No title

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

Author: Oxford University Press

8 downloads 370 Views 27MB 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 AN

INTERNATIONAL

SEMANTICS

JOURNAL FOR THE

INTERDISCIPLINARY STUDY OF THE

SEMANTICS OF NATURAL LANGUAGE

MANAGING EDITOR : PETER BoscH (IBM Scientific Centre, Heidelberg and University of Osnabriick)

ASSOCI AT E EDITOR S: NICHOLAS AsHER (University of Texas, Austin) RoB VAN DER SANDT (University of Nijmegen) REVIEW EDITOR : ANKE LiiDELING (University of Tiibingen) ASSISTANT EDITOR: ANKE LiiDELING (University of Tiibingen)

EDITOR I A L BOAR D: MANFRED BIERWISCH (MPG and Humboldt University Berlin) BRANIMIR BoGURAEV (Apple Computer Inc) MARIO BoRJLLO (University of Toulouse) KEITH BROWN (University of Essex) GENNARO CHIERCHIA (University of Milan) A.NN CoPESTAKE (Stanford University) OsTEN DAHL (University of Stockholm) PAUL DEKKER (University of Amsterdam) CLAIRE GARDENT (University of Saarbriicken) BART GEURTS (University of Osnabriick) MICHAEL HERWEG (IBM Scientific Centre, Heidelberg) LAURENCE R . HoRN (Yale University) JoACHIM jACOBS (University of Wuppertal) PHILIP N. JoHNSON-LAIRD (Princeton University) MEGUMI KAMEVAMA (SRI International Stanford)

HANS KAMP (University of Stutt g art) SEBASTIAN Li:iBNER (University of DiisseldorD SIR joHN LYONS (University of Cambridge) jAMES D. McCAWLEY (University of Chicago) MARC MoENS (University of Edinburgh) fRANCIS j. PELLETIER (University of Alberta) MANFRED PINKAL (University of Saarbriickcn) ANToNY SANFORD (University of Glasgow) hRINIM VON STECHOW (University ofTubingen) MARK STEEDMAN (University of Pennsylvania) ANATOLI STRJGIN (Max Planck Gesellschaft, Berlin) HENRIETTE DE SwART (University of Utrecht) BoNNIE WEBBER (University of Pennsylvania) HENK ZEEVAT (University of Amsterdam) THOMAS E . ZIMMERMANN (University of Stuttgart)

EDITORIAL ADD R ESS: Journal of Semantics, c/o Dr P. Bosch, IBM Germany Scientific Centre, Vangcrowstr. 18, D-691 15 Heidelberg, Germany. Phone: (49-6221-) 59-425 1 I4437· Tclefax: (49-622 1-) Email: [email protected]

©

59-3200.

Oxford University Press

All rights reserved: no part of this publication may be reproduced, stored in a retrieval system, or transmitted in

any form or by any means, electronic, mechanical, photocopying, recording. or otherwise without either the prior

written permission of the Publishers, or a licence permitting restricted copying issued in the UK by the Copyright

90 Tortenham Court Road, London WtP gHE, or in the USA by the Copyright Clearance 222 Rosewood Drive, Danvers, Massachusetts 01923, USA.

Licensing Agency Ltd, Center,

Journal of Semantics (ISSN o

167 s 133) is published quarterly in February, May, August and November by Oxford

University Press, Oxford, UK. Annual subscription is US$140 per year. Journal ofSemantia is distributed by

MAIL

America, 1323 Randolph Avenue, Avenel, New Jersey 07001, USA Periodical postage paid at Rahway, New Jersey, USA and at additional entry points.

US POSTMASTER: send address corrections to Journal of Semantia, c/o

Avenel. New Jersey

07001,

USA

For subscription information please see inside back cover.

MAIL America,

2J2J Randolph Avenue,

JOURNAL OF SEMANTICS Volume 15 Number 4

CONTENTS Special issue on Definites Guest Editors: Paul Dekker, Jaap van der Does, Helen de Hoop Editorial Introduction

301

PAUL DEKKER Speaker's Reference, Descriptions and Information Structure

30 5

EDITH KAAN Sensitivity to NP-Type: Processing Subject-Object Ambiguities in Dutch

335

EMIEL KRAHMER AND KEEs VAN DEEMTER On the Interpretation of Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches

355

J AAP VAN DER DoEs AND HELEN DE HooP Type-Shifting and Scrambled Definites

393

Please visit the journal's world wide web site at http://www.oup.eo.uk/semant

Subscriptions: The Journal of Semantics is published quarterly.

Institutional: UK and Europe

£78.oo; USA and Rest of World US$140.00. {Single issues: .

UK and Europe £23.00; USA and Rest of World US$4o.oo.)

Personal:* UK and Europe

£38.oo; USA and Rest of World US$71.00. (Single issue: UK and Europe £u.oo; USA and Rest of World US$21.00.)

*Personal rates apply only when copies are sent to a private address and payment personal cheque/credit card.

is

made by

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.

Back Issues.

The current plus two back volumes are available from the Oxford

University Press, Great Clarendon Street, Oxford OX2 6DP. Previous volumes can be obtained from Dawsons Back Issues, Cannon House, Park Farm Road, Folkestone, Kent Cf19 sEE. tel +44 (o)1303 850101, fax +44 (o)1303 850440. Volumes 1-6 are available from Swets and Zeitlinger, PO Box 830, 2160 SZ Lisse, The Netherlands. 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 University 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 the Journals Subscriptions Department, Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, UK. Telex 8 3 73 30 OXPRES, tel +44 (o)1865 267907, fax +44 (o)1865 26748 5.

JOURNAL OF SEMANTICS Volume

15 (1998)

CONTENTS

Articles NICHOLAS AsHER AND ALEX LASCARIDES Bridging

NICHOLAS AsHER AND ALEX LASCARIDES The Semantics and Pragmatics of Presupposition

239

REINHARD BLUTNER Lexical Pragmatics REINHARD Editorial

BLUTNER AND RoB VAN DER SANDT

PAUL DEKKER Speaker's Reference, Descriptions and Information Structure

I

305

PAUL DEKKER, JAAP vAN DER DoEs AND HELEN DE HooP Editorial Ecc Wh-questions in Underspecified Minimal Recursion Semantics

301

MARcus

STEVEN FRissoN, DoMINmK SANDRA, FRANK BrusARD, GERT VAN RILLAER AND HUBERT CUYCKENS

Flexible Semantic Processing of Spatial Prepositions

37

191

EDITH K.AAN Sensitivity to NP-Type: Processing Subject-Object Ambiguities in Dutch

335

EMmL KRAIIMER AND KEES VAN DEEMTER On the Interpretation of Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches

355

ANATOU STRIGJN Lexical Rules as Hypotheses Generators KEBs vAN DEEMTBR

Ambiguity and Idiosyncratic Interpretation

JAAP

5

VA N DBR OoBS AND HELEN DB HooP

Type-shifting and Scrambled Definites

393

JoANNIS VELOUDIS

'Quantifying' Superlatives and

Homo Sapiens

215

New fron1 Oxford Linguistics A

Theory of

Linguistic Signs

RUDI KELLER Translated by KIMBERLEY DUENWALD

Rudi Keller shows how signs emerge, function, and develop in the permanent process of language change. He recombines thoughts and ideas from Plato to the present day to create a new theory of the

Anaphora, Discourse. C�:Bd Understanding

Evidence from English and French

FRANCIS CORNISH

•

Frands Cornish sets out an original theory of anaphora and deixis, and proposes a way of integrating the potential in-context interpretations of given anaphors within an existing model of discourse representation. 0-1!Hl23648-4. 304 pp., a number ot basic line diagrams, Clarendon Press, April1999 £55.00

0

case Marking

Studies in EUtpsis and Gapping

and Reanalysis

Edited

GrammatiCal Relations from Old to Early Modem English

by SHALOM

LAPPIN

and ELABBAS BENMAMOUN

'This Is an excellent colltction. lt represents divme tluomfall pmpeatves and very much rejkcts the rurrent state of the art in the study of eUipsis.' Ivan Sag, CSU. Stanford University 0-19-512302.0, 320 pp., {OUP USA), Aprtl 1999 £50.00

Grammatical Constructions Their Form and Meaning

by MASAYOSHI SHIBATANI

and SANDRA A. THOMPSON

www.oup.co.ukJcatalogue

0-19-823733-2, 276 pp., 10 figures, October 1998 £45.00 0-19-823795-2, paperback £14.99

0

Fragments

Edited

For further information. contact Andrea Nicholls, Academic Marketing. Oxford University Press, Great Clarendon Street. Oxford OX2 6DP +44 (o)t865 556767 or email: [email protected] 24-hour credit card hotline +44 (o)t536 454534

meaning and evolution of icons and symbols. By assuming no prior knowledge and by developing his argument from first principles, Rudi Keller has written a basic text which includes all the necessary features: easy style. good o�tion.original scholarship. and historical depth.

This collection of previously unpublished papers by a cast of distinguished contributors is a response to and elaboration of Charles Fillmore's and Paul Kay's 'Construction Grammar'. 0-19-823871-1, 368 pp., 7 black and white figures, paperback, Clarendon Press. April 1999 £16.99

C'fJITHIA L ALLEH

'FruitjW interaction between historical Unguistia and linguistic thtory is rarely tht aim and even more rarely achieved ... Allen's work is a r) U d( t/J).The set b( ijJ) of binding variables of 4> is the set d( 4>) n B of binding variables in the domain of t/>. Static Interpretation: Before we tum to the definition of PL; -interpretation, let us first fix some notation. For any relation (or function) r we use d(r) for the domain {x l 3 y: (x; y ) E r} of r, and r{r) for its range { y l 3 x: {x, y) E r}. For-two assignments g and h, and for a set of variables X, we use g[X]h to indicate that g and h differ at most in their valuation of the variables in X, that is, 'Vx E V: if x r/. X then g(x) = h(x) (or both are undefined}. Furthermore, we say that r is a binding relation, if it is a relation r � B X F between bound and free variables. Each element {x, y) E r of a binding relation r is also indicated as Xr· Finally we say that g binds h under a binding relation r, g[r]h, iff d(g) = d(r) U d(h), g[d(r) U r(r)]h, and 'VXy E r: g(x) � h{y). Now we can tum to the definition of PL; -interpretation. All references to a model are suppressed, so the (extensional) interpretation of any predicate R will be assumed to be (exhaustively specified as) [R]. We also suppress reference to the domains from which the information aggregates draw their assignments, so it is left implicit that for all assignments g E [t/>] : d(g) = d( t/>). The definition runs as follows: =

[-.tJ>] [3Xy1/J] [t/> A t/J] (42) (4> � 1/J)

E [R] } {g l -.3h E [t/>] : h 2 g} = {gi 3 h E [1/J] : g[{xy}]h} = {g l 3f E [t/>] :g 2f & 3 h E [t/J]3h' [b {t/>)]h:g 2 h'} = -.(4> A -.tfJ ) ; (4> V 1/J) = (-.4> � 1/J); 'Vxy t/> = -. 3Xy-. tJ> =

Comments: The above definition is pretty straightforward, except, possibly, for the last two clauses. Atomic formulas are interpreted in the usual way. The negation -.4> of a formula 4> involves unselective binding of all the binding variables ·in t/>. Thus, the interpretation of -. 4> consists of all those assignments to the free variables which have no extension in the interpretation of t/>. In the clause dealing with an existentially quantified formula 3xy t/> we see that the information which 4> has about the value of the free variable y is effectively transmitted to the binding variable x. In the case of a conjunction a choice had to be made. We could have chosen to take those assignments to the variables in the domain of 4> and 1/J which extend assignments in the interpretation of both 4> and 1/J. This would have had the effect that the values of a binding variable which figures in the domains of both 4> and 1/J get identified. We don't want this to happen, and therefore we had to decide which occurrence of the variable-if indeed any one of them-would have to contribute to the whole conjunction. According to the above definition the first 'declaration' of x wins: we do not use extensions g of h in [1/J� itself. but extensi�ns of those assignments h' which can be obtained from h after quantifying away binding variables in 1/J which are also in the domain of t/>. Clearly, other options are available. PLi and First Order Predicate Logic: The presently sketched system of PLi does quite the right things for the purposes of the present paper. The system generates the kind of information aggregates needed for a treatment of reference and specificity along the lines sketched in section 5 · Besides, it is still intimately related to ordinary first order predicate logic (PL0). Consider a language of predicate logic ·built up using the relational constants from PL; and its set F of free variables. We define the following translation ( )b from PLi into PL0:

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

(41) [Ry, . . . y.] = {g i {g(y, ) , . . . , g(y. ))

328

Speaker's Reference, Descriptions and Information Structure

By induction on the construction of cjJ it can be seen that

(44) [l]PL•

=

{f E D F / 3g 2 f: g E [c/J]PL; }

Obviously, we can as well define a translation ( )a from PU into PLi, in which the only U non-trivial clause. has it that ( 3yc/J) = 3 .xy(c/J)a, for an arbitrary variable x E B not used in ( cjJ Then it is easily established that:

l

(4 s) [ct>DPL·

=

{f E D F / 3g 2f:g

Dynamic Interpretation:

E

[ct>#]PL; }

Since the formulas of

PL;

are interpreted as information

free variables in one aggregate are bound by which (binding) variables of another. To this end we use resolvers r which we called 'binding relations' above. We assume that resolution takes place when two aggregates merge, 30 and we therefore hang the resolver as an index on the conjunction sign which initiates the resolution. So conjunctions will now be of the form q, 1\, 1/J, where r is a resolver. The system of interpretation which uses this notion of conjunction is referred to as PL•. Before we define the interpretation of the resolving connective, we first settle the domain of the new type of formula ¢i I\, 1/J: d( c1> 1\, '1/J) = d( c/>) U d( r) U d( '1/J ). The interpretation is specified as follows:

(46). [c/> A, 'I/J]

=

{ g / 3f E [c/>] :g 2 / & 3 h

E

['I/J] 3h'(b( c/>)]h:g(rjh'}

Comments: Firstly, notice that we have interpreted our resolver as indicating which bound variables in the left conjunct bind which free variables in the right one. Of course, there are various alternative ways to go about here, but we have chosen to just stick to the practice of allowing left-to-right binding only, which is relatively standard in systems of dynamic semantics. Notice, secondly, that we have imposed no restrictions on our relation r. For this reason we may find both vacuous binding (e.g. when some xy is in r and y does not occur free in 1/J) as well as vacuous resolution (e.g. when Xy is in r and x does not occur in a binding position in 1/J). Neither possibility presents any problems, but it may have to be pointed out that

vacuous resolution amounts to in situ (existential) quantificatioiL Finally, notice that if r is the empty relation, then /\, equals the interpretation associated with I\ in PLi. This means that we are still dealing with a proper extension of ordinary

predicate logic. The reader may verify for herself that, nevertheless, this extension analysis enables an account of intersentential anaphoric relationships. Consider the following conjunction:

(47) (3x.(Fu I\ 3 z.., (Dw I\ Ouw)) 1\ {x.,z.} Bee) The second conjunct Bee here contains two free variables, which the resolver r =

{x" z, }

binds to the binding variables x and z, respectively. The conjunction thus turns out to be equivalent with the following existentially closed formula:

(48) 3x. (Fu I\ 3z.,(Dw I\ ( Ouw I\ Buw)))

PLa and Dynamic Predicate

Logic: The system of PL" enables a static treatment of

intersentential anaphoric relationships, and, therefore, it is also related to dynamic predicate logic

(DPL), this

in a fairly straightforward way. In order to simplify the


aggregates, their interpretation already displays sufficient structure to enable an account of intersentential anaphora. In order to work this out we need a device telling us which

Paul Dekker comparison, we build on three assumptions. In the first place we assume our resolvers

329 r

to

correspond to partial functions from their range to their domains. This is to say that .there are no pairs Xy and Zy in r such that x =/:- z. 3 ' In the second place we assume that resolvers only bind free variables to binding variables which have already been declared. That is to 4> say that for any conjunction 1\, '¢, the domain d( r) of the resolver r is a subset of that of

4>.3' In the third place we assume that binding variables are used at most once as a binder. This assumption enables us to avoid certain technical complications which are completely artificial when we look at things from the point ofview of a natural language semanticist. JJ Now consider the following translation ( ) t from PL" into DPL: (49) (Ry,

. . .y

(..., 4> )t

where

.

) t = Ry, · · · Yn = ..., (4>)t

[x/y]'¢

(3x ,P) t

(4> 1\1, ,p)t

=

3x([x/y]'I/J) t

=

(4>)t 1\ ([r] ,p)t

is ordinary substitution of

4>

of all

x for y

x

for y in '¢, and

such that

x1 E r

[r]'¢

involves the

Usingf =x g to indicate that f and g agree on the valuation of all the variables in X, it can now be seen that:

This equation may serve to show that dynamic semantic results can be obtained by a mere extension of classical predicate logic, an extension with aggregates and a resolver, that· is.

RE FERENCES Aloni, M. (I997), 'Quantification in dynamic semantics', in P. Dekker, M Stokhof & Y. Venema (eds.), Pro

ceedings of the Eleventh Amsterdam Collo quium, ILLC, University of Amsterdam,

Amsterdam. Boer, S. & Lycan, W.

(I986), Knowing Who,

MIT Press, Cambridge, MA. Chastain, C. (I975), 'Reference and context', in K. Gunderson (ed.), Language, Mind,

and Knowledge, University of Minnesota Press, Minneapolis. Dekker, P. (I996), 'The values of variables in dynamic semantics', Linguistics and Philosophy, 19, J, 2 1 1-57. (I999), 'Meaning

Dekker, P.

and

use

of

indefinite expressions', TS, University

of Amsterdam.

Donnellan, K. (I996), 'Reference and defi nite descriptions', Philosophical Review

75, 28 I-J04-

Donnel!an, K. (I978}, 'Speaker reference, descriptions and anaphora', in P. Cole (ed.), Syntax and Semantics, Vol 9: Prag matics, Academic Press, New York, 47-68.

Evans, G. (1982), The Varieties of Reforence, Clarendon Press, Oxford, edited by John McDowell. Fodor, J. & Sag, I. (I982), 'Referential and

Linguistics and Philosophy, 5, 3, 355-9 8. Frege, G. (I 892), 'Ober Sinn und Bedeutung', Zeitschrififur Philosophie und philosophische Kritik, NF IOO, 25-50. Geach, P. T. (I962), Reforence and Generality, quantificational indefinites',

Cornell University Press, Ithaca. Gerbrandy, J. (I997), 'Quantification in epistemic semantics', in P. Dekker, M Stokhof & Y. · Venema (eds), Pro

ceedings of the Eleventh Amsterdam Collo quium, ILLC, University of Amsterdam, Amsterdam, I 33-8 . Geurts, B. (1995), 'Presupposing', Ph.D. thesis, University of Stuttgart, Stutt-

gart. . Groenendijk, J. & Stokhof, M (1980), 'A pragmatic analysis of specificity', in F. Heny (ed.), Ambiguities in · Intensional Contexts, Reidel, Dordrecht. Groenendijk. J. & Stokhof, M (1991},


simultaneous substitution in

3 30 Speaker's Reference, Descriptions and Information Structure 'Dynamic predicate logic', Linguistics and

Philosophy. 14.

1,

39-100.

1 1: Presupposition, Academic Press, New York, 1-56. Kasher, A. & Gabbay, D. M (1976), 'On the semantics and pragmatics of specific and non-specific indefinite expressions',


Groenendijk. J., Stokhof, M & Veltman, F. (1995), 'Coreference and contextually restricted quantification', Proceedings of Theoretical Linguistics, 3, 1 45-90. SALT V, DMLL Publications. Heiin, I. (1983), 'File change semantics and Klein, E. ( 1979), 'On sentences which report beliefs, desires and other attitudes', the familiarity theory of definiteness', Ph.D. thesis, University of Cambridge, in R Bauerle, C. Schwane & A. von Cambridge. Stechow (eds.), Meaning, Use, and Inter pretation ofLanguage, de Gruyter, Berlin. Kripke, S. (1972), 'Naming and necessity', in D. Davidson and G. Harman (eds.), Heim, I. { 1989), The Semantics of Definite Semantics of Natural LAnguages, Reidel, and Indefinite Noun Phrases, Garland, Ithaca. Dordrecht, 254-3 S s, 763-69. Hendriks, H. {1993), 'Studied fleXibility', Kripke, S. (1979), 'Speaker's reference and semantic reference', in P. A. French, T. E. Ph.D. thesis, University of Amsterdam. Jackendoff, R (1972), Semantic Interpretation Uehling & H. K. Wettstein (eds.), Con temporary Perspectives in the Philosophy in Generative Grammar, MIT Press, of Language, Chs. 6-27, University of Cambridge, MA. Janssen, T. M. {1983). 'Foundations and Minnesota Press, Minneapolis. applications of Montague Grammar', Lewis, D. (1975), 'Adverbs of quantifi Ph.D. thesis, University of Amsterdam, cation', in E. L. Keenan (ed.), Formal Amsterdam. Semantics, Cambridge University Press, Kadmon, N. (1987), 'On unique and non Cambridge. unique reference and asymmetric Ludlow, P. & Neale, S. (1991), 'Indefinite quantification', Ph.D. thesis, University descriptions', Linguistics and Philosophy, 14, 2, 171-202. of Massachusetts, Amherst. Kamp, H. ( 1 984). 'A theory of truth Montague, R (1974), The proper treatment and semantic representation', in J. of quantification in ordinary English', in R Thomason (ed.), Formal Philosophy: Groenendijk, T. Janssen & M Stokhof (eds.), Truth, Interpretation and Information, Selected Papers of Richard Montague, Yale Foris, Dordrecht. University Press, New Haven, 247-70. Kamp, H. (1990), 'Prolegomena to a struc Neale, S. (1993), Descriptions, MIT Press, tural theory of belief and other atti Cambridge, MA. tudes', in C. A. Anderson & J. Owens Partee, B. H. (1972), 'Opacity, coreference, (eds.), Propositional Attitudes, CSLI, and pronouns', in D. Davidson & G. Stanford, 27-90. Harman (eds.), Semantics for Natural Kamp, H. & Reyle, U. (1993), From Discourse LAnguage, D. Reidel, Dordrecht, 41 5-4 1. to Logic, Kluwer, Dordrecht. Quine, W . (1956), 'Quantifiers and propo Kaplan, D. {1 989), 'Demonstratives: an essay sitional attitudes', journal of Philosophy on the semantics, logic, metaphysics, 5 3, 1 77-87. Reprinted in W. V. Quine and epistemology of demonstratives {1966), The Ways of Paradox and Other and other indexicals', in J. Almog, Essays, Random House, New York. J. Perry & H. K. Wettstein (eds.), Recanati, F. {1996), 'Domains of discourse', Themes from &plan, Oxford University Linguistics and Philosophy, 19, 445-75. Press, New York, 481-563. Rooth, M (1985), 'Association with focus', Karttunen, L. & Peters, S. {1979), 'Con Ph.D. thesis, University of Massachusetts, ventional implicature', in C.-K. Oh & Amherst. D. A. Dinneen (eds.), Syntax and Semantics Rooth, M (1992), 'A theory of focus

Paul Dekker 33 I interpretation', Natural Language Seman tics, I, I, 75-1 16. Russell, B. (I905), 'On denoting', Mind, XIV, 479-93· · Russell, B. (I957), 'Mr. Strawson on referring', Mind, 66, 3 85-9. Stalnaker, R (1974), 'Pragmatic presuppo sitions', in M Munitz & P. Unger (eds.), Semantics and Philosophy, New York

59. J2Q-44·

Scr:iwson, P. F. (r964), 'Identifying reference and truth-values', Theoria 30, 96- u 8. Vallduvi, E. (I992), The Informational Component, Garland, New York. van Deemter, K. (I992), 'Towards a generalization of anaphora', journal of

Semantics, 9, 27-5 I.

van Rooy, R (I 997), 'Attitudes and Changing Contexts', Ph.D. thesis, IMS, Stuttgan.

journal ofLogic, Longuage and Information, 2, 2I7-54· Vermeulen, C. F. (I996), 'Merging without mystery or: Variables in dynamic seman

tics, journal of Philosophical Logic, 24. 4, 405-50. Westerscihl, D. (I984), 'Determiners and context sets', in J. van Benthem & A. ter Meulen (eds.), Generalized Quantifiers in Natural Language, Foris, Dordrecht, 45-71. Whitehead, A. N. & Russell; B. (I9Io), Principia . Mathematica, Vol. I, Cambridge

University Press, Cambridge. Zeevat, H. (1989), 'A compositional approach to discourse representation theory', Linguistisc and Philosophy, I2,

95- I 3 I .

Zimmermann, E. (1998), 'Remarks on the episternic role of discourse referents', in L. Moss (ed.), Proceedings of ITALLC 9 6.

NOTES I I am quite generically referring to the well-known systems of discourse repre sentation theory (DR T, Kamp I 984; Kamp & Reyle I993), flle change semantics (FCS, Heim I983, I989), and dynamic predicate logic (DPL, Groenendijk & Stokhof I99 1 ), and all those akin in spirit or by offspring. 2 Their, causal and intentional, relation with the real individuals around us will be addressed in the next section. A note on terminology is in order, since we will be dealing with objects from different realms, or, as we would prefer to put it, with objects conceived from different perspectives. We try to consistently use the terms 'individuals' and 'entities' for the objects out there in the real world, and for the residents of the basic domains of our model theory. The term 'subjects' is used for the episternic representatives of individuals,

and the term 'objects' for things of both orders. The last term is meant to play a communicating pan. If subjects, which ar� panial objects, have grown into total objects, they correspond to real indi viduals. Notice that what are individuals from one's own episternic perspective may be · subjects from someone else's perspective or from the meta-theoretical perspective. 3 See Heim (I983) and Dekker (I996) for funher motivation and study. Alterna tive formulations, which fit the pur poses of this paper equally well, can be found in Zeevat (I989), Vermeulen (1993), Groenendijk et al. (1995) and Dekker (I 999). All of these formulations are evidently inspired and �otivated by Karnp (1984), Heim (1989) and Karnp

(1990)· 4 It is a useful simplification to think of these variables as the variables of a


University Press, New York. Stalnaker, R (I978), 'Assertion', in P. Cole (ed.), Syntax and Semantics 9: Pragmatics, Academic Press, New York, 3 I 5-32. Strawson, P. F. (I950), 'On referring', Mind,

Vermeulen, C. F. (I993), 'Sequence semantics for dynamic predicate logic',

332 Speaker's Reference, Descriptions and Information Structure

5

7

8

9

IO

II

I2

I3

I4 I5

I6

I7

first attack by Strawson: 'The mam purpose of the article [Strawson I950, PD) is to refute my theory of descrip tions. As I find that ·some philosophers whom I respect consider that it has achieved its purpose successfully, I have come to the conclusion that a polemical reply is called for' (Russell I957= 385). Simply observing that the explicit restrictions on quantifying noun phrases should be understood to be contextually restricted themselves is not the same as offering an analysis of course, and we do not offer one here. Let us suffice with pointing out that such mechanisms can be of various kinds. They may consist of contextually restricted quantification Westerst:ihl (I 984), possibly related to the dynamics of interpretation (Groenendijk et al I995); it may employ suppletion of 'elided' (syntactic) material (Ludlow & Neale I99I; Neale I993), bridging (van Deemter I992; Geurts I995) or other 'primary' pragmatic processes (Recanati I 996). The much less . widely known Strawson (I964) presents a m�ch more sophisti cated view of the matter; we will return to that paper in due course. Geach himself qualifies the kind of 'reference' at issue as 'of negligible importance for logic'. He mentions it 'only to get it out of the way' (Geach I962: 8). We will not go into the (non-)scopal properties of specific indefmites in this paper. Thus, our analysis goes against e.g. Donnellan (I978), Kasher & Gabbay (I976), Partee {I972), Fodor & Sag (I982), and is in line with e.g. Kripke (I979), Klein (I979), Groenendijk & Stokhof (I 980), Ludlow & Neale (I 99 I). If it turns out that none is missing. or more than one, clearly, the guard and the police will have to adjust their assumptions concerning this subject. For recent approaches to the issue of


6

language which is to be interpreted. However, as we will see in the next section, it does not at all matter what is used as a set of variables here. See Vermeulen (1996) for relevant discussion. We want to distinguish any two aggre gates which contain information about different sets of variables, 0 E Ex and 0 E E Y for any X =f. Y. We therefore stipulate, for any X, that there is one empty set 0x E Ex the domain d(0x) of which is X. In Montague (1974) such a distinction is made between two classes of variables (of type ( s, e )) for similar techno-philosophical reasons. Herman Hendriks aptly observed this, as well as the connection with the interpretation of anaphora {Hendriks 1993: Ch. I, fns. 5 and 20). This has not always been the idea behind subjects or 'discourse referents'. They have been conceived of as partial objects per se, by me, for instance. They have also been considered to be mere remnants of the noun phrases which have introduced them (c£ Heim 1989; Groenendijk et al. I 995). The outlook upon subjects expressed in this paper has its roots in Chastain {1975), Donnellan {I978) and Evans (I982), and in a more dynamic setting in Kadmon (I987) and Kamp (I990). For further motivation see van Rooy (1997), Zimmermann {1998) and Dekker (1999). See, among many others, the work of Boer & Lycan, Chastain, Donnellan, Evans, Kaplan, and Kripke to get an idea of the wealth of facts and concep tual problems which a full theory in the end should cover. In Frege (I892) definite descriptions are conceived of as compound proper names, the use of which presupposes reference. The possibility that such a presupposition is not fulfilled, is called an 'Unvolkommenheit der Sprache'. c£ e.g. Russell's (belated) reaction to the

Paul Dekker· 'knowing-who' within the context of a dynamic or epistemic semantics, c£ e.g. Aloni (I997) and Gerbrandy (I997).

IS

24

Original and motivating ideas behind such a distinction can be found in Jackendoff (I972: Ch. 6, linguistic), Stalnaker (I974, I978): philosophical), Karttunen & Peters (I 979: Montagovian), Rooth (I985, I992: focus), and Vallduvi (I992: discourse). It can be said to be anticipated by the early and under estimated Strawson {I964). Strawson

indefinite

2I

22

in the Appendix. Similarly, in reply to the question 'Who visited the exhibition yesterday?' the answer 'The king of France did.' is

For

many,

it

seems,

25

26 ·

& Peters-style

and to be saying of that man that he sells wooden shoes. This type of refer ence (Kripke I979) calls 'semantic reference'. In both cases we have construed the background as involving two objects, about which the mentioned predi cations are asserted. Alternatively, . it seems reasonable to construe the

(3 I)

A man who is a woman, is stretching his strings against an oak which is a birch.

27

One might add: .' . . . but I agree, he is old and he has been made full professor.'

28

So we say that an atomic formula Rt, . . . t. is satisfied relative. to a model M and an assignment g if and

only if the interpretation [t;JM, ( v) ,g of g any term t; is defined and [t,JM, (•), ' . . . ,

g g [t.JM,g(v),g stand in the relation R ing(v)

in M. Thus, a sentence 'Pegasus walks.' is deemed false, and its negation 'It is not true that Pegasus walks.' is deemed true.

theory of pre

have been developed afterwards, some

The utterance can be understood to be about the man, whoever he is, who is the one and only man in the domain of quantification who has a martini,

analyse the pronouns in Leo's replies. As Friederike Moltmann has observed, an E-type analysis certainly wouldn't do the job, since it would render Leo's reply two-way contradictory:

this problem

supposition, and all kinds of dynamic or discourse representational alternatives

example

utterance as being about a (perceived) situation. However, upon such an analysis it is not at all dear how to

has meant the death warrant for a Karttunen

the

someone succeeded George V, and to presuppose that (s}he had a hard time doing so.

false, intuitively speaking, while a state ment like 'As for the king of France, he visited the exhibition yesterday.' is pointless, intuitively speaking.

23

expressions,

29

Unlike truly free variables, they can be

bound by adverbs of quantification or by a negation operator.


20

if the term is 'topical'. We realize that more intricate divisions and classifications are needed, and have been made, and that the interaction between assertion and background and between background and context is much more involved than we can account for in the present context. What it means for a background to be 'appropriately connected' with the context can be spelled out in a variety ways: truth, definiteness, familiarity, binding or resolution . . . We will not commit ourselves to any specific implementation of the notion in this paper. In both examples the free variable x in the assertion must be matched with the existentially quantified variable in the background; a more sophisticated way of resolving free variables can be found

o f which relate back t o earlier work by Karttunen. Given our analysis of the use of

can just as well be seen to assert that

seems to have been the first to observe that a term presupposes reference (only)

I9

3 33

3 3 4 Speaker's Reference, Descriptions and Information Structure 30 Apparently, this is not the only sensible

option to be considered. Many (dynamic) theories of interpretation build on the assumption that resolution takes place before that; Bart Geurts (p.c.) has argued that it has to take place afterwards. 3 I This is a harmless assumption really because any resolver which is not functional in this sense, can be replaced, in a meaning-preserving way, by one

which is, simply by adding a couple of equations. 32 This, too, is a harmless assumption, because resolvers which do bind to undeclared variables can be replaced, in a meaning preserving way, by ones which don't, by adding existentially quantified formulas. 3 3 The interested reader is referred to Dekker ( I 996, I 999) for more discussion.


Joumt�l oJ&mantia

© Oxford University Press

I S: 3 l S-3S4

1 998

Sensitivity to NP-Type: Processing Subj ect-Obj ect Ambiguities in Dutch EDITH KAAN Utrecht Institute of Linguistics Abstract According to some theories of sentence processing, the human language processor relies

hi this p�per I

discourse related properties. Dutch 'which' clauses are at least locally ambiguous be�een a subject-object and an object-subject reading. On the basis of syntax-based parsing strategies (e.g. the Active Filler Strategy, Frazier

1987),

a subject-object preference is

expected. However, several on-line and questionnaire studies show that the type of �econd

NP affects the order preference: when the second NP is a non-pronominal NP the subject

object order is preferred, but more strongly so when the second NP is indefinite than when it is definite; when the second

NP

is a definite pronoun, in contrast, the object-subject

order is preferred. A corpus study yields the same pattern, except for the non-pronominal defmite NP cases: 'which' questions with a definite second NP more frequently occur in a object-subject rather than a subject-object order. This discrepancy can be explained in terms of the discourse status of the

NP referent.

I I NTRO D U CT I O N A major question in sentence comprehension research concerns the architecture of the sentence processing mechanism: Which kinds of processes are involved? Which kinds of information do these processes make use of? And what is the relative time course of these processes? One way to investigate these issues is by looking at how the parsing mechanism deals with structurally ambiguous input. An example of a structural ambiguity is the wh-question in Dutch. For instance, the fragment in ( I ) is ambiguous between a subject-object and an object-subject interpretation (among others) of the embedded clause.

(I) Ik vroeg welke man de vrouwen . . . I asked which man the women The fragment in (I) is disambiguated only when the finite verb is encountered. The finite verb can either agree with the wh-phrase, enforcing


mainly �n syntax-based strategies when dealing with structural ambiguities.

show that the parser is also sensitive to the nature of the noun phrases ' used and their

3 3 6 Processing Subject-Object Ambiguities

in

Dutch

a subject-object (SO) reading as in (2a), or with the second NP, enforcing an object-subject (OS) reading as in (2b). (2)

Ik vroeg welke man de vrouwen heeft gezien. I asked which man the women has-sG seen 'I asked which man saw the women.' b. Ik vroeg welke man de vrouwen hebben gezien. I asked which man the women have-PL seen 'I asked which man the women saw.'

a.

(SO) (OS)

2 THE SUBJE CT - OBJE CT AMBIGUITY IN D UTCH A number of studies on Dutch and German have shown that in temporarily ambiguous structures comparable to the one illustrated in ( r ), a subject object reading is preferred to an object-subject reading. For instance, Frazier


Although speakers of Dutch have no apparent problems in coping with these order ambiguities, careful experimental investigation shows that in most cases speakers prefer one interpretation of an ambiguous fragment to the other. Investigating how such ambiguities are resolved, that is, which reading is preferred before the point of disambiguation, and which factors determine the strength the preference may provide some insight in the operation of the sentence processing mechanism. In this paper I will focus mainly on the question of which information is involved in resolving order ambiguities. I will present some data suggesting that order preferences are not only driven by purely structural generali zations, but that the processing mechanism is also sensitive to the type of NPs involved. In section 2, some previous studies will be discussed, showing that in temporarily ambiguous structures similar to those in ( r), a subject-object reading (c£ 2a) is preferred to an object-subject reading (2b). Various proposals have been made to account for this processing preference in terms of structurally based strategies. However, in all previous studies, the first ambiguous NP was followed by a non-pronominal definite NP. In section J, I will show there are reasons to expect that the type of the second NP (pronominal NP, non-pronominal definite or indefinite NP) may also affect order preferences. Some data from corpus and experimental studies, among which a new off-line experiment, will be discussed that suggest that this is indeed the case (sections 4 and s). I will conclude with some implications for sentence processing theories and give some suggestions for future research in section 6.

Edith Kaan 33 7

( 1987) compared reading times for subject-relatives as in (3a) to object relatives as in (3b). (3)

Karl Karl b. Karl Karl a.

hielp helped hielp helped

de the de the

mijnwerkers mineworkers mijnwerkers mineworkers

die who die who

de the de the

boswachter vonden. (SO} forester found-PL boswachter vond. (OS} forester found-sc

increase in reading times was found for unambiguous object relatives (3b) compared to subject relatives (3a). This suggests that people have a preference for a subject-object order of the clause, and show signs of processing difficulty when e.g. number information at the verb enforces an object-subject order. In addition, Frazier's materials contained some relative clauses that remained structurally ambiguous. Responses to comprehension questions suggested that these fully ambiguous clauses were generally interpreted as subject-relatives. . These fmdings have been replicated by several other researchers using different experimental techniques (word-by-word reading: Van Gompel 1995; Schriefers, Friederici & Kii.hn 1995; Event-related potentials: Mecklinger, Schriefers, Steinhauer & Friederici 1995), and other sentence types that display the same kind of ambiguity: main clause declaratives and wh-questions (Bayer & Marslen-Wilson 1 992; Frazier & Flores d'Arcais 1989; Hemforth I 993; Kaan I 997; Lamers I996; Schlesewsky, Fanselow, Kliegl & Krems (to appear)); embedded declaratives in German (Bader I 994, I 996; Bayer & Marslen-Wilson I 992; Friederici, Steinhauer & Mecklinger I995; Scheepers, Hemforth & Konieczny I996), and embedded wh-questions (Bader & Meng, in press). This preference for a subject-initial reading is established very early in the process of parsing a sentence. For instance, Kaan (I9in: Experiment I) compared word-by-word reading times for main clause wh-questions such as (4), among other things: An

Welke atleten hadden de trainer niet getrakteerd which athletes had-PL the trainer not treated na de wedstrijd? after the game b. Welke atleten had de trainer niet getrakteerd which athletes had-sc the trainer not treated na de wedstrij d? after the game a.

(SO)

(OS)

Sentences were disambiguated at the fmite auxiliary immediately following the initial, ambiguous NP. An increase in reading times was already seen at the disambiguating auxiliary in the object-subject clause in (4h) relative to


(4)

3 3 8 Processing Subject-Object Ambiguities in Dutch

(s) a. de mijnwerkers b. de mijnwerkers

[die; [t; de boswachter vonden ]]] vond ]]] [die; [de boswachter [ t;

During sentence processing, the parser assigns a syntactic representation to the incoming sentence. Assuming a syntactic representation as in (s), a wh-phrase must be related to a trace downstairs in order to be syntactically integrated. According to the Active Filler Strategy, the parser initially relates a wh-phrase to the first possible gap position. As the functional subject position precedes the functional object position in Dutch, the parser will first relate a wh-phrase to a trace in the subject position, yielding a subject-object clause (c£ (sa)). When subsequent number information is not compatible with a subject-initial reading of the clause, the analysis must be revised, causing an increase in processing difficulty. This accounts both for the preference for a subject-object inter pretation of ambiguous clauses, and for the longer reading times seen for unambiguous object-subject relative to subject-object ordered clauses.

3 ORDER PREFERENCES AND THE TYPE OF S E C O ND NP According to the approach discussed above, the subject-object preference is motivated by the structural properties of the clause only (the position of the subject and object), that is, the properties of the NPs are not relevant.' There are, however, reasons to exP.ect that the nature of the NPs may affect the resolution of order ambiguities. In this section, some differences


the subject-object clause in (4a), indicating that already at this position, people had expected the first NP to be the subject of the clause (c£ also Hemforth 1993; Friederici et al. 1996; Schlesewsky et al., to appear). Several syntax-based proposals have been made to account for the processing difficulty seen for object-subject clauses. The most influential has been the Active Filler Strategy, first applied to the Dutch ambiguities by Frazier (1 987). This approach is based on the syntactic assumption that clause-initial wh-phrases such as welke man in (1), welke atleten in (4) or the relative pronoun die in (3), have been moved from their base position. Moved elements are assumed to leave a coindexed trace, t;, by means of which they can be interpreted as if they were still in their base position. This is illustrated in (s): (sa) is the structural representation of the subject object clause in (3a): the wh-phrase die is linked to a trace in subject position; (sb) corresponds to the object-subject relative (3b), with the wh phrase being coindexed with a trace in object position.

Edith Kaan

339

among three types of NPs (pronouns, full definite and full indefinite NPs) and their potential consequences for word order preferences will be discussed. ·

3.1

Pronouns, full definite and full indefinite NPs

Pronouns, full definite and full indefinite NPs differ with respect to the discourse status of the entities they. refer to. Consider for instance the paradigm in (6): (6)

a.

Even when the sentences are presented out of context, there is a difference between the sentences in (6a-c) with respect to what the addressee of the · utterance is assumed to know. The use of the pronoun in (a) implies that there is a highly �alient entity in the discourse context to which the pronoun refers. This is not the case for (b) and (c). In (b), the non pronominal definite NP the poet only presupposes that the hearer knows which poet is meant, either because a particular poet has been mentioned previously, or because it is part Of common knowledge. The .indefinite NP a poet in (c), finally, does not refer to an entity that is in some sense known to the hearer, but rather introduces a new entity into the discourse. This difference in use among the various types of NP has consequences for the likelihood with which they are encoded as subjects. Prince ( 1981, 1982) shows that NPs that refer to elements that have been mentioned in the linguistic discourse are more often encoded in the subject position than in any other position in the clause. This is especially true for NPs that refer to highly salient elements in the discourse (pronominal NPs). Results from experimental studies are in accordance with this finding: NPs that refer to given, salient entities are likely to be produced as the subject of the clause (e.g. Bates & Devescovi 1 989). Furthermore, reading times have found to be longer for sentences in which the subject NP introduces new information and the object refers to old information than vice versa (Yekovich, Walker & Blackman 1979). Given this tendency to encode old information as the subject and new information as the object of the clause, one would expect various types of NP to differ with . respect to the syntactic positions they typically occupy. Pronouns will be encoded as subjects more often t�n as non-subjects, .as they typically refer to given entities; indefinite NPs, which generally introduce new entities will be encoded more often in non-subject than


He was composing a song. b. The poet was composing a song. c. A poet was composing a song.

340 Processing Subject-Object Ambiguities in Dutch

3 .2

Potential ejfocts on the order preference

The various types of NP thus have different biases towards occupying certain positions in the clause. If this NP-specific information is made use of in resolving word order ambiguities in Dutch, one would expect order preferences to differ depending on the type of second NP. Pronouns favour a subject reading for themselves. One would therefore expect a pronominal second NP in a temporarily ambiguous clause to bias an object-subject reading of the Clause; this bias will be smaller for clauses containing non pronominal definite NPs; and an indefinite second NP would favour an object reading for itself and hence a subject-object reading of the clause. In the following sections we will see whether these NP-specific proper ties can indeed affect order preferences in Dutch. For practical reasons, the research is restricted to which clauses. First, some data will be reported concerning the relative frequency of subject-object and object-subject Table 1 frequency of occurrence in main clause subject and direct

object position for pronouns, full definite and full indefinite NPs (data &om Nieuwborg 1968). Percentage of subject is relative to the sum of subject and direct object occurrences for that type of NP. function of the NP in the

NP type

Subject (% subject}

Pronouns-total Pronouns-personal only full definite NPs full indefinite NPs

3 396 (79) 3082 (87) 1466 (68) 3 5 3 (40)"

d

Including r 1 7 indefinite subjects in presentational

main

clause

Direct object

er

'there' clauses.


in subject position; definite NPs are predicted to score in between, as can either be used to refer to old information or introduce new information. Although no extensive ·corpus studies on Dutch have been conducted to date, the sparse data available suggest that this indeed is correct. The figures in Table 1 are drawn from the Nieuwborg (1968) study based on sooo clauses drawn from Dutch and Flemish novels. The counts are restricted to subject- and object-initial main clauses. Pronouns, and especially personal pronouns, occur far more often as the subject than as the direct object of a main clause (87% subjects for personal pronouns). The percentage of subject use is smaller for non-pronominal definite NPs (68%). And indefinite NPs occur more often as the direct object than as the subject of the main clause {40% subjects).

Edith

which clauses in written Dutch. Next, in section will be discussed.

5,

Kaan 341

some experimental data

4 A CORPUS STUDY

Table

2 Absolute numbers (percentages) of subject and object initial welke-clauses for all predicates and (di)transitive predicates only.

Counts based on the INL

5

million corpus

Order of the welke-clause Subject-initial All predicates {Di)transitives only

Object-initial

Total

2 1 8 (36) 2 1 8 (79)

6o7 275


A corpus study was carried out using the INL corpus, covering 5 million words of written Dutch text. The purpose of this study was, first, to obtain some idea of the relative frequency of occurrence of subject-object and object-subject which-clauses; and, second, to investigate whether the type of the second NP has an effect on the relative frequency of the two orders. From the INL corpus, a sample of 607 welke, 'which' clauses was extracted and analysed for subject- or object-initiality (see Kaan 1997 for details of this study). Numbers of subject- and object-initial welke-clauses occurring in the corpus are given in Table 2. Taking all welke-clauses into consideration, that is, irrespective of the number of argument NPs in the clause, the subject-initial order is the most frequent (64%). However, if the counts are restricted to welke-clauses containing at least one other argument NP (transitive and ditransitive predicates), the object-subject order is the most frequent (79%). Under the assumption that written text is often edited and hence, also a product of compreheruion strategies, the object-subject preference for transitive clauses is in contrast to what would be predicted under the Active Filler Strategy. Furthermore, the relative frequency of the object-subject order is quite unexpected given the subject-object preference found in experimental studies on transitive clauses mentioned in section 2. Now let us see whether the type of second NP makes a difference. In Table 3 the frequencies of . transitive subject-object and object-subject welke-clauses are given as a function of the kind of second NP in the clause: pronominal, non-pronominal definite and indefinite NPs.

34-2 Processing Subject-Object Ambiguities in Dutch

Table

3 Absolute numbers (percentages) of (di)transitive subject object and object-subject welke-clauses as a function of the type of the second NP.

Order Type of NP2

Subject-object

Object-subject

Total

Pronoun Definite NP Indefmite NP

7 (6) 33 (3 1) 1 7 (4o)

1 20 (94) 72 (69) 26 (6o)

127 105 43

5

EXPERIME NTAL DATA

Let us now see whether the nature of the second NP can affect order preferences in experimental tasks. First, I will report some data from previous experiments on which-clauses containing a pronoun as the second NP. Next, a new off-line completion study will be conducted comparing which-clauses containing definite and indefinite NPs. 5.1 A

pronoun

as

the second NP

In a previous paper (Kaan 1 996) I reported several experiments on order preferences in Dutch wh-dauses that contained a pronoun as a second NP. In a first study, subjects were presented with an off-line questionnaire, containing fragments like the following:


As one can see, the object-subject order prevails, though more strongly so with pronouns than with definite and indefinite second NPs. The relative frequency of object-subject clauses is significantly different between clauses containing pronouns on the one hand, and clauses containing definite or indefinite NPs, on the other (p's

z

a child(z)

poss(x, z)

The consequent of the conditional contains an embedded DRS, repre senting the presupposition that John has a child. We mark a DRS as presuppositional by preftxing it with a 8. (The 8 operator was introduced in Beaver ( 1992); here

it is only used

syntactically to distinguish

presuppositional DRSs from ordinary, assertional ones.) Now van der Sandt's presupposition resolution algorithm is applied to this DRS, and starts looking for a suitable and accessible antecedent. Obviously, the discourse referent introduced for a child (i.e. y) is the ideal candidate. So

the presupposition can be bound. Binding a presupposition goes as follows: the presuppositional DRS is removed &om the DRS where it

originates (the source DRS, for short), and merged with another DRS (henceforth the target DRS), namely the DRS which introduces the


(s )

Emiel Krahmer

&

Kees

van

Deemter 3 59

antecedent to which the presupposition is bound. Furthermore, this target DRS is extended with an equality condition which equates the referent introduced in the presuppositional DRS with the referent of the antecedent · In this way the anaphor is 'absorbed' by the antecedent (van der Sandt 1992: 349). As van der Sandt (1992: 3 57) puts it: 'In the case of anaphoric binding the resolver puts in equations which link discourse markers and transfer the conditions associated with . the anaphoric expression to the binding site.' By binding the presupposition, (oRS r ) is transformed into (oRS 2) (DRS 2)

X r-------�

y. z child(y) poss(x, y) child(z) poss(x, z) y=z

happy(z) ==>

It is easily seen that (oRS 2) is equivalent with (oRs 3 ), which is also the DRS which would be used to represent (6}. In fact, the similarity between examples such as (s ) and (6) is one of the suggestive facts which van der Sandt discusses to motivate his presuppositions-as anaphors approach. (oRS 3}

x --------1 t--x = john

�••ld(y)

poss(x, y)

==

l

lwppy{y)

I

(6) If John has a child, it is happy. A

difference between presuppositions and pronouns shows up when

there is no suitable and accessible antecedent In that case, a presupposi tion can be accommodated. Consider the following example with its associated DRS:

(7} If John has an oriental girlfriend, his son is happy.


x = john

36o

Anaphoric Noun Phrases: Towards

(DRS 4)

a

Full Understanding of Partial Matches

X �-------1

x = john

y oriental(y) girlfriend(y) poss(x, y)

happy(z) =>

z

a son(z)

poss(x, z)

(DRS 5 )

X, Z �------------------------�

x = john son(z) poss(x, z)

y oriental(y) girlfriend(y) poss(x, y)

This results in a reading which may be paraphrased as john has a son; and ifJohn has an oriental girlfriend, he; is happy. As this paraphrase indicates, after accommodating the presupposition the resulting DRS entails that John has a son. In general: accommodating the presupposition in the main DRS yields a 'presupposing' reading (the presupposition is projected). By contrast, from (oRs 3 ) it does not follow that John has a child; the presupposition is not projected and this produces a 'non-presupposing' reading. An important element of the presuppositions-as-anaphors approach is that the algorithm associates a set of allowed resolutions to a DRS containing presuppositional DRSs. For instance, accommodation can in principle take place in every (sub-}DRS which subordinates the DRS containing the


Again, the resolver looks for an accessible and suitable antecedent to bind the presupposition that John has a son. There are two accessible antecedents aohn and his oriental girlfriend) but neither can qualify as suitable. Hence we accommodate the presuppositional DRS. If certain conditions (to be discussed below) are met, accommodation takes place in the main DRS (see van der Sandt 1 992: 345 for explanation). Technically, accommodating a presuppositional DRS amounts to removing it from the source DRS and merging it with the target DRS (which-under normal circumstances-is the main DRS). Thus:

Ernie! Krahmer & Kees

presuppositional

van

Deemter

361

DRS. The 'in principle' restraint refers to the fact ·that each

instance of accommodation should satisfy a number· of independently motivate� constraints. One of them is called the Consistency Constraint. This constraint says that accommodating a presupposition must never lead to an inconsistent

(8)

DRS.

Consider:

It is not the case that John is besotted with his oriental girlfriend, because John has no girlfriend.

oriental girlfriend which he is besotted with, because john has no girlfriend. Another constraint is the so-called Informativity Constraint, which

rules out interpretations that lead to excessive redundancy. Van der Sandt uses this constraint, among other things, to legislate against conditional representations where the information in the antecedent follows from material present in an accessible DRS (e.g. the main one). Thus, consider

(9)

If John has children, his children will be spoiled.

Suppose that the presupposition triggered by the children is accomodated in the top of the DRS, thus guaranteeing the existence ofJohn's children. Then the antecedent DRS of the conditional Uohn has children) violates the Informativity Constraint because it has become redundant given the accommodated information. As a result, this interpretation is ruled out. For more information on the background and formalization of these constraints we refer to van der Sandt ( 1 992: 367-9). So there may be various ways of resolving a presuppositional DRS, and all these possibilities together form the set of allowed resolutions. This brings us to a last, crucial ingredient of van der Sandt's theory: the definition of a preference order over permitted interpretations. One could argue that this preference order is what gives the theory content. Mter all, if all allowed readings were equally preferred, we would get an extremely liberal theory. For example, it would mean that for the following example the accommodation · reading (which intuitively is highly non preferred) would be just as 'good' as the obvious binding reading. (1o) Whenever a young cat and an old cat fight over a fish bone, the young cat seizes it. •

Notice incidentally that assigning preference orders is fairly common with theories dealing with ambiguities, perhaps most prominently with


Accommodating the presupposition thatJohn has an oriental girlfriend in the main DRS (and thus not within the scope ofthe negation) is easily seen to yield a contradiction. Therefore, global accommodation is ruled out; and the presupposition is accommodated locally (i.e. within the scope of the negation). The resulting reading can be paraphrased as It is not the case thatjohn has an

362 An.aphoric Noun Phrases: Towards a Full Understanding of Partial Matches

implementations of theories of anaphora resolution (e.g. Sidner I979), but also with theories of quantifier scope ambiguities (see e.g. Hobbs & Shieber 1987: 48-9 for some discussion). Van der Sandt {I992: 345, 3 57) defines a preference order based on the following general principles: DEFINITION I (Van der Sandtian preferences) 1. Binding to a suitable antecedent is preferred over accommodation. 2. Accommodation is preferred to occur as 'high' (far away from the source-DRS} as possible. 3· Binding is preferred to occur as 'low' (near the source-DRS} as possible.

( 1 I) A man was quietly walking down the street, when he was joined by another man. The man said: . . . According to van der Sandt (1992: 363) partial match examples also display a genuine ambiguity, and he claims that this is one of the phenomena that his theory can account for, while the satisfaction camp cannot. However, things appear to be somewhat more complicated.

3 THE PARTIAL MATCH P HENOMENON The intuitive idea underlying the notion of a partial match can be explained in anaphoric terms. Consider a definite NP, along with a potential antecedent. The plausibility of an anaphoric link between the two depends, among other things, on the 'similarity' between the two nouns. This is clear in extreme cases. For example, the woman cannot take a man as its


In most cases, these three preference rules order the set of admissable resolutions in such a way that there is one most preferred reading. For example, even though intuitively sentence (7) is not ambiguous, it is predicted to have three different interpretations: one obtained by global accommodation (oRS s ), one obtained by intermediate accommodation (where the target DRS is the antecedent DRS of the implicational condition), and one obtained by local accommodation {the target and source DRS are one and the same). However, this example is not classified as a genuine ambiguity because the preference constraints single out one most preferred interpretation. Following van der Sandt, we speak of a genuine ambiguity when there is no single most preferred reading. Examples are cases where there are two possible antecedents for binding which are introduced at the same level. Example ( u ) would be an illustration of this,. where nobody can tell which of the two men walking in the first sentence does the talking in the second.

Emiel Krahmer & Kees

van

Deemter 363

antecedent. A full or complete match corresponds with a high degree of similarity; a partial match corresponds with an intermediate degree of similarity; finally, one might speak of a complete non match when antecedent and anaphor are incompatible (as in the case a man; the woman). 3.I

The empirical facts

Let us first take · a closer look at what van der Sandt says about partial matches. Here are the examples put forward by van der Sandt (1992: 350-1). ( 12)

a.

According to van der Sandt, these sentences 'display a genuine ambiguity' (van der Sandt 1992: 3 so); they have no single, preferred reading and this is due to the fact that the respective presuppositions �may but need not be bound to a potential antecedent' (van der Sandt, ibid.). All cases in (12) display a similar structure. In each case there is, as Soames (1982) put it, 'a one-sided entailment between the antecedent and the presupposition of the consequent'. In other words, the antecedent is more 'informative' than the anaphor. Of course, this is only one of the possible relations between would-be antecedent and would-be anaphor. It is instructive to take a systematic look at the various pre-theoretic possibilities, which we loosely label as follows: I. antecedent is more 'informative' than anaphor, II. anaphor and antecedent are 'incomparable', III. anaphor and antecedent are equally informative, and IV. anaphor is more informative than antecedent. Restricting ourselves to NPs: A is more informative than B essentially means that the denotation of A is subsumed by the denotation of B. We make these notions formally precise in section 4·3·

Antecedent is more informative than anaphor. The examples in (12) may be placed in this category. We have discussed one of them in somewhat more detail, namely (12.a). We fully share van der Sandt's intuitions here, although the intuitions concerning example ( 1 2.a) might be a bit blurred due to a kind of lexical ambiguity in the word girlfriend (mistress/ companion in life). However, it is not difficult to find examples that do not suffer from this problem. I.

(13) If John buys a couple of cats, the pets won't be happy.


If John. has an oriental girlfriend, his girlfriend won't be happy. (= {I)) b. If John murdered his wife, he will be glad that she is dead. c. If someone at the conference solved the problem, it was John who solved it.

364 Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches

Here, the antecedent is more informative than the anaphor: the set of cats is a subset of the set of pets. Intuively, this sentence displays the same kind of ambiguity as (12.a). Thus ( I 3) has a presuppositional reading (paraphrasable as there is a set of petsi, and ifJohn buys a couple of cats, then they; will not be happy) and a non-presuppositional reading (ifJohn buys a couple ofcats;, they; will not be happy).

II. Anaphor and antecedent are incomparable. The following examples fall in this category:

Obviously, in both cases anaphor and antecedent are incomparable. Never theless, intuitively, these examples are ambiguous in the same way as the partial match examples discussed so far. Thus: example (I4.a) is ambiguous between a non-presupposing reading (paraphrasable as ifJohn has sons, the young ones among them will be spoiled) and a presupposing one Uohn has young children; and if he has sons, they; will be spoiled). Example (I 4.b) is ambiguous, in a similar fashion, between a presupposing reading (there is some group of children; and ifJohn talks to some partygoers, they; will laugh at him) and a non presupposing reading (ifJohn talks to some partygoers, the children among them will laugh at him). Of course, example (I4.b), (lik,e example (I4a)), can also have an identity reading (in which case all the partygoers are children). Hovy-ever, we contend that this is a special case of the last, non presuppositional reading. Notice that for the identity reading, the definite the children is typically deaccented. In general, and simplifying somewhat, identity anaphors tend to be deaccented, while all other noun phrases must be accented (c£ van Deemter I994 for the fine print). Hence, accenting can sometimes have a disambiguating function. In particular, the lack of an accent on children in (I 4.b) signifies that all the partygoers are children, whereas an accent on children signifies that the children are either a real subset of the set of partygoers or some other set of children-but not the entire set of partygoers that John talks to.

III. Anaphor and antecedent are equally informative. Consider the following examples: (I s)

a. If Fido catches a cat, the cat must be old. b. If Fido catches a cat and a mouse, he will chase the cat and devour the mouse.

The first example is a clear case of a full match (the set of cats is equal to the set of cats). Intuitively, sentences like ( I s.a) do not display an ambiguity.


(I4) a. If John has sons, his young children will be spoiled. b. If John talks to some partygoers, the children will laugh at him.


van

Deemter 365

Such examples, containing full descriptions, are slightly marked; a pronoun would sound more natural. However, it is easy to construct natural examples involving fully matching descriptions along the lines of {I s.b).

IV. Anaphor is more informative than antecedent. Notice that when an anaphor is more informative than its antecedent, it triggers a presupposition which presents new information.· The distribution of such examples is somewhat limited. For instance, it is difficult to . find conditional examples which fall in category IV. Consider:

For both these examples, the presupposing reading seems strongly preferred over the non-presupposing one, which is marginal at best. In other words, these sentences do not seem to be ambiguous. In Krahmer (I995: I6S) it is hypothesized that identity anaphora can only add information if the antecedent is interpreted specifically. Let us formulate this as the Informative Anaphors Hypothesis.2 INFORMATIVE ANAPHORS HYPOTHESIS {IAH) potential antecedent with a non-specific interpretation, which is less informative than the anaphor under consideration, does not qualify as a suitable antecedent for the anaphor, provided that the relation between anaphor and potential antecedent is one of identity. A

Thus, an (identity) anaphor can only add information about its antecedent when the antecedent has a specific interpretation, and this would account for the fact that the examples in (I 6) defy categorization as partial matches. The IAH only applies when the relation between anaphor and potential antecedent is one of identity, that is: when antecedent and anaphor refer to the same (set ofj object(s). This qualification is made, because it seems . possible for anaphors to add information about a subset of their antecedents. Consider: (I 8) If Barney owns cows, then he will feel sorry for the mad cows. Clearly, mad cows (where mad carries as accent) is more informative than cows. And contrary to examples such as (I 6), this example displays an intuitive, genuine ambiguity between a non-presupposing reading (para phrasable as ifBarney owns cows, then he willfeel sorryfor the mad cows he owns) and a presupposing one (there are mad cows;, and if Barney owns cows, then he will feel sorry for them;). Thus, examples of type N also display a partial


{I6) a. If John has a girlfriend, his oriental girlfriend won't be happy. b. IfJohn owns a donkey, he will be worried about the purple farmereating donkey on the loose. (after Beaver I995: 6I)

366 Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches match · ambiguity, as do types I and II, but factors such as the IAH complicate the picture somewhat. To summarize: we distinguished four kinds of relations between would be anaphor and would-be antecedent. When an anaphor is not equally informative as its would-be antecedent (i.e. the two have a different denotation) we have a partial match. Such sentences display a genuine ambiguity between a presupposing and a non-presupposing interpretation, but sometimes other factors may cause disambiguation. In particular, we have seen that continuations, deaccenting and the Informative Anaphors

)

Hypothesis (the IAH may eliminate one of the interpretations. Let us now look at the predictions the theory of van der Sandt (1 992) makes regarding

a number of problems.

3 .2

Van der Sandt's predictions

I. Antecedent is more informative than anaphor. Let us reconsider van der Sandt's own (1 2.a), and let us construct a DRS for this example. As before, we assume that his has already been resolved as Joht?'s.

(DRS 6)

X �------�

x = john

y oriental(y) girlfriend(y) poss(x, y) a

girlfriend(z) poss(x, z)

(

If we feed oRS 6) to van der Sandt's resolution algorithm, it will first start looking for an accessible and suitable antecedent for the presuppositional

DRS: it seeks a discourse referent which is accessible and which satisfies the conditions of being a girlfriend, and standing in the possessive relation with John. But such a referent is easily found: y meets all the conditions. So it is unclear how van der Sandt ( 1 992)'s algorithm can avoid binding the presupposition, which would make the non-presupposing reading (given in


the examples discussed above. As we shall see, if we apply the formal theory of van der Sandt (1 992) to the three kinds of partial matches, we encounter

Emiel Krahmer & Kees

van

Deemter

367

(2.a) ) the primary reading of ( 1 2.a) (by definition 1) and hence would predict that this example is not a genuine ambiguity after all. We do not consider this to be a point of great importance by itself since it is easy to modify the algorithm in this respect, but . it does indicate that van der Sandt's formal theory does not fully implement the intuitions sketched in the fi�t part of van der Sandt (1992).

II. Anaphor and antecedent are incomparable. The same problem applies as in category I. and other, more serious, problems apply in addition. Consider, for example, (14.a), repeated below for convenience.

(oRS 7) is the van der Sandtian representation for this example. Again, assume that his has been resolved as John's. We follow the notation for plurals used by van der Sandt ( 1992: 3 70) The capitals are discourse referents standing for sets of objects. It should be noted that all predicates in this article are 'strictly distributive' in the sense of Kamp & Reyle (1993, 407). E.g. young child(X ) has the intuitive interpretation that all elements of X are young children. In Kamp & Reyle (1993) this is denoted as young child* (X ). We omit the * superscript where this can be done without creating confusion. .

(DRS 7)

X

r-------�

x = john y

son(Y) poss(x, Y)

spoiled(Z) ===>

a

z

young child(Z) poss(x, Z)

If we feed (oRS 7) to the algorithm, it will look for an accessible, suitable antecedent for the presupposition triggered by his young children. It is not entirely clear to us whether sons is a suitable antecedent for his young children according to van der Sandt's algorithm (presumably it is), but it yields undesired results either way. Suppose the algorithm decides that Y (the sons) is not a suitable antecedent for Z (the presupposed young children). In that case, the van der Sandtian preference order given in definition 1 tells us that the presupposing reading, achieved by accommodating the presupposition, is


(19) If John has sons, his young children will be spoiled.


the single most preferred reading for (14.a). In other words, (14.a) does not display a 'genuine ambiguity' as partial match examples should do. If, by contrast, Y (the sons) is a suitable antecedent for Z (the young children), binding is preferred and, as before, no ambiguity results. In the latter case, there is an additional problem, which has nothing to do with preferences between interpretations. If the presupposition gets bound, it is 'absorbed by the antecedent', and this results in a reading which may be paraphrased as follows:3 (2o) If John has young sons, they will be spoiled.

(21) If John has children, he'll spoil the little bastards. We are well aware of the fact that epithets like little bastards have some pecular properties (see below). Nevertheless, they serve nicely to illustrate further the point about in situ binding. If we bind the presupposition triggered by the definite description in van der Sandt's way, we end up with a reading which may be paraphrased as ifjohn has children and they are little bastards, then he'll spoil them. That is: the children are only spoiled if they are little bastards. In our opinion, the right reading is something like ifjohn has children, they'll be little bastards and he'll spoil them. Moreover, we feel that, for the non-presuppositional reading of (14.a), there is no reason to assume that the young children should be identified with the sons (as is done in (2o) ). The only thing we can safely say is that the young children are a subset of the sons. In sum, we think that (14.a) has a binding reading paraphrasable as (22). (22) If John has sons;, the young ones among them; will be spoiled. said, we view the situation in which all the sons happen to be young as a special case of this interpretation, typically marked by a complete absence of accents in the NP his young children (c£ van Deemter 1 992). It is worth pointing out that the two problems we mentioned in connection with the paraphrase (2o) are independent of each other. In fact, some people informed us that, besides (22), they could also conceive of an interpretation of( 14.a) in which the anaphor is bound subsectionally but not in situ:

As

(23) If John has sons; and somej of them; are young, then thesej will be spoiled.


This reading seems wrong. First of all, in our opinion binding can also appear in situ, that is: the presupposition to be bound should not necessarily be merged with the target DRS, but may also be merged with the source DRS. Consider yet another example:

Ernie! K.rahmer & Kees van Deemter 369 We agree that (23) is a possible reading of sentence ( 1 4-a). Below we return to this issue.

ill.

Anaphor and antecedent are equally informative.

Above we argued that,

intuitively, the examples in this category (the full matches) do not display a genuine ambiguity. The theory ofvan der Sandt ( 1 992) correctly predicts that the binding interpretation of the examples in ( I s) is the most preferred one. It is interesting to note that the accommodation reading, while less preferred, is not ruled out. Reconsider example r s .b), and suppose that the presuppositions triggered by the cat and the mouse have been globally accommodated. The resulting reading is not ruled out by the informa� tivity constraint, since this constraint is defined in terms of redundant

(

tion, namely that Fido

chases a cat and a mouse. Nevertheless, it seems that

the accommodation reading for sentences like (r s.b) is not available. Therefore, Krahmer & van Deemter (1 997) have argued that some sort of informativity constraint should also apply at the level of NPs. However, a full discussion of this issue falls outside the scope of the current article.

IV Anaphor is more informative than antecedent.

As for the fourth group of

examples, the reader may easily verify that the same problems are

encountered here as in case II: van der Sandt's algorithm fails to predict the genuine ambiguity of the sentences in this category and binding is defined in a way that may yield incorrect interpretations.

4 M O D I FY I N G THE PRESUPPO S I T I O N S - A S - A N A P H O R S APPROACH In section 3 . 1 we argued that an anaphor and an antecedent stand in a partial match relation if the two are not equally informative, i.e. not co-extensive. Moreover, in the case of a partial match, a sentence is genuinely ambiguous between a presupposing and a non-presupposing reading. In other words, we support the

intuition sketched in van der Sandt (1 992: 3 49-5 1). However,

if we apply the formal theory (i.e. the presupposition resolution algorithm) of van der Sandt ( 1 992) to the partial match examples (as done in section 3.2), we encounter essentially two problems: (i) the algorithm does not generate the required genuine ambiguity in the case of a partial match, and (ii) not all the binding readings are correct. To remedy these problems, we propose a modified version of van der Sandt's resolution mechanism. One central ingredient is the use of so-called

context variables. Binding is viewed as contextually restricted quantification,


(sub-)DRSs, and the antecedent-DRS contains non-:redundant informa

370

Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches

where the relevant context is provided by the anaphoric antecedent. taken to be a contextually restricted variant of the usual accommodation procedure. To arrive at all the different possible (binding or accommodation) interpretations of a given sentence containing a presupposition, we exploit van der Sandt's resolution mechanism, with its use of unresolved representations. However, we make some modifications to the resolution mechanism as such, taking the notion of partial match into account by paying more attention to properties of potential antecedents. When antecedent and anaphor stand in a partial match relation, the algorithm generates a genuine ambiguity. This entails that our modification of the algorithm yields a modified, partial preference order between possible interpretations. One of the attractive features of van der Sandt (1 992) is that it aims at developing a uniform mechanism treating all presupposition triggers in exactly the same way, namely as anaphors looking for an antecedent. A disadvantage of this bird's eye view on presuppositions is that a number of details (such as the issues we are concerned with here) are not sufficiently spelled out. We take it that this is one of the factors which has lead to a certain confusion of tongues on the partial match phenomen'On in the first place. Therefore we opt for a frog's perspective on presupposition projection at first, focussing on one kind of presupposition triggers: definite descriptions. Later, we slightly broaden our horizon by taking all noun phrases into consideration. To start, however, we have to do some formal groundwork.

Accommodation is

Preliminaries

Van der Sandt (1 992) is mostly based on the DRT fragment as it was defined in Kamp (198 1). The kind of examples we are interested in, and the treatment we have in mind for them, calls for two extensions of this basic DRT fragment. The issue of plural quantification is known to be complex, and combining it with the dynamic interpretation underlying DRT only adds to the complications. In the following, we adopt the basic treatment of plurality and quantification outlined in Kamp & Reyle (1993, Ch. 4). Kamp & Reyle use an algebraic 'Link-style' (Link 198 3) interpretation of plurality, in which the domains contain atomic as well as non-atomic entities. Following the convention of Kamp & Reyle (1993), we use boldface lowercaps variables (x, y, z, . . . ) to range over both individual (or atomic) referents and plural (non-atomic) referents. Lowercase variables (x, y, z, . . . ) are used for individual referents, and uppercase variables (X, Y, Z, . . . ) for plural referents. This convention entails that general definitions contain boldface referents, and actual examples do not.

Plurality and quantification in DRT


4. 1

Erniel Krahmer & Kees

van

Deemter 3 7 I

We also adopt the treatment of generalized quantifiers in Kamp & Reyle ( 1993, Ch. 4) in terms of duplex conditions. In general, a generalized quantifier (which we shall denote as DET, short for determiner) is a relation between two sets of (atomic) entities, say A and B, and is represented by Kamp & Reyle as a condition of the following form, where A' is a DRS representing A, and B' a DRS representing B.

�

thesg (A) (B) is true with respect to a model M iff 3d E D(d E A & Vd' E D(d' E A ==? d' = d) & d E B) theP 1 (A)(B) is true with respect to a model M iff 3d E D(d E A & Vd' E D(d' E A ==? d' E B)) It is worth pointing out that Kamp & Reyle still distinguish indefinites from truly quantificational determiners Qike five, all ), and we follow this practice. Concretely, this means that indefinite NPs of the form DET CN, where DET is either a (n), some or empty (in the case of bare plurals) introduce a fresh discourse referent into the current DRS.

Context variables. In Westerscihl ( 1 9 8 5 ) the notion of contextually restricted

quantification is introduced, motivated by examples such

(24)

as

the following:

The children were having a lot of fun.

Clearly this is not a statement about all the children in the universe. According to Westersclhl, the defmite determiner acts as a 'context indicator which signals the presence of a context set C' (Westersclhl 1 9 8 5 : 6o} in such a way that the children denotes C n child, i.e. a contextually restricted subset of the set of all children. Westersclhl immediately" noted that there is a strong connection between discourse referents and context variables (Westersclhl 1 9 8 5 : 70). In our revision of the presuppositions-as-anaphors theory, we use context variables, which we represent as C, C', . . . These context variables are just designated discourse referents. Below, we let every NP introduce a fresh


The leftmost DRS of this duplex-condition is called the restrictor and the rightmost DRS the scope, the capsized box in the middle is the quantifier. The DET gets its usual interpretation as given in generalized quantifier theory (GQT, see e.g. Barwise & Cooper 1 9 8 1 ; for technical details on generalized quantifiers in DRT we refer to Kamp & Reyle 1 9 9 3 : 42 5-7). For example, here are the standard GQT-style definitions of singular and plural the (with d and d' atomic):4

372 Anaphoric Noun Phrases: Towards a full Understanding

of Partial

Matches

M Ff rt (x) iffj(x) E IM (TJ) nf(A ) M f=J TJ*A (X ) iffj (X ) E IM (TJ* ) nf(A ) 4.2

Definite descriptions and the presuppositions they trigger

Definite descriptions, phrases of the form [the CN] (where CN is a possibly complex common noun phrase), are generally assumed to trigger an existence presupposition, that is: the CN denotation is presupposed to be non-empty. We model this as follows: a definite description presupposes that there is some context variable C which has a non-empty intersection with the CN denotation. When the DRS construction algorithm encounters a definite description, the following rule (tempor ary version) is activated.6

I [DET CN]

Rule,

for DET

=

the

Upon encountering an S of the form af3 or a VP of the form {3a. with a a definite description (of the form the CN[± sg]), replace S or VP with the following presuppositional DRS and duplex condition, where and z are fresh discourse referents and C is a fresh context variable.

y

y

C�(y)

� w y

a� C� (z)

{3


ordinary discourse referent as well as a fresh context variable, and our modified presupposition resolution algorithm explicitly operates on the latter. Here, a context variable C is either equated with a previously introduced discourse referent, or with the entire domain of discourse. We have made this simplifying assumption in order to stick as closely as possible to van der Sandt ( 1 992). In any case, the output of our presupposition resolution algorithm is a DRS from which the context variable C can be eliminated. Thus, the use of context variables merely facilitates the resolution process. Besides introducing contextual variables, we also employ 'contextually restricted' predicates. That is, we use conditions like mane (john) which have as intuitive interpretation: John is a man and an element of the context variable C. Formally, if TJ is a noun representation:5

Ernie! Krahmer & Kees van Deemter 373

Here CN is the representation of CN (in singular form), and z is z or Z depending on the number of the CN.7 To illustrate this rule, consider example (1 4.a) again, here repeated as (25). (25) If John has sons, his young children will be spoiled.

(o� 8)

�x----------------------------------------� x = john y

son( Y)

poss(x, Y)

_ r; 'oun _g_c _ hil--, dc "'

young chi�(Z) possC(x, Z) C=Y

�oung chi�(v) (.2) spoiled(v) possC(v, x)

V


V

Emiel Krahmer

& Kees van Deemter 3 79

This DRS can be paraphrased as IfJohn has sons, then there are young children among them and these will be spoiled, or more concisely as (29). (29) IfJohn has sons, then the young ones will be spoiled.

The second reading comes about via a global application of ACCOMMODATE. The result is (oas 1 2).

(oRS

)

12

X, C, Z

y

son(Y) poss(x, Y)

�

v chilcJC(v) possC(x,v)

�

spoiled(v}

� -

This interpretation can be glossed as follows: (3o) John has young children;, and if he has sons, they; will be spoiled. Summarizing, if we feed the representation of example (25), (oas 10), to the modified resolution algorithm, it decides that there is a partial match between the presupposition triggered by the description his young children and its antecedent sons. The corresponding ambiguity is between (oas u ) and (nas 12) for the non-presupposing/binding and presupposing/ accommodation interpretation respectively. So far, we have discussed a refined and modified version of van der Sandt's presupposition resolution mechanism, and applied it to pre suppositions triggered by definite descriptions. We have assumed that definite descriptions trigger an existence presupposition, more precisely: the CN presupposes that the intersection of the CN denotation with a given context set is not empty. Compared with the resolution algorithm of van der Sandt (1992), our version is different in three respects: (i) it is more explicit in what counts as a potential antecedent (namely a pair consisting of a discourse referent and its value set), (ii) it is more explicit about the relation between presupposition and potential ante cedent, explicitly distinguishing full match, 'no match' and partial match, and generating the required ambiguity in the latter case, and (iii) binding is a different operation, defmed using contextual quantifica tion: it generalizes to non-identity cases and the presuppositional DRS


x = john young chilcJC(Z) posse (x, Z) C=D

3 80 Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches

remains in the DRS where it originated. Two well-known 'difficult' phenomena often associated with definite descriptions are bridging and epithets. How do these fit in with our proposals? Consider the following bridging example:

( 3 1) If John buys a car, he checks the motor first.

(32) Mary got some picnic supplies out of the car. The beer was warm. Even though the approach to world knowledge advocated by Krahmer/ Piwek is explicitly subjective, it is unlikely that the interpreter has a piece of world knowledge to her disposal to the effect that picnic supplies include beer. In general, world knowledge may allow the interpreter to infer food and beverage. Hence the presupposed beer will only match partially with the implied beverage. Epithets are definite descriptions like the poor woman, the little bastards, etc. They differ from ordinary descriptions in that they are not interpreted literally (the poor woman may be a millionaire and the little bastards can be legitimate children) and have a very strong preference for being bound. In particular, they do not appear to give rise to partial match ambiguities. Consider:

( 3 3) If John has an oriental girlfriend, then we feel sorry for the poor woman.

Example ( 3 3) does not display a partial match ambiguity; it only has a non-presupposing/binding reading. We claim that this is due to the fact that epithets essentially behave like pronouns, and as noted in the introduction,


The definite description the motor presupposes the existence of a motor. Clearly, the presupposition triggered by the motor does not even match partially with one of the potential antecedents Uohn and a car). Hence, (global) accommodation of the presupposition is predicted. The fact that cars have motors is not taken into account at all. In Krahmer & Piwek (1997) and Piwek & Krahmer (1997), it is shown how van der Sandt's presuppositions-as-anaphors theory can be combined with a constructive inference system, which allows for a formal interaction between presuppo sition resolution and world knowledge. What is relevant for our purpose here, is that a piece of world knowledge stating that cars have motors can be used to bind the presupposition to an implied antecedent (the motor of the car mentioned in the antecedent). In this case, the presupposition triggered by the motor fully matches the implied antecedent, but as one would expect, presuppositions can also match partially with implied antecedents. The more 'difficult' ca.Ses of bridging (such as (3 2), from Haviland & Clark 1 974: 5 1 4-1 5) illustrate this.

Emiel Krahmer & Kees van Deemter

381

pronouns cannot be·accommodated (we discuss this property below in more detail). One could say that epithets are pronouns that are emotionally dressed up by the speaker, ie. they express something of the speaker's attitude towards the designated referent. To hac� up this claim, consider the following example, which is Geurts' (1995: 47) variant of Karttunen's (1969) paycheck sentence. ·

(34) The man who fed his German shepherd on tandoori chicken was beaten by the man who restricted {its/the poor animal's/his dog's} diet to broccoli.

4·4

Extending the presuppositions-as-anaphors approach

In the previous section we have outlined a modified version of the presuppositions-as-anaphors algorithm and applied it to definite descrip tions. There seem to be few impediments to extending the resulting theory in several ways. For one thing, we would like to believe that the treatment of definite descriptions carries over to definite NPs in general. It is common practice to analyse possessives as definite descriptions, thus John's oriental girlfriend is seen as shorthand for the oriental girlfriend ofJohn. Moreover, various people have argued that proper names too can be seen as definite descriptions of the form the individual named such and such.' ' Thus, the proper name John can be understood as the individual named John'. It has been observed (e.g. by van der Sandt 1992: 3 75) that such a presuppositional analysis of proper names improves upon the standard DRT treatment, in which proper names are always inserted in the main DRS, in that it allows proper names to be 'bound' occasionally. Consider the following example, which does not appear to have a presupposing reading. (3 5) If parents decide to name their child 'Barf', we know in advance that Barf will have a hard time in school.

Notice that even though proper names can be bound, they do not seem to give rise to partial match ambiguities. Take example (36).


Focusing on the sloppy identity reading, Geurts notes that the pronoun it and the epithet the poor animal can be read as his German shepherd, but that such a reading is not available for the phrase his dog; in that case the breed of dog owned by the second man is left open. This, admittedly circumstantial, evidence indicates that epithets should be analyzed as pronouns rather than as full definite descriptions. Accordingly, we assume that they will always be bound.

382

Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches

{36) If a man enters the room, Bill whistles.

{37) If a doctor enters the room, he whistles. Here the situation is precisely the inverse of the proper name situation. Even though according to our criteria, the value sets associated with doctor and he (or: male individual) stand in a partial match relation (both men and women can be doctors), the presupposition associated with the pronoun cannot be accommodated. This follows from van der Sandt's claim that pronouns and 'other kinds of semantically less loaded anaphors' cannot be accommodated, since they do not carry sufficient descriptive content (van der Sandt 1992: 344-345). Van der Sandt founds his claim on pairs such as {38). (38)

a. Jack's dog is bald. b. It is bald.

According to van der Sandt, the presupposltlon that Jack has a dog is accommodated in (38.a) and hence this sentence receives a normal interpretation. Example (38.b), however, does not get an interpretation because the pronoun does not have an antecedent. We agree with van der Sandt's claim, and take it that this explains why the partial match in (37) is disambiguated in favor of the non-presupposing/binding reading. We believe that this claim should be placed in a wider perspective. For example, it seems to us that both (39.a) and (39.b) are uninterpretable without context, even though only the latter contains a pronoun. (39)

a.

The man is bald. b. He is bald.

In our opinion, this illustrates that van der Sandt's claim should be related to semantic content (e.g. maleness), rather than form. This might also explain why for certain sentences in which there is a partial match between anaphor and antecedent a binding interpretation seems preferred. For


Intuitively, example (36) only has a presupposing reading. The presupposi tion triggered by Bill cannot be bound to a man, even though the value sets for Bill and man presumably stand in a panial match relation {since Bill is typically a masculine name). We contend that this is due to our Informative Anaphors Hypothesis, which stated that an antecedent which is less informative than the anaphor does not qualify as a suitable antecedent. Finally, we adopt the fairly common treatment of pronouns as lightweight descriptions. Thus, a pronoun like he is analysed as the male individual (c£ e.g. Geurts 1995: 23). Interestingly, just like proper names, pronouns do not seem to give rise to panial match ambiguities. Consider:

Emiel Krahmer

&

Kees

van

Deemter

383

instance, Dekker (p.c.) pointed out that there is an intuitive difference between (2 7), repeated here as (4o.a), and (4o.b). (4o)

a.

If John has sons and a TV set, his children will watch a lot of

football. b. If John has sons and a TV set, the children will watch a lot of football.

NP PRESUPPOSmON SCHEME Suppose an NP is of the form [DET CN) and C is some context variable. If DET is strong or accented, then the NP triggers the presupposition that th� intersection of the respective denotations of C and CN is not empty. The distinction between strong and weak determiners is due to Milsark (1977), and is based on the observation that the latter, but not the former,


Even though both sentences are partial match cases and our modified resolution algorithm predicts both are genuine ambiguities, it seems that for example (4o.b) the anaphoric, non-presuppositional reading is preferred. To summarize: the observations made above indicate that the specific behavior of pronouns and proper names provide no obstacles for a generalization of the analysis of definite descriptions from the previous subsection to definite NPs in general. Let us now investigate the possibility of generalizing the treatment of definite description to all NPs. As was explained in section 4.2, our proposed treatment of definite NPs rests on the use of contextual restrictions on both the assertional and the presupposi tional part of what is expressed by a definite NP. The first restriction results in contextually restricted duplex conditions; the second restriction results in a presuppositional DRS of the form (C, y j CN c (y) ]. Moving from definite to non-definite NPs, it seems plausible enough that our treatment of the assertion (i.e. the contextually restricted duplex conditions) may be reused. However, in this section we would like to argue that the parallels between definite and non-definite NPs go even further than this, extending into the presuppositional part of what the NP expresses. We would like to make a rather strong claim: that a non-definite NP can express exactly the same (existential) presupposition as a definite NP. This assumption is exploited to arrive at an analysis of non -definite NPs along analogous lines as the analysis of definite NPs presented in the previous sectioiL In our opinion, an NP can trigger a presupposition in two cases: namely when the determiner is strong or if it is weak but accented. This is formulated in the following NP presupposition scheme.

3 84

Anaphoric Noun Phrases: Towards

a

Full Understanding of Partial Matches

can occur in postverbal position in there sentences. Thus, for instance, most is a strong determiner, while some is a weak one: (41)

a. *There are most politicians in this hotel. b. There are some politicians in this hotel.

(42) If a new teacher is hired, there are {No/FEw} girls in this class who immediately have a crush on him. In fact, they are primarily interested in the Backstreet Boys.

Here the weak NP {No/FEw} girls in this class triggers an existence presupposition. Intuitively, the presupposition is projected and this explains why the pronoun they in the second sentence succeeds in finding an antecedent, thereby referring to the set of girls in this class. Summarizing: the NP presupposition scheme leads to a generalization of the DRT construction rule for definite descriptions in section 4.2, applying to all NPs of the form DET CN where the DET is strong or accented. DET CN Rule, for DET = strong or DET = weak but accented

I

Upon encountering an S of the form a/3, or a VP of the form f3a, where a is of the form DET CN[± sg]) and DET is either strong or weak but accented, replaces S or VP with the following presuppositional DRS and duplex condition, where y and z are fresh discourse referents and C is a fresh context variable. y

CfiC(y)

�

{Jy

a� ctvC (z)

Note that NPs with a determiner that is either strong or accented (or both) can give rise to the expected partial match ambiguities. Consider the following example:


It is worth noticing that no definite NPs are allowed in postverbal position. For this reason, the phenomenon in (41) is also referred to as the definiteness effect. Partee (1984), de Jong (1987: 276} and Zucchi (1995), among others, have argued that all strong NPs trigger an existence presupposition. In certain situations weak NPs can also trigger an existence presupposi tion, namely when the determiner is accented (van der Sandt p.c.). Consider the following example (small caps indicate stress}:


(43)

van

Deemter 3 8 5

If the new teacher lectures some pupils, most girls immediately have a crush on him. b. If the new teacher lectures some pupils, at least THREE girls immediately have a crush on him. a.

(44) If thei new teacher lectures somej pupils, there is a group of girlsk among themj and {most I at least three} of themk have a crush on himi. When we accommodate the presuppositional DRS in van der Sandtian style, this yields a DRS paraphrasable as (45). (45) There is a group of girlsk and if thei new teacher lectures somej pupils, {most/at least three} of themk have a crush on himi. These predictions accord well with our intuitions.

s D ISCUSSION In this discussion section a number of issues for future research are mentioned (section s.r), and a comparison is made with van Deemter's approach to generalized anaphora (section 5.2). 5.1

Future research

We have opted for afrogs perspective in modifying van der Sandt's algorithm: that is, we have first worked out the details for definite descriptions, and subsequently argued that the same analysis carries over to NPs in general. Van der Sandt (1992) takes a bird's eye perspective: his approach is explicitly aimed at all presuppositions, and two of the three examples he gives of


The respective DRSs for (43.a) and (43.b) are structurally similar: both have a presuppositional DRS in the consequent presupposing the existence of a group of girls. There are two potential antecedents for this presupposition in the antecedent of which the new teacher is ruled out on the basis of number. If we assume that there is no relevant hearer knowledge, then there will be no full match between girls and pupils. However, there will be an H-model M in which [girlll M n [pupilll M i 0. In other words: our modified algorithm predicts that the examples in (43) display a partial match, and a genuine ambiguity between a binding and an accommoda tion reading is predicted. The binding reading for (43.a/b) can be paraphrased as (44).

386

Anaphoric Noun Phrases: Towards � Full Understanding of Partial Matches

partial matches involve non-NP presuppositions, repeated below. (46)

viz.

(1 2.b) and (1 2.c)

If John murdered his wife, he will be glad that she is dead. b. If someone at the conference solved the problem, it was John who solved it a.

5 .2

Van Deemter: generalization of anaphora

The revised theory proposed here has certain things in common with the proposal in van Deemter (1 992). In particular, both theories take a DRT based approach to anaphora and extend it to cover the context dependent interpretation of NPs of all major syntactic/semantic categories. Also, both theories use Westerscihl-style context sets as one of their basic mechanisms.


Intuitively, it is clear why these two sentences display a partial match ambiguity. In (46.a) the phrase be glad that triggers a presupposition that John's wife is dead. Since 'being murdered' is more informative than 'being dead', this would be a partial match of type I. However, various problems present themselves. The consequent contains a propositional presupposition triggered by glad to the effect that John's wife is dead. If such propositional presuppositions are anaphoric (which, according to some people, they are not; see e.g. Zeevat 1 992: 396ff ), then they refer to other propositions. This would require an extension of DRT to include discourse markers for propositions {maybe along the lines of Asher 1 993). How this is to be done, and what the consequences are for the modified resolution algorithm are important questions for future research. In section 3 we noted that presuppositions may but need not be bound in situ. The definition of BINning given in section 4.2 does not do full justice to this, as it leaves the bound presupposition always in situ. We feel that in many cases the in situ interpretation is the most preferred binding reading. "• However, this is not to say that other binding interpretations are necessarily unavailable (see e.g. (23) ). Strand (1 997) allows the binding site to be either the source DRS or the target DRS, or one of the sub-DRS sandwiched between these two extremes. While we think that this is an interesting idea, it also raises a number of questions. To name but one, given the variety of non-equivalent binding readings which become available in this way, which is the most preferred one? We do not have clear intuitions on this issue. To clarify it, we suspect that the distinction between presupposed and asserted information needs to be taken more seriously than we have done in this article. ' 3 In the mean time, we propose to stick to definition 2.

Ernie! Krahmer & Kees van Deemter 387 Yet, there are a number of vital differences, most of which derive directly from the work of van der Sandt. A relatively superficial difference lies in the fact that the present proposal is embedded in Kamp & Reyle's theory of plurals, rather than in that of van Eijck ( 1 98 3 ). Another nontrivial difference between the present account and that in van Deemter (1992) lies in the use of the Milsark-Partee perspective on weak/accented DETs, which allows us to treat non-definite NPs in the same way as definite NPs. A much more substantial difference lies in the fact that we use van der Sandt's unresolved representations (plus a version of his resolution mechanism) to make

6

C O N CLUS I O N

Let us take stock. One way of looking at our present endeavour is as a reconstruction of van der Sandt's theory of presuppositions. However, as stated in the introduction we think of this exercise as part of a wider research programma on the context-sensitivity of interpretation, in the spirit of Westerscihl ( 1 98s ), Partee (1984, 1987) and many others. We have primarily been concerned with the context-sensitive interpretation of noun phrases. Our approach rests on the assumption that every NP can be dependent on the linguistic context for its interpretation. Two complicating


predictions about interpretive preferences. Context variables play an essential role in the present account. In van Deemter (1992) it was assumed, for simplicity, that a given NP could take any suitable and accessible reference marker as its anaphoric antecedent. The choice between possible antecedents, and the choice between an anaphoric or a non-anaphoric interpretation, was not modeled by the theory. Instead, it gave a set of possible interpretations, all except one of which involved Westerstahl-style restricted quantification. In the account of the present article, by contrast, two stages can be distinguished. The first stage results in an unresolved (i.e. underspecified) representation, containing unresolved context variables (i.e. context variables that have not yet been equated to the domain of discourse or some previously established reference marker). The second stage delivers a partially ordered set of possible interpretations resulting from the different possible resolutions of the context variables, subject to the new rules governing interpretive preferences. Throughout this article we encountered various other factors which may cause interpretations to be ruled out, including, continuations, the IAH, hearer knowledge, and (de )accenting. Our understanding of the interpretive preferences predicted by the presuppositions-as-anaphors theory is that these preferences serve to choose among those interpretations that remain when these other factors have done their constraining work.


Acknowledgements Earlier versions of this work were presented at 'De Dag': Workshop on Definites (Groningen, November I 996), the DIP colloquium (Amsterdam, March I997), and the Eleventh Amsterdam Colloquium (December I997). We thank the audiences on these various occasions as well as David Beaver, Paul Dekker, Bart Geurts, Barbara Partee, Paul Piwek, Rob van der Sandt, Remko Scha, Mariet Theune and the reviewers for discussion, comments and suggestions. As always, we alone are responsible for all errors in this article. EMIEL KRAHMER Center for Research on User-System Interaction Eindhoven University of Technology PO Box 5 13, NL s6oo MB Eindhoven, The Netherlands e-mail: [email protected] uri: http://www.tue.nl/ipo/people/krahmer

IPO,

KEES VAN

DEEMTER Information Technology Research Institute (ITRI) University of Brighton Lewes Road Brighton BNz 4G), UK e-mail: [email protected] uri: http://www.itri.brighton.ac.uk/"'Kees.van.Deemter/

Received: I 8.o 1 .98 final version received: I 9.1 1 .98


factors are (i) that often the dependency relation between a would-be anaphor and a would-be antecedent is partial, and (ii) that the relation is by no means unrestricted. We have discussed a number of these restricting factors, in particular (de-)accenting, hearer knowledge and the Informative Anaphors Hypothesis (the IAH). We have argued that the presuppositions-as-anaphors theory of van der Sandt (1992) provides an excellent starting point for the current exercise. Beaver (1995: 61-2), discussing some problems related to the partial match phenomenon, supposes that these problems will be solved when the theory matures.'4 We indeed think that the problems with partial matches can be solved by refining and extending van der Sandt's algorithm, and we have tried to offer such a refinement. The resulting version of the presuppositions-as anaphors theory differs from the original proposal (van der Sandt 1 992) mainly in the following respects: ( 1 ) The new version contains a precise definition of the 'partial match' phenomenon; (2) we have modified the resolution algorithm in such a way that-in accordance with van der Sandt's intuitions-partial match sentences come out as genuine ambiguities; and (3) binding is redefined in such a way that non-identity anaphors receive adequate interpretations. Mter our exposition of the revised theory we have shown how it may be generalized to account for context-dependent interpretations of other classes ofNPs than those discussed by van der Sandt.

Emiel Krahmer & Kees van Deemter 389

NOTES 1 Langendoen Savin (1971: 54}: 'how [are] the presupposition and assertion of a complex sentence . . . related to the presupposition and assertion of the clauses it contains?' 2 There do exist some potential counter examples to the generalization proposed in the IAH Consider, for example the following 'politically correct' usage of the female pronoun. .

.

However, we are unsure whether exam ples such as (17) are real counter examples to the IAH For instance, it has been argued by various people that pronouns are essentially devoid of semantic content (by van der Sandt 1992, to give but one example), so to what extent can they add information? 3 We are aware of the fact that para phrasing DRSs in natural language is not always an easy matter. Both Geurts (p.c.) and van der Sandt (p.c.) objected to this paraphrase by saying that it blurs the distinction between asserted mate rial (sons) and presupposed material (young children). This is true, and we agree that this distinction is important (see e.g. Krahmer 1996, 1998). How ever, (i) the theory presented in van der Sandt (1992) makes no distinction between asserted and presupposed material in the resolved DRS. And (ii} the main point of our paraphrase is to indicate that the presuppositional material ends up in the antecedent of the conditional. 4 Here and elsewhere we assume that a model M is a tuple of the form (DM , IM ); where DM is the 'domain of discourse'. DM consists of a set of atomic entities D and an associated partial ordering C, together forming .

_


(17) If the reader has studied example (17), she might come to the conclusion that it constitutes a counterexample to the IAH

an atomic, &ee, complete upper semi lattice with zero, c£ Kamp & Reyle (1993: 425). IM is an interpretation function {combining NameM and PredM, Kamp & Reyle: passim). We drop the M subscripts on domain and interpretation function where this can he done without crearing confusion. s These clauses are variants of clause {ii : g : i) of definition 4-3·7 of Kamp & Reyle (1993: 426). Recall our remark that all predicates used in this article applied to plural (non-atomic) discourse referents are strictly distributive. That is: IM(17*) = {X I Vy E X : y E IM('IJ) }. 6 This rule is an instance of CRNP [Quant = +], Kamp & Reyle (1993: 3 1 8, 347). Below it is replaced by a more general rule. The referent z is an atomic referent when the CN is [+ sg] and a non-atomic referent when the CN is [- sg]. In other words: we assume that the child presupposes the existence of at least one child and that the children presuppose the existence of at least two children. Nothing hinges on this assump tion. Suppose for instance that one wants the existence presupposition always to be 'at least one', in that case z should always be an atomic referent. 7 Remember that CN(Z} and CN C (z) should be read as CN • (z) and CN •C ( Z) respectively. 8 This entails that an interpreter has to remember how an object was intro duced. Jarvella & Herman (1972) have shown that listeners are good at recall ing the last sentence they heard, but that recall of less recent material soon declines to 'nearly chance level'. Inter estingly, Bates, Masling & Kintsch (1978) have shown that when the func tion of an utterance is to introduce new riferents literal memory improves. 9 A more accurate modelling would make PA a partial order (because several dis course referents may be introduced at

390 Anaphoric Noun Phrases: Towards a Full Understanding of Partial Matches reading of the sentence comes about via a de re construction of the presup position trigger with respect to believe and that this presupposition is subse quently bound in situ. There is also a de dicto reading, but this reading is less preferred for the current example, probably because the description the little bastards is the 'responsibility' of the speaker {in the sense of Quine Kaplan-Lewis). Heim {I992) argues that for each presupposition trigger, a de re reading is always preferred over a de dicto reading, but in our opinion this is too strong a claim. We think that, at least for the presupposition triggers under discussion here, there is a sys tematic de re/de dicto ambiguity, where other factors (plausibility, world knowl edge, 'who is responsible for which description') may sometimes indicate a preference for one or the other. I 3 See Geurts & van der Sandt (I 997) for a similar plea. Krahmer {I996, I998) offers a version of the presuppositions as-anaphors approach in which presup positions remain presuppositional after resolution. I4 In fac�. we think that Beaver's examples are not problematic for van der Sandt's theory as such. The phenomena dis cussed by Beaver (an example is {I6.b)) all fall in category IV (anaphor is more informative than antecedent), and we have seen that these are often marked for independent reasons.

REFERENCES Asher, N. {I993), Reference to Abstract Objects in Discourse, Kluwer Academic Publishers, Dordrecht. Bach, K. {I987). Thought and Reference, Oxford University Press, Oxford. Barwise, J. & Cooper, R {I98I), 'General ized quantifiers and natural language', Linguistics and Philosophy, 8, I 59-2I 9. Bates, E., Masling, M. & Kintsch. W. {I978), 'Recognition memory for aspects

of dialogue', journal of Experimental Psychology, 4. I 87-97. Beaver, D. {I992), The kinematics of pre supposition', in P. Dekker & M Stokhof (eds.), Proceedings of the 8th Amsterdam Colloquium, ILLC, Amsterdam, I7-36. Beaver, D. {I995), 'Presupposition and assertion in dynamic semantics', Ph.D. dissertation, University of Edinburgh. to appear with CSLI Publications.


the same level and these are 'equally far away' from the source-DRS), but we ignore this here for the sake of simpli city. Nearness is formally defined in terms of subordination. Thus: a is nearer to a DRS ci> than b iff (i) a and b are introduced in different DRSs, (ii) the DRS in which b is introduced subordinates the DRS in which a is introduced, and (iii) the latter DRS subordinates ci>. a and b are equally near to ci> if neither a nor b is nearer to ci> than the other. IO The constant D refers to the domain of discourse: [DD M = IM(D) = DM. I I See e.g. Russell {I9I9: I74) and Kneale (1962). This proposal was criticized by Kripke {I98o: 68-73). Cresswell {I985) and Bach {I987) have proposed refuta tion of Kripke's criticism. See Geurts (1995: 23) for some discussion. I2 The situation is more difficult when presuppositions interact with inten sional operators like might or believe. Sometimes presuppositions appear to be bound outside the scope of an inten sional operator. Consider If John has children, his mother believes that he'll spoil the little bastards. In our opinion, the preferred reading of this sentence is If john ha.s children, they'll be little bastards and his mother believes that he'll spoil them (and not, as van der Sandt I 992 predicts, If John has children and they are little bastards, his mother believes that he'll spoil them). We think that the preferred

Emiel Krahmer & Kees van Deemter 39I Beaver, D. ( I996), 'Presupposition', in J. van Benthem & A. ter Meulen (eds.), The Handbook of Logic and Language, Elsevier, Amsterdam, 939- 1 008. Cresswell, M ( I 98 5 ), Structured Meanings:

The Semantics of Propositional Attitudes, MIT Press, Cambridge, MA. van Deemter, K. { I992), 'Towards a gen eralization of anaphora', journal of Semantics, 9, 27-5 I . van Deemter, K. { I994), 'What's new? A semantic perspective on sentence accent', repre sentation theory and plurality', in A. ter Meulen (ed.), Studies in Mode/theoretic Semantics, GRASS I , Foris, Dordrecht. Geurts, B. ( I995 ). 'Presupposing', Ph.D. dissertation, Osnabriick. Geurts, B. & van der Sandt, R ( 1997) , 'Domain restriction', in P. Bosch & R van der Sandt (eds.), Focus: Linguistic,

Cognitive and Computational Perspectives,

Cambridge University Press, Cambridge. Haviland, S. & Clark, H. ( I974), 'What's new? Acquiring new information as a process in comprehension', journal of

Verbal Learning and Verbal Behavior,

1 3,

5 1 2-2 1. Heim, I. ( 1983 ) , 'On the projection problem for presuppositions', in M. Barlows (ed.)

Proceedings of the West Coast Conference on Formal Linguistics, 2, Stanford Uni versity, Stanford, I I4-25. Reprinted, in S. Davis (ed.) ( ! 99 1 ), Pragmatics, Oxford University Press, Oxford, 397-405.

Heim, I. ( I992), 'Presupposition projection and the semantics of attitude verbs',

journal of Semantics, 9,

Hobbs, J.

&

algorithm scopings',

Shieber, for

I 8 3-222. S. ( I987),

generating

'An

quantifier

Computational Linguistics,

I

3,

47-63.

Jarvella, R & Herman, S. ( 1972) , 'Clause structure of sentences and speech

The Representation of(In)Definiteness, MIT Press, Cambridge, MA. 27o-8 5. Kamp, H. { I 9 8 I ), 'A theory of truth and semantic representation', in J. Groenendijk, T. Janssen & M Stokhof (eds.), Formal Methods in the Study of Language, Part I, Mathematical Centre 1 3 5 , Amsterdam, 277-322. Tracts Reprinted, in J. Groenendijk, T. Janssen & M. Stokhof (eds.), Truth, Interpretation and Information, Foris, Dordrecht. Kamp, H. & Reyle, U. { I 99J ) , From Discourse to Logic, Kluwer Academic Publishers, Dordrecht. Karttunen, L. ( 1969), 'Pronouns and vari ables', in Papers from the Fifth Regional

Meeting of the Chicago Linguistics Society, University of Chicago Press, Chicago,

108-1 5· Karttunen, L. ( 1974), 'Presupposition and linguistic context', Theoretical Linguistics, I,

1 8 I-94·

Kneale, W. { 1962) , 'Modality, de dicto and de re', in E. Nagel, P. Suppes & A. Tarski {eds.) Logic, Methodology and the Philosophy of Science, Stanford University Press, Stanford, 622-33Krahmer, E. ( 1995 ) , 'Discourse and presup position', Ph.D. dissertation, University of Tilburg. Krahmer, E. ( 1996) , 'Presuppositional discourse representation theory', in P. Dekker & M Stokhof (eds.), Proceedings ofthe Tenth Amsterdam Colloquium, ILLC, Amsterdam, 499-5 I 8.

E. ( I 998 ), Presupposition and Anaphora, CSLI Lecture Notes No. 89, CSLI Publications, Stanford. Krahmer, E. & van Deemter, K. ( I997), Krahmer,

'Partial matches and the interpretation of anaphoric noun phrases', in P. Dekker, M Stokhof & Y. Venema (eds.),

Perception and Psychophysics,

Proceedings of the Eleventh Amsterdam Colloquium, 205-10. Krahmer, E. & Piwek, P. ( 1997), 'Pre

J 8 I-J84de Jong, F. { I987), 'The compositional

supposition projection as proof con struction', in H. Bunt & R Muskens

processing', II,

nature of (in)definiteness', in E. Reuland

(eds.), Computing Meanings: Cu"ent Issues


journal ofSemantics, I I, I - J I . van Eijck, J. ( I983 ), 'Discourse

& A. ter Meulen (eds.),

392

in

Anaphoric Noun Phrases: Towards a Full Understanding of Parrial Matches

Computational Semantics,

Academic appear. Kripke, S.

Publishers,

Kluwer

Dordrecht,

to

( 1 980), Naming and necessity,

Blackwell, Oxford. Langendoen, D. & Savin, H ( 1 97 1 ) , "The projection problem for presuppositions', in C. Fillmore & D. Langendoen (eds.)

Studies in Linguistic Semantics, Holt, New York, s s -6o. Link, G. ( 1983 }, "The logical analysis of

Meaning, Use and Interpretation of Language, De Gruyter, Berlin. Milsark, G. ( 1977) , "Towards an explanation of certain peculiarities in the existential construction in English', Linguistic

Analysis, 3, 1-30.

Partee, B. ( 1984), 'Nominal and temporal anaphora', Linguistics and Philosophy, 7,

243-86.

Partee, B. ( 1987) , 'Binding implicit variables in quantified contexts', in Papers from

the 25th Annual Regional Meeting of the Chicago Linguistics Society, Chicago, 342-66. Piwek, P. & Krahrner, E. ( 1 997) , 'Pre suppositions in context: constructing bridges', in P. Brezillon & M. Cavalcanti (eds.), Formal and Linguistic Aspects of

Context,

Kluwer Academic Publishers,

Dordrecht, to appear. Russell, B. ( 1919), Introduction matical Philosophy, George Unwin, London.

to Mathe Allen &

et al. (eds.), Semantics arid Contextual Expression, Foris Publications, Dordrecht. van der Sandt, R ( 1 992) 'Presupposition ,

projection

as

anaphora

resolution',

journal of Semantics, 9, 3 3 3-77Sidner, C. ( 1 979), "Towards a

Compu

tational Theory of Definite Anaphora Comprehension in English Discourse', Ph.D. dissertation, MIT, Cambridge,

MA.

Soames, S. ( 1 982), 'How presuppositions are inherited: a solution to the projec tion problem', Linguistic Inquiry, I 3,

483-545·

R ( 1 973), 'Presuppositions', Journal of Philosophical Logic, 2, 447-457. Stalnaker, R ( 1974), 'Pragmatic presuppo sitions', in M. K. Munitz & P. K. Unger (eds.), Semantics and Philosophy, Stalnaker,

New York University Press, New York,

1 97-2 1 3 .

Strand, K. ( 1 99.7) , 'Accommodation and computability', draft Ph.D. dissertation, University of Oslo. Westerst:ihl, D. ( r985 ), 'Determiners and context sets', in J. van Benthem & A. ter Meulen (eds), Generalized quantifiers in natural language, Foris Publications, Dordrecht. Zeevat, H. ( 1 992) , 'Presupposition and accommodation in update semantics',

journal of Semantics, 9, 3 79-4 r 2.

Zucchi, A. ( 1 995 ), "The ingredients of definiteness and the definiteness effect',

Natural Language Semantics

3,

33-78.


plural and mass terms: a lattice-theore tical approach', in R Bauerle, C. Schwarze & A. von Stechow (eds.),

van der Sandt, R ( 1 989) , 'Presupposition and discourse structure', in R Bartsch

© Oxford University Press 1998

journal of&trulntics 1 5: 393-41 6.

Type-shifting and Scrambled Definites JAAP VAN DER DOES and HELEN DE H O O P ABN AMR O Bank N. V. and University of Utrecht Abstract

1

I NTRODUCT I O N

It is well known from the literature that some languages exhibit scrambling (which we use here as a descriptive term for the occurrence of an object to the left side of an adverb in an SOY-language), but whether or not scrambling is allowed to occur depends on several factors. The following examples will indicate the facts to be explained.


It has often been observed in the literature that when a language allo�s for scrambling defmite noun phrases may freely scramble whereas indefinites are subject to certain restrictions. Many have assumed that scrambling is related to the structure of the surrounding discourse such that familiar or topical (anaphoric) noun phrases must scramble whereas new or non-topical (non-anaphoric) are not allowed to scramble. We will show, however, that in similar discourse configurations certain indefmites are not allowed to scramble, whereas definites freely scramble. We account for the difference in scrambling behaviour between definites and indefinites within a version of what might be called Optimality Theoretic Semantics (OTS). Our version of OTS gives a new t\\fist to the strategy of Partee and Rooth (1983) to assign NPs a preferred type-often the simplest type possible-by using the preferences to phrase soft constraints. In this way one obtains a means to judge correct type-theoretic derivations as more or less optimal, and to vary the grammaticality judgment of the output accordingly. This is in stark contrast with the older strategy, where the output of correct derivations on the basis of unpreferred types would still be judged grammatical. More in particular our explanation of scrambling behaviour will be along the following lines. Different types of NPs come with different basic word orders, a variation which is consistent with the frequency of use of these constructions. Deviation of basic word order yields a shift in meaning. Defmites, unlike indefinites, allow for an interpretation in the scrambled and in the unscrambled position without change of meaning. Accordingly, scrambling of definites is truly optional. Scrambled indefinites, on the other hand, are either infelicitous or may induce a change in interpretation; they cannot be optional. This is especially clear for inputs consisting of an (in)definite with a light verb. Then, defmites are free to scramble, whereas indefmites cannot scramble. . We show how the interpretive tendencies that arise with scrambling can be modelled using a notion of optimal derivation, phrased in terms of the preferred types of (in)definites and yerbs. Two appendices provide the technical details of the seMantic insight on which our approach is based.

394 Type-shifting and Scrambled Definites

In ( r) the NP aile krakers 'all squatters' may occur both at the right-hand side and the left-hand side of the adverb gisteren 'yesterday', without any change in the meaning of the NP.

(r)

a.

dat ik gisteren aile krakers heb gesproken that I yesterday all squatters have spoken 'that I talked to all squatters yesterday' b. dat ik alle krakers gisteren heb gesproken that I alle squatters yesterday have spoken 'that I talked to all squatters yesterday'

(2) a. dat ik gisteren twee krakers heb gesproken that I yesterday two squatters have spoken 'that I talked to two squatters yesterday' b. dat ik twee krakers gisteren heb gesproken that I two squatters yesterday have spoken 'that I talked to two squatters yesterday' In (2b), there is a shift toward a quantificational reading of the NP, a shift which is less prominent in case of (2a). For singular NPs the situation is even clearer. The unscrambled singular indefinite in (3a) is perfectly in order, while the scrambled indefinite is (on the verge of being) ill formed. (3) a. dat ik gisteren een kraker heb gesproken that I yesterday a squatter have spoken 'that I talked to a squatter yesterday' b. dat ik ?een kraker gisteren heb gesproken that I een kraker yesterday have spoken 'that I talked to a squatter yesterday' By contrast, the singular definites in (4) are perfectly acceptable, and the scrambled and the unscrambled versions are synonyms.

(4)

a.

dat ik gisteren de kraker heb gesproken that I yesterday the squatter have spoken 'that I talked to the squatter yesterday' b. dat ik de kraker gisteren heb gesproken that I the squatter yesterday have spoken 'that I talked to the squatter yesterday'


However, scrambling may have an interpretive effect, and it is constrained by the kind of the NP involved (definite, indefinite . . . ). Although in (2) the weak NP twee krakers 'two squatters' may scramble, there is a noticable shift in possible interpretations.

Jaap van der Does and Helen de Hoop 395

In this respect, definite NPs and pronouns are different. Definite NPs are free to scramble, but pronouns are often forced to scramble. As a response to the question in (sa) or (sb), both (6a) and (6b) are felicitous. Heb je Jane gisteren het geld gegeven? have you Jane yesterday the . money given 'Did you give Jane the money yesterday?' b. Heb je Jane het geld gisteren gegeven? . have you Jane the money yesterday given 'Did you give Jane the money yesterday?' eergisteren (6) a. Ja, ik heb Jane al Yes I have Jane already the-day-before-yesterday het geld gegeven the money given 'Yes, I gave Jane the money already the day before yesterday' geld al eergisteren · b. Ja, ik heb Jane het Yes I have Jane already the money the-day-before-yesterday gegeven gtven 'Yes, I gave Jane the money already the day before yesterday' (s)

a.

(7) a. ?*Ja, ik heb Jane al eergisteren het gegeven yes I have Jane already the�day-before-yesterday it given 'Yes, I gave it to Jane already the day before yesterday' eergisteren gegeven b. Ja, ik heb Jane het al yes I have Jane it already the-day-before-yesterday given 'Yes, I gave it to Jane already the day before yesterday' c. Ja, ik heb het Jane al eergisteren gegeven yes I have it Jane already the-day-before-yesterday given 'Yes, I gave it to Jane already the day before yesterday' Note that the examples make it clear that the distribution cannot be explained in terms of the anaphoric behaviour of the object NPs, contra to what has been proposed in the linguistic literature. Against the backdrop of the general situation sketched by the previous examples, the aim of this paper is to provide an explanation for a very striking and clear linguistic puzzle concerning the possible syntactic positions that a certain type of definite and indefinite object may occupy in languages like Dutch and German. The definites and indefinites we are concentrating on look very similar. Intuitively, they form a semantic unity together with a light verb. Light verbs are verbs which do not have much semantic content of their own, which is indicated by the fact that they


By contrast, pronouns only allow the scrambled versions in (7b-c).


cannot bear contrastive stress. They combine with object NPs to form a meaningful whole. Yet the definite objects can occur to either the left or the right of an adverb (i.e. in scrambled or unscrambled position), whereas the indefinites may only occupy the unscrambled position, adjacent to the verb. The crucial difference is illustrated with respect to the Dutch pairs in (8)-(10). These pairs differ minimally in the (in)definiteness of the object, while the light verb they combine with remains the same. Consider de was doen 'do the laundry' versus een plas doen 'take a piss' in (8): (8)

a.

c.

d.

Or de bus nemen 'take the bus' versus een enkeltje nemen 'get a single' in (9):

(9)

dat ik altijd de bus neem that I always the bus take 'that I always take the bus' b. dat ik de bus altijd neem that I the bus always take 'that I always take the bus' c. dat ik altijd een enkeltje neem that I always a single take 'that I always get a single' d. dat ik een enkeltje altijd neem that I a single always take 'that I always get a single' a.

Or finally de mazelen hebben 'have the measles' versus kinderen hebben 'have children' in (xo): (xo)

a.

dat ik ook de mazelen heb that I also the measles have 'that I also have the measles'


b.

dat ik nog de was moet doen that I still the laundry must do 'that I still have to do the laundry' dat ik de was nog moet doen that I the laundry still must do 'that I still have to do the laundry' dat ik nog een plas moet doen that I still a piss must do 'that I still have to take a piss' * dat ik een plas nog moet doen that I a piss still must do 'that I still have to take a piss'

Jaap van der

Does

and Helen de Hoop 397

(10) b. dat ik de mazelen ook heb that I the measles also have 'that I also have the measles' c. dat ik ook kinderen heb that I also children have 'that I also have children' d. *dat ik kinderen ook heb that I children also have 'that I also have children'

2 S CRAMBLED DEFI NITES It has often been observed in the literature that when a language allows for scrambling, definite and other strong NPs may freely scramble, whereas indefinite and other weak NPs are subject to certain restrictions. The data in (8)-(10) above are in accordance with this observation. The question that springs to mind is why the indefinites in (8)-(10) are not allowed to scramble. A related question is whether NPs that do scramble share a certain characteristic: are there any properties of the object, the predicate, or the context, that actually trigger scrambling? Many recent approaches to scrambling phenomena argue that this is indeed the case, i.e. scrambling is not truly optional, it is driven.


Current :1nalyses of scrambling would not be able to account for the paradigm presented here, as we will point out in the next section. The fact that the definites in (8)-(10) behave exactly like other definites and unlike predicative indefinites indicates that a proper analysis of scrambling should not be based on a difference in properties of the objects, such as familiarity, anaphoricity, topicality and/or focus. Instead, the explana tion should be sought either in a syntactic difference between definites and indefinites, or in a semantic difference. We will argue in favour of the latter option and account for the scrambling differences between definites and indefinites within an Optimality Theoretic Semantic (OTS) variant of the type-shifting perspective, elaborating on certain insights of Partee (1987), de Hoop (1992), Diesing & Jelinek (1995), and van Geenhoven ( 1 996). In section 3 we will discuss the semantics of definites and indefinites and in section 4 we present a solution. to the puzzle by attributing the difference in syntactic behaviour between definites and indefinites to a difference in their Optimality Theoretic Semantics. But first we shall adduce some further evidence that definites are not forced to scramble.


( 1 I) weil ich selten die Katze streichle

since I seldom the cat pet 'since I seldom pet the cat' ( 1 2) weil ich nicht das Rosamunde-Quartett gespielt habe since I not the Rosamunde Quartet played have 'since I haven't played the Rosamunde Quartet' Diesing & Jelinek predict the sentences in (I I) and (12) to be ungrammatical. They actually mark the sentences with the grammaticality indication '*?'. Yet, the sentences in (I I) and ( 1 2) are not ill-formed at all, not even slightly. This means that the scrambling theory of Diesing & Jelinek that is supposed to cover indefinites as well as definites of a non quantificational type (since existential closure is only applied to free variables, it does not affect quantificational NPs) cannot account for the clear differences between definites and indefinites with respect to scrambling. With respect to the puzzle presented in the introduction above, one might argue that the definites in (8)-(10) are of a special kind and need not be familiar. Clearly, de bus 'the bus' in (9) does not have to be introduced in the discourse before and it is not like the sun either, a definite which denotes one and the same individual through contexts. The bus-type of definite is actually hardly referential. Therefore, the definites in (8)-(10) are not especially problematic for Diesing & Jelinek's analysis. What is, however, problematic for their analysis is that definites that are indeed referential or even anaphoric (i.e. definitely related to the previous discourse) do not obligatorily scramble either, as was illustrated by (s)-(6), and as is once again illustrated below:


Diesing & Jelinek (1995) claim that scrambling of definites is semanti cally driven. (In)definites are taken to be NPs of type e and (et ) that introduce free variables. Variables can get an existential interpretation in unscrambled (VP-internal) position, thanks to a default existential closure operator that is postulated at the VP-level. According to Diesing & Jelinek, definite NPs which receive a referential interpretation have to be in scrambled position. Otherwise, they would get bound by the default existential closure operator. This then would be a- violation of a novelty condition (c£ Heim 1982) that requires variables bound by existential closure to be new in the discourse. So, in languages like German and Dutch, referential definite NPs are predicted obligatorily to scramble in order to get out of the scope of the existential closure operator. Crucially, however, this prediction is not borne out. Referential definites do not obligatorily scramble. The sentences that Diesing & Jelinek claim to be ill formed are in fact perfectly well-formed:

Jaap ( 1 3)

van

der Does and Helen de Hoop 399

Paul heeft een kat die de laatste tijd een gespannen indruk maakt 'Paul has a cat that seems to be under stress, recently' a.

Misschien komt dat omdat Paul zelden· de kat aait

comes that because Paul seldom the cat pets maybe 'That's maybe because Paul hardly ever pets the cat' b. Misschien komt dat omdat Paul de kat zelden aait maybe

comes that because Paul the cat seldom pets

'That's maybe because Paul hardly ever pets the cat' To sum up, all types of definites, the highly referential and anaphoric ones as well as the weak ones and everything in between, freely scramble.

(6)

do not have to scramble (contra Diesing & Jelinek · 1 995, and others),

whereas non-topical, non-specific, non-anaphorically destressed definites like the ones in (8)-(10) may scramble (contra de Hoop 1 992; Choi 1 996; Neeleman & Reinhart 1998; and others). The conclusion must be that scrambling is truly optional for all definites, irrespective of the sentence or discourse context in which they occur.

3 S E M A N T I C S O F D E F I N I TES We would like to claim that the differences in scrambling behaviour between definites and indefinites is due to two interrelated semantic features: o

definites can be independent of their semantic context m a way indefinites cannot;

o definites are naturally viewed as having type

e

while indefinites are not.

To sustain this claim, we give a quick overview of the semantics of (in)definites in type

e

'referential', type

(et)

'predicative', and type

'quantificational'. Here we start from Partee's

(1987)

( (et)t)

seminal paper, adding

new observations as we go along.

Type e. As is well known, referential defmites have a natural interpretation in type e. For

P a property holding of a single entity, the meaning of 'the P' t(P). Here, L is the partial function o f type ((et)e), which

can be given as

returns the element of its argument provided this element is a singleton:


That is, definites may either occupy the scrambled or the unscrambled position relative to an adverb, and there does not seem to be a property of either the definite itself or the context in· general that forces or prohibits scrambling. Anaphoric, referential, familiar, topical definites like the one in

400

Type-shifting and

Scrambled Definites

( I4) �(x) = d, if X = { d} for some d. For instance, the VP in {I sa) {I S)

a.

b.

has

{I sh) as its semantics:

de bus nemen the bus take 'to take the bus' A.x.nemen(x, ty (bus(y) ) )

{I6)

a.

b.

een enkeltje nemen a single take 'to get a single' A.x{nemen(x, c:y{nemen(x, y)

&

enkeltje(y ) ) ) )

Here, c: means a choice from its argument provided this set is non-empty. This is of course equivalent to requiring that there be a single that one takes. It should be observed, however, that the choice cannot be from just the singles; for then, choosing a single that one does not take makes the corresponding variant of {I6b) false, whereas the meaning obtained from higher type NP denotation-for example, the quantificational one-could be true. By contrast, " can be applied independently of the verb; for each y in (I sh) there is a unique bus regardless of whether x takes this bus or not (c£ section 4 for a treatment of dependent definites). Due to the dependence on the verb, indefinites are not simply of type e. We therefore do not consider this option any further but follow the suggestion of Partee that indefinites live more naturally in the predicative type ( et). Type (et). The predicative meaning of an (in)definite NP is of type (et); it is the denotation of its noun. In the case of definites this denotation should be a singleton, and it should be non-empty in the case of indefinites. Clearly, for singleton properties P, the predicative meaning of a definite is essentially the same as its referential meaning:


with bus denoting a singleton. Partee suggests that a similar shift from ( et) to e is available for indefinites if these are viewed as 'novel' variables in the sense of DRT. However, variables are syntactic rather than semantic entities, and are therefore unsuitable as an ingredient of the required shift.' In seeking a semantic analogue of this idea, the use of choice functions seems the closest one can get (see Meyer-Viol I99S for an overview and development). But then one has to represent dependencies that are absent in the case of referential definites. The reason is that using the NP semantics in the lower type should not change the meaning of the sentence. Therefore, the semantics of (I6a) is (I6h):

Jaap van der Does and Helen de Hoop

401

(17) P = {d} iff t.(P) = d for any d. In Partee's terminology, the predicative definite in type (et) can be obtained by applying the total injective function

(I8) ident = .Xx.Xy (x = y) of type (e(et)) to t(P):

(I9) P = ident(t(P))

=

.Xy(t(P) = y).

Type ( (et )t ). First consider the prototypical singular (in)definites some N and the N for a noun N. In generalized quantifier theory, the semantics of some N is the set of all sets that have a non-empty intersection with the interpretation of N (here written as N):

(2o) some N := {X: N n X ;f= 0}. The semantics of singular the N can be seen as obtained from restricting the domain of quantification of some N to singletons. Just recall that the semantics of the N is usually given as: (2 I) the N := {X: N � X

&

INI = I }.

But this is equivalent to a 'restricted' version of the semantics of some N: (22) the N := {X: X E some N

&

INI = I }.

In general, quantificational definite NPs can be seen as 'domain restricted' indefinites, but we defer this discussion to Appendix A. We also trust the reader is familiar with how type-shifting relates the quantificational denotations with the referential and predicative ones (but see Appendix B for a slightly generalized view). 4 THE P UZZLE ·SO LVED

In this section we will provide an explanation for the difference between indefinites and definites in the examples (8)-( 10) above. Although the definites in (8)-(10) can intuitively be characterized as predicative, just like


In set notation: P = {t(P)}. Conversely, the partial, surjective t turns a predicative meaning of definites into a referential one. Note that at this level defmites can be seen . as 'restricted' indefinites; singletons are of course special instances of non-empty sets. Mutatis mutandis the same is true for the quantificational treatment of (in)definites.

402


their indefinite counterparts, they differ in syntactic behaviour: the definites can scramble whereas the indefinites cannot. 4. 1

Indefinite NPs

(23) a. Angunguu-p aalisagaq neri-v-a-a A-ERG fish.ABS eat-IND-(+TR)-3SG.3SG 'Angunguaq ate the/ a particular fish' b. Angunguaq neri-v-u-q AABS eat-IND-(-TR)- 3SG 'Angunguaq was eating' c. Angunguaq aalisakka- mik neri-v-u-q AABS fish-INST.SG eat-IND-(-TR)- 3SG 'Angunguaq ate fish' The transitive verb in (23a) is intransitivized in (23b) and (23c): in (23b) there is no object, in (23c) it bears weak (instrumental) case. Van Geenhoven argues that in West Greenlandic an instrumental object and the absence of object agreement are syntactic markers of the process of semantic incorporation (whereas noun incorporation is a morphological realization of this semantic process). Van Geenhoven extends her analysis of semantic incorporation to the semantic and syntactic properties of indefmites in languages like German and Dutch. A (pseudo-)transitive verb like eat - can combine with quantifi cational as well as predicative NPs in Dutch and German. This means that the verb can have two different semantic types, which seems to be in accordance with the fact that the corresponding verb in West Greenlandic can take two different morpho-syntactic forms (see (23) above). Either it is


In de Hoop ( 1992), the predicative reading of indefinites is argued to follow from the semantic type these NPs have, in particular the type of predicative modifier ( ( et ) ( et)) . But van Geenhoven ( 1 996) argues that predicative indefmites are of type ( et) . These predicates are incorporated or absorbed by an incorporating verb of type ( ( et ) ( et )) , and as such introduce a restriction on the individuals that the verb applies to. Their existential interpretation derives from the lexical semantics of the verb (following Carlson 1977). Van Geenhoven defends the view that West Greenlandic incorporated nouns are base generated in verb adjacent position. A semantically incorporated expression does not have to be realized as a syntactic morpheme, though: it can also be realized as a syntactic phrase bearing weak case (c£ also de Hoop 1992). Compare van Geenhoven's examples:


interpreted as an ordinary two-place relation between individuals or it is interpreted as an incorporating verb that combines with a predicative NP. As an illustration, (24a) represents the meaning of the non-incorporating predicate eat, and (24b) of its incorporating counterpart: (24)

.\y-\x ( eat(x, y) ) b. AP-\x [P (y) & eat (x, y)] a.

So an incorporating verb is the result of a shift from type ( e ( et)) to type ( (et ) (et)) as follows: (25) inc(R( t(d))) = AP(et) ,ut:Jt[P (y) & R(x) (y ) ]:

(26) a. wo yi xiaoshi nei xiao-le san ci bian I one hour within do-PERF three time piss 'I have taken a piss three times within the hour' b. wo yi xiaoshi nei xiao-le bian san ci I one hour within do-PERF piss three time 'I have taken a piss three times within the hour' This clearly indicates that the fact that predicative indefinites cannot scramble cannot be accounted for in terms of a syntactic adjacency requirement between the verb and its complement NP. Instead, we attribute the scrambling possibilities of different types of object NPs to different basic positions for these NPs. Deviation of the basic position induces a shift in interpretation (in accordance with the claims made by de Hoop 1992 and Adger 1994). For example, indefinites may scramble, but in those cases it can be argued that they have the type of a generalized quantifier. In other words, an indefmite that scrambles, shifts to a quantificational type and hence is not semantically incorporated:


Note that this shift must be restricted, since not all verbs allow for a predicative interpretation of an indefinite object (see also Diesing 1992 for discussion). According to van Geenhoven ( 1 996), an incorporating verb and a predicative indefinite have to be adjacent to each other in order to allow for the process of semantic incorporation. There can be no type-logical justification for this apparent adjacency requirement, however: the incor porating verb can be type-shifted to derive the same meaning when the verb and the indefinite are not adjacent. Moreover, in Chinese we find semantic incorporation interpretations for indefinites which are not allowed to be adjacent to the verb, as argued by Soh (1 998). In Mandarin, an object following a duration/frequency phrase (non-scrambled object) must bear focus, according to Soh, while an object preceding such a phrase, does not have such a restriction. Predicative indefinites that combine with light verbs, however, cannot occupy the scrambled position at all:

404


opgegeten heeft that Paul two cookies already eaten has 'that Paul has already eaten two (of the) cookies'

(27) dat Paul twee koekjes al

Output

Input

Indefinite NP +

iNP

in (et)

IV

in ((et)(et))

[> (et) ((et)(et))

Light verb

((et)t) (e(et))

The optimal candidate is marked with 4.2

*

*

' ' t> .

Negative indefinite NPs

Before we return to the problem of scrambled definites, let us pay some attention to monotone decreasing indefmites. Given that monotone decreasing indefinites do not license discourse anaphora, the common conclusion in DRT (e.g. Kamp & Reyle 1993) is that these NPs are always of the quantificational type,just like NPs such as everyfish. Ifmonotone decreasing NPs would always be quantificational, however, we cannot account for the fact that they behave like other indefinites in the following context: dat Fred ook geen kinderen heeft that Fred indeed no children has 'that Fred doesn't have children either' (28) b. *dat Fred geen kinderen ook heeft that Fred no children indeed has 'that Fred doesn't have children either'

(28)

a.


We assume that the unscrambled indefinites in (8)-(10) must be semantically incorporated. This is due to the light (non-contrastive) character of the verbs under consideration.z We claim that these verbs prefer to combine with a predicative reading of their object. For indefinite objects, this corresponds to their simplest type as we argued above. Scrambling the indefinite shifts its interpretation to the generalized quantifier type, violating its preferred interpretation as a predicate. This implies an additional deviation of the preferred interpretation of the light verb as incorporating.3 Clearly, scrambling would violate the optimal derivation, i.e. the derivation that yields grammatical output. The optimal derivation is the one in which both the indefinite object and the light verb denote in their preferred type. This information can be represented by means ofa tableau (c£ Prince & Smolensky 1997). The tableau takes as input the syntactic elements an indefinite NP (= iNP) and a light verb (= IV), and evaluates the type-theoretical derivation with respect to two interpretative soft constraints.

Jaap van der

Does

and Helen de Hoop 405

Van Geenhoven (1996) does not treat monotone decreasing indefinite NPs, but de Swart (1997) proposes that apart from a local existential closure ' operation for NPs of type ( et) that are derived from monotone increasing NPs, we also need a local universal closure operation for NPs of type ( et) that are derived from monotone decreasing indefinites. The weak inter pretation of the indefinite monotone decreasing NP in (28) can now follow from the (et)-type of the NP that restricts the individuals that the verb applies to. Universal closure is possible for predicate NPs derived from monotone decreasing NPs.. We can represent the meaning of the verb in this case as in (29b):

--t

4· 3

Definite NPs

Recall that definites other than indefinites can combine with light verbs and still scramble, as in (8)-(1o) above. This might indicate that the definites in (8)-(10) semantically incorporate after all, but even in these cases they freely scramble, just like other definites. Van Geenhoven (1996) argues that definites can only be understood as predicates of 'familiar' variables, whereas the internal argument's variable of a semantically incorporating verb is always 'novel'. In the case of do the laundry, take the bus and have the measles, however, this can hardly be the explanation, since the laundry, the bus, and the measles in these configurations do normally not refer to familiar discourse referents. In the previous section we pointed out that definites can be conceived of as restricted indefinites. This is illustrated in (Jo) for the predicative type of indefinite and definite:


{29) a. Ik eet geen vis I eat no fish 'I don't eat fish' •eat(x, y)] b. .AP.Ax\fy[P(y) · De Swart's local closure operations capture the generalization that weak interpretations of indefinites always involve the closure of a set of individuals which corresponds with the predicative use of the NP. The choice between an existential and a universal closure operation is made on the basis of the minimality/maximality property of the predicatively used NP, which is derived from the monotonicity properties of its underlying generalized quantifier denotation. Thus, we can account for the fact that weak monotone decreasing NPs behave like other predicative indefinites, but unlike definites, with respect to semantic incorporation.


(3o)

>.x(man(x) ) b. >.x (man(x) & Vy (man( y ) a.

-t

x = y) )

(3 I) de vis eten 'eat the fish' >.x(eat(x, Ly (fish(y) ) ) ) But shifting its referential meaning to the corresponding singleton predicate, it combines with the incorporating verb, too.

(32) de vis eten 'eat the fish' >.x3y [eat(x, y) & fish(y)] Moreover, since fish is supposed to be a singleton, (3 I) and (32) have the same meaning:

(33) >.x (eat(x, �y (fish( y) ) ) ) = >. x 3y [eat(x, y )

&

fish(y )J.

In other words, the two possible derivations result in the same meaning. The definite can therefore retain within its preferred type e; it does not have to be predicative. Since definites NPs (= dNP) are interpreted more freely than predicative indefinites, they have more freedom to scramble. This is true for definites as objects of light verbs as well. As noted before, light verbs have their preferred denotation in type ( ( et) (et) ) , but ari interpretation in type ( e (et )) is also available. Definites prefer a meaning in type e. Hence the


This semantic difference between indefinites and definites actually allows us to account for their difference in syntactic behaviour with respect to semantic incorporation and scrambling. While predicative indefinites are dependent on the verb for their existential interpretation-as in van Geenhoven (I 996) and de Swart (I 997)-definites are not. The uniqueness condition of definites is part of the semantics of the definite itself in all its types. It is strongly related to the view that a definite NP denotes in type e: it refers to an object of the kind indicated by its noun. Indeed, the L-operation that can be used to shift the type of a predicative definite in type (et ) to type e is not dependent on the verb. Since the L-operation is only defined in case the uniqueness restriction is fulfilled, it should be taken for granted that when one says one will take the bus, there will be one unique bus in the relevant context. Following the strategy in Partee & Rooth (I 98 3) to interpret an NP in as simple a type as is possible, definite NPs have their basic denotation in type e (as specified by means of the L-operation). As a consequence, they combine with non-incorporating verbs as usual:


combination of a light verb and a definite object involves one violation at least: either the definite has to be interpreted predicatively to combine with an incorporating verb, or the verb has to be interpreted transitively to combine with a referential definite. Consequently, two optimal derivations exist, which nevertheless result in the same meaning. Input

Output

dNP in e *

Definite NP +

t>

(et) ((et)(et))

Light verb

t>

e (e(et))

IV in ((et)(et)) *

(34) dat twee lingusten de bus nemen that two linguists the bus take 'that two linguists take the bus' Chierchia (1995), among others, holds that dependent definites are still independent of the verb; it is just the noun bus 'bus' which is functionally dependent on the linguisten 'linguists'. E.g. for quantificational dependent definites one has: (35) [two x: linguists] [the y : bus (x , y)] ( take (x, y ) )

a consequence, an explanation of why dependent definites scramble as freely as independent ones comes from a parameterized version of the argument given previously. As soon as the parameters of a definite are set, we obtain its familiar denotations in type e and (et). These combine with the incorporating and non-incorporating verb meanings as before to yield synonyms; e.g. (36) holds if bus is a functional noun: As

(36) >.x(take (x, t.(bus) (x) ) )

=

>.x 3y [take (x, y )

&

bus (x, y) ]

But the corresponding dependent indefinite NP prefers to live in type (e(et)) (i.e. the parameterized predicative reading); he1:1ce it cannot combine


We assume that the two constraints involved in determining the optimal derivation in the case of a definite are tied (which means that they are not ranked with respect to each other). This results in two optimal derivations. Note that if we had independen.t evidence that one of the two constraints were stronger than the other, our analysis would still hold, since in the case of definites both derivations yield the same interpretation (in fact, this even extends to the generalized quantifier type of definite). The explanation for the scrambling behaviour of dependent definites is similar, though a little more complicated. In (34) each linguist may take a different bus.


with a non-incorporating verb in the way required. The details of the parameterized Partee triangle can be found in appendix B.

5

WORD ORDER VARIATI O N AND MEA N I N G

(37) a. dat ik gisteren twee krakers heb gezien that I yesterday two squatters have seen 'that I saw two squatters yesterday' b. dat ik twee krakers gisteren heb gezien that I two squatters yesterday have seen 'that I saw two squatters yesterday' Pronouns occur most often in scrambled position. In this basic position they can get either an anaphoric or a deictic interpretation. For pronouns, the anaphoric interpretation is the most prominent. Using a non-basic (in this case, unscrambled) word order induces the deictic meaning which deviates from the preferred meaning.

(38)

a.

dat ik hem gisteren heb gezien that I him yesterday have seen 'that I saw him yesterday' b. dat ik gisteren hem heb gezien that I yesterday him have seen 'that I saw him yesterday'

For defmites there is no basic word order. Scrambling is truly optional and there is no shift i� meaning.


We already pointed out that the meaning of indefinites is shifted when they scramble. In general, we would like to claim that deviation of basic word order yields a shift in meaning (c£ de Hoop 1992; Adger 1 994). However, we do not want to argue that in the case of scrambling in Dutch and related languages, there is only one basic word order (the unscrambled word order) for all NP types. Rather, what is considered the basic word order varies among the different types of NPs and is consistent with the frequency of use of these constructions. That is, the most frequent word order seems to be the basic one: for indefinite objects, this is the unscrambled word order, whereas for pronouns this is the scrambled word order. Indefinites can be interpreted as predicates or as quantifiers in unscrambled position, but only as quantifiers when scrambled. Using a non-basic (in this case, scrambled) word order goes hand in hand with a deviation of the preferred meaning.


(39)

a.

dat ik gisteren de president heb gezien that I yesterday the president have seen 'that I saw the president yesterday' b. dat ik de president gisteren heb gezien that I the president yesterday have seen 'that I saw the president yesterday'

In fact, the same holds for objects that are unambiguously anaphoric, such as zichzelf 'herself/himself'. Because there is no shift in NP interpretation, there can be no constraint on word order either. ·

a.

dat de president gisteren zichzelf op t.v. zag that the president yesterday herself on t.v. saw 'that the president saw herself on t.v. yesterday' b. dat de president zichzelf gisteren op t.v. zag that the president herself yesterday on t.v. saw 'that the president saw herself on t.v. yesterday'

·

Scrambling zichze!f 'herself' in (4o) is truly optional as well.

6 CO NCLUS I O N In this paper we offered an explanation for the difference in scrambling behaviour between defmites and indefinites within an OTS variant of the type-shifting perspective. This variant allows one to distinguish optimal type-logical derivations based on preferred types. The difference in scrambling behaviour is especially striking in the case of definites as objects of light verbs, :;uch as the bus in Two linguists take the bus. Intuitively these definites are very similar to predicative indefinites such as a single in Two linguists get a single. However, we argued that whereas the predicative use of indefinites prefers to live in type ( et), definites have their basic denotation in type e as specified by means of the iota operation. So, in the case of a light verb, with the preferred type ( (et) (et) ) interpretation of incorporating verbs, ' a predicative indefmite object gives one optimal derivation. Defmite objects allow for two optimal derivations when combined with light verbs: either the definite or the verb denotes in its preferred type. In general, we would like to claim that deviation of basic word order yields a shift in meaning. What is considered the basic word order varies among the different types of NPs, and is consistent with the frequency of use of these constructions. ·

·


(4o)


This principle explains the free scrambling behaviour of definites. If there is no change in interpretation, there can be no constraint on word order either, and merging an incorporating verb and a predicative type of definite is synonymous with merging a non-incorporating verb and a referential type of definites. On the other hand, if an indefinite may scramble-e.g. in the context of an ordinary transitive verb in type ( e( et) ) it could change the interpretation of the phrases in which it occurs. This only happens when it has its non-preferred type ((et)t) interpretation as a quantifier. The different optimal derivations for the different combinations of (in)definite NP and verb reflect these judgements formally.

We would like to thank the audiences at De Dag: Workshop on Definites (Groningen, November 1996), the Tabu-dag (Groningen, June 1998), and Storage and Computation (Utrecht, October 1998) for helpful discussion. Two reviewers kindly commented upon an earlier version of this paper. The research for this paper was supported by the Netherlands Organization for Scientific Research. JAAP VAN DER DOES ABN AMRO Bank N.V. DNLS/LISA project {AA 9077) Bankrashof3 1 183 NP Amstelveen The Netherlands e-mail: [email protected]

Received: 18.01.98 Final version received: 1 9. 1 1.98

HELEN DE HOOP Utrecht Institute of Linguistics OTS University of Utrecht Trans 10 3512 JK Utrecht The Netherlands e-mail: [email protected]

APPE N DICE S A DEFINITE NPS AS D OMAIN RESTRICTED I ND EFIN ITE NPS ln the text we have recalled that the singular definite the can be seen as a domain restricted version of the indefinite some. This relation between definites and indefinites holds more generally. The idea can be visualized in an appealing way by means of the tree of numbers. To this end, we first recall the notion of a logical quantifier.


Acknowledgements

Jaap van der Does and Helen de Hoop 41 1

Definition A. I A logical quantifier is a Junctor D which assigns to each non-empty domain E a two piau relation among sets: DE E p(p(E) X p(E)) which satisfies three constraints: •

• •

Conservativity (CONS): DEAB iff DEAA n B; Extension (EXT): DEAB iff DE•AB,for all A , B � E � E'; Isomorphy (ISOM): for all bijections 7r from E onto !?, DEAB iff D E'7r[A]7r[B].

Fact A.3 On finite models, a logical quantifier D can be identified with a relation . natural numbers:

d among

di A - B I I A n B I iff DAB. Hence, quantifiers become subsets in the tree of numbers in Figure IA-B I

1.

n,m = IAnBI (0,0)

=

· · · ( n,m) · · · ..., Bf---- ---- 8!

Figure

I

Tree of Numbers

C£ Van Benthem (1986).

D

Quantificational indefinites Note that in general a quantiftcational proposition bAB depends on (the size of) two sets: A - B and A n B, not just on A n Bl Yet, some quantifiers, the so-called intersectives, only depend on A n B. This insight can be defined in several equivalent ways.

Definition A.4 A quantifier is intersective iff it is conservative and co-conservative: DAB iff DA n BB. is

conservative and symmetric: DAB iff DBA.

iff it


Fact A.z Conservativity and extension is equivalent to universality (UNW): DEAB iff DAAA n B. for UNIV D the first argument truly 'sets the stage'. formally, this means that we may forget about the domain E: DAB iff for some E DEAB; c£ van Benthem (1986) and Westerst:ihl (198 5).

412 Type-shifting and Scrambled Definites iff it is invariant under identical intersections:

A n B = A' n B ' implies: DAB iff DA'B '.

Keenan (1987) proposes to identify simple indefinite quantifiers with the intersectives. They are the basic quantifiers which occur felicitously in existential sentences. (41) There are just two/*most students at the.party When viewed as subsets of the tree of numbers, intersective quantillers are extremely well-behaved; see Figure 2. Indeed, on finite domains intersectives as a

Figure 2 Invariance of indefinites

relation between numbers is fully determined by its element on the righthand spine. They are essentially one- instead of two-dimensional quantifiers, since they can be written as (P)AB, with P � w a set of natural numbers, and: (P)AB iff lA n BIEP. For example, a singleton {n} gives just n. Also, the upward closed sets P with: if k E P and n 2: k, then n E P, are all of the form {n E w : m ::=; n} for some minimal m, which corresponds to at least m. Given the above invariance, it is clear that the following theorem holds: Theorem A. 5 The intersectives are precisely the left-oriented zebra's in the tree of numbers; see 0

Figure J.

Figure 3 Indefinites as left-oriented zebra's

Quantificational (in)deflnites Above we stated that definites are in a sense 'domain restricted' indefinites. Indeed, for P � w the definite can be defined by [PJAB, which holds iff: (P)AB and jA - Bj = o. Alternatively, in terms of a restriction on the domain of quantillcation A: [P]A := {B: A � B " I A I E P}, which gives definites as the principal filters in Barwise & Cooper ( 198 I ). For example,


only if:

Jaap van der Does and Helen de Hoop 4 1 3

Figure 4 Definites as spines of zebra.'s

Theorem A.6 In the tree of numbers, deflnites are just the spines of left-oriented zebra's; see 0

Figure 4·

The case discussed in section 3 is dearly a special instance of the pattern between defmites and indefinites given here. The next appendix is on dependent (in)definites.

B:

A GENERALIZED PARTEE TRIANGLE

Since a definite may depend on any finite number of elements, dependent definites live in several systematically related types. To make this precise, we define an auxiliary notion: Definition B.I Let ao = a,

and

a•+ • =

o:

be a type. The type o:. is defined recursively by: (e, a. ) .

That is, an object of type a. takes n elements of type e to return an object of type a. A dependent defmite lives in a type (e) •• n a natural number. It is a function/". such that for every sequence of objects x of length n,f(X) is the unique object of the kind given by its noun. For example, the dependent definite the bus in (34) denotes a function of type e' = (ee), which applied to an argument x gives the bus assigned to x.• A convenient way to specify this meaning uses a generalized t-operator that shifts the graph of an n-ary function-i.e. an object of type (et). coding a set of n + I -tuples,-to the function in type e. itself t(F) � , . . . , -

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