No title

JOURNAL OF SEMANTICS AN INTERNATIONAL JoURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MAN...

Author: Oxford University Press

5 downloads 385 Views 30MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

JOURNAL OF SEMANTICS AN

INTERNATIONAL JoURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MANAGING EDITOR: PETER BoscH (IBMScientific Centre, Heidelberg and University of Osnabriick) REVIE W EDITOR: TIBO R Kiss (IBMScientific Centre, Heidelberg) EDITORIAL BOARD:

N. AsHER (University of Texas, Austin) R. BARTSCH (University of Amsterdam) J. VAN BENTHEM (University of Amsterdam) M. BIERWISCH (MPG and Humboldt University Berlin) B. BoGURAEV (Apple Compucer Inc) M. BoRILLO (University of Toulouse) G. BROWN (University of Cambridge) 0. DAHL (University ofStockbolm) S. C. GARROD (University of Glasgow) B. GEURTS (University ofOsnabriick) M. HERWEG (IBMScientific Centre, Heidelberg) L. R. HoRN (Yale University) P. N. joHNSON-LAIRD (Princeton University) H. KAMP (University ofSrungarc) S. LEVINSON (MPI Nijmegen) S. WB NER (University of Diisseldorfj

SIR JoHN LYONS (University of Cambridge) A. MANASTER-RAMER (WayneState University) W. MARSLEN-WILSON (MRC, Cambridge) J. McCAWLEY (University of Chicago) M. MoENS (University of Edinburgh) F. J. PELLETIER (University of Alberta) M. PINKAL (University ofSaarbriicken) R. A. VANDER SANDT (University of Nijmegen) T.SANFORD (University of Glasgow) R. ScHA (University of Amsterdam) H.SCHNELLE (University of Bochum) A. V ONSTECHOW (University of Tiibingen) M.STEEDMAN (University of Pennsylvania) W. WAHLS TE R (DFKI,Saarbriicken) B. WEBBER (University ofpennsylvania) H. ZEEVAT (University of Amsterdam) T. E. Zimmermann (University ofSrungarc)

EDITORIAL ADDRESS : Journal ofSemantics, c/o Dr P. Bosch, IBM GermanySciennfic Centre, Vangerowstr. 18, D-6 9115 Heidelberg, Germany. Phone: (4
Congress Street, Salem, Mass

01970.

W1 P gHE,

UK by the Copynght L1censing 27

or in the USA by the Copyright Clearance Center,

The Journal of Semantics is publtshed quarterly tn February, April, August and November by Oxford Universiry Press. Subscription is

5133.

$119 per

year. Second class postage patd at Newark NJ and at additional mailing offices. ISSN

0167-

POSTMASTER: send address corrections to The Journal of Semantics, c/o Virgin Mailing and Distribution, Cargo Atlantic, 10 Camptown Road, Irvington NJ

07111-1105, USA.

OXFORD UNIVERSITY PRESS

¥J

NEW SERIES Oxford Studies in Lexicography and Lexicology Editors: Richard W. Bailey, Noel Osselton, and Gariele Stein This series provides a forum for the publication of substantial scholarly works on all issues of interest to lexicographers and lexicol ogists.

Tradition and Innovation in

Aspects of Semantic Opposition in English

Modern English

Arthur Mettinger

Dictionaries

Drawing on a corpus of data from contemporary novels and from Roget's Thesaurus, Dr Mellinger ana l yses the meaning relations that hold between antonyms (contrasting wo rd pairs) in English.

Dr Bejoint detects a conservatism in the fundamental design of dictionaries. He argues

0- I9-824269·7, 214 pp., Clarendon Press £25.00

�-.·

-i;io�:;:

· · ·- •

-"!!

. �Rls::;.f�':J�i��tts COf.1pu_i!Jl� � ..,_.__

�9MP· 1

,_

Henri Bejoint that this conservatism reflects the importance of dictionanes as cultural artefacts. 0-19-823919-X, 288 pp., Clarendo n Press £30.00

Research in Humanities Computing Edited by Susan Hockey and Nancy Ide 'for the humanities lay reader there are fascinating applications and ingenious use of basic computer techniques' Computer Books Review Each volume in this series prese n ts a selection of papers from the annual conference of the

�---· -·

....

Association fo r Computers and the Humanities and the Association for Literary and Lin guistic Computing. Research

in

Humanities Computing 3

Papers from the 1991 ACH-ALLC Conference

A Dictionary of English Collocations Based on the Brown Corpus Goran Kj ellmer Collocations (recurring and grammatically well-formed sequences of words) make up the building blocks of the speaker's mental lexicon. This book contai n s some 85,000 collocations from the American Brown Corpus, which contains one million words of rel atively recent texts from many different genres.

0-19-823903-3, Three volumes, 2,294 pp. £215.00

0-19-824064-3,304 pp., 1 0 81ine figures, Clarendon Press £37.50

Research in Humanities Computing 2 Papers from the 1990 ACH-ALLC Conference

0·19-824039-2, 272 pp., 761ine figures, 9 maps. Clarendon Press £35.00

•

Available from all good booksellers. To order the EASY way, please caD our

Credit-Card Hotline on (0536) 454534. For more infonnation please write to: A & R Marketing, OUP, Walton Street, Oxford OX2 6DP, UK.

JOURNAL OF SEMANTICS Volume

11

Numbers

1

&

2

CONTENTS

KEES vAN DEEMTER What's New? A Semantic Perspective on Sentence Accent LITA LUNDQUIST and RoBERT ]. JARVELLA Ups and Downs in Scalar Inferences

33

MARIA vALIOULI

Anaphora, Agreement, and Right Dislocations in Greek

55

PATRICK BLACKBURN Tense, Temporal Reference and Tense Logic

Review Articles FRANK BECKMAN Jane Grimshaw, Argument Structure

103

GABRIEL FALKENBERG Mark Richard, Propositions, Attitudes and Russellian Annotations

133

LANGUAGE

and COGNITIVE PROCF.SSFS Managing Editors: Lon-aine K. Tyler, Gen-y T.M. Altmann,

Kathryn Bock, Fran�ois Grosjean, William Marslen-Wilson

LangUilge and Cognitive Processes provides an international forum for the publication of theoretical and experimental research into the mental processes and representations involved in language

use.

Research relevant

to the psychological theory of language stems from a wide variety of disciplines and the content of LangUilge

and Cognitive Processes reflects

this interdisciplinary perspective. Apart from research in experimental and developmental psychology,

LangUilge

and Cognitive Processes

publishes

work derived from linguistics, philosophy, computer science, and AI.

RECENT AND FORTHCOMING PAPERS

Simultaneous Interpretation and the Recall of Source-language Sentences, William P. Isham and Harlan Lane Children's Application of Binding During Sentence

Processing, Cecil McKee, Janet Nicol and Dana McDaniel Phonological Activation of Semantic Competitors During Lexical Access in Speech Production, Trevor A. Harley Improving Written Communication Through Perspective-taking, Matthew J. Traxler and Morton Ann Gernsbacher Probabalistic Constraints and Syntactic Ambiguity

Resolution, M. MacDonald

SPECIAL ISSUES

1992: Discourse Representation and Text Processing, Editor: Jane Oakhill & Alan Garnham

1994: Morphological Structure, Lexical Representation, and Lexical Access, Dominiek Sandra & M arcus Taft

ABSTRACTS

Language and Cognitive Process es is covered by the following: Biosis; Current Contents; Ergonomics Abstracts; MLA International Bibliography & Directory of Periodicals; PsychiNFO; Sociological Abstracts SUBSCRIPTION INFORMATION

Volume9 (4 issues),1994ISSN01690965 Individuals:£44.00 (EC),US$86.00 (USA), £49.50 (Rest o{ World) Institutions:£132 (EC),US$264.00 (USA), £137.00 (Rest of World)

SEND ORDERS AND ENQUIRIES TO: Lawrence Erlbaum Associates Ltd 2:1 Church Road, Hove, E. Sussex, BN3 2FA, UK, Fax 0273 205612

LINGUISTICS

an interdisciplinmy journal of the language sciences

Editor-in-Chief:

Wolfgang Klein. Max-Planck-Institut,

Nijmegen, The Netherlands

LINGUISTICS was first published in 1963 as an international review. As of 1979, Volume 17, the journal was reconstituted under a new Editorial Board. In its new form the journal provides an interdisciplinary forum for research in the language sciences. LINGUISTICS publishes articles and short notices in the traditional disciplines of linguistics: pragmatics, semantics, syntax, morphology, phonology. In addi tion, it will accept articles and notices from neighboring disciplines that are engaged in the study of natural language: experimental phonetics, psycho linguistics, and neurolinguistics, and first and second language acquisition. LINGUISTICS also publishes occasional special issues and welcomes pro posals for these. LINGUISTICS accepts work from other disciplines, insofar as it is deemed to be of interest to linguists and other students of natural language; for example, cer tain studies in logic, artificial intelligence, social interaction, physiology, and neurology would meet this requirement. LINGUISTICS is published in one volume of six issues per year (approx. 1200 pages). As of 1979 the numbering was changed; Volume 32 corresponds to Issues 329-334. Subscription rates for

Volume

Institutions /libraries Individuals* Single issues

32

(1994):

OM 592.00 I oS 4618.00 I sFr. 560.00 (postage extra) OM 298.00 (includes postage)

OM 105.00 I oS 819.00 I sFr. 102.00

Prices in US dollars for subscriptions in North America only: $ 390.00 (postage extra) Institutions /libraries $ 190.85 (includes postage) Individuals* $ 65.00 Single issues • Subscriptions at the individual rate are for personal use only and must be prepaid and ordered directly from the publisher. Prepayment may be made by check or by credit card: MasterCard [Access]. EuroCard, Visa and American Express [AMEX may not be used in North America). Orders placed for institutions will be invoiced at the institutional rate. The individual rate is not available in the FRG. Switzerland or Austria.

Institutional subscriptions and single or back issues can be ordered from your local bookseller or subscription agent, or directly from MOUTON DE GRUYTER (a division of Walter de Gruyter) at the following addresses:

For North America:

Walter de Gruyter, Inc. 200 Saw Mill River Road Hawthorne, NY 10532 USA

For all other countries: Walter de Gruyter & Co. Postfach 30 34 21 D-10728 Berlin Federal Republic of Germany

Scope of this Journal The JOURNAL OF SEMANTICS publishes articles, notes, discussions, and book reviews in the area of natural language semantics. It is explicitly interdisciplinary, in that it aims at an integration of philosophical, psychological, and linguistic semantics as well as semantic work done in artificial intelligence and anthropology. Contributions must be of good quality (to be judged by at least two referees) and should relate to questions of comprehension and interpretation of sentences or texts in natural language. The editors welcome not only papers that cross traditional discipline boundaries, but also more specialized contributions, provided they are accessible to and interesting for a wider readership. Empirical relevance and formal correcmess are paramount among the criteria of acceptance for publication.·

Information for Authors Papers for publication should be submitted in 3 copies to the managing editor. They should be typed on

A4 (or similar format), one-sided, double-spaced, and with a wide margin and must

be accompanied by an approx. 200 word summary. Notes and bibliographical references must appear at the end of the typescript. All bibliographical references in the text by author's surname and year of publication. Diagrams must be submitted camera-ready. All papers submitted are subject to anonymous refereeing and are considered for publication only on the assumption that the paper has neither as a whole or in part already been published elsewhere nor has elsewhere been submitted or accepted for publication. Unless special arrangements have been made, copyright rests with Oxford University Press. Authors receive 20 offprints of their published articles and 10 offprints of their published reviews, free of charge. Larger numbers can be supplied at cost price by advance arrangement.

Subscriptions The Journal of Semantics is published quatterly.

Institutional: UK and Europe £62.00; USA and Rest of World US$1 19.00. (Single issues: UK and Europe £I8.oo; USA and Rest of World US$35.00.)

Personal:* UK and Europe £31.00; USA and Rest of World US$59.00. (Single issue: UK and Europe £9.00; USA and Rest of World US$17.00.)

• Personal rates apply only when copies are sent to a private address and payment is made by personal cheque/credit card. Prices include postage by surface mail or, for subscribers in the USA and Canada by Airfreight or in Japan, Australia, New Zealand and India by Air Speeded Post. Airmail rates are available on request.

Order Information Payment is required with all orders and subscriptions are accepted and entered by the volume. Payment may be made by cheque or Eurocheque (made payable to Oxford Uni�rsity Press), National Girobank (account 500 1056), Credit cards (Access, Visa, American Express, Diners Club), or UNESCO coupons. Please send orders and requests for sample copies to theJournals Subscriptions Department, Oxford University Press, Walton Street, Oxford, OX2 6DP, UK. Telex 837330 OXPRES.

Advertising Advertisements are welcome and rates will be quoted on request. Enquiries should be addressed to Jane Parker, Advertising Sales Manager, Oxford University Press, Walton Street, Oxford OX2 6DP, UK. Tel: +44 (o)865 56767 Fax: +44 (o)865 267782.

journal ofSemantics 11: 1-31

© Oxford University Press 1994

What's New? A Semantic Perspective on Sentence Accent KEES VAN DEEMTER Institutefor Perception Research (IPO) Eindhoven ,

Abstract This paper proposes semantic definitions of the twin concepts of'given' and'new' information,

in which the notion of an anaphoric relation is generalized beyond the case in which the denotation of the anaphor equals that of its antecedent. The resulting theory of sentence accent explains aspects of human speech production as well as understanding, and could be applied in automatic speech synthesis as well as, in principle, in automatic speech understanding.

I INTRODUC T IO N: ACCEN T A N D NEW I NFO R M A T IO N Until recently, many researchers were reluctant to ascribe systematic functional impact to intonation. A late representation of this school of thought was Bolinger, who remarked about intonation that 'its role may actually be ... limited and diffuse, more in keeping with affect and attitude than with transmitting a certain kind of information' (Bolinger I989: 38o-I). Bolinger concluded that attempts to bring 'truth-functional logic' to bear on the subtleties of intonation are bound to fail. This position may once have been justifiable, but times have changed. Although intonation might also have the sort of emotional impact that resists formal treatment, massive evidence now shows d·.e 'informational', and even the truth-conditional, relevance of intonation. For instance, experimental studies have shown how sentence melody can disambiguate various sorts of syntactic ambiguities (Lieberman I967; Lehiste I97 3; Price et al. I99I).Another well-established research tradition, dating back to Halliday (I967), has shown 1 how sentence melody, and accent in particular, can have clear semantic impact.By now, a variety of such cases have been brought to bear, and have been described with considerable success in a truth-conditional framework. by such authors as Mats Rooth (Rooth I992) and Amim von Stechow (von Stechow I99I). The phenomena described range from contrast and scalar implicatures to the semantics of focusing adverbs, question-answer sequences and counterfactuals.2

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

and shows how these definitions can be used to fill a gap in current theories of sentence accent. The semantic definitions proposed are based on a variant of Discourse Representation Theory

2 A Semantic Perspective on Sentence Accent

(I) My neighbour is afunny character. John is really nice though , John can either be my neighbour or an entirely different person. This leads to a clear-cut, truth-conditional difference in the meaning of the sentence. Note that the two readings may be disambiguated by prosodic means. For example, if john is accented-in which case I will often write +john , while writing -John to indicate that John is not accented-he will tend to be interpreted as distinct from my neighbour. Knowledge about the relation between informational status and accent can explain aspects of human speech production as well as understanding, and is potentially useful in various areas of natural language processing. In principle that is, if and when sentence accent can be recognized automatically automatic natural language understanding could make use of accenting to disambiguate spoken messages. Conversely, speech synthesis programs already do make use of the correspondence between accent and newness to improve accent assignment algorithms. Yet improvements over less enlightened predecessors are not as dramatic as one might expect. Why is that? One reason why 'informational status' has not yet led to much improvement in natural language processing systems is that few theoretical proposals so far have offered precise definitions of the relevant notions. This is clearly true for the older proposals, such as those of the Prague school, and those in Halliday {I967), Chafe (1976), and Clark & Haviland (I977), which were not based on a detailed theory of discourse and did not aim for formal precision, but it is true for later proposals as well.4 In addition, natural language processing systems have often interpreted givenness and newness at the wrong 'level' of linguistic

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The present paper deals with one way in which accent can have semantic impact, namely by marking which part of an utterance conveys 'new' information, and which pan conveys information that is already 'given'. This particular function of accent, in which it marks 'informational status', was already noted by Halliday {Halliday I967) and others, and has been studied extensively by psycholinguists. It has been established that high pitch accents mark new information, while opinions differ somewhat on the impact of low pitch accents.3 I will follow 't Hart et al. {I99I) in assuming that accent is basically one linguistic phenomenon that has different phonetic realizations, all of which can be used to mark 'new' information. Researchers in this tradition have found, for example, that, for a hearer, accent does more to distinguish between given and new Noun Phrases than syntactic cues such as the distinction between active and passive mood {Most & Saltz I979), or even fronting (Nooteboom et al. 1981; Nooteboom & Kruyt 1987). Although it has not even been noted, differences in the informational status of a Noun Phrase can even correspond with differences in truth conditions. For instance, in the discourse

Kees

van Deemter

3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

information. In Bell Labs' Newspeak, for example, a content word is marked as 'given' provided it has the same root as one of its accessible predecessors. I will argue that the required notion of givenness is a semantic one of the sort that can be based on a version of Discourse Representation Theory. The idea of bringing a theory of discourse to bear on accenting algorithms is not completely new. For example, there are hints in Hirschberg ( I 990) of how Grosz & Sidner's work on focus spaces may be used to arrive at an appropriate definition of newness, and van Kuppevelt (1991) has worked out another proposal. However, to my knowledge, Grosz & Sidner's ideas have not yet been worked out in sufficient detail actually to predict accents in context, and the same holds for van Kuppevelt's proposal, which relies on complex inferences about the hidden structure of a discourse into a sequence of question-answer pairs, where the question may be explicit, but more often only implicit. In addition, these theories have little to say about a class of cases that we will discuss at some length-cases in which the expression relies for its interpretation on some other expression, without having the same denotation as this expression. 5 The present proposal seeks to eliminate these limitations. Accenting is a complex issue. Even if all accents are used to highlight parts of an utterance, an expression may be highlighted for various reasons, such as: contrast with some other expression, an unusual or otherwise remarkable interpretation, strong emotion, or the mere desire to liven up the intonation of public speech. It would be extremely difficult to deal with all these factors in one paper. The present paper deals only with accents that highlight new information-in a pretty strict sense of that expression. The question of how newness differs from the related notion of contrast will be briefly taken up in section 3.2. The more general question of how factors other than newness may change predictions about accent will be discussed in the introduction to section 3 and in the Conclusion. One factor that I have largely disregarded, even though it has been argued to affect accent as well as other phenomena to such an extent that it should be made a part of the grammar (Oehrle 1988), is the factor of lexical 'weight', which predicts that certain words are more likely to be accented than others, even if they are in the same syntactic position.6 Here are some additional disclaimers. I will concentrate on accents in Eng lish Noun Phrases that are in a non-predicative position. NPs in a predicative position (e.g. '. . . is NP') must be considered in combination with their copula, and the question of novelty shifts to the resulting VP. Generalizations of the current proposal to VPs, as well as to other XPs, seem to pose no particular problems, but would require some additional work. For instance, Ewan Klein's ideas on VP-ellipsis, in which DRTs Reference Markers are used to collect the information that is conveyed in a VP can be straightforwardly applied to predict 'newness' accent in VPs, along the lines of the present paper (Klein 1987). Such an account would also have to explain how a VP (e.g. appointed a

4

A Semantic Perspective on Sentence Accent

2 T W O C OMPLEMEN T A R Y THEOR IE S: D R T AND F OCU S - A CCENT THEORY The present section will first show why and in what sense givenness and newness of information are semantic affairs at all, and why certain theories of anaphora present themselves as natural tools in this area. Then follows a discussion of the two main building blocks for the theory that I envisage. The first building block, sketched in section 2.2, is a suitably extended version of Discourse Representation Theory (Kamp I98I), here somewhat immodestly called Generalized Discourse Representation Theory(GDRT)(see van Deemter I992, I99I). GDRT will be used to define what it means to constitute given or new information. The second building block is a version of Focus-Accent(F-A) theory as it has been put forward by Baart(I987) and adapted by Dirksen(I992). As I view it, this theory aims to predict which parts of a sentence are accented,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

judge) causes its nominalization (e.g. judicial appointment) to become 'given'. Generalization of the present proposal to other Germanic languages seems straightforward, but I have no idea about their validity outside that domain. Further, nothing will be said about the interesting issue of accent placement within the word, or even within Noun-Noun compounds,' and in con formance with current practice in DRT, little will be said about limitations of memory. Also, except where recency of information becomes of the essence, we will only ask whether certain information is available in principle from a given piece of discourse. Further, we will remain mostly silent about the AI-hard problem of inferentially given entities, which seems beyond the reach of present theories. My last disclaimer is that the proposals made in this paper have not yet been tested experimentally,8 nor were they scrupulously checked against corpora of recorded text. Some aspects of informational status, including definitions of the key notions 'given' and 'new', are still so incompletely understood that the formulation of empirical hypotheses is hampered considerably. Therefore, theory building, in which the basic apparatus for a theory of givenness and newness is developed and in which initial hypotheses are put forward in terms of that apparatus, is still a very relevant activity. The present paper seeks to make some modest theoretical headway, wit..1. the emphasis on questions of broad semantics/phonetics architecture, rather than on empirical details. The structure of the rest of this paper is as follows. Section 2 outlines the two building blocks for the proposed approach to intonational focusing. Section 3 explains how these can be put together to build a bridge between meaning and intonation, and conclude with an extended example (section 3·3)· Section 4 draws some conclusions.

Kees van Deemter 5

given that some other theory decides which of its major constiruents constirute 'given/new information' ( i.e. -!+ Focus). In section happens when GDRT fills the slot in F-A theory.

2.1

3,

we will discuss what

The relevance ofsemantics

Whether a Noun Phrase must be accented or not depends on its denotation,

rather than on its form, or on its full meaning. To illustrate, consider the following examples:

(2) The president arrived. Mr. - clinton was late. thepresident

and Mr.

Clinton

refer to Bill C linton, the

different meaning.9 C onversely, if the same designator is used with different

denotations, accent seems obligatory:

(3) Clinton visited many towns; when hefinally arrived in + Clinton, he was late. ( It might be thought that sentences such as primacy of semantics:

(4 )

and

(s)

are exceptions to the

(4) Mary knows that the president is a powerful man, but she is unaware that Mr. + clinton is powerful. (s) His name is +Clinton, not + Clynton.

It seems, however, that such cases require some sort of opposition between the two items ( the president/C linton, and C linton/C lynton), which makes them amenable to a contrastive analysis of the kind proposed by Mats Rooth.10) More

precisely, accenting does not depend on acrual denotation, but on believed denotation. What counts is whether the speaker knows that the same

individuals are involved in the 'given' and in the 'new' expression. For example,

if the speaker does not know that Mr. C linton is the president, then he or she is

()

bound to accent the proper name in 2 . For a slightly more interesting example,

consider

(6) The crowd was approaching the castle. The women were very excited. For the

NP the women

to get rid of its accent, it is not sufficient that the crowd

consists of women. The fact that there are only women in the crowd has to be known to the speaker.

The examples given so far not only show that there is a semantic relevance to accenting, but they also suggest the relevance of a particular kind of semantic theory, namely the theory of anaphora. Here are some commonalities between

anaphora and absence of accent:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In a setting in which both

proper name will not normally be accented, even though he is designated by an expression that is physically different from the first occurrence and has a

6

A Semantic Perspective on Sentence Accent

Capitalizing on these commonalities, we will explore how a theory of anaphora, Kamp's Discourse Representation Theory (DRT), can be put to use in a theory of sentence accent. DRT was chosen because it is the most detailed representative of the family of so-called context-change theories of anaphora, which take the use of given (i.e. contextually provided) pieces of semantic information as their basic mechanism.12 �.2

(Generalized' Discourse Representation Theory

Since Discourse Representation Theory (DRT) is now one of the most influential semantics theories, I will not attempt to describe it in detail but refer to Kamp (1981) instead. DRT capitalizes on a distinction between two complementary classes of Noun Phrases: those that introduce an individual denoting reference marker (RM) into a discourse, and the anaphoric ones that can be identified with an already introduced RM. The first may be described as the 'new' NPs, and the latter as the 'given' ones. Kamp's original work was limited to singular NPs, but later work has extended DRT to discourse with plural NPs. As in the case of singulars, non-anaphoric plural NPs introduce their own RMs, denoting sets (or comparable algebraic structures) of indi viduals (van Eijck 1983; Roberts 1987). Anaphoric reference to plurals leads to additional conditions on already established RMs. The exact way in which a non-anaphoric NP introduces its RM, along with conditions that constrain its interpretation, depends on whether the NP is used collectively or distributively:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

1. Pronominal anaphora and absence of accent are legitimized if certain information, either in the linguistic or the extralinguistic context, has lent to the relevant item (the anaphor, or some constituent of which the accented item is a part) the status of'given information'. 2. Given information can be repeated, but 'reduced' anaphoric reference is usually more appropriate. Similarly, given information can be accented, but if it is, it tends to be processed more slowly by a hearer (Terken & Nooteboom 1987). 3· Both anaphoric reference and the absence of accent presuppose that the relevant information is linguistically accessible. In particular, accessibility tends to be blocked by logical operators.11 Similar observations hold for conditionals and for modal operators. 4· The distinction between given and new information is made on the level of known or believed referents, rather than, for instance, on a lexical level. That this applies to anaphora can be seen from examples in which a gender neutral NP becomes the antecedent for a non-gender neutral pronoun. For example, the anaphoric pronoun she can take the NP the student as its antecedent, provided the student is assumed to be female.

Kees van Deemter 7

Nonanaphoric NP-introduction: When a sentence y of the form NP VP is processed, a novel set RM, say X, is introduced in the principal DR along with a condition NP(X).LfNP is interpreted collectively, the condition y [NP :- X] is entered in the principal DR Lf NP is interpreted disrributively, a condition of the form b 1 � b 2 is entered, where b 1 consists of a newly introduced individual RM y and the condition y e X. The box b 2 consists of the condition y [NP:- y].

(7)

The men were angry

X

The

men (X)

�- -y e-le_m_:_nt-X� •�--M--gry--(y-)�

� 1 Figure 1

looks as in Figure r if tense is disregarded. Note chat y ranges over individuals, whereas X ranges over sets. The condition 'The men (X)' can be glossed 'X contains all the men in the universe of discourse, and nothing else'. To express the meaning of conditions of the form NP(X) in general, one can use embedding functions in the style of Kamp (I98 I), which map variables to suitable entities in the domain of quantification (c£ van Eijck I98 3 for motiva tion): NP-condition: An embedding functionf verifies DET CN (X) with respect to a modelM iffiiDET (CN)I!M (f(X)) & f{X) � IIC NJIM,

complemented as usual by the following definition of truth with respect to a model: Truth: A DRS is true with respect to a model M iff there is an embedding functionJ from the RMs of the DRS into the domain of M such thar, for all conditions¢ in the DRS,J verifies¢ with respect toM.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The condition NP(X) expresses the properties that X must have due to the meaning of the NP that introduced it (see below). The notation ¢ [cp := X] abbreviates the result of substitution of x for cp in¢. The condition b r � b 2 is the usual DRT-type implication which ensures that the variable y that is introduced in b I appears universally quantified. To exemplify. a DRS for the sentence

8 A Semantic Perspective on Sentence Accent

As a result, the variable X in our earlier example appears existentially quantified, and the DRS ends up saying 3X(The men(X) & 'Vy(y EX .... angry (y ))),

(8) Whenever a school class goes on a trip , one or two children will hate it . Subsectional anaphora turns out to be subject to largely the same constraints of syntactic and semantic structure as traditional anaphors.For instance, negation blocks both pronominal and subsectional anaphora (9, 10), but this does not affect NPs that outscope the negation, such as 'the team' in (10), which can be the antecedent for 'three players'. (9) There weren't any separate bedrooms in the house. *They/*Two very spacious

rooms were quite light, (10) The team did not play last Wednesday. Three players were injured.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

that is: all the men were angry. So much for the introduction of non-anaphoric NPs.For their anaphoric counterparts, the rough idea is that they are used to express new conditions on already established RMs,just as in the case ofKamp's singular RMs.Of course this rough idea must be refined in several ways to become fully accurate. In particular, accommodations have to be made for determiners expressing quantifiers (such as 'few', and 'less than two') that are not monotonically increasing. These quantifiers cannot directly be viewed as making a statement about a set of individuals and have to be're-analysed' along the lines of van Eijck ( 198 3). In addition, it has to be guaranteed that anaphoric reference to non-referring NPs in an earlier sentence-so-called e-rype anaphors-must refer to the maximal set for which this earlier sentence is trueY Such complications will, for now, be put aside, since their treatment is well known and less than crucial in the present setting. The present paper makes use of a proposal made in van Deemter (1992), where the notion of anaphoric reference was extended to include cases in which the 'anaphor' stands in relation to its antecedent other than that of simple identity. This notion of'anaphora' (henceforth the quotation marks are Oinitted) was applied to Noun Phrases of all kinds, including indefinites, quantifying NPs, and Proper Names.In particular, Generalized DRT accounts for cases of subsectional anaphora, in which the anaphor denotes a real part of its antecedent. The reason is that so many of the phenomena displayed by traditional anaphors can also be found in situations in which the denotation of an NP is context-dependent whilst, nevertheless, its denotation is not equal to that of some other NP in the same text.For example, sentence (8) shows a generalized case of donkey anaphora -a crucial motivating phenomenon for DRT-if the sentence is interpreted as saying something about one or two children in whichever school class goes on a trip.

Kees

van

Deemrer 9

Motivated by these and other similarities, a DRT-based treatment of subsec tional anaphors was outlined in van Deemter ( I992 . This treatment takes its departure from the proposals in Westerstahl

)

( I985),

in which the CN in the

familiar quantificational schema Q ( CN, B), where Q is the quantifier and CN and B are its relata, is intersected with some contextually given set. In order to

cope with syntactically simple NPs ( such as

everyone, something , but also proper

names and pronouns), this proposal was, in vanDeemter ( I992), extended to be directly applicable to entire NPs, as well as C Ns.14 In the sequel, I will sketch how

the contextualization of NPs is carried out. Firstly, all NPs introduce reference markers. The earlier rule of Nonanaphoric NP-introduction is restricted in its application to those NPs that are used non-anaphorically. If a Noun Phrase NP1

condition NP1 (X) to the antecedent RM Y. This is written ( NP1)Y (X). Informally, this relativized condition means that the 'ordinary' condition NP1 (X)

would hold if only individuals fromY were taken into account, and thatX�Y. More precisely:

Anaphoric NP-introduction : When a sentence y of the form NP VP is processed

which is anaphoric to an earlier RMY, a novel set-RM, sayX, is introduced in

the principal DR along with a condition NPY (X). If NP is interpreted collectively, the condition y(NP := X] is entered in the principal DR If NP is interpreted distributively, a condition of the form b I � b2 is entered, where b I consists of a newly introduced individual RM y and the condition y eX. The box b2 consists of the condition y (NP := y ].

Relativized NP condition : An embedding functionJ verifies (DET C N)Y (X) with respect to a model M iff 1 f (X) � II C NIIM&j (X) �j (Y) &II (DET C N)IIMrf(YJ (X). 5

M r A is straightforward restriction of the model M to the domain A : Let M- (D , I), where D is the domain and I the interpretation function. Then M r A = (DnA , I r A ), where I r A intersects all interpretations of constants with A : I r A (R" ) = I(R" ) n A X A X ... X A . (n times A ). Note that the two rules of NP introduction are very similar. In particular, an NP introduces an RM of its own, whether it is anaphoric or not. To illustrate the rules, consider the situation after the first sentence of (10), and assume that as a result a setY of players on the team has been introduced. Then when the subsequent sentence

( I I ) Three defenders were injured. is processed, the representation shown in Figure 2 results. Assuming that 'three' is used in the sense of 'at least three', the relativized condition defenders)Y (X)

(3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

with RMX has subsectional anaphora to another Noun Phrase NP2 with RMY, then Nonanaphoric NP-introduction for NP1 is modified by relativizing the

1 0 A Semantic Perspective on Sentence Accent

. Y

players-on-team (Y)

. X

(3 defenders) (X)

y

1- u u element X 1-l --l)lo�l

injured (u)

Figure 2

.------�

X

E

Defenders

Figure3

The commonalities notwithstanding, our extended notion of anaphora is subject to a few constraints that do not affect pronominal anaphora, or not so ostensibly. One constraint that is especially relevant here requires that if, on a given interpretation of an NP, the descriptive material in the NP makes requirements on denotations that cannot be fulfilled, then this reading is to be discarded. One might erroneously say of a female student, ( I2) The student succeeded where proftssionals hadfoiled. The guy is a real wizard , but only if the speaker believes in her own mistake. This means that the following consistency constraint must hold for DRS's {W denotes the set of those possible worlds that are compatible with the knowledge of the speaker):

Resolution constraint : Suppose K[m] is a DRS, and K[m + I] is the DRS resulting from K[m] after processing NP, and suppose that NP introduces the RM a; then the NP cannot take RM Y as its antecedent, leading to the condition NPY(a), if -.3X: NPY(X) holds in all wE Win which K[m] is true. { Compare van Deemter 1992.)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

means that X contains only elements that are also elements of Y and all of which are defenders, while at least three of X's elements are defenders. Should (II) have read All defenders were injured, then the relativized condition would have been (all defenders)Y(X), meaning that X contains only elements ofY and only defenders, while X contains all the defenders that can be found inY {see Figure 3).

Kees van Deemter

11

This constraint constitutes a powerful limitation on Noun Phrase interpretation because it exploits all the information that is available in the NP. For example, a reading of (I 3) in whichfivegirls is anaphoric to two children is ruled out by the fact that a set of five elements does not fit into a superset of two elements.

(I 3) Two children were up late. * Fivegirls had a party. As

Identity Anaphora : Let NP; correspond with RM X and NPj with RM Y in a

certain DRS. Then NP; has identity anaphora to NPj iff ( I ) the Anaphoric NP introduction rule has introduced the clause NPr{X) into the DRS; and (2) the conditions of the entire DRS, together with the speaker's background knowledge, imply logically that X� Y.

This definition allows identity anaphora to depend on whether the speaker is able to infer the identity of the two RMs. Actual equality is not enough: if the two RMs are equal, but the speaker cannot infer this fact, then the two count as different. Note that the question of whether a particular case of anaphora constitutes identity anaphora may be settled in various ways. Suppose NPi The C� and NP; The CN;. Then the speaker's knowledge to the effect that all the CNj are CN;, and hence X� Y, may derive from different sources. Sometimes, this may hold due to word meaning. For instance, if NP; The people and NPj The women , then the condition NPr{X) implies X Y, given that, by definition, all women are people. But the inclusion may also be a matter of world knowledge. For example, the speaker may have known beforehand, or be informed during the discourse, that all the CNj in the model happen to be CN;. or he/she may even know the exact denotations of CN; and CNi in the model. It is a consequence of the definition of Identity Anaphora that most occurrences of pronouns are identity anaphors if it is assumed that pronouns introduce RMs like any other NP does, along with a condition that is specific for the pronoun in question (van Deemter I992). For example, the pronoun she may be associated with a condition She(X) on its RM X saying that X is a singleton containing a female element. This would cause the relativized �

=

=

�

=

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

a consequence of the Resolution Rule, the use of an NP in which it is anaphoric to Ypresupposes that there is an X such that NPY(X).16 This holds for all NPs, no matter whether they are definite, indefinite, or quantifying. For example, if an NP two girls is anaphoric to an antecedent the children, then the fact that there are at least two girls among the children is presupposed. As always, some Lewis-style accommodation mechanism must be in place to account for presupposition failure. It is important to note that the generalized notion of anaphora subsumes the old one, which becomes a natural subclass of anaphors, namely those that have the relation of Identity Anaphora to their antecedents:

12

A Semantic Perspective on Sentence Accent

2.3

Focus-Accent theory

The main virtue of F-A theory, in the tradition of Liherman & Prince (1977) and Ladd (198o), is that it brings out the role of syntax in the determination of sentence accent. Syntax mediates between such informational affairs as given and new information on the one hand and linguistic form on the other. (See e.g. Nooteboom & Kruyt 1987 for an excellent explanation.) One of the most detailed versions of F-A theory that I have been able to find was put forward by Arthur Dirksen and Hugo Quene {Dirksen I 992; Dirksen & Quene 1991), in which some earlier ideas from Baarr (1987) are clarified and adapted to a computational setting. It is this version of the theory that will be presented here, in a simplified form that is subsequently adapted to suit present purposes. Dirksen uses a version of Phrase Structure grammar to build up metrical trees , which are like the trees that result from ordinary Phrase Structure rules, and in which each node has between o and 2 branches, each of which is marked as either Strong or Weak. At least initially, if a given node has one daughter, then it will he marked Strong, and if it has two daughters, then one of them is Strong and the other_Weak. In English, it will mostly be the right daughter that is marked Strong, but in languages with different word order, such as German and Dutch, the pattern varies considerably. It is assumed that

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

condition SheY(X) to mean that X is a singleton consisting of the only female element in X n Y. If Y itself is a singleton, this makes SheY(X) equivalent to X= Y, resulting in identity anaphora. A more conservative definition of She{X) would also require that the universe contains only one element. This reading would make identity anaphora obligatory for pronouns, ruling out subsectional uses, as in The married couple came out cifthe City Hall. She waspretty . Before moving on to matters of sentence accent, let me mention one more extension of the above-outlined framework that will have intonational relevance. This extension concerns relational NPs, as when an NP with a relational Noun expresses a relation (i.e. the relation that is contained in the Noun) with a contextually available antecedent. The treatment in van Deemter (1992) makes such NPs 'anaphoric' to the contextually expressed 'antecedent', thus extending the notion of anaphora even further than suhsectional anaphora. For example, if the children has introduced the antecedent RM Y for the relational NP the mother, then the mother introduces an RM X along with the relativized condition (The mother)Y(X), which expresses that X contains a unique individual that is the mother of all the elements in Y. We will later see how this analysis leads to the right predictions with respect to sentence accent. So much for the theory of discourse that we will make use o£ In the following section, the other building block for our theory of accent, Focus Accent theory {F-A theory), will be sketched out.

Kees van Deemter 13 some nodes are somehow marked for Fco us, using+ F if the node is known to be'interpreted as "new" or otherwise important addition to the discourse'. In

Dirksen's(1992) program, this is simply done by marking all major phrases(NP, VP, AdjP} as+ F. Accent is located by a recursiveCorrespondenceRule:

Correspondence Rule: For each nodeX,X

is accented if

(a) X is marked as+ F, or (b)X isStrong, and the node immediately dominatingX is accented. Strong nodesinherit accent from their parent node, while weak nodes do not have accent, unless they are marked+ F. For instance, consider the sentence

The version of F-A theory that was described and implemented by Dirksen predicts that there are two accents: theNP accent will, of course, land onMary; theVP accent lands onbook , since thisNoun i s the right sibling o fa (seeFigure

4)_17

I\s, +F Mary I\ s, +F w reads 1 \ w

w a

s

book

Figure 4

To arrive at the correct predictions in other cases, metrical trees are subject

to various transformations before the Correspondence Rule becomes active.

The most important of these is the Default Accent Rule, which determines

where an accent will land if its natural landing siteis somehow incapacitated or

'deaccented'.

InDirksen(1992 ), an expression isdeaccented if it is listed in the lexicon as one that'cannot be accented' (my gloss), which is true for words such asthe , a , and Default accent rule:

w

A

s

B

s

A

w B

b:

I\ -..I\s s w

B

Condition: B is deaccented. Figure 5

A

w

B

A

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(14) Mary reads a book ,

1

4 A Semantic Perspective on Sentence Accent

it,18

but we will use the te rm for any mar king on an e xpre ssion that cause s

accen t to be blocked . In add ition, it will be conve nie nt to use deaccented to de signate t he e ffe ct on acce nti ng that is caused by de acce nting through the

De fau lt Acce nt Rule , name ly that acce nt shifts from the de acce nted node to some where e lse . InDirk se n( 1992), any node that isde acce nted be come s - F.

Thus, inMary reads it, the fact that' it' isde acce nted cause sit to be marked - F and cause s the w and s labe ls on reads and the N oun P hr ase to be re ver sed , e l avin greads asStr ong and it asWe ak(Figure 6). The VP, be ing a major phr ase , is marked as + F, so the Corre sponde nce Rule w ill cause and it not to be acce nted .

reads

to be acce nted

reads

w,

it

Figure 6

I t has bee n sugge sted that the me chanism that take s care of e l xical de acce nti ng is also at work if a node isde acce nted for conte xtual re asons(B aart

1 987).

In se ction 3.2, we will e s e how thi s ide a can be wor ked out. Be fore

offeri ng a simple fl owd iagr am to vi sualize the orde r in which the rule s ofF-A the or y could apply , I propose tw o slight mod ifi cations inDirkse n' s for malism,

for the be nefi t of late r add iti ons. O ur goal here is to e l t+I- Fref le ct ne wI give n

status, cle aring it from contaminati on by e l xical (' cannot be acce nted' ) inform ation. The fi rst mod ifi cati on is to introd uce a se parate labe l, -L , for

le xical ite ms that are marked 'can not be acce nted'. As a re sult, it, in oure xam ple , wi ll be marked both+ F and L . N owin orde r to pre ve nt acce nt inMary -

reads it

from land ing both on read (d ue to the De faultAcce ntRule ) and on it

(d ue to the + F labe l), the Corre sponde nce Rule is mod ified as follows:

Modified Correspondence Rule : Fore ach nodeX,X

is acce nted if

(a) X is marked as+ F, butX isnot marked as- L , or (b)X isStrong, and the node immed iate ly d ominati ngX is acce nted .

As ar esult, -L will not tr igge r the De faultAcce ntRule , but also block acce nt l aving +1-F status unaffe cted. The on the word that is marked as -L , bur e follow ing simple fl ow char t refle cts this ne wd ivision of labour :

1 Con struct me tr ical tree 2 Assign+1-F(ocus) mar king, Assign+1-L(e xicalAcce ntabilir y ) marking

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

/ \s, +F Mary I \ s +F w

Kees van Deemter 1 s 3 Apply Default Accent Rule 4 Apply Modified Correspondence Rule In the following section, the connection between F-A theory and GDRT will be pointed out.

3 A D I S C O U RSE REPRE S E N T AT I O N T H E O RY F O R THE PRED I C T I O N AND I N TE R P RE T A T I O N OF A C C E N T

r.

An NP is marked + F if it introduces a new entity into the discourse, and it is

marked

2.

-F otherwise.

Deaccenting applies not only to certain lexically marked function words, but also to expressions that are somehow 'given'.

3· Baart's Default Accent Rule reverses the Strong!Weak labelling of sister nodes if the sister that was originally marked Strong happens to be accented due to (2). Rule r proposes a semantic explication ofBaart's ideas about focus, and suggests an implementation of focus that would allow us to do away with the + F

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

One reason why i t i s hard to come up with a plausible theory o fsentence accent is the considerable variability in accenting that has been shown in empirical studies, both between different speakers and with the same speaker. From the hearer's side, it has been observed that adding an accent to an utterance that is intonationally acceptable will still often lead to an utterance that is completely acceptable under the same circumstances (see e.g. Terken 1985). On the other hand, beyond a certain point, more accents cause processing difficulties. In particular, if given information is accented, processing by the hearer tends to slow down (Terken & Nooteboom 1987). This experimental fact is mirrored by the often-heard opinion that current accenting programs overgenerate accents. (This point is argued e.g. in Hirschberg 1990.) Given this situation, it seems prudent to start out with a minimalistic approach, describing an accenting pattern that is minimal given a certain context and a certain meaning.19 Later, in section 4, it will be shown how a more sophisticated account, indicating which additional accents may be added, can be superimposed on this minimal account. The current section outlines a theory of sentence accent on Noun Phrases, by showing how the two theories of the previous section can complement each other. In particular, it is shown how DRT can make F-A theory more precise in two respects: the definition of focus (+1-F), and the conditions under which deaccenting obtains. The version ofF-A theory that underlies Dirksen's work is modified by adopting versions of the following rules:

16 A Semantic Perspective on Sentence Accent marking on all major phrases. Rule

2

will extend Dirksen's notion of

deaccenting by proposing two semantic triggers for deaccenting, both of which once again have to do with givenness in a discourse. Rule 3 is a simple extension of the Default Accent Rule that takes this extended notion of deaccenting into account. The notions of givenness and newness that are used in these rules will be made more exact by means of the discourse theory that was outlined in section 2.2. Sections 3.I and 3-2 together will deal with point ( I ), and points (2) and (3) will be taken up in section 3.2.

3· I Plus or minusfocus} as modelled by DRT

(I s) The crowd was approaching the castle. The + women were very excited . In other words, a new conglomerate of elements counts as a 'new' discourse entity, even if all the elements are taken from an entity that constitutes given information. For example, if in the above sentence it is assumed that the crowd consists of women only, then it is better not to accent women , since the women has identity anaphora to the crowd . Moreover, what counts is not simply whether two NPs stand in a certain anaphoric relation, but whether they can be construed in such a way that they do so. For example, two subsequent occurrences of one and the same Proper Name, both of which denote the same individual, can, but need not be construed as standing in a relation of Identity Anaphora.20 A definition, therefore, has to say the following:

I ' Givenness. An occurrence NPi is 'given' due to NPi iffNPi can be construed

as standing in a relation of Identity Anaphora to NPj. An NP occurrence is given, or - F, iff it is given due to some other NP occurrence. An NP occurrence is 'new' if it is not given.

Given the GDRT framework of section 2.2, NPi can be construed as standing in a relation of ldentity Anaphora to NPi iff (a) NPi is accessible to NPi, (b) the anaphoric relation between NPi and NPj fulfils the Resolution Constraint, and (c) the respective RMs for NPi and NPj can be inferred to be equal. Before moving on to discuss some examples, let us look at some special classes of NPs. For example, the definition of Identity Anaphora causes

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Under what circumstances must the entity that was introduced by a Noun Phrase be regarded as 'new', and when as 'given'? One natural definition of givenness would define an occurrence NPi as 'given' if it is anaphoric to some NPi However, anaphoricity is not enough, since anaphora has now been stretched beyond Identity Anaphora. For example, as our earlier example (6) shows (repeated here as ( I s) ): if a certain crowd is known to consist of men and women, then accent on women is obligatory, even though the NP is anaphorically related to the crowd :

Kees van Deemter 1 7

principle I' to predict that pronouns will appear unaccented, unless they have subsectional anaphora-or, of course, unless some other factor such as contrastiveness or non-normal interpretation is involved (c£ section I}. And this prediction is borne out by the facts, for the identity anaphoric occurrence of she in (I6} is not normally accented, but if a subsectionally anaphoric occurrence of she in (I7} is possible at all, it certainly has to be accented: (I6} Mary walks. She talks . (I7} The married couple came out ofthe city hall. + She waspretty.

(I8} This time around, the Dutch soccer team won the match. Van +Basten scored the winning goal. Only in a case where the proper name refers to an entity that was itself already established (either by a proper name or by some other means) will our theory predict that no accent is required: (I9} Have you metJohn? -John is a goodfriend ofmine. (20) Have you met my neighbour? -John is a goodfriend ofmine .21 Finally, the theory predicts that relational anaphora will normally require accent, since it is normally used to introduce a new individual, or set of individuals, in the discourse. For example, in (2I) The children were hungry. The + mother was sick .

the mother is distinct from the children, and therefore has to be accented. Again, the prediction seems to be correct. Now that a notion of givenness has been defined for Noun Phrases, let us give some more detailed examples, in order to show how word meaning and sentence accent combine to constrain Noun Phrase interpretation. After these examples, we will turn to the topic of deaccenting. Consider

(22) The children were up late. Thegirls couldn't sleep . What are the possible readings of the second sentence? For simplicity, assume that word meaning equals word denotation. A Noun denotes a set of individuals. A Verb denotes a predicate P over sets S of individuals, with the understanding that, in the case of a distributive interpretation of the proposition P(S), a suitable first order version P' of P will be true of all the individual elements of S. Assume that the speaker has complete knowledge

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Similarly appealing predictions are obtained for proper names. When a proper name introduces a new individual into the discourse, it will require accent. This includes the case in which it 'zooms in' on one member of an already established set

18 A Semantic Perspective on Sentence Accent about the denotations of all the relevant Nouns, and that D the universe of discourse. Then:

I.

� {h I , . . ., h 10} is

Assume h I , h 2, and h 3 are the only children inD, so II Children II � { h I , h 2 , h 3}. Further, assume that some girls are too old to count as children. More specifically, in this particular domain, II Girls!! = {h I, h 2 , h 3. h 4. h s}. so all the children in D are girls. If no information about accenting is available, the sentence has a non-anaphoric interpretation, in which the Noun Phrase of the second sentence refers to all the girls in D . That is:

3X <;;;; D : The girls (X) & Vy(y E X - not-sleep (y)).

3X <;;;; D : (The girls)A (X) & Vy (y e X - not-sleep (y)).

3·

The clause (The girls)A(X) means that X contains all the children in A , and nothing more than that. But since all elements of A are children, X j ust equals A , so this is a case of identity anaphora. The girls represents given information, and no accent is required. The same assumptions as before, except that II girls II = {h 1, h 3, h 5}. Of course, a non-anaphoric reading of The girls couldn't sleep remains possible. An anaphoric reading, with The children as antecedent, will be obtained using the same formula as in (2 ). This time, however, the formula results in a subsectional anaphor, since the set of girls that are also children is a real subset of the set l!childrenl!. Therefore, Thegirls constitutes new information and must be accented, exactly as in (I)

3.2

Deaccenting ofotherwise-accented material

The proposals of the previous two sections determine whether a Noun Phrase must be marked as + F. One might think that this ought to be the end of this paper, since F-A theory can do whatever else there is to do, positioning all the accents in their appropriate places within major phrases that are marked + F. However, this is not true. For instance, if a + F NP contains a F NP, then F NP, and this is something accent cannot land anywhere within the -

-

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

2.

In accordance with section 2.2, the predicate The girls(X) means that X contains all the girls in D and girls only. Consequently, the utterance means that all the girls in D could not sleep. The Noun Phrase Thegirls constitutes new information, and consequently it must appear accented. Since girls is the right sibling of The (c£ section 2.3), the accent will land on girls . Given the same assumptions, there is also an anaphoric reading of the sentence, in which The girls is interpreted 'against the background' of the antecedent The children . Let A denote the set of children . Following the rules outlined above, the following interpretation results:

Kees van Deemrer

19

current implementations of F-A theory do not take into account. For example, suppose in the following example the town of Leyden is the topic of the discourse, so the town constitutes given informatioiL Then accent shifts from it to

atmosphere:

(23) He loved the + atmosphere ojthe -town . The Noun town is deaccented, losing its accent due to the fact that it is part of a

-F NP. To distinguish this case from a different kind of givenness to be discussed presently, we will call a F NP object-given . It will be shown in a -

-

(24) Thepeople in the hospital were in poor condition. The+ofd -people looked terrible . (25) Thegirls ate pizza. +Two -girls became ill afterwards. In both pieces of discourse, the subject NP of the second sentence introduces an NP that is 'object-new'. F-A theory would normally assign accent to the Noun, since this is the right sibling. How should this phenomena be accounted for? One approach, along the lines of Rooth (1 992), would be to construe old as being in contrast with some contextually understood other adjective (young), and to construe two as in contrast with either the , or with all stronger deterrniners.22 But sometimes, the word on which accent lands can hardly be said to stand in contrast with anything. For example, consider

(26) IfSusan owns a car, she must be rich. Well anyway, you don't +need a -car in New York City . Here, the second occurrence of car is deaccented, and accent shifts to need, but it would be highly implausible to attribute this shift to any contrastive force in need. Consequently, a 'contrastive' account cannot cover all the facts at hand. According to the present account, the dynamics is the other way round: car is deaccented since it is given information, and as a result-by another application of the Default Accent Rule-accent shifts to need. To make this idea precise, some notion of a givenness must be defined that makes the words men , girls , and car given, even if they are part of a + FNP . I will define such a notion and call it concept-givenness, in opposition to the notion of object-givenness that has been discussed so far. How should concept-givenness be defined? Evidence suggests that concept givenness affects words of all categories. Note that an antecedent that appears within the scope of a logical operator can legitimize the lack of an accent in a

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

while how deaccenting due to object-givenness can be accounted for in the same way as Lexical deaccenting, by a slight extension of the Default Accent Rule. A very similar mechanism can account for a kind of deaccenting that does not arise from object-givenness of a F NP that is embedded in a + F NP . Consider the following examples:

20

A Semantic Perspective on Sentence Accent

word that appears outside the scope. In (26) this happened with material in the scope of a conditional. (27) illustrates how negation has no 'blocking' effect either:

(27) Susan does not own a car. Even this + cheap -car is spoilt on her .23 Logical boundaries do , of course, affect object-given ness, which is, by definition, sensitive to all DRTs constraints on accessibility. For example, negation blocks object-givenness; wimess, for example, the obligatory accent on

cadillac

in

(28) Susan did not take a car to work. The brakes ojthe +cadillac werefaulty . givenness than in object-givenness. Therefore, a tentative bound on the 'memory limitations' for concept-givenness is hypothesized, even though the exact form of this limitation is a matter for further empirical research. So far, we have only studied concept-givenness in cases where the antecedent and the given word are identical. Note, however, that concept-givenness, like object-givenness, is a semantical affair: not the word itself, but the referent counts. A word can be deaccented because it is synonymous with a word in the previous sentence:

(29) juan owns a bicycle. You need a -bike ifyou work at Stanford. Moreover, if the reference of the first word (bicycle) is subsumed by that of the second (vehicle), then the second may be deaccented as well:

( 30) Juan owns a bicycle. You absolutely need a -vehicle ifyou work at Stanford . But here, the relation is asymmetrical: it is only the subsuming (i.e. exten sionally larger) word that can be deaccented due to the subsumed one. For instance, one cannot deaccent bicycle because of an earlier occurrence of vehicle .

( 3 1 ) juan owns a vehicle. * You absolutely need a -bicycle ifyou work at Stanford . Again, what counts is known or believed subsumption of the referent of one word by the referent of another. For example, if thepeople is known to refer to a group of women, then women may be deaccented since the referent ofpeople is subsumed by (in this case, equals) the referent of women :

(32) John pitied the poor people. +Two -women were severely ill . Deaccenting seems to be a directional process that operates strictly from left to right There are apparent exceptions of bidirectional deaccenting, but there some sort of contrast must always play a role, as has been argued in Rooth (1992). For example, in

(33) An +American -farmer was talking to a +Canadian -farmer.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Recency, on the other hand, seems to be a much stronger factor in concept

Kees van Deemrer

21

both occurrences o ffarmer lack an accent, but this is better explained by an obligatory (contrastive) accent on American and Canadian , which triggers the loss of an accent on farmer (see note 2 1 ). In light of these considerations, I propose to define concept-givenness as follows:

Concept-givenness: An occurrence w

of a word is concept-given if the same or the previous sentence contains, to the left of w, another occurrence w', of an expression,24 whose reference is known to be subsumed by that of w . The deaccenting principle 2 can then be given the following form:

2 ' Deaccenting .

Deaccent not only certain function words, but also all the

The Default Accent Rule would take concept-givenness into account in the following way. Suppose a Strong node has a Weak as well as a Strong daughter. Then if the Strong daughter is deaccented, a transformation applies that changes the Strong daughter into a Weak one and the Weak daughter into a Strong one. But what if both daughters of a node are deaccented? Double deaccenting arises in a sentence like (30) above, where the determiner a will be deaccented for lexical reasons, while the other daughter of the Noun Phrase, namely the Noun vehicle, will be deaccented since it is concept-given. An example in which both daughters of a Noun Phrase are deaccented due to concept-givenness is the following piece of discourse (which might be the beginning of a brainteaser):

(34) juan is the owner of two bicycles. Whenever he meets someone who owns -two -vehicles too , . . .

As before, the two Noun Phrases two bicycles and two vehicles are object-new, since they introduce potentially different sets of vehicles into the discourse. These examples suggest that if both daughters are deaccented, then neither of them needs an accent.

3 ' Default Acccent Rule . If the Strong daughter of a node is deaccented while the other one is not, then the labels Weak and Strong are reversed. If both daughters are deaccented, then both are relabelled Weak.25

'Deaccented' here means marked as either Lexically deaccented (Marking: L , for Minus Lexically Accentable); or Object-given (Marking: - F, for Minus Focus); or Concept-given (Marking: C for Minus Concept-new). In a diagram: -

-

1

2

,

Construct metrical tree Assign +1-F(ocus) marking,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

words in any occurrence of a Noun Phrase that is identity-anaphoric to another Noun Phrase (i.e. any expression that is object-given), as well as any occurrence of a word that is 'concept-given'.

22 A Semantic Perspective on Sentence Accent Assign +I-C(oncepr-newness) marking, Assign +1-L(exical Accentability) marking Apply new Default Accent Rule 4 Apply Modified Correspondence Rule

3

The order of the assignments of F, C, and L is arbitrary, since they do not depend on each other. Moreover, the Default Accent Rule does not distinguish between the three markings, so they could be lumped together into one single marking that does not distinguish between different sources of deaccenting. They were kept separate for expository reasons, however.

In the following, a few small pieces of discourse will be used to illustrate some of the key concepts introduced so far. We will use sentences in which there are no obstacles to anaphoric accessibility, and consequently there is no need to draws DRSs for them. In addition, it will be assumed that all the syntax rules that have two daughters have the right daughter marked as strong and the left one as weak.26 First, consider (3 5) followed by (36):

(3 s) The children were upstairs . (36) The girls cried . A simple phrase structure grammar might analyse (36) along the following lines (c£ Figure 7)P Now that we are dealing with a realistic example, we have to be specific about how to mark major phrases other than the NP for + or - F. Any VP will be marked as + Fprovided it encodes information that is, intuitively speaking, not contained in the previous sentence. In analogy with newness of NPs, newness of VPs thus depends on the newness of the entity itself, and not on whether the resulting predication is informative.28 Now first assume that both major phrases in (36) constitute new informa tion. Both are marked + F, while all the children in the domain happen to be The girls cried

I

\

VP, s

NP, w

I

I

V, s

DefNP, s

I \

Defdet, w

I

the, s

CN, s

I

girls, s Figure 7

I

cried, s

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

An extended example

3.3

Kees van Deemter

23

girls. In this case, the Default Accent Rule plays no role and the Correspon dence Rule will ensure that accent trickles down from DefNF to girls and from VP to cried 29 Next, change the example by assuming that (it is known that) all children are girls. In that case, there are two reasons why accent will not land on girls . Firstly, the word girls is now concept-given, since the girls subsume the children. Secondly, the DefNF as a whole is now object-given, and therefore - F. As a consequence, all movements within the NP become irrelevant, and only cried will be accented. Now for a slightly more complex sentence. Imagine the same context, The children were upstairs, but what has to be determined is the accents in

Three of the girls cried

\

I

NP, w

VP, s

I

I

NondefNP, s

\

I

Predet,

I

Nondefdet,

I

w

w

\

of, s

I

DefNP, s

I

Defdet,

I

three, s

V, s

the,

w

w

cried, s

\

CN, s

I

girls, s

Figure 8

Assume this utterance can be analysed as depicted in Figure 8. Again, first imagine a situation in which some of the children are not girls, so the girls constitutes new information. Then the Default Accent Rule is idle, and accent in the NondefNF, which is an indefinite NP and consequently + F,30 will trickle down to girls as a result of the Correspondence Rule. Assuming that the VP conveys new information as well, accent in the VP will, of course, land on the only word it comprises: cried . Now move on to the more interesting situation in which all the children are girls. As a result, the defNF thegirls is - F, so the Default Accent Rule causes the Weak/Strong values on Predet and DefNF, respectively, to be swapped. But now assume that all prepositions are listed in the dictionary as L , so the w and s labels on the daughters of the Predet are swapped as well. Then accent will trickle down from the NondefNF to the Nondefdet three . As in the earlier example, the fact that girls is now -

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(3 7) Three ofthegirls cried .

·

24 A Semantic Perspective on Sentence Accent

concept-given does not play a role. On the other hand, if this piece of discourse is followed by a third sentence saying {38) (On the other hand, ) three ofthegirls sang, then both daughters of the Predet Three of are concept-given, and therefore deaccented, while the DefNF thegirls is still -F. As a result, the entire subject of the sentence has to make do without an accent (due to rule 3 '), and the only obligatory accent lands on sang. 4

C O NC L U S I O N

(i) Whose knowledge determines whether a piece ofinformation counts as given or as new? -In both object-newness and in concept-newness, it is speaker knowledge that determines given/new status. {ii) At what lroel ofdescription is it that the entity is new? -Both notions have to be defined at the level of semantic denotations, while it is always the conglomerate of objects that counts, rather than the individual elements of the conglomerate. Hence the distinction between Identity Anaphors, which count as given, and Subsectional Anaphors, which count as new information. Further, (iii) Once an entity is given, how long, and under what structural constraints, will it remain so?-Object-givenness is subject to the same accessibility con straints as Noun Phrase anaphora. Concept-givenness, on the other hand, is much less sensitive to structural constraints, but has to fulfil very strict constraints on the recency of the legitimizer.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The paper has outlined a set o f rules expressing how sentence accent and semantic information mutually constrain each other. Given background information about the denotation of content words in a discourse, intended meaning restricts the opportunities for sentence accent in speech. Conversely, given the same background information, sentence accent determines the sort of anaphoric relation that can be expected to hold in a given piece of spoken discourse. In the process, two different distinctions between given and new information have been defined, which seem genuinely irreducible to each other. For a word to require newness accent, it has to occur in an object-new NP, and in addition it has to be concept-new itel£ Thus, concept-newness is no sufficient condition for accenting: a word may be concept-new, but unless it adds up to introducing a new discourse entity, no accent is required. Let us briefly compare these two notions of'newness of information'. When something is advertised as new information, at least the following questions are in order (here followed by answers that follow from the account that was given in this paper):

Kees van Deemter 25

(39) (a) Who didJ. insult when Max came in? (39) (b) ]. insulted +Max. Here, Max constitutes the answer to the question in (39a), and consequently it appears focused (c£ e.g. van Kuppevelt 1 991). The fact that Max is, at the same time, also concept-given (and object-given) does not block an accent on Max . Deaccenting will only operate inside a constituent that is answer-new; wimess the following example (which is, like the previous one, copied from Oehrle 1988): (40) (a) What didJ. do when Max came in? (4o) {b) ]. + insulted -Max.32 If these observations are correct, then answer-newness can be incorporated into the framework of the present paper by simply marking constituents that serve as the answers to explicit or implicit questions as + F. If this is done, a + F

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Note that this proposal makes newness and givenness discourse-based and knowledge-based at the same time: the key is always in the knowledge of the speaker, but what counts is his or her knowledge about the reference of certain expressions in the discourse. For example, when the first sentence of a discourse brings up the name Mr. Clinton , then this name must be accented even if the existence of its referent is common knowledge. Even a known entity can only become 'given' by linguistic, or possibly deictic means. The current proposal shows how theories originally designed to deal with the semantics of anaphora can also be instrumental in predicting whether a given expression constitutes given or new information and whether it requires pitch accent. Thus, intonation contributes to the 'dynamics of interpretation' that is currently of so much concern to semanticists (see e.g. van Benthem 1 99 1 for an overview), proving that traditional reservations about the 'informational' relevance of intonation as worded in Bolinger (1989) (see Introduction of the present paper) are less and less warranted. I should repeat that newness is only one possible source of accent, and that other sources of accents exist. Although a unified theory of how all these fit together is obviously beyond the scope of the present paper, here are some speculative remarks. First, there are some other 'objective' factors that can be decoded by a hearer. One example is the mechanism of'answer-newness' (my term) that was perhaps first noted in Kuno (1 972) and discussed at length in van Kuppevelt (1991), requiring that any constituent that functions as an answer to an explicit or implicit question is focused.31 Dick Oehrle has observed that an expression that is marked as focused due to answer-newness can never be deaccented due to concept-givenness or object-givenness. For example, consider the following question-answer pair.

26

A Semantic Perspective on Sentence Accent

(41) Bergkamp and Van Basten are both realgoa/getters. Frankly, I'm as much impressed by +Bergkamp as by Van +Basten . The notion of contrast that is accountable for these remaining phenomena must be very different from the issues of novelty and givenness that have been dealt with in this paper, since it involves some sort of syntactic/semantic symmetry between the two items involved (I am impressed by Bergkamp versus I am impressed by Van Basten ) and is, in most cases, non-directional and therefore, presumably, non-anaphoric in namre. This motivates why the notion of contrast has not been discussed at length in this paper. To what extent contrast, as well as such 'subjective' factors as liveliness and emotion, can be treated along the same lines as answer-newness, by allowing them to contribute only + F marking {in the case of contrast, emotion, liveliness) and never -F marking {in the case of a lack of contrast, emotion, liveliness), is a matter for further research. Subjective factors, of course, tend to introduce variability in accenting. However, the resulting variability seems to be constrained by a rule that says: subjective accents are only allowed if the intended interpretation can be recovered. For example, consider:

(42) The president visited many towns. Mr. +t-cfinton does not run out of steam quickly . This occurrence of Mr. Clinton constirutes identity anaphora, and therefore no accent is predicted. On the other hand, if the proper name is nevertheless accented, confusion is not likely to arise, as long as the circumstances make it clear that it is the president who is intended, rather than one of his namesakes. This suggests the hypothesis that subjective accents may be added as long as

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

marking on a constiment means that the constiment is either object-new or answer-new {or both). Thus, a constiment that is answer-new but object-given will still be marked + F and consequently, the Default Accent Rule will not be triggered. The same thing holds if a constiment is marked object-new but answer-given. On the other hand, the Default Accent Rule will be triggered if a part of an answer-new constiruent is deaccented due to objecr-givenness, concepr-givenness, or lexical deaccenting. Although the issue of incorporating answser-newness with the rest of our framework would merit further verification, the predictions of this preliminary account look plausible enough. Another objective factor that needs further smdy is the mechanism of contrast. For even though newness can, as was shown in section 3.2, account for many of the accenting facts that are usually explained by the notion ofcontrast, there is still work to do for a notion of contrast. For example, a contrastive analysis is still required for cases in which both items that are contrasted have the scams of given information, as in

Kees van Deemter 27

Acknowledgements The idea for this article arose from discussions at IPO with Jacques Terken and Arthur Dirksen, but most of the acrual paper was written during a very fruitful one-year stay at CSLI in Stanford that was made possible through a grant from the Netherlands Organization for Scientific Research (NWO), which is hereby gratefully acknowledged. Among the other people who contributed to this paper-some aspects of which are well beyond what I felt comfortable with, being a semanticist with little knowledge about prosody-! want to mention Cleo Condoravdi, Heleen Hoekstra, Jan van Kuppevelt, Dick Oehrle, Patti Price, Sieb Nooteboom, Ed Zalta, and Henriette de Swart, as well as a couple of anonymous reviewers, who corrected some remaining flaws. Needless to say, none of the people mentioned necessarily agrees with the positions defended here. An earlier version of this paper has appeared as CSLI Report no. CSLI-93-178, ISSN 1 042- 1 1 73. KEES VAN DEEMTER

IPO PO BOX 51) s6oo MB Eindhoven The Netherlands e-mail: deemter @ prl.philips.nl

Received: 0 1 .05.93 Revised version received: 26.1 1 .93

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

they do nor lead to irrecoverably wrong interpretations when they are interpreted by an 'objective' theory of the kind presented in this paper. But granted that subjective accents are sometimes allowed, their occurrence should still be expected to cost time and effort on the part of the hearer. Thus, if it is true, the hypothesis that was just formulated would explain Terken & Nooteboom's (1 987) finding that accents on given informacion tend to slow down processing (see section 3). The remarkable fact that, conversely, accents may not be deleted as easily by subjective factors may arise from the often claimed propensity of accent to highlight the non-obvious.If highlighting the non-obvious is the overaU function of accent, then the erasure of accents would be a hazard to successful communication, which is better avoided. This concludes my speculations beyond the confines that were set up in the Introduction. Implementing a semantic theory of accent, even in its present modest form, would be a formidable task, given the role of discourse theory and background knowledge in such a theory.But theories of discourse are making progress, and even though background knowledge in general is hard to model, knowledge about word denotations is available in most natural language understanding systems. For example, current question-answering systems typicaUy have complete (although, of course, application-dependent) knowledge about the denotation of the words in their domain, and this is, as we have seen, more than sufficient to determine whether two Noun Phrases can stand in the relation of ldencity Anaphora, which is the key notion of a theory of newness accent.

28 A Semantic Perspective on Sentence Accent

NOTES 7

8

9

10

II 12

13

See Liberman & Sproat ( I 992) for a treatment. Some preliminary experiments that involve testing of some of the hypotheses of the present paper are currently under way at the University of Utrecht under the direction of Wilma van Donselaar. For example, if meaning is formally understood as Montague's Intensions, then Mr. Clinton relates to a constant function and the president relates to a non constant function that maps indices on whichever individual is the president given with respect to them. The rwo are clearly different. See Rooth (I 992). It seems true that rwo items can be contrasted not only on the level of denotations, but on any level of linguistic analysis. The fact that, in ( s) and (6), both items that stand in opposition can be stressed speaks for an analysis in terms of contrast. See also the discussion on 'contrast' versus 'givenness' explana tions in section 3 .2. This will become clear later. See, for example, example (28). Other varieties of the context-change theory of anaphora include Heim's file change semantics (Heim I 982), Barwise's treatment of anaphora in situation semantics (Barwise I 98 s ), and Groenen dijk & Srokhof's theory of Dynamic Montague Grammar (Groenendijk & Stokhof I 99 I). For example, a piece of discourse such as

Some soldiers died. They were buried on the spot would usually be taken to imply that all soldiers who died were buried on the

spot. The formula 3X (some soldiers (X) & died (X) & buried(X)) would not have this implication. See van Eijck ( 1983) for a treatment. 14 Contexrualization of CNs remains neces sary in an NP that embeds another NP, if the rwo are contexrualized with respect to different context sets. For instance, in All

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I I will take accent to b e defined a s the prosodic phenomenon whereby part of a sentence stands out as prominent for listeners. It is assumed that pitch move ment is the most important determinant of accent, with loudness, and possibly other factors, such as speech tempo, playing a lesser role. See 't Hart et a/. (I99 I } for a precise framework. 2 Arguably, the notion of contrast is the central one in Rooth's work. An accented expression is typically construed as stand ing in contrast with one or more 'alternative' expressions. The main con tribution of accent is to help determine which alternative expressions are to be taken into account (Rooth I 992). In particular, a number of researchers, apparently using a slightly more liberal definition of accent than researchers in the tradition of 't Hart et a/. ( I 99 I }, have contended that low pitch accents can mark 'given' information (Pierrehumbert & Hirschberg I 990). 4 See, for example, van Kuppevelt ( I 99 I } for a n excellent survey and criticism of existing proposals in this area. This includes both the 'bridging' pheno mena that have been investigated in the psycholmguistic literature (discussed briefly as Relational Anaphora in section 2.2) as well as the phenomenon of so called Subsectional Anaphora (ibid.) 6 Rather than incorporating lexical weight into the grammar, I follow the conven tional practice of separating syntactic factors (which will be encoded in the phrase structure rules of Focus-Accent theory) from lexical ones. Moreover, the latter notion ts simplified by means of a rough dychotomy berween words that do and words that do not 'like' accent (i.e. the +I-L marking in my version of Focus Accent Theory, c£ section 2.3). A more accurate theory might be obtained by using a more discriminating approach.

Kees van Deemter 29

people who bought raincoats, the quantifica

1s

I

I

7

8

19

20

21

22

23

24

difference in meaning or contextual feli city. From a purely semantic point of view, nothing is gained from viewing the second occurrence as anaphoric to the first, since this does not affect the meaning of either occurrence. But the fact that they can be so viewed causes the second occurrence to lose its obligatory accent. As usual, the fact that an accent is not predicted does not mean that accent is per se impossible, but rather that it is not necessary for reasons of informational status alone. In the concludint; section, the issues of variability in accenting will be taken up. Presumably, the F-A mechanism here would be a reverse of the Default Accent Rule, saying that if a node has a Weak and a Strong daughter, while the Weak daughter stands in contrast with some thing else, then the Weak and Strong labels are swapped. Rooth does not dis cuss the mechanism by which accents land on one particular part of a +F NP. Moreover, the motivation of Rooth's (1992) theory of contrast (ibid., p. 8o) is intended to be restricted to symmetric contrast, even though the actual rules proposed there make wider predictions. Note that the absence of accents in -car cannot be explained by any plausible version of object-givenness, since the discourse entities in question are com pletely novel. This holds m (26} as well as in (27). Cases requiring some form of concept-givenness have already been mentioned in Chafe (1 976). Note that the 'antecedent' for concept givenness need not be a single word! An example is Children in rags . . . ; The -poor . ., where poor does not subsume chil dren , while it does subsume children in .

rags. 25 The same predictions are obtained if the parent node is relabelled F, but this seems less appropriate since it conflicts with the 'given information' interpre tation that was assigned to -F. -

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

16

tion over people may be restricted to some set of people who were mentioned earlier, while the raincoats are recruited from the entire domain of discourse (van Deemter I992). This definition of the meaning of a relativized condition deviates from the one in van Deemter (1 992) in form but not m substance, making use of an equality that is proved m the earlier paper. The definition of Anaphoric NP inrroduction was left implicit m the earlier presentation. This observation is due to Cleo Condo ravdi (pers. comm.). Note that the pre supposed information IS respected by negation, to mention one classic test for 'presuppositionhood'. In this simplified version of Dirksen's theory, there will always be an accent on the NP as well as on the VP. This will not be true for a version of the theory in which only those maJOr phrases are marked as + F which have 'new infor mation' starus (as will be proposed in section 3). But since the theory is based on +1-F markmg on all major phrases, this prediction will survive in so-called 'What's new?' contexts, where all the information in the sentence has 'new' status. (Prediction: What's new? +Mary reads a +book!, rather than What's new? +Mary reads book!) It is not immediately obvious how this flaw can be repaired-if it is a flaw, that is. To view the pronoun it as accented for lexical, rather than semantic reasons is, of course, not an optimal solution. It will become obvious in later sectiOns how this case of lexical deaccenting can be sub sumed under a semantic variety of deaccenting due to 'object givenness'. This is not necessarily a unique minimal accenting pattern. In particular, optional rules such as Baart's Rhythm Rule induce non-uniqueness to reflect that, for instance, one can either say a very +nice +girl or a +very -nice +girl, without

" 30 A semantic Perspective on Sentence Accent been rrue for the VP if it would have been -F. 30 It follows from the well-known famili arity constraint, of course, that indefinite NPs cannot have identity anaphora and, as a result, they must be marked as - F. For a discussion of how, in the setting of GDRT, the familiarity constraint follows from general Gricean considerations, see van Deemter ( 1992). 3 1 Note that this makes information of the hearer, rather than the speaker, the crucial issue for answer-newness; com pare point (i) above. 32 Here, the VP insulted Max is answer-new, and now the fact that Max is concept given (and object-given) leads to a shift of accent within the VP.

RE FERE N CE S Baart,J. L. G. ( I987), 'Focus, syntax and accent placement', unpublished Ph.D. thesis, Leyden University, The Netherlands. Barwise,J. ( I985) , 'Noun phrases, generalized quantifiers and anaphora', CSLI, Stanford, Informal Note. Barwise,J. & R. Cooper ( 1 980), 'Generalized Quantifiers and Natural Language', Lin guistics and Philosophy, 4, I 59-2 I 9· Benthem, J. van ( I 99 1 ), 'Language in action: categories, lambdas and dynamic logic', North Holland, Amsterdam. Bolinger, D. ( 1 972), 'Accent is predictable (if you're a mindreader)', Language, 48, 63344· Bolinger, D. ( 1 989), Intonation and Its Uses: Melody in Grammar and Discourse, Edward Arnold, London, Melbourne, Auckland. Chafe, W. L. (I 976), 'Givenness, contrastive ness, definiteness, subjects, topics and points ofview', in C. N. Li (ed.), Subject and Topic, Academic Press, New York. Clark, H. H. and S. E. Haviland ( 1 977), 'Comprehension and the given-new con tract', in R. 0. Freedle (ed.), Discourse Pro-

cesses: Advances in Research and Theory, Vol. I: Discourse Production and Comprehension ,

Erlbaum, Hillsdale, NJ. Deemter, K van ( 1991 ), 'On the compostion of meaning', Ph.D. thesis, University of Amsterdam. Deemter, K van ( 1 992), 'Towards a general ization of anaphora',journal ofSemantics, 9, I, 27-5 1 . Dirksen, A. ( 1 992), 'Accenting and deaccent ing: a declarative approach', Proceedings, Coling Conference, Nantes, France. Dirksen, A. & H. Queue ( 1991 ), 'Prosodic analysis: the next generation', in V. J. van Heuren & L. Pols (eds.), Analysis and

Synthesis ofSpeech: Strategic Research Towards High-quality Text-to-speech Generation , Mouton de Gruyter, Berlin. Eijck,J. van ( 1983 ), 'Discourse representation theory and plurality', in A. ter Meulen

(ed.), Studies in Mode/theoretic Semantics,

GRASS-I, Foris, Dordrecht. Groenendijk, J. & M Stokhof ( 1 991 ), 'Dynamic predicate logic', Linguistics and Philosophy, No. 1 4, pp. 39-100.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

26 In Dirksen's formulation, the rules would have the form a - {3/y, to denote the rewrite rule a - {3 + y, with the addi tional information that {3 is weak and y is strong. 27 As before, W equals weak, and S strong. The distinction between DefNPs (defi nite NPs) and NondefNPs (nondefinite NPs) serves no other purpose than to account for partitive constructions as in (37). 28 This intuitive notion can be refined by using Klein's extension of DRT in which he inl'l"oduces Reference Markers for VPs (Klein 1987). 29 The fact that the NP is the Weak daughter of the S does not affect the outcome, since the NP itself constitutes new information. The same would have

Kees van Deemter 3 1

·

'Accents, focus distribution, and the perceived distribution of given and new information: an experiment',]. Acoust. Soc. Am. , 82, 5, November. Nooteboom, S. G., J. G. Kruyt, & J. N. B. Terken {I98 1), 'What speakers and listen ers do with pitch accents: some explora tions', in T. Fretheim (ed.), Nordic Prosody II: Papersfrom a Symposium , Tapir, Trond heim, Norway. Oehrle, R. (1 988), 'Sources and structures of linguistic prominence in English', in S. Schiffer & S. Steele (eds), Cognition and Representation , pp. 209-41 , Westview Press, Boulder, Co. Pierrehumbert,J. & J. Hirschberg ( 1 990), 'The meaning of intonational contours in the interpretation of discourse', in Intentions in Communication , MIT Press, Cambridge, Mass. Price, P. J., M. Ostendorf. S. Sharruck Hufnagel, & C. Fong ( 1 99 1 ), 'The use of prosody in syntactic disambiguation', ]. Acoust. Soc. Am. , 90, 5, December. Roberts, C. (1 987), 'Modal subordination, anaphora and distributivity', Ph.D. thesis, UMASS. Rooth, M. (1 992), 'A theory of focus interpretation', Natural Language Semantics, no. 1 , 75-1 16. Stechow, A. von (199 1 ), 'Current issues in the theory of focus', in A. von Stechow & D. Wunderlich (eds), Semantics: An Interna tional Handbook of Contemporary Research , De Gruyter, Berlin. Terken, J. M. B. (198 5), 'Use and function of accentuation: some experiments', Ph.D. thesis, Leyden University. Terken,J. & S. Nooteboom ( 1987), 'Opposite effects of accentuation and deaccentuation on verification latencies for "given" and "new" information'. Language and Cognitive Processes, 2, 3/4, 1 45-63. Westerstahl, D. (r985). 'Determiners and context sets', in J. van Benthem & A. Meulen (eds), Generalized Quantifiers in Natural Language, Foris, GRASS-4, Dor drecht.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Halliday, M A. K. ( 1 967), 'Notes on transi tivity and theme in English', Journal of Linguistics, 3: 1 99-244. Hart, J. 't, R. Collier, & A. Cohen (1 990), A Perceptual Study of Intonation: An Experi mental Phonetic Approach to Speech Melody, Cambridge University Press, Cambridge. Heim, I. (1982), 'The semantics of definite and indefinite Noun Phrases', Ph.D. thesis, Amherst, Mass. Hirschberg, J. (1 990), 'Accent and discourse context: assigning pitch accent in synthetic speech', in AAAI 1 990, p. 9 5 3 · Kamp, H. (r98 1), 'A theory of truth and semantic interpretation', in J. Groenen dijk, T.Janssen, & M. Stokhof(eds), Formal Methods in the Study of Language , Mathe matical Centre Tracts no. 1 36. Klein, E. (1987), 'VP ellipsis in DR theory', in J. Groenendijk & M. Stokhof (eds.), Studies in Discourse-Representation Theory and the Theory of Generalized Quantifiers, GRASS8, Foris, Dordrecht. Kuno, S. (1972), 'Functional sentence pers pective: a case study from Japanese and English', Linguistic Inquiry, 3, 269-320. Kuppevelt, J. van (1991), 'Topic and com ment', Ph.D. thesis, Nijmegen University. Ladd, D. R. { 1 980), The Structure ofintonational Meaning: Evidence from English , Indiana University Press, Bloomington, Ind. Lewis, D. (1973), 'Phonetic disambiguation of syntactic ambiguity', Glossa , 7, 2, 1 07-2 1 . Lehiste, I . (1 969), Convention: A Philosophical Study, Harvard University Press, Cam bridge, Mass. Lieberman, P. (1 967), Intonation, Perception and Language, MIT Research Monograph No. 38, MIT, Cambridge, Mass. Liberman, M. & E. Prince (1 977), 'On stress and linguistic rhythm', Linguistic Inquiry. Liberman, M. & R. Sproat (1992), 'The stress and structure of modified noun phrases in English', in I. A. Sag (ed.), Lexical Matters , CSLI Lecture Notes No. 24. Most, R. B. & E. Saltz ( 1979), 'Information structure in sentences: new information', Lang. Speech , 2Z, 89-95. Nooteboom, S. G. & J. G. Kruyt ( 1987),

journal ofSemantlcs 1 1 : 3 3-53

© Oxford Umversu y Press

1 994

Ups and Downs in Scalar Inferences L I TA LUN D Q U I S T Copenhagen Business School R OBERT J . JARVELLA University ofMontreal and Lund University

We examine the notion (Ducrot 1 988; Anscombre 1 989) that use of scalar adverbs such as almost and only gives rise to systematic semantic inferences. Inferences of the so-called 'topos' rype were explored in sentence pairs in Danish and Finnish texts describing competitive events. An antecedent sentence describing a level of achievement obtained in terms of almost p or only p was predicted to give rise to a reasoning scheme involving two scales and relating the achievement to likelihood of an individual being in a winning or a losing position. Use of this inference scheme was examined from effort in u nderstanding, and the interpretation given to, an ambigu�us consequent sentence. In an overt question-answering task, consequent sentences were found to be interpreted in about 90 per cent of cases in the way predicted by the version of topos theory tested (Lundquist, 1 98 7). In a rimed reading task, whereas antecedent sentences were read in less time when scales taken to be activated in them were oriented upwards, con sequent sentences were read faster if they were co-oriented with the antecedent, and satisfied an inference which was expected to arise there. The results are discussed in terms of different linguistic accounts of scalar adverbs, and the origin of scales in cognition.

I N T RO D U C T I O N Some expressions in language involve scales in their interpretation. Such expressions comprise very heterogeneous linguistic material, including logical quantifiers (all, some , none, etc.), non-logical quantifying expressions (most, half of, Jew, aJew, etc.) and quantifying time expressions (always , often , never, etc.), but also scalar predicates (love , like , must , shall, can , etc.) and scalar adverbs, such as almost, only, more than , less than , hardly , scarcely . Various linguistic descriptions have been applied to scalar expressions, and the relevance of these descriptions psychologically has been the subject of some research. Broadly speaking, linguistic descriptions of scalar expressions have focused either on their informative content ('How many times is often ?', 'How many are some ?'), or on the pragmatic implicatures that can be drawn from their ordinary use. However, there is, in addition, a more recent and more radical kind of linguistic description, which argues that scalar expressions give rise to INFERENCES of a systematic kind.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Abstract

34 Scalar Inferences

In this paper, among the many scalar expressions to be found in language, we shall focus on the interpretation of scalar adverbs. With the help of data gathered in experiments on reading done in two dissimilar languages, we will lend some support to the third and last of the linguistic theories mentioned above, according to which use of scalar adverbs gives rise to systematic inferences. As these inferences are explained in terms of scales which are directed upwards or downwards, we shall further explore whether, in accordance with findings in cognitive semantics and cognitive psychology about directionality, sentences invoking scales directed upwards are simpler or more natural to use than ones invoking scales directed downwards.

S C ALAR E XPRE S S I O N S I N L I N G U I ST I C S

Scalar expressions have constituted a battlefield for linguists, who disagree not only about what to call them, but also where to put them, i.e. as to whether they should be treated in the domain of pragmatics or semantics. Most linguists, however, seem to agree to conceive of the expressions in terms of scales, whetlier this be in terms of STRENGTH on a scale (in the pragmaticist explanation), or in terms of DIRECTIONALITY (in the semanticist one). For a pragmaticist, expressions belonging to paradigms including quantifiers all some I none-any, epistemic adverbs such as certain-likely-possible , and qualitative predicates such as adore-love-like , boiling-hot-warm Ifreezing-cold-cool (Horn I 989), can be located on a scale according to their strength, i.e. to the quantity of the expression, all being stronger than some , certain being stronger than likely, etc. It is on the basis of such scales that the typical pragmatic reasoning scheme takes place and goes: 'Given an utterance p, first determine the truth conditional meaning of p, and second, describe possible additional meanings with the help of, e.g., the Gricean maxims of Quantity, Quality, etc.' It was, e.g., via such a pragmatic reasoning scheme that Sadock ( I 98 I ) arrived at the conclusion that an utterance almost p is equivalent to not p , because, and this is where a conversational implicature enters, had Sender known p , he would have said so (maxim of Quantity), and not used the weaker form almost p (schema I ): SCHEMA

I

stronger form weaker form

1

p almost p

-

not p

Opposed to these pragmaticists, there are linguists who plead for a new and quite radical SEMANTIC standpoint, which takes scalar expressions notjust to add

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I

Lira Lundquist and Robert J. Jarvella 3 5

some extra meaning to a basic one in terms of truth, but to contain in their own right and as a semantic primitive, an autonomous meaning which is inherently scalar and directional in narure. In this semantic explanation of scalarity in language, which has been elaborated and is still being extended by Oswald Ducrot and Jean-Claude Anscombre (e.g. 1 988, 1 989 respectively), expressions such as almost p and more than p activate a scale which is oriented in the same direction as p because they lead to the same type of conclusion as p . This reasoning scheme is supported by continuation tests, as exemplified below: (1) It's six o'clock. Hurry up! (2) It's almost six o'clock. Hurry up!

(3) It's not six o'clock. Take your rime.

(4) It's only six o'clock. Take your rime. The inappropriateness of combining these scales in the opposite manner is evident from examples ( s ) and (6). ( s ) ?It's only six o'clock. Hurry up! (6) ?It's almost six o'clock. Take your rime. The Ducrot-Anscombre explanation, illustrated in schema 2, is thus radically opposed to the truth-conditional, pragmatic one according to which, as we saw in schema 1 , almost p equals not p. SCHEMA 2

more than p I + almost p only pi less than p

l !

p

not p

In the semantic theory, however, scalar expressions activate not only the directionality of the scale on which they themselves are located. When used in contexts, they also activate a second scale to which they are in some way related. Thus, used in utterances such as (7) and (8),

(7) Nichols obtained almost 1 29 points

(8) Nichols obtained only 1 29 points

the scalar adverbs almost and only are taken to activate not only the kind of scale shown in schema 2, but also-and this is where the inference comes in-a second scale, of the kind shown on the right in schema 3·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Expressions such as only p and less than p activate a scale which is corres pondingly oriented in the direction of not p , since they permit the same kind of conclusion as not p :

36 Scalar Inferences

SCHEMA 3 many pomts

almost x points only xpoints few points

+

! j

+

1 j

more likely to win

less likely to win

(the more (the less) P Q, the more (the less) P Q') For examples (7) and (8) above, the inference-scheme might be the following symmetrical 'competition'-topos, with the more chosen for (7) and the less for (8): (the more (the less) points Nichols obtains, the more (the less) likely he will win) So to summarize, the scalar explanation in the Ducrot-Anscombre frame work2 differs from the pragmatic ones first in that it associates almost p with p and only p with not p , and second in that it operates with two scales to explain the specific graded type of inference that scalar expressions are claimed to activate. What remains common, however, to the two types of explanation pragmatic and semantic-is that scales are seen as being either positive or nega tive, and furthermore that positive is seen as being directed upwards and negative downwards. In this paper, we will refer correspondingly to up -scales and down -scales. The fact that there are three different kinds of notation which overlap, namely the scalar more-less , the vertical u�down , and the polar positive negative, may be due to a mere 'notational kinesthesia' (Sapir 1 944 quoted in Horn 1 989) felt especially by linguists, but it also may be due to a more funda mental aspect of how human beings perceive their world. The latter, at least, is what is claimed in cognitive semantics Qohnson 1 987; Lakoff 1 987), with which the semantic topos explanation has several points in

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The content of the second scale intervenes not via linguistic mechanisms but rather via extra-linguistic knowledge and norms.1 For instance, the two-scale relation in schema 3 is based on a norm stating that 'the more points you score (in a competition, an election, etc.), the more likely you are to win (that competition, that election, etc.)'. However, the inference relating the two scales to one another is taken to be activated by the scalar expression used in the utterance. Ducrot and Anscombre have formulated this specific inference scheme in terms of a 'topos', which is characterized by being GRADED (the grad ations coming from the scales) in the antecedent as well as in the consequent, and which as a result has the following canonical form:

Lita Lundquist and Robert J. Jarvella 37 common, and which we shall explore further in this article. In fact, scales, though not to our knowledge scales activated by adverbs, have been treated

(more-less ) has been related (positive-negative). There, the

abundantly in cognitive semantics, where scalarity to directionality (u�down ) and also to polarity

'more is up, less is down' relation has been explained as being a general pheno menon in our mental structuring of the world, stemming from a correlation in our daily life experience: whenever we add more of a substance, its level perceptively goes up, and vice versa (Lakoff I 987:

276).3 This regular perceptual

correlation in our experience between quantity and verticality has given way to a conceptual IMAGE SCHEMA which can explain, via a metaphorical 'source-to target-mapping', why the more abstract scales which underly X got almost/more

In cognitive psychology, there is also substantial agreement that the saliency and ubiquitousness of the up-down dimension (the gravitational vertical) is used extensively to structure not only our perception and mental representa

tion of space, but also our verbal descriptions (see e.g. Shepard & Hurwitz

I 98 s).

The correlation between verticality and polarity has been pointed out by Clark

( I 973, I 976), who maintains that upwardness (as well as aheadness and before

ness) is positive in the perceptual field of the Ego (the space normally visible in front of the body), whereas downwardness (and behindness and afterness) is

negative with respect to the perceptual field. Experiments by, for example, Clark & Chase I 972 have shown that it is usually easier and faster to verify a description such as A is above B against a picture than to verify one like A is

(

)

below B , supporting the idea

that expressions conveying upwardness may also

be linguistically more primary. In the experiments described below, we set out to test first whether use of scalar expressions in utterances in fact gives rise to inferences such as described by Ducrot and Anscombre. Second, we tried to determine whether, in accord ance with ideas within cognitive and psychological semantics on perceived directionality, readers find it easier to deal with up -scales activated by scalar expressions than with corresponding down -scales. And third, elaborating on the notion of topos, we wished to examine whether it is easier and/or faster for readers to understand and relate the content of two sentences whose scales go in the same direction (both upwards or both downwards) than sentences whose scales go in opposite directions. This prediction was based on the hypothesis (see below) that inferences activated by an antecedent sentence will invite the drawing of a simpler deduction from a consequent sentence if the scales under lying the two sentences are co-oriented.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

than n points are usually conceived of as ascending, while those underlying the form X got only/less than n points are conceived of as descending.4

3 8 Scalar Inferences 2

EXPERIMENTS

Four versions o f a single experiment were conducted, using a set o f narrative texts written about various competitive events for individuals. To evaluate the generality in language understanding of the cognitive phenomenon of scalar inference and the inferential schemes postulated here, we performed the work described in two unrelated and structurally highly dissimilar languages, Danish and Finnish. Competitions of the kinds described in our texts exist in many cultures, but, in radically different languages, the way in which people reason from scalar expressions and their general knowledge need not be the same.

theory. Secondly, the srudy was performed in both languages srudied (a) in a paper and-pencil version, in which subjects read texts from test booklets and answered questions about them, and (b) in a timed version, in which reading times were recorded as subjects advanced a text sentence by sentence. The research was aimed at testing the following hypotheses: r. 2.

Use of scalar expressions induces an inference5 based on the interrelating of a pair of scales, as stated in the topos concept. Up -scales will be simpler to handle than down- scales, especially in an antecedent (almost p , or only p) sentence. This hypothesis is derived from the

cognitive-semantic and perceprual criteria mentioned above. J. In a consequent sentence, a principle of directional congruence with the antecedent will operate, i.e. it will be simpler to draw a deduction from a sentence which is directionally congruent with its antecedent (up following up, down following down) than one which is incongruent with it (up following down, down following up). To approach the inference-generating mechanism being hypothesized, we inserted scalar expressions into contexts which were semantically ambiguous, and examined whether the presence of the expressions would lead subjects to disambiguate the contexts in a systematic way. The contexts we chose to study were REFERENTIALLY ambiguous. Embedded in narratives, they were made up of sequences of two sentences, where the first sentence in a pair contained a PROPER NAME, and the second sentence contained a DEFINITE NOUN PHRASE6 which could be interpreted as referring to the same individual as the proper name,7 or to a different individual:

(9) Nichols obtained 129 points. The red-haired Irish lad seemed to be leading.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Cross-linguistic comparison of such languages provides an arena for support ing, or rejecting, the semantic-cognitive explanation proposed here as a general

Lira Lundquist and Robert J. Jarvella 39

Introducing a scalar adverb into the first sentence in (9) seems to have the effect of disambiguating the sequence, such that (the aim of many competitions being to get a high score, or score a lot of points),8 the definite noun phrase in the second sentence in ( 1 0) below is interpreted as referring to the same individual designated by the proper name (i.e. a unified reading is made), whereas in (I I ), it is read as referring to a different individual (i.e. a disjoint reading is made):

( 1 0) Nichols obtained almost I 29 points. The red-haired Irish lad seemed to be leading.

( I I) Nichols obtained only I 29 points. The red-haired Irish lad seemed to be This difference is easily explained by postulating a topos, which we will, as above, term the 'competition topos', and which takes a different form in ( 1 0) and ( 1 1 ). In ( 10), the adverb almost activates an up -scale, which is consistent with the leading position described in the next sentence, provided the reader maintains the same riferent: 'competition' topos C : (the more points that X gets, the more likely i t i s X will be leading). In ( I I ), the adverb only activates a down -scale, which is only consistent with the leading position described in the next sentence, if the reader introduces a different riferent : 'competition' topos C : (the less points that X gets, the more likely i t i s Y will be leading). '

We shall call these two variants of the competition topos 'simple topos' and 'complex topos' respectively, referring to the fact that the first variant implies one referent but that the second one implies two referents.9 That is to say, simple topos should induce unified readings of our ambiguous sequences, whereas complex topos should induce disjoint readings of them. A topos can also be viewed, as we demonstrated in schema 3 above, as the relating of a pair of scales each of which can be directed upwards (as activated by scalar adverbs as almost p, more than p , etc.), or downwards (as activated by only p , less than p , etc.). Our experiment was designed around sentences in texts which should evoke these scales partly in order to test the hypotheses that up scales should be easier (faster) to handle than down- scales, and, secondly, that relating a pair of scales which go in the same direction will be simpler than relating a pair of scales which go in different directions. For the competition topos, in up-down terms, the simple variant can be seen as relating two co-directed scales, up-up or down-down , and the complex variant as relating

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

leading.

40 Scalar Inferences

two opposite-directed scales, u�down or down-up . Thus, as illustrated in schema 4, the simple topos: (the more points X gets, the more likely that X is leading) can be seen as relating two upwards-oriented scales, and the symmetric (the less points X gets, the less likely that X is leading) can be seen as relating two downwards-oriented scales. SCHEMA 4 +

the more points X gets +

the less likely X is leading

A complex topos, in contrast, employs scales oriented in opposite directions. Thus, the competition topos C': (the more points X gets, the less likely that Y is leading) moves from an up -scale to a down -scale, whereas the symmetric (the less points X gets, the more likely that Y is leading) moves from a down -scale to an up -scale, as shown in schema s: SCHEMA s the more points X gets

+

+

the more likely Y is leading

the less points X gets

4

�

the less likely Y is leading

2.1

t>
The linguistic material studied

Eight basic texts were constrUcted, each narrating some kind of competitive event for individuals (a beauty pageant, a cross-country ski race, a dog show, a baking contest, a swimming competition, filling of a job opening, a horse race, and a tennis championship). The characters who were described in narratives were invented. One of the eight texts, translated from Danish into English. is illustrated in Table 1. The original Danish and the corresponding Finnish text can be found in the Appendix. For each type of contest described, two competitors were identified by name early in the text as being favorites to win, or having the highest previous merits. In the beauty pageant text shown in Table I, these two characters are 'Karina

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the less points X gets

t �

t �

the more likely X is leading

Lira Lundquist and Robert]. Jarvella 4 1

Table

1

One o f the Danish competition texts studied, rendered into English

The Beauty Competition The year's big beauty competition took place in the Circus Theater in Copenhagen. Eighteen models participated, chosen from various provincial Danish cities. As journalists and other media people had expected, Karina Madsen and Louise Fernholm stuck out as the big favorites.

After the finals, all the girls received a gift check and a trip to Mallorca. It was BT that had sponsored the beauty contest. In your opinion, who was clearly behind/ahead? (write name) In your opinion, who ran away with the victory? (write name) Note : The two vemons of sentences sr and sz were combined in all four vers10ns. The scalar adverb in

SJ was always chosen by raking the scale in rhe direcnon opposite ro rhat invoked in s 1 .

Madsen' and 'Louise Fernholm'. The central portion o f the text was then manipulated against this background. Table I illustrates how four versions of passages used were constructed in the above context. The text continued with two pairs of sentences, S I -s2, and SJ-S4, similar to those in ( 1 0 and ( I I above. These two pairs of sentences occurred in

)

)

succession. In the first sentence, S I again referred to one of the two favorites by name (e.g. to 'Karina Madsen'), and included in its predicate a scalar adverb which led the predication in an upward or in a downwards direction. The following sentence s2 then contained a definite noun phrase and a predication describing the individual referred to by this NP as being in a leading (favored)

)

or a trailing (threatened) position (see Table 1 .

In the second sentence pair, SJ repeated the pattern from S I of using a competitor's name plus a scalar adverb in its predication. But this time it was the other favorite (e.g. 'Louise Fernholm') who was named, and the adverb used led the predication in the opposite direction as in S I . Finally, the following sentence s4 contained a second definite noun phrase and a predication which described the individual referred to by this NP as being victorious. Following s4, the text returned to a more general level of description, and was concluded in a few sentences. To summarize, across four versions of texts studied, two different scalar adverbs were used in s I ; in each of these contexts, two different states were

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

s i In the round with bathing suits, Karina Madsen received high marks from only afew/almost all of the judges. s2 The tall blond Arhus-girl was clearly behind/ahead. SJ In the final with evening-gowns, Louise Femholm scored points with over/ under half of the judges. 54 The 20-year-old office girl walked away with the victory.

42 Scalar Inferences

=

2.2

Paper-and-pencil task

The paper-and-pencil task was aimed at getting explicitjudgments from native speakers concerning the meaning of sentences s2 and s4, in the context of s I and s 3 respectively. Twenty native speakers ofDanish and 23 native speakers of Finnish served as subjects. The subjects were working or studying at Danish and Finnish institu tions of higher learning. The test was administered using individual test booklets, and subjects were tested in groups. The sequence of competitions de scribed was the same for all subjects. Versions of texts were assigned to subjects using a Larin square. Each subject read one version of all eight texts, and two texts in each of the four forms studied. Subjects were told they were taking part in a study of text understanding, and would be asked to read some short texts and answer a few questions about them. Each text was typed on a separate page, followed by three questions. We present results here from two of the questions, which concern the meaning of s2 and s4 (the third question concerned comprehensibility of the texts). We tried to phrase this pair of questions so that their answers would show what kind of readings subjects had given to these potentially ambiguous sequences. As shown at the bottom ofTable 1 , to do this we formed constituent (wh -) questions from s2 and S4 in which the definite NPs were replaced by interrogative pronouns. The instruction 'Write the name' was added following each question to discourage subjects from writing down a definite NP as an answer. The task thus called for subjects to identify the characters mentioned in the two definite NPs in s2 and s4 by name.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

described in s2; two different scalar adverbs were again used in s3, with the direction of the scale reversed from si; no manipulation of s4 was made, since we felt that it was more natural to describe the winner of the competition. The predicates actually studied were quite heterogeneous and included neutral predicates such as wrote almost/only 65 words a minute (about a short-hand test in a job interview), quantitative predicates such as received high marksfrom almost all/ only aJew ofthejudges , and qualitative predicates such as made an almost sublime/ just an acceptablefrosting and swam with a nearly ideal/only a mediocre rhythm . The pattern of results which emerged for these classes was very similar, however, and we will not delve into them further here. In Danish, mean length for sr and s3 was I r .8 words (SD = r .6) and 73 characters (SO = 8), while for s2 and s4, it was 7·5 words (SO = 1 .2) and 47 characters (SO = 6). For Finnish, mean length for si and s3 was 8.6 words (SD = r.8) and 75 characters (SD 1 3), while for s2 and s4, it was 5·3 words (SD = r.o) and 49 characters (SD = 8).

Lita Lundquist and Robert J. Jarvella 43 2.3

Results

DANISH

100

c: 0 >

50

'So "' 00 c:

:.a "' �

"0 0 ...:

·a

I

0 1 00

I FINNISH

::l

c 0 1::!

0 Cl.

50

'0

c.. ::l

6.. ::l

c:

�

0

"9 c: �

0 "0

•

•

0 "0

c �

c:

�

6.. ::l

c.. ::l

0 "0

c.. ::l

6.. ::l

I

c.. ::l

c �

0 "0

sl -s2 s3-s4 Scales implicit in sentence pairs Figure 1 Percent unified readings made in the paper-and-pencil task, as a function of che direction of the scales taken to be implicit in sentences.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Of 688 responses, a total of five had to be discarded because of a misprint made in one version of one text. In addition, one subject failed to respond to two questions in one text. All remaining responses identified one of the two favored competitors from a narrative by name. Responses were scored as signaling a unified or a disjoint reading. A unified reading here means that the same referent was assigned across the sentences in a test pair, a disjoint reading that different referents were assigned. It was predicted that unified readings would be given when the scales implied in si and s2, or in s3 and s4, were co-directional (both up or both down ), and that disjoint readings would be given when the scales went in opposite directions (one upwards and the other downwards). Figure 1 shows the percentage of unified readings obtained in the various experimental conditions. All readings not shown were disjoint. It can be seen

44 Scalar Inferences

from Figure I that distributions of responses tended to go very strongly towards, or away from, unified readings. In both languages, about 95 per cent of readings were unified in the case of simple topos, when the scales implicit in s I s2 o r in s 3-s4 went i n the same direction (only upward o r only downward). On the other hand, most readings made (75 per cent of those in Danish and 92 per cent in Finnish) were disjoint in the case of complex topos, where the scales implied were not co-directional. In the two conditions, subjects thus interpreted the potentially ambiguous sequences in essentially opposite ways. 2.4

Self-paced reading task Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Subjects' responses in the above paper-and-pencil task appear to support the activation and use of inference schemes, issuing from scalar adverbs, as proposed in the work of Ducrot and Anscombre. In fact, there seems little doubt that such reasoning schemes were brought into play, since the data are so consistent with their use. On the other hand, the answers given to questions in the above task do not cast much light on our second or our third hypothesis: it is not evident from the results shown in Figure r that up scales are easier to handle than down scales, and only in Danish did a change in the directionality of scales (in a complex topos) seem to lead to any uncertainty in making disjoint readings. Given the limited conclusions that can be drawn from a paper-and-pencil task, and the fact that subjects' answers in the task were also given in circumstances where they had the opportunity to re-read the text before answering, we changed the testing situation somewhat in the second version of the experiment conducted. In this case, we asked subjects to read each narrative one sentence at a time, advancing the text read from a computer screen by pushing the space bar on the keyboard. The time spent reading each sentence was recorded in milliseconds, and used to infer how easy the various sentences were to understand. Passages were shown in the same order and using the same design as previously, except that one narrative was added as a practice item. Subjects were encouraged to try to understand each text as they read it; at the end of each narrative, the text reappeared as a whole, and a yes/no question was asked about the competition's result. The subjects were 36 adult native speakers of Danish and 40 adult native speakers of Finnish from the same general population as before, who had not participated in the previous study. Subjects were tested individually, and were paid the equivalent of about ro dollars each for their time.

Lita Lundquist and Robert J. Jarvella 2.5

45

Results

=

=

DANISH

-;:::-

�"' lu

up up /; up

..<:

� g 01.) c:

70

:a "' � c: "'

up .

�

wo

up51"'

�

dowo

up

up

Q)

�

up �

down

dow

Q)

.g

FINNISH

down down

90

=

=

50

90

"P "P

70

50

sl

s2

s3

s4

sl

s2

s3

Relative position of test sentences

Figure 2 Mean reading rime (ms/character) for sentences in the experiment involving up and down- scales. Sequences tested are connected by lines.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The data oftwo subjects who failed to complete the experiment were discarded. Sixteen of 2,368 (74 subjects X 32 test sentences) remaining data points needed to be estimated because of an error made in editing two stimulus flles. Our hypotheses were evaluated from differences in sentence reading rimes tested using planned comparisons, following a repeated-measures analysis of variance. 10 To adjust latencies for differences in length of our antecedent vs. consequent sentences, rime spent per character read was used as the dependent variable. In Figure 2, we summarize rime spent by readers in the different conditions tested, with Danish and Finnish data shown side by side. We will focus on results which relate to our main hypotheses and which were obtained in both languages. First, sentences in the antecedent posicion in sentence pairs (s1 and s3) which contained scalar adverbs taken to activate up -scales were read in less time than the almost identical sentences in the antecedent invoking down -scales. This result is consistent with our hypothesis that, in the antecedent, up -scales would be simpler co deal with psychologically than down -scales. For Danish, the rime advantage in reading sentences involving up -scales was reliable for both s I (F(1,34) 1 5.87, MSe 674, p < o.oo1) and s3 (F(1,34) 7.38, MSe 847,

46 Scalar Inferences

p = om). For Finnish, although antecedent sentences in both positions S I and s 3 tended to be read faster when they involved up -scales, the effect reached significance only for s3 (F (I ,38) = IO.o6, MSe = 245, p = o.oo3); for S I , F ( 1,37) = I . I 9, MSe = 5 59, p 0.28). Second, for sentences in the consequent position in sentence pairs (s2 and s4), largely a different principle than upwardness vs. downwardness per se seems to have been operating: that of simple vs. complex topos. Take s2 first Sentences in . s2 involving up -scales were not read significantly faster than sentences involving down -scales (for Danish, F ( I ,34) 0.4 1 , p = 0.5 3, for Finnish F(1 .38) - 0.45, MSe = 546, p - 0.30). Rather, what seemed to matter in s2 was whether the scale was co-directional or not with those in s I . This effect can b e seen separately for up - and down -scales in s2, by following the branches in Figure 2 which lead from s I . A comparison of reading times for s2 showed chat a sentence in s2 describing a leading position (an up -scale) was read faster when si had implied upwardness (in Danish, F ( I ,34) = 7.6 I , MSe - 707, p - o.oo9; in Finnish, (1,38) - 1 0.4, MSe - 5 1 7, p = 0.003), but that a sentence in s2 describing a trailing position (a down -scale) tended to be read faster when si had implied downwardness (in Danish F ( I ,34) = 1.97, MSe = I I66, p = o.I 7; in Finnish F (I . 3 8) = 2.47, MSe = 6 10, p = 0. 1 2). A pooled analysis of the data (see also Table 2) showed that s2 was read significantly faster not only when it followed an SI oriented upwards and s2 described a leading position (F(I,72) = I 7.72, MSe = 607, p < o.ooi), but also . when s2 followed an si oriented downwards and it described a trailing position (F( 1 ,72) = 4·34. MSe = 873, p = o.o4). The differences in reading times observed for sentence s2 are consistent with our hypothesis that certain inferences are activated by scalar expressions, and that making use of these inferences in a consequent sentence is simpler in the case of simple topos, when the scales involved are co-directed. Note that a pattern similar to that observed for sentence s2 was observed for s4.11 Sentence s4, which described the winning of a competition, involved an up- scale. When S4 followed an up -scale hypothesized to be activated in s 3, it tended to be read faster than when S4 followed a down- scale in s3. This result is interpretable in =

=

down-scales and simple vs. complex topos

Topos type

Simple

Scale in s2

up

down

up

down

time in seconds ms/character

3.s s 80.3

4-00 83.6

4-36 9Q.8

4·43 92-4

Complex

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Table 2 Mean overall reading rate for consequent sentences s2 with up- vs,

Lira Lundquist and Roberr J. Jarvella

47

the same way as the pattern observed above for s2. It is whether the scale in the consequent sentence goes in the same direction as in the antecedent sentence which seems to be decisive: when the scales are co-directed, a consequent sentence is understood faster than when they are not. 3

DISCUSSION

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The present studies of reasoning from texts containing adverbs such as almost and only appear to have largely substantiated our three hypotheses about scalar inferences. First, the data show, in accordance with our first hypothesis, that subjects reason as the topos theory predicts: the presence of a scalar adverb in what we have called simple topos systematically (95 per cent of the time) led to unified readings of the referents in a sentence pair, whereas using a scalar adverb in a complex topos led to (more than So per cent) disjoint readings. Furthermore, as judged from speed in reading consequent sentences in texts, not only did the inferences predicted by topos theory appear to be exploited, but, in context, the deductions drawn from them were made much faster for what we have called simple topos than for complex topos. Second, and in agreement with our hypothesis 2, the findings show that sentences invoking up -scales are read significantly faster than quite similar sentences invoking down -scales. This effect was found mainly for antecedent sentences, which were read on average about 8 ms faster per character (6oo ms faster per sentence) when their predicates were oriented upwards than when they were oriented downwards. Third, however, for consequent sentences, a nonsignificant (2.5 ms per character or IIO ms per sentence) trend which was found favoring speed in reading upwards-oriented sentences was clearly overshadowed and out weighed by the effect found of directi<;mal congruence between an antecedent and consequent. The magnitude of the latter effect, of simple vs. complex topos, in sentence 2 was almost 10 rns per character (475 ms per sentence) and thus four times as large as the simple effect of directionality. Ease of interpreting a consequent seemed here to depend most on whether the scale invoked in it was co-directed with those invoked in the antecedent, and permitted a deduction which was supported by the inference scheme taken to be activated there. Thus, our third hypothesis could also be confirmed. Our hypotheses 1, 2, and 3 were furthermore supported for both Danish and Finnish,12 suggesting that similar deductions are made exploiting inferences based on up - and down -scales in reading highly dissimilar languages. In our view, these findings have considerable theoretical significance, both in that they support the semantic explanation given by the topos framework over

48 Scalar Inferences

a pragmatic one, and in that they lend support to an interrelation suggested in our introduction between the topos explanation and human cognition. 3.1

Topos and linguistic semantics

positively, since an ambiguous NP described as being in a leading position was dominantly taken in the context to refer to X, and the same NP when described in a trailing position was taken as not referring to X. Correspondingly, in sentences of the form X scored only n points , only p was evaluated negatively, consistent with X occupying a losing position rather than a winning one. Further fueling the suspicion that only p may lie cognitively, though not truth-conditionally semantically, closer to not p than to p , antecedent sentences here invoking down (only p )-scales were also read more slowly than quite similar sentences invoking up (almost p )-scales. From research in psycho linguistics (e.g. Clark 1976), it is known that processing of statements contain ing more linguistically marked or negative expressions typically takes longer than for ones containing unmarked or positive expressions. The correlation of up with positive, and down with negative, finds a cognitive explanation, e.g. in Clark ( 1 976), which in part is semantic: and in part perceptual. 3 .2

Topos and cognitive semantics

The semantic part of Clark's explanation can be summarized as follows: vertical adjectives such as tall-short, high-low have a common reference point which is an UPWARD measurement (from the earth's surface), high being of much HEIGHT, low of little HEIGHT. Upward is thus the unmarked direction, and downwards marked as negative, whereby-by derivation-upward becomes positive. Thus, one Inight be tempted to conclude that only p sentences were read more slowly here partly because their meaning was psychologically closer to that of not p than it was to p . We take predicates in our experiments describing leading or

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Tending to corroborate the claims advanced in the semantic linguistic explanation over the pragmatic one is first the fact that use of scalar expressions seems to induce the predicted reasoning schemes independently of language, context, and speech situation. The results obtained here were quite similar for Danish and Finnish, they were quite homogeneous over the set of texts smdied, and they were externally consistent between an explicit and an implicit comprehension task. Secondly, as judged from readers' interpretations as to who seemed to be winning or trailing a given competition, almost p was evidently interpreted as leading to the same or similar conclusions as the positive p, and only p to similar conclusions as the negative not p (see schema 2 above). One can conclude here that almost p in sentences of the form X scored almost n points was evaluated

Lira Lundquist and Robert J. Jarvella

49

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

winning to have been located on up -scales because they are evaluated as being positive, and predicates describing trailing/losing situations to have been located on down -scales because they are evaluated as being negative. We have argued here that the directionality of the scales is also fundamental in the topos concept. Our experimental results on the processing ofalmost/more than p expressions, as compared to only/less than p expressions, are consistent with those reported by Clark & Chase ( I 972) on the understanding of the expressions above and below. As in that research, where sentences that contained above were found to be verified faster than those that contained below, here almost/more thanp sentences were processed faster than only/less than p ones. Clark related these semantic facts and earlier results to certain fundamental human perceptual processes, which may also be dominant in the processing of scalar expressions, namely that space is perceived visually with the ego as the center: 'Normally the visual field . . . consists of the visible space that is in front of the ego and above theground' (Clark I 976: 6o). Therefore 'UPWARDNESS FROM THE GROUND IS POSITIVE . . . and . . . forwardness from the ego is positive' (ibid.: 6o, our emphasis). The convergence between results from Clark's experiments on the visually perceived above and below on the one hand, and ours on the conceptual up and down on the other, is consistent with the notion of directionality as invoked in the topos concept, although the scales studied in our work are ascending or descending in a metaphorical sense only. As stated earlier, the correlation between perceptual and metaphorical structures has been explained in cognitive semantics, as a metaphorical 'source-to-target mapping' based on the 'More is up; less is down' IMAGE SCHEMA. As Johnson (I989: I I6) states more generally, cognitive semantic analysis 'will involve at least two basic kinds of structures: ( I ) biologically-based, image-schematic structures tied to our sensorimotor experience . . . and (2) imaginative . structures-metaphors, metonymie� . . . by means of which we make and extend semantic and epistemic connections'. While cognitive semantics seems to be compatible with the topos-concept as far as the processing of scales is concerned, whether these be perceptual, conceptual or metaphorical, it should be recalled that the topos concept involves the directionality of two scales, rather than one (schema 3 above). In Table 2, we show reading time calculated for our consequent sentences s2. As mentioned earlier, the effect of the directionality of s2 ( I I O ms) is of a much smaller magnitude than that of its co-directionality with s I (4 75 ms). Since the two effects are independent, encountering one down -scale in the wake of another leads to faster reading than encountering an up -scale following a down -scale, i.e. the advantage in making a simpler deduction in s2 is greater than the advantage in processing a sentence in the same context which is oriented upwards. Thus, ups are not always faster to process than downs (c£ Clark I 976).

so Scalar Inferences

What we do not know from studies like the present ones is how immediately the inferences which are activated by scalar expressions in text are taken advantage of, in assigning referents to expressions. Is an ambiguous definite description, for example, disambiguated in a single pass for a simple ropos, but only with back-tracking for a complex topos? This question can best be approached in studies of reading using more on-line measures of language understanding than we have employed in the work reported here, such as those available from eye behavior (see e.g. Paterson 1 993). 3·3

Conclusion

Acknowledgements This research was supported by the Joint Committee of the Nordic Social Science Research Councils (NOS-S), and the Swedish Research Council m the Humanities and Social Sciences (HSFR). We wish to thank Herb Clark, Jukka Hyiina, Phil Johnson-Laird, Serge Larochelle, and Tony Sanford for commenting on this work, and Heli Hujanen and Leila Kalliokoski for their .help in performing it. UTA LUNDQUIST

ROBERT JARVELLA

Copenhagen Business School Lund University S-223 So Lund Dalgas have 15 DK-zooo Frederiksberg Sweden Denmark

Received: 26.02.9 3 Revised version received: 29.09.9 3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Very similar effects of scalarity emerged in the experiments on reading we report here for Danish and Finnish, two structurally very dissimilar languages. The generally high degree of convergence that was found in these two languages suggests that, in language use, scales of the kind we have investigated may play a quite fundamental role and, rather than be confined to the compartment of pragmatics, 13 should be given a privileged place in semantic theory, as suggested both by 'scalar linguists' such as Anscombre and Ducrot, and by cognitive semanticists such as Lakoff and Johnson. Whereas cognitive semanticists have suggested that image schemata such as scales are a natural, i.e. perceptual result of our 'embodiment', and that such fundamental image schemata extend by a metaphorical process to more abstract domains, scalar linguists have claimed, by linguistic reasoning, 14 that scalariry is inherent in a great many linguistic expressions, i.e. not only in adverbs but also in nouns and predicates. If it is the case, as postulated in these two semantic theories, that scales are involved in interpreting not only some, but rather most, and perhaps even all linguistic expressions, then scalar inferences of the types studied here play a more extended role in language understanding than has been thought hitherto.

Lira Lundquist and Robert J. Jarvella 5 I

APPE N D I X The Danish and Finnish versions of the beauty competition text Missikisat

Skonhedskonkurrencen Arets store sk0nhedskonkurrence fandt sted i Cirkusbygningen i K0benhavn. Der stil ledei 8 modeller op. valgt i forskellige danske

favoritter.

Typhoon-hallissa. I 8 kilpailijaa oli selviy tynyt mukaan ympari Suomea jarjestetyista alkukarsinnoista. Kuten lehdisto ja muu mediavaki oli ennustanut, erottuivat Kaarina Maki ja Liisa Niemi suurina vmttajasuosik keina.

sr I runden med badedragter fik Karina Madsen toppoints hos kun Jii aj7na?sten aile dommerne.

s i : Uimapukukierroksella Kaarina Maki sai

s2 Den h0je blonde Arhus-pige var klart

s2: Pitkanhuiskea helsinkilaistytto oli selvasti

huippupisteitii vain muutamalta tuomarilta/ huippupisteet liihes kaikilta tuomareilta . jiiljessiiledellii.

bagud/forud. s3 I finalen med aftenkjoler scorede Louise Fernholm points hos over/under halvdelen af dommerne. s4 Den 2o-arige kontorpige l0b af med sejren. Efter finalen modtog aile pigerne en gave check og en tur til Mallorca. Det var BT der

s3: Kisat paattaneella iltapukukierroksella Liisa Niemi sai pisteita yli!alle puolelta tuomaristoa. s4: 20-vuotias konttoristi vei voiton kotiin. Finaalin jalkeen jokainen tytto sai palkkioksi lahjashekin ja Mallorcan matkan. Missikisat oli kustantanut Mainostelevisio.

havde sponsoreret sk0nhedskonkurrencen.

N O TES I Reasoning using everyday knowledge has also been studied in psychology (see e.g. Johnson-Laird, Legrenzi & Legrenzi

I 972; Cheng & Holyoak I98 5). For a review of psychological work concerned with the description of quantities, see Moxey & Sanford ( I 993)-

2 In our presentation, we have omitted the first, 'argumentative' phase of Ducrot & Anscombre's theory of'argumentation in language' (Anscombre & Ducrot I 98 3 ) and have focused o n the rwo subsequent phases in which topos and scalarity have replaced the argumentative explanation. In their most recent works, scalarity is

extended to pertain to all kind oflinguis tic material, c£ 'Ia langue est fondamen talement scalaire . . . La signification d'une phrase est !'ensemble des topol qu'elle autorise a appliquer des lors qu'elle est enoncee' (Anscombre I 99 I : I 39); '. . . !'intention generate est . . . de loger Ia gradualite dans Ia signification meme des predicats' (Ducrot, to appear). 'Whenever we add more of a substance say, water to a glass-the level goes up . . . Remove . . . water from the glass, and the level goes down' (Lakoff I 987: 276). Lakoff's explanation of 'more is up; less is down' in terms of perceived gravity (I 987:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

provinsbyer. Som journalister og andre mediefolk havde spaet, skilte Karina Madsen og Louise Fernholm sig ud som de store

Taman vuoden missikisat jarjestettiin Turon

52 Scalar Inferences 276) seems somewhat problematic, how ever. In naive physics, the bigger and heavier something is, the faster it will fall. 4 This has been demonstrated in a series of later experiments conducted in French

s

(Jarvella & Lundquist, in press). That is, we wish to add a fourth type of inference to ones already described in psycholinguisrics (instrumental, elabora tive, and causal-see Rayner & Pollatsek

1989: 2 7 8-84): the scalar inference. 6 In Danish, these noun phrases also began with a definite article. Finnish does not

7 The semantics of proper names and definite noun phrases is not itself at stake here. Suffice it to say that whereas proper names are 'rigid designators' (Kripke 1 972), definite noun phrases have a full semantic content which may or may not apply to an individual previously men tioned.

1 1 It can be seen from Figure 2 that consequent sentences s2 and s4 were also read at a slower rate than sentences SI and SJ. The reasoning process whereby a reader actually uses scalar inferences made available by an antecedent sentence

to disambiguate a definite NP in the text should have its locus in s2 and s4. We take up this process in further detail in

Lundquist,Jarvella, & Hyona ( 1 99 3). 12 There was a modest advantage in favor of Finnish, both in proportion of inferences made as predicted and in speed of reading. Jarvella & Hyona (1 993) found thar Finnish readers also appear more flexible when returning to old thematic subjects in discourse. The two kinds of result may be related and reflect the fact that, being a language with a rich case

8 We are aware of the fact that this is not

marking system, Finnish allows an other

the case in all sports, e.g. in gol£ Similarly, in situations where the measure of performance is rime needed or errors made, a lower numerical value is evalua

wise more context-free interpretation of text.

ted positively. This issue will be dealt with in a forthcoming paper. 9 Lundquist (1 987) gives more substantial arguments for the simple-complex topos distinction, based partly on the main tenance of a coherent discourse represen tation. 10 ANOVAs were conducted with four

1 3 Indeed, the results show that, in the interpretation of scalar adverbs, some thing more fundamental is at stake than reasoning via conversational maxims of the type As Speaker says X and not Y, then

Speaker must mean Z.

14 Ducrot (to appear) explains the 'gram maticaliry' of II y a un probleme, meme dijficile as contrasted to the 'ungrammati caliry' of •II y a un probleme, memefacile by

orthogonal factors in each language: sentence pair (s 1 -2 vs. s3-4), relative position within sentence pair (first vs.

the fact that there is a scale (of difficulty) inherent in the noun probleme shown by its consistency with the intensifier meme

second), version of text read (of four studied), and position ofeach version read

difficile. facile .

and non-consistency with

meme

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

have articles, and definiteness of noun phrases there was ensured by their sentence-initial position and by using modifiers.

in the list of texts (first vs. second half of experiment). Though the number of different competition scenarios is limited, analyses done using texts as a random factor supported the same conclusions as those for subjects.

Lita Lundquist and Robert J. Jarvella s 3

RE FERE N CE S Anscombre, J.-C. ( I 989), 'Theorie de l'argu mentation, topoi et structuration argu

Revue de linguistique queoicoise,

Inquiry, vol. 9, nos. I-2-3, Montreal, Universite du Quebec a Montreal. Johnson-Laird, P. N., P. Legrenzi, & M. Legrenzi ( I 972), 'Reasoning and a sense of reality', British journal of Psychology, 63,

de Cerisy: L'Argumentation , Mardaga, Liege. Anscombre, J.-C. & 0. Ducrot ( I 983 ), L'A rgumentation dans Ia langue, Mardaga,

Kripke, S. ( I 972), 'Naming and necessity', in D. Davidson & G. Harman (eds.), Semantics of Natural Language , Reidel, Dordrecht,

Liege. Cheng, P. W. & K.J. Holyoak ( I98s ), 'Prag matic reasoning schemas', Cognitive Psy

Lakoff, G.

mentative', 18,

Universite du Quebec a Montreal. Anscombre,J.-C. ( I99 I ), 'Dynamique du sens et scalarite', in A. Lempereur (ed.), Colloque 1.

Cognitive Development and the Acquisition of LAnguage, Academic Press, New York. Clark, H. H. ( I 976), Semantics and Comprehen sion , Mouton, The Hague. Clark, H. H. & W. G. Chase ( I 972), 'On the process of comparing sentences against

pictures', Cognitive Psychology, 3, 472-5 I 7. Ducrot, 0. ( I 988 ), 'Topol et formes topiques',

Bulletin d'Etudes de Linguistique Jranraise,

no. 2 2, Tokyo. Ducrot, 0. (to appear), 'Les modificateurs derealisants', journa/ ofPragmatics .

Horn, L. R ( I 989), A Natural History ofNega tion , University of Chicago Press, Chicago. Jarvella, R J. & J. Hyi:ina ( I 993 ), 'Some effects of information flow in discourse on on line processing during reading', in L. Ahrenberg (ed.), Papersfrom the Jrd Nordic

Conference on Text Comprehension in Man and Machine, Linki:iping University, Lin ki:iping.

Jarvella, R J. & L. Lundquist (in press), 'Scales in the interpretation of words, sentences, and texts', .in A. J. Sanford (ed.), Special Issue on Psychological Semantics,journal of

Semantics.

Johnson, M ( I 987) ,

The Body in the Mind: The Bodily Basis of Meaning, Imagination and Reason , University of Chicago Press,

Chicago. Johnson, M

( I 989), 'Image-Schematic Bases Recherches simiotiques/Semiotic

of Meaning',

2 S 3 -3 S S·

( I 987), Women, Fire, and Dangerous Things: What Categories Reveal about the Mind, University of Chicago Press, Chi

cago.

Lundquist, L. ( I 987), 'Programme argumen tatif et desambigulsation referentielle', Revue Romane, 22, I6 3-8 I, Copenhagen.

Lundquist, L. ( I 990), 'Argumentation: semantics or pragmatics?', in L. Lundquist & L. Schack-Rasmussen (eds.), Pragmatics

and Its Manifestations in Language , Copen

I 3, Arnold Busck, Nyt Nordisk Forlag, Copenhagen. Lundquist, L., R. J.Jarvella, &J. Hyi:ina ( I99J ),

hagen Studies in Language, No.

'Topic, topos, and mental models', manu script submitted for publication.

( I 99 3 ), Com municating Quantities: A Psychological Per spective, Erlbaum, London. Paterson, K. ( I 993 ), 'The immediacy of Moxey, L. M. & A. J. Sanford

situation-based inferences in quantified texts', Symposium on Psychological Semantics, International Conference on the Psychology of Language and Com munication, Glasgow. Rayner, K. & A. Pollatsek ( I 989), The Psychol

ogy of Reading , Prentice-Hall, Englewood

Cliffs, NJ. Sadock,J. M. ( I 98 I ), 'Almost', in P. Cole (ed.),

Radical Pragmatics, Academic Press, New York. Shepard, R N. & S. Hurwitz ( I 985 ), 'Upward direction, mental rotation, and discrimi nation ofleft and right turns in maps', in S. Pinker (ed.), Visual Cognition , MIT Press, Cambridge, Mass.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

chology, 17, 39I -4I6.

Clark, H. H. ( I973 ), 'Space, time, semantics and the child', in T. E. Moore (ed.),

395-400.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

journal ofSmumtics

I I:

S S-82

© Oxford University Press 1994

Anaphora, Agreement, and Right Dislocations in Greek MARIA VAL I O ULI Aristotle University, Thessaloniki Abstract

G REEK RI GHT D I S L O C AT I O N S A N D T H E I R S PE C I F I C A T I O N S Greek right-dislocated constructions are employed for a number of distinct communicative purposes. Thus, the term Greek right dislocations does not necessarily apply to mechanisms of supplying some kind of afterthought information, but, most often, to special constructions, which are deliberately employed by the speaker for reasons that have to do with the pragmatic impact of the intended message. For example, Greek RDs can often be used when speakers want to project their empathy towards the referent of the dislocated NP in a casual, tacit mode.1 Before moving on to a discussion of the communicative functions of Greek right dislocations, however, I would like to specify the constructions in question firstly because Modem Greek is a relatively 'unknown' language, and secondly because right dislocations are largely considered to be instances of 'fancy' syntax (see Prince 1 98 5). The Greek right dislocations analysed in this paper have roughly the following structural analysis, which is a mirror image of the structural analysis of left dislocations as suggested by Prince (1 985).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

First, Greek right dislocations are specified and distinguished from other rightmost full fledged NP occurrences. Then they are classified into six categories, which reflect six differentiated intentions on the part of the speaker with respect to the pragmatic impact of his utterance in the discourse context. Finally, they are analysed in terms of the anaphoric relations and agreement possibilities that hold beween the preceding anaphor and the right-dislocated expression. It is argued that all right-dislocated constructions in Greek are used for an unemphatic projection of the thematic information of the anaphoric clause, and for a simultaneous unemphatic projection of the speaker's referential or attitudinal perspective. It is also suggested that the various concord possibilities between the anaphorically related elements reflect various referential or attitudinal differentiations towards the referent of the dislocated element. Finally, it is claimed that the different conditions on the speaker's 'freedom' to express through concord or non-concord his referential or attitudinal differentiations depend on the primary pragmatic function of the particular category of right dislocation used.

Agreement and Greek RDs

56

X 1 is either an object clitic pronoun, or the lexically unspecified subject that is included in the Greek finite verb suffix. X2 is non-vocative and a full-fledged

NP.

More specifically, X2 consists mainly of an article, most often a definite

one, and a proper name, which, quite usually, is a first name. It may also consist of an article and an epithet, or of an article, a noun denoting real life relations or titles or occupations, and a clitic possessive pronoun. X 1 and X2 are in an anaphoric relation. X2, however, is, quite often, a semi-autonomous, i.e. optional, co-referring expression that does not function as a reference maintenance element, as it refers to presupposed, i.e. active or latent, discourse

)

)

Examples ( 1 and (2 below are instances of Greek right dislocations. The expressions in small capitals (as in every example in this paper) receive the intonation peak of the utterance, unless stated otherwise. Note also that I have chosen not to use RDs for their English translations, as the outcome not only would be odd English, but, most probably, it would fail to recreate the 'swiftness' of the Greek RDs. (I)

Ta

VRJKA,

ta klidia .

[Them 1 found-1] - anaphoric clause [the keyst] = dislocated expression [s [s · [x 1 ta(=clitic pers. pro., neue., ace. case, pl.) x ] vrika(=verb, aorist, + I st 1 ps. sg. su£)5·] [x2 ta(=de£ art., neue., ace. case, pl.) klidia(=noun, neue., ace. case, pl.)xJ ] s I (have) found the keys.

(2) EFIGe, o adelfos mou .

[Left-he] = anaphoric clause [my brother] = dislocated expression

[s [s · efig-(-verb, aorist)+ [x 1 -e(- Jrd ps. sg. su£)x 1J s .] [x2 o(=de£ art., masc., nom. case, sg.)adelfos(=noun, masc., nom. case, sg.) mou(=clitic poss. pro., I st ps. sg.)x2]5] My brother (has) left. Apart from the specifications just stated, Greek right dislocations are characterized by canonical stress assignment in that there is always a clausal element-most often the last one-which is characterized by an intonation peak, i.e. it is the intonation centre, whereas x2 is unstressed (i.e. it is characterized by flat intonation). The intonation peak is placed on that clausal element which conveys information directly related to the communicative function of the particular category of RD used. If, for example, an RD is employed for the retopicalization of an issue already discussed, the intonation peak is placed on that element which, information-wise, constitutes the most

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

referents. S contains S' plus the dislocated NP, and S' is either a simple sentence or a clause.

Maria Valiouli 5 7

recent development of the issue that the speaker wishes to topicalize again. Moreover, a comma between S' and X2 quite often symbolizes a slight pause on the part of the speaker, who, especially when the intonation peak is placed on the last clausal element, has to stop a little in order to go on with unstressed (flat) X2• Lambrecht (I 98 I : 86) shows schematically the intonation contour of a typical New Standard French right-dislocated construction as follows:

Ils sont fous, ces Romains. Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The same schema can depict the intonation contour of a typical modern Greek RD, when the intonation peak is placed on the last element of the anaphoric clause. Furthermore, the last clausal element, whether it is stressed or not, is always characterized by non-falling or rather non-final pitch. The quality of the pause and the non-final pitch of the last clausal element indicate that such pauses should rather be interpreted, in Chafe's ( r 988) terms, as 'comma intonation' rather than as 'period intonation'. To be more precise, it seems that Greek right dislocated constructions, if we concentrate on S' plus X2, are what Chafe ( I 987, 1988) calls 'intonation units', which may end in non-final pitch. The non-final pitch functions as an instruction to the addressee to accept what follows, i.e. X2, as a somehow differentiated (as far as the speaker's perspective is concerned) intonation unit, which, however, is closely related to the previous one, i.e. S', which has not been closed. The intonation contour of Greek RDs, whether they are used discourse initially or not, and especially the requirement of flat intonation for the right most NP, which indicates presupposed information, i.e. information that can be omitted, is directly related to their contextual requirements. Greek RDs are not employed in emphasis- or contrast-inviting contexts. Example (I), for instance, is unacceptable if the NP TA KUDlA receives contrastive of emphatic stress. This constraint on its stress, of course, indicates that the full NP TA KUDlA refers to background information, as Joseph & Philippaki-Warburton ( I 987: 246) maintain for similar Greek examples. Example (2) is acceptable with an emphbatic or contrastive stress on the NP o ade/fos mou only if it is an answer to questions like Who has left?, What has happened?, What's up?, Has your sister left? in other words, to questions that invite answers in which the fully specified NP o adelfos mou would provide new information. (2) is unacceptable with this stress pattern, however, if it is a response to questions like How's your brother?, Where is your brother?, Is your brother here?, When will your brother leave?, Why hasn't your brother left yet?, Will your brother be stayingfor long?, What hasyour brother

58 Agreement and Greek RDs

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

finally decided to do? and, in fact, to any question that would presuppose that the referent in question, or the issue in question, or both are, or should be, somehow present in the interlocutors' current consciousness. An acceptable and felicitous answer to the second set of questions would presuppose either a different intonation contour, i.e. that of RDs, where the rightmost NP must be unstressed (see also Joseph & Philippaki-Warburton 1987: 109), or the verb e.fige on its own, something impossible with the first set of questions mentioned above. To be more precise, a question like Who has left? can be followed by the response 0 adelfos mou. , 0 adelfos mou e.fige. and Efige o adelfos mou. (with an emphatic stress on the elements char are in bold letters), whereas it cannot be followed by the answers Efige. , Efigan. , and Efigan, o ade/fos mou. (again with an emphatic stress on the bold-lettered elements). On the other hand, with a question like Where's your brother?, the situation is reversed: the latter three responses are acceptable and felicitous, whereas the first three responses are not. It should be noted also that the answer Efigan. (�verb, aorist, + 3rd ps. pl. su£) is not only acceptable as a response to the second question, but the 3rd person plural sufftx triggers an inferred extended reference to the active referent's family, as metonymy is quite usual in Modem Greek Thus, despite the fact that Greek is a free word order language, the rightmost Nom. case NPs we are dealing with in this paper are not postponed subjects conveying redundant information. Moreover, our observations also indicate that, in the case of Nom. case NP RDs, the Greek finite verb suffix (which always carries information as to the person and number of its subject, which is always understood to be in the nominative case), functions as an anaphor since (a) the rightmost NP can be omitted and (b) the verb sufftx can signal inferred extended reference. In the case of stressed Nom. case NPs, however, which cannot be omitted, the finite verb sufftx simply marks grammatical agreement with its subject. It rhus becomes obvious that the rightmost NPs in Greek RDs, whether in the nominative, accusative, or genitive case, are used for pragmatic reasons-the basic one being the speaker's need to empathize with their referents-that is, for reasons which have to do with the emotive meaning of the sentences they are part o£ This is true not only for simple-sentence RDs, but also for subordination, especially in those cases of subordination which would better be interpreted as instances of enhancing hypotaxis (as this term is analysed in Matthiessen & Thompson 1 988). With enhancing hypotaxis, which, according to Matthiessen & Thompson ( 1 988: 28 3-4), covers clause combining that involves 'some kind of circumstantial relation like condition, reason, purpose, and other kinds of cause, time, space, manner, and means', Greek RDs are not always obligatory and if they are preferred over SV occurrences, they reveal differentiated communicative, even rhetorical options on the part of the speaker:

Maria Valiouli 59

The speaker's surprise or contempt is realized in a casual way in (3b), whereas in (3a) Speaker B succeeds in projecting his feelings in a more emphatic, more marked way, despite the fact that the full NP is in the subordinate clause. We can have several occurrences similar to (3a) above in enhancing hypotaxis, as long as the pre-verbal Nom. case NP and the preceding anaphor are related by some kind of 'cause-effect' semantic link-which is determined by both general and specific background knowledge that helps the addressee compute the intended co-:referential reading. Yet there are cases in hypotaxis, which Matthiessen & Thompson (1988) would term as embedding, where RDs seem to be obligatory if the speaker wants to make use of a full-fledged Nom. case NP, as, for example, in reported speech. Nevertheless, even then we can have alternative options, rarely though, depending on the speaker's intent. Consider, for example, (4a) and (4b) below, which are both reported versions of Valiouli's self-referring utterance 'Mia Valiouli de lei POTE psemata.' (=A woman called Valiouli never tells lies-).3 (4) Speaker A:. Ti ipe i Valiouli; What did Valiouli say? (a) Speaker B: lpe oti mia Valiouli de lei POTE psemata. [Said-she1 that [one Valiouli1 not tell-she1 never lies]] [s ip-(=verb, aorist)+-e(=3rd ps. sg. su£) [s · oti (=complementiser) mia(=inde£ art., fern., nom. case, sg.) Valiouli(=proper name, i.e. surname,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 3) Speaker A:. Kale giati piasane tin pethera sou; Why did they arrest your mother-in-law? (a) Speaker B: Tin piasane giati i petheroula mou itane vAPoRAKI! [Her1 caught-they] [why little-mother-in-law-my1 was little-boat!] [s tin(=clitic pers. pro., fern., ace. case, sg.) piasane(=verb, aorist, + 3rd ps. pl. su£) [s · giati(=cause adverb clause conj.) i(=de£ art., fern., nom. case, sg.) petheroula(=noun, fern., nom. case, sg., +dimin. su£) mou(=clitic poss. pro., 1st ps.) itane(=verb, i.e. 'to be', past, + 3rd ps. sg. su£) vaporaki(=noun, +dim. su£)s ·] s] They arrested her1 because my mother-in-law1 was a drug trafficker.2 (b) Speaker B: Tin piasane giati itane VAPORAKI, i petheroula mou! They caught her1 [because was-she1 little-boat]=anaphoric clause [little mother-in-law my1]=dislocated expression. [s tin(=clitic pers. pro., fern., ace. case, sg.) piasane(=verb, aorist, +3rd ps. pl. su£) [s · giati(=cause adverb clause conj.)[x 1 itane(=verb, i.e. 'to be', past, +3rd ps. sg. su£) x 1] vaporaki(=noun+dim. su£)s ·Hx2 i(=de£ art., fern., nom. case, sg.) petheroula(=noun, fern., nom. case, sg., +dimin. su£) mou(=clitic poss. pro., 1 st ps.)x2] s] They arrested my mother-in-law because she was a drug trafficker!

6o Agreement and Greek RDs

fem, nom. case, sg.) den(=negative particle) le-(=verb, present)+-i(=3rd ps. sg. su£) pote(-neg. adverb). psemata(=noun, pl.) s·] s She said that no woman of the name Valiouli ever tells lies. (b) Speaker B: lpe oti den lei POTE psemata mia Valiouli. Said-she1 that [not tell-she1 never lies]-anaphoric clause [one Valiouli1] -dislocated expression [s ip-(-verb, aorist)+-e(=3rd ps. sg. su£)[s· oti (=complimentiser) den(-negative particle) le-(-verb, present)+[x 1 -i(=3rd ps. sg. su£)x1] pote(=neg. adverb) psemata(=noun, pl.)s ·Hx2 mia(=inde£ art., fem, nom. case, sg.) Valiouli(=proper name, i.e. surname, fern., nom. case, sg.)xJs] She said that no woman bearing the surname ofValiouli ever tells lies.

I

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(4b) is a casual description ofValiouli's feelings, whereas by employing (4a) the Greek speaker depicts Valiouli's sense of self-pride in a more marked, emphatic way. Example (4b) leads us to two further properties of Greek RDs. Firstly, on the basis of my data-which, naturally, are far more numerous than the examples analysed in this paper-Greek right-dislocated NPs are, for the most part, simple sentence- or clause-attracted elements rather than compound sentence phenomena. Of course, we do have right-dislocated expressions in compound sentences, but their most usual place is next to an anaphoric clause, wherever that might be, rather than at the rightmost end of the sentence. This should not be interpreted as an attempt to evade a more detailed analysis of subordination. Lambrecht (I 98 I : 8 ) maintains that French RDs, which he calls antitopics , are characterized by a similar behaviour. However, although Lambrecht ( I 987: 2 34) suggests a syntactic constraint, namely Ross's 'Right Roof Constraint'-according to which a right-dislocated element must appear immediately after the clause it is anaphorically related to and not after the intervention of some clause higher in the tree structure-it seems that this restriction is closely related to the importance of clauses and simple sentences in oral verbal communication. The importance of the clause as a linguistic phenomenon in contradistinction to that of the sentence has increased considerably in the recent years. For example, Matthiessen & Thompson (I 988: 287) refer to clauses as basic rhetorical units, and Pawley & Syder (1 983: S73) argue that the clause is the basic syntactic unit in spontaneous spoken discourse and even suggest (I 98 3: )64-5) the 'one-clause-at-a-time' constraint for spontaneous connected discourse. Finally, Chafe ( 1 987: 40 f£) maintains that clauses are constrained and determined by basic cognitive phenomena such as memory and consciousness, whereas sentences are independent of cognitive limitations and seem to be determined by factors such as rhetorical effect or control. Example (4b), however, is related also to the question of the definiteness of

Maria Valiouli 6 1

the dislocated NP. Although the overwhelming majority of Greek RDs are characterized by the definiteness of their dislocated NPs, definiteness it not an absolute requirement. Indefiniteness in Modem Greek can be marked either by the numeral one , by the absence of the article, or by a form of the indefinite pronoun used adjecrively. The first way can involve specified indefiniteness , as we read in Kazazis & Pentheroudakis (1 975). It seems that when there is specified indefi niteness involved, we can have indefinite right-dislocated NPs as in (5) below, which is an example taken from Kazazis & Pentheroudakis (1975), who, how ever, speak of object reduplication.'�

(5) would be an appropriate response to offers like How about some whisky?, How about a drink? , or What can we offer you?. In fact, Kazazis & Pentheroudakis claim that (5) can be uttered even if the speaker has not been offered anything, provided that his relation with the host is very friendly. The specificity of the indefinite NP ena whiskaki lies in the fact that it is part of a (explicitly realized or latent) discourse topic. Thus, the NP ena whiskaki is either a repetition or part of the current discourse topic if(5) is a response to the first offer, or a hyponym of the superordinate terms Drinks or Offers if it is a response to the second and third offers, respectively, even if these offers are not realized verbally.5 It should be noted that if the peaker employed a right dislocation with a definite NP, i.e. to whiskaki (=the whisky+dim. suf), that would presuppose a glass of whisky already filled and offered to the speaker, and, most probably, a different co-text such as Why don't you drink your whisky?, in which case the tense of the verb pino (-drink-1) would not correspond to a future tense but to a present tense with a progressive aspect.6 Before moving on to the analysis of Greek RDs in relation to the speaker's intentions, to anaphora, and to agreement, we should have a brief look at their contexts of occurrence, this rime with respect to their emotive meaning. As we have already seen, Greek right dislocations are optional constructions. Thus, although the examples analysed in this paper are RDs, it is obvious for those who know Greek that in the particular contexts of their occurrence, we could have alternative responses?

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(5) To pino EFHARISTOS ena whiskaki. [It1 drink-1 with-pleasure]=anaphoric clause [one whiskakii]=dislocated expresston [s [s · [x 1 to(=cliric pers. pro., neut., ace. case, sg.)x 1] pino(=verb, present with a future meaning, + 1 st ps. sg. suf) efharistos(=adverb of manner) s ·Hx2 ena(=indef art., neut., ace. case, sg.) whiskaki(=noun, neut., ace. case, sg., +dimin. suf)x2] s] I will gladly have a (glass oD whisky.

62

Agreement and Greek RDs

GREEK RI GHT D I SLOCAT I O N S A N D THE S PE AKER'S I NTEN T I O N S I n what follows, I will be using different terms for the various categories of right-dislocated constructions I propose, depending on their communicative functions, which will be presented rather briefly. I hope that these terrns will better reflect the speaker's intentions, which lie behind their use.8 Needless to say, I do not necessarily claim that the suggested categories exhaust the list. Greek right dislocations are primarily used for the projection of the speaker's {positive or negative) emotional attachment to the referent of the dislocated expression. To be more precise, the speaker may use a right dislocation simply because he feels the need to repeat the name, the title/occupation, or the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

For instance, instead of {6) below-which is one of the slogans about Menios Koutsoyiorgas, heard during his funeral, (Eleflerotypia , 2o/4f'91)-we could have, in theory, {6a) or {6b). {M. Koutsoyiorgas, a late Greek politician, had been prosecuted for embezzlement and, while defending himself in court during a trial that was shown live on the Greek national TV network, had a stroke and died a few days later.) If the crowd that cried out the slogan shown in {6) had omitted the right-dislocated NP, as in {6a), the clitic pronoun Ton (=him) would still have its specific referential value. However, their emotional attachment to Koutsoyiorgas would not be so obvious. Alternatively, if the clitic pronoun had been omitted, as in (6b), the utterance would be absurd, at best, as it would sound as an announcement of Menios Koutsoyiorgas' death during his own funeral! In other words, by omitting the right-dislocated NP , we fail to capture the emotive meaning of the particular sentence, whereas by omitting the elitic, we wrongly undertone the thematic continuity of its main referent and/or of the issue described. (6) Ton PETHANATE to Menio [Him1 died-you]-anaphoric clause [the Menio1]=dislocated expression [s[s ·[x 1 ton(=clitic pers. pro., masc., ace. case, sg.)x 1] pethanate{=trans. verb, aorist, +2nd ps. pl. su£)s·Hx2 [ ton(=de£ art., masc., ace. case, sg.) Menio(=proper name, i.e. first name, masc., ace. case, sg.)xJs] You caused Menios' death. (6a) Ton PETHANATE. Him(clitic) made-you die You caused his death. (6b) PETHANATE to Menio. Made-you die the Menio You caused Menios' death.

Maria Valiouli

63

(7)

de me ponai emena . [Nobody not me1 feels-pain]=anaphoric clause [me1]=dislocated expression [s [s· kanis(=inde£ pro., masc., nom.case, sg) de(=neg. particle) [x 1 me(=cli tic pers.pro., ace. case, Ist ps. sg) x1] ponai(=trans.verb, present, +1st ps sg. su£) s·Hx2 emena(=indep.pers. pro., ace.case, 1st ps.sg.)x ] s] 2 Nobody sympathizes with me, at all. KANIS

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

expression denoting his real life relation to the (already active) referent of the dislocated NP, and, through that repetition, project (unemphatically) his emotions towards the referent. The employment of proper names for subsequent mentions as a means of pushing the characters they refer to into a central role is also argued in Sanford, Moar, & Garrod (1988). Since there is some kind of re-naming involved, I have chosen the term Re namers for this category of right dislocations. Re-namers, just like all Greek RDs, are characterized by canonical stress assignment in that there is always an element, most often the last clausal one, in the anaphoric clause which receives the emphatic primary stress or nuclear stress of the utterance (i.e. intonation peak), whereas the rightmost NP must be unstressed. The element which receives the emphatic primary stress of the utterance is that linguistic expression which, according to the speaker, justifies, as it were, the main 'characterhood' of the referent of the dislocated NP. Example (6), which was analysed above, is an instance of a re-namer. However, it must be pointed out that (6) was uttered in a highly emotional context, and that is why the suggested alternatives could be characterized even as inappropriate.Under normal circumstances, a right-dislocated construction functioning as a re-namer is just one of the various contextually appropriate alternatives at the speaker's disposal.This last observation is not a contradiction to my analysis of either example (6) in particular or of Greek RDs in general, as empathy can be manifested in varying degrees. There is another category of right dislocations, which involves the use of the independent (or free) personal pronoun in the rightmost position. The independent personal pronoun implies a non-contrastive, non-emphatically used meaning of 'especially [the referent of the pronoun]', and its use assigns its referent some kind of pragmatic prominence.However, this prominence is, in a way, transferred to the force of the speech act of the predication (i.e. of the anaphoric clause). Thus, such right dislocations result in an intensification of the illocutionary force of the speech act of the predication, and that is why I have called them Intensifiers . Intensifiers are also characterized by canonical stress assignment, and in their case the emphatic primary stress of the utterance falls on that element of the anaphoric clause that 'co-justifies', as it were, the uniqueness of the speech act of the predication.

64 Agreement and Greek RDs

(8) Speaker : Tin PIRA, tifo usta . [It1 took-I)=anaphoric clause [the skirr1)=dislocated expression [s [s - lx 1 tin(=ditic pers. pro., fern., ace. case, sg.)x 1 ] pira(=verb, aorist, + 1 st ps. sg. suf.)s·Hx2 ti(=def. art., fern., ace. case, sg.)fousta(=noun, fern., ace. case, sg.)x 2] s ] I('ve) bought the skirt. Addressee: Kala ekanes. Yia valti na se do. Good. Why don't you put it on. I'd like to see you (in it). The speaker utters (8) as soon as she comes home from shopping. There has been a previous relevant discussion between the speaker and the addressee about the speaker's intention to buy a particular skirt. Thus the dislocation in (8) serves as a discourse topic re-establishing mechanism. Moreover, it should be noted that (8) and the overwhelming majority of Stressed-Focus antitopics are employed in contexts which guarantee, as it were, the latency of the re topicalized issue. For example, the speaker in (8) was holding the wrapped-up skirr in her hands and had a big smile on her face.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Example (7) can be a statement-fact or a statement-complaint depending on the colouring in the speaker's voice. Whether fact or complaint, however, in (7) the dislocated pronoun contributes to an unemphatic singling out of the referent and thereby to the intensification of the intended impact of the speech act of the predication. If, for example, we omitted the independent pronoun in (7), I would suggest Nobody sympathizes with me as its English translation. Greek right-dislocated constructions can also be used for the (re-)establishment of an issue about the referent of the dislocated NP, i.e. as Antitopics .9 To put it differently, they can be used as discourse topic re establishing or shifting mechanisms, i.e. as presentational constructions whose function is the introduction and establishment as the current discourse topic of an already discussed or taken-for-granted issue about the referent of the dislocated expression. As with all right-dislocated constructions, there is always a focal element in the predication, which receives the emphatic primary stress of the utterance and, which, information-wise, constitutes the recent development or exemplification of the issue already known or the default focus of the issue that is taken-for-granted. Since, however, both types of the presentational right-dislocated constructions differ in the amount and quality of the presupposed knowledge and concern on the part of the addressee, I shall be calling those that presuppose prior specific knowledge (and probably concern) on the parr of the addressee, Stressed-Focus antitopics, and those that do not presuppose any such knowledge on the parr of the addressee Informative-Presupposition antitopics.10 Examples (8) and (9) below are instances of Stressed-Focus and Informative-Presupposition antitopics, respectively:

Maria Valiouli 65

However, the addressee, who responds to the speaker's dislocation in (9) below, is totally ignorant of the issue of the predication. He is only aware of the fact of his cousin's existence. In other words, the speaker's dislocation in (9) is employed for the introduction of a totally new issue relevant to a mutually known referent as the suggested current discourse topic. Therefore, despite the difference in the amount and quality of the information shared between the speaker and the addressee, both (8) and (9) serve as mechanisms for the introduction of issues rather than of referents into the discourse model.

Both (8) and (9) illustrate a casual, tacit way of introducing a subject for discussion, and this is achieved through the pre-posing of the anaphoric clause, which implies presupposition, i.e. one of the strategies employed for the creation of politeness, as we read in Brown & Levinson ( 1 987: 2 1 7). The speaker may also use a right-dislocated construction when he wants to editorialize the information conveyed in the anaphoric clause (or even in the preceding context). The term I shall be using for those right dislocations which denote the speaker's judgement with respect to the reaction or behaviour of the 11 referent of the dislocated element is Tails . The addressee's response in (9) above is an example of a tail construction, through which the speaker (i.e. the addressee in the particular example) expresses his personal evaluation of the referred to entity with respect to what has been described in the preceding context. Naturally, the right-dislocated elements in the case of tails are epithets, which function anaphorically-connotatively (Bosch 1 983: 1 49), or even proper names which may also function anaphorically-connotatively (Marmaridou

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(9) Speaker: Ta mathes; Pire kenourgio AFIOKINITO , i ksade!fi sou. (Them leamed-you?] (Took-she1 new car)=anaphoric clause [the cousin your1) =dislocated expression. [ta(=clitic pers. pro., neut., ace. case, sg) mathes(=verb, aorist, +2nd ps. sg. su£)) [s [s · pir-(=verb, aorist)+[x 1 -e(= 3rd ps. sg. su£)x1 ] kenourgio(=adject., neut., ace. case, sg.) aftokinito(=noun, neut., ace. case, sg.)s ·Hx i(=de£ art., t fern., nom. case, sg.) ksadelfi(=noun, fern., nom case, sg.) sou�=elitic poss. pro., 2nd ps. sg.)xJ s] Have you heard the news? Your cousin (has) bought a new car. Addressee: Pali; POU vriski ta lefta to bastardo ; [Again?] (Where find-she1 the money)=anaphoric clause [the bastard1 ?)=dislocated expression [pali;(=adverb)] [s [s · pou(=adverb) vrisk-(=verb, present)+[x1 -i(=3rd ps. sg. su£)x1 ) ta(=de£ art., neut., ace. case, pl.) lefta(=noun, neut., ace. case, pl.)s ·] [xf to( =de£ art., neut., ace. case, sg.) bastardo(=epithet, neut., ace. case, sg.)x2) s Again? Where on earth does she fmd the money, the bastard?

66 Agreement and Greek RDs

1 989). Needless to say that tails are also characterized by canonical stress assignment. Notice that in (9) the elements which receives the emphatic primary stress of the utterance, i.e. POU (=where) is the one which, according to the speaker, justifies his evaluation most. Right dislocations are often employed as disjunctive mechanisms for the disambiguation of the referent of the anaphor in the anaphoric clause in cases where there is more than one active referent involved. The term I shall be using 12 for this category of right dislocations is Local Topic Reinstaters .

( 1 o) was uttered by Mr Tegopoulos-witness at a famous trial, which was shown live on Greek TV (ET1, 24/7/9 1)-because in the preceding context there were rwo male referents that were active, and rhus, if he had made use of anaphors only, we might not have been able to understand who would have crashed whom. However, by employing the particular construction in the particular context, the speaker clarified the particular referent, pointed out the thematic continuity of his utterance, and, at the same time, he chose to empathize with Koskotas rather than with the other active male referent. Example ( 1 0),just like any other instance oflocal-ropic reinstaters, is characterized by the intonation pattern we are familiar with by now. Finally, rightmost full-fledged NPs can also be employed as corrective or as intentional Afterthoughts. Afterthoughts can be described very sketchily as occurrences rather similar to the ones analysed so far, which, however, differ in rwo very important respects. Firstly, the final element of the anaphoric clause is characterized by final intonation, and, thus, the pause berween the last clausal element and the rightmost full-fledged NP is interpreted as 'period intonation' rather than as 'comma intonation'. Secondly, the rightmost NP is always stressed. Such occurrences can take place when the speaker realizes the absence of any referential value of the preceding anaphoric clause, and, therefore, with the help of a full-fledged NP, which, naturally, has to appear to the right of the anaphoric clause, he attempts to establish its referent (corrective afterthoughts). However, such 'disconnected' rightmost NPs can also be employed, especially when the rightmost NP is an evaluative expression, if the speaker deliberately decides to separate rwo intonation units-i.e. the anaphoric clause and the rightmost full-fledged NP-in order to assign, through the 'independence' of

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 10) . . . Tha ton TSAKIZE, ton Koskota . . . [Will him1 crashed-he)=anaphoric clause [the Koskorat) �dislocated element [s [s · tha(=future particle)[x 1 ton(�clitic pers. pro., masc., ace. case, sg.)x 1 ) tsakize(=verb, past progressive, + 3rd ps. sg. suf)s ·) [x2 ton(=de£ art., masc., ace. case, sg.) Koskora(�proper name, i.e. surname, masc., ace. case, sg.)x2] s] He would have crashed Koskotas.

Maria Valiouli 67

the evaluative NP, some extra prominence to his evaluation (intentional 'after thoughts', or rather intentional 'second-thoughts'). ' ST R I C T' A N D D I S C O U RSE A N A P H O RA

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

With regard to anaphora, Greek right-dislocated constructions of all categories are characterized by an almost a priori anaphoric relation holding between the clitic or the lexically unspecified subject of the anaphoric clause and the right dislocated element. However, the kind of anaphoric relation involved is not always the same. There are RDs whose clitic or lexically unspecified subjects are assigned their specific deictic values through the deictic value of the dislocated NPs with which they must always co-occur, and by deictic I mean 'non directional pointing', a term suggested in Lyons ( 1 975: 74). On the other hand, there are RDs whose anaphors are assigned their specific referential values before the occurrence of the right-dislocated expressions, which are redundant even as reference maintenance elements. Thus, we could suggest that, depending on whether the dislocated expression is necessary for the computation of the deictic value of the anaphoric clause or not, Greek right-dislocated constructions should be interpreted in terms of 'strict' or 'discourse' anaphora, respectively, as these two rypes of anaphora are analysed i� Cornish ( 1986). According to Cornish, 'strict' nominal anaphora 13 does not involve reference in the sense of co-reference, since the pronouns involved cannot be interpreted deictically without the obligatory co-occurrence of their antecedents (which Cornish calls controllers) in a strictly constrained context and, thus, 'strict' nomi nal anaphors function mainly as formal cohesion markers. Discourse nominal anaphors, on the other hand, function as pragmatic coherence markers since they are often used to signal the degree of focus or saliency which their refer ents have been assigned by the speaker, or, I would add, the thematicality , i.e. shared or latent information related to their referents. Discourse nominal ana phors, in contradistinction to 'strict' ones, can also be used for the introduction of a discourse entity into the discourse model. Thus, on the basis of this distinction, I can suggest that Informative-Presup position anti topic and local topic reinstating constructions (as well as corrective afterthoughts) should better be interpreted as instances of 'strict' anaphora, since their anaphoric clauses cannot be interpreted deictically without the obli gatory co-occurrence of their dislocated NPs. However, it seems that even in the case of Informative-Presupposition antitopics, the speaker's intention is to channel the addressee into an acceptance of the high degree of focus of the state of affairs described in the anaphoric clause. Thus, although the clitic and lexi cally unspecified subjects are better analysed as formal-cohesion markers, the

68 Agreement and Greek RDs

A G RE E M E N T RE LAT I O N S The most fascinating thing about Greek right dislocations, however, is related to the question of agreement. When anaphors and full NPs are involved in an anaphoric relation (whether it is a case of 'strict' or discourse anaphora), this relationship is manifested usually in their agreement or concord in terms of their gender, persor.., ntimber, and case feature values. 1 4 Yet concord is not an absolute requirement for the establishment of an anaphoric relation. Moreover, the presence or absence of concord, in the values mentioned above between the anaphorically related elements in the Greek right-dislocated constructions, reflects differentiations in the speaker's referential or attitudinal perspective upon the particular discourse referent, which can always be interpreted in accordance with the primary communicative function of the particular category of right-dislocated construction used. N O N - C O N C O RD I N T H E N U M B E R F E A T U R E V A L U E S A difference in the number feature value, for example, may be employed when the speaker wants to show some kind of differentiated referential or attitudinal perspective upon the discourse referent. Consider, for example, ( I I a) and ( I I b) below, which are instances of Stressed-Focus antitopic constructions. To be more precise, ( I I b) is an authen tic example, which was used as a re-topicalization device, whereas ( I I a) is its potential variation. The difference, as we are going to see, between ( I I a) and ( u b) is that in (1 1 a) the right-dislocated NP is a proper name denoting the family name of a particular family seen as a unit consisting of individuals (i.e.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

speaker wants the addressee to accept them as pragmatic-coherence markers signalling the high degree of focus of the overall information carried by the anaphoric clause. Furthermore, in the case of local topic-reinstating construc tions the dislocated NPs function as disjunctive markers since the preceding anaphors are clearly referential, but they have more than one referential value. Stressed-Focus antitopic and all tail, re-naming, and intensifying construc tions, on the other hand, are instances of discourse anaphora. Nevertheless, by discourse anaphora I do not mean the relation between the clitic or the lexically unspecified subject of the anaphoric clause and the dislocated element, but rather the relation holding between the former and a latent or previously estab lished discourse referent, as the dislocated element is a semi-autonomous, i.e. oprional, independent-although prosodically integrated with the anaphoric clause-co-referring expression uttered for pragmatic reasons only.

Mana Valiouli 69

something similar to the Smiths, the Browns, etc.), whereas in ( I I b) the dislo cated NP is a proper name of feminine grammatical gender.

Although in both ( I I a) and ( I I b) the verb METAKOMISAN has a 3rd person plural suffix, as we can see, in ( I I b) the right-dislocated NP i Marianthi is in the singu lar. This non-concord in the number feature values reflects the speaker's need to indicate that his utterance is Marianthi-oriented or Marianthi-centred or simply that Marianthi is the most important member in her family. I can suggest, therefore, that the absence of concord with respect to the number feature value in the case of antitopic constructions reveals an attitudinal differentiation on the part of the speaker. When there is such an attitudinal differentiation on the part of the speaker, we can say that there is a combination of two compatible intentions as to the pragmatic impact of the dislocated construction. The speaker primarily wants to inrroduce and (re-)establish an issue about the referent of the dislocated element, and, at the same time, he also projects his psychologically orientated emotional attach ment to the same referent (i.e. there is some kind of singling out of a particular referent for purely emotive reasons). A difference in the number feature values in the case of tail constructions, on the other hand, reflects another kind of differentiation on the part of the speaker. Let us consider the following examples: ·

(12) Pente hronia psonizo ap ton idio manavi ke pio to ofelos; I have been shopping at the same greengrocer's for five years, and do I get any special rreatment?

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(1 1 a) De mou les; METAKOMisan , i Tsilipakides ; [Not me tell-you?] [Moved-they1]-anaphoric clause [the Tsilipakides1 ?]-dislocated expression [De(-neg. particle) mou( -clitic pers. pro., gen. case, 1 st ps. sg.) les(-verb, +2nd ps. sg. su£)?] [s [s · metakornis-(verb, aorist)+ [x 1 -an(- Jrd ps. pl. su£)x,J s .] [x2 i(-de£ art., masc., nom. case, pl.) Tsilipakides(-proper name, i.e. surname, masc., nom. case, pl.)x,] s] Tell me. Have the Tsilipakoses moved (to their new house)? ( 1 1 b) De mou les; METAKOMISan , i Marianthi ; clause me tell-you?] [Moved-they1]-anaphoric [the [Not Marianthi1 ?]-dislocated expression . [De(-neg. particle) mou(-clitic pers. pro., gen. case, 1 st ps. sg.) les(-verb, +2nd ps. sg. su£)?] [s[s · metakornis-(verb, aorist)+ [x -an(- 3rd ps. pl. su£)x,J s .] [x2 i(-de£ art., fern., nom. case, sg.) Mariantk(-proper name, i.e. first name, fern., nom. case, sg.)x2] s] Tell me. Have Marianthi and her family moved (to their new house)?

70 Agreement and Greek RDs EM

In ( 1 2), when there is concord with respect to the number feature value between the dislocated tail NP o alitis and the suffixes of the preceding verbs, the tail NP is used both referentially, as it refers to the particular greengrocer, and attributively, as well, since the particular individual is simultaneously presented as a 'qualified' member of that class of people who can metaphorically be labelled as 'bums'. When, however, there is non-concord in the number feature values of the anaphorically related elements, the tail NP refers not only to the particular greengrocer, but to all greengrocers in general, who are all characterized as bums. Nevertheless, what is really interesting about the latter case is that there can also be simultaneous inferred reference to other members of closely related trades, such as grocers, butchers, even shop keepers in general, and all those inferred members together with the original referent, i.e. the particular greengrocer, are attributed the characteristics of those people who can metaphorically be called 'bums'. The question of how many trades or professions will be inferred is a function of the interlocutors' background knowledge, associated stereorypes, and expectations. I must point out, however, that when there is no concord, in contradistinction to when there is, there must be a discernible pause before the utterance of the dislocated element, although the emphatic primary stress of the utterance does not fall on the last clausal element. This fact, together with the non-final pitch of the last clausal element, indicates that the dislocated expression is meant to function not only as a casual evaluative comment, but probably as a suggested discourse topic, which, in this way, is separated without being disconnected from the predication. Moreover, this differentiation in the intonation pattern of the utterance may also give an indication as to the cognitive processes taking place in the speaker's mind, as there is a considerable change in the speaker's referential perspective, or, conversely, the intonation may function as an instruction to the addressee that what is going to follow needs extra processing effort.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

mou dose halia domates, EM mou fage ta resta o alitis/i a lites [Em me gave-he1 bad tomatoes, em me ate-he1 the change] =anaphoric clause [the bum/the bumst]=dislocated expression [s [s. em(=interj.) mou(clitic pers. pro., gen. case, 1 st ps. sg.) dos-(=verb, aorist)+[x 1 -e(= 3rd ps. sg. suf)x t ] halia(=determ., pl.) domates(=noun, fern., ace. case, pl.)s ·Hs · em(=interj.) mou(clitic pers. pro., gen. case, 1 st ps. sg.) fag-(=verb, aorist)+[x 1 -e(=3rd ps. sg. suf)x t ] ta(=def art., neut., ace. case, pl.) resta(=noun, neut., ace. case, pl.) s ·Hx2 o/i(=de£ art., masc., nom. case, sg./=def art., masc., nom. case, pl.) alitis/alites(=epithet, masc, nom. case, sg./=epithet, masc, nom. case, pl.)x 2] s] 15 He gave me bad tomatoes, and kept the change on top of that, the burn/ ?the bums.

Mana Valiouli

71

N O N - C O N C O R D I N T H E GE N DE R FEAT U RE V A L U E S In Greek we can also have non-concord in the gender feature values between anaphorically related elements. As far as tail constructions are concerned, the question of concord or non concord in the gender feature values between the tail and its coreferential anaphor in the predication is usually determined by morphological con siderations. In other words, if the dislocated element has masculine, feminine, and neuter forms, depending on the sex of the person referred to, or on the grammatical gender of the form that denotes the referent, there will normally be concord between the anaphorically related elements. If, on the other hand,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Thus, in the case of tail constructions, a difference in the number feature values of the anaphorically related elements reflects a different kind of differentiation on the parr of the speaker: the speaker's referential differentia tion, as the original reference is extended to other inferred referents, and the speaker's attitudinal differentiation, which is different from that of the antitopic construction, as the dislocated expression is used both referentially and attributively or connotatively. If we consider once more the communicative functions of the re-naming and intensifying constructions, we can understand that, as far as the question of non-concord in the number feature value is concerned, we must exclude the possibility of its occurrence, as in either case the dislocated element has to be an 'enriched variation' of that anaphor in the anaphoric clause whose referent is singled out for pragmatic-psychologically oriented-reasons only. In other words, in either case the reference made is unique and specific and should remain so with no extension or narrowing down or attribution or connotation involved, because any such referential or attitudinal differentiations would refute the suggested singularity, and, thus, they would go beyond the primary communicative purposes of the categories under discussion. Similarly, in the case of local topic reinstaters, since the primary function of the dislocated element is to assign the preceding anaphor, which has more than one referential value, a specific deictic value, the dislocated element should have the same number value, which guarantees identity of reference, with its co-referential anaphor. If, however, in the preceding context there are several families, then, I suppose, we could have instances of local-topic reinstaters manifesting non-concord similar to that of example ( 1 1 a) further above, since metonymy, i.e. the use of single referents to denote all the members of groups of people-whether they are grouped together on a permanent basis or not-is quite usual in Greek.

72

Agreement and Greek RDs

the dislocated epithet has only one grammatical gender, then, naturally, there will be non-concord as in: (1 3) Den katalava TIPOTA, o vlakas . [Not understood-I1 nothing]=anaphoric clause [the idioti]=dislocated expresswn [s [s·[den(=neg. particle) katalav-(=verp, aorist)+[x 1 -a(= I st ps. sg. suf)x 1 ]s.] [x2 o(=de£ art., masc., nom. case, sg.) vlakas(=epithet, masc., nom. case, sg.)x2]s] I haven't understood a single thing, idiot me.

( 14) Kitakse ti beba mas. KITAKSE tin pos kimate glika-glika to chriso mou I # i chrisi mou . [Look the baby(fem.) our.] [Look her1 how is-sleeping-she1 sweedy sweedy]=anaphoric clause [the golden(neut)mytfthe golden(fem.)my1] = dislocated expressions [kitakse(=verb, imperative, sg.) ti(=de£ art. fern., ace. case, sg.) beba(=noun, fern., ace. case, sg.) mas(=clitic poss. pro., I st ps. pl.)] [s kitakse(=verb, imperative, sg.) tin(=clitic pers. pro., fern., ace. case sg.)[s· pos(=adverb) kima-(=verb, present)+ [x 1 -te(=3rd ps. sg. suf)x 1] glika glika(=adverb of manner realised twice)s·Hx2 to/i(=de£ art., neut., nom. case, sg.)!(de£ art., fern, nom. case sg.)x2]s] Look at our baby girl. See how quietly she is sleeping, the little dear. In Greek, it is common to refer to babies by pronouns of neuter gender either because we think that they have not grown up into personalities yet, or sometimes because we want to avoid any sexual connotations associated with the 'male' or 'female' conceptual meanings of the masculine or feminine gender. 1 6 1t is even common to refer to grown-ups with the help of pronouns and epithets of neuter gender when we feel very close to and rather protective towards them, or when we are especially moved by their behaviour or, conversely, when we want to criticisze their childish, i.e. socially improper, behaviour, or when we want to express contempt. For example, in (9), further above, there is a clash between the sex of the cousin referred to and the gender of the evaluative epithet to bastardo . Nevertheless, in (14) the choice for the epithet of neuter gender is made because of the social connotations of the feminine gender epithet i chrisi mou , which is normally used-especially in Northern Greece-when we want to address grown-up women in a polite, but somehow detached way.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 3) can be uttered by either a man or a woman but in either case the tail will characterize the speaker. Yet sometimes the choice of a particular grammatical gender is determined by social considerations rather than by morphological ones:

Maria Valiouli 73

Let us now consider example (I s) right below:

(I S) Ta martirise OLA , i Katina .

[Them revealed-she/he1 everything]�anaphoric clause [the Katinai]� dislocated expression [s[s· ta(�clitic pers. pro., neut., ace. case, pl.) martiris-(�verb, aorist)+ [x -e(�3rd ps. sg. su£) x 1 ] ola(�quanti£ pro., neut., ace. case, pl.)s·Hx2 i(�def. art., fern., nom. case, sg.) Katina(�proper noun, i.e. first name, fern., nom. case, sg.)x ]s] S/He spilfed the beans, the silly woman/the silly man. Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In (I s), the referred to entity may very well be a man, although both the article and the head-noun of the tail NP are of feminine grammatical gender and, moreover, the tail NP is a feminine proper name. However, the particular proper name in the particular example is used connotatively/attributively and can assign its male referent certain qualities that are 'normally' (?) characteristic of females, because in Modern Greek the proper name Katina is often used as the epitome of the qualities allegedly displayed by low class, uneducated, gossipy, unsophisticated, socially incompetent females. Thus, if used for male reference, the particular example may sound a little provocative, yet it is not only acceptable but it reflects very vividly the speaker's evaluation of the behaviour of the person referred to in the particular context. In other words, in (Is) the speaker indicates that the behaviour of the referred-to male has made him a qualified member of that class of females that share the qualities mentioned above. We have just seen that, in the case of tails, although non-concord in gender between the anaphorically related elements is determined mostly by the mor phological possibilities ofthe dislocated NP, there are cases in which gender non concord, or clash, between the referent's biological gender and the grammatical gender of the referring expression-as in ( I s) when it is used for male reference reflects attitudinal differentiation, i.e. socially defined options, on the part of the speaker. However, in such cases there is always some kind of attribution/ connotation involved. We can, therefore, understand why, in the case of right dislocated constructions, there can be gender non-concord between the dislocated expression and the preceding anaphorically related element only when the connotation or attribution is intended by the speaker. In other words, we cannot have gender non-concord with the re-naming, intensifying, and local topic reinstating constructions, as connotation/attribution is incompatible with their primary communicative functions. As for antitopic constructions, we can have such a differentiation on the part of the speaker as long as the referred-to entity is latent and there is some kind of attribution involved as in (I6) below. However, it should be noted that (I 6) reflects a clash between the biological gender ofthe female referent and the grammatical gender ofthe right-disloc�ted

74

Agreement and Greek RDs

expression, rather than non-concord between the incorporated anaphor in the verbal suffix and the dislocated element.

(16) is an extract of a dialogue between two friends talking about their holidays. There had already been a preceding discussion about a trip to a particular island between speakers A and B, but the subject was left open because A and his wife did not know what to do with their cat, and B was not really keen on going away. Moreover, B's girlfriend disliked the idea of going by boat. B's answer, therefore, reintroduces the issue of their going away on holiday by boat, and presents its recent development. The referent (his girlfriend, who always organizes their holidays) is characterized as crazy because her need to go away has prevailed over her dislike for boats. The choice of the neuter grammatical gender epithet to o palalo instead of the feminine gender i palavi/palaviara reflects either B's tender feelings towards her or a polite 'weakening' of the attribution. In other words, in (16) we have a successful combination of two compatible intentions on the part of speaker B: a reintroduction of an issue already discussed together with its recent development, and an evaluation of the behaviour of the referent with respect to the same development. We can, therefore, come to the conclusion that with right-dislocated constructions non-concord in the gender feature values between the anaphorically related elements or gender clash between referents and referring expressions reflects an attitudinal differentiation towards the referent on the part of the speaker. Moreover, we see, once again, that the speaker's liberty to alter her attitude towards the referent depends again on its compatibility with the primary communicative function of the particular category of right dislocation employed. N O N - C O N C O R D I N T H E C A S E FEATURE V A L U E S Let u s now consider the question o f non-concord in the case feature value between the anaphorically related elements in right dislocations. Case marking can express both syntactic (i.e. Subject, Object) and semantic (i.e. Agent, Goal)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(16) Speaker A:. Ti cha kanete celika; Tha minete ke sis edo; What have you (two) finally decided to do? Will you also be staying here? Speaker B: Ti les kale; Evgale ISITIRIA , to palalo . What are you talking about? [Booked-she1 rickets] =anaphoric clause [the crazy1]=dislocated expression [5[5.[evgal-(=verb, aorist)+[x1 -e(=3rd ps. sg. suf)x 1 ] isiriria (=noun, neue., ace. case, pl.)s ·] [x2 to(=de£ art., neue., nom. case, sg.) palalo(=epithet, neue., nom. case, sg.)x2]5] What are you talking about? She (has) booked rickets, the crazy thing.

Maria Valiouli 75

( 1 7a) Tou tin FERANE, tou vlaka . 11

[His 1 her brought-they]=anaphoric clause [(of) the stupid1]=dislocated expression [s [s{x 1 tou(=clitic pers. pro., masc., gen. case, sg.)x 1 ] tin(=clitic pers. pro.,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

functions, and it is closely related to the issue of topicality as well. According to Giv6n ( 1 98 3: 22), there is a case-role hierarchy with respect to topicality, which reads as follows: AGT > OAT/BEN > ACC > OTHERS. In the case of antitopic constructions (of both types, since they are employed for the (re-)establishment of an issue in the discourse model rather than for the establishment of the referents of their dislocated NPs), the referents are seen as an integral part of the semantic information of the predication and as such they should retain the semantic roles which they are assigned in the anaphoric clause. Thus, the primary function of antitopics does not allow any dif ferentiation between their anaphorically related elements in the value of the case feature, as such a differentiation would result in a pragmatic 'promotion' of the referent of the dislocated NP and we would end up with a single utterance suggesting two discourse topics. Incidentally, Lambrecht ( 1987: 2 5 4) suggests a pragmatically motivated constraint in the form of a pragmatic maxim, namely 'Do not introduce a referent and talk about it at the same time', which can explain why the dislocated NP in antitopic constructions cannot be promoted to an independent discourse topic. Similarly, in the case of re-naming, intensifying, and local topic reinstating constructions, the dislocated element has to retain the same semantic and pragmatic roles of its anaphorically related anaphor, because none of those constructions is used for the establishment of the already active referent of the dislocated expression as the current discourse topic and, furthermore, the dislocated NP has to be an 'enriched variation' of the preceding anaphor for the reasons iihave already explained. In the case of tail constructions, however, things can be very different. A tail construction is employed primarily for the expression of the speaker's subjective evaluation of a discourse referent in a given context. Nevertheless, the speaker may wish to highlight his personal, subjective opinion of a particular entity through the presentation of some piece of information. In such a case, however, the speaker's evaluation (i.e. the tail expression) may end up functioning as the suggested discourse topic and the information conveyed in the anaphoric clause as 'supportive evidence' of that evaluation. Thus, depending on whether or not the speaker wishes to topicalize his evaluation, we can have cases of tail constructions which display non-concord in the case feature values between the preceding anaphor and the tail expression. Let us consider the following example and its alternatives, together with the interpretations that they suggest:

76 Agreement and Greek RDs

fern. ace. case, sg.) ferane(=ve�, a?rist, +3rd ps. pl. su£)s ·Hx tou(=de£ � art, masc., gen. case, sg.) vlaka(-eplthet, masc., gen. case, sg.)x2J s] He's been had, the silly bugger. (1 7b) tou tin FERANE, ton vlaka . The same as in ( 1 7a) except for: [(to) the stupid1]=dislocated expression The same as in ( 1 7a) except for: [x2 ton(=de£ art., masc., ace. case, sg.) vlaka(=epithet, masc., ace. case, sg.)x2]5] ( 1 7c) tou tin FERANE, o vlakas . The same as in ( 1 7) except for: [the stupid1]=dislocated expression. The same as in ( 1 7) except for: [x2 o(=de£ art., masc., nom. case, sg.) vla kas(=epithet, masc., nom. case, sg.)x2] s ] Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In ( 1 7a), where there is concord in the case feature values between the eli tic Tou and the tail epithet tou vlaka , the referent is characterized as silly only because of the unfortunate incident described in the predication. Moreover, the foolish ness of the particular referent is presented as a matter of course. In (I 7b) and ( I 7c), however, there seems to be some kind of differentiation in the speaker's evaluation in that the foolishness of the person referred to is no longer a matter of course justified only through the information conveyed in the predication, but something which is worth being discussed further. To put it differently, the dislocated epithets in both ( I 7b) and ( I 7c) function as suggested discourse topics, which ask for further elaboration. Nevertheless, (1 7b) and (I 7c) differ not only from (I 7a) but also between themselves as the dislocated epithet in (I 7b) is in the accusative case, whereas in (I 7c) it is in the nominative case. I am not quite certain as to what is intended by such a differentiation but I suggest that in ( I 7b) the speaker simply thinks that the foolishness of the referent is worth becoming the discourse topic, whereas in (I 7c) the speaker strongly believes it is worth being discussed further. I must point out, however, that some Greeks find ( 1 7c) quite unusual, although they would accept it with no reservation if it were a case of left dislocation. Another major difference between ( 1 7a) on the one hand, and ( 1 7b) as well as ( 1 7c) on the other is the obligatory occurrence of a discernible long pause before the utterance of the dislocated epithets in both (I 7b) and ( I 7c). Also in (I 7a) there is a slight pause before the occurrence of the dislocated epithet, but this is because the verb FERANE receives the emphatic primary stress of the utterance. In (I 7b) and (1 7c), however, the pause must be long and, furthermore, the dislocated epithet should be stressed too, in the sense that there is a slight but audible separation of the two syllables of the epithets, i.e. vla-ka and vla-kas. Yet, in all three cases, irrespective of the length of the pause, there is non-final intonarion before the dislocated epithet. Thus, ( 1 7b) and ( 1 7c) might be characterized as blends, i.e. cases which are somewhere in between the right-dislocated constructions we have been discussing so far, and the

Maria Valiouli 77

intentional 'afterthoughts' (or 'second-thoughts'), which were briefly men tioned earlier. If the interpretations I have suggested for ( 1 7b) and ( 1 7c) are correct, and if we accept Chafe's suggestions with respect to the function of the 'comma intonation', then we can say that, with tail constructions, non-concord in the case-marking reflects an increasing scale of the speaker's certainty as to the discourse topicalization of his evaluation, and, as I have already explained, topicalization of the speaker's evaluation is compatible with the primary com municative function of the tail construction.

Let us now move to the question o f non-concord i n the person feature values between the anaphorically related elements in Greek right dislocations. However, we shall discuss only self-reference cases of non-concord as they are the only ones I have been able to find. As we have already seen, in Modern Greek, as soon as reference to a particular entity is established, the speaker is, in theory, free to go on referring to the same entity by making use of anaphors, unless, of course, for pragmatic reasons, he wants to employ reference by full NPs or overtly present independent personal pronouns. Nevertheless, even if there are pragmatic reasons, excessive use of the independent personal pronoun ego (=I) for self reference is often considered to be a sign of egocentricity. Self-reference by the use of full name or tide, on the other hand, does not occur very often, and when it does, it can reflect various intentions on the part of the speaker; i.e. depending on the context, the speaker may want to keep some 'distance' from herself and assign her words generality or objectivity. The current Greek Minister ofEdu cation, Mr Souflias, for instance, quite often refers to himself as 'the Minister' or 'the Minister of Education', either when he wants to assign his point of view some extra status or when he wants to point out that he does not act as an indi vidual. In addition to self-reference by full name or tide, we can have a rather marked way of self-reference when the speaker makes use of a proper name, which is not his real name, and which is always used connotatively or attributively. Let us, for example, consider the following cases whose right dislocated NPs are proper names: 1 8

( 1 8)

ca kan o , i Mary Panayiotara . [Everything them do-I J ] =anaphoric clause [the Mary Panayiocara J ]=dis locaced expression [s [s · ola(=quanti£ pro., neue., ace. case, pl.) ca(=clitic pers. pro., neue., ace. OLA

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

N O N - C O N C O R D I N THE PER S O N FEATU RE V A L U E

7X Agreement and Greek RDs

case, pl.) kan-(=verb, present)+[x 1 -o(= 1 St ps. sg. su£)x1 J s .) [x2 i(=de£ art., fern., nom. case, sg.) Mary Panayiotara(=proper names, i.e. first name and surname, fern., nom. case, sg.)x2]s] I do every single job, a real Mary Panayiotara. ( I 9) owus tous peripioume , i Maria . [Everybody them attend-It] [the Maria1] [s [s· olous(=quanti£ pro., masc., ace. case, pl.) tous(=clitic pers. pro., masc., ace. case, pl.) peripi-(=verb, present)+ [x1 -oume(= I st ps. su£)x 1 ] s ·Hx2 i(=de£ art., fern., nom. case, sg.) Maria(=proper name, i.e. first name, fern., nom. case, sg.)x2]s] I attend to them all, a real maid. Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Although the main verb in both ( I X) and ( I 9) has a first person singular suffix, the right-dislocated NPs make reference not only to the speaker but to two women, called Mary Panayiotara and Maria, respectively. In ( I X), the NP Mary Panayiotara refers both to the speaker, whose name is not Mary Panayiotara, and to Mary Panayiotara, the main character of a famous Greek song, who has become the symbol of the working woman, mother, wife, and housewife who is wildly exploited by all the males in her social circle. Thus, the dislocated NP is deliberately used in a way seemingly referential and connotative/attributive at the same time. Similarly, in ( I 9), even if the speaker is called Maria, the marked way of self-referring employed marks a change in the speaker's referential attitude towards herself Since in Greece Maria is the proverbial female servant, the speaker wants to present herself as a maid. We can, therefore, claim that a difference in the person feature values again reflects a differentiated attitude towards the particular discourse referent on the part of the speaker, which may result in a topicalization of the suggested connotation. The fact, on the other hand, that I have not been able to find any similar examples with the other categories of right dislocations suggests that such an attitudinal and referential differentiation with the help of person non concord may only be possible with tail constructions. This, of course, once more confirms the conclusion that the speaker's 'freedom' to express referential or attitudinal differentiation is a question of functional compatibility. It should be noted, however, that when it comes to blends, which are somewhere in between antitopics and tails, we can also have cases of person non-concord. On the basis of the observed differentiations, we can come to the conclusion that Greek RDs can carry out simultaneously more than one of the pragmatic functions I have suggested, as long as these functions are compatible or applied in a coherent order, i.e. according to our expectations. To put it differently, the speaker can make use of non-concord in any of the feature values discussed as long as he orders his functional priorities in accordance with the primary function of the category of right dislocation used. Actually, apart from

Maria Valiouli 79

Lambrecht's ( 1987: 254) constraint, which I have already mentioned, Chafe ( 1987: 32) speaks of a similar cognitive constraint, namely the 'one new concept at a time' constraint. These two constraints together with the constraint of functional priority or compatibility I have suggested can account for the acceptable non-concord possibilities as far as Greek RDs are concerned.

CONCLUSIONS

Received: 1 1 - I-<) 3 Revised version received: 6-X-93

MARIA VALIOULI

>./ School ofEnxlish Faculty <>./Philosophy

Dept. < Theoretical and Applied Linxuistics

A ristotle University

54006 Thessaloniki Greece

N OTES 1 The term empathy is used m Kuno's sense { 1 976: 43 1 ), who defines it as 'the speaker's identification, in varying degrees, with a participant in an event'. 2 The idiomatic use of vaporaki in Greek here implies that the person referred to is not a drug dealer, and ignorant of carrying d rugs. In the particular context, the use of the

4

indefinite article mia (-one) in front of the surname implies 'all women bearing the surname Valiouli', Valiouli herself included, of course. It should be noted, as the authors point our that in order for (s) to be acceptable, the present tense of the verb pino has to be interpreted w.r.t. the future. Ashby ( 1 988: 2 1 s) discusses the case of a

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Greek right dislocations are used for a number of distinct communicative functions. Despite the differences, however, both in terms of the anaphoric relations and concord possibilities that hold between the preceding anaphor and the dislocated expression, and in terms of their primary communicative functions, all categories of Greek right dislocations are instances of the same phenomenon. All right-dislocated constructions of all categories are used for a tacit, casual, even polite projection of the thematicality of the anaphoric clause and for a simultaneous unemphatic projection of the speaker's referential or attitudinal differentiation towards the thematic information of the con struction-whether the thematically involved is justified by the discourse context or simply taken for granted. Thus, we can also conclude that the pre posed anaphoric clause is as a marker for pragmatic coherence, whereas the dis located expression is the indicator of the emotive meaning of the state of affairs described in the anaphoric clause.

8o Agreement and Greek RDs French RD with an indefinite right dislocated NP, which seems to be subject to contexrual conditions similar to the ones defined for (5 ). 6 Incidentally, I have come across the following RD in a wrinen dialogue of a Greek regional dialect, where the dis located NP is the indefinite pronoun tipotas (-nothing):

Ta mathes, tipotas; [Them1 learned you] -anaphoric clause [nothingi]-dis located expression

8 9

10

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7

The particular RD is employed by one of the rwo interlocutors just before she goes on with some piece of sensational news. In my dialect, however, which is the Modern Greek standard variety, one would probably use the expressions 'Ta mathes;' (-them learned-you?, i.e. Have you heard it?) or the RD ' Ta mathes ta nea;' ( -them(cliric)1 learned-you the news1?, i.e. Have you heard the news?) or possibly 'Emathes tipota;' (-Learned-you nothing?, i.e. Have you heard anything?) or even 'Emathes tipota nea;' (-Learned you nothing news?, i.e. Have you heard any news?). Although I do not know how often the RD with tipotas is used in the particular dialect or whether it sounds natural, I could suggest that it is a dialectal amalgamation of the four alter native expressions that are used in my dialect. Although all the RDs discussed in this paper can be further analysed in contra distinction to potential alternatives, I will not attempt such an approach for reasons which have to do with the length of the present paper. Both the classification and the terminol ogy used in this paper are fully developed m Valiouli ( 1 990). The term antitopic is used to denote a (re-)ropicalization strategy. It should be noted, however, that Lambrecht ( 1 98 1 , 1 987) uses the term antitopic for all French RDs. These rwo terms have been coined by

Prince ( 1 978), who proposes a similar distinction for It -Cleft sentences. More specifically, she claims that there are rwo varieties of It -Clefts, which differ With respect to their presupposed pans: (a) the Stressed-Focus It -Clefts whose presup posed parts represent information that the speaker assumes and that the addres see knows or can deduce, and (b) the Informative-Presupposition It -Clefts whose presupposed pans represent infor mation which is known to the speaker only. 1 1 The term tail has been chosen because of a specific connotation that the corres ponding Greek word oura implies when it is used in a saying that has to do with any definitive last-minute change of an overall picture. Thus, it should not be confused with the term tail in Dik (1978), which is a pragmatic function outside the predication presenting 'afterthought' mformation for the clarification or modificatiOn of a term in the predication. 1 2 The term local topic reinstaters is used 111 Cornish (1 988). 1 3 Cornish ( 1 986: 36) uses the term nominal 'in its more semantic sense as determining a first-order referent'. q The term concord is more appropriate than the term agreement as the latter presupposes syntactic relations rather than pragmatic ones (see also W1ese 198 3). 1 5 Em is an interjection which can be used for the introduction of rwo clauses which describe rwo instances of improper behaviour, the second always being worse than the first. 16 See Wiese (1983) for a detailed discussion of the conceptual meanings of personal pronouns. 1 7 Note that the expression tou Imou Ietc. tin Jerane (lit.: they brought her to him/me/ etc.) is an idiom meaning 'he's/I've/etc. been had'. Moreover, the indirect object of the ditransitive verbferno has to be in the genitive. 1 8 Examples (1 2), ( 1 3), ( 1 4), (1 5), (1 7a), ( 17b),

Maria Valiouli 8 1 ( 17c), ( 1 8) and (19) were also analysed in a paper entitled 'Right-Dislocated Ana phorically Functioning Nominals, Con cord, and Referential!Attitudinal Perspective', which I presented at the sth Annual Symposium on the Description

and/or Comparison of English and Greek, held by the Department of Theoretical and Applied Linguistics, School of English, Aristotle University, Thessaloniki, Greece, in March 199 1 .

RE FERE N CE S

Some Universals in Language Usage, Studies in Interactional Sociolinguistics, Cambridge University Press, Cambridge, New York, New Rochelle, Melbourne, Sydney. Chafe, W. (1 987), 'Cognitive constraints on information flow', in R. S. Tomlin (ed.),

Typological Studies in Language I I : Coherence and Grounding in Discourse , J. Benjamins

Publishing Co., Amsterdam, Philadelphia, 2 1- 5 1 . Chafe, W . (1 988), 'Linking intonation units in spoken English', in ]. Haiman & S. A. Thompson (eds), Typological Studies in

Language 1 8: Clause Combining in Grammar and Discourse,]. Benjamins Publishing Co.,

Amsterdam, Philadelphia, 1-27. Cornish, F. ( 1 986), Anaphoric Relations in

English and French: A Discourse Perspective, Croom Helm, London, Sydney, Dover, New Hampshire. Cornish, F. ( 1 988), 'Anaphoric pronouns: under linguistic control or signalling particular discourse representations?', Journal ofSemantics, 5, 23 3-60. Dik, S. ( 1 978), Functional Grammar, North Holland Publishing Co., Amsterdam. Given, T. ( 1 983), 'Topic continuity in dis course: an introduction', in T. Given (ed.),

Topic Continuity in Discourse: A Quantitative

Cross-Language Study, ]. Benjamins Publishing Co., Amsterdam, Philadelphia, 5-4 1 . Joseph, B. D. & I . Philippaki-Warburton, ( 1 987), Modern Greek , Croom Helm, London, Sydney, Wolfero, New Hamp shire. Kazazis, K. & ]. Pentheroudakis (1975), 'Reduplication of indefinite direct objects in Albanian and Modern Greek', Language, 52, 398-403. Kuno, S. (1976), 'Subject, theme and the speaker's empathy' in C. N. Li (ed.), Subject and Topic , Academic Press, London, New York, Toronto, Sydney, San Francisco, · P 7-44· Lambrecht, K. ( 1 98 1 ), Topic, Antitopic and Verb Agreement in Non-Standard French , ]. Ben jamins Publishing Co., Amsterdam. Lambrecht, K. (1 987), 'On the status of SVO sentences in French discourse', in R. S. Tomlin (ed.), Typological Studies in Language I I : Coherence and Grounding in Discourse, J. Be�amins Publishing Co., Amsterdam, Philadelphia, 2 1 5-6 1 . Lyons, ]. ( 1 97 5), 'Deixis a s the source of reference', in E. L. Keenan (ed.), Formal Semantics of Natural Language , Cambridge University Press, Cambridge, London, New York, Melbourne, 6 1 -83. Marmaridou, S. ( 1 989), 'Proper names in communication', journal ofLinguistics , 25, 3 5 5-72. Marrhiessen, C. & S. A. Thompson ( 1 988), 'The structure of discourse and subordina tion', in J. Haiman & S. A. Thompson (eds.), Typological Studies in Language I 8:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Ashby, W. J. ( 1 988), 'The syntax, pragmatics, and sociolinguistics of left- and right dislocations in French', Lingua , 75, 20329. Bosch, P. ( 1 98 3 ), Agreement and Anaphora , Academic Press, London, New York, Toronto, San Francisco. Brown, P. & S. C. Levinson ( 1987), Politeness:

X2 Agreement and Greek RDs Clause Combininx in Grammar and Discourse , J. Benjamins Publishing Co., Amsterdam, Philadelphia, 275-329. Pawley, A. & F. H. Syder (1 98 3), 'Natural selection in syntax: notes on adaptive variation and change in vernacular and hterary grammar', Journal of Pragmatics, 7, 5 51 -79· Prince, E. F. (1 978), 'A comparison of Wh clefts and It-clefts in discourse', Languaxe, 54. 8 8 3-906. Prince, E. F. ( 1 98 5 ), 'Fancy syntax and "shared

knowledge"', Journal of Pragmatics, 9, 6581. Sandford, A. J., K. Moar & S. C. Garrod ( 1 988), 'Proper names as controllers of discourse focus', Language and Speech , J I , Pan 1, 43-56. Valiouli, M. (1 990), 'Anaphora, Agreement and the Pragmatics of"Right Dislocations" in Greek', Ph.D. thesis, Aristotle Uni versity, Thessaloniki, Greece. Wiese, B. (1 983), 'Anaphora by pronouns', Linguistics , 21, 373-4 1 7. Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

journal ofSemantics 1 1 : X 3 - 1 0 1

© Oxford Umversiry Press 1 99�

Tense, Temporal Reference, and Tense Logic PATRICK BLACKBURN

Utrecht University

Abstract

1

I N T RO D U C T I O N

Few formal semanticists look with much favour upon Priorean tense logic. For a start, natural language tenses usually have referential import-bur reference to times is not possible in Priotean languages. Secondly, the focus of research on tense has shifted in recent years. The anaphoric properties of tense, and the role temporal constructions have to play in discourse, are now centre stage. Such issues seem alien to tense logic. These shortcomings are real; none the less there is a great deal of value in Prior's approach. Prior insisted on the primacy of the internal view of time. This view situates the speaker firmly inside the temporal flow: the speech-time centred ordering of past, present , and future , rather than the absolute earlier than \later-than relation, is considered central to the analysis of tensed talk. This is natural: we live in time, and the internal perspective is imprinted on natural language in many ways. Prior's insistence on the internal perspective led him to develop his temporal calculi as modal logics, not classical logics. Classical logic, with its explicit variables and binding, offers a God's-eye-view of temporal structure; modal logic, on the other hand, uses operators to quantify over structure 'from the inside'. Moreover, because modal logic rejects the complexities ofvariables and binding in favour of operators, the resulting systems are very simple. Prior's decision to work with modal languages has interesting logical consequences. For example, as we now know from modal correspondence theory (see van Benthem 1 983, 1 991 ), to work with a modal languages is essentially to

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This paper examines extensions of Priorean tense logic in which reference to times is possible. The key technical idea IS to sort the atomic symbols of Prior's language and to impose different interpretational restrictions on the different sorts. Among the sorts introduced are nominals (which permit Reichenbachian analyses of tense and tense-in-texts to be reconstructed m tense logic) and sorts which mim1c temporal index1eals and calendar terms. The possibilities ra1sed by sorting richer systems are briefly discussed.

84 Tense, Temporal Reference and Tense Logic

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

work with a very restricted fragment of classical logic, a fragment with many special properties. Thus if a problem can be modelled in a modal language, we already know something important: the full power of classical logic is not needed. Other logical aspects of Prior's system are also well understood. For example completeness, decidability, and complexity results for interesting classes of temporal structures have been obtained, and there is a growing interest in developing good theorem provers. Given the increasing interplay between formal semantics and disciplines such as knowledge representation, and the consequent emphasis on the role of inference, this logical insight is valuable. Thus there are many good reasons for attempting to extend Prior's approach rather than simply abandoning it. This paper introduces a number of extensions and shows that they are adequate for dealing with the problems mentioned above, as well as a number of other difficulties. Now, when extending Prior's system we must take care not to destroy those properties that made it attractive in the first place. The extensions proposed in this paper are all particularly simple and revolve around one central strategy: sorting the atomic symbols of the language. Different sorts of propositional symbols are introduced, and their interpretation is constrained in various ways to achieve the desired forms of temporal reference. This strategy preserves the simplicity of the Priorean model (both the syntactic and semantic changes introduced are confined to the atomic level) and does not tamper with the 'internal perspective' characteristic of modal logic. Moreover, the logical results known for systems of the kind considered here confirm that the strategy really is straightforward: completeness, complexity, and correspondence results have been given for such systems (see, for example, Gargov & Goranko (1993) or Blackburn (1993a, b) and these results typically extend the standard results for Priorean languages in a direct manner. Thus the proposed extensions preserve the desirable properties of Prior's system: it remains to be seen, however, whether they are of interest in natural language semantics. The purpose of the present paper is to show that they are. We proceed as follows. After a resume ofPriorean tense logic, we introduce our first sorted language: nominal tense logic . In this extension a second sort of propositional variable-the nominal -is introduced. Norninals are constrained to be true at exactly one time; thus they act as a name for the time at which they are true. This simple referential mechanism enables us to combine the ideas of Prior and Reichenbach in a natural way. It also allows us to deal with certain anaphoric phenomena; the key idea here is to use norninals as if they were discourse markers of temporal DRT. With the sorting strategy thus established, we turn to indexicals. Nominal tense logic is enriched with three new propositional symbols: yesterday, today , and tomo"ow, and these are interpreted using the semantic machinery of the

Patrick Blackburn Ss

California theory of reference. As we shall see, the exploitation of this machinery by means of sorted atomic symbols yields a clean account of some basic interactions between tense and temporal adverbiak Following this we introduce calendar terms, and then discuss the possibility of sorting richer systems. Some historical remarks are in order. The idea of sorting intensional languages can be traced back to work done in the late t9(Sos by Arthur Prior (1¢7, I9(}8) and Roben Bull (1970). This work seems

(O

have been largely

overlooked. The Sofia school of modal logic independently developed similar ideas in the I 98os, in the context ofboth Propositional Dynamic Logic (Passy & Tinchev 198 5, 1991 ), and modal logic (Gargov el a/. I 987; Gargov & Goranko temporal logic. With the exceptions ofPrior's discussion of the indexical 'now', most of this literature is technical and does not consider application of sorting to natural language semantics.

2.

PRIOREAN TENSE L O G I C

The language o f (propositional) Priorean tense logic is constructed out of the following primitive symbols. First we have a countably infinite collection VAR of propositional variables, which we officially write as p, q, and

r,

possibly

subscripted; however, when discussing examples drawn from natural language we shall frequently use more mnemonic symbols such asJohn run, Fido bark, and Tlze third door on tlze left be brown . In addition we have some truth functionally adequate collection of Boolean connectives (in chis paper we shall choose--. and /\), and the tense operators P and F. We form the well-formed forn1.ulas (or sentences) of Priorean tense logic from these symbols as follows. First we define ATOM, the set of atomic symbols of the language, ro be precisely VAR, the set ofpropositional variables. We then stipulate that WFF, the set ofwell-formed formulas ofthe language, is to be the smallest set containing ATOM that is closely under negation,

conjunction, and the application oftense operarors. We make free use of all the usual defined symbols. In particular, we define the other Boolean connectives and - in the usual manner, and in addition defme Of> to be -.F---¢, and H¢ to be -.P...¢ . , for all wffs ¢ .

-, V

All this will probably be familiar to the reader, with one (perhaps puzzling) exception: why did we bother introducing the set ATOM? This is dearly redundant: we could have defined WFF directly in terms of VAR In fact the introduction of ATOM is an anticipation of the later sorted languages we shall consider. Syntactically, sorting merely amounts to introducing different kinds of atomic symbols, and alterations of the definition of ATOM are the only

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I99J). Slightly later Blackburn (I990, 1993a) considered sorting in the setting of

X6 Tense, Temporal Reference and Tense Logic

M t= a [t] M t= -.rp [t] M t= ¢> 1\ lfJ [t] M F= 116 [t] M F= Prj> [t]

iff iff iff iff iff

t E V(a ), for all a E ATOM not M F= rj> ( t] M t= rj> ( t ] and M t= ljJ ( t] 3t '(t < t ' and M F= ¢> [ t ]) 3t '(t ' < t and M t= ¢> [t ']) '

Note the way this satisfaction definition captures the essence of the internal perspective. We evaluate formulas at points inside models (the evaluation point can be thought of as speech-time) and the operators F and P scan forwards and backwards relative to this deictic centre. In the work that follows we occasionally make use of the idea of logical equivalence. This is defined as follows. Two wffs ¢> and ljJ are logically equivalent iff for all models M and all points t in M: M t= rj> [t]

iff M t= l/J [t].

One final remark is worth making. Note that the base case of satisfaction definition is parameterized in terms of ATOM. As we proceed we are going to impose a syntactic sortal structure on ATOM. This syntactic sortal structure will have a semantic correlate: we are going to place constraints on how the different sorts may be interpreted. But such atomic level constraints are the only semantic changes we shall make; the satisfaction definition given above will not need changing when we consider nominal tense logic in the following section.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

syntactic changes made in what follows. Thus all the languages discussed in this paper have the same formation rules, those stated above. Prior's language is interpreted usingframes and Kripke models. A frame T is a pair ( T, <) where T is a non-empty set, and < is a binary relation on T. The elements of T are often called times or points, and < is usually called the precedence relation . In order to make this simple model more realistic we shall place some constraints on allowable precedence relations. In what follows we shall assume that < is always a strict total order, that is, a relation that is transitive, irreflexive, and trichotomous. In short, we are assuming a linear flow of time. This assumption will suffice for the first half of the paper; thereafter we will need to be a little more demanding. A Kripke model M is a pair (T, V) where T(� (T, <)) is a frame and V is a valuation on T. A valuation on T assigns a subset ofT to each atom; that is, such a valuation is a function from ATOM to Pow (T). Valuations thus impose 'information distributions' on frames, and Kripke models can be thought of as time flows decorated with information about what happens when. Given a Kripke model M( �( T, <, V)), a wff ¢> , and a point t E T the fundamental thing we are concerned with is whether M satiifies ¢> at t, or in symbolic form, whether M F= ¢> [t ]. This relation is defined as follows:

Patrick Blackburn 87

3 N O M I NA L TE N SE L O G I C As Priorean tense logic contains no mechanism for referring to times, its use as a

.

=

=

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

model of tense in natural language has severe limitations. Tenses aren't solely, or even primarily, quantificational in nature: they usually have referential import. For example, an utterance of john ran' doesn't mean that John ran at some completely unspecified past time, but that he ran at some particular (contextually determined) past time. The natural representation of this sentence in Priorean tense logic, P(lohn run ), fails to mirror the referential nature of the English original. Our solution to this difficulty is very simple. We make Priorean tense logic referential by means of the following extension. We add a new type of propositional symbol, called nominals, to our language. Nominals will typically be written as i ,j, k , and so on. They will be wffs, and we will be able to combine them freely with other wffs in the usual fashion to form new wffs. The point of introducing nominals lies in their interpretation: we shall insist that in any model nominals are to be true at exactly one time. Nominals thus 'name' the unique time at which they are true and our problem is solved. For example we will now have a simple representation for john ran' that captures its referential force: P(i A John run ) suffices, for in order for this to be trueJohn must run at the past time picked out by i . That's all there i s to it. Let u s now go through i t i n more detail and see where it leads. Syntactically we obtain the language of nominal tense logic NTL from the language ofPriorean tense logic by adding a denumerably infinite set NOM of fresh atomic symbols to the language. The elements of NOM are denoted by i , j, k . ., possibly subscripted, and they are called nominals. These symbols are 'fresh' in the sense that NOM n VAR 0. That is, nominals and propositional variables are distinct. We define the set of atoms ATOM of the language of NTL to be NOM u VAR, and we make the wffs of this enriched language using exactly the same formation rules as for the Priorean language: WFF is the smallest set containing ATOM that is closed under negation, conjunction, and application of tense operators. Thus the only syntactic difference is that ATOM is now VAR u NOM, instead of simply VAR (T, As with the Priorean language, we interpret NTL in Kripke models M V). As before T is a frame (T, <) and V is a function from ATOM to Pow(T). However, we now make a crucial demand: for all iE NOM, V(i) is a singleton subset of T. Functions V that do not obey this sortal restriction are not valuations. As with syntax, this atomic level change is the only change we make. Wffs of NTL are interpreted in Kripke models in exactly the manner familiar from Priorean tense logic.

88 Tense, Temporal Reference and Tense Logic

the representation of john ran' as P(i 1\ John run ) shows, the utility of nominals stems from the fact that they are a referential mechanism. Because i is true at one and only one point in any model, this representation makes a far stronger claim than P(]ohn run). In order for P(i 1\ John run ) to be true at some time of utterance t, not only must there be a time t ' preceding t at which John run is true, but in fact the unique time picked out by i must precede t and it must be true at this particular time thatJohn run . In short, the enriched language gives us the power to refer-and because temporal reference is such a pervasive phenomenon in natural language, this will enable us to model some interesting phenomena. For example, NTL provides a framework in which the ideas of Reichenbach and Prior can be naturally amalgamated. Reichenbach claimed that understanding tensed expressions involved understanding the temporal relations that held between three special points of time: the point ojspeech , the point ojevent , and the point ojreftrence. The first two points have the obvious meanings: the point of speech is the utterance time, and the point of event is the time at which the event being spoken of occurred. Note that Priorean tense logic, with its internal view of time, gives a clear account of these two points: the point of speech is the point in the model at which the utterance is evaluated, while the point of event is the point in the model where the event being spoken of is verified. But what of Reichenbach's novelty, point of reference ? This concept is best introduced by example. Consider the past perfect sentence john had run'. Note the way it works. Our attention is not directed immediately to the time at which John runs, rather we are first referred to some point t ' preceding the point of speech and told that at some point prior to this intermediate point John ran. Such an intermediate is called a point of reference. Reichenbach accounted for the variety of tenses found in natural language in terms of the different patterns of temporal precedence and coincidence these three points can exhibit. For example, in the past perfect we have the point of event E preceding the point of reference R , which in turn precedes the point of speech S, or, to use Reichenbach's diagrammatic notation, E-R -S. To give a second example, Reichenbach proposed that the function of simple past tense was to locate the point of event and point of reference at the same point in the past; thus his diagram for the simple past was E, R -S. Now this concept of a special intermediate point picked out by tensed sentences simply is not present in the Priorean framework. Equally clearly, once we have nominals in our language to refer to times, we can mirror this idea. Let us consider the matter systematically. Reichenbach's idea admits of thirteen possible patterns of precedence and coincidence. Table 1, the first three columns of which are taken from Comrie (1981: 2 5) , lists with examples the possibilities admitted by Reichenbach's analysis. The fourth column gives an NTL representation. We could sum up these entries by saying that we have factored the English tenses As

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Patrick Blackburn 89

Table 1 Name

English example

Representation

E-R-S E, R-S R-E-S R-S, E R-S-E E-S, R S, R, E S, R-E S-E-R S, E-R E-S-R S-R, E S-R-E

Pluperfect Simple past Future-in-the-past Future-in-the-past Future-in-the-past Present perfect Present Prospective Future perfect Future perfect Future perfect Future Future-in-the-future

I had seen I saw I would see I would see I would see I have seen I see I am going to see I will have seen I will have seen I will have seen I will see (Larin: abiturus ero)

P(i A Pr/J ) P(i A � ) P(i A R/J ) P(i A R/J ) P(i A R/J ) PrjJ � R/J F(i A Pr/J ) F(i A Pr/J ) F(i A Pr/J) F(i A � ) F(i A R/J )

into a 'shift' (performed by the tense operators) and a 'refer' (performed by the nominals). The 'shift' component is Prior's contribution, the 'refer' Reichen bach's. Some further comments on these representations are called for. First note that all three logically possible permutations of R , E, and S for both the future-in-the-past and the future perfect are represented by a single formula. Such a mode of representation seems to be preferred by linguists. Comrie, for example, argues that such non-committal representations are essential ( 198 I : 26-7); indeed he criticizes Reichenbach's diagrams for their specificity. His argument is that while one of the three possibilities may be favoured, and one largely ruled out, these preferences are merely Gricean implicatures: all three usages are possible, and a uniform representation is required. Second, note that the present perfect and the simple past are represented differently. The identical representations these two forms receive in standard tense logic is a source of dissatisfaction with the use of tense logic in the analysis of natural language. However, though it is pleasant that there is a distinction between the representations of these two tenses in NTL, note how it arises. It has been achieved by adding something-namely a Reichenbachian component-to Prior's account of the past tense. On the other hand we have added nothing to the (unsatisfactory) Priorean account of the perfect. The perfect poses difficult problems for formal semantics, and the machinery introduced in this paper does not solve them. Actually, even for English, Table I does not exhaust all the possibilities; for, as Prior observed ( 1¢7: 1 3), certain sentences seem to require more than one point of reference. Consider Prior's example, 'I shall have been going to see

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Structure

90 Tense, Temporal Reference and Tense Logic

John'. This seems to require two points of reference, R 1 and R 2. The pragmatically most plausible pattern they will exhibit is S -R 2-E -R 1 . The semantic possibilities are captured in NTL as follows:

F(i 1\ P(j 1\ F(I seeJohn )) ).

4 I N TE RSENTE N T I A L A N A P H O RA In this section we use nominals to model temporal intersentential anaphora. We approach this subject via Partee's paper 'Nominal and Temporal Anaphora' (1 984). Partee's discussion, which draws on earlier work by Hinrichs (1 982), is couched in the language of temporal DRT (Kamp 198 1 ; Kamp & Rohrer 1 98 3). In the remainder of this section we will see how Partee's examples can be dealt with in NTL. The key idea is to use nominals as discourse markers-or, to put a more Reichenbachian slant on it, to think of points of reference (as Reichenbach himself did) from the standpoint of intersentential anaphora. Consider the discourse 'The shutters were closed. It was dark.' As Partee points out, the tense in the second sentence (which is a state sentence) anaphorically picks out the time referred to by the tense of the first sentence (which is also a state sentence). We can represent this discourse in NTL as follows:

P(i 1\ The shutters be closed) 1\ P(i 1\ It be dark). Note that we have reused the i nominal provided by the first sentence in the representation of the second. This really does not give us what is desired, namely reference to a past time that is both a shutters-being-dosed-time and an it-being-dark time. Although the two occurrences of the i nominal are in

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Comrie (198 1 , 1 98 5: 1 22-30) develops this idea to its logical conclusion. He proposes a modification of Reichenbach's system which essentially amounts to allowing an arbitrary number of reference points. (Comrie also replaces Reichenbach's simultaneity relation with a language of intervalic overlap. This idea is easy to model in the sorted interval based languages discussed later in this paper.) The possible tenses are then defined to be linkages between the point of speech and the point of event mediated by a (possibly null) sequence of reference points. This idea is not made completely precise, but the following seems to capture Comrie's intention. Let � and tjJ be metavariables over wffs and n a metavariable over nominals. We take as primitive tensed forms all instances of P� , � , and Rj> . We then say that if tjJ is a tensed form then so are P(n 1\ '1/J ), ( n 1\ '1/J ), and F(n 1\ '1/J). Everything in the previous table falls under this definition and, according to Comrie, so do most of the tense forms exhibited in natural languages.

Patrick Blackburn 9 1

different conjuncts and embedded within the scopes of different past tense operators, we achieve the desired semantic effect, as it follows from the semantics of NTL that the above representation is logically equivalent to:

P(i 1\ The shutters be closed 1\ It be dark ).

P(i 1\ Mary wake up sometime during the night) 1\ P(i 1\ She turn on the light) is inadequate: clearly the illumination should follow the awakening. The event verb of the second sentence has advanced the referential focus. This is a typical feature of event sentences. Whereas state sentences elaborate on the time in focus, one of the functions of event sentences is to move the story along. (Of course, this is a simplification of a complex phenomenon. However, as my aim is simply to show that Partee's account can be reconstructed in NTL, I shall adopt her assumptions without further discussion.) In fact it is simple to advance the reference time in the required fashion: we need merely introduce a new reference marker {that is, a new nominal) and insist that it names a more recent time than did the old:

P(i 1\ Mary wake up sometime during the night) 1\ P(j 1\ Pi 1\ She turn on the light). Note how this works. The reference time of the first sentence is picked out by i. The second sentence then advances the reference time, and we capture the effect of this as follows. Instead of simply introducing a new nominal j we introduce the conjunction j 1\ Pi . The role ofj is to pick out the new reference time, and the conjunct Pi guarantees that the old reference time lies in the past. As a third example consider the following discourse which mixes stative and

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

That is, because nominals are 'hardwired' to a unique time, no matter how deeply they are embedded within the structure of our representations they are always available for reuse. This ability of nominals to 'float out of the scope' of tense operators means we can build up discourse representations online. Consider the above discourse once more. Imagine that we are processing this discourse, building up its representation as we go. First we process 'The shutters were closed'. The parser goes to work and builds P { i 1\ The shutters be closed). The second sentence is then analysed. Because it is a sentence in the simple past tense a representation of the form P(n 1\ It dark) must be built, where n is to be instantiated with some nominal. But which nominal should be used? Because the second sentence is a state sentence that follows hot on the heels of another, the parser reuses the i of the previous representation. As we have just seen, this natural strategy is semantically correct. Matters are more interesting when we have a sequence of two or more event sentences. Consider the discourse: 'Mary woke up sometime during the night. She turned on the light.' Representing this in NTL by:

92 Tense, Temporal Reference and Tense Logic

event verbs: john got up, went to the window, and raised the blind. It was light out. He pulled the blind down and went back to bed. He wasn't ready to face the day. He was too depressed.' We can represent this as follows:

In short, we can build the desired representations using a strategy that boils down to 'shift and refer' coupled with 'advance the reference time when an event verb is encountered'.

5 I NDEXICALS I n this section we use sorting to model the indexicals 'yesterday', 'today', and 'tomorrow'. We add three new symbols yesterday , today , and tomorrow -to NTL. These new symbols are atomic wfs and may be freely combined with other wffs to make more complex wffs. They will be used to represent English sentences containing indexicals. For example, we will be able to represent john climbed yesterday' as -

P(yesterday 1\ John climb), and this representation will correctly capture the fact that john ran yesterday' is true iff John did in fact run at some point during the day immediately preceding the day of utterance. But how are we to specify the semantics of these new items so that they correctly mimic the behaviour of the English originals? An elegant answer lies to hand: we shall make use of certain semantic machinery developed by Montague (1 970), Kamp (197 1 ), and Kaplan ( 1 977, 1979), namely the machinery underlying the California theory of reference. In spite of its age this theory remains one of the most detailed responses to the problems posed by indexicals. As we shall see, the main ideas of the Californian approach mesh smoothly with the techniques of the present paper, and in fact the major difference between this paper and the earlier work is not conceptual but technical. Hitherto the semantic machinery of the California theory of refer ence was exploited by means of additional modal operators. We exploit the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

P(i 1\ John get up ) 1\ P(j 1\ Pi 1\ go to the window) 1\ P(k 1\ Pj 1\ raise the blinds) 1\ P(k 1\ It be light out) 1\ P(i1 1\ Pk 1\ He pull blind down ) 1\ P(j1 1\ Pi1 1\ go back to bed) 1\ P(j1 1\ He be not ready toface the day) 1\ P(j1 1\ He be too depressed)

Patrick Blackburn

93

same machinery here, but not with operators: instead we sort. This leads to a cleaner account of the basic interactions between tense and indexical reference. Let us begin with the simplest task, defining the syntax of our language. This is NTL augmented with three new atomic symbols: yesterday, today , and tomorrow. The set of atomic symbols of this language is defined as follows: ATOM � VAR u NOM u {yesterday , today, tomorrow}.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We make wffs from this stock of symbols exactly as before. That is, WFF is defined to be the smallest set containing ATOM that is closed under negation, conjunction, and the application of tense operators. Now for the task of defining the semantics. First of all, we need to pay more attention to our choice of temporal ontology. These indexicals all presuppose the concept of dayhood; thus it is sensible to work with temporal structures rich enough to support a reasonably realistic picture of this assumed day structure. As there is no obvious single best choice, let us agree for the purpose of the present paper to work with a particularly simple option: Q(�(Q, <)), the rational numbers in their usual order. From now on we shall only consider models based on this frame. We can impose a reasonable looking day structure on Q as follows. Define the set of Days on Q to be { [z, z + I ) : z E Z(� Q)}. Here Z(�(z, <)) is the set of integers, and we are using them as a simple (and arbitrary) way to mark off the division between successive days. We also make use of the following three functions. The function day maps elements q of Q to the unique element of Days that contain q . That is, given a point of time, day answers the question 'What day does this point belong to?' The functions next and prev are both maps from Days to Days . In particular, for all integers z, next ([z, z + I )) � [z + I , z + 2) and prev([z , z + I )) � [z - I , z), so these functions take as input a day and return, respectively, the next day and the previous day. This is all we shall need to assume in the way of day structure for the purposes of this paper. With these preliminaries established, we are now ready to introduce the apparatus of the California theory of reference. The key idea of the California theory of reference is that context must be taken into account when evaluating the truth of utterance. This simple idea is made precise as follows: contexts are added to Kripke models, and when wffs are evaluated they are evaluated not only at a certain time, but also at (or in ) a certain context of utterance. Let's make this precise. Fix some non-empty set C, the set of contexts, or contexts of utterance, and specify a function g : C - Q . The function g is to be thought of as specifying the utterance time of each context of utterance. We call the quadruple ( Q , <, C, g) a contextualization of Q. Clearly this is an extremely meagre notion of context; none the less it suffices for the phenomenon we are going to consider. We are now ready to interpret our language. As always with sorted systems,

94

Tense, Temporal Reference and Tense Logic

the key points is to say what we mean by a valuation. Here is the definition. A valuation is a function V : ATOM X C - Pow(Q) that respects the following constraints. First, for each nominal i and each context c , V(i, c) must be a singleton. Second, for each context c: V(today , c ) = day(g (c)) V(tomorrow, c) = next(day (g(c ))) V(yesterday , c) = prev(day(g(c)))

M l= a [t, c] M I= ....,� [t, c ] M l= � l\ tJ! [t, c] M I= Rfi [t, c ] M I= J16 [t, c ]

iff iff iff iff iff

t E V(a , c ), for all a E ATOM not M I= � [t, c] M I= � [t, c] and M I= tJI [t, c] 3t '(t < t ' and M I= � [t ', c]) 3t'(t' < t and M I= � [t ', c])

Let us put these enrichments to work Consider the sentence john ran yesterday'. We could represent this as, P(yesterday 1\ john run ). This representa tion captures the temporal force that the 'yesterday' adds to john ran'. For sup pose we are in some context of utterance c, and suppose we evaluate P(yesterday 1\ john run ) at the pair [g(c), c]. That is, we are asking ourselves what happens ifwe 'utter this wff' in the context c at the utterance time for c . Then it follows from the above truth definition that this wff is true iff there is a time t ' in the day preceding the utterance day such that john run is true at [ t ' , c ] . In short, this wff is true iffJohn did run during the day classified as the yesterday of the context of occurrence. Now this is precisely what is required. Moreover, as many temporal semanticists have pointed out, it is something that multiple operator approaches have difficulty achieving. To take a recent example, Oversteegen (1 989: 3) complains that:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The intuition behind these constraints should be clear. In any context c the today atom is to be true at all points in the day containing the utterance time and at no others. That is, today picks out precisely those points belonging to the 'day of utterance'. Next, yesterday and tomorrow are to be true at, respectively, precisely those points in the day preceding, and the day succeeding, the day of utterance. Finally, note that within a fJXed context of utterance c, nominals (as before) pick out a unique time. This definition allows the possibility that in different contexts of utterance the same nominal i may be used to name different times. In giving these atomic level stipulations we have essentially completed our task The following clauses explain how to interpret arbitrary wffs, but as the reader familiar with the Californian theory will observe, they are standard: everything of importance has already taken place. So, define model M to be a pair (( Q , <, C, g), V) where ( Q, <, C, g) is a contextualization of Q and V is a valuation. Then for all times t E Q and all contexts c E C we define:

Patrick Blackburn

95

If temporal adverbs like yesterday are rreated as operators, there is a scope problem: both sequences, yesterday in the scope of the tense operator and the tense operator in the scope of yesterday, are obviously wrong.

P(i 1\ yesterday 1\ John run ) 1\ P(i 1\ It be wet and cold and nasty) Note that the representation of the first sentence contains two restrictors: the nominal i and yesterday . As described in the previous section, the nominal is contributed by the simple past tense of 'went'. It gives us a name for the reference point. The yesterday is contributed by the word 'yesterday'. Its function is to constrain when the running took place, namely to 'within yester day'. Together the two restrictors enable the temporal information encapsu lated in john went running yesterday' to be captured, and crucially part of this representation (the i ) is reused to correctly anchor the second sentence. Our discussion has become rather general. Let us return to specifics. Another pleasant feature of these restricting representations is that they handle certain semantically anomalous sentences rather nicely, namely those sentences where the shift demanded by the tense clashes with the reference demanded by the indexical. For example, both *John will run yesterday

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Such scoping problems simply do not arise with the sorting strategy. The reason is simple. Our indexicals do not in any sense quantify; rather they act as a range of restrictors on the quantification performed by the familiar temporal operators. For example the role of tomorrow in the wff F(tomorrow 1\ John run ) is to ensure that only those times classified as being 'tomorrow' are relevant to the trUth or falsity of the wff The familiar problems resulting from operator interactions are bypassed. This talk of restricting the range of temporal quantification bears on the work of the previous section. There we talked of factoring tenses into a 'shift' and a 'refer': the tense operators pointed us in the desired direction, and nominals filled the 'referential slot' giving us the desired specificity. But it is clear that the mechanism underlying both F(i 1\ John run ) and F(tomorrow 1\ John run ) is essentially the same. Not only do these wffs share a common syntactic form, i and tomorrow share the common semantic role of restricting the times of relevance to the F operator. 'Shift and refer' is a special case of'shift and restrict'. Thinking in this more general fashion in terms of quantifications and restrictions seems a useful way of thinking about temporal representations. This is particularly apparent when it comes to building up discourse representations. Consider the following discourse: john went running yesterday. It was wet and cold and nasty.' This can be represented as follows:

96 Tense, Temporal Reference and Tense Logic

and *John ran tomorrow are semantically anomalous for this reason. Let us consider what happens when we give the obvious 'shift and restrict' representations to these sentences, namely: F(yesterday 1\ John run ) and P(tomorrow 1\ John run ).

6 CALE N D A R TERMS This section sketches another application of the sorting strategy; modelling calendar terms such as january', 'Friday', 'March', and ' I 984'. We first extend the language by adding five further sorts: DAY = {Sunday , . . ., Saturday) {1, . . ., 31} DATE MONTH = Uanuary , . . ., December) YEAR = { I , 2, 3, . . . I 989, I 990,. . .) = {BC, ADJ ERA -

We define CAL to be DAY u DATE u MONTH u YEAR u ERA-we assume that all these sets are mutually disjoint from each other and from all the sorts we already have-and then define: ATOM = VAR u NOM u {yesterday, today, tomorrow) u CAL The wffs are made from ATOM in the usual fashion. As an example of how we are going to use this language, the sentence john ran on Monday 4th December I 989' will be represented by: P(Monday 1\ 4 1\ December 1\ I 989 1\ AD 1\ John run ). I shall now sketch the semantic definitions that ensure that this wff does its work properly. Recall from the previous section that we are only considering models based on the frame Q, and that we have defined what we mean by a day structure over Q. Thus it is relatively straightforward-if a little tedious-to

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Just as we would hope, neither wffis satisfiable in any model. That is, no matter what model we choose, no matter what pair [t , c] we evaluate at, both wffs evaluate to false. Our sorted language 'sees' the semantic anomaly; to some extent it is mimicking the way tense and temporal reference interact in English.

Patrick Blackburn

97

define the needed calendar structure to interpret our new atoms. The required calendar structure Cal is a 4-tuple: Cal � ( Weeks, Months, Yers , Eras ). Each of its four components, Weeks, Months , Years , and Eras , is a partition of Q . (A partition of Q is a subset P of Pow ( Q) such that U P � Q , each p E P is non-empty, and for all p , p ' E P, p of p ' implies p n p ' 0.) We assume without further ado the natural induced ordering of Weeks, Months , Years , and Eras . For example, if m , m ' E Months we say that m < m ' iff 'tJt E m 'tJ t ' E m '(t < t '). We want to place the obvious conditions on each of these partitions so that Cal really does look like a calendar. Consider the case of Years , for example. First, define a year to be the set of points contained in 36 s or 366 consecutive elements of Days . (Recall from the previous section that Days is the set of days in the day structure, and further recall that both successor and predecessor functions on days are defined as part of the day structure. Thus 'consecutive days' are defined in the obvious way using this apparatus.) We then define the set Years in the calendar structure to be any 'leap year structured' partition of Q such that every element of the partition is a year. By 'leap year structured' is meant that every fourth element in Years is a year containing 366 days, while the remaining three out of four years contain only 3 6 s days. We construct Weeks , Months , and Eras in similar fashion. Every week in Weeks is made up of the points in seven consecutive days; every month in Months is made up of the points in either 28, 29, 30, or 3 I consecutive days; and Eras partitions Q into two. However, as well as ensuring that each of these four sets Weeks , Months , Years, and Eras is correctly structured, we must also ensure that they 'hang together' correctly. For example, the end of the first era should coincide with the end of some year; the start of a year should coincide with the start of a month containing 3 I days, while the second month in any year should contain 28 days, unless the year is a leap year when it must contain 29; and so on. Stating these details completely would be tedious; I shall leave the details to the reader's knowledge of calendars and proceed. We must now describe what we mean by valuations. These are maps from �

which satisfy the constraints of the previous section, and which in addition satisfy the needed constraints concerning the sorts in CAL. These constraints, though obvious, are wearisome to spell out in detail, so I shall only give them for the simplest case, the two atoms BC and AD in ERA. Recall that Era is a binary partition of Q ; let us write this partition as {be , ad), where be < ad. If V is to qualify as a valuation it must satisfy the demands that:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

V :ATOM X C --+ Pow(T),

9H

Tense, Temporal Reference and Tense Logic

V(BC, c ) = be V(AD, c)= ad , for all c E C. I leave the task of stipulating the constraints needed for the other new sorts to the reader. We now have a language with a fairly wide referential repertoire. Further extensions in a similar vein are possible, and relatively routine to define. For example, we could add 'clock terms' to the language, and define 'clock structures' in a manner that would allow john reached the summit ofRuapehu at three o'clock' to be represented by

P(three o'clock 1\ John reach the summit ifRuapehu ). CONCL U D I N G REMARKS

I n this paper we have taken a rudimentary temporal language (namely Priorean tense logic) and, by progressively imposing sortal distinctions at the atomic level, have modelled various forms of temporal reference in natural language. The resulting systems are still very limited. For example, they throw no light on progressive aspect, adverbial modification, or the role of the present tense in English. None the less, we have made substantial progress using very simple tools, and it is natural to ask whether sorting over a richer ontology could prove fruitful. To close the paper we consider the matter. The most straightforward development of the present approach would be to sort an internal based system. A (fairly standard) choice of intervalic ontology is the following: an interval structure I is a 4-tuple ( T, <, 6 , o ), where T is a non empty set, the set of intervals, and <, 6, and o are all binary relations on T. These relations are thought of as the precedence , subinterval, and overlap relations respectively. Of course it is necessary to specify further properties of these relations and their mutual interactions if these interval structures are to look temporally realistic. We shall not do so here but refer the reader to van Benthem ( I 99 I ) for a thorough discussion of the various options. To talk about such interval structure we might use the following language. As primitive symbols we have a set VAR of atomic symbols, written p , q , r and so on, and the logical symbols --. , 1\, F, P, l and 0 . We make wffs from this collection of symbols in the obvious way. To interpret this language we again use Kripke models. In the present setting a Kripke model M is a pair (I, V) where I(=( T, <, 6, o)) is an interval structure and V is a valuation that assigns subsets of T to atomic symbols. We interpret wffs in models in the expected fashion. The clauses for the atomic symbols, and for .....,, 1\, F and P are precisely as for Priorean languages. The two new clauses are as follows:

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7

Patrick Blackburn

99

M l= l¢ [t] iff 3t '(t ' 6 t and M I= ¢ [t ']) M 1= (¥ [t] iff 3t '(t' o t and M I= ¢ [t '])

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

That is, !.¢ is true at an interval t iff¢ is true at a subinterval t' of t ; and C¥ is true at an interval t iff¢ is true at an interval t ' that overlaps t . Now this new language is certainly more expressive than the Priorean language, bur it also suffers from a familiar weakness; there is no mechanism for temporal reference. However (as should now be apparent), this weakness is easily removed. For example, let us add a second sort INOM, the sort of interval nominals, to the language. We shall write these new symbols as e, d, c and so on and insist that in any valuation these items are to be true at precisely one interval. Thus we could represent john ran' by P(e 1\ John run ). This asserts that John run is true at some past interval, namely the one picked out by e. Note that it is a somewhat better representation than that offered by NTL: the use of e which can be true over an extended interval-captures the fact that the speaker was referring to an extended period of time. Among other things this enables Comrie's modifications of Reichenbach's system to be accurately mirrored. More generally, the richer repertoire of interval operators, in combination with the various referential sorts, gives rise to finer grained analyses of tense and temporal reference, and it becomes possible to give a (not wholly risible) analy sis of progressive aspect. None the less, the improvements obtainable in the interval based setting seem to be, by and large, fairly obvious refinements of what has been achieved in the point based setting. A more exciting approach is to work with ontologies containing both interval and event structure. For example, Blackburn, Gardent, & de Rijke ( 1994) work with back andforth structures . These are Kripke models consisting of an interval structure homomorphically linked to an event occurrence structure. From the point of view of natural language semantics such two sorted models are natural: the interval structure provides a realistic temporal ontology, the event structure provides a domain in which aspectual distinctions can be drawn, and the morphic link systematically correlates temporal information with event information. Moreover, from the point of view of abstract modal logic, the extension involved is fairly straightforward. There are obvious modal languages for dealing with such structures (essentially interval languages with additional operators) and their logical theory can be developed in the standard manner. The work is still being developed, but already two points seem to be emerging. First, this ontology makes it possible to do interesting work on tense, aspect, adverbial modification and their interaction in a modal setting. Second, the natural way to exploit the ontological richness seems to be by working with languages of low complexity. That is, rather than working with languages with powerful collections of operators, it becomes natural to push ever more

IOO Tense, Temporal Reference and Tense Logic

information into 'structured atomic symbols' which are sensmve to sortal information. In a sense one takes the sorting strategy to its logical conclusion: the sorting does the bulk of the work, the operators relatively little. Only further work can determine just how far this approach can be pushed. In the meantime, I shall simply note that (as we have seen in this paper) a surprising amount of analysis can be done even within the point based setting. Although much criticized, Prior's work contains a kernel of technical and intuitive insights that is hard to dismiss. These insights can be extended in a natural manner to many current concerns of temporal semantics. Acknowledgements

PATRICK BLACKBURN

Received: o8- I0-93 Revised version received: OJ-O I -94

Department ofPhilosophy Utrecht University Heidelberglaan 8 3584 CS Utrecht The Netllerlands

RE FERE N C E S Bentham, J. van ( I 98 3), 'Correspondence theory', in D. Gabbay & F. Guenthner (eds.), Handbook of Philosophical Logic 2, Kluwer Academic Publishers, Dor drecht. Bentham, J. van ( I 99 1 ), The Logic ofTime, 2nd edition, Kluwer Academic Publishers, Dordrecht. Blackburn, P. (1 990), 'Nominal tense logic and other sorted intensional frameworks', Ph.D. thesis, Centre for Cognitive Science, Universiry of Edinburgh, Scotland. Blackburn, P. ( I 993a), 'Nominal tense logic', Notre DameJournal ofFormal Logic, 34, 5683. Blackburn, P. (1993b), 'Fine grained theories of rime', in Aurnague, Borillo, Borillo & Bras-Grivart (eds.), Working Papers of the

Fourth International Workshop on Space, Time and Movement, Chateau de Bonas, France. Blackburn, P., Gardent, C., & de Rijke, M. (1994), 'Back and forth through time and events' (extended abstract), to appear in

Proceedings of the Ninth Amsterdam Col loquium , ILLC, Amsterdam. Bull,

R

(I 970), 'An approach to tense logic',

Theoria , 1 2, 1 7 I -82.

Comrie, B. (1981), 'On Reichenbach's approach to tense', in Hendrich eta/. (eds.),

Papersfrom the Seventeenth Regional Meeting, Chicago Linguistics Society , Chicago, 24-30. Comrie, B. (1985), Tense , Cambridge Uni versiry Press, Cambridge. Gargov, G., Passy, S. & Tinchev, T. (1987), Modal Environment for Boolean Specula tions (preliminary report), in D. Skordev

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I would like to thank Inge Bethke, Claire Gardent, Hans Kamp, Alex Lascarides, and Jerry Seligman for their advice, criticism, and (not least) their patience in discussmg the ideas of this paper. The comments of the two anonymous referees on the penultimate version were also very helpful.

Patrick Blackburn 101

Methods in the Study of Longuage, Part

1,

Mathematisch Centrum, Amsterdam, 277-322. Kamp, H. & Rohrer, C. ( 1983), 'Tense in texts', in Bauerle, Schwarze, & von Stechow (eds.), Meaning, Use, and Inter pretation of Longuage, Walter de Gruyter, 25 cHi9. Kaplan, D. (1 977). 'Demonstratives: an essay on the semantics, logic, metaphysics and epistemology of demonstratives and other indexicals', Draft No. 2, MS, UCLA. Kaplan, D. (1 979), 'On the logic of demon-

stratives', in French, Uehling & Wettstein (eds.), Contemporary Perspectives in the Philo sophy ofLonguage, University of Minnesota Press, Minneapolis, 401 - 1 2. Montague, R ( 1970), 'Universal grammar', in Richmond Thomason (ed.), Formal Philo

sophy: Selected Papers of Richard Montague,

Yale University Press, New Haven and London. Oversteegen, E. (1989), Tracking Time: A

Proposal for the Representation of Temporal Expressions in Duuh , Proefschrift, Rijks universiteit te Utrecht. Passy, S. & Tinchev,T. (1985), 'POL with data constants', Information Processing Letters, zo, 3 5-4 1 . Passy, S. & Tinchev, T. (1991), 'An essay i n combinatory dynamic logic', Information and Computation , 93, 263-3 32. Partee, B. (1984), 'Nominal and temporal anaphora', Linguistics and Philosophy, 7, 243-86. Prior, A. (1 967), Past, Present and Future , Oxford University Press, Oxford. Prior, A. (1968), ' "Now" ', Nous, z, 1 01-19. Reichenbach, H. (1 947), Elements of Symbolic Logic , Random House, New York.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(ed.), Mathematical Logic and its Applications , Plenum Press, New York. Gargov, G. & Goranko, V. (1993). 'Modal logic with names', journal of Philosophical Logic, zz, 607-36. Hinrichs, E. ( 1982), 'Temporal anaphora in discourses of English', paper presented at the Ohio State University Conference on the Semantics of Tense and Aspect m Discourse. Kamp, H. ( 1 97 1), 'Formal properties of "now"', Theoria , 37, 227-3. Kamp, H. (1981), 'A theory of truth and semantic representation', in J. Groenen dijk, T.Janssen, & M. Stokhof(eds.), Formal

jollrnal ofSemantics 1 1 : 103- 1 3 1

© Oxford University Press 1 994

Review Article

Jane Grimshaw. Argument Structure . MIT Press, Cambridge, Mass.

(Linguistic Inquiry Monograph I 8), I990.1 FRANK BECKMANN

1

I NT RO D U C T I O N

2

A R G U M E N T S T R U C T U RE REPRE S E N T AT I O N : A N I N FORMAL OVERVIEW

Most generative approaches to the study oflexical knowledge assume a level of representation-called argument structure (henceforth a-structure )-which encodes the grammatically relevant properties of individual lexical items. One of the most important properties to be represented is the 'logical' or syntactic valency of the item, i.e. the number of arguments a predicator selects in order to yield well-formed structures on the phrasal and eventually on the sentential level. I would like to distinguish betwe�n basically four stages in the evolution of a-structure representations.2

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Studies in generative syntactic theory over the past decade are not only characterized by a fundamental shift from grammars as systems ofbase-generating phrase structure and construction-specific transformation rules towards a model if principles and parameters (e.g. Chomsky I 98 I, I 986), but also by a new understanding of the lexical component of linguistic knowledge. Whereas earlier versions of generative grammar-such as Chomsky's Aspects -model considered the lexicon as a place where the basic irregularities of a language are to be represented, more recent approaches emphasize the fact that sentence structure is-to a considerable extent-predictable from lexical properties of individual items. Furthermore, the lexicon is not conceived of as a simple list of lexical items, but as an organized structure, subject to independent principles of lexical knowledge. The study presented by Grimshaw (I 990) (henceforth G) is primarily concerned with a crucial aspect of lexical knowledge, namely the question: What are the syntactically relevant properties ifpredicators, and how are they to be represented?

1 04 Review Article

I . the 8-grid approach: a-structure is an unordered set ofE>-roles (represented by means of E>-role labels (c£ Stowell I98I)), e.g. PUT: (Agent, Theme, Location) 2.

the ordered E>-grid approach 1: diacritics are used to distinguish between external and internal arguments (Williams I 98 I), therefore encoding informations about the structural organization of the simple sentence; E>-role labels may be substituted by variables bearing no semantic content at all, e.g.

3·

the ordered E>-grid approach II: an additional diacritical distinction between direct and indirect internal arguments (Marantz 1984) is introduced, e.g. PUT: Agent ( Theme , Location)

4· the prominence theory of a-structure: a-structure is a hierarchically ordered representation along the dimensions of 'thematic' and 'aspectual' promi nence (G I 990). These various versions of a-structure representation differ sharply from traditional subcategorization frames; for example, they do not impose any restrictions on the part of speech a selected argument has to belong to, and a-structure does not encode any information about the linear precedence relations among the arguments whatsoever. Furthermore, a-structure is a level of representation distinct from , but via linking rules related to the semantic representation. G emphasizes that a-structure mediates between this level of meaning representation (specified in terms ofLexical Conceptual Structures; c£ Jackendoff (I 98 3, I 990)) and syntactic 0-srructure, operating as some sort of formal 'interface'. She points out that the 'organization of the a-structure for a predicate is taken to be a reflection of its lexical semantics, so that the a-structure of a predicate should be derivable from key characteristics of its meaning' (G: 3 ). This point raises the question why a-structure remains necessary as an independent level of representation at all if its main characteristics are 'predictable' from lexical meaning. G argues against reductionist solutions by pointing out that important restrictions on syntactic processes follow from the interaction of a-structure and independent sub-modules of syntax. One example is passivization: 'It is only by positing a-structure that we can explain the limits of passivizability' (G: 4; c£ foomote 2).3 In other words: a-structure is 'projected ' from the lexical meaning of an item (G: 1 ), but lexical meaning is not sufficient to explain its syntactic behaviour. It should be noted, however, that it

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

PUT: Agent (Theme, Location) or PUT: x(y, z)

Review Article 1 05

is nor only a-structure itself which might account for syntactic phenomena {such as an item's participation in certain rypes of transitivity alternations), but also linking relations holding between lexical semantic structure and a-structure. For example, word pairs with identical representations at the level of conceptual structure {semantic converses like the verbs give and receive) differ with respect to the variables of lexical conceptual structure (henceforth lcs) being mapped to their respective positions in a-structure (Speas I 990: 30), and studies on middle constructions show that there are lcs variables which remain unlinked to a-structure positions entirely {Hale & Keyser I 986; see also below).

In the second chapter ('The Structure of Argument Structure': 7-44), G argues for an internally organized representation of a-structure. The characteristic structure of this representation is derived from a fine-grained semantic analysis of the individual predicate, based on the notions of thematic and aspectual prominence . In the course of developing her theory, G eliminates thematic role labels from a-structure representation and defines the external argument as a concept based on the 'architecture' of the representation. Empirical evidence for structured a-structure is drawn from data concerning 0-marking in Japanese light verb constructions (which will not be covered in this review) and English synthetic compounding. Let us consider the notion of thematic prominence first. 3.1

The notion ofthematic prominence

Following Jackendoff's early proposal { I 972),4 G { I 990: 8) assumes a proto argument structure representing the Agent as the most and the Theme as the least prominent argument in a predicator's a-structure: (I) {AGENT{EXPERIENCER{GOAL/SOURCE/LOCATION{THEME)))) Due to the central claim of 0-prominence theory, the most prominent argument of a-structure (in agenrive predicates the argument bearing the Agent-role) is realized as the most prominent constituent in syntactic structure, i.e. as the subject of the sentence. A-structure itself does not contain any 0-role labels,5 the 'hierarchy itself plays only one role, and this is located stricrly in the interface between lexical conceptual structure and argument structure' {G: I o). Consequently, the syntactic functions which were originally ascribed to 0-roles are now attributed to thematic hierarchy, which has syntactic effects only indirectly via a-structure.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

3 T H E H I E RARC H I C A L O R G A N I Z AT I O N O F A - S T R U C T U RE

1 o6

Review Article

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This quite general property assigned to a-structure requires a deeper conceptual clarification. G states that the 'structural organization of the argument array is determined by universal principles based on the semantic properties of the arguments' (G 1990: 7). She neither elaborates on the 'universal principle' she is referring to, nor on 'the semantic properties' claimed to be relevant to the structural organization of the predicates a-structure. These important points are explained later in the chapter in terms of event-structure representations as far as the 'aspectual dimension' is concerned. The motivation for thematic analysis remains rather superficial from a semanticist's point of view. Thematic hierarchy, interpreted as the above mentioned 'unversal principle' might indeed function as some kind of 'organizing device' (e.g. in the sense of a well-formedness condition); one should bear in mind, however, that the arguments of a predicator do not have any semantic properties inherently, i.e. independently of the predicator. This fact makes it quite hard to defend a version of thematic a-structure organization based on 'the semantic properties of arguments', as the above quoted statement suggests.6 What G really means is the set of semantic (i.e. here: thematic) properties the predicator assigns to its arguments. Implicit to the assumption that 'the prominence relations expressed . . . are a reflection of the nature of the lexical semantic representations from which a-structure is projected' is the insight that merely stipulating thematic hierarchy in terms of a proto-argument structure is a rather weak hypothesis. As is well known from the literature, the very notion of e-role is poorly understood-both in terms of the semantic content or 'nature' of individual roles and even their number.7 This makes any conclusion based on the notion 'thematic role' equally unprecise. To summarize: G does not make explicit whether the hierarchy she proposes is some sort of grammatical, formal device (as suggested by her statement that 'it is located strictly in the interface between lexical conceptual structure and argument structure', 1 0), or whether it is a 'reflection' of the lexical semantic representation (which would be a conceptually rather different characteriza tion). The latter approach is based on the assumption that a-structure is 'preorganized' semantically-a position which makes it much harder to defend the existence of a level of a-structure as an independent level of representa tional organization of grammar. Lacking a well-founded theory of 8-roles is only symptomatic for a serious problem G's analysis faces: she has to base her theory upon semantic concepts (including not only thematic roles, but also the notions of 'events' and 'states', etc.; see below) without being able to provide the reader with a clearpicture ofthe theory ofsemantics she assumes. Claiming that 'the prominence relations expressed . . . are a reflection of the nature of the lexical semantic representations from which a-structure is projected' just promises that there might be a way to derive certain

Review Article 1 07

generalizations from the proper semantic representation, but G fails to illustrate how this derivation might proceed. In the context of thematic prominence theory, it might even be possible that thematic hierarchy is 'a ref lection' of extra-linguistic prominence constraints, based on general principles of cognition.8 3.2

Empirical evidencefor thematic prominencefrom synthetic compounding

(2) a. gift-giving to children b. *child-giving of gifts

In other words: 'When the head takes more than one internal argument, the least prominent must be inside the compound, and the more prominent outside' (G: 1 4- 1 s ). In a later paragraph (68-70), G convincingly shows that the observed differences in the behaviour of synthetic and root compounds can be reduced to their respective a-structure representation; since root compounds lack a e-marking head, they do not have an a-structure, hence constraints and principles of a-structure do not apply. This allows, for example, for the occurrence of subjects inside root compounds. Applying this empirical generalization to the description of the compounds with psychological predicates reveals that the direct object of frighten -class verbs does not participate in well-formed compounding: (3) a. God frightens man. b. *man-frightening god This fact is easily explained within the framework of G's theory: the Theme argument god is the least prominent argument in terms of thematic hierarchy, therefore it may not be expressed externally to the compound. Put in other words: since man is thematically more prominent than god, it must not be realized within the compound. This generalizes to the prediction that 'compounding of an external argument will be impossible when the predicate

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Let us proceed by reviewing some of the empirical syntactic evidence G uses to argue for thematic prominence dimensions, focusing on English synthetic compounding. English shows an interesting asymmetry with respect to the distribution of Theme and Goal arguments within synthetic (i.e. verbal) compounds: the Theme argument has to be realized within the compound (i.e. close to the e-marking verbal head), whereas the compound formed with the Goal argument yields an ungrammatical structure (cf. G: 1 4):

1 08 Review Article

3·3

The notion ofaspectual prominence

G starts her analysis with the observation that frighten -class verbs obviously contain a causative meaning component which fear-verbs are lacking. Since 'cause arguments of cause predicates are always subjects' (G: 23), this provides a simple explanation for frighten -verb constructions, given that we accept Pesetsky's analysis that 'the subject offrighten has the role "cause of emotion" ' (G: 2o). This explanation, however, draws from 'thematic reanalysis'-a strategy already rejected.10 A Principle ofThematic Uniqueness requires that the subject of frighten and the object of fear do bear the same 0-role. It is necessary, therefore, to express the insight that 'the arguments offear and frighten are thematically the same but . . . differ fundamentally in that for frighten the Theme is a cause, and in the fear-class it is not' (G: 24) independently from 0-structure considerations. This is achieved by introducing a (preliminary) 'causal tier' into the semantic analysis. The 'causal level' is eventually interpreted as expressing what G calls 'aspecrual' information. This conception is based on a level of analysis which is known from the literature as 'event structure analysis'. The version of event structure semantics G adopts is configurational in the sense that (certain types of) events are organized in terms of an internal hierarchical structure, where

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

takes an internal argument in addition to the external' (G: r 6), since the external argument is-almost by definition-the thematically maximally prominent argument.9 How does this apply to the fact that frighten -verbs always express the least prominent argument in subject position (cf (3a))? G argues that the explanation cannot be attributed to 0-marking mechanisms, since 'the VP does not define a domain for e-marking, perhaps because all subjects are generated and E>-marked inside VPs' (G: r6). This statement is hard to understand at first glance. What exactly is the domain of 0-marking? There is no explicit definition provided; just a stipulation is presented: '. . . synthetic compounds define two distinct E>-marking domains with a single predicate responsible for e-marking within them' (G 1 990: p. r6), namely the domain within and the domain outside the compound, with 0-marking having to apply inside the compound first. In order to account for the fact thatfrighten - andfear-class predicates behave differently, e.g. with respect to which semantic role is expressed as the external argument, G concludes that the lexical semantic structure of the respective predicators has to be considered in a more detailed manner. Thematic information is not sufficient to explain the asymmetric behaviour of the predicates under consideration.

Review Article 1 09

certain grammatical phenomena (e.g. instances of adverbial modification) are sensitive to this internal structuring. An example for an internally organized structure of an event is presented in (4):

(4) [EVENT [activity] [state/change of state]) (G: 40)

( s)

(Exp

(Theme)) 2

formally allows for a representation like (6)

(Exp 2

(Theme))

This leads to an undesirable result, since by the same logic structures like (7) a. (Agent (Theme)) 2

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This structure illustrates the structure of so-called 'accomplishment'-predicates (like the verb in sentences like 'Mary built a house'). In order to get a clearer pic ture of the event theory G pursues, the reader should consult Pustejovsky (I 99 I ). Pustejovsky develops an articulated theory ofevent structure representation and what he calls 'event composition', based on the idea that 'words lexically specify a �pecific sequence of sub-events organized by a well-defined geometry' (Pustejovsky I 991: 78). This is represented in (4) by the sequence of a state resp. change ofstate predicate following an activity . 1 1 G is not very explicit about the internal structuring of different event types and their respective motivation, probably because she considers this to belong to the domain oflexical semantics proper. However, if a theory draws so heavily from lexical semantics as G's theory does, it would have been useful to be more specific about the fundamentals of the approach. If we are willing to accept this rather intuitive motivation of event structure and the idea of 'sub-eventual properties' of verbs, this should allow for the configurational definition of certain notions, among them the notion of'cause argument': the cause -argument is defined 'configurationally' as the argument associated with the first sub-event of a complex event-structure associated with the predicate.12 In the case offrighten -verbs, the two dimensions of thematic and aspectual prominence do not coincide, i.e. the most prominent argument in terms of 8-hierarchy (the Experiencer) is not the most prominent one in terms of aspectual hierarchy. The possibility of 'mismatches' of this kind is predictable from the fact that there are two independent levels of semantic analysis which may combine unrestrictedly. Given a matching representation like

I Io

Review Article

b. (Agent (Theme)) 2

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

should exist. However, narural languages obviously do not exhibit patterns like (7b). The combinatorics of a-strucrure semantic levels is obviously too unrestricted since it does not rule our structures like (7b) which would violate the special property attributed to the Agent role: 'Agents and causes are always subjects . . . no matter what their properties might be' (G: 3 I ). At first sight, this generalization exempts Agent arguments from the generalization that 'the aspectual hierarchy determines which argument gets realized as subject' (G: 3 I ). If the latter assumption is correct, there must be a narural constraint ensuring that the Agent argument is always mapped to the aspectually most prominent position. To attribute this property to the very notion of agenthood or causality1 3 conflicts with the earlier assumptions that the observed misalignments of the semantic dimensions are in principle predictable on the grounds of the mere existence of two independent levels of semantic analysis. In other words: there is no convincing (empirical or formal) explanation for the statement that 'since Agents are always maximally prominent on the aspectual dimension, the two dimensions always coincide for Agents, and their subject status is assured' (G: 33), as long as deeper semantic insights into the structure of event strucrure are not taken into consideration. In case offrighten -verbs, where the two levels do not coincide, an external argument is completely lacking. There is a subject, however, since 'aspectual hierarchy determines which argument gets realized as subject' (G: 3 I ). This marks an important difference to unaccusatives which also lack an external argument, but derive the surface subject via NP-movement.I4 Thus, the thematic dimension appears to be irrelevant to subject realization, but it remains necessary in order to account for facts from compounding (see above) and preposition choice. G argues that the specific properties offrighten -class verbs 'are not a matter of lexical idiosyncrasy; it is a principled consequence of the dimensions of meaning in natural languages, and their interaction. Since there are two dimensions of prominence, there is no particular reason why they should always coincide' (G: 30-I). This holds for misalignments in the case of Expetiencer and Theme roles with respect to their aspecrual associations, bur it should also hold for Agent and Theme; G does not provide an argument from the combinatorics of semantic tiers ruling out the possibility of a misalignment in the latter case. In other words: an explanation is required why, for example, in the case of agentive predicates the 'dimensions of meaning' do coincide-it is not sufficient to point our the likelihood that 'the very notion of Agent encodes both aspecrual starus and thematic starus' (G: 3 3 ). Using this argument, the fine grained conceptual and semantic analysis is evenrually abandoned. I suspect

Review Article

111

that the explanation has to be found at the level of conceptual analysis of causation (in the sense of allowing certain event structure representations where sub-events stand in a causal relation). There are, however, important questions left unanswered: I . What is the exact nature of the proposed causal relation ? Where and how is this relation to be represented? 3· Agent and Cause are two very different notions; is the set of generalizations stated so far applicable to agentive predicates?

2.

The definition ofthe concept 'external argument'

One of the most important concepts mentioned above is the notion of'external argument', defined in G's theory as the maximally prominent argument along both semantic dimensions considered so far. Earlier accounts of argument externality depend heavily on notational devices (c£ Williams I 98 I ; Rappaport & Levin I 98 8). These approaches, however, do not offer an explanation of the fact wlzy a certain argument is external. G provides such an explanation in terms of an articulated semantic analysis. The existence of exactly one external argument follows from its definition; there can always be just one maximally prominent argument. It should be noted that G defines the notion of a different level than other authors did before. Whereas in Williams's theory the external argument is defined as a d-structure element, it figures as an a-structure construct in G's theory. This distinction on the basis of representational levels finally charac terizesfrighten -class verbs as lacking an external argument, although a subject is present. There remains a problem, however, with unaccusatives. They also lack an external argument, because the argument of an unaccusative verb presumably 'fails to reach maximal prominence either in the aspectual dimension or the thematic dimension (or both)' (G: 39). The problem is: prominence is a relational notion, and unaccusatives are often monadic. How can prominence be defined at all if there is just one element? In other words: what prevents it from being the maximally prominent argument? A semantic solution, based on lexical conceptual structures as proposed in the theory of Hale & Keyser (I 986), assumes a 'hidden' lcs argument with causal properties. An explicit account of what 'hidden argument' really means and what prevents it from being expressed overtly is not offered. The explanation G gives looks pretty much like the 'diacritic approach' she criticized earlier: 'We might define a maximally prominent argument as one surrounded by a single set of parentheses and propose that the Theme is always surrounded by two sets' (G: 39). The semantic motivation for this notational

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

3 ·4

1 12

Review Article

4 THE A R GU M ENT S T R U CT U RE O F N O M I N A L I Z AT I O N S 4· I

Ambiguous noun classes

The central claim of the chapter on nominalizations can be summarized as follows: so-called 'complex event nominals' select obligatory arguments in order to satisfy their a-structure requirements. Although all nouns have a lexical conceptual representation containing a set of semantic participants , not every noun projects all of these participants into a-structure, where they eventually become grammatical arguments .

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

procedure is derived from aspectual analysis. Unaccusatives belong to an aspectual class denoting either STATES or CHANGE OF STATES, i.e. event types which correspond-according to G-to the second sub-event of the event structure representation of so-called accomplishment predicates. Since the aspectually most prominent argument is defined as the argument being associated with the first sub-event (denoting ACTIVITY), it follows from event structure that an unaccusative verb must lack an external argument. G draws the final conclusion: 'The argument of an unaccusative can never meet the aspectual requirement for externality, because it intrinsically has the wrong status with respect to event structure' (G: 40). Let us briefly summarize the main topics addressed in the second chapter: the internal organization of a-structure is a projection from lexical representa tion. Crucial is the concept of 'external argument', defined on the level of a-structure, making use of the two independent 'tiers' of semantic analysis: thematic prominence and aspectual prominence, the latter being a reflection of the predicate's event structure analysis. An a-structure argument qualifies as external, if it is the maximally prominent one along both dimensions of semantic analysis. Agents are always realized as subjects due to the event structure analysis of the verb, associating the Agent argument with the first sub-event of a complex event structure. In the case offrighten -verbs, the thematically most prominent argument is not the aspectually most prominent one; therefore frighten -verbs do not have external arguments. Independent evidence shows that unaccusatives lack external arguments, too. To make this observation compatible with the theory of hierarchically organized a-structure G develops, we have to assume that the single argument of monadic unaccusatives fails to be maximally prominent in the aspectual dimension, a hypothesis well motivated on the basis of the event structure analysis of the verb, showing that the argument is associated with the second (and hence by definition the non-prominent) sub-event of the complex event structure.

Review Article

113

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Complex event nouns differ from result nominals and simple event nouns by having an internally complex event structure analysis. What appears to be an instance of 'optionaliry' in the expression of arguments reduces to this ambiguity within the nominal system: certain instances of nouns have different readings, where the complex event reading always requires the overt realization of the respective arguments which qualify as being obligatory , 15 the simple event or result reading being grammatical without the arguments being expressed. A prominent example for an ambiguous noun with this respect is examination . The difference between the two classes becomes obvious when the respective nouns are tested by means of various disambiguation techniques, among them event modiflcation , behaviour ofpossessive subjects, and the occurrence of subject-like by-phrases. In the case of event modification, 'the [event, F.B.] modifier forces the event reading' of an ambiguous nominal such as expression . Once this reading is forced, one observes that 'the object of the event nominal is obligatory' (G: so).16 G attributes this effect of event modifiers to the fact that they are licensed by the event structure of the nominal, i.e. whenever it can occur, the nominal being modified has an a-structure which has to be satisfied. However, there is no technical explanation offered as to how this licensing procedure might work, i.e. an articulated theory of event structure based licensing conditions is lacking.17 Ifwe take for granted that some kind oflicensing exisrs and regulates the co occurrence of nominals and event modifiers, how does this relate to the fact that 'constant and frequent [being typical instances ofevent modifiers, F.B.] have other uses . . . which are not associated with an event' (G: p. 5 1)? In these cases, the modifiers are no longer interpreted as event modifiers, i.e. they must be subject to an independent set of licensing conditions. This relates to the question what exactly makes constant and frequent event modifiers: is it inherent semantic properties, or a compositional intepretation being based on the event structure analysis of the modified nominal? If an event modifier is interpreted as an event modifier due to its inherent semantic properties, there must be an answer to the question why there are 'other uses' of the modifier at all. If the interpretation is triggered by the interpretation of the event structure of the nominal, the 'event modifier' should not be able to trigger the event reading. In other words: what is really needed is a convincing theory explaining what makes a modifier compatible with non-event readings of the noun. The behaviour of possessives is a further diagnosis for the distinction between complex event nouns and other nouns. The diagnostic is based on the hypothesis that subject-like possessives are licensed by a-structure, leading to the prediction that 'the presence of a possessive interpreted as subject will force the complex event reading of the noun, where the noun has an argument structure to be satisfied' (G: 5 1). Derailed arguments for the hypothesis that

114 Review Article

subject-like possessives are a-structure related is provided in a later chapter. Important, however, is again the notion of'licensing'; since subject-possessives are optional, one has to explain why they do not act like the other, a-structure related (object) arguments which are not only licensed, but required. Prenominal genitives show an interesting correlation between subjecthood and agenthood that should be mentioned. In a sentence like (8) The instructor's examination took a long time.

(9) The instructor's examination of the papers took a long time. Just like subject-like possessives, subject-like by-phrases are licensed by a-structure. By -phrases may occur, however, also as 'simple modifiers' if they are not construed as subject-like elements. As in the case of 'subject-like possessives', it would have been helpful if G had previously defined her notion of 'subjecthood'. The strucrure of the argument she develops suggests that the notion of subjecthood is rather closely related to the notion of agenthood, a correlation that has often been observed but does not play an explicit role within the theory of grammar, neither in the configurational conception of grammatical functions in standard GB theory, nor within the functional conception pursued in theories like Lexical Functional Grammar. 4.2

Ambiguous noun classes and their syntactic properties

A wide range of syntactic phenomena distinguishing the above-mentioned nominal classes is described on pp. 54-9· Differences in the behaviour concern the determiner system , the behaviour ofmodifiers , and aspectual properties . Let us briefly discuss the problem of 'unambiguous modifiers' (pp. 56-7). G argues that 'some phrases occur only as modifiers and never have an a-structure related interpretation' (p. 57). One example of such an unambiguous modifier is the post-nominal modifier. What seems to be unclear is the following statement: 'Certain possessives can never be interpreted as related to a-structure, since their meaning is such that they cannot contribute information about an argument position' (ibid.). The latter part of this statement opens a new perspective not considered before, namely the contribution which the meaning of a phrase provides 'about an argument position'. What kind of information is referred to here, and how does

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the possessive cannot be construed as a subject. The phrase 'the instructor's' cannot be interpreted as the Agent of the construction without making the sentence ungrammatical. An agentive interpretation makes the object obli gatory:

Review Article 1 1 5

it relate-if inherently being part of the meaning of the possessive-to an a-strucrure position opened by the predicator? The claim that the sentence ( 10) This semester's constant assignment of unsolvable problems led to a disaster (p. 57)

4·3

The lexical representation ofnominals

4.3.1 Event variables and the a-structure of nominals

The important issue of the lexical representation of nominals is addressed in paragraph 3·3· G proposes a non-thematic argument-comparable to the event

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

suggests that 'complex event nominals can take adjuncts but not modifiers' requires further clarification. Equally unclear is why the fact that 'these modifi ers (G obviously refers to temporal possessives, F.B.] are associated only with nouns with no a-structure' qualifies as 'another reflection of the fact that non argument taking nouns refer to individuals and simple events and argument taking nouns refer to complex events' (G: 57). With respect to aspectual modification, complex event nominals are also clearly distinguished from other nominals; whereas complex event nominals 'admit the same aspecrual modifiers as their verbal counterparts', result and simple event nouns lack an internal analysis ofevent structure and therefore do not license aspectual modifiers like infive hours. Although simple event nouns do denote events with temporal extent, they lack the internally organized com plex event strucrure necessary to license an aspectual modifier as quoted above. Again, the notion of'licensing' remains rather intuitive. G does not put for ward a well-articulated theory providing an explication of the interaction between event structure and the syntactic realization of aspectual modifier phrases. Furthermore, explaining the possibility of aspectual modification on the grounds of event structure rather than the level ofa-structure questions the initially presented general picture of the organization of grammatical levels. If d-strucrure is projected from a-strucrure, but aspecrual modifiers are not directly related to a-strucrure but event strucrure, they should not occur on the level of d-strucrure. This is compatible with the well-known conception of d-structure as a pure representation of a predicate's thematic structure; since adjuncts are not E>-marked, they should not occur on the level of d-strucrure. The odd thing is, however, that d-strucrure is the syntactic level most closely related to the lexicon (via a-structure), but event-structure as part of the lexical representation of the item licenses modifiers which occur quite 'far away' from the lexicon in a syntactic representation somewhere between d-structure and s-strucrure.

1 16

Review Article

variable known from a-strucrure representations ofverbs18-which serves as the external argument of nominals. She distinguishes between 1.

2.

a non-thematic argument R for simple event and result nouns which is involved in modification and predication; although members of this noun class lack an elaborated a-strucrure, an open argument defined as being external is present; a non-thematic argument Ev for complex event nouns which is present in addition to the 'thematic a-strucrure, an a-strucrure projection of their lcs participants' (G: 6s); as in case ofR, Ev is characterized as being external

-

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

in order to account for a wide range of syntactic phenomena being connected with the two subclasses of nominals. The explanation offered to account for the characterization of a non-the matic argument as external argument looks rather weak to me. G says that 'nouns always have an open argument. By this reasoning, all nouns have an external argument' (G: 64); why does it follow from the existence of an open argument that this argument figures as prominent? In this context, there is also an asymmetry compared to the definition of the notion 'external argument' provided in an earlier chapter: '. . . we define an external argument as an argument that is maximally prominent in both dimen sions', i.e. in the dimension of aspecrual and thematic hierarchy (G: 35). How does this definition apply to the non-thematic arguments of nominals? The argument seems to be circular when G states later (G: 66): 'Ev . . . is the external argument of a complex event nominal. In the prominence theory of a-struc rure, it must be the most prominent argument, since it is external.' In order to maintain the above-quoted definition of external argument as an a-strucrure notion, it is necessary to justify this in terms of prominence theory one cannot infer maximal prominence from the external starus of the argument, all the more since the notion of thematic prominence seems hardly applicable to an argument being inherently non-thematic. How can the distinction between Ev and R be put to use in order to account for the different properties of the two classes of nominals? 'A noun gets Ev as its external argument if it has an event strucrure' (G: 67); this raises the immediate question how this description is implemented in a technical sense: 'No noun with R as its external can ever have an event strucrure associated with it' (G: 67). Quite obviously, event strucrure is a precondition for the presence of Ev in a-scrucrure; but where does R come from? It seems unlikely to project it from the non-existence of event-scrucrure. In case of morphological derivations, the non-thematic argument is introduced by the respective affiXes: zero derivation unambiguously intro duces R, -ing-nominalization unambiguously introduces Ev and others are ambiguous with respect to R and Ev. Whereas thematic a-strucrure is

Review Article

1 17

4.3 .2 8-Marking properties of argument-taking nominals

3·5 discusses the important issue of E>-marking in argument-taking nominals. Nominals do not select bare NP arguments, a fact G attributes to the hypothesis that nominals are 'defective E>-markers' (G: 70) which require a preposition in order to yield a well-formed E>-marking configuration G (71). Both nominals and prepositions have an a-structure, but complex event nouns 'are not E>-markers', and the prepositions 'that combine with nouns for E>-marking purposes also have a-structure . . . but have no independent semantic roles to assign' (G: 70). This raises the question as to what really constitutes the E>-marking capacity of a lexical item. Assuming that it is the ability to assign an 'independent semantic role' which depends on the presence of a lexical conceptual structure, prepositions lack this ability, since they do not have an lcs representation, whereas complex event nouns should be E>-markers since they have an lcs representation. However, G explicitly says that complex event nouns are defective E>-markers. In other words, the problem is: what makes a preposition a non-defective E>-marker, even though it cannot assign a semantic role, and

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

projected from lexical conceptual structure, non-thematic arguments are present due to a very different procedure, they do nor seem to be projectable. I think the crucial question as to where they eventually come from remains unanswered. G shows convincingly how the differences in the behaviour of the nominals with respect to the determiner system can be reduced to the Ev/R distinction. Her theory is based on the assumption that determiners also have a non thematic a-structure; the determiner the-which can co-occur with both types of nouns-can introduce both Ev and R, i.e. the non-thematic external argument can principally be identified with the external argument of both noun classes. Other determiners unambiguously introduce R, allowing for an identification exclusively with the R argument of result and simple event nouns. What remains unclear, however, is the question how exactly event structure 'licenses modifiers like frequent and constant . . . and it is the event structure that licenses aspectual adjuncts like for a day and in a day'. Does licensing apply via the non-thematic Ev argument, or does it apply directly on the basis of event structure? A terminological detail should be mentioned: R is said to constitute a 'non thematic a-structure' of result and simple event nominals, therefore one should not say that 'complex event nominals have both an event structure and argument structure, and other nominals have neither' (G: 70). The crucial difference with respect to a-structure is that complex event nouns have a the matic a-structure which is lacking in others.

1 1 8 Review Article

1 . arguments which are licensed by a-structure, being obligatory satellites of complex event noun; 2. modifiers which are licensed by predication; they can be separated from the head by the copula, John's dog - the dog isJohn's ; 3· complements which are licensed 'by direct relationship to the lcs', i.e. the complement is-contrary to the modifier-in some way dependent on the lexical semantics of the head; they are found exclusively with simple event and result nouns. G holds lcs responsible for complement selection, a fact which explains why even nominals lacking a-structure have 'selectional control over their complements' (G: 94). She cannot offer a detailed theory of lcs representation, which none the less seems to be absolutely necessary in order to develop a convincing theory of the interaction between lexicon and syntax.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

what makes a complex event noun a defective e-marker even though it can contribute a thematic role due to the presence of an lcs representation? The reason seems to be the failure of complex event nouns to transmit the role directly to the argument, requiring the help of a transferring preposition; bur this transmission failure is merely stipulated, not justified or derived by means of a general principle or representational assumptions. G rejects standard GB explanations which refer to Case Theory as the relevant module, since CP arguments are absent from nominals, although they do not require case marking at all. 19 I will not go into the details of an analysis of CP complements which cannot be arguments, a fact which is concluded both from their optionality and their syntactic behaviour which suggests that 'they are not regulated by a-structure at all' (G: 74). This perfectly fits with the observation that CP complements co-occur with nouns which 'consistently act as result nominals or simple event nominals', i.e. nominals shown completely to lack a thematic a-structure. An unsolved mystery, however, remains the question why and in which sense nouns are 'defective e-markers'. The detailed discussion of passive nominals provides evidence that passive nominals do not rake arguments, a fact which is eventually also attributed to 'the limited e-marking capacity of nominals' (G: 88). However, this statement is merely a descriptive generalization without any explanatory value. To summarize: we need independent evidence not just to claim the 'defectiveness of nouns as 8-markers' (G 1 990: 89), but to justify and derive it from more general principles. One remaining important point concerns the '. . . various satellite phrases associated with non-argument taking nominals' (G: 9 1). G distinguishes three types of'stellites' on the grounds of their licensing conditions:

Review Article I I 9

5 A - STRUCTURE O F P A S S IVES A N D N O M I N A L S

s.r

A-structure and the licensing of'a-adjuncts }

The adequate treatment of adjuncts is one of the most challenging topics ofGB style syntactic theory. G introduces the main differences between adjuncts and arguments as follows: 1 . arguments are licensed by a-structure; 'they can occur only if they are 0-marked by a predicate' (G: 108); 2. adjuncts are not 'licensed by relationship to an a-structure' (ibid.). The immediate question is: what are the licensing conditions for adjuncts? G does not elaborate on this problem, but it seems to be quite clear that the lexical meaning of the predicate plays a crucial role, e.g. the event structure representation of the item in question.21 What requires further clarification, however, is the theoretical status of adjuncts compared to the status of complements. In paragraph 3.6, G defines complements as follows: 'A phrase acting as a complement must be licensed by direct relationship to the lcs' (G: 92), i.e. complements are not licensed by a-structure Gust like adjuncts). To show the difference between adjuncts and complements, we illustrate their different behaviour by examining their co occurrence restrictions: ( 1 1 ) a. b. c. d.

Bill broke the window with a hammer. Bill broke the window yesterday. *The window broke with a hammer. The window broke yesterday.

In the anti-causative alternation, the instrumental PP requires the presence of a cause element in the lcs representation of the verb. This lcs-element licensing the with -phrase cannot be combined with the anti-causative use of the verb. The PP qualifies as complement-due to its association with an lcs

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

G's hypothesis put forward i n this chapter is that the a-structure representa tions of passives and nominals differ from the representations of the corresponding active base verb in that they have suppressed argument positions which license the occurrence of optional satellite phrases, namely passive by phrases and possessive NP phrases. These expressions are called a-adjuncts in order to indicate their intermediate status: in terms of syntactic behaviour, they are similar to adjuncts, but they also resemble arguments because they 'contribute information about positions in the a-structure of predicates' (G: 109).20

1 20

Review Article

argument-whereas the temporal adverb is licensed by different conditions which are independent of the lcs representation. G is primarily interested in optional phrases which are licensed by a-structure. This poses a conceptual problem, since a-structure positions must be satisfied, a fact which contradicts their indisputable optionaliry. She proposes that the 'positions that can license a-adjuncts are those that are

5 .2

Nominalization and passivization: the role ofexternal argument

G provides further evidence to support her hypothesis that possessives and by phrases are not 'true arguments' which need not be reviewed here, but we should have a more detailed look at the topic of 'external arguments and suppression'. The basic assumption defended in paragraph 4·3 is the prediction that verbs lacking an external argument (namely non-agentive psych predicates and unaccusatives) fail to undergo the processes of passivization and nominalization, because 'passivization and nominalization suppress the external argument of a base verb' (p. I 1 2). Since passivization is sensitive only to external arguments, non-agentive psych-verbs (such as frighten ) cannot form verbal passives, but only adjectival passives_23.U G convincingly reanalyses Jackendoff's 'Thematic Hierarchy Condition'.25 The effects of the thematic hierarchy condition follow in G's theory from hierarchically organized a-structure and the 'hypothesis that the passive suppresses an external argument' {G: u s). The passive by -phrase qualifies as a-adjunct, licensed by the suppressed external argument position of the verb. Since the external argument is by definition the most prominent one, the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

lexically satisfied or suppressed' (G: 1 09), i.e. the positions are present in an a-structure representation, but 'not available for purposes ofE>-marking' (ibid.). The notion of 'suppressed positions' is introduced on a rather intuitive level, primarily using notational devices to indicate or mark the suppressed a-position, whereas a technical explication of what it means that an a-structure position is 'lexically satisfied or suppressed' (ibid.) is not at hand. Furthermore, it is not clear why the observed optionality of passive by-phrases and possessive NPs provides 'immediate support for the suppression proposal' (ibid.). What can be inferred from optionality is primarily a difference between a-structure representations of base verb and its nominalization of passivization, but it cannot favour, for example, the suppression approach over, say, an 'elimination approach'22 {or any other technical alternative to suppression). The a-structure framework to account for the optionality of satellite phrases, however, is superior to any approach which derives optionality from different 8-marking principles in nominals of passives. The a-structure theory 'allows us to maintain the same principle ofE>-marking for all argument-taking items' {G: I 1 0) and hence figures as a more unified account.

Review Article 1 2 1 argument figuring as derived subjects is-also by definition-always lower in the hierarchy than the by-phrase. On the one hand, this is a very convincing analysis, since the predictions based on the rather stipulative approach pursued byJackendofffollow from independently motivated assumptions ofa-structure

s .2. 1 -ing-nominalizations In this context,

-ing

nominalizations are quite interesting. They always have a

complex event reading, and since they are 'productively formed only by suppression of an external argument, they are not derivable for the frighten psych verbs' (G: I 2 I ). This should also hold for -ing nominalizations of monadic unaccusatives which lack external arguments, but 'apparently -ing nominalization is possible for certain verb classes with no suppression at all' (G: 122). G admits that it is 'not entirely clear what conclusions' ( I23 ) should be drawn from the data. A proposal to revise the theory she offers looks rather ad

hoc :

If the base has no external arguments, suppression is not necessary. Thus we could view the suppression of an external under nominalization as following from the addition of a new external, which forces a former external to suppress, rather than from the operation of nominalization itsel£ (G: 1 23) G does not give an explanation what is meant by 'addition of a new external', and I do not see any solution as to how this added external argument might 'force a former external to suppress'. This statement becomes clearer rather late in this chapter (G: I 4 I ), when she explains that 'nominalization both suppresses the original external argument and internalizes it in the sense that it adds a more prominent argument that then counts as the external'. In case of complex event nouns, this is Ev, i.e. a non-thematic argument. In this context, the same problem remains: how can a non-thematic argument figure as external in a prominence theory of a-structure? 5.2.2 Adjectival passive formation

A very interesting solution to adjectival passive formation (APF) is developed in paragraph 4·3·3· G presents an alternative to the analysis of APF proposed by

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

representation. On the other hand, the technical details are not spelled out. It is not self-evident, e.g. why the by -phrase must correspond to the suppressed agent argument which functions as licensing element. G shows that 'nominalizations resembles passivization in that the external argument of the base verb is suppressed in both cases' (G: I I 8); she convincingly argues that 'only verbs with external arguments' will form complex event norninalizations (G: I 2 I ). Result nominalizations offrighten -verbs are possible, as are nominalizations of agentive psych-predicates.

1 22

Review Article

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Levin & Rappapon (1 986) which is based on the asssumption that adjectival passives are formed from verbal passives in the course of a relabelling process changing the verb to an adjective, where an internal argument becomes externalized. This solution is not available in a prominence theory of a-structure, and the alternative solution she presents is quite impressive. She proposes a process of conversion which adds an external argument R (which is identified with an internal argument of the base) to the a-structure of the item. This process is inapplicable to items which do not have a genuine external argument. This also means that APF is possible with items which do not allow verbal passivization, since verbal passivization is possible only with verbs which have external arguments which are suppressed in passivization. Whereas verbal passives allow the occurrence of a-adjuncts (by -phrases) being licensed by the suppressed position, adjectival passives require the by -phrase 2 which satisfies an internal argument position of the head. 6 To summarize: a new external argument can be introduced only when a former external argument is either not present at all or suppressed. Assuming that adjectives cannot suppress argument positions, APF should be possible only with base verbs which do not have externals of their own. This seems to be plausible, but poses a problem with transitive verbs which obviously do undergo APF. G proposes that these APFs are formed on the basis of the corresponding verbal passive where the external was suppressed. This would explain why these APFs do not co-occur with obligatory by-phrases; the by phrases are optional, because they are licensed by the suppressed a-position. In case of adjectival passives formed from adjective verbs (which do not have an external argument), the by-phrases are obligatory, since the open argument position (i.e. the position of the internal y-argument) has to be satisfied. One observes, however, that by -phrases are extremely limited in adjectival passives only when they are formed on the basis of verbal passives; this leads G to the hypothesis that the 'optional' by -phrases co-occurring with adjectival passives derived from verbal passives are no a-adjuncts at all. By-phrases occur only rarely in adjectival passives. They are analyzed as internal arguments, which explains that they are obligatory (not optional, as would be expected if they were licensed by a suppressed a-position). A questionable generalization, though, seems to underlie the statement that these by-phrases 'occur only with verbs that do not form verbal passives' (G: 1 27). As G explained before, the adjectival passive is formed on the basis of a verbal passive, i.e. from an active base verb which must have an external argument. Thus, in adjectival passives, by -phrases are analyzed as internal arguments when they occur with adjectival passives. An open question concerns the optionality of PP phrases co-occurring with members of the frighten -class. Causative frighten -class verbs yield adjectival passives where adjectival passivization alters the underlying event structure representation from a

Review Article

I23

complex event structure (consisting of an activity causally related to a resulting state) to a simple state representation. This means that the activity-component is eliminated, i.e. it cannot be projected into a-structure, hence a by-phrase cannot be combined with the construction. The sentence can only have a non agentive reading. In case the by-phrase occurs, it is licensed by a-structure and therefore qualifies as an a-adjunct. 5 .2 . 3 The notion 'obligatory adjunct'

(12) a. *The house was built/designed. b. This house was built/designed by a French architect. These examples illustrate that the passivized verb requires an additional phrase whose categorial realization is nevertheless rather unrestrictedP A question is: does variability of the form suffice to characterize the expression as being an adjunct? One might argue that the uses of examples like (1 2a) are pragmatically highly marked. The information expressed by a sentence like (I 2a) is poor, since the predicate does not provide any further information compared to the informational content of the subject. This new information is presented by the by-phrase (or any other adjunct), which eventually makes-one might argue the sentence pragmatically acceptable. It is possible to construe contexts where a sentence like ( 1 2a) is perfectly acceptable, whereas this 'sentential context' is not related to the event structure of the predicate (see below). One example is:

( 3) This house was built, (and this one was made by magicians) I

G seeks a solution on the level of event structure, assuming that 'verbs that take obligatory adjuncts in the passive have a complex event structure' (G: I 3 3). By stipulation, each of the sub-events of the complex event structure must be syntactically identified, which is not the case in (I 2a) where the Theme argument identifies the resulting state, but the activity sub-event remains unidentified. On the other hand, in (I 2b) the by-phrase identifies exactly this complex event structure component, an observation which amounts to the following generalization: whereas a-structure is altered by passivization, the event structure of the item in question 'remains two-part and still requires the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Paragraph 4·4 critically evaluates approaches which assume that the a-structure of nominals and passives is not distinct from the a-structure of the base verb. According to these theories, either by -phrases or passive morphology satisfy the external argument position of the item. I will not go into the details of the evidence G uses to argue for her solution. However, the notion of 'obligatory adjuncts' she introduces requires a brief comment. The term 'obligatory adjunct' is used to describe the prepositional phrases found in data like G's (67) (G: I 32), quoted here as (12):

1 24

Review Article

identification of both the activity and the (resulting) state' (G: I 3 3 ). This description, however, holds only for verbal passives, since APF does change the event structure;28 this state-component is identified with the Experiencer in examples like 'Mary was frightened'. This is coherent in the light of G's theory. However, one feels slightly uncomfortable with a notion like 'obligatory adjunct' which sounds a bit like a 'round square', at least under the standard definition of the very notion of 'adjunct'.

6

A R G U M E N T STR U C T U RE A N D A N A P H O R A

The explanation of this description is based on the idea 'that the Romance clitics are not arguments but rather valency reducing morphemes added to verb complexes' (G: I 52-3), while pronominal clitization leaves the transitivity of the verb unaltered. Crucial to the valency reducing process is the assumption that a lexical binding process binds the external argument to the internal argument of the respective verb (practically making it intransitive), i.e. 'reflexive cliticization turns a verb with an external argument into a verb with no external argument: the syntactically expressed argument is the internal argument of the verb' (G: I S4)· In other words, reflexive cliticization satisfies an external argument which is no longer available for grammatical processes, a fact which implies the prediction that verbal passivization should not be possible with verbs which have undergone (reflexive) cliticization The paragraph on 'Local anaphora and thematic hierarchy effects' (G: I s8)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I n order to complete this overview, I will briefly discuss the final chapter o fG's book which is aimed to answer a further challenge to the 'structured a-structure approach'. She has to show that the specific anaphoric properties offrighten class verbs (with respect to reflexive cliticization, local and long-distance anaphora) can be explained in the framework she developed, since the configu rational solutions put forward in earlier papers are no longer at hand. One topic in this context is the phenomenon of reflexive cliticization in romance languages where reflexive (anaphoric) clitics cannot be bound by the subject of the respective Romance counterparts of the English frighten -class verbs. Belletti & Rizzi (I988) proposed for Italian that 'the anaphoric clitic si cannot be bound by a derived subj ect' (G: I 52). This descriptive generalization is no longer available in a structured a-structure approach, since here the members of the frighten -class have underlying subjects. G's alternative might be called the 'unaccusative analysis of clitic reflexiviza tion' (G: I S S)· Since frighten-verbs both lack external arguments and do not participate in reflexive cliticization, one might descriptively state that reflexive cliticization 'is possible only with verbs that have external arguments' (G: I 52).

Review Article 1 2 s

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

re-evaluates the notion of a-command which was introduced in order to account for the ungrammaticaliry of sentences where the object of a non agentive member of the frighten-class verbs is an anaphor.29 There are various problems (both theoretical and empirical in nature) with a solution along these lines; therefore G develops an alternative based on type theoretic considerations elaborated in a paper by Partee & Roath (198 3). With respect to their semantic type, the subject of the non-agentive frighten-verb belongs 'to the type of properties ofindividuals' (G: 16o) which is different from the individual type of the anaphor in subject position. This semantic typing applied to the example in ( 1 5) reveals the reason for the ungrammaticaliry independent from binding theoretical assumptions: the types do not match. However, this solution remains incomprehensible since the reader is not able to relate this idea to the topic discussed by Partee & Roath (198 3). These authors are concerned with type ambiguities in the context of conjunction theory; the theory they provide draws heavily from Montagovian-sryle type theory, i.e. a version of formal semantics which does not have any implications for the informal version of semantics G uses. Furthermore, it is by no means obvious how type theory (in the sense of Montague Grammar) might be related to a theoretical explication of anaphoric relations. In G's theory, the unusual behaviour with respect to anaphora naturally follows from the specific a-structure of non-agentive frighten -verbs: The subjects (i.e. Themes) of these verbs are never individuals, whereas Agents are always individuals-a fact which explains why in agentive readings of psych predicates and transitive/ditransitive agentive predicates the ungrammaticaliry is not observed. Psych -predicates differ from other verbs also with respect to long-distance anaphora in that the c-command requirement of Principle A of Binding Theory is obviously violated without yielding ungrammatical constructions. Rather than pursuing a configurational solution, she argues that binding might be sensitive to a-structure. Certain sets of data seem to suggest that long-distance anaphors require subjects as antecedents. However, this is not entirely correct; there are 'cases in which the antecedent is the object of one of the frighten-verbs'. The solution is simple in terms of prominence structure: the external argument must be the antecedent of the long-distance anaphor, i.e. the argument which is always the thematically most prominent one. Reference to the thematically most prominent argument allows the behaviour of'normal' predicates and non-agentive frighten-predicates to unify. In cases where no external argument exists, the most prominent argument in terms of thematic hierarchy functions as antecedent. The more basic criterion determining agenthood, however, is thematic prominence, since external arguments are by definition thematically maximally prominent.

1 26 Review Article

7 C ONCL U S I O N G focuses on the following topics in her attempt to develop a theory of a-structure which is meant to be a convincingly structured alternative to earlier unstructured proposals:

A number of detailed analyses presented in the book look quite promising at first glance: the notion of'external argument' is defined as an a-structure notion which figures prominently in a wide range of syntactic data, covering phenomena from nominalization, compounding, passivization, and anaphora, and the behaviour of psych -predicates seems to be described more adequately than in alternative theories which do not refer to a hierarchically organized a-structure. However, studied more closely, the reader will be disappointed: the introduction of notions such as 'a-adjunct' and 'obligatory adjunct' is empirically motivated, but remains unsatisfactory since the theoretical catego-

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

1 . The structure of a-structure . A-structure representations are structured in terms of thematic and aspectual hierarchy conditions. Empirical evidence is drawn from the behaviour of psychological predicates, their properties in English compounding and E>-marking properties in Japanese light verbs. Furthermore, in a-structure-based explication of the notion of 'external argument' is offered. 2. Nominalizations. Nominalizations are ambiguous with respect to their argument selecting properties. Nominals denoting complex events (where the notion 'complex event' is defined in terms of a configurational theory of event structure) do have an a-structure representation as part of their lexical entry and do require obligatory arguments in order to satisfy their a-structure requirements, whereas simple event nouns and result nominals do not. 3· The argument structure ofnominals and passives . Nouns are defective E>-markers, which accounts for the fact that they never select bare NP complements, but always need the presence of a preposition. In verbal passivization and nominalizarion, the external argument of the active base verb is suppressed, which explains data illustrating the distribution of by-phrases and posses sives which qualify as 'a-adjuncts'. A-adjuncts are licensed by suppressed argument positions. 4· A-structure and anaphora . G shows that specific anaphoric properties observed with the occurrence of psych -predicates can be explained ' by making binding sensitive to thematic hierarchy. Reflexive cliticization is described by means oflexical binding which changes a verb with an external argument into a verb lacking this argument.

Review Article I 27

FRANK BECKMANN

Received: I 5-03-93 Revised version received: I 6-07-93

Sprachwissenschafiliches Institut Ruhr Universitiit Bochum Universitatsstrasse D-4480 1 Bochum Germany

NOTES I I would like to thank Marrin Hoelter and Tibor Kiss for helpful comments on an earlier draft of this paper. 2 I will focus on the conception of a-structure as it is entertained within the framework of'principles and parameters'; alternative versions of a-structure, e.g. Bresnan's interpretation of a-structure within the model ofLFG, will be ignored; c£ Bresnan & Kanerva ( 1 992) for an outline of the 'Lexical Mapping Theory' of correspondences beween thematic structure and syntactic function, a theory relying on a notion of thematic hierarchy;

this concept is also a crucial feature ofG's theory (see below). Alsina ( 1 992) uses Mapping Principles based on thematic hierarchy for an analysis of causative constructions. G presents a theory of passivizarion where this grammatical process is sensitive to the presence of an external argument. 4 C( Jackendoff's Thematic Hierarchy Con dition as a constraint on the interpretation of by-phrases in passive constructions Oackendoff 1 972: 43). 'Thematic roles . . . are purely lexical conceptual labels and do not project into

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

ries as they were defined on the basis of syntactic tests (e.g. the elimination test) or theoretical axioms (8-theory) become fuzzy. This also holds for quite a number of other phenomena leading to terminological confusion and oxymor ons like G's notion of 'optional argument'. The major problem, though, is more serious: G fails to meet the challenge she formulated as the research programme of her study, namely: 'The fundamental goal of this enterprise is to derive a-structure from semantics and then to derive the lexical behaviour of a predicate and its d-structure from its argument structure representation' (G 1 990: 3). Essential semantic notions remain unexplained; this holds primarily for the presented version of event semantics which is not introduced and motivated systematically. Furthermore, G does not show how event structure is put to use in order to derive the syntactic facts she examines. The same is true for the Thematic Hierarchy she uses as the fundamental device for deriving facts concerning synthetic compounds in English orJapanese light verb constructions, etc.: one is forced to draw the final conclusion that the notion of 'prominence' in a-structure representations is not derived from anything, but merely stipulated. This conclusion, if correct, must be disastrous to the programme G pursued, namely developing a sound alternative to underived but merely diacritically enriched a-structure representations.

1 28 Review Article

6

8

9

10

II

I2

13 14

I5

'well-behaving' verbs, eventually forcing the conclusion that the very claim 'that the psychological predicates exhibit a cluster of significantly related gram matical properties must inevitably be denied' (G: 20). However, the basic idea of 'individualizing' thematic roles (as opposed to Thematic Role Types, cf. Dowry I 99 I : 5 50) by means of different sets of verbal entailments looks rather promising to me from a semantic point of VIeW. The example in (4) makes very clear that the version of event semantics G uses differs radically from what Pustejovsky ( 1 991: 48) calls the 'conventional view of events' usually pursued in linguistics based on the work of Davidson ( I 967). The conventional view is characterized by the assumption that the logical form of sentences contains an existentially quan tified event variable; for a recent elabora tion of this theory, the reader is referred to Parsons ( I 990). 'The generalization is that an argument which participates in the first sub-event in an event-structure is more prominent than an argument which participates in the second sub-event' (G: 26). Which would be natural in the sense that the causative sub-event always precedes the sub-event denoting the result. Unaccusative verbs like arrive are syntactically-intransitive verbs lacking an external argument; Burzio's General ization (Burzio I 986) predicts that these verbs fail to assign Accusative Case to their complement NP (i.e. their internal argu ment). This forces movement of the respective NP to subject position (to which nominative case is assigned) in order to satisfy the Case Filter, deriving well formed structures like [IPNPi[VPverb till· The very notion of being obligatory, however, seems to be rather fuzzy. G states: 'Of course, obligatory must mean the same for nouns as for verbs: capable in principle of being obligatory, but perhaps subject to lexical variation' (G: 49). What

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7

the grammatical representation. Thema tic role labels provide convenient termin describing for ology many lexico-syntacnc problems, but they do not figure in solving them' (G: 44). In this respect, G's theory is akin to the position defended by Rappaport & Levin (I988) who also argue that 8-roles have to be eliminated from a-structure. They define 8-roles in terms of positions in lexical conceptual structure. This is a well-known observation in thematic analysis. Assuming that a verb like hit semantically selects (in an intui tive sense) an agent and a patient argu ment inorder to yield a well-formed sentence like John hit Mary, neither John nor Mary bear their respective 8-role due to their inherent semantic properties, but only due to the 8-roles the verb assigns to them. Therefore, thematic roles are strictly relational, depending on the semantic content of the verb and the event described by the sentence. In terms of thematic structure, arguments do not have a 'semantic life of their own'. cf. work of Carlson ( I 984) and Dowry (I 989, I 99 1 ) for proposals concerning a model-theoretic explication of 8-roles. This seems to be the starting point for Fillmore who argues that semantic roles (his 'deep cases') are the reflection of perceptual abilities of humans, i.e. of 'certain types of judgements human beings are capable of making about the events that are going on around them' (Fillmore I 968: 24). Strictly speaking, the external argument is defined as the argument being maximally prominent along both dimensions of prominence, i.e. thematic and aspecrual; see section 3 ·4 of this review. Thematic Reanalysis claims that 'Exper iencer [or alternatively Theme, F.B.) is nor really a unified category: perhaps the thematic role involved forfoar is different from the one involved forfrighten ' (G: I 9). A solution along these lines would assimilate the frighten -verbs to the other,

Review Article I 29 is required in this context is an appro priate definition of 'lexical variation', combined with a set of convincing criteria to determine whether an argu ment is obligatory or not. 16. c£ G's example (7), p. so: a. *The frequent expression is desirable. b. The frequent expression of one's feel ings is desirable.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

17 'Licensing' is a crucial slogan in Chomsky's 'principles and parameters' approach to syntax, claiming that 'Every element that appears in a well-formed strucrure must be licensed in one of a small number of available ways' (Chomsky 1986: 93). The idea oflicensing is spelled out by showing that syntactic elements in strucrural descriptions have to be related in some principled manner to other elements within the same srrucrure without violating any condi tions on grammatical well-formedness. Taking event strucrure into considera tion, it remains to show how this semantically rather than syntactically based level of analysis fits into the syntactic research program of Chomsky's theory. 1 8 This relates to the Davidsonian concep tion of event logic cited above; the point is that Davidson (1967) posrulated an existentially quantified event argument for the a-strucrure representation of verbs in order to allow for certain cases of adverbial modification. This is to dis tinguish sharply from the configurational event strucrure theory posrulated by Pustejovsky (1991); c£ section J . J of this review. However, the event argument G seems to have in mind is quite obviously not interpretable configurationally, but interpretable in the sense of Davidson's approach. 19 With respect to Case Theory and Theta Theory one should bear in mind that a number of phenomena might well be language-specific. In German, nouns and adjectives are usually considered as

having case-assigning properties, e.g.'die Zersti:irung der Stadt', the destruction c!fthe city, with no preposition necessary in order to provide case. Since in German the bare NP argument of Zerstiirung is permitted, one might ask how this relates to the notion of 'defective 9-marker'. It might be possible to conclude that nominals like the one in the example are not defective with respect to e-marking in German. This raises the further ques tion in which sense and to which extent a-strucrure properties are language specific or universal. 20 In paragraph 4·5· G points out that possessive NPs and passive by-phrases underlie slightly different licensing conditions, whereas 'the possessive a-adjunct can be licensed by any sup pressed argument position' (G: 1 3 s). 'a by-phrase can be licensed by any external argument' (G: 1 36), which is, as G shows, suppressed in the case of verbal passiviza tion. 21 c£ (G: 26): 'event srrucrure . . . determines such things as which adjuncts are admiss ible'; I think this statement is too general-there are rypes of adjuncts which are possible due to licensing conditions independent of event srruc rure; c£ also G's hint that 'the roles taken by adjuncts . . . form a kind of secondary argument strucrure not associated with the lexical representation of individual predicates, but constiruting a template to which the adjunct strucrure of the clause must be accommodated' (G: 1 49). This seems to conrradict the above-mentioned dependence of adjuncts from the event srrucrure of the item. The latter charac terization might be resrrictable to senten tial adjuncts. 22 The idea of a-srrucrure position elimina tion is discussed later in paragraph 4·3·3 on adjectival passive formation; G argues that adjectival passives formed on the basis of verbal passive, i.e. the passive of a base verb which contains an external argument which is suppressed in the

1

30 Review Article

(x AGENT

(y)))

THEME

25 This says that a passive by-phrase must be higher in terms of thematic hierarchy than the derived subject. 26 However, the strucrure she presents in (43) (G: 1 25) is ill formed since the brackets do not match. The argument developed so far makes clear that the proper strucrure looks like ((x(y))), a structure which is similar to the strucrure of causative psych-predicates lacking an external argument. Compared to the rypology of a-strucrures on p. 4 1 , one asks

what the proper thematic labelling of 'rim -verbs' would be. In any case, since conversion is applied only ro verbs lacking externals, the strucrure must be as given above. The alternative solution to repair (43) would be to drop one of the opening brackets, e.g. (x(y)) which would make x the external argument, since the 'external argument is one surrounded by only one set of parentheses' (G: 41). 27 A wide 'range of other expressions, including adjuncts of rime, place, manner, and purpose' (G: I 32) can subsri rute the by-phrase. 28 This process G calls 'starivizarion'. 29 The relation of a-command is defined over a-strucrure, essentially saying that 'a more prominent argument asymmetri cally a-commands a less prominent argument' (G: I 59). In non-agenrivefrigh ten -verbs, exemplified in (I 4) ?Politicians depress each other (G: I 58) 'the binding relation violates a-command because the anaphor corresponds to the Experiencer, which a-commands the Theme antecedent, instead of the other way round' (G: 1 59).

REFEREN CE S Alsina, A. ( I 992) 'On the argument structure of causa rives', Linguistic Inquiry, 23, 5 I 756. Andrews, A. D. (I989), 'Lexical strucrure', in F. J. Newmeyer (ed.) Linguistics: The Cam bridge Survey. I. Linguistic Theory: Founda tions. Cambridge University Press, Cambridge, 6o-88. Bresnan,J. & J. M Kanerva ( I992), 'Locative inversion in Chichewa: a case srudy of factorization in grammar', in T. Stowell & ,

E. Wehrli (eds.), Syntax and the Lexicon, Syntax and Semantics, vol. 26, Academic Press, San Diego, 53-102. Belletti, A. & L. Rizzi (I988), 'Psych-verbs and 9-theory', Natural Language and Linguistic Theory, 6, 291-352. Burzio, L. (I986), Italian Syntax, D. Reidel, Dordrecht. Carlson, G. (I984), 'Thematic roles and their role in semantic representation', Linguis tics, 22, 259-79. Chomsky, N. (198 1), Lectures on Government

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

course of passivizarion, lose this sup pressed external argument completely. This fact allows to account for the obser vation that a-adjuncts, i.e. by-phrases which are licensed by a suppressed argument position, are not found very often in adjectival passives. 23 Agenrive readings of these psych predicates, however, do undergo passiv ization, since these verbs contain an Agent which figures as external argu ment which can be suppressed in passivi zarion. 24 The sample strucrure presented in (9) (p. 1 1 3), however, is ill formed: the structure should read as

Review Article Foris Publications, Dor

and Binding,

drecht.

( Ig86), Knowledge ofLanguage: Its Nature, Origin, and Use, Praeger, New

Chomsky, N. York. Davidson, D.

( I967),

'The logical form of

action sentences', in N. Rescher (ed.),

Lagic of Decision and Action ,

The

Clarendon

Press, Oxford.

Dowty,

D. R.

( 1989),

'On the semantic

Properties, Types, and Meaning,

( I99I ),

Kluwer,

'Thematic proto-roles

and argument selection',

Language,

67,

S 47-6Ig.

Fillmore, Ch. ( Ig68 ), 'The case for case', in E. Bach & R. Harms (eds.),

Linguistic Theory,

Universals in

Holt, Rinehart & Win

ston, New York, I -go.

Grimshaw, J. & R.-A. Mester verbs and 9-marking',

( Ig88),

Linguistic Inquiry,

Hale, K.. & S.J. Keyser ( 1 986), 'A view from the middle', Lexicon Project Working

10,

Center for Cognitive Science,

MIT, Cambridge, Mass. Jackendoff, R. ( I 972), Semantic Interpretation in

Generative Grammar,

MIT Press, Cam

bridge, Mass.

Jackendoff, R. ( 1 983 ), Semantics and Cognition, MIT Press, Cambridge, Mass. (Current

8 ). ( 1990), Semantic Structures,

Studies in Linguistics Series,

Jackendoff, R.

Inquiry 17, 623-662.

Levin, B. & M Rappaport ( I988) , 'Non-event -er

nominals: A probe into argument

structure', Linguistics, 2.6,

Marantz, A.

1 o67-1083.

( I984), On the Nature of Gram

matical Relations,

MIT Press, Cambridge,

(1 990), Events in the Semantics of English: A Study of Subatomic Semantics,

Parsons, T.

MIT Press, Cambridge, Mass.

Partee, B. & Rooth, M.

( 1983), 'Generalized

conjunction and type ambiguity', in R. Bauerle, C. Schwarze & A. von Stechow (eds.),

Meaning, Use, and Interpretation of Language, Walter de Gruyter, Berlin. Pustejovsky, J. ( 1 991 ), 'The syntax of event structure', Cognition , 41, 47-8 1. Rappaport, M . & B . Levin ( 1 988 ), 'What t o do with theta-roles', in W. Wilkins (ed.),

'Light

1 9, 205-32·

Papers

'The

Linguistic

MIT Press, Cambridge, Mass. (Current Studies in Linguistics Series, I 8.)

Syntax and Semantics, vol. 21, Thematic Roles, Academic Press, New York.

Speas, M J. ( 1 990), Phrase Structure and Natural

Language , Kluwer Academic Publishers, Dordrecht (Studies in Natural Language

and Linguistic Theory, 2 1 ). Stowell, T. ( I981 ), 'Origins

of

phrase

structure', doctoral dissertation, MIT.

Williams, E.

( 1981 ), 'Argument structure and

Linguistic Review, 1, 81-1 1 4. ( 1987), 'Ebenen lexika lischer Reprasentation', Linguistische Be richte, 108, 83-101. morphology',

Zubizarreta, M. L.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Dowty, D. R.

( I 986),

formation of adjectival passives',

Mass.

content of the notion thematic role', in G. Chierchia, B. H. Partee & R. Turner (eds.), Dordrecht.

Levin, B. & M Rappaport

131

journal ofSemantics

I I:

1 3 3 - 1 48

© Oxford University Press 1 994

Review Article

Propositions, Attitudes, and Russellian Annotations

A critical study of Mark Richard, Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them , Cambridge University Press, Cambridge (Cambridge Studies in Philosophy), 1 990, 275 pages.

Abstract Richard's

Propositional Attitudes

contains a novel theory of belief-sentences in the Russellian

tradition of'direct reference'. It distances itselfcritically from model-theoretic approaches, the tradition of Fregeau sense theory, and also from more psychologically orientated semantics. The theory can be described as a compromise between a referential and a linguistic view of propositions, taken to be fine-grained as in a Structured-Intensions approach. The way terms in a that-clause represent the 'how' of someone's belief are seen as determined by speakers' intentions, this being made explicit by staring restrictions on possible translations of those terms into different contexts. Propositions are formally reconstructed as Russellian annotated matrixes (RAMs), i.e. sequences of pairs consisting of sentential constituents and their respective Russellian interpretations. For a propositional attitude ascription to be true, the RAM of the ascriber's that- clause has to match, in accordance with the contextual restrictions in force, one of the

RAMs in the ascribee's belief system.

The theory is criticized on several counts, most importantly as being incomplete in not sufficiently spelling out how to get to the believer's RAM. Further, it is argued that-due to their hybrid nature-RAMs can hardly be taken as explaining the notion of a that-clause's content.

0

attitude ascription is a sentence whose principal verb is an attitude verb, and the latter is any verb taking singular terms as grammatical subjects and that clauses as objects (p. 7). This initial characterization is quite broad: not only believe , think , wish , etc., but also verbs like remark , assert , calculate , verifY , observe , imply , and others thus count as attitude verbs. (For some of the latter examples, this will invite opposition, since these do not share some of the typical 'psychological' properties of attitude verbs as usually conceived.) The book sets out to treat truth-conditions and ascription of some attitude sentences, mostly English. The only type of attitude considered in detail is An

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

GABRIEL FALKE NBERG

1 3 4 Review Article

belief, and-as the book makes clear-Richard thinks that the results can easily be generalized for all other attitudes. No type of content sentence other than that-clauses is dealt with. Both limitations are deplorable but are not uncommon in attitude semantics, and, of course, to some extent legitimate. What is misleading is the tide of the book. If only the author had simply had the courage to announce a treatment of one English attitude verb in its main grammatical construction. Richard's theory belongs, inevitably for a serious work in analytic philosophy, to the established tradition of Gotrand Fressel, that famous German-British philosopher who once lived, once upon a time and more or less happily ever after, somewhere between Jena and Cambridge. More precisely,

to Situation Semantics, the closest continuer of classical Possible-Wodds Semantics, there is a novel account of propositions, and an extended treatment of belief-retention. The work is, more specifically, connected with and influenced by ideas of David Kaplan, Scott Soames, and Nathan Salmon. Something like a reorienta tion in attitude semantics has been 'in the air' for the last couple of years, coming mainly from people working in the Russellian tradition of 'direct ref erence'. This reorientation distances itself from classical model-theoretic approaches, from the tradition of Fregean sense theory and also from more psychologically orientated conceptual-role semantics. (One may well wonder if this leaves any theoretical room to move.) Richard's book may serve as a welcome introduction to this line of reasoning in the field.

Here is a first, rough oudine of the basic tenets of the theory. Suppose someone says

Maggie thinks that Odile is tired. For this to be true, Maggie must have some 'representation' of Odile and of being tired, these representations being 'put together' somehow in an appropriate way (p. 2 . On the other hand, the speaker's sentence also contains

)

parts representing Odile, and being tired (i.e. the respective words Odile and is tired . So, in some sense,.what makes the propositional attitude ascription true is

)

the fact that there is a match between some parts of the ascriber's sentence and some corresponding things in Maggie's mind. (In fact, as will become clear,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the theory is Russellian with some concessions to Fregeanism, conceived in a Camapian way. (In fact, the term propositional attitude seems to have been 1 coined by Russell.) There is very good criticism of the work of others, there are some interesting ideas about contextual constraints which have some affinities

Review Article I 3 S Richard just reads the structure of the content sentence into the mind.) The corresponding parts on both sides-speaker's as well as thinker's-have to pick out, as it were, the same items in the world. Put in contemporary jargon, both must determine the same Russellian proposition. (Note that this does not yet differentiate the example ascription from other attitudes towards the same proposition.) Richard sidesteps attitude ascriptions containing empty and fictional names, as well as the general issue of 'quantifying out', which involves notorious problems as with sentences like Paul dreamt about the golden mountain, and Peter thought he even saw it, which have defied Russellian treatment so far. He

s),

astonishing in view of his heavily Russellian view on names that appears to determine, or at least to restrict, his position concerning the issue in advance. In fact, he is not able, in the following, to leave this important topic completely out of the picture.

As it stands, the Russellian condition is not sufficient. Frege's examples about the morning and evening star have amply demonstrated that some condition has to be added to the effect that the items represented must be united, for speaker and thinker, under an epistemic equivalence relation like 'same way of thinking', 'same Fregean sense', or something of the sort. Richard first strongly rejects any proposal of this kind, saying that ways of thinking are much to idiosyncratic for the purpose (p.

3). Indeed, in his

core theory, he chooses to

enrich Russellian propositions with linguistic expressions doing duty for whatever cognitive contact is deemed indispensable for belie£ But later, when dealing with Kripke's puzzles, he is compelled to employ a notion of a way of thinking, claimed to be 'thin' (p. 1 5 3) and not necessarily intersubjectively comparable. Furthermore, it is assumed that the match between ascription and things ascribed can vary from context to context. The upshot of this first rough outline is that Maggie believes that p is true in some context iff the content sentence p matches, in this context, something that constitutes a belief of Maggie's. What needs to be developed, then, are the two notions of match and beli constitution .

ef

It is clear that this view involves a strong form of semantic sententialism about the attitudes (p. 4): that-clauses are regarded as designating sentence-like

entities of a particular sort. It also involves a form ofpsychological sententialism

according to which what constitutes a belief is a psychological state that itself has some sort of sentential structure. Richard is prepared to defend both kinds of sententialism, although, concerning the second, he is at pains to be as uncommitted as possible to any particular proposal. What we are told throughout the book about belief states and their constitution is very meagre, and even this is open to criticism.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

maintains, or tries to maintain, strict neutrality on any particular theory about the semantic contribution of definite descriptions (p. which is somewhat

1 36

Review Article

2

Chapter I (Structure) of the book sets the stage by reviewing two other proposals about propositions presently on the market, both being developments in or out of classical Possible-Wodds Semantics. The first proposal is Robert Stalnaker's in Inquiry ( I 984). As is well known, classical Possible-Worlds Semantics like Montague's assumes that propositions have no internal structure; they are conceived simply as sets of circumstances, i.e. items that make sentences true. Two logically equivalent sentences must therefore designate the same proposition, and whoever believes the one must also believe the other. Put differently: any variant of this view requires the attitudes to be closed under logical consequence, something which they clearly are not. (At least, most of them. If to be entitled to know can be held to designate an attitude, it would presumably be one closed under logical consequence. But this case is surely the exception.) This requirement has hampered any real advance on the attitudes by Possible-Worlds Semantics, and has at times been pronounced by critics its theoretical tombstone. Today's communis opinio is that, in order to qualify as entities at which attitudes can be seen as being directed, propositions have to be more fine-grained. The question is only: how fine-grained must they be? As a guide in this search, the following methodological principle is enunciated which is in fact a sufficient condition for difference of things believed: In general, if it is possible that x believes that p be true while x believes that q not be true, then we have to assign the terms

that p

and

that q

to different things (pp.

I 7- I 8).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Richard refrains from giving an analysis of attitude ascnpnons (in the Moore-Church sense of individually necessary and jointly sufficient con ditions) which he deems 'probably impossible' to attain; rather, he rests content with stating, 'in a correct and illuminating way, exactly what situation-what worlds, if you like-are correctly characterized by various uses' of attitude sentences (p. 1 57). It appears that what have arisen above as two basic notions to be explained by the theory both centre around a suitable notion of proposition. Ifsuch a notion can be provided it may serve to elucidate what it is that is being believed, and, therefore, also to define the necessary match between speaker and believer. Or so Richard believes. Central to his account is the bold view that propositions 'have a structure that for all practical purposes is isomorphic to that of the sentences forming their canonical names' (p. 8).

Review Article I 3 7

Incidentally, from this condition it follows that no two belief sentences differing from one another only in the substitution of a synonym must designate the same proposition (since a case can always be constructed in which the believer may happen either not to know the expression q , or associate a different meaning with it). The principle makes the fine-grainedness of propositions towards which attitudes are directed dependent, so to speak, on the lowest limit of psychological narrow-rnindedness imaginable. (How dull, forgetful, logically short-sighted, etc. can we be while still being said to believe that p? It may not be clear whether this question receives a stable and workable

3 Propositions have to be internally structured in order to qualify as objects or contents of the attitudes. Richard therefore turns, as a working hypothesis, to the other extreme line of thought. It is one connected with the Structured Intensions approach developed by Max Cresswell in Structured Meanings ( I 98 5 ). This theory holds, roughly, that propositions designated by that- clauses also have their structure. Propositions are taken to be fine-grained in the same way that sentences are built up .from their constituent parts. This is in effect also what Richard believes, so there is really not much to criticize for him at this point, in a principled way. He rightly points out, though, along with other critics, that Cresswell's theory predicts many ambiguities for which there is no empirical evidence and, by claiming that the complementizer that is the real source of the trouble, locates the problem wrongly. That not any old change in a sentence seems to induce a change in the corresponding proposition was of course already noticed by Frege and, following him, Carnap. What we want is a clear demarcation between those sentential changes that do and those that do not preserve 'propositional identity'. But, as an excellent observation of Cresswell's has made very probable, no general way of drawing the line appears to be forthcoming because changes of wording are attitude-sensitive (p. 27). For instance, the verb say , as used for indirect quotation, seems propositionally indifferent to permutations around the conjunction and, so that in sentences of the form

·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

answer.) But, as a rule of thumb, the principle is certainly of value. A preferable formulation had been given by Gareth Evans ( I 982: I . 5). It is of course related to the Fregean attitude-based criterion for proposi tional identity: p is the same proposition as q iff it is necessary that any sufficiently reasonable creature (in grasp of both p and q) who believes that p also believes that q, and conversely.

1 3 8 Review Article A said that m and r A said that r and m the same proposition can be held to be designated by the two that-clauses. On the other hand, with the verb deduce this is not possible, since one can deduce that r and m without at the same time deducing that m and r (precisely because

4 In my view, Cresswell's point is much underrated and does have important consequences for any theory of the attitudes. First, if the observation is not confined to one isolated example-and it seems that it is not-it opens up the exciting avenue of ordering attitude verbs according to their 'logical strength', relative to some operations in the content sentence. This is not undertaken in the book under review. Attention is only called by Richard, without any elaboration, to those structural operations that do not touch (as it may be called) 'propositional isomorphy' in the attribution of belief: conjunction and disjunction permuta tion, permutation of adjectival or adverbial modifiers, movement of material into or out of a relative clause, permutation of arguments around symmetric predicates (p. 32). Attention should, I think, be paid to the fact that all these 'belief-preserving operations' in the content-sentence involve the same lexical material being moved around (perhaps with the exception of a few syncategorematic morphemes). This is in line with the earlier methodological principle about propositional nonidentity of belief, and its consequences (see section 2 above). A more thorough investigation would be a promising topic of

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

one deduces the first from the second). Suppose, for example, that A, a pupil in an elementary logic course, is asked to derive r 1\ m , given m 1\ r. This provides a reason for denying that the proposition that m and r is the proposition that rand m , at least in the vicinity of deduction, for you can do something to the first proposition (i.e. deduce it) without doing it to the second (p. 30). I will call this observation, henceforth, Cresswell's point. Thus, whereas the verbs say and deduce are both closed under, for example, some adjectival permutation in the content sentences they govern, say is closed under conjunction permutation while deduce is not. This seems to indicate that the second attitude requires a finer notion ofproposition than the first, and that both require a finer than, for example, verify or make it happen (my examples; note that all these are attitude verbs according to Richard's very broad usage recorded at the outset). Although only the first of them appears to be opacity inducing, both seem to permit what neither say nor deduce allow, i.e. existential generalization with respect to contained singular terms.

Review Article 1 39

research for the semantics ofword-order; it is, moreover, in accord with Frege's lifelong attempt at formulating a criterion of sense-identity (his very first eXamples having been active/passive-pairs; c£ van Heijenoon 1977). On the other hand, Cresswell's point may be used to shed sceptical light on the whole enterprise of a theory of attitudes, as it is usually undenaken, since the notion ofproposition sansphrase has apparently to go overboard. Instead, we would have a family of proposition notions, each for a different but related purpose. This would, of course, complicate the whole business of attitude theory enormously; but although Richard refrains from taking this step, it is, perhaps, just what is needed.

Chapter 2 (Some Cognitive Theories of Content) gives Fregean theories of sense and cognitive content their due. After pointing out the difficulties Frege has with quantifying-in and the way the sense of an expression is given and identified, two extended arguments are developed. First, Fregeans have problems with iterated (multiply embedded) attitudes, since, with every new embedding of a that-clause, its designation changes. Second, their treatment makes that-clauses flaccid (as Richard aptly puts it) in a way that excludes proper treatment of valid arguments involving propositional quantification (A believes that p , B believes everything that A believes, ergo B believes that p ) Both shoncomings depend on the same feature of Frege's theory, i.e. that-clauses invariably switch their designation with every new embedding governed by an attitude verb. These sections are well taken and eminently readable; although the arguments as such are known from the literature, the extended way they are put here may well become standard; in any event, they pose serious challenges to a Fregean. (The only criticism I have is that Richard tends to mix Frege's theory with a much more psychological one in which senses are idiosyncratic ways of thinking; as Richard is of course aware, Frege nearly everywhere underscores the objectivity of sense.) Richard also prepares here the way for evaluating his oWn later account, since he claims that it allows us to treat attitude iteration and propositional quantification more successfully. .

6 Richard comes to his own theory in Chapter 3 (Ascribing Attitudes), sections 1 and 2 of which contain the central parts of the book. Anyone versed enough in the literature who is only interested in Richard's core theory will not miss too

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

5

1 40

Review Article

much if(s)he sticks only to these fifty pages or so. One can describe the theory as broadly Carnapian in spirit in being a compromise between a referential and a linguistic view of propositions. A compromise, that is, in the most liberal manner conceivable: propositions are nothing but referential and linguistic items put together peacefully side by side. On this count, propositions are not entities sui generis , but logical constructions out of other entities. At the same time they are claimed, apparently without argument, to be objective and mind independent (p. 3 7). First, Richard argues that attitude verbs must be seen as indexical. By this he means that, for example, the sentence can be both true and false at the same time, only in the mouths of different speakers in different contexts. Here is the story, told in two parts. One ofOdile's friends, Mutt, knows that Odile lists Twain under dead Americans; thus he says she believes that Twain is dead. Another friend of Odile's, Jeff, is asked by someone who couldn't understand why Odile, pointing to a picture of Twain, wants to meet him Doesn't she realize that Twain is dead? Jeff answers: no, Odile does not believe that Twain is dead. According to Richard, Mutt and Jeff are both right, each in his own way. Of course, one may say, the example only works because Odile has two different 'dossiers' of the same man in her 'mental archive' without connecting them, and that Mutt's and Jeff's uses of Twain just trade on these different understandings of the name. Richard denies, though, that the sentence is semantically ambiguous. In some respects possibly important for our intuitions, the example seems underdescribed by Richard. The men could only be held to be both right, it seems, if the latter is assumed to share Odile's false opinion that the man in the picture is not dead. Moreover, Richard more than once switches back and forth, as he does here, between believing and rea lizing,without taking into account the subtler semantic differences between these verbs which seem to play a role. In this case, the example begins to lose, at least for me, some of its initial dramatic impact since the name Twain is, in each part of the story, part of a different 'chain of transmission'. And one may grant that names can be, in some sense, indexical; but from this it does not seem to follow, as Richard wants, that attitude verbs are indexical. Of course, according to some authors, nearly everything in language from natural-kind terms to the truth predicate is 'indexical'. But this is only an inflated way of speaking to which Richard is not party. In fact, I am at a loss to understand, after all, what it could mean for a verb to be indexicaP Decisive for Richard's theory is the way the variations in the two parts of the story are reconstructed a!!d brought to bear on the meaning of attitude

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 ) Odile believes that Twain is dead

Review Article

1.p

(2a) Context Mutt Odile : Twain - Twain (2b) Context Jeff Odile : Twain - that man (in the picture) Each contextual restriction contains three things: a person, an input expression, and an output expression.3 The restriction tells us that, in evaluating an attitude ascription to someone (here: Odile), we are restricted to using correlations that map a particular input expression of ours to a particular output expression (i.e. of Odile). The role of the context in Richard's theory is mainly to provide a collection of constraints on translations. Noteworthy is that context plays a strictly semantic role, not a merely supplementary (waste-basket 'pragmatic') one. By contributing restrictions on how the beliefs of others are identified in our ascriptions, our intentions can be seen to affect the use and meaning of attitude verbs. This comes out well in Richard's remarkable treatment of belief retention (pp. 22o-44), which is the most sophisticated I have encountered and too rich to be summarized here, but largely independent of his particular theory of content. It is, rather, one development of a line of thought from an important earlier paper (Richard I 98 I).

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

ascriptions. What is said to vary, in the two contexts exemplified, is 'what counts as an acceptable translation of the sentences Odile accepts, in a broad sense of "sentence' " (p. I07). This is of course still putting things quite metaphorically. The idea is that there are substantive restrictions on translation at work here: in Mutt's context, Odile's acceptance of a is dead (for some standard name a of Twain) makes sentence ( I ) above true; whereas in Jeff's context, it seems necessary chat she accepts I am pointing at a before Twain can translate a . In the first context, Twain translates Twain ; in the second, Twain translates that one. So Twain and that one have the same Russellian interpretation guaranteed. (Of course, the whole idea of a translation hinges on how the idealization that Odile 'accepts some sentence' is spelled out in the end.) It is assumed that the interests and intentions of the speaker, and to some extent of his audience, are what fundamentally determine the way that terms in a that-clause represent the 'how' of someone's belie£ This can be made explicit by seating the restrictions constraining the translations of those terms into different contexts. In Mute's context, a constraint like 'Twain (Mutt) translates Twain (Odile)' is operative; in Jeff's, it is something like 'Twain Geff) translates that one (Odile)'. Formally:

1 42 Review Article

7 The next step is Richard's development of a novel account of propositions. The proposition a sentence determines is said to contain not only its usual Russellian designations (objects, properties, etc.) but, in addition, all the constituents of the sentence itsel£ We have a sequence of pairs, each pair (also called an annotation ) consisting of a sentential constituent and its Russellian interpretation. These hybrids are called RAMs (Russellian Annotated Matrixes). Thus, the RAM of the sentence

Twain is dead

is

(In fact, this is also the RAM of the

that- clause that Twain is dead.

Richard

nowhere, as far as I can see, provides for an interpretation of the com plementizer

that.)

Every sentence has at least one RAM. The set of all the

believer's RAMs that encodes the totality of facts about her that are relevant to the truth and falsity of belief ascriptions to her is called the believer's

representational system . Since the sentences treated are all of very simple structure, higher grammatical constituents are nowhere dealt with; but in a full statement of the theory, such constituents, together with the grammatical rules that generate them, would appear mandatory. A related, more serious defect is that one does not really see how the insight drawn from Cresswell's point (see section

3

above) is integrated into the theory. The sentences John believed that a fot, ugly man came and John believed that an ugly,fot man came would, on the face of it, receive two different RAMs since order of constituents is decisive in a RAM. One would therefore have expected some notion of 'propositional iso

RAMs)

at work, with respect to the class of morphy' (i.e. an equivalence class of belief-preserving verbal operations. But we are given nothing of this kind and are consequently at a loss to treat linguistically somewhat more sophisticated examples, like the two just cited. In short, the notion of a RAM cannot yet be the notion of proposition wanted for the attitudes; it can only do preliminary duty for it by applying to a particularly chosen diet of examples, and this is the only nourishment we are given in Richard's book. In general, more complex matters oflinguistic form are eschewed, the remaining ones often being treated in a very cavalier manner simply as variants of logical form.

8 In ascribing to Odile the belief that Twain was dead using our above problem sentence 1 ), Odile believes that Twain is dead, what is it that makes this ascription

(

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(3) ((is dead, being dead), (Twain , Twain)).

Review Article 143

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

true, according to the theory? The RAM of our that-clause has to match one of the RAMs in Odile's representational (belief) system, and match it, moreover, in accordance with the contextual restrictions in force. In the example scenario the that-clause of sentence (I) above designates the same RAM in both contexts, namely (3). In fact, the RAM of a that-clause remains the same in all contexts. What changes is not what the that-clause designates, but what it correlates with, i.e. which of Odile's RAMs the speaker's RAM is able to match. Everything, therefore, depends on the explanation of the notions of match and restriction employed. Our RAM (3) above matches Odile's RAM iff (a) the two RAMs amount to the same Russellian proposition (i.e. are both identical as to their objectual parts, given in the same order), and (b) this relation ofRAMs observes all restrictions on our context (expressible as mappings between linguistic expressions) (p. I 40). Thus match would be defined via existential generalization over correlations, i.e. partial functions that map annotations to annotations preserving reference. This looks deceptively simple. Well, it is. What it conceals is the most important problem of all: how and whence do we get Odile's RAM, primarily its linguistic parts? In fact, at no time does Richard spell out the metaphor of Odile conversing in a language of thought to which we as theoreticians have open and unlimited access. Thus we find Jeff's contextual restriction (2b) above also stated as follows: Twain translates the 'names of Twain Odile associates with her current perceptual experience ofTwain' (p. I48). But apart from the fact that it is highly idealized to assume that Odile necessarily associates names everywhere with what she sees, we are never told what these names are and how they are to be identified. (Of course, Richard cannot assume that our RAM, together with our restrictions, suffices to reconstruct the matching RAM of the believer.) At other places, we are told that to believe something is to accept a sentence or a sentence-like entity (p. 47), and that the notion of acceptance is used technically here, but this sense is in fact nowhere further explained. Even in later passages, Richard sticks to his 'let's pretend' way of talking about acceptance as 'a matter of having sentence tokens written inside of the head' (p. I 8 I ). The critical notion of two expressions representing 'the same way of thinking' presupposes the notion of acceptance as already explained (p. I s 3 ). Without a separate theory about Odile's RAM-in other words, without something more of an account of the nature of mental representation-the notion of a match between speaker's and believer's RAMs runs idle, and with it the truth-conditions of the whole attitude ascription. This is my first main objection to the theory; I am only astonished to find that nowhere in his book does Richard openly address the difficulty. The official theory developed later is that believe and other verbs of

1 44

Review Article

pr op os ti i on al atti tu de ar e not simp ly r elati ons taking singu lar terms andRAMs,

bu t actu ally tri adic r elati ons with corr elati ons bu ilt into them as an extr a ar gu mentp lace(p. 1 42):

3 f B(a, thatp ,

f).

Pr op os iti onal qu antifi cati on involves qu antifyi ng over corr elati ons (p. 1 48); these ar e also br ou ght to bear on a solu ti on toKr p i ke's P ierr e Puzz le. Bu t in many cases, they can besuppr essed for simp licity.

9 p hilosop hicalpr ogr ess, or whether it amou nts to a son of'closet Fr egeanism',

with lingu isti c expr essions doing du ty for Fr egean senses, and is evenpr ep ar ed

to gra nt the ep ithet(p. 1 49). In fact, u j st like theFr egean, he sees the tru th of an atti tu de ascr p i ti on as being sensiti ve to facts abou t the way someone thinks

abou t the objects andpr op erti es her belief is abou t. Bu t he seems to over state his cas e when he wri tes: Whether Mutt and Odile associate similar ways of thinking ofTwain with their uses of Twain is wholly irrelevant to the question Does, or Could, Mutt use of

Twain ? In fact, one can

Twain

to represent Odile's uses

comfortably hold the sort of view I am urging and insist that, in

general, interpersonal comparisons of a sentence's cognitive role or sense can't be made. (p.

1

so)

This si pu tti ng matter s too str ongly. O f cours e, Richar d makes nou se of non r efer enti al content or Fr egean sense. Thi s si what allows him, inter alia , a

tr eann ent of qu antify ing- in that is not op en toFr egeans. Instead of sens es, he u ses the noti on of a match between sp eaker 's and believer 's RAM s, sup

p lemented by r estr icti ons ontr ans lati on. B oth match and tr anslati on cer tainly

pr esupp ose a standar d of comp ari son betw een sp eaker and beli ever , i. e. betw een

pu bli c langu age and the believer 's 'langu age of thou ght'. In vi ew of the sad incomp leteness of Richar d's exp lanati on of exactly those noti ons that ar e supp os ed to do the wor k of cogniti ve sens e in his theor y , he is r eally not in a

p ositi on thatpu ts him well aboveFr egeani sm. An additi onalpr oblem whi chis conn ected with evalu atin gRichar d's vis -a

vis a Fr egean accou nt is bur ied in one of the 'techn ical niceti es ' secti ons. It comes ou t wi th iter ated atti tu dessu ch as

(4) J ohn

believes thatH enry beli eves thatp.

T he RAM of its highes t that- claus e servi ng as a r epr esentati on of whatJ ohn

beli eves contains the wor d believe, s ince the second occurr ence of the ver b, the

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Ri char d does anti cip ate thequ esti on as to whether his theor y consti tu tesr eal

Review Article 1 4 5 one inside the that-clause, must go into one of the annotations of the highest that-clause. Richard regards this as a 'technological problem' (p. 1 45), solvable by stipulating a hierarchy of semantic levels. Thus, according to him, there is actually not one verb believe in English, but indefinitely many: believe1, believe2, believe3, etc., each with a slightly different meaning. The theory makes the attitude verb indefinitely ambiguous, as does the orthodox treatment of the truth-predicate. The meaning of believe seems to vanish from our eyes by hierarchical segmentation. No general statement is even possible about all verbs

believen. I find this move counter-intuitive in the extreme, as I find the corresponding level-treatment of the antinomies, for which other, more natural solutions have

is not open for us to use, in representations of the meaning of this verb, the verb itself, in this very meaning, only to claim a moment later that in fact the second verb isn't actually the same, but a slightly different one, and so on indefinitely (and that anyway technology will somehow take care of it). This wavering between downright circularity and endless regress does not seem to me a serious attempt to set the meaning of the propositional attitude verb straight.

10 I think that the example (4) above and Richard's reasoning about it indicate that

·

there is something fundamentally wrong with the notion of a RAM really being, as it aspires to be, a sound and fruitful explication of the notion of a that clause's content. One is inclined to think that any such ontological bastard lacks the necessary internal unity required for this role. This is my second main complaint. Let me explain. Frege held that senses are those entities that we know if we correctly understand the expressions having these senses. When we grasp these senses, and grasp what they are senses of, we know the content of the corresponding expressions. This Richard evidently denies. In his theory, the content of a that clause is a RAM, something in which we find all the pieces of the very that clause which is held to designate that RAM So, according to Richard, not only do expressions such as that-clauses designate things which contain those expressions themselves, but the content of an expression is allowed literally to contain all the parts of the expression itsel£ It must be granted that these are curiously hypertrophic notions of 'designation' and 'content'. The whole language goes into the units of its own semantic interpretation! I think doubts about the legitimacy of this undertaking would remain even if s�mehow it .

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

been proposed in recent years (Kripke, Gaifman 1992). But this is not my main complaint. Remember what we are doing here, as I see it: we are doing natural language semantics, trying to describe the meaning of the verb believe . I think it

1 46 Review Article

(sa) Louis believes that Hector is a Hellene (sb) Louis believes that Hector is a Greek (sc) Louis glaubt, daB Hektor ein Grieche ist Richard allows the second to be false and the third to be true! A sentence and its natural, in fact synonymous, translation into another language are therefore prevented by the theory from saying the same thing any more (c£ also Richard 1 990, esp. iv). Richard is thus prepared to sacrifice the generalization that translation must preserve truth-value (p. 161). This is in line with his view that he is not supplying us with a paraphrase-type of analysis but aims only at giving truth-conditions. Thirdly, the notion of RAM seems to be of doubtful value when applied to other operators on propositions, e.g. it is true that or it is necessary that. The respective RAMs contain extra material that apparently has to be disregarded in the formalization of arguments involving those operators. If not, and if it is held to deal with propositions, logic would no longer be language-independent on this count; instead, we would be faced with a separate logic for each natural language. RAMs are simply too rich to be suitable bearers of truth-values. Where the extreme language-relativity of RAMs also gets Richard into · trouble is, fourthly, in connection with the beliefs of speechless creatures (animals and small children). He discusses these in the book's last section. According to him, successful belief ascription requires 'referential agreement' between the ascriber's that-clause and the believer's state. If the that-clause contains expressions (other than verbs of attitude) which have a particular Russellian interpretation, then the ascription is true only if the ascribee is in a state with a part that determines the Russellian interpretation (p. 252). This remains vague, since no attempt is made to spell out what it is for a part of a state to determine a particular referential value. What is more, Richard grants that this makes most of our workaday attitude ascription to animals false

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

technically works. For me, the general relation between natural-language expressions and their RAMs remains the basic semantic mystery in Richard's theory. What is more, this position makes the notion of content language-relative in an extreme way. The sentences Snow is white and Schnee ist weifi never, according to Richard, designate the same proposition: the first designates an English RAM, the second a German one.4 At first sight, this seems to prevent us from saying that two expressions of different languages have the same content at all. Actually, it only makes it more complicated to say so: Richard needs an additional step, the comparison of the two RAMS and the stripping-off of their linguistic parts. Secondly, his theory is open to Church-like translation arguments: for example, in the following sentences (the first one is assumed to be true)

Review Article

1 47

Acknowledgement I am indebted to Graeme Forbes, Mark Richard, and an anonymous referee for helpful remarks on an earlier version, and

to

Colin Foskett for stylistic advice.

GABRIEL FALKENBERG

Received:

Bergische Universitdt-Gesamthochschule Wuppertal Sonderforschungsbereich zSz 'Theorie des Lexikons' D-42097 Wuppertal Germany eMail:Jalkesjb @ wrcd I.urz.uni-wuppertal.de

Revised version received:

1 5-02-93 30-06-93

N O TE S 1

excellent entry Indexicality in Bright's International Encyclopedia of Linguistics

'I mean by "propositional attitude" any one of those whose expression involves

( 1 992).

propositions-desiring, believing, doubt ing, etc.' (Russell

1 9 1 8/J 986: 268).

The

3

first occurrence of the term in print seems lO

tations; I simplify for expository reasons.

have been in Russell's Introduction to

Ludwig Wittgenstein's genstein

1 922: 1 9);

Tractatus

Ostwald (in

(Witt

Annalen

der Natur- und Kulturphilosophie , 14, 1 92 1 :

95) rranslated 'Satz-Einstellungen'.

2 Richard does not claim

this in his

Actually, restrictions map annotations (as explained in the next section) into anno

4

Incidentally, the German examples that there are in the book abound with grammatical mistakes; their main value seems ornamental.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(p. 259). For example, The dog believes that this is a bone will tum out false, as there fails to be referential agreement with respect to bone and a relevant state or sub state of the dog. Richard does not consider the possibility of relaxing the requirement of referential agreement. He allows some animal concepts and attitudes, though, construing The dogs thinks that the bone is behind the couch in a very uncommitted way as For some x and some y, the dogs thinks that x is behind y . This strongly imputes to the dog only the concept of something being behind another thing; but I do not see the theoretical resources in Richard's account to back up even this very modest claim. Richard's strategy at this point is the one he often follows in the book: refuting some strong arguments against the position he is developing and then letting the matter stand, rather than going on to advance reasons in favour of his theory. That his theory of propositions and propositional attitudes is not downright false or incoherent is one thing; that it is hard to believe, another.

148 Review Article

R E F E RE N CE S

International

Richard, Mark (1 990), 'Comments on Schif

Oxford Uni

fer's Remnants of Meaning', Pacific Philo sophical Quarterly, 71, 223-39.

Structured Meanings.

Russell, Bertrand ( 1 9 1 8/I 986), 'Propositions

Bright, William (ed.) (1 992),

Encyclopedia of Linguistics, versity Press, Oxford. Cresswell, Max J. (1985),

Bradford/MIT Press, Cambridge Mass. Evans, Gareth (1 982),

erence, J. M

The Varieties of Rtf

McDowell (ed.), Clarendon

Press, Oxford. Gaifman, Haim (1 992), 'Pointers to

truth',

MIT Press, Cambridge, Mass.

Heijenoon, Jean van (1 977), 'Frege o n sense identity', journal

of Philosophical Logic, 6,

IOJ-8.

Philosophicus, London.

Richard, Mark (1981), 'Temporalism and etemalism',

Wittgenstein, Ludwig ( 1 922),

Philosophical Studies, 39,

1-1 3·

Routledge

Tractatus Logico & Kegan Paul,

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Journal ofPhilosophy, 89, 223-6 1 .

(Ms), in B. Russell, The Philosophy oJLogical Atomism and Other Essays 1 9 14-19 , J. G. Slater (ed.) (Collected Papers, vol. VIII), Allen & Unwin, London. Stalnaker, Robert (1 984), Inquiry. Bradford/

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents