SV and SVO order in English. Wilder’s analysis, therefore, nicely accounts for why the Right Edge Restriction needs to be observed in RNR. His analysis can derive the Right Edge Restriction by showing that the linearization is not possible when the RNR target does not reach the right edge of its conjunct.

(58) Jane-un wine-ul

[e], kuliko Mary-nun orange juice-lul masi-ess-ta.

J.-Top wine-ACC [e], Conj

M.-TOP

(Korean)

orange juice-ACC drink-PAST-DEC

‘Jane drank wine, and Mary drank an orange juice.

As in the deletion accounts, the concerns for the movement analyses do not arise in the multiple dominance analyses. First of all, movement constraints are not relevant here because the RNR target stays in-situ at all time. Therefore, the insensitivity of RNR to island violations is correctly predicted as in (59a-b), as is the insensitivity of RNR to preposition stranding in (60).

66 (59) a. wh-islands [[CP1Susan wonders when John ordered

], and

[CP2Bill wants to know when he returned

[DP the tickets for the opera]]].

b. Complex NP constraint [[CP1John likes a professor who lectured on

], and

[CP2Mary likes a graduate student who debunked

[DP a recent theory of RNR]]].

Recall that Irish does not allow preposition stranding if the complement of the preposition undergoes movement as shown in (18), repeated here in (60a), but a preposition can be stranded in RNR which indicates the complement of the preposition does not move in (19), repeated in (60b). The multiple dominance account can capture the grammaticality of (60b) on the grounds that the RNR target is shared by both conjuncts without any movement.

(60) a. *Bhí mé a éisteacht le t inné Was I listen(prog) with yesterday

[DP clár

mór fada ar an ráidió faoin toghachán]

program great long on the radio about-the election

‘I was listening yesterday to a great long program on the radio about the election.’ b. Nil sé in aghaidh an dlí a thuilleadh a bheith ag éisteacht LE Is-not it against the law anymore

be(-fin) listen(prog) with

nó ag breathnu AR or look(prog)

on

[DP ráidió agus teilifís an Iarthair] radio and television the West(gen)

‘It is no longer against the law to listen, or to watch, Western radio and television.’ (McCloskey 1986: 184-185)

67 The issues of non-constituency in RNR do not seem to arise under the multiple dominance accounts. One could imagine that there can be multiple instances of shared material between the conjuncts, each of which forms a constituent individually. Let us consider (23) again, repeated in (61). Note that the embedded vP forms a constituent, and so does the matrix clause.

(61) [AJ-nun [TP Mary-ka
cip-e ka-ess-ta>],

M.-NOM come-because home-to go-PAST-DEC

kuliko [Sue-nun [TP John-i oa-seo] CONJ

S.-TOP

J.-nom come-because

cip-e kas-ess-ta.] home-to go-PAST-DEC.

‘AJ went home because Mary came, and Sue went home because John came.’

We assume that a copy of the embedded vP is shared by both conjuncts, and independently the same thing happens for the matrix clause. The simplified structure is drawn in (62).21

(62)

21

&P qp TP1 &’ 3 3 AJ-TOP vP & TP2 eu 3 TP : Sue-TOP vP 3 t ri Mary-NOM vP TP : t 3 2 John-NOM vP went home 2 came-because

To linearize a multiple dominance structure, it is crucial to assume that the underlying structure of SOV languages is SVO, and that SOV is derived by movement of the internal argument (Kayne 1994). Otherwise, the verb would precede the internal argument since it asymmetrically c-commands the internal argument. In (61), I assume that the embedded TP undergoes movement, so that it asymmetrically ccommands the matrix verb.

68 The linearization of the structure in (62) works as follows. The asymmetrical c-command relationship among the terminal nodes determines the word order of each conjunct in (63a-b). The shared constituent is not in the image of the coordinate structure, so (63c) results.

(63) a. TP1: AJ-TOP>Mary-NOM>came-because>went-home b. &’: &>Sue-TOP>John-NOM>came-because>went-home c. TP1> &’: {AJ-TOP>Mary-NOM}>{&>Sue-TOP>John-NOM}

The only place where the embedded and the matrix vP are linearized is the rightmost position as shown in (64). Therefore, with multiple shared constituents, the multiple dominance account can account for (61). In fact, under this view, the constituency issue for (61) does not arise.

(64) AJ-TOP>Mary-NOM>&>Sue-TOP>John-NOM>came-because>went-home

Finally, there may be a way to resolve the problems of relational modifiers in RNR constructions with the multiple dominance analyses. 22 Let us consider (65) where the shared material contains one of the relational modifiers, similar. The reading we are interested in here is the distributive reading in which there is a similar type of painting that Sue likes but Mary hates.23

22

See also section 2.1.1 of this chapter. This interpretation requires the shared material to “see” the coordinate structure, but it is not clear how. We will return to consider this issue in section 3.2.1.

23

69 (65)

&P qp TP1 &’ 3 3 DP VP & TP2 Sue t 3 V DP VP likes Mary ry V DP hates 5 a similar painting

A concrete attempt to account for relational modifiers in RNR under the multiple dominance account is made by Chung (2004) and Yoon and Lee (2005) with Korean RNR data. Let us examine their proposals. Korean has markers that are similar to relational modifiers in English, such as the Dummy Plural Marker (DPM) –tul and the Distributive Marker (DM) kakkak ‘each, respectively’. The dummy plural marker in Korean can be optionally attached to any phrase except the subject. The dummy plural marker can be attached to many categories within a sentence, such as an adverb, a verb, or even a complementizer. (66) shows a maximal case, with the dummy plural marker attached to every phrase (except the plural argument itself) within a sentence.

(66) John-kwa Mary-ka satang-(ul)-tul

ppali-tul mekki-tul hass-ta-tul.

J.-Conj M.-Nom candy-(A)-DPM fast-DPM eat-DPM do-Dec-DPM ‘John and Mary ate candy fast’

Choe (1988) claimed that the dummy plural marker must meet the following requirement to be licensed: the marker must be c-commanded by a plural subject in the local domain. Let us

70 compare (66) with (67). (67) is ungrammatical because the dummy plural marker does not meet the plurality requirement: the local subject is not plural (although the non-local subject is).

(67) *Bill-kwa Mary–nun [John-i B.-and M.-Top

swuhak-ul cal-tul

[J.-Nom

han-ta-ko] sangkak han-ta.

math-Acc well-DPM do-Dec-C] think-Dec

‘Bill and Mary think that John does mathematics well.’

Chung (2004), and Yoon and Lee (2005) claim that the distributive marker shares an interesting property with the dummy plural marker; it also needs a local plural subject, as shown in (68). In (68a), the DP conjuncts in the subject position meet the plurality requirement. While the DP conjuncts are in the matrix subject in (68b), they are not in local with respect to the distributive marker. Thus, (68b) does not satisfy the plurality requirement.

(68) a. John-kwa Mary-nun chayk hankwon-ul J.-and M.-TOP

kakkak ilk-ess-ta.

book one-CL-ACC each read-PAST-DEC

‘John and Mary each read a book.’ b. *John-kwa Mary-nun [Bill-i J.-and M.-TOP

chayk-ul kakkak ilk-ess-ta-ko] tul-ess-ta.

B.-NOM book-ACC each

read-PAST-C heard-DEC

‘John and Mary heard that Bill each read a book.’

However, the plurality requirement for the distributive marker needs to be revised to accommodate empirical facts that the requirement also seems to be satisfied by coordination of VP.24 In this sense, the distributive marker does not behave exactly the same way as the dummy

24

This plays an important role in the next chapter, section 3.2.2.

71 plural marker. Let us compare (69a-b). Note that there is a singular subject John-TOP in both sentences. Neither the distributive nor the dummy plural marker would be able to appear, if only a plural subject could satisfy the requirement. However, (69a) is grammatical, and the reason is that it takes scope over the plural conjuncts of VP (Carlson 1987). On the other hand, the dummy plural marker does need to be in the local domain of the plural subject, as the ungrammaticality of (69b) shows.

(69) a. John-un [&P [VP hansi-ey cemsim-ul mek-ko] [VP neysi-ey kansik-ul]] kakkak mek-ess-ta. J.-TOP

1 o’clock lunch

eat-CONJ 4 o’clock snack-ACC each eat-PAST-DEC

‘John ate lunch at 1 and snack at 2, respectively.’ b. *John-un [&P [VP hansi-ey cemsim-ul mek-ko] [VP neysi-ey kansik-ul]] manhi-tul mek-ess-ta. J.-TOP

1 o’clock lunch eat-CONJ

4 o’clock snack-ACC a lot-DPM eat-PAST-DEC

‘John ate lunch at 1 and snack at 2.’

With this background, let us consider the dummy plural marker and distributive marker in a RNR context (70-71). Chung (2004), and Yoon and Lee (2005) claim that under the multiple dominance account, the plurality requirement of the dummy plural marker and distributive marker can be explained by assuming that a single copy of those markers is simultaneously in the scope of the subject of each conjunct (70-71). Therefore, plurality conditions for the markers are satisfied.

(70) Sue-nun yagwu-lul, kuliko Jane-un nonggwu-ul S.-TOP baseball-ACC, CONJ J.-TOP

cal-tul han-ta

basketball-ACC well-DPM do-DEC.

‘Sue plays baseball well, and Jane plays basketball well.’

72 &P qp TP1 &’ 3 3 DP VP & TP2 Sue-TOP t 3 DP DP VP baseball-ACC John-TOP ry DP VP basketball-ACC3 well-tul do-DEC

(71) Mary-ka chayk-ul, kuliko John-i shinmun-ul

kakkak il-ess-ta.

M.- NOM book-ACC, CONJ J.-NOM newspaper-ACC each read-PAST-DEC. ‘Mary read a book, and John read a newspaper, respectively. &P qp TP1 &’ 3 3 DP VP & TP2 Mary-NOM t 3 DP DP VP book-ACC John-NOM ry DP VP newspaper-ACC3 DM V kakkak read-PAST-DEC

To summarize this section 2.2, we have followed arguments for non-movement analyses. The non-movement analyses can solve many of the problems for the movement analyses by not assuming that there is no movement involved. Therefore, movement constraints, such as island violations, preposition stranding, and the Right Roof Constraint, are not relevant any longer. The non-movement analyses can derive the right-edge restriction in RNR. For the deletion accounts, Hartmann (2000) reduces this edge constraint to a prosodic requirement of RNR in which the

73 nuclear accent needs to be aligned to the rightmost edge position. Thus, the immediately following unaccented RNR target must be incorporated into the preceding phonological phrase by deletion or deaccenting. The multiple dominance accounts appeal to the fact that RNR sentences can be linearizable only when it is in the rightmost position of the coordinate structure.

74

CHAPTER 3 RIGHT NODE RAISING IS ELLIPSIS

In the last chapter, we examined the previous accounts of Right Node Raising (RNR): movement and non-movement analyses – strict phonological deletion and multiple dominance. I concluded that movement analyses, including Sabbagh’s (2007) cyclic linearization analysis, are not on the right track. As alternatives, non-movement analyses seem to be promising, since they are not sensitive to any movement constraints, such as island violations, preposition stranding, or the Right Roof Constraint. In this chapter, we will take a close look at the non-movement accounts, and then I will argue that they are problematic on their own. Those problems disappear if we assume that RNR is an ellipsis phenomenon. My proposal is that the RNR target in the first conjunct is deleted at PF as ellipsis, and I will call it the ellipsis account of RNR. 1 Let us first consider problems of the deletion and the multiple dominance accounts.

3.1 Problems of the previous accounts This section is concerned with problems for the strict phonological deletion and the multiple dominance accounts. There are some problems that both accounts suffer from and others that are specific to one of the accounts. I will start discussing problems that apply to both accounts:

1

The ellipsis account is different from the strict deletion accounts, and I will discuss the differences in section 3.1.6.

75 Vehicle Change effects and morphological mismatches. And then I will move to problems for the multiple dominance accounts, which include sloppy identity and parallelism. After pointing out the problems for the previous accounts in each subsection, I will argue that the problems do not arise under the ellipsis account. Note that, however, the discussion of the ellipsis analysis of RNR in this chapter is limited to the demonstration of the similarities between RNR and forward ellipsis. The actual ellipsis account – semantic licensing conditions of RNR – will be presented in chapter 5.

3.1.1 Vehicle Change effects in RNR Let us consider the following examples in (1). The embedded vP is not exactly the same between the conjuncts in (1a). That is, the proper name John is in the second conjunct, but its pronominal form is realized in the first conjunct. In (1b), the vP in the first conjunct contains a reflexive, but the second conjunct has its pronominal form. A crucial question is whether the vP can be RNRed despite the form mismatches. Since the non-movement accounts rely on the identity of phonological strings, matching phonological forms is crucial. Therefore, under those accounts, the embedded vP is predicted not to be RNRed.

(1)

a. Hei hopes Susan won’t fire himi (at the end of this year), but the secretary knows that she will fire Johni at the end of this year. b. Johni couldn’t nominate himselfi, so I nominated himi.

Under the strict phonological deletion accounts of RNR, the deletion of the embedded vP fire him should not be possible because the matching antecedent in the second conjunct has a different

76 form: fire John. The embedded vP could be deleted if the proper name replaces the pronoun in (2). However, this should give rise to a Principle C violation since John is bound.2

(2)

Johni hopes Susan won’t , but the secretary knows that she will fire Johni at the end of this year. (*Principle C)

Under the multiple dominance accounts, the same problem occurs. If the proper name John is shared, a Principle C violation should result, and if one conjunct contains John and the other contains him, the constituents are not eligible to be shared. The matrix subject of the first conjunct binds the proper name John within the shared material, as shown in (3).

(3)

&P qp TP &’ 6 3 Johni hopes Susan won’t & TP 6 The secretary knows that she will vP 6 fire Johni at the end of this year (*Principle C)

This problem can be solved if we assume that RNR arises from ellipsis. It has been observed in the literature that Principle C violations can be avoided in VP ellipsis contexts. This 2

Note that the “strict” phonological deletion account considers that RNR is strictly a PF-phenomenon, so no communication from or to the other interfaces is assumed. This assumption differs from the ellipsis account in which RNR is licensed by interface requirements.

77 phenomenon is due to an interpretive option, proposed by Fiengo and May (1994) and referred to as Vehicle Change. A simplified generalization of Vehicle Change effects is given in (4).

(4)

Vehicle Change (simplified version, Fiengo and May 1994) As long as indices remain constant, proper names and their pronominal correlates are considered equivalent.

Let us examine how Vehicle Change works with an example (5). If the elided copy were phonologically identical to its antecedent, we would expect a Principle C violation to occur since the subject of the main clause binds the R-expression in the second conjunct in (5).

(5)

Mary loves Johni, and hei thinks Sally does , too. (Fiengo and May 1994:220)

However, Fiengo and May (1994) observe that (5) is a grammatical sentence, and propose that proper names and pronouns share a binary feature [± pronoun]. Proper names have the feature [- pronoun], but the reconstruction of the ellipsis is not sensitive to the value of the feature, so it can be realized as [+ pronoun]. In (6), the proper name is converted into a pronoun by Vehicle Change which bleeds the Principle C violation.

(6)

Mary loves Johni, and hei thinks Sally does , too

Fiengo and May propose that pronouns and reflexives are similarly nondistinct, hypothesizing that reflexives are composed of a pronoun and a separate morpheme –self which only has a

78 syntactic function. For example, a Principle A violation would be expected in (7a) since the reflexive is not bound within its binding domain. However, by hypothesis, himself and him are the same argument for reconstruction, so the reflexive can be shifted into a pronoun, thereby avoiding a Principle A violation, as in (7b).

(7)

a. Joshi didn’t vote for himselfi, but Mary did . b. Joshi didn’t vote for himselfi, but Mary did .

I argue that the similar Vehicle Change effects are observed in RNR constructions. Let us consider (8a-b). The acceptability of (8a) indicates that no Principle C violation has occurred; the proper name has been shifted into a pronoun. Similarly, the pronoun in (8b) must have been shifted into a reflexive to avoid a Principle B violation.

(8)

a. Hei hopes that Susan WON’T , but the secretary knows that she WILL –

fire Johni at the end of this year. b. Johni COULDN’T , so I NOMINATED himi

It is clear that the non-movement analyses do not account for Vehicle Change effects of (8). The multiple dominance account has to assume that there exists only one occurrence of the target vP fire John at the end of this year, which is shared by both conjuncts. Thus, there is no need or possibility of a reconstruction operation. Vehicle Change crucially depends on changing the value of the [± pronoun] feature during reconstruction, so this account predicts that no Vehicle Change effect should occur in RNR, contrary to fact. The strict phonological deletion accounts assume no

79 involvement of any other interfaces but PF. Thus, the flexibility of form under semantic identity is not relevant to the accounts. There is a puzzle, however. There have been previous observations with similar examples, but leading to the opposite conclusion. Levine (1985) and Swingle (1993) claim that RNR shows a ‘still-there’ effect; for example, when the RNR target contains a proper name and the first conjunct contains the co-indexed pronominal subject, a Principle C violation seems to be unavoidable in (9). This claim follows that the proper name inside the RNR target cannot undergo Vehicle Change, since, if it could, because the proper name would have been shifted into a pronominal form and the Principle C violation could be avoided. Therefore, there seems to be some conflict in judgment between (8) and (9).

(9)

*Shei SAID , and I happen to AGREE – that Maryi needs a new car. (Swingle 1993)

However, when we take a closer look at (9), it is not difficult to see that the sentence in (9) is ungrammatical for an independent reason. The pronoun she in the matrix subject of the first conjunct is co-indexed with Mary in the second conjunct. Notice that this is a violation of Backwards Anaphora Constraint (Langacker 1969), which prevents an anaphor from preceding its antecedent under coordinate structure. One way to factor out the constraint from (9) would introduce Mary in the previous context, so that the antecedent of the pronoun can be found in discourse, not in the subsequent clause. Let us consider (10).

80 (10) [Context] Mary is very frugal. She always tries to use products as much as she can. Recently, her car has stopped on the street and it turns out that it costs more to repair her car than to buy a new used car. [Target sentence]: At this point, shei ADMITTED , and I definitely AGREE – that Maryi needs a new car.

My informants find (10) to be marginally acceptable, but definitely better than (9) without context. The marginality of (10) can be attributed that the sentence bears an unfixable problem with respect to information structure. Note that even the fully pronounced version in (11) is equally marginal. This is due to the introduction of the proper name following a series of the pronominal forms. The point I want to make here is that the “still-there” effect with respect to Vehicle Change effect is not conclusive.

(11) ?At this point, shei admitted that shei needs a new car, and I definitely agree that Maryi needs a new car.

A similar Vehicle Change effect can be found in the alternation between some and any. Klima (1964) claims that some and any are different morphological realizations of the same underlying element. In (12), the Negative Polarity Item (NPI) in the ellipsis clause can be reconstructed with a Positive Polarity Item (PPI). Therefore, it is not (12a) but (12b) that is reconstructed for the sentence.

81 (12) Mary didn’t read any of the articles, but John did. a. *Mary didn’t read any of the articles, but John did read any of the articles. b. Mary didn’t read any of the articles, but John did read some of the articles.

In the spirit of Fiengo and May’s Vehicle Change, we can suppose that an NPI and a PPI share a binary feature, [± polarity indefinite]. An NPI has the feature [- polarity indefinite]. According to Vehicle Change effects, reconstruction is not sensitive to the value of the feature, so it can be realized as a PPI with [+ polarity indefinite]. Let us consider RNR examples in (13-14). In (13), the RNR target contains an NPI but only the second conjunct provides environment to license it.3 (13), nevertheless, is grammatical. Under the ellipsis account, this is straightforward because some flexibility of morphological realization is assumed to be possible in the ellipsis clause by Vehicle Change. Thus, the NPI is shifted into a PPI in the RNR clause, as shown in (13b).

(13) John read, but he hasn’t understood – any of my books. a. *John read any of my books, but he hasn’t understood any of my books. b. John read some of my books, but he hasn’t understood any of my books.

However, the flexibility is unexpected under both the strict phonological deletion and the multiple dominance accounts. Those accounts predict (13) to be ungrammatical, due to the failure of the first conjunct to provide environment for licensing the NPI in the RNR target. One might think that it all depends on the view of NPI licensing. One might hypothesize that the NPI licensing under coordination can be satisfied if it is licensed by at least one conjunct.

3

The example in (13) is from Cann, Kempson, Marten, and Otsuka (2005).

82 However, this hypothesis cannot be right, because if so, (14) should also be possible. In the case of (14), the NPI is overtly pronounced in the non-NPI-licensing environment (i.e. the second conjunct) and the first conjunct has negation that can license the NPI. The empirical generalization would be, rather, that the conjunct containing the NPI must be able to license it, and (14) is ungrammatical because the second conjunct does not provide licensing conditions for the NPI.

(14) * John hasn’t understood, but he has read – any of my books. a. *John hasn’t understood any of my books, but he has read any of my books. b. *John hasn’t understood some of my books, but he has read any of my books.

Some previous analyses make an independent assumption, such as an adjacency requirement, for the licensing of NPIs in coordinated contexts (Van Riemsdijk cited by Hartmann 2000: 71, fn. 8). With an adjacency requirement, the contrast (13-14) may be explained. In (13), the NPI is adjacent to the negation, but it is not in (14). However, it is not clear why we should have such a requirement. Meanwhile, the ellipsis account can capture why (14) is ungrammatical without any further assumption. In (14), the copy in the first conjunct is not problematic since the NPI is bound by the negation. On the other hand, the NPI in the second conjunct cannot be licensed, so the sentence is ruled out. In addition, since the second conjunct is not under ellipsis, it is not eligible to undergo Vehicle Change, the fact that explains the difference in grammaticality between (13) and (14).

83

3.1.2 Morphological mismatches in RNR The empirical facts demonstrate that the RNR target does not have to morphologically agree with one of the conjuncts. Let us consider examples in (15).4 Notice that, in all the examples, the second conjunct contains a verb with either past inflection or a participle morpheme, but the RNR clause requires the verb to be a root form. Therefore, there is a mismatch in verbal morphology between the RNR and antecedent clause.

(15) a. John WON’T , but Mary already HAS – [negotiated her salary with the company]. b. Mike told his advisor that he NEEDS TO , and we all knew that he SHOULD’VE – [chosen his dissertation topic] sooner. It’s too late to meet the deadline to submit the prospectus this semester. c. Bill DIDN’T , but MIKE solved the puzzle.

The non-movement accounts predict such a mismatch to be unacceptable. Under the strict phonological deletion accounts, the deletion of the root form should be impossible because the target word string is different between the conjuncts. Under the multiple dominance accounts, sharing should be unavailable for the same reason. Agreement problems arise whether a root or inflected form is used for the shared material. That is, if the root form is used for the shared material, then the second conjunct cannot agree with it morphologically. And if the inflected form is used, the first conjunct cannot agree with it. However, this is also straightforwardly explained under the ellipsis account simply because ellipsis allows this type of mismatch in verbal morphology. I will discuss the question about why

4

As discussed in chapter 1, backwards VP-ellipsis is treated as RNR.

84 ellipsis allows such flexibility in chapter 5. Here, I will just assume that morphology of lexical verbs does not have to match under ellipsis (Warner 1986, Lasnik 1999, and Lightfoot 1999). Let us consider a forward VP-ellipsis example in (16). Notice that there is a mismatch in verbal morphology between met and meet. This indicates that the identity of verbal morphology does not have to match in VP-ellipsis.

(16) Bill met Prof. Barnes yesterday, and I will this afternoon.

It is natural to assume that we should see the same effect in RNR, if RNR is a type of ellipsis. I argue that this is exactly what happens in (15). The morphological mismatch for the lexical verbs can be tolerated in RNR, meaning that the ellipsis of the root form is licensed by the inflected form in the antecedent clause. There is a restriction on this morphological mismatch. It is also well known that when the copula be (and also an auxiliary have) is involved in mismatches, it must be realized in forward VP-ellipsis. Let us consider (17). The tense of the copula in the antecedent clause is past and in the ellipsis clause the copula needs to be realized as an infinitive, so there is a mismatch in morphology. (17a) is acceptable with the copula overtly pronounced, and (17b) is not with the ellipsis of the copula. This indicates that, unlike for lexical verbs, morphological mismatch for the copula is not tolerable under ellipsis.

(17) a. Jane was here, and I will be , too. b. *Jane was here, and I will , too.

85 The same distribution holds in RNR (Bošković 1997, 2004). The copula needs to be overtly pronounced when the morpheme does not match. That is why (18a) is not acceptable, but (18b) is.

(18) a. *John WON’T , but MARY already HAS been picked for the team. b. John WON’T be , but MARY already HAS been picked for the team.

Interestingly, however, Vries (2007) notes that morphological identity needs to be strictly observed in Dutch RNR constructions. 5 For instance, in (19), if the lexical verb of the first conjunct is inflected as plural but that of the second conjunct as singular, the deletion of kop-en ‘buy-PL’ in the first conjunct is blocked. Notice that in the same circumstance, English would allow RNR of the lexical verb even when the morphology does not match, as we observed in (15). Thus, this is an interesting puzzle, which I will leave open for the future research.

5

Bartos (2001) claims that the tense morphemes do not match in the conjuncts in (ia), but the ellipsis appears to be still licensed in Hungarian. Thus, this shows morphological mismatches under ellipsis in Hungarian as well. On the other hand, (ib) shows the same explanation does not hold for RNR. Then, the question arises why RNR cannot tolerate a tense mismatch in Hungarian, which I do not have an answer for at this point. (i) a. Péter tegnap vásárolt, én pedig tegnapelőtt yesterday shopped I and yesterday-before shopped b. *Péter tegnap , én pedig tegnapelőtt vásárolt yesterday shopped I and yesterday-before shopped ‘Peter did the shopping yesterday, and I did the day before yesterday.’

(Ellipsis) (RNR) (Bartos 2001:(8-9))

86 (19) *… dat jullie

boeken en ik CD’s koop

… that you-PL books buy-PL and I CDs buy-SG ‘that you buy books and I buy CDs.’ (Vries, 2007: (19c))

3.1.3 Sloppy identity in RNR Let us consider a RNR example in (20) where the RNR target contains a pronoun. It is possible for the pronoun to refer to the subject in either the first or the second conjunct, as in (20a-b). Such an interpretation is also known as a strict identity reading.

(20) Mary LOVED, but Sue HATED – her twin sister. a. Mary loved Mary’s twin sister, but Sue hated Mary’s twin sister. b. Mary loved Sue’s twin sister, but Sue hated Sue’s twin sister.

The multiple dominance accounts can derive the readings in (20a-b). A single occurrence of the pronoun merged to both conjuncts simultaneously can refer to by the subject in the first conjunct for (21a) or the one in the second conjunct for (21b). Therefore, the interpretations of (20a-b) are derived from each tree structure of (21a-b), respectively.

(21) a.

&P wo TP TP 6 6 Maryi LOVED Sue HATED

b.

DP 6 heri twin sister.

&P wo TP TP 6 6 Mary LOVED Suej HATED DP 6 herj twin sister

87 There is another possible reading in (20). The pronoun can refer to the subject of its own conjunct, which results in a sloppy interpretation in (22) (Jacobson 1999).

(22) Maryi LOVED, but Suej HATED heri/j twin sister. = Mary loved Mary’s twin sister, but Sue hated Sue’s twin sister.

(Sloppy identity)

The availability of sloppy identity is an insurmountable problem for the multiple dominance accounts in that a single occurrence of the pronoun must be bound by each DP subject of both conjuncts at the same time, as shown in (22). As a result, the pronoun has only one occurrence but multiple meanings, hence uninterpretable. The same challenge holds for the multiple dominance accounts in Korean in (23) where the only possible interpretation is a sloppy reading. 6 To represent the sloppy reading under the multiple dominance accounts, the reflexive must be bound simultaneously by the subjects of both conjuncts. It creates that the same problem as with the English pronominal variable – the reflexive casin would be one occurrence having multiple binders.

(23) Johni-un ahop-si-ey, (kuliko) Maryj-nun yel-si-ey J.-TOP 9 o’clock-at (and)

M.-TOP

[casini/j-uy cip-ulo kass-ess-ta].

10 o’clock-at self’s

house-to go-PAST-DEC

‘John went to John’s house at 9, and Mary went to Mary’s house at 10.’ (Sloppy identity)

Let us turn to how the deletion accounts – strict phonological deletion and ellipsis – handle strict and sloppy interpretations of RNR. It is well-known in the forward ellipsis literature that the

6

Hestvik (1992:1) claims that reflexives do not allow strict identity readings in VP-ellipsis (cf. Sag 1976).

88 ellipsis site can have multiple interpretations that are unavailable when overtly pronounced. The classic example of sloppy identity in the forward ellipsis context is shown in (24), due to Sag (1976), Williams (1977), and Reinhart (1983). The sentence in (24) is multiply ambiguous: strict identity (24a), the third-party reading (24b), and sloppy identity (24b).

(24) Mary loves her twin sister, and Sue does, too. a. Mary loves Mary’s twin sister, and Sue loves Mary’s twin sister (Strict identity) b. Mary loves Jean’s twin sister, and Sue loves Jean’s twin sister

(Third-party identity)

c. Mary loves Mary’s twin sister, and Sue loves Sue’s twin sister

(Sloppy identity)

The multiple readings in (24) are assumed to be available because the ellipsis site has a covert counterpart of the antecedent at the point of interpretation. The pronoun can be treated as a referential, so the pronoun refers to the subject of the first conjunct (24a), or to the third individual from the context (24b). The pronoun can also be a bound variable, and sloppy reading arises when the pronominal variable is bound by the subject of the second conjunct (24c). Under an ellipsis account, the availability of sloppy readings in a sentence where a bound pronoun is involved in the RNR target can be easily explained. Let us consider (20) again, repeated here in (25), which yields multiple readings. The ellipsis account assumes that there are two occurrences of the RNR target, which in turn means there are also two occurrences of the pronoun: one for the antecedent DP, and the other for the elided DP. Being referential, the pronoun in both positions can refer to either (25a) or (25b). For the sloppy reading, the only assumption needed is that the pronoun in the elided DP can serve as a variable bound by the subject of the first conjunct, which is exactly the same as VP ellipsis.

89 (25) Mary LOVED , but Sue HATED her twin sister. a. Mary loved Mary’s twin sister, but Sue hated Mary’s twin sister.

(Strict identity)

b. Mary loved Sue’s twin sister, but Sue hated Sue’s twin sister.

(Strict identity)

c. Mary loved Mary’s twin sister, but Sue hated Sue’s twin sister.

(Sloppy identity)

Likewise, the strict phonological deletion accounts have no problem with the ambiguities because two instances of the RNR target are assumed – the one pronounced and the other unpronounced. At PF, the deletion of the RNR material is licensed because the phonological forms strictly match. As shown in (25), the form of the RNR target matches, so RNR is possible under the accounts. Note that in forward VP-ellipsis, the pronominal variable cannot refer to Sue in (26). In this respect, it is surprising that (25b) is available in RNR.

(26) Mary loved her twin sister, and Sue did, too. ≠ Mary loved Sue’s twin sister, and Sue loved Sue’s twin sister

I propose that the Backwards Anaphora Constraint (Langacker 1969) blocks the pronoun her in the first conjunct from referring to Sue. Meanwhile, the constraint does not apply to (25) since there exists no overtly pronounced pronoun preceding its binder in coordination; Rather, the pronoun here is elided. As we have observed in section 3.1.1, some flexibility in identity is allowed between the proper name and its pronominal correlates under ellipsis (cf. Fiengo and May’s (1994) Vehicle Change). In other words, the elided pronominal in the first conjunct can be identified as its proper name correlate Sue for the reading (25b).

90

3.1.4 Parallelism in RNR Wilder (1999) argues that the multiple dominance accounts can predict that a sentence like (27) is possible, where the two conjuncts are not structurally parallel. He predicts that, as long as linearization works, even non-parallel RNR structures under coordination should be available. For example, the object DP the book in (27) can be linearized as long as it occupies the right edge position of the first conjunct.

(27) John should fetch , and give the book to Mary. TP 3 John T’ 3 T &P should wo vP1 &’ r 3 fetch & vP2 3 give vP 3 DP vP the book 3 V to Mary (Wilder 1999: 595)

Wilder assumes that it does not matter where it occupies in the second conjunct. Note that the RNR target in the second conjunct is not in the right edge position. The first (vP1) and the second (vP2) conjunct are linearized as in (28a-b), respectively, and the linear order between the conjuncts is given in (28c).

91 (28) a. vP1 = fetch>the>book b. vP2 = give>the>book>to>Mary c. vP1>&’ = {fetch}>{and>give>to>Mary}

The only place that the RNR target the book can be linearized concurring the previous orders (28a-b) is between give and to. Therefore, the whole linearization in (29) is compatible with Wilder’s multiple dominance account.

(29) John>should>{fetch}>and>{give>the>book>to>Mary}

However, Wilder acknowledges that his account also predicts (30) to be allowed, but it is not in fact grammatical. Since the shared DP occupies the right edge of the first conjunct, this sentence should be linearizable under Wilder’s multiple dominance account. He stipulates that parallelism is operative in (30), but not in (27).

(30) *John met , and Mary laughed. (Wilder 1999: 598)

Norvin Richards (p.c.) notes that an example in (31) that is structurally identical to (27) is not acceptable, and suggests that (27) may not be an ordinary RNR example. However, the multiple dominance hypothesis predicts (31) to be grammatical, since it has exactly the same linearization as (27).

92 (31) *Mary congratulated , and gave the winner a prize. (Norvin Richard, p.c.)

Then the question arises why (27) is acceptable, but not (31). Paul Hagstrom (p.c) points out that one possible explanation would be that (27) involves “reanalysis” of fetch and give as a complex predicate. In support of this, if we pair two verbs that are not as easily understood as a single action, such as critique and give in (32), the sentence is dramatically degraded.7

(32) ?*John should critique , and give the book to Mary.

To sum up, Wilder’s multiple dominance account encounters paradox because even a nonparallel RNR structure is linearizable, such as (27), and others are not, such as (30). This indicates that the right edge effect alone cannot account for the examples above, and that it seems safe to conclude that parallelism also needs to be assumed under the multiple dominance account. This is a good place to introduce Bachrach and Katzir’s (2006a) account. Bachrach and Katzir have taken up the problematic (27) with another version of the multiple dominance account and claimed that in fact (27) is not linearizable. Their multiple dominance analysis is based on Fox and Pesetsky’s (2005) cyclic linearization analysis where the linear order needs to be preserved between spellout domains (i.e. vP and CP). As Wilder (1999) did, they assume that the spellout of the RNR target will be delayed if it is not completely dominated. Complete Dominance is defined in (33).

7

Since should appears in (32), it cannot be just the modality of should that allows (27) and not (31).

93 (33) Complete Dominance A node X completely dominates a node Y iff (a) X is the only mother of Y, or (b) X completely dominates every mother of Y. The set of nodes completely dominated by X will be called the Complete Dominance Domain of X. (Bachrach and Katzir 2006a: (41))

Let us consider the definition with an example in (34). At the point that the derivation reaches vP1 and vP2, the RNR target is not completely dominated by anything in the structure; i) neither vP1 nor vP2 is the only mother of the target or ii) neither of them completely dominates every mother of the shared DP. Thus, even if vP in each conjunct is a spellout domain, the shared DP is not spelled out, as shown in (34a-b).8 For the same reason, neither TP1 nor TP2 completely dominates the shared material. It is &P that completely dominates the RNR target for the first time because &P is the first mother that completely dominates every mother of the shared DP.9 Bachrach and Katzir claim that the phrase merged immediately to the coordination will be able to spell out the RNR target. Thus, after CP is merged, the RNR target will be spelled out, as shown in (34c).

8

Note here that unlike Wilder, Bachrach and Katzir assume that the shared material which is not completely dominated is not even linearized in each conjunct (34a-b). 9 Note that any node above the &P also achieves complete dominance for the shared DP.

94 (34)

CP 3 : &P qp TP1 &’ 3 3 DP vP1 & TP2 Bill t 3 V DP vP2 wrote John ry V DP reviewed 5 her paper a. vP1: b. vP2: c. CP:

(plus to be linearized)

Preserving the linear order in the previous spellout domains, the RNR target can be linearized at the end of the sentence, as shown in (35a). This is how their account captures the right edge effect in RNR. (35b) is a linearization where the shared DP is linearized in both places. Bachrach and Katzir rule out (35b) on the grounds that reflexivity violation results; her paper precedes and follows itself.10

(35) a. b. * 10

There is another possibility that Bachrach and Katzir did not consider and is in fact problematic for their account. Their account does not rule out (i). Since the RNR target is the first time to be linearized and it can be placed anywhere. This problem is partly due to their assumption that &P is not a spellout domain. Consequently, the linear order between the first and the second conjunct has not yet been determined when &P is complete. Therefore, the linear order between the conjuncts and the RNR target would be set at the same time in the CP phase, so the RNR target can be linearized in (i). (i) *

95 Let us return to (27) and examine how Bachrach and Katzir’s analysis can explain why (27), repeated in (36), is not linearizable. Each conjunct of (27) will be spelled out as (36a-b). Upon completion of the coordinate structure, CP is the first phase that completely dominates the shared materials, as in (36c). Thus, the shared DP can now be linearized in the CP spellout domain.

(36) John should fetch , and give the book to Mary. a. vP1: b. vP2: c. CP:

(plus to be linearized)

To account for the non-linearizability of (36), Bachrach and Katzir assume that each conjunct is spelled out before the coordinate structure is formed, except the shared constituent. Therefore, fetch in the first and give to Mary in the second conjunct have already been spelled out and linearized, as in (37a-b). They further assume that once linearized, the linear order cannot not be modified any more (“No Modification”, Bachrach and Katzir (2006a)).

(37) a. vP1 = fetch b. vP2 = give-to-Mary

When the shared DP is spelled out in the next spellout domain, it cannot be linearized between give and to Mary, because placing the book between give and to Mary would lead to

96 modification of a string that has been already linearized. By hypothesis, this is a violation of “No Modification”. Therefore, the derivation crashes and linearization of (36) is not licit.11 However, Bachrach and Katzir’s analysis still does not capture why (30), repeated in (38), is not linearizable. The completion of the coordinate structure would be able to spell Mary out, and the RNR target can be linearized in the subject position of the second conjunct, as shown in (39). Note that this is not a violation of “No Modification” because the coordinator and the second conjunct have not yet been linearized. Their linear order would be determined at the same time when the RNR target is linearized (i.e. CP phase). Therefore, Bachrach and Katzir’s analysis would predict the linear order of (39) to be preserved, leading to grammaticality of (38). On the contrary, (38) is not an acceptable sentence.

(38) *John met , and Mary laughed. a. TP1 = John-met b. TP2 = laughed c. CP =

(plus to be linearized)

(39) John>met>and>Mary>laughed. a. TP1 = John-met b. TP2 = Mary-laughed c. CP =

11

This implies a slightly more complicated view of linearization. For this to work, linearization cannot simply be saying “this precedes that”, but also “this is adjacent to that”. It must be “this immediately precedes that”.

97 This indicates that Bachrach and Katzir’s analysis needs to assume parallelism to rule out (38). If this is true, however, it means that the ungrammaticality of the other examples, such as (32) and (36), can also be attributed to the lack of parallelism; the first conjunct is a transitive and the second conjunct is a ditransitive construction. In other words, regardless of the preservation of linear order, those examples have independent reasons to be ruled out, so they do not achieve the goal to show that the RNR target must be linearized in the right edge position. It is not clear that an appeal to parallelism will succeed more generally for other examples from Bachrach and Katzir (2006a). Let us consider (40). The direct object is merged to the right edge position, so that (40) can be linearized. However, notice that (40) does not seem to obey parallelism any more than (32) did.12 The paradox still remains in Bachrach and Katzir (2006a), too.

(40) John enjoyed reading and gave to his wife – the Harry Potter series.

Let us turn to the deletion accounts – the strict phonological deletion and the ellipsis accounts. Parallelism is a well-known requirement for deletion; the conjuncts must be parallel for ellipsis to

12

There are other problems for Bachrach and Katzir (2006a). The only way to derive a linear order out of the set of linearizations in (36a-c) is to place the shared material in the rightmost edge of the sentence, as shown in (i). (i) *John congratulated and gave the prize – [the winner]. a. vP1 = congratulate b. vP2 = gave-the-prize-[the winner] c. CP = The sentence in (i) is linearizable under Bachrach and Katzir (2006a), but it is not a grammatical RNR sentence. They are silent on this issue but this is rather a serious problem for their account since a similar linearization would be possible as in (ii). Bachrach and Katzir need to put the RNR target in (ii) at the end of the sentence to preserve ordering among spellout domains. However the same strategy cannot be applied to (i). In chapter 5, I will provide a solution with ellipsis for why (ii) is a good RNR sentence but not (i). (ii) Josh will donate to the library, and Maria will donate to the museum – each of these old novels.

98 be licensed. 13 If RNR underwent deletion, we expect that parallelism also holds in RNR. Therefore, there is no need to make any particular assumption for the deletion of the RNR target. The deletion accounts assume parallelism (Ha 2007, Hartmann 2000), so they predict that any RNR example will be ruled out where the conjuncts do not meet syntactic or semantic identity requirements. 14 Therefore, the accounts correctly rule out examples, such as (36) and (38). However, they cannot account for the acceptability of (40). Here I will just leave an account of (40) open, since it requires a further discussion of licensing conditions and the structure of ditransitive constructions. The issues on parallelism and identity conditions will be thoroughly discussed in chapters 4 and 5. And in chapter 5, I will argue that (40) is actually parallel at some level. To summarize so far, I have presented problems for the previous non-movement accounts, and argued that those problems do not arise in the ellipsis account. The comparison among the non-movement accounts is summarized in Table 1.

Table 1. Comparison between non-movement accounts Vehicle Change Morphological Mismatches Sloppy Identity Parallelism

Multiple Dominance No No No No

Strict PF deletion No No Yes Yes

Ellipsis Yes Yes Yes Yes

3.1.5 Another argument for ellipsis: Certain dialects of British do An interesting argument comes from certain dialects of British English. Chalcraft (2006) observes that in those dialects, a non-finite form of do can replace the site of VP ellipsis.

13

I will refer to parallelism requirements as identity conditions. While Hartmann assumes syntactic identity, Ha (2007) assumes semantic identity. This is what makes the ellipsis account differ from the strict phonological deletion accounts. 14

99 (41) a. John said he would resign today. We will have to see if he does do. b. First, Mary talked to John, and then Bill did do, too. c. I don’t know whether to go. I might do. (Chalcraft 2006: 7)

Chalcraft asserts that this usage of do can only be used in an ellipsis context, but not in other contexts, such as those in (42). The sentences in (42) are not acceptable when the VP is pronounced.15

(42) a. *They might do leave tomorrow. b. *He could have done finished by now. c. *It was doing blowing a gale. (Chalcraft 2006: 8)

Interestingly, this usage of do can also be found in RNR constructions, as shown in (43). Given the evidence that do in these British dialects can only be used in ellipsis in (41-42), the availability of do in RNR is naturally explained if RNR involves ellipsis. In fact, there is no other way to explain this if RNR is not ellipsis. Therefore, this set of examples supports the argument that RNR is an ellipsis phenomenon.

15

At this point, it cannot be determined whether this do should be treated as a licensor for eliding its complement VP or an anaphor for the VP predicate. It appears that one could claim either that there is an unpronounced syntactic derivation following do or that do behaves like so in do so anaphora. Since an overtly pronounced VP cannot co-occur with this usage of do in (42), whereas VP can be pronounced in normal VP ellipsis, the anaphoric do analysis seems to be more plausible. For the purpose of our discussion, the similar behavior between RNR and ellipsis with respect to this usage of do is enough to support that RNR is an ellipsis phenomenon.

100 (43) a. Tom said he would do, and Bill actually did – audition for the choir. b. Mary should have done, and Mike probably already has – lit the fire. (Chalcraft 2006: 8)

3.1.6 Ellipsis vs. Strict phonological deletion The ellipsis account differs from the strict phonological deletion analysis since the RNRed material is not just deleted under a matching form-identity with its antecedent, but allows for various kinds of changes in the RNRed site, such as sloppy dentity, and Vehicle Change, as discussed in the previous sections. These relative flexibilities support an account for RNR based on ellipsis, rather than just strict phonological deletion. On the other hand, there are some cases which favor the strict phonological deletion accounts, but potentially disfavor the ellipsis account. Those cases include below word-level deletions, an elision of DP, and non-constituent deletion. I will just list these cases in this section, and return to them in chapter 5, where ellipsis features for RNR are proposed. There, I will argue that these cases are not problematic for the ellipsis account. First, the below-word-level deletion would not be a problem for the strict phonological deletion analysis since by hypothesis, only PF is affected in RNR. Therefore, any deletion (even of morphemes) could be done if the RNR target has an identical and structurally parallel (subword) constituent in the antecedent. However, under an ellipsis account, such deletions are hard to conceptualize, due to the assumption that RNR is predicted to observe the same distribution as forward ellipsis. Therefore, the contrast in (44) challenges the ellipsis view of RNR since RNR allows the ellipsis of sub-word constituents in (44a) but there is no below word-level forward ellipsis in (44b).

101 (44) a. This hypothesis suffers from both UNDER- and OVER-generation. b. *This hypothesis suffers from both UNDER-generation and OVER-.

Second, one of the main differences between RNR and ellipsis is whether a DP alone can be unpronounced. RNR allows an object DP to be elided, but forward ellipsis does not, as shown in (45a-b). For both the strict phonological deletion and the ellipsis accounts, it needs to be explained why it is only possible to elide an object DP in RNR (45a), but not in forward ellipsis in (45b). In the strict deletion point of view, there is not much further to say about this because the analyses have been developed just to account for the deletion of the RNR target in the first conjunct. Thus, the distinction between (45a) and (45b) is simply out of the question. On the other hand, in the ellipsis point of view, the differences raise a serious challenge because if RNR is an ellipsis phenomenon, RNR is also predicted to observe the same distribution as forward ellipsis.

(45) a. John enjoyed , but Jane disliked – the TV show. b. *John enjoyed the TV show, but Jane disliked .

It has been observed in the ellipsis literature that only constituents are eligible to be elided (Postal 1986, Baltin and Postal 1996, Phillips 1996, 2003). The examples in (46) show that pseudogapping is not allowed for non-constituents. 16 The verb-preposition sequence does not form a constituent, so the sentence is ungrammatical.

16

Pseudogapping is assumed to be involved in vP-ellipsis.

102 (46) a. *John didn’t stand near Cathy, but Mary did *(near) Sue. b. *Paul talked about constituency in vP-ellipsis, and Jason didn’t *(about) case-matching in sluicing (Phillips 2003: 48)

On the other hand, as introduced in the previous chapters, even non-constituents appear to be elided in RNR in (47). The RNR target, which consists of some part of the embedded and matrix predicates, can never form a constituent, but (47) is a perfect RNR sentence.

(47) [Bill-un B.-TOP

[Mary-ka ] kuliko M.-NOM math-ACC well do-C

[John-un

[casin-i

J.-TOP

self-NOM math-ACC well do-C

think-PRES-DEC

CONJ

swuhak-ul cal hanta-ko] sayngkak-han-ta]. think-PRES-DEC

‘Bill thinks that Mary does math well, and Johni thinks hei does.’ (Korean)

Thus, the question would be if RNR is an ellipsis phenomenon, why it appears to be exceptionally able to target non-constituents. The strict phonological deletion account naturally allows for the possibility of non-constituent deletion since it assumes that RNR is just a PF condition without syntactic constraints. There seem to be two paths that we could tackle this challenge under the ellipsis account. The first path would be just denying the claim that only constituents can be elided in ellipsis. In that connection, it does seem that the case for a constituent condition on ellipsis is not as strong as it might appear. In the case where forward ellipsis of a non-constituent is disallowed, RNR of the

103 non-constituent is also disallowed; see (48). So far, I have not found any pair of examples that distinguish impossible non-constituent ellipsis from possible non-constituent RNR in English.

(48) a. *John DIDN’T Cathy, but Mary DID stand near Sue. b. *Paul DIDN’T constituency in vP-ellipsis, and Jason DID talk about casematching in sluicing.

The second path, which I will eventually pursue in chapter 5, is to accept the claim that ellipsis indeed targets constituents and propose that RNR does, too. I will suggest that RNR also targets constituents by introducing multiple ellipsis features. To summarize this section 3.1, I have argued that the previous non-movement accounts encounter many problems but the ellipsis account can account for them. In section 3.2, I will discuss challenges to non-movement accounts posed by a movement analysis, and make an attempt to resolve them under ellipsis.

3.2 Potential problems for the ellipsis account and solutions Sabbagh (2007) presented interesting cases which only movement analyses would be able to account for. The evidence includes i) the RNR target, containing relational modifiers, and ii) scope ambiguity. Those are properties that are difficult to account for in non-movement accounts, but I will argue that the ellipsis account has solutions for each of them.

3.2.1 Relational Modifiers Abbott (1976), Jackendoff (1977), and Gazdar (1981) have claimed relational modifiers, such as similar, together, different, respectively, are problematic for the deletion accounts of RNR. Let us

104 first examine how the relational modifiers behave. Relational modifiers require plural subject NPs, as shown in (49a-b). While (49a) is acceptable with plural NPs in its subject position, (49b) is not with a singular subject NP.

(49) a. John and Mary enjoyed reading the same book. b. *John enjoyed reading the same book.

Sabbagh (2007) claims that the plurality requirement for relational modifiers is also satisfied if the modifiers take scope over conjunction of VP or TP. That is why, he claims, the rightward ATB movement analysis of RNR can explain cases where the RNR target includes the modifiers. Let us consider how the movement accounts can explain those cases with an RNR example in (50). Given the assumption that (50b) is the underlying form of (50a), the puzzle is why only the RNRed sentence is grammatical in (50a). First, it is straightforward why (50b) is ungrammatical, because the relational modifier does not have plural subject NPs in each conjunct. Sabbagh claims that RNR can salvage the grammaticality in (50a) by overtly moving the target out of coordination; the plurality requirement of the relational modifier is satisfied after rightward movement of the RNR target because it takes scope over the conjunction of TP.

(50) a. Peter sings and Mary whistles – a similar tune. b. *Peter sings a similar tune and Mary whistles a similar tune.

How can the non-movement accounts solve this puzzle? Sabbagh claims that the puzzle poses serious challenges for the non-movement accounts. The challenges for the accounts are based on the fact that the RNR target stays in-situ, so that there is no way that relational modifiers can take

105 scope over the conjuncts. Let us first consider what the multiple dominance analyses need to say about this. Recall that under the multiple dominance hypotheses, there exists only one occurrence of the shared RNR constituent. That way, the RNR target can be linearized at the rightmost position of the whole sentence. The tree structure of (50a) can be drawn as (51). Note that the relational modifier in-situ does not take scope over the conjuncts, so the plurality requirement cannot be satisfied.

(51)

&P qp TP1 &’ 3 3 DP vP1 & TP2 Peter t 3 V DP vP2 sings Mary ry V DP whistles 5 a similar tune

The multiple dominance accounts cannot assume “overt” rightward movement of the shared material, as in (52a), because the shared DP would asymmetrically c-command everything, meaning that it should precede both conjuncts according to the linearization algorithm as in (52b).17

17

The possibility that the relation modifier contained in the shared material undergoes “covert” movement will be discussed in section 7.5.2.

106 (52) a.

&P qp &P DP qp 6 TP1 &’ a similar tune 3 3 DP vP1 & TP2 Peter t 3 V DP vP2 sings Mary r V whistles

b. a>similar>tune>Peter>sings>Mary>whistles

The puzzle in (50a-b) poses challenges for the deletion accounts, too. Since (50a) is derived from (50b) by deleting the RNR target in the first conjunct, the relational modifier stays in-situ in each conjunct, as shown in (53). Therefore, (50a) is predicted to be as ungrammatical as (50b). They both fail to satisfy the plurality requirement of the modifier.

(53) *Peter sings
and Mary whistles a similar tune.

As one of the strict phonological deletion accounts, Hartmann (2000) acknowledges but dismisses the problem that (50a-b) pose, pointing out that it is subject to cross-linguistic and individual variation. Hartmann observes that the equivalent German example is not grammatical, shown in (54), and that not every English informant accepted (50a).18

18

Without appealing to cross-linguistic and individual variations, I think the phenomenon in (50a-b) can still be within reach of the deletion accounts, on the grounds that it is possible to assume that the first verb in (50a) is intransitive. (50a) is not derived from (50b), so it is not truly a RNR example. Rather, it means something closer to Peter sings, and then Mary starts whistling with a similar tune. This hypothesis predicts that a structure in which the verb in the first conjunct is unambiguously and necessarily transitive will be bad. The prediction appears to be borne out in (i).

107 (54) *Hans singt und Maria pfeift – ein ahnliches lied.

(= (50a))

Hans sings and Mary whistles – a similar tune. (Hartmann 2000:79)

Abels (2004) points out that the availability of the distributive reading in (55) is similarly problematic for the deletion analyses. The reading we are interested in is the distributive reading (55a), although the collective reading is also possible (55b). In addition, the intransitive use of sang for the first conjunct is available in (55c).

(55) Carrie sang, and Mike recorded – two very different songs. a. Carrie sang a song, and Mike recorded a song, and they are different from each other. b. Carrie sang two different songs (with someone), and Mike recorded two very different songs (with someone), too. c. Carrie did some singing, and meanwhile Mike recorded two songs, one quite different from the other. (Abels 2004: (16))

Abels compares (55) with the VP-ellipsis examples in (56), and claims that the distributive reading is not available in VP-ellipsis (56a), but only the collective reading is available in (56b). It indicates that only collective reading in (55) can be derived by ellipsis, not the distributive reading. Therefore, the distributive reading of (55a) should be explained in a different manner.

(i) Mary and I have very different taste in fine art. We have seen a lot of paintings in the last couple of days. For example, *?I like, but Mary absolutely hates – a similar painting.

108 (56) Carrie sang very different songs, and Mike did , too. a. *Carrie sang a song, and Mike sang a song, too. The two songs were different. b. Carrie sang two different songs (with someone) and Mike sang very different songs (with someone), too. c. *Carrie sang two different songs and Mike sang something, too.

A conclusion we can derive from the examples in (55) and (56) is that PF-deletion, including ellipsis, does not involve the distributive reading of those examples. This implies that the different readings in (55) do not all arise from the same source, but actually from a kind of structural ambiguity. One source is the RNR-type deletion (55b), another source has an intransitive verb in the first conjunct (55c), and the third source may have an additional coordinate structure (55a), which is left open in the deletion accounts. Would there be a way for the ellipsis account to capture why the distributive reading is available in RNR but not in VP-ellipsis? I propose that there is. The full explanation, however, requires detailed discussions of what Across-The-Board constructions are, which will be discussed in chapters 6 and 7. Here I will start the discussion by questioning whether Sabbagh’s assumption is correct; namely, that only overt movement of the relational modifier can explain the grammaticality of (50a). It is relatively easy to make the case that the relational modifiers do not have to move overtly. Let us consider (49a), repeated here in (57a). If the modifiers must move overtly, the tree structure of (57a) would look like (57b).

109 (57) a. John and Mary enjoyed the same book. b.

TP wo TP RM 6 the same book John and Mary enjoyed t

The evidence that such a string vacuous movement is blocked comes from scope interactions between the moved constituent and the rest of the conjuncts. The fact that inverse scope is not available in (58) indicates that the universal quantifier cannot undergo string vacuous rightward movement (p.c. Danny Fox).

(58) Some student thought that Mary loves every professor.

(*∀>∃)

We can conclude that the plurality requirement of relational modifiers must be considered at LF, and that is actually what the previous literature suggests. Carlson (1987) argues that the modifiers need to take scope over the plural NPs at LF (cf. Beck 2000, Gawron and Kehler 2004). The LF form of (57a) would be (59) where the relational modifier takes wide scope over the plural subject NPs.

(59) [The same book x] [John and Mary enjoyed reading x]

If this is on the right track, we lose the exclusive benefit of the overt movement analyses of RNR to account for the behavior of relational modifiers in RNR. The real issue is then whether nonmovement analyses have ways to move relational modifiers out of conjunction covertly. Any

110 non-movement accounts would be able to explain the puzzle (50a-b), repeated in (60a-b), if they can move the relational modifier at LF in (60a) but block such movement in (60b). I will propose that this is what the ellipsis account predicts. The proposal requires discussion of Across-TheBoard constructions, and the availability of Across-The-Board movement at LF must be reevaluated (cf. Bošković and Franks 2000). Therefore, I will return to how the ellipsis analysis can capture the differences between (60a-b) in section 7.5.2.

(60) a. Peter sings and Mary whistles – a similar tune. b. *Peter sings a similar tune and Mary whistles a similar tune.

The multiple dominance accounts would also be able to account for (60a-b) if they can show that a single occurrence of the shared material would be able to scope over in covert syntax. This is in fact what Bachrach and Katzir (2006b) claim. Their claim will be reviewed in chapter 7, too.

3.2.2 Korean relational modifier kakkak As introduced in chapter 2, there is a distributive marker kakkak ‘each, respectively’ in Korean, which behaves similarly to English relational modifiers. And the marker has been argued to be problematic for any deletion accounts of RNR (Chung 2004, Yoon and Lee 2005). 19 Let us consider (61a-b), for example. Like English relational modifiers, the Korean distributive marker needs to satisfy plurality requirement. It needs to take scope over the conjuncts, and (61a) does not satisfy this condition. Therefore, if (61b) were simply derived from (61a) by the deletion of the distributive marker and the verb in the first conjunct, it is puzzling that (61b) is grammatical.

19

I will account for the Korean dummy plural marker with ellipsis in chapter 5.

111 (61) a. *Mary-ka chayk-ul

kakkak

il-ess-ta

kuliko

M.-NOM book-ACC respectively read-PAST-DEC CONJ John-i sinmwun-ul

kakkak

il-ess-ta.

J.-NOM newspaper-ACC respectively read-PAST-DEC b. Mary-ka chayk-ul, kuliko John-i shinmun-ul

kakkak il-ess-ta.

M.- NOM book-ACC, CONJ J.-NOM newspaper-ACC respectively read-PAST-DEC. ‘Mary read a book, and John read a newspaper, respectively.

The movement analyses can account for why (61b) is grammatical since they assume that the marker overtly moves out of coordination where it takes wide scope, satisfying its plurality requirement. Chung (2004), and Yoon and Lee (2005) claim that the multiple dominance account can explain the puzzle, too. They claim that the plurality requirement is satisfied if the distributive marker is within the scope of the conjuncts. Let us see the tree structure of (61b) in (62). The VP, including the distributive marker (DM), is assumed to be shared. Given their assumption, the marker is in the scope of the two conjuncts, thus satisfying the plurality requirement.

(62)

&P qp TP1 &’ 3 3 DP VP & TP2 Mary-NOM t 3 DP DP VP book-ACC John-NOM ry DP VP newspaper-ACC3 DM V kakkak read-PAST-DEC

112 However, this explanation falls apart for the following examples in (63a-b). In (63a-b), there are two distributive markers in each conjunct.

(63) a. MARY-NUN Chelswu-ka NONMWUN-UL, kuliko SUE-NUN John-i M.-TOP

C.-NOM

article-ACC,

CONJ.

S.-TOP

SINMWUN-UL

John-NOM newspaper-ACC

(??kakkak1) ilk-ess-tako (kakkak2) mal-hass-ta. (respectively) read-PAST-C (respectively) said-DEC ‘Mary said that Chelswui read newspapers, and Sue said that John read articles, respectively.’ b. John-uy emeni-nun John-i hansi-ey, kuliko Mary-uy emeni-nun Mary-ka twusi-ey John’s mother-TOP J.-NOM at 1pm CONJ. Mary’s mother-TOP M.-NOM at 2pm (??kakkak1)

cip-ey on-ta-ko (kakkak2) sayngkak-hass-ta.

(respectively) home come-C (respectively) thought-DEC ‘John’s mother thought John would come home at 1pm, and Mary’s mother thought that Mary would come home at 2 pm, respectively.’

It is interesting to note that the distributive marker can appear in the matrix VP (DM2), but crucially not in the embedded VP (DM1). The multiple dominance hypothesis, however, predicts that the distributive markers in both places would be possible. The hypothesis correctly explains why DM2 can exist in the matrix clause, because DM2 is within the scope of coordination of matrix TP. However, the hypothesis incorrectly predicts that DM1 would also be possible to exist in the embedded VP because DM1 is in the scope of coordination of embedded CP. Let us consider the structure of (63a) in (64).

113 (64)

&P qp TP &’ 3 wp M.-TOP vP & TP r 3 CP S.-TOP vP 6 ro C.-NOM article-ACC CP VP 6 3 J.-NOM newspaper-ACC DM2 said-DEC respectively VP 3 ??DM1 read-DEC-C respectively

Note that (64) is not different from (62) in that the distributive marker (DM1) can see the subject in each CP. This satisfies the plurality requirement of the distributive marker, and the presence of DM1 is predicted to be possible, contrary to fact. This may not be a problem for the traditional movement analyses of RNR, but this is a problem for Sabbagh’s movement analysis. Let us ignore any issues about how the embedded VP as well as the matrix VP in (63) can move out of the conjuncts at the same time. Let us just assume that the embedded VP and matrix VP can be linearized, with the linear order preserved between spellout domains by cyclic rightward movement. Even if that is the case, the movement analysis also predicts that both DM1 and DM2 would take scope over the coordination, hence satisfying the plurality requirement. Let us first consider the interpretation of DM2 under the traditional movement analyses. Recall from chapter 2 that the distributive marker needs to obey locality. Let us consider (65a-b). The examples in (65) demonstrate that the distributive marker needs to take scope over plural conjunction in the local domain.

114 (65) a. John-kwa Mary-nun chayk hankwon-ul J.-and M.-TOP

kakkak ilk-ess-ta.

book one-CL-ACC each read-PAST-DEC

‘John and Mary each read a book.’ b. *John-kwa Mary-nun [Bill-i J.-and M.-TOP

chayk-ul kakkak ilk-ess-ta-ko] tul-ess-ta.

B.-NOM book-ACC each

read-PAST-C heard-DEC

‘John and Mary heard that Bill each read a book.’

The distributive marker of (65a) covertly moves and takes scope over its immediate clause containing the marker, as shown in (66a). However, the marker in (65b) needs to move to the matrix clause since the embedded clause does not contain any conjunction. The ungrammaticality of (65b) indicates that the distributive marker obeys locality, in that it cannot move further to the matrix clause containing the conjunction of DP where it can take scope over the conjunction, as shown in (66b).

(66) a. [Distributive Op] [TP John-kwa Mary-nun chayk hankwon-ul J.-and M.-TOP

kakkak ilk-ess-ta].

book one-CL-ACC each read-PAST-DEC

b. * [Distributive Op] John-kwa Mary-nun [CP Bill-i chayk-ul kakkak ilk-ess-ta-ko] tul-ess-ta. J.-and M.-TOP

B.-NOM book-ACC each read-PAST-C heard-DEC

Bearing this locality restriction in mind, let us return to (63) and see if the traditional movement analyses can tell why DM1 cannot be realized. The RNR target containing the DM1 needs to move across the embedded CP, so the rightward movement of DM1 out of the coordination would not be considered as a local movement, as shown in (67). Since DM1 is not local to the embedded clause, it does not satisfy its plurality condition. Therefore, it cannot be

115 realized. On the other hand, DM2 is local to coordination of the matrix clause, so that it can appear.

(67)

&P qp &P VP qp 6 &P VP DM2 said 3 6 TP &’ *DM1 read 3 3 M.-TOP vP & TP 3 3 CP t S.-TOP vP 6 3 C.-NOM CP t article-ACC t 6 he-NOM newspaper-ACC t

Note that, however, the locality is captured differently in Sabbagh’s analysis. Sabbagh assumes that it is okay to undergo unbounded rightward movement as long as the linear order is preserved. Given that the RNR targets move to the rightmost edge of the coordination, the linear order of the whole sentence is preserved between spellout domains. This, in turn, means that the locality of DM1 we discussed above is not relevant to Sabbagh’s movement analysis of RNR. Since the linearization works in (67), under Sabbagh’s analysis, the DM1 is predicted to be possible to be realized, contrary to fact.

116 In chapter 7, I will propose that the ellipsis analysis can account for why only DM2 can appear in (63a-b). Again, this needs to follow discussions of the RNR account of Across-TheBoard constructions, so I will return to this puzzle in section 7.5.2.

3.2.3 Scope differences A similar issue arises when the RNR target contains a quantifier. The question is whether the quantifier can take scope over the entire pair of conjuncts by means of Quantifier Raising. As introduced in chapter 2, Sabbagh (2003, 2007) raises challenges for the non-movement views based on scope differences between the RNRed (68a) and overtly pronounced (68b) versions of the same sentence. Sabbagh observes that (68a) allows an inverse scope reading, but (68b) does not.20 It should be clear now why (68a-b) would pose problems for non-movement analyses; If the underlying structure of the deletion view is (68b), and the DP bearing the universal quantifier in the first conjunct is deleted under the identity with its antecedent, then the deleted version must take the same scope with the underlying structure, meaning only surface scope reading should be possible in (68a), too.

(68) a. Some policeman arrested t, but ended up releasing t – every teenager who was near the crime scene.

(∃ > ∀ or ∀ > ∃)

b. Some policeman arrested every teenager, but ended up releasing every teenager. (Only ∃ > ∀)

The challenge will be able to be resolved under the ellipsis account in chapter 7. This issue is closely related to the question of whether ATB movement is possible in covert syntax (cf. 20

Bošković and Franks (2000) claim that there is no LF ATB movement. In (68b), the universal quantifier cannot undergo ATB movement at LF, so inverse scope is unavailable.

117 Bošković and Franks (2000)). In chapter 7, ATB constructions will be revisited and analyzed under a new analysis, and I will propose that constructing ATB at LF is possible. In a nutshell, I will argue that the ellipsis of the RNR target in the first conjunct makes the quantifier in the second conjunct move covertly out of coordination and take wide scope as in (68a). This covert movement is not possible when the object quantifier in both conjuncts is overtly realized as in (68b). To conclude this chapter, I reviewed the non-movement analyses of RNR, and showed that they also encounter many problems, such as Vehichle Change, sloppy identity, and parallelism. It is interesting to note that those problems could be solved if we assume that RNR is a type of ellipsis. Thus, I proposed that RNR be best explained by an ellipsis account. The discussion of licensing conditions for ellipsis is the core part of any literature on ellipsis. If RNR is ellipsis, a question arises as to what licenses ellipsis. In the next two chapters, I will argue for semantic licensing conditions for RNR.

118

CHAPTER 4 LICENSING RIGHT NODE RAISING

Through the last two chapters, I reviewed analyses of RNR found in previous literature, which include movement, strict phonological deletion, and multiple dominance. In chapter 2, I examined the movement analyses. The traditional rightward movement accounts face many problems because extracting the RNR target out of the conjunct would violate locality – for example, islands and the Right Roof Constraint. Adopting Fox and Pesetsky’s (2005) cyclic linearization, Sabbagh (2007) proposed a new movement account where he claimed that movement is immune to locality in case it preserves the linear order between spellout domains. While intriguing, Sabbagh’s analysis also got into both empirical and conceptual problems (see section 2.1.4 for detail). As an alternative to the traditional movement analyses, other researchers have pursued the idea that the RNR target does not move. The crucial benefit of the non-movement analyses was that the locality violations do not arise because the target stays in situ at all times. However, those alternatives face different types of problems, which were discussed in chapter 3. I proposed that RNR is best explained as an ellipsis phenomenon, and showed that RNR shares many of the ellipsis properties, such as morphological mismatch, sloppy identity, and Vehicle Change. And previous non-movement accounts suffered from such ellipsis properties as RNR exhibits. If RNR is a type of ellipsis, the next question to ask would be what licenses RNR. And that is the main concern in this chapter: the licensing conditions for RNR. In particular, I will address whether RNR obeys the same licensing conditions as the other types of ellipsis phenomena, such

119 as VP-ellipsis and sluicing. Previous literature on RNR has rarely addressed the semantic licensing conditions since RNR has been assumed to be a PF phenomenon. The multiple dominance views do not rely on deletion, so they do not predict substantive licensing conditions to hold of RNR. Similarly, the strict phonological deletion views have not attempted to characterize the conditions determining eligibility of an RNRed element to be deleted, but just stipulated that deletion can occur in the first conjunct in case there is a phonologically identical part of strings in the antecedent clause. As far as I know, Hartmann (2000), and Féry and Hartmann (2005) are the only analyses that have proposed licensing conditions for RNR, and their view (henceforth, the Hartmann-Féry analysis) will be reviewed in this chapter. The Hartmann-Féry analysis is based on Rooth’s (1992b) licensing conditions for VP-ellipsis. I will first introduce Rooth (1992b) and describe what makes deletion and deaccenting different in terms of licensing conditions in section 4.1. There are also other competing accounts for licensing VP-ellipsis and deaccenting. In section 4.2, I will discuss one of them: Merchant’s (2001) e-GIVEN. I will compare these two different analyses in VP-ellipsis and discuss predictions that each account can make. At the end of this section, I will conclude that Merchant’s analysis is convincing, on the grounds that it successfully points out problems for Rooth’s analysis and captures (cross-linguistically) a wider range of empirical facts. In sections 4.3-4.4, we will turn to RNR licensing conditions and review the Hartmann-Féry analysis and I will suggest that Hartmann-Féry’s licensing conditions falls into the same kinds of problems as Rooth’s for VP-ellipsis. In the next chapter, I will propose licensing conditions for RNR, based on Merchant’s (2001) e-GIVEN.

120

4.1 Structural Isomorphism To understand licensing conditions of deaccenting and ellipsis, we need to introduce Rooth’s (1985, 1992a) alternative semantics for focus (cf. von Stechow 1989, Kratzer 1991). Rooth (1985) argues that focus introduces a set of alternatives which is built up compositionally. Let us consider (1a), for example, in which the subject is focused. A set of alternatives is generated by the focused subject, which in turn generates numerous propositions that are built up by replacing the subject with each of the subject’s alternatives in turn, as shown in (1b).

(1)

a. [JOSH]F came to the party. b. {Josh came to the party, Mike came to the party, Chris came to the party…}

With this brief background for alternative semantics for focus, let us consider what Rooth (1992b) claims about licensing conditions for deaccenting and ellipsis. Rooth assumes that ellipsis and deaccenting are the prosodic realizations of the semantically redundant material, and argues that an entailment relationship between the clauses can determine which phrase can be marked as redundant. Let us first consider what licenses deaccenting. He claims that deaccenting is licensed when the antecedent clause is a subset of alternatives of the subsequent clause. This can be rephrased in terms of entailment as in (2).

(2)

Rooth’s (1992b) deaccent licensing Deaccenting is licensed if the preceding clause entails propositional alternatives of the following clause.

121 Let us consider how the condition in (2) works with an example (3). Our interest is whether the VP in the second conjunct is licensed to be deaccented by its antecedent in the first conjunct.1

(3)

JOHN drove his new Corolla to the market, and SUE WENT TO THE MARKET.

Note that the subject of each conjunct is focused, and focus generates a set of alternatives. Therefore, the second conjunct generates propositional alternatives, as in (4).

(4)

{Sue went to the market, Mary went to the market, Jane went to the market, John went to the market, Chris went to the market, Bill went to the market…}

According to the condition in (2), the deaccenting of VP in the second conjunct would be licensed if the antecedent entails the propositional alternatives (4). Since JOHN drove his new Corolla to the market entails ∃x [x went to the market], it satisfies the condition in (2). Therefore, the VP in the second conjunct can be deaccented. Rooth observes that while the condition in (2) is sufficient to license deaccenting of the VP, it is not enough to license VP-ellipsis. Let us take (5a-b). The examples show that ellipsis is only licensed when the syntactic structure of the antecedent is identical with that of the elided clause, which is called syntactic isomorphism. In other words, in addition to the entailment relationship between the two conjuncts in (2), ellipsis requires another condition: syntactic isomorphism.

1

A deaccented constituent is represented with ITALICIZED SMALL CAPS in this dissertation.

122 (5)

a. JOHN drove his new Corolla to the market, and SUE did , too. b. *JOHN drove his new Corolla to the market, and SUE did , too.

The condition is stated in (6). While (5a) satisfies the conditions (6i-ii), (5b) satisfies only one of the conditions (6i). (5b) does not satisfy (6ii) because the VPs are not structurally identical.

(6)

Rooth’s (1992b) ellipsis licensing Ellipsis is licensed when i) The antecedent entails propositional alternatives of the ellipsis clause, and ii) The two clauses are structurally isomorphic.

4.2 Semantic identity A number of problems have been uncovered for proposals that appeal to a requirement of structural isomorphism in ellipsis. Merchant (2001) gives the following evidence that structural isomorphism is not required in ellipsis: i) even if the antecedent and the elided clause are not structurally equivalent, ellipsis is still allowed in (7a), ii) a correspondence between structures can be observed, so a gerund antecedent can license the deletion of an infinitive in (7b), and iii) it is hard to explain cases where morphological flexibility is allowed in the reconstruction of the elided clause. One type of example that shows such a flexibility are those showing Fiengo and May’s (1994) Vehicle Change effect, in which R-expressions and pronouns are considered equivalent under ellipsis in (7c).

123 (7)

a. Mary was singing, but I don’t know WHAT . b. Changing oil for your car is easy if you know how <*changing oil for your car/ to change oil for your car>. c. The police arrested Mikei, though hei thought they wouldn’t .

Notice that under Rooth’s analysis, ellipsis in the examples in (7) is predicted not to be licensed, contrary to fact. This in turn indicates that structural isomorphism does not need to hold in ellipsis resolution. Merchant (2001) basically follows Rooth’s (1992b) licensing condition for deaccenting; the antecedent clause must entail the deaccented clause, modulo ∃-type shifting, which will be discussed at length shortly. However, these two accounts differ with respect to what licenses ellipsis. While Rooth claims that ellipsis requires structural isomorphism, Merchant proposes what licenses ellipsis is a stronger entailment relationship. In addition to the antecedent entailing the ellipsis clause, the reverse also needs to be true; that is, the antecedent and the ellipsis clause must mutually entail each other. Crucially, Merchant’s analysis does not rely on syntactic isomorphism, but only on semantic licensing conditions.2 Let us examine Merchant’s semantic licensing condition for deaccenting and ellipsis in detail. Merchant’s condition is based on Schwarzschild’s (1999) GIVENness constraint (8). 2

In his subsequent work (2005, 2006), Merchant argues that both syntactic and semantic identity must be observed, in order to account for the following cases, due to Chung (2005). The grammatical (ia-b) and ungrammatical (ic) indicate that preposition stranding is possible in the ellipsis site only if the preposition has a correlate in the antecedent clause. This supports his contention that syntactic information also plays a crucial role for licensing ellipsis. (i) a. Bill is upset. Guess about what . b. Bill is upset about something. Guess what . c. Bill is upset. *Guess what . (examples from Merchant 2005) However, assuming the revised version of Merchant’s licensing conditions is not necessary for our discussion here, so I will adhere to the earlier version of Merchant’s view.

124 Schwarzschild’s GIVENness is a constraint on interpretation that requires that constituents that are not focus-marked to be interpreted as GIVEN. In reverse, constituents that are not GIVEN need to be focus-marked.

(8)

GIVEN An utterance U counts as GIVEN, iff U has a salient antecedent A, and modulo ∃-type shifting, A entails the F-closure of U. (Schwarzschild 1999: 148)

Let us consider how Schwarzschild’s GIVENness works with an example in (9). An utterance A provides an antecedent for B. These two utterances are minimally different with respect to the object DP. Since the utterance B has a salient antecedent A, we try to determine whether any constituent in an utterance B is GIVEN. Let us first consider whether TP is GIVEN. Suppose that pear is focus-marked and the focused constituent provides an open variable that needs to be existentially closed, which Schwarzschild calls F-closure. The F-closure of TP is ∃x [he also ate a x]. Since the utterance A entails the F-closure of TP, it satisfies the GIVENness constraint in (8). Thus, the TP in the utterance B counts as GIVEN. Now, let us consider whether VP is GIVEN. Modulo ∃-type shifting, F-closure of the VP would be ∃x∃y [x ate y], which is entailed by the antecedent. Thus, VP is GIVEN, too. Schwarzschild (1999) argues that constituents that are GIVEN cannot be focus-marked (AvoidF), so the consequence is that focus-mark is limited to the object NP.

(9)

A: John ate an apple B: He also ate a PEARF.

125 Merchant (2001) brings the GIVENness constraint into licensing ellipsis, proposing that only the given information can be elided or deaccented. First, Merchant argues that deaccenting is licensed only when Schwarzschild’s GIVENness constraint is satisfied. Consider (10) and whether the second conjunct is GIVEN. In (10), contrastive focus is assigned on the subject in each conjunct. By F-closure of the subject, the second conjunct yields ∃x [x went to the market]. Given that the antecedent clause entails F-closure of the second conjunct, it satisfies the condition in (8). This also indicates that the second conjunct counts as GIVEN. By Merchant’s hypothesis, deaccenting in this context is licensed because the GIVENness condition is satisfied, so what follows the focused subject in the second conjunct can be deaccented in (10).

(10) JOHN drove his new Corolla to the market, and SUE WENT TO THE MARKET. John drove his new Corolla to the market  ∃x [x went to the market].

Note that the GIVENness for licensing deaccenting here is basically the same with Rooth’s conditions where an antecedent needs to be a subset of focus value of the deaccented clause. What makes Merchant’s analysis differ from Rooth’s is the ellipsis licensing conditions. Merchant proposes that in addition to this unidirectional entailment relationship between the utterances, the elided clause also needs to entail the F-closure of the antecedent. In other words, the antecedent and the elided clause must mutually entail each other for ellipsis to be licensed. Merchant dubs it e(lliptical)-GIVEN, which is formalized in (11).

126 (11) e-GIVEN An expression E counts as e-given iff E has a salient antecedent A and, modulo ∃-type shifting, (i) A entails F-clo (E), and (ii) E entails F-clo (A). (Merchant 2001: 26)

Let us take (12). The VP-ellipsis in (12) is licensed since e-GIVEN is satisfied. The antecedent and the elided clause are shown in (12a-b), respectively. The subject in each conjunct is assumed to be contrastively focused. To evaluate whether the VP in the second conjunct counts as e-GIVEN, the focused part of each conjunct is replaced by a variable, which is subject to be bound by existential closure (12c-d). Given that A entails F-clo (E) and E entails F-clo (A), e-GIVEN is satisfied. Therefore, the VP-ellipsis is licensed in (12).3

(12) JOHN bought an apple for Mary, and BILL did too. a. A = JOHN bought an apple for Mary. b. E = BILL bought an apple for Mary. c. F-clo (A) = ∃x [x bought an apple for Mary]. d. F-clo (E) = ∃x [x bought an apple for Mary].

Merchant also addresses how sluicing is licensed with examples, such as (13). The wh-word in the second conjunct is focused and thus replaced by a variable for the purpose of computing eGIVENness. 3

Existential closure of this variable gives (13d). The indefinite in the antecedent clause

Merchant’s (2001) e-GIVEN is a semantic licensing condition. Merchant also proposes syntactic and phonological licensing conditions for ellipsis. I leave those aside since they are not relevant here. In chapter 5, I will lay out those aspects of licensing conditions when the complete analysis for RNR is proposed.

127 is existentially closed which yields (13c). e-GIVEN is satisfied on the grounds that the antecedent entails F-clo (E), and the elided clause entails F-clo (A). Therefore, sluicing is licensed in (13).

(13) I met someone in the street yesterday, but I don’t remember who . a. A = I met someone in the street yesterday. b. E = I met t in the street yesterday. c. F-clo (A) = ∃x [I met x in the street yesterday]. d. F-clo (E) = ∃x [I met x in the street yesterday].

Recall that Rooth’s structural isomorphism condition predicts that ellipsis would not be licensed in (7), repeated here in (14). Now let us consider whether Merchant’s e-GIVEN analysis can account for the non-isomorphic ellipsis in (14).

(14) a. Mary was singing, but I don’t know what . b. Changing oil for your car is easy if you know how <*changing oil for your car/ to change oil for your car>. c. The police arrested Mikei, though hei thought they wouldn’t .

First, let us examine the detailed computation of (14a) in (15). Since there are no focused constituents in the antecedent clause, the F-closure of the antecedent remains the same as the antecedent in (15c). 4 The wh-word is focused in the second conjunct, so (15d) results after

4

The computations about ellipsis are all done relative to a constituent that should be undergoing ellipsis. The computation for VP-ellipsis are all done relative to the VP, so the external argument is not part of the structure being evaluated and so is computed as an existentially closed variable. On the other hand, slucing

128 existential closure of the focused constituent. Because the antecedent entails F-clo (E) and the elided clause entails F-clo (A), e-GIVENness is satisfied and sluicing is licensed. Thus, the eGIVENness

conditions can account for the structural difference between the antecedent and the

ellipsis clause.

(15) Mary was singing, but I don’t know what a. A = Mary was singing. b. E = Mary was singing twhat. c. F-clo (A) = Mary was singing. d. F-clo (E) = ∃x [Mary was singing x].

A category mismatch in (14b) is also resolved under e-GIVEN. Let us take (16). I assume that an arbitrary PRO takes the subject position in both gerund and infinitive constructions (16a-b). PROarb is interpreted as an unbound variable, so modulo ∃-type shifting F-closure of A and E yields (16c). A entails F-clo (E) and E entails F-clo (A). Thus, despite the differences in syntactic category of the antecedent and elided clause, e-GIVEN is argued to be satisfied.

(16) Changing oil for your car is easy if you know how a. A = PROarb changing oil for your car. b. E = PROarb to change oil for your car. c. F-clo (A) = F-clo (E) = ∃x [x changes oil for your car].

cases are done relative to the TP, so the external argument, tense and modality are included. Therefore, (15c) is different from (12c) because in (12c) the subject is not within the constituent being equated for ellipsis (i.e. the VP), but in (15c) the subject is there.

129 Finally, Fiengo and May’s (1994) Vehicle Change effects can also be explained within Merchant’s e-GIVENness. Notice that in (14c), repeated here in (17), the R-expression in the elided clause would be bound by co-indexed pronoun in the matrix subject position (in violation of Principle C), if Mike were realized in the elided site. Merchant argues that Vehicle Change effects can be derived under e-GIVENness. As long as the index i gets the same interpretation, arrested Mikei and arrested himi will entail each other. Assuming that pronoun serves as a bound variable, him2 in the elided clause yields g(2). Therefore, it is possible to delete arrested himi under identity with arrested Mikei, because e-GIVEN is satisfied.

(17) The police arrested Mikei, though hei thought they wouldn’t a. A = The police arrested Mike. b. E = They arrested g(2). c. F-clo (A) = ∃x [x arrested Mike]. d. F-clo (E) = ∃x [x arrested g(2)].

(where g(2) = Mike)

One advantage of Merchant’s method of deriving Vehicle Change is that it explains why the constituents that seem to be able to change are what they are – they are the constituents that eGIVENness

predicts will be equivalent. Therefore, it is not an ad hoc theory of which things can

undergo Vehicle Change, as the one in Fiengo and May (1994) is. In addition to the empirical advantages, Merchant’s analysis is superior to Rooth’s on conceptual ground as well. Merchant’s semantic licensing account can explain the differences between ellipsis and deaccenting without relying on independent conditions. That is, deaccenting can be captured by weakening the same licensing conditions that govern ellipsis. While deaccenting is licensed when the antecedent entails F-clo (E), ellipsis requires a stronger semantic

130 licensing condition: the ellipsis must entail F-clo (A) as well. On the other hand, Rooth’s (1992b) account needs two independent sources of constraint; in addition to the entailment relationship between the conjuncts, structural isomorphism has to be called upon for ellipsis. In the next section, I will turn to the Hartmann-Féry analysis of RNR which is based on Rooth’s structural isomorphism (Hartmann 2000, Féry and Hartmann 2005), and show that their licensing conditions of RNR falls into the same kind of troubles as Rooth’s ellipsis analysis.

4.3 The Hartmann-Féry analysis In this section, I will review previous studies of RNR whose licensing conditions appeal to Rooth’s alternative semantics for focus. As far as I know, only Hartmann (2000) and her subsequent work, in collaboration with Féry (Féry and Hartmann 2005), have an explicit proposal about the licensing conditions for RNR.5 The Hartmann-Féry analysis proposes that a specific focus and prosodic structure licenses PF-deletion of RNR (and gapping), and that their unique pitch contour signals the missing material.6 The central proposals of their analysis are: i) The coordination determines the focus structure of the conjuncts, meaning that one conjunct contributes to the determination of the focus structure of the other conjunct. ii) There is a specific prosodic structure that makes the PF-deletion of the RNR target possible. The pitch contour shows “register scaling” between intonational phrases and there is an extraordinarily clear boundary tone in front of the RNR target.7 We will evaluate each proposal in this section.

5

Hereafter, Hartmann-Féry will refer to Hartmann (2000), and Féry and Hartmann (2005). Although they present both RNR and gapping data, here I will just review their RNR data. Given that their analysis applies to both constructions uniformly, RNR would be the only one relevant to our discussion. 7 Féry and Hartmann (2005) define “register scaling” as the relationship between tones inside of a single IP and across phrases, and claim that downstep, upstep, and reset would be the phrasings signaling ellipsis. 6

131

4.3.1 Focus structure and RNR Hartmann-Féry assume that any conjunct, either preceding or following a clause involving RNR, can serve as an antecedent in a parallel syntactic structure. Let us consider an example in (18). The focus on the preposition would make the whole PP F-marked, due to the focus assignment rules (Selkirk 1995: 555). 8 However, under the assumption that each conjunct provides background for the other conjunct, dem sofa ‘the sofa’ in the first conjunct can count as given, and similarly dem sofa in the second. Thus, narrow focus is assigned on the preposition in each conjunct.

(18) Q: Where are Halma and Mikado? A: Halma ist [PP AUFF dem sofa] und Mikado ist [PP UNTERF dem sofa] Halma is

ON the sofa

and Mikado is

UNDER the sofa. (Féry and Hartmann 2005: (13))

Hartmann-Féry claim that the DP the sofa in the first conjunct would be deaccented in the answer of (18) because the second conjunct would entail the focus value of the first conjunct ‘there is someone who is in some spatiotemporal relationship with the sofa.’ Furthermore, the conjuncts are structurally parallel, hence according to Rooth (1992b), the sofa in the first conjunct also meets the environment to be elided. At the PF side, they assume that PF-deletion occurs right after the focused preposition in the first conjunct in (19) due to a prosodic requirement. This prosodic requirement is how they attempt to capture the Right Edge Effect in RNR. I will return 8

F-assignment Rules (Selkirk 1995: 555) a. Basic Focus Rule An accented word is F-marked b. Focus Projection (i) F-marking of the head of a phrase licenses the F-marking of the phrase. (ii) F-marking of an internal argument of a head licenses the F-marking of the head.

132 to this issue in section 4.3.2. Here I will be only concerned with their syntactic-semantic licensing conditions.

(19) Halma ist [PP AUFF ] und Mikado ist [PP UNTERF dem sofa].

Hartmann-Féry propose the syntactic-semantic licensing conditions listed in (20). Notice that parallel syntactic and focus structure are both crucial for RNR to be licensed under their analysis. They claim that RNR is licensed only if all the conditions in (20i-iv) are satisfied.

(20) Hartmann-Féry’s syntactic-semantic licensing conditions for RNR i) The conjuncts must be structurally identical, ii) The pre-RNR elements in the first conjunct and the elements with which they contrast in the second must be focused, iii) The focused elements create sets of alternatives (Rooth 1992), and the sets of alternatives for both conjuncts must be identical. iv) The deletion occurs immediately after the contrastively focused pre-RNR element.

Let us consider how each condition is satisfied in detail with an example (21). In (21), the syntactic structure is parallel since each conjunct is a transitive structure, thus satisfying (20i). The subject and the verb in each conjunct are focused. Notice, in particular, that the pre-RNR element in each conjunct, SOLD and BOUGHT, is contrastively focused, satisfying (20ii). The focused constituents generate sets of alternatives, forming parallel focus structure. As shown in (21a-b), the focus value of TP1 and that of TP2 are identical (i.e. [[TP1]]f ↔ [[TP2]]f), which satisfies

133 the condition (20iii). The RNRed object DP the car is licensed to be elided at PF by virtue of being in the position just after the contrastive focus (20iv).

(21) [TP1[JOHN]F [SOLD]F ], and [TP2[MARY]F [BOUGHT]F the car]. a. [[TP1]]f = λp [∃z∈ALT(John′) & ∃Q∈ALT(sold′) & p=Q(ιx.car(x))(z)] b. [[TP2]]f = λp [∃z∈ALT(Mary′) & ∃Q∈ALT(bought′) & p=Q(ιx.car(x))(z)]

Identical syntactic structure is crucial to the Hartmann-Féry analysis, since it is responsible for ruling out (22), on the grounds that the first conjunct has a transitive, compared to the ditransitive in the second conjunct. The contrastive focus is assigned on the verb but not on the indirect object Mike in the second conjunct. According to the condition (20iv), the RNR target is determined to immediately follow the focus. Thus, the RNR target in the second conjunct would be Mike the book, which is different from the target determined by the first conjunct. Since the RNR target does not match, RNR is illicit in (22). There are cases where the conjuncts are not structurally identical, but RNR is still licit. I will discuss them in section 4.4.

(22) *Bill BROWSED THROUGH , but Peter SENT Mike – the book. (Hartmann 2000:120)

Consider the requirement of contrastive focus in condition (20iv). Hartmann-Féry provide examples which would be ruled out just because (20iv) is not satisfied. Let us consider (23-24). In the answer of (23), the pre-RNR element Maria’s is identical between the conjuncts, violating the requirement of contrast contained in condition (20iv). In (24), a prefix is attached on the pre-RNR

134 element in the second conjunct, differentiating it from the pre-RNR element in the first, but it is not enough to satisfy (20iv), since agreement morphemes are not truth-conditionally distinct.

(23) Q: Whose car did Peter and his brother sell? A: *Peter kaufte MARIASF und sein Bruder kaufte MARIASF auto. Peter bought Maria’s

car and his brother bought Maria’s

car

(24) Q: What did Jonas and Claus do with the letters? A: *Jonas SCHICKTE und Claus VERSCHICKTE einen brief. Jonas sent

a letter

and Claus PREFIX-sent

a letter (Féry and Hartmann 2005: (18))

Similarly, examples in (25) are ruled out since focus in the pre-RNR position is not contrastive. On the other hand, when different verbs are used, allowing focus to be contrastive, RNR is possible in (26).9

9

Phillips (1996) claims that an ECM complement and double object constructions cannot be RNRed. The judgements given below are as Phillips reports them. Regarding (i-ii), notice that the pre-RNR elements are not contrastively focused. In other words, RNR seems to be incorrectly targeted. If we change the example so that the RNR target includes to believe in (i) and prevent in (ii), enlarging the RNR target with the result that the pre-RNR elements can be contrastively focused, acceptability is predicted to improve. i) *I find it easy to believe, but Joan finds it hard to believe – Tom to be a dishonest person. ii) *I wanted to prevent, but I couldn’t prevent – Rob from meeting Sally. iii)*John offered, and Harry gave – Sally a Cadillac. (*Double object) iv) John offered, and Mary actually gave – a gold Cadillac to Billy Schwartz. (Dative) In fact, the prediction seems to be borne out in (v-vi). On the other hand, (iii) does not seem to be bad, and in fact parallel examples have been cited in previous literature as being acceptable (e.g. Sabbagh 2007). I suspect that those who have judged (iii) unacceptable might not have read with correct prosody; with contrastive focus on both offered and gave, (iii) seems to be grammatical. v) I find it EASY, but Joan finds it HARD – to believe Tom to be a dishonest person. vi) I wanted to, but I couldn’t – prevent Rob from meeting Sally.

135 (25) a. *John BELIEVES , but Frank does not BELIEVE – Tom to be a fool. b. *John GAVE
, but Bill didn’t GIVE – a gift to Mary.

(26) a. John BELIEVES , but Frank CONSIDERS Tom to be a fool. b. John DID , but Bill DIDN’T give a gift to Mary. (Hartmann 2000: 112-113)

4.3.2 Prosodic structure of RNR As briefly discussed in chapter 1, RNR shows unique prosodic properties. The contrasting constituent prior to the RNR target is focused and bears a nuclear accent, and the first conjunct always ends with a prominent rising tone with a relatively long pause. The RNR target itself usually does not bear pitch accent when it is a simple DP, but it can if it is large enough to form its own phonological phrase (PhP). Féry and Hartmann (2005) propose prosodic structures of RNR in a more specific way. They observe that the pitch contour of RNR conforms to the following patterns: each conjunct forms an intonational phrase (IP), and the second IP contains two PhPs. The first IP is upstepped, and the second IP is downstepped, as shown in figure 4. They argue that this upstep-downstep pitch track, together with extraordinarily high boundary tone at the end of the first IP, signals RNR.

First IP Second IP ` First PhP Second PhP Figure 4. The upstep-downstep pitch contour in RNR (Féry and Hartmann 2005: 102)

136 Let us consider the pitch contour of RNR with a German example in (27) and figure 5. In (27), the contrasting pre-RNR element in each conjunct is focused, licensing RNR of the object DP herden ‘shirts’ and non-finite verb trägt ‘wear’ in the first conjunct. Note that RNR displays a unique pitch contour in the first IP, in particular, its final position. It is higher than the initial accent word. Féry and Hartmann (2005) note that the speakers did not start the conjunct at their highest pitch level in the beginning. Rather, at the end of the first IP, the final high accent is “rescaled” to even higher than the level of the initial pitch accent. They claim that this upstep in the first IP-final position signals that the utterance is not complete, and some material has been elided. They also claim that prosodic phrasing between boundaries is extraordinarily clear in RNR, and that both the upstep-downstep pitch contour and the clear phrasing are the key prosodic features of RNR. Furthermore, these prosodic patterns have been hypothesized to be applicable to the other ellipsis cases, and they try to support their hypothesis in gapping.10 I will present their gapping data in section 4.3.

(27) What kind of things do the men wear who Claudia and Maria know? [IP[IP[PhP MARIA] [PhP kennt einin Mann] [PhP der GELBE]] [IP[PhP und Claudia] Maria

knows a

man

who yellow

and Claudia

[PhP kennt einen Mann] [PhP der ROSA Hemden trägt]]. knows a

man

who pink shirts

wear

‘Maria knows a man who wears yellow shirts, and Claudia knows a man who wears pink shirts.’ (Féry and Hartmann 2005: 101)

10

Féry and Hartmann (2005) assume that gapping is an ellipsis phenomenon (cf. Johnson 1996/2003).

137

Figure 5. Upstep in a RNR sentence (Féry and Hartmann 2005: 104)

Féry and Hartmann claim that this pitch contour and phrasing are by-products of semantic focus structure as well as prosodic requirement; namely, nuclear accents need to be aligned at the edgemost position in prosodic phrase. This is also a way to derive the Right Edge Restriction in RNR. Narrow focus needs to be assigned on the pre-RNR constituents, gelbe ‘yellow’ and rosa ‘pink’, which receive nuclear accents. Those nuclear-accent constituents are not in the edge position, so RNR is licensed to allow the edge alignment condition to be satisfied. In other words, by deleting the RNR target in the first conjunct, the nuclear accent can be in the right edge position of the first IP. Therefore, the pitch contour of the first IP can be understood by the combination of narrow focus on the pre-RNR constituent and prosodic alignment condition. The narrow focus makes the pre-RNR prominent, and the prosodic requirement makes it aligned to the right edge, so the pitch contour of the first IP is derived as in figure 4. On the other hand, it is unclear how the Hartmann-Féry analysis can derive the pitch contour of the second IP in figure 4, and in fact problematic for their analysis, a point which will be elaborated in section 4.4. Here I will briefly demonstrate their analysis and point out problems. They assume that there are two

138 PhPs in the second IP and the PhPs are divided in-between the pre-RNR constituent and the RNR target. A narrow focus is assigned on the contrastive pre-RNR constituent, and this is the place where a nuclear accent is also assigned. Consequently, a nuclear accent is not in the rightmost edge position of the second IP, and what follows would be licensed to be deleted. The problem is that the RNR target is not deletable in the second conjunct, as in (28). In section 4.4, I will entertain a possible alternative to get around this problem maintaining the Hartmann-Féry analysis. And I will argue that the alternative cannot be right, either.

(28) *[IP[IP[PhP MARIA] [PhP kennt einin Mann] [PhP der GELBE Hemden trägt]] [IP[PhP und Maria

knows a

man

who yellow shirts

wear

and

Claudia] [PhP kennt einen Mann] [PhP der ROSA ]]. Claudia

knows a

man

who pink

To sum up, the Hartmann-Féry analysis provided an interesting proposal for licensing RNR. The licensing conditions imposed syntactic and semantic requirements on the one hand, and a prosodic requirement on the other. The conjuncts must be structurally parallel, which guarantees that the focus value of the conjuncts mutually entail each other. And a nuclear accent on the preRNR constituent in each conjunct is a by-product of focus structure. The unique prosodic contour of RNR can be derived by prosodic requirement, which rules the nuclear accent to be aligned in the right edge position of each IP. In section 4.4, I will demonstrate some RNR examples that cause problems for the HartmannFéry analysis. One type of challenges arises due to their assumption that syntactic isomorphism must be satisfied in RNR. I will show cases where RNR is still acceptable even when the conjuncts are not structurally but semantically parallel. Another type of problems arises when the

139 RNR target determined by focus and prosodic structure shows discrepancies between the conjuncts. I will show that the target determined by the first conjunct and the one by the second can be different. Finally, I will present that the edge alignment condition seems to hold only in backwards ellipsis cases and not applicable to any forward ellipsis.

4.4 Problems for the Hartman-Féry analysis In this section, we will discuss cases where the Hartmann-Féry analysis would face problems. We have observed that structural mismatches between the conjuncts make RNR impossible. Let us consider (22), repeated here in (29). The explanation given by the Hartmann-Féry analysis is that the predicate of the first conjunct is transitive and that of the second is ditransitive. As a result, RNR is predicted to be unavailable.

(29) *Bill BROWSED THROUGH , but Peter SENT Mike – the book.

However, the structural isomorphism requirement seems to undergenerate. Let us compare (30a) with (30b). (30a) is structurally the same as (29). The first conjunct is a transitive and the second is a double object construction. The RNR target is the direct object. (30a) confirms that RNR does not tolerate this kind of mismatches. However, if we use a dative construction in the second conjunct instead of the double object construction as in (30b), RNR is acceptable. Notice that the Hartmann-Féry analysis predicts (30b) to be ungrammatical for the same reason that (30a) is; that is, the two conjuncts are equally structurally different.

140 (30) a. ?*Bill and Mary FILLED IN, and their assistant SENT the IRS – the tax document (by express mail). b. Bill and Mary TALKED TO, and their assistant BOUGHT candies FOR – the orphans.

One remark worth making here is that the focus pattern differs in (30b). While the immediately preceding constituent prior to the RNR target – which is the IRS – cannot be focused in the second conjunct in (30a), the focus is assigned on the immediately preceding constituent – which is for – in (30b). This will be one of the keys to solving the puzzle later in the next chapter. Let us consider another example in (31). Here the conjuncts are not syntactically identical, either. The auxiliary had in the first conjunct and the copular isn’t in the second are contrastively focused, and by hypothesis the constituents that immediately follow the focus would be determined to be the RNR target. Thus, the RNRed target in the first conjunct would be the embedded VP driven this car, and the one in the antecedent clause would be the DP a car he has driven. 11 Consequently, the RNR targets are not syntactically identical, violating HartmannFéry’s licensing conditions for RNR. Contra their prediction, however, (31) is grammatical.

(31) Bill might wish he HAD , but this ISN’T – a car he has driven. (Johnson 1996: 6)

Another type of challenge for the Hartmann-Féry analysis is that their semantic conditions incorrectly rule out some grammatical RNR sentences. (32a) is similar to Hartmann’s unacceptable (25). Here the identical verb has no reason to be focused, which in turn means the conjuncts cannot generate a set of alternatives. PF-deletion of the RNR target relies on the focus 11

It is interesting to note that the RNRed embedded VP in the first conjunct must be understood driven this car, not driven a car, which is puzzling, since the antecedent does not include the demonstrative.

141 semantic structure. If the verb in each conjunct cannot be contrastively focused, the nuclear accent cannot be aligned on it. Therefore, the deletion of the RNR target should not be licensed at all in (32a). However, if we assign narrow focus on DIDN’T in the second conjunct (32b), instead of on meet, the sentence becomes acceptable. The problem is that (32b) is still predicted to be unacceptable under the Hartmann-Féry analysis, since by hypothesis the RNR target is determined to immediately follow the contrastive focus (20iv). If MET is focused, the RNR target should be , while its antecedent would include larger constituents . This is due to the fact that DIDN’T is focused in the second conjunct. Therefore, the condition (20iv) needs to be revised.

(32) a. *CATHY MET, but MARY didn’t MEET her husband at the train station. b. CATHY MET , but MARY DIDN’T [meet her husband at the train station].

Apart from the structural mismatch between the antecedent and the elided part, the alternative sets in (32) do not appear to match either. Assuming that focus is assigned on the subject and the verb in the first conjunct, a set of alternatives would be (33a), which would look like ∃x∃R [x Red her husband at the train station]. This formula is basically derived by existential closure of the focused subject and verb. Meanwhile, narrow focus is assingend on the subject and the auxiliaries in the second conjunct. The set of alternatives for the second conjunct would be (33b) by existential closure of the focused constituents: ∃x [x {did, did not} meet her husband at the train station]. For now I will assume that the focus alternatives of auxiliaries include anything that occupies T, which also includes modals with [±polarity]. The sets of alternatives in (33a-b) do not entail each other, which violates Hartmann’s condition (20iii).

142 (33) a. {Cathy met her husband at the station, Mary met her husband at the station, Cathy called her husband at the station, Mary called her husband at the station, Cathy found her husband at the station, Mary found her husband at the station…} = ∃x∃ ∃R [x R-ed her husband at the train station] b. {Mary did meet her husband at the station, Cathy did meet her husband at the station, Mary didn’t meet her husband at the train station, Cathy didn’t meet her husband at the train station…} = ∃x [x {did, did not} meet her husband at the train station].

There is a way to get around the problem for (32). It has been argued in syntax literature that a phrase is needed upon which polarity focus can be realized, which could be dubbed VerumP (Gleitman 1969, Laka 1990, Höhle 1992, Romero and Han 2004, among others). Let us assume that there is a VerumP between TP and vP. Then, we can argue that the locus of contrastive focus assignment is not the verb, but VerumP in the first conjunct of (32b), which is illustrated in (34).

(34) Cathy [VerumP] , but Mary DIDN’T[Verum focus] [meet her husband at the train station].

Assuming that verbs in English do not move out of the vP, no phonological material fills VerumP in the first conjunct. However, the focus feature in the VerumP still needs to phonetically realized, so the verb picks it up, as shown in (35). The crucial assumption here is that the actual RNR target in (34) is the whole vP, including the focused verb, so that no mismatches result. Assuming that a focused constituent cannot be deleted by ellipsis (MaxElide, Fiengo and May 1994, Kennedy 2002, Takahashi and Fox 2005), PF-deletion of the vP spares the focused verb in (35).

143 (35) Cathy [VerumP] , but Mary DIDN’T[Verum focus] [meet her husband at the train station].

I will return to this issue in chapter 5, and provide more a detailed analysis, after the ellipsis feature of RNR is introduced. Finally, Hartmann-Féry’s premise that the register rescaling and clear prosodic boundary tones signal ellipsis encounters overgeneration problems; their premise is too weak to generalize as a prosodic requirement for other deletion cases, due to the fact that the upstep-downstep pitch pattern does not always guarantee that deletion has occurred. In support of their proposal, Hartmann-Féry claim that gapping is another case in which the upstep and phrasing signal ellipsis. Notice that gapping occurs in the medial position of the second conjunct in (36). The verb lesen ‘read’ in the second conjunct is missing.

(36) What kind of things are the children and the parents reading? [IP[IP[PhP Die KINDER] [PhP lesen einen dicken ROMAN]] [IP[PhP und die ELTERN] The children

read

a

thick

novel

and the parents

[PhP eine langweilige BALADE] a boring

ballade

‘The children are reading a thick novel and the parents a boring ballade.’ (Féry and Hartmann 2005: 108)

They argue that the location of the gap leads to different expectations as to the overall prosodic structure. In a broad sense, the upstep-downstep pitch contour is observed between the first and the second IP as in RNR. However, in gapping, the clear prosodic phrasing is formed

144 right after the first PhP of the second conjunct, signaling a missing verb follows. Figure 6 demonstrates these empirical facts.

Figure 6. Upstep and phrasing in gapping (Féry and Hartmann 2005: 109)

Let us consider another gapping example (37) and its pitch contour in figure 7. As seen in figure 7, the second IP appears to be reset. Clearly, it is not downstepped. The pitch contour of gapping in (37) appears to be an exception to the generalization that upstep-downstep pitch contour is a reliable indicator of ellipsis.

145 (37) [IP[IP[PhP Anna] [PhP bastelt ihrer ENKELIN] [PhP ein SEGELBOOT]] Anna

makes her granddaughter

a sailing boat

[IP[PhP und MARIA] [PhP ihrem SOHN] [PhP ein MÜLLauto]]] and Maria

her

son

a garbage truck

‘Anna is making her granddaughter a sailing boat and Maria her son a garbage truck.’

Figure 7. Gapping without downstep (Féry and Hartmann 2005: 108)

There is a more critical conceptual problem. Let us reconsider (27-28). The sentence before deletion is repeated in (38). Notice that a nuclear accent in the second IP is assigned on ROSA ‘pink’, which is not at the right edge position. Recall that the reason of deleting the RNR target in the first IP is to align the nuclear accent to the right edge position of the IP. The question here arises whether the same thing can be done in the second IP deleting the RNR target. As we observed in (28), this is not possible.

146 (38) [IP[IP[PhP MARIA] [PhP kennt einin Mann] [PhP der GELBE Hemden trägt]] [IP[PhP und Maria

knows a

man

who yellow shirts

wear

and

Claudia] [PhP kennt einen Mann] [PhP der ROSA Hemden trägt]]. Claudia

knows a

man

who pink shirts

wear

‘Maria knows a man who wears yellow shirts, and Claudia knows a man who wears pink shirts.’

Hartmann (2000) provides a short speculative remark on this puzzle, but it is not clear what it means, or how the prediction is generated. There is no known consequence of deleting lexical accents.

Due to the implication of the rising tone associated with the prenuclear focus in the first conjunct, the target may be totally reduced there. In the second conjunct, post-focal deletion leads to the disappearance of even the lexical item – this is a prediction which will be checked empirically with acoustic experiments. I leave this issue open for the future research. (Hartmann 2000: 108)

There might be an alternative way to capture this particular puzzle within Hartmann-Féry’s framework but this will not be an ultimate solution for it because it also faces problems on its own. Let us suppose that RNR consists of three intonational phrases, instead of two (Swingle 1993). The first intonational phrase converges to the first conjunct, and the second one is the second conjunct, excluding the RNR target. And the RNR target forms its own intonational phrase, which is similar to Swingle’s (1993) analysis. Swingle claims that the RNR target needs to stand alone as an intonational phrase. If the target is prosodically too weak, then RNR is ruled out as in (39). In (39a-b), the RNR target is just composed of a pronoun and cannot form an

147 intonation phrase by itself. Thus, RNR is disallowed. On the other hand, the object DP in (39c) can stand alone as an IP, so RNR is allowed.

(39) a. *[IP Alice composed] [IP and John performed] [IP it]. b. *[IP Mary bought] [IP and Fred stole] [IP that]. c. [IP John bought] [IP and Fred stole] [IP the car]. (Swingle 1993: 94)

Let us suppose that the RNR target exists in the first IP, too. Concretely, the three intonation phrases of (39c) would look like (40a), and the RNR target in the first IP undergoes PF-deletion to obey the edge alignment constraint. Notice that after the deletion, the nuclear accent in each IP is aligned at the right edge as in (40b).

(40) a. [IP John BOUGHT the car] [IP and Fred STOLE] [IP the CAR]. b. [IP John BOUGHT < >] [IP and Fred STOLE] [IP the CAR].

However, there are a couple of problems with this alternative. One is, as Hartmann (2000) has already observed, that the RNR target is not required to bear a nuclear accent. Let us consider the question-answer pair in (41). Given that the RNR target counts as given, the pitch contour of the target would be flat, which in turn implies that this RNR target cannot form its own IP. Rather it would be incorporated into the previous one. Therefore, the answer in (41) would be predicted to be as bad as (39a-b), contrary to fact, because it fails to form three IPs.

148 (41) Q: What do you and Mary think play the third base in Red Sox? A: [IP I think that JOHN ], [IP and Mary thinks that BILL] [IP plays the third base for Red Sox].

The second problem is that there is nothing that prevents the RNR target from forming an IP in each conjunct – for example – in (42). If this is possible, then we lose the motivation for deleting the RNR target, because here the nuclear stress already occupies the right edge position without deletion. Even if an answer for that were found, it would still be unclear why the one in the first conjunct needs to be deleted, rather than the one in the second conjunct.12

(42) [IP John BOUGHT] [IP ] [IP and Fred STOLE] [IP the CAR].

The Hartmann-Féry analysis appears to be correct that the focus structure determines the prosodic structure, and all the RNR cases seem to uniformly fit in their phonological phrasings. 12

In his dissertation, An (2007) proposed something similar in this line. He assumes that the RNR target in each conjunct needs to form its own intonation phrase, as shown in (i). And IP2 would be licensed to be deleted. (i) [IP1 A

B] [IP2 C

D

E] [IP3 X

Y] [IP4 C

D

E].

An (2007) attempts to provide an account of why IP2 must be deleted on the basis of ordering preservation. Let us assume that (ii) is a linear order of an RNR sentence A>B>and>X>Y>C>D>E where CDE is an RNR target. The linear order of each conjunct is (iia-b), and the whole sentence prior to deletion is (iic). He claims that the deletion of the RNR target is only acceptable if the previously established linear orders are not disturbed. While (iid) is acceptable, (iie) is not because (iib) indicates that C>D>E follows X>Y. Thus, the only place that the linear order is consistently observed is (iid). (ii) a. 1st conjunct: A>B>C>D>E b. 2nd conjunct: X>Y>C>D>E c. The underlying sentence: A>B>C>D>E>and>X>Y>C>D>E d. RNR: A>B>C>D>E>and>X>Y>C>D>E e. *RNR: A>B>C>D>E>and>X>Y>C>D>E However, this account has a fatal empirical problem since (iic) can be overtly pronounced without any deletion. Note that (iic) also violates linear order because here too the RNR target in the first conjunct precedes X>Y.

149 That is, we can clearly see high boundary tone at the end of the first IP and the downstep in the second IP. However, it does not seem to be possible to generalize this to other ellipsis cases, in particular any forward ellipsis cases. Here, I take this to indicate that prosodic structures should be considered as only by-products of the semantic focusing, and that the prosodic requirement of the edge alignment cannot be a licensing condition for deletion. To summarize this section, Hartmann-Féry’s syntactic and semantic licensing conditions for RNR seem to encounter several problems. The conditions are not strong enough to predict that RNR is grammatical between structurally different but semantically identical conjuncts, such as in (31), and they are too strong to capture acceptable sentences such as (32b). In addition, the Hartmann-Féry analysis attempts to generalize RNR and ellipsis with specific prosodic structure, but yet the prosodic patterns sometimes do not follow from their generalization. In chapter 5, I will revise Hartmann’s syntactic and semantic licensing conditions, to avoid those problems and capture further empirical evidence, and I will propose more concrete and comprehensive licensing conditions for RNR that cover the syntactic, semantic, and phonological aspects of RNR.

150

CHAPTER 5 THE ERNR ANALYSIS

In chapter 4, we have reviewed Hartmann-Féry’s proposal for the RNR licensing conditions and I have argued that their analysis faced both empirical and conceptual problems. Many of the problems are due to their appeal to Rooth’s (1992b) ellipsis licensing conditions in which both structural and semantic identities need to be fulfilled between the conjuncts. An alternative proposal will be explored and defended in this chapter, which is based solely on semantic identity. I will address the syntactic, semantic, and phonological aspects of RNR licensing conditions by introducing an E(llipsis) feature for RNR, a variant of the E features introduced by Merchant (2001).

5.1 Ellipsis features 5.1.1 Merchant (2001) Merchant (2001) proposes that ellipsis is licensed by an elliptical feature [E]. For example, when the E feature for sluicing (henceforth, ES) enters the derivation, it forces TP to be deleted. In order to have an ES feature, certain requirements must be met: syntactic, semantic and phonological requirements. Let us first consider the syntactic requirements on the English sentence in (1). Merchant (2001) argues that ES is [uwh*, uQ*], a feature of T. ES is an uninterpretable feature that needs to be checked during the syntactic derivation. When the C head bears the [+wh, +Q] feature, it checks ES via T-to-C movement, as illustrated in the tree structure (1b).

151 (1)

a. Matthew met a Red Sox player near Fenway, but he doesn’t remember WHICH ONE. b.

&P wo TP1 TP2 6 3 M. met a RS player he T’ near Fenway 3 doesn’t vP 3 remember CP 3 WHICH ONE C’ 3 C[+wh,+Q][Es] TP 3 T[Es] vP

The ES feature, now attached to C, is interpreted at PF as an instruction not to pronounce its complement. Therefore, the TP of the second conjunct is forced to be unpronounced in (2).

(2)

… and [TP he doesn’t remember [CP WHICH ONE C[Es] <[TP Matthew met t near Fenway]>

As the semantic requirement for the inclusion of ES, e-GIVENness must be met, which is defined in chapter 4: roughly, an expression E is e-GIVEN iff there is an antecedent A which entails E and which is entailed by E, modulo ∃-type-shifting. Let us examine how the eGIVENness

of the example (1) is satisfied. The antecedent is the first conjunct, Matthew met a Red

Sox player near Fenway, and the corresponding elided clause is the embedded TP of the second conjunct, [TP Matthew met t near Fenway]. By existential-closure of the indefinite, the first conjunct yields ∃x [Matthew met x near Fenway] (∃-clo (A)). Given that the fronted wh-phrase leaves a wh-variable, it would also be existentially closed. The second conjunct also yields ∃x [Matthew met x near Fenway] (∃-clo (E)). Since A entails ∃-clo (E) and E entails ∃-clo (A), e-

152

GIVEN

is satisfied in (3). Since the syntactic, phonological and semantic licensing conditions are

satisfied, sluicing in (1) is licensed.

(3)

e-GIVENness of (1) a. A = Matthew met a Red Sox player near Fenway. b. E = He met tWH near Fenway. c. ∃-clo (E) = ∃x [he met x near Fenway]. d. ∃-clo (A) = ∃x [he met x near Fenway].

5.1.2 The ERNR feature Adopting Merchant’s (2001) idea of introducing an E feature during the derivation, I propose that a variant of the E feature is introduced for deletion of the RNR target. I will call this variant ERNR, and the remainder of this section will be concerned with the various licensing conditions that the ERNR feature imposes in syntax, semantics and phonology. Given the empirical generalization that the pre-RNR constituent is always focused, I will first hypothesize that a focused constituent that occupies pre-RNR position enters the syntactic derivation bearing ERNR, and the feature instructs PF to leave the RNRed element unpronounced. And the derivation converges at LF if the RNR target is e-GIVEN. Our first definition is summarized in (4).

153 (4)

Ellipsis feature of Right Node Raising (the first pass) a. A focused constituent in the pre-RNR position of the first conjunct can bear the ERNR feature in syntax. b. ERNR instructs the PF-interface not to pronounce its sister. c. Merchant’s (2001) e-GIVENness conditions must be satisfied by the conjuncts.1

Let us see how the definition in (4) works with examples in (5-6). In (5), the pre-RNR element is focused and the ERNR feature enters the derivation with the one in the first conjunct chose.2 The E feature forces its sister to be unpronounced at PF, i.e. a couple of books. Now let us consider whether the first conjunct is e-GIVEN. The focused constituents in the first conjunct turn to variables at LF and are existentially closed by F-closure, so F-closure of the first conjunct would yield ∃x∃R [x R-ed a couple of books]. The antecedent in the second conjunct entails F-clo (RNR). The F-closure of the antecedent yields the identical focus value ∃x∃R[x R-ed a couple of books], and the first conjunct entails it. Given that the antecedent entails F-clo (RNR), and the RNR clause entails F-clo (A), e-GIVEN is satisfied. Since all the three conditions in (4) are fulfilled, RNR in (5) is licit.

1

An interesting question arises what prevents a sentence like (i). If it is supposed that the wh-word generates a set of alternatives that is directly comparable to those generated by the focused constituent (see also Beck 2006), the F-closure of the first and the second conjunct both would be ∃x∃R[x R-ed the book]. Then, the antecedent entails F-clo (RNR) and the RNR clause entails F-clo (A), satisfying e-GIVENness. Therefore, it would seem that (i) is predicted to be grammatical, contrary to fact. (i) *JOHN SOLD and WHO BOUGHT the book? I will speculate that the solution is provided by the different clause-types in the conjuncts. Assuming that [clause-type] is an uninterpretable feature on the C head, it needs to be valued by merging to TP (Cheng 1991). Since the second conjunct is interrogative, C would be valued as [clause-type: interrogative]. There would be two scenarios at this point. First, the first conjunct, merged to the specifier position of &P, bear a mismatching clause type, which is declarative, so that the derivation would crash. Alternatively, we can assume that the type mismatches do not matter, but the wh-word in-situ fails to check its wh-feature in C. 2 I assume that the adverbial phrase before going to bed is adjoined above the conjuncts.

154 (5)

THE CHILDREN CHOSE
, and THEIR MOTHER READ – a couple of books before going to bed. a. RNR = THE CHILDREN CHOSE a couple of books. b. Antecedent = THEIR MOTHER READ a couple of books c. F-clo (RNR) = F-clo (A) = ∃x∃R [x R-ed a couple of books].

Let us turn to (6). The pre-RNR element is focused, so the ERNR feature enters the derivation with the verb liked. The feature would make the deletion of the RNR target possible at PF if (6) satisfies e-GIVEN. The focused constituents turn into variables subject to existential closure. The F-closure of RNR yields ∃x∃R [x R-ed her advisor], and the F-closure of A yields ∃x∃R [x R-ed her]. The RNR clause does not entail F-clo (A), and the antecedent does not entail F-clo (RNR). Thus, the e-GIVENness is not satisfied. Even if the RNR target her advisor in the first conjunct finds its antecedent in the second conjunct, the deletion of the RNR target cannot be licensed since one of the conditions in (4) is not fulfilled.

(6)

*JEANNETTE LIKED but HER ADVISOR HATED her. a. RNR = JEANNETTE LIKED her advisor. b. A = HER ADVISOR HATED her. c. F-clo (RNR) = ∃x∃R [x R-ed her advisor]. d. F-clo (A) = ∃x∃R [x R-ed her].

The conditions in (4) are empirically borne out in the examples above. Let us turn to examples in (7-8). Compared with (7) where the pre-RNR element in each conjunct bears

155 contrastive focus, this is not the case in (8). There may be some type of focus on the pre-RNR element in (8), but the nature of the focus is not contrastive.

(7)

a. John DID , but Mary DIDN’T enjoy the movie b. John LIKES , but Mary HATES running in cold weather. c. John DOESN’T , but Bill LIKES going to the Red Sox game with his sister.

(8)

a. *John DID , and Mary DID enjoy the movie, too. b. *John LIKES , and Mary LIKES running in cold weather, too. c. *John DOES , and Bill LIKES going to the Red Sox game with his sister, too.

The contrast between (7) and (8) indicates that a certain type of contrast between the two conjuncts is required to license RNR. It is not just any focused constituent that licenses RNR, but contrastive focus can bear the ERNR feature and license RNR. While RNR is licensed in the examples in (7) because the pre-RNR element is contrastively focused, RNR in (8) is not since the pre-RNR element bears no contrastive focus. The system overgenerates so far. Suppose that the lexical item with contrastive focus in the pre-RNR position bears the ERNR feature. This implies that the pre-RNR element in each conjunct may have the feature in (9); that is, both MADE and ATE bear the ERNR feature. Then, our analysis predicts three possibilities of deletion of the RNR target: the RNR target is deleted in the first conjunct (9a), or in the second conjunct (9b), or in both conjuncts (9c). However, only (9a) is a possible RNR construction, and (9b-c) are not. This indicates that we need to further constrain the

156 position of the feature, so that only the feature in the first conjunct has an effect. To eliminate the possibility (9c), only one ERNR feature should be allowed to have an effect in a sentence, and to block the possibility (9b), the feature in the second conjunct should not have an effect.

(9)

JOHN MADE[ERNR] the spaghetti, and BILL ATE[ERNR] the spaghetti. a. JOHN MADE[ERNR] , and BILL ATE the spaghetti. b. *JOHN MADE the spaghetti, and BILL ATE[ERNR] . c. *JOHN MADE[ERNR] , and BILL ATE[ERNR] .

It is not clear how the ERNR feature survives only in the first conjunct, but here is one possibility. Suppose that there is a probe outside the conjuncts, which must be valued by an ERNR feature in the coordinate structure. The closest ERNR feature, which is the one in the first conjunct, would be able to value the probe, and only this feature can license RNR.3 The other ERNR feature in the second conjunct would be void, so no RNR takes place in the second conjunct. 4 We suppose that the ERNR feature is interpretable since this is basically contrastive focus, so it can remain without crashing the derivation. Let us consider this hypothesis in more detail. Suppose that the ERNR feature must enter the contrastively focused pre-RNR constituent in both conjuncts, and that there is a head X outside the coordination that can bear a feature F and this feature probes for the ERNR feature.5 Each ERNR feature has a value; the one in the first conjunct has the value [α], and the one in the second conjunct has [β]. The ERNR feature in the first conjunct is the goal closer to the probe than the one in the second conjunct, so it values the F feature in X. Thus,

3

I assume that the probe cannot be valued multiple times. This would rule out the possibility of (9c). See also (11c). 4 I use “void” in a descriptive term. It simply means that it has no effect at PF; that is, the sister of the voided ERNR feature cannot be deleted at PF. 5 The X head could be either C or FocusP. I also assume that each sentence can have an X.

157 F is now valued as [α], which is shown in (10b). Once it is valued, the agreement relationship is established between the probe and goal, so the F feature agrees with the ERNR feature in the first conjunct. Any other ERNR features that fail to agree with F would be void in derivation, as marked in double strikethrough.

(10) a.

XP wo X F[ ]

b.

&P wo TP &’ 6 eo … ERNR VAL[α]… & TP 6 … ERNR VAL[β]…

XP wo X F[α]

&P wo TP &’ 6 eo … ERNR VAL[α]… & TP 6 … ERNR VAL[β]… AGREE

Now it becomes clearer why only (9a) is grammatical, but not (9b-c). (11a-c) are the illustrations of (9a-c) in which agreement relationship is shown between the F feature and the valued ERNR features in the conjuncts. In (11a), the F feature probes the closest ERNR feature and is valued. The ERNR feature in the first conjunct enters agreement with F and has the RNR target to be deleted at PF. On the other hand, (11b-c) show illicit derivations. (11b) is ungrammatical because F probes the E feature in the second conjunct, rather than the closer one in the first in (11b). (11c) is ruled out since F is valued twice.

158

(11) a. [XP F[α] [&P JOHN MADE[ERNR][α] , and BILL ATE[ERNR][β] the spaghetti.]] b. *[XP F[β] [&P JOHN MADE[ERNR][α] the spaghetti, and BILL ATE[ERNR][β] .]] c. *[XP F[α][β] [&P JOHN MADE[ERNR][α] , and BILL ATE[ERNR][β] .]]

There is another possibility where the RNR target is pronounced in both conjuncts in (12). I suppose the the F feature in X is optional. The ERNR feature is attached to the contrastively focused verbs, but in this case no agreement can be established due to the absence of the F feature in X. Thus, both of the ERNR features is void and RNR does not take place.

(12) John MADE the spaghetti, and Bill ATE the spaghetti. [XP ø [&P JOHN MADE[ERNR][α] the spaghetti, and BILL ATE[ERNR][β] the spaghetti.]] No Agreement

To summarize, the ERNR feature must enter the derivation with a contrastively focused preRNR constituent in both conjuncts. The F feature in X probes the closest ERNR feature to be valued, so only the ERNR feature in the first conjunct can agree with F. Once valued, the ERNR feature has an effect at PF and LF interfaces. The ERNR feature instructs PF not to pronounce the sister of the feature, and LF converges if e-GIVENness is satisfied.6 I suggest that this is how RNR occurs only in the first conjunct. Syntactic, phonological and semantic requirements of the ERNR feature are summarized in (13).

6

Here I assume that the RNR target forms a constituent, contrary to the common contemporary assumption. I will return to this issue in section 5.4.

159 (13) The ERNR feature a. Syntax of ERNR: The ERNR feature enters the derivation with the contrastively focused preRNR constituent in the first conjunct. &P wo : VP 3 Y[ERNR]

: VP 6 … Z …QP

b. Phonology of ERNR QP  Ø/ ERNR ____. c. Semantics of ERNR: e-GIVEN must be observed in RNR. i) RNR  F-clo (A) ii) A  F-clo (RNR).

5.2 Consequences of ERNR In chapters 3 and 4, I challenged previous non-movement analyses on the grounds that their licensing conditions, strictly based on PF-identity, do not capture empirical facts. Those facts include sloppy identity, Vehicle Change, and morphological mismatches between antecedent and RNR clauses. In this section, I will lay out consequences of the ERNR analysis for RNR, which is based on semantic identity. First, the ellipsis properties found in RNR will be reanalyzed in terms of ERNR, such as sloppy identity, Vehicle Change effects, and morphological mismatches. Second, we will discuss different predictions of licensing RNR between double object constructions and dative constructions. Third, we will look at cases where the antecedent clause and the RNR clause are structurally mismatched. Finally, in section 5.3, we will see that the ERNR analysis is able to predict cases where the RNR target is larger than what is actually left unpronounced.

160

5.2.1 RNR and Deaccenting Under Merchant’s (2001) analysis, the difference between ellipsis and deaccenting is explained by a different strength of the requirement imposed by the e-GIVENness condition. While ellipsis must satisfy mutual entailment relationship between the conjuncts, deaccenting is licensed when a sentence satisfies weaker licensing conditions in (14).

(14) Deaccenting A entails F-clo (E), but E does not need to entail F-clo (A).

Let us first consider how the strength of e-GIVENness can license VP-ellipsis and deaccenting in (15). The antecedent is KIM enjoyed a Korean novel. In all examples (15a-c), the antecedent entails F-clo (E).7 By hypothesis, VP-ellipsis is licensed only when both conjuncts mutually entail each other. VP-ellipsis is licensed in (15b), but not in (15a), on the grounds that the elided clause cannot entail F-clo (A). In the case of (15a), only deaccenting is licensed, as shown in (15c).

(15) VP-ellipsis vs. deaccenting a. *KIM enjoyed a Korean novel, and even BILL did[E] .

(*VPE)

b. KIM enjoyed a Korean novel, and even BILL did[E] . (√VPE) c. KIM enjoyed a Korean novel, and even BILL READ A BOOK.

(√deaccented VP)

If the ERNR feature is a variant of E, we expect that the same conditions would hold for the difference between RNR and deaccenting. I argue below that this is the case. Let us examine (16). The ERNR feature enters the derivation with the verb, and the RNR target is two lobsters (in a 7

Given that the subject in each conjunct is contrastively focused, F-closure of the antecedent and the elided clause yields ∃x [x enjoyed a Korean novel].

161 seafood restaurant). Both the subject and the verb are focused, so the focus value of each conjunct yields ∃x∃R [x R-ed two lobsters in a seafood restaurant], as shown in (16c). (16a) entails (16c), and so does (16b). Therefore, e-GIVENness is satisfied, and the deletion of the RNR target is licensed.

(16) MARY BOUGHT[ERNR] , and JOHN DEVOUERED – two lobsters in a seafood restaurant. a. Antecedent = Mary bought two lobsters in a seafood restaurant. b. RNR = John devoured two lobsters in a seafood restaurant. c. F-clo (A) = F-clo (RNR) = ∃x∃R [x R-ed two lobsters in a seafood restaurant].

Let us consider deaccenting cases in (17a-b). The target some shellfish in a seafood restaurant is okay to be deaccented in (17a), but it cannot be completely silent, as in (17b). It is crucial to notice that the object in the second conjunct is more specific than that in the first conjunct. In other words, eating two lobsters entails eating shellfish, but not vice versa. This indicates that the premise in (14) holds for RNR. As shown in (18a-d), the antecedent entails Fclo (RNR), but the first conjunct does not entail F-clo (A), which satisfies the licensing condition for deaccenting, but not for RNR. Therefore, we can conclude that deletion and deaccenting of the RNR target can be captured under the ERNR analysis.

(17) a. MARY BOUGHT SOME SHELLFISH IN A SEAFOOD RESTAURANT, and JOHN DEVOURED two lobsters in a seafood restaurant. b.*MARY BOUGHT[ERNR] , and JOHN DEVOURED – two lobsters in a seafood restaurant.

162 (18) Computations for (17a-b) a. Antecedent = John devoured two lobsters in a seafood restaurant. b. RNR = Mary bought some shellfish in a seafood restaurant. c. F-clo (A) = ∃x [x devoured two lobsters in a seafood restaurant]. d. F-clo (RNR) = ∃x [x bought some shellfish in a seafood restaurant].

5.2.2 Ellipsis properties 5.2.2.1 Sloppy Identity The semantic licensing condition of ERNR, i.e. e-GIVENness, requires that the antecedent and the RNR clause must mutually entail each other, after F-closure. The question we are interested in is whether the first conjunct is e-GIVEN in (19) by the antecedent in the second conjunct, even if the pronominal in the RNR target has a different index from the one in the antecedent clause. The direct answer is yes. Let us take a closer look at why e-GIVENness does not care about an indexical difference for pronominals. First, the F-closure of the antecedent and the RNR clause is identical. The nuclear contrastive focus is assigned on the pre-RNR lexical items, and the subject in each conjunct is also contrastively focused. These narrowly focused lexical items are turned into focus variables and existentially closed at LF. The result is ∃x∃R [x R-ed his father]. To account for how sloppy identity arises, we need to assume that the pronoun in the RNR target may serve as a bound variable.8 I assume that the antecedent would be Bill λx [x hates x’s father]

8

Sloppy identity has been argued to be insensitive to phi-features in VP ellipsis, but it is not the case for RNR. Let us compare VP ellipsis (i) with RNR (ii): (i) Johni think hei’s going to win the race, and Maryj does , too. (ii) *Johni believed , but Maryj almost knew – that shej would win the race. If RNR is ellipsis, it is reasonable to ask why only RNR is sensitive to phi-features in sloppy identity. I do not have a principled reason why they differ in this matter, but it seems that the requirement of contrastive focus in RNR constructions is relevant here. Notice that phi-feature mismatches in the sloppy reading are as

163 and the RNR clause would be John λy [y likes ]. The result of F-closure of the antecedent and the RNR clause would still be identical, ∃x∃R [x R-ed x’s father]. Given that the antecedent entails F-clo (RNR) in (19a), and the RNR clause entails F-clo (A) in (19b), eGIVENness

is satisfied. Therefore, we predict that sloppy identity in RNR is available.9

(19) JOHNi LIKES[ERNR] , but BILLj HATES – hisj father. a. Bill λx [x hates x’ father] entails ∃y∃R [y R-ed y’s father]. b. John λy [y likes y’s father] entails ∃x∃R [ x R-ed x’s father].

5.2.2.2 Vehicle Change Vehicle Change effects (Fiengo and May 1994) can be derived in a similar manner. Let us take (20), for example. The ERNR feature enters the derivation with the contrastively focused verb SUBMITTED in the first conjunct and licenses the deletion of the object DP. The next step is to see if the first conjunct is e-GIVEN. The crucial issue here is whether the RNR target containing a pronominal form co-indexed with John can be counted as e-GIVEN. In other words, we want to know whether the e-GIVENness hypothesis can capture that the RNR target fire himi at the end of this year would count as e-GIVEN by the antecedent fire Johni at the end of this year. I argue that this is the case. The RNR target in the elided portion is e-GIVEN only if the pronoun refers to John; that is, [[ John8 ]]g = [[ him8 ]]g must be established for any g, so the pronoun in the RNR clause yields g(8). Note that the embedded VP is the largest constituent that is eligible to be eGIVEN

because the embedded TP contains different modals which in turn cannot be GIVEN. Thus,

hard in VP ellipsis in (iii) as they are in RNR (ii) when the two conjuncts stand in contrast with one another. (iii)??Johni thinks hei’s going to win the race, but Maryj doesn’t . 9

The strict reading is predicted to arise in RNR when the pronoun is interpreted as referential.

164 the antecedent is the embedded VP in the second conjunct, which is Shej fire Johni (at the end of this year), and the RNR clause is Susan fire himi (at the end of this year). Modulo ∃-type shifting, the F-closure of (RNR) is ∃x [x fire himi at the end of this year], and so is the F-closure of (A). The antecedent entails F-clo (RNR) as in (20a), and the RNR clause entails F-clo (A) as in (20b). Thus, e-GIVENness is satisfied. Therefore, the R-expression in the antecedent can license the deletion of the co-indexed pronoun in RNR.

(20) JOHNi hopes that Susanj WON’T , but THE SECRETARY knows that shej WILL – fire Johni at the end of this year. a. [VP Shej fire Johni (at the end of this year)] entails ∃x [x fire g(8) at the end of this year]. b. [VP Susan fire g(8) (at the end of this year)] entails ∃x [x fire John at the end of this year]. (g(8) = John)

Let us consider another type of Vehicle Change effects in (21), introduced in chapter 3: some/any alternations. Assuming that some and any are indefinites, they would be subject to be existentially closed at LF. By F-closure, the focused constituents would turn into variables. As a result, both the F-closure of (RNR) and that of (A) yield the same formula:∃x∃y∃R [x R-ed y of my books]. The antecedent VP entails F-clo (RNR) and the RNR clause entails F-clo (A), thus eGIVENness

of the VP constituents in the first conjunct is fulfilled. The ERNR feature enters the

derivation in the contrastively focused verb READ and licenses the deletion of its sister, object DP.

165 (21) John READ , but he hasN’T UNDERSTOOD – any of my books. a. John hasn’t understood any of my books entails ∃x∃y∃R [x R-ed y of my books]. b. John read some of my books entails ∃x∃y∃R [x R-ed y of my books].

Therefore, appealing to semantic identity, we can explain how a string of the RNR target in the RNR clause different from the one in the antecedent clause still can be licensed for deletion.

5.2.2.3 Morphological mismatches In chapter 3, we discussed the fact that morphological identity of lexical verbs does not have to be observed in forward VP-ellipsis. Thus, the root form of the verb meet in the second conjunct can be deleted by its antecedent with its inflected form as in (22a). It is also interesting to notice that the copula be cannot tolerate such a morphological mismatch and must be overtly realized. This is shown in (22b-c).

(22) a. Bill met Prof. Barnes yesterday, and I will this afternoon. b. Jane was here, and I will be , too. c. *Jane was here, and I will , too.

The ERNR analysis predicts that RNR would show the same distribution with those forward ellipsis cases in terms of morphological identity. That is, we predict that the mismatched lexical verbs would be able to be elided, but the mismatched copula be must be overtly realized. Although the prediction seems to be borne out empirically, it is not immediately clear how to conceptualize it under the current analysis. Let us first consider (23a). Although the verb in the antecedent is a form of participle, the root form of the verb can be RNRed in the first conjunct.

166 Now let us compare (23b) with (23c). The contrastive focus is assigned on the modal in the first conjunct, so the ERNR feature is assigned on the modal. The RNR target, which is determined to be the sister of the feature, would be be picked for the team as in (23b). However, (23b) shows that the deletion of the full RNR target results in ungrammaticality, but rather the copula must be pronounced as in (23c). By this empirical evidence, we can conclude that RNR does what VPellipsis does.

(23) a. John WON’T[ERNR] , but Mary has ALREADY negotiated her salary with the company. b. *John WON’T[ERNR] , but MARY already HAS been picked for the team. c. John WON’T[ERNR] be , but MARY already HAS been picked for the team.

Let us first see how this fact has been explained in forward VP-ellipsis. Lasnik (1999) proposes that lexical verbs and the auxiliaries in English enter the numeration in different ways. He argues that lexical verbs enter the derivation as a root form, and affixes are realized on verbs only at PF. Let us take (24), for example. How could the root form of the verb in the second conjunct be elided based on the inflected form of the verb in the first conjunct in (24a)? During the derivation, the root form of the verb enters the derivation as in (24b), so the elided vP finds an identical vP in the antecedent. The form stopped that appears on the surface is a result of morphological interpretation on the way to the PF interface, but is syntactically still separated into a tense morpheme in T and a lexical verb in V. Thus, under Lasnik’s approach, there is no morphological mismatch between the antecedent and the elided material.

167 (24) a. John stopped by my office, and Mary will , too. b. John –ED stop by my office, and Mary will , too.

On the other hand, Lasnik (1999) argues that the copula be is numerated as fully inflected. Let us consider (25). Compared with the lexical verbs, the copula enters the derivation with fully inflected. As a result, the copulas cannot be deleted as in (25a) since they are not morphologically identical. VP-ellipsis can only target the adjectival predicate in (25b).

(25) John was asleep, and Mary will be asleep, too. a. *John was asleep, and Mary will , too. b. John was asleep, and Mary will be , too.

This approach predicts that if the copula is realized with an identical morpheme, the deletion of the copula is possible. The prediction is borne out in (26).

(26) John should be asleep, but Mary shouldn’t .

However, it is not clear how Merchant’s e-GIVENness can account for why a tense morpheme in the lexical verbs does not matter and why the morpheme in the copula does. In fact, Merchant’s analysis pretends that this kind of problems does not exist, ignoring any tense difference between the conjuncts (Merchant 2001). It is also puzzling that there is a contrast between (25b) and (26). To account for the contrast, the necessary assumption would be that the location of the E feature for VP-ellipsis must be different in these cases. For (25b), the E feature must be in v, so that the copula is not deleted. However, for (26), the E feature must be in T, so

168 that the deletion includes the copula. For a loose end, I will just assume that the E feature is possible to include in either T or v, as shown in (27), and that there is some unsolved mystery about the undeletability of the copula when the morphology is different from its antecedent. Obviously, further research is necessary here.

(27) a. John was asleep, and Mary will be v[E] , too. b. John should be asleep, but Mary shouldn’t T[E] .

(= 25b) (= 26)

Lasnik’s analysis appears to be applicable to RNR. We assume that lexical verbs enter the derivation without morphemes, and the copula does with morphemes. Therefore, the morphology of the lexical verb matches in RNR in (28a), on the grounds that the root form of the verb in the second conjunct matches the one in the RNR clause. On the other hand, the copula needs to be realized in the second conjunct if it is not identical to the first conjunct. The fully inflected copula in the RNRed target would be undeletable in (28b). I will assume that the ERNR feature in the modal migrates to the copula as in (28c), and the deletion does not include the copula. Just as VPellipsis, when the morphology of the copula is the same, it is deletable in (28d).

(28) a. John WON’T[ERNR] , but Mary already HAS -ED [negotiate her salary with the company]. b. *John WON’T[ERNR] , but MARY already HAS been picked for the team. c. John WON’T be[ERNR] , but MARY already HAS been picked for the team. d. John WON’T , but Mary WILL be promoted.

169

5.2.3 Double object-Dative asymmetries Double object constructions differ from dative constructions with respect to the availability of RNR. Let us consider the contrast in (29a-b). It appears that in a dative structure (29a), RNR can occur even when it does not target the rightmost edge position in the first conjunct. On the other hand, such a possibility is not available in double object construction (29b). The possibility of RNR in (29a) appears to be problematic in any non-movement analyses, in particular, if the accounts assume a right edge effect. The RNR target does not appear at the right edge position, so it is a violation of the right edge restriction.

(29) a. John’s mom gave to MARY, and his dad gave to BILL – an interesting book. b. *John gave a BOOK, and Bill gave Mary a PRESENT.

Under the ERNR analysis, the ERNR feature is assumed to enter with the PP to Mary. However, the direct object an interesting book is not within its c-command domain, so a question also arises how the deletion of the RNR target is possible in (29a). To account for (29a), I assume that dative constructions are derived from the theme DP, originally base-generated below the goal PP, undergoing movement to its surface position (Dryer 1987, Aoun and Li 1993, Pesetsky 1995, Takano 1998, Sauerland 2000, and Sauerland and Elbourne 2002). With this assumption, it follows that the underlying structure of the first conjunct in (29a) is (30a), and the linear order is derived by movement of the object DP an interesting book, passing the PP, as shown in (30b).

(30) a. John’s mom gave to Mary an interesting book. b. John’s mom gave an interesting book to Mary t.

170 I propose that the ERNR feature, attached to to MARY, licenses the ellipsis of the object DP an interesting book in the base position of the first conjunct. This is shown in (31).

(31) John’s mom gave to MARY[ERNR] , and his dad gave to BILL – an interesting book.

Normally, the direct object undergoes movement to be linearized between the verb and the PP as in (30b), but this does not happen in RNR. I suggest that this is due to economy. If grammaticality can be achieved without movement, then the computational system prefers not to move.

(32) *John’s mom gave to MARY, and his dad gave an interesting book to BILL t.

On the other hand, the double object construction is base-generated exactly as shown in the surface structure of (29b). One type of evidence comes from NPI licensing. Barss and Lasnik (1986), Larson (1988, 1990), Aoun and Li (1989, 1993), and Marantz (1993) argue that the NPI licensing possibilities indicate that the indirect object asymmetrically c-commands the direct object position in (33). In particular, the ungrammaticality of (33b) tells us that anyone is not below the direct object at any time of the derivation.

(33) a. John gave nobody any valuable present. b. *John gave anyone nothing.

171 With this background, let us return to (29b). The direct object in each conjunct is contrastively focused, hence eligible to bear the ERNR feature. Since the direct object cannot move, there is no way for the ERNR feature to license the elision of the indirect object in the first conjunct. Thus, (29b) is ruled out. The distinction between (29a-b) indicates that the double object construction is structurally less flexible than the dative construction. While internal arguments can move in the latter, this is not the case for the former.10 Another interesting asymmetry between the two constructions arose in the previous chapter. Hartmann’s (2000) analysis predicted that there would be no differences between them because they both count as ditransitive and generate a different set of alternatives from a transitive predicate. However, the prediction was not borne out. It appears that the direct object of the dative object construction can be RNRed by the antecedent of the transitive construction, but it does not hold between double object and transitive constructions, as shown in (34a-b).

(34) a. ?*Bill and Mary FILLED IN, and their assistant SENT the IRS – the tax document (by express mail). b. Bill and Mary TALKED TO, and their assistant BOUGHT candies FOR – the orphans.

The ungrammaticality of (34a) is predicted by the current analysis, since the first conjunct is a transitive but the second conjunct is a ditransitive construction. The semantic entailment relationship would not be established between them, so that e-GIVENness would not be satisfied. Consequently, the deletion of the RNR target would not be licensed. However, it is interesting to note that (34b) is acceptable, even if the second conjunct of that sentence is also a ditransitive.

10

The issue centered around wh-movement of internal arguments in those constructions will be discussed in chapter 7.

172 The e-GIVENness conditions would not be satisfied, which in turn means that RNR should not be licensed. To account for the puzzle around (34a-b), the current analysis needs to show that the verb and the direct object are a semantic unit in the second conjunct and the unit can generate the same set of alternatives with a verb in the first conjunct. The analysis also needs to show that the verb and the indirect object are not a semantic unit. I argue that the answer can be found in how double object and dative constructions are structurally formed. In both cases, the direct object forms a constituent with the verb before the verb moves, and they are a semantic unit even after the verb moves. Let us consider the structures for (34a-b), presented in (35-36). In (35), the RNR target is the direct object. Since the verb does not form a constituent with the indirect object at any level, no semantic unit may be established between the verb and the indirect object. Therefore, two internal arguments exist in the second conjunct which would make it impossible to establish mutual entailment relationship with the first conjunct.

(35) Double object construction (= 34a) vP 3 v’ 3 v VP sent 3 DP1 V’ 63 the IRS V DP2 6 the tax document

On the other hand, the sister of the PP is the RNR target in (36). We assume that the direct object merges as a sister of V, and the PP merges to a specifier position of VP1 (Dryer 1987,

173 Aoun and Li 1993, Pesetsky 1995, Takano 1998, Sauerland 2000, and Sauerland and Elbourne 2002). The direct object moves to the specifier position of VP2 and the verb undergoes head movement to v, as shown in (36). Given our assumption that the verb and the direct object behave as a semantic unit, bought candies for can count as a single predicate.

(36) Dative construction (= 34b) vP 3 v’ 3 v VP2 bought 3 DP V’ 6 3 candies V VP1 3 PP V’ 6 3 for the orphans V DP

If this is true, it follows that the two conjuncts mutually entail each other. Let us see how in (37). The crucial issue here is whether bought candies for in the antecedent clause (37b) can generate the same set of alternatives as talk to in the RNR clause (37a). I suggest that this is the case as in (37c). Thus, RNR entails F-clo (A), and A entails F-col (RNR), which makes e-GIVENness satisfied.

174 (37) Dative a. RNR = Bill and Mary [talked to] the orphans. b. Antecedent = Their assistant [bought candies for] the orphans. c. F-clo (RNR) = F-clo (A) = ∃x∃R [x R-ed the orphans].

On the other hand, double object constructions do not satisfy e-GIVENness with a transitive predicate. Let us consider (38). Assuming that a contrastive focus is assigned on the pre-RNR material, both the verb and the indirect object turns to a focus variable, so F-closure of the antecedent clause would be (38d). The RNR clause cannot entail (38d), and neither does the antecedent clause entail F-clo (RNR) in (38c). Thus, e-GIVENness is not satisfied.

(38) Double object a. RNR = Bill and Mary [FILLED IN] the tax document. b. Antecedent = Their assistant [SENT] [THE IRS] the tax document. c. F-clo (RNR) = ∃x∃R [x R-ed the tax document]. d. F-clo (A) = ∃x∃y∃R [x R-ed y the tax document].

5.2.4 Korean Dummy Plural Markers In chapter 2, I introduced Chung’s (2004) argument that the Korean dummy plural marker (DPM) –tul poses challenges for the deletion account. The issue was that the dummy plural marker needs a plural subject, but the dummy plural marker in each conjunct does not have one. Thus, it is puzzling that the deletion of the RNR target improves grammaticality in (39), given the assumption that there is a copy of the dummy plural marker deleted in the first conjunct.

175 (39) Sue-nun yagwu-lul[ERNR] , kuliko Jane-un nonggwu-ul S.-TOP baseball-ACC well-DPM do-DEC, CONJ J.-TOP

cal-tul han-ta

basketball-ACC well-DPM do-DEC.

‘Sue plays baseball well, and Jane plays basketball well.’

In this section, I will demonstrate that the ERNR analysis can account for RNR with the dummy plural marker. As we can imagine from the name of the “dummy” plural marker, I will assume that it makes no semantic contribution. 11 Consequently, it has no influence on entailment relationships. The elision of the RNR target that I defend in this dissertation is licensed under conditions of mutual entailment, nothing directly arising from the structure or phonology. Therefore, the structure of (39) could be (40), with the DPM only in the second conjunct rather than in both conjuncts.

11

Whether the DPM makes a semantic contribution has been a matter of some debate. Choe (1988), and Chung (2004), among others, proposed a syntactic analysis of the DPM, claiming that the DPM can freely appear wherever the syntactic conditions are met, and that it has no effect on the semantics. Yim (2003), Kim (2004), and Joh (2005) each claim that the DPM contributes distributivity over plural subjects to the semantics, and is thus not meaningless. Here, I assume that the DPM does not contribute to the semantic interpretation, for the following reasons: First, the distributivity of the DPM is not as strong as the semanticists (Yim 2003, Kim 2004, and Joh 2005) have claimed. Multiple occurrences of DPM do not seem to influence the meaning of the sentence compared with non-occurrence of the DPM. If the DPM contributes to meaning differences, each occurrence would be expected to do so. Second, there is no difference in meaning accompanying the difference in position in (ia-b). (i) a. John-kwa Mary-ka satang-(ul)-tul ppali mekki-tul hass-ta. John-Conj Mary-N candy-(A)-DPM fast eat-DPM do-Dec b. John-kwa Mary-ka satang-(ul)-tul ppali-tul mek-ess-ta. John-Conj Mary-N candy-(A)-DPM fast-DPM eat-Past-Dec ‘John and Mary ate candy fast’

176

(40)

[SUE-nun YAGU-lul[ERNR] ], kuliko [JANE-un NONGGU-lul cal-tul han-ta.] S.-TOP baseball-ACC well do-DEC,

CONJ

J.-TOP basketball-ACC well-DPM do-DEC

a. Antecedent = Jane plays basketball well-DPM. b. RNR = Sue plays baseball well-∅. c. F-clo (A) = ∃x∃y [y plays x well-DPM]. d. F-clo (RNR) = ∃x∃y [y plays x well-∅].

It should be clear now why RNR is possible in (40) since the antecedent and RNR clause mutually entail each other (modulo F-closure). The antecedent entails (40d) and the RNR clause entails (40c), so e-GIVENness is satisfied. There is one remaining question, although this is not specific to the current analysis; that is, it is not immediately clear how the dummy plural marker in the second conjunct can see the subject in the first conjunct and fulfill its plurality requirement. I leave this question for future research.12

5.2.5 Structural Mismatches Under the ERNR analysis, e-GIVENness is the only necessary licensing condition for RNR. How this analysis differs from the other non-movement analyses is that structural identity does not need to hold. The crucial example we introduced in the last chapter was that RNR is acceptable if eGIVENness

is satisfied but the conjuncts are not structurally isomorphic. In this section, I will

show that the current analysis can account for such a case. Let us consider (41). The pre-RNR

12

Suppose that TPs are coordinated in (40), and C is merged to the coordinate structure. One way to solve the problem is to propose that the plurality requirement of the dummy plural marker needs to be satisfied within CP. This implies that a combination of the subjects in the conjuncts would satisfy the requirement as long as it is within CP. Thus, the singular subject in the second conjunct does not violate the requirement of the dummy plural marker because the derivation has not reached CP yet. It can look for its clause-mate subject in the first conjunct. The combination of the clause-mate subjects inside the TP conjuncts satisfies the plurality requirement of the dummy plural marker.

177 contrastive focus is assigned on the auxiliary in each conjunct, so has is assumed to bear the ERNR feature. The RNR target is the sister constituent of the feature, which is driven this car. It is crucial to see if the target is e-GIVEN. The possible salient antecedent can be found in the second conjunct, which is This ISN’T a car he has driven. The antecedent is logically equivalent to ‘It is not the case that he has driven this car.’ That is, if this isn’t a car he has driven then he has not driven this car, and if he has not driven this car, then this isn’t a car he has driven.

(41) Bill might wish [he HAD[ERNR] ], but [this ISN’T a car he has driven]. a. RNR = He had driven this car b. Antecedent = He has driven this car c. F-clo (RNR) = F-clo (A) = ∃x [x has driven this car].

The RNR and antecedent clauses are (41a-b), respectively. Modulo ∃-type shifting, each clause yields the same predicate logic, as in (41c). 13 The RNR clause entails F-clo (A), and the antecedent entails F-clo (RNR). Therefore, e-GIVENness is satisfied and the deletion of the RNR target is licensed.

5.3 ERNR in Verum focus In chapter 4, we posed a challenge for the Hartmann-Féry’s analysis in that the analysis cannot account for the grammaticality of the RNR sentence in (42). Under their analysis, the category of the pre-RNR element is different from each other – i.e. the verb in the first and the auxiliary in the second conjunct – so a different set of alternatives would be generated. The focus value of the first conjunct would be (42a), and that of the second conjunct would be (42b). (42a-b) do not 13

As Merchant (2001) does, I ignore tense difference here.

178 entail each other, so RNR in (42) would be prohibited. Nevertheless, RNR seems to be acceptable in (42).

(42) CATHY MET , but MARY could NOT [meet her husband at the train station]. a. ∃x∃R [x R-ed her husband at the train station]. b. ∃x [x {could, could not} meet her husband at the train station].

In this section, I will argue that the ERNR analysis can explain the grammaticality of (42) and that it also makes further predictions about where the pre-RNR focus can be assigned in the conjuncts. To account for (42), we need to determine where exactly the pre-RNR focus is assigned in each conjunct. I assume that there is a VerumP between TP and vP where polarity stands, so the negation can be placed here, as shown in (43).

(43)

TP 3 T’ 3 T VerumP = PolarityP 3 Verum ModalP Aff/Neg[F] 3 Modal vP

In case of the second conjunct of (42), the modal could first merges to the Modal head, and then moves to T. Thus, the conjunct is derived as shown in (44).

179

(44) [TP MARY could [VerumP NOT [ModalP t [vP meet her husband at the train station]]]].

On the other hand, the first conjunct in (42) has no modal, so nothing overtly occupies the Verum head. I assume that there still exists an affirmative focus which is in contrast with the negation. The conjunct is derived, as in (45).

(45) [TP CATHY [VerumP [affirmative]F [vP meet her husband at the train station]]].

The contrastive focus is assigned on the Verum head in the first conjunct, so Verum would be able to bear the ERNR feature, as shown in (46). The ERNR feature in the first conjunct would delete its sister at PF, which is the VP meet her husband at the train station.

(46) [&P CATHY [VerumP [affirmative]F[ERNR] ], and MARY could NOT meet her husband at the train station].

There are two more things to consider; i) we need to make sure that the whole first conjunct is eGIVEN,

and ii) we also need to decide if, and if so, where the affirmative focus with the ERNR

feature would be phonetically realized. The TP of the first conjunct is not e-GIVEN due to the existence of a modal in the second conjunct. The VP, however, is e-GIVEN. The antecedent VP is (47a), and the VP in the RNR clause is (47b). Modulo ∃-type shifting, F-clo (VPA) and F-clo (VPRNR) yield the identical formula, as in (47c). (47a) entails (47c), and so does (47b). Therefore, the VP is e-GIVEN.

180 (47) a. VPA = CATHY meet her husband at the train station. b. VPRNR = MARY meet her husband at the train station. c. F-clo (VPA) = F-clo (VPRNR) = ∃x [x meet x’s husband at the train station].

What if the modal exists in the first conjunct, too, as in (48)? Of course, the VP is e-GIVEN, as we have seen in (47), but would the whole conjunct be e-GIVEN? I argue that the answer is yes. I assume that the Verum itself is replaced by a variable when focused in the evaluation of F-closure, as in (48c). Here, the first conjunct entails F-clo (TPA) and the second conjunct entails F-clo (TPRNR), hence satisfying the e-GIVENness conditions.

(48) CATHY could [affirmative]F[ERNR] , but MARY could NOT [meet her husband at the train station]. a. TPA = CATHY could NOT meet her husband at the train station. b. TPRNR = MARY could meet her husband at the train station. c. F-clo (TPA) = F-clo (TPRNR) = ∃x∃V [x meet could [VerumP V ] x’s husband at the train station].

Let us turn to our second question regarding when the affirmative focus in the Verum head is overtly realized. When it comes to the negation in the Verum head, the focus is obligatorily assigned on the negation. However, there is no particular overt constituent that the affirmative focus must be assigned on. Notice that the affirmative focus does not have to be phonetically realized. For example, supposing that an affirmative focus exists at default in (49a), the focus could but need not be realized in (49b).

181 (49) a. John [VerumP [affirmative]F [vP saw a giant whale in his recent trip]]. b. John saw/SAW a giant whale in his recent trip.

However, I propose that the affirmative focus must be overtly pronounced if the Verum head bears the ERNR feature, and that the focus percolates down to the closest c-commanding head. Let us consider how it works for the first conjunct of (46), which is shown in (50a). The closest head within the c-commanding domain of the Verum head is v, thus at PF the verb must be pronounced with narrow focus, as shown in (50b).

(50) a.

TP 3 CATHY T’ 3 T VerumP ei Verum[ERNR] Aff[F] 3 v VP met 3 [MET] DP V’ her husband3 V PP at the train station

b. CATHY [VerumP [ERNR][vP MET her husband (at the train station)]]

During the derivation, the ERNR feature in the Verum head licenses the deletion of its complement, vP. And the VP is forced to be unpronounced at PF. I argue, however, that the whole VP cannot be deleted because focused constituents are not eligible to be deleted. Focused, the verb must be

182 pronounced.14 And only the object DP is deletable, as shown in (51a) (cf. MaxElide, Takahashi and Fox 2005).15 The structure in (51b) demonstrates that the RNR target determined by the ERNR feature during the derivation could be different from what is actually deleted at PF.

(51) a. CATHY [VerumP [ERNR] <[vP MET her husband (at the train station)]>]. b.

(at PF)

&P qp TP TP 3 6 Cathy VerumP 3 Verum[ERNR] The RNR target 3 tsubj v’ 3 v VP The deletable target MET 3 DP V’ her husband3 V PP at the train station

Meanwhile, (48) is a case where the RNR target coincides with the deletable target at PF. The closest head to the ERNR feature within its c-commanding domain is the modal in (48). Thus, the verum-focus is assigned on the modal. When it moves to T, this will look like (52a). The ERNR feature in the Verum head instructs PF not to pronounce its complement, ModalP, and all the word strings within the ModalP are deleted at PF, so (52b) results. 14

One remaining question is the direction of the deletion at PF – i.e. whether PF-deletion of the word strings begins with her husband and proceeds until the right edge of the RNR target (left-to-right deletion), or whether it begins with station up to the left edge of the RNR target (right-to-left deletion). In the next section, I will address this issue in detail and argue for the latter. 15 Takahashi and Fox (2005) argue that the target of ellipsis must be the largest within some Parallelism Domain (PD). An elided clause is considered a PD when it is semantically identical to an antecedent clause. It is important to notice that the target of ellipsis within PD may differ from the actual deletable target due to independent factors. Takahashi and Fox (2005) discuss this is in the context of re-binding phenomena of ellipsis.

183 (52) a. CATHY COULD [affirmative]F[ERNR] [ModalP t ] b.

&P qp TP TP 3 6 Cathy T’ 3 T VerumP COULD 3 Verum[ERNR] The RNR target = The deletable target 6 meet her husband at the train station

Now let us consider (53). Compared with the previous cases, the negative verum-focus is assigned on the first conjunct and the affirmative verum-focus is assigned on the antecedent conjunct. It is interesting to note that the affirmative verum-focus does not have to be overtly realized. The verb in the second conjunct can be either focused or not focused.

(53) John DIDN’T , but BILL shared/SHARED pizza with his sister.

Note that there is no ERNR feature in the second conjunct. Given our assumption that the affirmative verum-focus is required to be realized at PF only when Verum bears the ERNR feature, the focus does not have to percolate down to its closest head in the second conjunct in (53). That might explain why the affirmative verum-focus can be unrealized in the second conjunct. I suggest that it can still percolate down to its closest head, and the head picks up the focus. Therefore, shared could also have focused pronunciation. The negative verum-focus realization is not affected by the presence of the ERNR feature because negation always picks up the negative verum-focus. This is shown in (54). The verumfocus in the second conjunct must be realized even when the ERNR feature is not present.

184 (54) BILL [affirmative]F[ERNR] , but JOHN did [VerumP NOT [share pizza with his sister]].

To sum up, in all the RNR cases above, the Verum head hosts the ERNR feature. Under focus, Verum generates a set of alternatives – affirmative and negative – and the affirmative focus percolates down to its closest head for focus pronunciation when Verum hosts the ERNR feature. When e-GIVENness is satisfied, the ERNR feature instructs PF not to pronounce the RNR target, which is the VP here. However, in cases like (42) where a modal is not present, the verb picks up the focus and PF-deletion needs to stop at the focused verb.

5.4 Non-constituency in RNR revisited We observed in the first chapter that deletion in RNR does not seem to be constrained by constituent structure, and this is puzzling given the current assumption that ellipsis targets only constituents. The cases that clearly show the deletion of a non-constituent RNR target are found in SOV (55a) or V2 languages (55b). Here part of the matrix clause and part of the embedded clause are elided through RNR. In the Korean example (55a), the verb of the embedded clause and vP of the matrix clause are elided, and in the German example (55b), the NP Katzen ‘cats’ and the verb inside the relative clause and the verb of the matrix clause are elided. Those elements do not form constituents at any time during the derivation, but RNR appears to be still possible.

185 (55) a. [TP AJ-nun [CP MARY-ka ] ], kuliko AJ-TOP

M.-NOM come-because home-to go-PAST-DEC CONJ

[TP SUE-nun [CP JOHN-i S.-TOP

wa-seo]

cip-e kas-ess-ta.]

J.-nom come-because home-to go-PAST-DEC.

‘AJ went home because Mary came, and Sue went home because John came.’ (Korean, Ha 2007) b. [CP Ich habe einen Mann, [CP der DREI ], ], und I

have

a man

who three

cats

owns

knows

and

[CP Sie hat eine Frau, [CP die VIER Katzen besitzt], gekannt.] she has

a woman who four cats

owns

knows

‘I have known a man who owns three cats, and she has known a woman who owns four cats.’ (German, Wilder 1997)

Let us suppose that multiple ERNR features may enter the derivation in the first conjunct as long as the potential RNR target is e-GIVEN and the pre-RNR material bears contrastive focus. Given that the subject of the matrix and embedded clause are each contrastively focused in (55a), we assume that each subject can bear the ERNR feature. Thus, there are two independent deletion processes in the first conjunct, as shown in (56). The lower ERNR feature in MARY licenses elision of its sister, so only the embedded verb is elided. The upper ERNR feature in the matrix subject AJ elides the matrix VP. Notice that the RNR target determined by the upper ERNR feature also includes the entire embedded clause including Mary. This makes the lower ERNR feature actually have no effect because the RNR target by the lower ERNR feature is contained by the one by the higher feature.

186 (56) a. [TP AJ-nun[ERNR] [CP MARY-ka[ERNR] ] ]… AJ-TOP b.

M.-NOM

come-because home-to go-PAST-DEC CONJ

TP wo AJ[ERNR] wp CP VP 6 6 MARY[ERNR]

An immediate question arises about the non-deletion of the embedded subject even if it is within the RNR target of the upper ERNR feature. Let us first consider how the RNR target would be deleted at PF. I assume that the conjunct is shipped to PF and linearized at spell-out. (56) would be linearized as (57a). The shipment also includes which constituent is the RNR target, and the RNR constituent is marked by the subscript E, as shown in (57b). The string of the RNR target undergoes deletion at PF. Notice that there is a focused word within the RNR target. Assuming that focused words are not deletable, PF-deletion cannot include Mary. That is why Mary is pronounced in (56).

(57) a. AJ-NUN > (MARY-KA> (wa-seo)E > cip-ey > kass-ess-ta)E The RNR target b. AJ-NUN > (MARY-KA> (wa-seo)E > cip-ey > kass-ess-ta)E

(at PF)

This raises a question about the direction of deletion within the RNR target. There are two options: the one is that deletion goes from the rightmost edge up to the left edge of the RNR string, so it begins from kass-ess-ta and then proceeds until Mary-ka in (57). Finally, the deletion has to stop prior to the focused Mary-ka, which is unable to be deleted. The deletion of the RNR

187 target by the lower ERNR feature would have no effect. This option provides a correct result. I will call this right-to-left deletion. The other option is that deletion goes from the leftmost edge to the right edge of the RNR string, which I call left-to-right deletion. The RNR target determined by the upper ERNR feature has no effect on deletion because the string deletion would be immediately blocked by the focused Mary-ka, so it would not be able to delete any string, as shown in (58).

The RNR target (58) AJ-NUN > MARY-KA> wa-seo > cip-ey > kass-ess-ta Blocked deletion

Meanwhile, by the lower ERNR feature, the embedded verb wa-seo is marked as the RNR target, so it would be deleted at PF, as shown in (59). The string deleted both by the upper and lower ERNR features results in (59), and this is not the right consequence. Therefore, this indicates that the right-to-left deletion is correct, and crucially that deletion cannot skip over an intervening focused material. The RNR target (59) AJ-NUN > MARY-KA> wa-seo > cip-ey > kass-ess-ta

If the right-to-left deletion hypothesis is correct, it would predict that any unfocused material between the upper and the lower ERNR feature would survive. The prediction seems to be borne out in (61). The adverb ilccik ‘early’ between the ERNR features in the first conjunct cannot be deleted even if it is within the RNR target determined by the upper ERNR feature.

188 (60) [TP AJ-nun [CP ilccik MARY-ka ]
early M.-NOM come-because home-to go

[TP SUE-nun [CP ppali JOHN-i S.-TOP

kaya hay-ss-ta>],

wa-seo]

early J.-nom come-because

cip-e

kuliko

must-PAST-DEC CONJ

kaya hay-ss-ta.]

home-to go must-PAST-DEC.

‘AJ had to go home because Mary came early, and Sue had to go home because John came early.’

The non-deletion of the adverb provides us with two important theoretical implications; First, it supports the claim that PF-deletion operates from right to left. The upper ERNR feature would be able to delete the adverb if the deletion worked the other way around. Second, PF-deletion stops when it encounters any undeletable material and goes no further, so the adverb is not deleted. Let us now turn to the German case (55b). Here the object NP MANN ‘man’ and the numeral in the relative clause are contrastively focused, so let us assume that they both can bear the ERNR feature. Since the nature of the ERNR feature is contrastive focus, I argue that the ERNR feature follows focus projection rules in (61) (Selkirk 1995).

(61) Focus Projection (Selkirk 1995) a. F-marking of the head of a phrase licenses the F-marking of the phrase. b. F-marking of an internal argument of a head licenses the F-marking of the head.

I assume that the ERNR feature can project if it satisfies both focus projection and e-GIVENness conditions. I argue that the upper ERNR feature, which enters the object NP, projects to DP in (62). It follows the focus projection rules in (61) in the following ways: by the second condition (61a), the focused head N F-marks the D head, and then by the first condition (61a), the head F-marks

189 the phrase DP. Thus, we can assume that the phrase DP is considered to bear the ERNR feature and its sister, matrix vP, is a RNR target in (62). The N head bears the ERNR feature, and its sister CP is a RNR target, too. The numeral quantifier DREI ‘three’, which the lower ERNR feature enters with, is an adjunct, so it cannot project further. Therefore, it only licenses the deletion of its sister NP.

(62) a. [TP Ich habe einen MANN[ERNR], [CP der [DP DREI[ERNR] ] besitzt], ]... I b.

have

a

man

who three

cats owns

knows

vP wp DP[ERNR] 3 6 D[ERNR] NP known a 3 NP[ERNR] MAN 3 who : vP 3 DP VP 3 6 AP[ERNR] own THREE cats

Overall, there are three ERNR features independently eligible to delete their sisters at PF in (62). (62) is linearized in (63). The RNR target determined by the ERNR feature in DP is the matrix verb as in (63a). The target determined by the feature in N is the relative clause, as in (63b). Note that this RNR target contains a focused word, thus deletion stops in front of it. Thus, thus only the ‘cat > own’ sequence can be deleted. The RNR target determined by the ERNR feature in AP may delete its sister, NP, in (63c). Combination of those three deletion processes at PF, we obtain the correct RNR linearization in (63).

190 (63) einen > MANN > der > DREI > Katzen > besitzt > gekannt. a. gekannt

(by DP[ERNR])

b. der > DREI > Katzen > besitzt

(by N[ERNR])

c. Katzen

(by AP[ERNR])

To summarize this section, I propose that constituency is observed in RNR. The problematic cases have been revisited here and argued to be resolved by the introduction of multiple ERNR features in the first conjunct. We have observed that each ERNR feature deletes its own sister, and the combination of each deletion has yielded correct consequences in the given examples. I have argued that the direction of PF-deletion must proceed from the right edge of the RNR target to the left edge. As evidence, we have observed that the deletion stops if there is an intervening focused material, and that strings between the focused word and the left edge of the RNR target are pronounced intact. Before ending this section, I will show another prediction that the multiple ERNR feature account can make. Since the combination of the RNR targets, independently motivated by multiple ERNR features, makes PF-deletion possible, we predict that RNR would be ruled out if some ERNR feature is missing. I will focus on the deletion of the matrix verb in the examples above, which can only be licensed by the highest ERNR feature. Thus, if the highest ERNR feature is not licensed – e.g. due to the lack of contrastive focus –, we predict RNR to be ruled out. The schema we are considering is, for example, (64), where the matrix subject does not bear the ERNR feature.

(64) *[TP Subject[ø] [CP Subject[ERNR] ] verb] and …

191 Our prediction seems to be borne out. If the matrix subject in each conjunct is co-indexed, it is not contrastive, so no ERNR feature would be able to enter with the matrix subject. Let us compare (55a), repeated in (65), with (66). While the matrix subject is contrastively focused in (65), it is not in (66). Therefore, we assume that there exists no ERNR feature in the matrix subject of the first conjunct in (66). As a result, the sister of the subject, which is the matrix verb, is not elidable, and only the embedded vP can be elided.

(65) [TP AJ-nun [CP MARY-ka ] ], kuliko AJ-TOP

M.-NOM come-because home-to go-PAST-DEC CONJ

[TP SUE-nun [CP JOHN-i S.-TOP

wa-seo]

cip-e kas-ess-ta.]

J.-nom come-because home-to go-PAST-DEC.

‘AJ went home because Mary came, and Sue went home because John came.’

(66) *[TP AJ-nuni [CP MARY-ka ] AJ-TOP

M.-NOM come-because home-to go-PAST-DEC CONJ

[TP ku-nuni/proi [CP JOHN-i hei-TOP/proi

cip-e ka-ess-ta], kuliko

wa-seo]

cip-e kas-ess-ta.]

J.-nom come-because home-to go-PAST-DEC.

‘AJi went home because Mary came, and hei (also) went home because John came.’

The source of the illicitness of the RNR in (66) is a violation of the Right Edge Restriction since the right edge of the first conjunct must be unpronounced in RNR. In the next section, I will define the right edge restriction under the ERNR analysis.

192

5.5 The Right Edge Constraint under the ERNR analysis One of the motivations for the multiple dominance accounts was that they can capture the robust empirical fact that the RNR target needs to be linearized and pronounced at the right edge of the final conjunct; otherwise, the conflict of linearization between the spellout domains would arise, and ordering preservation is violated. As discussed at the end of chapter 4, the previous deletion accounts have attempted to capture the right edge effects in RNR, and I showed that they encounter both empirical and conceptual problems. For example, Hartmann (2000) claims that there is a general prosodic requirement in that the nuclear accent needs to be aligned on the edge position of prosodic phrase. Given that the contrastive focus is assigned in the middle of the intonation phrase in RNR configuration, the prosodic deprivation of the RNR target is applied as a strategy of the computational system to comply with the prosodic requirement. In other words, PF-deletion of what follows the contrastive focus in the first conjunct till its right edge makes the nuclear accent be aligned at the right edge. However, Hartmann’s analysis cannot be the ultimate answer to the right edge effect because the same prosodic requirement should be applied to the final conjunct in which deletion cannot occur. Right edge effects in RNR do seem to be problematic for the ERNR analysis, too, and I cannot think of any way that can motivate silence of the right edge position only in the first conjunct. The best I can do is to assume that the Right Edge Restriction is a RNR-specific constraint which imposes on PF that the RNR target must be deleted at the right edge of first conjunct. Thus, I define the Right Edge Constraint as in (67).

(67) Right Edge Constraint The RNR target must be deleted at the right edge of the first conjunct.

193 It is important to note that this PF constraint has nothing to do with syntactic, semantic, and phonological requirements of the ERNR feature. In other words, even if those requirements of ERNR are satisfied, RNR still can be illicit, due to a violation of the Right Edge Constraint. I show that (66) is such an example. The RNR of the embedded VP is licensed on the grounds that the ERNR feature enters the derivation with the contrastive embedded subject and marks its sister as the RNR target. The RNR target is e-GIVEN, because the embedded clauses mutually entail each other, modulo F-closure of the focused constituents. And the RNR target is unpronounced at PF. The only thing that goes wrong is that the position of the target is not at the right edge of the first conjunct, which is a violation of the Right Edge Constraint.

5.6 Summary of the ERNR analysis In the analysis, I adopt a variant of Merchant’s (2001) ellipsis feature, which is dubbed ERNR. The ERNR feather enters the derivation along with contrastively focused lexical item and values the interpretable E in the C head. Upon the agreement between the ERNR feature and E in C, the sister of the ERNR feature is defined as the RNR target. During the syntactic derivation, the RNR target is categorically determined, with which we maintain the view that only constituents can be elided, and RNR is not an exception. The RNR target must also satisfy semantic and phonological requirements that the ERNR feature imposes. Semantically, the RNR target must be e-GIVEN: Modulo F-closure, the conjuncts must mutually entail each other. Phonologically, ERNR is interpreted as an instruction to leave the RNR target unpronounced. I also argued that PF-deletion goes from the right edge of the RNR target to the left edge, and an intervening focus stops the deletion. All in all, the final version of ERNR is summarized in (68).

194 (68) ERNR (The final version) a. Syntax of ERNR: The ERNR feature enters the derivation with the contrastively focused preRNR constituent in the first conjunct. &P wo : VP 3 Y[ERNR] abc

: VP 6 … Z …QP abc

b. Phonology of ERNR i) Rule: QP  Ø/ ERNR a b c. ii) Direction of PF-deletion: The deletion begins with c to a. iii) Intervening focus: If b is focused, only c is deleted at PF, leaving a pronounced. c. Semantics of ERNR: e-GIVEN must be observed in RNR. i) RNR  F-clo (A) ii) A  F-clo (RNR).

I also pointed out that RNR has a construction specific PF constraint that the RNR target must be unpronounced at the rightmost edge of the first conjunct. I conclude the ellipsis analysis of RNR in this chapter. The next two chapters will be concerned with the application of this analysis to another domain. I will propose a new analysis of Across-The-Board constructions in light of what I have proposed in this chapter.

195

CHAPTER 6 ACROSS-THE-BOARD CONSTRUCTIONS

In the previous chapters, I argued that RNR is an ellipsis phenomenon and licensed by ERNR. In the next two chapters, I will discuss the connectivity between RNR and Across-The-Board (ATB) constructions and argue that there are no ATB-type of extractions in grammar; what appears to be a simultaneous extraction from both conjuncts is just illusory. I will propose that the apparent ATB movement can be derived from the RNR construction, followed by successive cyclic whmovement only from the second conjunct. There have been some previous attempts that have shown connections between RNR and ATB constructions, and those accounts are based on multiple dominance (Citko 2005, Bachrach and Katzir 2006a, among others). I will review the multiple dominance accounts of ATB constructions and argue that they face problems similar to those they faced for RNR. In the end, I will propose an alternative account in which ATB constructions are derived with two steps: first, the ERNR feature deletes the RNR target in the first conjunct, and then the target in the second conjunct undergoes movement cyclically out of the coordinate structure. I will call this a RNR analysis of ATB constructions. The contents of this chapter are as follows. In section 6.1, I will introduce some traditional concepts and characteristics for ATB extractions. In section 6.2, I will discuss various problems confronted by the movement view since Ross (1967) and evaluate these challenges. In section 6.3, I will review various alternative accounts attempting to resolve those problems. I will show that

196 those alternatives are incomplete and encounter new sets of problems. In chapter 7, I will propose a RNR account for the ATB constructions.

6.1 Coordinate Structure Constraint Coordinate structures are well-known to behave like a syntactic island for movement. Extraction of a conjunct itself or a constituent from (one of) the conjuncts generally results in ungrammaticality. Ross (1967) first notes this restriction and proposes the Coordinate Structure Constraint (CSC) as in (1).

(1)

Coordinate Structure Constraint In a coordinate structure, no conjunct may be moved, nor may any element contained in a conjunct be moved out of that conjunct. (Ross 1967)

Examples in (2a-b) show that the movement of one of the conjuncts is illicit. Here the DP is coordinated, and ungrammaticality results if either of the conjuncts moves out of the coordination. The examples in (3) indicate that wh-movement derived from a single conjunct also results in ungrammaticality.

(2)

a. *Which book did John like to read [the article and t]? b. *Which book did John like to read [t and the article]?

(3)

a. *Which car did John want to sell t, and Mary want to buy the bicycle? b. *Which car did John want to sell the bicycle, and Mary want to buy t?

197 What makes the generalization in (1) complicated is that there is an exception to the CSC. Ross observes that the extraction is acceptable if an identical constituent moves from both conjuncts simultaneously. Notice that compared with (3), wh-movement seems to occur out of the second conjunct in (4) too, and this would make wh-movement leave a trace in each conjunct. Under the generalization in (1), (4) constitutes a violation of the CSC, but it is still acceptable. To account for those exceptions, Ross stipulates that simultaneous movement out of the conjuncts is immune to the CSC, and he dubs this type of movement Across-The-Board (ATB) movement. I will call this configuration an ATB construction.

(4)

Which car did John want to sell t, and Mary want to buy t?

6.2 Challenges for the CSC Ross’ treatment of ATB movement raises many conceptual and empirical questions, centering around the nature of simultaneous movement. For example, in case of ATB wh-movement, each conjunct contains an identical wh-phrase, and each wh-phrase leaves a trace in its conjunct but takes the same specifier position of CP as the other wh-phrase. This is conceptually odd on the grounds that the two independent movements of the copy end up with a single component. In this section, I will introduce previous studies that have challenged Ross’ view of ATB movement as an exception to the CSC (Munn 1993, Franks 1992, Bošković and Franks 2000, Fox 2000, Citko 2005, among others).

198

6.2.1 Multiple copies from movement Citko (2003) argues that Ross’ view on ATB movement does not prevent cases like (5-6). Note that a wh-phrase originates from each conjunct and takes multiple SpecCP positions in (5), and under Ross’ account, this appears not to be a problem because simultaneous extractions from both conjuncts are assumed to be immune to the CSC violation. Of course, English has an independent reason to block a derivation like (5) because it disallows multiple specifier positions of CP.

(5)

*Which cari which carj did John want to sell ti and Mary want to buy tj?

However, the fact that even multiple wh-fronting languages, such as Polish in (6), does not allow multiple specifiers for the ATB construction, supports the argument that the landing site for the ATB wh-movement must be a single SpecCP position.

(6)

*Kogoi kogoj Jan lubi a ti Maria kocha tj? whom whom Jan likes and Maria loves ‘Whom does Jan like and Maria love? (Citko 2003: 3)

6.2.2 Single-identity reading In most cases, ATB movement yields a single-identity reading. 1 Although the wh-phrase is extracted from both of the conjuncts, the referent for both wh-phrases must match. For example,

1

There are some cases where non-ATB readings are possible, due to Höhle (1991), Jacobson (1999), Munn (1999), Nissenbaum (2000). Note that in (i), the paired reading is available, and that the function interpretation for the answer in (ii) is possible. I will discuss the non-ATB reading possibilities in section 7.3.2. (i) Which of his(i,j) parentsk does every Americani love tk BEST, and every Germanj love tk LEAST

199 the identity of the wh-phrase must be the same with regard to the events of each conjunct in (7). Therefore, the question denotes ‘what is x, such that John sold x, and Mary bought x.’, not ‘what are the x and y, such that John sold x, and Mary bought y.’

(7)

What did John sell t and Mary buy t?

An immediate question that arises is how the single-identity interpretation is obtained from the two wh-traces. One possibility is that the moved wh-phrase plays the role of an operator and binds the variables. However, the idea appears to violate Koopman and Sportiche’s (1983) Bijection Principle in (8), which has taken to be crucial for explaining Weak Crossover effects.

(8)

Bijection Principle An operator must bind exactly one variable.

To see the motivation for the Bijection Principle, let us consider (9a-b). (9a) observes Bijection Principle since the pronoun his is bound by the trace of the wh-phrase, so it does not count as a variable. Thus, the wh-phrase as an operator binds only one variable, its trace. On the other hand, (9b) is ruled out since the operator binds the two variables, its trace and the pronoun, hence violating Bijection Principle.

(9)

a. Whoi ti loves hisi mother most? b. *Whoi did hisi mother like ti most?

(ii) Q: What did John like but Bill hate? A: His own car.

200 With this background, let us return to (7) and consider if Bijection Principle is satisfied. Given the assumption that a single occurrence of wh-phrase arrives in SpecCP by wh-movement leaving a wh-trace in each conjunct, the two traces can be considered to be locally A’-bound variables. Thus, the moved wh-phrase is a single operator that needs to bind these two variables as in (10), which constitutes a violation of Bijection Principle.2

(10) What Opx did John sell x and Mary buy x?

6.2.3 Extraction out of only one conjunct There are a significant number of cases where movement is transparently derived from only one of the conjuncts, but still the sentence is fully acceptable (Goldsmith 1985, Lakoff 1986, Kehler 1996). Let us consider the following examples (11-13).

(11) How much can you drink and still stay sober?

(Kehler 1996)

(12) That’s the stuff that the guys in the Caucasus drink and live to be a hundred. (Lakoff 1986) 2

A possible solution to avoid the problems with the Bijection Principle under Ross’ framework would be to assume that two copies of wh-phrase exist in SpecCP, one of which would be deleted under identity (Bošković and Franks 2000). However, we may still assume that there are two operators, assuming that PFdeletion does not affect the status of operators. Therefore, each operator would bind only one variable in (i). (i) What What Opx Opy did John sell x and Mary buy y? Bošković and Franks (2000), however, raise questions for such an analysis, on the grounds that such a possibility does not seem to be available to wh-phrases that move covertly. Let us take (ii). If the two insitu wh-phrases covertly moved to SpecCP, followed by deletion of a copy of wh-phrase at LF, it would predict (ii) to be acceptable, since an operator for each variable still exists. (ii) *Who thinks [(that) Jane detested who] and [(that) Harry adores who]? Bošković and Franks (2000: 123)

201 (13) What did Harry buy, come home, and devour in thirty seconds?

(Ross 1967)

In all these examples (11-13), the extracted constituent in either a wh-question or a relative clause moves only from one of the conjuncts. Thus, the examples above constitute violations of the CSC. They do not fall under the ATB exception because they clearly demonstrate that there is no extraction site in the second conjunct of (11-12). (13) is an interesting example since there are multiple conjunctions. Of the conjuncts, only the first and third conjuncts would allow for a whtrace, and this does not happen in normal ATB constructions

6.2.4 Lack of LF ATB movement Bošković and Franks (2000) argue that ATB movement is prohibited in covert syntax. Let us consider (14). For the existential quantifier to take wide scope, ATB quantifier raising would need to be involved. QR from only one of the conjuncts should be ruled out since it would violate scope parallelism (Sag 1976, Williams 1977, Hirschbühler 1982). The fact that the only surface scope is available in (14) indicates that covert ATB movement is not an available operation.

(14) Every philosopher read some paper and every linguist reviewed some paper. (∀>∃, *∃>∀)

Citko (2003), citing Cho and Zhou (1999) and Wu (1999), provides data from Chinese and Korean that supports Bošković and Franks’ (2000) generalization, and argues that there is a parallel lack of ATB wh-movement in wh-in-situ languages (15-16).3

3

The CSC appears to be observed in wh-in-situ languages, as shown in (i). The wh-phrase in the first conjunct is assumed to move out of the conjuncts at LF, which constitutes a violation of the CSC.

202 (15) Zhangsan xihuan shenme ren Lisi taoyan shenmo ren?

(Chinese)

Zhangsan like which person Lisi hate which person ‘Which person does Zhangsan like and which person does Lisi hate?’

(16) John-i enu salam-ul cohaka-ko Maryka-ka enu salam-ul miweha-ni? John

which person like-and

Mary

(Korean)

which person hate-Q

Which person does John like and which person does Mary hate?

In (15-16), the single-identity reading is not available. These sentences only allow paired readings. To obtain a single-identity reading in these languages, the wh-phrase in each conjunct must find a way to move out of the conjuncts in overt syntax. In Chinese, which lacks scrambling, identical wh-phrases may undergo ATB topicalization in (17). In the case of Korean, ATB scrambling is permitted as in (18).

(17) Shenmo ren Zhangsan xihuan Lisi taoyan? which person Zhangsan like

(Chinese)

Lisi hate

‘Which person does Zhangsan like and Lisi hate?’

(18) Enu salam-ul John-i cohaka-ko Mary-ka miweha-ni? which person John

like-and Mary

(Korean)

hate-Q

‘Which person does John like and Mary hate?’

(i) *John-i enu salam-ul cohaka-ko Maryka-ka Chelswu-lul miweha-ni? J.-Nom which person-Acc like-and M.-Nom C.-Acc hate-Q ‘Which person does John like and which person does Mary hate?

(Korean)

203 To sum up, Ross’ account of ATB movement faces both conceptual and empirical challenges. ATB movement has been described in terms of simultaneous movement, but without any specific restriction on the landing site of the ATB constituent. A more serious puzzle is that the landing site must hold only one position, regardless of how many traces have been left in the original sites. This seems to contradict other accepted syntactic principles, such as the Bijection Principle. There have been several attempts to resolve those puzzles, which I will review in the next section.

6.3 Alternative accounts To avoid the conceptual oddness and empirical problems of ATB constructions, many alternatives have been proposed (Bachrach and Katzir 2006a, Citko 2003, 2005, 2006, Fox 2000, Hornstein and Nunes 1999, Munn 1992, 1999, 2001, Nunes 2004, Postal 1993, Williams 1990, among many others). I will introduce three alternative accounts in this section: sideward movement, parasitic gap, and multiple dominance.

6.3.1 Sideward Movement Inspired by Chomsky’s (1995) decomposition of movement into Copy and Merge, Nunes (1995, 2004) proposes that a copied constituent from one clause can be merged with an unconnected clause, the operation which he calls Sideward Movement. Let us examine (19) under Nunes’ account.

(19) Which view did John like but Mary hate?

204 The wh-phrase is merged with the verb hate in the second conjunct (20a) (a syntactic object with label K), and then an identical copy is created. The copied wh-phrase is merged with the verb like in a different clause L (20b), and this is what constitutes sideward movement.

(20) a. K = [Mary hated [which view]i] Copy which view from K and merge it within L b. L = [John liked [which view]i]

K and L can then be connected by coordination as shown in (21). When the C head is merged with [+wh, +Q], a third copy of the wh-phrase is merged with SpecCP.

(21) Which viewi did [&P [K Mary hate which viewi] and [L John like which viewi]

Then, the wh-phrase in the topmost position forms a chain with its copy in each conjunct, as shown in (22). By Chain Reduction, only the highest copy is pronounced and the copies within the conjuncts are deleted (that is, left unpronounced). As to why the highest copy alone survives until pronunciation, Nunes claims that movement is motivated by feature checking, requiring that the highest copy (local to the feature that motivated the movement) must be preserved to check the uninterpretable wh-feature.

(22) Which viewi did [&P [K Mary hate which viewi] and [L John like which viewi]

Nunes (1995, 2001, 2004) also claims that parasitic gaps can be explained in the same way in terms of sideward movement. Under his analysis, a sentence which has been traditionally

205 considered to include a parasitic gap in the adjunct clause in (23) (due to Engdahl, 1983) contains a real gap. The copies in the matrix and adjunct clauses are not distinguished from each other with respect to whether they are real or parasitic gaps because they are all derived by sideward movement followed by chain reduction.

(23) Which report did John file without reading? a. K = [without reading [which report]i] b. L = [John filed [which report]i]

The wh-phrase, which report, in the adjunct clause K is copied. A copy is merged with the matrix clause as in (23b), and the two clauses are connected. A third copy of the wh-phrase is merged with the matrix clause, followed copy reduction of the two lower copies of the wh-phrase. The complete derivations are shown in (24).

(24) Which reporti did [K John file which reporti] [L without reading which reporti]

Sideward movement analyses predict strong parallels between parasitic gaps and ATB gaps, but yet there are differences, such as those observed by Postal (1993): i) constituent limitation, ii) finite subjects, iii) passivizability constraints, iv) pronominalization constraints, and v) predicate nominals. (25) shows that parasitic gaps are limited to the DP category. When the licensing gap is not a DP, the parasitic gap is not licensed (25a). On the other hand, there is no such restriction for ATB gaps (25b).

206 (25) a. How sick did John look and Betty say he actually felt? b. *How sick did John look without actually feeling? (Postal 1993: 736)

(26) demonstrates that parasitic gaps cannot occur in the subject position of finite clause, a placement which is allowed for ATB gaps.

(26) a. Which patientsi did he convince you ti were already doctors and ti were going to become psychiatrists? b. *Which patienti did he convince ti (that he had agreed) (that) pgi should visit him? (Postal 1993:737)

A verb, like bother, does not permit passivization as shown in (27b), and yet wh-movement appears to be possible from this position, as shown in (27d). (27c) shows parasitic gap constructions are not compatible with verbs that allow passivization. The fact that a comparable ATB construction in (27e) allows extractions from the final clause indicates that parasitic gaps and ATB gaps differ.

(27) a. Greg bothered Lucy with his marital problems. b. *Lucy was bothered by Greg with his marital problems. c. *Whoi did they convince ti that Greg would bother pgi with his marital problems? d. Who did he bother ti his marital problems? e. Whoi did Tony respect ti and (Arnold) constantly bother ti with his marital problems? (Postal 1993:739)

207 (28) shows that parasitic gaps are not allowed where pronouns are prohibited, but there is no such constraint for ATB gaps. This constraint goes along with the DP-like property of parasitic gaps.

(28) a. *What kind of spidersi did he praise ti before learning there were pgi in the soup? b. The kind of spidersi that he found ti in the chicken soup yesterday and there will be ti in the bean soup today are hairy. (Postal 1993: 744)

Predicate nominals can be ATB gaps, but not parasitic gaps, as shown in (29).

(29) a. *What kind of zombiei did they convince ti that Elizabeth had turned into pgi? b. What kind of zombiei did Tom turn into ti and Mark subsequently study ti? (Postal 1993: 747-8)

The asymmetries between ATB and parasitic gaps tell us that parasitic gap constructions are more restricted than ATB constructions. Hornstein and Nunes (2002) claim that the differences are attributed to how flexibly each construction can be licensed by sideward movement. Hornstein and Nunes acknowledge that not every asymmetry Postal (1993) catalogues can be captured under a sideward movement analysis, but they suggest that ATB and parasitic gaps must be licensed in other ways in addition. That is, while parasitic gaps are licensed by Last Resort (Chomsky 1995), ATB gaps can be licensed either by last resort or by the parallelism requirement. Let us examine their proposal in more detail.

208 Last resort requires that movement always have a motivation. It can be understood to restrict both copy and merge. In (23), repeated in (30), the copy of which report is triggered to be merged to the label L, since the verb filed needs to assign θ-role to its complement.

(30) Which report did John file without reading? a. K = [without reading [which report]i] b. L = [John filed [which report]i]

Therefore, last resort can naturally capture the empirical fact that parasitic gaps must be argument DPs. In (25b), repeated in (31), copying a non-argument how sick is not motivated since the adjunct cannot satisfy the theta-role assignment of the verb, so the copy and merge to the label L is not motivated. Therefore, sideward movement of how sick does not take place in (31).

(31) *How sick did John look without actually feeling? a. K = [actually feeling [how sick]] b. L = look

Compared with (31), ATB gaps are licensed in a similar environment in (32). Hornstein and Nunes argue that ATB gaps can be licensed not only via lexical requirement (as parasitic gaps can be) but also in order to satisfy the parallelism requirement. The idea of the parallelism requirement is intuitively understandable as a demand that coordinate structures are parallel between the conjuncts.4 The parallelism requirement Hornstein and Nunes propose for movement under coordination requires that if there is movement in one of the conjuncts, then there must be

4

However, it is not so clear why the coordinate structure need to be parallel.

209 movement in the other. It seems to me, however, that this formulation of the parallelism requirement is simply a restatement of the fact that the CSC be violated only when extractions occur in both conjuncts. For the technical details, let us examine (32). The sentence in (32) fails to meet last resort, since how sick is not an argument. Suppose that ATB gaps can also be licensed by parallelism. L can be syntactically parallel to the embedded clause of K if copying of the whphrase takes place. Thus, Hornstein and Nunes assume that parallelism forces the computational system to copy the wh-phrase and merge it with L.

(32) How sick did John look and Betty say he actually felt? a. K = [and Betty did say he actually felt how sicki] b. L = [

John did look how sicki]

However, the analysis of ATB gaps in terms of sideward movement driven by parallelism encounters many of the same problems that the conventional ATB movement analyses do. For example, it does not account for why in some cases movement out of one conjunct can be licensed even if it fails to meet the parallelism requirement, as shown in (10), repeated in (33).5

5

Hornstein and Nunes (2002: 40) actually touch upon some of the cases where extraction only one of the conjunct is grammatical, acknowledging that they do not have an account why and when the parallelism requirement can be suspended. They bring up an interesting contrast in (i-ii). (i) Which student did Nora go to the drugstore, come home, and talk to t for an hour? (ii) a. *To which student did Nora go to the drugstore, come home, and talk t for an hour? b. *How long did Nora go there, come home, and talk to that student t? c. *Very fat though Nora went to Italy, ate a lot, and become t. (Hornstein and Nunes 2002: 40 (eg. (34)-(35))) Hornstein and Nunes argue that the contrast between (i) and (iia) demonstrates that when the parallelism requirement fails to apply (for whatever reason), extraction appears to be licensed in the same way as in parasitic gap constructions. Recall that parasitic gaps are only allowed to be arguments, as established earlier, and it looks like an argument/adjunct distinction here: the gap in (i) is a DP and the gap in (iia) is a PP. Therefore, they claim that (i) is licensed just like parasitic gaps. However, the (i) vs. (iia) contrast is not really enough to make that case solid.

210 (33) How much can you drink t and still stay sober?

(Kehler 1996)

In addition, there are some asymmetries between ATB and parasitic gap constructions that Hornstein and Nunes’ analysis fails to capture. Munn (2001) points out that both ATB and parasitic gap constructions are sensitive to strong crossover in (34), but insensitive to weak crossover in (35-36), as will be discussed in the next section. The examples in (34) show that strong crossover effects are observed in ATB and parasitic gap constructions on the grounds that movement either from the first or second conjunct equally violates strong crossover. If copy and merge combinations count as a syntactic movement, sideward movement can account for the ungrammaticality in (34) since it leaves a trace of movement in the final clause.

(34) a. *Whosei report did we criticize tj after hei submitted tj. b. *Whosei report did we criticize tj and hei never replied to tj? c. *Whosei paper did hei file tj after we reviewed tj? d. *Whosei paper did hei file tj but we not review tj?

Now we move on to some constructions that do not seem to fit into a sideward movement analysis. The sideward movement analysis does not predict any asymmetry between the conjuncts with respect to movement out of coordinate structures, but there is some reason to believe there is such an asymmetry. Let us consider weak crossover in the sentences in (35-36), due to Munn (2001).

In chapter 7, I will argue that (i) is licensed by information structure (i.e. narration), due to Kehler’s (1996, 2002) Coherence theory.

211 (35) a. Which mani did you visit ti just before hisi boss fired ei? b. *Which mani did hisi boss fire ti just after you visited ei?

(36) a. Which mani did you visit ti and hisi boss fire ei? b. *Which mani did hisi boss fire ti and you visit ei? (Munn 2001: 374)

(35) are parasitic gap constructions, and (36) are coordinate structures. Munn (2001) judged that the (a) examples in (35-36) are acceptable, but not the (b) examples. If this is correct, the data points to the conclusion that in both parasitic gap and ATB constructions, the movement occurs from the first conjunct or main clause, but not from the second conjunct or adjunct clause. If the movement also occurred in the final clause, weak crossover effects should be predicted to appear in all of the examples, so the (a) examples in (35-36) are predicted to be ruled out, contrary to the judgments. Given the assumption that the chain is also formed between the wh-phrase in the highest position and its trace in the second conjunct, the sideward movement analysis predicts the (b) examples should also result in a weak crossover violation. For example, let us examine (36a) in more detail, whose derivation is illustrated in (37). After sideward movement, the wh-copy in the topmost position forms a chain with the one in the first and the other in the second. The lower copy would be deleted under identity, as shown in (37a). At LF, the topmost copy would be an operator and the lower copy would serve as a bound variable. Since the co-indexed pronoun inside the chain is also bound by the operator, the second conjunct also constitutes a weak crossover violation, as shown in (37b).

212

(37) a. Which mani did you visit which mani and hisi boss fire which mani? b. *OPx, … and x’s boss fire x.

(*Bijection Principle)

6.3.2 ATB gaps are parasitic gaps: Parasitic gap analysis To explain such asymmetries as (35-36), Munn (1992, 2001) proposes that ATB gaps are parasitic gaps (See also Pesetsky 1982, Haïk 1985, Huybregts and van Riemsdijk 1985, Williams 1990, among others).6 Just like parasitic gap constructions, Munn claims that overt movement only comes out of the first conjunct for ATB constructions. Following Contreras (1984) and Chomsky (1986), he assumes that a null operator moves in the second conjunct. The derivation of ATB construction would look like (38).7

(38) Which article did Mary like and Bill hate? CP 3 Which articlex C’ 3 C TP did 3 Mary &P wo vP and CP 6 3 like tx OPx TP 6 Bill hate ex

6

Contra Munn (1993, 2001), William (1990) argues that parasitic gaps are ATB gaps. Note that the categories in coordination are different in (38); the first conjunct is vP and the second conjunct is CP. Given that coordination is considered as an adjunct clause under this analysis, it must be assumed that coordination can join different categories. 7

213 The parasitic gap analysis predicts that the place where a weak crossover violation would be triggered is in the first conjunct and the other clauses would be immune to the weak crossover violations. This appears to be what we have observed in (35-36). The (a-b) examples are acceptable on the grounds that the weak crossover environment is in the final clause. On the other hand, weak crossover violations are triggered in the (c-d) examples since the environment is in the first conjunct. The asymmetry is further supported by binding effects in (39). Munn (2001) argues that (39a) would violate Principle A if the movement occurs in an ATB fashion since the subject in the second conjunct cannot bind himself. The grammaticality of (39a) indicates that there is no movement involved in the second conjunct. (39b) is degraded because wh-movement in the first conjunct would leave no binder for the reflexive herself.

(39) a. Which picture of himselfi did Billi sell t and Mary buy? b. *Which picture of herselfj did Bill sell t and Maryj buy?

However, Munn’s (2001) analysis encounters problems explaining why strong crossover is sensitive to movement from the second conjunct. Under Munn’s analysis, it is mysterious that (34b), repeated here in (40), is ruled out despite the fact that the no strong crossover environment exists in the first conjunct.

(40) *Whosei report did we criticize tj and hei never reply to ej?

214 It would be possible to make an assumption that a null operator moves inside the second conjunct and the parasitic gap counts as an R-expression in (41). Then, Principle C is violated, when the pronoun binds the parasitic gap.

(41) *… and [OPx hex never reply to x?]

(*Principle C)

However, this stipulation would face a couple of new problems. If the assumption is correct, then weak crossover effects should also be observed in the environment where the bound pronoun exists in the second conjunct. Let us consider (42). If the null operator moves over the bound pronoun in the second conjunct, it would bind two variables, which also constitutes a violation of Bijection Principle.

(42) Which mani did you visit ti and [OPx hisx boss fire x?]

Second, the asymmetry effect of Principle C seems to be found in other cases, as shown in (43). Note that, in particular, (43b) is judged acceptable. If the parasitic gap contained an R-expression, a Principle C violation would be expected in (43b) and ungrammaticality is predicted, contrary to fact.8

(43) a. *Which picture of Johni did hei like, and Mary dislike? b. Which picture of Johni did Mary like, and hei dislike?

8

(Citko 2005: 494)

We will revisit the asymmetry in (43) in chapter 7. In 7.4.1, I will present that the grammaticality of (43b) is peculiar, by showing that such asymmetry as Principle C has not been found in other ATB constructions, such as ATB topicalization, or ATB scrambling.

215 Movement out of only one conjunct is a canonical violation of the CSC, as shown in (44). Munn’s analysis clearly runs counter to the CSC since in his framework movement must always come out of the first conjunct structure. If nothing like the CSC constrains syntactic representations, however, it raises the question of what blocks the derivation in (44a-b). One might argue that (44a) does not count as a parasitic gap construction since there is no gap in the second conjunct.

(44) a. *What did Bill buy t and Mary sell the book? b. *What did Bill buy the book, and Mary sell t?

However, if we turn the second conjunct into an adjunct clause, it is still acceptable in (45). This indicates that the gap in the first clause is crucial for licensing parasitic gaps in the adjunct clause but the adjunct clause does not have to contain a gap. The contrast between (44a) and (45) tells us that ATB and parasitic gap constructions use different mechanisms.

(45) What did Bill buy t after Mary made a speech?

6.3.3 Multiple Dominance 6.3.3.1 Citko’s (2003, 2005) Parallel Merge The multiple dominance views for RNR have received a fair amount of attention since Wilder (1999). Recently, Citko (2003, 2005) has extended this view to ATB constructions. She proposes that the wh-constituent is first merged with both conjuncts in a parallel fashion (46), which she calls Parallel Merge.

216 (46)

&P wo TP1 and TP2 3 3 John VP Mary VP rp wu V DP V likes who hates

Then, the single occurrence of the wh-phrase inside the shared elements is remerged with SpecCP, so that the ATB wh-question is completed, as shown in (47).

(47)

CP wi Who does &P wo TP1 and TP2 3 3 John VP Mary VP rpwu V DP V like twho hate

Citko’s analysis departs from Wilder’s (1999) with respect to how to linearize a structure like (46). Recall that Wilder needed to modify Kayne’s LCA to avoid conflicts in linearization. In (46), the wh-constituent precedes the verb hates since the first conjunct asymmetrically ccommands the second conjunct. Furthermore, hates precedes the wh-word since it asymmetrically c-commands the wh-word. Modifying the LCA, so that only fully dominated constituents are linearizable, causes who in the first conjunct not to be in the image because it is also dominated by the second conjunct. This modification prevents who from asymmetrically c-commanding hates. By contrast, Citko preserves the original version of Kayne’s LCA, and claims that (46)

217 would not be linearizable due to the ordering conflict. This conflict triggers movement of the shared wh-phrase out of the coordinate structure. The wh-word is remerged with C and pronounced in the topmost position as in (47). Assuming that only pronounced copies count for linearization, this movement resolves the linearization problems that (46) has had because who in SpecCP now asymmetrically c-commands every constituent; it precedes everything in linear order. Citko’s parallel merge account is similar to sideward movement analyses since both mechanisms appeal to chain reduction, where the lower copy must be unpronounced. They differ in what motivates the movement and why the topmost copy must be pronounced. While featurechecking is the driving force for the sideward movement analyses, for the parallel merge account, avoiding conflicts in linearization is the main reason that the shared constituent must move upward, so that it can asymmetrically c-command the rest of the clauses.

6.3.3.2 Bachrach and Katzir’s (2006a) Delayed Spellout Another variant of the multiple dominance accounts has been proposed by Bachrach and Katzir (2006a). As discussed in chapter 3, the core of Bachrach and Katzir’s delayed spellout analysis is that the RNR target is not spelled out until the coordinate structure is completed. Then, the immediate spellout domain above the coordinate structure would be able to spell out the target. In the case of (46), when CP is merged to the coordinate structure, the shared constituent would be spelled out for the first time. Bachrach and Katzir claim that the shared material can be spelled out either the right or left edge of the sentence because these are only places where no ordering conflict occurs. The C head probes a wh-phrase in English, so the shared wh-phrase is remerged into SpecCP and linearized, as in (47).

218 Bachrach and Katzir claim that the delayed spell-out hypothesis can capture the empirical fact that some ATB constructions are insensitive to island violations. ATB movement is generally subject to island violations. For example, in (48), the ATB movement violates the complex noun phrase constraint, resulting in ungrammaticality.

(48) *Whoi did [a man who loves ti sing], and [a woman who hates ti dance]?

Interestingly, ATB movement appears to be able to cross island sites if the wh-phrase is derived from the right-most constituent in each conjunct. Let us consider (49). For the ATB movement of the wh-word to be possible, it must escape the complex noun phrase island in each conjunct. Therefore, the improvement of the judgment in (49), compared with (48), is striking.

(49) Which booki did [TP1 John meet the man who wrote ti], and [TP2 Mary meet the woman who published ti]?

Let us follow how the delayed spellout account works, and then examine whether it can capture the difference between (48) and (49). Let us first consider the good case (49). For an explicit discussion, the tree structure of (49) is provided in (50). In (50), TP is assumed to be coordinated and the ATB constituent is parallel-merged in the base-position. After CP is merged to the coordinate structure, the shared constituent is remerged to SpecCP.

219 (50)

CP wo Which booki C’ 3 C &P did wo TP1 and TP2 2 2 John VP Mary VP 2 2 meet DP meet DP 2 2 the man CP the woman CP 2 2 who VP who VP t 3 wrote published

Bachrach and Katzir claim that each conjunct, TP1 and TP2, would be linearized as in (51). In order to be linearized, a syntactic object must be completely dominated. 9 Since the wh-word has not been completely dominated by the time the structure of each TP is fully constructed, it cannot be linearized.

(51) a. TP1: b. TP2:

9

Complete Dominance is defined in the following: A node X completely dominates a node Y iff (a) X is the only mother of Y, or (b) X completely dominates every mother of Y. The set of nodes completely dominated by X will be called the Complete Dominance Domain of X. (Bachrach and Katzir 2006a: 17)

220 When TPs are coordinated, Bachrach and Katzir assume that the linearization would yield (52). Given the assumption that a constituent cannot precede or follow itself (i.e. reflexivity), only one occurrence of the wh-phrases may remain to be linearized.

(52) TP1 & TP2:

Since TP is not a spellout domain (and neither is &P), the shared wh-constituent has not been spelled out at the point where &P is constructed. The next step is merging C to the coordinate structure, and the shared material can be remerged (i.e. moved) to the topmost position as in (53). Since CP is a spellout domain, the string (53) is spelled out, and here the ATB wh-constituent can be linearized for the first time. Notice that the linearization in (53) preserves the linear order from the previous spellout domains. Therefore, the sentence Which book did John meet the man who wrote and Mary meet the woman who published? is derived.

(53) Wh & TPs:

Like Sabbagh (2007), Bachrach and Katzir also assume that islands can be reduced to a question whether the linear order is preserved among spellout domains. To be more concrete, let us examine what would happen if a complex NP island is involved in a normal wh-extraction (54). The wh-word moves in English, so it cannot be spelled out in its original position. First, the whword moves to the edge of Spec,vP. The wh-phrase in Spec,vP is completely dominated since it cannot undergo any further movement due to the presence of who in the intermediate Spec of CP.

221 Therefore, the wh-phrase which book is forced to be linearized in the vP spellout domain, which makes it a linearization to be unavailable (54).

(54) *Which booki did John meet the man [CP who [vP t’i wrote ti.]

On the other hand, in (49), the spellout of the shared constituent is assumed to be delayed, so that islands inside would not force the shared material to be spelled out. Therefore, the island insensitivity for the example (49) follows. Let us turn to (48) and see if the delayed spellout hypothesis can correctly capture the fact that the example is not linearizable. The short answer is that the analysis actually predicts that (48) can be linearized without violating ordering preservation. Let us examine in more detail with (64), which is the structure of (48).

(55)

CP 3 Who C’ 3 did &P qp TP1 &’ 3 3 DP vP & TP2 3 sing 3 a NP DP vP 3 3 dance man CP a NP 3 3 who vP woman CP r 3 loves who vP 3 hates who

222 Let us compute the linearization of each spellout domain. Each conjunct would be linearized, as in (56). The shared wh-constituent is not spelled out here since it is not completely dominated.

(56) a. TP1: b. TP2:

The next step to consider is the conjoined TPs in (57). A condition on reflexivity rules out the two occurrences of the shared constituent, so only one occurrence may remain to be linearized. Since the coordinate structure is not assumed to be a spellout domain under Bachrach and Katzir’s analysis, the shared constituent is not linearized yet.

(57) TP1 & TP2: < a, man, who, loves, sing, and, a, woman, who, hates, (who), dance>

Once C is merged to the coordination, the shared wh-phrase can be spelled out. The shared constituent is remerged to the left edge of the sentence and precedes the rest of the sentence in linear order in (58). It is important to notice that the linearization in (58) does not violate ordering preservation because it does not contradict the linearizations of the previous spellout domains. The shared wh-phrase is linearized for the first time. Thus, as long as it is linearized at either the left or the right edge, no ordering contradiction would be incurred. Consequently, the sentence in (48) is predicted to be acceptable, contrary to fact.

(58) Wh & TPs:

223 Bachrach and Katzir discuss a similar case in their paper and claim that their analysis can capture the grammaticality of the sentence (59a), compared with its ungrammatical RNR counterpart (59b). (59b) is an unacceptable RNR example because the RNR target is not at the right edge, which is a violation of the Right Edge Restriction.

(59) a. Which horsei did [John meet the man who trained ti] and [Mary meet the woman who claimed ti won the race]? b.*[John met the man who trained], and [Mary met the woman who claimed the black new thoroughbred won the race]. (Bachrach and Katzir 2006a: (75-76))

However, the delayed spellout of the shared wh-constituent in (59a) allows the wh-word to be linearized in the SpecCP position, so no ordering preservation problem arises. Thus, this correctly explains why (59a) is grammatical under the delayed spellout account. The point here is that (59a) is not much different from (48) on the grounds that the shared constituent does not have to occupy the right-edge position before it is linearized. I have argued that the delayed spellout analysis for ATB construction fails to account for some island sensitivity. In the next section, I will discuss conceptual problems for the multiple dominance-based accounts – parallel merge and delayed spellout. And in chapter 7, I will discuss empirical problems for those accounts.

224

6.3.4 Conceptual problems for the multiple dominance accounts 6.3.4.1 Problems for the delayed spellout account Following Fox and Pesetsky’s (2005) cyclic linearization proposal, the predictions of the delayed spellout hypothesis centers on the question of whether the linear order is preserved between spellout domains. Therefore, it is crucial to determine which phrases are spellout domains for the feasibility of this account. If I correctly understand Bachrach and Katzir (2006a), only vP and CP are the spellout domains for their account. Let us assume that this is true and consider the linearization of the sentence in (49) again. Bachrach and Katzir assume that TP is coordinated for the sentence, and show the linearization of each conjunct in (51). Given the assumption that TP is not a spellout domain and that only a spellout domain can be linearized, it is not clear how to interpret the “linearization” of TP in (51), since under their system, the material between vP and CP is not linearized until the whole CP is spelled out.10 Furthermore, after the TPs are coordinated, Bachrach and Katzir assume another linearization occurs at &P in (52), despite the fact that &P is not hypothesized to be a spellout domain. They claim that the shared material is completely dominated by &P, and a single occurrence of which book should be placed on the right as in (52). Since the shared wh-constituent is not spelled out at this point, it is not clear why it matters where the shared wh-phrase should be placed in either TP1 or TP2. In other words, no constraint would prevent the linearization (60) since the wh-phrase has not been spelled out.

(60) TP1 & TP2:

10

Also, notice that the matrix subject in each conjunct should not appear in the linearization yet since TP is not a spellout domain.

225 To prevent this possibility in (60), Bachrach and Katzir should assume that &P is a spellout domain. Although this might rule out the linear order in (60), a new problem arises. To preserve the linear order from the previous spellout domains, the shared wh-phrase must be linearized at the end of the second conjunct. However, notice that in that case, remerging the shared wh-phrase to SpecCP would no longer be possible because this would cause a violation of the linear order preservation. The shared constituent would be linearized in the rightmost edge position in the &P spellout domain, but remerge would linearize the constituent to the topmost position in the next spellout domain preceding the rest of the sentence, which is ruled out by the linear order preservation. There is another critical problem, which involves the linearization status of modals and auxiliaries. The linearization of dummy do is ignored in Bachrach and Katzir’s account. In fact, it is problematic. Let us consider (61) where dummy do is assumed to merge to T. There might be two possibilities – the one is to assume that the auxiliary exists in each conjunct, and the other is to assume that a single occurrence of the auxiliary is parallel-merged – and either option leads to ordering conflicts with the subsequent spellouts. The former option yields ordering conflicts outright because the auxiliary is already spelled out in each conjunct and cannot undergo any further movement without violating ordering preservation. For the latter option, we need to assume that the auxiliary holds up spellout until the coordination is completely dominated, so that it is not linearized in (61a-b).

(61) a. TP1: b. TP1:

226 Now let us compare (61) with the later linearization (62), which is TP1 & TP2. At this point, the auxiliary cannot be linearized since it is not in the edge position. Thus, wherever it is placed, ordering conflicts will necessarily arise, and the linearization fails under Bachrach and Katzir (2006a).

(62) *TP1 & TP2:

The second remark on Bachrach and Katzir’s (2006a) analysis concerns superiority in ATB extractions. Under Bachrach and Katzir’s analysis, superiority violations are also predicted to be possible if the coordinate structure includes the superiority violations. This prediction is not borne out, as can be seen in (63).

(63) a. *What did who take a picture of, and who buy an article of? b. *What did who buy, and who eat?

Since the first spellout domain includes the whole sentence, the sentences in (63) should be linearizable. Bachrach and Katzir (p.c.) must rely on the superiority constraint being an independent rule, rather than a consequence of attract closest to the edge of a spellout domain.

6.3.4.2 Problems for the parallel merge account The parallel merge account appears to solve some problems that Ross’s movement analysis had. Since a single constituent is merged with two clauses in a parallel fashion, no question arises why a sentence like (6), repeated in (64), is not available. Recall that under Ross’s movement analysis

227 the movement is predicted to be licit in (64) since the identical wh-phrase moves out of the conjuncts simultaneously.

(64) *Kogoi kogoj Jan lubi a ti Maria kocha tj? whom whom Jan likes and Maria loves ‘Who does Jan like and Maria love? (Citko 2003: 3)

Under the parallel merge account, the wh-phrase does not leave a trace in two different positions, so it is crucial to assume that in numeration there should be one occurrence of the whphrase. Then, lubi ‘like’ and kocha ‘love’ is merged with the wh-phrase in its underlying structure. The wh-movement of the shared constituent will derive a structure in (65).

(65) a. Kogoi Jan lubi a ti Maria kocha ti? whom Jan likes and Maria loves ‘Who does Jan like and Maria love? b.

CP wi Kogo &P wo TP1 a TP2 3 3 Jan VP Maria VP rpwu V DP V lubi tkogo kocha (Citko 2003)

228 Although the parallel merge account captures the grammaticality of (65), it does not account for the ungrammaticality of (64). To explain why (64) is an unavailable option, it is necessary to come up with a good theoretical reason for what makes the computational system force parallel merge in (65) during the numeration. The answer cannot be that two identical wh-words are disallowed in the numeration, since it would also rule out an interrogative, like Kogoi Jan lubi a ti kogoj Maria kocha tj? ‘Who does Jan like and who does Maria love?’ This suggests that the number of wh-copies is not pre-determined for derivation in (64-65). No answer within the parallel merge account has been provided to date for this conceptual problem. To summarize, the main purpose of this chapter was to review the previous accounts for movement out of a coordinate structure. In particular, it focused on an unusual type of movement, called ATB movement. ATB constructions are unique since they would seem to violate Ross’s (1967) CSC unless encoded in the constraint itself as an exception. Since Ross’s proposals for a movement analysis for ATB constructions, there have been several alternatives attempting to demystify ATB constructions with commonly used syntactic mechanisms, such as (Parallel) Merge, Copy, or Chain reduction. In this chapter, I have evaluated four contemporary accounts of ATB constructions – sideward movement analyses, parasitic gap analyses, and multiple dominance accounts – and shown that each of those alternatives is incomplete and encounters new sets of problems. In the next chapter, I will revisit the CSC and introduce an analysis in which the CSC is stated in terms of the LF representation. And then I will propose a new analysis of deriving ATB constructions in combination with the new definition of the CSC. I will then argue that the underlying structure of ATB constructions is RNR, and that the RNR target in the second conjunct undergoes successive-cyclic movement.

229

CHAPTER 7 A RNR ACCOUNT OF ATB CONSTRUCTIONS

In the last chapter, I reviewed previous analyses of ATB movement. The previous research suffers from both conceptual and empirical problems. In this chapter, to resolve those problems posed for the previous analyses and account for new sets of data, I will propose a new analysis of ATB constructions. This new analysis is based on the ERNR analysis of RNR. I will argue that there is no such syntactic operation as ATB movement, unlike many of the previous analyses assume (Hornstein and Nunes 2001, Munn 2001, Citko 2005, among others), but movement is derived only out of the second conjunct following the ellipsis of the RNR target in the first. I suggest that the underlying structure of the apparent ATB movement is RNR, and when C is merged to the coordinate structure, the RNR target in the second conjunct undergoes successive cyclic whmovement to SpecCP. I will call this a RNR analysis of ATB constructions. It will also be necessary to adopt an alternative account of the CSC to allow movement out of the coordination. And I will introduce a multidimensional account of the CSC (Ruys 1993, Fox 2000, Lin 2001, and Kato 2006). The RNR analysis is directly contrary to Munn’s (2001) parasitic gap analysis of ATB constructions. I will compare my analysis with Munn’s in detail, and conclude that the RNR account is empirically superior to the parasitic gap analysis. In addition to theoretical challenges for the multiple dominance analyses, I will also show that the RNR analysis is more explanatory than the multiple dominance analyses (Bachrach and Katzir 2006a, Citko 2006, among others).

230 The scope of the discussion will be limited to ATB wh-movement, but I assume that the RNR analysis of ATB wh-movement can be applied to the other types of ATB movement, such as ATB scrambling, ATB topicalization, or ATB focus movement. In section 7.7, I will briefly discuss Johnson’s (1996) ATB analysis of verb gapping whose configuration is different from the others; in gapping, the head undergoes ATB movement, but in the other ATB extractions above, an XP undergoes movement. And I will defend the ellipsis account of gapping constructions (Coppock 2001, among others). The RNR analysis can account for the apparent ATB extractions out of (indirect or direct) object positions, given the fact that these are the only positions where RNR can be constructed. In section 7.6.2, I will argue that the subject ATB movement is derived in different manners.

7.1 Williams’ (1978) insight The idea that RNR is the underlying structure for the apparent ATB movement was entertained only in passing – and eventually discarded – by Williams (1978). Let us consider (1).

(1)

Who did Mary like, but John dislike?

By hypothesis, the base construction for (1) is considered to be Mary liked who and John disliked who. Williams assumes that the wh-constituent in each conjunct undergoes rightward ATB movement, just as the movement analysis of RNR hypothesizes, as illustrated in (2).

(2)

Mary liked t, but John disliked t – who.

231 Then, the RNRed wh-phrase undergoes wh-movement to SpecCP, which is schematically shown in (3). This results in the correct linearization for (1).

(3)

CP 3 Who C’ 3 C &P did rp &P t 3 TP TP 6 6 Mary like t John dislike t

However, Williams rejects the RNR analysis for ATB extractions, due to the fact that it would need to assume idiosyncratic movement to be possible in such a case as (4).

(4)

Who does [TP John hate t] but [TP we hope [CP t will win the election]]? (Williams 1978: 34)

With right-node-raising the wh-phrase out of the coordinate structure in (4), the wh-word would leave a trace both in the right-edge position of the first conjunct and in the subject position of the embedded clause in the second conjunct (5). Subsequently, the wh-phrase undergoes leftward movement to SpecCP.

(5)

Who does [&P [TP John hate t] but [TP we hope [CP t will win the election] t]?

232 Williams points out that the ATB movement out of the second conjunct is not licit for the following reasons:

It would require RNRing who out of the subject position of a tensed S, and RNR cannot perform this kind of operation. Thus this kind of derivation cannot be considered an alternative to ATB rule application. (Williams 1978: 34, fn. 4)

Non-movement analyses are first proposed in the early 80s (Wexler and Culicover 1980, McCawley 1982, for example), so only the movement analysis of RNR would be available to Williams (1978) when he attempts to derive ATB movement from RNR. About thirty years later, it is my contention that I will revive the Williams’ insight with the ERNR account of RNR and show that the ATB movement is derived from RNR after all.

7.2 RNR and the apparent ATB movement 7.2.1 Similarity between RNR and ATB In chapters 3 and 5, RNR is argued to be an ellipsis phenomenon and licensed by a variant of the ellipsis feature, ERNR. With the ERNR analysis, I will suggest that constructing RNR is the prerequisite condition for ATB constructions. In other words, the underlying structure for ATB constructions is RNR. RNR and ATB constructions have something in common. For either to be licensed, contrastive focus is required on the constituent just prior to the RNR target or ATB target (Hartmann 2000, Ha 2006, Citko 2006).1 Let us examine (6-7). As discussed in chapter 5, one of the licensing conditions for RNR is that the constituent immediately before the RNR target and its correspondent in the second conjunct must be contrastively focused. (6a) is grammatical in 1

Citko (2006) calls it a distinctness requirement.

233 accordance with the condition as the verbs are contrastive, but (6b) is ungrammatical as contrastive focus cannot be assigned on identical elements.

(6)

a. JOHN LIKED , but MARY HATED – the opera. b. *John liked , and Mary liked – the opera.

The same holds for ATB constructions. It appears that the element prior to the ATB constituent needs to be contrastively focused with its correspondent in the other clause (7a). If this requirement is not satisfied, ungrammaticality results (7b).

(7)

a. What did John LIKE t, but Mary HATE t? b. ??What did John like t, and Mary like t?

Independently, Citko (2006) reports that a distinctness requirement must be met in Polish ATB wh-movement. In (8a-b), the verb and the subject are not contrastive between the conjuncts, respectively. Only (8c) is acceptable since both the subject and the verb are contrastive.2

2

It is surprising, under the view of the ellipsis account, that even the subject needs to be contrastive, because RNR is argued to be licensed by contrastive focus of the pre-RNR element in the first conjunct. We have not discussed any constraint about the subject so far.

234 (8)

a. *Jaką książkęi Jan przeczytał

ti a Maria przeczytała ti?

what book Jan read

[Polish]

and Mary read

‘What book did Jan read and Maria read?’ b. *Jaką książkęi Jan przeczytał what book

ti

Jan read

a

Jan odłożył ti?

and Jan filed

‘What book did Jan read and Jan file?’ c. Jaką książkęi Jan przeczytał what book

ti

Jan read

a

Maria odłożyła ti?

and Maria filed

‘What book did Jan read and Maria file?’ (Citko 2006: 230)

It is important to point out that (8b) has been judged unacceptable in Polish. Similarly, corresponding English examples in (9a), and others in (9b-c) are not acceptable either. Also, their underlying RNR structure is not acceptable in (10a-c).

(9)

a. *What book did John read and John file? b. *Which movie did John watch and John hate? c. *What did John like and John dislike?

(10) a. *John read and John filed the book. b. *John watched and John hated the movie. c. *John liked and John disliked the TV program.

On the other hand, if the coordinate structure does not include the subject in (9-10), grammaticality improves. This is shown in (11). As for the (c) examples in (9-10), they contain

235 semantic failures; it is a straight contradiction. Therefore, grammaticality would not improve even if the size of the coordination were altered.

(11) a. Which book did John read and file? b. Which movie did John watch and hate? c. John read and filed the book. d. John watched and hated the movie.

One conclusion from the contrast in those examples above is that Citko’s example (8b) convincingly argues that the subject – if present – needs to contrast in both RNR and ATB constructions. It does not necessarily jeopardize the status of the RNR analysis, which argues that having a contrastively focused pre-RNR element is assumed to be the only requirement for RNR licensing. The fact that the coordination is possible when the subject is excluded indicates that contrast in the subject is not obligatory for licensing RNR. I suggest that identical subjects under coordination should be avoided, but obviously more work should be followed in the future. To summarize, the fact that there are similar licensing conditions on these two different constructions, ATB and RNR, is presumed not to be a coincidence. The similarity can be captured by the assumption that ATB constructions are derived from RNR. We will explore how it works in section 7.2.2.

7.2.2 ERNR and successive cyclic wh-movement In chapter 5, I introduced ERNR, an ellipsis feature, which licenses RNR. At the point that the coordinate structure is completed in (12), the computational system evaluates whether the ERNR feature is licensed.

236 (12) [&P [TP1 JOHN LIKED[ERNR] what] and [TP2 MARY HATED what]]

The ERNR feature enters the derivation with the contrastively focused verb LIKED in the first conjunct. To satisfy the semantic licensing condition for the ERNR feature, the conjuncts need to mutually entail each other, modulo ∃-type shifting (e-GIVEN, Merchant 2001). If the condition is satisfied, the ERNR feature instructs PF not to pronounce the sister of the feature, as in (13).

(13) C [&P [TP1 JOHN LIKED[ERNR] ] and [TP2 MARY HATED what]]

After RNR is constructed, I propose that the RNR target in the second conjunct undergoes successive-cyclic movement to SpecCP, as in (14). It stops at the edge of the vP phase in the first conjunct and then directly moves to the edge of CP.3 This is how ATB wh-movement is derived under the RNR analysis.

(14) [CP What did[+Q,+wh] [&P [TP1 John LIKE] and [TP2 Mary [vP t’ HATE t]]]

The wh-movement is driven by a need to check the uninterpretable wh-feature in the C head, just as in simple wh-questions. The motivation of the wh-movement in the RNR analysis is distinguished from that under multiple dominance accounts, where wh-movement is driven by a need to resolve the conflicts in linear order of the base position.

3

I will assume that &P is not a phase.

237

7.2.3 The CSC revisited: Multidimensional analysis The new analysis faces a violation of the CSC, which blocks movement out of the coordinate structure. Movement of the RNR target from the second conjunct constitutes a violation of the CSC in (14). To resolve this problem, I will adopt an alternative analysis that captures the CSC not as a movement constraint, but as an LF condition (Ruys 1993, Fox 2000, Lin 2001, and Kato 2006). Before evaluating the LF representation analysis of the CSC, I will begin with the traditional analysis of the CSC at LF interface. May (1985) and Ruys (1993) claim that QR obeys the CSC. An example given to support the claim includes the scope disambiguation effects by a scopally uninformative second conjunct, which prevents a quantifier in the first conjunct from undergoing QR. 4 For example, (15a) is ambiguous since it contains multiple quantifiers. Each quantifier takes scope via QR and the existential quantifier can take either wide or narrow scope with respect to the universal quantifier. On the other hand, the scope ambiguity disappears in (15b): only surface scope is available. The reason is that the object DP the president in the second conjunct is not eligible to undergo QR. Therefore, every student cannot undergo QR alone, since it would violate the CSC.

(15) a. Some teacher likes every student. b. Some teacher likes every student but hates the president.

(∃>∀, ∀>∃) (∃>∀, *∀>∃)

However, the fact that even with a quantifier phrase in the object position of the second conjunct fails to yield inverse scope in (16) casts doubt on this explanation of the unacceptable inverse scope reading of (15b), since the CSC should be exceptionally obeyed by the ATB dependency (cf. Bošković and Franks 2000). 4

Fox (2000: 41) uses the term scopally (un)informative to indicate whether the sentence is scopally ambiguous or not.

238 (16) Every philosopher read some paper and every linguist reviewed some paper. (∀>∃, *∃>∀)

The real question is, then, why the violation of the CSC cannot be salvaged by ATB movement at LF, and this question will be revisited later in section 7.5. Let us consider (17). Since there is no quantifier in the second conjunct, the QR of the universal quantifier in the first conjunct would violate the CSC, so inverse scope is predicted to be unavailable. Contrary to the prediction, inverse scope is possible. To account for this, Ruys (1993) stipulates that the CSC is exceptionally obviated when a pronoun exists in the second conjunct bound by the universal quantifier.

(17) Some student likes every teacheri and wants himi to be the next dean. (∃>∀, ∀>∃)

Fox (2000) takes up Ruys’ observation and proposes a new way of looking at the CSC. Fox argues that extraction out of the conjunct is possible if each conjunct independently obeys all the grammatical constraints (Fox 2000, Lin 2001, Kato 2006, among others). The two components of the coordinate structure for inverse scope are the following in (18).

(18) a. Every teacheri, some student likes ti. b. Every teacheri, some student wants himi to be the next dean.

He argues that the pronoun serves as a variable bound by the universal quantifier at LF. Therefore, (18b) does not involve in vacuous quantification. (18b) is comparable to (15b). The LF

239 configuration of each conjunct in (15b) is shown in (19a-b). Notice that (19b) involves vacuous quantification.

(19) Some teacher likes every student but hates the president.

(*∀>∃)

a. Every studenti, some teacher likes ti. b. *Every studenti, some teacher hates the president.

(Vacuous quantification)

This view of the CSC is not limited to QR. Fox (2000) claims that overt movement also follows from the representational view of the CSC. Let us examine (20-21). The component structures of (20) obey every grammatical constraint given the fact that (20a) and (20b) each are grammatical as they stand. On the other hand, one of the component structures of (21) violates the ban on vacuous quantification, i.e. (21a). Therefore, (21) is ungrammatical.5

(20) What do you think John likes and Mary hates? a. What do you think John likes t? b. What do you think Mary hates t?

5

It is interesting to note that in ATB wh-movement, putting a bound pronoun in either conjunct does not seem to save the grammaticality. Consider (ia-b). If multidimentional analysis of the CSC is correct, however, the sentences in (ia-b) are predicted to be grammatical because vacuous quantification should not arise, just like (17). This seems to be a problem for the multidimensional analysis, but I will ignore it in this dissertation. i) a. ?*Whoi will John marry and Bill fire heri father? b. ?* Whoi will Bill fire heri father and John marry?

240 (21) *What do you think John likes the cake, and Mary hates? a. *What do you think John likes the cake? b. What do you think Mary hates t? (Fox 2000: 50)

The representational view of the CSC fits in both overt and covert syntax, and Fox (2000) generalize it as the following (22).

(22) Multidimensional analysis of coordination a. Extraction out of a coordinate structure is possible only when the structure consists of two independent substructures, each composed of one of the coordinates together with material above it up to the landing site. b. Grammatical constraints are checked independently in each of the component structures. (Fox 2000: 50)

7.2.4 ERNR and the multidimensional view of the CSC With this new view of the CSC, let us revisit our example in (14), repeated in (23). Our derivation of ATB constructions in section 7.2.2 would no longer violate the CSC. By the assumption of the multidimensional view, the CSC is satisfied when subcomponents of the coordinate structure obeys independent grammatical constraints. Recall that the ERNR feature enters the derivation in the verb, and the RNR target in the second conjunct undergoes movement to SpecCP. That is how the ATB construction in (23) is derived.

(23) [CP What did[+Q,+wh] [&P [TP1 John LIKE ] and [TP2 Mary [vP t’ HATE t]]]

241 To see if the CSC is observed, we divide the conjuncts into two subcomponents. Each subcomponent of (23) is shown in (24a-b).

(24) a. What did John like ? b. What did Mary hate t?

Notice that the wh-phrase in the original position differs in (24a-b). While an elide copy of the wh-phrase occupies the object position in (24a), its trace does in (24b). In the case of the whtrace, it is bound by the wh-operator at LF, yielding (25). However, there is no clear prior understanding of how the elided copy of the wh-phrase should be interpreted at LF, yet it is clear how it must be interpreted under the present analysis. Therefore, I assume that the elided wh-copy serves as a variable, and to be bound by the wh-phrase that has moved to a position outside the coordination. (24a) thus yields (25), and the quantification is not vacuous. Since each conjunct is grammatical, the CSC is obeyed. Therefore, the sentence in (23) does not constitute a violation of the CSC.

(25) What λ8 did John like g(8)?

Let us now consider (26). In (26), an argument remains overt in the object position of the first conjunct and there is only one trace position for wh-movement. RNR is not licensed because a cake is not e-GIVEN, so the ERNR feature does not enter the derivation. The two subcomponents of the coordination are provided in (26a-b). The second conjunct (26b) is grammatical; the wh-trace is bound by the operator at LF. However, the wh-operator in (26a) has no variable to bind, so this

242 is a violation of the ban on vacuous quantification. As a result, the CSC is not satisfied and the sentence is ungrammatical.

(26) *What did John BUY a cake, and t’ Mary EAT t? a. *What did John like cake?

(Vacuous quantification)

b. What did Mary hate t?

What is different here from the traditional view of the CSC is that the wh-movement itself is licit in both (23) and (26); that is, a constituent can move out of the coordinate structure. The CSC here is posited as a constraint that needs to be satisfied at LF. (26) is ungrammatical not because the wh-phrase moves out of only one conjunct but because it results in vacuous quantification in one of the conjuncts. The same distinction holds of ATB scrambling in Korean. Let us take (27-28). Under the traditional movement analysis of the CSC, (27) is ruled out because the object DP in the second conjunct moves out of the coordinate structure, while (28) is acceptable as an exception to the CSC.

(27) *Nonmun-ul [[Mary-ka chayk-ul Article-ACC

ponay-ess]-ko [John-i t il-ess-ta]]

M.-NOM book-ACC send-PAST CONJ. J.-NOM

read-PAST-DEC.

‘Mary sent a book and John read an article.’

(28) [CP Nonmun-ul [&P [Mary-ka t PONAY-ESS ]-ko [ John-i t ILK-ESS-TA]] Article-ACC

M.-NOM

send-PAST

‘Mary sent an article and John read it.’

CONJ.

J.-NOM

read-PAST-DEC.

243 First, let us explore how (28) can be explained under the RNR analysis of ATB constructions. I assume that the base structure for ATB scrambling is RNR, and the deletion of the RNRed element follows. The object DP is e-GIVEN, and the verb in the first conjunct, PONAY-ESS-TA ‘send-past-dec’, can bear the ERNR feature. The feature allows its sister on its left side, i.e. the object DP, as shown in (29).

(29)

&P wo : : vP vP 3 6 Mary-ka v’ John-i nonmun-ul ILK-ESS-TA 3 VP v 3 V[ERNR] PONAYESS[ERNR]

The semantic licensing condition of the ERNR feature is satisfied since the antecedent clause entails F-clo (RNR) and the RNR clause entails F-clo (A), as shown in (30). The phonological aspect of the ERNR feature is also satisfied by leaving unpronounced any material immediately following the feature to the right edge of the first conjunct.

(30) a. RNR = Mary sent an article. b. Antecedent = John read an article. c. F-clo (A) = F-clo (RNR) = ∃x∃R [x R-ed an article].

244 After the RNR construction is established, the object DP may scramble out of the conjunct in a successive-cyclic way (Fox and Pesetsky 2005, Ko 2007), targeting SpecCP, as in (31).6 And that is how ATB scrambling is derived.

(31) [CP Nonmun-ul [&P [Mary-ka PONAY-ESS]-ko [ t’ John-i t ILK-ESS-TA]] Article-ACC

M.-NOM send-PAST-CONJ.

J.-NOM

read-PAST-DEC.

‘Mary sent an article and John read it.’

Let us return to the issue of whether (31) satisfies the CSC. The subcomponents of (31) are listed in (32a-b). I assume that the elided copy of the scrambled DP serves as a bound variable, so that the first conjunct does not violate the ban on vacuous quantification in (32a). Since the second conjunct obeys all the grammatical constraints in (32b), the CSC is satisfied. Thus, the scrambling in (31) is acceptable.

(32) [CP Nonmun-ul [&P [Mary-ka PONAY-ESS]-ko [ t’ John-i t ILK-ESS-TA]] a. 1st conjunct: Article-ACC λ8 Mary-NOM send-PAST g(8) b. 2nd conjunct: Article-ACC John-NOM t read-PAST-DEC.

Let us return to (27), repeated in (33). The object DP in the first conjunct cannot be RNRed because it is not e-GIVEN. Thus, there is no deletion in the first conjunct and the object DP is scrambled out of the second conjunct. The subcomponents of the coordination are listed in (33ab). I argue that scrambling violates the CSC in (33), not because the object DP moves out of only one of the conjuncts, but because the representation of the first conjunct in (33a) violates a 6

I assume here that the ATB constituent moves to SpecCP, although nothing really hinges on the specific destination of the scrambling movement.

245 grammatical constraint in Korean; that is, a clause cannot contain more than one direct object without coordination.

(33) *Nonmun-ul [[Mary-ka chayk-ul

ponay-ess]-ko [John-i t il-ess-ta]]

a. *1st conjunct: Article-ACC Mary-NOM book-ACC send-PAST-DEC b. 2nd conjunct: Article-ACC John-NOM t read-PAST-DEC.

Recall that in the last chapter we argued that the following example in (34) would be problematic for Ross’ account of ATB constructions. The argument is based on the definition of the CSC as a movement constraint and ATB is an exception to it. In (34), identical whconstituents move out of the conjuncts simultaneously, so Ross’ account would seem to predict (34) to be an licit ATB movement, contrary to fact.

(34) *Kogoi kogoj Jan lubi a ti Maria kocha tj? whom whom Jan likes and Maria loves ‘Whom does Jan like and Maria love? (Citko 2003: 3)

Under the present analysis of the ATB construction, it is now clear why (34) is ruled out: it violates the CSC. The two subcomponents of the coordination are provided in (35a-b). There are two fronted wh-phrases, but just one wh-trace exists in each position. Consequently, one of the operators would have no variable to bind in both (35a-b), which is banned by vacuous quantification. Therefore, the CSC is violated, and wh-movement is not possible in (34).

246 (35) a. *Kogoi kogoj Jan likes ti. b. *Kogoi kogoj Maria loves tj.

(*Vacuous quantification) (*Vacuous quantification)

7.2.5 ERNR and weak crossover effects One challenge this account might face is that successive-cyclic movement of the RNR target should constitute a weak crossover violation, on the grounds that the moved wh-phrase as an operator would bind the deleted copy in the first conjunct as well as the wh-trace. Let us consider (36). If the deleted what in the first conjunct is a bound variable, the wh-operator would bind it as well as the intermediate wh-variable. Since the operator binds the two variables, it is a violation of Bijection Principle. And this would predict that (36) should show a weak crossover effect, contrary to fact.

(36) What OPx did John LIKE and Mary tx’ HATE t?

It seems to be true, however, that an elided copy and a trace of the moved material behave differently with respect to weak crossover. Let us consider the sentence in (37), which is ambiguous (Merchant 2002). What we are interested in is the wide scope reading of each man with respect to the existential quantifier. To obtain the wide scope reading, each man needs to undergo QR across the ellipsis site where a variable is co-indexed with it, which in turn should constitute a weak crossover violation. Therefore, the wide-scope reading in (37) is predicted to be ruled out, contrary to fact. At this point, I do not know why the elided copy behaves differently from A-bar traces with respect to weak crossover and I will leave this question open for future research.

247 (37) Someone who shouldn’t have kissed each man. = [each man]i [someone who shouldn’t have kissed ti ]

(∀>∃) (Evans 1988: (14))

7.2.6 ERNR and ATB extractions from double object positions Let us consider the examples in (38-39). We observed in chapter 5 that the indirect object cannot be RNRed in double object construction while the direct object can. This is shown in the (a) examples in (38-39).

(38) a. *JOHN gave AN APPLE[ERNR], and BILL gave Mary A PIZZA. b. *Who did JOHN give AN APPLE[ERNR], and BILL give t A PIZZA?

(39) a. John gave MARY[ERNR] , and Mary gave CHRIS – ten dollars. b. How much money did John give MARY[ERNR] and Mary give CHRIS t?

The contrast between (38a) and (39a) has been explained by the location of the ERNR feature. The direct object in the first conjunct bears the ERNR feature and the indirect object is merged higher than the direct object in (38a). Since the indirect object is not c-commanded by the ERNR feature, RNR is not possible. We suppose that the direct object cannot move over the indirect object, which prevents the possibility that the indirect object would be deleted after movement of the direct object, as shown in (40).7 Thus, RNR in (38a) is not licensed. On the other hand, the ERNR

7

This may be due to superiority effects between direct and indirect objects in double object construction (Bruening 2001). This is a crucial difference from dative constructions where the direct object is able to be moved over the indirect object (see the discussion of (42)).

248 feature rests in the indirect object Mary, and is able to license the deletion of its sister ten dollars. Therefore, RNR is properly licensed in (39a).

(40) JOHN gave AN APPLE[ERNR] t, and BILL gave Mary A PIZZA

Since the derivation of the underlying RNR structure crashes in (38a), we predict that ATB whmovement of a direct object would not be acceptable.8 The prediction is borne out in (38b). On the other hand, since RNR is possible in (39a), The ATB construction is derived by extracting the RNR target – i.e. the the indirect object – out of the second conjunct, as shown in (39b). Let us turn to dative constructions. Recall that in chapter 5, I introduced the argument that a dative construction is derived from an underlying structure where the direct object is basegenerated lower than the indirect object (Dryer 1987, Aoun and Li 1993, Pesetsky 1995, Takano 1998, Sauerland 2000, and Sauerland and Elbourne 2002). Then, the direct object undergoes movement over the indirect object as shown in (41).

(41) John gave the book to Mary t, and Bill gave the book to John t.

8

Some native speakers of English find (38b) grammatical, and they also accept a wh-extraction out of the indirect object position in double object construction. (i) Who did Bill give a pizza? There is a controversy concerning the grammaticality of the example in (i). For an interesting debate on the shaky judgment for the sentence, refer to Wasow and Arnold (2005). I will follow judgments, made by Fillmore (1965), Culicover (1976), Wexler and Culicover (1980), and Jacobson (1982), that it is ungrammatical for the indirect object to wh-move or passivize in double object construction (iia-c). (ii) a. John bought Sue a book. b. *Who did John buy a book? c. *Who was bought a book. (Baltin 2001: 252, fn. 2)

249 With this syntactic assumption about the dative structure, let us consider the following examples in (42-43). The deletion of the direct object is derived when the indirect object in the first conjunct bears the ERNR feature. The feature licenses the deletion of the direct object in its base position, as shown in (42a). On the other hand, the deletion of the indirect object is derived when the direct object bears the ERNR feature. The feature deletes the indirect object after the movement for its surface position, as shown in (42b).

(42) a. John gave to MARY[ERNR] , and Mary gave to BILL – the book. b. John gave a BOOK[ERNR] , and Bill sent a present – to Mary.

Let us imagine that the RNR target is a wh-phrase, as in (43). ATB wh-questions in (43a-b) are derived from (42a-b), respectively. In (43a), the direct object wh-phrase in the first conjunct is RNRed in its base position, and the one in the second conjunct undergoes successive cyclic movement to SpecCP. Likewise, the PP wh-phrase in (43b) undergoes successive cyclic whmovement from the second conjunct.9

(43) a. What did John give to MARY , and Mary give to BILL – t? b. To who(m) did John give a BOOK , and Bill send a PRESENT – t?

9

Another variant of (43b) would be the one with a preposition stranded in (i).

(i) Who did John give a BOOK to , and Bill send a PRESENT to t. This is a potential challenge to the present analysis because it is not clear why to survives if the ERNR feature is on BOOK. I suggest that the preposition is cliticized to book, making it part of the pronunciation of book and therefore not deletable.

250 To summarize section 7.2, I have argued that ATB extraction is just an apparent phenomenon. The conditions that license the apparent ATB extraction are the same as those that license RNR, which implies that ATB constructions are derived from RNR structures. I have provided evidence from ATB wh-movement and ATB scrambling. Important to this explanation was the multidimensional analysis of the CSC, a constraint on interpretation rather than a constraint on movement.

7.3 Consequences In this section, I will discuss consequences of the RNR analysis for ATB constructions, combined with the multidimensional view of the CSC.

7.3.1 Single-identity vs. Paired answers As introduced in chapter 6, the answer to ATB wh-questions usually yields a single identity reading. In other words, the paired reading is not available as an answer to the question in (44).

(44) Who does John admire, but Mary dislike? a. J.F.K. b. *John admires J.F.K, but Mary dislikes President Wilson.

However, Munn (1999) argues that there are some ATB cases where the paired reading is available (45-46). Each component question can be answered with the respective sentence.10 In this section, I will argue that the RNR account can capture both of the interpretations.

10

All of the examples in (45) use adjunct wh-words. However, paired answers do not seem to be specific to adjunct wh-phrases, since the argument wh-phrase in (i) also triggers paired answers.

251 (47) a. Where did Mary vacation and Bill decide to live? b. How tired did Bill look and Mary seem? c. Why did Bill leave and Fred arrive?

(48) a. Mary vacationed in Paris and Bill decided to live in Toronto. b. Bill looked exhausted and Mary looked OK. c. Bill left because Fred arrived and Fred arrived because he had a meeting. (Munn 1999: 421)

7.3.1.1 Deriving paired answers Let us first consider how ATB can be answered in paired. Since the RNR account assumes two occurrences of the wh-phrase, the possibility of paired answers can be easily captured. Let us take the question-answer pair in (47). The ATB construction is derived when the wh-phrase in the first conjunct is elided, and the one in the second conjunct undergoes successive cyclic movement in (47a).

(47) a. Where did Mary VACATION[ERNR] , and Bill decide to LIVE – t? b. Mary vacationed in Paris and Bill decided to live in Toronto.

The answer (47b) is derived by two independent LF representations. I assume that in this case each subcomponent of the conjuncts is answered independently and as a combination of these subcomponents, the answer (47b) is derived. Let us first consider the second subcomponent of the

(i) Q: Which man did John kill on Thursday, and Bill murder on Friday? A: *John killed the terrorist on Thursday, and Bill murdered him on Friday. A: John killed the terrorist on Thursday, and Bill murdered his co-worker on Friday.

252 conjuncts in (48). The interpretation of the answer arises when the moved wh-phrase is reconstructed in its base position at LF.

Reconstruction (48) a. Where did Bill decide to live where? b. Bill decided to live in Toronto.

As for the first subcomponent of the conjuncts, the elided copy is bound by the moved whphrase in this conjunct, and it can be answered as (49b).

(49) a. Where did Mary vacation ? b. Mary vacationed in Paris

The point here is that the ATB wh-question in (47) would be divided into two independent questions and each subcomponent can be independently answered. This division is available by the multidimensional analysis of the coordination. Then, the whole answer would be a coordination of the answers to the questions. Therefore, the non-ATB interpretation results from this possibility with the combination of (48b) and (49b): Mary vacationed in Paris and Bill decided to live in Toronto.

7.3.1.2 Deriving single-identity answers Let us turn to single-identity answers. The ATB question can be answered with a single-identity answer, such as in either A1 or A2. A1 is assumed to be derived by focus movement out of its underlying structure in A2.

253 (50) Q: Where did Mary vacation and Bill decide to live? A1: Nantucket Island. A2: Mary vacationed (in Nantucket Island), and Bill decided to live in Nantucket Island.

Under the RNR analysis, there is an elided wh-copy in the first conjunct. However, in this case, the partition of the conjuncts does not arise as we assumed for paired answers.11 The whphrase in the second conjunct undergoes successive cyclic movement (51a) and the one in the first conjunct serves as a variable bound by the moved phrase.12 The variable is bound by the whphrase of the second conjunct since the moved wh-copy has been derived from the second conjunct (51b). Thus, the single-identity reading is derived under the RNR account.

(51) a. Where did Mary vacation and Bill t’’ decide to t’ live t? b. Where λ8 did Mary vacation g(8) and Bill decide to live g(8)?

7.3.2 Functional readings of bound pronominals The RNR analysis can also account for the availability of functional readings, which relies on the partitioning of the conjuncts into subcomponents as described in section 7.3.1.1. Let us consider the following question-answer pair (52).

11

The single-identity answer is not just a situation where the answer to each of the conjunct questions happens to be the same because not all questions with coordination allow paired answers. Thus, the single identity answer could not be derived from the paired answer. At this point, I leave open the issue of why different mechanisms need to be exploited for paired and single-identity answers. I will just suppose that both options are available for the interpretations. 12 As noted earlier, (51) would be a typical case for a weak crossover violation, but I argued that the elided copy behaves differently from the trace with respect to weak crossover effects.

254 (52) Q: Whose car does John LIKE, but Bill HATE? A: His (own) car.

(53-54) show how functional readings of (52) can be derived. Following Merchant (2004), I assume that fragment answers undergo focus movement and the E feature licenses the elision of the remnant in (53).

(53) Q: Whose car does Johni LIKE , but Billj HATE – t? A: Hisi/j (own) car [CP C[E] [John LIKES , but Bill HATES t]]].

For interpretation of the fragment answer, reconstruction must occur at LF in (54). Note that the reconstructed copy exists only in the second conjunct. I assume that the elided copy in the first is simply interpreted at LF. The functional interpretation is derived when each subcomponent is divided into two subcomponents, presented in (54a-b). The pronoun in each conjunct serves as a bound variable, so that it can be bound by the subject of its own clause. This allows each pronoun to be realized with different identities as in (54a-b). Thus, the functional answer results.

(54) His car [John LIKES , but Bill HATES t]. a. 1st conjunct: Johni LIKES hisi car

(by interpretation of the elided DP)

b. 2nd conjunct: Billj HATES hisj car.

(by reconstruction of the moved DP)

255

7.3.3 Movement out of the second conjunct In this subsection, I will provide evidence that shows movement is derived only out of the second conjunct. Given that this analysis is directly contrary to Munn’s (2001) parasitic gap analysis, I will compare these two analyses in detail afterward, in section 7.4.

7.3.3.1 Selectional properties The RNR account makes an interesting prediction with respect to selectional properties. Let us first consider (55-56). Since the verb think cannot select an interrogative complement, a wh-word cannot remain in the specifier of a complement CP. Therefore, the indirect question is not possible in (55a), but the direct question in (55b) is possible, even if the wh-word once occupied the specifier of the intermediate CP, so long as the wh-word moves further to the matrix CP.

(55) a. *Bill thought who went to Prof. William’s talk last night. b. Who did Bill think went to Prof. Williams’ talk last night?

On the other hand, the verb wonder selects an interrogative complement, yielding the reverse pattern. In (56a), the wh-word is allowed to remain in the specifier of the embedded CP. The ungrammaticality of (56b) indicates that further movement out of the embedded clause is not available.

(56) a. John wondered who went to Prof. Williams’ talk last night. b. *Who did John wonder went to Prof. William’s talk last night?

256 With this contrast between (55) and (56), let us consider (57a-b). I argue that (57a) is ruled out, based on the underlying structure in (58). The embedded clause in the first conjunct is elided by the ERNR feature, and the wh-phrase in the second conjunct tries to move to derive an ATB construction. However, the movement of the wh-word is blocked for the same reason as in (56b), because the verb wonder selects a question. Let us consider (57b). The verb think in the second conjunct cannot select a wh-phrase, so it needs to move further. Therefore, the wh-phrase can move to SpecCP and the sentence is grammatical, as shown in (59). This distinction supports that the hypothesis movement is derived from the second conjunct.

(57) a. *Who did JOHN THINK, but BILL WONDER went to Prof. Williams’ talk last night? b. Who did JOHN WONDER, but BILL THINK went to Prof. Williams’ talk last night?

(58) Who did JOHN THINK[ERNR] , but t’ BILL WONDER t went …last night?

(59) Who did JOHN WONDER[ERNR] , but t’ BILL THINK t went …last night?

7.3.3.2 Crossover effects The RNR account predicts strong crossover effects, regardless of whether the pronoun coindexed with the wh-phrase exists in the first conjunct or in the second. Let us consider (60a). The whphrase in the first conjunct is RNRed, and the one in the second conjunct moves to SpecCP. The subject in the second conjunct is coindexed with the wh-trace of the moved wh-phrase, so it binds the trace, resulting in a strong crossover violation (Postal 1971, Wasow 1972, Chomsky 1976,

257 1981, 1982). (60b) is structurally different from (60a), in that the coindexed pronoun is in the first conjunct, rather than in the second. The subject in the first conjunct does not bind the wh-trace in the second conjunct, so strictly speaking, there is no strong crossover effect in (60b).

(60) a. *Whoi did Mary love in the world, but hei disappoint of ti? b. *Whoi did hei love most in the show, but we not like ti at all?

However, (60b) is independently ruled out by the Bijection Principle, on the grounds that the moved wh-phrase binds two variables, i.e. the co-indexed pronoun and its own trace (61).13

(61) *Whoi OPx did x love x most in the show, but we not like x at all?

The RNR account also predicts that we would find weak crossover effects in ATB constructions. Contrary to this prediction, previous literature, especially Munn (2001), observes that a weak crossover violation appears to occur only when the pronoun exists in the first conjunct, which he took as evidence that movement occurs only out of the first conjunct (62).

(62) a. Whoi did you gossip about ti but hisi mother vouch for ei? b. *Which mani did hisi boss fire ti and you visit ei (Munn 2001: 374)

Therefore, the extraction site considered in Munn’s analysis is directly the opposite of the RNR analysis. I will return to this issue in section 7.4, and compare the RNR account with parasitic gap 13

As noted earlier, the elided copy of the wh-phrase is not engaged in crossover effects. Thus, I ignore it being bound with respect the violation of Bijection Principle.

258 accounts, and I will present evidence that the Munn’s arguments for the first conjunct asymmetry are not conclusive.

7.3.3.3 Island effects The RNR account predicts that movement out of the second conjunct would be sensitive to syntactic islands. (63a) is an instance of a violation of the complex noun phrase constraint. (63b) is ruled out for the same reason; the complex NP constraint is violated during successive cyclic movement of the wh-phrase, as shown in (64).

(63) a. *What animal did John know someone who hated t? b. *What animal did John hate and Mary know the man who washed t?

(64) *What animal did John hate and t’’ Mary know the man who t’ washed t?

The analysis further predicts that the movement would be licit when the location of the islands is in the first conjunct (65), because no movement is involved in the first conjunct. However, the judgment for this prediction seems to be noisy. Although (65) is not as bad as (64), it is not perfect. I have nothing further to discuss why this is so.

(65) ??What animal did Mary know the man who washed , and John know Mary hates t?

259 In chapter 6, we reviewed the claim made by Bachrach and Katzir (2006a) that some ATB extractions are not sensitive to syntactic islands, and this can be captured under their delayed spellout analysis. I have two remarks on their claim. First, it is interesting to notice that D-linked wh-phrases are involved in all the examples that they present for the insensitivity in island conditions, such as (66).14

(66) a. Which animali did John say that Mary knew [a man who wrote], and [a woman who published] – an encyclopedia article about ti? b. Which booki did [John meet the man who wrote], and [Mary meet the man who published] ti? (Examples from Bachrach and Katzir 2006a)

If we switch the D-linked wh-phrases into non-D-linked ones, the sentences are considerably degraded (67). Therefore, the cause of the island insensitivity might not lie on the delayed spellout, but in special properties of the wh-phrase involved in (67).

(67) a. ?Whati did John say that Mary knew [a man who wrote], and [a woman who published] an encyclopedia article about ti? b. *Whati did [John meet the man who wrote], and [Mary meet the man who published] ti? 14

Pesetsky (1987) argues that some wh-phrases can be D(iscourse)-linked and behave differently from nonD-linked wh-phrases. The most crucial evidence comes from superiority facts in English in (i). (i) a. *What did who sell? b. Which book did which store sell? Compared with (ia), the superiority violation can be salvaged if D-linked wh-phrases are used in (ib). Pesetsky (2000) argues that the upper D-linked wh-phrase which store only undergoes feature movement. Although it is not clear what role D-linking plays, it seems that it provides some flexibility on syntactic constraints. This is in the similar line with what we have discussed in chapter 5 that discourse coherence can make syntactic mismatches available.

260 Bachrach and Katzir compare (66a) with (68), where extraction is not derived from the coordinate structure, and judge (68) to be degraded. Thus, it is still a mystery that the ATB construction improves grammaticality.

(68) ??Which animali did John say that Mary knew a man who caught ti?

I suspect that the better acceptability for (66a) arises from the complexity of the structure. In general, complex structures seem to be judged more acceptable, perhaps because they provide more cues and semantic content, so that the meanings of the sentence can be recovered with relative ease. Many of my consultants found the sentences to be degraded while processing the first conjunct, just as in (68), but upon further processing discovered that the sentence seemed to be comprehensible as they neared the end of the sentence. The distinction between (66) and (67) shows us that the grammaticality does not arise from the structure. That is, it is not ATB constructions that improve the grammaticality for (66). It is rather the recoverability in connection to the semantically detailed complex structure. Thus, I conclude that if D-linking is factored out, ATB extractions still obey island constraints as a normal movement does. Below are more examples that show island effects in ATB constructions, where an embedded clause is RNRed in (69). It is possible that the main verb in the first conjunct bears the ERNR feature and license the deletion of the embedded clause in (69a), meaning that the ATB construction looks like (69b).

(69) a. John THOUGHT[ERNR] , and Bill KNEW Mary loved Tom. b. Who did John THINK[ERNR] , and Bill KNOW Mary loved t?

261 The deletion of the embedded clause does not save the grammaticality even when the islands have been deleted in the first conjunct in (70a-c). This is due to the island violations, still incurred in the second conjunct.

(70) a. *What did John wonder and Mary ask which guests gave him t? b. *What did Mary run , but Bill call animal controls after John hit t? c. *Who did John spread , and Bill hear of the rumor that Chris will get married to t soon?

Bachrach and Katzir (2006a) predict examples in (70) to be grammatical as well since the only difference between examples in (66) and (70) is the size of the shared material. Notice that the shared material in both (66) and (70) includes the defective syntactic islands, and there is no island beyond the shared material. Therefore, if islands are transparent before spellout, it is puzzling that sentences in (70) are ruled out under their analysis. It further supports my counter arguments to the delayed spellout hypothesis that the improvement of the acceptability in (66) does not come from any syntactic peculiarities of the shared material under coordinate structure, but from easier recoverability provided by more accessible complex structures.

7.3.3.4 Binding facts The successive cyclicity assumed in the RNR account can capture multiple binding possibilities of a reflexive when it stops by intermediate CPs. Let us consider (71). The reflexive himself can be bound by Chris, Bill, and John.

262 (71) [Which picture of himself]i did JOHN SAY[ERNR] <[which picture of himself]j that Chris would disregard tj>, and BILL HOPE ti’ that Chris would disregard ti?

Assuming that the embedded clause is elided by RNR and that the wh-phrase leaves its trace in the intermediate CPs during successive cyclic movement, two readings can be derived. The first reading available from (71) is that the reflexive is bound by Chris in its original position in (72a). The second reading arises when the wh-phrase in each conjunct moves to the embedded CP and is bound by the matrix subject of each conjunct in (72b). Notice that if movement only came out of the first conjunct, as Munn assumes, the reflexive would not be able to be interpreted as Bill, because it cannot move to the intermediate CP in the second conjunct.

(72) a. Which picture of himself did JOHN SAY that Chrisk would disregard which picture of himselfk, and BILL HOPE that Chrisk would disregard which picture of himselfk? b. Which picture of himself did JOHNi SAY which picture of himselfi that Chris would disregard t, and BILLj HOPE which picture of himselfj that Chris would disregard t?

To summarize section 7.3, I have discussed some predictions that the RNR account makes, and found out that the paired and functional reading in ATB constructions support the view that there is an elided RNR copy in the first conjunct and reconstructed at LF. Also, movement only out of the second conjunct is supported by selectional properties, strong crossover effects, island effects, and multiple binding of reflexives.

263

7.4 Comparing analyses In this section, I will compare the RNR analysis with two other analyses – parasitic gap and multiple dominance. The RNR account differs from Munn’s (2001) parasitic gap analysis with respect to where movement is derived from; the former argues for movement out of the second conjunct and the latter argues for the first. In section 7.4.1, I will discuss this contrast with weak crossover, reflexive binding, Principle C, and the use of resumptive pronouns. The RNR account differs from multiple dominance views with respect to how many copies for the ATB constituent are assumed; the former assumes two copies and the latter assumes one. In section 7.4.2, I will address different predictions that the two analyses make.

7.4.1 The source of movement: evidence against parasitic gap analyses Munn (2001) provides evidence for a similarity between ATB gaps and parasitic gaps. Parasitic gap constructions are derived when overt movement occurs in the main clause and a parasitic gap exists in the adjunct clause. The similarity with parasitic gaps would imply that ATB constructions also involve movement only out of the first conjunct, and there is no movement involved in the second conjunct. On the other hand, the RNR account proposed in this dissertation asserts that parasitic gap constructions differ from ATB constructions in terms of the originating site of the movement; here ATB constructions involve movement out of the second conjunct. Therefore, there is tension between the two analyses. Here is how this section will be organized: I will introduce Munn’s analysis of the argument for the parasitic gap analysis first. Then, I will counter that view by pointing out that the analysis would overgenerate, and that it fails to predict some of novel data introduced below.

264 The following examples in (73–75) show asymmetries between the first and the second conjunct with respect to Principle C effects, weak crossover, and reflexive binding. 15 Munn (2001) claims that these effects only arise inside the first conjunct (i.e. the first conjunct asymmetry). On the other hand, the RNR analysis predicts that there would be no asymmetry for (73-74) and the second conjunct asymmetry for (75).

(73) a. *Which picture of Johni did hei like and Mary dislike? b. ?Which picture of Johni did Mary like and hei dislike?

(74) a. *Whoi did hisi boss fire and John hire? b. Whoi did John hire and hisi boss fire?

(75) a. *Which picture of herselfj did John like but Maryj hate? b. Which picture of himselfi did Johni like, but Mary hate?

First, let us consider Principle C effects. Although it appears that there is an asymmetry in movement out of the first and second conjunct, it does not apply to all the ATB constructions.

15

Another argument for the asymmetries comes from the use of resumptive pronouns in (i). Munn (2001) reports that if the first conjunct contains a gap in Hebrew, the second conjunct may contain a gap or a resumptive pronoun. Therefore, (ia) is ruled out since the resumptive pronoun occurs in the first conjunct, but (ib) is correct since the gap occurs in the first and the resumptive pronoun occurs in the second conjunct. (i) a. *ha ?is še Rina roca ?oto ve ?ohevet yoter mikulam the man that Rina wants him and loves more-than anyone b. ha ?is še Rina roca ve ?ohevet ?oto yoter mikulam the man that Rina wants and loves him more-than anyone ‘the man that Rina wants and loves more than anyone’

(Munn 2001: 375-376)

At this point, I have nothing particular to say about the status of this asymmetry under the ellipsis account. Therefore, I will limit myself to discussing Principle C, weak crossover, and reflexive binding in this section.

265 (76-78) show that Principle C is violated wherever the pronoun is located in an ATB relative clause, an ATB topicalization, and an ATB scrambling in Korean.

(76) a. *The studenti who Mary likes but hei dislikes works on Right Node Raising. b. *The studenti who hei likes but Mary dislikes works on Right Node Raising.

(77) a. *President Bushi, every Democrat criticizes, but hei admires. b. *President Bushi, hei admires, but every Democrat criticizes.

(78) a. *John-uli Mary-nun silhe ha-ko, kui-nun coha han-ta. J.-ACC M.-TOP

hate do-CONJ he-TOP like do-DEC.

‘Mary hates John, but he likes himself.’ b. *Johni-ul kui-nun coa ha-ko, Mary-nun silhe han-ta. J.-ACC he-TOP like do-CONJ M.-TOP

hate do-DEC.

‘John likes himself, but Mary hates him.’

The RNR account has no problem with explaining the ungrammaticality of these examples above. Let us first consider (77). The pronoun in the second conjunct binds the co-indexed DP when it is reconstructed in (77a), thus violating Principle C. The elided copy is bound by the coindexed pronoun in the first conjunct in (77b), which also constitutes a violation of Principle C. One might wonder whether Vehicle Change could turn the proper name into a reflexive (i.e. himself) in the elided material. If so, a violation of Priciple C would not occur. However, Fiengo and May (1994) argue that this is prohibited. Let us consider an example in (79). The proper

266

name does not change into a reflexive in the elided VP of the second conjunct. Therefore, Principle C is violated here.

(79) *Louise is proud of Franki, but hei isn’t . (Fiengo and May 1994)

Let us now turn to (78). The verb in the first conjunct bears the ERNR feature which elides John, as illustrated in (80a). The object DP scrambles out of the second conjunct and takes the top position. Since John in the original position is bound by the pronoun, a Principle C violation results. Similarly, in (78b), John is elided in the first conjunct, and its copy in the second conjunct undergoes scrambling out of the conjuncts, as shown in (80b). The pronoun does not c-command the object DP in the second conjunct, so no Principle C violation results here. However, the elided copy of John in the first conjunct is bound by the pronoun, which violates Principle C. It is not eligible to undergo Vehicle Change for the same reason as in (77b).

(80) a. *Johni-ul Mary-nun silhe ha-ko , t’ kui-nun coha han-ta [Johni-ul]. J.-Acc M.-Top

hate do-Conj

he-Top like do-Dec.

b. *John-ul kui-nun coa ha-ko , t’ Mary-nun silhe han-ta t J.-Acc he-Top like do-Conj

M.-Top hate do-Dec.

The second piece of evidence for the first conjunct asymmetry that Munn shows is a weak crossover effect. Let us take (81). (81a-b) show that weak crossover effects do not result when the

267 bound pronoun is located in the second conjunct. On the other hand, (81c-d) indicate that the bound pronoun in the first conjunct incurs weak crossover violations.

(81) a. Whoi did you gossip about ti but hisi mother vouch for ei? b. Which mani did you visit ti and hisi boss fire ei? c. *Whoi did hisi mother gossip about ti and you have vouched for ei? d. *Which mani did hisi boss fire ti and you visit ei? (Munn 2001: 374)

(81a-b) have been judged as grammatical, but not perfect, although they do sound better than (81c-d). Furthermore, this type of contrast can also be found in non-ATB contexts. Let us consider (82). (82a-b) are slightly marginal, but sound better than (82c-d) just as with the comparison between (81a-b) and (81c-d).

(82) a. ?Which employeei did Mary think t’ that hisi boss would fire t next week? b. ?Which mani did Mary say to her friends t’ that hisi boss would fire t next week? c. *Which mani did hisi boss think t’ that Mary would love t very much? d. * Which mani did hisi boss tell Bill t’ that Mary would love t so much?

This tells us that the asymmetry shown in (81) is not very conclusive. What is relevant in (8182) seems to be the judgment becomes fragile when the distance is longer between the moved wh-phrase and the pronoun that incurs weak crossover violations. The point I wish to make here is that sentences in (81-82) are all ungrammatical, but the ungrammaticality of the (a-b) examples seems to be alleviated due to proximity effects (Williams 1990). The RNR analysis correctly

268 predicts examples in (81-82) are ungrammatical due to a weak crossover violation. This is so because wherever the pronoun exists, successive cyclic wh-movement from the second conjunct guarantees that the wh-operator would bind its trace and the pronoun. Finally, let us consider reflexive binding in (75), repeated in (83). The data seem to indicate that the reflexive in the wh-phrase can be bound by the subject in the first conjunct, but not in the second conjunct.

(83) a. *Which picture of herselfi did John buy, and Maryi sell? b. Which picture of himselfi did Johni sell, but Mary buy?

This contrast is puzzling under the RNR account since it predicts exactly the opposite; that is, (83a) is predicted to be grammatical and (83b) is not. While I have no conclusive answer to why this should be the way it is, there is evidence showing that Munn’s analysis is not correct, either.16 Let us consider (84). In (84), a reflexive only exists in the second conjunct. Therefore, Munn’s analysis predicts (84) to be a violation of Principle A, on the grounds that there is no bindier for the anaphor. The prediction does not seem to be borne out, since the sentence is grammatical.

16

Citko (2004) provides one possible answer to this puzzle, assuming that the reflexive is a logophor. A logophor is not subject to the standard binding conditions and usually take any closest (subject or nonsubject) antecedent (Kuno 1987, Pollard and Sag 1992, Reinhart and Reuland 1993). This is shown in (i). The logophor finds its closest antecedent in the previous sentence in (ia). (ib) indicates that the antecedent does not have to be the subject of the clause but a closer object DP. Citko asserts that it is not surprising that (83) shows the proximity effect. This answer is not really satisfactory, however, since Mary in (83a) is the closest antecedent in the sentence with respect to the logophor. Thus, the logophor would be predicted to be bound by Mary. (i) a. Johni was going to get even with Mary. [That picture of himi/himselfi] in the paper would really annoy her, as would the other stunts he had planned. (Pollard and Sag 1992) b. Peteri heard from Andrewj about the picture of himselfj on the wall. (Kaiser, Runner, Sussman and Tanenhaus 2004)

269 This sentence, however, is expected to be fine under the RNR analysis because this shows that a reflexive is bound by the subject in the second conjunct.

(84) Which news of himselfi did the NY times write, and President Bushi hate?

7.4.2 The elided copy: evidence against multiple dominance analyses In chapter 6, I have pointed out some conceptual problems for the multiple dominance analyses of ATB constructions. Furthermore, earlier this chapter, I have already addressed some issues of how the problematic cases for the multiple dominance accounts are handled under the RNR account. The purpose of this section is, therefore, to explore some remaining empirical issues in detail.

7.4.2.1 The paired interpretation The single-identity interpretation as an answer for ATB wh-questions follows under the multiple dominance accounts. Since only one wh-phrase is assumed to be merged with multiple clauses, the wh-phrase is predicted to be interpreted as a single individual, which is correct for the singleidentity interpretation. However, as we saw in sections 7.3.1-7.3.2, a paired interpretation is sometimes possible. For example, the pronoun can have a sloppy identity interpretation in the question-answer pair (85) (See also Höhle 1991, Jacobson 1999, Nissenbaum 2000). The RNR account was able to account for the sloppy identity in 7.3.2. However, under the multiple dominance account, it is puzzling that a single occurrence of the pronoun can be simultaneously bound by the subject of each conjunct. Recall in chapter 2 that the same problem arose for RNR.

270 (85) Q: Whose car does John LIKE, but Bill HATE? A: His (own) car. (Sloppy interpretation)

7.4.2.2 ATB LBE and the right edge condition The Left Branch Extraction (LBE) constraint (Ross 1967) is not respected in some Slavic languages, such as Polish. For example, only the wh-portion of the wh-phrase can move to SpecCP stranding the NP remnant in the original position, as shown in (86).

(86) Którego Kowalski polecił which

t studenta?

Kowalski recommended

student

‘Which student did Kowalski recommend?’

However, Citko (2003, 2006) observes that the LBE constraint must be observed in ATB extraction, as shown in (87). The full wh-phrase must move out of the coordination in ATB extraction.

(87) a. *Którego Kowalski polecił Which Kowalski recommended b. Którego studenta Kowalski polecił

t studenta i

firma

zatrudniła t studenta?

student and company employed t i

firma

zatrudniła t?

which student Kowalski recommended and company employed ‘Which student did Kowalski recommend and the company hire?

student

271 Recall that under Citko’s parallel merge analysis, the motivation for ATB movement is the preservation of linear order. The shared wh-phrase in the original position would cause a conflict in linear order. Thus, it needs to move to the topmost position so that it can asymmetrically ccommand other constituents. Assuming that the wh-portion of the wh-phrase is shared by the conjuncts in (87a), the shared part would cause a linearization problem because the shared constituent is not in the right-edge position. Therefore, the wh-part needs to be remerged to the topmost position by ATB LBE. Given the fact the LBE is possible in Polish, this wh-portion can be remerged to SpecCP and be spelled out there. Thus, (87a) is predicted to be possible under Citko’s analaysis. To accommodate the ungrammaticality of (87a), Citko posits another independent constraint: the conjuncts must contain some contrasting elements (a distinctness condition). Note that the NP remnant studenta is identical between the conjuncts in (87a). Citko observes that if the remnant NPs contrast with each other in coordinate structure, grammaticality improves (88).

(88) Ilei

Maria napisała ti książek a Jan przeczytał ti artykułów?

how-many Maria wrote

books and Jan read

(Polish)

articles

‘How many books did Maria write and how many articles did Jan read? (Citko 2006: 229)

However, the distinctness constraint is not strong enough to distinguish the unacceptable (89a) from the acceptable (89b). Citko disregards the fact that (89b) is significantly improved where ksiazek ‘books’ in the first conjunct is unpronounced. The only difference between (89a) and (89b) is that the shared constituent in (89b) is the entire wh-phrase, while only the wh-portion is shared in (89a). The same reason that has been used for the unacceptability of (89a) should be

272 applied to (89b) as well, on the grounds that (89b) still violates the distinctness condition. Under her analysis, the wh-portion still needs to move out of the conjuncts to avoid the conflict in linearization. However, with respect to the distinctness condition, (89b) commits the same violation as (89a) since the remnant NP books is shared in both conjuncts. Therefore, (89b) is predicted to be as bad as (89a), contrary to fact.

(89) a. *Ilei

Maria przeczytała ti książek a Jan odłozył ti książek?

how-many Maria read

t books

and Jan filed

t books

‘How many books did Maria read and Jan file?’ b. ?[DP Ile ti]j

Maria przeczytała t j a

how-many Maria read

Jan odłozył t j książek?

and Jan filed

books

‘How many books did Maria read and how many books did Jan file?’ (Examples from Citko 2006)

Let us now consider these examples in (89a-b) from the perspective of the RNR account in (90a-b). It is crucial to note that the object DP is e-GIVEN, and the ERNR feature enters the derivation with the verb in the first conjunct. Therefore, the object DP in the first conjunct is licensed to be deleted at PF. Now it is clear why (90a) is ruled out. It is ungrammatical because the remnant NP book in the first conjunct is not deleted at PF. This predicts the improved judgment of (90b) where the whole RNR target is deleted at PF.

273 (90) a. *Ilei

Maria przeczytała[ERNR]

how-many Maria read b. Ilei

ksiazek a

books

Maria przeczytała[ERNR]

how-many Maria read

a

Jan odłozył ti ksiazek?

and Jan filed t books Jan odłozył ti ksiazek?

and Jan filed t books

As seen in (86), LBE is available in Polish, so it is possible to move the wh-portion from the second conjunct to SpecCP as in (91). The derivation of (90b) is illustrated in (91) using the English glosses.

(91) How many Maria read[ERNR] and t’ Jan filed t books

Now a question arises as to what would happen when the remnant NP is distinct, hence undeletable, as in (88). To account for this, I speculate that the NP in the first conjunct undergoes movement out of the DP first, and then the ERNR feature deletes its sister VP, as shown in (92).17

17

I am agnostic to whether the movement of the NP remnant in the first conjunct is rightward. For the purpose of exposition, I will suppose that it moves rightward.

274 (92)

CP ei How many TP qp TP1 & TP2 3 Maria vP Jan read t article 3 vP books 3 wrote[ERNR] 3 6

7.4.3 Interim summary In section 7.4, I have compared predictions between the RNR analysis and two previous analyses – the parasitic gap and multiple dominance analyses. The points of the comparison differ from theory to theory. First, the RNR account differs from the parasitic gap analysis in terms of the conjunct in which the movement originates. Munn (2001) provides evidence of asymmetric movement out of the first conjunct with weak crossover, reflexive binding, Principle C, and the use of resumptive pronouns. In section 7.4.1, I argue that two pieces of Munn’s evidence, Principle C and weak crossover effects, are not very conclusive since the effect is neither attested in other similar ATB constructions nor limited to the ATB constructions. Second, the RNR account differs from multiple dominance views in terms of the (non-)existence of the wh-copy in the first conjunct. The multiple dominance accounts seem to nicely re-address the issue of ATB extractions being exceptional cases for the CSC. There is only a single wh-movement of the constituent shared by the coordinated structures, so neither is there simultaneous extraction out of the coordination nor any need for an exception clause to the CSC. However, the analyses fail to

275 predict the possibility that the wh-determiner can move out of the coordination, leaving the NP remnant only in the second conjunct. On the other hand, the RNR account predicts this to be possible. I also pointed out that there would be no clear way for multiple dominance views to derive paired interpretations of ATB construction. Although some empirical data have been left unexplained, overall, it is clear that the RNR account can capture a broader range of empirical facts than the other accounts.

7.5 ERNR and LF ATB construction 7.5.1 ATB Quantifier Raising In this section, I will address ATB constructions in covert syntax. As noted in chapters 2 and 6, Bošković and Franks (2000) claim that ATB movement is not possible at LF. The typical example showing this claim is presented in (93).

(93) Every philosopher read some paper and every linguist reviewed some paper. (∀>∃, *∃>∀)

However, interestingly, the RNR version of (93) allows the inverse scope reading (cf. Bachrach and Katzir 2006b).18 Let us consider (94). The existential quantifier can QR out of the conjunct and take scope over the universal quantifier, so that it is possible to interpret as there is a single paper that every philosopher read and every linguist reviewed.

(94) Every philosopher read and every linguist reviewed some paper.

18

(∀>∃, ∃>∀)

Bachrach and Katzir (2006b) claim that the shared existential quantifier is not spelled out until it is completely dominated, which allows the existential quantifier to scope over the universal quantifiers. Thus, inverse scope is possible.

276 The possibility of inverse scope in (94) indicates that ATB quantifier raising is theoretically possible. It is important to note that the difference between (93) and (94) is in whether a single overt copy of the ATB constituent exists in the sentence. Note that the same holds for overt ATB constructions in that the two identical phrases in the underlying structure leave a single overtly pronounced copy in surface. Thus, let us make an observational generalization about what ATB constructions entail in (95).

(95) Interpretation of ATB constructions A single-identity interpretation is available if a single overt copy of the ATB constituent is obtained.19

Compared with (93), which has two copies of the existential quantifier, RNR provides only one overt copy of the existential quantifier in (94). To take inverse scope, the existential quantifier in the second conjunct undergoes QR in the same way that a wh-phrase moves in ATB constructions, as shown in (96).

(96) a. [TP Some paper [&P [TP every philosopher read ] and [TP t’ every linguist reviewed t] b. There is some paper x, such that every philosopher read x and every linguist reviewed x.

The inverse scope reading in (96) implies that reducing two identical constituents to a single copy by deletion makes possible the reading that is not available in its underlying structure. This fact is also confirmed in ATB wh-questions. We have already observed that ATB constructions

19

This idea comes out of discussion with Jairo Nunes, to whom I am very grateful.

277 induce a single identity reading as in (97a). And this is not possible when the two wh-phrases are overtly pronounced as in (97b). Thus, overt ATB wh-questions seem to fit in the hypothesis (95).

(97) a. Who did John like but Mary hate? b. Who did John like but who did Mary hate?

(A single-identity reading) (A paired reading)

An immediate question arises as to why a single copy of the ATB constituent is important; namely, what blocks cyclic movement of the existential quantifier in (93). Suppose that an overt copy of the existential quantifier exists in each conjunct, and the one in the second conjunct moves out of the conjuncts, as shown in (98). Following the multidimensional analysis of coordination, we divide the coordination into two subcomponents, as in (98a-b). Let us consider (98a). Since the existential quantifier in the first conjunct is not elided, it cannot count as a bound variable, so (98a) violates a vacuous quantification. Therfore, wide scope interpretation is not available when the ATB constituent is overtly pronounced in the first conjunct.

(98) [Some paper] [&P Every philosopher read some paper and t’ every linguist reviewed t] a. *Some book every philosopher read some paper.

(Vacuous quantification)

b. Some book every linguist reviewed t.

In fact, the elided copy of the existential quantifier in the first conjunct is the source that makes wide scope interpretation possible in (94). The elided copy of some paper serves as a bound variable, so that no vacuous quantification arises, as illustrated in (99a-b).

278 (99) [Some paper] [&P Every philosopher read and t’ every linguist reviewed t] a. Some paper λx every philosopher read x. b. Some paper λx every philosopher read x.

To summarize so far, we can conclude the following things. The observational generalization we made for (95) holds up. An ATB construction is a strategy to obtain a single-identity reading and overt movement does not have to be involved. The evidence comes from the empirical facts that covert movement (i.e. QR) of the ATB constituent derives wide scope interpretation so long as a single overt copy of the ATB constituent is obtained; here, RNR makes it possible to delete one of the ATB constituents. I argue that Sabbagh’s (2007) ACD examples from chapter 2 can be captured in this vein. Let us examine his examples in (100). The crucial reading is (100b) where the ACD part seems to scope over both conjuncts. As Sabbagh claims, this interpretation cannot be obtained if the RNR portion in the first conjunct stays in situ and is interpreted there. In such a case, only the (100a) reading arises.

(100) The doctor tried to give a flu shot to, and administer a blood test for every patient that the nurse did e. a. e = [[give a flu shot to t], and [administer a blood test for t]]. b. e = [tried to [give a flu shot to t], and [administer a blood test for t]]. (Sabbagh 2007: 367)

279 However, given that QR does seem to be available, after all, where RNR occurs in the first conjunct, the RNR target can undergo QR and take scope over both conjuncts, as shown in (101a). Therefore, the ACD gap can include the matrix vP in the first conjunct.

(101) a. [every patient that the nurse did e] [&P The doctor tried to give a flu shot to , and t’ administer a blood test for t. b. e = [tried to [give a flu shot to t], and [administer a blood test for t]].

In the next section, we will revisit relational modifiers which have been problematic for nonmovement analyses of RNR in chapter 3. With the assumption that a single copy of the target constituent may trigger a reading that its underlying structure does not, I will propose that there is a way to explain relational modifiers in RNR under our current analysis.

7.5.2 Relational Modifiers in RNR revisited In section 3.2.1, it was mentioned that relational modifiers, such as similar, together, different, respectively, are problematic for non-movement analyses of RNR. The main problem is that a RNR sentence can have an interpretation that its underlying structure cannot. For example, the distributive reading is only available in (102a), but not in (102b).

(102) a. Peter sings and Mary whistles a similar tune. b. Peter sings a similar tune and Mary whistles a similar tune.

I assume that relational modifiers have quantificational force (cf. Carlson 1987), so the single copy of a similar tune undergoes covert movement, as shown in (103). The modifier moves out of

280 the coordinate structure and takes scope over the whole conjunct. Thus, the distributive reading arises.

(103) [A similar tune] [&P Peter sings and Mary whistles]

Note that this LF movement is only possible due to a single copy of the relational modifier. As discussed earlier in (98), such a movement is blocked for (102b) where a second copy of the modifier is pronounced; were this movement to happen, a violation of the representational CSC would necessarily result. We also observed in chapter 3 that the Korean distributive marker kakkak ‘each, respectively’ must take scope over the conjuncts. Let us consider (104a-b). It is puzzling that the RNR sentence is grammatical in (104b) while its underlying sentence is ungrammatical. This has been considered to be a significant problem for non-movement analyses.

(104) a. *Mary-ka chayk-ul

il-ess-ta

kakkak

kuliko

M.-NOM book-ACC respectively read-PAST-DEC CONJ John-i sinmwun-ul

kakkak

il-ess-ta.

J.-NOM newspaper-ACC respectively read-PAST-DEC b. MARY-KA CHAYK-UL[ERNR]< kakkak M.- NOM book-ACC

il-ess-ta>

kuliko

respectively read-PAST-DEC CONJ

JOHN-I SHINMUN-UL kakkak

il-ess-ta.

J.-NOM newspaper-ACC respectively read-PAST-DEC. ‘Mary read a book, and John read a newspaper, respectively.

281 I argue that the mystery can be solved with the present analysis. Suppose that the distributive marker undergoes covert movement, just like the case of the relational modifier in (103). A single occurrence of the distributive marker in the second conjunct, obtained by RNR, moves out of the coordinate structure and takes scope over the whole conjunct at LF, as in (105).

(105) [Kakkak] [&P MARY-KA CHAYK-UL kuliko JOHN-I SHINMUN-UL il-ess-ta] Respectively M.- NOM book-ACC

CONJ

J.-NOM newspaper-ACC read-PAST-DEC

It should be clear by now why (104a) is ungrammatical. If the distributive marker is also pronounced in the first conjunct, the covert movement is prohibited, due to a violation of the representational CSC. Recall that in section 3.3.2, I also introduced cases where multiple distributive markers are present in each conjunct, and I argued that these cases pose problems for Sabbagh’s movement analysis and the multiple dominance accounts. Let us see if the present analysis can correctly account for these cases. Let us consider (106).

(106) MARY-NUN[ERNR] Chelswu-ka NONMWUN-UL[ERNR]<(??kakkak1) … (kakkak2) …>, M.-TOP

C.-NOM

article-ACC,

kuliko SUE-NUN John-I SINMWUN-UL (??kakkak1) ilk-ess-tako (kakkak2) mal-hass-ta. CONJ.

S.-TOP

J.-NOM newspaper-ACC (respectively) read-PAST-C (respectively) said-DEC

‘Mary said that Chelswui read newspapers, and Sue said that John read articles, respectively.’

282 Suppose that the ERNR feature enters the derivation with the matrix subject and the object DP in the first conjunct, and that its sister would be deleted at PF. The upper ERNR feature in the matrix subject would delete its sister, but only the matrix verb and embedded verb are allowed to be deleted, due to focus intervention of PF-deletion; that is, the focused object DP in the embedded clause prevents further deletion.20 What is interesting in this example is that the distributive marker can only exist in the matrix clause (kakkak2), but it cannot exist in the embedded clause (kakkak1). I propose that this is due to the fact that the covert movement of kakkak1 is bounded by the CP that immediately contains it, which is the embedded clause, as shown in (107). Since kakkak1 cannot move out of the coordinate structure, it does not take scope over the coordination of embedded CP, either. Consequently, Kakkak1 does not satisfy the plurality requirement, so it cannot exist in the embedded CP.

(107) … and [TP Subject … [CP [CP … kakkak1 …]] kakkak2 … verb]

The present analysis can account for why kakkak2 can exist in the matrix clause. Suppose that the matrix TPs are coordinated in the example. The schema I have in mind is shown in (108).

(108) [CP C [&P [TP

] and [TP [CP

] … kakkak2 …]

Kakkak2 can move out of the conjuncts because the immediate CP is outside the coordination. Thus, it moves and takes scope over the coordination of TP, which satisfies the plurality condition for the distributive marker.

20

See section 5.4 for detail.

283

7.6 ATB constructions irrelevant to RNR In the next two sections 7.6-7.7, I will address some ATB constructions whose underlying structure is not RNR. In section 7.6, I will argue that DP coordination and subject ATB constructions are not derived from RNR constructions. In section 7.7, I will discuss Johnson’s (1996/2003) verb gapping.

7.6.1 DP coordination and the CSC In cases where DP is coordinated, all or any subcomponent cannot move out of the coordinate structure. Let us consider (109). It would not be surprising under traditional views of the CSC that (109a-b) are unacceptable since it violates the CSC. On the other hand, (109c) appears to fit in the exception to the CSC on the grounds that the identical constituent is simultaneously extracted out of the conjuncts. To prevent a case like (109c), Grosu (1976) proposes an independent constraint, called Null Conjunct Constraint, which basically states that the conjuncts cannot be empty.

(109) a. *Which book did you read [DP __ and Harry Potter]? b. *Which book did you read [DP Harry Potter and __]? c. *Which book did you read [DP __ and __]?

From the ellipsis point of view, DP coordination is not relevant to RNR. The size of the coordinate structure is too small to posit a contrastively focused constituent in each conjunct, prior to the RNR target. I argue that movement out of DP coordination is ruled out, assuming that DP is a phase (Sabbagh 2003, Svenonius 2004, Hiraiwa 2005). Therefore, any movement out of the DP coordination is blocked, as illustrated in (110a-b).

284 (110) a. *Which book did you read [DP __ and Harry Potter]? CP qo : Phase DP 3 DP &’ 5 3 Which book & DP 5 Harry Potter b. *Which book did you read [DP Harry Potter and __]? CP qo : Phase DP 3 DP &’ 5 3 Harry Potter & DP 5 Which book

7.6.2 Subject ATB constructions The underlying structure of subject ATB constructions cannot be RNR since in most cases, the subject takes the highest position in a clause. Thus, there is no place for the ERNR feature to enter in each conjunct and license the ellipsis of subject in the first conjunct. I propose that what appear to be subject ATB constructions are derived from VP coordination, and crucially the subject does not originate within the coordinate structure. Therefore, the underlying structure for the sentence Who liked candies and hated marshmallows? would be represented in (111), where the subject who is merged beyond the conjuncts.

285 (111) a. Who liked candies but hated marshmallows? b.

vP 3 Who v’ 3 v VP wo VP and VP 5 5 liked candies hated marshmallows

Compared with (double) object ATB constructions, the subject ATB construction in (111) contains only one occurrence of wh-phrase. As was discussed in section 7.3, having two copies of wh-phrase is understood to be responsible for the paired reading, such as (112).

(112) Q: Where did Mary vacation and Bill decide to live? A: Mary vacationed in Paris and Bill decided to live in Toronto. (Munn 1999: 421)

Since there is only one occurrence of the wh-word in (111), the paired interpretation is predicted to be unavailable.21 The prediction is borne out in (113).

21

The unavailability of inverse scope in (i) can be understood in a similar vein. Supposing that the subject leaves no trace within the coordinate structure and cyclic movement of every cat in the second conjunct is blocked at LF, the universal quantifier cannot take scope over the existential quantifier. (i) [vP Some girl [&P [VP kicked every doll] and [VP hugged every cat.]]]

(∃>∀, *∀>∃)

If the subject were within the coordinate structure, the prediction would be different. Since there would exist two copies of the subject quantifier, and assuming that the coordinate structure is partitioned into two subcomponents for LF interpretation as in (iia-b), inverse scope is predicted to be available. (ii) [&P [vP Some girl kicked every doll] and [vP some girl hugged every cat.]] a. [vP Every doll [vP some girl kicked t]] b. [vP Every cat [vP some girl hugged t]]

286 (113) Q: Who liked candies but hated marshmallows? A: My son A: *My son liked candies but my daughter hated marshmallows.

7.7 Gapping: ATB movement vs vP ellipsis Johnson (2006, 1996/2003, inter alia) argues that gapping constructions in English are derived when identical verbs in the coordinate structure undergo ATB movement. Then, the subject from the first conjunct moves out of the conjunct, as illustrated in (114b).

(114) a. John likes apples and Mary pears. b.

TP 3 John PredP 3 likes &P wo vP and vP 3 3 t VP Mary VP 3 3 t apples t pears

Since my analysis does not utilize a movement analysis of ATB constructions, I need to account for gapping in some other way. One may argue that vP is elided after the object DP of the first conjunct moves out of the vP (cf. Merchant 2006).22 This is shown in (115). Since the object 22

Merchant (2006) does not argue for the view that gapping is a case of vP ellipsis. However, he argues that this is how pseudogapping is derived. His account is related with voice mismatching possibilities in VP ellipsis and pseudogapping. He argues that the object DP undergoes focus movement to a FocusP between TP and vP in pseudogapping, and the focus head bears the E feature deleting vP. Thus, pseudogapping does

287 DP pears is focused, it can move to SpecFocP, which is in between TP and vP. Following Merchant’s suggestion, I will suppose that the Foc head can bear an E feature and elide vP. See also different versions of the deletion analysis of gapping (Ross 1967, Hankamer and Sag 1976, Sag 1976, Hartmann 2000, Coppock 2001, Lin 2002, Féry and Hartmann 2005, among others).

(115) [TP John likes apples] and [TP Mary [FocP pears Foc[E] [vP .]]]]

I will not review Johnson’s ATB analysis and the other ellipsis accounts for gapping in this dissertation. See Coppock (2001), for more detailed analysis arguing for ellipsis. The point I would like to make here is that there is a way to analyze gapping with the ellipsis account.

7.8 Exceptional cases and Information Structure 7.8.1 Williams’ example revisited Earlier in this chapter, I introduced the example (5) that led Williams (1978) to discard a RNR approach to the ATB construction. With our present RNR account, let us reconsider (5), repeated here in (116). Suppose that the ERNR feature enters the verb in the first conjunct, so who would be the RNR target for the first conjunct. On the other hand, the RNR target in the second conjunct is the embedded clause, which is the sisiter of the matrix verb hope. Consequently, the RNR target and its antecedent do not mutually entail each other, and RNR is not possible. Since the underlying RNR is not licensed, extraction of the RNR target out of the conjuncts would be illicit.

not allow voice mismatches since the v(oice) head is deleted as well. On the other hand, gapping does not allow voice mismatches between conjuncts in (ia-b). Thus, if Merchant’s analysis is right, such focus movement exploited in (115) should not be available. (i) a. *A Honda was driven by John, and Mary a Toyota. b. *Roses were brought by Mary, and Sue lilies.

288 The sentence is, however, grammatical, so it poses a challenge for the present analysis. At this point, I do not have an answer. This will be left for the future research.

(116) Who does JOHN HATE[ERNR] , but t’ WE HOPE t will win the election?

In fact, it is impossible to analyze this structure with any other contemporary accounts. One could account for (116b) at the expense of parallelism, but that analysis would immediately encounter overgeneration problems since examples like (117) would be predicted to be possible, contrary to fact.

(117) *Who will John call and hang up on him?

Under the RNR account, the unavailability of (117) can be captured by failure of mutual entailment relationships between the conjuncts. Let us consider the underlying structures of the first and the second conjunct in (118a-b). The conjuncts do not mutually entail each other, so RNR is not licensed. Therefore, wh-movement would be illicit.

(118) a. John called who. b. Who hang up on him.

On the other hand, the RNR account can predict the acceptability of (119), since the underlying structures of the conjuncts mutually entail each other. The underlying structure of the first conjunct is YOU SAVE who and that of the second conjunct is someone ACCUSED who.

289 After F-closure, the conjuncts mutually entail each other, so the ERNR feature attached in SAVE elide the wh-word. Then, wh-word in the second conjunct undergoes successive cyclic movement.

(119) Who did YOU SAVE and Bill say t’ was ACCUSED t?

7.8.2 Information structure In the last chapter, I mentioned that some exceptional cases have been observed that do not fit into the standard ATB analysis (Goldsmith 1985, Lakoff 1986, Williams 1989/1990, Kehler 1996). These cases are involved in non-parallel conjuncts in most cases, there appears to be no trace of the moved constituent in one of the conjuncts, and there is a particular information structure relationship between the conjuncts. Let us examine (120-122). In (120), I assume that the VP is coordinated and the wh-phrase moves only out of the first conjunct, so it violates the CSC. However, the sentence has been judged acceptable, contra the prediction (Goldsmith 1985). The multidimensional interpretation of the CSC does not help in this case since the wh-movement results in vacuous quantification in the second conjunct, as illustrated in (120b).

(120) How much can you [&P [VP drink t] and [VP still stay sober]]? a. How much can you drink t? b. *How much can you still stay sober? (Violated Expectation) [Kehler 1996]

The same observation holds for (121) (Lakoff 1986); movement is derived only from the first conjunct in (121a) and from the second in (121b). Thus, those examples should be treated as true exceptions to the CSC.

290 (121) a. That’s the stuff that the guys in the Caucasus drink t and live to be a hundred. [Lakoff (1986), attributed to Farley] b. What kind of toothbrushes did Mary go to the store and buy t? (Cause-effect)

Lakoff (1986) also notes that extraction can also violate the CSC when the relation among conjuncts is Narration in (122). Of the three conjuncts, there is no wh-trace in the second conjunct, but the sentence is acceptable.

(122) What did Harry buy t, come home, and devour t in thirty seconds? (Narration) [Kehler 1996]

The generalization I will assume for the examples above is based on Kehler’s (1996, 2000, 2002, inter alia) Coherence Theory. Kehler (1996, 2000) observes that there are special kinds of information structure built in among the conjuncts in (120-122). He divides information structure into two categories: resemblance and non-resemblance. A resemblance relationship would be one with parallel or contrasting information structure, whereas a non-resemblance relationship would be one involving cause-effect, explanation, violated expectation, or denial of preventer (See Table 2).

291 Table 2. Resemblance vs. Non-Resemblance Resemblance Non-Resemblance • Parallel: • Cause-effect: e.g. Bill likes to play golf. Al enjoys e.g. Bill was about to be impeached. He called surfing the net. his lawyer. • Contrast: • Explanation: e.g. John voted for Clinton, but Mary e.g. Bill called his lawyer. He was about to be voted for Dole. impeached • Violated expectation: e.g. Bill was about to be impeached, but he didn’t call his lawyer. • Denial of preventer: e.g. Bill didn’t call his lawyer, even though he was about to be impeached. (Kehler 2000: 541-543)

Kehler proposes that the non-resemblance information structure allow the syntax to dodge the CSC violation. Let us first consider an example in (123), which constitutes “resemblance” discourse relationship between the conjuncts. Kehler’s generalization predicts the CSC must be observed, and the prediction seems to be borne out. The conjuncts in (123a) do not allow flexibility in applying syntactic constraints and the violation of the CSC is not allowed. Therefore, (123a) is ruled out.

(123) a. *Which newspapers did John buy and read the magazine? b. Which newspapers did John buy and read? (Parallel)

Compared with (123), the examples in (120-122) establish non-resemblance discourse relationships in coordination. Of many subcategories of non-resemblance in discourse structure, Kehler further categorizes (120) as involving violated expectations, (121) as involving a causeeffect relation, and (122) as involving narration. CSC violations seem to be allowed in these

292 examples. Although Kehler’s coherence theory manages to describe the distribution of these examples, it remains unclear why – and how – these types of information structures allow the CSC to be violated.

7.8.3 ATB out of an ECM subject position In this section, I report another ATB construction for which RNR is not the underlying structure. Let us consider (124). Under the RNR account, the wh-phrase is assumed to be derived from Spec,vP in the embedded clause in the second conjunct.

(124) Which candidate did the CIA believe to be loyal but the FBI believe t’ to t be disloyal?

However, if we assume that (125) is the underlying structure of (124), there does not appear to be anything contrastively focused between the conjuncts to license the elision of the wh-copy in the first conjunct. The only candidate for contrast is (dis)-loyal. I assume that the VPs, be loyal and be disloyal, are contrastively focused. In order for this focus to license ellipsis with the ERNR feature, the VP in the first conjunct, bearing the ERNR feature, must move to SpecFocP, as illustrated in (126). The ERNR feature would then license the ellipsis of vP in the first conjunct.

(125) The CIA believes which candidate to be loyal, but Melvin believes which candidate to be disloyal.

(126) Which candidate did the CIA believe [TP to [FocP [VP be loyal][ERNR] ]], but Melvin believe t to t be disloyal?

293 However, this analysis does not account for every example where an ECM subject as the RNR target is involved. In (127), the embedded clause to be intelligent is not contrastive in each conjunct, so there can be no focus movement. Instead, the main verb is contrastively focused in each conjunct and EXPECTED can bear the ERNR feature. The feature licenses the deletion of TP in the first conjunct, as in (128), which is what the RNR account can predict.

(127) This is the student everyone EXPECTED to be intelligent and John recently SHOWED to be intelligent.

(128) Everyone EXPECTED[ERNR] [TP ] and John recently SHOWED the student to be intelligent.

On the other hand, (127) indicates that the deletion of only the ECM subject appears to be possible, contra the prediction as illustrated in (129). Under the RNR account, it is impossible to delete only part of the RNR target since it would violate the phonological licensing conditions of ERNR.23

(129) This is the student everyone EXPECTED[ERNR] [TP to be intelligent] and John recently SHOWED t to be intelligent.

To conclude this chapter, ATB constructions have been traditionally treated as involving simultaneous movement out of the coordinate structure, leaving a trace in each conjunct. In this chapter, I proposed an alternative analysis of ATB constructions. I argued that the ATB 23

However, it seems that the unelided portion of TP in the first conjunct must be deaccented.

294 construction is derived from a pre-built RNR structure, which explains why RNR licensing conditions must be satisfied in ATB as well. As I argued in previous chapters, RNR is an ellipsis phenomenon and licensed by the ERNR feature. While the feature elides the RNR target in the first conjunct, the target in the second conjunct undergoes movement out of the conjunct and take SpecCP in case of ATB wh-constructions, which makes the RNR target the same as the ATB constituent. In this chapter, I also compare the RNR analysis with the other two accounts – parasitic gap and multiple dominance. Although there was some reason to believe that the ATB construction is derived only out of the first conjunct (parasitic gap analysis) or parallel-merged in the base position (the multiple dominance accounts), I concluded that the RNR analysis better captures the facts, and with fewer conceptual problems.

295

CHAPTER 8 CONCLUDING REMARKS

This dissertation has addressed various issues involved in RNR and ATB constructions. Given that these two constructions have not been well-understood under generative syntax, the main goal of this dissertation has been an attempt to demystify them by providing evidence that they can be accounted for as a form of ellipsis. I argued that RNR is a type of ellipsis, and that the derivation of ATB constructions would not be anything particularly exotic if RNR is assumed to be their underlying structure. That is, it establishes a connection between these constructions and the more familiar ellipsis phenomenon, reducing RNR and ATB to special cases of backwards ellipsis under coordination. Chapter 2 was concerned with syntactic structure of RNR. I reviewed three derivational approaches of RNR: movement, strict phonological deletion, and multiple dominance. Among various movement analyses, this dissertation has mainly concentrated on Sabbagh’s (2007) recent proposal that ordering preservation between spellout domains would be able to explain the unbounded nature of rightward ATB movement. The main criticism of his analysis was that ordering is preserved in cases where he shows RNR is impossible. Non-movement analyses – the strict phonological deletion and the multiple dominance accounts – could solve many problems that movement analyses suffer. However, the currently popular multiple dominance hypothesis still faces a number of problems, both conceptual and empirical. A single copy of the RNR target makes it difficult to derive a sloppy identity reading

296 when the target contains a bound pronominal. Such analyses need to make an additional assumption that the pronoun can be interpreted twice. Furthermore, some RNR cases show that the target does not agree with all the conjuncts. For example, an NPI inside the RNR target is licensed even if the first conjunct does not provide the NPI licensing environment. On the other hand, an NPI is not licensed if the second conjunct does not provide the environment. It is not immediately clear how the NPI can be licensed in that case and why there is an asymmetry for the first conjunct. Chapter 3 provided us with an answer: RNR is an ellipsis phenomenon. I presented evidence that those problems faced by previous accounts do not arise in the ellipsis account. The evidence includes sloppy identity, Vehicle Change, morphological mismatches, and parallelism. Those are not special under the ellipsis account because they are common properties of ellipsis. In addition, I argued that challenges faced by the deletion accounts in previous literature can be resolved with ellipsis. I showed some possible solutions for the derivation of the distributive readings where relational modifiers involved in both English and Korean RNR examples. After we established our claim that RNR is ellipsis, we turned into the next question; namely, what licenses RNR. The purpose of chapter 4 was to review the Hartmann-Féry account (Hartmann 2000, Féry and Hartmann 2005) in which the licensing conditions for RNR were proposed. Since the analysis is based on Rooth’s (1992a, 1992b) focus semantics and its role in the licensing conditions of ellipsis, I started this chapter with the review of Rooth (1992a, 1992b). Rooth’s account has been argued against in previous literature (e.g. Merchant 2001), due to its appeal to structural parallelism between the antecedent and the ellipsis. I pointed out that there are also some good RNR cases where structural isomorphism is not observed, which the HartmannFéry analysis fails to capture. Thus, I suggested that Hartmann-Féry’s licensing conditions for RNR need to be relaxed. Rather than assuming both syntactic and semantic parallelism, the only

297 necessary licensing condition for ellipsis is a semantic one: a mutual entailment relationship between the antecedent and the elided clause. Of those various accounts in this line, I adopted Merhant’s (2001) e-GIVENness conditions and applied to RNR. With e-GIVEN, we can capture the differences between deaccenting and RNR with the strength of application of the e-GIVENness conditions. At the same time, it can capture the cases that were problematic under the HartmannFéry analysis, where the structure of RNR is different from that of the antecedent. Chapter 5 was concerned specifically with the adaptation of Merchant’s e-GIVENness conditions to RNR. I proposed a complete set of licensing conditions for RNR by laying out syntactic and phonological aspects of licensing conditions. I argued that there is an elliptical feature (i.e. ERNR) that enters the syntactic derivation along with contrastively focused lexical item in the first conjunct in syntax. The sister of the feature becomes the RNR target, with which we maintain the view that only constituents can be elided, in RNR as in other forms of ellipsis. Also, the semantic licensing conditions, i.e. e-GIVENness, must be satisfied: After F-closure, the conjuncts must mutually entail each other. At PF, then, the syntactically licensed ERNR feature is interpreted as an instruction to leave the RNR target unpronounced to the maximum extent possible, by deleting material in the target from right to left. The deletion proceeds from the right edge of the RNR target up to the ERNR feature. The evidence for the leftward nature of the deletion comes from a Korean RNR example where the deletion can be interrupted by the intervening focused material. In the end of the chapter, I generalized the ERNR analysis in (1).

298 (1) The ERNR feature a. Syntax of ERNR: The ERNR feature enters the derivation with the contrastively focused preRNR constituent in the first conjunct. &P wo : VP 3 Y[ERNR] abc

: VP 6 … Z …QP abc

b. Phonology of ERNR i) Rule: QP  Ø/ ERNR a b c. ii) Direction of PF-deletion: The deletion begins with c to a. iii) Intervening focus: If b is focused, only c is deleted at PF, leaving a pronounced. c. Semantics of ERNR: e-GIVEN must be observed in RNR. i) RNR  F-clo (A) ii) A  F-clo (RNR).

The final two chapters were concerned with ATB constructions and their correlation with RNR. The main purpose of chapter 6 was to review the previous accounts for movement analyses of ATB constructions. The chapter began with a review of the unusual nature of this type of movement; ATB constituents move out of a coordinate structure which would violate Ross’ (1967) Coordinate Structure Constraint, and they need to take a single position for its landing site even if it leaves two traces in the conjuncts. Since Ross’ proposals for a movement analysis for ATB constructions, there have been several alternatives attempting to demystify ATB constructions with commonly used syntactic mechanisms, such as (Parallel) Merge, Copy, or Chain reduction. I have evaluated three contemporary accounts for ATB constructions – sideward

299 movement analyses, parasitic gap analyses, and multiple dominance accounts and point out problems each analysis face. In chapter 7, I proposed a new analysis of ATB constructions where the underlying structure of ATB constructions is argued to be RNR, which in turn means that ATB also involves in ellipsis. I argued that the ATB constructions are derived from a pre-built RNR structure, which explains why contrastive focus is crucial for both constructions. As proposed in chapter 5, the ERNR feature enters the derivation with a contrastively focused lexical item in the first conjunct. Then, while the ERNR feature elides the RNR target in the first conjunct, the target in the second conjunct undergoes successive-cyclic movement to SpecCP. Thus, ATB structures are constructed. I called it the RNR analysis of ATB constructions. I compared the RNR analysis with the other two accounts – parasitic gap and multiple dominance. While the RNR account predicts that movement effects should be seen only in the second conjunct, the parasitic gap analysis predicts the opposite; we would see the effects only in the first conjunct. I argued that the evidence for movement out of the first conjunct is not conclusive, since the first conjunct effects are not found in other ATB constructions. Also, the evidence from weak crossover effects could also be captured under alternative explanations, such as proximity effects. Multiple dominance accounts assume that there exists a single ATB copy, compared with the RNR analysis in which an identical ATB constituent exists in each conjunct. Interesting issues of interpretation arise when we compare these two analyses. The RNR analysis can capture both a single-identity and a paired interpretation, but a single-identity reading is the only one that multiple dominance account can explain. Thus, I concluded that the RNR analysis is better at capturing the empirical facts and is more conceptually palatable. We also looked at the question of whether LF ATB movement is possible, and I presented new data showing LF ATB constructions are actually possible (contrary to some prior claims) if there is a way to obtain a

300 single copy of the ATB constituent. In other words, constructing ATB does not require overt movement. Rather, what ATB entails is the existence of a single ATB constituent. This means, then, that there is no difference between RNR and ATB. ATB is simply a RNR structure with an additional movement, and the motivation for this movement is independent of ATB. The ATB constituent moves because some probe higher attracts it.

301

BIBLIOGRAPHY

Abbott, Barbara. 1976. Right node raising as a test for constituenthood. Linguistic Inquiry 7: 639642. Abels, Klaus. 2004. Right Node Raising: Ellipsis or ATB movement? In Keir Moulton, and Metthew Wolf (eds.), Proceedings of the 34th North East Linguistics Society, 44-59. GLSA Publications, Umass, Amherst. Ahn, Hee-Don, and Yongjoon Cho. 2006. Semantic identity vs. phonological identity: Some implications for verb-less coordination in Korean. To appear in Toronto Working Papers in Linguistics. Akmajian, Adrian. 1975. More evidence for an NP cycle. Linguistic Inquiry 6: 115-129. An, Duk-Ho. 2006. Right Node Raising as PF deletion. To appear in Proceedings of the 16th Japanese/Korean Linguistic Conference. An, Duk-Ho. 2007. Syntax at the PF interface: Prosodic mapping, linear order, and deletion. Ph.D. dissertation. University of Connecticut. Aoun, Joseph, and Yen-hui Audrey Li. 1989. Scope and constituency. Linguistic Inquiry 20: 141172. Aoun, Joseph, and Yen-hui Audrey Li. 1993. Syntax of Scope. Cambridge, MA: MIT Press. Bachrach, Asaf, and Roni Krazir. 2006a. Right Node Raising and Delayed Spell-Out. Ms., MIT. Bachrach, Asaf, and Roni Krazir. 2006b. Right Node Raising and QR. Ms., MIT. Baltin, Mark. 1987. Do antecedent-contained deletions exist? Linguistic Inquiry 18: 279-295.

302 Baltin, Mark. 2001. A-movement. In Mark Baltin and Chris Collins (eds.), The Handbook of Contemporary Syntactic Theory. Blackwell Publishers. Barss, Andrew, and Howard Lasnik. 1986. A note on anaphora and double objects. Linguistic Inquiry 17: 347-354. Bartos, Huba. 2001. Sound-form non-insertion and the direction of ellipsis. Acta Linguistica Hungarica, 48: 3–24. Beck, Sigrid. 2000. The semantics of different: Comparison operator and relational adjective. Linguistics and Philosophy 23:101-139. Beck, Sigrid. 2006. Intervention effects follow from focus interpretation. Natural Language Semantics 14: 1-56. Booij, Geert. 1985. Coordination reduction in complex words: a case for prosodic phonology. In Harry van der Hulst and Noval Smith (eds.), Advance in Nonlinear Phonology, 143-160. Dordrecht: Foris. Bošković, Željko. 1997/2004. Two notes on right node raising. In Miguel Rodriguez Mondonedo and Emma Ticio (eds.), University of Connecticut Working Papers in Linguistics 12, 1324, Department of Linguistics, University of Connecticut, Storrs. Bošković, Željko, and Steven Franks. 2000. Across-The-Board movement and LF. Syntax 3:10728. Bresnan, Joan. 1971. Sentence stress and syntactic transformation. Language 47:257-281. Bresnan, Joan. 1974. The position of certain clause-particles in phrase structure. Linguistic Inquiry 5: 64-619. Bruening, Benjamin. 2001. QR obeys superiority: Frozen scope and ACD. Linguistic Inquiry 32: 233-273.

303 Cann, Ronnie, Ruth Kempson, Lutz Marten, and Masayuki Otsuka. 2005. Right Node Raising, coordination and the dynamics of language processing. Lingua 115: 503–525 Carlson, Greg. 1987. Same and different: Some consequences for syntax and semantics. Linguistics and Philosophy 10:531-565. Chalcraft, Faye. 2006. Right Node Raising as ellipsis: evidence from (what the) British do. Snippets 12: 7-8. Chao, Wynn. 1987. On ellipsis. Ph.D. dissertation, University of Massachusetts-Amherst. Cheng, Lisa. 1991. On the typology of wh-questions. Ph.D. dissertation, MIT, Cambridge, MA. Cho, Sungeun, and Xuan Zhou. 1999. The interpretation of wh-elements in conjoined whquestions. Ms., State University of New York, Stony Brook. Choe, Hyun-Sook. 1988. Restructuring parameters and complex predicates: Transformational approach. Ph.D. dissertation, MIT, Cambridge, MA. Chomsky, Noam. 1976. Conditions on rules of grammar. Linguistic Analysis 2: 303-351. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. 1982. Some Concept and Consequences of the Theory of Government and Binding. Cambridge, MA: MIT Press. Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press Chomsky, Noam. 1995. Minimalist Program. Cambridge, MA: The MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale. A Life in Language, 1-52. Cambridge, MA: MIT Press. Chomsky, Noam. 2005. On phases. Ms., MIT. Chung, Dae-Ho. 2004. Multiple dominance analysis of Right Node Raising constructions. Language Research 40: 791-812.

304 Chung, Sandra. 2005. Sluicing and the lexicon: The point of no return. To appear in Proceedings of the Berkeley Linguistics Society. Citko, Barbara. 2003. ATB wh-questions and the nature of Merge. In Makoto Kadowaki, and Shigeto Kawahara (eds.), Proceedings of the 33th North East Linguistic Society, GLSA Publications, Umass, Amherst. Citko, Barbara. 2005. On the nature of merge: External merge, internal merge, and parallel merge. Linguistic Inquiry 36: 475-497. Citko, Barbara. 2006. On the interaction between Across-The-Board wh-movement and left branching extraction. Syntax 9: 225-247. Contreras, Heles. 1984. A note on parasitic gaps. Linguistic Inquiry 15: 704-713 Coppock, Elizabeth. 2001. Gapping: In defense of deletion. In Mary Andronis, Christoper Ball, Heidi Elston, and Sylvain Neuvel (eds.), The Thirty-Seventh Annual Meeting of the Chicago Linguistic Society, 133–148. University of Chicago. Culicover, Peter. 1976. Syntax. Academic Press, New York. Dryer, Matthew. 1987. On primary objects, secondary objects and antidative. Language 62: 808845. Engdahl, Elizabeth. 1983. Parasitic gaps. Linguistics and Philosophy 6: 5–34. Erteschik-Shir, Nomi. 1987. Right Node Raising. MIT Working Papers in Linguistics 9: 105-117. Evans, Frederic. 1988. Binding into anaphoric verb phrases. In Joyce Powers and Kenneth de Jong (eds.), Proceedings of the Fifth Eastern States Conference on Linguistics, 122–129. Columbus: Ohio State University. Féry, Caroline, and Katharina Hartmann. 2005. The focus and prosodic structure of German Right Node Raising and gapping. The Linguistic Review 22: 69-116. Fiengo, Robert, and Robert May. 1994. Indices and Identity. Cambridge, MA: MIT Press.

305 Fillmore, Charles J. 1965. Indirect Object Constructions in English and the Ordering of Transformations. The Hague, Mouton. Fox, Danny. 2000. Economy and Scope Interaction. Cambridge, MA: MIT Press. Fox, Danny and David Pesetsky. 2005. Cyclic linearization of syntactic structure. Theoretical Linguistics 31: 1-45. Franks, Steven. 1992. A prominence constraint on null operator constructions. Lingua 87:35-54. Gawron, Jean Mark, and Andrew Kehler. 2004. The semantics of respective readings, conjunction, and filler gap dependencies. Linguistics and Philosophy 27: 169-207. Gazdar, Gerald. 1981. Unbounded dependencies and coordinate structure. Linguistic Inquiry 12: 155-184. Gengel, Kirsten. 2006. Contrastivity and ellipsis. Paper presented at LingO 2006.

Giannakidou, Anastasia, and Jason Merchant. 1998. Reverse sluicing in English and Greek. Linguistic Review 15: 295-333. Gleitman, Lila. 1969. Coordinating conjunctions in English. In D. Reibel and S. Shane. (eds.), Modern Studies in English: Readings in Transformational Grammar, 80-112. Englewood Cliffs, NJ: Prentice-Hall. Goldsmith, John. 1985. A principled exception to the coordinate structure constraint. In Papers from the Twenty-First Annual Regional Meeting of the Chicago Linguistic Society, 133143. Chicago Linguistic Society, Chicago. Goodall, Grant. 1983. A three-dimensional analysis of coordination. In Papers from the nineteenth Annual Regional Meeting of the Chicago Linguistic Society, 146-154. Chicago Linguistic Society, Chicago. Goodall, Grant. 1987. Parallel structures in syntax: Coordination, Causatives, and Restructuring. London: Cambridge University Press.

306 Grosu, Alexander. 1973. On the nonunitary nature of the coordinate structure constraint. Linguistic Inquiry 4: 88–92. Grosu, Alexander. 1976. A note on subject raising and object to Right Node Raising. Linguistic Inquiry 7: 642-45. Ha, Seungwan. 2006. Contrastive focus: Licensor for Right Node Raising. To appear in Proceedings of the 37th North East Linguistic Society. GLSA. Amherst, Umass. Ha, Seungwan. 2007. On ellipsis features and Right Node Raising. To appear in Proceedings of the 15th annual Conference of the Student Organization of Linguistics in Europe. University of Leiden. Haïk, Isabelle. 1985. The syntax of Operators. Doctoral dissertation, MIT, Cambridge, MA. Halle, Morris and Alec Marantz. 1993. Distributive morphology and the pieces of inflection. In Ken Hale and Samuel J. Keyser (eds.) The View from Building 20. Cambridge, MA: MIT Press Hankamer, Jorge. 1971. Constraints on deletion in syntax. Doctoral dissertation, Yale University, New Haven, CT. Hankamer, Jorge and Ivan Sag. 1976. Deep and surface anaphora. Linguistic Inquiry 7: 391-428. Hardt, Daniel. 1992. VP ellipsis and semantic identity. In Chris Barker and David Dowty (eds.), Proceedings of SALT II, 145–161. Columbus: Ohio State University, Working Papers in Linguistics. Hardt, Daniel. 1993. VP ellipsis: Form, meaning, and processing. Ph.D. Dissertation, University of Pennsylvania. Hartmann, Katharina. 2000. Right Node Raising and Gapping: Interface Conditions on Prosodic Deletion. John Benjamins: Amsterdam.

307 Hestvik, Arild. 1992. Strict reflexives and the subordination effect. In Steven Berman and Arild Hestvik (eds.), Proceedings of the Stuttgart Workshop on Ellipsis, 1-50. University of Stuttgart. Hiraiwa, Ken. 2005. Dimensions of symmetry in syntax: Agreement and clausal architecture. Ph.D. dissertation, MIT, Cambridge, MA. Hirschbuhler, Paul. 1982. VP-deletion and across-the-board quantifier scope. In James Pustejovsky and Peter Sells (eds.), Proceedings of the 12th North East Linguistic Society, 132-139. GLSA, University of Massachusetts, Amherst. Höhle, T. N. 1991. Über Verum-Focus im Deutschen. In J. Jacobs, (ed.), Informationsstruktur und Grammatik, Linguistische Berichte, Sonderheft 4/1991-92, Westdeutscher Verlag, Opladen. Holliday, Michael. 1967. Notes on transitivity and theme in English (Part 2). Journal of Linguistics 3: 199-244. Holmberg, Anders. 1999. Remarks on ‘Holmberg’s generalization’. Studia Linguistica 53: 1-39. Hornstein, Nobert and Jairo Nunes. 2002. On Asymmetries between parasitic gap and across-theboard constructions. Syntax 5: 26-54. Hudson, Richard. 1976. Conjunction reduction, gapping, and Right Node Raising. Language 52: 535-562. Jackendoff, Ray. 1977. X' syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press. Jacobson, Pauline, 1982. Evidence for gaps. In Pauline Jacobson and Geoffrey K. Pullum (eds.), The Nature of Syntactic Representations, 187–228. Dordrecht. Jacobson, Pauline. 1999. Towards a variable free semantics, Linguistics and Philosophy 22: 117184.

308 Jayaseelan, Karattuparambil. 2001. IP-internal topic and focus phrases. Studia Linguistica 55: 3975. Joh, Yoonkyung. 2005. The semantic analysis of the Korean plural marker tul. In Emar Maier, Corien Bary and Janneke Huitink (eds), Proceedings of Sinn und Bedeutung 9. Johnson, Kyle. 1986. A case from movement. Ph.D. dissertation, MIT, Cambridge, MA. Johnson, Kyle. 1996a. When VP go missing. Glot International 2: 2-9. Johnson, Kyle. 1996b/2003. In search of the English middle field. Ms., University of Massachusetts, Amherst. Johnson, Kyle. 2001. What VP ellipsis can do, what it can’t, but not why. In Mark Baltin and Chris Collins (eds.), The Handbook of Contemporary Syntactic Theory, 439-479. Malden, MA: Blackwell Publishers. Johnson, Kyle. 2007. LCA+alignment=RNR. Paper presented at Workshop on Coordination, Subordination and Ellipsis, University of Tübingen, June 2007. Kaiser, Elsi, Jeffrey Runner, Rachel Sussman, and Michael Tanenhaus. 2004. Susceptibility to discourse/semantic factors: An experimental approach to short-distance pronouns and logophoric reflexives. Paper presented at Workshop on Semantic Approaches to Binding Theory, ESSLLI, Nancy. Kato, Takaomi. 2006. Symmetries in coordination. Ph.D. dissertation, Harvard University. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Kehler, Andrew. 1996. Coherence and the Coordinate Structure Constraint. In Proceedings of the 22nd Annual Meeting of the Berkeley Linguistics Society, 220-231. Berkeley Linguistics Society, University of California, Berkeley. Kehler, Andrew. 2000. Coherence and the resolution of ellipsis. Linguistics and Philosophy 23: 533-575.

309 Kehler, Andrew. 2002. Coherence, Reference, and the Theory of Grammar. CSLI Publications. Kennedy, Chris. 2000. Comparative (sub)deletion and ranked, violable constraints in syntax. In Proceedings of the 30th North East Linguistic Society, GLSA Publications, Amherst. Kennedy, Chris. 2002. Comparative deletion and optimality in syntax. Natural Language and Linguistic Theory 20: 553-621 Kennedy, Chris and Jason Merchant. 2000. Attributive comparative deletion. Natural Language and Linguistic Theory 18: 89-146. Kim, Chonghyuk. 2004. On the syntax and semantics of Korean non-nominal plural marker tul. Ms., University of Delaware. Kim, Soowon. 1999. Sloppy/strict identity, empty objects, and NP ellipsis. Journal of East Asian Linguistics 8: 255-284. Klima, Edward. 1964. Negation in English. In Jerry A. Fodor and Jerrrold J. Katz (eds.), The Structure of Language, Englewood Cliffs, NJ: Prentice Hall. Ko, Heejeong. 2007. Asymmetries in scrambling and cyclic linearization. Linguistic Inquiry 38: 49-83. Koopman, Hilda and Tim Sportiche. 1983. Variables and the bijection principle. The Linguistic Review 2: 139-160. Kratzer, Angelika. 1991. The representation of focus. In Arnim von Stechow and Dieter Wunderlich (eds.), Semantics. An International Handbook of Contemporary Research, de Gruyter, Berlin/ New York, 831-834. Kratzer, Angelika and Junko Shimoyama. 2002. Indeterminate pronouns: The view from Japanese. Paper presented at the 3rd Tokyo Conference on Psycholinguistics. Kuno, Susumu. 1987. Functional Syntax: Anaphora, Discourse and Empathy. Chicago.

310 Laka, Itziar 1990. Negation in syntax: On the nature of functional categories and projections. Doctoral Dissertation, MIT, Cambridge, MA. Lakoff, George. 1986. Frame semantic control of the Coordinate Structure Constraint. In Papers from Pragmatics and Grammatical Theory, 152-167. Chicago: Chicago Linguistic Society. Langacker, Ronald. 1969. On pronominalization and the chain of command. In David A. Reibel and Sanford A. Schane (eds.) Modern Studies in English. New Jersey: Prentice-Hall, 160186. Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19: 335-391. Larson, Richard. 1990. Double objects revisited: Reply to Jackendoff. Linguistic Inquiry 21: 589632. Larson, Richard, and Robert May. 1990. Antecedent containment or vacuous movement: reply to Baltin. Linguistic Inquiry 21: 103-122. Lasnik, Haward. 1999. Verbal morphology: Syntactic structures meets the Minimalist Program. In Haward Lasnik (ed.), Minimalist Analysis, 97-119. Malden, MA: Blackwell Publishers. Lebeaux, David. 1988. Language acquisition and the form of the grammar. Ph.D. dissertation, University of Massachusetts, Amherst. Levine, Robert. 1985. Right Node (non-)Raising. Linguistic Inquiry 16: 492-497. Lightfoot, David. 1999. The Development of Language: Acquisition, Change, and Evolution. Malden, MA: Blackwell Publishers. Lin, Vivian. 2001. A way to undo A-movement. In Proceedings of the 20th West Coast Conference on Formal Linguistics, 358-371. Somerville, MA: Cascadilla Press. Lin, Vivian. 2002. Coordination and sharing at the interfaces. Doctoral dissertation, MIT, Cambridge, MA.

311 Lobeck, Anne. 1995. On Ellipsis: Functional Heads, Licensing, and Identification. Oxford Uiversity Press. Marantz, Alec. 1993. Implications of asymmetries in double object constructions. In Sam A. Mchombo (ed.), Theoretical Aspects of Bantu Grammar, 113-150. Stanford, CA: CSLI Publication. Marling, Joan. 1972. On gapping and the order of constituents. Linguistic Inquiry 1: 101-108. May, Robert. 1977. The grammar of quantification. Doctoral dissertation, MIT, Cambridge, MA. May, Robert. 1985. Logical Form: Its Structure and Derivation. Cambridge, MA: MIT Press. McCawley, James. 1982. Parentheticals and discontinuous constituent structure. Linguistic Inquiry 13: 91-106. McCawley, James. 1988. The Syntactic Phenomena of English. Chicago: The University of Chicago Press. McCloskey, James. 1986. Right node raising and preposition stranding. Linguistic Inquiry 17: 183-186. McCloskey, James. 1999. On the right edge in Irish. Syntax 2: 189-209. Merchant, Jason. 2001. The Syntax of Silence. Oxford: Oxford University Press. Merchant, Jason. 2002. A WCO and ACD puzzle. Ms. University of Chicago. Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and Philosophy 27: 661 – 738. Merchant, Jason. 2006. An asymmetry in voice mismatches in VP ellipsis and pseudogapping. Ms. University of Chicago. Merchant, Jason. To appear. Variable island repair under ellipsis. In Kyle Johnson (ed.), Topics in Ellipsis, Cambridge University Press: Cambridge. Moltmann, Friederike. 1992. Coordination and comparatives. Doctoral dissertation, MIT, Cambridge, MA.

312 Mukai, Emi. 2003. On verbless conjunction in Japanese. In Makoto Kadowaki and Shigeto Kawahara (eds.), Proceedings of the 33rd North East Linguistic Society. 205-224. GLSA. Amherst, Umass. Munn, Alan. 1992. A null operator analysis of ATB gaps. The Linguistic Review 9: 1-26. Munn, Alan. 1993. Topics in the syntax and semantics of coordinate structures. Doctoral dissertation, University of Maryland, College Park. Munn, Alan. 1999a. First conjunct agreement: Against a clausal analysis. Linguistic Inquiry 30: 643-668. Munn, Alan. 1999b. On the identity requirement of ATB movement. Natural Language Semantics 7: 421-425. Munn, Alan. 2001. Explaining parasitic gap restrictions. In Peter Culicover, and Paul Postal (eds.), Parasitic gaps, 369-392. Cambridge, MA: MIT Press. Nissenbaum, Jon. 2000. Investigations of covert phrasal movement, Ph.D. dissertation, MIT, Cambridge, MA. Nunes, Jairo. 1995. The copy theory of movement and linearization of chains in the Minimalist Program. Ph.D. dissertation, University of Maryland, College Park. Nunes, Jairo. 2001. Sideward movement. Linguistic Inquiry 32: 303-344. Nunes, Jairo. 2004. Linearization of Chains and Sideward Movement. Cambridge, MA: MIT Press. Otani, Kazuyo and John Whitman. 1991. V-raising and VP-ellipsis. Linguistic Inquiry 22: 345358. Pesetsky, David. 1982. Paths and categories. Ph.D. dissertation, MIT, Cambridge, MA.

313 Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In Eric J. Reuland and Alice G. B. ter Meulen (eds.), The Representation of (in)definiteness, 98-129. Cambridge, Mass: MIT Press. Pesetsky, David. 1995. Zero syntax: Experiencers and cascades. Cambridge, MA: MIT Press. Pesetsky, David. 2000. Phrasal Movement and its Kin. Cambridge, MA: MIT Press. Phillips, Colin. 1996. Order and structure. Ph.D. dissertation. Cambridge, MA: MIT. Phillips, Colin. 2003. Linear order and constituency. Linguistic Inquiry 34: 37-90. Pierrehumbert, Janet. 1980. The phonology and phonetics of English intonation. Ph.D. dissertation. Cambridge, MA: MIT. Pollard, Carl, and Ivan Sag. 1992. Anaphors in English and the scope of Binding Theory. Linguistic Inquiry 23: 261-303. Postal, Paul. 1971. Cross-over Phenomena. New York: Holt, Reinhart and Winston. Postal, Paul. 1974. On Raising. Cambridge, MA: The MIT Press. Postal, Paul. 1993. Parasitic gaps and across-the-board phenomenon. Linguistic Inquiry 24: 735754. Postal, Paul. 1998. Three Investigations of Extraction. Cambridge, MA: The MIT Press. Reinhart, Tanya. 1983. Pragmatics and Linguistics: An Analysis of Sentence Topics. Reproduced by the Indiana University Linguistics Club. Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24: 657-720. Rizzi, Luigi. 1997. Locality and left-periphery. In Liliane Haegeman (ed.), Elements of Grammar, 281-337. Dordrecht: Kluwer. Rochemont, Michael. 1992. Bounding rightward A’-dependencies. In Helen Goodluck and Michael Rochemont (eds.), Island Constraints, 373-397. Dordrecht: Kluwer.

314 Romero, Maribel, and Chung-hye Han. 2004. On negative yes/no questions. Linguistics and Philosophy 27: 609-658 Rooth, Mats. 1985. Association with focus. Ph. D. dissertation. University of Massachusetts at Amherst. Rooth, Mats. 1992a. A theory of focus interpretation. Natural Language Semantics 1: 75-116. Rooth, Mats. 1992b. Ellipsis redundancy and reduction redundancy. In B. Steve and A. Hestvik (eds.), Proceedings of the Stuttgart Ellipsis Workshop. Report 29 of the Series “Sprachtheretische Grundlagen fuer die computerlinguistik”, IBM Heidelberg. Ross, John. 1967. Constraints on variables in syntax. Doctoral dissertation, MIT, Cambridge, MA. Ruys, Eddy. 1993. The scope of indefinites. Doctoral dissertation, Universiteit Utrecht. Sabbagh, Joseph. 2003. Ordering and linearizing tightward movement. In Proceedings of the 22nd West Coast Conference on Formal Linguistics, 436-449. Somerville, MA: Cascadila Press. Sabbagh, Joseph. 2007. Ordering and linearizing rightward movement. To appear in Natural Language and Linguistic Theory. Sabbagh, Joseph. to appear. Right Node Raising and extraction in Tagalog. Linguistic Inquiry. Sag, Ivan. 1976. Deletion and logical form. Doctoral dissertation, MIT, Cambridge, MA. Sauerland, Uli. 1998. The meaning of chains. Doctoral dissertation, MIT, Cambridge, MA. Sauerland, Uli. 2000. Syntactic economy and quantifier raising. Ms., University of Tubingen. Sauerland, Uli, and Paul Elbourne. 2002. Total reconstruction, PF-movement, and derivational order. Linguistic Inquiry 33: 283-319 Schwarzschild, Roger. 1999. Givenness, AvoidF and other constraints on the placement of accent, Natural Language Semantics 7: 141-177.

315 Selkirk, Lisa. 1995. Sentence prosody: Intonation, stress, and phrasing. In John Goldsmith (ed.), Handbook of Phonological Theory, 550–569. Oxford: Blackwell. von Stechow, Arnim. 1989. Focusing and backgrounding operators. Technical report 6, Fachgruppe Sprachwissenschaft, Universitat Konstanz. Svenonius, Peter. 2004. On the edge. In David Adger, Cécile De Cat, and George Tsoulas (eds.), Peripheries: Syntactic Edges and their Effects. Dordrecht: Kluwer Academic Publishers. Swingle, Kari. 1993. The role of prosody in Right Node Raising. Ms., University of California at Santa Cruz. Takahashi, Shiochi and Danny Fox. 2005. MaxElide and the re-binding problem. To appear in Proceedings of Semantics and Linguistics Theory 15. Takano, Yuji. 1998. Object shift and scrambling. Natural Language and Linguistic Theory 17: 817889. Thompson, Ellen. 1994. A minimalist approach to the syntax of temporal adverbs. Maryland Working Papers in Linguistics 2: 221-237. Thompson, Ellen. 1995. The structure of tense and the syntax of temporal adverbs. In Proceedings of the 13th West Coast Conference on Formal Linguistics, 499-513. Somerville, MA: Cascadilla Press. Uriagereka, Juan. 1999. Multiple Spell-Out. In Samuel Epstein and Norbert Hornstein (eds.) Working Minimalism, 251-282. Cambridge, MA: MIT Press. de Vos, Mark and Luis Vicente. 2005. Coordination under Right Node Raising. In Proceedings of the 24th West Coast Conference on Formal Linguistics, 97-104. Somerville, MA: Cascadilla Press. Vries, Mark de. 2007. Internal and external remerge: on movement, multidominance, and the linearization of syntactic object. Ms. University of Groningen.

316 Warner, Anthony. 1986. Ellipsis conditions and the status of the English copula. York Papers in Linguistics 12: 153-172. University of York, Heslington. Wasow, Thomas. 1972. Anaphoric relations in English. Doctoral Dissertation, MIT, Cambridge, MA. Wasow, Thomas, and Jennifer Arnord. Intuitions in linguistic argumentation. Lingua 115: 14811496. Wilder, Chris. 1997. Some properties of ellipsis in coordination. In Artemis Alexiadou and Alan Hall (eds.), Studies on universal grammar and typological variation, 59-107. Amsterdam/Philadelphia: John Benjamins Publishing Company. Wilder, Chris. 1999. Right Node Raising and the LCA. In Proceedings of the 18th West Coast Conference in Formal Linguistics, 586-598. Somerville, MA: Cascadilla Press. Williams, Edwin. 1977. Discourse and logical form. Linguistic Inquiry 8: 101–139. Williams, Edwin.1978. Across-the-board rule application. Linguistic Inquiry 9: 31-43. Williams, Edwin.1989/1990. The ATB theory of parasitic gaps. The Linguistic Review 6: 265-279. Williams, Edwin. 1997. Blocking Anaphora. Linguistic Inquiry 28: 577-628. Winkler, Susanne. 2000. Silent Copy and Polarity Focus in VP Ellipsis. In Kerstin Schwabe and Ning Zhang (eds.), Ellipsis in Coordination, 221-247. Niemeyer, Tübingen. Winkler, Susanne. 2005. Ellipsis and Focus in Generative Grammar. Berlin: Mouton De Gruyter. Wexler, Ken and Peter Culicover. 1980. Formal Principles of Language Acquisition. Cambridge, MA: MIT Press. Yim, Changkook. 2003. The semantics of non-nominal tul in Korean. Ms. Cornell University. Yoon, James. 1994. Korean Inflection and Checking Theory. In Colin Phillips and Heidi Harley (eds.), MIT Working Papers in Linguistics, 251-270. Department of Linguistics, MIT.

317 Yoon, James, and Wooseung Lee. 2005. Conjunction reduction and its consequences for noun phrase morphosyntax in Korean. In Proceedings of the 24th West Coast Conference on Formal Linguistics, 379-387. Somerville, MA: Cascadilla Proceedings Project. Wu, Jianxin. 1999. A minimalist account of quantification in Chinese. Doctoral dissertation, University of Maryland, College Park. Zoerner, Edward. 1995. Coordination: The syntax of &P. Ph.D. dissertation. University of California, Irvine.

BOSTON UNIVERSITY GRADUATE SCHOOL OF ...

grammar for my conference abstracts, term papers, manuscripts, and this dissertation, ...... For example, in (21), the antecedent of the elided VP go to the ball.

1MB Sizes 4 Downloads 190 Views

Recommend Documents

The University of Tokyo Graduate School of Information ...
sources. Semantic annotations represent information contained in text documents in a structured format which are more amenable to applications in data mining,.

Manipal University Welcomgroup Graduate School ... -
131401148 Karthik S Ballal. 3. 4. 9. 16. 55. 11. 66. 13. 40. 45. 131401150 Sanket Raj. 3. 4. 9. 16. 59. 12. 68. 14. 42. 46. 131401152 Rahul Samson Rebello. 3. 4.

Education Stanford University, Graduate School of ...
ALEXANDER A. NEZLOBIN. Haas School of Business,. University of California, Berkeley. Berkeley, CA 94720. Tel: +1 (650) 862-8875. E-mail: [email protected] sites.google.com/site/alexanderanezlobin/. Education. Stanford University, Graduate Sc

boston university
opadhyay, Udit Raha, Anupam Mukherjee, K. V. M. Naidu, and many more. ...... is usually used instead of multiple access when multiple (virtual) sources share a.

BOSTON UNIVERSITY COLLEGE OF ENGINEERING ...
then put back together on a remote host using recent graph-theoretic techniques. We present analyses ... gossip protocols and content delivery networks. We provide .... 2.5 CPIsync vs. slow sync, fixed number of differences . . . . . . . . . . . 29.

BOSTON UNIVERSITY COLLEGE OF ENGINEERING ...
68. 4.5 Average delay: Comparison between the oracle and the real modes . . 70. 4.6 CCDF: Comparison between the oracle and the real modes at load=20. 70.

Summer 2010 Information - Boston University
featured on the BSO website in a short film about a piece written ... further my own technique and musicality. My future plans are not completely clear to me, but.

BOSTON UNIVERSITY COLLEGE OF ENGINEERING ...
BOSTON UNIVERSITY. COLLEGE .... 4.2 Mask-length and maximum graph node degree . . . . . . . . . . . . . 70 ...... data is converted to data readable by our Palm program using PRC-tools [49]. It is ...... 22, (Copper Mountain Resort, Colorado), pp.

Summer 2010 Information - Boston University
include students up to age 20 to increase options available to college ... Boston Symphony Orchestra (BSO) for 38 years .... in California with her husband and.

Boston University - WhatIsCS.pdf
Whoops! There was a problem loading more pages. Retrying... Whoops! There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. Boston University - WhatIsCS.pdf. Boston

Welcomgroup Graduate School of Hotel ... -
ATTENDANCE PERCENTAGES TILL FEBRUARY 20, 2018. FIRST SEMESTER BHM SECTION A ... 57.89. 2. 171401002 MOHAMMED HARSHAD BHAVA. 19. 19. 100.00. 3. 171401005 SIDDANTH RAINA. 19. 17. 89.47. 4. 171401008 AJAY JOSEPH JAIN. 19. 18. 94.74. 5. 171401012 LAVANYA

Stanford Business - Stanford Graduate School of Business
1. I'm excited to write about reinvention because it is a process I think about often ..... School of Business for 12 years. The class ... at trade shows to lend a hand.

The Marion F. Gislason Award - Boston University
For Excellence in the Field of Leadership Development ... EDRT is a dynamic, peer-based learning forum and research center open to all organizations that view.

UNIVERSITY POST-GRADUATE DIPLOMA- 2nd ... -
X CODE. DENOMINATION OF THE MASTER'S COURSE. LEVEL. 1105. INTERNATIONAL HEALTH AND MEDICINE FOR COOPERATION WITH. DEVELOPING COUNTRIES. II. To that end, I declare to be in possession of the following requisites for the participation in the above ment