Ellipsis and Focus in Generative Grammar 9783110890426, 9783110186017

Citation preview

Ellipsis and Focus in Generative Grammar

w DE

G

Studies in Generative Grammar 81

Editors

Henk van Riemsdijk Harry van der Hulst Jan Köster

Mouton de Gruyter Berlin · New York

Ellipsis and Focus in Generative Grammar

by

Susanne Winkler

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter G m b H & Co. K G , Berlin.

The series Studies in Generative G r a m m a r was formerly published by Foris Publications Holland.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication

Data

Winkler, Susanne, 1 9 6 0 Ellipsis and focus in generative grammar / by Susanne Winkler. p. cm. — (Studies in generative grammar ; 81) Includes bibliographical references and index. ISBN-13: 978-3-11-018601-7 (cloth : alk. paper) ISBN-10: 3-11-018601-2 (cloth : alk. paper) 1. Grammar, Comparative and general — Ellipsis. 2. Focus (Linguistics) 3. Generative grammar. I. Title. II. Series. P291.3.W55 2005 415—dc22 2005031448

Bibliographic

information published by Die Deutsche

Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at < h t t p : / / d n b . d d b . d e > .

ISBN-13: 978-3-11-018601-7 ISBN-10: 3-11-018601-2 ISSN 0167-4331 © Copyright 2005 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. N o part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing f r o m the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Preface

This monograph presents the results of an in-depth syntactic and focustheoretical investigation of ellipsis in generative grammar. The syntactic analysis of ellipsis is couched in the Minimalist Program. The focus analysis addresses the question of the prosodic realization of elliptical constructions in relation to deaccentuation and deletion. This multidimensional account shows that ellipsis is an interface phenomenon which results from the complex interaction of the core grammatical components with the information structural component. The central hypothesis is that there are two types of ellipses in English, one sentence-bound and the other discourse-bound, and that their different syntactic derivations correlate with their specific information structural functions. This hypothesis is based on a revised model of grammar in which focus and information structure play a crucial role. Considering linguistic research on empty categories and focus from over more than three decades the book develops an account of ellipsis based on parallel computation, which is shown to be a natural consequence of the division of labor between the syntactic, the information structural and the interpretive components. Empirical evidence for this account comes from a detailed analysis of discourse-bound ellipsis, such as VP-ellipsis in English and its less well-known instantiations in German, and an intensive investigation of the syntax and information structure of gapping, a case of sentence-bound ellipsis. Both empirical analyses provide evidence for the claim that the information structural component functions as a relay station between syntax and the interpretive components on the one hand and between phonology and pragmatics on the other. The aim of this book is to explain on the basis of modern linguistic theory how it is possible that we understand more than we actually hear. The answer developed throughout this book is that ellipsis is an interface phenomenon which can only be explained on the basis of the complex interaction between syntax, semantics and information structure. This book is based on my 2003 University of Tübingen postdoctoral thesis. The topic of ellipsis and focus arose in connection with my DFGproject Ellipse und Informationsstruktur im Englischen of the Sonderforschungsbereich 441 Linguistische Datenstrukturen: Theoretische und empirische Grundlagen der Grammatikforschung at the University of

vi Preface Tübingen. The research program presented here has been developed as an integral part of the project work. First and foremost, I would like to thank my committee, Bernie Drubig, Kyle Johnson, Uwe Mönnich, Marga Reis and Wolfgang Sternefeld, for encouraging me to put my thoughts on ellipsis on paper and thus bring the period of postdoctoral studies to an end. A special thank you goes to Ewald Lang for his detailed and constructive comments on the original manuscript. The remaining mistakes are mine. I would further like to gratefully acknowledge the fruitful discussions on ellipsis, syntax and focus I have had with the following people: Bernie Drubig, Kirsten Gengel, Remus Gergel, Edward Göbbel, Carlos Gussenhoven, Dan Hardt, Jutta Hartmann, Katharina Hartmann, Klaus von Heusinger, Kyle Johnson, Chris Kennedy, Wolfgang Klein, Ekkehard König, Angelika Kratzer, Tony Kroch, Ewald Lang, Luis Lopez, Jason Merchant, Valeria Molnär, Jürgen Pafel, Ellen Prince, Marga Reis, Kerstin Schwabe, Lisa Selkirk, Arnim von Stechow, Mark Steedman, Wolfgang Sternefeld, Satoshi Tomioka, Michael Wagner, and Ellen Woolford. Special gratitude goes to Kirsten Brock, Ladi Bursik, Juliane Möck, and Tina Schäfer who started editing this study the moment the latest version of each chapter left the printer. I also thank the DFG for supporting the research presented in this book with a special grant (Habilitationsstipendium) from 2001 to 2002 and with support for the DFG-project since 2002. This book is dedicated to Jan, Ben and Ladi.

Contents

Prologue Chapter 1: Ellipsis and focus: An introduction 1. Introduction 2. The derivational model 3. The syntax-semantics interface 3.1. The cyclic derivation of surface semantic interpretation 3.2. The double-cycle information structure hypothesis 3.3. Information focus and contrastive focus: A derivational approach 3.3.1. Information focus 3.3.2. Contrastive focus 4. The hybrid focus analysis of ellipsis 5. Conclusion 6. Distinguishing sentence-bound and discourse-bound ellipsis: A preview Chapter 2: Ellipsis at the interfaces: A proposal 1. Introduction 2. Deletion vs. deaccentuation 2.1. Previous accounts 2.2. The parallel computation account 2.2.1. Accentuation and deaccentuation in autosegmental theory 2.2.2. Sentence-bound ellipsis 2.2.3. Discourse-bound ellipsis 3. The pragmatics-PF interface 3.1. Anaphoricity and the traditional notion of givenness 3.2. Schwarzschild's proposal: The notion of GIVENness 3.3. Objections 3.4. E-GIVENness and VP-/IP-ellipsis 3.5. Contrastivity and VP-ellipsis 4. The syntax-phonology interface 4.1. The derivation of intonation by phase 4.2. The derivation of ellipsis by phase 5. Conclusion

1 10 10 14 25 25 27 28 30 33 35 38 39

43 43 44 45 54 58 60 67 77 78 81 83 94 99 103 104 110 117

viii

Contents

Chapter 3: VP-anaphora and ellipsis of VP in English and German 1. Introduction 1.1. In search of VP-ellipsis in German 1.2. Roadmap 2. The representation of ellipsis 2.1. The proform hypothesis vs. the PF-deletion hypothesis 2.2. VP-ellipsis: Arguments for an empty proform account 2.3. German VP A patterns with English VPE 2.3.1. The data 2.3.2. Givenness and polarity focus: A proposal 2.3.3. Problems for the polarity focus account 2.4. Evidence from English for the derivational account of VPE 2.4.1. Syntactic arguments: Evidence from extraction 2.4.2. Evidence from the syntactic identity condition and the binding theory 3. German auch -ellipsis 3.1. Ellipsis of VP in German 3.2. Auch-ellipsis is a case of contrastive remnant ellipsis or stripping 3.2.1. Syntactic arguments: Parallel configuration and locality constraints 3.2.2. Information structure analysis of contrastive topic ellipsis 3.2.3. Contrastive remnant ellipsis: The paradigm 3.3. A sideward movement account of contrastive remnant ellipsis 3.3.1. Contrastive topic ellipsis 3.3.2. Contrastive focus ellipsis 3.4. Contrastive remnant ellipsis and the syntax-phonology interface 4. Conclusion

119 119 120 124 125 125 127 134 134 137 142 145 145 149

Chapter 4: Gapping: A sideward movement account 1. Introduction 2. The syntax of gapping 2.1. Deletion vs. ATB-movement accounts 2.2. The proposal: Gapping as sideward movement of vP 3. Evidence for the sideward movement account 3.1. The information structure of gapping: Paired contrastive remnants

181 181 183 183 186 191 191

151 151 153 154 159 163 166 169 174 177 179

Contents

3.2. 3.3. 3.4. 3.4.1. 3.4.2. 3.4.3. 4.

Scope facts as evidence for vP coordination Intonational evidence for vP coordination Evidence for an A'-position in vP Wh-phrases at the edge of vP A'-position in vP hosts topicalized phrase Evidence for a focus position at the edge of vP in German Conclusion

ix

195 196 207 208 209 212 217

Chapter 5: Conclusion and final remarks 1. The informational structural proposal for ellipsis 2. Challenges for further research 3. Closing remark: Or a note on Wittgenstein's problem

218 218 227 228

Notes References Subject Index Author Index

230 247 286 291

Prologue

"In any field find the strangest thing and then explore it." - John Archibald Wheeler

This study is concerned with a puzzle of the human language system that Robert D. Ladd described more than 20 years ago with the following light and shadow metaphor: "If we shine the spotlight on one actor, everything else on stage is in shadow in comparison" (Ladd 1979: 111). This metaphor expresses the idea that prosodic phenomena in language can be better understood in terms of the concepts of visual perception: "perhaps accents go where they do both in order to highlight what they are on and to cast in shade what they are not on" (Dwight Bolinger quoted by Ladd 1979: 110). The puzzle of prosodic highlighting and backgrounding still has not been solved, let alone the relation between highlighting and the degrees of backgrounding - from prosodic flatness up to the point of omitting prosodic realization altogether. Therefore, let me start out by investigating the visual metaphor more closely and see how it pertains to the topic of this book.. Now You See It - Now You Don't. Rubin's well-known demonstration of the visual reversal of figure and ground in figure 1 allows us to perceive either a vase or two faces looking at each other.1 Our perception changes depending on what is highlighted and thereby moved into focus. If grey models the background, and we focus on black, the vase will be perceived as salient. If the background is black, and we focus on grey, the faces become prominent.

2

Prologue

Figure 1. (taken from Rubin 1915)

If the grey background is omitted, as in figure 2, the vase is the only thing we see.

ι

Figure 2. (adopted from Rubin 1915)

Prologue

3

Sometimes, our brain must supply missing information, as in figure 3 below. Spot, a Dalmatian, comes alive if we reconstruct the missing parts of the picture.

L JM/

Figure 3. (taken from Simon 1998: 31)

If Spot appears on a background as in figure 4, the identification is much easier.

*

Figure 4. (background added)

^

4

Prologue

Note, however, that if the black spots are moved further apart or if some of them are removed altogether, it becomes more and more difficult to fill in the missing information and Spot cannot be recognized.

Figure 5. (changes added) Add a background of black and white dots, and perceptual recognition again becomes more difficult, as illustrated in figure 6. If you see a dog walking away from you, then you succeeded in organizing a mass of black and white shapes into Spot.

Figure 6. (taken from Goldstein 1996: 183, fig. 5.13)

Prologue

5

Intuitively, the optical illusions presented above are qualitatively different from each other: Figure 1 shows that visual perception is sensitive to foregrounding via focusing and backgrounding via defocusing; that is, the interpretation of figure 1 changes depending on what we select as highlighted or focused, and what as background. Omitting the background, as in figure 2, not only causes the vase to be perceived as salient, but removes the ambiguity almost completely. Figures 3 and 4 show that the process of image interpretation can operate on incomplete information. Missing information can be reconstructed up to a certain point (see figure 5). Additional information, such as adding a homogeneous background, as in figure 4, can facilitate the search. Adding a heterogeneous background, as in figure 6, makes the search more difficult (because no particular part of the picture is highlighted). Three initial observations seem relevant for the metaphoric extension of the visual system to the prosodic phenomena in language:2 1.

2. 3.

Visual perception is sensitive to different mechanisms of focusing and defocusing/backgrounding, highlighting and putting into shade. Reconstruction of missing visual information is rule governed. Reconstruction of missing information is dependent on the mechanisms employed in focusing and backgrounding. For example, a previously focused figure can be reconstructed more easily. Also, a figure with a homogeneous background can be reconstructed more easily than one with a heterogeneous background.

Now You Hear It - Now You Don't. The claim implicit in the light and shadow metaphor is that similar observations hold for the prosodic phenomena in language. There are various ways of foregrounding and backgrounding in language. One immediately accessible way is accenting one word and reducing another, as described by Bolinger's quote in the opening passage above. This quote, in addition to establishing a metaphoric relationship between the visual system and prosody, proposes that the function of accents in language is two-fold: their presence has the function of foregrounding one part of the utterance, while their absence has the function of backgrounding the other part. The discourse in (1) shows that this implied complementarity of foregrounding and backgrounding is

6

Prologue

not complete. (Capitalization signals foregrounding/focusing via high pitch accent assignment, and italics signal backgrounding via deaccentuation. (1)

(Cookie Monster to Bert from behind a Venetian blind:) a. Someone is going to eat a COOKIE. And you must GUESS exactly WHO. b. Now you hear ERNIE eating a cookie. c. Now you HEAR it. d. Now you DON'T.

Although the instructions of the game in (la) are presumably completely new to Bert, a single accent on cookie is required in the opening statement. In the coordinate sentence, there are two accents, one on the verbal head of the embedding sentence guess and another one on the wh-word who. In the first sample in (lb), we learn that ERNIE is doing the eating. This information is new and focused, while the rest of the sentence can go without pronounced pitch accents. The immediate intuition is that the phrase eating a cookie is uttered with a low flat intonation because it is repeated and therefore backgrounded. However, even repeated elements like hear can be accented, as in (lc), if what follows cannot be accented, as in the case of the pronoun it that refers to the process of Ernie eating a cookie. (Id) shows that the complete deletion of hear it brings out the meaning that what we hear is silence. Let us assume that the game "Identify the cookie-eater by the munching sound" continues with the following statements: (1)

(Continuation after (1 d)) e. Now you hear ME eating a cookie. f. Now you HEAR it. g. Now you DON'T.

The statement in (le) is parallel in structure to (lb): the subject of the eating process, in this case ME, carries a pitch accent, while the verb phrase eating a cookie is deaccented. Although (lb) and (le) differ only in the instantiation of the subject, the role within the game is different. While in (lb) the agent of the eating process, namely Ernie, is merely identified, in (Id), the information is that in contrast to Ernie, now Cookie Monster is eating the cookie. Note that although we can assume by now that the game is about cookie-eating, the instructions would be imprecise if not

Prologue

7

misleading, if (le) did not repeat eating a cookie. The idea of the game described in (1) requires Bert to identify the respective cookie-eater by the sound he makes eating a cookie.3 As in the case of image interpretation the language system uses foregrounding and backgrounding for interpretation. And as in the optical illusions looked at above, the language game in (1) also shows that there is not just one way of foregrounding and backgrounding; the function and the means of foregrounding and backgrounding can differ. In addition to accentuation and deaccentuation, which might be thought of as complementary concepts, there are focus movement, pronominal ization and deletion. All of these are not static but dynamic notions as implied by Ladd's light and shadow metaphor in the context of a stage play. They are highly discourse dependent and they play a particular function at each point in time in the language game. The comparison of examples (Id) and (le) also shows that, as in the case of image interpretation, the language system has the ability to reconstruct missing information up to a certain limit. An intuitively accessible initial hypothesis is given in (2): (2)

Initial hypothesis: Whatever is backgrounded via deaccentuation (spoken without audible pitch accents) in a sentence can be deleted.

The examples in (3) show that this preliminary hypothesis is not correct (deletion is signaled by strike-through): (3)

a. b. c. d. e. f. g.

Now you hear ERNIE eating a COOKIE. Now you DON'T hear Ernie eating a cookie. *Now you DON'T hear Ernie eating a cookie. #Now you DON'T hear Ernie eating a cookie. #Now you DON'T hear Ernie eating a cookie. *Now you DON'T hear Ernie eating a cookie. Now you DON'T hear Ernie eating a cookie.

(3a, b) establish that deaccentuation of the string hear Ernie eating a cookie is possible. If we cut into the sentence from the end, it becomes obvious that other constraints are at play, too. (3c) shows that if we delete the last unaccented word the sentence becomes highly ungrammatical. The sentence turns into a grammatical string if we delete the complete DP, as in

8

Prologue

(3d), but the meaning has changed. If we delete another word, the sentence remains grammatical but again means something different: what is communicated is that we don't hear Ernie. Removing the DP Ernie leaves us with an ungrammatical sentence in (3f), which turns grammatical again with the additional deletion of the verb in (3g). Moreover, (3g) does not feel like an incomplete sentence; it is a perfect continuation of (3a). Let us revise the hypothesis in (2) as in (4): (4)

Deaccented phrases can be deleted.

Let us consider a case which shows that not all deaccented phrases can be deleted, as in (5), and then one that shows that even seeming nonconstituents can delete, as in (6a-c): (5)

a. b.

Now you hear ERNIE eating a COOKIE, *Now you SEE Ernie eating a cookie.

Example (5) shows that the deaccented DP cannot be deleted despite the fact that it is properly deaccented. The examples in (6) show that apparent nonconstituents can delete: (6)

a. b. c.

COOKIE Monster was and ERNIE was eating COOKIE Monster was and ERNIE was eating COOKIE Monster has has eaten CHIPS.

eating a PEAnut-butter cookie, a CHOcolate-chip cookie. eating HIS peanut-butter cookie, his peanut-butter cookie, TOO. eaten more COOKIES than ERNIE

The deaccented words that are deleted in (6a-c) do not seem to form phrases and thus constitute a problem for the hypothesis in (4). Moreover, as in the discussion of the optical illusions (fig. 1), there is an ambiguity in (6b): it is not completely clear whether Ernie is eating his own peanutbutter cookie, or in fact Cookie Monster's. Hypothesis (4) cannot account for these two interpretations. We can conclude from this initial discussion that deletion process is rule governed. It does not seem possible to just delete whatever is deaccented. Deaccentuation is a necessary, but not a sufficient condition for deletion, as seen in (3) and (5).

Prologue

9

Economy and Default Strategies in the Cognitive Organization of the Mind: The discussion of Ladd's light and shadow metaphor, correlating the concepts of visual perception with the functioning of the prosodic system, can be summarized by three hypotheses that constitute the starting point of the present study. (7)

Initial Hypotheses: a. The prosodic system is sensitive to different mechanisms of focusing and backgrounding. The processes of focusing by prominence assignment and backgrounding by reduction of the phonological prominence (or complete omission) are not complementary processes. b. Reconstruction of missing information is possible and rule governed. c. Hypotheses 1 and 2 are systematically related.

Before I begin with the investigation of these hypotheses and others in chapter 1, a word of caution is in order here: translating the visual metaphor into initial hypotheses of language might be considered unorthodox, and the underlying hypothesis of this prologue, namely the parallel internal organization of the visual system and the language system, might turn out not to be tenable. However, the fascinating and hitherto unanswered question that researchers in both areas will have to tackle is: why do we perceive more than we see, and why do we understand more than we hear? An answer will have to be sought in the maximally economic cognitive organization of the mind. The mind is equipped for using default strategies of information processing which integrate information from all the interfaces (cf. Seuren 2003). Focusing on the language system, the answer lies in the most economic division of labor between the interfaces. Explaining ellipsis means explaining the sound of silence at the interfaces.

Chapter 1 Ellipsis and focus: An introduction

1.

Introduction

The main aim of this study is the development of an interdisciplinary account of ellipsis. More specifically, I will explore the syntax and information structure of a subset of the set of elliptical constructions in English given in (la) to (If): (1)

a. b. c. d. e. f.

Manny plays the piano and Anna the flute. Manny plays the piano but Anna doesn't. Manny plays the piano and Anna does the flute. Manny plays the piano and Anna, too. Someone's playing the piano but I don't know who. Manny played a solo with one hand and Anna with two.

The term ellipsis, from Greek elleipsis, most generally, refers to the omission of linguistic material, structure and sound. In each of the elliptical constructions in (1) linguistic material is omitted, deleted or simply left unpronounced. Nevertheless, the silent string is understood in each case. The silent sting in the second conjunct in (la) is interpreted with the so-called Ellipsis Remnants, Anna and the flute, as ...and Anna plays the flute} The verb play is gapped, therefore the construction is called Gapping. In (lb) the second conjunct is interpreted as ...but Anna doesn't play the piano. Here the verb phrase after the auxiliary is elided, forming a case of VP-Ellipsis (VPE). The interpretation of the second conjunct of (lc) as ...and Anna does the flute constitutes a case of Pseudogapping. The construction in (Id), where the second conjunct is interpreted as ...and Anna plays the piano, too is known as Stripping, (le) where the missing material is understood as ...but I don't know who is playing the piano is Sluicing, and the example (If) which is interpreted as ...and Anna played a solo with two hands combines Noun Phrase Ellipsis (NPE) with gapping. I investigate the interaction between the syntactic, the prosodic and the semantic derivations of elliptical sentences in the framework of generative grammar and the information structural component. In particular, I address

Introduction

11

the problem of deriving adequate phonological and semantic representations from the computational system of human language (CHL) IN correspondence with the syntactic theory of focus that is rooted in the Anglo-American tradition of information structure theory (as developed by Halliday 1967a, b, Bolinger 1972, Jackendoff 1972, 2002, Kuno 1972, Chafe 1976, Prince 1981, 1986, 1988, Rochemont 1986, Gussenhoven 1992, 1999, Vallduvi 1992, Selkirk 1995, Ladd 1996, Winkler 1996, Culicover and McNally 1998, Kiss 1998, Steedman 1996, 2000, Drubig 2003, among others). Thereby I am going to concentrate on issues of the interpretation of the elliptical construction from three perspectives: first, I will focus on the issues related to the syntactic derivation and the surface syntactic interpretation (such as the interpretation of Contrastive vs. Information Focus, the informational relation between the gap and the remnants), second the interpretation of the ellipsis itself (such as strict and sloppy readings, Backward Anaphora Constraint, Binding Principles, different scopal readings, the interaction of scopal readings under negation), and finally on the interaction between the phonology and discourse-pragmatics (such as deaccentuation vs. deletion, intonational disambiguation, and different focus readings). Throughout this investigation, I am interested in the interaction between the syntactic derivation of ellipsis, the focusing of the remnants and the phonological reduction of the elliptical material and its interpretation. The main goal is the formalization of the interrelatedness between syntax, surface semantic interpretation, focus and deaccentuation in the derivation and interpretation processes of ellipsis. Considering the set of examples in (1), I focus essentially on two types ellipsis, VPE on the one hand and gapping and stripping on the other. The starting-point of the present investigation is captured by three main questions, given in below: i. ii.

iii.

What is the role of focus in the derivation of VPE vs. gapping and stripping? What are the principles that regulate the interrelatedness between deaccentuation of the elliptical material and focusing of the remnant(s)? What role does phonological disambiguation play in the interpretation of VPE and gapping and stripping?

Recent influential work on the theory of ellipsis (Dalrymple et al. 1991, Hardt 1993, 2003, Hartmann 2003, Johnson 2004, Kehler 2000, Lappin

12

Ellipsis andfocus: An introduction

1996, Merchant 2001, Romero 2003, Tomioka 2003, among others) and new developments in the semantic/pragmatic theory of focus (Rooth 1992, Schwarzschild 1999) seem to suggest that the meaning of omitted elements is a purely interpretive process taking place in direct correspondence between phonology and semantics/pragmatics without recourse to syntax. My aim here, however, is to show that core computational processes and operations - such as the syntactic theory of displacement - do in fact play an important role in the process of deriving and interpreting elliptical constructions in the information structural component at LF and PF. In particular, I will show that ellipsis provides evidence for the hypothesis that displacements have a direct effect on the Information Structure (IS) of a sentence. The term IS stems from Halliday (1967b: 200) and refers to the hypothesis that "the distribution of information specifies a distinct constituent structure on a different plane; this 'information structure' is then mapped on to the constituent structure as specified in terms of sentences, clauses and so forth (...)." Many different research programs have since explored aspects of IS. In the present study, I will concentrate on the interrelatedness between the syntactic structure of a sentences, its derivational history (including movement and anaphoric processes) and its information structural interpretation. I use IS to refer to both the constituent structure and its respective interpretation that results from topic and focus movement and the distribution of given and new information. When I am referring to the interface at which syntax affects IS, I use the term Surface Semantic Interpretation (SSI), a term introduced by Chomsky (2000, 2001: 15) to describe the subcomponent of LF that is responsible for the interpretation of syntactic displacement. Further, I will also show how syntax, information structure, intonational phonology and discourse-pragmatics connect in deriving elliptical sentences in a parallel effort. I am developing an account of the architecture of grammar in which the syntactic theory of contrastive focus and Topic Movement is recast more formally in the theory of syntactic displacement which operates in parallel with the interpretational and the phonological component. More straightforwardly, I am putting to test an account of ellipsis at the interfaces. Most linguists working in the generative framework today agree that ellipsis and information structure is an interface phenomenon. However, the question about the actual division of labor between the components is more controversially discussed; for example, how do the components of

Introduction

13

grammar divide up the work between them so that at the end all that needs to be said is nothing? To narrow down this question: (i) What exactly happens when, for example, a VP is left unpronounced? Does the deletion process take place in syntax or in phonology? (ii) What exactly is the role of LF? Does it interface only with syntax or also with phonology? When and where does interpretation of the elided element(s) take place? Is the parallelism requirement and the identity constraint checked at different levels? (iii) What exactly is the role of PF? Is it an extension of the syntactic component? How does it identify focused/defocused material and how does it derive intonational contours? The issue of the division of labor between syntax, SSI, LF and PF is the main focus of this work. 2 Specifically, I will propose that the syntactic theory known as the Minimalist Program (Chomsky 1995, 2000, 2001, 2004, 2005) enriched by the theory of displacement, formerly known as the Movement Theory of Focus (cf. Rizzi 1997, Kiss 1998, Drubig 1994) allows us to derive the SSI and the prosody of Sentence-Bound Ellipses (SBE) directly from its syntactic encoding (cf. Phillips 2003, Uriagereka 1999, Platzack 2001, Lopez and Winkler 2003, Lopez to appear). For the Discourse-Bound Ellipses (DBE) additional interface correspondence rules pertaining to anaphora are required to derive the appropriate interpretation (Hardt 1993, Kamp and Reyle 1993). Thus, the ultimate goal is deriving a theory of ellipsis from a syntactic theory of focus and thereby clarifying the function and interaction of the interfaces with respect to focus, deaccentuation, anaphoricity and deletion. Let me caution the expectations of a unified theory of ellipsis at this point. The analysis that I propose is a Hybrid Focus Account of Ellipsis. It takes the distinction between SBE and DBE, which is inspired by Williams' (1977a) original differentiation in sentence grammar and discourse grammar, as one essential indication of the different information structural functions that ellipsis can assume. However, the method of investigation is essentially the same for both types of ellipses. In each case, I start out investigating the contribution of the syntactic derivation to the IS and SSI on the one hand and to the semantic interpretation and intonation on the other. Before I start testing this essentially syntax-driven account of ellipsis, I will introduce the grammatical model, the core cases of ellipsis discussed in this book and their intonational realization, as well as the basic hypotheses that explain their behavior as the result of mapping syntactic structures to SSI.

14

Ellipsis and focus: An

introduction

2.

The derivational model

The basic syntactic framework that I adopt for my analysis of ellipsis is that of the Minimalist Program (MP) (Chomsky 1995) with its current further developments (Chomsky 2000, 2001, 2004, 2005) keeping in mind its roots in the Principles and Parameters approach (PandP) (Chomsky 1981). I will concentrate in this introductory chapter on two main areas of this theory which are essential for the leading idea of this study. The first concerns the concept of the basic grammatical model that underwent a change in recent years from a basically representational model, known as the T-Model (Chomsky and Lasnik 1977), to a D(erivational)-Model (cf. Chomsky 2000, 2001, 2004, 2005). More precisely, the question that will be central to this and the next chapter is how is ellipsis explained in the Dmodel. The second area concerns the precise nature of the interfaces involved. In this chapter, I concentrate on the interaction of core computational processes and focus interpretation at SSI, leaving the discussion of these interfaces with phonology to chapter 2. Now, let me start with the discussion of the D-model. In the classic view set out by Chomsky (see Chomsky 1965, 1981, 1995), the syntactic component of the grammar accounts for the matching of sound and meaning (see Jackendoff 1997, 1998 for an opposite view). More precisely, syntactic structures are interpreted at two different levels: at the Phonological Form (PF) and at the Logical Form (LF), which constitute interfaces with other systems, the articulatory-perceptional and the conceptual-intensional system (Chomsky 1995: 168). Under this conception, it is one of the most important issues to find an explanation of how it is possible for speakers to produce ellipses and for listeners to interpret them in the absence of form. One central hypothesis of this study is that the intonation of the sentences is relevant for an answer to this question. In examples (2) to (7), pitch extraction contours of a prototypical intonational realization of attested examples are provided below. The contour description uses Pierrehumbert's (1980) notation, as it is modified in Beckman and Pierrehumbert (1986). Intuitively, the idea is that intonational contours consist of only two phonemic tones, H(igh) and L(ow). The prosody of a sentence containing a gap, such as example (2), is made up of several possible sequences of pitch accents (H*, L*, L+H*, L*+H, H+L*, H*+L) which are associated with a lexical item (for a more detailed discussion of the inventory of tones the reader is referred to Chapter 2.2.2.1.).

The derivational model

(2)

15

Gapping She wrote long [...] letters, which she sent to her sister and she to my mother. (Ruth Prawer Ihabrala (1975) Heat and Dnst\ quoted from Prince 1988)

The intonational contour depends on three parameters: the relationship of the tone to the baseline, the degree of prominence expressed by the pitch range, and the relationship of the present tone to the preceding tone. The concatenation of the single tones makes up the melody of the sentence. The intonation of the embedded gapping example in (2) is graphically represented in figure 1:

SYSTTFP*

H*+L Ctata

CFTJPTÜIMS·

H*L"H% ΜΪ©Μ

I-fink

H*+LH"

91KM

SPNIK

LYSSE·

Jrtithl

I tei· •othi • ι™ mfirtacniFi ι ;lti Kent La her s I ϊ- t e r η it :; h e rtw t & «« >

H*LT% EJfct T^s MADITO L O A H.HHHHiK - |

Li . BDEKMI tp my in c t h e t«.fifjiiriH·; S13> I

ι Γ-„ chl

Χ5ΪΜ

Figure 2a.

CftjptvMt

Rata

H 1 - 5 7 - 2 . IT; F

UIchj

!H* L"H% Link

8 1 km*

H* L A rail « j a «

RiSt

H* L"L% f-fty

Haar«

The derivational model

H*

L"H%

H*+L

17

H* L"L%

Figure 2b.

The pitch track in figure 2a shows an H* pitch accent on flowers and a downstepped H* pitch accent on Linda (!H* according to Beckman and Ayers 1997). At the end of the first intonation phrase, which corresponds to the first conjunct, we see a small rise signaling continuation. In the second intonation phrase, both the negative element not and the DP loveletter are assigned an H* pitch accent. The intonation contour corresponds to the interpretation that of all possible things that he could have given to Linda, he gave her the flowers but not the love-letter. The contour in figure 2b also shows an H* pitch accent on the correlate flowers in the antecedent clause, but the prepositional phrase to Linda is characterized not by an H* but by an L" phrase accent and an H% boundary tone. Thus, whereas the contour of the first conjunct in figure 2a could also have qualified as a gapping contour, figure 2b clearly marks only the DP the flowers as focus in the first conjunct, which is contrasted with the DP a love-letter in the second conjunct. The affirmative instantiation of sentence polarity, too, is also assigned an H* pitch accent. Example (4) contains two cases of ellipsis:

18

(4)

Ellipsis and focus: An

introduction

VP- and NP-Ellipsis [Why does Betty think I was trying to kill myself?] On the principle that one swallow doesn't make a summer, but two probably do, dear girl. (Walters (1996) The Dark Room. London: Pan Books, p. 42.)

Example (4) involves both a VPE and an NPE in the second conjunct: but two swallows probably do make a summer. The remaining functional heads, number head two in the case of the NPE and Infi do in the case of the VPE, both bear highly modulated peak accents (H*) with a subsequent fall (L), as can be seen in the pitch track representation in figure 3. H*+L Sy«-*«·* C-p.jptur-e· Itata

H* L"L% Link

Show·

H*+L freely®«*

OOülil · IFft UWHSCKIFT jrirae a v. ad ] ow docsit m ; ^ a s u m m e r li tit Τinu is BO?

t w o

H* L"L% Edit

ϊ^βΐ

M-aar^o Lob

ρ r ob & bl jr

U.aüBüU fl u·

tf.MHNMrt;

l-96>

Figure 3.

The VPE and NPE realized in figure 3 differ from gapping and stripping in that the H* accents target functional elements immediately preceding the ellipsis site and not phrasal remnants. The question what exactly triggers the accent realization will be further discussed in section 1.3.3, where it is proposed that different focusing mechanisms are at work in VPE/NPE and in gapping and stripping. In particular, the assumption that accent realization is directly associated with focus feature assignment in languages like English is replaced by the Phasal Head Prominence Principle, which accounts for sentences, in which the head of a phrase, in

The derivational model

19

our case the functional heads Number and Inflection, are prominent because the elided material in (4) is redundant or anaphorically given. The examples of pseudogapping, as in (5), seem to constitute a mixed form. In (5a), an accent is realized on the functional head didn't and on the remnant The Great American Novel. In (5b), again a fall (H*+L) is realized on the modal would and two accents on the complex remnant on a straight policy for the same amount. (5)

Pseudogapping a. Manny read The Facts, but he DIDN'T read The Great American NOVEL. b. Third and most important, Amex would charge me a far higher premium than other reputable companies WOULD on a STRAIGHT POLICY for the same AMOUNT. (Penn Treebank #9)

The sluicing examples in (6) are characterized by the omission of the IPconstituent following the w/j-element. In (6a) the IP he read t is omitted. In (6b), the IP it exists, which follows the w/z-phrase where, is omitted. A pitch extraction contour of (6b) is given in figure 4. (6)

Sluicing a. Manny read The Facts, but I don't know what else, [IP he read t j . b. There is a lot of talk about freedom. It's like the Holy Grail, we grow up hearing about it, it exists, we're sure of that, and every person has his own idea of WHERE. (Winterson 1987 The Passion. Penguin Books, p. 154.)

20

Ellipsis and focus: An

H*+L

introduction

H*

!H*

!H* L-L%

Figure 4.

In example (6b), sluicing is licensed by the presence of C°, which is specified for [+wh]-features, as originally proposed by Fukui and Speas (1986) and later captured as Specifier-Head Agreement in the MP (see the detailed proposal by Merchant 2001). The remaining overt wh-element where bears a downstepped H* pitch accent immediately followed by an L" phrase accent and an L% boundary tone. An interesting ambiguity arises in (7) that can be intonationally disambiguated, as observed by Merchant (2001: 77). Depending on the intonation of the first conjunct, who else is either interpreted as a subject or an object expression, as shown in (7a) and (7b). (7)

Manny called Ben an idiot, but I don't know who else. a. MANNY called Ben an IDIOT, but I don't know who ELSE called Ben an idiot. b. Manny called BEN an idiot, but I don't know who ELSE Manny called an idiot.

In each of the above elliptical constructions in English, part of the second conjunct is omitted but still interpretable at LF. In this respect the mere occurrence of ellipsis is a challenge to our understanding of the architecture of grammar, conceived of in terms of the MP-version of the T-

The derivational model

21

Model as in (8), where the most powerful constraint, the Principle of Full Interpretation (PFI), requires all PF/LF symbols to have interpretations at the PF/LF interface levels (Chomsky 1995). (8)

The T-Model of the MP (Chomsky 1993, 1995) Lexicon overt syntax (Spell-Out) covert syntax PF

LF

Part of the answer to the question of how words or phrases can be understood without being pronounced, or otherwise referred to, already lies in the way we assume that elliptical material is represented and where. Three questions that are central to the study of ellipsis are listed in (9): (9)

Core questions of the study of ellipsis: i. Does ellipsis have internal structure? ii. How is the elliptical structure interpreted? iii. Is ellipsis the result of a deletion operation or is it a basegenerated empty category?

The answers to (9i) and (9ii) determine the answer to (9iii), namely the answer to the question of whether ellipsis can be derived by transformational operations or whether it is to be analyzed as a basegenerated empty category. Within the T-Model of the early MP, three prevailing research paradigms can be isolated: the Phonological Deletion Theory, the Syntactic Displacement Theory, and the Semantic Theory. The phonological deletion theory assumes that elliptical material is fully syntactically represented, but deleted at a certain point in the derivation of the sentence. 4 The syntactic displacement theory investigates the computational system proper and aims at an answer to the question of whether ellipsis can be reduced to movement or other independently existing syntactic processes. 5 Although the starting point of any syntax-first methodology is, as is self-evident,

22

Ellipsis and focus: An

introduction

essentially syntactic, the goal is to find out exactly at which stage the syntactic derivation must be handed over to the interfaces. The semantic theory assumes that semantic mechanisms governing the recovery of meaning from discourse can be put to work in the analysis of elliptical constructions. 6 A problem that all three theories face is that the place of the interpretation and the place of the actual deletion diverges. This problem results from the architecture of the T-Model in (8), still an essentially representational model, in which syntax interfaces with LF and PF, not allowing any other correspondences. Therefore, I will depart from the model in (8) and propose an elaboration of the derivational model of grammar in (10), as conceptualized by Chomsky (2000, 2001, 2004, 2005). 7 There are two innovations represented in the derivational model in (10): First, the syntactic, semantic and phonological components derive sentence structures in parallel, cyclically or - in modern terms - in phases. Second, IS figures prominently as a subcomponent of LF mediating between syntax and semantics proper. At this point, the role of pragmatics within grammar is still an open issue. (10)

Elaboration of the D(erivational) Model of Grammar

The derivational model

23

Let me discuss the core differences between the D-model in (10) and the TModel (8) in more detail. Most importantly, the grammatical derivation in the D-model is built up in parallel in the individual components (CHL, PF, LF). The implementation of parallel computation is that CHL builds up larger chunks. The derivation in (10) starts with the operations merge and agree. A given structure is then sent to LF where its interpretability is determined. Uninterpretable features are marked as such and are then eliminated in a communicative act between syntax and LF. The model in (10) proposes that each component interacts with the other components in the derivational process. In the case of elliptical constructions, syntactic derivation precedes interpretation which is heavily dependent on the semantico-pragmatic component before the structure is sent to PF. The role of IS at SSI and where exactly ellipsis takes place will be discussed in more detail in Chapter 2 (cf. Jackendoff 1998).8 Furthermore, the parallel derivation of syntax and semantics in the Dmodel solves one of the most crucial problems of the theory of ellipsis: all previous work on ellipsis faced the dilemma that the location of the actual deletion of linguistic material and the location of the interpretation of the elliptical material did not coincide. This problem is solved in the D-model proposed here. The account, later referred to as Parallel Computation Approach, derives the different types of ellipses in a system where interpretation and deletion happens cyclically (see chapter 2 for discussion). More precisely, the grammatical derivation of a specific sentence is subdivided in smaller units, known as Phases that are sent to the interfaces. Chomsky (2000, 2001) defines a phase as "the closest syntactic counterpart to a proposition" and identifies it as vP or CP. The phase is a modern instantiation of what used to be called a Complete Functional Complex in GB theory. The notion of a phase as a unit that functions independently at the interfaces (can be independently manipulated, forms a phonological constituent and a sense unit) is crucial for the theory of focus introduced in section 1.3.3. For a given phase the interfaces determine its interpretation. It has been proposed by Chomsky (2000) that direct correspondence exists between LF and CHL with respect to uninterpretable features (i.e. formal features). They are marked, and the structure is sent back to the CHL where agree removes the uninterpretable features. In case the target structure is a complex sentences, a larger phase is built around the previous phase, thereby closing off the smaller phase for further operations (cf. PhaseImpenetrability Condition).9 Although recent work on phases has concen-

24

Ellipsis andfocus: An introduction

trated on the syntax-LF interface, a comparable correspondence must be assumed for the systems interfacing with PF. More precisely, it is argued in chapter 2.4 that PF is derived cyclically, or in phases. The assumption is that PF has access to both, the syntactic derivation of the phase (uninterpretable features remain visible at PF) and the semantic interpretation thus providing evidence for the D-model in (10) where LF and PF derive structures in parallel. The theory of ellipsis developed here, is partially based on the concept of movement operations of the remnants, also referred to as syntactic dislocation or displacement. Displacement in the CHL is triggered by uninterpretable features and constitutes only "an apparent imperfection" (Chomsky 2001: 3). In fact, it is an instruction to the LF-interface that the SSI differs from an identical derivation without movement. The underlying function of displacement can be considered an ingenious move of the envisaged super-engineer to signal differences in information structure that can be hypothesized to be active from early on in the language acquisition process. 10 However, movement operations in the phonological component (displacement rules like Zubizarreta's (1998) p-movement, head movement and some deletion rules) should have comparably little semantic effect, as expressed by the hypothesis "Surface semantic effects are restricted to narrow syntax" (Chomsky 2001: 15), a point to which I will come back in connection with deletion and deaccentuation rules. The D-model expresses the Multiple Spell-Out hypothesis that is contingent on agreement/feature-checking operations. The underlying idea is that there is only a single cycle, thus no covert movement. The basic architecture of the PF component also follows the idea of the single cycle. That is, there is no differentiation between overt and covert operations. Instead there is cyclic Spell-Out and all operations at PF are cyclic. The assumption is that the intonation of a sentence is derived from the interaction of various phonological rules: The pitch accent assignment rules which operate automatically over a specific domain (the phase) and are sensitive to anaphoric deaccentuation and those that are sensitive to formal features and trigger displacement (transformational cycle). The core ideas of phonological rules operating cyclically over phases are introduced in section 2.4.11 One of the central questions I wish to address in this book concerns the properties of the different types of ellipsis represented in (1) to (7). More precisely, the hypothesis is that there is a systematic difference in the derivation, the intonation, the interpretation and discourse function of

The syntax-semantics interface

25

SBEs and DBEs. I approach this issue by investigating the hypothesis that movement of a phrase, in the case of ellipsis an overt remnant, in an identifiable A'-position results in a contrastive topic/focus reading of this remnant. 12 However, if ellipsis does not involve such syntactic displacement of the remnant(s), the interpretation is completely dependent on functional aspects of information structure theory and accessibility theoiy hosted at the component of (discourse-)pragmatics. Thus, the basic claim is that the two types of elliptical constructions involve different syntactic, (surface) semantic, phonological and pragmatic derivations.

3.

The syntax-semantics interface

The idea of proposing different types of ellipses on syntactic, semantic and pragmatic grounds is by no means new (cf. Williams 1977a, b, Chao 1988, Lobeck 1995, among others). However, what is new is that I propose that the main characteristics result from C HL interfacing with LF/PF and from general assumptions about information structure and focus.

3.1. The cyclic derivation of surface semantic interpretation The guiding idea of the D-model in (10) is that the syntactic, information structural, phonological and pragmatic components derive the sentences in parallel. The components are subject to modularity in the derivation and interpretation of the phase, but they communicate with each other at the interfaces by handing on and manipulating the phase. The core question that needs to be answered is of how IS (and here particularly the notions of given vs. new, topic vs. focus, anaphoricity, contrastivity, specificity, etc.) and syntax interact and how it contributes to the SSI in general and to the interpretation of ellipsis in particular. There are essentially two different general approaches, the formal grammar and the discourse grammar account (see Molnär and Winkler to appear): the proponents of the formal grammar account assume that information structural notions are integrated into the formal system of language. 13 The proponents of the discourse grammar account assume that information structural notions are represented in the pragmatic component primarily governed by contextual factors. 14 The formal grammar account assumes that within the grammatical model sketched in (8), LF (the syntax-

26

Ellipsis and focus: An

introduction

semantics interface) is responsible both for the representation of scopal relations and for the interpretation of information structural and discourse notions. The opposite position assigns the explanation of the discourse functions, such as those expressed by these notional pairs and dichotomies, to (discourse-) pragmatics, which either interfaces with LF or forms a separate component. 15 My claim that elliptical constructions form two distinct natural syntactic classes, one sentence-bound and the other discourse-bound seems to cause a dilemma at first sight since it forces me to assume a different account for each phenomenon. However, the research program pursued in this monograph aims at answering the question that is crucial to future research in the interface discussion of ellipsis: how much of the information structural notions and exactly what is formally integrated in the computational system of grammar, how much of it and exactly what falls in the realm of (discourse-)pragmatics and how exactly does it influence the derivation of intonation. The answers to these questions require a refinement of the grammatical model with respect as to how exactly the elliptical phenomena (DBEs like VPE vs. SBEs like gapping and stripping) are derived, interpreted and intonated. Concentrating on the mapping relations between syntax and interpretation in the D-Model, the central hypothesis is that there is a direct relation between syntactic form and SSI. Since the Chl derives structures that involve movement and others that don't, this hypothesis implies that there is not a homogenous information structural component but rather that it consists of at least two cycles - an essentially functional cycle and a grammatical cycle. The grammatical cycle hosts the interpretive rules that operate at the syntax-semantics interface. The functional cycle includes rules that operate at the interface to pragmatics. These include rules that determine discourse structuring, anaphoricity, (in)definiteness, reference and event management. The grammatical cycle and the functional cycle contribute to SSI in different ways. While the grammatical cycle interfaces with syntax directly (phases are transferred back and forth), the functional cycle takes syntax as input and interfaces with (discourse-)pragmatics. Assuming this to be correct, the notion of interface is far more dynamic than previously conceived of and is strongly connected to the notion of phase.

The syntax-semantics

interface

27

3.2. The double-cycle information structure hypothesis The claim presented above is that information structure involves two separate cycles: a functional cycle and a grammatical cycle. The first cycle (cycle 1) is the default cycle that automatically derives the Information Focus (IF) of the phase that is handed on to LF, and checks the functional notions of information structure in correspondence with pragmatics. The smallest phase constitutes the domain of IF. Furthermore, cycle 1 determines whether a phase needs to be sent on to the second cycle on the basis of the presence or absence of formal features. The second cycle (cycle 2) checks the grammatical status of the constituents in the phase. It identifies uninterpretable features that lead to syntactic displacement and word order variation. As a general rule for English, cycle 2 interprets displaced constituents to a sentential initial position as either Contrastive Focus (CF) or Contrastive Topic (CT) at SSI. The first cycle typically derives unmarked sentence structures, whereas the second cycle derives marked sentence structures. The derivational history (i.e. whether a phase passes the first cycle without being sent on to the second cycle from where it is sent back to CHL) has direct consequences for the SSI. The assumption that two different cycles operate in the derivation of information structure is comprised in the Double-Cycle Information Structure Hypothesis (DC-ISH).16 The DC-ISH spells out the claim that the default functional cycle and the grammatical cycle contribute to SSI in a cyclic fashion. The information focus hypothesis, here referred to as In Situ hypothesis, is given in (11a). The direct relation between syntactic displacement and surface semantic interpretation is stated in ( l i b ) by the Displacement Hypothesis: (11)

The Double-Cycle ISH: a. In Situ Hypothesis (cycle 1): Cycle 1 operates automatically over the smallest phase sent to LF/SSI and locates the information focus in situ. b.

Displacement Hypothesis (cycle 2): Cycle 2 checks the phase for displaced material. Material moved into an A'-position at the edge of the phase (spec,vP, or spec,CP) in narrow syntax has a direct effect on surface semantic interpretation at LF/SSI.

28

Ellipsis andfocus: An introduction

The idea is that the IS component identifies the different types of foci before handing the material on to SSI for interpretation. The information focus identified by the first cycle receives the default interpretation as new IF at SSI. The displaced material identified by cycle 2 is interpreted as CF or CT (depending on cycle l). 17 The consequences of the derivation of intonation at PF will be spelt out in more detail in Chapter 2. The basic idea is that the information focus and the displaced constituents are assigned pitch accents at PF. The reader will have noticed that the in situ case of contrastive focus has not been treated here. I will come back to it in the next section. 18 There are three questions that need to be answered in connection to focus in general and as it pertains to ellipsis: first, what are the basic assumptions of focus and how is the concept of focus defined. Second, what role does anaphoricity (and givenness) play in the theory of focus? Third, how exactly should we conceive of the working of focus at the interfaces (syntax-PF, semantics-PF) and what kind of focus hypothesis is relevant for the discussion of ellipsis? An initial answer to all three questions is provided in the remainder of this chapter, leaving the thorough investigation for the chapters to come.

3.3. Information focus and contrastive focus: A derivational approach Recently many controversial proposals that have been forwarded in the search of an empirically and theoretically adequate account of the relation between focus, intonation and interpretation. Here, I will propose a focus theory that assumes with recent work on focus typology that at least two types of foci must be distinguished: information focus and contrastive focus (Drubig 1994, 2002, Hetland and Molnar 2003, Kiss 1995a, 1998, among many others). 19 According to these authors, the major syntactic distinction of these two focus types is seen in the way these foci are licensed: while information focus, which marks new information in the sentence, is assumed to involve in situ licensing as graphically represented in (12a), contrastive focus is assumed to involve LF-movement of the focus phrase in an appropriate Α-bar position as in (12b):

The syntax-semantics

(12)

a. b.

interface

29

information focus: licensed in situ: [ X; [...Focusj...]] contrastive focus: formation of an operator-variable chain at LF [Focus; [ ...t; ...]]

While I follow these authors in maintaining the essential differentiation of these focus types, the implementation will be spelt out differently. Recent developments in the conceptualization of the interfaces suggest that these two types of foci are not subject to different licensing conditions in English and German, but that they play different roles in the cyclic derivation of sentences at the syntax-phonology interface (cf. Chomsky 2000). The central claim with respect to contrastive focus/topic and information focus is that only the former is assigned a formal feature (a E(dge)-feature [F] or [T]), while the latter is defined per default over a specific domain, namely the smallest phase, which is derived in the CHL and sent to the syntax-semantics interface. 20 In the case of contrastive focus/topic, the Ε-feature on a word-level category in the phase has direct effects on surface semantic interpretation. More precisely, the contrastive focus or topic interpretation at SSI is connected to the process that erases the Ε-feature at the syntax-semantics interface. The Ε-feature is a formal feature and is erased either by an agreement operation or by a displacement operation of the constituent that contains the Ε-marked element. To the extent that this claim goes through, it constitutes another reason to assume that phases are real. (13)

a.

contrastive focus/topic:

b.

The [F]/[T]-feature must be erased either via agreement or via movement, information focus: [ a ] is the domain of information focus, where α is defined as the smallest phase.

[A

XP[F]/[T]

] E -feature

The derivation of information focus is the topic of the next section, while the discussion of contrastive focus/topic is taken up in section 1.3.3.2.

30

Ellipsis andfocus: An introduction

3.3.1. Information

focus

The concept of information focus (IF) is defined by three parameters: its domain of application, its interpretation and its phonological realization. With respect to the first parameter, the domain of information focus is the smallest phase that is derived by CHL and sent to LF. The smallest phase is vP. With respect to the second parameter, information focus is pragmatically defined as new information (cf. Halliday's 1967b original definition). Here, the notion of new information is implemented by the functional cycle (cycle 1) of the DC-ISH in interaction with discourse factors. With respect to intonation at PF, information focus is signaled by pitch accent assignment in intonational languages like English and German (cf. Selkirk 1995, Fery 1993). More will be said about the location of the pitch accent (PA) in the phase, as I proceed. An example of information focus is given in (14): 21 (14)

(15)

A: B:

What's going to happen next? Someone is eating a COOKIE. is [ a someone eating a cookie].

The first step involves the derivation of the phase α in (15) via the operation merge, α is identified as vP, the smallest phase that is sent to LF. Cycle 1 identifies the location of the IF per default. Cycle 2 at LF checks if α contains uninterpretable [F]/[T]-features. If it doesn't the phase α is interpreted as the domain of information focus per default. Anaphoric ity and givenness check is applied. Since (14A) is a discourse initial sentence, it does not contain any anaphoric or c-construable/given elements. Therefore the smallest phase α is confirmed as domain of information focus at SSI. So far, nothing has been said about the distribution of focus features, or about pitch accent assignment. In SSI, information focus is not a feature on a focused word level category, but a default assumption that the smallest incoming phase α (vP) is the domain of new information. The central idea is that the vP phase constitutes the domain of information focus much in the spirit of Diesing's 1992 Mapping Hypothesis and the literature growing out of this proposal (cf. Winkler 1996 for an overview, and Drubig 2003 for a recent implementation). As will be shown later, this hypothesis requires that the domain of information focus can contain Given

The syntax-semantics interface

31

Information, as has been assumed by various authors (cf. e.g. Ladd 1980, Selkirk 1984, 1995, Rochemont 1986, Drubig 1994, Winkler 1996, Zubizarreta 1998), and formally spelt out by Schwarzschild (1999). Pitch accent assignment rules (PAR) apply at PF. As suggested by the grammatical model in (10), α is sent to PF as an independent unit. However, PF has access to the results of the information structure cycles 1 and 2 at SSI. More precisely, PF has access to the derivational history of the phase, as hypothesized by Chomsky (2000: 131, 2001). That is, when α of example (15) is sent to PF, PF reads off it that it does not contain eliminated formal [F]/[T]-features ( a was sent to PF right after passing cycle 1) and that α only contains new information. Under the assumption that PF interacts directly with SSI that has access to anaphoricity, reference, specificity and event management in the pragmatic component, then nothing special must be said about the location of the accent in the first cycle. It will follow naturally from a rule that assigns a PA to the head of the phase α if and only if its internal argument is given, as in (16). 22 (16)

PhasaJ Head Prominence Rule (PHP-Rule) The head of the phase is assigned a pitch accent, iff its internal argument is given.

This principle is modeled after the Head Prominence Principle proposed by Drubig (1994: 16, 1997: 7) but it is defined over phases and connects the absence of the accent on the internal argument to the notion of anaphoricity and givenness. 2 3 Moreover, prominence is defined in terms of pitch accent assignment for intonational languages like English and German. The PHP-Rule in (16), thus, has similar effects as Selkirk's (1995) focus projection mechanism (see Winkler 1996), but achieves these without the assignment of focus features and the operation of focus projection. It operates over a specific domain, namely the smallest phase that is built by the CHL and is interpreted by LF as the domain of information focus. 24 In essence, the PHP-Rule in (16) achieves similar results as the sophisticated versions of the NSR that derive sentence prosody on the basis of syntactic analyses paired with pragmatic considerations like anaphoricity, accessibility and givenness (cf. Bresnan 1971a, 1972a, Reinhart 1996, Zubizarreta 1998). However, the overall proposal developed here is a purely derivational one that is conceptually superior to previous proposals since it derives information structurally unmarked structures (information focus) like (16) via the functional cycle

32

Ellipsis and focus: An

introduction

at PF (the PA on cookie is assigned according to the PHP-Rule) and derives contrastive focus and contrastive topic on a separate cycle, namely the grammatical cycle, thus connecting the syntactic derivation of phases and their surface semantic interpretation to the presence or absence of formal features. For the purpose of this introduction, I limit myself to the discussion of some core cases, ignoring focus regularities in adjuncts, double objects, prepositional objects etc. Rule (16) accounts for the pitch accent location in (14B) and also for the accent location in (17), where the object cookies is given. (17)

Cookie-Monster LIKES cookies.

Given information is defined as "recoverable either anaphorically or situationally" (Halliday 1967b, 211); in Rochemont's (1986) terminology given information is defined as C(ontext)-construable and in Pesetsky's (1987) terminology as D(iscourse)-linked.25 These notions are discourse notions and hold at the LF-pragmatics interface. If applied to (lb) and (lc) of the prologue, here repeated as (18a, b), the whole infinitival phrase eating a cookie is given by virtue of being mentioned in the preceding utterance (14B) Someone is going to eat a COOKIE. (18)

a. b. c.

Now you hear ERNIE eating a cookie. Now you HEAR it. Now you DON'T.

The question arises as to why the system assigns a PA to Ernie? Intuitively, the answer is straightforward: if the intended reading is an information focus reading, no other element can bear the focus. More precisely, under the assumption that α is the smallest vP that is sent to LF [ p you hear [ α Ernie eating a cookie]] and furthermore that the head of the perception verb complement and its object are given, the accent is realized on the agent subject. 26 (18a) readily answers the question: What do we hear noM>? The same argument applies for the more intricate cases in (18b) and (18c). In (18b), the internal argument of the perception verb is pronominalized and thus counts as D-linked and is anaphorically deaccented. 27 Therefore, the head of the larger phase hear is assigned a PA according to (16). Note, that hear in (18b) is assigned a PA, despite the fact that it has been mentioned in the preceding utterance (18a), which qualifies as the well-

The syntax-semantics

interface

33

known default accent phenomenon (Ladd 1980, 1996). These cases constitute an argument for the assumption that the domain of information focus may contain given information. 28 Example (18c) is a genuine case of VP-ellipsis. Although more will be said about cases like (18c) where the complete VP-constituent is missing (see chapter 3), we can assume that (16) applies blindly to the next higher phase (β) and assigns a pitch accent to the head of the phase don't. The question what the computational system provides as input to SSI and PF in cases like (19c) (either sentences with complete VP structures/ or with empty pronominal VPs), will be discussed in chapter 2 together with problems of anaphoricity and the process of phonological deaccentuation and complete reduction. The example is understood as Now you don't hemit where hear it is understood as hear Ernie eating a cookie.

3.3.2. Contrastive

focus

I now turn to the notion of contrastive focus and the question of how it interacts with the concept of information focus: In terms of CHL, contrastive focus and contrastive topic are edge features (also referred to as peripheral features) that are associated with word level or phrasal categories in the lexical subarray. In terms of interpretation, the notion of contrastive focus is informally defined as evoking a suitable "subset of the set of contextually or situationally given elements for which the predicate phrase can potentially hold; it is identified as the exhaustive subset of this set for which the predicate phrase actually holds." (Kiss 1998: 245). Once the phase α is assembled and sent to SSI, the material marked with a [F]/[T]-feature is detected in the grammatical cycle. Since formal features cannot be interpreted at LF, the phase is sent back to CHL- AS with cyclic vt'/z-movement, the focus feature triggers the assignment of an Efeature to the head of the phase. Then, there are basically two options: either, the formal feature on the contrastive constituent is erased via agreement (contrastive focus in situ), or the feature is erased via movement to the edge of the phase (vP) (cf. the displacement hypothesis in ( l i b ) above) where it is accessible to further movements (e.g. movement to a higher position). The optionality of the operation (agreement or displacement) is not an imperfection of the system, but a skillful and economical implementation of the formal Ε-feature to v/C of the super-

34

Ellipsis and focus: An introduction

engineer that Chomsky (2000, 2001) is envisaging, as I will illustrate below and in more detail in Chapter 2. An example for the erasure of the formal Ε-feature via agreement is given in (19a), and via displacement to the edge of the phase by entering into a cleft construction in (19b): (19)

a.

Now you hear [ME]F eating a cookie. It is [ME,]f that you hear t; eating a cookie now.

At stage α of the derivation, the pronoun me, that is assigned a contrastive focus feature, is merged into the structure. (20)

[β you hear [ a [ me]K eating a cookie]]

In a next step, α is sent to LF/PF. At LF, the focus feature cannot be interpreted. Therefore, the phase is sent back to CHL· The focus feature on me triggers the assignment of an Ε-feature to the head of the phase v. Then there are basically two technical implementations of feature checking, as given in (21): (21)

a.

b.

Agreement (feature checking without movement): Agree can take place between the Ε-feature (here the focus feature [F]) and an identical feature [F1] in the domain of F. The domain of F is the complement of F. This is the case in which the contrastive focus feature on me is checked by the feature on the phase in situ. Contrastive Focus Displacement The Ε-feature of the phase can be formally erased by movement of the focused pronoun me to the edge of the phase, here spec,vP, where it is accessible to further displacement.29

In the case of feature checking via agreement, the resulting structure at SSI does not have any Ε-features. However, these features would be visible for PF. In (20), PF is able to read the deleted [F]-feature on the pronoun and assign a pitch accent to me. In the case of feature erasure via displacement, these features are still visible for PF. PF assigns a pitch accent to the displaced constituent.

The hybrid focus analysis of ellipsis

35

Let me end this section on contrastive focus with a brief note on the distinction between contrastive focus, topic, and contrastive topic by taking a look at example (22). (22)

a.

Mary bought the book about BATS in Amherst... (...not the one about RATS) The book about BATS Mary bought in Amherst. The book about BATS Mary bought in AMHERST... (...the one about SQUIRRELS in BOSTON)

Mary in (22a-c) is a Topic in the sense of Reinhart (1982), usually defined as a constituent that occurs in an aboutness relation. A contrastive in situ focus is realized on BATS in (22a) placing it in opposition to the DP the one about RATS. The contrastive focus DP the book about BATS has been moved to the left periphery of the sentence in (22b). The same DP is displaced in (22c), but in this case functions as a contrastive topic, where the contrastive focus is realized on the PP in Amherst. There has been an ongoing debate between those who analyze the contrastive topic as a subtype of topic, for example Kuno (1973: 44-49), Kiss (1987) and Biiring (1997), and those who assume that contrastive topics are actually foci, for example Krifka (1989) and Selkirk (1984). I follow the proposal made by Molnär (1998) that contrastive topics show a "combined effect of topicality and focussing" (p. 135) and that they occur in "obligatory combination [...] with an additional focus in the sentence" (p. 135). The main reason why I classify contrastive focus together with contrastive topic here, lies in the fact that both fulfil the criteria of contrastiveness and the criteria of movement out of the domain of information focus. Chapter 3 on stripping and chapter 4 on gapping take up the issue of contrastive topic and contrastive focus in more detail.

4.

The hybrid focus analysis of ellipsis

An immediate consequence of the Double Cycle-ISH in (1 la, b) for ellipsis is that sentence-bound grammatical processes, like gapping and stripping, and discourse-bound processes like VPE are treated differently in each of the information structural cycles. Consider gapping in (23a), Stripping in (23 b) and VPE in (23 c).

36

(23)

Ellipsis and focus: An

a. b. c.

introduction

Leon read The FACTS and Manny read The Great American NOVEL. Leon read The FACTS, but he has not read The Great American NOVEL. John said that Manny read The FACTS. But Mary - who knows that Manny has never read a book by Philip ROTH said that he HASN'T [read The Facts],

Gapping in (23 a) involves movement of contrastive remnants to the edge of the phase (as will be argued in much detail in chapter 4). Stripping in (23b) involves the displacement of a single contrastive remnant. In both cases, the contrastive constituents bear an Ε-feature that is detected in cycle 1. The presence of a formal feature requires that the respective phases enter the grammatical cycle. In each case, the second cycle identifies the formal feature(s) in the phase and sends it back to CHL to check these features before it can be interpreted at SSI. In gapping and stripping, the formal features are erased via movement of the remnant(s) to the edge of the phase (spec,vP). VPE on the other hand, can - at least in instances such as (23 c) (that is those that do not involve displacement) - be interpreted by the functional cycle per default without being sent back to CHL· The information focus is realized on the functional head HASN'T, as predicted by the PHP-rule in (16). The interpretation can access larger discourse structures and check the construction for accessibility. The question occurs as to how the DC-ISH treats pseudogapping, a construction that has been classified syntactically in the linguistic literature as a subcase of VPE (cf. Lasnik 1999a, Boeckx 2000, Boeckx and Stjepanovic 2001), as well as a subcase of gapping that leaves a contrastive remnant behind (Levin 1978, 1986, Kuno 1981). Let me reconsider example (5a), repeated here in (24): (24)

Manny read The Facts, but he DIDN'T read The Great NOVEL.

American

The information structural analysis of pseudogapping requires that the object remnant The Great American NOVEL in (24) is contrastively focused. That is, the DP carries an Ε-feature (F) that requires the phase hosting the object to pass through both cycles, checking off the uninterpretable feature by syntactic displacement to the left edge of the phase. After this movement operation, the smallest vP can be omitted.

The hybrid focus analysis of ellipsis

37

Thus, syntactically, that derives pseudogapping involves the omission of a vP constituent, however, information structurally, the construction must pass cycles 1 and 2 of the IS component and thus classifies together with gapping and stripping as a construction that isolates a contrastive remnant. The result of applying the DC-ISH to SBE and DBE is given in (25a, b): (25)

a.

b.

Sentence-Bound Ellipsis: The information structural function of SBE is to isolate of (narrow) contrastive foci. Discourse-Bound Ellipsis: The information structural function of DBE is to mark the elliptical material as anaphoric or given.

The research strategy that I am pursuing here derives SBEs within narrow syntax, to which the grammatical cycle (cycle 2) is applied. DBEs, however, are derived outside narrow syntax under the application of the discourse functional cycle (cycle 1) of the ISH. That is, the two accounts discussed at the beginning of this section, the functional grammar account and the discourse grammar account are represented by different cycles. 30 Recent analyses that have been given in the literature for ellipses as in (1) fall into two basic types: those in which all types of ellipsis are uniformly treated (either PF-deletion or Proform Account) and those in which they are not (Hankamer and Sag 1976, Williams 1977a, Chao 1988, Lobeck 1995, see Winkler and Schwabe 2003 for an overview of the diversity of approaches). In this study, I will provide evidence for the Hybrid Focus Account of Ellipsis that results as a direct consequence from the DC-ISH introduced in section 1.3.2. Establishing syntactic movement (movement of contrastive topic and focus) out of the ellipsis site as the relevant feature of the classification, the data in (23c) is accounted for by the proform account while the data in (23 a,b) and (24) are accounted for by a PF-deletion account, which will be specified in chapter 2. The hypothesis that forms the basis of the hybrid focus account of ellipsis is given in (26): (26) a.

b.

Hybrid Focus Hypothesis of Ellipsis: (first approximation) A PF-deletion account must be assumed for elliptical constructions in which a contrastive focus/topic is isolated by syntactic displacement. Α proform account is feasible when no movement is involved, as in the core cases of VPE.

38

Ellipsis and focus: An

introduction

That is, whenever syntactic displacement (A'-movement) out of the ellipsis site occurs, the moved constituent is interpreted as a contrastive focus/topic and the elliptical site must have syntactic representation for obvious reasons. When no movement is involved, as in the core cases of VPE, the construction can be better explained by the proform account where the verbal anaphora serves the function of marking the deleted string as given material. In this case, remnants are typically free to assume various discourse functions depending on the context. That is, the Hybrid Focus Hypothesis of Ellipsis in (26) can be captured structurally as in (27). (27) assigns a PF-deletion analysis to SBEs in which the remnants have undergone syntactic displacement for reasons of contrastive focus/topic, and it assigns DBEs a proform account that marks the elided material as highly accessible in the discourse context. (27)

Hybrid Focus Analysis of Ellipsis: a.

SBE:

[... [XPJCF/CT [Α-..t,

b.

DBE:

[ XP X [„ ·· - e

]]

]]

PF reads off the incoming phase whether it passed the functional or also the grammatical cycle. The remnants [XP;] of SBEs, which passed the grammatical cycle, are always assigned a contrastive focus or topic accent at PF, whereas α is phonologically reduced. In the case of DBEs, PF assigns PAs according to the PFIP-rule; that is, the PA on the functional head in VPE (and in some cases the PA on the subject) follows from the functional cycle of the DC-ISH. At PF, accents are assigned to these remnants according to discourse structure. The phrase a , on the other hand is an empty proform (e) with no phonological realization at PF.

5.

Conclusion

In this introductory chapter, I have proposed a nonuniform treatment of focus and of ellipsis. In particular, I have claimed that SBEs and DBEs form two natural classes of ellipses that differ with respect to their syntactic derivation and their surface semantic interpretation. I have proposed a hybrid focus analysis of ellipsis and suggested that it results naturally from the implementation of the syntax-SSI relation in grammar. The present proposal incorporates the idea that although the phenomena of

Distinguishing

sentence-bound

and discourse-bound

ellipsis: A preview

39

focus, intonation and information structure are interface phenomena with consequences at PF, they are syntax/semantics-driven. I further proposed a nonuniform treatment of focus. In essence, I presented a theory of focus that acknowledges the existence of at least two different types of foci: information focus and contrastive focus. As I showed above, information focus is connected to the so-called in situ focus defined over a certain domain (traditionally called unmarked or normal intonation), whereas contrastive focus (and contrastive topic) is crucially connected to the assignment of a formal feature [F] that is either checked by syntactic displacement or agreement. I proposed that information focus and contrastive focus are active in the derivation of an expression in different cycles, as spelt out by the DC-ISH. I made a conscious effort to keep this introductory chapter as brief as possible by concentrating on the implications of the interaction between syntax and surface semantics interpretation in a derivational model. What I did not do is offer an account of the role which the other interfaces play in the derivation of ellipsis. This is what I will do next: I will provide evidence for the derivational model of grammar that computes the syntax and the semantics/pragmatics of a given sentence in parallel feeding readily analyzed phases to PF. Thus, I concur with Jason Merchant (2001), who concentrated in his book on ellipsis on The Syntax of Silence. But this is not the whole story. The next chapter is about the hypothesis that meaning and syntax are computed in parallel and that the result provides us with the complementary component: The Sound of Silence.

6. Distinguishing sentence-bound and discourse-bound ellipsis: A preview Chapter 2 investigates previous proposals to ellipsis and to information structure, raises objections and finally proposes an account, referred to as Parallel Computation Account (PCA), that is a natural consequence of the D-model executing the division of labor between syntactic derivation and surface semantic interpretation. The central idea of the PCA is that syntax, semantics and phonology derive structures in parallel, thus resolving the dilemma of earlier proposals that resulted from the separation of the location of deletion and the interpretation of ellipsis before they emerged as silence at PF. In implementing parallelism and cyclicity into the system, the PCA relies heavily on the concept of phase. The differences between

40

Ellipsis and focus: An

introduction

SBEs and DBEs fall out from the DC-ISH at SSI and the parallelism constraint and anaphora constraint stated at the syntax-semantics interface. Considered from today's perspective, the PCA, which proposes that deletion (or in case of DBE, the application of the Phase-Invisibility Condition) happens in CHL in correspondence with the semantic component, reconciles the strict transformational position (Ross 1967, 1969, Postal 1970, 1972, Hankamer and Sag 1976) with the strict interpretive position (Wasow 1972, 1979, Shopen 1972, Williams 1977, Kennedy 2003) since structures are computed in parallel. While section 2.2 provides an answer to the question of whether deletion and deaccentuation is syntactic, semantic, pragmatic or phonological in terms of the PCA, sections 2.3 and 2.4 concentrate on the question as to whether PF interfaces with only the semantic/pragmatic component or also the syntactic component. More precisely, section 2.3 discusses Schwarzschild's (1999) theory of GIVENness and points out the conceptual and empirical problems that arise. It is shown that these problems do not only apply to general deaccentuation phenomena, but reoccur in theories of ellipsis that are based on semantic/pragmatic theories of focus which do not address syntactic dislocation phenomena. Section 2.4 offers initial evidence for the PCA by spelling out the derivation of intonation by phase showing that syntactic displacement has an immediate effect on intonation. Thus, the basic idea is close to Bresnan's (1971a, 1972a) original hypothesis that PF reads off syntax directly. However, the implementation is crucially based on the parallel computation of the phase. In developing this account, I lay out the derivational model of grammar paying special attention to the interaction of the computational system of human language with surface semantic interpretation and the PFcomponent in deriving elliptical constructions. Chapter 3 is the first of two chapters providing empirical evidence for the PCA of ellipsis concentrating on the differences between SBE and DBE. More precisely, I explore the syntax and information structure of VP-ellipsis in English and its less well-known instantiation(s) in German. There are three possible candidates, which I descriptively refer to as esconstruction, ^«c/z-ellipsis and J£er-ellipsis. Since these constructions - as to the best of my knowledge - have not been systematically discussed up to now, the better part of this chapter provides a thorough empirical investigation of their distribution, syntax and information structure. In doing this, I will address two related questions that have been central to all generative and information structural investigation of ellipsis: first, does

Distinguishing

sentence-bound

and discourse-bound

ellipsis: A preview

41

ellipsis have internal structure that is deleted in the course of the derivation, or is it a type of anaphora underlying the usual mechanisms of recovery of meaning without recourse to syntactic structure of the ellipsis site? Second, is ellipsis a means of contrastive focus marking of its remnant(s) or is it a means of givenness marking of the silent site? The evidence from German argues for the DC-ISH introduced in chapter 1.3.2. On the basis of the investigation in German, I will argue for the Hybrid Focus Account of Ellipsis: whenever syntactic displacement out of the ellipsis site occurs, the moved constituent is interpreted as a contrastive focus/topic and the ellipsis site must have syntactic representation, as in the case of Contrastive Remnant Ellipsis (CRE). The CRE comprises both the Contrastive Focus Ellipsis and the Contrastive Topic Ellipsis and is analyzed as an instance of Sideward Movement (cf. Nunes 1995 and subsequent work). When no movement is involved, as in the core cases of VPE, the construction can be better explained as an anaphoric construction that serves the function of marking the deleted string as given material by applying the Phase Invisibility Condition (cf. chapter 2). The esconstruction is the corresponding construction in German. In these cases, remnants are typically free to assume various discourse functions depending on the context. The chapter ends with a brief discussion of the question when, why and exactly where heads move and CRE takes place. Chapter 4 discusses the syntax and information structure of gapping. In particular, I show that gapping also classifies as an instance of CRE, in this case of paired contrastive remnants. The goal of the chapter is to show that the syntactic and information structural restrictions on gapping follow from the PCA and the DC-ISH. In section 4.2,1 discuss two contrasting analyses of gapping, namely the deletion vs. the Across-The-Board (ATB) movement accounts. I propose a Sideward Movement (SM) account of gapping (cf. Nunes 1995, 1998, 2001) that is basically recasting Johnson's (1996) ATB-movement account in a derivational model. In developing this account, I rely heavily on the prominent function of the phase in the PCA. Section 4.3 presents evidence for the SM-account. First, evidence for the assumption that gapping can involve coordination of vP-phases stems from the interaction of prosodic phrasing with scope ambiguities. Second, evidence for the existence of A'-positions at the edge of the vP that hosts contrastive gapping remnants in English stems from information structural considerations in connection with the dislocation of w/z-phrases and topicalized phrases in gapping. Finally, I provide initial data that shows that negation cannot be gapped in German. More generally, contrastively

42

Ellipsis andfocus:

An

introduction

focused elements in gapping must occur in the scope of an overt negation particle. Chapter 5 contains the conclusion of the research program developed in this study. By looking back, this final chapter summarizes the major hypotheses presented here and by looking ahead it sketches the directions for further research which emerge from them.

Chapter 2 Ellipsis at the interfaces: A proposal

1.

Introduction

In the traditional T-model of grammar, syntactic representations were the sole input to PF. In recent proposals that analyze ellipsis from an information structural perspective, however, two controversial theories can be distinguished: first, the syntactic theory of focus investigates the interation of syntax with information structure and phonology, assuming that notional pairs like focus and topic, or given and new have a specific grammatical implementation (traditional position defended, e.g., in Bresnan 1971a, b, 1972a, b, Chomsky 1972, Jackendoff 1972, Rochemont 1986, Winkler 1996, Zubizarreta 1998, among many others); second, the semantic theory of focus investigates the interation of semantics and pragmatics with PF, seeking generalizations without addressing the role of syntax in detail (von Stechow 1991, Kratzer 1991, Rooth 1992a, b, Krifka 1993, 2001, Schwarzschild 1999, among many others). Concentrating on the rules that actually determine whether deaccentuation or deletion of the phonological information can take place, the proponents of these two theories hold different views accordingly: The first group assumes that rules of focus and topic movement, as well as anaphoric interpretation, anaphoric deaccentuation and particularly the identity and accessibility conditions on ellipsis, hold at the syntax-information structure interface with the immediate effect of producing silence at PF (Lopez and Winkler 2000, 2003, among others). The second group proposes that anaphoric rules (like anaphoric redundancy rules and semantic identity conditions) hold at the semantics/pragmatics component and interact directly with PF without recourse to syntax (Rooth 1992a, Schwarzschild 1999, Hartmann 2000, 2003, Merchant 2001, 2003, 2004). 1 The Double-Cycle Information Structure Hypothesis (DC-ISH), built into the LF-component (see chapter 1,(11)), integrates both positions and is defended here. More concretely, in chapter 1 I have concentrated on the division of labor between syntax and information structure. I have proposed a D(erivational)-model of grammar which implements the grammatical and functional aspects of information structure at LF. In this chapter, I

44

Ellipsis at the interfaces: A proposal

investigate the derivation of ellipsis, concentrating on two central questions: First, where does elision actually take place?2 And second, where does semantic interpretation of the unrealized material take place? In answering these two questions, I will discuss different theories of ellipsis relating them to the generative enterprise today, raise objections with respect to the uniform and semantically and pragmatically driven PFdeletion accounts and finally offer a proposal that computes syntax, semantics/pragmatics and PF in parallel. This proposal, referred to as the Parallel Computation Account (PCA), relies heavily on the concept of phases as a device that implements cyclicity into the system. Note that the PCA is a necessary consequence of the D-model executing the division of labor between syntactic derivation and surface semantic interpretation in the form of the DC-ISH. With respect to ellipsis, the proposal spells out different aspects of the computation already implicit in the hybrid focus hypothesis of ellipsis. The central idea is that the two classes of ellipsis, SBE and DBE, fall out as a direct consequence of the derivation by phase guided by the interaction between syntax and SSI as specified by the DCISH. Thus, the functional cycle, which interacts with the pragmatics-PF interface, is crucial for the derivation of the DBEs, while the grammatical cycle, which interacts with the syntax-PF interface, is crucial for the derivation of the SBEs. The PCA proposes then that deletion in the case of SBEs (or alternatively, the application of the Phase Invisibility Condition in the case of DBEs) takes place in CHL after the interpretation of the phase has taken place in the semantic component. Thus, the syntactic and the semantic/pragmatic components, at which the identity and the anaphora constraints (cf. Sag 1976, 1977, Rooth 1992a, Tancredi 1992, Williams 1997, among many others) are formulated, work together in deriving phonological silence at PF. In section 2.2, I introduce the PCA. In section 2.3, I discuss the proposals which depend on the interaction of pragmatics and phonology, before I offer initial evidence for the PCA in section 2.4.

2.

Deletion vs. deaccentuation

In recent publications on ellipsis and information structure (see, for example, the publications in Schwabe and Winkler 2003), at least two controversial theories can be distinguished: approaches that assume that ellipses and their deaccented counterparts are fully syntactically repre-

Deletion vs. deaccentuation

45

sented before they are deaccented or deleted and those that do not. The first approach is referred to as the Complete Syntactic Representation account (CSR account) and the second as the No Syntactic Representation account (NoSR account). A third approach, introduced by the D-Model in chapter 1 and here referred to as the Parallel Computation Approach (PCA), proposes that the parallel computation of syntax and semantics offers a solution to the problem that all previous theories have been confronted with, namely the separation of the location of deletion of syntactic material from the location of its interpretation at the interfaces. More precisely, the PCA, developed on the basis of recent work by Chomsky (2000, 2001, 2004, 2005), targets the interaction between syntax and SSI with the effect of producing input structures to PF in parallel. The DC-ISH interprets the incoming phases cyclically and separates those that are sent on to PF (cycle 1) from those that still contain uninterpretable features and are sent back to CHL· Concentrating on ellipsis: the interpretation happens at the syntaxsemantics interface, where also the conditions on ellipsis are formulated. The idea is that the result of the interpretation (topic/focus readings, parallelism of interpretation, anaphoricity, redundancy, scopal readings) is marked in the form of features on the respective phase that is sent to PF. PF derives phonological structures, phase by phase, and in the case of ellipsis, silence. One suggestion of how to implement the difference between deaccentuation and deletion at PF stems from Klein (p.c.), who proposes that deaccentuation results if the suprasegmental level is null (SO), while deletion results if both the segmental and the suprasegmental levels are null (SS0). 3 The PCA implements this idea of viewing the difference between deletion and deaccentuation in terms of the Null Suprasegmental Feature Hypothesis and the Null Segmental Feature Hypothesis in section 2.2.2.3 below.

2.1. Previous accounts In the following, I will briefly investigate the most relevant approaches to ellipsis, classifying them based on two parameters, formulated as questions in (1) below: (1)

i.

Where do deaccentuation and deletion take place? Or more precisely, where does grammar assign the features that are interpreted as [± segmental] / [± suprasegmental] at PF?

46

Ellipsis at the interfaces: A proposal

ii.

Where does semantic interpretation take place? And more precisely, where are core semantic and anaphoric processes formulated and checked?

The questions in (1) not only address the general interface problem of where the features [+ segmental] / [± suprasegmental] are assigned and where semantic interpretation takes place, but they address two more specific issues: what conditions cause deaccentuation or deletion, and how do they interact with information focus and contrastive focus/topic. Let me start with a brief discussion of the CSR account. Each approach establishes a relation between two different levels, the level of deletion and the level of interpretation, as summarized in (2) and graphically represented in (3). (2)

Complete Syntactic Representation Accounts: i. The PF-deletion account (based on Tancredi 1992): The level of deletion is PF, the level of interpretation is PF with access to the semantics/pragmatics component. ii. The syntactic deletion account (based on Sag 1976, 1977): The level of deletion is Surface Structure, which branches off the level of interpretation, called Shallow Structure (SS). iii. The Delta-interpretation account (based on Wasow 1972): The level of deletion is ZYee/;y-Structure, interpretation involves reconstruction.

The picture in (3) provides an overview of the three approaches in one glance. The locations of deletion and interpretation are circled. (3)

CSR Accounts I)S

Syntactic deletion account

Delta-interpretation account

/

PF-deletion account

Deletion vs. deaccentuation

47

The NoSR-account of ellipsis is summarized in (4): (4)

No Syntactic Representation Accounts: i. The Proform account (cf. Hardt 1993, 2003): Proform insertion occurs in syntax, interpretation at LF. ii. The Proform account and reconstruction (cf. Lobeck 1995): Proform licensing in syntax, interpretation and reconstruction at LF.

The goal of both the CSR and the NoSR accounts is to derive structures that are interpreted at PF as phonologically null. The central question, then, which guides the discussion of different theories in this chapter and the discussion of VPE, stripping and gapping in chapters 3 and 4 is the following: are the different theories empirically distinguishable? The PCA predicts that since syntactic derivation and interpretation proceed in parallel, that is, in terms of phases, the accounts in (2i, ii) and (4ii) fall together. More crucially, if this turns out to be the case, research objectives must be formulated differently: the question of exactly how the work is divided between the different components must be more forcefully addressed. Let me look at the CSR accounts, more closely answering the questions posed in (1) as we go along. The CSR account has basically three instantiations, each of which targets different components at which deletion and interpretation take place. Note that although all three theories propose that the complete sentence is derived by the computational system of human language, they employ different hypotheses of how exactly the process of the nonrealization of phonological material at PF should be understood. The PF-deletion account in (2i) is based on the Phonological Reduction Hypothesis (e.g., Tancredi 1992, Chomsky and Lasnik 1993, Klein 1993, Lasnik 1999a) given in (5): (5)

Phonological Reduction Hypothesis (PRH): Elliptical sentences are formed by a rule of the PF-component that deletes the phonologically redundant information which is characterized by a "distinguished low-flat intonation". (Chomsky and Lasnik 1993: 564)

The proponents of this theory assume that deletion takes place at PF and that deletion or deaccentuation follows from identity conditions stated at PF

48

Ellipsis at the interfaces: A proposal

itself or that the relevant constraints follow from focus considerations (Tancredi 1992, Klein 1993, Hartmann 2000, 2003, Merchant 2001, Schwabe 2003, Romero 2003), as represented by the dot-dashed line in (3). Tancredi's (1992) major claim is that the constraints on VP-deletion can be reduced to those on deaccentuation, where deaccentuation is a process that takes place at PF. However, the process of deaccentuation is triggered by parallel interpretation based on the concept of focus-related topic marking of presupposed material, as in (6). Italics signal deaccentuation, strikethrough signals deletion: 4 (6)

John likes flying planes but Bill doesn't like flying planes. a. ... but Bill doesn't [ w like [NP flying planes]]. a'. ... but Bill doesn't [Vp like flying planes]]. b. ... but Bill doesn't [ w like [5 PRO flying planes]]. b'. ... but Bill doesn't [vp Hke P-RO flying planes]]. (examples from Sag 1976)

Following the basic insights of Sag (1976, 1977), Tancredi observes that although the first conjunct is two-ways ambiguous between an NPcomplement reading and a gerund reading, the resulting elliptical examples in (6a', b') are also only two-ways (and not four-ways) ambiguous. Tancredi's particular claim is that similar conditions hold not only with respect to ellipsis but also with respect to deaccentuation, as in (6a, b). Furthermore, he investigates strict and sloppy readings of pronouns in deletion and deaccentuation contexts and observes that similar readings result in both reduction contexts. Therefore, he claims that "VP-ellipsis is no more than an extreme case of deaccenting where a VP ceases to be audible altogether" (Tancredi 1992: 120). In section 2.2.2.3 below, I investigate Tancredi's hypothesis that the conditions on VP-ellipsis can be reduced to those on deaccentuation more closely, concentrating on phonological parameters. Furthermore, I address an aspect hitherto not discussed in the literature on ellipsis, namely that deaccentuation and deletion differ with respect to the degree of accessibility of the antecedent, in the sense of Ariel (1990). The syntactic deletion account in (2ii) (also known as the strict transformational position), here modeled after Sag (1976, 1977), is based on the Shallow Structure Hypothesis, stated in (7).

Deletion vs. deaccentuation

(7)

49

Sag's Shallow Structure Hypothesis: Given ... a theory that countenances deletion rules, it is certainly the case that semantic interpretation must take place at a level prior to deletion. The picture of linguistic systems that emerges, then, is one that countenances a level which we might call "shallow structure" ... which serves as the input to rules of semantic interpretation. (Sag 1977: 68)

Thus, Sag proposes that interpretation of (6a', b') takes place at the level of shallow structure, and that the deletion rules and stylistic rules apply in the derivation of surface structure, which could be understood as PF in modern terms (see Sag's model in (12) below). The Delta-interpretation account in (2iii) (also known as the strict interpretive position; see Bach 1970) was originally proposed by Wasow (1972), and later elaborated by Williams (1977a). A modified version is introduced here for DBEs. It assumes that deletion (here understood as replacement of the terminal elements by deltas Δ) takes place at Dstructure, and interpretation is achieved via reconstruction, as schematically represented by the dotted line in (3). The Delta-interpretation account is based on Wasow's (1979: 105f) Empty Structure Hypothesis (ESH), stated in (8): (8)

Wasow's Empty Structure Hypothesis: Null anaphors have all the structure of their antecedents, lacking only phonetic material. This hypothesis presupposes that lexical insertion is always optional. This makes it possible to generate structures with all of their normal syntactic properties, but lacking any phonological or semantic material. If the anaphora rule or rules are allowed to associate such an empty structure with an antecedent, then the reading of the antecedent can be associated with the empty nodes. Surface structures containing uninterpreted empty nodes would be regarded as semantically anomalous. (Wasow 1979: 109)

According to the ESH, sentence (9a) will be interpreted like (9b):

50

Ellipsis at the interfaces: A proposal

(9)

NP Aux

Anna will (9)

b.

go

to

the lecture

Mary

might

Anna will go to the lecture and Mary might go to the lecture.

The ambiguity displayed by (6) would be solved by the ESH providing two different structures, as in (10a, b): (10) a. b.

John likes flying planes but Bill doesn't like flying planes. ... but Bill doesn't [VP [Δ] PRO [Δ] [Δ]]. ... but Bill doesn't [νρ[νΔ] [ορΔΔ]].

This brief discussion of the CSR approaches exemplifies the problem with which the deletion accounts in (2i, ii) are confronted, namely, that deletion of syntactic material and its interpretation occurs at separate levels. This difficulty has been preprogrammed in the structure of grammar assumed in the syntactic literature since the Extended Standard Theory (Chomsky 1965), in which LF and PF branch off the syntactic level (S-structure). Consequently, a deletion based theory can only do one of two things: it can either preserve the T-model based on the concept of modularity and state all necessary constraints on deletion directly at the level at which deletion takes place (e.g., initial proposal of Tancredi 1992) or it can change the model in such a way that ellipsis is interpreted before deletion takes place (e.g., Sag 1976, 1977, and also the PCA discussed below).5 There is still another way in which the T-model can be preserved, namely by choosing an interpretive account. There are two general proposals. Wasow's (1979)

Deletion vs.

deaccentuation

51

account, in which the structure is fully syntactically represented, and the accounts mentioned in (4). The NoSR accounts in (4) assume that ellipsis is not syntactically represented. There are basically two instantiations. The first is the dynamic theory of ellipsis interpretation, which assumes ellipsis is licensed in syntax as an empty proform and interpreted at the semantics interface in correspondence with discourse representation structures and focus structure (Hardt 1993, 1979, 1999, 2003, Asher, Hardt, and Busquests 2001, Hardt and Romero 2004), as in (11). (11)

Proform Account: John likes flying planes but Bill doesn't [ype]. Two ways of reconstruction at LF: a. John [likes [PRO flying planes]] but Bill doesn't [Vp like [s PRO flying planes]]. b. John [likes [ DP flying planes]] but Bill doesn't [yp like [ DP flying planes]].

A variant of this approach, which assumes the proform hypothesis plus reconstruction (4ii) and which is particularly concerned about the syntactic licensing conditions of empty categories, is proposed by Chao (1988), Lobeck (1995, 1998, 1999) and Lopez and Winkler (2000) for VPE, and by Chung, Ladusaw, and McCloskey (1995) for sluicing. Both variants of the proform account assume that the proform is licensed in syntax and interpreted at LF. In the P&P approach, in early Minimalism (e.g., Lobeck 1995, Lopez 1995) and in computational work (e.g., Hardt 1993), the prevailing assumption has been that VPE in English is an anaphoric phenomenon. I reformulate this claim in Ariel's (1990) Accessibility Theory, which she developed for nominal anaphors, and propose that an empty verbal proform functions as a high accessibility marker in discourse (cf. Ariel 1990: 58, fn3: 224). According to this hypothesis, VPE constitutes a verbal anaphor that has similar effects on the focus structure of a sentence as other anaphoric elements. The functional cycle of the DC-ISH interprets the VPanaphor with respect to its highly accessible verbal antecedent in discourse and sends the phase, which is coextensive with the VP-anaphor, to PF, where it remains silent. Additional material is added to the phase in the derivation of the complete sentence. The remaining overt element (viz. the licensing functional polarity head Σ 0 in Lopez and Winkler 2000) governs

52

Ellipsis at the interfaces: A proposal

the empty category pro. It is most often accented and cannot occur in its reduced form. At PF, it is assigned a pitch accent by the PHP-Rule, and at LF, it is interpreted as information focus. In Lopez and Winkler (2000), we argue at length for the possibility of subsuming VPE under other default accent cases (see Ladd 1980, 1996). As will be further discussed in chapter 3, there is also the possibility of a contrastive focus interpretation of the functional head. As is the case with other information focus constructions, the interpretation is dependent on the discourse structure, as predicted by cycle 1 of the DC-ISH. In more recent years the paradigm has shifted. It has turned from licensing theories of ellipsis in an essentially representational model of grammar (as those mentioned above - Lobeck 1995, Lopez and Winkler 2000) to PF-deletion theories in a derivational model (cf. discussion in chapter 1). The hypothesis that ellipsis is a process of phonological reduction located at PF, as suggested by the PRH in (5), has the advantage that it leaves the syntactic and lexical representation of the sentence completely intact for the interpretation at LF. We saw in (2) that there are various other hypotheses of how exactly the deletion process that leads to phonologically unexpressed structures at PF could be understood. The PFdeletion approach based on the PRH in (2i) proposes that ellipsis is the endpoint on a continuum ranging from deaccentuation to complete silence (Tancredi 1992, Chomsky and Lasnik 1993, but see Winkler 1997 and below). PF-deletion, as understood in (2ii), is achieved in the derivation of surface structures. That is, Sag's (1976, 1977) account classifies as a syntactic deletion approach which uses identity of interpretation as a precondition to deletion in syntax with the effect of producing silence at PF. Furthermore, this proposal requires that the grammatical model be changed in such a way that the interpretive component feeds Surface Structures, as in (12). Finally, the Delta-interpretation account in (2iii) derives the effect of phonological silence by assuming that ellipsis has abstract features of phonetic and phonological representation, and that grammar contains a means of blocking pronunciation (cf. Wasow 1972, 1979, Kennedy 2003). A variant of this proposal is that late insertion of phonological features is blocked for the elements in the ellipsis site (Halle and Marantz 1993).

Deletion vs. deaccentuation

(12)

53

Sag's (1976, 1977) Syntactic Deletion Account:

Base Rules

Initial Phrase Markers Transformation

Shallow Structures Deletion Rules, Stylistic Rules, etc.

/ / ^

Surface Structures

\ \

Rules of Semantic Interpretation Logical Forms

Historically, the PCA proposed below brings together both Sag's syntactic deletion account, in which interpretation is a precondition for deletion, and Wasow's Delta-interpretation account, in which "null anaphora underlie the same restrictions as other anaphors with the difference that null-anaphors are generated with no phonetic realization" (Wasow 1979: 105). Thus, the PCA constitutes a synthesis of those approaches that assume that deletion occurs in syntax but is driven by semantic interpretation, thereby causing silence at PF. That is, the PCA reconciles the strict transformational and the strict interpretational positions and thus solves the dilemma resulting from the separation of deletion and interpretation in earlier proposals.

54

Ellipsis at the interfaces: A proposal

2.2. The parallel computation account The Parallel Computation Account is a natural consequence of the derivational conception of grammar in the Minimalist Program (Chomsky 2000, 2001, 2004, 2005) that computes syntax and semantics cyclically, or in phases. In this model both modules, syntax and semantics, derive structures in parallel. Thereby, it is possible to collapse both the location of deletion and the interpretation of ellipsis with the effect of causing silence at PF. The PCA proposes that incoming phases are interpreted before they are read by PF. Moreover, identity and parallelism constraints apply in the interpretation process of phases and not directly at PF, as for example assumed by the PF-deletion account which is based on the phonological reduction hypothesis. The PCA is implemented by the parallel effort of syntactic derivation and semantic, pragmatic and phonological interpretation as shown in the elaborated D-model in (13) below.

Deletion vs.

(13)

Elaboration of the D-Model of Grammar

deaccentuation

55

56

Ellipsis at the interfaces: A proposal

More concretely, the PCA proposes that the constraints on ellipsis are the result of the DC-ISH applying at the syntax-semantics interface, which figures prominently in the middle column in (13). The idea is that the DCISH applies to the smallest phase of each sentence, separating those that involve unchecked features from those that don't. The basic claim of the DC-ISH is that phases that do not involve Ε-features are automatically interpreted by the in situ hypothesis (see chapter 1 (11a)) as information focus, while those phases that do are sent back to the CHL for erasure of the Ε-features by the displacement hypothesis (see chapter 1 ( l i b ) ) . Then, the fully derived and interpreted phase is sent to PF for phonological interpretation. A first concrete proposal of how to derive PF by phases is provided for the famous Bresnan (1971a, 1972a, b) examples in section 2.4.1. The PCA proposes that the difference between SBE and DBE falls out from the DC-ISH stated at the syntax-semantics interface. More explicitly, accounting for the parallelism of interpretation in SBE and DBE has two different values: in SBE the parallel interpretation is required of remnants that occur in a specific information structural relation to each other (see the Contrastive Topic and Focus Principle in (24)) for the construction to be felicitous. In DBE parallel interpretation is required of the anaphoric or given VP/IP constituent that holds across utterances (cf. (15a, b)). The claim is that in the PCA this difference falls out from the appropriate implementation of the DC-ISH at the syntax-semantics interface, which also takes recourse to pragmatics. The task of PF, then, is to derive the most appropriate and economical phonological form of each incoming phase. Two cases are relevant for an initial discussion of ellipsis: first, a case of SBE, here gapping, as in (14a, b), and second, a case of DBE, here VPellipsis, as in (15a), and sluicing, as in (15b). (14)

a. b.

She wrote long [...] letters, which she sent to her sister and she to my mother. (Ihabrala 1975 Heat and Dust) It's probably true ... that the system is so hierarchical that only the assistant manager can talk to the manager and the manager to the general manager ... (Perm Treebank, #...wsj_0037.mrg 805)

Deletion vs. deaccentuation

(15)

57

a.

("Even a joke should have some meaning - and a child's more important than a joke, I hope. You couldn't deny that even if you tried with both hands." "I don't deny things with my hands," Alice objected.) "Nobody said you did," said the Red Queen. "I said you couldn't if you tried."

b.

Co-Pilot schlägt Passagier Axt in den Kopf [...] co-pilot hits passenger axe in the head Buenos Aires - Bislang weiß niemand warum: Buenos Aires so far knows nobody why

(Carroll 1961 Through the Looking

Glass)

Als die Maschine [...] in der Luft war, begann Pablo Moreira gegen die when the plane in the air was began Pablo Moreira against the Tür des Cockpits zu treten... Schließlich überwältigte ihn ein Steward, door of the cockpit to kick.. Finally overpowered him a steward, ein Fluggast und der Co-Pilot. Letzterer griff nach einer kleinen Axt, a passenger and the co-pilot. The last-mentioned took a little axe. und schlug dem 28-jährigen Banker aus Uruguay in den Kopf. and hit the 28 year old banker from Uruguay in the head (Spiegel-online, 11.02.2002)

All four examples in (14) and (15) are naturally occurring examples and exhibit the complexity of the required analysis. To mention only the major differences: The gaps in (14a, b) must occur in coordinate sentences leaving behind contrastively accented remnants with correspondents in the first conjunct. In addition, gapping cannot occur across discourse. This is different with VPE and sluicing. For example in (15a), two instances of VPE occur in successive sentences, both referring to different antecedent VPs. 6 The first VPE occurs in "Nobody said you did" and takes the highly accessible VP "deny things with my hands" as antecedent, requiring a sloppy interpretation of the possessive pronoun. The second VPE occurs in "I said you couldn't if you tried," which refers to an earlier utterance of the Red Queen "You couldn't deny that even if you tried with both hands." Note that the PF-deletion theory of (2i) could not account for the fact that VPE can obviously select an antecedent VP as referent further back in the discourse. 7 Regarding the information structure of this dialogue, the antecedents of the VPEs are wrongly considered highly accessible (or given) by the speaker. The subject remnants are contextually given, and pitch accents are realized on the functional elements did and couldn't, and on the verb tried.

58

Ellipsis at the interfaces: A proposal

The sluicing example in (15b) is another interesting case that challenges the PF-deletion account. Although the interpretation of the sluiced sentence "Bislang weiß niemand, warum:" as so far nobody knows why the co-pilto hit a passenger with an axe is readily available from the headline (and explainable under a PF-deletion account), the intended reading so far nobody knows why Pablo Moreira began kicking the door of the cockpit requires a cataphoric interpretation and is only available under an anaphoric account of ellipsis. Taking up the discussion of SBE vs. DBE in sections 2.2.2.2 and 2.2.2.3 below, I will first briefly address the issue of the phonological representation of accentuation and deaccentuation in autosegmental theory. I am aiming at an explanation of the derivation of the phonology of the phase. The CHL component starts out building the phase, then it sends it to SSI, where it is interpreted with respect to information structure and semantic structure proper before it is sent on to PF. The idea is that PF remains maximally lean with respect to syntactic operations and merely translates the features of the incoming phase.

2.2.1. Accentuation and deaccentuation

in autosegmental theor}>

The phonological framework developed by Pierrehumbert (1980) and the modifications in Beckman and Pierrehumbert (1986), Pierrehumbert and Hirschberg (1990), Silverman et al. (1992), and Beckman and Ayers (1997) are based on the idea that intonational tunes are structured strings of only two phonemic tones, H(igh) and L(ow). An intonation phrase may consist of several possible sequences in which the pitch accent (H*, L*, L+H*, L*+H, H*+L, H+L*) is associated with the stressed syllable of a lexical element to which it is assigned (according to certain rhythmical and focus structural considerations) and functions as an anchoring point for the fundamental frequency (f 0 ) contour. Another fixed point is the phrase final boundary tone (H%, L%) and optionally an H% initial boundary tone. The phrase accent (or phrase tone) controls the fo between the last pitch accent and the boundary tone. The inventory of accents and tones of an intonational contour in English is summarized in (16).

Deletion vs. deaccentuation (16)

59

Inventory of tones of an intonational contour in English: T*: two monotonal pitch accents: H*, L* four bitonal pitch accents: L+H*, L * + H , H * + L , H + L * T": two phrase accents: H", L" T%: two boundary tones: H % , L %

Pierrehumbert's (1980: 29) original proposal assumes that well-formed tonal sequences are generated by the following finite-state grammar: (17)

Pierrehumbert's (1980) Finite-State Grammar: Boundary Pitch Accents Phrase Tone Accent

Boundary Tone

T w o major modifications to Pierrehumbert's original system have been proposed by B e c k m a n and Pierrehumbert (1986): first, the H*+H" pitch accent has been eliminated f r o m the set of seven tones in (17); and second, t w o phrasal levels have been suggested over which the intonational contour is defined, the intonational phrase (ip) and the intermediate phrase (pp). The idea is that an ip is composed of one or more phrase-like prosodic units that are smaller than the ip and larger than the prosodic word. These intermediate phrases provide a structure for an intonational phrase, as in (18). (18)

[ip (pp)---(pp)]

60

Ellipsis at the interfaces: A proposal

More precisely, the intermediate phrase consists of one or more pitch accents, plus a simple high or low tone, namely the phrase accent, which marks the edge of the intermediate phrase. The phrase accent is now defined as the terminal tone of the intermediate phrase, while the boundary tone marks the edge of the intonational phrase. Let me consider the intonation of the relevant examples in the next subsections.

2.2.2. Sentence-bound

ellipsis

Let me turn to gapping first, of which an example is given in (20), and concentrate on the three questions stated below. (19)

(20)

i. ii. iii.

Does the gapping site have internal structure? How is gapping interpreted in the semantic component? What role does information structure play in the structural representation and interpretation of gapping? She wrote long [...] letters, which she sent to her SISTER and SHE to my MOTHER.

The first question - does the gapping site have internal structure - can be straightforwardly answered in the affirmative. Two arguments underlie this claim: one based on the Parallel Structure Hypothesis, the other based on Island Constraints (Ross 1967). The first argument is based on the assumption that if the antecedent of a gap in the first conjunct contains a wh-trace, the gap itself contains a wÄ-trace, as is the case in (21) (missing material appears in strike-through): (21)

She wrote long letters;, whichj [she k [sent t,| to her SISTERj] and [SHEj [sent tL] to my MOTHER].

The second argument is based on the observation that gapping is subject to island constraints and must therefore contain syntactically represented material, just like w/z-constituents. Example (22a) exemplifies gapping of the verb asked and sluicing of we wrote in the second conjunct. The ungrammaticality in (22b) results from the violation of the Complex NP Constraint.

Deletion vs. deaccentnation

(22)

a. b.

61

SHE asked which LETTERS we wrote and HE which BOOKS. *SHE discussed my question which LETTERS we wrote and HE which BOOKS.

It is not possible for a w/z-phrase to be moved out of a complex NP in (22b). Island constraint violations are generally assumed to constitute a strong argument for movement, and if movement out of an elliptical construction is barred, it is an argument for the assumption that the elided material still has a syntactic effect. Now let me turn to questions (19ii-iii) above: How is gapping interpreted, and what role do focus and information structure play? Consider (14b), the gapping example found in the Penn Treebank (cf. Marcus, Santorini, and Marcinkiewicz 1993).8 A typical pitch extraction contour is given in figure 1. The contour shows that there are four focus accents in the utterance: a falling accent on ASSISTANT and on MANAGER in the first conjunct, and a falling accent on MANAGER and on GENERAL manager in the second conjunct. Beckman and Ayers' (1997) notation is given in (23) below (the reflex of pitch accents is signaled by capitalization):

62

Ellipsis at the interfaces: A proposal

(23)

H* L" H* L" L% only the ASSISTANT manager can talk to the MANAGER H* L" H* L" L% and the MANAGER to the GENERAL manager •Hal

' AmK· Γ*. Tt, H»bi La»tyun.VrAf fcU

Ufa ti* hW

Ü*m *J*J üLSiaiid s a i Jsd aiHiai si I . LfO' 1 ν i. •

• • :: ϊ: V q : . Ε

'J Ε 0 tlfl IH BCfl 4 Γ L

I

αΰ i~L^Lj

•.cm· L ί r : ItJ qrtttjgal aa^üjif

·

1 Ol'»:·;'f.

•« sali.·!

Μ

[iL

Figure 1.

The PCA makes it possible to connect the syntactic derivation of SBE to the surface semantic interpretation of SBE and the specific information structural relation of the remnants. The underlying idea is that in addition to a parallel syntactic structure, gapping also requires a parallel focus structure, as expressed in the Contrastive Topic and Focus Principle in (24) (cf. also Lopez and Winkler 2003 for a slightly different version): (24)

Contrastive Topic and Focus Principle: In gapping, the first remnant is a contrastive topic, the second remnant a contrastive focus. The gapped elements must be given.

Deletion vs. deaccentuation

63

The principle in (24) applied to the manager example in (14b) provides the following picture: there are parallel topics and foci, and the remnants (MANAGER, GENERAL manager) occur in a contrastive relationship to their correspondents in the first conjunct (ASSISTANT manager, MANAGER). A straightforward test which shows that the remnants of gapping constructions behave like a pair-list answer to a multiple w//-question that has an exhaustive topic-focus structure, as originally stated by Kuno's (1982) Sorting Key Hypothesis, is given in (25). 9 The answers in (25) also show that the modal verb can and the matrix verb talk are actually redundant: (25)

A: B:

Who can talk to whom? The ASSISTANT manager can talk to the MANAGER and the MANAGER can talk to the GENERAL manager.

The w/z-phrases in (25) specify the sets of alternatives from which the relevant set must be chosen; in the case of (25B), the answer specifies two pairs, namely the pair assistant manager and manager and the pair manager and general manager. In each pair, the first mentioned element represents the sorting key (or topic) and the second the corresponding focus. Note that up to this point, I have intentionally ignored the interpretive function of only in (14b). Let me look at it more carefully in (26). (26)

It's probably true ... that the system is so hierarchical that only the ASSISTANT manager can talk to the MANAGER and the MANAGER to the GENERAL manager ...

It is generally assumed that the focus sensitive particle only is associated with the focus (cf. semantic work ranging from Jacobs 1983a, b, 1986, over Rooth 1985, 1992a, and Tancredi 1990, to von Stechow 1991, and syntactic work ranging from Anderson 1972, over McCawley 1996 and Kayne 1998, to Biiring and Hartmann 2001). That is, only is a focus particle that takes as its associated focus a prosodically marked element that is implicitly contrasted with alternatives, as for example in (27): (27)

a. b.

Only JOHN came. John only introduced BILL to Sue.

(but not BILL) (but not his MOTHER)

64

Ellipsis at the interfaces: A proposal

Under the PCA, the syntactic derivation is essentially based on the derivation of phases. Thus, I will explicitly argue against a derivation of (26) that assumes that syntax derives complete syntactic representation of IPs, as given in the simplified structure in (28). Note that the representation follows Rooth's (1985: 28) original assumption that only is part of the DP constituent: (28)

S

S

DP

Μ

only the ASSISTANT manager

and

VP

S

DP

Μ

VP

can talk to the eafy the eaa talk to the GENERAL MANAGER MANAGER manager

In (28), deletion would have to affect the focus particle only, the modal verb can and the main verb talk in the second conjunct. The question that I have not addressed so far is: What is a possible paraphrase of the manager example as represented in (28)? A possible paraphrase is given in (29): (29)

PI:

As for the MANAGER, only the ASSISTANT manager can talk to him; and as for the GENERAL manager, only the MANAGER can talk to him.

If the paraphrase in (29) models our interpretation, then this sentence contains a pair of contrastive topic/focus phrases, as well as two foci which are associated with only. However, for most informants example (26) allows only the paraphrase in (30):

Deletion vs. deaccentuation

(30)

P2:

65

There are only two pairs (x, y) with the property that -i can talk to y - namely, the pair (assistant manager, manager) and the pair (manager, general manager) and no other pair.

We can see immediately that the paraphrases PI and P2 actually differ when we test them with an additional topic/focus pair, such as the pair (assistant manager, executive director). This pair is excluded by the paraphrase P2, but not by PI. The relevant observation is the following: the reading P2 cannot be explained under an account that assumes the deletion of the focus sensitive particle as represented in (28).10 However, it can be explained under the PCA, which is based on the displacement hypothesis (cycle 2 of the DCISH). The displacement hypothesis proposes that contrastive topic/focus phrases must be moved to an A'-position in the parallel derivation of the syntax and semantics of the phase. A detailed account of gapping constructions is given in chapter 4. Here, I introduce only the core idea, as it is derived from the DC-ISH. The PCA, then, requires for SBE the parallel syntactic derivation of phases, and the parallel information structural interpretation of the dislocated elements (remnants) deriving the appropriate SSI. The condition that requires the displacement of the remnants out of the vP-phase is defined in (31). (31)

Phase Impenetrability Condition: In phase α with head H, the domain of Η is not accessible to operations outside a , only Η and its edge are accessible to such operations. (Chomsky 2000: 108 (21))

The Phase Impenetrability Condition (PIC) requires that constituents like topic and focus constituents that may be subject to further movement operations and which play a role at SSI must be moved to the edge of the phase. The PIC applied to example (26) provides the structure in (32): (32)

[vP1 the manage^ [vP2 to the general managerj [vP3 tj talks tj]]]

Anticipating the gapping analysis in chapter 4, the proposal there is the following: Independent evidence suggests that gapping involves the coordination of vPs and not of sentences (see Johnson 1996 as well as Lopez and Winkler 2003). Thus, gapping is derived by two independent

66

Ellipsis at the interfaces: A proposal

movement operations: topic/focus movement of the remnants into a spec,vposition with subsequent movement of the vacated vP. More precisely, the derivation of gapping is argued to involve a movement operation that copies the vacated vP of the second conjunct into the first conjunct. Applied to (32), the vP 3 is copied into the first conjunct, where it merges with the correlates, the assistant manager and the general manager. The process, then, that marks the vP 3 as identical to its corresponding phase in the first conjunct must be understood as a process taking place between syntax and SSI, whereas the process which marks the vP 3 in the second conjunct as phonologically null is a Chain Reduction process that suppresses the segmental and suprasegmental features of the lower copy in the derivation of PF. (33)

Chain Reduction: In syntax, the lower copy in a chain is marked [- segmental] and [- suprasegmental], which is translated as phonological silence at PF (SSO).

Turning to (26) again, the focus sensitive particle only in (26) is not adjoined to the focused DP assistant manager as assumed by Rooth (1985), but is adjoined to the higher vP and functions as a quantifying sentential adverb in the sense of Jacobs (1983a) and Biiring and Hartmann (2001). From this position, only takes scope over the vP-coordination and, thus, over the pairs (ASSISTANT manager, MANAGER) and (MANAGER, GENERAL manager), providing the correct interpretation, in which the communication between the pairs is fixed. The question addressing the issue of the division of labor at the interfaces is straightforwardly answered. The main work load in the derivation of phonologically silent structures falls on the syntax-semantics interface, and more specifically, on syntactic operations interacting with information structure at LF, displacing the contrastive remnants to the edge of vP and marking the vacated vP (lower copy) as deletable at PF. The idea is that the PF-component interprets the incoming phase directly. In SBEs, only the edges of the vP-phases are phonologically represented; the vacated phase is a syntactic copy which is reduced by (33).

Deletion vs. deaccentuation

2.2.3. Discourse-bound

67

ellipsis

In this section, I investigate the DBEs. The PC A based on the DC-ISH makes three major claims: (i) The PRH in (5) does not apply to DBE. That is, the constraints on VPE cannot be subsumed under those on deaccentuation. (ii) The location of the conditions on deaccentuation and deletion is the pragmatic component corresponding with the functional cycle of the DC-ISH. (iii) If the accessibility condition on VPE is fulfilled, the syntax-semantics interface marks the complete phase as invisible ([- suprasegmental] and [- segmental], abbreviated as (SSO)). This idea is implemented by the Phase Invisibility Condition (PhIC), which marks the phase as a proform, in the sense of Wasow's (1972) ESH. In the case of deaccentuation, however, the elements in the phase are merely marked as [- suprasegmental] (SO) and the PhIC does not apply. Let me turn to the first claim and consider the paradigm in (34) (cf. Tancredi 1992: 28): (34)

a. b. c. d. e.

Manny believes he is brilliant and Leo does too. ... because Leo does. ... and Leo believes he is brilliant. ... and Leo believes he is a smart guy. ... because Leo believes he is a smart guy.

Tancredi's central hypothesis is that the case of VP-ellipsis in (34a, b) is a more extreme case of the deaccentuation which can be observed in the second occurrence construction in (34c) and in such cases as given in (34d, e). With respect to examples like (34d, e) Tancredi claims that "identity is not required for deaccenting" (p. 33). Before reporting on three studies on deaccentuation and deletion, let me briefly consider the phonological parameters of these phonological processes. Departing from an earlier discussion (see Winkler 1997), I propose that viewing the difference between deletion and deaccentuation in terms of the absence of suprasegmental features (deaccentuation) or of both suprasegmental and segmental features (deletion), as proposed by Klein, is promising. Nonlinear phonology disassociates the connection between segments, tones and prominence and postulates that tones are represented independently of segments or syllables and specifically that they occupy their own tier, in parallel with the segmental tier (cf. Goldsmith 1990,

68

Ellipsis at the interfaces: A proposal

Selkirk 1984). The tonal tier itself also consists of independent segments, called antosegments, as represented in (35): (35) a.

b.

segmental tier: Γι/

/η/

!ά!

/λ/

!YJ

Ι\Ι

/θ/

/η/

[·.·]

[•·.]

[•·•]

[·•·]

[•··]

[···]

[·•·]

[·•·]

tonal tier:

[+High]

On the basis of the theory of nonlinear and autosegmental phonology I propose that deaccentuation results from the Null Suprasegmental Feature Hypothesis, as in (36a), and deletion from the Null Segmental Feature Hypothesis, as in (36b). (36)

a.

b.

Null Suprasegmental Feature Hypothesis: Deaccentuation results from marking the suprasegmental level as null (SO), Null Segmental Feature Hypothesis: Deletion results from marking the suprasegmental and the segmental level as null (SSO).

Employing the standard analogy of the melody (suprasegmental tier) with the words of the song (segmental level), marking the suprasegmental tier as phonologically null will result in the text spoken without the melody. Marking both levels as phonologically null will result in silence. Note that the hypotheses in (36) apply at PF. The features SO and SSO, however, are instructions for PF assigned by the syntax-semantics interface. Three studies were conducted, two production and one perception experiment, which are reported in Winkler (1997) in more detail. The experiments were designed to empirically verify the following hypothesis on Second Occurrence Expressions (SOEs) in (37):11 (37)

Second Occurrence Expression Hypothesis: Complete phonological reduction only occurs in proper second occurrence sequences. Changes in the second occurrence expressions are signaled phonologically by audible pitch movements in the intonational contour.

Deletion vs. deaccentnation

69

The testing of examples like (34) resulted in three generalizations: First, if the conjuncts contain identical VPs, as in (34c), the second occurrence of the VP does not bear an accent. Second, if variations in the lexical and semantic content are present, as in (34d, e), accent assignment occurs. And third, pronouns behave differently in proper SOEs: they maintain their focus accent in the SOE if they are accented in the first conjunct. Next, I address the question of whether the intonational strategies employed to express the bound variable interpretation differ phonological ly from the strategies employed to express the referential interpretation in ambiguous proper and quasi SOEs. Particular attention is paid to the phonological disambiguation strategies that speakers employ to distinguish the different pronominal readings, concentrating on two questions: (i) Can Tancredi's prediction that the second conjunct, which contains redundant or "quasi" redundant material, is deaccented be confirmed? Second, what generalizations can be established with respect to the accent patterns of the pronominal forms in each conjunct? The relevant paradigm is given in (38): (38)

a.

b. c.

d.

M a n n y glaubt, dass er; wunderbar ist und LeOj glaubt, das erj wunderbar ist. 'Manny believes that he is brilliant and Leo believes that he is brilliant.' Manny ( glaubt, dass er, wunderbar ist und LeOj glaubt, dass er; wunderbar ist. Manny j glaubt, dass er, wunderbar ist, und Leoj glaubt, dass erj ein Genie ist. 'Manny believes that he is brilliant, and Leo believes that he is a genius.' M a n n y glaubt, dass er, wunderbar ist, und LeOj glaubt, dass er ; ein Genie ist.

In examples (38a, b) the embedded CP of the matrix verb in the second conjunct is identical to the embedded phrase of the first conjunct. The pronoun er in each conjunct may be interpreted either as a bound variable pronoun (also referred to as the sloppy reading) or as a referential pronoun (also referred to as the strict reading). The intended interpretation of (38a) is that Manny believes that Manny is brilliant and Leo believes that Leo is brilliant. (38b) can be paraphrased as Manny believes that Manny is brilliant and Leo also believes that Manny is brilliant. In (38c, d), the embedded phrase is not a proper SOE, but a quasi SOE. In (38c), Manny

70

Ellipsis at the interfaces: A proposal

believes that Manny is brilliant and Leo believes that Leo is a genius. In (3 8d), Manny believes that Manny is brilliant and Leo believes that Manny is a genius. 12 The results of the intonation experiments confirm the SOE hypothesis in (37) and show that the first question raised above must clearly be answered with no. The important generalization which can be made with respect to the intonational realization of pronouns is that in sentences such as those given in figure 2 and figure 4, in which the speaker stresses the pronoun in the first conjunct, he or she also stresses it in the second conjunct. Only in sequences in which the pronoun is unstressed in the first conjunct does it also occur unstressed in the second conjunct, as can be seen in the pitch extraction contours in figure 3 and figure 5.

Figure 2. Bound variable interpretation

Deletion vs. deaccentuation

Figure 3. Referential interpretation

•A>chl I D1-3B-4.NSP

0.00000
PITCH

Figure 4. Bound variable interpretation

0.00000 ·:·

1

•

^'"^S^ISBSIm

Figure 1. Narrow-scope reading The narrow-scope reading was triggered by a multiple w/z-question with negation Who can't eat what? The response to multiple w/z-questions is typically a distributive reading (see Kuno's 1982 SKH, section 4.3.1, and also chapter 2). The typical intonational phrasing of the answer is such that the conjoined sentences both start out with a rise on the first remnant and end with a fall on the second remnant. The perception effect is that there are two complete phonological units, in which the contrasting elements are paired and the verb plus negation is gapped in the elliptical conjunct. Next, consider the wide-scope reading of the example (39), given in figure 2 with the contour description in (42b).

200

Gapping: A sideward movement account

(42)

a. b.

[Leon CAN'T eat CAVIAR and Anna BEANS] (H*) H*+L H*+L Η" (H*) H*+L H%

«FRIFALI?!

'^mßmm

Ν

; JJ Y

Ι

ι»

r

rid

Ρ

Η Π

*

B. .R

a * -•

9

-i 7[ I

^

L

^

I

J

L

'

F

F

R

fl

J

1

T

Mwr

Figure 2. Wide-scope reading The wide-scope reading was triggered by the paraphrase given in (40b). The prototypical intonational contour is one in which the first correlate is realized with a pitch accent followed by a steep fall on the negated auxiliary and a fall-rise on the second correlate caviar that strongly signals continuation. In the second conjunct, the first remnant is again realized with a pitch accent followed by a fall on the second remnant. Often the wide-scope reading of negation over the conjunct is rather difficult to obtain without contextual manipulation. I used examples like (43) in the intonational disambiguation test. Example (43), adopted from Johnson (1999), has the two readings in (44).

Evidence for the sideward movement account

(43)

201

The left eye can't go up and the other one down.

(44)

a. b.

The left eye can't go up and the other one can't go down. It can't be the case that the left eye goes up and the other one down.

The trigger for the narrow-scope reading was the question in (45). A prototypical contour of the narrow-scope reading is given in figure 3, with the context description in (46b): (45)

What can't Lenny's eyes do?

(46)

a. b.

Γ& E d

V«ft

The LEFT eye can't go UP and the OTHER one DOWN, H*+L H*+L% H*+L H* L"L% r»A

4V

Hkjk

ugi j k n ü n V h f e w

SIJIBI *|*| . ι ^ ΐ ' ί ^ 'Β-«! .-•!»! I

Ii

•

:

I

jT

»'«ι

·ΐ| '

ii

|*| - '·· >: " *|«| ι

SfMäf.

Τ

1

*

203

·''

rrv

•:'

ϋ·Αώ

ft

bog Wnotw Hi» ·ί|-Ε|;3;| ttpmt

1:

Γ

,|Hl

I i i rs"

iftfI I

ur

II

J

·:-Μ

:ι

ill

•

ES

C3K5a r i Figure 6. Wide-scope reading The wide-scope reading in (44b) was triggered by presenting first the figures 4 and 5 above, and then (47a). The resulting contours showed a fall on left, a steep fall on the negated auxiliary, a pitch accent on up. almost no pause between the conjuncts, and little pitch movement on the initial material in the second conjunct until the contour is closed off by a prolonged fall.7 The relevant generalization is the following: An intonational contour consisting of two or more intonational phrases supports the interpretation in which negation has narrow scope relative to the connectives. It typically pairs up the topics (first correlate and first remnant) with the focus correlate and focus remnant: LEON can't eat CAVIAR and ANNA can't eat BEANS. Note that the list is understood as an exhaustive list (see Garrett 1996). The question to which such a contour occurs as an answer is a multiple whquestion. An intonational contour, however, in which the two conjoined constituents are integrated under the heavily accented and negated auxiliary (and are not broken up by a pause) facilitates the wide-scope reading.

204

Gapping: A sideward movement

account

On the basis of the intonational data, I propose two hypotheses that bring together the derivation of intonational contours with the SM-account: the Single Intonational Phrase Hypothesis (SIP), which enhances the widescope reading, and the Multiple Intonational Phrase Hypothesis (MIP), which enhances the narrow-scope reading, given in (48) and (49) respectively: (48)

Single Intonational Phrase Hypothesis: A wide-scope reading becomes available if the highest vP is interpreted by SSI as the smallest phase and is sent to PF. PF derives an intonational contour for each incoming structure.

(49)

Multiple Intonational Phrase Hypothesis: A narrow-scope reading becomes available if the two (or multiple) coordinated vPs are interpreted by SSI as individual phases and are sent to PF individually. PF derives an intonational contour for each incoming structure.

Let me consider the division of labor between syntax and SSI and SSI and PF more closely. There are basically two different approaches to the question of how syntactic structure is mapped onto phonological structure. The proponents of the first group have suggested allowing phonological rules to access syntax directly, as in the Direct Reference Hypothesis (e.g., Kaisse 1985), whereas the proponents of the second group only allow phonological rules that refer to prosodic constituent structure (Inkelas 1989, Truckenbrodt 1999: 221). Under the assumption that syntax and information structure influence prosodic phrasing, the Indirect Reference Hypothesis, as given in (50), is the hypothesis pursued here: (50)

Indirect Reference Hypothesis: Phonological rules refer only to prosodic constituent structure.

With respect to (50), the following question arises: What type of structure is actually sent to PF? The basic hypothesis underlying the focus framework developed in chapters 1 and 2 is that the smallest phase, namely vP, is coextensive with a phonological unit in instances of information focus. 8 I follow Hayes and Lahiri (1991) in the claim that the intonational phrase and the intermediate phrase (cf. Beckman and Pierrehumbert 1986)

Evidence for the sideward movement account

205

correspond to the domains of the prosodic hierarchy (Selkirk 1984, Nespor and Vogel 1986), specifically the international phrase (ip) and the phonological phrase (pp). The prosodic hierarchy of Hayes (1990: 86) is given in (51).9

foot ε syllable σ Although the claim that intonational and segmental domains coincide is not uncontroversial (cf. Gussenhoven 1984, Ladd 1996: 94), evidence from Bengali reported by Hayes and Lahiri (1991) and Lahiri and FitzpatrickCole (1999) seems to support it. Assuming that utterances are divided up into hierarchically structured phonological units like the phonological word, the phonological phrase and the intonational phrase and assuming further that prosodic phrasing is derived by a set of phrasing algorithms which derive phonological domains from syntactic structure (see Hayes 1990: 85, Hayes and Lahiri 1991, Fitzpatrick-Cole 1994, 1996), then the Coextension Hypothesis in (52) formulates the claim that the phonological phrase is directly derived from - or is coextensive with - the smallest syntactic phase that constitutes the domain of information focus. (52)

Coextension Hypothesis: The smallest syntactic phase (i.e., vP) is coextensive with a phonological phrase iff it does not contain any [F]/[T]- features.

I propose that whatever phrase constitutes the input to PF must be identified by some kind of independence at the interfaces (see Chomsky 2000, 2001, 2002, 2004, 2005). That is, if the vP is a phase and if it

206

Gapping: A sideward movement

account

constitutes the domain of information focus as proposed here and in chapters 1 and 2, it satisfies the sense unit condition (Selkirk 1984) and constitutes the input to a phonological phrase. More precisely, I claim that the vP that functions as the domain of new information constitutes a phonological phrase. In cases, however, where the derivation of the phase involves formal [F]/[T]- features (i.e., contrastive focus/topic), as in the case of gapping, this claim does not hold. A moved focus/topic constituent triggers a tone group boundary (phonological phrase boundary) to its right (see Truckenbrodt 1999: 225). A silent copy (subject to CR), as in the case of gapping, however, is ignored by the mapping constraint (see Chomsky 1995: 202ff). (53)

Insertion of Phonological Phrase Boundary: A displaced constituent triggers a phonological phrase boundary to its right.

Let me turn to the narrow-scope reading first. The ΜΙΡ-hypothesis in (49) states that the two (or multiple) coordinated vPs are interpreted by SSI as individual phases and are sent to PF individually. Since the vPs contain [F]/[T]- features, the displaced remnants introduce phonological phrase boundaries. Thus, (53) requires that the individual displaced constituents in the elliptical conjunct form their own phonological phrases, as shown in (54): (54)

U

PP L+H* LEON can't eat

PP PP H* L- L% L+H* CAVIAR and ANNA

PP H* L- L% can't oat BEANS

For the wide-scope reading, the situation is different. The SIP-hypothesis in (48) above states that a wide-scope reading becomes available if the highest vP is interpreted by SSI as the smallest phase and is sent to PF. PF derives an intonational contour for each incoming structure. In this case, there is no intermediate phrase boundary (Beckman and Pierrehumbert 1986) after caviar. The sequence eat caviar and Anna beans forms a phonological unit, namely a phonological phrase.

Evidence for the sideward movement account (55)

207

Wide-scope reading

PP

PP

(H*) H*+L Leon CAN'T eat

H*+L Η" (H*) H*+L H% CAVIAR and Anna BEANS

The question is what exactly does this imply for the distribution of labor at the interfaces? Let me assume that in the wide-scope reading the syntax-LF interface selects the highest vP as smallest phase, as is shown by the circled vP in (56) below: (56)

vP

ConjP Conj

vP

The intonational evidence from the different scopal readings shows that the coordination, at least in the wide-scope case, must involve vPs. As the next step, I address the question as to which position the remnants in the second conjunct move into.

3.4. Evidence for an A'-position in vP The SM-account that I have proposed in section 4.2.2 is crucially based on the derivation by phase (Chomsky 2000, 2001, 2002, 2004, 2005) and the PC A discussed in chapter 2. Much recent effort has been geared towards providing empirical support for the concept of the phase (cf. Lopez and Winkler 2003, Legate 1998, 2002, Nunes 1995, Epstein and Hornstein 1999, Uriagereka 2000, and the present study). Here, I'll provide empirical evidence for the hypothesis that there is an A'-position at the edge of vP that receives semantic interpretation at SSI, as suggested by the displacement hypothesis of the DC-ISH. There are altogether three pieces

208

Gapping: A sideward movement account

of evidence for the existence of an A'-position in vP, which are presented in the following three subsections. In 4.3.4.1, I review the evidence given in Fox (2000) and Chomsky (1999, 2001, 2002) that w/?-phrases adjoin to vP on their way to CP. In a second step, I'll provide evidence of an A'-position in vP from w/?-remnants in gapping. In 4.3.4.2, I'll show that evidence from topicalization in gapping suggests that at least the first remnant has raised. Finally, in 4.3.4.3, I'll provide evidence from German that suggests that there is a position at the edge of vP in German that hosts contrastive elements.

3.4.1. Wh-phrases at the edge of νΡ Chomsky (1986) already hypothesized that w/?-phrases adjoin to VP on their way to spec,C (see also Chomsky 1995: 302). 10 More recently, Fox (2000: chapter 5) has confirmed Chomsky's intuition. Consider the sentence in (57): (57)

[Which of the papers that he; wrote for Mrs. BrownJ did every student; get her, to grade t?

In (57), the w/z-phrase must reconstruct so that the variable, here the pronoun he,, is c-commanded by its binder, the quantifier every student,. The question is to which position should the w/z-phrase reconstruct? Reconstruction to the original position t causes a condition C effect: the pronoun her, would c-command Mrs. Brown,, in violation of Condition C. However, the w/z-phrase must reconstruct somewhere, so that the quantifier can bind the pronoun. It follows that the wh-phrase must reconstruct to an intermediate position, higher than the object and lower than the subject. Notice that the same result obtains for focus-topicalization instead of whextraction in (58): (58)

Only ONE of the papers that he; wrote for Mrs. Brownj, did every student; get herj to grade.

The data in (57) and (58) count as initial evidence that the wh-phrase and the focus phrase have stopped at the edge of vP. I conclude, essentially following Fox (2000) and Chomsky (1999, 2001, 2002), that there is an A'position at the vP-edge for displaced phrases.

Evidence for the sideward movement account

209

Although it is generally assumed that the position at the vP-edge cannot be the final landing site for A'-movement in English, the position is interpretively relevant, since the focus/w/z-phrase must reconstruct in spec,v. Therefore, the inevitable conclusion is that ν must have F to trigger movement of w/z-phrases to spec,ν also in English, as has been assumed under the displacement hypothesis of the DC-ISH. Now, the interesting question is: Is there empirical evidence for an A'position at the edge of vP that hosts w/z-phrases overtly? Evidence from gapping shows that remnants can also be w/z-phrases. Under the assumption that in gapping vPs are coordinated, as argued above, the w/z-phrases in (59) occur overtly at the edge of vP ((59a) from Pesetsky 1982: 646). (59)

a. b.

Bill asked which books I gave to Mary and [yp which records to John], Bill asked which books I gave to which students and [Vp which bones to which dogs].

Prima facie, (59) looks like evidence that gapping does involve CP coordination (Pesetsky 1982: 646, Hartmann 2000). 11 However, appearances are misleading. In its negated version in (60a) negation takes scope over both conjuncts, providing the reading paraphrased in (60b) (see the discussion above and also Johnson 1996: 65): (60)

a. b.

Bill asked which books I didn't give to Mary or which records to Susan, Bill asked which books I didn't give to Mary and which records I didn't give to Susan.

In the next subsection, we consider topicalized phrases at the edge of vP.

3.4.2. A'-position in vP hosts topicalized

phrase

In this section, I address the status of the non-wh remnants, which we have defined as contrastive topic and contrastive focus. The question that arises is whether these topic/focus remnants are in situ or may have raised vPinternally. Evidence suggests the latter. That topicalization is possible in gapping constructions such as in (61) has been repeatedly observed in the

210

Gapping: A sideward movement

account

literature (Sag 1977: 265, Hankamer 1979: 151, Wilder 1994, Johnson 1996: 66, Vanden Wyngaerd 1998: 8): (61)

a. b. c.

At our home we play poker and at Mary's house, bridge. The beans, Harry cooked, and the potatoes, Henry. During dinner, my father talked to his colleagues from Stuttgart and at lunch time to his boss.

The contour description of (61a) is given in (62b). The pitch extraction contour of (61a) is given in figure 7. (62) '

a. b.

: .

At OUR home we play POKER and at MARIA's BRIDGE, H* L" H* L'L% H*+L H" H*+L% 'J J..,,

·

if«.

. .

-w*.

f'Ä.

»1*1 ^ ^ ί . ? ! * ! Jyi jfelg·) ji

^

w

mm*

i-

—•·#·— -

..

f?·:

J3.

-J:': -./.v.· -J::: -J:': -./.v.· -J::: -J:': -./.v.· -J::: -J:': -./.v.· -./.v.· -J:': -./.v.· -./.v.· -J:': -./.v.· -./.v.· -J:':

: iwyiirtifJ ;; - (SiiwjA

Figure 7.

The pitch extraction contour clearly shows the peak accents on the contrastive remnants and their correlates. Example (63), taken from Lopez and Winkler (2003), shows that also an NPI is also possible in a topicalized position in the second conjunct under gapping.

Evidence for the sideward movement account

(63)

Al: B: A2:

211

Was your father in a bad mood last night? Why? Did he do anything strange? During dinner he didn't annoy Manny or at any time Amanda, for that matter.

A prototypical intonational contour of (63A2) is given in figure 8, and a contour description in (64b): (64)