Audiatur Vox Sapientiae: A Festschrift for Arnim von Stechow [Reprint 2015 ed.] 9783050080116, 9783050036724

Citation preview

Fery, Sternefeld (eds.) Audiatur V o x Sapientiae

studia grammatica Herausgegeben von Manfred Bierwisch unter Mitwirkung von Hubert Haider, Stuttgart Paul Kiparsky, Stanford Angelika Kratzer, Amherst Jürgen Kunze, Berlin David Pesetsky, Cambridge (Massachusetts) Dieter Wunderlich, Düsseldorf

studia grammatica 52

Caroline Fery Wolfgang Sternefeld (eds.)

A u d i a t u r ^ u u m i u ·

Vox Sapientiae A Festschrift for Arnim von Stechow

Akademie Verlag

Die Deutsche Bibliothek-CIP-Einheitsaufnahme Audiatur vox sapientiae: a festschrift for Arnim von Stechow / Caroline F6ry; Wolfgang Stemefeld (eds.) - Berlin: Akad. Verl., 2001 (Studia grammatica; 52) ISBN3-05-003672-9

ISSN 0081-6469 © Akademie Verlag GmbH, Berlin 2001 Das eingesetzte Papier ist alterungsbeständig nach DIN/ISO 9706. Alle Rechte, insbesondere die der Übersetzung in andere Sprachen, vorbehalten. Kein Teil dieses Buches darf ohne schriftliche Genehmigung des Verlages in irgendeiner Form - durch Photokopie, Mikroverfilmung oder irgendein anderes Verfahren - reproduziert oder in eine von Maschinen, insbesondere von Datenverarbeitungsmaschinen, verwendbare Sprache übertragen oder Ubersetzt werden. All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form - by photoprinting, microfilm, or any other means - nor transmitted or translated into a machine language without written permission from the publishers. Druck und Bindung: GAM Media GmbH, Berlin Printed in the Federal Republic of Germany

Contents

INTRODUCTION

10

JOSEF BAYER

Asymmetry in Emphatic Topicalization

15

SIGRID BECK AND WILLIAM SNYDER

The Resultative Parameter and Restitutive Again

48

DANIEL BÜRING

What Do Definites Do That Indefinites Definitely Don't?

70

CATHRINE FABRICIUS-HANSEN

"Wi(e)der" and "Again(st)"

101

GISBERT FANSELOW

When Formal Features Need Company

131

CAROLINE FERY

Focus and Phrasing in French

153

CORNELIA HAMANN, KATRIN LINDNER AND ZVI PENNER

Tense, Reference Time, and Language Impairment in German Children

182

8 IRENE HEIM

Degree Operators and Scope

214

GRAHAM KATZ

(A)temporal Complements

240

KATALIN E . KISS

Focussed Number Phrases

259

WOLFGANG KLEIN

Time and Again

267

MANFRED KRIFKA

For a Structured Meaning Account of Questions and Answers

287

MANFRED KUPFFER

Learning French from a Dictionary: An Essay on Utterance-Dependent Meaning

320

CECILE MEIER

Multihead Comparatives and Result Clause Constructions with "Split Antecedents"

348

RENATE MUSAN

Narrowing Down the Extended Now

372

IRENE RAPP

The Attributive Past Participle: Structure and Temporal Interpretation

392

MONIKA RATHERT

Anteriority versus Extended-Now: Theories o f the German Perfect

410

Contents

9

KJELL JOHAN S / E B 0

Necessary Conditions in a Natural Language

427

PIRMIN STEKELER-WEITHOFER

The Way of Truth: Parmenides' Seminal Reflection on Logic, Semantics and Methodology of Science

450

WOLFGANG STERNEFELD

Partial Movement Constructions, Pied Piping, and Higher Order Choice Functions

473

DIETER WUNDERLICH

Prelexical Syntax and the Voice Hypothesis

487

THOMAS EDE ZIMMERMANN

Unspecificity and Intensionality

414

List of Publications

534

List of Supervised Dissertations

541

Introduction1

The papers in this volume are dedicated to a friend and an outstanding personality in the linguistic community. They all relate to aspects of his work and are presented to Arnim on the occasion of his sixtieth birthday. The selection of articles has been guided by our limited awareness of what work in progress is related to Arnim's interests, and by the requirement of keeping the size of the volume manageable. Given the very number of Arnim's pupils and people he has influenced intellectually, and the even larger number of his friends, it turned out to be impossible to make such a volume representative in every respect. Nonetheless we believe that the contributions to the present volume do cover at least the topics that are most central to his current interests. Arnim's "official merits" as a linguist are too obvious to be mentioned in this introduction. Instead we believe that his influence as a human being, his ability to fascinate people, and his distinguished aura as a teacher are what should be revealed in an introduction that is meant to address not only a wider audience, but also Arnim himself. His enormous influence on our way of thinking, writing and even speaking is perhaps more hidden than his public accomplishments, but becomes apparent upon reflection. When it comes to describing his inner qualities more explicitly, one of Arnim's most outstanding features is his reliability. Those who have climbed mountains in his company are impressed by his stamina, and by the dependability that emanates from him. But he is also reliable in a long-term way, thinking of how he can help friends in crucial times, sharing experiences and providing jobs when he is in the position to do so. Since he has many friends and many occupations, he organizes his time in a way that can be frustrating to some, but in the end, the goodness and kindness and the real interest that he communicates often compensate for his lack of time. To be loved by him is a gratifying experience, not only because of the faithfulness of his friendship, but also because he is somewhat picky about whom he chooses to invest his affection in. He seems to understand the heart of humankind, to be perceptive of what drives most people. He can disarm or embarrass you by uttering a comment that shows that he has understood a part of you you thought was hidden, or that you weren't '

We would like to thank Ede Zimmermann, who has helped us extensively with the redaction of this introduction. Cicile Meier also deserves our gratitude since she provided the long-sought title and photo. Daniela Lentge and Anja Mietz have invested an enormous amount of time in formatting, correcting, changing and verifying all papers, and finally Kirsten Brock and Tonio Green have corrected the English of nearly everybody. We express our deep gratitude to them.

Introduction

11

even aware of. His interest in other people is real and deeply anchored and combined with a genuine curiosity about what motivates people, about their warmth and generosity, all qualities that he himself possesses in profusion, and that he likes in others. Professionally, he takes care of people, making sure that his students and colleagues achieve their goals, often before thinking of himself. To talk with him is an enjoyable activity. In a recent e-mail he sent for the turn of the century, he writes that he lost the art of conversation many years ago, a typical von Stechowian comment. He often attributes to himself just the opposite quality from what is obvious to everyone else. The truth is that he is a master of the art of conversation. He is talkative, interesting, witty, and can adapt to nearly everyone's conversation, and that in many different languages. He is impressively cultured and knowledgable about innumerable subjects, including the churches of Lviv, the history of Europe, chemistry, literature, opera, mechanics, or philosophy, to name just a few fields that he can talk about for hours. He can also talk about trivialities in an interesting way: how he cut his hand in repairing the roof of his Tessin cabin, cooking recipes, and what he saw on TV. He seems to have an inexhaustible memory for songs, poems, languages and linguistic facts, among other things. He talks to everybody in her or his own language, recalling words and expressions he may have last heard twenty years before. He is mostly joyful, but often feels that he is not the right person for his job, that his career is an accident, and that if people knew better they would know how incompetent he is. It is difficult to be sympathetic when he is in such a mood, since no one could be better. But of course this is not what he wants to hear at such times. However, when he expresses real sadness, something that he also does better than most of the rest of us, it is often because he is thinking of dear friends or members of his family who died unexpectedly or too young. This is something that deeply moves him, and at such times you want to cry with him. An important aspect of his personality is the appearance of untidiness, chaos and confusion that he delights in taking on, especially when entering a place in which he is expected. When you come to know him better, you realize that this is just external, an illusion, and that the surface of his desk, severely clean and tidy, reflects his personality much better than the condition of his sweaters. Another characteristic is his childlike interest in new toys. Computers have certainly changed his life, as well as new cars, his new Vespa, his new palm pilot. These are the kind of objects upon which he can focus his attention for several weeks. When acquiring such an object, he explores the qualities and properties of the machine, showing it to everybody he meets, and sometimes breaks it, a bit the way a six-year-old child does. In these circumstances he will never admit that it was his own incompetence that brought about the problem, but it will always be the fault of the manufacturer, the vendor or the student who tried to implement the special wish he had. His qualities as a linguist are such that he partly controls the evolution of modern linguistics. This is mostly true for semantics, where his influence is clearly felt, as evidenced by the impressive number of his brilliant and influential students. This influence is due to the quality of his teaching, to his ability to explain things, to his profound understanding of the field, to his countless ideas, to the knowledge he has accu-

12 mulated, and to his demanding and critical spirit - even with respect to his closest friends - towards the knowledge he has accumulated. Semantics, however, is not the only linguistic domain in which his influence is felt. Syntax has also been influenced by his work, and even phonology through some of his students. There are quite a few papers written by Arnim that may rightly be called influential, or even pioneering. Yet his deepest influence on the discipline, we feel, cannot really be pinned down to the many individual contributions he has made; nor can it be objectively measured by, say, a citation count. It is much subtler, but nonetheless clearly manifest. The influence is seen in his general approach to semantics, in his methodological rigor and his high standards of precision and clarity - much of which he has managed to pass on to his many students, who themselves have helped to spread the lore. Reading recent contributions to semantic journals, evidence of the influence of Arnim's thought is apparent more often than not - be it in the framework, in matters of descriptive detail, in the notation even, or more generally and more importantly in the quality. From the very beginning of Arnim's academic career, the logical analysis of meaning and its relation to syntactic structure appear to have been one of his major scientific concerns. To be sure, much of his early work, rooted in the tradition of "mathematical linguistics", dealt with syntactic formalisms and their complexity - the opus magnum of that period being a two-volume introduction to formal syntax co-authored with Angelika Kratzer and Peter Pause. But already at a time (around 1970) when formal linguistics was almost exclusively devoted to the expression side of language, Klaus Brockhaus and Arnim von Stechow had started exploring the content side in their pioneering paper "Formal Semantics", which contained a fully specified deductive apparatus for predicting sense relations among the sentences of a context-free fragment of German. Though this particular approach was soon to be overtaken by model-theoretic accounts of meaning, it still stands out as one of the first systematic applications of formal logic to the interpretation of natural language. Impressed by the formal rigor and predictive power of Montague Grammar and its cognates, around 1973 Arnim began studying Montague's and Cresswell's possible world semantics. His first original contribution to that theory again came in the form of a small, almost context-free fragment of German and it contained the first development of what would later be called "cross-categorial techniques": his 1974 paper "ε-λ-kontextfreie Sprachen" has long been recognized as the locus classicus of generalized Boolean operations, which form a central part of semantic theories from Partee & Rooth's "Generalized Conjunction and Type Ambiguity" (1983) to Winter's 1998 dissertation "Flexible Boolean Semantics". An equally important, though maybe less well-known contribution to extensional semantics is his 1980 paper "Modification of Noun Phrases", which constitutes one of the first formal investigations into the meaning of determiners and noun phrases - i.e., what would later become Generalized Quantifier Theory. Neither of these early pieces is directly addressed in the contributions to this festschrift, but their status as classical semantic papers is beyond any doubt.

Introduction

13

Three leitmotifs may be discerned in Arnim's many-faceted semantic work: • The situational dependence of reference Long before it became fashionable, Arnim saw the importance of Kaplan's and Lewis's work on intension and indexicality for the analysis of natural language. His contributions to this theory - starting with a joint paper with Angelika Kratzer ("Äußerungssituation und Bedeutung", 1977) - are substantial and include an account of referential variation in pronouns in terms of occurrences ("Occurrence-Interpretation and Context-Theory", 1979), a solution to puzzles of informativity and rigid designation ("Structured Propositions and Essential Indexicals", 1984), as well as his extensive work on tense and temporal deixis, among which there is a widely read, though never properly published introduction to formal semantics that is centered around temporal indexicality and intensionality and that has acquired the status of an underground textbook classic among German students of linguistics ("Intensionale Semantik - eingeführt anhand der Temporalität", 1992). Many of the contributions to this festschrift directly relate to Arnim's work on temporal reference, context dependence, and, most prominently, the semantics of again; cf. the contributions by Beck & Snyder, Fabricius-Hansen, Katz, Kupffer, Klein, Musan, Rapp, Rathert, Stekeler-Weithofer, Wunderlich and Zimmermann. •

The structuring of propositional information

Among philosophers of language, Arnim is probably best known for his solution to the problem of logical omniscience ("De Re Belief Generalized", with Max Cresswell, 1982). The solution is based on a detailed account of structuring propositions into properties and things, a structuring that Arnim subsequently identified with the focus/background distinction needed for the interpretation of answers to wh-questions and focus-sensitive operators ("Current Issues in the Theory of Focus", 1991). Together with Mats Rooth's alternative semantics, this structured meaning approach, independently developed by Arnim and Joachim Jacobs, is now one of the standard theories of focus interpretation; Krifka's contribution to this festschrift compares the two theories with respect to their performance on question answer dialogues. The influence of focus on the structure of a proposition is also addressed by Kiss. •

The architecture of the syntax/semantics interface

Although certainly impressed by Montague's surface compositional approach, Arnim has always been too much of a syntactician to adopt it himself. Around the early eighties, his work on German led him to the Chomskyan framework, which he soon started connecting with the logical tradition in semantics. Arnim's most ambitious contribution to the Handbook of Semantics (edited by himself and Dieter Wunderlich) was a paper called "Syntax und Semantik" (1991), which contained the first systematic model-theoretic interpretation of Chomskyan LFs, alongside with a detailed compari-

14

son with the then influential theory of (generalized) categorial grammar. An abstract framework for interpretation was already implicit in his "Comparing Semantic Theories of Comparison" (1984), one of the most influential pieces he has written. In later work, Arnim extended this abstract view on interpretation with the framework of Transparent Logical Forms - a term and theory whose precise history appears to be unknown, but which is clearly Stechowian in spirit. More recently, Arnim's logical forms have become ever more abstract, including a level of lexical decomposition as motivated by classic scope ambiguities with causatives. This development has not always been appreciated, and some of the discussion of Arnim's lexical decompositions is continued in the contributions of Wunderlich and Fabricius-Hansen. His much less controversial comparative semantics stands in the center of Meier's and Heim's papers, while his interest in modal logic and modality in natural language is reflected in the contribution of Saehe. Arnim's interest in the method of choice functions as a means to interpret wh-operators in situ is adopted by Sternefeld in his contribution to the present volume. Although the three leitmotifs cover a lot of his semantic work, there is still much that does not fall under any of the above headings. And then there is also his non-semantic work, among which his contributions to syntax - which include the 500-page book "Bausteine syntaktischen Wissens" (1988, with one of the editors) - is most influential. Relating to Arnim as a syntactician, Bayer discusses a focus-based syntactic projection in Bavarian and Büring and Fery continue focus-related work done by Arnim, while Hamann, Lindner & Penner contribute to the issue of functional projections and their learnability. Finally, Fanselow addresses a topic that has arisen in the minimalist framework, but which had been discussed by Arnim with one of the editors before minimalism came into being: the issue came up in connection with Mark Baker's "Incorporation without Incorporation." Arnim proposed (p.c.) that what is usually called LF movement should rather be analyzed as a kind of invisible chain-formation at Sstructure that moves grammatical features only and leaves behind phonetic and, since the relevant case was head movement, semantic information. At that time (probably in 1986) the idea was considered too wild to be taken seriously. Its time came only ten years later. It is our conviction that many of Arnim's other flashes of wit are living with us now in a still embryonic state and will only fully develop at some time in the future. Arnim will continue to contribute to the field.

Asymmetry in Emphatic Topicalization

Josef Bayer "Die Betonung des Satzeingangs aber entspringt der Erregung des Sprechenden. " Otto Behagel (1932), Deutsche Syntax, Vol. IV

1.

The Phenomenon

Southern German of the Austro-Bavarian type is known for its disrespect of the Doubly-Filled-COMP Filter (DFCF). Wh-questions and relative clauses are generally formed by moving the wh-phrase or the relative operator to the specifier of an overt complementizer. The examples in (1) and (2) from Bayer (1984) illustrate the construction: (1)

a. I woaß ned wer daß des döa hot I know not who that this done has Ί don't know who did this'

indirect question

b. I woaß ned wiavui daß-a kriagt I know not how-much that-he gets Ί don't know how much he will get' (2)

a. Der Mantl den wo i kafft hob the coat which that I bought have 'The coat which I bought'

relative

b. Des Audo des wo i mecht the car which that I want 'The car which I would like to have' In Bayer (1984) the phrase structure of these constructions was represented in the S, S', S" system, and it was assumed that there are two complementizers, one of which (COMP2) hosts the operator while the other (COMP1) is filled with the complementizer. The shortcomings of this analysis are easily avoided under the phrase structure which Chomsky (1986a) suggests in Barriers. This phrase structure extends X'-syntax to functional elements such as COMP (henceforth C). Under this implementation,

Josef Bayer

16

C0MP1 corresponds to C and COMP2 to SpecCP, the specifier of the maximal projection of the head C. Bavarian and Standard German can then be argued to have the same underlying structure, the difference being that Standard German rejects simultaneous filling of C and SpecCP. Since Bavarian does not respect the DFCF, it is not surprising to see that movement to SpecCP is not confined to wh- and relativization operators. 1 Consider the following examples, of which (3a,b) are from Bayer (1984), (3c) is from Weiß (1998) and (3d) is a datum from a corpus which is reported in Lutz (1997): (3)

a. Da Xaver daß anMantl kafft hot hot neamad the Xaver that a coat bought has has nobody 'As for Xaver, nobody believed that he bought a coat'

glaubt believed

b. An Mantl daß da Xaver kafft hot hot neamad glaubt a coat that the Xaver bought has has nobody believed 'As for a coat, nobody believed that Xaver bought one' c. Da Hans ob kummt woaß-e ned the Hans whether comes know-I not 'As for Hans, I don't know whether he will come' d. An Fünfer daß-e kriag häid-e ned g'moant a five that-I get had-I not thought 'As for a grade five, I didn't think I would get one' The phenomenon central to the present article is that - contrary to the examples in (1) in which the DFC-complement appears in post-verbal or "extraposed" position - this kind of topicalization is only possible if the CP-complement is topicalized itself. It is strictly impossible in CPs which occupy the post-verbal position (in traditional terminology the Nachfeld). The data in (4) are all the more puzzling because the Nachfeld is the canonical position for a finite complement and is considered by many syntacticians to be its base position:

1

The construction seems to also be possible in other dialects which don't observe the DFCF with respect to wh-movement, such as Alemannic and more northeastern varieties such as Franconian and Thuringian, but seems to be far less widespread in these dialects. For recent discussion, see Weiß (to appear).

Asymmetry in Emphatic a. *Neamad hot nobody has

17

Topicalization glaubt, believed

b. *Neamad hot glaubt, nobody has believed ned, da not the

Mantl kafft hot bought has coat

da Xaver daß an the Xaver that a an a

c. *I I

woaß know

Hans Hans

d. *I I

häid ned g'moant, an had not thought a

Mantl coat ob whether

daß da that the

Xaver kafft hot Xaver bought has

kummt comes

Fünfer daß-e five that-I

kriag get

This restriction on topicalization across C has remained unexplained in the sense that no agreement has so far been reached among syntacticians as to the description of the phrase structure.2 The rest of this article is organized as follows: In section 2 I present a review of two classes of earlier accounts. In section 3 I refer to the differences between topicalization as movement to SpecCP and left dislocation. In section 4 it is suggested that the topicalization in question is an indicator of illocutionary force which I argue is implemented in a functional projection. The core of my analysis and some of its consequences are presented in section 5. Previous accounts have mainly been concerned with a parasitic-gap construction that is intimately connected with the kind of topicalization seen in (3). I will show in section 6 that the account developed up to section 5 is compatible with and provides new evidence for the ATB approach to parasitic gaps. The results are summarized in section 7.

2.

Previous Attempts at an Explanation

The simplest imaginable analysis would be one in which the topicalization facts in (3) simply result from the Bavarian disrespect of the DFCF. In this case, it is crucial to assume that the moved XP targets SpecCP in the same way as a wh-phrase targets SpecCP. But then the difference between (l)/(2) and (4) must be made to follow from something else. Alternatively, one may argue that the parallelism between topicalization and wh-movement is only apparent, and that the underlying structures are in fact rather different. Both directions have been explored already. The argumentation has in various cases been determined by the fact that the observed type of topicalization seems to be responsible for parasitic gap constructions which are found in Bavarian, and to which we will turn in section 6. Proponents of the SpecCP theory are Felix (1985), Grewendorf (1988) and Bayer (1988). Proponents of an alternative theory are 2

One reviewer tells me that according to an Austrian informant constructions such as those in (4) are acceptable for "some speakers of Upper Austrian". Given the sharp ungrammaticality that arises for myself and all the other Bavarian speakers I could consult, I find this surprising. At this moment I can only leave this for further investigation.

Josef Bayer

18

von Stechow & Sternefeld (1988) as well as Weiß (1998), who closely follows them. They suggest that the topicalized constituent in (3) is not in the specifier of its immediate CP but is in fact part of the matrix clause. In my presentation I partially rely on a review of the earlier work by Lutz (1997). I take the liberty of leaving Felix (1985) aside because his primary concern is the parasitic gap phenomenon, whereas my primary concern here is an explanation of the topicalization asymmetry (TA). 3

2.1

Top in SpecCP

Grewendorf (1988:254ff.) as well as Bayer (1988) follows the conservative assumption that XPs which are topicalized to the immediate left of C are in SpecCP. While Bayer (1988) does not try to come to grips with the TA, Grewendorf does. He suggests that the asymmetry derives from a restriction on government according to which SpecCP of a clausal complement in post-verbal position is governed, whereas it is ungoverned if CP is topicalized., Grewendorf proposes a principle according to which only a lexically ungoverned Vorfeld position may be occupied by XP, and which he makes sure does not apply if XP is +wh and undergoes wh-agreement. Attractive as such a solution may appear within the GB-framework, its attractiveness relies on a central relation that has been given up in the successor of that theory, the Minimalist Program (Chomsky 1995): government. If one is justified in assuming that government has ceased to play the role it used to play, it cannot be responsible for the restriction at hand. But let us concentrate more on empirical arguments. In Bavarian, like in any other variety of German, certain verbs allow V2-complements such as Ich glaube [den Hans ι hat die Polizei ti erwischt] ( Ί believe Hans has been caught by the police'). Since den Hans arguably occupies SpecCP (in which the finite verb has raised to C), extra provisos would have to be invoked to prevent government of the Vorfeld here. Such provisos do not appear to be quite natural or plausible. A conclusive argument against an explanation based on government comes, however, from cases which exhibit the very same TA while not involving government according to any standard theoretical implementation. Consider the examples in (5) and (6), both of which involve adjunct clauses of a certain type, namely if- and os-clauses: (5)

a. Da Xaver wenn hoam kummt kriagt-a wos z' essn the Xaver if home comes gets-he something to eat 'As for Xaver, if he comes home, he will get something to eat' b. *Da Xaver kriagt the Xaver gets

3

wos something

z' essn der to eat he

wenn if

hoam home

kummt comes

Of course, I believe that a proper account will also have to say something about the parasitic gap construction, and I hope to be able to say something sensible at the end of this article, but I want to emphasize that parasitic gaps are not in the focus of my discussion.

19

Asymmetry in Emphatic Topicalization (6)

a. An Xaver wia-s g'seng hom hom-sa-se recht the Xaver as-they seen have have-they-REF really 'As for Xaver, when they saw him they were really happy'

g'freit rejoiced

b. *Sie hom se recht g'freit an Xaver wia-s g'seng hom Since adjunct CPs as initiated by wenn or wia (Standard German wie) are ungoverned by definition, Grewendorf s explanation can hardly be tenable. I suspect that the TA is independent of government and thereby of lexical selection, argument or adjunct status.

2.2

Topic/Focus in the Matrix CP

Von Stechow & Sternefeld (1988:387f.) propose an entirely different solution which cannot rely on Grewendorf s central point because the data von Stechow & Stemefeld are looking at primarily involve topicalizations in adjuncts. Their considerations are based on examples in Felix (1985) such as the following:4 (7)

a. I bin glei b'suffa [wenn-e des Bier no trink] I am immediately drunk if-I this beer still drink 'I'll be drunk right away if I continue drinking this beer' b. *I bin glei b'suffa [[des Bier]i wenn-e no t] trink] c. I bin [wenn-e des Bier no trink] glei b'suffa d. *I bin [[des Bier]i wenn-e no ti trink] glei b'suffa e. [[Des Bier]i wenn-e no ti trink] bin-e glei b'suffa

(7c,d) show that the same contrast as before occurs if the dependent CP is in clausemedial position. The wew/j-clause in (7c) is ok when parenthetically inserted, but as (7d) shows, it ceases to be ok when an XP has been moved to the specifier of the wennclause. Only when the topicalization in front of C occurs in a CP which is likewise topicalized do we find a grammatical result. In their account, von Stechow & Sternefeld consider the topicalized constituent to be focused. This is in an interesting conflict with Grewendorf s representations, in which the very same type of constituent is considered to be a topic. If topic and focus are mutually exclusive notions, one of the analyses should be wrong. A more benevolent interpretation is possible, however, according to which the notion focus is not necessarily restricted to quantificational structure and the standard analysis in terms of variable binding. According to Erteschik-Shir (1997:70), a new topic is introduced in outof-the-blue sentences by a phonologically stressed phrase. Thus, phonological promi4

I take the liberty of turning examples from published sources which do not strike me as quite idiomatic for one reason or the other into what I take to be proper colloquial standard in my own dialect.

20

Josef Bayer

nence, which is clearly required in the construction under investigation, should not prevent us from considering the preposed elements to be topics (of some sort). 5 And indeed, the preposed elements pass classical tests for both topic- and focushood (cf. Erteschik-Shir 1997:14). Discourses in which the preposed element has already been mentioned are well formed, i.e., the usual aboutness relation which topics are subject to is observed. To see this, consider the following discourse: (8)

A:

Wos hom-s g'sagt, wia-s an Xaver wieder g'seng hom? what have-they said as-they the Xaver again seen have 'What did they say when they saw Xaver again?'

B:

An Xaver wia-s g'seng hom hom-sa-se recht g'freit 'As for Xaver, when they saw him they were really happy'

(=6a)

It is also possible to test in the opposite direction. Potential foci (in the sense of quantificational structure) must correspond to a variable as initiated by a constituent question. Normally, the constituent that can be questioned is focus. Consider now the dialogue in (9): (9)

A:

Wos fir-a Notn host gmoant what for-a grade have(-you) thought 'Which grade did you think you would get?'

B:

An Oanser daß-e kriag how-e a one that-I get have-I 'Grade one I thought I would get'

daß-st that-you

kriagst? get

g'moant thought

It appears here that the preposed XP an Oanser, which supplies the new information, can also function as the binder of a variable trace, i.e., as a focus. This fact, or perhaps the mere fact that the preposed element is prosodically prominent, has led various researchers to consider it to be a focus; cf. von Stechow & Sternefeld (1988), Lutz (1997) and Weiß (1998). However, given that one test shows that the preposed XP is a topic, while the other test shows that it is a focus, there is a contradiction because something cannot be topic and focus simultaneously. I think this conflict can be resolved if it is realized that the movement to the immediate left of C is not focus-driven movement but something else. Notice that the construction in (9B) is not at all essential for answering (9A) successfully. The answer could equally well be I hob g'moant, daß-e an Oanser kriag ( Ί thought that I would get a one'). The focality of the filler an Oanser (in the sense of "new information") is already checked in the middle field; thus, the constituent must have been raised still higher for an independent purpose. I will return to this issue. At this point we conclude that where the preposed element has been termed focus this notion makes sense only if it is identified with prosodic prominence,

5

Non-salient elements which can never bear stress such as es ('it') or man ('one') are banned from this position.

Asymmetry in Emphatic

Topicalization

21

and not if the position is associated with the semantic aspect of the notion "focus". 6 The motivation for looking at the preposed and stressed XP as a focus constituent seems to have yet another source. As mentioned earlier, the construction under consideration has been considered to be intimately connected with the parasitic gap phenomenon. Starting from Felix's (1985) original analysis, von Stechow & Sternefeld as well as Lutz (1997) consider it essential that the extracted element c-commands both the parasitic gap (e) and the real gap (t) in (10):7 (10) An Karli wenn-eti dawisch daschlog-e ei the Karl if-I get kill-I 'If I get hold of Karl, I'll kill him' Von Stechow & Sternefeld (1988:387) conclude on the basis of these data that the Bavarian root clause has an additional Vorfeld position for focused constituents. Their analysis provides an insight which is of central importance to my own account, namely that this additional position is only available in root clauses. The proposed structure of (7e) is given in ( l l ) . 8 (11)

S"

If this point can be established, the deviant examples (4a-d) and (7b,d) are ruled out as desired. Unfortunately, the von Stechow & Sternefeld analysis does not specify how the relatedness of the root clause to this position could be derived. It is left as a stipulation, and as such it reduces to a restatement of the facts. Technical problems with their 6

7

8

In short, the pre-C position is as neutral w.r.t. topic and focus as the pre-Vf,„ position of an ordinary German V2 clause. For concreteness I assume that the focus feature of a constituent may have been checked before another feature of the same constituent is checked in a higher domain. This is absolutely standard, as one can see in those cases where an NP first moves to an Α-position to get its Case feature checked and then moves on to get its wh-feature or some other feature checked. It seems obvious that the trace is in the topicalized clause, while the parasitic gap is in the root clause. Only the latter can be replaced by a personal pronoun. However, cf. Lutz (1997: section 3.4) for a controversial discussion of this issue. Von Stechow & Sternefeld use the S, S', S" system. More recent considerations of functional categories in phrase structure suggest that the finite verb of the root clause bin is in C° and the adjunct CP as well as the focus phrase des Bier are in two specifiers of C°.

Josef Bayer

22

analysis have been noticed by Lutz (1997): One problem is that the analysis violates the Adjunct Condition according to which nothing may be extracted from an adjunct. However, the Adjunct Condition, is respected in Bavarian in the same way as in Standard German. Another problem is that the preposed/focused constituent is predicted to successfully bind into the domain c-commanded by it. But the deviance of the examples in (12) signals that this prediction is in disagreement with the facts: (12) a. *K6a Mensch ι wenn ti b'suffa is foit eami no man if drunk is falls him intended: 'Nobody has good ideas when he is drunk'

wos g'scheids el something useful in

b. *A jeder Menschi wenn ti niachdan is foit eami an every man if sober is falls him g'scheids e! useful in intended: 'Everyone has good ideas when he is sober'

wos something

Contrary to what (11) may suggest, quantified NPs (QNPs) such as in (12) cannot bind a pronominal variable in the matrix clause. Considerations of this kind have led Lutz (1997) to propose two separate structures for what he refers to as the "focus construction" and the "parasitic gap construction". Let me continue to concentrate on the former and postpone discussion of the latter till section 6. Lutz follows Brandt et al. (1992), who claim that the displaced element cannot be in SpecCP because SpecCP may be occupied by a wh-phrase, and that therefore the only possibility is adjunction to CP. On the basis of this, in my view unmotivated, suggestion he proposes the structure in (13): (13) [CP FOCUS, [ C p...t, ...]] The argument against movement to SpecCP obviously rests in part on a confusion about the word wann. While in Standard German wann always means 'at what time', in

23

Asymmetry in Emphatic Topicalization g

many Bavarian varieties it is homophonous with the conditional particle wenn ( ' i f ) . If trace-binding cannot take place under any other relation than c-command, as Lutz assumes, i.e., if m-command plays no role here, (13) successfully prevents the derivation of the examples in (12). On the other hand, however, the fact that (13) is confined to the left periphery of the matrix clause must still be made to follow from some other principle. Lutz briefly considers Rizzi's (1997) proposal of a more articulate structure of the left periphery of the clause in which more functional heads are available than under the received standard analyses of German in terms of CP, and he also considers Chomsky's (1995) suggestion that there may be multiple specifiers, if one head carries different features which need to be checked by different phrases. His remarks are too short to allow any firm conclusions, but they point in a promising direction as I will try to show in the following. Lutz's adjunction analysis in (13) suffers from a conceptual problem, because it leaves unexplained why only Bavarian but not Standard German too should have this construction. The conservative approach according to which movement targets SpecCP, however, connects to the well-known difference with respect to the DFCF. The arguments that can be raised against von Stechow & Sternefeld's as well as Lutz's analyses indicate that the conservative solution in terms of movement to SpecCP may have been abandoned too quickly. What follows should be seen as an

9

For instance, example (i), which is a phonetic variant of (7e), should not lead to the conclusion that the preposed XP is beyond SpecCP. The example is clearly a conditional. (i) [Des Bier], wann-e no t) trink bin-e glei b'suffa this beer if-I still drink am-I immediately drunk Both Lutz (1997) and Weiß (1998) point to (constructed) examples in which wann must be interpreted as a wh-phrase, and as such should occupy SpecCP rather than C. Given that for the speaker the conditional wenn and the temporal wann may not be fully distinct due to the ambiguity of the form /van/, one would have to search for more convincing examples. The only other case that seems to be frequently attested is wia: (ii) [B Muatter]! wiat] des g'heat hot is-s ganz naarisch woan the mother as this heard has has-she completely furious become 'As/when mother heard this, she became really furious' But in this case, wia is clearly not interpreted as 'in which manner' but rather as the particle 'as'. While some of the details may remain murky, there can be no doubt that whenever an unambiguously maximal wh-phrase occupies SpecCP, preposing - adjunction to CP - leads to an ungrammatical result, e.g.: (iii) a. *[Da Xaver] ι wohl t, zong is woaß-e ned the Xaver where-to moved has know-I not b. *[Da Xaver] ι in wäichan Haus t, wohnt woaß-e ned the Xaver in which house resides know-I not Therefore I think that in the cases at hand adjunction to CP is neither necessary nor desirable. The minority of examples which seem to invite such a solution are either built on misunderstandings or should be investigated more carefully before an adjunction analysis is adopted. For instance, movement across warum ('why') leads to a more acceptable construction than those in (iii), but this class also shows special behavior elsewhere; cf. Rizzi (1990), Hornstein (1995) and § 3.1 of Bayer (to appear), among others.

24

Josef Bayer

attempt to push the conservative approach as far as possible. If successful, the analysis will enable us to retain the closest possible connection between the DFC- and the top/foc-construction, because the latter will simply be an instantiation of the former.

3.

Topicalization versus Left Dislocation

It is widely known that German has left dislocation (LD), i.e., cases in which an XP seems to be left-adjoined to a V2-clause in which SpecCP hosts a pronoun that is coindexed with XP. (14) shows some examples of V2 root clauses with LD. The examples in (15) show that no TA arises here: (14) a. Den Hansi, deni kennt er kaum the Hans him knows he hardly 'As for Hans, he hardly knows him' b. [An dieser Sonate]ι, [an der]i hast du bestimmt viel at this sonata at this have you certainly much 'As for this sonata, you will certainly have much fun with it' (15) a. Ich glaube [den Hansi, deni kennt er I believe the Hans him knows he

Freude fun

kaum] hardly

b. Ich glaube [[an dieser Sonate]], [an der]i hast du bestimmt viel I believe at this sonata at this have you certainly much Freude] fun Topicalization in the face of embedding as in (15) is obviously possible because the V2-complement has root properties.10 The TA-effect can, however, be witnessed in those cases in which LD applies to a C-headed CP. In Standard German such cases are only possible with a focused resumptive pronoun in situ. The examples in (16) and (17) demonstrate this with argument and adjunct CPs respectively. In each case, the LDtopic carries a rising contour and is separated from CP by a prosodic break, which I indicate with a double slash: (16) a. [Den Hansi // [daß er deni kennt]] glaube ich the Hans that he him knows believe I 'As for Hans, that he knows HIM, I don't believe' b. *lch glaube nicht [den Hansi // [daß er deni kennt]]

10

Cf. Heycock (2000).

nicht not

Asymmetry in Emphatic Topicalization (17) a. [Den the mich me

25

Hans ι//[wenn du den ι siehst]] sag ihm er Hans if you him see tell him he anrufen call

soll should

'As for Hans, if you see HIM, tell him he should call me' b. *Sag ihm er soll mich anrufen [den Hansi // [wenn du den] siehst]] As already suggested by the prosodic break, the LD-topic must be outside SpecCP, which is expected in those varieties of German which respect the DFCF. This impression is corroborated by the fact that such constructions are not possible with quantified NPs (QNPs) in LD-position which cannot receive a referential interpretation. QNPs which give clear results are those which involve the operators jeder ('each') and kein ('no'): (18) a. * [Jeden/keinen Studenten]//[daß er each/no student that he

deni kennt]] glaube ich nicht him knows believe I not

b. * [Jeden/keinen Studenteni // [wenn du deni siehst]] dann each/no student if you him see then der Anna, sie soll mich anrufen the Anna she should me call

sag tell

Although the LD-examples from Standard German show the TA, the Bavarian topicalizations under consideration are different in showing neither the prosodic break nor the semantic restriction nor the resumptive pronoun. We have already seen that the Bavarian examples which give rise to TA do not invoke resumptive pronouns. There is also no indication of a prosodic break. According to prosodic information, the topic is entirely part of the CP proper. The following examples show that QNPs can be topicalized too: (19) a. Neamadi wenn ti kummt, no is-s aa ned recht nobody if comes then is-it also not good 'If nobody shows up at all, it isn't ok either' b. Α jeden daßti so deppert is glaub-e a everyone that so stupid is believe-I Ί don't think that everybody is that stupid'

ned11 not

Thus, the construction is distinguished from LD by three factors: prosody, lack of a resumptive pronoun, and the semantic nature of the topic. It can be no accident that according to this list the Vorfeld patterns exactly with the Bavarian topics: The fact that there is an indefinite determiner in front of jeder does not affect the semantics. In Bavarian the use of determiners is in general much more widespread and obligatory than in the standard language.

26

Josef Bayer

(20) a. [Den the

Hans, (??//) [kennt er ti nicht]] Hans knows he not

NO PROSODIC BREAK

Hansi Hans

[kennt er knows he

(•ihn,) nicht]] (him) not

: NO RESUMPTIVE PRONOUN

c. [Jeden/keinen each/no

Studenten] student

[kennt erti]] knows he

: NO REFERENTIALITY RESTRICTION

b. [Den the

I take this to be strong evidence that the Bavarian topics which give rise to TA are indeed in SpecCP and not in a higher position such as an LD-position.

4.

Emphatic Topicalization as an Indicator of Illocutionary Force

We have seen above that both the notions topic and focus may be slightly confusing in connection with the construction under consideration. The preposed element must bear stress but seems to be neutral with respect to topicality or focushood in the semantic sense. What is the common feature that is shared by all the elements that undergo such preposing into SpecCP? I want to suggest here that it is emphasis, and that the proper notion should be "emphatic topicalization", where "topicalization" simply refers to the fact that the effect of emphasis is achieved by preposing to SpecCP. I admit that emphasis - or more appropriately Erregung ('excitement') in Behagel's terminology - is not a well-defined notion in linguistic theory, as it can be achieved by a variety of grammatical means. Nevertheless it seems to be true that this form of topicalization is the grammar's reflex of the speech act to be performed and is as such on a par with German constructions involving modal particles like aber, denn, doch, ja, etc. Modal particles supply features which interact with other features such as [wh], yielding a wide range of illocutionary forces. 12 As Doherty (1985:64) points out, modal particles are root clause indicators of illocutionary force or, in her terminology, epistemic meaning. Embedded clauses which are modified in this way lead to illformedness unless the root predicate is a verb (e.g., of speaking) that permits a quotational reading:

Cf. Bayer (1991) and Hasegawa (1999). The special properties of root clauses were, of course, noticed very early; cf. Emonds (1969) and Hooper & Thompson (1973) for English, and Heycock (2000) for an overview that covers various Germanic languages.

Asymmetry in Emphatic

27

Topicalization

(21) a. Konrad ist {ja/doch} Konrad is PRT/PRT 'Konrad is out of town' b. *Nina nimmt an, Nina assumes

verreist traveled-away

daß Konrad {ja/doch} that Konrad PRT/PRT

c. Nina {sagte/meinte}, Nina said/gave-the-opinion

verreist traveled-away

daß Konrad {ja/doch} that Konrad PRT/PRT

ist is

verreist traveled-away

sei besubj

Obviously, particles like ja and doch are indicators of an attitudinal meaning that cannot be activated unless the particle has access to a root(like) clause. According to Cinque's (1998) universal hierarchy of functional heads which provide positions for adverbs, the highest projection is reserved for speech act indicators. If emphatic topicalization belongs to the class of grammatical means of force projection in the sense of Rizzi (1997), its root clause property and strict left peripherality are not surprising. (22) shows that a V2-clause with emphatic topicalization can be in the scope of a manner-of-speaking predicate just like the complement with a modal particle in (21c). (22) B' Fanny hot g'moant [[da Hans ob kummt]wissat i ned] the Fanny has thought the Hans whether comes knewsuhjl not 'Fanny uttered the opinion that as for Hans I wouldn't know whether he would show up' Thus, although it seems to be difficult to elucidate the semantics and pragmatics of emphasis and emphatic topicalization, it seems equally clear that the syntax reserves for them a specific space in phrase structure. If we adopt the view that this space is defined by a functional head, the head responsible for emphatic topicalization must be to the extreme left. In the next section I will develop an account which specifies the activation of emphasis by topicalization to SpecCP. The relevant feature will be dubbed [etop].

5.

The Syntactic Nature of Emphatic Topicalization

5.1

The Feature [etop]

If we want to push the conservative approach of movement to SpecCP to its limits, we have to assume a feature in C to which a phrase from within the clause is moved to check a corresponding feature. A phrase which bears the feature [etop] is moved to C if the language permits C to bear the feature [etop] too. The DFC-phenomenon of Ba-

28

Josef Bayer

varian syntax shows that something similar is independently needed if we want to keep to the widely accepted analysis according to which a wh-phrase may move to SpecCP although the C-position is filled with a complementizer. Recall that neither wh-movement to SpecCP nor emphatic topicalization is allowed in the standard language if C is lexicalized. Let me therefore propose a low-level parametric difference between Standard German and Bavarian according to which the complementizers of Bavarian may bear the features [wh] and [etop], whereas the complementizers of the standard language cannot. 13 As we were able to demonstrate in (16) and (17), the standard language does show emphatic topicalization, thus giving rise to the TA-effect. The difference is that the topicalized phrase cannot have been moved to SpecCP as shown by various tests. The topicalized phrase seems to be adjoined to CP. From a Minimalist perspective it is more plausible to assume a spec-head relation in this case too, albeit one in which the feature [etop] is associated with a zero head. 14 This form of emphatic topicalization can also be used in Bavarian. Let me therefore suggest that the feature [etop] may be associated with a zero head in all varieties of German, whereas it can be associated with a class of overt complementizers only in Bavarian. This class of complementizers consists of daß, ob, wenn/wann, wia and bai (corresponding to daß, ob, wenn/wann, wie and sobald in the Standard language). These elements are lexically compatible with the feature [etop] such that this feature may be added in the numeration or not; daß may in addition be enriched with the feature [wh]. If daß is endowed with [wh], a wh-phrase will move toward SpecCP, checking its wh-feature against daß. If daß, ob, wenn/wann, wia and bai are endowed with [etop], an etop-phrase will move toward SpecCP checking its etop-feature against the respective head. The parametric difference that exists between Bavarian and Standard German with respect to the DFCF is summarized in table 1, in which we adopt the convention that α and β can freely adopt the values + or - 1 5

Within Rizzi's (1991) Criterion approach one could say that Bavarian complementizers are either positively or negatively specified for [wh] and [etop], whereas those of the standard language are always negatively specified for these features. Here I assume that there are only positive specifications of features. Negative specification amounts to the absence of a feature, which means that no checking can take place. Notice that in the embedded clause of Standard German a wh-phrase moves to SpecCP, whose head C must be zero. In terms of the present theory, the minus value amounts to the absence of the feature; thus, to say that daß is [ßetop] is a way of saying that daß can bear the feature [etop] or not.

Asymmetry in Emphatic Topicalization

29

Table 1:

0 daß ob wenn/wann wie (so)bald

Standard German

Bavarian

awh, ßetop

awh, ßetop awh, ßetop aetop aetop aetop aetop

-

-

-

-

-

Table 1 determines that both Standard German and Bavarian allow sentences in which the wh-phrase corresponds to no overt head. Both varieties equally allow constructions such as those in (16a) and (17a). According to the present proposal, (16a) has the partial structure in (23), in which the etop layer is completely above CP:16 (23)

[ETOPP

den Hansi [etop· [6,ορ 0 ]

[CP

daß er deni kennt]]]...

Bavarian ob, wenn/wann, wie and (so)bald are incompatible with a wh-feature, but are compatible with an etop-feature.17 This amounts to structures in which CP coincides with EtopP, the emphatically topicalized phrase now being in SpecCP/SpecEtopP. Example (3a) then receives the partial structural description in (24): (24)

[cp/etopp

da Xaven

[Cyetop· [c/etoP

daß] [ ίΡ ti an Mantl kaffi hot]]]...

With respect to the feature [etop] the rest of the set of complementizers seen in table 1 behaves accordingly in Bavarian.

16

17

The question here is how the feature [etop] can be checked in this case. There being a pronoun in situ, it is not plausible to assume that the left-dislocated phrase is moved. It rather seems to be base-generated where we see it. It remains to be shown how it acquires its Case and how it can be subject to feature checking for the feature [etop], I have to leave these questions for future research. One reviewer suggests that [Den Fritz, [wer t, kennt]] weiß ich nicht (the Fritzacc who knows, know I not) is grammatical in certain varieties of Bavarian. In my speech it clearly is not. The only grammatical option for me would be to resort to the resumptive pronoun strategy as in [Den Fritz ι [ 0 [wer den, kennt]]] weiß ich nicht. As Hoekstra (1993) points out, there are Dutch dialects in which the corresponding interrogative complementizer of is compatible with the wh-feature. Since ob and of are interrogative, a more fine-grained distinction would have to distinguish between [+interr, -wh] for all German varieties and Standard Dutch and [+interr, +wh] for the Dutch dialects Hoekstra describes.

30

5.2

Josef Bayer

Feature Combination and Wh-Scope

The feature specification of Bavarian daß seen in table 1 suggests that α and β may choose the positive value simultaneously. The result would be an emphatic wh-construction. Given that in Bavarian both [etop] and [wh] can be associated with SpecCP and the moved phrase binds a trace (and not a resumptive pronoun as in the LD-cases), no semantic conflict with the non-referentiality of a wh-phrase is expected. As example (25) shows, the combination of the features [etop] and [wh] indeed leads to a wellformed result. The emphatic nature of the question is signaled by stress and is indicated here with capital letters on the wh-phrase: (25) [WERi [daß [ti sein Hund ned fiadat]]]2 hot-a who that his dog not feeds has-he 'WHO did he think does not feed his dog?'

g'moantt 2 ? thought

This example differs from cases of regular wh-movement in interesting ways. 18 Notice first that according to the conservative approach which I am trying to defend here, the wh-phrase has not been moved into the root clause; it rather remains part of an embedded CP which has undergone movement to SpecCP of the root clause. As it stands, it is not at all obvious why the entire clause should be an interrogative. But notice that it has to be an interrogative. If it were an assertion, it could never be grammatical. The reason is that the attitude verb moana ('mean', 'think') is incompatible with a whcomplement as seen in the ungrammaticality of *Er hot g 'moant [wer daß sein Hund ned fiadat] ('*He thought who does not feed his dog'). Indeed, (25) turns into an ungrammatical declarative as soon as the stress on the wh-element is removed and shifted elsewhere, e.g., to Hund as in (26): (26) *[Weri [daß [ti sein HUND ned fiadat]]] who that his dog not feeds 'He thought who does not feed his dog?'

hot-a g'moant? has-he thought

The topicalized CP is then interpreted in its pre-movement position, and wh-scope gets activated in this lower position and not at all in the root. Thus, somehow (25) must be such that the wh-operator has scope over the entire clause. We have the interesting situation that the wh-phrase reaches matrix scope without moving into the matrix itself. Wh-scope is rather obtained by a combination of CP pied-piping and stress on the whphrase. I would like to suggest here that this is due precisely to the fact that the wh-

Regular wh-movement would be as in (i): (i) [Wer, [hot [-a g'moant [t,* [daß [t, sein Hund ned fiadat]]]]]]? who has -he thought that his dog not feeds Notice that (25) is not an echo question. The topicalized clause is subordinated, and the whphrase is in SpecCP. This is normally not the constellation which allows an echo interpretation. Thanks to Susanne Trissler for pointing this out to me.

Asymmetry in Emphatic

31

Topicalization

phrase may in addition to [wh] carry the feature [etop]. If emphatic topicalization is a root phenomenon as we have argued in section 4, and if the phrase which gives rise to emphatic topicalization is itself a wh-phrase, we expect matrix wh-scope. However, the missing link is still to understand how [etop] and [wh] can be activated in the root clause without the phrases which carry these features actually moving there. To fill this gap, I will in the next section develop an account in terms of pied-piping which will also enable us to give an explanation of the TA-effect.

5.3

Pied-Piping 1 9

We can substantiate the claim that [etop] is a root feature by considering recent proposals as to a more fine-grained articulation of the C-system. According to Rizzi (1997), the C-system is "the interface between prepositional content (expressed by the IP) and the superordinate structure (a higher clause or, possibly, the articulation of discourse, if we consider a root clause)." The structure proposed by Rizzi is given in (27):20 (27)

[ForceP •••[Force 0 [τορΡ*—[τορ°

[FOCP—[FOC°

[τορΡ* —[τορ° [FinP —[Fin 0

IP]]]]]]]]]]

Most relevant for our concern is the suggestion that ForceP is the highest projection of the clause. If emphasis is a pragmatic feature that can be imposed on the primary illocutionary force of the utterance, its place in ForceP is expected. There is an important difference, however, between the features that are subsumed under Force. The quote from Rizzi (1997) suggests that in a subordinate clause the highest projection of the Csystem mediates between the subordinate clause and its minimally superordinate clause, while in a root clause the highest projection of the C-system mediates between the root clause and the discourse. Closer scrutiny reveals, however, that force in the narrower sense of speech act theory can only be the latter.21 Imperatives, exclamatives,

20

21

Thanks to reviewer 2 for helping me remove certain unclear formulations in this section w.r.t. the role of features in the derivations I propose. My proposal is certainly not fully compatible with standard minimalist assumptions in Chomsky (1995). For lack of space I cannot discuss the differences here. This phrase structure has mainly been developed to account for word order data in Italian and French. The asterisk on TopP indicates that TopP can be iterated. For wh-sentences a sharp distinction has to be drawn between the formal feature [wh] and the contentive feature for interrogativity. Various clause types which are subsumed under [wh] have nothing to do with interrogativity. There are likewise interrogatives which appear in declarative mood (with rising intonation). Clear examples are also root yes/no questions, which differ formally from their embedded counterparts which are introduced by a complementizer such as if or whether (German ob).

32

Josef Bayer 11

etc. are generally not selected by predicates but depend directly on discourse. Assuming that [etop] is part of the force features associated with the C-system, we expect feature checking by movement to SpecCP (or in Standard German to the specifier of a separate projection for a left-dislocated e-topic above C). If we are right in assuming that [etop] is a root feature, the result of such simplex movement does not yield a grammatical output as long as the CP in which this movement takes place is not identical with the root clause. The examples in (4), which have the structure seen in (28), are such that emphatic topicalization has taken place in the embedded CP, while the root clause remains unaffected.

The feature [etop] resides in C° and therefore projects to C' and CP, but no further. An important question is this: If [etop] is a root feature which arguably has interpretive import, why should it be part of the feature structure of C? In my view, the answer must be that [etop] cannot be an interpretable feature in C. This makes it similar to a moved wh-phrase in a non-root clause of a partial-movement construction where we also find a wh-phrase in a position in which its wh-feature cannot be interpretable.23 I assume that [etop] is a formal feature that may be freely inserted in C, be it interpretable there or uninterpretable. Once in C, [etop] will attract XP for checking, a process after which XP loses its corresponding feature. This process leaves the embedded CP As has been pointed out in the text already (cf. section 4), the complexity of the issue is enhanced by the fact that certain predicates in German permit V2-complements which may show root properties. But it is debatable whether we are dealing with genuine subordination in these cases. (i) Was glaubst du [wen ich gesehen habe] what believe you whom I seen have 'Who do you believe I have seen?' (ii) *Du glaubst [wen ich gesehen habe] (i) shows that wh has moved to SpecCP of the embedded clause, although it cannot be interpreted there, as suggested by the ill-formedness of (ii). Obviously the wh-part has to be attracted by the root wh-element because only there can it be interpreted.

Asymmetry in Emphatic Topicalization

33

with the feature [etop] which is likewise uninterpretable because it resides in a constituent distinct from the force system associated with the root clause. In order to activate [etop] in the root clause, the embedded CP has to move to the specifier of the root clause. This leads to the grammatical output seen in the examples in (3), the structure of which is given in (29): (29)

rOOtCP[et0p]

The requirement that [etop] be a root feature has been met here by pied-piping the entire CP (which already bears the so far uninterpretable feature [etop]) to the specifier of the root CP. Since force can only be associated with the root CP (or its equivalent in a quasi-quotative context), checking the [+etop] CP against the head C[et0p] leads to the feature's interpretability. In the head of the root CP, [etop] has become part of the force system in the same way as [wh] may have become part of interrogative force in a root question. This analysis, which I believe is the only feasible one, leads to two questions in connection with the minimalist theory I am subscribing to. The first question is how piedpiping should be possible at all after the phrase marked [etop] has been checked against the complementizer of the embedded clause. The second question is why instead of the minimal phrase marked [etop] the entire CP should move, i.e., why pied-piping should be allowed if the same effect could be achieved by movement of a smaller phrase. Let me try to answer these questions in the following two sections. 5.3.1

How is Pied-Piping Possible?

According to our assumptions, the movement seen in (28) is motivated by feature checking: both C and XP bear the feature [etop], and XP will get rid of [etop] as soon as it is in SpecCP. As a result, it is now the embedded CP which bears the feature [etop]. This feature cannot be interpreted, however, unless it is the feature of a root CP. Assume now that [etop] may also be part of the feature complex associated with the head of the root CP. This is the place to which the finite verb moves. We know that the V2 property is strongly connected to clausal typing: Imperatives, yes/no interrogatives and conditionals in German rely on I-to-C movement, and obviously the feature of whclauses which specifies the ForceP for interrogativity is also dependent on I-to-C. I

34

Josef Bayer

take this as a motivation for proposing the principle in (30), which makes our earlier assumption that emphatic topicalization is a root phenomenon more precise: (30) [etop] cannot be interpreted unless it is part of the featural complex associated with [ c I]. The Principle of Full Interpretation (Chomsky 1986b) dictates that any information that remains in a representation be interpretable. As we saw, [etop] in the daß-CP is not. Thus, the daß-C? has to undergo movement to the specifier of the root CP (which is headed by the raised V+I). This kind of process is familiar: In the course of the derivation the formal features are deleted up to interpretability. If we are right, movement can be triggered by purely formal requirements before the final step is reached in which [etop] - now being interpretable - remains. 5.3.2

Why Pied-Piping?

The next question that needs to be answered is why the pied-piping option can be chosen if the same effect could be achieved by simply moving the phrase which primarily bears the feature [etop] to the specifier of the root CP. 24 Notice that roughly the same semantic effect as in (3) can be achieved by simply moving the e-topic phrase alone. (31) a. Da Xaveri hot neamad glaubt [daßti anMantl the Xaver has nobody believed that a coat 'As for Xaver, nobody believed that he bought a coat'

kafft hot] bought has

b. An Mantl] hot neamad glaubt [daß da Xaver ti a coat has nobody believed that the Xaver 'As for a coat, nobody believed that Xaver bought one'

kafft hot] bought has

c. Da Hansi woaß-e ned [ob ti kummt] the Hans know-I not whether comes 'As for Hans, I don't know whether he will come' d. An Fünferi häid-e ned g'moant [daß-e ti kriag] a five had-I not thought that-I get 'As for grade five, I didn't think I would get one' If pied-piping is a last-resort operation that applies in avoidance of a violation of principles of grammar, we would expect (31) rather than (3). Why is it then that both options exist? One reason might be that pied-piping enables us to apply emphatic topi24

I am not aware of an economy principle that favors the movement of small constituents over the movement of larger constituents of the same syntactic type, but it is not unlikely that such a principle could play a role in considerations of derivational economy. A principle like that could be seen as a natural extension of the idea that covert movement or feature movement be preferred over move-alpha.

Asymmetry in Emphatic Topicalization

35

calization even in cases in which movement of the smaller XP would be blocked by the Adjunct Condition. This can be seen in the difference between (32a) (which is identical to (5a)) and (32b): (32) a. Da Xaveri wenn ti hoam kummt kriagt-a wos z' essn the Xaver if home comes gets-he something to eat 'As for Xaver, if he comes home, he will get something to eat' b. *Da Xaverj kriagt(-a) wos the Xaver gets(-he) something

z' essn [wenn t] to eat if

hoam home

kummt] comes

Under the present theory, pied-piping of the adjunct clause as seen in (32a) could be seen as an option which is chosen if direct movement of the critical XP (here da Xaver) would violate the restriction against movement from adjunct clauses. But even then, the question remains as to why the pied-piping option may still be chosen where XPmovement is legitimate, as is the case in the examples in (31). My suggestion is that the difference between (3) and (31) rests on a difference in the numerations with which the respective derivations start. In examples of type (3), in which we see pied-piping, the complementizer of the embedded CP bears the feature [etop], which, however, cannot be interpreted in this position whence the entire CP undergoes movement to the left periphery of the root CP. The difference in the numeration underlying examples of type (31) is that the complementizer of the embedded CP does not bear the feature [etop]; only the C-position of the root CP does. As a consequence, XP does not move for checking to the specifier of the CP which minimally contains it. It rather moves directly to the specifier of the root CP to check the feature [etop]. In order to make this possible we must assume the operation Form Chain, which has been suggested by Chomsky & Lasnik (1993).25 This operation is distinct from move-alpha because under the Form Chain approach, long movement may rest on a single instance of movement, independent locality conditions (such as Minimal Links) taking care of the wellformedness of the resulting chain. The difference between direct movement of XP to SpecCP of the root clause and movement of the embedded CP to SpecCP of the root clause by pied-piping then boils down to a difference in the respective numerations underlying the derivations, which can be described as in (33):

Reprinted as chapter 1 of Chomsky (1995); see p. 44.

36

Josef Bayer

(33) a. The Form Chain option [rOOtCPetop ... [CP ... XPe,op ... ]]

4

I

b. The Pied-Piping option [rOOtCPetop ... [CPetop ... XP e ,op -

]]

One may ask why XP cannot move out of the lower CP and directly move on to the root CP such that pied-piping would still be avoided, despite the difference in terms of numerations. There is a theoretical as well as an empirical answer. The theoretical answer is that once XP has checked its feature against [etop] in C, it becomes inert for further movement as conditioned by [etop]. The empirical answer is that whenever movement of XP to SpecYP is observed, further movement of XP (to SpecZP) appears to be blocked. Since Ross (1967) this has been known as the Freezing Effect?6 Movement of the domain in which XP-movement has taken place is widely attested, however. In the present account this is expected since the feature that functions as the "checker" may still remain active in the sense that it must undergo movement itself in order to reach a position at which it can attain full interpretation. If we are right in arguing that emphatic topicalization is a root phenomenon, pied-piping of CP to the specifier of the root clause is a way of satisfying the semantic needs of the feature [etop]. 5.3.3

Recursive Pied-Piping

Grewendorf (1988:256) observes that emphatic topicalization may apply recursively. His example is given in (34). According to our theory, it has the derivation seen in (35a) through (35d): (34) Da Peter daß bled is,daß-e g'sagt hom soi, is glatt g'long the Peter that stupid is that-I said have should is outright lied 'As for Peter, it is an outright lie that I am supposed to have said that he is stupid'

See also more recent considerations of the Freezing Effect in Müller (1997) and Bayer (1999). With respect to semantics, the clearest effect of freezing is the scope of operators. Once an operator has reached a scope position, its scope can never be altered. For details see Bayer (1996) as well as Bayer & Grosu (2000).

Asymmetry in Emphatic

31

Topicalization

(35) a. [cp is glatt g'long [cp daß-e g'sagt hom soi [CP daß da Peter bled is ]]] b. [cp is glatt g'long [cp daß-e g'sagt hom soi fcp da Peteri daß ti bled is ]]] c. [CP is glatt g'long [CP [cp da Peteri daß ti bled is]2 daß-e g'sagt hom soi t2 ]] d. [CP [CP [CP da Peteri daß ti bled is]2 daß-e g'sagt hom soi T2]3 is glatt g'long Tß] The derivation is driven by repeated checking of the feature [etop], which is attributed to C in each cycle. Since [etop] can only be interpreted at the root CP, the process of topicalization must carry on until the root is reached. The steps (35a) through (35c) cannot surface because the feature [etop] cannot be interpreted yet; only (35d) can. In section 2 we have seen that emphatic topicalization is allowed in certain adjunct clauses. The example in (36) shows that recursive pied-piping may also apply under the involvement of adjuncts of this type: (36) An Fünfer daß-e kriagt hob wenn der Vater ned merkt bin-efroh a five that-I gotten have if the father not notices am-I happy 'As for grade five, I am happy if father does not notice that I've gotten one' The recursivity observed in constructions like (34) and (36) strongly supports the analysis in terms of pied-piping of CP. 5.3.4

Free Ride

Example (25) has demonstrated that the scope of [wh] can be extended to the root CP, although - according to my proposal - the wh-phrase remains in the specifier of the embedded CP. The example is repeated here for convenience: (25) [WER, [daß [t,sein Hund ned fiadat]]]2 who that his dog not feeds 'WHO did he think does not feed his dog?'

hot-a has-he

g'moantt 2 ? thought

The embedded CP is headed by daß. Since this CP is selected by a verb which is incompatible with a wh-complement, daß can be assumed to lack a wh-feature. On the other hand it may contain the feature [etop]. If the wh-phrase wer is marked [wh, etop], it will move to SpecCPet0p giving [wh] a "free ride" (cf. Chomsky 1995). The embedded CP is now marked [wh, etop] and will move on to the specifier of the root CP, where the feature [etop] can be interpreted. Under the realistic assumption that the features [etop] and [wh] do not exclude each other, the root CP can in addition bear the feature [wh]. Thus, if the head of the root CP is marked [wh], the relevant part of the raised CP's representation can appear in reconstructed position in such a way that the wh-part of wer is stripped off, essentially leaving an indefinite. Thus, wh-scope has been obtained by pied-piping the embedded CP due to a feature that is entirely independent of the feature [wh]. We see a property of the system here which I think speaks

38

Josef Bayer

favorably for the modular character of a grammar in which semantic effects may be achieved as a consequence of morphosyntactic necessity.

6.

Parasitic Gaps

Let me finally consider parasitic gaps (PG) as briefly introduced above by example (10), which is repeated here: (10) An Karli wenn-eti dawisch daschlog-e the Karlacc if-I get kill-I 'If I get hold of Karl, I'll kill him'

ei

As Lutz (1997) notes, it is unclear how the PG ei can be licensed if its binder - the emphatically topicalized phrase - remains in SpecCP of the topicalized clause. This seems to challenge the analysis we have presented above. In the rest of this article, I want to sketch a solution that preserves the present analysis while removing the observed incompatibility. Let me start with an observation from Bayer (1988) about the Bavarian PG-constructions which I take to be of central importance for an adequate account. As the minimal pair in (37) shows, the presence of the resumptive adverb dann ('then') destroys the licensing of the PG. (37a) is grammatical because of the presence of the clitic pronoun -n (which is the reduced form of German ihn). The PG-construction in (37b), in which we see a gap instead of the clitic pronoun, is, however, sharply ungrammatical: (37) a. An Karli wenn-e ti dawisch dann the Karlacc if I get then 'If I get hold of Karl, then I'll kill him'

daschlog-e-nj kill-I-him

b. *An Karli wenn-e ti dawisch dann daschlog-e ei As I have argued in Bayer (1988), the primary licensing of the PG can therefore not be due to the topicalized phrase in the preposed CP, but must be connected to the position in which the element dann occurs. This leads us to the account suggested in Contreras (1984) and Chomsky (1986a), according to which the PG is bound by a null operator (OP), and there is a process of chain composition as indicated by the dotted line in (38): (38)

XP, ... t, ... [OP 2 ... e 2 ...]

where 1 = 2

39

Asymmetry in Emphatic Topicalization

Applied to (37b), the argument would be that if the position for OP is occupied by dann, the PG (i.e., e 2 in (38)) remains unbound. The examples in (19) have shown that a quantified NP can be moved to SpecCP by emphatic topicalization. The failure of binding which was observed in (12) demonstrates that the QNP is not in a position "high enough" for c-commanding the pronoun which could be a potential variable. The examples are repeated here: (19) a. Neamadi wenntikummt, no is-s aa ned recht nobody if comes then is-it also not good 'If nobody shows up at all, it isn't ok either' b. A jederi daß tj so deppert is glaub-e a everyone that so stupid is believe-I Ί don't think that everybody is that stupid'

ned not

(12) a. *Köa Menschi wenn ti b'suffa is foit eami wos g'scheids el no man if drunk is falls him something useful in intended: 'Nobody has good ideas when he is drunk' b. *A jeder Menschi wenn ti niachdan is foit eami an every man if sober is falls him g'scheids ei useful in intended: 'Everyone has good ideas when he is sober'

wos something

In spite of emphatic topicalization, the scope of the QNP is confined to the minimal clause from which it has been moved. This is exactly what the theory presented so far makes us expect. But consider now the following examples of PG-constructions with QNPs, which are taken from Lutz (1997): (39) a. A jeda Hund ti wann ti sei Fuada kriagt a every dog when his food gets 'Every dog, when it gets food, wags its tail'

wedelt ei mit'n wags with-the

b. Köa Hund ti wenn ti g'schlong wead traut no dog when beaten becomes ventures no amoi zur Tür rel yet once to-the door in 'No dog, if beaten, dares to enter the door once again'

Schwanz tail

se ei REFL

As Lutz observes, the QNP now has scope over the matrix clause, a fact which at first sight seems to be in blatant conflict with our suggestion that the QNP is an emphatic topic which is in the specifier of the topicalized adjunct CP. This conflict has led Lutz to propose different phrase structures for the PG-construction and for what he calls the "focus construction" (cf. (13) above). These two constructions are shown in (40):

Josef Bayer

40 (40) a. The focus construction according to Lutz (1997) CP,

b. The PG-construction according to Lutz (1997) CP,

For reasons I cannot discuss here, Lutz assumes that in the PG-construction the PG is in the adjunct clause, and the actual trace is in the matrix clause (see also note 6). He remains neutral as to the attachment of XP3 and CP2, which 1 represent here as part of a multiple specifier construction for merely expository reasons. It is important for Lutz to express the fact that the leftmost XP c-commands both the PG and the real gap. He manages to retain a connection with the construction in (40a) by postulating an empty operator (OP) in the position where normally the emphatically topicalized (in Lutz's terminology "focus") phrase is found. The adjunct CP must somehow be in the left periphery. Otherwise examples like those in (41) - which are modeled after (39) - would be expected to be grammatical not only with a pronoun (here with the clitic -a for 'he'), but also with a gap (here e,):

41

Asymmetry in Emphatic Topicalization (41) a. A jeda Hundi wedelt - wann-{ai /*ei} a every dog wags if-it ti mit'n Schwanz with-the tail b. Köa Hundi traut no dog ventures ti no amoi zur yet once to-the

sei Fuada his food

kriagtgets

se - wenn-{ai /*ei} g'schlong w e a d REFL when-it beaten become Tür rei door in

(41a,b) show that the adjunct clause can be used parenthetically, but never under the condition of PG-licensing. How would the left-peripherality of the adjunct clause in the PG-construction follow if the structure were as in (40b), i.e., if the adjunct were essentially a parenthetical? My suggestion is to keep the parenthetical constructions in (41) apart and account for the PG-construction by deviating as little as possible from the general analysis, according to which the emphatically topicalized XP remains in the specifier of the adjunct CP. In order to do this, I return to an analysis of the PG-construction that has been proposed by Huybregts & van Riemsdijk (1985). These authors suggest analyzing PGconstructions in terms of an Across the Board (ATB) rule application. ATB rule application has been intended for coordinate constructions. 27 An application to the PG-phenomenon requires two things: Firstly, the clauses from which simultaneous leftward movement takes place must conform to the parallelism that normally holds in coordinate constructions, and secondly, there must be independent evidence that the hypotactic construction resembles or can be reanalyzed as a paratactic construction. 2 Parallelism as observed in the Dutch and English cases Huybregts & van Riemsdijk have in mind is clearly given in the Bavarian cases too. The grammaticality difference between (37a) and (37b) shows that the PG has to be bound by a null operator. If it is bound by a null operator, the emphatically topicalized XP and its trace form a parallel chain. As to the second requirement, there is evidence that the conditional introduced by wenn,

27 28

Cf. Williams (1978). For the kind o f reanalysis required here, Huybregts & van Riemsdijk (1985:10) have coined the term "insubordination''.

42

Josef Bayer 29

which forms the core of the Bavarian PG-constructions, shows signs of parataxis. If I was right about the core configuration of the Bavarian PG-construction in Bayer (1988), the topicalized conditional CP cannot occupy SpecCP but must be either adjoined to CP or merged into the specifier of another head. Given our overall account, let us assume that the latter is the case. SpecCP is then filled by the null operator OP, and there is a higher projection ZP which licenses the interpretability of the feature

The chains (XP3, t3) and (OP4, e4) show the required parallelism, and the positioning of the two clauses CP2 and CPi conforms to the phrase structure that is standardly assumed for coordinate constructions in antisymmetric syntax and its predecessors. 3 My suggestion is that ATB applies to CP2 and CP] in such a way that the head of the original chain, here XP3, is adjoined to ZP, licensing the null operator, here OP4. According to Lasnik & Stowell (1991), who adopt the Contreras/Chomsky chain composition analysis of PG, OP is not a quantifier but a name-like null operator, in their terminology a null-constant or a null R-expression. Assuming that by moving a non-referential

30 31

Consider the examples (i) through (iii) from Standard German, of which (iii) is taken from König & van der Auwera (1986). In (i), the conditional occupies SpecCP, while in (ii) and (iii) it precedes SpecCP: (i) Wenn das mein Hund wäre, bekäme er keinen Zucker if that my dog were got he no sugar 'If that were my dog, it would not get any sugar' (ii) Wenn das mein Hund wäre, dann bekäme er keinen Zucker (iii) Wenn das mein Hund wäre, er bekäme keinen Zucker In the terminology of König & van der Auwera (1986), (i) shows integration of the conditional by having it right in the Vorfeld of a V2-clause; (ii) shows resumption by dann, which arguably occupies the position which hosts the conditional CP in (i); (iii) shows non-integration by putting a topic in SpecCP instead of the resumptive dann. In this case, the conditional appears to be completely separated from the root. For support of the idea that conditionals are topics, see Haiman (1978). Cf. Kayne (1994) and references given therein.

Asymmetry in Emphatic

Topicalization

43

phrase (QNP, wh-phrase), ATB identifies OP as the head of the parasitic chain, and (42) turns into the structure seen in (43):

It is important to notice here that the leftward adjunction of XP to ZP is not the result of emphatic topicalization but the result of ATB. We predict that it applies only in cases where two clauses are involved which exhibit the necessary parallelism. The deviance of examples like (12a,b) is expected because the required parallelism is lacking, and the ATB rule cannot apply. In (12a,b) we find a pronoun instead of a gap, which would be bound by OP. Thus, since OP is lacking, the ATB process cannot apply, and the QNP of the conditional clause remains in SpecCP. As a consequence, the bound variable interpretation of the pronoun is unavailable. Bavarian quantified sentences provide independent evidence for the correctness of this analysis. As predicted by Lasnik & Stowell, movement of a QNP gives rise to crossover or weak crossover (WCO) effects, but movement of the null operator does not (or only gives rise to "weakest" CO). To see that this prediction is borne out, consider the pair of examples in (44): (44) a. ?*An jedn Hundi wann sei] Herr ti ziagt [OPi an every dog if his master pulls intended: 'Every dog gets enraged if its master pulls it'

packt grabs

ei d'Wuat] the rage

b. An jedn Hundi wann-st ti ziagst [OPi druckt sei] Hoisbandl ei ] an every dogacc if-you pull presses his throat-band 'Every dog is troubled by his collar if you pull it (=the dog)' Movement of the QNP to the specifier of wann in (44a) elicits a WCO violation. However, movement of OP to the specifier of the root clause in (44b) does not.

Josef Bayer

44

To summarize, we have shown that the dilemma that has been revealed by Lutz (1997) can be avoided by having an ATB rule apply to the output of emphatic topicalization. To the extent that the ATB approach to the PG-phenomenon is tenable in general, it seems to serve as the most parsimonious and straightforward extension of my analysis of emphatic topicalization as the phrase structure for emphatic topicalization can be monotonously expanded.

7.

Summary

The empirical starting point of this article was a topicalization asymmetry (TA) which holds in Bavarian and similar Southern German dialects. The topicalization of this kind applies in a dependent clause, but is only possible if the embedded clause itself is topicalized to the left periphery of the root clause. My conjecture was that the observed topicalization is possible because Bavarian lacks the Doubly-Filled-COMP Filter (DFCF). Once the phenomenon is connected to the lack of the DFCF, however, there seem to be two major questions. One question is why there should be the TA (while there is no such asymmetry with respect to regular wh-movement); another is how this kind of topicalization can license parasitic gaps (PGs) if the moved element remains in the embedded clause and thus fails to c-command the PG. To approach the first question, I suggested that the topicalization involved is an emphatic topicalization, and should as such be characterized as a property of the illocutionary force of the utterance. Since illocutionary force is syntactically a root phenomenon and is - at least in the western languages - anchored in the left periphery of the root CP, it could be argued that emphatic topicalization is only semantically successful if it can access the root CP. We have argued at length that there is a formal feature [etop] which can be checked by a C-head but can only be interpreted if it is associated with the force system, which by definition is part of the root. The feature [etop] can access the root clause although it remains in the specifier of the embedded CP if CP moves to the specifier of the root clause - essentially a case of pied-piping. To answer the second question, we suggested that in Bavarian PG-constructions the basic configuration of emphatic topicalization remains the same, the difference being that an Across the Board (ATB) rule applies to both CPs simultaneously. The viability of this approach is supported by the fact that SpecCP of the root clause (in which we located the PG) must be able to host a null operator by which the PG can be locally bound. Once SpecCP is occupied by the resumptive element dann, no PG can be licensed. The conceptual advantage of the present approach is that hardly any tampering with the concept of the German Vorfeld was necessary, and virtually all the relevant differences between Standard German and Bavarian could be derived from a low-level parametric difference according to which the standard language obeys the DFCF while Bavarian does not.

Asymmetry in Emphatic

Topicalization

45

Acknowledgments The present work was supported by a research grant from the Deutsche Forschungsgemeinschaft within the Leipzig/Jena Forschergruppe Sprachtheoretische Grundlagen der Kognitionswissenschaft. I profited from discussions with Markus Bader, Uli Lutz, Susanne Trissler and Helmut Weiß, who, however, must not be blamed for any shortcomings. Many thanks also to two anonymous reviewers for valuable comments and to Anne Breitbarth and M.T. Hany Babu for their help in preparing the final version. And finally: Many thanks to Arnim for having been a great teacher and a great source of inspiration and fun.

References Bayer, Josef (1984). COMP in Bavarian Syntax. The Linguistic Review 3. 209-274. Bayer, Josef (1988). Fortschritte der Syntaxtheorie. Anmerkungen zu Henk van Riemsdijk und Edwin Williams. Introduction to the Theory of Grammar. Cambridge, MA: MIT Press. 1986. Linguistische Berichte 117. 410-426. Bayer, Josef (1991). German Particles in a Modular Grammar Neurolinguistic Evidence. In: Werner Abraham (ed.) Discourse Particles. Amsterdam: Benjamins. Bayer, Josef (1996). Directionality and Logical Form: On the Scope of Focussing Particles and Wh-in-Situ. Dordrecht: Kluwer. Bayer, Josef (1999). Final complementizers in Hybrid Languages. Journal of Linguistics 35. 233-271. Bayer, Josef (to appear) Wh-in-Situ. SynCom. University of Utrecht. Bayer, Josef & Alexander Grosu (2000). Feature Checking Meets the Criterion Approach: Three Ways of Saying Only in Romance and Germanic. In: Virginia Motapanyane (ed.) Studies in the Comparative Syntax of Romanian. Amsterdam: Holland Academic Graphics. Behagel, Otto (1932). Deutsche Syntax. Eine geschichtliche Darstellung. Vol. IV. Heidelberg: Carl Winters Universitätsbuchhandlung. Brandt, Margareta, Marga Reis, Inger Rosengren & Ilse Zimmermann (1992). Satztyp, Satzmodus und Illokution. In: Inger Rosengren (ed.) Satz und Illokution. Vol.l. Tübingen: Niemeyer. 1-90. Chomsky, Noam (1986a). Barriers. Cambridge, MA: MIT Press. Chomsky, Noam (1986b). Knowledge of Language: Its Nature, Origin and Use. New York: Praeger. Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam & Howard Lasnik (1993). The Theory of Principles and Parameters. In: Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld & Theo Venne-

46

Josef Bayer

mann (eds.) Syntax: An International Handbook of Contemporary Research. Vol.1. 506-569. [Reprinted as chapter 1 of Chomsky, 1995]. Cinque, Guglielmo (1998). Adverbs and Functional Heads: A Crosslinguistic Perspective. Oxford: Oxford University Press. Contreras, Heles (1984). A Note on Parasitic Gaps. Linguistic Inquiry 15. 698-701. Doherty, Monika (1985). Epistemische Bedeutung. Berlin: Akademie-Verlag. Emonds, Joseph (1969). Root and Structure-Preserving Transformations. PhD dissertation, MIT. Erteschik-Shir, Nomi (1997). The Dynamics of Focus Structure. Cambridge: Cambridge University Press. Felix, Sascha (1985). Parasitic Gaps in German. In: Werner Abraham (ed.) Erklärende Syntax des Deutschen. Tübingen: Narr. 173-200. Grewendorf, Günther (1988). Aspekte der deutschen Syntax. Tübingen: Narr. Haiman, John (1978). Conditionals are Topics. Language 54. 564-589. Hasegawa, Kei (1999). Exklamativsätze im Deutschen und im Japanischen. Μ A thesis, Friedrich-Schiller-Universität, Jena. Heycock, Caroline (2000). Embedded Root Phenomena. SynCom, case #035, University of Utrecht. Hoekstra, Eric (1993). Dialectal Variation Inside CP as Parametric Variation. In: Werner Abraham & Josef Bayer (eds.) Dialektsyntax. Opladen: Westdeutscher Verlag. 161-179. Hooper, Joan Β. & Sandra A. Thompson (1973). On the Applicability of Root Transformations. Linguistic Inquiry 4. 465-497. Hornstein, Norbert (1995). Logical Form: From GB to Minimalism. Oxford: Blackwell. Huybregts, Riny & Henk van Riemsdijk (1985). Parasitic Gaps and ATB. Tilburg Working Papers in Language and Literature 76. Kayne, Richard (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT Press. König, Ekkehard & Johan van der Auwera (1986). Clause Integration in German and Dutch Conditionals, Concessive Conditionals and Concessives. In: John Haiman & Sandra A. Thompson (eds.) Clause Combining in Discourse and Grammar. Amsterdam: John Benjamins. 101-133. Lasnik, Howard & Timothy Stowell (1991). Weakest Crossover. Linguistic Inquiry 22. 687-720. Lutz, Ulrich (1997). Parasitic Gaps und Vorfeldstruktur. In: Franz-Josef d'Avis & Ulrich Lutz (eds.) Zur Satzstruktur des Deutschen. Arbeitspapiere des Sonderforschungsbereichs 340, No. 90. 55-80. Müller, Gereon (1997). Incomplete Category Fronting. Dordrecht: Kluwer. Rizzi, Luigi (1990). Relativized Minimality. Cambridge, MA: MIT Press. Rizzi, Luigi (1991). Residual Verb Second and the Wh-Criterion. Technical Reports in Formal and Computational Linguistics 2. University of Geneva. Rizzi, Luigi (1997). The Fine Structure of the Left Periphery. In: Liliane Haegeman (ed.) Elements of Grammar. Dordrecht: Kluwer. 281-337. Ross, John R. (1967). Constraints on Variables in Syntax. PhD dissertation, MIT.

Asymmetry in Emphatic

Topicalization

47

Stechow, Arnim von & Wolfgang Sternefeld (1988). Bausteine syntaktischen Wissens. Ein Lehrbuch der generativen Grammatik. Opladen: Westdeutscher Verlag. Weiß, Helmut (1998). Syntax des Bairischen. Studien zur Grammatik einer natürlichen Sprache. Tübingen: Niemeyer. Weiß, Helmut (to appear) On Two Types of Natural Languages: Some Linguistic Consequences. Theoretical Linguistics. Williams, Edwin (1978). Across-the-Board Rule Application. Linguistic Inquiry 9. 31-55.

The Resultative Parameter and Restitutive Again Sigrid Beck and William Snyder

1.

Introduction

This paper provides evidence for a semantic parameter. The parameter consists in the (un-)availability of a principle of semantic composition creating accomplishments. The evidence comes from crosslinguistic variation in the availability of restitutive readings for again. Von Stechow (1995) proposes a special interpretation principle that handles the interpretation of resultative constructions. It introduces a causation relation between two predicates and thereby constructs an accomplishment type predicate syntactically, from an activity (the matrix predicate) and a state (the result predicate). We call this principle (R). Snyder (1995, 2000) argues that the availability of such an interpretation principle is subject to parametric variation. Not all languages permit resultative constructions. This is viewed here as a semantic parameter: lack of a matching interpretation principle renders the constructions uninterpretable, hence ungrammatical. The focus of (von Stechow 1995) is repetitive vs. restitutive again. He provides an analysis of restitutive again with both syntactic and lexical accomplishments (relying on the resultative interpretation principle and on decomposition, respectively). In this paper we examine crosslinguistic variation in the availability of restitutive readings. We find that languages that do not have resultatives, i.e., do not allow the construction of accomplishments in the syntax, permit restitutive readings with a more limited range of predicates than languages that do have resultatives. The former disallow restitutive readings with predicates like 'walk to the summit', which we will call goal-PP constructions; they do permit restitutive readings with lexical accomplishments. We conjecture that the interpretation of goal-PP constructions involves the parameterized interpretation principle (R) in the languages that have it: An activity matrix predicate is combined with a result predicate to form an accomplishment. They do not denote an accomplishment in (R)-less languages. Our research supports the view that the proposed parameter consists in the availability of an interpretation principle: Goal-PP constructions are grammatical, but not interpretable as accomplishments in languages without (R). It also supports von Stechow's (1995) view that the analysis of lexical accomplishments should be different from the analysis of syntactic accomplishments. Finally, the crosslinguistic data indi-

The Resultative Parameter and Restitutive Again

49

cate that the result state for restitutive again needs to be linguistically specified. It seems that again does not have access to an independent, conceptually given target state. We first present the evidence from Snyder (1995, 2000) for a resultative parameter (section 2). In section 3, we introduce von Stechow's (1995) analysis of resultatives and extend it to goal-PP constructions. Section 4 discusses von Stechow's analysis of restitutive again with lexical accomplishments and with resultative constructions (section 4.1). We then report the behaviour of restitutive again crosslinguistically (section 4.2) and analyze it in section 4.3. Section 5 concludes the paper.

2.

The Resultative Parameter

The starting point for the present investigation is the observation that the availability of resultative constructions like (1) varies across languages. While English, for example, allows them, they are ungrammatical in Spanish. (1)

Mary beat the metal smooth.

(2)

Maria Mary

golpeo el beat the

metal metal

(*liso) (*smooth)

This crosslinguistic variation in the availability of resultative constructions is tied to variation in other points of grammar. Snyder (1995, 2000) brings forth both crosslinguistic and acquisitional evidence that relates resultatives (among other things) to verbparticle constructions, pi/i-locatives and notably, the availability of root compounding as a productive means of word-formation. He reports that in a crosslinguistic survey, languages permit the direct equivalent of (1) only if they allow novel compounds such as (3). Thus, in contrast to English, Spanish (2), (4) disallows both. (3)

worm can ('container for the storage of fishing worms')

(4)

bote ""(de) can *(of)

gusanos worms

The resultative construction is relatively infrequent in spoken English, but verb-particle constructions such as (5) are quite common, and show a similar association with productive root compounding. (5)

Chris picked the book up.

In a study of ten children learning English, Snyder found that each child began producing verb-particle constructions at almost the exact age when she or he began producing novel (non-lexical) noun-noun compounds.

50

Sigrid Beck and William Snyder

The association with root compounding is unidirectional, however: Languages with resultatives or verb-particle constructions consistently have productive root compounding, but the opposite does not hold. Basque is an example of a language that allows productive root compounding, but disallows both resultatives and verb-particle constructions. Compounding thus seems to be a prerequisite for allowing resultatives. We propose that the availability of resultatives depends both on root compounding, and on a parameter of semantic interpretation, which we term the Resultative Parameter. In languages with the positive setting of the Resultative Parameter, accomplishment predicates such as 'beat smooth' can be created in the syntax, by combining a verb describing a simple activity like beat, with an adjective phrase describing a result state. Below we will formalize the Resultative Parameter in terms of von Stechow's (1995) analysis of resultatives. First, however, we follow Snyder (1995) in proposing that the Resultative Parameter is also responsible for another point of crosslinguistic variation between English and Spanish observed in Aske (1989). In English one can create an accomplishment predicate by combining an activity verb such as walk with a goal PP such as 'to the summit'. Hence, a delimiting temporal phrase 'in an hour', which is possible with accomplishments but not with activities, is impossible in (6) but fully grammatical in (7). (6) *Paul walked in an hour. (7)

Paul walked to the summit in an hour.

In contrast, in Spanish the temporal delimiter is impossible with an activity verb even when a goal PP is present: (8)

Pablo Pablo

camino hasta walked up-to

la the

cima (""en una hora). summit (*in one hour)

We propose that the crucial difference between English and Spanish is again that English permits construction of accomplishments in the syntax - in the present example from an activity verb and a goal PP - while Spanish does not. This combination of an activity verb with a PP specifying a goal is what we call a goal-PP construction.

3.

Von Stechow (1995) on Resultatives

3.1

Resultatives and the Composition of an Accomplishment Predicate

Von Stechow (1995) provides an interpretation principle specifically for resultative constructions. Below is a German example and the structure von Stechow associates with it:

The Resultative Parameter and Restitutive Again (9)

51

a. Olga ihren Sohn gesund betete. Olga her.Acc son.Acc healthy prayed 'Olga prayed her son healthy' b. Fritz das Eisen.Acc flach hämmerte. Fritz the.Acc iron.Acc flat hammered 'Fritz hammered the iron flat'

(10) [[das Eisen] [ l [ v p Fritz [ v , tl [ v , [ SC PR01 flach] hämmerte]]]] The small clause does not combine with the matrix verb in a straightforward way. Von Stechow suggests the following interpretation principle for such structures:1 (11) If a = [ s c ß VT] and β' is of type and γ' is of type =.001 to this result. In other words, the probability of obtaining this degree of association simply by chance is one in a thousand. A possible worry might be that some of the languages in this study are fairly closely related to one another. Among the twelve languages we are currently considering are six Indo-European ones (English, French, German, Hindi/Urdu, Serbian/Croatian and Spanish). Note that the availability of (R) divides the Indo-European languages by language family. Thus it might make more sense to count the language families and consider those six languages as the Germanic languages versus the Romance and Slavic languages and Hindi/Urdu, counting the evidence from English and German only once and similarly for French and Spanish. This leaves us with ten language families/languages, for which we still find the association of (R) and the restitutive reading at a significant level. (Four language families are positive for both, and six are negative for both, with no mixed results; this yields a probability p=.005, or five chances in a thousand, by Fisher Exact Test.) We conclude that the behaviour of restitutive again corresponds to the availability of resultatives, and provides further evidence for the proposed parameter.

1.3

Interpretation: No Result State Available in (R)-less Languages

Let us think about the theoretical consequences of our results regarding the interpretation of goal-PP constructions. The case of (R)-languages is clear: We have given a derivation of the two readings of 'Sally walked to the summit again' in section 4.1. This is what we propose happens in all (R)-languages. Given that we observe a perfect correspondence of (R) and restitutive readings, unavailability of (R) must be sufficient to lead to unavailability of a restitutive reading. The question, then, is how the lack of (R) leads to the unavailability of that reading. On von Stechow's theory, a restitutive reading is possible if again can modify a result state. Thus, our crosslinguistic results imply that in (R)-less languages, there is no result state accessible to again in goal-PP constructions. This subsection is a discussion of how we can design an appropriate semantics for goal-PP constructions that has this property. We should stress that the discussion is by no means complete.

The Resultatixe Parameter and Restitutive Again

65

Informally speaking, our idea for what goes on in (R)-less languages like Spanish is this: In (R) languages, the predicate 'walk to the village' makes available at least two different eventualities: the accomplishment predicate as a whole, and the result state of being at the village. In (R)-less languages, however, there is only one eventuality that could have happened again: walking to the village (whatever the exact meaning of that predicate is). (R) enables one to interpret 'to the village' as the description of an independent event. Lacking (R), the goal PP can only be a regular modifier, and we do not expect an ambiguity any more than we do with 'walk in the park'. We followed von Stechow's (1995) formal theory, which does not introduce eventualities (but compare von Stechow 1996). Instead of modifying a property of events (as in our informal description above), again modifies a proposition, in this theory. Translating our idea for what goes wrong in (say) Spanish into this framework could look like this: Again applies to a proposition. Interpreting 'to the village' in Spanish as a proposition results in uninterpretability, due to lack of (R). The goal PP must be a modifier (say, type « e , < s , < x , t » > , < e , < s , < x , t » » ) . The only proposition to which again can apply is the whole proposition 'Suresh walk to the village'. This leaves open the question of what modifier precisely 'to the village' denotes, and how it comes about that the interpretation of the simple sentence 'Suresh walked to the village' seems so similar in Spanish and English. (After all, it was only the behaviour of 'in two hours' and restitutive again that made us suspect that there was any difference at all.) We tentatively suggest (77). (This is specifically for Spanish; we have not tried to determine the precise meaning for every (R)-less language, and there might be some variation.) (77) [[to the village]](P)(x)(w)(t)=l iff P(x)(w)(t)=l & χ is at the village at rb(t) in w. (78) [[Suresh walk to the village]](w)(t)=l iff Suresh walks in w at/during t and Suresh is at the village in w at the end of t. Thus (78) means that Suresh walks and ends up at the village, which seems appropriately similar to English. The difference between to and at, in either Spanish or English, would be that to makes reference to the temporal endpoint, so at could look like (79): (79) [[at the village]](P)(x)(w)(t)=l iff P(x)(w)(t)=l & χ is at the village at/during t in w. The change in the semantics of to is compatible with our story about English; we simply c h a n g e ' t ' to 'rb(t)' wherever we gave a semantics for to. In (80), below, is the reformulation of (14). (80) Sally walked to the summit —» λΛνλΐ. CAUSE w ,t walk w ',t' (Sally), Xw"Xt". BECOMEw» j» (Xw*Xt*. at w *,' r b(t*) (the summit) (Sally))

(kwxa.

66

Sigrid Beck and William Snyder

The crucial difference, in our view, between English and Spanish is that in English a prepositional phrase can denote a proposition, while in Spanish it cannot. Yet, it does not follow that we have to use to and not at in an English goal-PP construction. At present we have no explanation for the impossibility of at in a goal-PP construction. This is a possible analysis of goal PPs that would indeed preclude modification by again. We would like to stress that the crucial aspect of our suggestion is that the goal PP in 'walk to the summit' is treated in an analogous way to the PP in (81). (81) Sally walked in the park again. Example (81) is unambiguous; in particular, (82) is not a possible reading. (82) Sally walked, and she was once more in the park. This reading would come about if again could modify just the PP 'in the park'. This is impossible. We suggest that 'to the summit' in Spanish (78) has the same semantic status as 'in the park' in (81). It is worth stressing this point because technically our explanation is that there is a simple type mismatch, and type mismatches can be overcome in a number of ways. Example (81) shows that we do not want to do this in the present case: PP modifiers cannot in general be modified by again. If 'to the summit' is simply a PP modifier, we expect that it cannot be, either. There is one more possibility for deriving a restitutive reading for goal PPs that we should consider. (Thanks to an anonymous referee for pointing this out to us.) Rapp & von Stechow (1999) propose a decomposition analysis of PPs with to that is very similar to the decomposition discussed above for verbs like open. (83) to = BECOME + at (84) to the summit = BECOME [PP at [the summit]]] This decomposition makes available a result state 'at the summit'. It is thus incompatible with our crosslinguistic results. We can think of two reasons why such a decomposition should be impossible: One possibility is that decomposition is unavailable for prepositions, in contrast to verbs (contra Rapp & von Stechow 1999). Alternatively, decomposition could result in a type for the PP that leads to uninterpretability, due to a type mismatch. Note that our BECOME operator yields a proposition (type ) which does not combine with walk. Either way, our empirical results imply that this decomposition analysis is not available in (R)-less languages. Finally, the story on goal PPs in (R)-less languages should lead to an explanation for the ill-formedness of temporal modifiers such as 'in two hours'. We speculate that the absence of BECOME is responsible. Dowty (1979) observes that 'in two hours' not only requires that there be a unique culmination point in the extension of the predicate with which we combine 'in two hours', but moreover it must be guaranteed by the semantics that this is so (independent of the actual facts). This is true in English because

The Resultative Parameter and Restitutive Again

67

of the presence of BECOME, but not in Spanish. (Suresh only needs to end up at the village, but it does not follow from the semantics that this has to be the unique point of reaching it.)

5.

Conclusion

Our research has brought together results from work by von Stechow and Snyder, and has found confirmation of important aspects of both. Our obvious conclusion is that we have supported the parameter proposed in Snyder (1995, 2000), because we found another test which divides languages into the same two groups. Moreover, our results support Snyder's intuition that the parameter is about the formation of accomplishments in the syntax, by combining an activity with a result state. The parameter has effects on an interpretational level: Goal-PP constructions are well-formed, but they are not interpreted in the same way in languages lacking (R). This shows that the resultative parameter is not purely structural, excluding resultatives by ruling out certain structures morphosyntactically. Our proposal places the parameter in the interpretation component (availability of a principle of composition). This has consequences for both grammaticality and interpretational possibilities, which is more adequate for our case. We thus conclude that we have found a semantic parameter. We do not mean by this that there is variation as to what meanings a language can express, but that there is variation with respect to how a language can express a certain interpretation. Having or not having (R) is a parameter of compositional interpretation. We believe that we have also found support for an important feature of von Stechow's theory of restitutive again: There must be a linguistic expression corresponding to the result state of an accomplishment for the restitutive reading of again to be possible. It is not sufficient that there is conceptually a culmination point to the event described in a sentence. That is obviously the case in Spanish etc., where speakers assure us that for Suresh to have walked to the village he needs to have ended up there. What is lacking is a semantic object corresponding to that result state which again could operate on. If this reasoning is sound, it lends support to von Stechow's approach of doing decomposition in the syntax. Certainly, we have found support for another aspect of his (1995) theory: giving a different analysis to lexical accomplishments than to syntactic accomplishments. Restitutive readings are accepted much more widely with lexical accomplishments than with goal-PP constructions. An interesting question is to what extent they are possible universally. Inuttut suggests that decomposition might not be universally available, although this remains to be tested with a larger set of data. We will have to leave this question for future research. This paper raises a few other questions, besides the one about universality of restitutive readings. One concerns the semantic nature of goal PPs, which we have not discussed exhaustively by any means. We have yet to tie in our results with work on paths (e.g., Krifka 1986, 1998). Another issue is compatibility with alternative theories of

68

Sigrid Beck and William Snyder

restitutive again (in particular Fabricius-Hansen 1983 and Fabricius-Hansen, this volume) and alternative semantic theories of resultatives (for example Rothstein, to appear). Finally, there is the more general question of decomposition adverbs and their parameters (Rapp & von Stechow 1999). We hope to be able to address at least some of these issues in future work.

Acknowledgements The following people have given us invaluable help with collecting and understanding our data: Shin-Sook Kim, Luisa Marti, Deborah Chen Pichler, Katie Plavetic, Yael Sharvit, Arthur Stepanov, Koji Sugisaki and Douglas Wharram. Our (other) informants have been extremely helpful, thorough and patient; thanks to Xabier Artiagoitia and Ricardo Etxepare (Basque); Cedric Boeckx, Caroline Fery and Michele Bacholle (French); Ron Artstein, Nurit Assayag, Daphna Heller and Orr Ravitz (Hebrew); Miriam Butt, Veneeta Dayal, Tipu Fateh, Utpal Lahiri and Anees Shaik (Hindi/Urdu); Maria Mikaratsuk Dicker and Ellen Ford (Inuttut); Takako Iseda, Keiko Ishii, Nobu Miyoshi, Fumi Niinuma, Masashi Nomura and Toshiko Oda (Japanese); Raymond Lamm (Khmer); Inkie Chung, Bosook Kang, Sei-Rang Oh and Bum-Sik Park (Korean); Haldor Heimer (Lingala); Liang Chen, Pan Ning Chen and Kirsten Y. Lin (Mandarin); Eva Bar-Shalom, Evgeni Gousev, Alex Lavrentiev, Arkady Polyak and Natasha Rakhlin (Russian); Zeljko Boskovic, Nada Jevtic, Marin Pichler, Mihajlo and Sandra Stepanovic (Serbian/Croatian); Adolfo Ausin, Rene Builes, Katrin, Regine and Paco Garcia, Patricia Molina, Lara Reglero and Emma Ticio (Spanish). We are also grateful to our audience at a University of Connecticut colloquium and to two anonymous referees for comments and feedback.

References Aske, Jon (1989). Path Predicates in English and Spanish: A Closer Look. Proceedings of the Annual Meeting of the Berkeley Linguistics Society 15. Beck, Sigrid & William Snyder (to appear). Complex Predicates and Goal Phrases: Evidence for a Semantic Parameter. Proceedings of the 25th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press. Dowty, David (1979). Word Meaning and Montague Grammar. Dordrecht: Kluwer. Fabricius-Hansen, Cathrine (1983). Wieder ein wieder? Zur Semantik von wieder. In: Rainer Bäuerle, Christoph Schwarze & Arnim von Stechow (eds.) Meaning, Use and Interpretation of Language. Berlin: de Gruyter. 97-120. Fabricius-Hansen, Cathrine (this volume). "Wi(e)der" and "Again(st)".

The Resultative Parameter and Restitutive Again

69

Krifka, Manfred (1986). Nominalreferenz und Zeitkonstitution. Zur Semantik von Massentermen, Plural-Termen und Aspektklassen. PhD dissertation [published 1989, Munich: Wilhelm Finck], Krifka, Manfred (1998). The Origins of Telicity. In: Susan Rothstein (ed.) Events and Grammar. Dordrecht: Kluwer. 197-235. Lewis, David (1973). Causation. The Journal of Philosophy 70. 556-567. Rapp, Irene & Arnim von Stechow (1999). Fast 'Almost' and the Visibility Parameter for Functional Adverbs. Journal of Semantics 16. 149-204. Rothstein, Susan (to appear). Secondary Predication and Aspectual Structure. In: Cathrine Fabricius-Hansen, Ewald Lang & Claudia Maienborn (eds.) Approaching the Grammar of Adjuncts: Proceedings of the Oslo Conference, September 22-25, 1999. ZAS Papers in Linguistics 17. 287-308. Snyder, William (1995). Language Acquisition and Language Variation: The Role of Morphology. PhD dissertation, MIT. Snyder, William (2000). On the Nature of Syntactic Variation: Evidence from Complex Predicates and Complex Word-Formation. Ms, University of Connecticut. Stechow, Arnim von (1995). Lexical Decomposition in Syntax. In: Urs Egli, Peter E. Pause, Christoph Schwarze, Arnim von Stechow & Götz Wienold (eds.) Lexical Knowledge in the Organization of Language. Amsterdam/Philadelphia: John Benjamins. 81-118. Stechow, Arnim von (1996). The Different Readings of Wieder 'Again': A Structural Account. Journal of Semantics 13. 87-138. Washio, Ryuichi (1997). Resultatives, Compositionality and Language Variation. Journal of East Asian Linguistics 6. 1-49.

What Do Definites Do That Indefinites Definitely Don't?1 Daniel Biiring

1.

Introducing the Problem

This paper investigates how (in)definiteness influences word order; more specifically, how it influences the ordering of objects in the Mittelfeld of German double-object constructions. As a starting point I take what I'll call the Indefinite Puzzle.

1.1

The Indefinite Puzzle

According to Lenerz (1977), dative-accusative (henceforth dat-acc) order in German is unmarked, which means it can occur with all F(ocus)-patterns. Accusative-dative (accdat) order is marked, reflected by the fact that it can only occur with selected F-patterns. The reader is referred to Lenerz (1977) or Biiring (2000) for the full range of data, but a summary of the facts is given in the following table: F-marked: order: dative-accusative accusative-dative

Dat only

Acc only

both

neither

ok ok

ok

ok

ok

*

*

*

Of particular interest here is the optionality in word order in the first column of the table (only the dative object is F-marked). The following example illustrates this case (I use capitals to indicate pitch accents, both primary and, where relevant, secondary; where more than one accent is indicated within a sentence, the last one will be the primary, or nuclear, accent; focus patterns are elicited by lead-in questions or other leadin material):

1

I'd like to thank Summer Kern, Line Mikkelsen and an anonymous reviewer for their useful and detailed comments on a draft version of this article.

What Do Definites Do That Indefinites Definitely Don't? (1)

71

Wem hast du das Buch gegeben? 'Who did you give the book (to)?' a. Ich habe dem I have the-DAT

SCHÜler das Buch student the book

b. Ich habe das Buch dem I have the book the-ΌΑΊ Ί gave the book to the student'

gegeben. given

SCHÜler gegeben. Student given

It is important to be clear about the sense of the term "unmarked" used here: (lb) as an answer to the question given is no less acceptable to native speakers than (la). The reason Lenerz (1977) calls acc-dat order "marked" relates to the grammar of German as a whole: Dat-acc order (the "unmarked" one) can appear without any other factors such as focus/background order recommending it; acc-dat order on the other hand, is only possible where some other factor favors it. I will therefore speak of dat-acc order as the lexico-syntactically unmarked order. Paraphrasing Lenerz (1977) in this way, the optionality in (1) arises because both forms have something to recommend them: (la) displays (lexico-syntactically) unmarked dat-acc order, while (lb) displays unmarked background-focus (as opposed to focus-background) order. The reader can verify that this "tie" between lexico-syntactic unmarkedness and focus-structural unmarkedness is not found in any of the other columns in the table above. In Büring (2000) I propose to derive these facts, among others, from a general theory of prosodic phrasing, focus and word order; this account will be summarized in section 2. My aim in this paper is to address a set of additional facts only noted in passing in that earlier work, a representative datum for which is (2): (2)

Wem hast du ein Buch gegeben? 'Who did you give a book (to)?' a. Ich habe dem I have the-DAT b. *Ich habe ein I have a

Buch book

SCHÜler ein Buch student a book dem the-DAT

gegeben. given

SCHÜler gegeben. student given

Ί gave a book to the student' Notice the contrast between (2), which doesn't allow for acc-dat order, and the earlier (1), which did. The only difference between the two is that the accusative object in (2) is indefinite, where its counterpart in (1) is definite. The immediate diagnosis for this case, it would seem, is that definite NPs want to precede indefinite NPs even more than focused NPs want to follow unfocused ones. The question I will explore in this paper is: Just what is the status of this tendency in grammatical theory? Three possible explanations will be discussed:

72

Daniel Büring

• A purely morphosyntactic constraint: "Definites precede Indefinites" • A pragmatic conspiracy: Definites are often discourse-old, indefinites are often discourse-new. New material is focused, old material is not, so indefinites will follow definites because foci follow the non-foci (the background). • A semantic constraint: Indefinites have no quantificational force of their own, so they need to be in a position that is mapped onto the nuclear scope (NS) of a tripartite quantificational structure (if they are to be interpreted existentially); this mapping is in turn regulated by structural constraints which locate material that is to be mapped onto the NS towards the end of the sentence. Each of these factors and its relevance for object ordering in flexible word order languages such as German has been proposed somewhere in the literature, either without regard to the others, or as an alternative to one of them. My conclusion in this paper will be that indeed all of these factors seem to be active in German, and that none of them is sufficient to explain the data alone. Accordingly, I will provide a way of intergrating them in a unified model, using ranked violable constraints.

1.2

Morphosyntax Isn't All

Let me start by arguing against a purely morphosyntactic account, which would strive to explain the contrast between (1) and (2) by postulating a general prohibition against the order indefinite-definite; note that (2b), but not (lb) would violate this prohibition. In terms of ranked constraints, this solution would postulate the following ranking: 2 (3)

definite before indefinite » before focus

dative before accusative « »

background

I submit, however, that such a constraint cannot be the solution to the Indefinite Puzzle. Observe with Lenerz (1977) that the same asymmetry between definite and indefinite accusatives shows up if the focused dative itself is indefinite: An unfocused accusative can precede it, but only if the accusative is definite: (4)

2

Wem hast du das/ein Buch gegeben? Who did you give the/a book (to)? a. Ich I

habe einem SCHÜler das have α-DAT student a

Buch book

gegeben. given

b. Ich I

habe einem SCHÜler ein have α-DAT student α

Buch book

gegeben. given

I presuppose familiarity with OT-type analyses and the standard notational devices used therein.

What Do Deflnites Do That Indefinites Definitely Don't? c. Ich habe das Buch I have the book

einem SCHÜler gegeben. α-DAT student given

d. *Ich habe ein I have a

einem SCHÜler gegeben. α-DAT student given

Buch book

73

This example aptly provides two arguments against using a general prohibition against indefinite-definite order. First, (4a) demonstrates that indefinites can precede definites given dative-accusative order. This should be impossible, if (3) were correct. Second, (4d) shows that an unfocused indefinite accusative cannot precede a focused indefinite dative any more than it could precede a focused definite dative in (2b). Clearly, this cannot be captured by any constraint that alludes to the definite-indefinite contrast. I conclude that reference to the morphological definite-indefinite distinction - even though active in German, as we will see - will not help us to solve the Indefinite Puzzle.

1.3

The Relation Between Definiteness/Indefiniteness and Background/Focus

In this subsection I will briefly outline the connection between (in)definiteness and focusing, and then show why the Indefinite Puzzle cannot be solved by it either.3 The connection I am talking about can be illustrated by the following reasoning: (5)

a. Definites and indefinites differ with respect to the familiarity of their referents: i. Indefinites introduce new discourse referents. ii. Definites refer to old discourse referents. b. New material must be in the focus of a sentence, old material constitutes the background. c. Therefore, indefinites are always focused, while definites are always in the background. d. To the extent that focused elements follow background elements (say, for prosodic reasons), it follows that indefinites follow definites.

While this reasoning is correct over all and might in fact be held responsible for some of the ordering restrictions on indefinites, it is simply not strict enough to account for the Indefinite Puzzle. The reason is that several steps in the argument are valid for the majority of cases, but not all. Let me make caveats about three of them here:

3

I assume here the kind of theory of focus advocated in Selkirk (1984), Selkirk (1995), and Schwarzschild (1999), among others; see von Stechow (1981), von Stechow (1989), von Stechow (1991), and the references therein for a discussion of the semantics of focus.

74

Daniel Büring

Deflnites refer to old discourse referents. While true in tendency, there are numerous exceptions to this, as critics of the so-called "familiarity theory of definites" continue to point out, among them unique definites such as the sun, the number 2 or the first man on the moon and dependent definites as in John lost his keys/the keys to his house. New material must be in the focus of a sentence, old material constitutes the background. I know of no reason to doubt the first half of this claim (under a reasonably liberal construal of what can pass as non-new, at least), but the second half is clearly too strong. Consider a question-answer pair such as Who did John's mother praise? She praised JOHN., a variation on Schwarzschild's (1999:145) ex.(l 1). Here John in the answer is discourse old (i.e., the word John has just been spoken in the question) and refers to an old discourse referent (the same John that was mentioned in the question), yet it is focused. For an enlightening discussion of such cases, sometimes though I believe misleadingly - called "contrastive foci" (e.g., Rochemont 1986), see Schwarzschild (1999). Therefore, indefinites are always focused, while definites are always in the background. I have just discussed two cases in which this equation breaks down for definites, namely definites that introduce a new discourse referent (the keys to his house) and definites that are in focus despite being discourse-old. On top of that, indefinites, even if introducing a new discourse referent, can be in the background if their lexical material has been previously mentioned, as in the following English examples, where a rose is not focused (it is unaccented despite being in the sentence-final position): (6)

(After Dirk had conveyed to Amber how much he likes roses,) she Finally BOUGHT him a rose.

The indefinite roses in the lead-in doesn't introduce a discourse referent, so the use of an indefinite in the answer is possible; but it makes the phrase a rose given, so that a rose doesn't need to be focused in the answer. Thus, while it is likely to be true that indefinites, in declaratives at least, always introduce new discourse referents, it is clearly false that they always have to be in focus.4 As far as the Indefinite Puzzle is concerned, all three of the above caveats are potentially relevant here. First, the dative object dem Schüler in (1) and (2), although definite, is in focus. We interpret it either as truly discourse new (in which case it must be construed as a dependent definite or as discourse-unique in some other way); or, more likely, we accommodate a prior discourse in which its discourse referent has been introduced, in which case it is an instance of a given definite which is nonetheless focused. Second, the indefinite accusative object ein Buch, even though presumably introducing a discourse referent, is not focused, because its descriptive content is given in

4

I am less certain that indefinites in questions never introduce discourse referents, given examples like Is there an ATM around here? - It's right behind you. Note that, unlike in an example like Did you see a big black cat? - It went that way, it seems implausible to call an ATM "specific" (even in scare quotes) in this example.

75

What Do Definites Do That Indefinites Definitely Don't?

the context (here: the question). Thus the link between definiteness/indefiniteness and background/focus breaks down in these examples, yet the fact remains that the indefinite can't precede the definite. I conclude that the Indefinite Puzzle cannot be solved by recourse to the focus/background distinction alone.5

1.4

A Semantic Mapping Effect

We have seen that neither a morphosyntactic "definites precede indefinites" constraint, nor recourse to background/focus structure can explain the Indefinite Puzzle. I believe, though, that something along the lines explored in the previous section is true: While they don't need to be in focus, indefinites do need to be with a certain semantic domain, namely in a place where they can get existential closure, to borrow a term from Heim (1982). As noted there, indefinites do not have quantificational force of their own; rather, their quantificational force is determined by their environment. If semantically an indefinite gets interpreted within the restrictive clause of a quantifier, it inherits the quantificational force of that quantifier. Only if the indefinite is interpreted within the nuclear scope of a quantifier does it receive an existential interpretation. Kratzer (1995) and Diesing (1992), elaborating on this idea of Heim's, have proposed that even sentences without overt quantificational elements are mapped onto a tripartite quantificational structure involving a nuclear scope and a restrictive clause. I'd like to explore the idea that indefinite accusatives such as ein Buch in (2) cannot precede the dative because they have to be in a syntactic position which will be mapped onto the nuclear scope, rather than the restrictive clause, of a quantificational structure, and that position is following the dative. Evidence for this general line of reasoning comes from sentences such as (7): (7)

Wem würdest Du ein Buch schenken? 'Who would you give a book?' a. Ich würde ein I would a

Buch book

einem SCHÜlerp α-DAT student

b. Ich würde ein Buch dem I would a book the-ΌΚΊ 'I'd give a book to a/the student' 5

SCHÜlerp student

schenken. give schenken. give

An alternative that comes to mind is to count unstressed indefinites as "focused" in some abstract sense and then insist that acc-dat order is impossible with an unstressed (yet allegedly focused) indefinite accusative object, given that the resulting structure will always violate the "background precedes focus" constraint to at least the same degree as the dat-acc structure. Such a strategy is of course impossible to adopt in an approach like mine, which tries to reduce effects of focus to effects of stressing (unless I want to claim that indefinites are also always "stressed" in some abstract sense). The approach I present below, however, is as close as I can get to this strategy by claiming that indefinites do need to be within a certain domain, and that that domain wants to coincide with the focus.

76

Daniel Büring

The surprising thing about the two answers in (7) is that the unfocused indefinite ein Buch precedes the focused dative ein/dem Schüler. Why, then, are (7a) and (7b) possible, but (4d) and (2b) are not? The answer, I believe, is that the indefinites in (7) are interpreted not as existential, but as generic. The sentences can be paraphrased as: (8)

typically, if I had an χ which is is a book, I'd give χ to a/the student

As opposed to that, the sentences in (2) involve an existential indefinite; their paraphrase is something like: (9)

there is an χ which is a book, and I gave χ to a/the student

It seems that indefinite accusatives can precede a dative only if they are to receive a generic interpretation. If they are to be interpreted existentially, they have to follow the dative, because that is the domain which will be interpreted as part of the nuclear scope, and hence receive existential force. The reason (2b) and (4d) are unacceptable is thus a complex one: Their overall form (indicative mood, perfective tense) makes it pragmatically hard to understand them as generic statements. Their generic reading would be something as far-fetched as "for typical x, if χ was a book, I gave χ to a/the student" (people like me, who have trained themselves in getting far-fetched readings, do actually get this reading for (2b)/(4d)). And their word order, in particular the acc-dat order, makes it impossible to give the indefinite accusative ein Buch the pragmatically plausible existential interpretation. The question obviously is: Just what constitutes the syntactic counterpart to the semantic nuclear scope? In our particular case, why can an accusative following a dative get mapped onto the nuclear scope, whereas an accusative preceding a dative cannot? Different authors have given different answers to these questions: According to Kratzer (1995) and Diesing (1992), it is the VP that is mapped onto the nuclear scope; an accusative preceding a dative would have to be generic if we were to assume that acc-dat order can only arise through VP-adjunction of the accusative. Krifka (1995), on the other hand, argues that the focus is mapped onto the nuclear scope, a position slightly modified by Eckardt (1996), who postulates an abstract category Fl-focus, which does not necessarily coincide with the focus marked by accenting (Eckardt's F2-focus), to be the structural counterpart to the nuclear scope; under this view, the acc-dat order must entail that the accusative is outside of the (F1-) focus. My proposal is perhaps most similar to that in Eckardt (1996). It crucially differs from it, however, in that I assume that the relevant unit for the syntax-semantics mapping is in fact a prosodic one, the accent domain to be introduced in the next section. The gist of the proposal is that having an existential indefinite accusative in front of the focus cannot yield a structure which both constitutes an improvement over the dat-acc order in terms of focus structure, and at the same time allows for the indefinite to be mapped onto the nuclear scope. Such an acc-dat order is thus possible only if the accusative is either a generic indefinite, as in (7), or a definite NP, which doesn't care whether it is in the nuclear scope or not, as in (1). To demonstrate how this accounts

What Do Defmites Do That Indefinites Definitely Don't?

77

for the Indefinite Puzzle, I'll first have to introduce the prosody-based account of focus-related word order variation proposed in Büring (2000), which will be done in section 2. Following that, I will present and motivate a prosody-based account of existential closure in section 3. With all that in place I can then give an account of how (in)definiteness, focus and the existential/generic dichotomy influence prosodic phrasing, accenting and word order in German double object constructions in section 4, where, among other things, a solution to the Indefinite Puzzle will be proposed.

2.

Variation in Object Order - A Prosody-Based Account

The cornerstone of the analysis I outline in this section (essentially that proposed in more detail in Büring 2000) is that word order is determined by at least two kinds of constraints: Constraints that refer to the lexico-syntactic properties of a constituent, such as its case, its animacy or its (morphological) definiteness; and constraints that refer to the properties of a constituent in an utterance type, such as whether it is focus or background, and what kind of semantic interpretation it receives. Languages that give more weight to the former, such as English, are traditionally called fixed word order languages (since the word order seems invariable if viewed from the morphosyntactic point of view, even though it is of course quite flexible regarding the question of where, say, a focus can occur), while those that give more weight to the latter, such as the Slavic languages, are often called free word order languages (though their word order is rather fixed if viewed from a focus/background perspective). A second hypothesis I explore here is that focus- and interpretation-related constraints do not directly influence linear order, but only indirectly, via their effect on prosodic phrasing. Prosodic phrasing will in turn be related to word order by markedness constraints. In German, optionality between two word orders arises if the two kinds of constraints favor different linear orders; strict word order, on the other hand, is the result of agreement between them. A helpful way to think about this is that German has two operative grammars, one lexico-syntactically driven, one prosodically driven (and hence heavily influenced by such things as focus). For each grammar, there is an optimal form; we thus have a lexico-syntactic winner and a (possibly distinct) prosodic winner. Formally, I will implement this via a constraint tie.

2.1

Lexico-Syntactic Factors

In many approaches, ditransitive verbs are taken to lexically specify an "unmarked" order amongst their objects. In this paper, I follow an alternative line of analysis, according to which the unmarked order is determined by the interplay of three factors: Animacy, Case, and Definiteness. I adopt the specific analysis in Müller (1998), ac-

78

Daniel Büring

cording to which these are encoded in the form the three constraints in (10), ordered as shown in (11): (10) Lexicosyntactic Constraints: a . ANIMACY

Animate NPs precede inanimate NPs. b . DATIVE

Datives precede accusatives. c . DEFINITENESS

Defmites precede indefinites ( 1 1 ) ANIMACY »

DATIVE »

DEFINITENESS

As these factors are not my primary concern in this paper, I refer the interested reader to Müller (1998) for a discussion and justification of them. For the purpose of my presentation I will almost exclusively consider cases involving animate datives and inanimate accusatives. That way, I can simply talk about the lexico-syntactically optimal candidate (the one with dat-acc order) and ignore cases where animacy and case push in different directions, as well as whatever subordinate effects morphological definiteness might have. In the tableaux to follow I will summarize these three constraints as D-Α, reminiscent of "dative precedes accusative", but this is no more than a convenient abbreviatory convention. (12) abbreviatory convention: D - A = DEF ANIMACY »

DATIVE »

DEFINITENESS

I should mention that this system, as developed by Müller, does not predict forms to be ungrammatical, but merely degraded. That is, ceteris paribus, an inanimate dative preceding an animate accusative will be degraded compared to an animate accusative preceding an inanimate dative (since A N I M A C Y outranks DATIVE), but not unacceptable; and even an inanimate accusative preceding an animate dative will be deteriorated, but not hopeless. These graded judgements, whatever their actual implementation, can be imported into the system here, but they will be immaterial for most all of the examples considered.

2.2

Focus

Let us now turn to those constraints that regard non-lexical properties of constituents, beginning with focus. Why do foci tend to follow non-foci? I suggest that the answer involves two steps: First, German (just like English) has an unmarked prosodic structure in which more or less each phrasal constituent of a clause corresponds to an accent domain (AD), the rightmost one of which becomes the head of the intonational phrase

What Do Definites Do That Indefinites Definitely Don't?

79

(iP), and thereby most prominent among the ADs.6 The unmarked structure thus looks like (13): (13) ( Χ

Χ

X ) IP

(XP)AD(YP)AD(ZP)AD

Second, German (like many languages) adheres to a constraint that requires foci to be prosodically prominent, called FOCUSPROMINENCE or FP for short (Truckenbrodt 1995; 1999). For example, if an AD contains an F-marked constituent, it wants to achieve prominence by becoming the head of the iP. Thus in (13), ZP could legitimately contain a focus, since it is the head of iP and thereby most prominent. Alternatively, YP and ZP could both contain foci, in which case YP violates FP (it is not the head of iP), but not fatally, since any alternative phrasing (e.g., one in which YP, rather than XP, is the head of iP) would violate FP as well (e.g., because XP isn't the head of iP) plus possibly additional constraints.7 What cannot happen is that YP, but not ZP, is a focus, because this involves an avoidable violation of FP. To understand this we have to look at the constraints in more detail, though: Accent Domain Formation To regulate accent domain formation we import the following two constraints, more or less directly from Truckenbrodt (1995): (14) A(CCENT)D(OMAIN)F(ORMATION) a. STRESSARG(UMENT)

A thematic argument bears phrase-level stress. b. WRAP XP

An XP is contained within one AD. STRESSARG prevents two non-overlapping NPs (or any two non-overlapping constituents) from being mapped into one big AD, because that way only one of them would receive phrase-level stress (qua being the head of that AD). To see this we have to pay attention to secondary accents, as in the following example, where meiner Mutter 'my 6

7

The prosodic correlate of an accent domain is the presence of phrase level stress (marking its head), which in turn is a necessary condition for association with a pitch accent. Quite presumably, ADs are identical to the phonological phrases of Selkirk (1984) or Truckenbrodt (1999) and show other prosodic correlates such as boundary tones, final lengthening, breaks etc. More investigation of this issue is needed, though. The prosodic correlate of the intonational phrase is a final boundary tone and, most prominently, an obligatory pitch accent on its head, the nuclear accent. The reader might wonder if YP and ZP cannot simply form one AD which then becomes the head of iP, thereby avoiding any violation of FP, as in (XP)AD (YP ZP) AD , where boldface indicates maximal prominence. The answer here is that this phrasing violates the formal version of FP at the AD level, because the prosodic word corresponding to YP fails to become the head of AD, despite containing a focus. And it's turtles all the way down: If a phrase contains two or more F-marks with neither dominating the other, a FP violation will inevitably occur at some level.

80

Daniel Büring

mother', must bear a (secondary) pitch accent, which in turn is indicative of phraselevel stress, i.e., the presence of an AD:8 (15) Warum why

warst were

du auf der you at the

Post? post-office

a. Ich will meiner MUtter eine POSTkarte I want my mother a postcard

schicken. send

b. #Ich will meiner Mutter eine POSTkarte schicken. The following tableau derives this:9 i:

XP YP

a.

(XP)AD(YP)AD

b.

(XP

STRESSARG

*!

YPHO

W R A P X P , on the other hand, prevents XPs from being unduly fractured. For example, a verbal predicate will be in the same AD as an adjacent argument, given that separating them would violate W R A P X P for X P = V P . This is the effect Jacobs (1992) calls integration, cf. also von Stechow and Uhmann (1986):

(16) Gudrun Gudrun

möchte Astronautin werden, wants astronaut become

und Peter and Peter

will... wants

a. ...DAMPFschiffe bauen. steam-boats build b. #...DAMPFschiffe BAUen. c. #...Dampfschiffe BAUen. 'Gudrun wants to become an astronout, and Peter wants to build steam boats' i: a. b. c.

8

9

NP otec ,V (NP v) AD (NPWV14D (np V)AD

STRESSARG

WRAPXP *!

*!

My account predicts that XPs preceding the head of iP will almost always form an AD, due to STRESSARG; (15) illustrates this with a case in which my mother is also focused. It is less obvious that unfocused XPs necessarily form an AD; they certainly do not need to bear a pitch accent, even though they can. As noted above, the concept of an AD thus doesn't have a necessary prosodic correlate, but only entails the possibility of a pitch accent. Here and throughout, the a. and b. numbers in the tableaux refer to the example sentence immediately preceding the tableau. Where there is more than one candidate structure for a single example sentence, I'll use a., a.', a.", etc.

81

What Do Definites Do That Indefinites Definitely Don't?

Notice, too, that in an argument-head complex, the argument, rather than the head, will be the head of the AD, hence prominent (indicated by (NP v), as in (a), as opposed to (np V), as in (c), in the tableaux). This follows from STRESSARG as well, given that NP but not V is an argument. The overall effect of these constraints is that predicates form an AD with their adjacent argument, while all other constituents form their own AD (see, once again, Truckenbrodt 1995). Nuclear Stress and Accent: As mentioned above, ADs aren't the highest level of prosodic structure. Simple sentences like the ones I am concerned with in this paper are mapped onto an intonational phrase (iP). The only relevant constraint in connection with the present investigation is that the head of iP is right-peripheral in German (again, just as in English). This accounts for the fact that in a sentences with more than one pitch accent, such as (15) above, the final one is most prominent, cf. (18) (where χ marks AD-level stress and X marks iP-level stress): ( 1 7 ) IPHEADRIGHT ( I P H R )

An iP and the AD that is its head are right-aligned. (18) a. ( χ X )iP (meiner MUtter )AD (eine POSTkarte schreiben )AD b. *( Χ χ )iP (meiner MUtter )AD (eine POSTkarte schreiben )AD In the following tableaux I use boldface for the constituent that is the head of iP (I continue to use upper case letters for the head of an AD): i: a. b.

NP NP V ( N P W NP v)AD (NPWNPvW

IPHR

STRESSARG

WRAPXP

Focus Prominence: The final constraint I borrow from Truckenbrodt ( 1 9 9 5 ) is FOCUSPROMINENCE: ( 1 9 ) FOCUSPROMINENCE (FP)

Focus is most prominent.10 Space does not allow me to discuss the effects of FP with a wider variety of F-patterns, cf. Biiring (2000), but let me briefly illustrate three basic cases: Focus on a rightmost XP, focus on two XPs, and focus on a penultimate XP. 10

As already mentioned, the most prominent prosodic constituent within a larger constituent is defined as the head of that constituent. The formal version of FP is: If Α is a prosodic constituent at level η which contains a syntactic node that is F-marked, α is the head of the prosodic category at level n+1 that contains a .

82

Daniel Büring

If the rightmost phrase in a clause is F-marked, the standard pattern (13) will arise, in compliance with all constraints discussed: i: a. -> b. c.

XP Y?h (XPWYPW) (xp YP)An (XPyp ) An

IPHR

FP

STRESSARG

WrapXp

*!

*1

*

The same prosodic structure emerges if two XPs are focused. One of them will inevitably violate FP, but this will be unavoidable: In (a) below, (the AD containing) XP fails to become the head of iP, in (b) (the prosodic word containing) it fails to become the head of AD, and in (c) the same happens to YP. Which structure is chosen thus depends on STRESSARG alone: i: a. b. c.

XP/rYP/r (XPWYPW (xp ΥΡΧ,ο (XPyp )Al>

IPHR

FP

STRESSARG

WrapXp

*

* *

*l *t

The final and by far most interesting case arises if a non-final YP is focused. What we observe is that the non-final YP receives the nuclear accent, and that no secondary accents can be on the XP(s) following YP: (20) Wem who

hast du eine Postkarte geschrieben? - Ich have you a postcard written I

habe... have

a. ... meiner MUtter eine Postkarte geschrieben. my

mother a

postcard

written

b. #... meiner MUtter eine POSTkarte geschrieben. c. #... meiner Mutter eine POSTkarte geschrieben. I interpret this to imply that we have one AD that spans from the left edge of the focus to the end of the sentence/iP. This, obviously, is an imperfect prosodic structure. Why is it chosen? Note that in a case like this, STRESSARG, FP and IPHR are in conflict. If perfect ADs are formed, IPHR and FP cannot simultaneously be met. Either the final AD becomes the head of iP and thereby most prominent, as in (b), which violates FP (given that the focus sits in the penultimate AD); or the non-final AD (the one containing the focus) becomes the head of iP, meeting FP but violating IPHR, as in (b1). Alternatively, ADF could be sacrificed, as in (a), with the benefit of reconciling IpHR and FP (because now the focus is within the rightmost AD). Evidently, this is what

83

What Do Definites Do That Indefinites Definitely Don't? happens in German. The conclusion is that and FP:11 i: XP/. Y P a. -> ( X P YP^O b. b.'

c.

IPHR

(XPWYPI,D

STRESSARG

FP

must be outranked by

STRESSARG *

YPW)

WRAPXP

•

;

*T

( X P W Y P U

(xp

;

IPHR

I

*

:

This "destructuring" effect of non-final foci has been observed for various languages, among them German in Uhmann (1991:237ff.), where a similar rationale for it is offered, and Japanese (see, among others, Nagahara (1994), Uechi (1998), and the references therein). The essentials of the analysis adopted here are due to Truckenbrodt (1995: chapter 5), where it is set in the context of a broader typology of focus-alignment effects. The fact that non-final foci lead to marked prosodic structures is central to the explanation of focus-related word order variation in Büring (2000) to be presented in the next subsection; it has also been adopted for the analysis of focus-related word order variation in Spanish in Gutierrez-Bravo (1999). In the remainder of this paper I will summarize the constraints W R A P X P and STRESSARG as ADF (reminiscent of accent domain formation) in the tableaux, which is violated whenever an AD contains less than an XP, or more than an XP plus its predicate. (21) abbreviatory convention: ADF = jef STRESSARG, W R A P X P

2.3

Focus Related Word Order Variation

We just saw that focus on a non-rightmost XP leads to a prosodic structure with an "extra-large" AD, namely one that extends from the beginning of the focused XP to the end of iP (here: the sentence). Only in this way can the focus be maximally prominent and be in the rightmost AD in iP. Notably, this extra-large AD could be avoided while respecting all other constraints if XP and YP could be freely reordered (the use of { ... } in the input specification of the following tableaux means that the input is unspecified for word order):

I assume for the sake of this exposition that WRAPXP is subordinated, too, though this hasn't been demonstrated.

Daniel Büring

84 FP

IPHR

i:

{XPFYP}

a.

(XP YP)^

b.

(ΧΡ)ΛΟ (ΥΡ)Λ£>

b.'

(XPWYPX4D

d. ^

(YPWXP)

ADF *!

*! *!

This idea provides the basis of an account of focus related word order variation: If an XP is unfocused, having it in front of a focused one will provide for a perfect prosodic structure which is nonetheless in keeping with FP. This isn't quite German, however, because while German allows non-canonical word order in such cases, it doesn't require it. To implement this, we go back to the lexico-syntactic constraints discussed in subsection 2.1 above and summarized there as D-Α ("dative before accusative"). These constraints will be used to counter-balance the effects of the prosodic constraints, assuming a ranking as in (22), where prosodic constraints and lexico-syntactic constraints are tied:12 (22) IPHR, FP » ADF « » D-A The effect of this tie is that both the prosodically optimal candidate and the lexicosyntactic candidate are grammatical, in other words: optionality (where the two are different, that is). To demonstrate the effects of this, let us go back to our initial example (1), repeated here: (23) Wem hast du das Buch gegeben? 'Who did you give the book (to)?' a. Ich habe dem SCHÜler das Buch I have the-DAT student the book b. Ich habe das Buch I have the book

gegeben. given

dem SCHÜler gegeben. the-DAT student given

Ί gave the book to the student' i:

{dat/racc}

IPHR

a.

( D A T acc^O

a.'

(DAT^D

a." b.

12

(Α(Χ)ΛΟ (DAT)^O

FP

ADF

D-A

*

:

(ACC)AD

(DATHN(ACCHN

:

*!

*I

:

1

*

The alert reader will have noticed that what is tied here are not two constraints but rather two sub-hierarchies of constraints, which, it must be admitted, constitutes a significant deviation from what is normally considered an ordering or ranking.

What Do Definites Do That Indefinites Definitely Don't?

85

As said earlier, we can think of such a tie as encoding two different grammars. Thus the tableau above abbreviates the two tableaux below, which present the resolution of the tie into the "prosodic grammar" and the "lexico-syntactic grammar", respectively: i: a. a.'

{dat/r acc} (DAT acc^o (DATWACCW

a."

(DATWACCW

FP

IPHR

D-A

ADF

*L *

b.(ACCWDATlu, i: a. a.

1

{dat/r acc} (DAT acc)^0

FP

IPHR

D-A

ADF *

*!

(DATWACC)^

a." ( D A T W A C C ) , o b.-> ( A C C W D A T W

*!

We thus have implemented the optional non-canonical word order. And we derive that optionality emerges only if the lexico-syntactically unmarked order yields an imperfect prosodic structure, compare (l)/(23) to (24), which corresponds to the second column in the very first table on page 1: (24) Was what

hast du dem Schüler have you the student

a. Ich habe dem Schüler I have the student

gegeben? given

das BUCH gegeben. the book given

b. #Ich habe das BUCH dem Schüler gegeben. Here, since the lexico-syntactically unmarked dat-acc order also allows for a perfect prosodic phrasing, no word order variation is possible: I: a. a.' b. b.

1

{dat acc/.} (DATWj (ACC)„„ (dat \CC)AD (ACC datW (ACC),ο (DAT),,,

IPHR

FP

ADF

D-A

*!

*ι *L

*i *

Our next task is to show why acc-datp order is impossible if the accusative is indefinite. As shown above, this cannot just be due to a lexico-syntactic constraint that wants indefinites to follow definites. I argued that in order to understand what kind of constraint is operative here, we have to realize that the incrimated order is possible, but only if the indefinite is generic. It is thus an effect of being non-generic, rather than

86

Daniel Büring

being indefinite per se that we observe here. Let us therefore examine the generic/existential contrast more closely.

3.

Generic and Existential Indefinites

As announced at the end of section 1, I want to explore the idea that the domain of existential closure, the nuclear scope, corresponds to a prosodic domain. As we saw in the previous section, a sentence consists of a linear sequence of accent domains (ADi)(AD2)...(ADn). With respect to the mapping onto the tripartite structure, I propose that, going from left to right, the mapping onto the nuclear scope can start at any accent domain ADm, and then continues until the end of the sentence. As an additional restriction, the nuclear scope must be left-aligned with an AD which contains a focus. Consider the abstract structure in (25), which represents a sentence with four accent domains, the last two of which contain foci. I will use : 3 to mark where the mapping onto the nuclear scope starts (=the domain of existential closure). According to the above idea, (25a) and (b) represent well-formed mappings, but (25c) and (d) do not, because in the latter, accent domains which do not contain a focus are mapped onto the nuclear scope: (25) (AD1)(AD2)(AD3F)(AD4F) a.

(AD1)(AD2)(AD3F):3(AD4F)

b. ( A D 1 ) ( A D 2 ) : 3 ( A D 3 F ) ( A D 4 F ) c. * :3(A D L ) ( A D 2 ) ( A D 3 F ) ( A D 4 F ) d. * ( A D L ) : 3 ( A D 2 ) ( A D 3 F ) ( A D 4 F )

I will implement this as in (26): ( 2 6 ) ALIGN NUCLEAR SCOPE ( A N S )

The nuclear scope consists of complete accent domains, all of which contain focus. On a speculative note, it is perhaps justified to think of (26) as an iconicity constraint, whose objective it is to mark a domain of content, the nuclear scope, by aligning it with a domain of form, ADs containing pitch accents. Be that as it may, indefinites are specified as either existential or generic in the input, a specification which cannot be overridden, due to an undominated constraint F A I T H ( 3 / G ) : ( 2 7 ) FAITH(3/G) ( F 3 G )

An indefinite specified as existential (generic) in the input is interpreted existentially (generically).

What Do Definites Do That Indefinites Definitely Don't?

87

In the representations that follow I will use :3 as above in the candidates, and subscript indefinites with 3 or G in the candidates and the input. Viewed this way, (26) is an alignment constraint (because it seeks to align the nuclear scope with a focus), and (27) is an injjut/output faithfulness constraint (because it prohibits change of 3 to G and vice versa). (26), in tandem with (27), will have two distinct effects: Existential indefinites will need to be in an AD that contains a focus (so that :3 can precede that AD), while generic indefinites need to be in an AD that precedes at least one AD containing a focus (so that :3 can follow the indefinite while still preceding a focus-containing AD). To motivate this, it is instructive to study the generic/existential contrast under circumstances where it does not interact with the other object-ordering constraints. Consider (28):14 (28) Wenn if

man in die one into the

USA US

a. ...VORstrafen previous-convictions

einreisen will, enter wants

muss must

man one

ANgeben. list

b. ...VORstrafen angeben. 'If you want to enter the US, you have to list previous convictions' These two sentences differ only in their prosodic shape; in (28a) we find two ADs, as witnessed by two accents (the latter of which is the most prominent one, due to I P H R ) . It expresses the generic reading of this sentence, which happens to be true: If you travel to the US and you have previous convictions, you have to list them. In (28b) we find Eventually it might be advantageous, however, to view Ξ / G and :3 as notational shorthands for aspects of the interpretation, not parts of the syntactic or prosodic representation. On this view, a candidate will consist of an interpretation alongside with prosodic and perhaps syntactic structure(s), and (26) as well as (27) are correspondence rules that hold between the different representations that make up a candidate (cf. Jackendoff 1997). Inspired by this perspective I refrain from giving a constraint that prohibits existential NPs preceding the :3-boundary; there can be no existential indefinites outside of the nuclear scope, because being existential and being in the nuclear scope are one and the same thing. All there can be is an indefinite that was specified as existential in the input, but winds up generic in the output. Note that most of the examples I present in this subsection involve focused generic indefinites. Notice the temptation to reduce the influence of the generic/existential contrast to focus or familiarity along the following lines: Generic indefinites are prime candidates for staying unfocused, because they can be repeated in a discourse, in order to refer to the genus or kind they name, again and again, whereas an existential indefinite cannot be repeated in order to refer to the same individual again (that's where you use a definite instead). Repeated things (generic indefinites or definites) are unfocused, so we derive that generic indefinites pattern with definites. The cases of focused generics wam us not to give in to that temptation: Focused generics behave differently from focused existentials, just as unfocused generics behave differently from unfocused existentials, as 1 will show later on.

88

Daniel Büring

integration, i.e. object and verb form one AD whose head is the object, in accordance with the principles discussed above. It expresses an existential reading, "if you want to enter the US, there must be previous convictions for you to list", which is of course false. Let us start by deriving (28b), which is run-of-the-mill. The prosodic constraints in ADF favor the integrated structure. F3G and ANS aren't involved here, since the sentence doesn't contain a generic NP, and since the indefinite is itself part of the focus (which is VP or some higher constituent) (I will henceforth leave out the AD subscript in the candidates for perspicuity; note that all parentheses in the candidates represent ADs): i:

AccV/T

IPHR

FP

a.

:3(ACC3)(V)

*

b.

:3(ACC3V)

*

F3G

ANS

ADF

D-A

*!

What about (28a)? This structure will be the optimal realization for a generic NP, provided that we rank F3G and ANS higher than ADF: i:

Acc/. (; V/r

a.

(ACC g ): 3 (V)

a.' b. b.'

IPHR

:3(ACC3)(V)

FP

ANS

ADF *

*

*

*

: 3 ( A C C 3 V)

F3G

*

D-A

*!

*

( A C C G : 3 V)

We have thus captured the connection between semantics and prosody in (28). A generic NP will force separation into two ADs, against prosodic constraints (an observation made e.g. by Krifka 1999: section 1.4.7). An existential NP will integrate (as NPs generally prefer to do). Each of the resulting prosodic patterns is the optimal candidate for their respective NP-type. There is a one-to-one correlation between prosodic structure and reading. This correlation, however, is not always observed, because crucially, ANS itself is not inviolable. Notice that formation of an object-only AD necessitates formation of a verb-only-AD. Since iP is right-headed, the verbal AD will bear the main prominence. This is fine in (28a), since the V is focus. It violates FP since the accusative, too, is focus but not most prominent within iP; but one violation of FP is unavoidable and thus not fatal. The alternative structure (28b) violates FP as well (this time on behalf of V), but looses out on ANS. This picture will change if V is not focused: (29) Stimmt es, is-correct it - D a s nicht, - that not

dass that aber but

man one man one

sämtliche Knöllchen all parking-tickets muss must

a. #...VORstrafen ANgeben.

angeben list

muss? must

What Do Definites Do That Indefinites Definitely Don't?

89

b. ...VORstrafen angeben. 'Do you really have to list all your parking tickets? - Not quite, but you have to list previous convictions' The former winning candidate, (29a), is unacceptable here. This is because it violates FP in making the non-focused verb angeben, rather than the focused NP Vorstrafen prominent in iP. We predict the correct form (29b) if we rank ANS lower than F3G: i:

AccF,G V

a.

(ACCg):3(V)

b.

( A C C G : 3 V)

IPHR

FP

F3G

ANS

ADF

D-A

*

*! *

Note incidentally that (29b) is optimal for the same F-pattern with an existential indefinite object, too - as in (28b) (the first tableau). The generic-existential contrast is thus prosodically neutralized in these narrow focus cases. It is interesting to note that the generic indefinites in (28a) and (29b) bear stress, in the latter case even main stress. Data like these have been noted in Biiring (1996:4, ex. (6)), and Eckardt (1996:60, ex. (31), attributed to I. Kohlhof, p.c.), where it is also noticed that they pose serious problems for focus-based accounts of the generic/existential contrast such as Krifka (1995) and Eckardt (1996). It should also be noted that the present analysis is not committed to any phrase-structural difference between the sentences with different types of indefinites, as proposed in de Hoop (1992) and Diesing (1992). A thorough comparison to these theories is beyond the scope of the present investigation, however. It is quite conceivable that a similar constraint-pattern holds for English. Since English is VO, the difference between (V)AD(0)AD and (V 0)AD does not result in a shift of the nuclear accent, as it does in the German cases in (28), and is thus less easily detectable. It has been observed, though, that sHfy'eci-integration, i.e., forming a single AD out of a subject and an intransitive verb, interacts with genericity. Consider the following contrast from Halliday (1967), reported in Rooth (1996:273): (30) a. SHOES must be worn, b. DOGS must be CARried. Rooth comments: If you bring along no dog at all, you obey the second regulation, but if you bring no shoes at all, you violate the first. If you carry one dog and bring another on a leash, you violate the second regulation; but if you wear one pair of shoes and carry another pair in a shopping bag, you obey the first. (Rooth 1996:2)

It should be easy to see that the English subject-verb pattern is entirely parallel to the object-verb patterns observed in the German examples in (28):

90

Daniel Büring (ARGUMENT

predicate)^

(ARGUMENT^oCPREDICATE^o

satisfies A D F , is compatible with A N S if argument is existential, but violates it if argument is generic violates ADF, but satisfies ANS if argument is generic

In fact, the same relative ranking of ANS and ADF would account for these English facts, too, even though there doubtlessly are more complications. That this parallelism might not be coincidental is also suggested by the fact that the same "neutralization" observed in (29) above occurs in English: (31) Hey, you've got to carry your cat here. That's what the regulations say! - No, dude,... a. DOGS must be carried (, CATS can go on a LEASH). b. #DOGS must be CARried The same reasoning applies here: The generic indefinite dogs in (31) wants to form its own AD, on behalf of ANS, as the one in (30b). But then the rest of the sentence must form an AD, too, which would be the rightmost one and therefore receive main prominence; and that violates the higher constraint FP. The result is unacceptable as seen in (31b). Therefore, the sentence will be squeezed into one big AD as in (31a). Summing up, we have seen that generic indefinites, unlike existential ones, like to form an AD of their own. I have proposed to capture that by a constraint that regards the mapping between prosodic structure and interpretation which governs the "cut-off point" for the domain of existential closure, the nuclear scope. This constraint will affect not only generic indefinites (by forcing them to precede that point), but also existential indefinites (by forcing them to follow it). With this constraint, we finally have all the pieces in place to return to the placement of indefinites in double object constructions, and the Indefinite Puzzle in particular.

3.

Indefinites In Double Object Constructions

3.1

Existential Indefinites in the Background: Solving the Indefinite Puzzle

We are now in a position to solve the Indefinite Puzzle: Why can't an unfocused indefinite precede a focused dative, where unfocused definite accusatives can?

91

What Do Definites Do That Indefinites Definitely Don't? (32) Wem who

hast du das / ein have you the a

Buch book

gegeben? given

a. Ich habe das Buch I have the book

dem SCHÜler gegeben the student given

b. *Ich habe ein I have a

dem SCHÜler gegeben the student given

Buch book

Recall that an unfocused accusative precedes the dative to improve prosodic structure: (DAT/.· a c c V)AD ( A C C ) A D ( D A T / . V)AD

(lexico-syntactically u n m a r k e d )

(prosodically unmarked)

To understand the peculiar behavior of indefinite accusatives, or existential indefinite accusatives, to be precise, note that the acc-dat-structure provides no basis for properly inserting the :3-boundary in that case. Inserting it in front of the accusative would violate ANS (because the AD following it doesn't contain a focus), inserting it after the accusative would leave the indefinite accusative without existential force, violating F3G. The following tableau illustrates this: i: (2)

a.

( 3 2 ) b. b.'

b."

{Acc3i Dat/r} acc3 v)

IPHR ;

FP

F3G

ANS

: 3 ( A C C 3 ) ( D A T V) ( A C C G ) : 3 ( D A T V)

:3(acc3 DAT v)

ADF

D-A

*

:3(DAT

*

*!

*

*! *!

*!

We have thus solved the Indefinite Puzzle: Using acc-dat order to improve prosody makes it impossible to get an existential reading for the accusative indefinite. And the one candidate which displays acc-dat order and doesn't violate any of the indefiniterelated constraints, (b."), is neither lexico-syntactically nor prosodically unmarked; it violates both ADF and D-A. Let me review the logic of this account once more. It does not say that an existential indefinite generally has to follow the focus; nor does it say that an existential indefinite accusative cannot occur with acc-dat order. It merely says that an existential indefinite has to form an AD with a focus, and that the order of the indefinite and the focused argument within that AD will be determined by the lexico-syntactic constraints ANIMACY, DATIVE and DEFINITENESS - alone, which, in the example above, all favor the outcome dat-acc. The remainder of this subsection is devoted to demonstrating that this is indeed the correct generalization. First, an existential indefinite can precede the focus, if that is what the lexico-syntactic constraints favor. Consider (33), in which we have an unfocused existential dative. DEFINITENESS doesn't apply, and both DATIVE and ANIMACY favor dat-acc order,

92

Daniel Büring

even though that implies that the unfocused indefinite precedes the focus. And indeed the opposite order in (33b) sounds rather awkward: (33) Obwohl der Verkauf von Schusswaffen although the selling of guns verboten ist, haben Sie am 28.11. prohibited is have you on 11/28

an to

Mindeijährige minors

a. ...einer Mindeijährigen eine GASpistole verkauft. a minor a gas-gun sold b. #...eine GASpistole einer a gas-gun a

Mindeijährigen verkauft. minor sold

Let us next see what happens if two lexico-syntactic constraints conflict. In (34), isn't relevant, and DATIVE and DEFINITENESS pull in opposite directions. It seems to me that the acc-dat order in (34b) is much better than in the previous example:

ANIMACY

(34) (Rainer saw a girl at a party we went to, who he wants to see again. He expects me to know her name, because he saw me introduce her to an Italian looking guy, so he asks:) Wen hast du einem Italiener who have you an-Dat Italian a. Ich habe einem Italiener I have an-Dat Italian

vorgestellt? introduced

MARION vorgestellt. Marion introduced

b. Ich habe MARION einem Italiener I have Marion an-Dat Italian

vorgestellt. introduced

Is this expected under the present account? If DATIVE strictly outranked DEFINITENESS, only (34a) should be grammatical. Recall from subsection 2.1, though, that in the original conception in Müller (1998), the lexico-syntactic constraints derive degradation rather than ungrammaticality. Under that assumption, (34b) would be degraded, but much better than any of the examples to which I gave a # above. The candidate corresponding to that sentence is marked by in the tableau below: i: Acc/rDat3 V a. : 3 (dat 3 ACCv) a.' (date ):3(ACC V ) a." :3(dat3)(ACC ν ) b. iJ : 3 (ACC f dat3 ν ) b.' :3(acc)(DAT3 ν )

IPHR

FP

F3G

ANS

ADF

D-A

*

DEF

*!

DEF *!

DEF *

*!

*

DAT DAT

93

What Do Definites Do That Indefinites Definitely Don't?

Note then that (34b) constitutes a (rather rare and curious) case in which a focused accusative can precede an unfocused dative. As just discussed, the present system predicts this, given that accent domain formation isn't relevant in these cases, because the non-focused NP is an existential indefinite. 15 It likewise predicts the sharp contrast to the cases of accp dat in which the dative is definite and could thus form its own AD: (35) Peter Peter Dabei albeit

weigert refuses habe ich have I

sich, mir SELF me-DAT

a. ...dem Blödmann the-DAT jerk b. #...MArion Marion

seine his

MArion Marion

dem Blödmann the-DAT jerk

Schwester sister

vorzustellen! to-introduce

vorgestellt.16 introduced vorgestellt. introduced

To sum up, an unfocused existential indefinite has to form an AD with the focus. This violates ADF, but it is necessary to meet the higher constraints F3G and ANS. Therefore, prosodic considerations will not play a role in choosing between acc-dat and datacc order; only lexico-syntactic-constraints will. In standard cases as Lenerz' (2), the lexico-syntactic-constraints will unanimously favor dat-acc, accounting for the Indefinite Puzzle. In other cases, like (34), the lexico-syntactic constraints conflict with each other and a certain degree of word order freedom is predicted.

1.2

Generic Indefinites

In this subsection and the next I will look at those cases which are not part of the Indefinite Puzzle, but for which the system developed so far makes predictions. Let us start by checking how the constraints formulated so far account for generic indefinites in double object constructions. 4.2.1

Generic Accusatives that Obligatorily Precede Datives

A generic indefinite, unlike the existential one in (2), can precede a focused co-argument and thus allow for a perfect prosodic structure, cf. (36a). In fact, this acc-dat order is obligatory here, as (36b) shows:

They present serious challenges for both phrase-structure and focus/background-structure based accounts of the generic/existential distinction, though, as well as to accounts which directly relate word order variation to focusing. To see that dem Blödmann, 'the jerk', is not focused in this context, consider a sentence in which its accent would be nuclear, such as a continuation like ...und trotzdem mag ich den Blödmann (' ...and still I like the jerk'); the nuclear accent has to sit on mag, absolutely not on Blödmann.

94

Daniel Büring

(36) Bisher haben wir Ladendiebe nicht so far have we shopliftings not der neuen Regelung... the new regulation

gemeldet, aber nach reported but according-to

a. ...müssen wir Ladendiebe dem GeSCHÄFTSfuhrer melden. must we shoplifters the manager report b. #...müssen must

wir dem GeSCHÄFTSfuhrer Ladendiebe melden. we the manager shoplifters report

'So far we haven't reported shoplifters but according to the new regulations, we have to report shoplifters to the manager' This behavior is predicted, given that for a generic interpretation to obtain, the indefinite must be in the restrictive clause, i.e., preceding the :3-boundary. The : 3 in turn must precede an AD containing the focus, which means that the indefinite must precede a focus: i: a.

ACCG Dat/. V ->

IPHR

FP

FBG

ANS

ADF

*!

*

*

*

D-A *

(ACCG):3 ( D A T Ν )

b.

(DAT acco : 3 ν )

b.'

(DAT)(ACCG):3(V)

*!

The exact same pattern obtains if the generic indefinite is itself focused, too: (37) Bisher so-far werden will

haben wir Ladendiebstähle have we shopliftings wir... we...

a. ...WiederHOlungstäter der PoliZEI repeat-offenders the police b. #...der PoliZEI the police

nicht not

gemeldet, aber ab sofort reported but as-ofnow

melden. report

WiederHOlungstäter melden. repeat-offenders report

'So far we haven't reported shopliftings, but as of now we will report repeat offenders to the police' This is again as expected, given that the indefinite needs to get the : 3 between itself and the end of the sentence (recall from the discussion of cases like (25a) above that : 3 doesn't need to precede every AD containing a focus, but merely that it precedes only such ADs). This example also shows that optimization of prosodic structure in (36) was a welcome side-effect, but not the driving force behind the obligatoriness of accdat order, for there is a structure for the unacceptable (37b), namely (b.'), whose prosodic structure is perfect, too. But it is ungrammatical, due to the violation of F3G:

What Do Definites Do That Indefinites Definitely Don't? I:

a. b. b.' b." 4.2.2

ACCG.F Dat/r V

IPHR

(ACCG):3 (DAT v) (DAT)(ACC 0 : 3 v) (DAT):3 (ACC 3 v) (DATXACCc) : 3 (V)

FP

:

F3G

95 ADF D-A

ANS

*

J

*! I

*! *

*

Optional Acc-Dat Order with Generic Indefinites

In the examples discussed in the previous subsection, there were two constraints that favored acc-dat order; ADF, since the dative was narrow focus and thus should be sentence-final (modulo the verb, that is), and ANS/ FBG, since the generic accusative must have a prosodic phrase boundary following it (for : 3 to align with), which means it has to precede the focus. Note that the former is a weak motivation, since ADF can be overruled by the lexico-syntactic constraints on one resolution of the tie. It is the latter motivation that is responsible for the obligatoriness of acc-dat order, because ANS and F3G dominate the lexico-syntactic constraints. The prediction is thus that if we manage to "switch o f f ' ANS and F3G, we would observe optionality between the two orders again. This predication is borne out: If the generic has a chance to form an AD on its own while following its co-argument, both word orders are possible: (38) Damit eine Seite so-that a page

wie diese funktioniert, ist like this- one functions is

es it

wichtig... important

a. ...dass man verALtete LINKS dem WEBmaster meldet. that one outdated links the web-master reports b. ...dass man dem WEBmaster verALtete LINKS MELdet. that one the web-master outdated links reports 'For a page like this to work it is important that you report outdated links to the web-master' The present account predicts this, given that the generic accusative in (38b) can have the : 3 following it and preceding a focus (namely the verb), unlike in (37b), where the verb is unfocused (see the structure in the tableau below). While this structure violates ADF (the verb doesn't integrate with its adjacent argument), it is optimal under lexicosyntactic considerations. Acc-dat order, on the other hand, allows for optimal prosody at the expense of lexico-syntactic markedness, so candidate (a) is grammatical, too, just as in (36) and (37) above:

96 i:

Daniel Büring IPHR

AccG,F Dat/7 \LF

FP

(ACC G ): 3 (DAT ν ) (ACC G ): 3 (DAT)(V) (DAT)(ACC G ): 3 ( V )

**

b.'

( D A T ) ( A C C G :3 Ν )

**

b."

( D A T ) :3 ( A C C 3 V )

**

a. a.1 b.

F3G

ANS

ADF

D-A *

** **

*

*! *!

In the system developed here, the word order freedom in (38) arises from the same tie known from double object constructions that do not involve generic NPs, namely that between a perfect prosodic structure (generic)(nongeneric verb) and a lexico-syntactically unmarked structure (dative)(accusative)(verb). The prediction, then, is that like in the simple cases, the optionality in word order should disappear if prosody and lexicosyntax favor the same outcome. Again, this seems to be a correct prediction: (39) Damit eine Seite so that a page

wie diese like this-one

a. ...dass man ERSTbenutzern that one novices-DAT b. #...dass man die that one the

funktioniert, ist functions is die the

NUTzungsrechte terms-of-use

es it

wichtig... important

NUTzungsrechte terms-of-use

erklärt. explains

ERSTbenutzern novices-DAT

erKLÄRT. explains

'For a page like this to work it is important that you explain the terms of use to first time users' This result follows in the same manner: The optimal structure (39a) meets both ADF and DAT, while the alternative order yields (39b), which violates both: i:

Accf Datf o Vf

IPHR

FP

a. -> ( D A T G ) : 3 ( A C C v )

**

b.

( A C C ) ( D A T g ) :3 (V)

**

b.'

( A C C ) ( D A T G :3 V )

**

1.3

Existential Indefinites in Focus

F3G

ANS

ADF *!

*!

D-A *! *

I close this section with a look at existential indefinites in focus. This is a rather boring endeavor, because these behave just like definites in focus: The reason is that a focused indefinite can always form an AD without running the risk of violating FP, and : 3 can then precede that AD in keeping with ANS, and thus guarantee an existential reading. Since a focused accusative wouldn't precede a dative for prosodic reasons, we expect to see focused existential indefinites wherever the lexico-syntactic constraints prefer

What Do Definites Do That Indefinites Definitely Don't?

97

them to be. The examples below illustrate two such cases. In (40) all lexico-syntacticconstraints favor the dat-acc order, which, accordingly, is the only one possible: (40) Peter Peter

wurde fur was for

schuldig guilty

befunden,... found

a. ...einem

KolLEgen / seinem colleague / his-Dhl geschickt zu haben. sent to have α-DAT

CHEF eine BOMbe boss a bomb

b. #...eine BOMbe einem KolLEgen a bomb α-DAT colleague geschickt zu haben. sent to have

/ seinem /his-DAT

CHEF boss

'Peter was found guilty to have sent a colleague / his boss a bomb' Example ( 4 1 ) is one of the sort I didn't consider much in this paper; DAT and ANIM conflict (with DEF, presumably irrelevantly, siding with DAT); since ADF is neutral on the issue, the acc-dat order, as preferred by the highest lexico-syntactic constraint ANIM, wins: (41) Die Sache wurde the thing became a. ...einen a-Acc b. #...dem the-Dat

kriminell, als criminal when

GeFANgenen prisoner

dem the-Dat

Lügendetektortest lie-detector-test

sie... they Lügendetektortest lie-detector-test

einen a-Acc

GeFANgenen prisoner

aussetzen wollten. subject wanted aussetzen wollten. subject wanted

'The whole thing got criminal when they wanted to expose a prisoner to the lie detector test' ACC/.i3, ,a„ Dat/ra.a„ WF

IPHR

F3G

i: a.

: 3 ( A C C 3 . + O N )(DAT a . on v)

**

DAT

b.

: 3 (DAT 3 . Q „ )(ACC 3 i + £ ,„ v)

**

ANIM!

FP

ANS

ADF

D-A

We see, thus, that the definite/indefinite distinction is void if the pertinent NP is in focus.

98

5.

Daniel Büring

Conclusion

This paper explored in what ways the defininte/indefinite distinction influences word order in the German Mittelfeld. I have found three distinct factors to be relevant, the morphosyntax, focusing, and interpretation. The analysis developed models each of them and shows how they interact. Particular attention was devoted to the interpretation-related constraints that regulate the formal realization of the generic/existential distinction. I took as my starting point a simple generalization about the phrasing-behavior of generic vs. existential indefinites, which was motivated outside the realm of double object constructions; I then demonstrated how that very generalization, when combined with a theory of stress-related word order variation such as that of Büring (2000), yields a wide range of correct predications about double object constructions. I have tried to control for the various parameters such as focus/givenness, animacy, case, definiteness etc. as scrupulously as I could, and I believe the picture presented in this paper to have a fine-grainedness and accuracy that exceeds that of previous studies. Nonetheless, I could present here only a fraction of the different combinations of parameters that the theory makes predictions for (about 430, I believe), and I won't pretend to have been able to reliably test all the others in the privacy of my office. Also, while I've been careful to steer clear of some potentially relevant factors such as scopal dependencies or more complex F-patterns within the argument NPs, I have no doubt that there are others which make themselves felt in the examples discussed in this paper and lead me to wrong interpretations of my findings (I can't shake off the feeling, for example, that some of the more complex double object constructions may involve a more articulated inventory of information structural categories than just focus and background; I resisted the temptation of introducing any further features such as "contrast" or "topic", because in the absence of clear criteria to test those, they would amount to no more than arbitrary features used to trigger certain word order anomalies). More work is waiting to be done. Accepting the conclusion drawn in the present paper, that prosody, morphosyntax and semantics are all irreducible forces influencing the ordering of arguments, it is worthwhile to note that all three of them pull in the same direction in the majority of cases, often masking one another; often, definites are in the background and indefinites are focused, and if an indefinite isn't focused, that is often because it is generic and as such can be repeated. In other words, the constraints regularly converge. We could easily imagine and construct a grammar in which the prosody wants foci to follow the background (as in German), but in which, say, the background ADs, rather than the focus ADs, are mapped onto the nuclear scope. Such a grammar would produce a very different language from German, presumably one without a clear rule of thumb such as "indefinites tend to follow definites", i.e., without convergence of the constraints. To take another example, it seems likewise "natural" that definiteness and animacy should converge in that sense, assuming that we speak about humans and animals more often than we do about inanimate things, and given that that which we speak about would generally be encoded as a definite. Formal grammars such as the one used in the

What Do Definites Do That Indefinites Definitely Don't?

99

present paper do not offer an explanation for this convergence. To the extend that such convergences are common in grammars, they perhaps hint at something like "usability": A language (as perhaps most systems) is simply more stable and usable if little changes don't yield big effects, that is, in which principles, often redundantly, converge. This is at least a conceivable alternative to a reductionist (or "minimalist") approach, according to which convergence must be attributed to one grammar-internal force; and, if the findings of this paper are on the right track, it is perhaps the empirically more accurate one.

References Büring, Daniel (1996). Interpretation and Movement: Towards an Economy-Theoretic Treatment of German 'Mittelfeld' Word Order. Ms, Frankfurt University. Büring, Daniel (2001). Let's Phrase It! - Focus, Word Order, and Prosodic Phrasing in German Double Object Constructions. In: Gereon Müller & Wolfgang Sternefeld (eds.) Competition in Syntax. Amsterdam: Benjamins. 101-137. Diesing, Molly (1992). Indefinites. (Linguistic Inquiry Monograph No. 20.) Cambridge, MA: MIT Press. Eckardt, Regine (1996). Intonation and Predication: An Investigation in the Nature of Judgement Structure. Technical Report 77, Sonderforschungsbereich 340, Stuttgart & Tübingen. Gutierrez-Bravo, Rodrigo (1999). Focus, Phonological Phrasing and Word Order in Spanish and English. Ms, UC Santa Cruz. [Presented at the 30th Linguistic Symposium on Romance Languages, Gainesville, FL.] Halliday, Μ. A. K. (1967). Notes on Transitivity and Theme. Journal of Linguistics 3. 199-244. Heim, Irene (1982). The Semantics of Definite and Indefinite Noun Phrases. PhD dissertation, University of Massachusetts, Amherst. Hoop, Helen de (1992). Case Configuration and Noun Phrase Interpretation. PhD dissertation, Groningen. [Published 1996, Garland.] Jackendoff, Ray S. (1997). The Architecture of the Language Faculty. Cambridge, MA: MIT Press. Jacobs, Joachim (1992). Integration. Technical Report 14, Sonderforschungsbereich 282, Düsseldorf, Köln & Wuppertal. Kratzer, Angelika (1995). Stage Level and Individual Level Predicates. In: Greg N. Carlson & Francis J. Pelletier (eds.) The Generic Book. Chicago: University of Chicago Press. 125-175.

100

Daniel Büring

Krifka, Manfred (1995). Focus and the Interpretation of Generic Sentences. In: Greg N. Carlson & Francis J. Pelletier (eds.) The Generic Book. Chicago: University of Chicago Press. 238-264. Krifka, Manfred (1999). Focus, Prosody and Syntax, (part of forthcoming book on focus). Lenerz, Jürgen (1977). Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Narr. Müller, Gereon (1998). German Word Order and Optimality Theory. Technical Report 126, Sonderforschungsbereich 340, Stuttgart & Tübingen. Nagahara, Hiroyuki (1994). Phonological Phrasing in Japanese. PhD dissertation, UCLA. Rochemont, Michael (1986). Focus in Generative Grammar. Amsterdam/Philadelphia: John Benjamins. Rooth, Mats (1996). Focus. In: Shalom Lappin (ed.) The Handbook of Contemporary Semantic Theory. London: Blackwell. 271-297. Schwarzschild, Roger (1999). GIVENness, AvoidF and Other Constraints on the Placement of Accent. Natural Language Semantics 7. 141-177. Selkirk, Elisabeth (1984). Phonology and Syntax: The Relation between Sound and Structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth (1995). Sentence Prosody: Intonation, Stress, and Phrasing. In: John A. Goldsmith (ed.) The Handbook of Phonological Theory. London: Blackwell. 550-569. Stechow, Arnim von (1981). Topic, Focus, and Local Relevance. In: Wolfgang Klein & Willem Levelt (eds.) Crossing the Boundaries in Linguistics. Dordrecht: Reidel. 95-130. Stechow, Arnim von (1989). Focusing and Backgrounding Operators. Technical Report 6, Fachgruppe Sprachwissenschaft, Universität Konstanz. Stechow, Arnim von (1991). Current Issues in the Theory of Focus. In: Arnim von Stechow & Dieter Wunderlich (eds.) Semantik - Ein internationales Handbuch der zeitgenössischen Forschung. (Vol. 6 in HSK). Berlin & New York: de Gruyter. 804-825. Stechow, Arnim von & Susanne Uhmann (1986). Some Remarks on Focus Projection. In: Werner Abraham & Sjaak de Meij (eds.) Topic, Focus and Configurationality. Amsterdam: John Benjamins. 295-320. Truckenbrodt, Hubert (1995). Phonological Phrases: Their Relation to Syntax, Focus, and Prominence. PhD dissertation, MIT. [Published 1999, MITWPL.] Truckenbrodt, Hubert (1999). On the Relation between Syntactic Phrases and Phonological Phrases. Linguistic Inquiry 30. 219-255. Uechi, Akihiko (1998). Focus Phrasing in Japanese - An Optimality Approach. In: Elena Benedicto, Maribel Romero & Satoshi Tomioka (eds.) Proceeding of Workshop on Focus. UMass Occasional Papers in Linguistics. Vol. 21. Amherst, MA: GLSA. 259-276. Uhmann, Susanne (1991). Fokusphonologie. Tübingen: Niemeyer.

"Wi(e)der" and "Again(st)"1

Cathrine Fabricius-Hansen

1.

Preliminaries

It is by now a well-established fact that the occurrence of German wieder and English again may express that there has been a former event of the type described in the sentence like, e.g., the opening of the door, the leaving of a guest in (la)/(2a), or that with the event described in the sentence, a state which has formerly held is restored, like, e.g., the door being open, the absence of a certain participant in (lb)/(2b). 2 These two readings have come to be known as "repetitive" and "restitutive" wieder/again, respectively. (1)

a. [Jemand öffnete die Tür um neun Uhr.] Eine halbe Stunde später wurde die Tür wiEder geöffnet. '[Somebody opened the door at nine o'clock.] Half an hour later the door was opened aGAlN' b. [Hans Schloß langsam die Tür.] Sie wurde jedoch sofort wieder geöFFnet. '[Hans slowly shut the door.] But it was immediately Opened again'

(2)

a. Heute ist WiEder ein Teilnehmer abgereist. 'Today again a participant left' b. Heute ist ein Teilnehmer wieder ABgereist. 'Today a participant has left again'

This so-called ambiguity of wieder and again caught the attention of theoretical linguistics in the seventies (McCawley 1972, Fabricius-Hansen 1975, Dowty 1979). After a certain low in the eighties (Fabricius-Hansen 1980, 1983), interest in the subject of 1

2

The German preposition wider 'against' is etymologically identical to the adverb/particle wieder 'again' (cf. section 3). I thank two anonymous reviewers for very useful comments on the first version of this paper and Torgrim Solstad for assisting me with the German-English data (see section 4). Capitals mark main sentence stress (focus accent).

102

Cathrine

Fabricius-Hansen

wieder has revived since the early nineties as witnessed by Kamp & Roßdeutscher (1994), Blutner & Jäger (to appear), Jäger & Blutner (to appear), Pittner (2000), Egg (2000), and Klein (in this volume) in addition to von Stechow (1995, 1996) and unpublished contributions of my own.3 The discussion concentrates on two issues:4 (i)

What are the sentence-internal factors determining the interpretation of wieder in a given sentence?

(ii)

How can we explain the behaviour of wieder, in particular, the relationship between the two readings?

As far as the first, empirical question is concerned, there seems to be general agreement (a) that a repetitive reading is obligatory when wieder precedes the finite verb in V2-position, and obligatory or at least strongly preferred when wieder precedes an object, the sentence subject or an adverbial modifier (at least of a higher type 5 ) in the so-called middle-field, with some semantically determined exceptions pointed out by von Stechow (1996:109ff.) and Pittner (2000); (b) that a genuine restitutive interpretation is allowed only with telic change-of-state predicates; and (c) that nuclear stress on wieder, indicating narrow focus, blocks the restitutive reading in cases where it is not excluded by (a) and (b). In the examples in (3) the restitutive interpretation is excluded by one or the other of these constraints (underlined). (3)

a. WlEder Again

musste Arnim sein altes Auto reparieren, (a) had Arnim his old car to-repair.

b. Arnim hat wieder das Dach Arnim has again the roof c. Ich I

bin wieder zu have again too

d. Arnim hat die Arnim has the

repariert, (a) repaired.

spät aufgestanden, (a) late got-up.

Kategorialgrammatik wieder kritisiert, (b) Categorial-Grammar again criticized.

e. Die Tür wurde eine halbe the door was an half

Stunde später hour later

WlEder geöffnet, (c) again opened

The question of how to explain this observed distribution of repetitive and restitutive wieder and how to describe the relationship between the two readings of wieder/again has caused considerably more controversy. The majority of the proposals represent 3

4 5

Referred to as Fabricius-Hansen (1994a, 1994b) by von Stechow (1996) and as Fabricius-Hansen (1995) by Klein (in this volume), Blutner & Jäger (to appear), and Pittner (2000). Cf. Kamp & Roßdeutscher (1994:190). Under this category I subsume adverbials that are higher than so-called process/state-related adverbials in the adverbial hierarchies outlined by, e.g., Frey & Pittner (1999), Haider (to appear),, and Pittner (2000).

103

"Wi(e)der" and "Again(st)"

what one might call a reductionist (and "repetitionist") view: restitutive wieder/again is described as repetitive wieder/again having narrow scope, i.e., having scope only over the state resulting from the event described by telic change-of-state predicates (achievements and accomplishments in Vendlerian terms) like ÖFFNEN 'to open' (transitive) and ABREISEN 'to leave' which are assigned the result states described by '(be) open' and '(be) not present', respectively. This analysis involves decomposition of such verbs, at least at the level of semantic representation (cf. Dowty 1979). The verb OPEN must be analysed as something like [CAUSE TO] (BECOME(OPEN)) in order for again to modify the state predicate OPEN in its restitutive reading.6 In his papers from 1995 and 1996, von Stechow strongly advocates a reductionist explanation of the repetitive-restitutive ambiguity within a theory of lexical decomposition in syntax. Like other decompositional approaches,7 his theory predicts that ambiguity arises only with predicates that call for such decomposition ("resultatives"); cf. observation (b) above and von Stechow (1996:105ff.). And delegating lexical decomposition to syntax, it accounts for observation (a) in terms of operator scope only, i.e., in purely structural terms. It predicts that ambiguity may arise only if wieder/again occurs to the right of nominal complements, as in (lb), (2b) and (3e). In this case, wieder/again may occupy the structural position below (CAUSE + ) BECOME which is correlated with the restitutive reading, or it may be adjoined to the phrase headed by (CAUSE + ) BECOME, thus giving rise to a repetitive reading. This structural ambiguity is displayed in (4), which is taken from von Stechow (2000:288). (4)

weil Fritz das Fenster because Fritz the window

wieder öffnete again opened

a.

Ugrs Fritzi [ A g r o the window2 again open]]]]]] (repetitive)

[VoiceP

b.

[ A g rs Fritzi [ A g r o the window2 open]]]]]] (restitutive)

ti

[voiceP

ti

[voice

[voice

CAUSE [ V p BECOME [ X P t2

CAUSE [ V p BECOME

again [ XP t2

If, however, wieder/again precedes the subject or an object, it necessarily occupies a position higher in the tree than VoiceP and consequently must receive a repetitive interpretation. Von Stechow (1996), however, has little to say as far as the generalization in (c) above is concerned.8

6

7

8

It is a problem for this analysis that evidence for wieder having intermediate scope between CAUSE and BECOME apparently cannot be found (cf. von Stechow 1996, 2000). Cf. McCawley (1972), Fabricius-Hansen (1975), Dowty (1979). Egg (2000), Blutner & Jäger (1999, to appear) and, as far I can judge, Klein (in this volume) represent approaches that are nondecompositional but nevertheless reductionist or "repetitionist", taking repetitive wieder with its apparently more simple meaning to be basic. They are all considerably less explicit than von Stechow (1995,1996) as far as syntax goes. This issue is taken up by Blutner & Jäger (1999, to appear), Egg (2000), and Klein (in this volume).

Cathrine Fabricius-Hansen

104

The main advantages of von Stechow's (1996) theory can be summarized as follows: It gives a principled explanation for the repetitive/restitutive ambiguity which is easily incorporated into a general framework of compositional interpretation, it can do so without stipulation of a lexical ambiguity of again, and it is able to account for disambiguating word order effects. (Jäger & Blutner to appear: section 3)

Blutner & Jäger (1999, to appear) and Jäger & Blutner (to appear) (henceforth Blutner & Jäger), have challenged von Stechow (1996). They argue, followed by Egg (2000), that restitutive readings are possible with wieder/again preceding an object (5) or a subject (6), contrary to generalization (b), which von Stechow's theory is meant to explain.9 (5)

weil Peter because Peter

wieder ein again a

FENster window

öffnet opened

(6)

Es siedeln sich wieder Delawaren in New Jersey an. EXPL settle REFL again Delawares in New Jersey PART 'Delawares are settling in New Jersey again'

In addition, Blutner & Jäger point out that in (6), we get a restitutive reading without reference identity between the "theme" or "holder" of the state resulting from settling in New Jersey ("live in NJ") and the presupposed prior instantiation of the same state. In von Stechow's strictly structural approach, this possibility is excluded because it implies a scope clash between wieder and the indefinite subject NP. Blutner & Jäger view wieder/again as semantically underspecified (or rather polysemous?) and take the two readings to be inherent interpretation possibilities that are ranked differently in different contexts by a set of (bidirectional) Optimality Theoretic principles concerning word order, information structure, and scope. In his answer to Blutner & Jäger, von Stechow (2000) suggests some modifications of his theory that may overcome some, if not all, of their objections, and shows that their OT explanation of interpretation preferences in given contexts can be implemented into his own theory. I will not exclude that it may be possible to explain the distribution of the repetitive and the restitutive reading and the relationship between them along the lines von Stechow proposes. But I am not convinced that it should be done that way, i.e., that it is an adequate "type" of explanation. This is partly because I adhere to the principle underlying Blutner & Jäger's as well as Egg's approaches (and Optimality Theory in general) that one should not let syntax do more work than absolutely necessary as far as semantic interpretation is concerned but should seek to place the explanatory burden

9

See also Pittner (2000). One could argue that the identity of the referents in question is, in a way, irrelevant in these cases. What matters - and is restored - is that not all windows are closed and that there are Delawares in New Jersey.

"Wi(e)der" and "Again(st) "

105

with semantics and pragmatics, maximizing the role of general semantic-pragmatic interpretation principles. But my reservations also have to do with two further questions which the repetitive-restitutive dichotomy gives rise to, but which solutions taking repetitive wieder to be basic do not even try to answer: (iii) How does it come that wieder and again, being of different origin, have apparently undergone a parallel semantic development from a preposition/adverb meaning "contra", "against" or the like into the restitutive-repetitive range of readings illustrated above?10 (iv) Why is a restitutive reading normally not found with other "repetitive" adverbs/particles like English once more and German noch einmal, abermals, erneut? According to von Stechow (1996), scope restrictions prevent them from occurring in the position of restitutive wieder. But why should they be restricted in this way, given that they have the repetitive meaning attributed to wieder as well?" The fact that wieder and again have generally been treated as semantically equivalent in the theoretical discussion raises another question that has not been discussed so far: (v)

Do wieder and again behave in the same way in actual discourse? And if not, what would be the consequences - if any - for the semantic description of again as opposed to wieder?

In what follows, I shall not try to argue in any detail against von Stechow's decompositional solution to the repetitive-restitutive problem. My aim, rather, is to contribute to answering the questions that have not been asked. At this stage of a discussion that has been going on for approximately 30 years, by and large neglecting these questions and being based primarily on (pairs of) sentences without a natural context,12 it may seem more rewarding to focus on the relation between repetition and contrast, on the one hand, and the role wieder plays in natural discourse, on the other hand. In section 2, I discuss the nature of the entities wieder modifies when used as an "event adjunct" (repetitively) and as a "process adjunct" (restitutively) in the sense of Pittner (2000), and the semantic contribution of wieder in what could be called standard instances of repetitive and restitutive readings, i.e., cases where wieder occurs with a telic change-of-state predicate but one of the two readings is excluded or strongly disfavoured under unmarked focus conditions. Section 3 aims at showing that 10

11

12

The issue is discussed tentatively by Fabricius-Hansen (1980, 1983) and Kamp & Roßdeutscher (1994:197, fn. 17). In fact, I do not see how assigning the same (repetitive) meaning to wieder and, e.g., erneut but excluding by stipulation the latter from occurring in the structural positions that give rise to a restitutive reading of the former has more explanatory power than viewing wieder as polysemous and partially synonymous with erneut. Kamp & Roßdeutscher (1994) are a notable exception to this tendency.

Cathrine Fabricius-Hansen

106

the semantic variation wieder exhibits can be explained quite naturally a posteriori if we take the interpretation pattern observed in the standard restitutive case illustrated, e.g., in (lb) and (2b), to be basic; by the same token, this section suggests an answer to questions (iii) and (iv) above. Section 4, which focuses on the function of wieder and again in discourse, presents evidence to the effect that although these two adverbs are (and probably should be) taken to be equivalent at some level of description, they differ in interesting ways when it comes to their actual use. This can be seen as an argument in favour of the description presented in section 3. Section 5 contains a brief conclusion.

2.

Syntactic and Semantic Aspects of the RepetitiveRestitutive Dichotomy

Pittner (2000), Frey & Pittner (1999) and Haider (1999), among others, argue that the structural position of an adverbial adjunct reflects its semantic "domain", i.e., the type of entities it is applied to. As far as the syntactic behaviour of wieder is concerned, Pittner (2000) shows that the repetitive variety patterns with what she calls event adverbs, which "c-command the base position of the highest argument" and delimit the range of existential closure with the effect that indefinite NPs occurring to the left of the adverb receive a "strong" (specific) interpretation (Pittner 2000:21 Of.). Restitutive wieder, on the other hand, "shares many properties with process adjuncts, minimally ccommanding the final verb" (Pittner 2000:215); that is, it patterns in certain respects with so-called manner adverbs. It takes position close to the (final) verb to the right of sentence negation and, as we have seen, it normally does not appear to the left of an object or the subject.13 (7) will receive a restitutive interpretation unless wieder itself carries the focus accent. (7)

Arnim hat das Haus Arnim has the house

in Tessin wieder verkauft. in Tessin again sold

I shall assume that this is an adequate account of the difference between repetitive and restitutive wieder as far as syntax is concerned and that the V-projection wieder modifies is existentially closed under/below repetitive wieder but above/outside the domain of restitutive wieder. As far as semantics is concerned, it is generally agreed that the presence of wieder does not contribute to the assertoric content of a sentence but triggers a presupposition that is determined by what wieder modifies at the level of semantic representation.14

14

Exceptions like those discussed in section 1 are instances of so-called "integration" of the object or subject into the predicate. See Pittner (2000) for further references. Cf. von Stechow (1996:95), Fabricius-Hansen (1980, 1983), Kamp & Roßdeutscher (1994), Blutner & Jäger (to appear).

"Wi(e)der" and "Again(st)"

107

And what wieder modifies, as we have seen, is mediated more or less unambiguously by surface word order: material to the right of wieder is in the scope of wieder and enters into the presupposition together with the sentence predicate, whereas constituents to the left of wieder belong to the assertoric part (Kamp & Roßdeutscher 1994). This means that quantifiers following wieder are copied into the presupposition, so to speak - they cannot bind variables in the presupposition. Quantifiers, etc., in a subject or an object preceding wieder in the surface sentence, on the other hand, will bind variables both in the assertoric and the presuppositional part of the sentence content, leading to identity of reference between corresponding arguments.15 (8)

In diesem Jahr hat Arnim wieder einige Arbeiten über Tempus in this year has Arnim again some studies on tense heftig kritisiert. vehemently criticized Ass.: This year, Arnim has vehemently criticized some studies on tense. Pres.: At least once before, Arnim vehemently criticized some studies on tense.

(9)

Arnim hat einige Arbeiten über Tempus jetzt wieder heftig Arnim has some studies on tense now again vehemently kritisiert. criticized Ass.: Arnim has vehemently criticized some studies (X) on tense. Pres.: At least once before Arnim vehemently criticized X.

For the reason outlined above, then, the difference between external and internal binding of variables in the presupposition goes hand in hand with the restitutive-repetitive dichotomy: a standard restitutive interpretation correlates with identity of reference between the participant(s) of the state resulting from the asserted event - i.e., the "theme", "patient", or "holder" of the state, or whatever one chooses to call such a role16 - and its (their) counterpart(s) in the presupposition. (10) (repetitive) patterns with (8), and (11) (restitutive) with (9). (10) Arnim hat wieder ein Haus in Tessin verkauft. Arnim has again a house in Tessin sold 'Arnim has sold another house in Tessin' (11) Arnim hat ein Haus in Tessin wieder verKAUFT. Arnim has a house in Tessin again sold 'Arnim has sold a house/one of his houses in Tessin again' 15 16

Cf. Fabricius-Hansen (1980, 1983). Von Stechow (1996:109) uses the term "holder" for the subject argument of so-called holder + object result verbs like verlassen 'leave' where the result state involves both arguments. These are the verbs that allow or prefer a restitutive interpretation even when wieder precedes the object.

108

Cathrine

Fabricius-Hansen

Von Stechow (1996) defines the meaning of wieder as shown in (12) - von Stechow's (3-7) - where the first sentence specifies the presupposition. (12) Let Ρ be a property of eventualities and let e be an eventuality, ||again||(P)(e) is defined only if 36'[|ΜΑΧ||(Ρ)(ε') = 1 & e' < e]. Where defined, ||again||(P)(e) = 1 iff P(e) = l. 17 The presupposition requires that there be an eventuality prior to e having the property P. Under a restitutive reading, Ρ is the result state property expressed by the small clause restitutive wieder is adjoined to according to von Stechow's theory, say the door being open - OPEN(d) - in (lb) or Arnim not owning the particular house in Tessin -•POSSESS(a,h) - in (11). (11), then, might be paraphrased as shown in (1Γ). (1 Γ) Ass.: Arnim has sold a house (h) in Tessin. Pres.: Before the selling event there was a maximal eventuality characterized by Arnim's not being in the possession of h. (12) presents the presupposition triggered by wieder as indefinite or existential. So it does not do justice to the fact that this presupposition is of an anaphoric nature, like all other presuppositions, 18 a fact that Kamp & Roßdeutscher (1994) have demonstrated convincingly - and which von Stechow (1996:133, note 2) does recognize after all. In particular, (12) cannot account for the intricate merging of information induced by wieder and information stemming from the modified predicate that yields the socalled restitutive reading in a standard case like (lb) (repeated below). What we infer from (lb) is not simply that the state characterized by the door being open, which is the result of the opening event described in the sentence, has held once before that event. We infer that the door in question stayed closed during the whole (short) period of time lying between the closing event described in the first sentence and the opening event of the second sentence so that the latter restores a state that must have held at the onset of the former, viz., the door being open. (lb) Hans Hans Sie it

18

schloß langsam slowly shut

die the

Tür. door

wurde jedoch sofort wieder geÖFFnet was but immediately again opened

MAX(P) holds of e' iff e' is a maximal P-eventuality, i.e., iff there is no proper part of e' such that P(e') = 1; cf. von Stechow (1996:96). The presence of the maximality operator in the presupposition clause of (12) prevents the two instantiations of Ρ (e and e') from being part of one larger P-instantiation; they have to be separate eventualities. Cf. van der Sandt (1992), Saebe (1996), Geurts (1999).

109

"Wi(e)der" and "Again(st)"

The contribution of restitutive wieder in the standard case, then, is more adequately described as follows (cf. also Fabricius-Hansen 1980, 1983): The central conception conveyed by restitutive wieder is that the process which is implicitly or explicitly asserted by the sentence in which it occurs was preceded by an opposite process whose effects the later process undoes, thereby restoring the state of affairs which obtained when the first process began. (Kamp & Roßdeutscher 1994:195)

Thus, (11) tells us that the asserted selling event (e), the result of which is that Arnim does not own h (i.e., the house he sold), has been preceded by an event (e') by which he came into possession of h, i.e., an event which like selling involves a succession of the two complementary states POSSESS(a, h) and ^POSSESS(a, h) but in the opposite order; and the result state of this event (e') is identified with the pre-state of the selling event itself (the state characterized by POSSESS(a, h)) with the effect that the result of the latter - -WSSESS(a, h) - is conceived as a restoration of the state holding at the onset of the former: the two states characterized by ^POSSESS(a, h) are separated by a single (maximal) state of the complementary type (POSSESS(a, h)). In other words, the presupposition is that there was a prior event that caused this very state to come into existence, whose effect is now undone. In (lb), then, the pre-state of the opening event, whose existence is asserted in the sentence containing wieder, is identified with the result state of the closing act whose existence is asserted in the preceding sentence. In (11"), likewise, the presupposed "opposite" change of state e' is identified with the buying act described in the relative clause. In both cases, the interpretation provides a maximally coherent text without "gaps" in the chain of eventualities; cf. figure 1. (11") Arnim hat ein Haus in Tessin, das er vor ein paar Jahren Arnim has a house in Tessin that he ago a couple years gekauft hatte, schon wieder verkauft, bought had already again sold 'Arnim has already sold a house in Tessin (again) which he bought a couple of years ago' Fig. 1: Interpretation of (lb) (restitutive) (pre-state of e) (result state of e')

(pre-state of e')

(result state of e)

s": OPEN(d) s':-OPEN(d) s: OPEN(d) /////////////////////////////| 11\\\\\\\\\\\\\\\\\\\\\\\\\\\| 11///////////////////////////// e' e

t

BE-CLOSED(d)

t

BE-OPENED(d)

110

Cathrine Fabricius-Hansen

This pattern of alternating complementary states - of i having the property that a maximal instantiation of P(xi, ..., x„) is located within them or that they fall under two different maximal instantiations of that type. Focussing on two such intervals alone, however, thereby in a way abstracting from what happens between them, is what characterizes a repetitive 24

25

Correspondingly, restitutive wieder and "again" often co-occur with zurück and back or the prefix re-, respectively; cf. section 4. The semantic relationship between restitutive wieder and expressions denoting continuation without interruption (i.e., the complement of interruption) is discussed in more detail in Fabricius-Hansen (1975). It can be exemplified by the semantic equivalence between Die Tür wurde nicht wieder geöffnet 'The door wasn't opened again' and Die Tür blieb geschlossen 'The door stayed closed' as alternative continuations of the first sentence of (lb).

"Wife)der" and "Againfst)"

117

interpretation as opposed to a restitutive interpretation, which yields a with a continuous chain of eventualities; cf. figure 2 vs. figure 1. Obviously, it is not a very big step from the notion of "iteration" or "repetition" at the level of basic eventualities, involving one and the same "theme" referent, to the notion of iteration or repetition applied at a higher level, with respect to "larger" eventualities, time spans or situations. But it is a step that implies using wieder as a higher modifier. And at that level there is no room for an interpretation following the basic counterdirectional-restitutive pattern. The interpretation pattern in (13) cannot be applied since time spans, etc. are not events that can be defined by characteristic result states or fall under the counterdirectionality relation. From the fact that a time i contains an instantiation of the event type SELL(a, h), where h is a specific house in Tessin, and consequently has the property λε^ΞχΒε^ο^ε,η.^χ) & SELL(a, x)(e) & e c e*), we can infer nothing as far as the time spans bordering on i to its left and right are concerned. But time spans can have contrasting properties - e.g., if one does and the other does not contain an event of a certain type. We can conclude, then, that it is the nature of the modified entity that blocks the basic or prototypical interpretation pattern when wieder functions as a higher adjunct. The interpretation pattern we get in this case, however, is part of the basic pattern, viz., the pair zero; see also (27): (34) Der Mann ist zu Boden getaumelt [...] und dann sofort wieder aufgestanden, ohne daß ich ihm auch nur die Hand gereicht hätte. (PH1) The man fell to the ground [...], then instantly stood up without my even offering him a helping hand. (35) Ich nahm nicht den Heimweg hinunter in die Ebene, sondern verschwand in eine Gasse, so schmal, daß niemand da hätte neben mir gehen können. Die Gasse beschreibt eine Schleife und mündet nach einem Steilstück wieder in den Hauptweg zur Innenstadt. (PH1) Instead of heading across the plain on my way home, I turned into a street so narrow that no one could have walked beside me. This street describes a loop which, after a steep hill, leads to the main road to the Old City. (iib) wieder 1' b. ^exactly two inches -er than one foot]|= λΡ. max(P) = 1' + 2"

(6)

max(P) := id. P(d) = 1 & Vd'[P(d') = 1

d' < d]

Being of type , these complex DegPs cannot be interpreted in situ, but must move for interpretability to a position above the adjective's subject (not necessarily above the surface subject, if there are lower covert subjects). The movement leaves a trace of type d and creates a λ-abstract of type , which makes a suitable argument to the DegP. Here is a sample derivation. (7)

John is taller than 4 feet.

SS minus extraposition: John is [ap [DegP _er than 4 ft] tall] LF: [DegP * e r than 4 ft]i John is [AP ti tall] interpretation:4 max{d: tall(j,d)} > 4'

2 3

4

I treat degrees as a separate basic type (label "d"). This is not the DegP of Abney, Kennedy, and other proponents of the "DegP hypothesis". For them, AP is an argument of Deg, whereas my DegP is an argument of A. Abbreviations: tall(x,d): = χ is tall to degree d. j: = John.

Degree Operators and Scope

217

So John is taller than 4 feet is true iff the maximal degree to which John is tall exceeds 4 feet.

2.

DegP-Scope and Truth-Conditions

In the syntax and semantics just presented, I deliberately highlighted a close analogy between DegPs and DPs. In both categories, we find referring (and bound variable) phrases with denotations of a basic type (d or e) as well as quantificational phrases of the corresponding generalized quantifier type ( or ). And in both, the quantificational phrases can move at LF, leaving behind basic-type traces and creating complex λ-predicates. In the case of DPs, this movement is known as QR (Quantifier Raising) and is standardly assumed to be the source of truth-conditional ambiguities. If DegPs are so similar in their LF-syntax and semantics, shouldn't we find analogous scope-ambiguities due to multiple choices of landing-sites for DegP-movement? For example, when a DegP originates in the surface-c-command domain of a quantificational DP, negation, or intensional predicate, it should be able to scope either below or above that item - at least as long as there is nothing in its path that is known to create islands for movement. In the present section, we will try to test this prediction by comparing the range of truth-conditionally distinct readings that we can generate by DegP-movement in our system with the readings that are actually possible.5

1.1

Caveat: Systematic Equivalences and Anomalies

Not every scope ambiguity is a truth-conditional ambiguity. For example, when every boy scopes above every girl in the sentence Every girl saw every boy, the truthconditions are just the same as when it scopes below it. The methodological implications of this fact are familiar. If we want to study the properties of QR, we must choose our examples judiciously. The fact that this particular sentence has only one interpretation doesn't show that objects can't move over subjects. When it comes to quantifiers over degrees rather than ordinary individuals, this point is especially pertinent. Due to the ordered structure of the domain of degrees and the monotonicity property of adjective meanings, there are lots of cases where the relative scopes of a DegP and another quantifier are indistinguishable. To get a feel for what I am talking about, consider a comparative with a universal subject DP. (8)

5

Every girl is taller than 4 feet.

This whole section closely follows the reasoning in Kennedy (1997). 2.1 and 2.2 more or less just replicate his examples, arguments, and conclusions. 2.3 and 2.4 contain differing facts and conclusions, but even there I will mostly mimic his logic of argument.

218

Irene Heim

If the DegP -er than 4 ft is free to scope either below or above the subject, we have two LFs: a low-DegP LF as in (9a) and a high-DegP LF as in (9b). (9)

a. [every girl]! [-er than 4 ft] 2 ti is t2 tall b. [-er than 4 ft]2 [every girl] ι ti is t2 tall

On the semantics specified above, these structures get the truth-conditions in (10a,b) respectively. (10) a. Vx[girl(x) —»max{d: tall(x,d)} > 4'] b. max{d: Vx[girl(x)

tall(x,d)]} > 4'

(10a) says that each girl's (maximal) height is above 4'. (10b) is less transparent at first: We form the set of degrees d such that every girl is tall to d. Given the monotonicity of the tall-relation, this is precisely the set of degrees to which the shortest girl is tall (or the shortest girls, if two or more are tied for shortest). So the maximum of that is the (maximal) height of the shortest girl(s), and (10b) says this is above 4'. But that's the same claim as (10a)! If every girl is taller than 4', than the shortest girl is, and if the shortest girl is taller than 4', then every girl is. Something similar happens when the subject is an existential quantifier. If we construct two LFs for (11) analogous to the ones in (9a,b), we derive the truthconditions in (12a,b). (11) Some girl is taller than 4 feet. (12) a. 3x[girl(x) & max{d: tall(x,d)} > 4'] b. max{d: 3x[girl(x) & tall(x,d)]} > 4' In (12b), we have the set of degrees d such that some girl is tall to d. This turns out to be the set of degrees to which the tallest girl is tall (or the tallest girls in the event of a tie). So (12b) says that the (maximal) height of the tallest girl(s) is above 4'. But if some girl is taller than 4', then the tallest one is, and if the tallest girl is taller than 4', then some girl is. So again (12b) winds up being equivalent to (12a).

6

What about infinite domains? If there were infinitely many girls, there might be no shortest ones and/or no tallest ones. In that case, the maxima referred to in (10b) or (12b) would be undefined. So my real claim, more accurately, is that pairs like (10) and (12) are equivalent whenever these maxima are defined. For my methodological point, this qualification makes no difference. I assume (see right below) that the high-DegP LF is automatically ruled out when its interpretation refers to an undefined maximum. So even if we consider universal and existential quantifiers with infinite domains, there can be no detectable scope ambiguity.

Degree Operators and Scope

219

How pervasive is this kind of equivalence? Quite. It generalizes from comparatives to equatives. 7 It generalizes from universal and existential to all monotone increasing quantifiers. And it generalizes from quantifiers over individuals to quantifiers over possible worlds. This means that we would be wasting our time trying to detect scopeambiguities, not only in examples like (8) and (11), but in any of the following as well. (13) a. Every girl is as tall as John is. b. Some girl is as tall as John is. (14) a. More than 20 girls are taller than 4 feet. b. Most of the papers are longer than 5 pages. (15) a. The paper is required to be longer than 10 pages, b. The paper is allowed to be longer than 10 pages. Pick (15a), for instance. Required is a necessity operator, a universal quantifier over the set of accessible worlds (for some contextually given accessibility relation, plausibly here a deontic one). If the DegP -er than 10 pp scopes below required, our semantics delivers the truth-condition in (16a); if the DegP scopes above required, we predict (16b).8 (16) a. Vw e Acc: max{d: long w (p,d)} > lOpp b. max{d: Vw e Acc: long w (p,d)} > lOpp (16a) transparently says that the paper is longer than lOpp in every accessible world. (16b) instructs us to determine the maximal degree to which the paper is long in every accessible world. That is its (maximal) length in those accessible worlds where it is shortest (the "minimal compliance" worlds, so to speak). This, (16b) says, exceeds 1 Opp - which is tantamount to (16a), since if the paper exceeds 1 Opp in the accessible worlds where it's shortest, it exceeds lOpp in every accessible world, and vice versa. The reasoning is just as for every girl in (8). By the same token, the possibility operator allowed in (15b) behaves just like some girl in. (11). I leave it to the reader to establish the pertinent equivalences for the remaining examples.

7

8

This is assuming the (standard) "at least"-interpretation of equatives, on which Mary is as tall as John is true (though under-informative) when she is actually taller than him. A suitable meaningrule for equative DegPs analogous to (5a) is: [as as one foot\ = λΡ. max(P) > Γ. As for exactlyequatives, see 2.2 below. Abbreviations: p: = the paper. long w (x,d): = χ is long to degree d in world w. Acc(w): = the set of worlds accessible from w. The world argument of Acc is suppressed when it is the utterance world.

220

Irene Heim

So what have we learned so far about DegP-movement? Nothing yet. The examples we have analysed here are consistent with everything we have assumed plus the assumption that DegP can move to any position where it is interpretable. They are also consistent with the possibility that DegP never moves beyond the lowest position where it is interpretable. We just can't tell. Our next set of examples will still not allow us to tell how far DegPs can move, but this time for a different reason. The high-DegP LF will not be equivalent to the lowDegP LF, but it will be semantically deviant and thus ruled out by independent principles. Consider scoping a DegP over a negation, as in the second LF given below for sentence (17a). (17) a. Mary isn't taller than 4 feet. b. not [-er than 4'] Mary is t tall max{d: tall(m,d)} > 4 ft c. [-er than 4'] not [Mary is t tall] # max{d:-i tall(m,d)} > 4 ft The predicted interpretation for (17c) refers to the maximum of the set of degrees to which Mary isn't tall. But this set has no maximum. 9 So if such an LF can be generated, it is a presupposition failure. Plausibly, this suffices to explain why only one reading (the one in (17b)) is attested. For all we know, (17c) may or may not be generated by the syntax; even if it is, it will be filtered out. This observation generalizes from plain negation to other implicitly negative or monotone decreasing operators. (18) a. At most two girls are taller than 5 feet. b. [at most two girls] ι [-er than 5']2 ti are t2 tall I {x: girl(x) & max{d: tall(x,d)} > 5'} | < 2 c. [-er than 5']2 [at most two girls] ι ti are t2 tall # max{d: |{x: girl(x) & tall(x,d)}| < 2} > 5' (19) a. (Mary works 60 hrs a week, and) she refuses to work harder than that. b. she refuses [[-er than that] PRO to work t hard] Vw e Acc: -i max{d: m works d-hard in w} > 60hrs/wk c. [-er than that] she refuses PRO to work t hard # max{d: Vw e Acc: -i m works d-hard in w} > 60hrs/wk 9

This way of ruling out reading c admittedly depends on the specific semantics for -er that I chose, which makes reference to maximality. (Maximalization in the interpretation of the i/iaw-clause is standard (see von Stechow 1984, Rullmann 1995b), but in the matrix clause it is not.)

Degree Operators and Scope

221

As the reader can verify, the b-LFs capture the (only) available readings, and the c-LFs refer to undefined maxima. If scoping over upward monotone operators always gives equivalent readings, and scoping over downward monotone ones always implies presupposition failure, we are left with the non-monotone ones. (20) a. Exactly two girls are taller than 5 feet. b. [exactly two girls]i [-er than 5']2 ti are t2 tall |{x: girl(x) & max{d: tall(x,d)} > 5'}| = 2 c. [-er than 5']2 [exactly two girls] ι ti are t2 tall max{d: |{x: girl(x) & tall(x,d)}| = 2} > 5' The maximum referred to in (20c) turns out to be well defined. In fact, the maximal degree to which exactly two girls are tall coincides with the maximal degree to which at least two girls are tall. So here we finally seem to have found a case that tells us something. The fact is that sentence (20a) unambiguously means (20b) and does not share a reading with the sentence At least two girls are taller than 5 feet. But if DegP is allowed to scope over the quantifier, as in (20c), then such a reading is generated. So the analysis makes inadequate empirical predictions unless supplemented by some constraint on DegP-movement. This is a fine argument given the assumptions I have adopted in this paper. It may not go through, however, in the context of a more comprehensive theory which incorporates an account of polar opposition along the lines of Bierwisch (1987) and Kennedy (1997). Briefly, (20c) may be ruled out by an independently motivated prohibition against sets of degrees which fail to be initial or final segments of the scale. Not being able to go into this, I leave it open here whether non-monotone operators can tell us anything about DegP-movement.10

1.2

Exactly-Differentials, Less-Comparatives, and Kennedy's Generalization

The equivalences observed in 2.1 hold for simple comparatives (and equatives) and also for comparatives with simple differential phrases, but they no longer hold for comparatives with a differential phrase containing at most or exactly (or for equatives modified by at most or exactly). Nor do they hold when we switch from morecomparatives to /ess-comparatives. For instance, we saw that if the shortest girl is taller than John, then every girl is. But if the shortest girl is exactly one inch taller than John, 10

The set {d: |{x: girl(x) & tall(x,d)}| = 2} in (20c) fails to be an initial segment of the set of all heights whenever there are more than 2 girls, because it does not contain the degrees to which the third-tallest girl is tall. The constraint in question is needed to account for what Kennedy calls "cross-polar anomaly".

Irene Heim

222

it doesn't follow that every girl is. And if the shortest girl is less tall than John, it also doesn't follow that every girl is. These and analogous non-equivalences will enable us to construct examples where the low-DegP and high-DegP LFs have clearly distinct yet equally non-deviant truth-conditions, and where we can thus get empirical evidence about the availability of each LF. Starting with exacify-differentials, consider the sentences in (21a) and (22a) and their predicted truth-conditions for LFs with low-DegPs (b) and high-DegPs (c). (21) a. (John is 4' tall.) Some girl is exactly 1" taller than that. b. [some girl]] [exactly 1" -er than 4']2 ti is t2 tall 3x[girl(x) & max{d: tall(x,d)} = 4' + 1"] c. [exactly 1" -er than 4']2 [some girl]] ti is t2 tall max{d: 3x[girl(x) & tall(x,d)]} = 4* + 1" (22) a. (...) Every girl is exactly 1" taller than that. b. Vx[girl(x) -» max{d: tall(x,d)} = 4' + 1"] c. max{d: Vx[girl(x)

tall(x,d)]} = 4' + 1"

The sentences are judged unambiguous, and the b-LFs correctly represent their intuitive meanings. What about the c-LFs? (21 c) says, in effect, that the tallest girl is exactly 4' 1", which appears not to be a possible reading. Since it is a stronger claim than (21b), though, it is a bit hard to prove that it's not an alternate reading. (Speakers may be reluctant to call something false when it's true on another reading.) But with the universal example (22a), we don't have this methodological problem. (22c) expresses a weaker claim than (22b). So we can construct scenarios in which c is true while b is false: just imagine that the shortest girl is exactly 4' 1" but some other girls are taller. If c were a possible reading of the English sentence, then speakers should sometimes be willing to judge it true in this situation. But it clearly is false. So (22c) cannot be a grammatical LF. The DegP apparently cannot scope over the quantificational DP. Test cases with less lead to the same conclusion. Let us first make explicit the meaning of /ess-headed DegPs. The obvious rule is (23), which differs from the rule for -er just in the reversal of the >-relation. (23) [/ess than one fooi\= XP- max(P) < 1' Now consider low-DegP and high-DegP derivations for the sentence in (24a).

Degree Operators and Scope

223

(24) a. (John is 4' tall.) Every girl is less tall than that. b. [every girl] ι [less than 4']2 ti is t2 tall Vx[girl(x) -> max{d: tall(x,d)} < 4'] c. [less than 4']2 [every girl], ti is t2 tall max{d: Vx[girl(x) tall(x,d)]} < 4' (24b) transparently captures the intuitive meaning of the sentence. (24c) says that the shortest girl is less than 4' tall, which is a weaker claim than b and clearly not an available reading. Again, we must conclude that DegP doesn't scope over DP. This holds not only for DPs in subject position. (25) has a universally quantified object, but still the DegP seems unable to scope over it. (25) (The frostline is 3 and a half feet deep.) Mary set every post exactly 2 feet deeper than that. If exactly 2' -er than that took scope over every post, the sentence would be true if Mary set just the least deep post exactly 2' below the frostline, but she set the other posts deeper. Intuitively, (25) is false in this case. Even in (26), where the quantified DP is an internal argument and to the right of the graded adjective, the DegP cannot scope over it. (26) (John gave every candidate an A.) ?Mary was less impressed with every candidate than that. This cannot be true just because the candidate who impressed Mary the least didn't impress her enough to get an A. The generalization appears to be the one in (27), which I will refer to as "Kennedy's generalization". (27) If the scope of a quantificational DP contains the trace of a DegP, it also contains that DegP itself.11 The most reasonable diagnosis at this point is that (27) holds generally, not just for the examples in this section, where DegP-scope happens to affect truth-conditions. The This formulation is intended to allow DPs that take scope inside an argument of the graded adjective. In (ia,b), the underlined DPs can have narrower scope than DegP, but since they don't intervene between the DegP and its trace, they are not counterexamples to Kennedy's generalization. (i) a. Every student showed up less often than that. 'it happened less often than that that every student showed up' b. An earthquake is more likely than a snowstorm. The only potentially genuine counterexamples to (27) that I could come up with involve indefinite complements to adjectives of distance, as in (ii). (ii) JafFrey is closer to an airport than it is to a train station.

224

Irene Heim

examples in the previous section, where high DegPs yielded interpretations that were either truth-conditionally equivalent to low DegPs or semantically deviant, did not provide evidence for (27). But they were, of course, consistent with it as well.

2.3

Intensional Verbs and Stateva's Argument

In the last section we only looked at DegP-scope in relation to DP-quantifiers. What about intensional verbs? The logical point is the same: when we add exactlydifferentials or switch from more to less, we ensure non-equivalence between low and high-DegPs. Let's look at some examples and their predicted readings. (28) a. (This draft is 10 pages.) The paper is required to be exactly 5 pages longer than that. b. required [[exactly 5 pp -er than that] the paper be t long] Vw e Acc: max{d: long w (p,d)} = 15pp c. [exactly 5 pp -er than that] [required [the paper be t long]] max{d: Vw e Acc: long w (p,d)} = 15pp (28b) says that the paper is exactly 15pp long in every acceptable world. This implies that it is not allowed to be longer than 15pp. (28c) says that the paper is exactly 15pp long in those acceptable worlds where it is shortest. This leaves open whether it might also be allowed to be longer than 15pp. The English sentence (28a) can be understood in either one of these two ways. It seems to be ambiguous, in just the way that we predict if the DegP can move either below or above the necessity operator. The same thing happens with a possibility operator. (29) a. (...10 pp.) The paper is allowed to be exactly 5 pages longer than that. b. 3w e Acc: max{d: long w (p,d)} = 15pp c. max{d: 3w e Acc: long w (p,d)} = 15pp (29b) says that the paper is exactly 15pp in some acceptable worlds, leaving it open that it might also be allowed to be other lengths. (29c) says it is exactly 15pp long in the acceptable worlds where it is longest, which means it is not allowed to be longer than 15pp. Again, these two distinct readings are both available for the English sentence (29a). Examples of intensional verbs with /ess-comparatives make the same point, as previously argued by Stateva (2000).12

Stateva talks about superlatives with least, but the point is the same.

Degree Operators and Scope

225

(30) a. The paper is required to be less long than that. b. required [[less than that] the paper be t long] Vw e Acc: max{d: long w (p,d)} < lOpp c. [less than that] [required [the paper be t long]] max{d: Vw e Acc: longw(p,d)} < lOpp (30b) says that the paper is less than lOpp in every acceptable world; so it's not allowed to be longer. But c merely says that it is less than lOpp in the acceptable worlds where it's shortest, c is equivalent to "the paper is not required to be as long as that". Both of these clearly distinct readings are available. Similarly for the possibility sentence: (31) a. (... 10 pp.) The paper is allowed to be less long than that. b. 3w e Acc: max{d: long w (p,d)} < lOpp c. max{d: 3w e Acc: long w (p,d)} < lOpp (31 c), which means that the paper is not allowed to be as long as 1 Opp, is a possible reading for (31 a). The availability of the c-readings in these examples indicates that DegP is not, after all, always confined to its narrowest possible scope. Intervening intensional verbs seem to differ from intervening quantificational DPs in this respect, and Kennedy's generalization does not carry over to them. Or is there a different explanation for these ambiguities? These intensional verbs are known to participate in another strange phenomenon, the so-called scope-splitting of negative and other non-monotone-increasing quantifiers.13 For example, No deposit is required means that it is not necessary to make a deposit, and At most three attempts are allowed means that it is not permitted to make more than three attempts. Maybe the c-readings of our comparative examples above are obtained not by DegP-movement, but by the same mechanism that splits scope in these cases. But what is that mechanism? Scope-splitting phenomena are not well understood and their analysis is controversial. One approach involves decomposition into a wide-scope negation and a narrow-scope residue, e.g., no deposit = NEG + a deposit, at most three attempts = NEG + more than three attempts, exactly 5pp -er than that = NEG + more than 5pp -er than that, less than that = NEG + as as that. But these decompositions are not all morphologically transparent (e.g., the last one involves a switch from the comparative to its dual, the equative). Another idea is to split off different wide-scope pieces in different cases, negation in some, focus-sensitive adverbs like only, exactly in others. But this would not seem to cover the less-cases. What I would like to suggest instead is that scope-splitting (at least sometimes) is DegP-movement. So I agree that the two phenomena are the same, but I question that there is a good analysis of scope-splitting

See, e.g., Jacobs (1980), Rullmann (1995a), Larson et al. (1997), and de Swart (2000).

226

Irene Heim

that is a genuine alternative to DegP-scoping. Much more work is required to substantiate this suggestion. But assuming that I am on the right track with this, I tentatively conclude (with Stateva) that DegPs are able to scope over intensional verbs. Does this hold for all intensional verbs? A wider range of examples reveals a mixed pattern. A few more verbs act just like require and allow, e.g., need and be able. (32a,b) are ambiguous in the same way as our examples above, and in particular permit the indicated high-DegP readings. (32) a. John is able to run less fast than that. 'he is not able to run as fast as that' b. The paper needs to be exactly 5 pp longer than that. 'the paper's required minimum length is exactly 5pp longer than that' But the verbs below behave differently in that the high-DegP readings indicated are not attested. (33) The paper might be less long than that, ""it's not possible for it to be as long as that' (34) The paper should be less long than that, ••"it's not required for it to be as long as that' (35) The paper is supposed to be less long than that. *'it's not required for it to be as long as that' (36) I want the paper to be less long than that. *'I don't require it to be as long as that' Why the difference? There are two imaginable reasons. Either the intensional verbs in (33-36) just don't allow the DegP to move over them. Or there is something in their semantics that masks the truth-conditional effect of this movement. The second kind of explanation would be more interesting. In (33), we have an epistemic modal, and maybe the absence of the high-DegP reading can be related to a general resistance of epistemic operators against anything scoping over them.14 In (34-36), all the verbs are so-called neg-raising verbs: their outer negations are systematically understood as inner negations. Perhaps this could have something to do with why the relevant readings fail to surface, but we would need a clearer understanding of neg-raising to substantiate this.15 At this point, I am unable to spell out any concrete explanations for the unambiguity of (33-36), and it is only a hope that it will follow without specific stipulations about DegP-movement.

See von Fintel & Iatridou (2001). In the SALT-version of this paper, I made an attempt at this , but Roger Schwarzschild showed me that it didn't work.

Degree Operators and Scope

2.4

227

DegP-Scope and De Re/De Dicto Ambiguity

Some of the older literature on comparatives took for granted that the famous Russell ambiguity in (37) was a matter of DegP-scope. (37) John thinks the yacht is longer than it is. When the DegP scopes below the verb think, it was assumed, we get the contradictorythought reading, and the sensible reading is due to the DegP scoping above think. Von Stechow (1984) was the first to see clearly that this was a mistake. Analysing examples like (38), he showed that a high-DegP LF would not only be an island violation, but also misrepresent the truth-conditions. (38) If Mary smoked less than she does, she would be healthier. We conclude with von Stechow that the Russell-ambiguity is a matter of de re or de dicto interpretation of the ί/iaw-clause. There is merely a one-directional connection with DegP-scope: a de dicto interpretation of the /Aan-clause is possible only if the DegP is in the scope of the relevant intensional verb, but a de re interpretation is compatible with either wide or narrow DegP-scope. To spell this out a bit more, I need to fill in a treatment of clausal complements to than (which I have avoided in the examples so far). Following standard practice, I take ί/jaw-clauses to be derived by wh-movement of a covert operator from the degreeargument position of an adjective. The trace is interpreted as a variable over degrees. The wh-clause as a whole may be treated as a definite description of a maximal degree (von Stechow 1984). (39) flwA/ the bed is t, long]|= max{d: long(the bed,d)} (39) in conjunction with our meaning for -er predicts that the sentence John is taller than the bed is long is true iff the (maximal) degree to which John is tall exceeds the (maximal) degree to which the bed is long. As for the de relde dicto distinction, I assume that LF-representations contain explicit world-arguments for each predicate, and that the world-argument of the restrictor of a quantifier may, but need not, be bound locally. So even if we only consider narrow scope for the DegP, sentence (37) has two LFs, which differ in the choice of world-variable in the //jaw-clause.16

16

The free variable w here stands for the utterance world (the evaluation world for the whole sentence). The idea that quantifiers can have narrow scope even when their restrictors are interpreted de re is widely accepted nowadays, not just for comparatives but (especially) for DP-quantifiers. Abusch (1994), for example, makes this point about an example like If every senator were a rancher instead, I'd be happy (which is transparently parallel to von Stechow's (38)). See also Percus (2000).

228

Irene Heim

(40) a. John thinks w Xw'[[-er than wh it is t longw>] the yacht is t longw>] Vw' e Acc(w): max{d: long w '(y,d)} > max{d: long w '(y,d)} b. John thinks w Xw'[[-er than wh it is t long w ] the yacht is t longw>] Vw' e Acc(w): max{d: long w '(y,d)} > max{d: long w (y,d)} (40a) expresses the contradictory de dicto reading and (40b) the sensible de re reading. As we see here, both readings are compatible with narrow DegP-scope, and therefore, Russell ambiguities cannot be used as evidence for scopal mobility of DegPs.17 But notice that the ambiguities presented in the previous section were not Russell ambiguities. I have deliberately been using phrases after than that rigidly pick out degrees, so the issue of de re vs. de dicto didn't arise in the first place. Even though they are restricted to intensional contexts, the ambiguities we saw with exactlydifferentials and /ess-comparatives cannot be explained away as involving de re without wide DegP-scope. We can, of course, contrive examples in which there are both DegP-scope and de relde dicto ambiguities. (41) The box is required to be less wide than it is tall. We expect three possible readings, and this, I think, is what we find. (41) can mean that, in order to satisfy the requirements, the box must be taller than wide. This reading is generated by leaving the DegP low and interpreting the than-clause de dicto. (41) can also mean that the box should be less wide than the height that it actually happens to be. On this reading, the requirements don't regulate the box's proportions, only its width. We obtain this interpretation by means of a low-DegP and a de re than-c\&use. Finally, (41) can mean that the box isn't required to be as wide as its actual height. This reading has a high-DegP and (therefore) a de re i/iaw-clause. There can be no such thing as a high-DegP, de dicto iAa«-clause reading. In this respect, (41) contrasts with (42). (42) The box is not required to be as wide as it is tall. Recall that when the than-phrase is rigid, the high-DegP reading of less coincides in meaning with the negation of the corresponding equative. Given a de re reading of the than or as-clause, the high-DegP reading of (41) is equivalent to (42). But (42) also

In later work (1993, 1998), von Stechow reversed his earlier position and decided that (40b) was not, after all, sufficient to capture the sensible reading of (37). The problem he saw is that (40b) has a comparison in the complement of think, but intuitively the thought reported by (37) need not have a comparative content. For example, John might just be thinking "the yacht is 14' long" (when in fact it is 13' long). I don't see the force of this objection (at least not given the semantics I am assuming here). (40b) just says that in every doxastic alternative of John's, the yacht is longer than it is in the actual world. This is true if he thinks it is 14' long. Despite its form, (40b) does not actually say he has a comparative thought.

Degree Operators and Scope

229

allows a reading not shared by (41), on which the as-clause is de dicto: (42) can mean that it isn't required for the box to be as wide as tall (i.e., that it is allowed to be taller than wide). These judgements may be too tenuous to build strong conclusions on. But if they are real, then we may also have an argument here against at least one version of a scope-splitting analysis, which would say that less is decomposed into negation and equative, and only the negation can outscope the intensional verb.

3.

DegP-Scope and Syntactic and Semantic Ellipsis

3.1

Syntactic Ellipsis: VP-Deletion

A standard argument for QR (covert movement of DPs) is that it is needed to license ellipsis in examples which on the surface suffer from antecedent containment (henceforth AC). For example, the deleted VP in (43a) seems to have an antecedent (the matrix VP) which contains it and therefore is not identical to it.18 But after QR has applied, as in (43b), AC is resolved and the two VPs match perfectly. (43) a. I read every book wh that you did read t. b. [every book wh that you did read t] I read t A completely analogous argument can be put forward for covert DegP-movement. In the SS of (44a), the deleted VP is contained in its antecedent, the matrix VP. If covert DegP-movement applies as in (44b), the result is a structure with no AC and properly matched VPs. (44) a. Mary ran faster than wh John did run t fast. b. [-er than wh John did run t fast] Mary ran t fast The idea that DegP-movement is involved in licensing ellipsis in comparative clauses has a long tradition. 19 One recent argument in its favour was presented by Wold (1995). Carlson (1975) had observed that antecedent-contained deletion (ACD) in relative clauses is degraded when the head of the relative clause is a weak indefinite, especially an existentially read bare plural.20

18 19 20

I use strike-out to represent an elided VP with a certain intended interpretation. See Bresnan (1973) and Sag (1976), among others. There is some doubt about the correctness of Carlson's descriptive generalization. According to Pesetsky (pc), at least some examples of this sort improve with the addition of too in the elliptical clause. This may undermine Wold's argument.

230

Irene Heim

(45) *John was climbing trees that Bill was. Diesing (1992) explained this by her thesis that existential bare plurals do not QR, but remain inside VP to get bound by existential closure. Thus they cannot get out of the VP to resolve AC, and ellipsis cannot be licensed. But (as also noted by Carlson) existential bare plurals do allow ACD in comparative clauses. (46) a. John was climbing higher trees than Bill was. b. John was climbing more trees than Bill was. Diesing left this as a potential challenge for her account. Wold showed that there is no problem if AC can also be resolved by DegP-movement instead of QR. (47) [-er than wh Bill was climbing t high trees] John was climbing t high trees The bare plurals in (46) can stay in VP, low enough to get caught by existential closure, while DegP moves high enough to license the ellipsis. In the first part of this paper, we found some semantic and syntactic constraints limiting where DegPs can move. If DegP-movement is indeed a prerequisite for ellipsis licensing, these constraints should affect the range of possible ellipses and their interpretations. Let us see whether this is true. One thing that we saw above was that DegP-movement cannot cross over a quantificational DP (Kennedy's generalization). So what will happen if we construct an example in which, on the one hand, DegP must move out of VP to resolve AC, but, on the other hand, there is an intervening DP-quantifier in that VP? (48) is a case in point. (48) John pushed every weight higher than Mary did. What we expect is that the only grammatical derivation will be one in which the DP every weight first moves out of the VP, so that then the DegP can also move out of VP without crossing over it. (It will only cross over its trace, which is fine.) So we get the LF in (49). (49) every weight] [-er than wh Mary did push t+1 high]2 John pushed ti t2 high (49) says that for every weight x, John pushed χ higher than Mary pushed x. This is what the English sentence in (48) means, and indeed the only reading that it appears to have. The general prediction is that quantificational DPs must scope "out of the way" whenever they are in the path of a DegP containing an antecedent-contained elided VP. This has intricate empirical implications, which for the most part remain to be tested. For example, we predict an interference with sloppy-identity construals of pronouns in examples like (50). (50) I planted one of my trees deeper than you did.

Degree Operators and Scope

231

If we are on the right track, (50) should only allow a pragmatically odd reading: I planted one of my trees deeper than you planted that same tree. It is not obvious that this is correct, but also not easy to say what other readings are possible. More work is called for here.21 We also observed that DegP doesn't scope over negation and other monotone decreasing items, since the resulting readings would systematically be presupposition failures. If DegP-movement is needed to resolve ACD, it follows that (51) is ungrammaticai on any reading that takes the matrix VP headed by refuse to be the antecedent of the elided VP. (51) * John refused to run faster than Mary did refuse to run t fast. This is correct. (The only good reading for this surface string is that John refused to run faster than Mary ran.)2 However, it turns out that we would be making this correct prediction regardless of whether DegP-movement plays any role in ellipsis licensing. The reading indicated in (51) also involves a presupposition failure within the thanclause: there is no maximal degree d such that Mary refused to run d-fast. This in itself rules out the intended construal of the elided VP, independently of any assumptions about how ellipsis is licensed. So the judgement in (51) is merely consistent with the hypothesis that DegP-movement is needed for ellipsis licensing; it does not help support it. With those intensional verbs that we saw a DegP can cross, we get ambiguity of antecedent-size, as expected. (52) is from Williams (1974). (52) a. My father tells me to work harder than my boss does. b. My father tells me to work harder than my boss does work t hard. c. My f. tells me to work harder than my b. does tell me to work t hard.

To make sense of the facts in this area, it will be necessary, for one thing, to take a systematic look at the behaviour of quantifiers inside the /Aa/z-clause. There are well-known puzzles here, and a recent paper by Schwarzschild & Wilkinson compellingly argues that they can only be solved if we change some of the basic assumptions that I have unquestioningly taken over from the literature. Obviously, I need to make my reasoning consistent with their insights somehow, and before I manage to do this, there is reasonable doubt as to whether this paper is even asking the right questions. This is a reading where the elided VP is headed by run and is interpreted de re. That is, we are saying about Mary's actual speed that John refused to surpass it. It may be marginally possible also to get a de dicto reading, meaning that John refused to outrun Mary. That reading is easier if the auxiliary is changed from did to would, or (even better) if we switch to the Comparative Ellipsis variant with no auxiliary at all.

232

Irene Heim

(53) a. John needs to drive faster than Mary does. b. John needs to drive faster than Mary does drive t fast. c. John needs to drive faster than Mary does need to drive t fast. To bring out each of the two readings of (53), imagine two contexts. In the context for reading b, we are talking about what it takes for John to win the race. Here there is no absolute minimal requirement for the speed he needs to attain; what matters is that his speed exceed Mary's. In the context for reading c, we are discussing John's and Mary's chances of making it to Ithaca by 8 pm tonight (at the latest). Since John is currently 300 miles away from Ithaca, whereas Mary is only 200 miles away, the minimal speed for him to make it exceeds the minimal speed for her to make it. (Notice that there is no requirement here on their relative speeds.) Interestingly, this ambiguity of antecedent-choice carries over to examples with verbs which we found not to show truth-conditionally distinct high-DegP readings, like the neg-raising verb want. (54) is the variant of Williams' example analysed in Sag (1976). (54) a. Mary's father wants her to work harder than her boss does. b. Mary's father wants her to work harder than her boss does work t hard. c. M's f. wants her to work harder than her b. does want her to work t hard. This indicates that DegP can, after all, scope over these verbs, and that the reason why we didn't see truth-conditional reflexes of this scoping (when looking at /ess-DegPs) must be in their semantics. So we really need to identify the link between neg-raising behaviour and lack of DegP-scope ambiguity that I hypothesized at the end of section 2.3. Williams and Sag used these examples to point out a correlation between ellipsis and availability of de dicto readings. The non-elliptical (or less elliptical) (55) allows a de dicto reading of the than-clause that is not shared by its elliptical counterpart (52c). (55) My father tells me to work harder than my boss tells me to. (55) can describe a state of affairs where my father tells me: "I don't care how many hours you work, but make sure it's more than whatever your boss demands. If he says you have to work 20 hours, work at least 21; if he tells you to work 30, work 31; etc." (52a) cannot describe this scenario on any of its readings. The only reading it shares with (55) is a de re reading, where the embedded te/Z-clause is evaluated in the utterance world, not in the worlds conforming to what my father demands. That is, (52c) has to mean that my boss imposes a certain minimum requirement for how hard I have to work, and my father imposes another one which is higher than that. Williams and Sag took this fact to support the hypothesis that ellipsis licensing depends on DegP-scope. They reasoned on the basis of a naive identifcation of de re and de dicto

Degree Operators and Scope

233

with wide and narrow scope, respectively. As we saw in 2.4 above, this is not justified; there is only a one-directional implication, namely that if the DegP's scope is wide, the than-clause must be read de re. But fortunately, it is only this direction of the connection that is required for their argument. If the DegP must scope over the matrix verb in order to license the broad construal of the ellipsis, then the than-clause gets carried along out of the matrix verb's scope and therefore can only be read de re. This is just what was observed. We can confirm the correlation between ellipsis-size and DegP-scope even more directly by changing Williams' kind of example to a variant with less. (56) Mary needs to drive less fast than John does need to drive t fast. When the elided VP is understood to be headed by need (as indicated), the DegP must scope over need, and the truth-conditions therefore are those of a negated equative ("she doesn't need to drive as fast as he needs to drive"). This prediction is also correct. What have we learned in this section? There is more fieldwork to be done in the area where ACD in comparative clauses interacts with Kennedy's generalization. Apart from this, it appears that our previous conclusions about DegP-movement are at least consistent with the view that ellipsis licensing relies on DegP-movement. Our last set of facts, about DegP-scope and ellipsis-size in examples with intensional verbs, shows moreover that this combination of assumptions makes some interesting correct predictions. It is difficult, however, to draw stronger conclusions at this point. We have not given any consideration to alternative hypotheses, which might account for ellipsis licensing without appeal to DegP-movement. Before we examine concrete versions of such alternative theories, we cannot know whether they will have any difficulty making the same correct predictions also. What such alternatives might there be? That depends, for one thing, on our background assumptions about VP-ellipsis in general. We have tacitly taken for granted here that an elided VP and its antecedent must have LFs that are identical up to possible differences in indices.23 If we consider the possibility that some less restrictive parallelism condition might be sufficient, there may be ways to establish the required parallelism without DegP-movement.24 Even if we stick to the strict (near-)identity

24

See Rooth (1992) and work based on this for more detail on the assumptions about ellipsis licensing that underlie this requirement. For example, it may suffice to move just the f/iaw-clause or the wh-clause after than: (i) [wh John did run t fast] Mary ran [-er than t] fast The VPs are still not identical here, but AC is resolved. And as Kennedy (pc) points out, the existential closures of the two VPs are semantically equivalent ^ x . 3 d [x ran d-fast] = λχ.Ξά [χ ran faster than d]), which suffices for ellipsis licensing according to a recent proposal by Merchant (1999). But this idea runs into a problem with differential phrases, as in Mary ran 5 km/h faster than John did, where the analogous equivalence fails. (Xx.3d[x ran dfast] Φ λχ.Ξά [χ ran 5 km/h faster than d].)

Irene Heim

234

condition we have been assuming, it is not obvious that DegP-movement is the only way to resolve AC. What about moving the entire AP, as in (57)? (57) [[-er than wh John did fun-ί] fast] Mary ran t For this to be a viable alternative, we would need appropriate interpretive mechanisms to deal with the AP-sized traces and their binding in this structure. It is not at all obvious what exactly these should be, how they would affect the reasoning in section 2, and whether this path in the end would lead to something substantively different from DegP-movement. 25 It is incumbent upon me to deal with these questions, but not in the space of this paper.

3.2

Semantic "Ellipsis": Superlatives and Too

As we just saw, it is difficult to make a strong argument for DegP-movement on the basis of facts about VP-ellipsis, because one needs so many auxiliary assumptions. Part of the difficulty is that it is controversial just how this kind of ellipsis is licensed and how close the match between the missing VP and its antecedent has to be. To side-step this controversy, it would be useful if we could look at cases of what I will call here "semantic ellipsis". By this I mean a construction in which there aren't two parallel pieces of syntactic (LF) structure which stand in an anaphor-antecedent relation, but rather there is a single piece which, however, is used twice in the semantic calculation. Analyses of this kind have often been proposed as competitors for analyses in terms of syntactic ellipsis, notably for so-called Comparative Ellipsis,26 but there we would be getting onto even more controversial ground. The constructions I will look at here are ones for which (to my knowledge, at least) a semantic-ellipsis analysis is the only kind of analysis on the market. Consider the superlative. (58) John screamed (the) loudest. (58) can be paraphrased by a comparative: John screamed louder than anyone else did. The LF of the comparative paraphrase contains two degree predicates, one in the

One proposal for interpreting structures like (57) is worked out in Kennedy (1997). I will discuss this on another occasion. See, e.g., the "direct analysis" of phrasal comparatives in Heim (1985), and similar analyses for Comparative Ellipsis as well as related constructions like Bare Argument Ellipsis and Gapping that are defended by Reinhart & Rooth (1986), Reinhart (1989), Kennedy (1997), and others. See also Lechner (1999) for critical discussion.

Degree Operators and Scope

235

matrix clause and one in the than-clause. The superlative has only a matrix clause and still manages to convey the same meaning. The following analysis reveals how.27 (59) a. [[-es*]]= XR^x. max{d: R(x,d)} > max{d: 3y / x: R(y,d)} b. LF for (58): John -esti [[scream t| loud]] Notice that R (the denotation of -esf s complement) is "used twice" in the calculation prescribed by entry (59a). We use it to determine one set of degrees for John, and another one for everybody else. Recall now our example (53) (repeated here as (60a)) and compare it with the analogous superlative in (60b). (60) a. John needs to drive faster than Mary does, b. John needs to drive fastest. The two readings we observed in (60a) are replicated in (60b). To bring out each reading, just amend the scenarios that I gave above by the addition of a third person, say Bill. Then (60b) might mean that John needs to outdo both Mary and Bill (say, in order to win the race). Or it might mean that the minimal speed that John needs (to get to Ithaca by 8 pm) is higher than the minimal speeds needed by Mary or Bill (since they are already nearer to Ithaca). In the comparative example (60a), the ambiguity clearly involves different-sized choices of antecedent for the VP-ellipsis. On the view that ellipsis licensing requires LF-identity - but not independently of this view - this implies that the two readings also differ in the scope of the DegP. In the superlative variant (60b), there is no ellipsis to resolve. So how do we generate the analogous two readings? The obvious way - and here the only one available to us - is by scoping the DegP. As the reader can verify, if we scope -est below need, as in (61a), the entry in (59a) gives us the first reading, and if we scope it above, as in (61b), it gives us the second one. (61) a. John] needs [ti -est2 to drive t2 fast] b. John -est2 λι [needs [tj to drive t2 fast]] Our analysis of the ambiguity in the superlative examples is parallel to our analysis for the comparative variant: in both cases, we invoke DegP-movement across the intensional verb to obtain the second reading. So what is new here, what do we learn from the superlative example that we haven't already learned from the comparative one? The new point is that the evidence for DegP-movement needn't be contingent on assumptions about ellipsis licensing. Whether or not you accept the relatively strict Heim (1985). In Heim (1999), I propose a variant that makes the semantic type the same as for the comparative and equative DegPs (by eliminating the external argument). This amendment would not affect the present reasoning.

236

Irene Heim

parallelism conditions on VP-ellipsis that imply that DegP must be able to move above need in (60a), you can't get around this movement if you want to account for all the readings of (60b). For the same methodological reason, it may be useful to look at gradation constructions with so ... that, too, and enough?* These also seem to involve "semantic ellipsis", in the sense that the material in the scope of the degree-operator is used twice in the semantic calculation. For example, too tall roughly means "taller than is compatible with certain (contextually given) goals or desires". If John is too tall, he is taller than it is acceptable for him to be tall. Notice how the clause "John (be) tall" appears twice in this paraphrase, once unmodalized to the left of than, and another time under a possibility operator to its right. The following entry for too makes this transparent in the double occurrence of P.29 (62) |[ioo]|w = XP· max(P(w)) > max{d: 3w'e Acc(w): P(w')(d) = 1} Consider now an example like (63). (63) John needs (to have) too much money. (63) has one reading where it means that what John needs is to have too much money. Imagine he is the type of person who only feels secure when he has more money than he can spend. Unless he has too much money (for him to spend), his (emotional) needs are not satisfied. But (63) also has another reading, where it means that John's financial needs are too high. Imagine that he has many financial obligations (mortgage, alimony, gambling debts, ...). He considers taking a job as a linguistics professor, but thinks better of it, since his needs regarding salary are too high. He needs too much money for him to be a linguist. These two readings differ in the contextually supplied accessibility relation for the implicit possibility operator associated with too, but that is not their only difference. They also differ in the scope of too with respect to need. For the first reading, the LF in (64a) is adequate, but for the second one, it must be as in (64b). (64) a. Johni needs [too2 [ti to have t2 much money]] Vw G Accneed'· max{d: j has d-much money in w} > max{d: 3w' e Acc/ 00 (w): j has d-much money in w'}

See von Stechow (1984) for a treatment of too, and Meier (2000) for a detailed recent investigation and semantic analysis of all three constructions. The same point still holds when we consider examples where too selects aybr-infinitival. The for-clause supplies an explicit restrictor for the modal operator introduced by too, but it does nothing to spell out its nuclear scope. So we still need to use DegP's sister twice, as shown in the following rule for too + (for-)CP.

(i)

Itoo cpf = λΡ. max(P(w)) > max{d: 3w'e Acc(w): [CP^' = 1 & P(w')(d) = 1}

Degree Operators and Scope

237

b. too2 [Johni needs [ti to have t2 much money]] max{d: Vw e Kcc n e e ( j: \ has d-much money in w} > max{d: 3w e Acctoo'· Vw' e AccMeed(w): j has d-much money in w'} Accneed is meant to pick out the worlds in which John's needs are met. Acc/ 0 0 is different for each reading. For the intended interpretation of (64a), let's say it maps each w to the set of worlds which are like w in the respects that determine spending limits and in which John spends all his money. So (64a) says that John's needs are met only if he has more money than for him to have is compatible with him spending it all. For the intended interpretation of (64b), let Acc too m a P each w to the set of worlds in which people obey the laws in effect in w (without starving) and in which John is a linguist. Then (64b) says that John's financial needs are greater than for them to be is compatible with him being a linguist. The important point for our purposes is that what is incompatible with being a linguist is not having lots of money, but needing lots of money. This is why we need to scope too over need in order to capture this second reading properly. Due to the multiple layers of modality and concomitant context-dependency, the analysis of this type of example is rather complex. Still, I think we can isolate the contribution that is made by the degree operator's scope in relation to the intensional verb. If so, then we have another way here of detecting DegP-scope without taking a stand on licensing-principles for syntactic ellipsis.

4.

Conclusion

We have probed the scopal behaviour of degree quantifiers, using direct evidence from truth-conditional ambiguity as well as more indirect evidence from the interpretation of elided material. The argumentation was incomplete and tentative in a number of places, but suppose the picture that began to emerge is right. Then DegPs are mobile, but they are not allowed to cross over quantificational DPs. This appears to be a syntactic constraint of some kind, because the prohibited scopings would not in any way be semantically deviant. And if our treatment of ellipsis licensing is right, it is not a constraint on the absolute length of movement, since non-quantificational DPs in the same domains (including QR-traces of quantifiers) can be crossed over. Rather it is a constraint that specifically targets interveners of a certain kind. The situation is thus highly reminiscent of the intervention effects in wh-questions studied by Beck (1996), although the precise connections remain to be explored. More data need to be investigated to decide whether the parallel really holds up. And most importantly, we need to work out a syntactic analysis in which the observed constraint has principled reasons.

238

Irene Heim

Endnotes This paper appears in almost identical form in B. Jackson & T. Matthews (eds.) Proceedings of SALT 10, CLC Publications, Cornell University. I had hoped to revise it substantially, but have not been able to do so in the available time. For help with the preparation of the SALT-version, I would like to thank Sigrid Beck, Rajesh Bhatt, Danny Fox, Martin Hackl, Roumi Izvorski, Chris Kennedy, Lisa Matthewson, Cecile Meier, Yael Sharvit, Penka Stateva, Arnim von Stechow, Zoltän Szabo, and Karina Wilkinson. More recently, I received very detailed and thoughtful written comments from Cecile Meier, Roger Schwarzschild, and an anonymous reviewer. They have convinced me that the present version suffers from many inadequacies, not just in matters of exposition, but also with regard to the descriptive claims, analyses, and argumentation. I have every intention of following up on what I have learned from their criticism, and I apologize to them and to Arnim that this is not reflected in my contribution to this volume.

References Abney, Steven (1987). The English NP in its Sentential Aspect. PhD dissertation, MIT. Abusch, Dorit (1994). The Scope of Indefinites. Natural Language Semantics 2. 83-135. Beck, Sigrid (1996). Quantified Structures as Barriers for LF Movement. Natural Language Semantics A. 1-56. Bierwisch, Manfred (1987). Semantik der Graduierung. In: Manfred Bierwisch & Ewald Lang (eds.) Grammatische und konzeptuelle Aspekte von DimensionsAdjektiven. (Studia Grammatica 16.) Berlin: Akademie-Verlag. 91-286. Bresnan, Joan (1973). Syntax of the Comparative Clause Construction in English. Linguistic Inquiry 4. 275-343. Carlson, Gregory (1975). Amount Relatives. Language 53. 520-542. Diesing, Molly (1992). Indefinites. Cambridge, MA: MIT Press. Fintel, Kai von & Sabine Iatridou (2001). On the Interaction of Modals, Quantifiers, and If-Clauses. Ms, MIT. Fox, Danny & Jon Nissenbaum (2000). Extraposition and Covert Movement. Ms, Harvard University & MIT. Gueron, Jacqueline & Robert May (1984). Extraposition and Logical Form. Linguistic Inquiry 15. 1-31. Heim, Irene (1985). Notes on Comparatives and Related Matters. Ms, University of Texas, Austin. Heim, Irene (1999). Notes on Superlatives. Ms, MIT. Jacobs, Joachim (1980). Lexical Decomposition in Montague Grammar. Theoretical Linguistics 7. 121-136.

Degree Operators and Scope

239

Kennedy, Christopher (1997). Projecting the Adjective. PhD dissertation, University of California, Santa Cruz [also Garland Press 1999]. Larson, Richard, Marcel den Dikken & Peter Ludlow (1997). Intensional Transitive Verbs and Abstract Clausal Complementation. Ms, SUNY Stonybrook. Lechner, Winfried (1999). Comparatives and DP-Structure. PhD dissertation, University of Massachusetts, Amherst. Meier, Cecile (2000). Konsekutive Konstruktionen und relative Modalität. PhD dissertation, University of Tübingen. Merchant, Jason (1999). The Syntax of Silence. PhD dissertation, University of California, Santa Cruz. Percus, Orin (2000). Constraints on Some Other Variables in Syntax. Natural Language Semantics 8. 173-229. Reinhart, Tanya (1989). Elliptic Conjunctions - Non-Quantificational LF. In: Asa Kasher (ed.) The Chomskyan Turn. Oxford: Blackwell. 360-384. Reinhart, Tanya & Mats Rooth (1986). Taking Tiger Mountain by Strategy. Ms, Tel Aviv University & University of Stuttgart. Rooth, Mats (1992). Ellipsis Redundancy and Reduction Redundancy. In: Stephen Berman & Arild Hestvik (eds.) Proceedings of the Stuttgart Ellipsis Workshop. University of Stuttgart. 1-26. Rullmann, Hotze (1995a). Geen eenheid. Tabu 25. 195-197. Rullmann, Hotze (1995b). Maximality in the Semantics of WH-Constructions. PhD dissertation, University of Massachusetts, Amherst. Sag, Ivan (1976). Deletion and Logical Form. PhD dissertation, MIT. Schwarzschild, Roger & Karina Wilkinson (1999). Interval Semantics for Scalar Predication. Ms, Rutgers University. Stateva, Penka (2000). In Defense of the Movement Theory of Superlatives. In: R. Daly & A. Riehl (eds.) Proceeding of ESCOL 1999. (Distributed by Cornell Linguistics Club). 219-226. Stechow, Arnim von (1984). Comparing Semantic Theories of Comparison. Journal of Semantics 3. 1-77. Stechow, Arnim von (1993). Die Aufgaben der Syntax. In: Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld & Theo Vennemann (eds.) Syntax - A Handbook of International Research. Berlin: de Gruyter. 1-88. Stechow, Arnim von (1998). Some Remarks on Ch. Kennedy's Syntax and Semantics of Comparison. Ms, University of Tübingen (presented at Workshop Syntax and Semantics of Comparatives, Berlin). Swart, Henriette de (2000). Scope Ambiguities with Negative Quantifiers. In: Klaus von Heusinger & Urs Egli (eds.) Reference and Anaphoric Relations. Dordrecht: Kluwer. 109-132. Williams, Edwin (1974). Rule Ordering in Syntax. PhD dissertation, MIT. Wold, Dag (1995). Antecedent-Contained Deletion in Comparative Constructions. Ms, MIT.

(A)temporal Complements

Graham Katz

1.

Introduction

Untangling the factors that go into the temporal interpretation of complement clauses has provided work for formal semanticists for a number of years (Ladusaw 1977; Dowty 1982; En? 1987; Ogihara 1989; Abusch 1994; von Stechow 1995b; Kusumoto 1999). Balancing the considerations of semantic innocence - that meanings be the same in root as in embedded contexts - with those of compositionality - that the meaning of the whole should be composed out of the meanings of the parts - has proved a difficult challenge. Much attention has been focussed on embedded finite clauses and the fact that the tense such clauses bear sometimes appears not to be interpreted. This is the so-called "sequence of tense" phenomenon. In this paper, however, I will be concerned with clauses that do not bear tense - infinitival clauses - and the fact that they sometimes appear to be interpreted as if they did. This might be called the "specification of tense" phenomenon. The problem I will address can be presented in puzzle form, and the puzzle is this: Why, if (lb) is an acceptable variant of (la), is (2b) not an acceptable variant of (2a)? (1)

a. Fritz expects that Arnim will like his gift, b. Fritz expects Arnim to like his gift.

(2)

a. Fritz believes that Arnim will like his gift, b. Fritz believes Arnim to like his gift.

(la), (lb) and (2a) can all be used to describe Fritz's attitude toward a potential future situation, while (2b) can only be used to describe his attitude toward a current situation. More generally, verbs such as expect, promise and predict - what I will call "futureoriented" verbs - appear to require that their infinitival complements have a future interpretation, while verbs such as believe, think and understand - "present-oriented" verbs - require that their infinitival complements have a present interpretation. In recent work, Abusch (to appear) has noted three properties which distinguish infinitival complements of future-oriented verbs from their present-oriented counterparts.

(A)temporal Complements

241

The first - illustrated by the contrast in (3) - is that infinitival complements of futureoriented verbs can be non-stative, while infinitival complements of present-oriented verbs must be stative (compare (2b) and (3b)). (3)

a. Fritz expects Arnim to unwrap his gift, b. *Fritz believes Arnim to unwrap his gift.

The second - illustrated by the contrast in (4) - is that infinitival complements of future-oriented verbs can be modified by future time-frame adverbials, while infinitival complements of present-oriented verbs cannot be. (4)

a. Monika expects Arnim to be in Tübingen tomorrow, b. * Monika believes Arnim to be in Tiibingen tomorrow.

And the third - illustrated in (1) - is that infinitival complements of future-oriented verbs are equivalent with future-tensed complements, but infinitival complements of present-oriented verbs are equivalent with present-tensed (or sequence-of-tense past tense) complements. So while (2b) is not equivalent with (2a), it is equivalent with (5). (5)

Fritz believes that Arnim likes his gift.

In short, infinitival complements of future-oriented verbs behave as if they are futuretense clauses, and infinitival complements of present-oriented verbs behave as if they are present-tense clauses. To account for this contrast, Ogihara (1989) - in the first model-theoretic treatment I am aware of - took infinitival clauses to be ambiguous between a "present tense" interpretation and a "future tense" interpretation. The difference between future-oriented verbs and present-oriented verbs, then, is simply that future-oriented verbs select for future-tense complements and present-oriented verbs select for present-tense complements. For future-oriented verbs this account is quite appealing, as, in addition to selecting for a covert "future tense" in infinitival complements, these verbs select for an overt future tense in finite complements. As we see in (6), neither present-tense nor past-tense finite clauses can comfortably appear as complements of expect·} (6)

a. ??Amim expects that Monika is writing a very long dissertation, b. ??Arnim expects that Monika was writing a very long dissertation.

I am aware of dialects of English in which there is a sense of expect meaning 'strongly believe' which can appear with such complements. Insofar as non-future readings of expect's infinitival complements are also available in these dialects, there is nothing of interest to say about them.

242

Graham Katz

For these verbs, then, the interpretation of infinitival complements and the restriction on acceptable finite complements can be attributed to the same general selectional restriction: the complement clause must be future tensed. For present-oriented verbs, however, such an ambiguity-plus-selection account is less attractive. Present-oriented verbs do not select for present-tense finite complement; they take future-tense complements and past-tense complements as well. Thus, the "present tense" interpretation of infinitival complements cannot be derived from general restrictions on the tense of complement clauses and must be independently stipulated. The complement must be "present tensed" if it is infinitival, but there are no restrictions on the tense if it is finite. This is clearly less than desirable. In later work Ogihara (1996) adopted a "lexical" account which avoided this problem. On this account, a form of which Abusch also adopts, infinitival complement clauses are taken to have only a relative-present interpretation, with the lexical semantics of the embedding attitude verb specifying the time with respect to which the complement clause is to be interpreted: Present-oriented verbs specify a present time, and future-oriented verbs specify a future time. In essence, future-oriented verbs are taken to introduce into the semantics of the sentence a covert future-tense operator that "shifts" the evaluation time of the complement clause into the future. In fact, Abusch argues explicitly that - since a sentence of the form χ expects y to VP is true exactly when the corresponding sentence of the form χ believes that y will VP is - expect should be interpreted as believe-will. This lexical-future analysis runs into trouble, however, when it comes to the treatment of finite complements, as these complements aren't future-shifted when, perhaps, they should be. More on this later. The interplay between the temporal interpretation of infinitival complements and the temporal interpretation of finite complements makes the puzzle presented by the data in (1) and (2) more challenging than it might at first seem, particularly if one is looking for a solution that respects compositionality considerations. In this regard, there are four properties that we would hope an ideal solution to our puzzle would exhibit: I.

Infinitival clauses should have a uniform interpretation in all contexts.

II.

Finite clauses should have a uniform interpretation in all contexts.

III. Attitude verbs should have a uniform interpretation in all contexts. IV. Attitude verbs should combine with their complements in a uniform manner. As we have seen, the ambiguity analysis lacks the first property. As we will show later, the lexical analysis lacks at least one of the final two. In what follows I will more thoroughly review the puzzle presented by the contrast in (1) and (2) and the problems that exist for solutions to it of the type Ogihara and Abusch have suggested. I will then sketch an account that avoids these problems and respects the compositionality considerations just mentioned. This will involve adopting a standard Lewis-style de se analysis of attitude verb meanings, generalizing it to the treatment of temporal interpretation, but significantly altering the anaphoric account of tenses that usually accompanies such a treatment. I instead adopt a two-dimensional

(A)temporal Complements

243

relational analysis of tense interpretation, embedded in a "dynamic" framework that allows tenses to act like predicates on the one hand and variables on the other, depending on their context of use. This would appear to be a radical departure from standard treatments of tense semantics, but, in fact, is in keeping with some of the most interesting recent proposals (von Stechow 1995a, Kusumoto 1999).

2.

Tenses and Tensed Complements

The meaning of tensed sentence is usually factored into two components: the tense and the remainder of the sentence. This remainder has been called variously a "tenseless sentence" (Εης 1986), a "sentence radical" (Galton 1984), or a "basic proposition" (Katz 1995). Intuitively, the tenseless remainder is what the sentences (7a-c) all have in common. (7)

a. Ede lived in Germany. b. Ede lives in Germany. c. Ede will live in Germany.

This remainder is invariably interpreted as that which must be combined with a tense to yield a tensed-clause meaning. I take it to be desirable, if not essential, that infinitival clauses be assigned the same types of meanings, since - at least pre-theoretically - this would seem to be what they are: clauses that lack tense. But what is tense? Or, better, how are tenses interpreted? Are they sentential operators which "shift" the evaluation time of a sentence, as the traditional Montagovian analysis would have it (Montague 1974)? Or are they referential elements that fill a verbal argument role, as argued by En? (1986) - perhaps with anaphoric properties (Partee 1973; Partee 1984)? The referential-anaphoric account of tenses has become increasingly popular (En? 1987; Abusch 1994; Heim 1994; Kratzer 1998; Kusumoto 1999). On this account, tenses refer to times, and may be "bound" by other time-denoting elements in the sentence or discourse. For example, the embedded tense in (8) may either be bound by the higher tense, as indicated in (9a), or interpreted freely (potentially to be discoursebound), as in (9b). (8)

Monika believed that Arnim was in Constance.

(9)

a. Monika PAST] believe that Arnim PAST] be in Constance, b. Monika PAST] believe that Arnim PAST2 be in Constance.

The logical form (9a) corresponds to the prominent "same time" reading of (8), on which Monika's belief was that Arnim was in Constance at the time she had her belief,

244

Graham Katz

while (9b) corresponds to the less-prominent "back shifted" reading, on which Monika's belief was that Arnim was in Constance at a time before she had her belief. Tenses bear grammatical features - analogous to gender features associated with thirdperson pronouns - which reflect presuppositional requirements on their reference. "Sequence of tense" is simply taken to be an agreement phenomenon. The pastness of the embedded tense in (9a), for example, simply reflects the binding relation that holds between it and the main clause past tense. It need not indicate semantic anteriority. As Arnim von Stechow (1984, 1995a, 1995b) has been careful to point out, however, it is a fundamental semantic error to bind variables in the complement clause of an attitude verb, as such binding improperly smuggles information into the content of the attitude that should not be there. In the case of temporal information, the problematic consequences of such illicit binding are quite clear. For example, if the coindexation in (9a) is interpreted as true coreference, then the sentence would attribute to Monika the belief (at some past time) that Arnim was in Constance at that time. The problem is that Monika may have had no idea what time it was or may have been mistaken about the time. To take an unlikely context, if Monika's watch was running so slowly that when Arnim called from Constance at noon, she thought it was only eleven in the morning, (10a) might be true, but (10b) false. (10) a. At noon, Monika believed that Arnim was in Constance. b. At noon, Monika believed that Arnim was in Constance at noon. The embedded tense cannot be taken to refer to the same time as the main clause tense in such constructions, or we risk predicting synonymy. But what, then, does it refer to? One way of answering this question is to adopt a temporal analog of Lewis' (1979) treatment of de se attitudes (von Stechow 1984; Ogihara 1989; Abusch 1994; Heim 1994; von Stechow 1995a; von Stechow 1995b). To review briefly, Lewis considers the problem of how to distinguish belief about the facts from indexical belief. In Kaplan's famous sentence (11), for example, we need to distinguish the belief that Jones' pants are on fire from the indexical belief - what Lewis calls the de se belief - that the believer has burning pants. (11) Jones thinks his pants are on fire. If Jones is confused about his own identity, these two beliefs do not coincide. If he thinks that he is Smith, for example, we might want to deny that Jones has the belief that Jones' pants are on fire while affirming his belief that his own pants are on fire. Lewis gives a clever account of the distinction by suggesting that the objects of attitudes are not normal propositions - sets of possible worlds - but rather properties, or "centered" propositions - functions from individuals to sets of possible worlds. The embedded expression his pants are on fire is not interpreted simply as the set of possible worlds in which Jones' pants are on fire, but rather as the function from (male) individuals to the set of worlds in which that individual's pants are on fire. The verb believe, then, is translated as a predicate of type ((e{st))\e(st))) with the following

(A)temporal Complements

245

(atemporal) interpretation (the function Doxatemporai returns, for each individual χ and world w, the set of world-individual pairs that χ entertains in w as possible alternatives for who he might be and what world he might be in, i.e., x's doxastic alternatives in

(12) Hbelieve,tempoi»i]l(w,*,^>)= 1 iffV(w',t',x') = 1.

(A)temporal Complements

247

abstract over the "bound" tense. Doing so yields the structure in (18a), which is interpreted as in (18b).6'7 (18) a. At midnighti Fritz PAST] believe λ 1 [it PAST, be raining] b. V(w',t') e Dox(wo,midnight,Fritz) be-raining(w',t') = 1 (15) is true, then, if at midnight Fritz located himself at a time in a world in which it was raining, that is, if Fritz's doxastic alternatives at midnight were a subset of the set {(wi, 11:30pm), (wi,12:30am), (w 2 ,midnight), (w3,l 1:30pm)}. If, in addition, Fritz knew what time it was, then he located himself in world w2. The de se analysis of the temporal component of attitude reports has become a standard accompaniment to the anaphoric treatment of tense (and vice-verse). The parallel it presents between pronouns on the one hand and tenses on the other is quite appealing, particularly in the context of the interpretation of "sequence of tense" past tenses. As we have seen, the same interpretive mechanisms can be applied to these bound tenses as are applied to bound pronouns in attitude complements (for discussion of these parallels, see Kratzer 1998). For us, however, the real appeal lies in the straightforward way this analysis can be applied to the treatment of infinitival complements of attitude verbs, as they can literally be interpreted as tenseless clauses.

3.

Infinitival Complements

On the referential account of tense, interpreting infinitival clauses as tenseless means interpreting them as intensionalized properties of times. And as such, they are intrinsically of the right semantic type to appear as the clausal complement of believe. The infinitival complement clause in (19), for example, is interpreted as the predicate XtXw.be-raining(w,t), which of course is simply the meaning of the VP be raining. (19) Fritz believes it to be raining. Combining this with our analysis of believe gives us the truth conditions in (20). (20) V(w',t') e Dox(w0,to,Fritz) be-raining(w',t') = 1 In other words, (19) is true if and only if all the world-time pairs in Fritz's doxastic alternatives are worlds and times at which it is raining. In the state of affairs we de-

6

7

It is an open question whether it is possible to have a de re interpretation of bound tense. While I think that it is, we will not consider that question here. The distinguished variables w 0 and to are taken - when free - to refer to the actual world and the speech time, respectively (Heim 1994; Stechow 1995b).

248

Graham Katz

scribed above, (19) would be true if and only if Fritz' doxastic alternatives were a subset of the set {(wi,l 1:30pm), (wi,12:30am), (w2,midnight), (w3,l 1:30pm)}. The belief attributed to Fritz in (19) and that attributed to him in (15), then, are identical. The fact that in one case the complement clause contains a de se-interpreted past tense and in the other case it is an infinitival clause doesn't play a role, since in both cases the temporal argument of the embedded predicate is filled by the temporal parameter of the doxastic alternative pair, i.e., by the believer's "subjective present." The only difference is that in the case of tensed clauses, the tense must be abstracted over before the embedded clause can be combined with the attitude verb. To put it another way, the "same-time" reading is what we call the interpretation on which the subjective present introduced by the attitude verb fills the temporal argument of the embedded verb. Tensed complements only have the same-time reading, of course, when the tense is interpreted de se, i.e., when it is abstracted over, and not when it is free. Infinitival complements, however, should always have a same-time interpretation. Our puzzle, of course, is that they don't. In contrast to (19), for example, (21) has a future shifted interpretation. (21) Fritz expects it to be raining. (21) is true if Fritz locates himself at a time in a world at which it will be raining later. How, then, do we explain the fact infinitival complements of future-oriented verbs such as expect have such a future-shifted interpretation? Ogihara (1989) suggested that the difference between (19) and (21) lay in the meaning of the complement clause, and, in particular, in the meaning of the infinitive marker to. In (21) we have tOfU,r - a future tense operator - while in (19) we had topres a present tense operator. In the introduction I pointed out one of the undesirable consequences of such an account, namely that we would need to stipulate that tOfU,r can never appear in the complement of present-oriented verbs such as believe. Since in principle believe can take future-tensed complements, there is no other way to prevent believe's infinitival complements from getting a future-shifted reading. A further reason to doubt that the temporal interpretation of infinitival complements is tied to the interpretation of the infinitival marker to is that the same present/future contrast in interpretation exists for ηοη-ίο-marked complements: (22) a. Francis believes them lost, b. Francis expects them late. Clearly in (22a) we have a present interpretation, while in (22b) we have a future interpretation. Whatever is responsible for the present/future contrast, then, must be unrelated to the infinitival marker. Accordingly, following Ogihara (1996) and Abusch (to appear), we will provide infinitival clauses themselves with a univocal meaning and derive the differences in complement-clause interpretation from the lexical semantics of the embedding attitude verbs. In particular, we will take expect and other future-oriented verbs to introduce

(A)temporal Complements

249

lexically a time-shifting future-tense-like operator into the semantics of the sentence. We (provisionally) translate expect, then, as a predicate of type {{i{st)){e{i{st)))) - the ο same type as believe - but with the following interpretation: (23) UexpectprovKxU*,/') = 1 iff V t' & P(w',t") = 1] Whereas to believe Ρ is to locate oneself at a world and time at which Ρ holds, to expect Ρ is to locate oneself at a world and a time at which Ρ will come to hold. For expect the complement clause is applied to a time that the attitude holder takes to be in his future, not, as in the case of believe, to a time he takes to be his present. The truth conditions for (21) are, then, as follows: (24) V(w',t') e Dox(w0,to,Fritz) 3t" [t" > t' & be-raining(w',t") = 1] In other words, (21) is true if and only if in each of his belief worlds Fritz locates himself at a time before a time at which it is raining. In the state of affairs described above (21) would be true if and only if Fritz's doxastic alternatives were a subset of the set {(wi,l 1:30pm), (wj,midnight), (w2,11:30pm)}. The advantages of this type of analysis are clear. It confines the difference between present-oriented verbs and future-oriented verbs to the lexical semantics of the verbs themselves, it involves no stipulated ambiguity, and, at least to a first approximation, it captures the intuitive relationship between believe and expect·. To expect something is simply to believe that it will happen. Such an analysis runs into trouble, however, when it is applied to tensed complements. As is well known, a present tense clause in the scope of a future tense can have a future-shifted reading, as in (25). (25) Arnim will buy a fish that is alive. So, the fish might be alive at the (future) time at which it is bought but not (yet) alive at the time of speech. If the lexical semantics of expect truly contains a future-shifting operator, then we would expect this operator also to shift the interpretation of presenttense clauses in its scope. Consider, however, sentence (26): (26) ??Fritz expects that it is raining. It is clear that the complement clause does not have a future-shifted reading, even for those speakers who take the sentence to be acceptable. Unfortunately, the theory as so far outlined provides (26) with exactly such a future-shifted interpretation. 8

In order to capture the fact that one can have misconceptions about the temporal relationships that hold between times, the temporal ordering relation should be parameterized with respect to the world of evaluation, i.e., t" > t' should read t" >w· t'. We simplify for readability's sake.

250

Graham Katz

Following the de se analysis of tensed complements we presented in the last section (26) would be assigned the LF in (27). (27) Fritz PRESi expect λ 1 [it PRESi be raining] As we noted above, when the embedded tense is abstracted over, the tensed complement clause is interpreted as a property of times, i.e., it has the same interpretation as an infinitival clause. This means that when the complement clause in (27) is combined with a future-oriented matrix verb such as expect, the complement clause should be future-shifted by the covert temporal operator associated with the verb. The truth conditions for (26), then, should be as in (28). (28) V(w',t') e Dox(w0,to,Fritz) 3t" [t" > t' & be-raining(w',t") = 1] (26) should mean that Fritz thinks it will rain. But, of course, it doesn't. Both Ogihara (1996) and Abusch (to appear) handle this problem by simply stipulating that the covert future-shifter is only introduced when the complement clause is an infinitival. Such a stipulation, however, violates one of the basic compositionality conditions that we are trying to hold to. It effectively means that attitude verbs are provided with different interpretations depending on whether or not their complement clause is tensed. In addition, such a stipulation makes it impossible to relate the futureshifted interpretation of infinitival complements of expect to the selectional restrictions on expect's finite complements. To treat future-oriented verbs in more satisfactory way we will need to adopt a more sophisticated analysis of tensed complement clauses than the one which we have been assuming. We can no longer assume that embedded tenses are simply interpreted as variables. We need to confront the fact that tenses have semantic content.

4.

A New Semantics for Tensed Complements

In our earlier discussion of embedded tenses I made a slight omission when I suggested that the back shifted reading of a sentence such as (29) is a case of a "free" tense. (29) Fritz believed that Arnim climbed Monta Rosa. This is not really correct. A tense embedded under an attitude verb is rarely, if ever, free.9 Even if it is not bound, it is typically still interpreted with respect to the attitude holder's subjective present. On the back-shifted reading of (29), for example, the embedded past tense is interpreted with respect to the time Fritz thought it was when he 9

Potential exceptions are de re interpretations of tense such as the so-called "double access" interpretations of sentences like John said that Mary is pregnant (Ogihara 1989, Abusch 1994).

251

(A)temporal Complements

had his belief. The sentence is interpreted to mean that Fritz located himself at a time after a time at which Arnim climbed Monta Rosa. This relativity of interpretation has the effect that sometimes an embedded past tense doesn't end up referring to a past time at all. In (30), for example, the embedded past tense is interpreted with respect to Wolfgang's future doxastic alternatives and might therefore refer to a time that, while in the past with respect to this future time, is actually after the speech time. (30) Wolfgang will believe that the students turned in their papers on time. Relative tense interpretation presents well-known problems for the anaphoric analysis of tense meaning, since it shows that tenses do behave (at least sometimes) like "shifting" operators. In what follows, I would like to develop on a suggestion made by von Stechow (1995a) for how to treat relative tenses in the context of an anaphoric theory and show how this new account will help us address the problems raised above with the interpretation of tensed complements of future-oriented verbs. Von Stechow suggested that tenses be treated as relational operators. A tense takes three arguments: two times - the perspective time and the event time - and a temporal predicate. The perspective time is the time with respect to which the truth of the clause as a whole is evaluated - usually either the speech time or the subjective present of the attitude holder - and the event time is a time which satisfies the open temporal argument of the predicate. Formally, tense operators will be subscripted with two indices. The left index is interpreted as the perspective time and the right index as the event time. The interpretations for the tenses given in (31) should make this clear. (31) a. lPRESjf(w,P)

= 1 iff g^) = g(0 and P(w,g0)) = 1

b. [LPASTJF(w,JP) = 1 iffgO) < g(>) and P(w,g0)) = 1 c. ILFUTRJV,/>) = 1 iff#(/) > g(i) and P(w,g(j)) = 1 Semantically, then, tenses require that a particular temporal relation hold between the perspective time and the event time and that the temporal predicate be applied to the event time. Under this view of tense interpretation, tensed clauses are treated as formulae with possibly free variables. The indices on a tenses may be bound by other time-denoting elements in the sentence or discourse, and may themselves act as binders. Although highly non-standard, this proposal is very much along the lines of the analysis put forward recently by Kusumoto (1999). There are two operations that can apply in the course of the derivation to bind the otherwise free variables (or to rebind those that are improperly bound): existential closure and abstraction. These will be taken to be operations on LFs, as indicated in (32). (32) a. Existential Closure: CP[v] b. Abstraction:

CP

=>

[3 ν CP[v]]

=>

[λ ν CP]

252

Graham Katz

Existential Closure must bind a free variable. Abstraction need not. Abstraction applies - as we have seen - in order to avoid type-clashes. Existential closure simply provides a strategy for interpreting otherwise unbound variables (as has become standard since Heim 1982). Both operations are clause-bound. With respect to abstraction, note that variables bound by operators outside the CP must be abstracted over when the CP is a complement of an attitude verb in order to avoid the semantic problems associated with binding into an opaque context. We follow von Stechow in adopting the convention that that these variables are abstracted over last. Vacuous abstraction is not ruled out, and, as we mentioned above, in the case of normal propositional belief vacuous abstraction must apply in order to preserve interpretability. We will, however, rule out vacuous abstraction when there are free variables that could be abstracted over. In other words, if abstraction applies, then it should bind a variable if there is one to bind. Existential closure applies after abstraction to bind any variables that remain unbound. Note that variables which are deictically bound, such as the indexical pronouns and the speech time variable 0 are not bound by existential closure. While this may seem an ad hoc set of principles, in fact what we see is that the temporal variables assoicated with the tenses are being treated as indefinite pronouns, analogous to those Chierchia (2000) has recently been investigating in the nominal domain. When they are not bound, they are treated like indefinites, and when they are bound they are treated like pronouns. Since abstraction and existential closure are used to treat de se pronouns on the one hand and indefinite noun phrases on the other, it is not surprising that both operations end up playing a role in tense interpretation. Von Stechow assumed that tense operators were ambiguous between a relative interpretation and an anaphoric interpretation. I will instead treat all tenses as relative and incorporate a kind of sequence-of-tense rule into the syntax, in the style of Stowell (1993). That is, I assume that the actual morphological spell-out of a tense operator is determined by two factors: the identity of the tense operator itself, and whether or not it is marked with the morphological feature [past]. The [past] feature is lexically associated with the PAST operator, and is passed on to other tense operators whose perspective time index is "bound" by the event time index of a [past]-marked operator. This feature passing is accomplished via the Tense Agreement Rule given in (33). (33) Tense Agreement Rule [... TNS[past],... [... ,-TNS ...]...] => [... TNS[past],... [... ,TNS[past]...] ...] The tense operators themselves are spelled out morphologically as indicated below: (34) The Tense Operators PRES [past] -ed FUTR[past] would PAST[past] -H> -ed PRES - » 0 FUTR - » will

(A)temporal Complements

253

Presumably, English morpho-syntax takes care of the proper spellout of these abstract morphemes. To illustrate the workings of the analysis, let us consider the treatment of (35). (35) John believed that Bill would leave. The LF for (35), which indicates the operators and their binding relations, is (36). (36) John0 PAST[past]i believe that [Billi FUTR[past]2 leave] The [past] feature on the FUTR operator indicates that this operator is (syntactically) bound by a [pastl-marked matrix tense. Note that the [past] feature does not carry any semantic value.1 To "break" the binding relation between the higher tense and the lower tense, abstraction must apply to (36), yielding the structure in (37). (37) John0 PAST[past]i believe that λ 1 [Billi FUTR[past]2 leave] Note that had we abstracted over the free variable 2 instead of 1, the illicit binding relation between the higher and the lower tense would remain. Were we to abstract over both 1 and 2, then we would have an object of the wrong semantic type to appear as the object of believe. Only the structure in (37) satisfies our conventions concerning abstraction and is interpretable. Interpreting the LF in (37) - applying existential closure at each clausal level but not to the distinguished variable 0 - gives us the truth conditions in (38). (38) 3t [t < to & V e Dox(w0,t,John) 3t" [t" > t' & leave(w',t",Bill) = 1]] In other words, the claim is that at some time in the past John located himself in a world at a time at which Bill subsequently left. This seems correct. We might note at this point that we have already accounted for part of our puzzle, namely the fact that (2a) and (2b) cannot be synonymous. Since infinitival complements of believe are always applied directly to the believer's subjective present, but, as we have just seen, future-tense complement clauses are always future-shifted with respect to the subjective present, a Z»e/;eve-sentence with an infinitival complement can never be synonymous with the corresponding sentence with a future-tense complement. The real puzzle, of course, is to account for the fact that for expect this is not the case. Before we turn to this, however, let me illustrate how my treatment of sequence of tense works by deriving the two interpretations of the ambiguous sentence (39). As Kusumoto has noted, on accounts in which would is decomposed into a PAST operator plus a FUTR operator (cf. Abusch 1988), there is no explanation for the fact that would never gives rise to a "back shifted" reading, i.e., that (35) never means John thought that Bill had been going to leave. This problem does not arise here, as would only appears when its perspective index is bound by a higher PAST operator, as explicitly indicated in (36).

254

Graham Katz

(39) Fritz believed it was raining. As is typical for sequence-of-tense analyses, the LF for the same-time interpretation contains an embedded present tense operator, while that for the back-shifted interpretation contains an embedded past tense operator: (40) a. Fritz 0PAST[past]i believe λ 1 [it ,PRES[past]2 be raining] b. Fritz oPAST[past]i believe λ 1 [it iPAST[past]2 be raining] Note that both PRES [past] and PAST[past] are spelled out by the same morpheme. The truth conditions for (40a) and (40b) are different, of course, as we see in (41a) and (41b). (41) a. 3t [t < to & V(w',t') e Dox(w0,t,Fritz) 3t" [t" = t' & be-raining(w',t") = 1]] b. 3t [t < to & V(w',t') e Dox(w0,t,Fritz) 3t" [t" < t' & be-raining(w',t") = 1]] (40a) has an underlying present tense operator, so the perspective time index and the event time index must refer to the same time, but (40b) has an underlying past tense operator, so the event time index must refer to an earlier time. We are now in a position to give a new semantics for future-oriented verbs that does not run afoul of the problems that we raised above for the simple lexical account.

5.

Tensed Complements of Future-Oriented Verbs

1 propose that future-oriented verbs differ from present-oriented verbs not just in their temporal orientation, but in the logical type of their clausal complements. As we have seen, to interpret tensed complements of present-oriented verbs such as believe we need only abstract over the perspective time index. The event time index is typically otherwise bound, usually existentially. To interpret future-oriented verbs such as expect, however, it appears that we need to abstract over both the perspective time and the event time. I propose, then, that expect be translated as a predicate of type {(i{i{st))){e{i{st)))) with the following interpretation: (42) öexpectKHUx,/') = 1 iff V e Dox(w,t,x) 3t" [t" > t' & P(w',t",t') = 1] Whereas previously expect took as its clausal argument an intensionalized property of times, now it takes an intensionalized property of properties of times. Expect and believe now differ in the logical type of their complement clauses. In addition, expect specifies the temporal relation that holds between the times that fill the complement clause's temporal argument positions.

(A)temporal Complements

255

It may be easiest to understand how this new semantics for expect combines with our system for tensed-clause interpretation by way of example. Consider (43a), the logical form of which is given in (43b). (43) a. Fritz expects that Arnim will laugh. b. Fritzo PRESi expect [Arnimi FUTR.2 laugh] In order that the clausal argument of expect have the right semantic type, we need to abstract twice. First we abstract over the free variable 2, since an abstractor should bind a variable if there is one to bind. Then we abstract over the improperly bound variable 1, since it must be abstracted over, and by convention it must be abstracted over last. This yields the structure in (44): (44) Fritzo PRESi expect λ 1 λ 2 [Arnimi FUTR2 laugh] This structure can be interpreted, and the truth conditions for it are given in (45). (45) V(w',t') ε Dox(w0,to,Fritz) 3t" [t" > t' & laugh(w',t",Arnim) = 1]] Note here that the existential force associated with the variable t" - the time at which Arnim laughs - is contributed by the lexical semantics of the predicate expect. Furthermore, note that the relationship between the subjective present - t' - and this time of laughter is doubly specified, once by the lexical semantics of the attitude verb itself and once by the tense. This double specification accounts for the requirement that tensed complements of expect must be future-tense clauses. If the complement were a present- or past-tense clause, the temporal specification associated with the attitude verb would be in conflict with that associated with the tense.11 Consider, for example, (46a), whose logical form is (46b). (46) a. ??Fritz expects that it is raining. b.

Fritzo PRES] expect λ 1 λ 2 [itj PRES2 be raining]

While (46b) is formally interpretable, it is necessarily false, since the temporal relation introduced by expect and that introduced by the embedded present tense are in conflict. This can be seen by inspecting the computed "truth conditions" given in (47). (47) V(w',t') e Dox(w0,to,Fritz) 3t" [t" > t' & t" = t' & be-raining(w',t") = 1]]

It might be the case that these specifications have somewhat different force. The marginal acceptability of non-future tensed finite complements indicates that the temporal specification associated with the lexical entry for expect might, in fact, be some sort of default presupposition, which an overt tense can override.

256

Graham Katz

Naturally, no time can be both after t' and identical with it. Thus the lexical specification account explains the fact that only future-tense clauses can appear as complements of future-oriented verbs. As desired, the same account explains the future-shifted interpretation of infinitival complements. Again, let us taken an example. Consider an infinitival version of (43a): (48) Fritz expects Arnim to laugh. Given our new semantics, the infinitival clause is of the wrong type to appear as the complement of expect. Infinitival clauses are interpreted as properties of time, but expect needs a complement that is a property of properties of time. In order to get the right type, then, we need to apply vacuous abstraction. The logical form of (48), then, is (49). (49) Fritzo PRES! expect λ 2 [Arnim to laugh] The binder variable 2, like those variables introduced by Lewis for non-de se belief, will play no role in the specification of the truth conditions of the sentence, as it will not bind anything. The complement clause meaning will first be applied to the subjective present variable t', which will, in a sense, disappear. Then it will be applied to the future time introduced lexically by expect, and it is this time that will be interpreted as the laughing time. The truth conditions for (48), then, are those given in (50). (50) V(w',t') e Dox(w0,to,Fritz) 3t" [t" > t' & laugh(w',t",Arnim) = 1]] This, of course, is identical with (45). In the course of calculating (50), however, the temporal relation between t' and t" is specified only once, and this is by the lexical semantics of the attitude verb. Thus the lexical requirements of the embedding verb simultaneously account for the future-shifted interpretation of infinitival complements and the selectional restrictions on finite complements. Going back to our puzzle, we now have accounted for the synonymy of (la) and (lb) and the non-synonymy of (2a) and (2b). And, since we have explicated the meanings of expect and believe with the same Dox operator, we have also accounted for the synonymy of (la-b) and (2a). Furthermore, our account appears to have each of the properties we enumerated at the outset that a compositional account should have. Both verbs are given univocal interpretations, as are both classes of complement clauses, and the mechanism of verb/complement-clause combination is uniform. Only the flexibility which we have built into this mechanism might be seen as not-quite-compositional. It is precisely this flexibility, however, that allows us to relate the temporal interpretation of infinitival complements to the selectional restrictions on finite complements, this being the final desideratum that we had hoped our analysis would satisfy. I would venture to suggest, then, that the brute empirical facts surrounding the temporal interpretation of complement clauses might simply show us how necessary a flexible, dynamic approach to natural language semantics is.

(A)temporal

6.

Complements

257

Conclusion

Deriving the temporal interpretation of finite and non-finite clausal complements of attitude verbs such as believe and expect presents many challenges for a compositional semantic theory. Owing to the groundbreaking work of von Stechow, Heim, Ogihara, Abusch, and others, we have been able to come to grips with the scope of these challenges. The simple facts that need to be accounted for - that infinitival complements of believe have a present interpretation while infinitival complements of expect have a future interpretation and that finite complements of believe can be past, present, or future tensed, but finite complements of expect must be future tensed - do not, in the end, appear amenable to a classical analysis. We have been forced to adopt a more sophisticated approach, on which tenses display both anaphoric and operator-like properties. This may appear to many to be quite strange. I hope that the strangeness isn't off-putting and that the considerations that have led us to adopt the analysis are convincing. After all, superficial oddity should not trouble us - here of all places - if the account can be shown to be motivated and to work.

References Abusch, Dorit (1988). Sequence of Tense, Intensionality and Scope. In: Hagit Borer (ed.) Proceedings of the 7th West Coast Conference in Linguistics. Stanford Linguistics Association. Abusch, Dorit (1994). Sequence of Tense Revisited: Two Semantic Accounts of Tense in Intensional Contexts. In: Hans Kamp (ed.) Ellipsis, Tense and Questions. Dyana2 Esprit Basic Research Project 6852, Deliverable R2.2.B. Abusch, Dorit (to appear). On the Temporal Composition of Infinitives. In: James H. Higginbotham, Alessandra Giorgi & Fabio Pianesi (eds.) On the Syntax and Semantics of Tense and Mood Selection. Oxford: Oxford University Press. Chierchia, Gennaro (1989). Anaphora and Attitudes De Se. In: Renate Bartsch, Johan van Benthem & Peter van Emde Boas (eds.) Semantics and Contextual Expression. Dordrecht: Foris Publications. 1-31. Chierchia, Gennaro (2000). Indefinite Pronouns. Paper presented at the Seminar fur Sprachwissenschaft, Tübingen. Cresswell, Max J. & Arnim von Stechow (1982). De Re Belief Generalized. Linguistics & Philosophy 5. 503-535. Dowty, David R (1982). Tenses, Time Adverbs and Compositional Semantic Theory. Linguistics and Philosophy 5. 23-55. Εης, Murvet (1987). Anchoring Conditions for Tense. Linguistic Inquiry 18. 633-657. Εης, Murvet (1986). Towards a Referential Analysis of Temporal Expressions. Linguistics and Philosophy 9. 405-426.

258

Graham Katz

Galton, Antony (1984). The Logic of Aspect. Oxford: Clarendon Press. Heim, Irene (1982). The Semantics of Definite and Indefinite Noun Phrases. PhD dissertation, University of Massachusetts, Amherst. Heim, Irene (1994). Comments on Abusch's Theory of Tense. Ms. Heim, Irene & Angelika Kratzer (1998). Semantics in Generative Grammar. Oxford: Blackwell. Katz, Graham (1995). Stativity, Genericity and Temporal Reference. PhD dissertation, University of Rochester. Kratzer, Angelika (1998). More Structural Analogies Between Pronouns and Tenses. In: Proceedings of SALT VIII. (Cornell Working Papers in Linguistics.) Kusumoto, Kiyomi (1999). Tense in Embedded Contexts. PhD dissertation, University of Massachusetts, Amherst. Ladusaw, William (1977). Some Problems with Tense in PTQ. Texas Linguistic Forum 3. 89-102. Lewis, David (1979). Attitudes De Dicto and De Re. The Philosophical Review 88. 13-43. Montague, Richard (1974). Formal Philosophy: Selected Papers of Richard Montague. New Haven: Yale University Press. Ogihara, Toshiyuki (1989). Temporal Reference in English and Japanese. PhD dissertation, University of Texas at Austin. Ogihara, Toshiyuki (1996). Tense Attitudes and Scope. Dordrecht: Kluwer Academic Publishers. Partee, Barbara (1973). Some Structural Analogies between Tense and Pronouns in English. The Journal of Philosophy 70. 601-609. Partee, Barbara (1984). Nominal and Temporal Anaphora. Linguistics and Philosophy 7. 243-286. Stowell, Timothy (1993). Syntax of Tense. Ms, University of California, Los Angeles. Stechow, Arnim von (1984). Structured Propositions and Essential Indexicals. In: Fred Landman & Frank Veltman (eds.) Varieties of Formal Semantics. Dordrecht: Foris Publications. 385-^04. Stechow, Arnim von (1995a). On the Proper Treatment of Tense. In: Mandy Simons & Teresa Galloway (eds.) Proceedings of SALT V. (Cornell Working Papers in Linguistics.) 362-386. Stechow, Arnim von (1995b). Tense in Intensional Contexts: Two Semantic Accounts of Abusch's Theory of Tense. In: Fritz Hamm, Joachim Kolb & Arnim von Stechow (eds.) The Blaubeuren Papers, Universität Tübingen, Seminar für Sprachwissenschaft. 379-433.

Focussed Number Phrases

r

Katalin E. Kiss

1.

Introduction

The type of focus that occupies a designated Α-bar position in the left periphery of the sentence, for example, the Hungarian preverbal focus or the English cleft constituent,1 is understood to express exhaustive identification from among a set of alternatives present in the domain of discourse. The intuitive notion of identification associated, e.g., with the Hungarian structural focus essentially corresponds to that represented by the structured meaning of focus structures in Arnim von Stechow's theory (see von Stechow 1982, 1991, Creswell & von Stechow 1982, etc.). Consider a Hungarian sentence containing a preverbal focus (presumably sitting in the specifier of a focus projection (FP), with the V preposed into the F head), and the formula describing the meaning of the FP in von Stechow's framework: (1)

a. Mari [FP

ARNIM

CIKKET, [F· olvastas

[Vp t j

el

t,]]]

Mary Arnim's paper read PREF 'As for Mary, it was Arnim's paper that she read' b. Α\Ν(χ)(Μ)], THING)

I will follow here the latter notation convention, which will be slightly more convenient. I will call the first part of a question meaning its background, the second part its restriction. Polarity questions then will be represented as follows: (8)

Did Mary read Die Kinder der Finsternis? afIf(READ(KF)(M))], {λρ[-,ρ], λρ[ρ]}>

The answers that we have considered so far are term answers (cf. von Stechow & Zimmermann 1984). They are to be contrasted to full answers:

290 (9)

Manfred Krifka A:

Who did Maiy see?

Β:

Mary saw John.

The meaning of a full answer is simply the question meaning applied to the term answer.

1.3

The Proposition Set Approach

The proposition set analysis goes back to Hamblin (1958, 1973) and has been further developed and refined by Karttunen (1977) and Groenendijk & Stokhof (1984). The essential idea is that the meaning of a question is the set of its possible full answers, that is, a set of propositions. (10) A:

B:

Who read Die Kinder der Finsternis?

{p|3x[PERSON(x) λ ρ = READ(KF)(x)]} = {READ(x)(KF) | PERSON(x)} = {READ(KF)(J), READ(KF)(M), ...} Mary read Die Kinder der Finsternis. READ(M)(KF)

It is important to keep in mind that p, or READ(M)(KF), stands for propositions (sets of possible worlds), and not for their descriptions. There is a variety of the proposition set analysis which assumes that the meaning of a question is the set of descriptions of those propositions in some representation language (cf. Higginbotham & May 1981), which is a more expressive framework, as it would allow for syntactic operations on these descriptions that could not be defined for sets of propositions. I will generally assume a restrictive denotational framework here, not a representational one. In this I follow the general tenet of truth-conditional semantics that semantic analysis cannot consist in the mere translation of one language into another one (cf. Lewis 1970). The natural answer type of the proposition set analysis is a full answer. In Who read Die Kinder der Finsternis? - Mary, the answer has to be spelled out first as Mary read Die Kinder der Finsternis before it can be interpreted. Proponents of the proposition set analysis have rarely discussed how this spelling-out of full answers can be achieved. Presumably, it is by some syntactic process that deletes material recoverable from the question, essentially, all the material except the question constituent. This material has to be retrieved; presumably, the deleted material is interpreted anaphorically in some way or other. (11) A: Β:

Who [read Die Kinder der Finsternis] ι ? Mary 0 ι .

Spelling out: Mary [read Die Kinder der Finsternis]ι The proposition set analysis deals with polarity questions as follows:

For a Structured Meaning Account of Questions and Answers (12) A: B:

Did Mary read Die Kinder der Finsternis! Mary didn't read Die Kinder der Finsternis.

291

{READ(KF)(M), -IREAD(KF)(M)} -IREAD(KF)(M)

As before, the natural answer of a polarity question is a full answer. But now it is especially difficult to see how the term answers yes and no could be dealt with in this system. The reason is that the two propositions in the question meaning cannot be identified as "positive" and "negative", as one proposition is just the complement of the other - recall that all we have is sets of possible worlds, not their descriptions. We will therefore have to treat yes and no as devices that refer to a syntactic object that is constructed when formulating the question. We could analyze yes as an anaphoric element that refers to the proposition embedded in the polarity question, and no as an anaphoric element that refers to its negation: (13) A:

Did Mary read Die Kinder der Finsternis?

LF:

QU [Mary read Die Kinder der Finsternis]

Interpretation:

[QU]([Mary read Die Kinder der Finsternis}) = λρ{ρ, -,p}(READ(KF)(M)) = {READ(KF)(M), -|READ(KF)(M)}

Β:

Yes Ι.

[Yes Ι ] = [Antecedens of

Noi.

[Noi] = -i[Antecedens of

1] =

READ(KF)(M)

1 ] = -.READ(KF)(M)

There are a number of variations on the general theme of the proposition set analysis of questions. Karttunen (1977) proposed that the meaning of a question should just be the set of its true answers. (In the following representation, v p can be read as "p is true in the actual world".) (14)

What did Mary read? { Ρ | 3x[ v p Λ xe THING Λ Ρ = R E A D ( X ) ( M ) ] } If Mary read Die Kinder der Finsternis and nothing else: = { R E A D ( K F ) ( M ) }

The reason for this move was to obtain a proper analysis of inferences involving embedded questions, such as Mary knows whether it is raining; it is raining => Mary knows that it is raining, to which I will return in section 5. Notice that Karttunen's theory does not provide for an easy solution for term answers either. Groenendijk & Stokhof (1984) proposed another variant, according to which the meaning of a question is the set of its exhaustive answers. This is how they would represent our example question, with possible worlds made visible: (15) What did Mary read?

XjXi[Xx[READ(i)(x)(M)] = Xx[READ(j)(x)(M)]], = Θ

292

Manfred Krifka

We get a relation between possible worlds j and i that obtains if the things that Mary read in i are the same as the things that Mary read inj. This is not a set of propositions yet, but it comes close enough - it is an equivalence relation, which stands in a one-toone correspondence with a partition on the set of possible worlds: (16) Partition corresponding to the equivalence relation ©: {p|3i[iepAp={j|©(i)(j)}]} If only Die Kinder der Finsternis and Das Totenschiff are under discussion, we get the following set of four propositions: ( 1 7 ) {READ(KF)(M) A READ(TS)(M), READ(KF)(M) Λ -iREAD(TS)(M), —iREAD(KF)(M) Λ READ(TS)(M), -IREAD(KF)(M) A -IREAD(TS)(M)}

As equivalence relations contain exactly the same information as their corresponding partitions, the account of Groenendijk & Stokhof clearly belongs to the proposition set analysis. But interestingly, to derive this propositional interpretation in a compositional way, Groenendijk & Stokhof have to go through a construction step that involves question meanings of the structured meaning type, called "abstracts". For question (15), we derive from a zero-place abstract (18a) a one-place abstract (b), which then is turned into the question meaning (c): ( 1 8 ) a. [READ(iXx)(M)] b. X.x[READ(i)(x)(M)]

C. Xj λί [X.x[READ(i)(x)(M)] = λχ[READ(j )(x)(M)]] In general, we can derive a propositional interpretation from a structured meaning interpretation, but not vice versa - structured meaning interpretations are more expressive. One of the aims of this paper is to show that we indeed need this additional expressiveness.

2.

Problems with Focus in Answers

It is an old observation (e.g., Paul 1880) that the prosodic contour of the answer relates an answer to the corresponding question. In particular, the focus of the answer, which is marked by a prosodic peak, corresponds to the meaning of the question: (19) a. A: Β: B':

Who read Die Kinder der Finsternis? Mary read Die Kinder der Finsternis. *Mary read Die Kinder der Finsternis.

For a Structured Meaning Account of Questions and Answers b. A: B: Β':

293

What did Mary read? Mary read Die Kinder der Finsternis. * Mary read Die Kinder der Finsternis.

The focus pattern of the answer has largely been neglected in the development of theories of question meaning. But it can be integrated. The two frameworks for question meanings suggest different treatments of focus.

2.1

Answer Focus in the Proposition Set Framework

The proposition set framework for questions fits with the Alternative Semantics analysis of focus, as developed by Rooth (1985) for contrastive focus, for focus-sensitive operators like only, and for focus-sensitive quantification. Rooth (1992) also deals with focus in answers to questions. In Alternative Semantics, expressions have, in addition to their ordinary meaning, a set of alternative meanings. The alternatives are introduced by expressions in focus (indicated by a syntactic feature, F, that is realized by accent). This is illustrated in the following examples. expression

ordinary meanings

alternatives

(20) [Mary]F Μ [Märy]F read Die READ(KF)(M) Kinder der Finsternis.

A L T ( M ) , = {M, J . . . }

(21) [Die Kinder der KF Finsternis]? Mary read [Die READ(KF)(M) Kinder der Finsternis]^.

ALT(KF)

{ R E A D ( K F ) ( x ) I XG A L T ( M ) } = {READ(KF)(M), READ(KF)(J),...}

= { K F , TS, . . . } {READ(x)(M) I x e A L T ( K F ) } = {READ(KF)(M), READ(TS)(M),...}

For meanings a, ALT(a) stands for the alternatives of a. The alternatives can either be taken broadly (e.g., ALT(a) = all objects of the same type as a ; this is the option that Rooth 1992 chooses), or restrictively (ALT(a) = the set of relevant alternatives in a particular context); they must contain at least two elements. Example (20) shows that the ordinary meaning of Märyρ read Die Kinder der Finsternis is the proposition that Mary read Die Kinder der Finsternis, and the alternatives are propositions of the type "x read Die Kinder der Finsternis", where χ ranges over persons. Both the ordinary meaning and the alternatives can be derived compositionally from their parts; the rule for the alternatives follows in a straightforward way from the rule for the ordinary meanings. Notice that while the ordinary meanings of (20) and (21) are the same, their alternatives are different, and correspond to the question meanings. In particular, for congruent answers, the alternatives are supersets or subsets of the question meaning, depending on whether we understand ALT(a) in a broad way or in a narrow way. This

294

Manfred Krifka

is illustrated in (22). The answer MäryF read Die Kinder der Finsternis is congruent to the question Who read Die Kinder der Finsternis, as the alternatives of the answer are either supersets (i) or subsets (ii) of the question meaning. (22) A: B:

Who read Die Kinder der Finsternis ?

{READ(KF)(x) | PERSON(x)}, = { READ(KF)(M), READ(KF)(J), . . . } ,

[Mary]p read Die Kinder der Finsternis.

{READ(KF)(x)

= ©

| xe ALT(M)}, = Φ

The answer is congruent, as we either have (i) or (ii): (i)

©

C

©, if A L T ( M ) = all objects of the type of M,

w e h a v e Φ = {READ(KF)(M), READ(KF)(J), READ(KF)(MOON),...}

(ii)

Φ c Φ , if ALT(M) = a contextually restricted set of persons, e.g., {M, J} w e h a v e Φ = (READ(KF)(M), READ(KF)(J)}

On the other hand, the answer Mary read [Die Kinder der Finsternis]^ is not congruent, as its alternatives are neither a superset nor a subset of the question meaning; they just overlap in one element. (23) B:

*Mary read [Die Kinder der Finsternis]F.

{READ(x)(M) | χ e ALT(KF)} = {READ(KF)(M), READ(TS)(M), . . . } , = ®

The answer is not congruent: (i)

® cz ©

(ii)

Φa ®

(but notice that Φ

Η

®=

{READ(KF)(M)})

One necessary qualification of the superset-or-subset criterion is that the subset contains more than one element. Otherwise we will have ® c © or © c ® even with noncongruent answers, as they overlap in one proposition. We can plausibly require that both the question and the alternatives of the answer contain more than one proposition: A true question must leave at least two possibilities open, and the contrast indicated by a focus would not be a true contrast if the focused expression does not have at least one proper alternative. This suggests another, simpler criterion: the meaning of the question and the alternatives of the answer have more than one element in common (notice that #(© η ®) > 1 but #(© η (D) = 1). For non-congruent answers, the meaning of the question and the alternatives of the answer have only one element in common, the ordinary meaning of the answer. This suggests the following three criteria for congruent question-answer pairs:

For a Structured Meaning Account of Questions and Answers

295

(24) Possible criteria for congruent question-answer pairs Q - A : a. [ β ] c [Λ]Α, provided that #([0]) > 1 b. [Λ]Α C [ 0 ] , provided that #(MA) > 1 C. # ( M A Η

[ρ]) >

1

There is a fourth criterion, which may appear to be the most obvious one. We may assume that congruent question-answer pairs identify the meaning of the question and the alternatives of the answer, i.e., [ 0 ] = M A · This identification is not asserted, but presupposed; every answer to a question presupposes that it is a congruent answer. Being a presupposition, the identity statement can be accommodated. We would have to assume an additional restriction for question meanings, RESTR, that is subject to accommodation just like ALT. (25) A: Β:

Who read Die Kinder der Finsternis?

{READ(KF)(X) | PERSON(x),

[Märy]F read Die Kinder der Finsternis.

{READ(KF)(X)

X: Β = Β'and F e R. This predicts that (29a - b) is a congruent question-answer pair, as the background of the question and the background of the answer are identical, and the focus is an element of the restriction of the question. But (29a - c) is not congruent, as the backgrounds differ, and (29a - d) is not congruent either, as the focus is not an element of the restriction. (29) a. Who did Mary see?

(Xx[SEE(x)(M)], PERSON)

b. Mary saw [Johnjp.

(Xx[SEE(x)(M)], J), where J E PERSON

c. *[Mary]p saw John.

(Xx[SEE(J)(x)], M), where Μ E PERSON

d. *Mary saw [Die Kinder der Finsternis]^.

(Xx[SEE(x)(M)], KF), where KF £ PERSON

We conclude that there are ways to represent the contribution of the focus of the answer in both the proposition set framework and the structured meaning framework. But let us turn now to cases that present a problem for the proposition set analysis.

2.3

Underfocused and Overfocused Answers

Consider the following question-answer pairs: (30) A:

What did Maiy do?

B:

Mary [read Die Kinder der Finsternis]^.

Β':

*Mary read [Die Kinder der Finsternis]^.

Β":

*Mary [reäd]F Die Kinder der Finsternis.

For a Structured Meaning Account of Questions and Answers

297

Clearly, A - Β is congruent, but A - B' and A - B" are not. Intuitively, the expressions that are focused are too small; I will call such answers underfocused. This does not readily appear in the case A - B' due to an ambiguity of focus marking: If a VP consisting of a verb and an argument is focused, this focus is typically expressed by accenting the argument (cf. Schmerling 1976, Gussenhoven 1983, von Stechow & Uhmann 1986); but when just this argument is in focus, it will also receive accent. The two cases differ, however, when the argument is given, e.g., pronominal, because then the argument does not receive accent when part of a larger focus: (31) B: B':

Mary [read that]F. *Mary read [that]F.

The converse problem to the one just discussed appears in the following example: (32) A:

What did Mary read?

B:

Mary read [Die Kinder der Finsternis]?.

Β':

*Mary [read Die Kinder der Finsternis]?.

In (32B'), the focus of the answer is too large; it is an overfocused answer. As focus marking is ambiguous, this does not result in ungrammaticality; the answer Mary read Die Kinder der Finsternis can be interpreted as (32B) instead. But in case the object is pronominal, we do find that the answer corresponding to B' is ungrammatical: (33) B: B':

Maiy read [that]F. *Mary [read that]F.

Another instance of overfocusation is illustrated by the following example: (34) A:

What did Mary do with Die Kinder der Finsternis?

Β:

Mary [reäd]F Die Kinder der Finsternis.

B':

*Mary [read Die Kinder der Finsternis]?.

We would expect that an adequate theory of questions and congruent answers excludes overfocused and underfocused answers. Let us check whether this is the case with the two frameworks that we have discussed. We start with the proposition set framework, in Hamblin's version. Take the following example of an underfocused answer:

298

Manfred Krifka

(35) a. What did Mary do? (P(M) I Ρ e ACTIVITY} = {SANG(M), DANCED(M), LISTENED TO(BACH)(M), LISTENED TO(BEETHOVEN)(M), READ(KF)(M), READ(TS)(M), BOUGHT(KF)(M), BOUGHT(TS)(M),...}, = Φ

b. ""Mary [read]? Die Kinder der Finsternis. Alternatives: {R(KF)(M) | R e ALT(READ)} = {READ(KF)(M), BOUGHT(KF)(M), . . . } , = ©

The question is whether any one of the three criteria for congruent question-answer pairs listed in (24) rules out this answer. Notice that we have © c ©, and also #(© η ©) > 1, hence the criteria (24b) and (c) will not do. We might see this as evidence that we should take the first criterion, (24a), as we don't have © c ©. But now we run into the problem of overfocused answers: (36) a. What did Mary do with Die Kinder der Finsternis? {R(KF)(M) I R e DIRECTED ACTIVITY} = {READ(KF)(M), BOUGHT(KF)(M), . . . } , = Φ

b. *Mary [read Die Kinder der Finsternis]F {P(M) I Ρ e ALT(READ(KF))} = {SANG(M), DANCED(M), LISTENED TO(BACH)(M), LISTENED TO(BEETHOVEN)(M), READ(KF)(M), READ(TS)(M), BOUGHT(KF)(M), B0UGHT(TS)(M),...}, = ©

According to the first criterion, (24a), this should be a congruent answer, which it is not. We run into problems with all three criteria for the proposition set theory. The identification strategy (25) doesn't work either: (37) a. What did Mary do with Die Kinder der Finsternis? {R(KF)(M) I Re directed activity, Re RESTR}, = Φ b. *Mary [read Die Kinder der Finsternis]F {P(M) I Ρ e ALT(READ(KF))}, = ©

Notice that accommodation Φ = ©, with #(©) > 1, is possible, leading to Φ = © = {READ(KF)(M), BOUGHT(KF)(M) }

Hence (25) does not predict either that this question-answer pair is not congruent. This is not a problem specific to Hamblin's version of the proposition set theory. It would appear in Karttunen's version as well, which differs from Hamblin's only insofar as the propositions in the question meaning are reduced to the ones that are true. It

For a Structured Meaning Account of Questions and Answers

299

also appears in Groenendijk & Stokhof s theory. For our example (35), that theory would give us a similar representation, except that we should understand the propositions in the question meaning as holding exhaustively. Presumably, the alternatives created by focus in the answer should be interpreted exhaustively as well. But then the problems of over- and underfocused answers as illustrated in (35) and (36) appear. The danger of over- and underfocusation in Alternative Semantics has been observed with contrastive focus by Schwarzschild (1994). He draws attention to the fact that (38a,b), but not (c), are adequate contrastive statements in the given context: (38)

A:

John picked strawberries at Mary's farm.

a. B:

No, John picked strawberries at Sandy's? farm.

b. B:

No, John picked blueberries? at Sandy's? farm.

c. B:

#No, John picked strawberries? at Sandy's? farm.

Sentence (c) is overfocussed. But if the rule for contrastive statements (in Alternative Semantics) is just that the target sentence (here, A) should express a proposition that is an element of the alternatives of the contrastive statement, then (c) should be fine. Schwarzschild's solution is to appeal to a general principle that favors minimal focus. This is a consequence of general pragmatic considerations: focus means added complexity (the introduction of alternatives), and therefore should be used only when necessary. Focus on blueberries in (b) is necessary because the contrastive statement differs in this constituent from the target sentence, but focus on strawberries in (c) is not necessary to satisfy the conditions for contrastiveness, and so (a) should be preferred over (c). We may try to apply Schwarzschild's Razor - "no complexity without necessity!" to congruent question-answer relations. For example, for Who read Die Kinder der Finsternis, the answer Märy? read Die Kinder der Finsternis is preferred over the answer Märyf read [Die Kinder der Finsternis]? because the alternatives of the first answer form a proper subset of the alternatives of the second answer. Similarly, for the question What did Mary read?, the answer Mary read [Die Kinder der Finsternis]? is better than the answer Mary [read Die Kinder der Finsternis]^ because the alternatives of the first answer form a proper subset of the alternatives of the second. In general, then, we would predict that focus is assigned only when necessary, and on as small a constituent as possible, in order to satisfy the condition for congruent question-answer pairs. The problem of this pragmatic account is that it makes wrong predictions for underfocused answers, as in (35). Here the alternatives of the non-congruent answer form a proper subset of the alternatives of the congruent answer, which would give preference to the non-congruent answer. In the contrastive cases that Schwarzschild considered, underfocused statements as in (39B) are excluded because in this case the meaning of the target sentence (A) is not an element of the set of alternatives (B).

300

Manfred Krifka

(39) A: B:

John picked strawberries at Mary's farm. *No, John picked [blueberries^ at Sandy's farm, (where Mary Φ Sandy)

Hence, there is a natural limit for the pragmatic rule that prefers minimal focusation in the case of contrastive statements. But there is no such limit if we apply it to answers to questions. Also, the general rule, "Focus minimally!", which might work fine for cases with multiple focus as in (38), does not work for single foci. It is not quite clear how to apply this principle to single focus, but either interpretation leads to problems. If we interpret it as saying, "Use a focus as large as possible", then we would run into problems with overfocused answers, and if we interpret it as saying, "Use a focus as small as possible", we would have similar problems with underfocused answers. What about overfocused or underfocused answers in the structured meaning framework? The problem simply does not arise. Consider the following example of an underfocused answer: (40) a. What did Mary do?

ρ] and λρ[ρ] form the restriction, which enables the answers no and yes. In the meaning of the alternative question (b), the propositions themselves form the restriction, which enables answers like Mary read Die Kinder der Finsternis and Mary didn't read Die Kinder der Finsternis. Once more, the structured meaning framework allows for more distinctions than the proposition set framework, and it appears that these distinctions are necessary.

4.

Problems with Multiple Constituent Questions

4.1

Multiple Constituent Questions

A third area in which the structured meaning framework fares better concerns multiple constituent questions, that is, questions with more than one question constituent. We have to distinguish a number of types of such questions (cf. Wachowicz 1974, Comorovski 1996). The first type is matching questions. These are questions that presuppose a list answer. Examples are given in (51) to (53): (51) A: B: (52) A: B: (53) A: B:

Who read what? Mary read Die Kinder der Finsternis, and John read Das Tötenschiff. Who left when? Mary left at four, and John left at four-thirty. Who went where when? Mary went to Spain in September, and John went to France in July.

A different type of multiple constituent question is illustrated in the following examples. These questions do not presuppose a list answer, and they are expressed differ-

304

Manfred Krifka

ently from matching questions, usually involving a conjunction. They are commonly called conjoined questions. (54) A: B: (55) A: B:

Who left, and when? Mary left, at four. Who read something, and what? Mary read something, namely Die Kinder der Finsternis.

There is a type of question that appears quite similar to matching questions in its syntactic form, but which presupposes that there is only a single answer, and hence is similar to conjoined questions in its semantics: (56) A:

What maestro is conducting what orchestra in the performance of what symphony at tonight's concert?

B:

Bernstein is conducting the Boston Symphony orchestra in the performance of Beethoven's Fifth Symphony.

(57) A: B:

Which French general invaded which Eastern-European empire in which year? Napoleon invaded Russia in eighteen twelve.

Example (56) is from Aqvist (1965), who took it to show that multiple questions need not be answered by a list. However, Wachowicz (1974) points out that these questions are typically pronounced with an intonation phrase after every question phrase, which she takes as indicating that these questions are to be interpreted conjunctively, and not as matching questions. Wachowicz also observes that questions of this type are often used in quizzes, that is, the person that asks the question already knows the answer. Hence she calls them quiz questions. This is indeed a typical use of such questions, but it is probably not a defining criterion; a question like (56) can be uttered as a sincere request for information. But what is typical for these questions is that they are about a particular event (in our examples, tonight's concert and a particular event in history). This makes them good candidates for quizzes, which often ask for information about a particular event. Another type of multiple constituent question is echo questions, as in the following example: (58) A: B:

Esmeralda needs a bandoneon. Who needs what?

Of course, echo questions need not be multiple constituent questions; Esmeralda needs whät? or Who needs a bandoneon? are proper echo questions in reaction to (58A) as

For a Structured Meaning Account of Questions and Answers

305

well. The question constituent in echo questions asks for the specification of phonological information of the preceding sentence, either because it was not understood properly, or because the hearer is reluctant to believe what is expressed by the first sentence. In a sense, echo questions are a variety of quiz question because both question types are about a particular event - standard quiz questions about an event that the speaker assumes to be identifiable by the hearer, and echo questions more specifically about a speech event, the utterance of the preceding sentence. Of course, they differ in other respects; echo questions are just about phonological material and hence can disregard many restrictions on wA-elements otherwise found. In the following, I will mainly concentrate on matching questions, but it is critical to first establish them as a type that is separate from conjoined questions, including multiple quiz questions and echo questions. There are indeed a number of differences. First, for matching questions there is a restriction that the wA-constituents must be clausemates. Kuno & Robinson (1972), building on work by Baker (1970), observe the following minimal pairs: (59) a. I don't know who expects who to marry Mary. b. *I don't know who expects that who will marry Mary. (60) a. I cannot tell you who I like better than who. b. *I cannot tell you who I like better than I like who. The syntactic nature of this constraint is not quite clear, as root questions in which the second question constituent is embedded in a complex noun phrase are fine (cf. Which student bought the book that which professor recommended?). It may be that clausemate cases are preferred over non-clausemate cases with a similar meaning; notice that there is no alternative clausemate construction if the question constituent occurs in a complex noun phrase. In any case, the clausemate constraint does not obtain for echo questions and quiz questions: (61) Quiz show moderator: Now, for 500 dollars, who expects that who will marry Mary? (62) A: B:

(almost inaudible): John expects that Bill will marry Mary. What did you say? Who expects that who will marry Mary?

Second, question formation is known to lead to ungrammaticality in weak-crossover contexts: (63) *Whoi does hisi mother adore ti? But this does not hold for echo questions (cf. Wachowicz 1974, an observation attributed to Cole 1972):

306

Manfred Krifka

(64) A: B:

(almost inaudible): You must know, his mother adores Jonathan. Whoi does hisj mother adore?

Weak-crossover effects are absent in quiz questions as well, including multiple quiz questions: (65) Which Roman emperori did which of hisi female relatives despise ti so much that she had him poisoned? This contrasts with matching questions, where we find weak-crossover effects: (66) A: B:

Which Roman emperori did which of hisi female relatives despise ti? ""Tiberius, his daughter despised, and Nero, his mother.

A further difference is that matching questions on the one hand and conjoined questions and quiz questions on the other are sometimes expressed differently in some languages (cf. Comorovski 1996:134ff, for Romanian). (67) a. Cinei si cei [ti ti-a spus t2]? who and what to.you-has told 'Who told you what?' (expects single answer) b. Cinei cei [ti ti-a spus t2]? who what to.you-has told 'Who told you what?' (list answer preferred) For non-arguments we find a similar construction in German: (68) a. Wann und wo hast du Urlaub gemacht? when and where have you vacation made 'When did you go on vacation, and where?' (single answer) b. Wann hast du wo Urlaub gemacht? when have you where vacation made 'When did you go on vacation where?' (list answer preferred) However, in both cases the pattern identified for matching questions, (67b) and (68b), can also be used for quiz questions, though not as the preferred form.

For a Structured Meaning Account of Questions and Answers

4.2

307

Conditions for Matching Questions

We have seen that one important condition for matching questions is that they require a list answer. An answer with a single sentence is not felicitous; the question presupposes that there is more than one answer. (69) A: B:

Who saw Bob when? *Mary, at four. (o.k. in quiz interpretation, Who saw Bob, and when?)

In case a list answer is excluded for extraneous reasons, matching questions are infelicitous, as pointed out by Wachowicz (1974). (70) a. *Who killed John F. Kennedy when? o.k. only in "quiz" interpretation. b. Who killed which American President when? Another condition for matching questions is that one of the question constituents is linked to a contextually given set; cf. Comorovski (1985) and Pesetsky (1987). They are D(iscourse)-linked. Typically, the D-linked phrase is the first question constituent. This explains the following contrast (cf. Comorovski 1996). (71) a. Who on earth broke the expensive Chinese vase? b. *Who on earth broke what? The phrase on earth indicates that the domain for who should be taken as widely as possible, hence it marks the phrase as not D-linked. This is in conflict with the requirement of matching questions that the first question constituent be D-linked. It is typically the subject that is D-linked, and non-D-linked expressions cannot be moved in front of D-linked expressions, an effect known as superiority (cf. Chomsky 1973). (72) a. Who read whät? b. "What did who read? But there are well-known exceptions. Which-phrases, which presumably are always Dlinked, can violate superiority (cf. Karttunen 1977; cf. (73)), and appropriate contexts can also lead to acceptable violations of superiority (cf. Bolinger 1978a, Pesetsky 1987; cf. (74)): (73) Which girl did which boy kiss?

Manfred Krijka

308

(74) I know that we need to install transistor A, transistor B, and transistor C, and I know that these three holes are for transistors, but I'll be damned if I can figure out from the instructions where what goes. Bolinger (1978a) suggests that D-linking induces fronting of a question constituent, and that question constituents that are not D-linked remain in an interior position. However, it seems that the subject is an exception to this rule, as it has to be understood as D-linked even when occurring in a non-fronted position. This was pointed out by Comorovski (1996:85) with the low acceptability of a question like the following: (75)

??

Which book did how many people buy?

The subject, how many people, cannot normally be interpreted as D-linked, even if it occurs in an interior position. It refers to a number, and numbers are normally not given by the context. But, of course, other wA-constituents in an interior position need not be D-linked. It seems that a matching question in which one wA-constituent is Dlinked is "about" the antecedent set of this constituent. Consequently, Bolinger (1978a) identifies the first wA-phrase of a matching question as the topic of this question. Kuno (1982) suggests another way of describing the unequal role of the question constituents in a matching question. According to him, the fronted constituent represents the sorting key to access the relevant pieces of information in the answer. (76) A: B:

(77) A: B:

Which of these climatic conditions occur in which countries? Typhoons occur in Japan, Korea, and China; hurricanes occur in the Carribean and in the southern US;... In which countries do which of these climatic conditions occur? In Japan, typhoons and early-summer rain spells occur; in Thailand, they have monsoons;...

Notice that answer (76B) would be odd for question (77A), and answer (77B) would be infelicitous for question (76A). This is because in (76A) the sorting key is the set of climatic conditions, whereas in (77A) the sorting key is the set of countries, and the answer of a matching question should be from the perspective of the sorting key of the question. Kuno's account can be related to D-linking by the plausible assumption that sorting keys must be given, i.e., D-linked. Another condition worth mentioning is that matching questions usually presuppose that every element in the set denoted by the D-linked question constituent is part of one answer in the answer list. That is, matching questions are interpreted distributively. This explains the following contrast:

For a Structured Meaning Account of Questions and Answers

309

(78) We had a list of criminals, and we knew for sure that at least some of them were involved in the crimes. The question we had to solve, then, was: a. Which crime was committed by which criminal. b. Which crime did which criminal commit. c.

??

Which criminal committed what crime.

In the context given, the questions are about the crimes, not about the criminals; hence, the crimes are D-linked and the sorting key. This explains why (a), a form in which the phrase which crime occurs in subject position, is preferred over (c), and why even superiority violations as in (b) are allowed. Finally, there is a curious contrast that shows up with embedded questions (cf. Karttunen 1977): (79) a. Mary knows which student read which book, b. *Mary knows whether John read which book. Whether marks embedded polarity questions (or alternative questions, cf. whether John wants tea or coffee; we have noticed this formal similarity already for root questions above). We can analyze whether as a question constituent that ranges over the truth value of sentences, but then it remains to be explained why it is different from other question constituents in not allowing for matching questions. In light of the discussion above, the reason why (79b) is out is the following: whether is the first question constituent, which is typically interpreted as D-linked, but truth values are never D-linked.

4.3

The Semantics for Matching Questions

What do the observations of the previous section tell us about the semantics of matching questions? They clearly make it impossible to defend an analysis for which the question constituents are treated on a par. It is useful to compare matching questions in natural languages with the semantics of queries in a database language like Prolog (cf. Clocksin & Mellish 1981), which can be given as follows: (80) A question with multiple wh-elements [whi.. .wh2.. .wh3...] asks for which tuples (Χι, X2, X3) the sentence [Xi.. .X2.. .X3...] is true. It is easy enough to implement this in both the proposition set analysis and the structured meaning framework:

310

Manfred Kriflca

(81) Who read what? a. Proposition set analysis:

{READ(y)(x) | xe PERSON, ye THING} = {READ(KF)(M), READ(TS)(M), . . . READ(KF)(J), READ(TS)(J),...}

b. Structured meaning analysis:

(X(x,y)[READ(y)(x)], PERSON χ THING)

In the proposition set analysis, both question constituents introduce alternatives, resulting in the set of propositions (81a). One simple way of implementing matching questions in the structured meaning approach is to assume that the background is a function over pairs (or triples, quadruples...), with the domain defined as the Cartesian product of the domains of the question constituents. But this does not explain the condition we found for matching questions. In particular, it does not rule out answers that consist of single pairs (cf. (69)) or answers in which the first question constituent is not D-linked (cf. (71)), and it does not capture the fact that one question constituent acts as a sorting key (cf. (76)). There are certainly ways to implement these additional restrictions, and we will turn to this issue below. But the question that arises here is: Why do natural languages not have questions of the type of Prolog queries as in (80)? I suggest that a database query like (80) is cognitively too complex to be carried out in one go, as it asks for two things simultaneously, and we can answer only one thing at a time. The query asks us to inspect the whole database, taking note of any pairs or triples that satisfy the condition expressed by the question. If η is the number of entities in the database, and x, is the i-th entity, then what (81) requires us to do is the following: (82) For i := 1 to η do: Is Xj a person? If yes: For j := 1 to η do: Is Xj a book? If yes: Did Xj read Xj? If yes: Print x„ xj! Such searches in two or more dimensions appear to be too demanding, at least when applied to the whole knowledge base. What seem to be licit cases of asking for more than two things at a time actually turn out to be cases of asking only one thing at a time. This is obvious for conjoined questions, which unfold into two or more questions: (83) a. Conjoined question: Who came, and when? b. Question 1: Who came? Answer: Mary. Question 2: When did Mary come? Answer: At noon. c. Conjoined answer: Mary, at noon.

For a Structured Meaning Account of Questions and Answers

311

Answering this question involves the following steps: (84) a. For i := 1 to η do: Is Xj a person that came? If yes: Print Xj and stop loop. b. For j := 1 to η do: Is Xj a time at which x, came? If yes: Print Xj and stop loop. What about quiz questions like (56) and (57)? They are about a particular event that is left unspecified with respect to one or more participants or parameters, and they are understood by the hearer as being about a particular event. This means that they can be answered without going through the whole database; rather, the hearer is asked to focus on a single, unique event, and specify the parameters left open in the question. Question (56) is answered as follows: (85) Identify in your database the unique event in which a French general invaded an Eastern-European empire. Print the name of the general, the empire, and the year of the event. Similarly, echo questions are about a particular event - a preceding utterance that was incompletely understood. What, then, about matching questions? I would like to argue that they, too, are answered by only one "thing", but that this thing is a function. Of course, if we see a function extensionally as a set of tuples, asking for more than one thing at a time and asking for a function would be the same. But I would like to restrict the cognitively relevant sense of "function" to mapping procedures from a given and identifiable domain to values, and claim that such procedures, though complex, can be considered one "thing" that we can ask for. To illustrate, answering the matching question (51 A) involves the following steps: (86) For i := 1 to η do: Is ί a function from the persons mentioned to (readable) things? If yes: Does it hold for every χ in the domain of fj that χ read fj(x)? If yes: Print fj and stop. Several observations about matching questions that we made earlier follow from this account. If cognitive functions require a given domain (a mapping procedure must apply to some set), then the question element that defines the domain must be Dlinked. The elements in the domain naturally are the "sorting key", as cognitive functions are nothing but a way of identifying certain entities by way of other entities. If cognitive functions are mapping procedures for a domain, then it follows that they should map each element of the domain to something, hence, have a distributive inter-

312

Manfred Krifka

pretation. And the domain should contain more than one element; a domain with one or zero elements would result in a degenerate function that is not "really" a function, just as a line or a point, which qualify as triangles in the mathematical sense, are not "really" triangles. If matching questions ask for a mapping procedure, then the question constituents should be used to construct this procedure. This is possible with the structured meaning account, as domain and range of the function are accessible. Currently, our theory would give us the following interpretation for Prolog queries: (87) Who read what?

(X(x,y)[READ(y)(x)], PERSON Χ THING)

This is a question that asks for more than one thing at a time and cannot be answered. But it can be transformed into a question that asks for a mapping procedure by using the following operators: (88) a. FUN(R) = XfVxfxe DOM(f) —» R((x, f(x)))], the set of functions f such that every χ in the domain of f stands in R-relation to f(x) b. FUN'(A χ B) = the set of functions from A to Β Applied to our example, we get the following result: (89) (FUN(X(x,y>[READ(y)(x)]), FUN'(PERSON x THING)), where FUN(X, such that for any aeD 0 f(a) = |oT,8[a/v1·

iv.

If ν e Varc and ae Exp, and ο is an occurrence of Vva, then |o|°'8 is defined iff |oT' 8 is. If it is defined, then |o|c·8 = n a€Do |o a | g[a/v] .

We may now go on and define, analogously to the above Kaplanian definition of "character of an expression", the character of an occurrence. This character may be called the meaning of the occurrence. But it is the meanings of expressions semantics is interested in. So what is the meaning of an expression in the present framework? If we learn the meaning of an expression we at least acquire the ability to understand arbitrary occurrences of that expression. Therefore I will define the meaning of α wrt. an assignment g to be that function ω from those pairs of occurrences ο of α and contexts c, such that |o|c,g is defined into contents of the appropriate type, such that if is in the domain of ω, then 0)(o,c) is |o|c'8.

3.

The Less Certain Principle

3.1

The Problem

Above I have provided for the possibility that Vc may be partial, that it may not define values for every occurrence ο. I have done so for a very good reason. Let us say "o is 7

Cf. von Stechow (1979:33If.). The main difference lies in the fact that Arnim's valuations are relativized to an expression a . This is meant to be an expression the occurrence to be evaluated occurs in (if it does not occur in a, the valuation will be undefined). The extra relativization leads to the puzzling possibility that what is intuitively the same occurrence may receive different interpretations relative to different occurrences it is a part of (regarded as expressions it occurs in). A further difference is that where we define, solely for the purposes of exposition, the simpler notion of "content relative to a context", Arnim gives a recursive definition of context-dependent meanings, with a resulting different technical treatment of undefinedness.

Learning French from a Dictionary

331

uttered in context c" if and only if there is an expression α such that ο occurs in α and α is uttered in c. Now the only cases where there is a non-arbitrary way to define Vc(o) are those where ο is uttered in context c. Then, and only then, can we uniquely identify an utterance u which may be taken to be the utterance of ο in the context. And this allows us (in most cases) to determine the values of a 0 , 10, and t 0 in that context, with the help of our general recipe on p. 10. On the other hand it is totally unclear how to determine Vc(o) in contexts where something different from ο is uttered (say, some item of another language) or, even worse, nothing is uttered at all. At least this holds for all conceptions of context I am aware of. Contexts may be identified with η-tuples of contextual parameters, with centered worlds or stages of possible individuals, or with utterances themselves. For all these conceptions it is possible to define the notion "o is uttered in context c"; but given the straightforward definitions for each case, it is still incomprehensible how to identify a 0 , 10, and t 0 for contexts in which ο is not uttered. (Like Arnim I will not settle the issue of what contexts really are, and I will take the notion "o is uttered in c" as a primitive.) So let us take Vc to be partial, defined only for those ο which are uttered in c. This will lead to the following fact about meanings: (PM)

The meaning of α is only defined for pairs of occurrences ο and contexts c, such that ο is uttered in c.

We may now define an utterance of an expression α to be a pair of an occurrence ο of a and a context c, such that ο is uttered in c. (So far I have only dealt with utterances in a less technical sense. In the following I will mostly deal with the technical concept defined just now. 1 hope no confusion arises; the two concepts are closely related and may be put to equivalent use.) Then we can restate (PM) in the following form: (PM)

The meaning of α is only defined for utterances of a.

Something quite like (PM) is adopted in Arnim's paper, too. On p. 330 he defines the notion "σ-meaning" (possible meaning of type σ). σ-meanings are functions from contexts, construed as world/time/place triples , into contents. He notes that σmeanings are partial functions and continues: This is so, because if a particular σ-meaning ω is the meaning of some expression a then ω is defined for at most those indices such that a is uttered in w at t and p.

(PM) is a very natural principle, given the utterance-dependence of reference. If the determination of content is all there is to meaning, and if the content of a word at a 8

Even if ο is uttered in context c, there may be cases where a 0 , 10, and t 0 are undefined, e.g., if the speaker changes during o; however, I have assumed that the speaker never changes during the utterance of a lexical item, so these cases can safely be ignored for the purpose of the definition of

Ve.

Manfred Kupffer

332

particular occasion is dependent on the properties of the utterance of the word at that occasion, then how could the meaning be applicable to occasions where the word is not uttered? - But now there is a problem.

The problem is that, according to (PM), no two different expressions could ever have the same meanings. This is so because no two different expressions have identical sets of utterances (this follows from Fact 2 above), therefore the domains of their meanings will differ. That no two expressions could have the same meaning is a somewhat appalling consequence. Ede Zimmermann, who has been willing to bear it, has called it the "less certain principle (LCP)". 9 But this seems to be a mild epithet. We simply know that some expressions are synonymous, for example, French je and English I, or French eau and English water, or J'aifaim and lam hungry. And we also know (SYN) Synonymy is sameness of meaning. So according to intuition and (SYN) there are cases of different expressions which have the same meaning. On the other hand our semantics implies that there are none. How should we react? • Not by defying the intuitions. They are deep-rooted. Furthermore, they have explanatory value. Firstly, that I and je are synonymous may be taken to explain why they are intertranslatable.10 (On the other hand, if synonymy and intertranslatability are equivalent (or even only equivalent ceteris paribus) this seems to be a fact that is itself in need of explanation. Also note that sometimes interpreters don't give synonyms, e.g., if they put what they translate in indirect speech.) But secondly the existence of cases of synonymy (or near synonymy) explains why it is that we can sometimes learn the meaning of a word from a bilingual dictionary. Most of the time lexicographers don't define literal meanings. Maybe they don't even intend to give synonyms. But if you take a certain entry in your dictionary to tell you that je and I mean the same, at least what you think this tells you is something true. If you rely on that information you have learned how the content of any arbitrary utterance of je (except in a quotation) is determined. So we should take the intuitions seriously. But how? • Not by trying to define a better notion of synonymy. Any definition other than that synonymy is sameness of meaning just misses the point. • Maybe by invoking the notion of extrapolation. Let me explain.

10

Zimmermann (1997:147). Versions of the LCP have also been defended by Nelson Goodman (1949,1953) and Catford (1965). Or so it is said - e.g., by Putnam (1988:29): "But meaning, we should recall, if it is anything, is what we try to preserve in translation."

Learning French from a Dictionary

3.2

333

Extrapolating MEANINGS

Maybe we could say / and ich are identical in meaning, but in another sense of "meaning", a sense that is not in conflict with, but derivable from occurrence-interpretation. You may object that speakers are not aware that there are different senses at issue. But in the study of meaning it ought to be a familiar phenomenon that there is no single precise theoretical notion that matches all the pre-theoretical intuitions. And often, this is so without the layman's awareness that there are different senses of "meaning" involved. Sometimes "meaning" is best explicated as truth conditions, sometimes as cognitive significance, sometimes as character, sometimes as propositional concept.11 The theoretician has analysed the layman's use of "meaning" into several different notions, and he has been compelled to do so because there is nothing that answers all the layman's intuitions, taken together. So there seems to be nothing wrong if we add just another notion which yields the intuitively correct predictions about synonymy. Let this notion be called MEANING. But, since we think occurrence-interpretation is essentially correct, these MEANINGs ought to be dependent on meanings. The LCP is a result of the fact that meanings of different expressions have different domains. The domain of the meaning of I is the set of utterances of I, the domain of the meaning of ich is the set of utterances of ich. A straightforward way to get rid of this problem would be to extend these domains. Why not take the union of the two sets as the common domain for the two MEANINGs? Only then do / and ich have even the chance to be synonymous. Still this would only help for I and ich, but not for je, so we should join all relevant domains. In effect, in order to avoid any acting ad hoc, we propose to take the same domain for every MEANING, namely the set of all utterances of arbitrary possible expressions. But now the problem arises of how the values of the respective MEANINGs for the extra elements in their domains are defined. There is a straightforward answer to this question, namely "by extrapolation". Simply extend the old functions to the new elements. Aren't they basically the same? In the case of I the function is basically: take the utterer. In the case of ich, again, the function is basically: take the utterer. Of course, the two functions are only basically the same, since they have different domains. But if we abstract from this, so the story of the extrapolationist goes, no difference remains. And extrapolation to arbitrary utterances is abstraction from this difference. Unfortunately the notion of "extrapolation" is particularly unpopular. Both Goodman's "new riddle of induction" (Goodman 1955: chapter 3) and "Kripke's Wittgenstein" (Kripke 1982) are based on a deep scepticism about it. It seems as if we can't specify rules by just pointing to a few samples and then add "go on in the same way." But that is just what I have been proposing! Now others have defended extrapolation. For example, David Lewis proposes to rescue extrapolation by appealing to naturality. He points to the fact that only some extensions of a given series may be called natural, and some are more natural than others. Therefore it is at least not completely undeter-

Cf. the variety of notions of content discussed in Perry (1997:599-609).

334

Manfred Kupffer

mined how a series might be extended, see, e.g., Lewis (1983, 1992). (Lewis claims that we need some notion of metaphysical naturalness here. Now what is metaphysically natural may not be accessible to the individual. But since we want to explain our faculty to extend a function from utterances of / to utterances ofje we need standards of naturalness that are within our control. I conclude that one needs a notion of psychological naturalness here.) One also needs naturalness in order to explain how it is that children can grasp the meaning of I in English from a few samples. Meanings are acquired by extrapolation, Goodman's and Kripke's scepticism notwithstanding. But if this is true, why say that first meanings are extrapolated and then, only afterwards, MEANINGs are extrapolated rather than that both are extrapolated in one fell swoop? We shouldn't have MEANING besides meaning. Let us have it instead. Note also that your possessing meanings (and MEANINGs alike) should not be regarded as you having the functions stored as infinite lists somewhere in your head. It should rather be regarded as you having certain semantic faculties. For example, possessing the meaning of / consists in part in the faculty to understand utterances of I. Possessing the MEANING of I consists in the former faculty and also in the faculty to learn the meaning ofje from a bilingual dictionary. Since we have no clue that the two faculties are acquired at different stages, and since the mechanism of acquiring them is the same, there is no reason to separate them. Summing up, resorting to extrapolation is not entirely misguided: we need extrapolation to explain language learning, or so I think. But it seems much more sensible to assume that our meanings have already been extrapolated than to extrapolate them for my purpose of explaining how it is that we can learn the meaning ofje from a dictionary. So finally I have reached the conclusion that my notion of meaning was too restricted after all.

3.3

Kaplanian MEANINGs

Kaplanian semantics fares better in this respect. This is so because Kaplan does not restrict his characters to utterances. The Kaplanian meaning of I that we sketched above, e.g., is totally expression independent. We can give German ich and French je exactly the same meanings, for instance. This is so, because here we are not required, for the purpose of evaluation of an arbitrary expression, to stick to those contexts where the expression is uttered. One can define the latter notion within Kaplan's framework. Let's say an expression a is uttered in context c iff in cw, c a counts as the speaker of a at time t and place p. Kaplan uses the term "utterance" for context/expression-pairs in which the expression is uttered (quite in agreement with our terminological decision above); but arbitrary context/α-pairs he calls occurrences of α in a context. (This decision may be a bit confusing in the context of the present paper.) Let us call those Kaplanian occurrences where nothing is uttered in the context at all or where the expression uttered is not the expression uttered in the context "impossible utterances". These have been branded as useless in Zimmermann (1997:145): "I cannot imagine why any theory would want to

Learning French from a Dictionary

335

make any (in principle untestable) predictions about what a sentence expresses in a (possible) situation in which it is not uttered." I don't think this critique is justified. First, we have already seen a decent motivation for impossible utterances: they are what enables Kaplan to have synonymy. The introduction of impossible utterances can simply be seen as a means of getting rid of the reference to the particular expression in the meaning of that expression. But what about the problem that predictions about impossible utterances are in principle untestable? What is meant here is presumably not an empirical test, but rather the kind of test we perform when summoning our intuitions about what an expression would express in twin earth. However, Zimmermann is right in saying that these intuitions (the "tests") never concern impossible utterances. Now, does this mean all choices of contents for impossible utterances are equal? Such that "for utterances of /take the utterer; take the addressee for any other occurrence in which something is uttered; take Manfred for all other occurrences" would be as good a rule for / as the above "take whoever counts as the speaker in the context"? Then, for any two expressions there would be (a) pairs of meanings which make them synonymous and (b) pairs of meanings which make them non-synonymous and both the (a) and the (b) assignments of meaning would be as equally correct. But this is not so, since synonymy itself is a fact of the matter. That / and je are synonymous is a fact of the matter since this explains, among other things, how we can learn the meanings of French words from a dictionary. Therefore the above crazy rule for I is incorrect, because it predicts that any Kaplanian rule for je that is in accordance with what we know of the content of utterances ofje disagrees with the above rule on those utterances. So Kaplan was on the right track. His meanings may be termed "already extrapolated". On the other hand his meanings are empirically incorrect, because they admit of no occurrence-dependence. The remaining question is how to find a semantics that has the virtues of the Kaplanian semantics without its vices.

4.

Simple Expressions

Now there seems to be a trivial solution to the problem at hand. Simply proceed as Kaplan did! Let us make our valuations total rather than partial and let the content of every occurrence be defined for all contexts. Then, our meanings could indeed be identical for I and je, because the domains no longer disagree. Unfortunately, it can't be all that simple. Such a solution would require our contexts to possess magical linguistic powers. How could a context be able to determine for any occurrence that does not occur in the context who counts as the utterer of that occurrence? The situation is different from the Kaplanian case, because at least on some conceptions of context, the distinguished items (agent, location, time, and world) can be directly extracted from the

336

Manfred Kupffer

context without reference to expressions or occurrences. (Of course Kaplan has to pay for this advantage in terms of applicability. It is not entirely clear in his framework how to determine for a given utterance what the context of that utterance is.) I will now show how to remove the expression-dependence from occurrence-interpretation. In this section I will concentrate on simple expressions. For example, we want I and je to come out synonymous, therefore the relevant meanings should share their domain and agree on the shared domain. As indicated above, I will construct a domain as big as possible by letting the meaning of I and je be defined for any utterance (in the technical sense) which is equipped with someone that counts as its speaker. Still, I am aiming at a generalization of our definition of meaning in the foregoing section. I can no longer define MEANINGS in terms of a prior definition of contents though, because now the content of an utterance will depend on the MEANING considered, e.g., has a different content if evaluated wrt. the MEANING of / or if evaluated wrt. the MEANING of here. Now what are the MEANINGS of I, here, now, and actually? As I have said, for some utterances what counts as the speaker, time, and place may not be defined. So I have to assume that the MEANINGS of I, now and here are truly partial. But for every utterance where they are defined, they will be defined by the familiar functions into contents. In particular, • If a c 0 is defined, |[/J(o,c) = a c 0 , |[/l(o,c) is undefined else. • If lc0 is defined [Aere]|(o,c) = l c 0 , [AereJ(o,c) is undefined else. • If tc0 is defined e |«OW](o,c)(p) iff Ep, [HOW1(O,C) is undefined else. • e lactuallyJ(o,c)(p)

5.

iff ep.

Complex Expressions

Let me now turn to the question of how the MEANINGS of complex expressions are determined. We will again want to extend the domain of the MEANING of a beyond utterances of a. This is so because (8)

I am now finished

and (9)

Ich bin jetzt fertig

should come out synonymous but wouldn't if the two MEANINGS had different domains - viz., the utterances of (8) and (9), respectively.

Learning French from a Dictionary

337

But now we face an immediate problem. It is the problem of defining relevant suboccurrences. Let me explain. Let me first say how we evaluate utterances of (8). Let I be the unique occurrence of I in (8). Now e [(8)](o,c) iff at w and t, [/](/,c) = a c , is finished. But if we extend the domain of [(8)1, the question arises of how to treat an utterance that is in the domain of [(8)1 without being an utterance of (8). We need to know where to apply the MEANING of I. What replaces reference to I here? So the problem is how to define, for any occurrence o' in (8), the sub-occurrence of ο which is relevant for the evaluation of o'. Given our two example sentences there is a very simple solution that suggests itself. Presented with an utterance of (9) one could take the relevant sub-context for the evaluation of the content of now to be the utterance of the item that has the same structural position in the sentence, namely jetzt. Following this line of reasoning, one would have to restrict synonymy to cases of intensional isomorphism. But this is far too restrictive. For example, it should at least in principle be possible to express the same meaning with a different number of words. Indeed, it is even tempting to say that the French sentence (10) Fini! is synonymous to (8) and (9) although it lacks structure. (Although later on I will find a reason to say that (10) is not synonymous to either (8) or (9), I will still use this example for purposes of illustration. It should be easy to substitute other examples of sentences which are synonymous, but yet not identical in structure.) You might try to find other solutions to the problem. I doubt that there are any. Anyway, I think that already the search is misguided. So far my extension of domains has been motivated by the observation that one can learn languages from a bilingual dictionary. If we are told that I and je are synonymous, we learn how to evaluate any arbitrary utterance (given we know who counts as the producer of ο in c). But if we are told that (8) and (9) are synonymous, things are different. Suppose you don't know German and are told that (8) and (9) are synonymous. Are you now able to determine the content of any arbitrary utterance of (9)? I don't think so. Even if you know who counts as the speaker of ο in c and the times that count, in c, as the times of all sub-occurrences of o, you still can't say what proposition has precisely been expressed unless you are told that now and jetzt are synonymous and hence, that the relevant sub-occurrence for now is o's sub-occurrence of jetzt. To sum up, the problem of defining the relevant sub-occurrences has no solution and ought to have no solution. But then we're stuck with our search after a notion of MEANING which allows different sentences to have the same MEANING, at least if we want these MEANINGs to be defined in a compositional way. Should we give in? In the next section I will propose a way out.

338

6.

Manfred Kupjfer

MEANINGs and Synonymy

This section comes in three parts. In the first subsection I will informally present the main ingredients of the recursive definition of MEANINGS which will then be given in the second subsection. That system will still suffer from the LCP, though. Only in the last subsection will I indicate how to take the MEANINGs defined in the second and obtain a notion of synonymy which does not suffer from the LCP.

6.1

Preliminaries

The upshot of the last section was that there is no principled way to define relevant sub-occurrences. Instead I want to propose that relevant sub-occurrences are stipulated, and that contents are assigned to utterances only relative to stipulations. Formally, we will treat stipulations as functions that map occurrences onto occurrences, e.g., the suboccurrences of (8) onto the sub-occurrences of (9). Now there are usually different possible stipulations for the same pair of sentences. For example, they might differ as to what is the relevant sub-occurrence for the evaluation of now in (9). Of course in this case there is only one correct choice, but you might not know which one. If you are told that (8) and (9) are synonymous and you only know English and no German, you might not know to which occurrence in (9) to apply the meaning of now. In effect, I am proposing that there is an additional component to linguistic knowledge which has largely gone unnoticed so far. While people are aware that in order to know a language you need to know its syntax and its meanings, they seldom pay attention to the fact that if they want to apply a meaning to an utterance they also need to know to which component utterances to apply which meanings. The reason that this is usually overlooked is that the answer is trivial in the case of one's own language. You only need this kind of information if you want to determine the content of an utterance of a foreign language, and the only thing you know about that utterance is that it is an utterance of a sentence which is synonymous to some sentence of your own language. By saying this is an additional component of linguistic knowledge, I also want to apologize for not treating it here in any formal detail. I will only define meaning proper, and therefore, I will not say what it means to know to which component utterances to apply which meanings. Consequently, my MEANINGs will almost never allow the determination of a unique content for any utterance. Even if we are evaluating an utterance of the sentence itself, it will be possible to choose different stipulations. Nevertheless, we should impose at least one constraint on the range of possible stipulations. It is that the relative order of indexicals in the expression to be evaluated should be respected by the stipulation. For example, if we want to evaluate

Learning French from a Dictionary

339

(11) Now is before now vis ά vis an utterance of (12) Jetzt ist vor jetzt 'Now is before now' we shouldn't be able to stipulate that the relevant sub-occurrence of ο for the evaluation of the first occurrence of now is the second occurrence of jetzt and vice versa. If we would, we could hardly distinguish (12) and (13) Jetzt ist nach jetzt 'Now is after now' neither could we account for the odd flavour of (12). 12 1 will therefore assume that, for every occurrence o, its sub-occurrences of constants are linearly ordered by c . 1 3 As usual, "OEO"' abbreviates "OEO' or o= o"'.

6.2

MEANINGS

Let me now introduce the formal system that embodies these ideas. The reader should be warned that the definitions have now reached a complexity that may make them not easy to read. For example, my MEANINGs will take four arguments before they yield a content. So the definitions are somewhat overcrowded with subscripts and arguments. This does not hold for partial valuations. These are defined as in 2.2. I will repeat the definitions for convenience. A partial valuation V is a partial function from the set Con into the set D, such that the following holds: • if c e Con a and V(c) is defined, then V(c) e D„, • VH(p) = WxI\p, • V(&Xp)(q) = pnq, • e V( • )(p) iff for all ve W, e p. • For no ae D^ V(exists)(a)=0 and if e V(/i-a/)(lXa), then e V(exists)(a). • There is a distinguished individual a*, a distinguished location I*, a distinguished time t*, and a distinguished world w *, s.t. V ( / ) = a * , V(here) = 1*, e V(«ow)(p) iff e p, and e V(actually)(p) iff e p.

I owe these examples to Ede Zimmermann (p.c.), as well as the whole point presented in the paragraph. Of course this relation should be formally defined, but I will skip the details of a rather obvious, but tedious definition, here.

340

Manfred Kupffer

Our present idea of context dependence is now simply that the distinguished items a*, 1*, t*, and w* are not fixed once and for all, but vary with the utterance, such that every utterance determines a valuation V . I will, as before, refer to the distinguished items a*, I*, t*, and w* of an utterance as a c 0 , lc0, t c 0 , and w* c , respectively. Now let us define our stipulations. Definition: For any expression a let a„ be the unique occurrence of a in a. Let α be an expression and ο be any occurrence, f is an a/o-map iff f is a function from the set of sub-occurrences of a^ into the set of sub-occurrences of o, such that if o' and o" are occurrences of indexical constants14 in a, then O'EO" iff f(o*)sf(o"). f is a stipulation if it is an ct/o-map for some expression a and occurrence o. Let us say expression a occurs in β iff there is some occurrence of a in ß. In the following I will try to evaluate an expression α wrt. to ß/o-maps for some expression β such that α occurs in β, the reason being that both the utterance evaluated and the stipulation made should stay constant during the evaluation of a complex expression, so stipulations should stay applicable even if we turn to subexpressions of the expression we have begun with. 1 will say that f is admissible for α and ο iff f is an ß/o-map for some expression ß, such that α occurs in β. Obviously, if f is admissible for α and 0, then it is admissible for yand o, where / i s any expression that occurs in a. Following our former use a / will be the unique immediate constituent of c^ which is an occurrence of β, i.e., the topmost occurrence of β in a. Now we can define, for arbitrary expressions a, stipulations f, assignments g, and utterances , such that f is admissible for a and o: [ aJ f ' 8 (o,c), the content of a wrt. f, g, and . 1.

Let o' = f(Oo). If α is a constant, then [ a] f ' 8 (o,c) = v " represents the greater-than-or-equal-to relation. This is possible if we assume that every degree is represented by a real number.

350

Cecile Meier

egy can be applied to derive the truth conditions of split-antecedent result clause constructions.

2.

Multihead Comparatives

Chomsky (1981:81) was the first to group multihead comparatives and split-antecedent result clause constructions together. He disputes that the surface structures of sentences such as (1) are derived by extraposition of the that-clause from some so-phrase internal position and proposes "a rule of construal in the LF component that associates [the that-clause and the so-phrases]." Von Stechow (1984a:43) provides an explication of how this rule of construal may be understood: as a rule of reconstruction. For a sentence like (6a), von Stechow proposes the (rudimentary) Logical Form in (6b). (6)

a. More dogs ate more rats than cats ate mice. b. More dogs [than cats ate mice] ate more rats [than cats ate mice]

The underlying principle is the duplication of the than-clause, each copy of which is then associated with one comparative morpheme. This process of reconstruction can only be captured as an instance of syntactic lowering of (a) the material base-generated in sentence final position and (b) its copy. I favor an account of this process of construal in terms of base-generation of all copies of the than-clause as complements of the degree phrases. Only one copy is raised to a sentence final position to derive the surface structure. This movement has to have all properties of extraposition; see Müller (1995) for an account of extraposition as movement to the right that is subject to a version of a chain uniformity condition. All other copies are deleted under identity and recovered at the level of LF, ready for interpretation.3 Gueron & May (1984) propose a similar account in terms of deletion and extraposition for split-antecedent result clause constructions. I will present their account in section 3.2 in more detail. Their main hypothesis is that extraposition is not an instance of move-α in the proper sense, contra Chomsky's original assumption. The sentence in (6a) implies, according to von Stechow (1984a), the conjunction of two comparisons, as in (7). (7)

3

4

The number of dogs that ate rats > the number of cats that ate mice and the number of rats that were eaten by dogs > the number of mice that were eaten by cats.

Note that there are two kinds of comparative constructions: so-called elliptic and non-elliptic structures. Only elliptic structures allow for multiple comparison; see Chomsky (1981) and Haider (1995:247) for more details on non-elliptic comparatives. ">" represents the greater-than relation.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 351 How can this paraphrase be derived compositionally? Preliminarily, I assume an extensional typed language in the style of Gallin's (1975) language Ty2 to express truth conditions. In addition to the common basic semantic types e (for entities), t (for truth values) and s (for possible worlds), I propose introducing a type d for extents. The elements of possible denotations in D j are all sets of degrees. The inputs for semantic interpretation are phrase structure trees that conform to the requirements of GB-Theory. The interpretation of terminal nodes is specified in the lexicon. The interpretation of unary nodes is equal to the interpretation of their daughter nodes. The branching nodes are interpreted by functional application or conjunction, by λ-abstraction and/or by a mechanism of existential closure. Logical representations are interpretable if every node is interpretable. To secure a strictly compositional interpretation, I assume with von Stechow (1993) that traditional Logical Forms must be enriched with arguments that are implicit at the surface of the sentences (all predicates take an extra world argument, for instance). These enriched syntactic representations are called Transparent Logical Forms (TLFs). The comparative morpheme -er has the meaning in (8). In this approach, the comparative is assumed to be a quantifier that takes a set of degrees (the denotation of the fAaw-clause) to form a nominal.5'6 (8)

[ - e r f = f: D«d,t>,«d,t>,t>>

For any extent predicates E,, E2 e D: f(E2)(E,)= 1 iff MAX(£/) > M AX(£ 2 ) Furthermore, I assume that more is a lexicalization of the adjective many and the comparative morpheme -er (Bresnan 1973). The iAaw-clause is a complement of the comparative morpheme (von Stechow 1993). The constituents headed by the comparative morpheme are base-generated in an adjunct position of the many-AP and moved to a sentence initial position (von Stechow 1993). With these assumptions in mind, we may state the elaborate LF-version of (6a) as in (9).7

5

The truth conditions presuppose an extent semantics in terms of von Stechow (1984b); see Kennedy (1999) and Meier (2000) for an introduction. Von Stechow (1984a) already argued for the fact that both components of the comparison must be a maximum of a set of values (see Heim 2001 for a recent discussion of additional data). The definition is also different from von Stechow's original version (1984a), in which the than-clause expresses a definite description and the comparative morpheme itself is not mobile. That is, the scope constellations in my approach are not the same as in von Stechow's original approach to multihead comparatives.

6

I am using the notion of a quantifier in a broader sense. The comparative denotation does not meet some general conditions on ordinary quantifiers (i.e., determiner denotations). It is, for example, not conservative. Compare this LF to the one proposed by von Stechow (1984a:44).

7

[CP [[NP2 more dogs][Cp than4 e 4 many cats ate e many mice]] 2 [CP [[NPI more rats][Cp than3 e many cats ate e 3 many mice]]!... [yp t2 ate ti]]] In (i), the plural dogs has wide scope with respect to the comparative morpheme headed by the object NP more rats. (')

352 (9)

Cecile Meier [cpfoegP2-er [cp than4 e4 many cats ate e many mice]]4 [Cp [Degpi -er [cp than3 e many cats ate e 3 many mice]]3 [cpi [np2 ti many dogs]2 [cp [npi t3 many rats]i ... [vpt2 atet,]]]]]

In the first step, the plural NPI and NP2 are adjoined to CP leaving the traces ti and t2, respectively. These movements are type-driven and an instance of genuine quantifier raising. Next, DegPl and DegP2 are raised to a sentence initial position. This movement is also type-driven and leaves the traces t3 and tj· 8 The NPs e3 many mice, and e4 many cats occur with and without superscripts. If the degree phrase is base-generated in the subject of the main clause, the empty element of the subject of the /Aa«-clause is coindexed with than. If the degree phrase is base-generated in the object of the main clause the empty element of the object of the iAa«-clause carries the superindex of than. In other words, which empty element of the than-clause carry the superindex is determined by the syntactic function of the constituent that heads the degree phrase in the main clause. These superscripts on the empty element e control a binding mechanism that is central for the derivation of the multihead-problem. The superscripts have a meaning only in the scope of a potential binder than with the same superscript. In the following, I will show in more detail how to derive the formula in (10) from the LF in (9). The derivation of the DegPs and the NPs are of particular interest. The rest is routine calculation. (10) -er^n*.3d[n*-many cats ate d-many mice])(Xn.-er^m*.3d[d-many cats ate m*-maiiy mice])(Xm.n-many dogs ate m-many rats)) The meaning of NP2 t4 many dogs in (9) may assumed to operate on a set of individuals. The interpreted by the variable η over extents that denotation of the NP and the denotation of the NPI ti many rats analogously.

8

be derived as in (11). The plural NP is trace t4 of the comparative movement is gets bound later in the derivation. The AP are conjoined. I represent the plural

The exact properties of this type of movement are not totally clear and will be left to future investigation.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 353 (11) a. [np2 t4 many dogs] b. NP2

XQ3XflX| > η & dogs(X) & Q(X)]

D ® t c « e , t > , t > , « < e t> t>

t » ',t>

λΡλΟΒΧ[Ρ(Χ) & Q(X)]

DegPd t4 n

A i « e t> t » f(X)(Y) = 1 iff VxVy[x e X and y e in Υ => χ ate y], The logical representation of the NPs is derived analogously to (11) above.

354

Cecile Meier

(12) a.

[OEGP2

-er [CP than4 e4 many cats ate e many mice]]

b. DegP«d,t>,t> er(Xn*3d.n*-many cats ate d-many mice)

The process of existentially quantifying all free variables at some point of the derivation is the same kind of mechanism that plays a key role in the interpretation of asymmetric donkey binding; see, e.g., Kamp & Reyle (1993:314) for an implementation of such a mechanism in a different framework.11 This existential closure is central for capturing the following parallelism observed by von Stechow (1984a): a ίΑαη-clause associated with the subject dogs of the main clause denotes a set of cardinalities that the plural subject cats of the ίΑα/j-clause may have. And, a /Aa«-clause associated with the object rats of the main clause denotes a set of cardinalities that the plural object mice may have.12 The derivation of the DegPl is analogous to the latter representation.

I thank the anonymous reviewer for pointing out this fact. In von Stechow's framework the than-c\ause [ C P than cats eat mice] just expresses the degree predicates λη.η cats eat mice or Xm.cats eat m mice. That is, he treats one plural as a bare plural and therefore does not take into consideration the binding mechanism proposed here.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 355 The formula in (10) may be interpreted as in (13). The key characteristic of this interpretation is that one comparison is embedded within the first component of the other comparison. 13 (13) a. [[-er(Xn*.3d[n*-many cats ate d-many mice])( λη.-βΓ(λπι*. 3d[d-many cats ate m*-many mice])( ληι.η-iiiany dogs ate m-many rats))] = 1 iff (by the denotation of the comparative -er) b. [the max n:[the max m: n-many dogs ate m-many rats > m*] > n*] where: m* - the maximal m*: 3d[d-many cats ate m*-many mice] and n* = the maximal n*: 3d[n*-many cats ate d-many mice]14 The next task is to show that (13b) indeed implies something like the paraphrase in (7). We may reformulate the occurrence of "the max n" containing a ι-operator, as in (14), on the basis of the definition of the maximality operator. (14) a. in[[the max m: n-many dogs ate m-many rats > m*] & Vn'[the max m: n'-many dogs ate m-many rats > m* => n' < n]] > n* b. m[[the max m: n-many dogs ate m-many rats > m*] & η is maximal] > n* Russell's theory of definite descriptions provides us with a method for translating formulas containing a ι-operator into formulas containing quantifiers familiar from predicate logic. (15) a. 3n[Vn'[[the max m: n'-many dogs are m-many rats > m*] & n' is maximal η' = η] & η > η*] b. 3n[n is unique & η is maximal & [the max m: n-many dogs ate m-many rats > m*] & η > n*]

Von Stechow's version of the interpretation of this sentence may be rewritten as in (i). (i) [the max n: (n-many dogs) [the max m: (m-many rats) (X ate Y) > m*] > n*] where: m* = the maximal m*: 3d[d-many cats ate m-many mice] and n* = the maximal n*: 3d[n*-many cats ate d-many mice] The anonymous reviewer observed that the scope relations are equivalent in von Stechow's version and my version - basically because both kinds of scope taking operators involved (maximum and group quantification) are union operators. This is only true if we assume, as in the intended reading, that the plural n-many dogs does not distribute over m-many rats. I will discuss scope interaction between the plurals and degree operators later in this section in more detail. The maximal η: φ" denotes the maximum of the set ληφ".

356

Cecile Meier

By using the explicit definitions for maximality and definite description for the expression "the max m", we derive the formula in (16a). (16a) is equivalent to (16b) since m is not free in the conjuncts "« is maximal" and is unique". (16) a. 3n[n is unique & η is maximal & 3m[m is unique & m is maximal & nmany dogs ate m-many rats & m > m*] & η > η*] b. 3nBm[n is unique & η is maximal & m is unique & m is maximal & n-many dogs ate m-many rats & m > m* & η > n*] Therefore (13) is true if there is exactly one maximal η and exactly one maximal m such that «-many dogs ate /n-many rats and m is greater than m * (the cardinality of mice eaten by cats). And, η is greater than η * (the cardinality of cats eating mice). In other words, these truth conditions imply that there are maximal cardinalities of dogs (n) and rats (m) such that «-many dogs ate m-many rats and that m outnumbers the cardinality of mice eaten by cats and that η outnumbers the cardinality of cats that eat mice. This is the paraphrase proposed by von Stechow in order to capture the denotation of multihead comparatives. It is false if three dogs share a rat and one cat eats two mice. Hendriks (1994) criticizes the approach of von Stechow (1984a)15 by pointing out that (some) constructions with "non-identical comparative operators" (i.e., -er is different from more in this account) are less acceptable than constructions with identical comparative operators. Such constructions include constructions with negative polar adjectives as in (17a) or predicative constructions as in (17b). (17) a. ??Fewer dogs ate more rats than cats ate mice, b. More doors are higher than windows are wide. Hendriks (1994) claims (i) that von Stechow's analysis may not predict the marginality of (17a) and (ii) that von Stechow's analysis gives false truth conditions for examples like (17b). These claims, however, do not seem to apply to my version of von Stechow's analysis. The truth conditions for (17a) imply that there is a minimal cardinality of dogs («) and a maximal cardinality of rats (m) such that «-few dogs ate m-many rats and that m outnumbers the cardinality of mice eaten by cats and that η is outnumbered by the cardinality of cats that eat mice. The fact that (17a) is somehow odd is just a reflex of the fact that it is hard to imagine a situation in which the comparison construction might be relevant. Let us look at a parallel example. If we think of organizing a birthday party for our 6-year-old child and if we are trying to calculate the right amount of candies

She argues for the generalization that a sentence may not contain more than one sentence-internal comparison but may contain infinitely many discourse comparatives. In discourse comparatives as in John makes people prettier the standard of comparison is supplied contextually.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 357 and pizzas we have to order dependent on the number of boys and girls that are invited, we might utter (18) without difficulty. (18) Last time, fewer boys ate more pizzas than girls ate candies. But (18) has the same structure as (17a) and is fully comprehensible. It is defined if there is exactly one minimal η and exactly one maximal m such that « - f e w boys ate mmany pizzas and m is greater than m * (the number of candies eaten by girls). And, η is smaller than η * (the number of girls that ate candies). For (17b), Hendriks (1994) claims that the meaning predicted by von Stechow's theory is as given in (19). (19) below, however, clearly does not meet our intuitions. (19) The number of doors that are high is greater than the number of windows that are wide and the height of doors is greater than the width of windows. The problem with this paraphrase is that (19) refers to a standardized height for doors and a standardized width for windows, as Hendriks (1994) points out. It is, however, unclear to me how such a paraphrase may be derived in the framework o f von Stechow. Hendriks does not go into details and von Stechow himself does not discuss examples of this sort. Let me illustrate how to calculate the truth condition of this sentence in the framework presented in this paper. It will be shown that (17b) is ambiguous (due to scope interaction between the plural and the degree operator associated with the adjective wide) and that it has only restricted interpretations in the reading with wide scope of the degree operator. I assume that (17b) may be represented by the LF given in (20). The degree operator associated with wide has wide scope over the plural N P ti-many doors. (20) [cp [üegP2 -er [cp than3 e3 many windows are e wide]]3 [cp begPi -er [cp than1 e many windows are e1 wide]]i [Cp [NP2 t3 many doors]2 [CP [vp h is ti wide]]]]] If we evaluate the LF-structure given in (20), according to our assumptions we get the truth conditions in (21), in analogy to the cats-and-dogs example. (21) [the max n: [the max d: n-many doors are d-high > d*] > n*] where: d* = the maximal d*: 3n[n-many windows are d*-wide] and n* = the maximal n*: Bd[n*-many windows are d-wide] Let us concentrate on the embedded comparison. Here, I assume that the predicate be d-high distributes over the plurality (in the sense o f Link 1983) denoted by n-many doors. Therefore, (22a) - the first component o f the embedded comparison - is

358

Cecile Meier

equivalent to (22b).16 (22) a. the max d: n-many doors are d-high b. the max d: 3X[doors(X) & |X| > η & Vx(x e X

χ is d-high)]

Now, assume that the doors in the relevant domain all have a different height and that there are relevant windows. Under this assumption, (22b) gives us the height of the highest door and the cardinality of the set containing the highest door is 1. The only value that is admissible for the number of relevant windows is zero. And this conclusion leads to a contradiction with our assumption that there are windows. If there were no windows, d* would not be defined. Therefore, (17b) is only defined if there are doors with the same height and it is true if the group with the highest height is higher than the widest window if the cardinality of this group outnumbers the cardinality of a group of windows. This is predicted to be a possible reading of the sentence in question. There is another reading that all the doors are higher than all the windows are wide (or in other words: that the smallest door is higher than the widest window) and that there are more doors than windows. In order to derive this reading, we would have to assume that the plural n-many doors and in particular the distributive operator that hides universal quantification has wide scope with respect to the degree operator associated with the adjective. This reading has the truth conditions in (23). 7 (23) [the max. n: 3X[doors(X) & |X| > η & Vx(x e X [the max. d: χ is d-high > d*])] > n*] where: d* = the maximal d*: 3n[n-many windows are d*-wide] and n* = the maximal n*: 3Y[windows(Y) & |Y| > η & V χ (χ e Y => 3d[y is d-wide]) Hendriks (1994) does not discuss a suitable paraphrase for the sentence in (17b). She only discusses paraphrases that the sentence is not allowed to have. Furthermore, Hendriks cannot explain why the sentence is acceptable for some people. My explanation for the variations in acceptability is that the sentence is not acceptable because its evaluation is not defined when we have a situation in mind where all doors have different heights and some windows are wider than some doors are high. But it is defined (and true) in other situations.

See Heim (2001), Schwarzschild & Wilkinson (1999) and Rullmann (1995) for a related discussion with respect to the scope interaction of the universal quantifier and the degree operator. I refer the reader to an analysis in terms of an interval semantics for degrees as proposed in Schwarzschild & Wilkinson (1999) for a derivation of this reading without movement of the plural and the distributive operator. However, these authors do not consider the derivation of multiple comparatives to back their analysis.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 359 In the following, I will show how a similar analysis may be applied to result clause constructions with split antecedents. First, I will introduce a semantics for result clause constructions with a single occurrence of so. The version proposed here is different from the version proposed in Meier (2000).18 Whereas I previously assumed that constructions with so and the like are cases of semantic ellipsis, I will argue here that they are cases of syntactic ellipsis.19 This solution is necessary in order to capture the meaning of the split-antecedent cases correctly.

3.

Result Clause Constructions

3.1

Single so

Basically, result clause constructions are modalized constructions. As in Meier (2000), I assume that the sentential complement, i.e., the result clause, is (explicitly or implicitly) modalized. Implicit modals tend to have universal force and are usually interpreted with respect to a purely epistemic conversational background. This generalization remains unexplained but may be observed for genuine conditionals as well (Kratzer 1978). The modalization process is a syntactic process and takes place only if the result clause is not explicitly modalized. Modals in result clauses are considered to be 4-place relations, as if they occurred in a conditional. They relate a world, a conversational background, and two propositions. The type ρ abbreviates the type of propositions, and the type h of conversational backgrounds abbreviates the type . A modal with universal force, for example, has the truth conditions in (24). I am adopting here the standard version of modal interpretation of Kratzer (1991) in order to keep matters as simple as possible. However, in Kratzer's original version the conversational background is introduced as a parameter on the interpretation function and not as an argument of the modal. (24) [must] = f: D m*] > n*] where: m* = the min m*: 3d[if d-many people like m*-many pictures, the show will go on, given the regulations] and n* = the min η*: 3d[if n*-many people like d-many pictures, the show will go on, given the regulations]

Reconsider our situation discussed in the introduction. We can show that Chomsky's sentence (33a)/(l), respectively, is true if we assume that, in fact, 5000 people liked 153 pictures out of 170 and the board of the exhibition agreed to prolong the exhibition if more than 4000 people liked more than 150 pictures ("the regulations"). 5000 people (the max. n) is more than 4000 people (the min. η*) and 153 pictures (the max. m) is more than 150 pictures (the min. m*). They also predict (33a) to be false if 5169 people only liked one and the same picture since 1 picture is not more than 150 pictures (the minimally admissible number of liked pictures). Let us look at another example. For Liberman's example in (4) repeated here in (36a), we derive the truth conditions in (36b) if we assume that the ί/ιαί-clauses are (implicitly) modalized by must and if we assume that the sentence is interpreted with respect to a realistic background.26 (36) a. John hit his car so hard with such a big hammer that it started. b. [the max n: [the max m: J. hit his car η-hard with an m-big h. > m*] > n*] where: m* = the min m*: 3d[if John hits his car d-hard with an m*-big hammer, the car must start, given the facts] and n* = the min η*: 3d[if John hits his car η *-hard with an d-big hammer, the car must start, given the facts] Assume a situation in which we have five sizes of hammers. If John hits his car with the hammer of size 5, the car will be ruined no matter how hard he hits it. Moreover, assume that John's physical abilities allow him to hit his car with a force between 1 and 4 newtons. If he hits the car with 4 newtons of force, the smallest hammer is sufficient to start the car. If he hits it with the second smallest hammer he has to hit his car with at least 3 newtons to get it started. If he takes the second biggest hammer, he has to hit with at least 2 newtons and for the biggest hammer the force may be only 1 newton.27 Now, assume that John actually hit his car with 2 newtons of force and with a hammer of size 2. In this case, we want to predict that our sentence is false in the given scenario. Either the hammer is too small or the force is too small.

26

27

The LF is analogous to the one of the previous sentence. Interpreting the LF we use the usual semantic rules plus existential closure. We may think of these conditionals as causal laws. In reality these laws are more complicated.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 367 In order to determine truth conditions of our sentence in this scenario, we first determine the standards of comparison m * and η *. If we choose the maximally possible force 4, the minimal size of the hammer (m *) such that the car still starts will be 1. And if we choose the biggest hammer, the minimal force (n *) we are allowed to use to get the car started will be 1 newton, according to the facts. However, since m* and n* are smaller than the actual force and the actual hammer size the sentence is predicted to be true, contrary to our intuitions. What went wrong? The answer to this question is the following: In determining the standards of comparison we are not allowed to choose the value of the existentially quantified variable hypothetically. We are obliged to choose the actual hammer size in order to determine the minimal possible force and we have to choose the actual force with which the car was hit by John in order to determine the minimally possible hammer size in question. In other words, we are not looking for the absolute minimum in determining the standard of comparison for (36), but the minimum depending on actual values. Technically, this restriction may be introduced as a restriction on the world argument of the adjective: the world argument of the adjective whose extent argument is existentially quantified over must remain free at the level of the interpreted LF. The values for m* and n* are therefore the following.28 (37) a. m* = the min m*: 3d[must@g(h)(Xw.the car@ startedw)(Xw.John hitw his car@ d-hard@ with an m*-bigw hammerw)] b. n* = the min η*: 3d[must@g(h)(Xw.the car@ startedw)(Xw.John hitw his car@ n*-hardw with a d-big@ hammer@)] With this clarification in mind, we may determine the standards of comparisons m * and n* as follows: If we choose the actual force 2, the minimal size of the hammer such that the car still starts (m *) is 3 according to the causal laws. And, if we choose the actual size of the hammer (2), the minimal force we are allowed to use to get the car started («*) is 3 newtons. The minimal hammer size 3 is not smaller than the actual hammer size and 3 newtons is not smaller than the force actually used. Therefore the sentence is formally predicted to be false and this prediction meets our intuition. Finally, note that not all constructions with more than one so-phrase have to be interpreted as split-antecedent result clauses. Consider example (38a). The truth conditions of the split-antecedent reading may be schematized as in (38b). I assume that the conversational background is epistemic.

I assume here that the world arguments of the adjectival, nominal or verbal predicates are represented as subscripts. The parameter represents the world of utterance.

368

Cecile Meier

(38) a. So many people ate so many pizzas that they ran out of yeast at Luigi's. b. [the max n: [the max m: n-many people ate m-many pizzas > m*] > n*] where: m* = the min m*: 3d[if d-many people eat m*-many pizzas, they must run out of yeast, given the rules for pizza production and what we know] and n* = the min η*: 3d[if n*-many people eat d-many pizzas, they must run out of yeast, given the rules for pizza production and what we know] Consider a situation (a) where 20 people ate 50 pizzas, (b) where Luigi forgot to order new yeast, (c) where we know that Luigi still had 250 g of yeast left and (d) where we need 5 g of yeast for one pizza. In such a situation it does not matter how many people ate the pizzas. Our truth conditions imply that the cardinality of pizzas that were, in fact, eaten by people outnumbers (or is equal to) some m*. Now since Luigi has only 250 g of yeast left, he can produce at most 50 pizzas. Therefore m * equals 50. The minimal number of people that can eat 50 pizzas is 1 person. Therefore, the outer comparison cannot be false if the set of people who ate pizzas is not empty. The truth conditions of the single-so reading may be schematized as in (39), where g is any variable assignment. The first appearance of so receives a deictic interpretation. (39) [the max m: g(ni)-many people ate m-many pizzas > m*] where: m* = the min m*: [if g(ni)-many people eat m*-many pizzas, they must run out of yeast, given the rules for pizza production and what we know] In sum , result clause constructions with multiple occurrences of so may be interpreted along the lines of multihead comparatives. Multiple comparatives and split-antecedent result clause constructions only differ with respect to (a) the determination of the standard of comparison (minimum vs. maximum) and (b) the characteristics of the comparison relation (ordering "greater-than" vs. "greater-than-or-equal-to")· In order to derive the interpretable logical forms for multiple-ίο constructions, we have to multiply the that-clause. (40) shows that it is (at least marginally) possible to have more than one result clause in constructions with more than one so-phrase. This is predicted under the cur29 rent assumptions. (40) So many lectures that we had no time to attend them all were given by such boring professors that we didn't want to attend them all.

The anonymous reviewer suggested that (40) might have a distributive reading so that each of the professors gave so many lectures. This reading may be derived by giving the plural professors and the distributive operator wide scope with respect to the degree phrase so that we had no time to attend them all. See the discussion in section 2 for a clarification of this point.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 369 This sentence is assumed to mean that the number of lectures that were held by boring professors was responsible for the fact that we had no time to attend all lectures. And the reason for the fact that we did not want to attend them all is the teaching skills of the professors that gave the lectures. More specifically, the truth conditions may be captured by (41). Here, I assume that the modal of the first that-clause is evaluated against a buletic conversational background and the second against an epistemic conversational background. (41) [the max n: [the max m: n-many lectures were given by m-boring professors > m*] > n*] where: m* = the min m*: 3d[if d-many lectures are given by m*-boring professors, we don't want to attend them all, given what we want] and n* = the min η*: 3d[if n*-many lectures are given by d-boring professors, we have no time to attend them all, given what we know] Result clause constructions with more than one so-phrase and more than one thatclause in sentence final position are not grammatical, as shown in (42). (42) *So many books have been published by so many authors recently that I haven't been able to read them all that I've run out of money to buy them. (Gueron & May 1984:27) Therefore, there seem to be restrictions on the number of adjunctions of the same type of clauses in the sentence final position. I have no explanation for this fact; see Haider (1995) for discussion on multiple extraposition. This example does not show that splitantecedent constructions only have one result clause at LF, as Gueron & May (1984) claim.

4.

Conclusion

The current derivation of the meaning of split-antecedent result clause constructions relies on the following assumptions: (1) There is one ίΑαί-clause associated with each occurrence of so, unless so has a deictic interpretation. (2) The sentential complement of a result clause construction is assumed to represent a hidden conditional. (3) This hidden conditional is recovered syntactically and not semantically, as previously assumed (e.g., Meier 2000). (4) The antecedents of the conditionals are open propositions that are identical to the proposition expressed by the main clause. (5) An indexing mechanism (superscripts) on that and some extent variable of the antecedent of the (hidden) »/•clause determines the binding properties of the variables. (6) Variables with superscripts get bound by λ-abstraction. (7) Variables without superscripts get bound

370

Cecile Meier

by an existential quantifier. The superscripts depend on the syntactic function of the first maximal phrase that dominates the degree phrase. In other words, the maximal phrase that dominates a degree phrase in the main clause and the phrase that contains an extent variable bound by λ-abstraction in the antecedent of a hidden conditional have the same syntactic function. These assumptions enable us to derive the meaning of split-antecedent result clause constructions compositionally and in analogy to the derivation of the meaning of multihead comparatives.

Acknowledgments I would like to thank Irene Heim and Stephan Hirzel for extensive discussion on the topic and an anonymous reviewer for his insightful and clarifying comments and criticism. Erez Levon was so kind as to correct my English.

References Bresnan, Joan W. (1973). Syntax of the Comparative Clause Construction in English. Linguistic Inquiry 4. 275-343. Büring, Daniel & Katharina Hartmann (1995): All Right! In: Uli Lutz & Jürgen Pafel (eds.) Extraction and Extraposition. Amsterdam/Philadelphia: Benjamins. 279-311. Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foris. Gallin, Daniel (1975). Intensional and Higher Order Modal Logic with Applications to Montague Semantics. Amsterdam: North-Holland. Gueron, Jacqueline & Robert May (1984). Extraposition and Logical Form. Linguistic Inquiry 15:1. 1—31. Haider, Hubert (1995). Downright Down to the Right. In: Uli Lutz & Jürgen Pafel (eds.) Extraction and Extraposition. Amsterdam/Philadelphia: Benjamins. 245-271. Heim, Irene (this volume). Degree Operators and Scope. Heim, Irene & Angelika Kratzer (1998). Semantics in Generative Grammar. (Blackwell Textbooks in Linguistics 13.) Maiden, MA: Blackwell. Hendriks, Petra (1994). Multiple Head Comparison and Infinite Regress. In: Joel Ashmore Nevis & Vida Samiian (eds.) Proceedings of the Western Conference on Linguistics (WECOL 92). Vol. 5. Department of Linguistics, California State University, Fresno. 117-131. Kamp, Hans & Uwe Reyle (1993). From Discourse to Logic. Dordrecht: Kluwer.

Multihead Comparatives and Result Clause Constructions with "Split Antecedents " 371 Kennedy, Christopher (1999). Projecting the Adjective: The Syntax and Semantics of Gradability and Comparison. (Outstanding Dissertations in Linguistics.) New York: Garland Publishing, Inc. Kratzer, Angelika (1978). Semantik der Rede: Kontexttheorie, Modalwörter, Konditionalsätze. Königstein: Scriptor. Kratzer, Angelika (1991). Modality. In: Arnim von Stechow & Dieter Wunderlich (eds.) Semantik: Ein internationales Handbuch zeitgenössischer Forschung. Berlin: de Gruyter. 639-650. Liberman, Mark Y. (1974). On Conditioning the Rule of Subj.-Aux.-Inversion. In: Ellen M. Kaisse & Jorge E. Hankamer (eds.) Papers from the Fifth Annual Conference of the North Eastern Linguistics Society. Harvard University, Cambridge, MA. 77-91. Link, Godehard (1983). The Logical Analysis of Plurals and Mass Terms. In: Rainer Bäuerle, Christoph Schwarze & Arnim von Stechow (eds.) Meaning, Use and Interpretation of Language. Berlin: de Gruyter. 302-323. May, Robert (1985). Logical Form: Its Structure and Derivation. Cambridge, MA: MIT Press. Meier, Cecile (2000). Konsekutive Konstruktionen und relative Modalität. PhD dissertation, University of Tiibingen. Müller, Gereon (1995). On Extraposition and Successive Cyclicity. In: Uli Lutz & Jürgen Pafel (eds.) Extraction and Extraposition. Amsterdam/Philadelphia: Benjamins. 213-243. Müller, Gereon & Wolfgang Sternefeld (1993). Improper Movement and Unambiguous Binding. Linguistic Inquiry 24. 461-507. Rullmann, Hotze (1995). Maximality in the Semantics of Wh-Constructions. PhD dissertation, University of Massachusetts, Amherst. Schwarzschild, Roger & Karina Wilkinson (1999). Interval Semantics. Ms, Rutgers University, New Brunswick. Stechow, Arnim von (1984a). Comparing Semantic Theories of Comparison. Journal of Semantics 3. 1-77. Stechow, Arnim von (1984b). My Reaction to Cresswell's, Hellan's, Hoeksema's and Seuren's Comments. Journal of Semantics 3. 183-199. Stechow, Arnim von (1993). Die Aufgaben der Syntax. In: Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld & Theo Vennemann (eds.) Syntax: Ein internationales Handbuch zeitgenössischer Forschung. Vol. 9:1. Berlin: de Gruyter. 1-88.

Narrowing down the ExtendedNow

Renate Musan

1.

Introduction ... die Wolkenkratzer ragen nicht mehr als schwarze Türme vor der gelben Dämmerung, nun hat die Nacht gleichsam ihre Körper verschluckt, und was bleibt, sind die Lichter darin... ('the skyscrapers do not anymore like black towers loom in front of the yellow twilight, now the night has as it were their substance erased, and what remains are the lights therein') - Max Frisch, Stiller -

What do these lines speak about? The adverb nun makes it very clear - it is the situation "now": the silhouette of the skyscrapers is no longer visible in the evening light. But it seems no less clear that the event described - the night's erasing the silhouette of the skyscrapers - is in the past. In a way, therefore, the text speaks about the now and the time which leads up to this now - about an "ExtendedNow". It is the perfect which gives the sentence this particular temporal contour. In his comprehensive review of various theories of the English perfect, McCoard (1978) characterizes approaches which take this fact to be the central property of the perfect as ExtendedNow theories, and he favors them over other, competing analyses, such as theories based on the notion of indefinite past or of current relevance. Dowty (1979) was the first to give it a precise semantic formulation. In the German tradition, the notion of the "ExtendedNow" has never played a significant role, with the exception of Fabricius-Hansen (1986), who speaks of a "unechter Vergangenheitsbereich" ('improper past interval'). But now, Arnim von Stechow has worked out an account of the German perfect, as well (von Stechow 1999). In this paper, I will critically examine the approach and confront it with a different analysis. More specifically, von Stechow proposes the following account of the German present perfect: Under its canonical (i.e., non-futurate) interpretation, a present perfect clause in German involves a reference time R (more or less in the Reichenbachian 1947 sense) that is identical to the time of utterance, i.e., the "now" of the clause. The present perfect opens up an interval that includes the reference time R and extends it into the past. Since in the case of the canonical present perfect, the reference time is

Narrowing down the ExtendedNow

373

identical to the time of utterance or the "now", one may say that the construction provides thus an "ExtendedNow". At its left edge, this interval is limited by a time that may be contextually given or determined by certain temporal adverbials (p. 88). Finally, at least a truth interval of the event time or situation time of the VP embedded in the perfect construction is located properly within the ExtendedNow. These basic assumptions of the analysis are illustrated with regard to the examples (1.1a) and (1.2a) with the pictures in (1.1b) and (1.2a), respectively. In (1.1a), the time that limits the ExtendedNow on its left side must be provided by the context somehow. Or, when (1.1a) is uttered out of the blue, the ExtendedNow can presumably extend over all the time before now that could possibly be a time where Arnim is looking for a staircase in the Empire State Building (ESB), i.e., his lifetime from a certain age on. In (1.2a), the limiting time is provided by the adverbial seit drei Stunden ('since three hours ago'). (1.1) a. Arnim hat im Arnim has in-the

Empire State Building eine Treppe Empire State Building a staircase

A's searching ooooooo

b.

R = now Ι

gesucht. searched

.

ExtendedNow (1.2) a. Arnim hat seit drei Stunden Arnim has since three hours Treppe gesucht, staircase searched b.

im in-the

Empire State Empire State

Building eine Building a

A's searching R = now oooooooooooooooooooooooo| three hours ago

ExtendedNow

Note that while the searching for the staircase in (1.1) takes up only part of the ExtendedNow, in (1.2) it takes up all of it. We will come back to this difference below. How is the semantics of the present perfect in this analysis composed? - Leaving aside details that are irrelevant for the present purpose (e.g., tenses that are interpreted as variables), the tense of a clause denotes a contextually determined time interval which corresponds to the reference time. For example, the past tense denotes a time before the time of utterance. And the present tense denotes a time that includes the time of utterance; this also holds of the present tense contained in a present perfect clause. The perfect auxiliary haben/sein denotes the ExtendedNow interval. The ExtendedNow is combined with the untensed clause by the help of a quantificational adverb that may

374

Renate Musan

be overt or phonetically empty. In the latter case, it receives an existential default interpretation. The ExtendedNow (EN) serves as a restrictor of the quantificational adverb. The nuclear scope of the quantificational adverb contains the participle VP, which may or may not contain an aspect, too. The participle morpheme itself is semantically empty. The quantificational adverb is in principle free to quantify over times (as in (1.3a)) or over events (as in (1.3b)). Both the times and the events are combined with the ExtendedNow by a proper inclusion relation IN. Hence, (1.1a), for instance, may be schematically represented as in (1.3a) or as in (1.3b). (Von Stechow assumes with Herweg 1990 that stative VPs are properties of times while non-stative VPs are properties of events. Hence, strictly speaking the VP in (b) is reinterpreted as an event.) (1.3) a. EXt [IN (EN (t))] [Arnim eine Treppe im ESB such- (t)] = 'There is a time t properly included in the ExtendedNow such that Arnim looks for a staircase in the ESB at t.' b. EX e [IN (EN (e))] [Arnim eine Treppe im ESB such- (e)] = 'There is an event e properly included in the ExtendedNow such that e is an event of Arnim looking for a staircase in the ESB.' The future perfect and the past perfect receive the same interpretation, the only difference to the present perfect being that the reference time cannot be identified with the time of utterance, i.e., now. Instead, the position of the reference time must somehow be provided by the context. Moreover, the future tense is analogously analyzed with an ExtendedNow that reaches into the future instead of the past. For reasons of space, however, we will in this paper concentrate on von Stechow's analysis in as far as it concerns the present perfect. The remainder of this paper is organized as follows. Section 2 shows that von Stechow's approach accounts for many readings the present perfect can in fact have and correctly excludes some readings it cannot have. This supports the analysis proposed by von Stechow. Still, the predictions concerning the semantic composition of the present perfect do not seem to be correct. Moreover, the approach does not quite adequately hit what one may consider the spirit of the present perfect in German. Thus, section 3 presents counterarguments to the analysis. Section 4 sketches a more adequate analysis. Finally, building on section 4, section 5 explains where the impression that the present perfect introduces an ExtendedNow comes from.

2.

Achievements of the Analysis

It is well known that perfect clauses can give rise to numerous readings which differ from each other in various ways. In his 1999 paper von Stechow focuses on two main readings, the so-called universal reading and the existential reading. He attributes these readings to semantic factors. More subtle reading differences may result from the effects of pragmatic principles (cf. Musan 1999). In von Stechow's analysis, which of

375

Narrowing down the ExtendedNow

these two main readings arises depends on the choice of the quantificational adverb. In the remainder of this section, we will look in more detail at these readings, at how they come about, and at their special effects as they are described by von Stechow.

2.1

Existential Perfect Readings

We have already seen examples of the existential perfect reading in section 1. Recall (1.1a) and its readings that were represented in (1.3), repeated here as (2.1) and (2.2). (2.1) Arnim hat im Arnim has in-the

Empire State Building eine Treppe Empire State Building a staircase

gesucht. searched

(2.2) a. EX, [IN (EN (t))] [Arnim eine Treppe im ESB such- (t)] = 'There is a time t properly included in the ExtendedNow such that Arnim looks for a staircase in the ESB at t.' b. EXe [IN (EN (e))] [Arnim eine Treppe im ESB such- (e)] = 'There is an event e properly included in the ExtendedNow such that e is an event of Arnim looking for a staircase in the ESB.' In both representations, there is an existential adverb of quantification. In (2.2a), it quantifies over times while in (2.2b), it quantifies over events. According to von Stechow, the existential quantification in both cases leads to anteriority effects, i.e., Arnim's searching for the staircase must be located before the time of utterance. Moreover, in the case of event quantification, it triggers a completedness effect, i.e., Arnim's searching for the staircase must be over at the time of utterance. How do these effects arise? First, in both (2.2a) and (2.2b), Arnim's searching for the staircase must obviously be located before the time of utterance because it must be located within the ExtendedNow, which in turn is a time interval reaching from the time of utterance into the past, simply because of the semantics of the auxiliary haben/sein. Second, the completedness effect in (2.2b) arises because an event of a certain kind, e.g., an event of Arnim looking for a staircase in the Empire State Building, is identified with the maximal time which has the characteristics of this kind of event. That is, the event quantification in (2.2b) provides a time interval at which Arnim looks for the staircase and which is limited both on its left side and on its right side by times at which Arnim does not look for the staircase. The time quantification in (2.2a), however, only requires a time within the ExtendedNow at which Arnim looks for the staircase; it allows for the searching, however, to go on beyond the boundaries of the ExtendedNow interval.

376

2.2

Renate Musan

Universal Perfect Readings

The universal perfect reading arises in at least two cases, first, when the perfect clause contains the quantificational adverb immer ('always') as in (2.3a), and second, when the perfect clause contains a duration adverbial like seit drei Stunden ('since three hours ago') as in (1.2a), repeated here as (2.3b). (2.3) a. Das habe that have

ich I

doch (schon) immer gesagt. indeed (already) always said

b. Arnim hat seit drei Stunden Arnim has since three hours eine Treppe gesucht, a staircase searched

im in-the

Empire State Building Empire State Building

In (2.3a), the situation time of the VP is at least preferably understood as reaching from some time in the past up to the time of utterance; when schon ('already') or expressions like bis einschließlich jetzt ('up to and including now') are added, this effect is obligatory. In (2.3b), it is obligatory, too. In this respect, the meaning of present perfect clauses differs from the meaning of past tense clauses. Von Stechow concludes from this that the present perfect cannot be interpreted as a direct or indirect temporal relation between the situation time of the VP and the time of utterance - if it could, how should one be able to capture the intuition that in some cases, the situation time of the VP is located before the time of utterance (e.g., in the existential readings in (1.2)) but in other cases can reach up to the time of utterance (e.g,. in the universal readings in (2.3))? In the ExtendedNow analysis, both kinds of readings can be captured easily. Under the existential reading, the ExtendedNow contains a possibly short VP-situation-time or VP-event. Under the universal reading, however, the VP-situation-time covers all of the ExtendedNow interval. This can be represented roughly as in (2.4) for the examples in (2.3). (Examples like (2.3b) are not explicitly dealt with in von Stechow 1999. The representation in (2.4b) is thus what I consider a plausible approach within the ExtendedNow account. The interpretation of the adverbial seit drei Stunden is sensitive to the reference time of its clause, i.e., in present perfect clauses the time of utterance.) (2.4) a. ALL, [IN (ExtendedNow (t))] [ich das sag- (t)] = 'For all times t that are properly included in the ExtendedNow, I have said itatt.' b. ALL, [IN (ExtendedNow (t)) & seit drei Stunden (t)] [Arnim im ESB eine Treppe such- (t)] = 'For all times t that are properly included in the ExtendedNow and contained in the three hour interval counting backwards from now, t is a time where Arnim looks for a staircase in the ESB.'

Narrowing down the ExtendedNow According to this analysis, the occurrence of existential and universal perfect readings is thus due to the choice of the quantificational adverb: existential perfect readings arise with existential, and universal perfect readings with universal adverbs of quantification. Recall that the inclusion relation IN that connects the situation time of the VP to the ExtendedNow interval requires proper inclusion. Does this amount to a problem for the derivation of the universal readings in the ExtendedNow account? One may assume at first sight that if the situation time of the VP has to be properly included in the ExtendedNow interval, then perhaps it is not allowed to reach up to the time of utterance in this account either. There are, however, at least two conceivable ways to avoid problems resulting from the semantics of IN. First, the proper inclusion may be guaranteed on the left side of the ExtendedNow interval, i.e., there may be a short subinterval contained in the ExtendedNow interval where the situation time of the VP has not yet begun. For instance, in the case of (2.3b), the ExtendedNow may exclude the very left edge of the three-hour-interval mentioned. Second, the proper inclusion may be guaranteed on the right side of the ExtendedNow interval. This can be achieved if the ExtendedNow does by stipulation not reach from the right side of the reference time (or: the time of utterance) inclusively into the past but from its right side exclusively. This requires, of course, that the time of utterance (or: the reference time) is not represented as a point of time but as a time interval. But since this appears quite plausible in any case, it is unproblematic for the analysis. Interestingly, Rathert (1999:18) argues that the ExtendedNow in fact excludes the time of utterance: when seeing a certain red Mercedes for the second time, one can say (2.5). (2.5) Diesen roten this red

Mercedes habe Mercedes have

ich I

schon einmal gesehen. already once seen

If the ExtendedNow did not exclude the time of utterance, then one should say instead: "Diesen roten Mercedes habe ich schon zweimal ('twice') gesehen." In the situation described, however, (2.5) sounds much more natural.

3.

Counterarguments

Adequate as the ExtendedNow analysis seems so far, there are some points that raise doubt about it. In order to present the criticism, it is useful to look at some characteristics of the ExtendedNow analysis separately: first, the idea that there is a time interval constructed backwards from the time of utterance to a salient time in the past within which the situation time of the VP is located; second, the choice of the quantificational adverb; third, the hypothesis that the anteriority component within the perfect is associated with the auxiliary haben/sein; and fourth, the situation type that is denoted by perfect constructions.

378

3.1

Renate Musan

The Extension of the Now

In many concrete cases it is not at all clear how the construction of appropriate ExtendedNow intervals works and how it interacts with the functions of temporal adverbials in perfect constructions. Consider, for instance, (3.1a), an example that is constructed out of (1.1a) by supplementing it with the position adverbial an jenem Tag ('on that day'). Clearly, one would not want to claim that the adverbial in (3.1a) determines the left edge of the ExtendedNow: if it did, then (3.1a) should be judged true whenever the situation time of the VP is located anywhere on the stretch of time reaching from (some time contained in) yesterday up to now. This, however, contradicts the intuition that no interval of Arnim's looking for a staircase is able to make (3.1a) true. Rather, the adverbial locates the situation time of Arnim's looking for a staircase on the day in question as is indicated in the paraphrase in (3.1b). This may be achieved in a representation like (3.1c), where an jenem Tag is intersected with the ExtendedNow. (3.1) a. Arnim hat anjenem Tag im Arnim has at that day in-the gesucht. searched

Empire State Building eine Treppe Empire State Building a staircase

b. = 'There is a time t that is included in that day such that Arnim looks for a staircase in the ESB at t.' c. EXt [IN (EN (t)) & anjenem Tag (t)] [Arnim im ESB eine Treppe such- (t)] Hence, an jenem Tag presumably has two functions: it must intersect with the ExtendedNow as well as provide the left side of the ExtendedNow by introducing the left side of yesterday. Similarly, sert-adverbials do not in general directly name the left side of the ExtendedNow either. Rather, they introduce an interval whose right side is given by the reference time, but - in the case of seit drei Stunden - the left side of the ExtendedNow can only be determined by calculating back the three hours from the reference time into the past. In fact, with the possible exception of adverbials like seit Anfang Mai ('since the beginning of May') no adverbial in perfect constructions seems to be able to directly name the left side of an ExtendedNow. Moreover, in the case of universal perfect readings, it is not obligatory that the situation time of the VP reaches up to now. Thus, in an appropriate context as in (3.2a), a perfect clause containing immer ('always') can have a reading as paraphrased in (3.2b), where the situation time of the VP does not reach up to now. This reading may be represented as in (3.2c), where the contextual restriction given by the preceding clause is intersected with the ExtendedNow.

379

Narrowing down the ExtendedNow (3.2) a. [Schon already Arnim Arnim

als Kind hatte Arnim einen as (a) child had Arnim a hat immer Donald Duck geliebt. has always Donald Duck loved

guten good

Geschmack:] taste

b. = 'For all times t that are times of Arnim's childhood, Arnim loves Donald Duck at t.' c. ALLt [IN (EN (t)) & t is contained in Arnim's childhood] [Arnim Donald Duck lieb- (t)] Although we get the correct truth conditions by the representations in (3.1c) and (3.2c), it seems unintuitive that the implicit or explicit adverbials that specify the situation time of the VP are associated with the interpretation of the perfect auxiliary.

3.2

The Quantificational Adverb

Recall that according to the present analysis, the occurrence of existential or universal perfect readings depends solely on whether the obligatory quantificational adverb in the semantic representation receives an existential or a universal interpretation. Although this gives us the correct readings, it does not seem very convincing - why should an adverbial like seit drei Stunden ('since three hours ago') only be able to cooccur with a universal quantifier but not with an existential one? Rather, one would expect that any adverbial can in principle combine freely with both existential and universal quantifiers. But if this were the case, then an example like (2.3b), Arnim hat seit drei Stunden im Empire State Building eine Treppe gesucht should be able to have a reading like "There is a time t that is properly included in the ExtendedNow and contained in the three hour interval counting backwards from now, such that t is a time of Arnim looking for a staircase in the Empire State Building." Such a reading, however, is not available. Of course, there may be independent factors that rule out the missing reading. I cannot see, however, a plausible explanation for why the reading should be missing. Hence, the dependence of the obligatory universal reading of (2.3b) on the presence of the universal adverb of quantification appears to be stipulated.

3.3

The Anteriority Component

A more crucial point of criticism aims at the assumptions concerning the semantic composition of the perfect meaning out of its morphosyntactic components. More specifically, it is not convincing to associate the anteriority meaning component of the perfect construction with the auxiliary. Rather, it seems more plausible to associate it with the past participle. (For a more detailed argument, see Musan 1998 and Musan, to appear.)

380

Renate Musan

This is so because the interpretation of the past participle in other constructions remained quite mysterious under the assumption that its semantics does not express anteriority. The crucial point is that in all other constructions in which the past participle can appear, it implies an anteriority component - either synchronically or diachronically. Hence, associating the anteriority meaning in perfect constructions with the auxiliary amounts to imposing a meaning on the auxiliary that one could get cheaper by associating it with the participial verb. In addition, one would have to stipulate that the past participle in perfect constructions lost its anteriority component. (3.3) lists constructions containing past participles which obviously imply an anteriority meaning component. Note that especially the attributive constructions (3.3c,d) strongly suggest to associate the anteriority with the past participle form: in attributive constructions, past participle forms appear without any other item that might be able to carry the anteriority component; yet they imply anteriority. The escaping in (3.3c) must have taken place before the time of utterance of the nouns, and also the covering of the cabin by snow mentioned in (3.3d) fits into the anteriority pattern insofar as the covering of the cabin must have started before the time of utterance. The same holds for (3.3e,f): the assertions in (3.3e,f) both concern the time before the time of utterance. (3.3) Perfect constructions: a. Arnim hat einen Arnim has a b. Der Baum the tree

ist is

Nussbaum nut-tree gewachsen. grown

Attributive constructions: c. Die ausgerissene Lore the escaped Lore d. Die von Schnee the by snow Stative passive: e. Das ist nicht that is not

gepflanzt. planted

von from

Singapore sitzt auf dem Dach. Singapore sits on the roof

bedeckte Hütte covered cabin

ist is

urgemütlich. extremely-cosy

ausgearbeitet. out-worked

f. Die Briefe sind sicher aufbewahrt. the letters are safely kept The only potential problem is the occurrence of the past participle in "eventive" passive constructions: eventive passives as in (3.4) contain past participles, too, but do not seem to express anteriority at all. (3.4) Eventive passive: Christus wird Christ becomes/is

erschlagen. slain

('He is slain')

Narrowing down the ExtendedNow

381

But still, eventive passives do not provide a good counterargument against the generalization that past participles generally imply anteriority, because eventive passives have developed historically from a past participle construction that clearly contained an anteriority component. Specifically, the predecessor of the eventive passive construction in Old High German is a semantically compositional werc/ew-construction as illustrated in (3.5) (.Isidor 471; cited after Granvik 1986:24). The construction in (3.5) expresses a futurate change of the object, i.e., Christ, by getting into a state caused by the action, i.e., the slaying (cf. also Grimm 1960:258, Oubouzar 1974:87 and others; the construction serves as a translation of a Latin future tense construction; cf. Granvik 1986:23). Analogously to today's ingressive predicative wer\you climb y)) It also yields the correct result as regards the fact that the wA-phrases cannot have scope independently of each other.

7.

Conclusion

Above I showed that a semantics that uses alternative sets and functions that choose among them can solve a number of problems related to the islandhood of certain constructions. We did not, however, discuss other syntactic problems mentioned at the end of section 2.1, which concern specific language particular differences. It seems to me that these have to be treated by syntactic stipulation. That is, we have to stipulate that with the constructions from Hungarian, a higher order choice is possible, whereas the ungrammatically of analogous constructions in German (or Japanese) requires that this kind of choice function is unavailable in these languages. Moreover, the overt choice functions, interpreted as "scope markers" in German and Hungarian, require CP com-

Partial Movement Constructions, Pied Piping, and Higher Order Choice Functions

485

plements, whereas the silent choice functions that interpret pied piping seem to take complements on the subclausal level only. This way, the asymmetry mentioned in the last paragraph of section 2.1 can be accounted for.

Acknowledgments For comments and valuable criticism I would like to thank Irene Heim, Katalin Kiss, and Uli Sauerland.

References Dayal, Veneeta (1996). Scope Marking: In Defense of Indirect Dependency. In: Uli Lutz & Gereon Müller (eds.) Papers on 'Wh-Scope Marking'. (Arbeitspapiere des SFB 340, Bericht Nr. 76.) Universität Tübingen. 107-130. Dayal, Veneeta (2000). Scope Marking: Cross Linguistic Variation in Indirect Dependency. In: Uli Lutz, Gereon Müller & Arnim von Stechow (eds.) Wh-Scope Marking. Amsterdam: Benjamins. 157-194. Hamblin, Charles L. (1973). Questions in Montague English. Foundations of Language 10.41-53. Horvath, Julia (1997). The Status of 'WH-Expletives' and the Partial WH-Movement Construction of Hungarian. Natural Language and Linguistic Theory 15. 509-572. Horvath, Julia (2000). On the Syntax of 'Wh-Scope Marker' Constructions: Some Comparative Evidence. In: Uli Lutz, Gereon Müller & Arnim von Stechow (eds.) Wh-Scope Marking. Amsterdam: Benjamins. 271-316. Lutz, Uli & Gereon Müller (eds.) (1996). Papers on 'Wh-Scope Marking'. (Arbeitspapiere des SFB 340, Bericht Nr. 76.) Universität Tübingen. Lutz, Uli, Gereon Müller & Arnim von Stechow (eds.) (2000). Wh-Scope Marking. Amsterdam: Benjamins. Nishigauchi, Taisuke (1990). Quantification in the Theory of Grammar. Dordrecht: Kluwer. Reinhart, Tanya (1994). Wh-in-Situ in the Framework of the Minimalist Program. OTS Working Papers, University of Utrecht. Rooth, Mats (1985). Association with Focus. PhD dissertation, University of Massachusetts at Amherst. Srivastav, Veneeta (1991). Wh-Dependencies in Hindi. PhD dissertation, Cornell University.

486

Wolfgang Sternefeld

Stechow, Arnim von (1996a). Against LF Pied-Piping. Natural Language Semantics 4. 57-110. Stechow, Arnim von (1996b). Partial Wh-Movement and Logical Form. In: Uli Lutz & Gereon Müller (eds.) Papers on 'Wh-Scope Marking'. (Arbeitspapiere des SFB 340, Bericht Nr. 76.) Universität Tübingen. 1-36. Stechow, Arnim von (2000a). Partial Wh-Movement, Scope Marking, and Transparent Logical Form. In: Uli Lutz, Gereon Müller & Arnim von Stechow (eds.) WhScope Marking. Amsterdam: Benjamins. 446—478. Stechow, Arnim von (2000b). Some Remarks on Choice Functions and LF-Movement. In: Klaus von Heusinger & Urs Egli (eds.) Reference and Anaphoric Relations. Dordrecht: Kluwer Academic Publishers. 193-228. Sternefeld, Wolfgang (to appear). ^-Expletives and Partial ^-Movement: Two NonExisting Concepts? In: Werner Abraham & Jan-Wouter Zwart (eds.) German in Typological Context. (Linguistik Aktuell.) Amsterdam: Benjamins.

Prelexical Syntax and the Voice Hypothesis

Dieter Wunderlich All the world is queer save me and thee, and even thou art a little queer. (attributed to Robert Owen 1828)

This paper relates to two articles by Arnim von Stechow, whom I have always admired for his semantic expertise but with whom at the same time I have always quarrelled in questions on the overall architecture of grammar. I first present an argument showing that morphology restricts adverbial scope differently from syntax. I will then provide a couple of arguments playing with the Voice hypothesis, which claims that only Voice makes an Agent (which, after all, is a wonderful metaphor). More specifically, I accept (with one important exception) Arnim von Stechow's highly sophisticated semantic decompositions, but I take issue with his view that the internal structure of words is part of syntax. Moreover, from a cross-linguistic point of view I will come to the conclusion that morphological voice and the introduction of Agent are largely independent of one another.

1.

Introduction1

The advocates of Generative Semantics proposed that words are constituted by syntactic trees made up of VPs (or Ss at that time) and NPs, where the terminal symbols encode partial predicates such as DO, CAUSE, BECOME, and CLOSED, which together determine a verb such as close. By successive predicate raising (which is now known as "head movement"), these predicates are adjoined to each other under one single V node and are then spelled out by an appropriate phonological matrix. This proposal aimed at a specification of interlinguistic differences as to what lexical items are possible. One major motivation was the fact that adverbs often only modify a constituent that is part of a word, as in the following examples taken from McCawley (1971, 1973:348).

1

I am grateful to Ingrid Kaufmann, Albert Ortmann, and Barbara Stiebels for various comments, especially to Ingrid Kaufmann for the permission to include some of her results on the middle. I also thank Anja Latrouite and Yi-chun Yang for providing me with Tagalog and Chinese examples, as well as the two anonymous reviewers for their helpful remarks.

488 (1)

Dieter Wunderlich a. The door opened, and then I closed it again, b. *The door opened, and then I kicked it again.

(2)

a. I closed the door temporarily, b. *I kissed Susan temporarily.

The idea that lexical items should be decomposed into smaller predicates standing in structural relations to each other (rather than listed) has made its way into several approaches to the lexicon, although these decompositions are mostly conceived of as semantic and not syntactic characterizations of words (see Dowty 1979, Bierwisch 1983, Jackendoff 1990, Wunderlich 1997a,b, among others). However, the original proposal has also survived, for instance, in the work of Hale & Keyser (1994), and especially in the work of Arnim von Stechow (henceforth: AvS), even though the conception of underlying syntax has changed dramatically since the times of Generative Semantics. AvS (1996) assumes that the Logical Form (LF) of a verb consists of a VP, which is optionally embedded into a Voice Phrase adding an agent or holder (in this respect following Kratzer 1994),2 which itself is part of higher functional projections such as AspP, AgrOP, and TP (relating to aspect, object agreement, and tense). Being myself a follower of the lexical decomposition hypothesis couched in a semantic framework, I find many results of AvS's investigations conclusive, but, at the same time, I am astonished at an account in which LF configurations within a word are expressed by syntactic phrases. I am deeply convinced that predicates within a word are clumped together in a different fashion from predicates within a clause, in other words, that morphology and syntax are two independent domains of surface structure. The fact that close is a verb which is semantically structured even without possible segmentation into visible morphemes in my view is a morphological rather than a syntactic idiosyncrasy. If a verb with the meaning of close can be transparently decomposed into more than one morpheme, I take this as a morphological fact that changes nothing in its status as a verb. If, however, a sentence containing close can be paraphrased by means of words meaning 'cause', 'become' and 'closed', I would take this as a syntactic fact of quite a different nature. Words can undergo movement in a sentence, can show agreement with other words, and can have optional scope over argument expressions, whereas morphemes normally cannot be moved in a word, cannot agree with other parts of the word, and cannot have scope over anything else than argument variables. From a formal point of view, close is indivisible in both morphology and syntax. My suggestion is that divisibility in the morphology would lead to no ef-

2

It is interesting to note that agentivity of a verb has been represented by means of the predicate DO by the generative semanticists (Ross 1972). Dowty (1979:113) suggests "that both Stative and active verbs are constructed from the same homogeneous class of primitive stative predicates, thus the presence of DO is the only thing that distinguishes the meaning of a stative from that of an active verb". In this respect, even the Voice hypothesis is based on an old concept, but it also allows for stative verbs to be represented by a value of Voice (namely, Holder).

Prelexical Syntax and the Voice Hypothesis

489

fects in the use of this word, whereas divisibility in the syntax would lead to major effects. 3 In this paper, I will briefly discuss two of AvS's central assumptions. First, I consider directional phrases that are inside or outside of the lexical meaning of a verb; both are visible for again, but only the internal variant is visible for re-, the bound variant of again. Second, I will discuss various types of voice, some of which hardly go together with the Voice hypothesis. I combine these two topics because it is precisely them that form the explanative framework in AvS's work on adverbial readings. My aim is to raise questions rather than to offer solutions. Throughout the paper, I will use abbreviated semantic representations; although both AvS and myself use situation (or event) variables in the representation of verbs, I mostly neglect them in order to concentrate on the major points.

2. Morphology and Syntax Differ in Scope

2.1

The Structural Context for Adverbial Ambiguities

In their paper on German fast 'almost' and the visibility parameter of adverbs (1999:3 If.), Irene Rapp & Arnim von Stechow illustrate that verbs of motion may take a directional phrase either syntactically, as in (3), or lexically, as in (4): (3)

a. Sie rannte zum Haus. 'She ran to the house' b. Sie stieg auf den Hügel. 'She climbed onto the hill'

(4)

3

Sie bestieg den Hügel. 'She climbed the hill'

If close is decomposed in syntax, I cannot see what prevents us from paraphrasing (i) by (ii): (i) John closed the door again. (ii) John did something which made the door to become closed again. We are then confronted with the problem that (ii) is logically weaker than (i): (ii), but not (i), can be true in a situation where John removes a chair and then, after a while, the door gets shut by a sudden blast of wind. While close refers to a single coherent situation, the paraphrase may be about two situations. This difference could be implemented in a syntactic framework only if it is marked already whether the predicates are morphologically merged or not, for instance, by the absence or presence of an operator that binds a situation variable.

490

Dieter Wunderlich

According to Rapp & AvS, there is no difference in representation. Both (3b) and (4) are represented by the same structure, rendered in (5). (5)

VoiceP VP

& Ag(x) VP

BECOME

I

XP

CLIMB

X' LOC.ON

the hill

From a representation such as (5), one cannot see whether the PP is syntactically added, as in (3), or lexically integrated, as in (4). If an adverb such as wieder 'again' is added syntactically, no difference arises; both sentences in (6) allow a restitutive or repetitive reading: either she was first on the hill, climbed down and then returned to the hill (restitutive), or she climbed on the hill a second time (repetitive). The latter reading correlates with stress on the adverb. (6)

a. Sie stieg wieder auf den Hügel. 'She climbed again on the hill' b. Sie bestieg wieder den Hügel. 'She climbed again the hill'

However, as we will see in section 2.3, these two variants of the verb behave differently with respect to bound adverbs such as re- in English. AvS (1996), as well as Rapp & AvS (1999), explain the two readings of wieder 'again' by the assumption that wieder either adjoins to the XP, which gives the restitutive reading ('be again in that state'), or to the VoiceP (or even higher in the tree), which gives the repetitive reading ('perform the same action'). Precisely this possibility to explain the semantic ambiguity in terms of structure was the driving force behind AvS's approach. 4

4

AvS's approach is attacked by Jäger & Blutner (1999); these authors assume that wieder 'again' is lexically ambiguous with respect to the presupposition that is made. As AvS (2000) points out, the account by Jäger & Blutner is not unproblematic either.

Prelexical Syntax and the Voice Hypothesis

2.2

491

Sortal Conditions for Adverbial Ambiguities

Given the structure in (5), one would expect even four readings to be possible, as can be seen from the abbreviated representation in (7): (7)

wieder [Ag(x) & wieder [CLlMBlNG(e) & wieder [BEC wieder [LOC.ON(x,y)]]]]

If wieder is adjoined to the VP, no interpretation is available because the repetition of a climbing event with no Agent is impossible - this fact makes it doubtful whether the separation of Voice from the VP is reasonable (see also section 3). If, however, wieder adjoins to the PP, a reading saying that a pure transition event is repeated should be possible, although it is hard to grasp. Let us assume that the constituent to which wieder adjoins must meet the sortal requirements of the adverb, otherwise the reading is ruled out semantically; wieder seems to be compatible with any kind of state or event, but it is not implausible that pure transitions are excluded. Other adverbs have different sortal conditions: kurz 'shortly' restricts a state or an atelic process/activity, hence (8a) has only one reading, which is the most internal one (kurz [LOC.ON(x,y)]), whereas fast 'almost' presupposes a scale which can be realized by different kinds of entities. In (8b), the scale is contributed by stages of becoming, and the most internal reading obtains (fast [LOC.ON(x,y)]). By contrast, in the counterfactual sentence (8c) the subjunctive contributes another scale (e.g., circumstances under which she would or would not climb the hill): here, the most external reading (which is still in the scope of the subjunctive) is preferred, but the internal reading is possible, too. (8)

a. Sie stieg kurz auf den Hügel. 'She climbed for a short time on the hill' b. Sie stieg fast auf den Hügel. 'She climbed almost on the hill' c. Sie wäre fast auf den Hügel gestiegen. 'She almost climbed on the hill'

AvS is well aware of the semantic conditions originating from the meaning of adverbs. If there were only these readings that 'look into' a lexical item, he probably would not have carried out his huge work specifying the structural prerequisites for interpretation. Rather, he was also convinced that these different readings have positional effects in the sentential surface. I am, however, sceptical in this respect because intonational patterns are possible that do not correlate with a different linear order of syntactic constituents (see footnote 6 below). Positional effects of semantic factors should, in principle, be visible in other languages, too, but very little is known in this respect from a cross-linguistic perspective. Given the flexibility of semantic coercion in other fields (Pustejovsky 1995:115ff.), one could as well claim that nothing follows from the study of adverb-verb readings for the internal syntactic structure of lexical items. Note that it is the semantic type of begin, want or believe (together with additional qualia consider-

Dieter Wunderlich

492

ations for the respective objects) that forces us to complete the sentences in (9) stereotypically. One does not need to assume a syntactic ellipsis. (9)

a. She began a book.

['She began reading/writing a book']

b. She wanted a beer.

['She wanted to drink a beer']

c. She believed the author.

['She believed what the author wrote']

begin is a function that requires an event of some sort rather than a physical object; the reading of (9a) adapts to this condition by type coercion; similar things hold for (9b,c). One could similarly argue that fast is a function that requires a scale of some sort, and the readings of (8b,c) adapt to this condition. Thus, the interpretations of the sentences in (8) could be constructed on the basis of semantic knowledge, without any regard to an internal syntax of the verb. Of course, one knows that the semantic representation of besteigen 'climb' involves some complexity in terms of semantic types. I assume that this semantic complexity is relevant for the way in which the arguments are realized by agreement and case,5 but not necessarily for their position in the clause (Wunderlich 1997a,b). Arguments are obligatory complements of the verb, whereas adjuncts are optional. If the position of adverbs refers to the internal structure of verbs, than the way in which the arguments are realized should do as well. In this respect, AvS's work is more ambitious than mine. However, in his work the realization of arguments does not seem to depend on the internal syntax of verbs.

2.3

Morphological and Syntactic Adjuncts have Different Scope

While a syntactic adverb can have access to the whole clause (although this is restricted by position or stress), a bound morpheme can only have access to the stem to which it attaches. Although the syntactic adverb again and the prefix re- have the same meaning, one expects that the latter interacts with the verb's meaning more restrictively. This is indeed the case. As shown by Wechsler (1989), the English equivalents of the sentences given in (6) above lead to different grammaticality judgements if again is replaced by re-: 5

The fact that in examples such as the following the substance being drunk cannot be expressed in syntax (at least not by an argument expression with morphological case) can be explained by the Restriction on Structural Arguments ( R S A ) , which is defined for a lexical SF representation in Wunderlich (1997a,b). It is unclear how syntax can explain this "loss" o f internal arguments. (i)

Er trank den Kühlschrank (*von Wein) leer. 'He drank the fridge empty' (of wine) y is nonstructural in {DRiNK(x,y) & BECOME EMPTY(z)} according to RSA.

(ii)

Er trank sich den Frust ( ' m i t Bier) weg. he drank himself the frustration (with beer) away

(iii)

Er betrank seinen Sieg (mit Champagner), he be-drank his victory (with champagne)

Prelexical Syntax and the Voice Hypothesis

493

(10) a. *She reclimbed on the hill. *She reran to the station. b. She reclimbed the hill. She reentered the station. A syntactic PP cannot be in the scope of re-, whereas a lexical "PP" can, a fact that is not captured by the tree structure (5). If one assumes that re- is restricted to result verbs, and furthermore assumes the lexical representations in (11), it is clear that the avariant of climb must reject re-prefixation, even though it is subcategorized for a directional PP to appear in syntax, whereas the b-variant of climb admits re- because it is characterized as a result verb lexically. As indicated in (12), word-internal re- (again) allows either wide or narrow scope. (11) a. steigen/climb\

λΡ

λχ {CLIMB(x) & P(x)}

+dir b. besteigen/climb: (12)

reclimb:

λy λχ {CLlMB(x) & BECOME LOC.ON(x,y)} λy λχ again{CLIMB(x) & BECOME again [ LOC.ON(x,y)]}

In either case, re- must be in the scope of any operator that binds the variable y relating to the direct object. This is not necessarily the case with syntactic adverbs. Compare the following sentences: (13) a. Arnim again climbed a hill, b. Arnim reclimbed a hill. (13a) can have the reading that at the second event Arnim climbed another hill than at the first event, which is shown in (14a), whereas (13b) cannot have this reading; in this case there is only one hill which is climbed twice (or climbed down and up), although not necessarily by Arnim, see (14b). That is, the existential operator associated with a hill cannot be in the scope of re- in (13b). One therefore has to conclude that the syntactic adverb interacts with the internal structure of climb differently from the prefix another aspect which is not expressed by the tree in (5). 6

6

Note that (13b) allows for both a repetitive and a restitutive reading, with the presupposition that is was this particular hill that was occupied before. Thus, the scope paradox observed by Jäger & Blutner (1999) cannot arise. (In contrast to A Delaware settled in New Jersey again, which is Jäger & Blutner's counterexample to AvS's theory, the sentence A Delaware resettled in New Jersey cannot mean that the Delaware tribe originally comes from New Jersey and some recent member of the tribe moved to the home of his ancestors.) Nevertheless, re- would have to be lexically ambiguous in the same way as again in Jäger & Blutner's account.

494

Dieter

Wunderlich

(14) a. again [ 3 y (HILL(y) & CLIMB(arnim,y))] b. λχ 3y (HILL(y) & again [CLlMB(x,y)]) (arnim) That a PP can be in the scope of again but not in the scope of re- was demonstrated by Wechsler with the following pair of examples: (15) a. John reswam the English Channel with flippers. b. John swam the English Channel with flippers again,

(this time with flippers) (both times with flippers)

If again in (15b) is moved right in front of the verb, it is still possible to get the reading that John swam the English Channel both times with flippers, whereas exactly this reading is excluded in (15a). Wechsler also contributed evidence that the bound affixes in Greek and Chichewa behave exactly like English re-. Consider his examples from Chichewa (Wechsler 1989:429): (16) a. Mu-lembe=nso chimangirizo [ndi nthenga]pp. you-write=again essay with feather 'You write the essay again, with a quill (this time)' b. Mu-lembe-re=nso nthenga chimangirizo. you-write-APPL=again feather essay 'You write the essay with a quill again' The verb in (16b) results from that in (16a) by the applicative operation, which is similar to the locative alternation observed in (1 lb)/(13b). Another property of re- has first been observed by Keyser & Roeper (1992): re- is possible with the resultatives in (17), but not with those in (18). (Keyser & Roeper's examples are (17a,b), and (18e) - the others are mine.) (17) a. First John painted the house yellow and then he repainted the house red. [John painted the house.] b. John rehammered the nail flat,

(first he hammered the nail crooked)

c. John rewiped the table clean.

(when he first wiped the table it remained dirty)

(18) a. Anna again ran her shoes threadbare.

[*Anna ran her shoes.]

b. *Anna reran her shoes threadbare. c. John again drank the fridge empty.

[*John drank the fridge.]

d. *John redrank the fridge empty. e. *John redrove the man crazy.

[*John drove the man. (in the intended sense)]

Prelexical Syntax and the Voice Hypothesis

495

(17) illustrates weak resultatives (in the sense of Kaufmann & Wunderlich 1998), where no further argument is added and the resultative AP only specifies a result that already is implied by the base verb, whereas (18) illustrates strong resultatives, where, by contrast, a further argument is added and the resultative AP specifies a result that is not implied by the base verb. This difference is captured by the semantic representations in ( 1 9 ) , in which BECOME P(y) represents the resultative extension. (19)

{Ag(x) & {RUN(x)

BECOME WITH.PAINT(y) & BECOME P ( y ) } & BECOME P ( y ) }

The condition for re-, then, is that it forbids a resultative extension in its scope; in other words, re- has only access to the base verb and not to the resultative extension. That rethus has to 'precede' the resultative extension might be explained by a syntactic account of resultatives but only if it is admitted that re- itself is added morphologically; it is hard to see how the differences between the two types of resultatives with respect to re-prefixation are captured within a single syntactic tree. Note that Keyser & Roeper's (1992) ungrammatical example (18e) has a straightforward explanation under the morphological account: here, re- would have to apply to a syntactic idiom (drive someone crazy), which is ruled out if re- is a prefix. Although our subject certainly calls for further study, the preliminary insights suggest that morphology and syntax differ in their scopal properties. This fact may constitute one of the reasons why there is morphology at all, in addition to syntax.

HIS MASTER S VOICE

3.

Voice and the Introduction of Agent

In this section, I discuss to what extent the various appearances of voice interact with the introduction of Agent (or other thematic argument roles). In the first two subsections I consider problems that immediately arise from AvS's work, and in the further subsections I bring additional voices to ear from a cross-linguistic perspective. Since many voices do not automatically form a chorus, one hopes that something like the Voice hypothesis can harmonize the noise coming about.

496

3.1

Dieter Wunderlich

Lexical Selection by Voice

AvS states (1996:108): (20) a. Unaccusative verbs do not project Voice (they are voiceless). b. Unaccusative verbs have the semantic structure BECOME + Stative. This does not exclude that unaccusative verbs can add Voice. More precisely, there are three classes of verbs that have to be distinguished. (21) a. (Ag(x) &) BEC P(y) schmelzen 'melt' P=LIQUID, brechen 'break' P—BROKEN b. Ag(x) & BEC P(y) schließen 'close' P=CLOSED, fällen 'fell' P=DOWN c.

BEC P(y) ankommen 'arrive' P=LOC.HERE, fallen 'fall' P=DOWN

(21 a) is the class of alternation verbs, which optionally allow for Voice, (21 b) is the class of inherently causative verbs, which require Voice, and (21c) is the class of proper unaccusative verbs, which forbid Voice. If Voice is part of the lexical entry of verbs, it would be easy to make the distinction between (21a-c); but then Voice would be superfluous. If Voice is a functional category that selects lexical items, one needs to know in advance whether the item to be selected permits Ag(x) or not. In other words, the lexicon already makes the distinction whether Voice is possible or not; it would do so most effectively if Ag(x) already is part of the lexical entry.

3.2

Iteration of Voice

Considering result verbs such as verlassen 'leave', AvS (1996:109-112) observes that the ambiguity between restitutive and repetitive readings of wieder 'again' is preserved if the adverb precedes the accusative object, in contrast to result verbs such as öffnen 'open', where, according to AvS's jugdments, only the repetitive reading obtains:7 (22) a. Als Anna wieder das Haus verließ, war es dunkel. when Anna again the house left, it was dark b. Als Anna wieder die Tür öffnete, war es dunkel. when Anna again the door opened, it was dark 7

(restitutive/repetitive)

(only repetitive)

According to my informants, the restitutive reading is merely the preferred one in (22a), while the repetitive reading is preferred in (22b), which may, but need not, have a structural explanation. All of my informants judge a restitutive reading of (22b) to be possible, too, at least if the verb bears a LH accent, just as (22a) does per default; this is even clearer with wieder die Tür ZUmachte 'again the door closed'. If these intonational judgments can be more generally confirmed, the positional effects of wieder might be seen in a different light.

Prelexical Syntax and the Voice Hypothesis

497

Verbs such as öffnen 'open' are called "object result verbs" because only the object participates in the result state, whereas verbs such as verlassen 'leave' are called "subject-object result verbs" (in Rapp & AvS 1999:180) because both subject and object participate in the result state. To account for this difference, AvS assumes that AgrOP dominates the BECOME-VP in the object result verbs, whereas AgrOP is dominated by the BECOME-VP in the subject-object result verbs. This allows a restitutive reading, even if wieder precedes the accusative NP. Moreover, AvS assumes two subsequent Voices for the latter class of verbs, an inner Voice (=Holder(pro)), stating that the subject participates in the result state as a holder, and an outer Voice (=Ag(x)), stating that the subject behaves as an agent. This is shown in (23) (AvS 1996:111): (23) verlassen ('leave')

The semantic representation is given in (24a), which can, however, be abbreviated as in (24b), if one allows more than one occurrence of the subject argument variable: ( 2 4 ) a. { A g ( x ) & BECOME ( H o l d e r ( p r o ) & LOC.OUT(y)}} b. { A g ( x ) & BECOME LOC.OUT(x,y)}

The motivation to introduce the inner Voice (=Holder(pro)) is to ensure that the subject performs two roles in the verb. The category Voice thus has to fulfill functions for which morphological exponents of voice have not yet been found. If there is some other means to guarantee the identification of Agent with an internal argument, it should certainly be preferred. The possibility to iterate Voice within one verb enlarges the generative power to an undesirable extent. AvS's proposal in (23) constitutes the most drastic departure from the traditional view of word meanings: BECOME, being part of the meaning of the verb, now c-com-

Dieter Wunderlich

498

mands functional categories such as AgrO and Voice, 8 which is undesirable under the perspective that lexical items are syntactic atoms. In the following, we see other instances where Voice must be iterated if it is given the function to introduce an agent.

3.3

Causative and Passive

The Active Voice morpheme only adds Ag(x) to a verb, whereas the Passive morpheme, according to the Voice hypothesis, adds Ag(x) and, simultaneously, licenses prOcase in Spec Voice and thus absorbs one of the possible internal cases licensed by the verb. Semantically, Passive simply adds Ag(pro) (see AvS 1996:103-105). The Active Voice morpheme seems to be identical with a causative morpheme, which also adds Ag(x) to a verb. Since in many languages Causative also applies to agentive verbs, the operation of Agent introduction must be iterative. That this indeed is possible is obvious from languages such as Malayalam or Turkish, where Causative is iterative. (25) a. Active:

Voice Ag(x) = Causative

b. Passive:

Voice Ag(pro case )

Let us now consider the case where Causative is followed by Passive. For this case, the Voice hypothesis claims that a first agent is introduced by Causative, and a second agent (differently from the first one) by Passive. However, what in fact happens is that Passive demotes the agent introduced by Causative, as demonstrated by the examples (26c,d) from Yucatec Maya (Bricker 1978:22). The possible semantic representations are indicated below the examples; a pro argument has to be bound existentially. (26) a. k=u

kan

-ik

'he is learning it'

INCOMPL=3 .ERG learn -IMPF ACTIVE: Xy λ χ { A g ( x ) & BEC KNOW(y)}

b. k=u

ka'an

-al

'it is being learned'

INC0MPL=3.ERG learn.PASS -IMPF PASS: λ y {Ag(pro) & BEC KNOW(y)}

c. k=u INCOMPL=3.ERG

ka'an

-s

-ik

learn.PASS

-CAUS

-IMPF

'he is teaching it'

PASS-CAUS: Xy λιι {Ag(u) & Ag(pro) & BEC KNOW(y)}

8

One might see this as a relic of word-formation, because all clear subject-object result verbs in German are prefixed (be-treten 'enter', er-reichen 'reach', ver-lassen 'leave', etc.), and there might be evidence that it is the prefix that contributes the predicate BECOME. But even under this perspective, it is unexpected that BECOME can c-command AgrOP, in particular when the verbs are clearly lexicalized.

499

Prelexical Syntax and the Voice Hypothesis d. k=u INC0MPL=3.ERG

kä'an

-s

-a?al

learn.PASS

-CAUS

-PASS.IMPF

'it is being taught'

PASS-CAUS-PASS: Xy { A g ( p r o 2 ) & A g ( p r o i ) & BEC KNOW(y)}

If Passive follows Causative (see (26d)), together they introduce only one instance of Ag(pro). Since there is no indication that CAUS-PASS is fused into one morpheme, here Passive must set an agent that already is present to pro. How can this be ensured?

3.4

Passive and Antipassive

The ingenious part of the Voice hypothesis is that Passive does not only demote an agent (by setting it to pro), but demotes what it introduces; this correctly predicts that Passive only applies to agentive verbs. Many languages also exhibit an operation which is termed Antipassive: it demotes the lowest or the affected object in the context of an agent. As it stands, the Voice hypothesis would have to claim that first an invisible Active operates, and then Antipassive, as in (27a). This combination would yield a representation such as that in (27b): (27) a. Antipassive:

Voice Aff(pro case )

b . { [ A g ( x ) & VERB(e)] & A f f ( p r o ) }

One problem is to determine which case is absorbed by Antipassive. Since most languages with a morphological Antipassive exhibit ergative structure, one could establish the following rule: Passive absorbs accusative, whereas Antipassive absorbs ergative. (This is equivalent to saying that Passive and Antipassive are alternative operations for intransitivizing transitive verbs.) Yucatec Maya is a language that exhibits both Passive and Antipassive, but has only one marked case, which, according to Krämer & Wunderlich (1999), is the ergative. One might think that both operations absorb ergative, which, however, is not quite true. Yucatec has a split-aspect system in intransitive verbs: imperfect correlates with ergative, and perfect with nominative, regardless of whether these aspects are inherent or marked. Passive yields inherently perfective verbs (and it is only in this sense that it absorbs ergative), while Antipassive yields inherently imperfective verbs (and therefore does not absorb ergative). As a consequence, Passive and Antipassive of Yucatec can only be distinguished in terms of the thematic roles involved: Ag(pro) in the Passive, and Aff(pro) in the Antipassive. Yucatec Passive is marked by glottal-stop insertion together with vowel lengthening, while Antipassive is marked by vowel lengthening combined with low tone. This is illustrated in the following examples with the transitive verb hek' 'break' (from Bricker & Yah 1981:xi). All examples are given in the imperfective aspect, which requires an ergative clitic for intransitive verbs.

Dieter Wunderlich

500 (28) a. k=in INC0MPL=1.ERG ACTIVE: λ y λ χ

hek'

-ik

break

-IMPF

Ί am breaking it/something'

{Ag(x) & BEC BROKEN(y)}

b. k=u INCOMPL=3.ERG

he'ek' break. PASS

-el

'it is being broken'

-IMPF

PASS: λ y { A g ( p r o ) & BEC BROKEN(y)}

c. k=in INCOMPL=L.ERG

heek' break. ANTIP

Ί am breaking'

a n t i p : λ χ { A g ( x ) & BEC BROKEN(pro)}

Both Agent and the affected object can also be introduced by independent morphological means. Yucatec thus exhibits the fully symmetric system shown in (29): (29) a. Causative:

Voice Ag(x)

b. Passive:

c. Affected Object (AO): Voice Aff(y)

Voice Ag(pro)

d. Antipassive: Voice Aff(pro)

The following examples (from Straight 1976:193-94) illustrate an instance where Affected Object (30b) is followed by Passive (30c); as in (26) above, the agent must be present prior to the application of Passive. (30) a. k=a

k'ooy

INCOMPL=2.ERG ACTIVE:

'you dig'

dig

λχ {Ag(x) & DIGGING(e)}

b. k=a INCOMPL=2.ERG (ACTIVE)-AO:

k'ooy

-t

-ik

dig

-AO

-IMPF

'you dig it (up)'

λy λχ {Ag(x) & DIGGING(e) & Aff(y)}

c. k=u INCOMPL=3.ERG

k'ooy

-t

-a'al

dig

-AO

-PASS.IMPF

'it gets dug (up)'

(ACTIVE)-AO-PASS: λ y { A g ( p r o ) & DIGGlNG(e) & A f f ( y ) }

Since at most two structural arguments are possible in Yucatec, a verb can undergo Causative or Affected Object only if it is intransitive ("unaccusatives" may undergo Causative, and "unergatives" may undergo Affected Object), 9 and it can undergo Passive or Antipassive only if it is transitive. Following the Voice hypothesis, all Yucatec verbs would have to be zero-valent underlyingly, but with a different potential to undergo the above-mentioned operations: Some verbs are lexically intransitive, and others are lexically transitive. In order to distinguish between lexical and morphologically 9

Here I disregard the positionals, which have a special morphology; see Krämer & Wunderlich (1999:449ff.).

501

Prelexical Syntax and the Voice Hypothesis

derived meaning, some of these operations have to operate in the lexicon (without any morphological reflection), and some in the morphological-syntactic part of the language. The three major classes of lexical verbs, then, have the structures shown in (31).10 (31) a. "unergatives": Voice' VP

b. "unaccusatives": Voice' VP

Ag(x)

Aff(y)

c. transitive verbs: Voice' or VoicepSVS>Ag(x) VP

Aff(y)

Voice' VoiceP~^Aff(y)

VP

Ag(x)

Passive only demotes an agent in the context of an affected object, and Antipassive only demotes an affected object in the context of an agent. It is irrelevant whether the required argument is introduced lexically (by an invisible operation) or morphologically (by a visible operation). The Voice hypothesis, as it stands, is confronted with the problem of generating the required number of arguments to be present: all demoting voice operations must be preceded by a complementary argument-adding voice operation, and vice versa." (32) a. Causative: Voice'

/ \ / \ VP Aff(y)

VoiceP Ag(x)

b. Passive: Voice'

c. Affected Object: Voice'

VoiceP Ag(pro)

VoiceP

VP

Aff(y)

Aff(y)

Ag(x)

d. Antipassive: Voice'

VoiceP Aff(pro) VP

Ag(x)

The actual situation is even more intricate. As already shown, both Causative and Affected Object can be followed by Passive, where an agent is already present. Moreover, Antipassive alternates with noun incorporation, which can be followed by Affected Object, as shown in (33) (Krämer & Wunderlich 1999:467). 10

Since hek' 'break' can undergo Passive and Antipassive (see (28)), it may have started o f f from either (31a) or (31b), and results in one o f the structures in (31c).

"

Jelinek (1998), in her analysis o f the Uto-Aztecan language Yaqui, assumes that Voice only introduces the highest argument, while another functional category, namely Tran (below o f Voice), introduces internal arguments; the underlying verb is considered to be zero-valent, and Tran can be iterative. If a transitive verb is causativized, there are two occurrences of Tran (where the second, the higher Tran, introduces the original Agent), followed by the causativizing Voice. The same structure arises with the applicative introducing a Beneficiary argument; here, however, the Beneficiary is introduced by the second Tran, and the original Agent is introduced by Voice. In this account it is stipulated whether the particular morphology on the verb is associated with Voice or with Trans. Yaqui is a language that has Passive but lacks Antipassive; therefore the problem of demoting an internal argument does not arise.

502

Dieter

Wunderlich

(33) taan=u

kon -lol -t -ik -et/ sell -flower -AO -IMPF -2.NOM 'he's selling you flowers' (lit. 'he's flower-selling you')

INCOMPL=3.ERG

These facts suggest that the voice system is not fully symmetric: Affected objects behave differently from agents; it is generally assumed that a noun can be incorporated only as the lowest argument of the verb (Stiebeis 1998). It is therefore doubtful whether a structure in which an affected object c-commands an agent is possible at all. Let us assume that the only legitimate structure is that in (34), and the operations either add or demote one of the arguments. (34)

Voice' Ag(x) VoiceP VP

Aff(y)

It is questionable whether such a structure is compatible with the Voice hypothesis because now not all of the operations considered so far are licensed in the syntax. In any case, (34) is compatible with the semantic representations given earlier (26-30). Yucatec also exhibits a third voice, the Middle, marked by vowel lengthening combined with high tone, as illustrated in (35) (Bricker & Yah 1981 :xi). (35)

k=u

heek ?

-el

INCOMPL=3.ERG

break.MiD

-IMPF

'It is breaking': λ y BEC BROKEN(y)

The Middle results in an unaccusative verb. However, contrary to what is predicted by the Voice hypothesis, heek is not a voiceless instance of the verb hek (see (28)), rather it is morphologically marked. How can the application of Middle yield something that looks like the underlying representation for both Active and Passive? I will return to this problem in section 3.6.

3.5

Tagalog Voice

Tagalog exhibits several voice operations on the verb, each reflecting the role of the nominative argument: Active Voice (AV), Objective Voice (OV), Dative/Locative Voice (DV), Instrumental Voice (IV), and Benefactive Voice (BV). This voice system is illustrated in the following examples, taken from Foley & Van Valin (1984:135). (Note that some prefixes of the verb surface as infixes.)

Prelexical Syntax and the Voice

Hypothesis

503

(36) a. B-um-ili ang=lalake ng=isda sa=tindahan. PERF.AV-buy NOM=man GEN=fish DAT=store 'The man bought fish at the store' b. B-in-ili ng=lalake ang=isda sa=tindahan. PERF.OV-buy GEN=man NOM=fish DAT=store 'The man bought the fish at the store' c. B-in-ilh-an ng=lalake ng=isda ang=tindahan. PERF-buy-DV GEN=man GEN=fish NOM=store 'The man bought fish at the store' d. Ip-in-am-bili ng=lalake ng=isda ang=pera. IV-PERF-buy GEN=man GEN=fish NOM=money 'The man bought fish with the money' c. I-b-in-ili ng=lalake ng=isda ang=bata. BV-PERF-buy GEN=man GEN=fish NOM=child 'The man bought fish for the child' IV and BV may be thought of as operations that introduce an instrument or a beneficiary, respectively (see (36d,e)), and it is also conceivable that DV introduces a goal (recipient, or location), which is otherwise realized by dative. But there is no evidence that an agent or affected object (=theme) is introduced or demoted by any of the voices. Regardless of the alternation in voice, these core arguments of the verb are realized either by nominative or by the so-called genitive. Although OV is the preferred voice for definite objects (see (36b)) and often less marked morphologically than the other voices, the agent is marked by genitive, which is a structural rather than an oblique case. Therefore, the agent is not pro, and OV cannot be identified with Passive. Moreover, the same set of case markers appear with AV and OV, that is, no case absorption can be observed. These case markers also appear in event nominalizations marked with pag- rather than with a voice marker, see (37) from Schachter & Otanes (1972:163). Thus, the Voice hypothesis does not seem to have much to say about Tagalog. (37) Masyadong mabilis ang=pag-tugtug ng=lalake too fast NOM=NOML-play GEN=man 'The man's playing of the piece is too fast'

ng=piyesa. GEN=piece

That genitive must be a structural case in Tagalog has been shown by several tests (Schachter 1976, Kroeger 1993). Consider the most common type of control constructions in Tagalog, where both the controller and the controllee must be Agent. As the data in (38) (taken from Kroeger 1993:39) show, an alternation in the voice category and the case marking in the dependent verb does not affect the control relation; the gap in the dependent clause can be either understood as nominative, as in (38a), or as genitive, as in (38b).

504

Dieter Wunderlich

(38) a. B-in-alak niya=ng mag-bigay PERF.OV-plan 3sg.GEN=COMP AV-give 'He planned to give money to Mother' b. B-in-alak niya=ng i-bigay PERF.OV-plan 3sg.GEN=COMP IV-give 'He planned to give the money to Mother'

ng=pera GEN=money

sa=Nanay. DAT=mother

sa=Nanay DAT=mother

ang=pera. NOM=money

The control relation is also independent o f the Voice realization in the matrix verb; the controller can be nominative, as in (39a,b), or genitive, as in (39c,d) (Dell 1981:17). ako=ng a. Um-iwas PERF.AV-avoid lsg.NOM=COMP

t-um-ingin AV-look.at

kay=Lorna DAT=Lorna

ako=ng b. Um-iwas PERF.AV-avoid lsg.NOM=COMP

tingn-an look.at-DV

si=Lorna NOM=Lorna

c. In-iwas-an ko=ng PERF-avoid-DV lsg.GEN=COMP

t-um-ingin AV-look.at

kay=Lorna DAT=Lorna

ko=ng d. In-iwas-an PERF-avoid-DV lsg.GEN=COMP Ί avoided looking at Lorna'

tingn-an look.at-DV

si=Lorna NC)M=Lorna

The fact that both controller and controllee can be genitive shows that genitive cannot be an oblique case. There is no evidence that the agent in the dependent verb is introduced by a voice operation. Another set o f data illustrates that the Tagalog voice system and the introduction o f arguments are independent o f each other; consider the following imperatives (from Schachter 1976:506). (40) a. Mag-bigay ka sa=kaniya AV-give 2sg.NOM DAT=3sg 'Give him some coffee'

ng-kape. GEN=coffee

b. Bigy-an mo siya give-DV 2sg.GEN 3sg.NOM 'Give him some coffee'

ng-kape. GEN=coffee

c. I-bigay mo IV-give 2sg.GEN 'Give him the coffee'

ang-kape. NOM=coffee

sa=kaniya DAT=3sg

Although the adressee o f an imperative must be an agent for semantic reasons, (40a) is not the preferred choice. Preferably, either the recipient or the theme is selected as nominative (40b,c), while the agent is realized by genitive. DV or IV on the verb nei-

Prelexical Syntax and the Voice Hypothesis

505

ther introduce nor demote the agent, and there is no morphological evidence that AV operates before DV or IV in order to introduce the agent. Therefore, the most natural assumption is that all three voices shown in (40) operate alternatively on the Semantic Form of the given verb. Tagalog verbs show many idiosyncrasies: for instance, the voice that advances an argument to nominative is often lexicalized; for some reason, the voice that advances the affected object=theme of bigay 'give' is IV rather than OV. For a principled description of the Tagalog data in (40), let us assume that (42a) represents the verb bigay 'give'. The structural positions of the theta-roles are annotated by the features [+hr] for 'there is a higher role', and [+lr] for 'there is a lower role' (see Joppen & Wunderlich 1995, Wunderlich 1997a,b). In the usual accusative or ergative systems, accusative is specified by the feature [+hr], ergative by [+lr], and dative by [+hr,+lr], and they can link only to theta-roles with compatible values. Correspondingly, the highest argument (the agent) is [-hr] (so it cannot be realized by accusative or dative), and the lowest argument (the theme) is [-lr] (so it cannot be realized by ergative or dative). Let us further assume that the Tagalog voice system is neutral with respect to accusative and ergative, but rather uses the feature [+hs] for 'there is a more salient role'. The most salient argument of a clause ([—hs]) must be definite and functions as the syntactic subject (according to the tests given by Kroeger 1993: it is subject to relativization, raising, and conjunction reduction, and can control a secondary predicate); only this argument can become the referential argument of a nominalization, which is often only indicated by intonational means, see the examples in (41). (41) a. mag-sä-säkaw AV-fiit-farm 'x will farm'

b. mäg-sa-säkaw AV-fut-farm 'farmer'

c. hugäs-an wash-DV 'wash at x'

d. hugas-än wash-DV 'sinkw'

The voices, then, determine the most salient argument by assigning the feature [-hs] to the designated argument role, and [+hs] to all others. If Tagalog dative is specified as [+hr,+lr] (like in other systems), genitive as [+hs] ('non-salient for reference'), and nominative is left unspecified, 1 the case distribution shown in (42b-d) arises. This distribution also holds for causativized transitive verbs and is therefore the default pattern for ditransitive verbs. In such a mixed system, dative remains the most specific case responsible for the medial argument (more specific than genitive). Since only one argument can be the most salient one, it is obvious that only one nominative is allowed in a clause. The genitive, by contrast, is compatible with more than one argument.

Latrouite (2000) argues that genitive is the default case, while nominative is marked for t+salient]. However, this proposal is problematic for topic structures: a topicalized argument is always nominative, even if the position it is related to is not; hence, there can be more than one nominative in a sentence based on one verbal clause.

506

Dieter Wunderlich

(42) a.

+hr -lr

Xy +hr +lr

-hr +lr

b. AV +hs gen

+hs dat

-hs nom

c. DV +hs gen

-hs nom

+hs gen

d. IV

+hs dat

+hs gen

λζ

-hs nom

λχ

{Ag(x) & BEC

POSS(z)(y)}

Although many details of the Tagalog phenomena are still very little understood, a first inspection of its voice system reveals that it seems to be at odds with the Voice hypothesis; in particular, none of the various voices sets the agent to pro.13 Since we are dealing with a language in which the categorial distinction verb vs. noun is problematic, and Active Voice is not the preferred choice at all, an alternative hypothesis claiming that voice renders an argument the most salient for reference is not implausible. If one implements this idea into a syntactic framework, the Tagalog voices rather function analoguously to AgrS in that they assign nominative to the selected argument. However, the selection function itself probably cannot be determined syntactically.

3.6

Middle Voice

A further problem for the Voice hypothesis arises from the fact that although the middle presupposes an agent, the latter can never be expressed. It is a well-known fact that the agent even cannot be referred to implicitly in the middle construction, in contrast to the passive. Consider the minimal pair involving purpose clauses in (43). (43) a. A fork was broken in order to impress the audience, b. A fork broke (*in order to impress the audience). For English, one could simply state that unaccusative break forms a lexical pair with the causative verb; the ungrammaticality of the purpose clause in (43b) then follows from the assumption that unaccusatives are voiceless. However, in many languages similar instances are marked morphologically. Sells (1998) argues that the Philippine voices assign a null pronominal to a particular argument. This null pronominal is the real subject. Within the clause, the nominative NP is coindexed with the pronominal subject, and behaves itself as an A-bar-category. In this account, each type of voice sets one particular argument to pro, which seems to be in accordance with the Voice hypothesis. However, differently from the Voice hypothesis, the respective argument is designated for its syntactic function rather than demoted from syntactic function.

507

Prelexical Syntax and the Voice Hypothesis (44) a. German

b. Spanish

c. Fula

Das Fenster

öffnet

sich.

the

opens

REFL

window

La

cuerda se

rompe.

the

rope

splits

Yolnde maßß-ake. door

d. Classical Greek

REFL

close-MID.COMPL

Dii-omai. sink-lsg.MID Ί am sinking d o w n '

Apart from this so-called decausative reading, the Middle usually allows for other readings, too, depending on the meaning of the verb. Those readings are illustrated by the following examples from Classical Greek and Fula (adapted from Kaufmann

2000):14 CIGreek

Loii-omai. wash-lsg.MID Ί wash myself

b. indirect reflexive reading:

CIGreek

Hair-omai moiran. take-lg.MlD part Ί take a part for myself

c. causative reflexive reading:

CIGreek

Didäsk-omai. teach-Isg.MID Ί let myself be taught'

d. modal reading:

Fula

Nde-loot-oto.

(45) a. direct reflexive reading:

SBJ.CL9-wash-MID.INCOMPL 'It is washable'

All these readings require the presence of an agent in the semantics. According to Kaufmann (2000), the common core of all the readings of the Middle is a marked control condition for the event. Both in Active and in Passive voice, the agent is assumed to control the event, that is, it is the only one who is able to instantiate or to finish the event. In the Middle, this default control condition is specified differently so that the agent (the default controller for the event) is excluded. Kaufmann's analysis is in the spirit of the Voice hypothesis insofar as the middle introduces a highest argument, which, however, is not the agent. In the following, I will slightly modify Kaufmann's

All these readings are possible for Spanish se and the Fula middle. For some reason, Classical Greek seems to lack the modal reading, but adds some further readings. Some of the readings in (45) are also possible in German and Russian inherent reflexives, as well as in Icelandic.

508

Dieter

Wunderlich

analysis. Let us assume that the Middle introduces Contr(u) (for 'u is the controller'), thereby demoting Agent, with the option for different instantiations for u, all of which are in the complement set of Agent. 1 (46) Middle:

Contr(u) & Ag(pro) u is identical with one of the internal arguments of the verb, or u is generically bound, or u is empty, that is, bound by -i3.

The different readings of the middle, then, arise by the different instantiations for u, and by a choice of binding pro (either to u, or existentially). 16 In the following, the readings illustrated in (44) and (45) are briefly discussed. (47) a. direct reflexive reading: {Contr(u) & Ag(pro) & BEC CLEAN(y)} with u=y. The event is controlled by the affected person. (In a washing event the person who undergoes washing may control the event). By default, pro is identified with u. Xy {Contr(y) & Ag(y) & BEC CLEAN(y)} b. indirect reflexive reading: {Contr(u) & Ag(pro) & BEC POSS(z,y)} with u=z. The event is controlled by the Recipient/Beneficiary. (In a transfer event the person who profits from the transfer may control the event). By default, pro is identified with u. - 4 λy λ ζ {Contr(z) & Ag(z) & BEC POSS(z,y)} c. causative reflexive reading: {Contr(u) & Ag(pro) & BEC KNOW(z,y)} with u=z. The event is controlled by the affected person, the recipient or beneficiary. This can be done by influencing another person, who functions as the agent. Therefore, in this reading pro is existentially bound. λy λζ 3x {Contr(u) & Ag(x) & BEC KNOW(z,y)} with u=z.

If one is willing to enlarge the ontology of objects, the second option could assume GEN, and the third option could assume 0 as values for u. No other instantiations are in the relevant complement set for the agent: objects that are not participants of the event are irrelevant, and both the existential and the general quantifier would include the agent. The default reading is that pro is bound to u. This is not hard to explain: Even if the agent is identical with the person who controls the event, it is not the agent who controls the event, but another participant of the event. (A reflexive action may be controlled from the point of view of the acting person or from that of the affected person.) If pro is existentially bound, an agent other than the controller is presumed; nevertheless this agent cannot be expressed in syntax (by an oblique agent phrase), probably because it is no longer the highest argument.

Prelexical Syntax and the Voice Hypothesis

509

d. modal reading: {Contr(u) & Ag(pro) & BEC EATEN(y)} with u generically bound. If the controller of an event is bound generically, the event itself is bound generically, and a modal meaning arises. By default, pro is identified with u. (pro cannot be existentially bound because then the agent would be the controller, contrary to what the middle requires.) - > λ y GEN u {Contr(u) & A g ( u ) & BEC EATEN(y)}

e. decausative reading: {Contr(u) & Ag(pro) & BEC CLOSED(y)} with u bound by -.3. By default, pro is identified with u. (pro cannot be existentially bound because then the agent would be the controller, contrary to what the middle requires.) As a result, it turns out that a part of the semantic representation becomes irrelevant; strictly speaking, however, it is not deleted. λ y - i 3 u { C o n t r ( u ) & A g ( u ) & BEC CLOSED(y)} = Xy {BEC CLOSED(y)}

This treatment of the middle is compatible with the Voice hypothesis if the controller of an event is introduced as a more general notion than simply "Agent". The multiplicity of readings obtains because the controller has to be taken from a set of instantiations next to the agent.

3.7

Chinese Cause Voice

Another voice that operates after Agent is established is illustrated by the following examples from Chinese. (Note, though, that there is no morphological indication of this voice.) (48) a. nei-ben shu kan-hua-le Lisi-de yanjing. that-CL book read-blur-ASP Lisi-POSS eye 'By reading that book Lisi's eyes got blurred' lit.: That book read by someone made Lisi's eyes blurred (by inference: Lisi read the book) b. zhe-geyao chi-huai-le ta-de-wei. this-CL medicine eat-get.ruined-ASP he-POSS-stomach 'By taking this medicine his stomach got ruined' lit.: This medicine eaten by someone ruined his stomach. c. nei-ping jiu he-zui-le Zhangsan. that-CL wine drink-get.drunk-ASP Zhangsan 'By drinking this bottle wine Zhangsan got drunk' lit.: That bottle wine drunk by someone made Zhangsan drunk.

Dieter Wunderlich

510

d. neichang qiu lei-de dajia mei weikou chi-fan. that ballgame get.tired-DE everyone not appetite eat 'That ballgame made eveiyone so tired that they had no appetite to eat' Chinese allows for resultative constructions (V-V compounds or ί/e-constructions) to express as the subject of the construction a participant which is not the agent. In a spirit similar to the Middle just discussed, one can assume, following Kaufmann & Wunderlich (Τ 998), that a Cause can be specified by Cause Voice, thereby setting the agent to

(49) Cause Voice:

Cause(u) & Ag(pro) u is identical with one of the internal arguments of the verb.

The first three examples of (48), then, can be represented as follows: (50) a. -{Cause(u) & READ(pro,y) & BECOME BLURRED(z)} with u=y > λ ζ λ y 3x {Cause(y) & READ(x,y) & BECOME BLURRED(z)} b. {Cause(u) & EAT(pro,y) & BECOME RUINED(Z)} with u=y λζ λy 3x {Cause(y) & EAT(x,y) & BECOME RUINED(Z)} c. {Cause(u) & DRlNK(pro,y) & BECOME DRUNK(z)} with u=y &z=pro λ ζ λ y { C a u s e ( y ) & DRlNK(z,y) & BECOME DRUNK(z)}

That the construction illustrated in (48) is different from Passive is shown by the fact that it can itself be passivized.18 The Passive of (48a,b) is given by (51a,b); bei is a preposition that introduces the demoted subject. (51) a. Lisi-de yanjing bei nei-ben shu kan-hua-le. Lisi-POSS eye BEI that-CL book read-blur-ASP 'Lisi's eyes got blurred by (his) reading that book' b. ta-de-wei bei zhe-ge yao chi-huai-le he-POSS-stomach BEI this-CL medicine eat-get.ruined-ASP 'His stomach got ruined by (his) taking this medicine' Since the Cause construction is morphologically unmarked, we would expect it to be restricted, and indeed it is only possible with resultatives. This construction corresponds to the Voice hypothesis insofar as a further thematic role is added to the verb in the highest position. Both the Middle and the Cause construction add an entity (or a more general instantiation function) that puts the agent out of action, but from different 17

18

For obvious reasons, the default interpretation for pro must be existential binding then. However, it is possible to identify pro with another internal participant. However, (48c) cannot be passivized, probably because here the result object ('z') is conceptually identified with the agent of the verb.

Prelexical Syntax and the Voice Hypothesis

511

perspectives: the former focuses on the control of an event, while the latter focuses on a causal relation which is uncontrolled. Obviously the Chinese Cause construction still has to be investigated in more detail. Du hörst den Ruf? Wir sind am Ziel. (Richard Wagner)

4. Concluding Remarks Though I am convinced that words can be decomposed, I do not believe that they can be decomposed in the syntax. Several differences between again and re- show that these adverbial elements operate quite differently: again can, but need not, have scope over PPs and all the argument expressions, while re- never has scope over PPs, nor strong resultatives, nor any operators associated with the argument expressions. It is questionable whether these differences can be captured within a syntactic representation which is blind against the morphology-syntax distinction. The Voice hypothesis strengthens the decomposition-in-syntax account by claiming that the functional category Voice interacts with word meaning in that it adds an agent in the syntax (even if it is demoted). This hypothesis correctly predicts that only "agentive" verbs can be passivized. However, during my cross-linguistic expedition I have found many more voices than at first hearing may be thought of. Even if the Voice account were formulated in terms of morphology, it could only be a first attempt, since too many voices appear which are unexpected from this scenario. Nevertheless, one aspect of the Voice hypothesis has been confirmed: nearly all of these voices presuppose the existence of an agent. Some voices (such as Antipassive) render internal arguments less salient than the agent, while other voices (such as those in Tagalog) render internal arguments more salient than the agent. Particularly interesting are the Middle and the Chinese Cause voice: they add an argument role which puts the agent out of action but cannot be instantiated independently from the participants of the event. Here again, as in the Passive and the Tagalog voices, an internal argument may become more salient then the agent. It is, however, clear that in order to synchronize the several voices reflected in natural languages much future work has to be done.

512

Dieter Wunderlich

References Bierwisch, Manfred (1983). Semantische und konzeptuelle Repräsentationen lexikalischer Einheiten. In: Rudolf Ruzicka & Wolfgang Mötsch (eds.) Untersuchungen zur Semantik. Berlin: Akademie Verlag. 61-99. Bricker, Victoria R. (1978). Antipassive Constructions in Yucatec Maya. In: Nora C. England, Collete G. Craig & Louanna Furbee-Losee (eds.) Papers in Mayan Linguistics. Columbia: Museum of Anthropology, University of Missouri. 3-23. Bricker, Victoria R. & Eleuterio Po'ot Yah (1981). Yucatec Maya Verbs (Hocaba Dialect). New Orleans Latin American Studies Center, Tulane University. Dell, F r a n c i s C. (1981). On Certain Sentential Complements in Tagalog. Philippine Journal of Linguistics 12. 11-28. Dowty, David R. (1979). Word Meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montagues PTQ. Dordrecht: Reidel. Foley, William & Robert Van Valin (1984). Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Hale, Kenneth & Samuel Jay Keyser (1994). On Argument Structure and the Lexical Expressions of Syntactic Relations. In: Kenneth Hale & Samuel Jay Keyser (eds.) The View from Building 20. Cambridge, MA: MIT Press. 53-109. Jackendoff, Ray (1990). Semantic Structures. Cambridge, MA: MIT Press. Jäger, Gerhard & Reinhard Blutner (2000). Against Lexical Decomposition in Syntax. In: A. Z. Wyner (ed.) Proceedings ofLATL 15. University of Haifa. Jelinek, Eloise (1998). Voice and Transitivity as Functional Projections in Yaqui. In: Miriam Butt & Wilhelm Geuder (eds.) The Projection of Arguments: Lexical and Compositional Factors. Stanford: CSLI Publications. 195-224. Joppen, Sandra & Dieter Wunderlich (1995). Argument Linking in Basque. Lingua 97. 123-169. Kaufmann, Ingrid (2000). Medium Voice. Working Papers Theorie des Lexikons 117. University of Düsseldorf. Kaufmann, Ingrid & Dieter Wunderlich (1998). Cross-Linguistic Patterns of Resultatives. Working Papers Theorie des Lexikons 109. University of Düsseldorf. Keyser, Samuel Jay & Thomas Roeper (1992). Re: The Abstract Clitic Hypothesis. Linguistic Inquiry 23. 89-125. Krämer, Martin & Dieter Wunderlich (1999). Transitivity Alternations in Yucatec, and the Correlation between Aspect and Argument Roles. Linguistics 37. 431-479. Kratzer, Angelika (1994). The Event Argument and the Semantics of Voice. Ms, University of Massachusetts, Amherst. Kroeger, Paul (1993). Phrase Structure and Grammatical Relations in Tagalog. Stanford: CSLI Publications. Latrouite, Anja (2000). Case- and Voice-Marker Alternations in Tagalog. Paper read at the 22nd Annual Meeting of the Deutsche Gesellschaft fur Sprache. University of Marburg.

Prelexical Syntax and the Voice Hypothesis

513

McCawley, James D. (1971). Prelexical Syntax. Monograph Series on Language and Linguistics 24. 19-33. [22nd Annual Roundtable Meeting, Georgetown University. Reprinted in: James D. McCawley (1973). Grammar and Meaning: Papers on Syntactic and Semantic Topics. Tokyo: Taishukan Publ. Co. 343-356.] Pustejovsky, James (1995). The Generative Lexicon. Cambridge, MA: MIT Press. Rapp, Irene & Arnim von Stechow (1999). Fast 'Almost' and Functional Heads. Journal of Semantics 16.149-204. Ross, John R. (1972). Act. In: Donald Davidson & Gilbert Harman (eds.) Semantics of Natural Languages. Dordrecht: Reidel. 70-126. Schachter, Paul (1976). The Subject in Philippine Languages: Topic, Actor, ActorTopic, or None of the Above. In: Charles N. Li (ed.) Subject and Topic. New York: Academic Press. Schachter, Paul & Fe Τ. Otanes (1972). Tagalog Reference Grammar. Berkeley: University of California Press. Sells, Peter (1998). The Function of Voice Markers in the Philippine Languages. In: Stephen G. Lapointe, Diane K. Brentari, & Patrick Farrell (eds.) Morphology and Its Relation to Phonology and Syntax. Stanford: CSLI Publications. 111-137. Stechow, Arnim von (1995). Lexical Decomposition in Syntax. In: Urs Egli et al. (eds.) Lexical Knowledge in the Organization of Language. Amsterdam: Benjamins. 81-118. Stechow, Arnim von (1996). The Different Readings of Wieder 'Again': A Structural Account .Journal of Semantics 13. 87-138. Stechow, Arnim von (2000). How Are Results Represented and Modified? Remarks on Jäger & Blutner's Anti-Decomposition. In: Cathrine Fabricius-Hansen, Ewald Lang & Claudia Maienborn (eds.) Approaching the Grammar of Adjuncts. ZAS Papers in Linguistics 17. Berlin: ZAS. 287-308. Stiebels, Barbara (1998). Complex Denominal Verbs in German and the MorphologySemantics Interface. In: Geert Booij & Jaap van Marie (eds.) Yearbook of Morphology 1997. 265-302. Straight, H. Stephen (1976). Decompositional Structure in Yucatec Verbs. In: Marlys McClaran (ed.) Papers in Mayan Linguistics. Vol. 1. Los Angeles: UCLA and American Indian Studies Center. 189-201. Wechsler, Stephen (1989). Accomplishments and the Prefix Re-. NELS. 419-434. Wunderlich, Dieter (1997a). Cause and the Structure of Verbs. Linguistic Inquiry 28. 27-68. Wunderlich, Dieter (1997b). Argument Extension by Lexical Adjunction. Journal of Semantics 14. 93-142.

Unspecificity and Intensionality

Thomas Ede Zimmermann

1.

Introduction

An Unexpected Birthday Present Franzis enters a wine store. She is looking for a bottle of decent Riesling-Sylvaner.1 "Of the twenty customers before you today, every single one bought a bottle of Cacter's Champers to celebrate don 't-ask-me-what, " says the wine merchant. "I guess the wine is for yourself. You know what? I'll give you two bottles for the price of one one is for you and one is for your husband. " Now Franzis and Arnim each own a bottle of excellent white wine — to be consumed as soon as an appropriate occasion arises. [to be continued]

I did not say where this scene is set - be it out of discretion, be it because I forgot (not to say it - but where it happened). In particular, I did not specify any particular wine store. Rather, the wine store remains unspecified. However, this does not mean that Franzis entered an unspecific wine store. In fact, assuming the story is true as told, there would have been quite a specific, particular wine store in which it took place.2

1

2

I must admit that, until very recently, I used to think Riesling-Sylvaner is a blend made of Riesling and Sylvaner grapes - cf. von Stechow (1984:392, example 3) - but it is the same hybrid as Müller-Thurgau (http://www.ccf.at/wein/weisswein/sorten/riesling.htm): "Frisch, süffig, spritzig und versehen mit einem Muskatton präsentiert sich dieser RIESLING SYLVANER und kaum einer weiß, daß es ein MÜLLER-THURGAU ist." [Fresh, palatable, crisp, and with a hint of nutmeg is how this RIESLING SYLVANER presents itself, and hardly anybody knows that it is a MÜLLER-THURGAU.] On the other hand, genetic tests have brought to light that RieslingSylvaner (= Müller-Thurgau) is not a hybrid of Riesling and Sylvaner (although such hybrids do exist), but probably of Riesling and Madelaine royale. I am indebted to an anonymous referee for re-opening the debate and to viniculturist Rolf Blaich of the University of Hohenheim for settling it by supplying the genetic background information. Could there have been more than one such store? Yes, but only one that the story would have been about. I will ignore this complication in what follows and assume that indefinite reference amounts to existential quantification. See, e.g., Dekker (to appear) for some criticism of this orthodoxy.

Unspecificity and Intensionality

515

I did not say which bottle of wine Franzis was looking for either. However, this time, neither discretion nor forgetfulness was the reason for my omission, but love of truth: there just was no particular, specific bottle that Franzis had been looking for. She was simply looking for some bottle or other. In other words, there is no bottle of MüllerThurgau that has been left unspecified in the second sentence of the story. Rather, what Franzis had been looking for, according to the intended and natural construal of that sentence, was an unspecific bottle of Müller-Thurgau. The difference between unspecified and unspecific objects is as important as it is well studied. The former derive from the semantics and pragmatics of indefinites, the latter are the result of the semantic interaction of a peculiar kind of verb and its (usually indefinite) object. Hence, while there may be a way of construing the object in the second sentence as relating to an unspecified bottle, there is no way of reading the first sentence as being about an unspecific shop. For unlike look for, enter does not belong to the class of unspecificityinducin| verbs; it is, as I will from now on say, (referentially) transparent, i.e., not opaque. The wine merchant in the above story did not tell Franzis which bottle it was that all the other customers had bought. He too had a good reason for not doing so: there just was no particular, specific bottle that each customer bought. In other words, there is no bottle of champagne that has been left unspecified by the shopkeeper. Rather, according to the intended and natural construal of the merchant's utterance, depending on each customer there is a corresponding - specific - bottle that he or she took home. Corresponding, dependent objects are at least as well studied as unspecific objects. Usually, they are taken to be the result of one quantifier taking scope over another one. Again, while there may be a way of construing the bottle in the first sentence of the shopkeeper's utterance as being left unspecified, there is nothing that the shop in the first sentence of the story could depend on. And there are more parallels between unspecificity and dependence. In fact, according to an influential doctrine,4 the unspecificity created by referential opacity is but a special case of scopal dependence. I did not say which bottle of wine Franzis and Arnim own. Again the reason is obvious: there just is no particular, specific bottle of wine that both Franzis and Arnim own (according to this story). Rather, Franzis owns one bottle of Müller-Thurgau, and Arnim owns another one. Does this mean that the bottles are, respectively, dependent on Arnim and Franzis? No: it is not true that, depending on each member of the couple, there is a corresponding - specific - bottle of Müller-Thurgau that he or she owns. For the wine merchant did not specify which of the two bottles is to be whose - and until they are somehow distributed, neither really is anybody's. But it is still true that both Franzis and Arnim own a bottle of Müller-Thurgau. Does this mean that they each own an unspecific bottle of Müller-Thurgau? If so, own would be referentially opaque. Well, is it? It is this question the present paper is mainly concerned with.

3 4

The terms derive from Quine (1953), where they are used in a more general sense, though. In this form the doctrine goes back to Montague (1969, 1970), but the basic idea is due to Quine (1956).

Thomas Ede Zimmermann

516

2.

Setting the Stage

Let me make two basic assumptions concerning semantic analysis. The first one is a bias and may be taken as purely terminological: (i) ordinary predicates express relations between individuals. For instance, the ordinary transitive verb kiss expresses the binary relation of kissing, i.e., it denotes the set of ordered pairs of kissers and corresponding kissees. The second assumption is substantial and not undisputed: (ii) singular indefinites express existential quantification. For instance, the indefinite a bottle existentially quantifies over bottles, i.e., it denotes the set of those sets of individuals that overlap with the set of bottles (= the extension of bottle). I am making the second assumption largely for the sake of definiteness, orthodoxy, and simplicity. In particular, a dynamic view, according to which indefinites express properties, would also do for my present purposes (and may be more adequate in general), but it would complicate matters in a distracting way. Given certain natural ways of combining meanings, the assumptions (i) and (ii) have an immediate consequence for the truth conditions of sentences of the form (*) χ Rs an N, where χ is an individual-denoting expression (e.g., a proper name or a definite description), R is a transitive verb, and TV is a singular count noun. If R is an ordinary transitive verb, (*) is true if, and only if, the extension of Ν overlaps the set of individuals to which the referent of χ stands in the relation expressed by R. In particular, if R is an ordinary transitive verb, then sentences of the form (*) unequivocally support the following inferential schemata (where M i s an arbitrary count noun): Existential

Impact5

From χ Rs an Ν infer: There is at least one N. Extensionality6 From χ Rs an N, Every Ν is an M, and Every Μ is an Ν infer: χ Rs an M. Specificity From χ Rs an Ν infer: Some (specific) individual is Red by x. The three schemata do not fully characterize ordinary transitive verbs. From a logical point of view, there may be verbs that obey each of them without being (reducible to) 5

6

Existential Impact, as it is understood here, may hold even if the indefinite object fails to quantify over existing objects. As a case in point, Julius may worship a Greek goddess, even though goddesses (of any nationality) are mythical beings and as such do not exist. For, somewhat paradoxically, the indefinite a Greek goddess may be taken to denote an existential quantifier quantifying over non-existent objects - thus preserving Existential Impact. Such a view presupposes a Meinongian ontology ä la Parsons (1980) or Zalta (1988); to avoid terminological hardship, one may follow the latter and distinguish Existential Impact from Existential Impact. In any case, I will ignore reference to Meinongian objects for the rest of this paper. Extensionality implies, but is not implied by substitutivity of extensional nouns salva veritate (which may not be an inference schema).

Unspecificity and Intensionality

517

binary relations between individuals. But all known counter-examples to ordinariness appear to violate at least one of the inferences. In fact, the most popular ones, seek and owe(-to-z), violate all three. Moreover - and this will become important in the sequel the inference schemata are independent of each other. From a logical point of view, there may be verbs that violate any one (or two) of them without violating the other(s). As a matter of fact, various combinations of violations are attested. For instance, remember preserves only Specificity, whereas raise supposedly7 violates only Extensionality. According to the standard Montagovian approach to opacity, failure of Extensionality and Specificity show in the (lowest) types of denotations of the verbs - and more specifically in the part of the types of the second (direct object) argument. If a transitive verb violates Extensionality, there must be something intensional about its argument type. And, as was mentioned above, if it violates Specificity, it would have to take scope over the indefinite object - which means that the argument type must be at least as high as that of a quantifier. Using familiar notational devices, I will let the letters t, e, and s stand for the (basic) types of truth values, individuals, and situations (= world/time pairs), respectively, and have pairs ab of types indicate the type of functions from type a to type b. In particular, the type of a (unary) quantifier will be (et)t; and whenever the extension of an expression is of some type a, then its intension will be of type sa. Given these conventions, the preceding paragraph boils down to this: (a) whenever a transitive verb lacks Extensionality, then the type of its second argument will contain an s; and (b) whenever it violates Specificity, then that type will be at least (et)t. In the former case, the type in question need not start with an s 8 because, e.g., for all practical purposes, the types s((et)t) and (et)(st) are equivalent; in the latter case the type need not be exactly (et)t because lack of Specificity may also arise when an intensional predicate takes scope over the quantifier, thus resulting in an argument type s((et)t). It is worth pointing out that the reversals of (a) and (b) do not hold. For one thing there are (highly artificial) relations between individuals and quantifiers that satisfy Extensionality, although in general their second argument is index-dependent.9 More interestingly, there are essentially higher-order verbs that induce specificity - one of them is remember, which relates an individual χ and a quantifier Q at an index i iff there is an index j before i such that the set of y individually remembered by χ at / is in the extension of the quantifier at j. Using type-logical notation, this truth condition may be expressed as follows:

7

8 9

... adapting the analysis of rise in Montague (1973) - rise as in Prices are rising, of course, not as in The sun is rising (for which vide Montague 1969). It should be noted that rise does satisfy extensional substitutivity (for which vide fn. 6). It need not be quantificational either, as the case of rise shows, where it is simply se. The relation λίλζ)Αχ( Ξ/)[ι ί j a ß(/)(Av-L)] (defined using type-logical notation) is a case in point: it trivially satisfies Extensionality but can distinguish between certain co-extensional quantifiers - like XjkPP(a) and = i * λ P ( a ) ] ν [ / ' * / ' * λ / 3 = λγ±]] (where a and i* are some fixed individual and index).

518

(R)

Thomas Ede Zimmermann (3j)[j < / a

Q(j)(XyR{i){x,y))]

where < is temporal precedence between (world-sharing) indices and R is the relation that holds between two (specific) individuals at a given index iff, at that index, the first of them remembers the second one. If Q happens to be an existential quantifier, (R) implies that there is a specific individual to which χ stands in the relation of remembrance;10 in other words, remember satisfies Specificity. At the same time, however, (R) does not render remember transparent, i.e., even if (R) is valid, there need not be a relation S such that (R) is true whenever the following holds: (S)

m&ySMx,y))

Of course, according to (R), remember also violates Extensionality - thence the s - but, judging from its Specificity behaviour, its higher-orderness may come as a surprise. On the other hand, the violation of Existential Impact shows that, on top of its intensional character, there is something odd about the verb, which is reflected by the analysis (R); for (R) implies that the quantified argument, and hence in particular its relativization, 1 contributes its intension to the extension of the sentence. It is this shift in the relativization, then, that leads to higher-orderness despite Specificity. To sum up the observations made so far, there is a certain match between the inference schemes considered above and the argument types of transitive verbs. The match is not perfect,12 but aberrant inference behaviour invariably leads to predictable type complications. The situation can be depicted by the following table:

10

" 12

Note that (as was already pointed out in Montague 1968) y does not have to exist for R(i)(x,y) to be true. In this respect remember is analogous to worship (cf. fn. 5 above) and may thus be called temporally extranuclear (adapting the terminology of Parsons 1980), thus replacing the term temporally opaque, as it was used in an earlier version of von Stechow (to appear). The relativization of a quantifier Q is that set Μ such that Q = D(M), where D is a (conservative and invariant) determiner. Unlike D, Μ is uniquely determined: cf. Johnsen (1987). This need not be an imperfection of the choice of inference schemata but may also be due to the expressiveness of language. For instance, as was shown in Zimmermann (1985), transparency does not correspond to any inference scheme expressible in the PTQ fragment of Montague (1973).

Unspecificity and Intensionality

519

Table 1: Nominal argument types of verbs Existential Impact + Specificity + +

Extensionality + —

—

+

—

—

Argument Type(s) e se (et)t s((et)t),((se)t)t,s(((se)t)t)

The table, I take it, is self-explanatory. The three options in the last row result from an indetermination of the source of intensionality by the inference schemata. This will be of no concern in what follows. Instead, I would like to point to the penultimate row, where a failure of Existential Impact or Specificity (or both) and a simultaneous preservation of Extensionality leads to extensional higher-orderness. Given the above classification, one would clearly expect such verbs to exist. What do they look like? What would they look like? There are three cases. Case 1: Extensional verbs that fail on both Existential Impact and Specificity are easy to imagine. Any relation R between individuals and quantifiers can be turned into one by evaluating its argument at the same index as the predicate itself: faXQhcR(i)(x, yQ(i)). In the case of look for one would thus obtain a reading that licenses inferences like the following: 13 Arnim There There .·. Arnim

is looking for a cow with white ears. are no cows with white ears. are no unicorns. is looking for a unicorn.

No doubt, the first premise can be construed so as to make the inference valid - but this is a special reading (or use) of look for which, other things being equal, should not be explained by lexical ambiguity. 14 And no verb seems to force the above inference. Other extensional relations that defy both Specificity and Existential Impact are readily dreamed up. For instance, one may imagine a verb aviod which worked a bit like avoid, only that a sentence like Arnim aviods a pub would mean that there is no pub that Arnim enters. The relevant reading could be obtained using the following analysis of aviod·. XiAjQXx—iQ(XyE{i)(x,y)), where - maybe as a result of contextual

The example is inspired by an example in an earlier version of von Stechow (to appear), where the following surprising claims are made: 'Elisabeth Engdahl once told me that she had never seen a cow with white ears and put forward the generalization that there were no such cows. For several years I tried to falsify the generalization, which I could not.' The reading is discussed in Forbes (2000:166, fn. 43). I think it is best analysed as a special case of the phenomenon first discussed in Bäuerle (1983) - independence of quantifier scope and evaluation index.

Thomas Ede Zimmermann

520

enrichment - Ε denotes the relation of entering. Again, no such verb, and no other extensional, unspecific verb without Existential Impact, seems to exist. Case 2: An extensional verb that respects Specificity but not Existential Impact seems hard to imagine. In order to see why, one may restrict attention to indefinite objects (existential quantifiers) so that an extensional relation boils down to a relation between individuals and sets (the restrictors of indefinite objects). Specificity then demands that any relation between an individual χ and a set Μ is recorded by a relation between χ and a singleton {7} (= the individual in the conclusion of Specificity); the problem is to make a natural choice for y in case Μ is empty - which would have to be true for at least one x, lest Existential Impact comes out valid. Now while such a constellation can certainly be realized by purely formal means,15 it is not obvious that there exist any natural, practically interesting or relevant, relations of that kind - which may ultimately explain the lexical gap. Case 3: Finally, there ought to be extensional verbs that relate subjects to unspecific objects and at the same time guarantee Existential Impact. They are easy enough to imagine. In fact, they have been said to exist: The following example, due to Mats Rooth, is a case in point: (7)

Mats owns 75% of the ball bearings in the basement.

(7) can be true if Mats holds a 75% share of the objects mentioned without thereby owning any particular ball bearing; it thus has a nonspecific reading in addition to the more obvious one concerning certain individual objects. On the other hand, (7) seems to allow for substitution of extensional equivalents: if the only metallic objects in the basement are ball bearings, it is obvious that (7) implies: (8)

Mats owns 75% of the metallic objects in the basement. Zimmermann (1993:152) 16

As it stands, the example does not quite fit the format of the above inference schemata; but then the final sentence of the introductory story does. As expected, all of the types listed in table 1 appear to be realized by some verb or other, thus confirming the analysis of inference failure in terms of abstractness of denotation types. However, things are not that simple.

The following relation between individuals and quantifiers turns out to do the trick: hXQXx(3y)[Bi (x,y) λ [ - . β ( λ ζ ζ = y) ν Q = (λΡ />(>>))]], where Β is a (not too trivial) relation between individuals. A footnote concerning the question of whether (7) has a specific reading has been omitted; I will address the issue in due course.

Unspecificity and Intensionality

3.

521

Extensional Ownership

An Unexpected Birthday Present fcont. 7 On her way home from the wine store, Franzis meets her friend Christiane, who wants to know where she bought the two bottles. "I only bought one of them and got the other one for free, " Franzis explains. "One is for Wladimir, though. " "Which one?" asks Christiane, whereupon Franzis replies: "Whichever I choose; his is the bottle that is not mine. "

[to be continued]

In the context of Franzis's reply, it seems fair to read the possessives as being about possession; the content of her reply is thus preserved by the following reformulation: (1)

Arnim owns the bottle that Franzis does not own.

Franzis is not only witty in her reply, she is also right. So the paraphrase (1) of her utterance should come out as true under the circumstances; it should be assigned at least one true reading, that is. Following Mats Rooth's suggestion, (1) ought to have two readings. On the first, specific one (la), the definite description takes scope over the verb; on the other, unspecific reading (lb), the quantifier denoted by the description is an argument of the verb: 7 (la)

(iy:B(y)A^0(f,y*))0(a,y*)

(1 b)

0(a,(iy.B(y)A^0(f,y*)))

Of course, (la) is false: there are (at least) two bottles that Franzis does not own, so that (la) already fails on the restrictor of the Russellian description. But the scope does not fare much better, at least under the (admittedly unlikely) assumption that Arnim happens to own no bottles apart from the (unspecific) one donated by the wine merchant. How about (lb), then? Since no assumptions about the precise truth-conditions induced by the unspecific use of own have been made, it may seem as if the truth of (lb) could just be stipulated, as a matter of lexical meaning: whatever the relation Ο may be, given the above scenario, it should relate Arnim to the quantifier denoted by the bottle that Franzis does not own. Alas, at the end of this reasoning absurdity is on the lurk. For what is that quantifier? It applies to a given set of individuals if that set contains the unique bottle that Franzis does not own; since there are (at least) two bottles Franzis does not own, the quantifier does not apply to any set. In other words, the

In case they are not obvious, here is a quick tour through the notational conventions employed in this section: ι and 3 are the Russellian and binary existential quantifiers (determiners); χ * is [the extensional version of] Montague's type lift, i.e., λΡ P(x)', and if Q is a binary quantifier, ' (Qx : (py and ' (Qx : φ)ψ' respectively abbreviate ' Q(Ax (VJ')[C(J)

[Rj(x,Q)

{Qjy)Oj{x,y)}]

It thus turns out that there is something circular about this (largely Moltmannian) analysis of ownership. For although the details may be negotiable, it is obvious that any principle excluding unspecific ownership at claim-compliant indices would have to involve the relation R of unspecific ownership. And this is where circularity is lurking: taken together, (FM) and (CP) cannot be used as explications of the meaning of own?5 Apart from circularity, more obstacles would have to be removed to turn (FM), or anything like it, into a coherent analysis. In fact, there is reason to doubt that this is even possible. It stems from the appearance of extensionality in own, which motivated Rooth's analysis. After all, in many cases, truth-preserving substitution of co-extensional objects under own is possible even if the verb is construed unspecifically. Arnim, for instance, owns a blue BMW (a German car) ba. This is a case of specific ownership, established by some contract ca. According to the intensional analysis of own, the following formula should thus be true at a present and actual index /': (9)

(3Y : BMW(y) Α Βί(Γ))(3α)[α(ε)

A (V/)[c(/)

Oj{a,y)]]

Despite being inadequate analyses, the principles may still be true. Cf. Zimmermann (1999:553), for a similar case ("unspecific" transparency postulates).

Unspecificity

and Intens

527

ionality

Under the assumption that ba and ca satisfy the relevant predicates, (9) is true, provided that the contract specifies ba, which I take to be the case.26 So the analysis reflects the fact that (10) is true: (10)

Arnim

owns a blue

BMW.

However, does it also reflect the fact that (10) is not false? For this to be the case, (10) must not be assigned any false reading. In particular, (10) would have to be true on its non-specific reading, i.e.: (11) (3c)[C,(c) λ (Υ/)[φ·) -> (By: BMWj(y)

A

Blj(y))Oj(a,y)]]

That (11) is true under the circumstances is not at all obvious: the contract specifies the marque, the chassis number, and various other details about the car, but not its colour. Hence (11) does not appear to be true in virtue of the content of ca. However, it may still be true as a result of an implicit understanding of compliance with a claim to possession. The following general principle would see to that: (GP) [ΛΓ.ΟΟ Λ C.(c)

Λ

(V/)K/)

Oj(x,y)]]

(V/)[C(y)

Nj(y)]

(GP) says that, whenever some specific object y with property Ν is owned in virtue of a claim c, then c specifies that - i.e., it only complies with indices at which -y has property N. Apart from being ad hoc, (GP) does not seem to be general enough. For the kind of reasoning that leads from Arnim's ownership of ba to the truth of (10) can also be applied in cases of unspecific ownership. Indeed, since the two bottles of RieslingSylvaner happen to be the objects in Franzis's bag (to adapt Mats Rooth's example), Arnim owns an object in Franzis's bag, although the wine merchant certainly did not put it that way. However, generalizing (GP) to such cases of unspecific ownership just does not work: [N. = M. A C.(c) Λ (V/)[cC0 - > ßY:

N ^ O ^ y ) ] ]

(Vj)[c(j)^(By.M.(y))Oj(x,y)]

This principle would have the effect that, whenever someone owns an unspecific N, he or she, by virtue of the same claim, also owns an unspecific M, provided that Ν and Μ are co-extensional. However, by the reasoning in the previous section, this cannot be right: if Ν is the property of being a bottle in Franzis's bag and Μ is the property of being an Ν that Arnim does not own, then Ν and Μ are co-extensional; but at no index Actually 1 don't - because of general reservations about simplistic de re ascriptions - but then my qualms would only distract from the main point about the purported de dicto reading (11), which could not be corrected by a more sophisticated de re analysis (along the lines of Kaplan 1969, say).

528

Thomas Ede Zimmermann

complying with what the wine merchant said could Arnim possibly own a bottle with property M. The examples also show that determining the indices (or "situations") quantified over by the verb own is not just a matter of enriching ownership claims with factual information. For one would still need a criterion for distinguishing facts about the colour of Arnim's car from facts about his possessions; invoking the latter would lead to familiar trouble, whereas the former appear to be harmless when it comes to establishing claims like (10). Nor is it a matter of restricting indices to actuality. For it would still be true that Arnim owns a bottle even if - God forbid - Franzis dropped them before they could ever be distributed - in which case there would be no actual 27 situation in which Arnim eventually owns a particular bottle. Maybe compliance with claims for possession is a special case of counterfactuality 28

based on similarity. After all, given the circumstances, (12) could be roughly paraphrased as (13), which in turn may be formalized along the lines of (14): (12) Arnim owns a bottle of wine. (13) If Arnim's claim were redeemed, there would be a specific bottle of wine that Arnim owned. (14) (3c : C.(c))(Y/)[$;(c)(7) -> (3y:

BWj(y))Oj{a,y)\

The formalization comes with an underlined restriction over counterfactual indices selecting those scenarios among the claim-compliant ones that are maximally similar to the evaluation index i. Given (CP), claim compliance guarantees that only indices are taken into account at which ownership relations are completely specific; and maximal similarity sees to it that everything else is left as is. Sentence (12) does not say anything about what would have happened if the bottles had actually been distributed. Given the circumstances, such a distribution is most likely to be made by Franzis so that in the most realistic scenarios in which Arnim owns a particular one of the two bottles, he owns the bottle that Franzis handed over to him. But even given the circumstances, (12) does not imply that Arnim owns a bottle to be handed over to him by Franzis. Rather, the hypothetical distribution is to be thought of as taking place without anyone performing it. Moreover, realistically, distribution takes time. But (12) does not say that Arnim owns a future bottle. Rather, the hypothetical distribution is to be thought of as taking place at the index time. It thus seems that a peculiar similarity relation (between co-temporal indices) would have to be associated with unspecific ownership. In particular, differences in unspecific ownership

I am indebted to Joachim Sabel for bringing up this horror scenario. I spare the reader the usual references (to Stalnaker 1968 and Lewis 1973, that is). The idea of employing the technique came up in a conversation with Arnim von Stechow; cf. fn. 23.1 do not recall whether either of us defended it. I am indebted to an anonymous referee for pointing out an error in the original formulation of (14) below.

Unspecificity and Intensionality

529

would have to weigh less than other differences. The following assumption about claims c and the underlying system $ of similarity spheres would have this effect: (AS)

$, (c) = {/' I Ρ, = Pj, for any ownership-independent P}

In (AS), the notion of ownership-independence must be taken so as to include properties like being a bottle in Franzis's bag while at the same time excluding the property of not belonging to Arnim. A precise definition of this notion seems hard to come by. Here is an approximation: (Ol)

(Vj-.O^Oj )(3k:Ok=Oj )Pk=Pj

If satisfied at arbitrary indices /, the criterion says that arbitrary changes in (unspecific) ownership are consistent with Ρ staying put. For instance, distributing the bottles among Arnim and Franzis is consistent with them being the objects in Franzis's bag, but not with them being the objects in the bag that do not belong to Arnim. Whatever the precise notion of similarity may be, the idea behind the counterfactual analysis of unspecific ownership should be clear enough: it is ownership with respect to those indices that only differ from the origo in that all belongings have been distributed. This approach has at least two advantages: • It can account for the appearance of extensionality in own. • It can handle the counter-examples to Rooth's extensional analysis. The first point is due to the fact that, surely, most predicates would have to come out as ownership-independent - whatever that may turn out to be. Hence, for most predicates it would hold that their extension at claim-compliant indices would coincide with their actual extension (or, more precisely, their extension at the evaluation index). Consequently, if two ownership-independent predicates extensionally coincide, then they will also coincide at the indices relevant in determining unspecific ownership - which suffices for a salva veritate substitution. Still, and this is the second point, not all predicates are ownership-independent. In fact, the troublemakers of the previous section certainly are not. There are ways of changing - i.e., specifying - ownership that necessarily affect the extension of, say, the predicate bottle that neither Franzis nor Arnim owns. In fact, substitution of bottle by this particular (co-extensional) predicate does not preserve the truth of the proposition expressed on the counterfactual account of ownership: the claim-compliant indices are all such that they contain a bottle that Arnim owns without containing one that he owns but neither Franzis nor Arnim owns. Hence, matters of circularity notwithstanding, the intensional analysis of unspecific ownership - particularly in its counterfactual guise - clearly outperforms its extensional rival. But it has its price. For it rests on the assumption that unspecific ownership can be specified - and hence change - with everything else remaining the same. For instance, Arnim's unspecific ownership of a bottle of Riesling-Sylvaner is licensed by

Thomas Ede Zimmermann

530

the wine merchant's donation, the content of which specifies certain counterfactual situations in which Arnim owns one of the bottles. These situations only differ from the actual (purported) course of events in that somehow, magically and clearly unbeknownst to any of the participants, the bottles have been assigned to their respective owners. Ownership, it would seem, does not supervene on any physical, let alone social, constellations. This may sound absurd. But I think it is slightly less absurd than it sounds. A detailed look at the (as yet unfinished) story may clear this up. The story describes a possible scenario (an index) i in which Arnim unspecifically owns one of two bottles, which means that there are scenarios j (and k) which are just like / except that Arnim owns § and Franzis | (and vice versa). Surely, i, j, and k are mutually distinct, because they disagree in matters of ownership. One thing Franzis knows (in /') is that she is not in j: the bottles have not yet been distributed, and so Arnim does not own either of them. But, then, how does she know this? How could she know this, given the fact that i and j only disagree on ownership-dependent matters? The answer is strikingly simple: Franzis knows that there is not magical distribution. In other words, for Arnim to own a specific bottle, someone would have had to distribute the two bottles, which - as Franzis knows - has not happened. 29 So Franzis can safely exclude scenarios like j and k - even though they are metaphysically possible and, given the special system $, strikingly similar to /. The price to be paid for the counterfactual analysis of unspecific ownership lies in the acceptance of certain remote metaphysical possibilities.

5.

Unspeciflable Ownership

An Unexpected Birthday Present fcont. 7 After Arnim's phone call Franzis takes one of the bottles out of her bag and puts it in the fridge for her dinner, leaving Arnim's bottle on the kitchen table. [to be continued] Now she is in the know. She knows which of the bottles belongs to Arnim and which is hers. It was her decision alright, but once she made it, the die was cast: Arnim owns § and hers is The ownership relations thus are exactly as in the spontaneous distribution situation j. It is now tempting to go for a little Ockhamian reduction on the excessive metaphysical price by trying to eliminate unspecific ownership altogether: Couldn't it be that Franzis had been in / s world all along? In other words, couldn't it be that, of the above three indices, only j and k constitute real metaphysical possibilities? If so, there would be no need for an additional index i at which the ownership is undetermined. For all she knew before her bold decision, Franzis could have been in j or in k. But now she

She could be wrong about this, but I assume that she is not. One obvious consequence of this train of thought is that knowledge must be ownership-dependent - if only for its factivity.

Unspecificity and Intensionality

531

knows it was j, not k. Of course, she could have decided otherwise. But that just means that she could have been in k all along. The metaphysical and methodological profit of this turn is not to be underestimated: no unspecific ownership, no magical distribution, no mysterious similarity correlated with an ill-understood notion of independence. The new view may cause some initial indisposition - just what is it that makes the specific ownership relations hold before distribution? - but maybe that is just a matter of sharpening intuitions and dropping habits.30 What happens to unspecific ownership if there are no situations in which it holds as a relation? In a sense, it is no longer with us. In another sense, it still is. Take j, the situation that Franzis was in at the beginning of the story, i.e., after being presented the bottles. At that time, (4) was true: Arnim owned one of the two bottles - on this all theories agree. However, the present minimalist approach has it that Arnim's owning a bottle was unequivocally true, and true in virtue of him owning g. Of course, no distribution had as yet taken place, which is why nobody knew, nobody could have known, that Arnim owned A; on the other hand, at the time it was known to Franzis that Arnim owned either g or So, according to this minimalist approach, Franzis's knowledge was truly disjunctive. She was thus in a position to claim (4) without being able to specify any bottle owned by Arnim. The bottle is unspecified. Of course, this in itself does not make the ownership relation unspecific. The particular flavour of so-called unspecific ownership derives from the fact that, in the case at hand, no one could have specified the bottle owned by Arnim. It was unspecifiable, for epistemic reasons. The principal problem of the counterfactual approach was the duplication of indices - the assumption that possible situations could differ with respect to ownership and nothing else. Unfortunately, the reduction of unspecificity to unspecifiability still suffers from the same defect, if to a lesser extent. But duplication does occur whenever, by accident, distribution does not occur. The construction of a pertinent scenario is left to the reader. The upshot is that the elimination of unspecific ownership is not the M25.31 There is another, more radical reduction of unspecific ownership to unspecifiability.32 According to it, the wine merchant's speech act does not establish any ownership relation. In other words, Arnim would not own any bottle - specifically or unspecifically - before Franzis picks out hers. All he has is a claim to a bottle, but that does not mean he owns one. In other words, sentence (4) would be literally and unequivocally false even after the merchant's present. Appearances to the contrary would have to be explained, presumably pragmatically. I do not know of any such pragmatic explanation, but I guess the strategy would be to somehow re-interpret (4) as It is almost as if Arnim owned a bottle. In the absence of any specific analysis, I find the proposal hard to judge. In any case, the problems in the semantic analysis of own seem to reappear as problems with the complex predicate have a claim to. 30

31 32

The reduction of de diclo to de re attitudes (Slater 1963) tends to cause the same kind of nausea as does the epistemic approach to vagueness (Williamson 1992). The road to Ockham, that is. Martin Stokhof (p.c.) suggested something along these lines to me.

Thomas Ede Zimmermann

532

6.

Conclusion

The driving force behind the present paper was the question of the relation between the two concepts that make up its title. According to standard semantic analysis, the two should be independent of each other, and yet empirically they appear to co-occur. However, the correlation has been called into doubt, with reference to a purported counterexample, own. The chief aim of the present paper was to dispel the doubt: either own is intensional, or else it does not give rise to unspecific readings. What remains is the correlation, and a need to explain it. An Unexpected Birthday Present, The End Arnim comes home just after midnight. He sees the bottle on the table. Franzis enters. "This is yours, " she says. "Iput mine in the fridge. " "No, this is yours too. I am giving it to you. After all, it's your birthday today. Well, then: Happy birthday to you, Franzis!" Yes, and happy birthday to you, Arnim.33

References B ä u e r l e , R a i n e r ( 1 9 8 3 ) . P r a g m a t i s c h - s e m a n t i s c h e A s p e k t e der N P - I n t e r p r e t a t i o n .

Manfred Faust et al. (eds.) Allgemeine Sprachwissenschaft, Sprachtypologie linguistik. Tübingen: Narr. 121-131. Dekker, Paul (to appear). M e a n i n g and U s e of Indefinite Expressions. Journal

In:

und Textof

Logic,

Language, and Information. Forbes, G r a e m e (2000). Objectual Attitudes. Linguistics

and Philosophy

23. 141-183.

This paper grew out of a seminar that I co-taught with Roger Schwarzschild at Rutgers University in Spring 2000. I am greatly indebted to Roger, the participants of the seminar, two anonymous referees, and especially Arnim von Stechow, who came visiting to present the abovequoted earlier version of von Stechow (to appear) and talk about all things opaque. While writing the present paper, I frequently felt the urge to call him or send him an email, which - for obvious reasons - I couldn't do; I am sure the paper (and I) would have immensely profited from his criticism. I would like to apologize for the clumsiness of the title: Opacity may have been slightly better than Unspecificity (which even my word processor rejects), but then traditionally opaque and intensional are synonyms. Maybe 1 should have applied de Morgan's Law to obtain Specificity or Tensionality.

Unspecificity and

Intensionality

533

Johnsen, Lars (1987). There-Sentences and Generalized Quantifiers. In: Peter Gärdenfors (ed.) Generalized Quantifiers: Linguistics and Logical Approaches. Dordrecht: Reidel. 93-107. Kaplan, David (1969). Quantifying In. In: Donald Davidson & Jaakko Hintikka (eds.) Words and Objections: Essays on the Work of W. V. Quine. Dordrecht: Reidel. 206-242. Lewis, David K. (1973). Counter/actuals. Oxford: Blackwell. Moltmann, Friederike (1997). Intensional Verbs and Quantifiers. Natural Language Semantics 5. 1-52. Montague, Richard (1968). Pragmatics. In: Raymond Klibansky (ed.) Contemporary Philosophy. Vol. I: Logic and the Foundations of Mathematics. Florence: Nuova Italia Editrice. 102-122. Montague, Richard (1969). On the Nature of Certain Philosophical Entities. Monist 53. 159-195. Montague, Richard (1970). Universal Grammar. Theoria 36. 373-398. Montague, Richard (1973). The Proper Treatment of Quantification in Ordinary English. In: Jaakko Hintikka, Julius Moravcsik & Patrick Suppes (eds.) Approaches to Natural Language. Dordrecht: Reidel. 221-242. Parsons, Terence (1980). Nonexistent Objects. New Haven: Yale University Press. Quine, Willard Van Orman (1953). Reference and Modality. In: Willard Van Orman Quine (ed.) From a Logical Point of View. New York: Harper and Row. 139-159. Quine, Willard Van Orman (1956). Quantifiers and Propositional Attitudes. Journal of Philosophy 53. 177-187. Slater, B. Hartley (1963). Talking about Something. Analysis 93.49-53. Stalnaker, Robert (1968). A Theory of Conditionals. In: Nicholas Rescher (ed.) Studies in Logical Theory. Oxford: Blackwell. 41-55. Stechow, Arnim von (1982). Book Review: John Lyons 'Semantics'. Beiträge zur Geschichte der Deutschen Sprache und Literatur 104. 256-267. Stechow, Arnim von (1984). Structured Propositions and Essential Indexicals. In: Fred Landman & Frank Veltman (eds.) Varieties of Formal Semantics. Dordrecht: Foris. 385^403. Stechow, Arnim von (to appear). Temporally Opaque Arguments. In: Festschrift for Andrea Bonomi. Williamson, Timothy (1992). Vagueness and Ignorance. Proceedings of the Aristotelian Society 66. 145-162. Zalta, Edward N. (1988). Intensional Logic and the Metaphysics of Intentionality. Cambridge, MA: MIT Press. Zimmermann, Thomas E. (1985). A Note on Transparency Postulates. Journal of Semantics 4. 67-77. Zimmermann, Thomas E. (1993). On the Proper Treatment of Opacity in Certain Verbs. Natural Language Semantics 1. 149-179. Zimmermann, Thomas E. (1999). Meaning Postulates and the Model-Theoretic Approach to Natural Language Semantics. Linguistics and Philosophy 22. 529-561.

List of Publications

Articles Stechow, A. von (1969). Zum theoretischen Status von Buchstabenübersetzungsregeln. In: D. Wunderlich (ed.) Linguistische Reihe - Probleme und Fortschritte der Transformationsgrammatik. Munich: Max Hueber. 101-108. Stechow, A. von (1970a). Aspekte zur Bewertung von generativen Grammatiken. Linguistische Berichte 9. 19-28. Stechow, A. von (1970b). Sprachwissenschaft vs Linguistik - Bemerkungen zu Leo Weisgerbers 'Hat das Wort "Muttersprache" ausgedient?'. Muttersprache 80. 396-399. Brockhaus, Κ. & A. von Stechow (1971). On Formal Semantics: A New Approach. Linguistische Berichte 11. 7-35. Stechow, A. von (1971a). Unverständnis oder Unverständlichkeit - Replik zu Leo Weisgerber. Muttersprache 81. 102-105. Stechow, A. von (1971b). Zur Theorie der Präsupposition. Münchener Papiere zur Linguistik 1. 1-18. Stechow, A. von (1971c). Ein Formalismus zur Darstellung von einfachen Übersetzungszusammenhängen. Folia Linguistica 4. 215-225. Stechow, A. von (1972). Grammatik als Theorie der Argumentation in natürlichen Sprachen? Arbeitspapiere des Instituts für Sprachwissenschaft der Universität Bern 4. 41-52. Klein, W. & A. von Stechow (1973). Introduction - Functional Generative Grammar in Prague. In: W. Klein & A. von Stechow (eds.) Functional Generative Grammar in Prague. Kronberg: Scriptor. VII-XXX. Stechow, A. von (1973). Syntactic Analysis of Italian. In: A. Zampolli (ed.) Linguistica Mathematica e calcolatori. Florence: Olschki. 541-562. Brockhaus, K. & A. von Stechow (1974). Mathematische Verfahren in der Linguistik. In: Η. L. Arnold & V. Sinemus (eds.) Grundzüge der Literatur- und Sprachwissenschaft. Munich: DTV. 61-90. Pause, Ε. & A. von Stechow (1974). Grammaire comparee transformationelle de Pallemand et du fran9ais. In: C. Rohrer & N. Ruwet (eds.) Actes du colloque Franco-Allemand de Grammaire Transformationelle. Tübingen: Niemeyer.

List of Publications

535

Stechow, A. von (1974). ε-λ-kontextfreie Sprachen - Ein Beitrag zu einer natürlichen formalen Semantik. Linguistische Berichte 34. 1-33. Stechow, A. von (1975). Transformational Semantics. In: Ε. L. Keenan (ed.) Formal Semantics of Natural Language. Cambridge: Cambridge University Press. 205-214. Radvila, R. & A. von Stechow (1975). Zur formalen Syntax und Semantik des Deutschen. Arbeitspapiere des SFB 99 der Universität Konstanz. Kratzer, Α. & A. von Stechow (1977). Äußerungssituation und Bedeutung. Zeitschrift für Linguistik und Literaturwissenschaft 23/24. 98-130. Stechow, A. von (1977). Semantische Präsuppositionen. Arbeitspapiere des SFB 99 der Universität Konstanz. Stechow, A. von (1978). Direktionale Präpositionen und Kontextheorie. In: Μ. E. Conte (ed.) Wortstellung und Bedeutung. Tübingen: Niemeyer. 157-166. Stechow, A. von (1979a). Occurrence-Interpretation and Context-Theory. In: D. Gambara, F. Lo Piparo & G. Ruggiero (eds.) Linguaggi e Formalizzazioni. Rome: Bulzoni. 307-348. Stechow, A. von (1979b). Visiting German Relatives. In: R. Bäuerle, U. Egli & A. von Stechow (eds.) Semantics from Different Points of View. Berlin: Springer. 226-265. Stechow, A. von (1979c). Deutsche Wortstellung und Montague-Grammatik. In: J. Μ. Meisel & Μ. D. Pam (eds.) Linear Order and Generative Theory. Amsterdam: Benjamins. 317^190. Bäuerle, R. & A. von Stechow (1980). Finite and Non-Finite Temporal Constructions in German. In: C. Rohrer (ed.) Time, Tense and Quantifiers. Tübingen: Niemeyer. 375—421. Stechow, Α. von (1980a). Modalverben in einer Montague-Grammatik. In: D. Clement (ed.) Empirische Rechtfertigung von Syntaxen. Bonn: Bouvier. 126-152. Stechow, A. von (1980b). Modification of Noun Phrases: A Challenge for Compositional Semantics. Theoretical Linguistics 7. 57-110. Stechow, A. von (1980c). Wie interessant ist die Syntaxforschung heute? Das Studium der Linguistik 8/9. 32-59. Stechow, A. von (1980d). Notes on Topic and Focus of Interrogatives and Indicatives. Arbeitspapiere des SFB 99 der Universität Konstanz. Stechow, A. von (1980e). Zwei Fragen an die Montague-Grammatik. In: H.-H. Lieb (ed.) Oberflächensyntax und Semantik. Tübingen: Niemeyer. 51-69. Stechow, A. von (1981a). Topic, Focus and Local Relevance. In: W. Klein & W. Levelt (eds.) Crossing the Boundaries in Linguistics. Dordrecht: Reidel. 95-130. Stechow, A. von (1981b). Presupposition and Context. In: U. Mönnich (ed.) Aspects of Philosophical Logic. Dordrecht: Reidel. 157-225. Klein, W. & A. von Stechow (1982). Intonation und Bedeutung von Fokus. Arbeitspapiere des SFB 99 der Universität Konstanz. Stechow, A. von (1982a). Three Local Deictics. In: R. J. Jarvella & W. Klein (eds.) Speech, Place and Action. Chichester: Wiley & Sons. 73-99.

536 Stechow, Α. von (1982b). Structured Propositions. Arbeitspapiere des SFB 99 der Universität Konstanz. Cresswell, M. J. & A. von Stechow (1983). De Re Belief Generalized. Linguistics and Philosophy 5. 503-535. Stechow, A. von (1983). Sind 'groß' und 'klein' Prädikate oder Relationen? - Ein Interview mit Aristoteles. In: M. Faust et al. (eds.) Allgemeine Sprachwissenschaft, Sprachtypologie und Textlinguistik - Festschrift für Peter Hartmann. Tübingen: Narr. 105-120. Stechow, A. von & Ε. T. Zimmermann (1984). Term Answers and Contextual Change. Linguistics 22. 224-241. Stechow, Α. von (1984a). Structured Propositions and Essential Indexicals. In: F. Landman & F. Feldman (eds.) Varieties in Formal Semantics. Dordrecht: Foris Publications. 385-404. Stechow, A. von (1984b). Comparing Semantic Theories of Comparison. Journal of Semantics 3. 1-77. Stechow, A. von (1984c). My Reaction to Cresswell's, Hellan's, Hoeksema's and Seuren's Comments. Journal of Semantics 3. 183-199. Ghisla, S. & A. von Stechow (1985). Stanno nascendo una grammatica e un vocabulario del dealetto di Mergoscia. In: Eco di Locarno. Locarno. Uhmann, S. & A. von Stechow (1986). Some Remarks on Focus and Projection. In: W. Abraham & S. de Meij (eds.) Topic, Focus and Configurationality, Amsterdam/Philadelphia: Benjamins. 295-320. Stechow, A. von (1986). Zwei Kritiken: Cresswells Strukturierte Bedeutungen und Vennemanns Anti-Grimm. Arbeitsberichte des SFB 99 der Universität Konstanz. Stechow, A. von (1987). Semantik im Sonderforschungsbereich 99 'Grammatik und sprachliche Prozesse'. Konstanzer Blätter für Hochschulfragen 92/93. 8-27. Stechow, A. von (1988a). Syntax und Semantik. Arbeitspapiere der Fachgruppe Sprachwissenschaft der Universität Konstanz. Stechow, A. von (1988b). Drei semantische Analysen. In: Μ. T. Schepping & A. von Stechow (eds.) Fortschritte in der Semantik - Abschlußbuch des Sonderforschungsbereichs 99 'Grammatik und sprachliche Prozesse'. Weinheim: VHC Verlagsgesellschaft. Stechow, A. von (1988c). Fortschritte in der Semantik. In: Μ. T. Schepping & A. von Stechow (eds.) Fortschritte in der Semantik - Abschlußbuch des Sonderforschungsbereichs 99 'Grammatik und sprachliche Prozesse'. Weinheim: VHC Verlagsgesellschaft. Fabricius-Hansen, C. & A. von Stechow (1989). Explikative und implikative Normalerweiterungen. Zeitschrift für Sprachwissenschaft 8. 173-205. Stechow, A. von (1989). Distinguo - Eine Antwort an Dieter Wunderlich. Linguistische Berichte 122.330-339. Stechow, A. von (1990a). Ringvorlesung Linguistik 1989/90: Linguistik, Morphologie und Syntax. Arbeitspapiere der Fachgruppe Sprachwissenschaft der Universität Konstanz.

List of Publications

537

Stechow, A. von (1990b). Status Government and Coherence in German. In: G. Grewendorf & W. Sternefeld (eds.) Scrambling and Barriers. Amsterdam: Benjamins. 143-198. Stechow, A. von & D. Wunderlich (1991). Semantik. Vorwort/Preface. In: A. von Stechow & D. Wunderlich (eds.) Semantik - Ein internationales Handbuch zeitgenössischer Forschung. Berlin/New York: de Gruyter. I-X. Stechow, A. von (1991a). Syntax und Semantik. In: A. von Stechow & D. Wunderlich (eds.) Semantik - Ein internationales Handbuch zeitgenössischer Forschung. Berlin/New York: de Gruyter. 90-148. Stechow, A. von (1991b). Current Issues in the Theory of Focus. In: A. von Stechow & D. Wunderlich (eds.) Semantik - Ein internationales Handbuch zeitgenössischer Forschung. Berlin/New York: de Gruyter. 804-824. Stechow, A. von (1991c). Focusing and Backrounding Operators. In: W. Abraham (ed.) Discourse Particles. Amsterdam/Philadelphia: Benjamins. 37-84. Stechow, A. von (1991d). Lexical Decomposition in Syntax. Arbeitspapier der Fachgruppe Sprachwissenschaft der Universität Konstanz. Stechow, A. von (1992a). Kompositionsprinzipien und grammatische Struktur. In: P. Suchsland (ed.) Biologische und soziale Grundlagen der Sprache. Tübingen: Niemeyer. 175-248. Stechow, A. von (1992b). Intensionale Semantik - Eingeführt anhand der Temporalität. Arbeitspapier der Fachgruppe Sprachwissenschaft der Universität Konstanz. Stechow, A. von (1993). Die Aufgaben der Syntax. In: J. Jacobs et al. (eds.) Syntax Ein internationales Handbuch zeitgenössischer Forschung. Vol 2. Berlin/New York: de Gruyter. 1-88. Stechow, A. von (1994). Modular Morphology and the Scope of 'Wieder' and 'Again'. SfS-Report der Universität Tübingen. Stechow, Α. von (1995a). Lexical Decomposition in Syntax. In: U. Egli et al. (eds.) Lexical Knowledge in the Organisation of Language, Amsterdam/Philadelphia: Benjamins. 81-177. Stechow, A. von (1995b). Tense in Intensional Contexts: Two Semantic Accounts of Abusch's Theory of Tense. In: F. Hamm & A. von Stechow (eds.) Blaubeuren Conference of Semantics. 379-433. Stechow, A. von (1995c). On the Proper Treatment of Tense. In: T. Galloway & M. Simons (eds.) Proceedings of SALT V. 362-386. Nohl, C. & A. von Stechow (1995d). Einführung in die Semantik. SfS-Report der Universität Tübingen. Stechow, Α. von (1996a). The Different Readings of Wieder 'Again': A Structural Account. Journal of Semantics 13. 87-138. Stechow, A. von (1996b). Against LF-Pied-Piping. Natural Language Semantics 4. 57-110. Stechow, A. von (1996c). Partial Wh-Movement and Logical Form. In: G. Müller, U. Lutz & A. von Stechow (eds.) Proceedings of the Tübingen Workshop on Partial Wh-Movement.

538 Stechow, Α. von (1997). Funktionale Köpfe des Satzes aus semantischer Perspektive. In: F.-J. d'Avis & U. Lutz (eds.) Zur Satzstruktur im Deutschen. Arbeitspapiere des Sonderforschungsbereiches 349, Nr. 90. 263-290. Stechow, A. von (1998). Introduction. In: G. Katz, S.-S. Kim & H. Winhart (eds.) Proceedings of the 1997 Tübingen Workshop in Reconstruction. Arbeitsberichte des Sonderforschungsbereichs 340, Tübingen. 1-8. Rapp, I. & A. von Stechow (1999). Fast "Almost" and the Visibility Parameter for DAdverbs. Theorie des Lexikons. Arbeiten des Sonderforschungsbereichs 282. Stechow, A. von (1999a). Eine erweiterte Extended-Now-Theorie fur Perfekt und Futur. Zeitschrift für Literaturwissenschaft und Linguistik 113. 86-118. Stechow, A. von (1999b). Das deutsche Perfekt. Zeitschrift für Literaturwissenschaft und Linguistik 29. Stechow, A. von (1999c). German Participles II in Distributed Morphology. To appear in the proceedings of the Bergamo Conference on Tense and Mood Selection. Stechow, A. von (1999d). Temporally Opaque Arguments in Verbs of Creation. In: B. Cecchetto (ed.) Festschrift per Andrea Bonomi. Milano. Gero, Ε. C. & A. von Stechow (2000). Tense in Time: The Greek Perfect (to appear). Rapp, I. & A. von Stechow (2000). Fast "Almost" and the Visibility Parameter for Functional Adverbs. Journal of Semantics 16. 149-204. Sauerland, U. & A. von Stechow (2000). The Syntax-Semantics Interface. In: B. Comrie (ed.) International Encyclopedia of Social and Behavioral Sciences (to appear). Stechow, A. von (2000a). Partial Wh-Movement, Scope Marking, and Transparent Logical Form. In: G. Müller, U. Lutz & A. von Stechow (eds.) Partial WhMovement. Amsterdam: Benjamins. 409—446. Stechow, A. von (2000b). Some Remarks on Choice Functions and LF-Movement. In: U. Egli & K. von Heusinger (eds.) Reference and Anaphoric Relations. Dordrecht: Kluwer. 193-228. Stechow, A. von (2000c). How Are Results Represented and Modified? Remarks on Jäger & Blutner's Anti-Decomposition. In: C. Fabricius-Hansen, E. Lang & C. Maienborn (eds.) Adverbs: Proceedings of the Oslo Conference on Adverbs, (to be published). Paslawska, A. & A. von Stechow (2001). Perfect over Perfective in Russian and Ukrainian, (submitted).

Edited Books Stechow, A. von (ed.) (1970). Beiträge zur generativen Grammatik - Referate des Fünften Linguistischen Kolloquiums. Braunschweig: Vieweg. Klein, W. & A. von Stechow (eds.) (1973). Functional Generative Grammar in Prague. Kronberg: Scriptor.

List of Publications

539

Bäuerle, R., U. Egli & A. von Stechow (eds.) (1979). Semantics from Different Points of View. Berlin: Springer. Bäuerle, R., C. Schwarze & A. von Stechow (eds.) (1983). Meaning, Use and Interpretation of Language. Berlin/New York: de Gruyter. Grewendorf, G. & A. von Stechow (eds.) (1985). Linguistische Berichte. Wiesbaden: Westdeutscher Verlag. Schepping, Μ. Τ. & Α. von Stechow (eds.) (1988). Fortschritte in der Semantik Abschlußbuch des Sonderforschungsbereichs 99 'Grammatik und sprachliche Prozesse'. Weinheim: VHC Verlagsgesellschaft. Stechow, A. von & D. Wunderlich (eds.) (1991). Semantik - Ein internationales Handbuch zeitgenössischer Forschung. Berlin/New York: de Gruyter. Jacobs, J., W. Sternefeld, T. Venneman & A. von Stechow (eds.) (1993). Syntax - Ein internationales Handbuch zeitgenössischer Forschung. Berlin/New York: de Gruyter. Egli, U., C. Schwarze, & A. von Stechow (eds.) (1994). The Lexicon in the Organisation of Language. Amsterdam: Benjamins. Hamm, F., J. Kolb & A. von Stechow (eds.) (1995). The Blaubeuren Papers Proceedings of the Workshop on Recent Developments in the Theory of Natural Language Semantics, 9-16.10.1994, SfS-Report, University of Tubingen. Lutz, U., G. Müller & A. von Stechow (eds.) (2000). Wh-Scope Marking, Linguistic Aktuell/Linguistics Today, vol 37. Amsterdam: Benjamins.

Monographs Stechow, A. von (1968). Zur Theorie der Finite-State-Sprachen mit einer Anwendung auf das Italienische. Unpublished PhD dissertation, Universität Münster. Kratzer, Α., Pause, Ε. & A. von Stechow (1973). Einführung in Theorie und Anwendung der generativen Syntax. Frankfurt: Athenäum. Sternefeld, W. & A. von Stechow (1988). Bausteine syntaktischen Wissens - Ein Lehrbuch der generativen Grammatik. Wiesbaden: Westdeutscher Verlag.

Reviews Stechow, A. von (1972). Rezension von: Heringer H. J., Formale Logik und Grammatik (Tübingen 1972, Niemeyer). Linguistik und Didaktik 3.311-318. Stechow, A. von (1982). Rezension von: Lyons, J., Semantics (München 1980/83, Beck). Beiträge zur Geschichte der deutschen Sprache und Literatur 2. 256-267. Stechow, A. von (1984). Book-Review: Gunnar Bech's Government & Binding Theory - GB's GB-Theory. Linguistics 22. 3-40.

540 Stechow, Α. von (1985). Rezension von: Cresswell, M. J., Structured Meanings - The Semantics of Propositional Attitudes (Cambridge, MA 1985, MIT Press). Journal of Semantics 4. 165-191. Stechow, A. von (1986). Notizen zu Vennemanns Anti-Grimm. Beiträge zur Geschichte der deutschen Sprache und Literatur 108:2. 159-171. Stechow, A. von (1988). Rezension von: H. van Riemsdijk & E. Williams, Introduction to the Theory of Grammar (Cambridge, MA 1988). Linguistics 26. 325-332. Stechow, A. von (1989). Categorial Grammar and Linguistic Theory. Rezension von: E. Bach, R. T. Oehrle & D. Wheeler (eds.) Categorial Grammar and Natural Language Structure (Dordrecht 1988, Reidel). Studies in Language 14. 433-479. Sternefeld, W. & A. von Stechow (1991). GB-Fallen: Eine Replik zu Tibor Kiss' Rezension der 'Bausteine syntaktischen Wissens'. Zeitschrift für Sprachwissenschaft 10. 134-144.

Translations Klein, W., A. Kratzer & A. von Stechow, Translation of: R. Wall (1973). Einführung in die Logik und Mathematik für Linguisten, Vol. 2. Kronberg: Scriptor. Stechow, A. von, Translation of: A. Naess (1975). En del elementaere logiske emner. Universitetsforlaget: Oslo-Bergen-Tromso. Erschienen in: Kommunikation und Argumentation. Kronberg: Scriptor. Stechow, A. von, Translation of: J. Lyons (1991). Theories of Meaning. In: Α. von Stechow & D. Wunderlich (eds.). Semantik - Ein internationales Handbuch zeitgenössischer Forschung. Berlin/New York: de Gruyter.

Unpublished Manuscripts Sternefeld, W. & A. von Stechow (1981). A Modular Approach to German Syntax. Konstanz. Stechow, A. von (1986). Alternative Semantics for Questions. Wellington. Stechow, A. von (1990). Morphologie, Syntax und Semantik: Ein Plädoyer für abstrakte Syntax. Tübingen. Stechow, A. von (1996). Tense, Aspect and Resultativity. Tübingen. Stechow, A. von (1998). Deriving Hitzeman 's Νon-Ambiguity. Tübingen.

List of Supervised Dissertations

Mönnich, Uwe (20.11.1975). Untersuchungen zum Problem der Grammatikäquivalenz. Radvila-Hottiger, Regula (21.7.1976). Pronominale Syntax. Schpak-Dolt, Nikolaus (23.6.1977). Zur Semantik von Tempus und Aspekt des Russischen im Rahmen einer λ-kontextfreien Sprache. Engesser, Kurt (8.12.1980). Untersuchungen zur Montaguegrammatik. Sternefeld, Wolfgang (3.9.1984). On Case and Binding Theory: A Comparative Study in the Syntax of German and English. Oh, Ye-Ok (13.2.1985). Wortsyntax und Semantik der Nominalisierungen im Gegenwartsdeutsch. Zimmermann, Thomas Ede (17.7.1987). Intensional Logic and Two-Sorted Type Theory. Fery, Caroline (24.7.1989). Prosodic and Tonal Structure of Standard German. Kim, Cheong-Min (19.7.1991). Bindung und Inkorporation: Syntax der Affigierung im Verbalsyntagma des Koreanischen. Lee, Hyun-Hee (23.2.1992). Zur Syntax des Koreanischen: Phrasenstruktur, Scrambling und Kasus. Müller, Gereon (5.11.1993). On Deriving Movement Type Asymmetries. Haas-Spohn, Ulrike (15.7.1994). Versteckte Indexikalität und subjektive Bedeutung. Choi, Myung-Won (7.7.1995). Kasustheorie und Transparenz im Koreanischen: Drei Fallstudien. Geenhoven, Veerle van (10.7.1996). Semantic Incorporation and Indefinite Descriptions: Semantic and Syntactic Aspects of Noun Incorporation in West Greenlandic. Biiring, Daniel (17.7.1996). The 59th Street Bridge Accent: On the Meaning of Topic and Focus. Beck, Sigrid (23.7.1996). Wh-Constructions and Transparent Logical Form. Kang, Jung-Goo (22.1.1997). Consecutio Temporum, Aspekt und transparente LF. Lüdeling, Anke (18.11.1998). On Particle Verbs and Similar Constructions in German. Meier, Cecile (8.2.2000). Konsekutive Konstruktionen und relative Modalität. Geuder, Wilhelm (26.7.2000). Oriented Adverbs: Issues in the Lexical Semantics of Events.