JOURNAL OF SEMANTICS AN INTERNATIONAL JoURNAL FOR
THE INTERDISCIPUNARY Sroov OF THE
SEMANTICS OF NATURAL LANGUAGE
MANAGING EDITOR . PETER BosCH (mM Scienufic Centre, Heidelberg and Uruversity of Osnabnlck) ASSOCIATE EDITORS. NICHOLAS AsHER (Uruvemty of Texas, Austin) RoB VAN Dmt SAmrr (Uruvemty of NJJmegen) REVIEW EDITOR: AND Uloi!LING (Uruvemty of Stuttgan) ASSISTANT EDITOR AND! UloELING (Uruvemty of Stuttgan) EDITORIAL BOARD· MANFRED BIEIIWISCH (MPG and Humboldt Uruvernty Berlm) BRANIMJR BoGUIL\EV (mMWatson Research Center) Mum BmuLLO (Uruvemty of Toulouse) Kmn! BROWN (Uruvcmty of Essex) GENNAIIO CHIBII.CHIA (Uruvemty of Milan) ANN CoPP.STilll (Stanford Uruvernty 6s-rnN DAHL (Uruvemty of Stockho ) PAUL DFxxa (Uruvemty of !uruterdam) CLAIIlE GAII.DB.NT (Uruvemty of Surbrucken) BAliT GEUim (Uruvemty of Osnabrock) M!CHA£1. Hl!i.WEG (mM Saenufic Centre, Heidelberg) LAUill!NCB R HollN (Yale Uruvemty) JOACHIM JAcoBs (Uruvcmty of Wuppertal) PHIUP N joHNsoN-LAIIlD ( Pnnceton Uruvemty) MBGUMI K.uu!YAMA (SRI lnternanonal Stanford)
L
HANs !CAMP (Uruvemty of Stuttgut) SEBASTIAN UIBNEII (Uruvemty ofDUsseldorf) Sill joHN LYON S (Univemty of Cambndge) jAMES D. McCAWLEY (Uruvemty of Chicago) MAlle MoBNs (Uruvemty of Edmbnrgh) fuN CIS J. Pn1.l.1!111!1l (Uruvemty of Alberta) MANFlll!D Pmv.L (Uruvemty of Saarbnicken) ANToNY S ANPOII.D (Uruvemty of Glasgow) AilrNIM VON STECHOW (Uruvemty ofTubmgen) MAn STEEDMAN (Uruvemty of PenruylvanLa) ANATOU STRJGIN (MD: Planck Gesellschaft, Berlm) HBmmrrm DB SwAliT (Uruvernty of Utrecht) BoNNIE WBBBmt (Uruvemty of Penruylvarua) Hilla Zlmvt.T (Uruvemty of Anuterdam) THoMAS E. ZIMMI!IlMANN (Uruvemty of Stuttgan)
EDITORIAL ADDRESS Journal of Semanncs, c/o Dr P Bosch, ffiM Germany Saennfic Centre, Vangerowstr. 18, D-6<}IIS He!delbe� , Germany Phone- (4g-6m-) S9-42.SL Telefax. (49-62.2.1-) S9-J2.00 Errrnl. pbosch@de 1bm.com
©
Oxford Uruvcmty Press
All nghts reserved. no put of thLS pubhaoon nuy be reproduced, stored m a retneval syttem. or tru:LSmJtted m any form or by any means, eleetroruc, mech.arucal, photocopymg. reeordmg. or otherwue Without etther the pnor wntten pcrmuswn of the Publishers, or a hcence perrrutnng restncted copytng wued 1n the UK by the Copynght Ltcennng Agency Lrd, \)0 Tonenlum Coun Road, London W1P 9HE. or m the USA by the Copynght Cleannce �n�r. lll Rosewood Dnvc, Danvers, �usertl ot�J, USA.
journal of &m.lllla (!SSN o 167 s 133) u pub! !.Shed quarterly m Febnury, May, August and November by Oxford Umvc:mty Press, Oxford, UK. Annual subscnption LS US$140 per year. ]0>1rMl ofSmttuwa LS dLStnbured by MAIL Arneru:a, �J.l3 Randolph Avenue, Avenel, New Jersey 07001, USA. PenodLC11 postage pa>d at R.thway, New Jersey, USA and at additional entry pomu. US POSfMASIER: send address correctioru
Avc:ncl, New Jersey 07001, USA.
to jowrMI
of &manila, c/o MAIL AJtu,rica, �323 Randolph Avenue,
JOURNAL OF SEMANTICS Volume IS Number
I
CONTENTS Special issue on Underspecification and Interpretation Guest Editors: Reinhard Blutner and Rob van der Sandt REINHARD BLUTNER AND RoB VAN DER SANDT Editorial Preface KEES VAN DEEMTBR
s
Ambiguity and Idiosyncratic Interpretation MARcus EGG Wh-questions NICHOLAS Bridging
in Underspecified Minimal Recursion Semantics
AsHER AND
ALEx
37
LASCARIDES 83
(to be continued in Volume IS Number 2)
The contents page published in the journal of &mantus, Volume IS Number I: February 1998 w as in error. We apologise for this mistake and request that you please insert this sheet instead of the faulted page. Please visit the journal's world wide web site at http://www.oup.eo.uk/sema.nt
JOURNAL OF SEMANTICS Volume 15 Number 3
CONTENTS STEVEN FiussoN, DoMINIEK SANDRA, FRANK BRISARD, GBRT VAN RILLAER AND HUBERT CUYCKENS Flexible Semantic Processing of Spatial Preposinons IOANNIS VBLOUDIS 'Quantifying' Superlatives and
Homo Sap1ens
NICHOLAS AsHER AND ALEx LAsCARIDBS The Semantics and Pragmatics of Presupposition
215
239
Ple2Se visit the journal's world wide web site at http://www.oup.eo.uk/semant
Subscriptions: The Journal of Semantics is published quarterly. lnstituuonal: UK and Europe £-,&.oo; USA :md Rest of World US$140.00. (Single issues:
UK :md Europe £23-00; USA and Rest of World U$$4-o.oo.) Pmonal:* UK and Europe £J8.oo; USA and Rest of World US$71.00. (Single issue: UK and Europe £u.oo; USA and Rest of World US$21.00.) * Pmorsal ra� apply only whm copit5 art m�t to a pnJJa� aJdrw and payrnntt is madt by
pmonal chtqudcrtdit card.
Prices mclude postage by surface mail or, for subscribers in the USA and Canada by Airfreight or in Japan. Awtralia, New Zealand and India by Air Speeded Post. Airmail rates are available on request. Back bsues. The current plus two back volumes are available &om the Oxford University Press, Great Clarendon Street, Oxford OX2 6DP. Previous volumes can be obtained &om Dawsons Back Issues, Cannon House, Park Fann Road. Folkestone, Kent CT19 sEE. tel +44 (o)1303 85o1oi, fax +44 (o)1303 850440. Volumts 1-6 are available from Swets and Zeitlinger, PO Box 830, 216o SZ Lisse, The Netherl:mds. Paymttrt is required with all orders :md subscriptions are accepted and entered by the volume. Payment may be made by cheque or Eurocheque (made payable to Oxford Uruversity 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 Subscnptions Department, Oxford Uruversity Press, Great Clarendon Street, Oxford OX2 6DP, UK. Telex 8373 30 OXPRES, tel +44 (o)1865 267907, fax +44 (o)1865 26748 5 .
Scope of this Journal The JOURNAL OF SEMANTICS publiShes anicles, notes, discussions, and book reviews m the area of natural language semantics. It is explrcttly mterdisctplinary, in that 1t aims at an integration of phtlosoprucal, psychological, and linguiStic semantics as
well as semannc work done m amficial mtellrgence and anthropology. Contributions must be of good quality (to be JUdged by at least two referees) and should relate to quesnons of comprehensiOn and interpretanon of sentences or texts in natural language. The editors welcome not only papers that cross traditional disciplme
boundanes, but also more speaalrzed contributions, provided they are accessible to and interesting for a wider readership. Empirical relevance and formal correctness are paramount among the cntena of accepunce for publication.
Information for Authors: Papers for publrcanon should be submitted in 3 cop1es to
the managing editor. They should be typed on A4 (or sunilar format), one-s1ded, double-spaced, and with a w1de rnargm and must be accomparued by an approx. 200 word summary. Notes and biblrographical references must appear at the end of the typescnpt. All bibliographical references m the text by author's surname and year of
publrcation. Diagrams must be submitted camera-ready. The submissiOn should be accompanied by a Postscnpt fUe of the paper on diSkette. Final versiOns after acceptance of a paper for publication must be accompanied by a file
m
source code.
All papers submitted are subject to anonymous refereeing and are considered for publrcation only on the assumption that the paper has neither as a whole or m part already been published elsewhere nor has elsewhere been subrrutted or accepted for publication. Unless speaal arrangements have been made, copynght rests With Oxford Umvemty Press. Authors receive 20 offpnnts of their publiShed anicles and 10 offpnnts of their publiShed reviews, free of charge. Larger numbers can be supplred at cost pnce by advance arrangement. It is a condition of publication in the Journal that authors assigD copyright to Oxford Univemty Press. This ensures that requests from third parries to
Copyright:
reproduce articles are handled efficiently and consistently and will also allow the article to be as widely disserrunated as possible. In assigning copyright, authors may use thm own material m other publrcations provided that the Journal IS acknowledged as
the ongmal place of pubhcation, and Oxford University Press m advance.
Advertising:
IS
notified
m
wntmg and
AdvertiSements are welcome and rates will be quoted on request.
Enqmries should be addressed to Helen Pearson, Oxford Journals Advertising, PO Box 347, Abmgdon SO, OX14 sXX. UK. Tel/fax: +44 (o)1235 201904· Email: oxfordad
[email protected].
(0 Oxford Umvenity Press
1998
Flexible Semantic Processing of Spatial Prepositions STEVEN FRISSON, DOMINIEK SANDRA, FRANK BRISARD, GERT VAN RILLAER AND HUBERT CUYCKENS
University of An�
Abstract users spontaneously apply semantic extenston pnnciples to novel usages, or whether they treat word meanings as discrete, ngidly defined entities. In Experiment 1, readers made a timed deci51on on the correctness of a sentence. Rejecting a cognitively plausible yet unattested extension of a preposition (*Pmr IS standmg by tk county) took longer and lead to more errors than rejectmg a pLamly incorrect usage of that preposition (*Pmr is standing by tk subtJtk� ThU result was obtained for two novel extensions of two different preposinoru. Experiment 2 mcluded a preceding context in order to fix the referent of the target word and to exclude singular mterpretanoru. The results were consutent With those of Experiment 1, although the overall number of errors dropped considerably. Taken together, these experiments indicate that, even in a wk dJSCOuragtng the use of flexible processing, subjects are not able to suppress the apphcanon of extension principles. We conclude that interpreting the mearungs of words in a flexible way is an mherent property of a semantic processing system.
This article presents two experiments investiganng whether language
INTRODUCTION
Word meanings have rwo characteristics. First of all, they have an element of conVffl tionality, such that they behave as relatively stable linguistic units. This conventional aspect makes it possible for children and foreign language learners to learn how to use words in the way adult native speakers of the language do, thereby conforming to the norms that members of the language community have implicitly agreed upon. For the same reason, lexicographers can describe those meanings in their dictionaries. In short, the notion of lexical meaning entails the existence of conventional usage norms, which one has to respect in order to engage in successful communication. At the same time, however, word meanings are characterized by a degree of jkxibility. This property makes it possible to extend the range of applicability of words to situations in which they are not conventionally used. This potential of meaning extension is the driving force behind the historical evolution of a word's meaning, ie. the process by which novel
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
1
192
Flexible Semantic Processing of Spattal Preposinons
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
usage possibilities are added to the usage potential of a word. For instance, the word wing was certainly not used in expressions like t� wing of an airpla� before the emergence of flying techniques, but the original meanmg of the word made it possible to extend its usage to such a new context. Examples like this can be found in abundance in the diachronic evolution of many words in any language. This readiness of language users to extend word meanings also explains the high frequency of different forms of figurative language, like metaphor and metonymy (which, incidentally, are often the processes behind diachronic semantic evolution, as in the example above), irony, etc. This also manifests itself in the realm of grammatical or so-called function words, in so far as the same principle is responsible for the emergence of new, related usage types. Prepositions, the class of grammatical items we will focus upon in the present paper, get extended semantically by virtue of unusual combinations with other lingmstic items, in particular with the types of objects that follow them. Accordingly, wing acquires another meaning either through combination with an explicit linguistic item belonging to the semantic domain of, say, 'airplanes', or at least through the realization that an intended referent of which wing is attributed as a part is not, in any literal sense, a bird. Of course, for prepositions the combination with unusual referents and/or linguistic expressions will tend to have a bearing on the whole of a proposition, since the function of such grammatical items is exactly to establish a coherent scene in which all participants are given a specific role. The fact that meaning extension is so pervasive in language indicates that language usm are very good at it Speakers apparently have no problems using words in a novel way Oanguage production) and listeners do not seem to encounter problems of interpretation Oanguage comprehension). This is an interesting observation from a psycholinguistic point of view. It suggests that the flexibility of word meanings actually reflects a property of the language user's processing system. That system must be designed in such a way that adapting word usages to the communicative needs of the moment is a very natural and effortless process. In the present paper, we will investigate one aspect of this 'naturalness'. We will test the hypothesis that, despite the highly conventionalized nature of lexical meaning, language users are unable to deal with word meanings as conventional linguistic entities only. Rather than treating their stored semantic knowledge in a rigid, inflexible way, they will spontaneously and automatically extend it to novel usage contexts. In order to test this hypothesis, we designed an experimental technique which explicitly invited language users to reject non-existing but possible semantic extensions of word usages. Indeed, the best way to demonstrate the automatic nature of the process of semantic extension is to show that it cannot be brought under
Steven
Frisson,
Dominiek Sandra, Frank Brisard, Gert V:m Rillacr, Hubert Cuych:ns
193
2 EXPERIMENT 1 The hypothesis that language users automatically extend word usages beyond the set of existing ones was tested in a speeded decision task with the Dutch spatial prepositions aan 'by, near, at' and naast 'next to'. Dutch-speaking subjects were asked to decide whether they thought sentences containing these prepositions could be used in Dutch or not.
2.1 Method 2. I. 1 Materials and design For the purpose of this experiment, only those instances of aan and naas t were used that express the spatial relational component 'proximity'. This semantic overlap enabled us to test similar extensions in two different prepositions. The results for aan and naas t will be presented separately. The use of aan and naas t expressing a spatial proximity relation imposes restrictions on the size of their landmarks (LM) relative to their trajectors (TR).' This allows us to create extensions in which the difference in size of LM with respect to TR is manipulated. Extension principles, such as metaphorical and metonymic mappings, end-point focus, etc., are frequently based on some manipulation of an original or prototypical semantic configuration. in which, at least in the case of prepositions, the relation between TR and LM is considered crucial Out of all the possible
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
subjects' conscious control, even in a situation where attending to extensions is irrelevant for the task at hand and actually interferes with task performance. The basic technique was to present sentences in which a word-always a preposition-was used incorrectly, but where this usage was a cognitively plausible (though unattested) extension of the actual word meaning. Subjects were expected to classify such sentences as incorrect sentences in the language. If the process of semantic extension is indeed automatic, subjects will be unable to inhibit processing of the extension and, accordingly, will find it relatively difficult to reject these sentences. This will be reflected in longer reaction times and possibly higher error rates relative to the data for plainly incorrect sentences. The technique used here is sitrular to the one that Glucksberg and his colleagues have successfully used in their research on metaphor processing (e.g. Glucksberg, Gildea & Bookin 1982).
194
Flexible Semantic Processing of Spatial Preposttiom
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
criteria that can be changed (properties of TR, of LM. or of the type of relationship linking the two), only one, i.e. the relative size of LM, was selected for this particular experiment, in order to make sure that no interference from other semantic variables was measured. If flexibility in the processing of lexical-semantic configurations is as pervasive as the experimental hypothesis purports it to be, it should manifest itself when the manipulation of this criterion leads to incorrect, yet quite conceivable, usages of a preposition. In a relation of spatial proximity as lexicalized by aan or naast, LM is usually larger than TR (e.g. Ptter [=TR] stond aan het bruggetje [=IM], 'Peter was standing by the bridge), but LM can also be allowed to be somewhat smaller than TR (e.g. Hij stond naast htt barluukje, 'He was standing next to the bar stool). The important point is that the difference in size between LM and TR should not exceed a certain limit. This limit is not absolute but is to be defined in terms of LM's relative size with respect to TR A small TR, for instance, puts a much stricter limit on the size of the LM it is spatially related to than a large TR does. Consider, in this respect, Marokko ligt naast de Salutra 'Morocco is located next to the Sahara Desert' vs. De speld ltgt naast htt huis The pin is lying next to the house'. The former sentence is more acceptable than the latter, even though the landmark in the former sentence is much larger than that in the latter. The restriction on TR's and LM's relative size has (cognitive) advantages for the hearer, in that it allows her to identify TR's precise location with respect to LM In this sense, it is true that the magnitudinal constraint present in these prepositions' m eanings constitutes an instantiation of a more general, functionally inspired criterion, which states that reference points (i.e. LMs) need to be salient enough-where salience is a relative, not an absolute notion-to serve as the reference point or anchor of a spatial identification process (see also Vandeloise 1991 ). If LM is too large with respect to TR (e.g. Gtrt bMndt zich aan htt graafsclutp, 'Gert is located by/near the county}, TR's location cannot be sufficiently identified, simply because the perimeter of the landmark in question is too extensive to warrant any precise localization of the trajector, i.e. the hearer has too many options from which to select the actual location of the target at hand. Conversely, if LM is too small with respect to TR (e.g. Dt stoel staat aan dt spdd, The chair is standing by/near the pin'), LM is not salient enough within the scene described, which also impedes easy localization ofTR It is exactly this sort of manipulation of LM with respect to TR that we have performed in order to create the extensions used in the following experiment. It is true that the normative size of TR and LM is due to pragmatic requirements, but this does not invalidate its being seen as part of the
Steven frisson, Dominie� Sandra, Frank Brisard, Gert Van Rillaer, Hubert Cuycken.s 195
semantic characterization of the prepositions concerned. For one thing, similar restrictions on size do not seem to hold for that subset of locative prepositions which defines a relatively specific spatial relation between TR and LM, such as 'support' (for on) or 'inclusion' (for in), such that the principle referred to above is by no means a general one that can be indiscrim.inately applied to all instances of localization through the use of spatial prepositions., The materials consisted of three categories of word usages:3
Examples of sentences for each of the three categories and the two prepositions are given in Table 1. The stimuli used in each of the three categories were attested by means of a pre-test, resulting in a categorization not dependent on experimenters' intuitions. In this pre-test, four different sets of So sentences expressing an aan/naast-relation between TR and LM were presented to 100 subjects (all of them first-year business undergraduates), each randomized set being presented to 25 subjects.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(i) Corr�d usages (COR}: the preposition's usage falls within the range of existing usages, i.e. it indicates an acceptable and lexicalized spatial relation between TR and LM (e.g. Het kind bevond zich aan d� roltrap, 'The child was located at the escalator'). The size of LM relative to TR is neither too small nor too large, i.e. it does not exceed an implicitly agreed upon and conventionalized limit. The required response to sentences containing these usages is yes. (ii) Plainly incorrect usages (INC}: the preposition's usage falls outside the range of existing usages, in that it indicates an unacceptable spatial relation, either because the relation is impossible in the physical world (e.g. H�i bevindt zich naast tk hoofdletter, 'Heidi is located next to the capital letter') or because it indicates a spatial relation with an abstract NP (e.g. De man staat naast het verbruik, 'The man is standing next to the consumption'). The required response to sentences containing these usages is no. (iit} Exwuied usages (EXT): the preposition's usage falls outside the range of established usages, yet it can be perceived as a possible usage, i.e. it indicates a spatial relation between TR and LM which is physically possible, but which happens not to be lexicalized in the language (e.g. Gm bmndt zich aan het graafschap, 'Gert is located by the county'; Els bevindt zich naast het zakmes, 'Els is located next to the pocketknife'). Two subcategories are distinguished, one in which LM is too small (EXT1) and another in which it is too large (EXT2) with respect to an implicitly agreed on limit. The required response type to sentences containing extended usages is still no.
I
96 Flexible Semantic Processing of Spatial Preposltloru
Table
I
Example sentences for the different condittons used in Expenment
Category
Example
COR
Htt
I
fond btvond zuh aan dt roltrap 'The cluld w.u located at the escalator'
Anntmit staat naast dt bromfltts
EXT EXT!
'Annemie is standing next to the motorbike'
110
Ptttr sto nd aan
110
htt scllttmlts
EXT2
Anntmlt bmndt zuh naast dt oostltust
'Annemie
u
located next to the
HI} staat aan dt ondtrt�UI
INC
east
coast'
'He is standing by the subtitle' wrbruik 'The man IS standmg next to the consumption'
Dt man staat naast htt
To make the task as natural as possible, subjects were informed that some of the sentences were produced by learners of Dutch as a foreign language while others were deliberate mistakes made up by the experimenters themselves. Since learners of a foreign language often produce mistaken though plausible word usages and language users recognize such errors, the subjects were offered a natural criterion to discriminate between possible extensions and impossible ones. (The fact that learners of Dutch would not actually produce such sentences because the relations expressed are not sanctioned in any language is rather immaterial to the argument, since our perception of non-native speakers in the course of learning a language is such that we expect them to make plausible mistakes that just 'sound funny'. It is this feeling of a plausible strangeness which we wanted to create so that subjects would provide the type of reaction needed for the test, without prompting them to inquire into the actual possibility of such mistakes occurring.) Subjects were then asked to indicate, for each sentence, whether the sentence was
correct
(i.e. 'whether a native speaker of Dutch
would use that sentence') or whether it was incorrect. For each sentence classified as incorrect, subjects had to state (i) whether they thought this
sentence could have been produced by a learner of Dutch as a foreign language, i.e. whether they thought it liluly that the learner of Dutch could
this particular sentence, or (ii) whether they thought this unlikely, i.e. whether they suspected that the sentence had been made up
have come up with
by the experimenters.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'Peter was standmg by the razor' Hans btvond zich naast dt sptld 'Hans was located next to the pin' Gtrt btvindt zuh aan htt graafichap 'Gen is located by the county'
Steven
Fri.sson,
Dominiek Sandra, Frank Brisard, Gert Van Rillaer, Hubert Cuyckens
197
The stimuli used in the experiment were selected from the sentences in the pre-test on the following basis: (i) sentences predominantly classified as correct in the pretest made up the COR-category in the experiment (So sentences, 40 for aan and 40 for naast); (ii) sentences predominantly classified as incorrect and as unlikely mistakes made up the INC-category in the experiment (40 sentences, 20 for aan and 20 for naas t); (iii) sentences predominantly classified as incorrect but as likely mistakes made up the EXT-category in the experiment (40 sentences, 20 for aan and 20 for naas t, equally divided between EXTI and EXT2).
categories had to be matched on word length and frequency. As far as word length is concerned, the matching had to involve the category of COR items as well, in order to make sure that subjects could not use word length as a cue for making the distinction between positive and negative response types. The means are shown in Table 2. A one-way ANOVA, in which category (COR, EXT,. INC) was treated as the independent variable and word length as the dependent one, was non-significant (F < I). A separate
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The sentences in the INC and EXT categories were clearly perceived as incorrect, both for aan (mean number of mcorrect-responses: INC: I94 (out of 20), EXT: 17.6) and for naas t (INC: I9.4, EXT: I6.J). The pattern observed for the EXT items pertained to each extension type (for aan: EXTI = 94 (out of 10), EXT2 = 8.2; for naas t: EXTI = 7.8, EXT2 = S.s ). For both prepositions, the number of [,July-responses differed significantly between the INC and EXT categories, as assessed in an analysis of variance (ANOVA), where category was treated as the independent variable and the number of likely-responses as the dependent one: F[I, 24) = I40.88, p < o.ooi for aan (INC: 3.6, EXT: I 1.2) and F[I, 24) = I4J·Is. p < o.ooi for naast (INC: 2.7, EXT: I 1.3). Items in the INC category were pre dominantly considered to be unlikely usages, whereas those in the EXT category were predominantly considered to be likely ones. For both prepositions, this contrast was also significant when the two extensions were analysed separately (aan: INC vs. EXTr: F[I, 24] = 30.42, p < o.oi, INC vs. EXT2: F[ 1, 24] = 2J.I7, p < o.oooi; naast: INC vs. EXTI: F[I, 24] = I0.2S, p < O.OI, INC vs. EXT2: F[I, 24] = 64-00, p
198 Flexible Semannc Procc:sstng of Spari21 Prepositions
ANOVA. in which the two extension types were treated as separate levels (COR, EXTr, EXT2, INC}, also fell short of significance (F < r). The INC and EXT categories were also matched for frequency. A one-way ANOVA in which category was treated as the independent variable and frequency as the dependent variable was non-significant for both preposi tions:aan:F[r, 38] = 2.22, p > o.ro, naast:F[r, 38] = 1.54, p > o.Io. When the two extension types were treated as separate levels in the ANOVA (EXT r,
EXT2, INC}, the result remained non-significant: F[2, 37] = 2.67, p > o.os for aan, F < 1 for naast. Items in the COR category were drawn from the
same frequency region as those in the EXT and INC categories. All sentences in the experiment (i) showed the same grammatical
hutsdtur, 'Luk is standing by the door';johan stand naast de bureaustoel johan was standing next to the desk chair'). To avoid having to establish a limit on TR's and LM's difference in size for each TR-LM configuration, the TR and the verb used in the sentences were kept as constant as possible. This means that (1) TRs all referred to entities of the same size, i.e. persons (tk student 'the student', d� jongen 'the boy', personal pronouns, proper nouns), and (ii) the verb expressed a static spatial relation Qexicalized by staan 'to stand' or zich bevinden 'to be located', both used in the simple present and simple past the same number of times over the various conditions). To force subjects to read the whole sentence, they had to answer yes/no-questions about their contents at regular intervals. The experimental list was divided over
8
blocks of
20
sentences. An
additional block, containing 20 filler items, was added at the beginning of the experiment to familiarize subjects with the experimental procedure; its results were not analysed. In each block, there was an equal number of
aan-
and naast-sentences. Each block consisted of
incorrect ones, and s extensions, i.e. there were
10
correct sentences, s
10 yes and 10
no
Table � Mean word length (WL) and frequency (Freq) for each condition AAN
NAAST
Category
WL
Freq
WL
Freq
COR EXT EXT I EXT2 INC
8.2 7·9 8.] 7·6 8 .]
77 95 8] 107 So
7·8 7·7 7·6 7·8 84
95 99 97 100 87
responses.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
structure: NP-V-PP, and (ii) expressed the spatial relation 'proximity' lexicalized by aan or naast-between TR and LM (e.g. Luk staat aan de
Steven Frisson, Dominiek Sandra, Frank Brisard, Gen Van Rillaer, Hubert Cuyckens
199
The experimental blocks were presented in two different orders. In each version, the order of the blocks was fixed, but within each block, the order of the sentences was randomized. Between blocks, there was a pause of maximum ro seconds. Subjects pressed a key
if they
wished to continue
with the experiment before the end of the pause or the computer proceeded with the program when the 10 seconds had elapsed.
2.1.2 Procedure Before the experiment started, subjects were instructed-both orally and in writing-about the relevant aspects of the experimental procedure. In
they thought native speakers could use the sentence in Dutch or not. If they thought the sentence was possible in Dutch, they were invited to strike the� key; if they thought the sentence was not possible, they had to strike the no key. They used their preferred hand for� responses. Subjects were told to
decide as quickly as possible while making as few mistakes as possible. Finally, they were informed that, at regular intervals, a question about the previous sentence would be asked, which they had to answer by using the ylS or no key.4 During the experiment, subjects were sitting in front of a computer screen in a darkened room. Before the first block of sentences was presented, a summary of the instructions appeared on the monitor. For each sentence, subjects had to proceed as follows: after seeing the first part of the sentence (presented in lowercase letters) up and until the preposition tUtn
or
naast, they had to push a key in between the two response keys on the
button box to make this part of the sentence disappear. Mter an interval of
250 ms, a fixation point (+) appeared in the middle of the screen for 250 ms, which was in turn followed (again after an interval of 250 ms) by an NP (presented in uppercase). Subjects had to strike the � or no key, depending on whether they thought the whole sentence could be used in Dutch or not. Reaction times were measured from the onset of the NP up to the subject's response. Mter making their decision, either the message 'Next sentence' appeared on the screen, in which case subjects could strike the }'6 or the no key, or a question appeared, which they had to answer with the )'fi or no key. The next trial was then initiated.
2.1 .3 Subjects Thirty first-year undergraduate language students volunteered for this experiment (each order of blocks was presented to 15 subjects). Their native language was Dutch. None of the subjects had participated
in the pre-tests.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
particular, they were asked to decide, for each of a number of sentences, whether the sentence could be considered as correct Dutch, that is whether
200
flexible Semantic Processing of Spatial Prepositions
2.2 Results The results will be presented separately for aan and naast. Since only the results for the INC and the EXT categories are relevant to our hypothesis, the data for the COR category will not be discussed.
2.2.1 AAN
Table 3 Mean reaction times (R'Ij, standard devtatlom {SD) and error percentages for a:ms
Category
RT
SD
Errors
COR EXT EXT I EXT� INC
Il S 3 1729 1479 1886 126o
286 468 3 77 S l7 25 0
7 26 21 30
(nu)
(%)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Table 3 presents the mean RTs and error percentages for the different categories of aan. ANOVfu; were performed on both the RT data and the error data. Reaction ti� analyses. Response times above 4000 ms were excluded from the analysis (s%). In the analysis of the RT data, responses which did not correspond to the expected response type were also excluded (13% ). A one-way ANOVA {EXT vs. INC} was performed on the mean subject and item reaction times. The higher RT that was found in the EXT category (difference: 469 ms) differed significantly from the RT in the INC category {Fs[1,29] = 34-27, p < o.ooi; F,[ 1, 38] = 49.10, p < 0.001). When the two extension types were kept separate in the analysis, the outcome of the ANOVA remained significant {INC vs. EXT 1 vs. EXT2; Fs(2, 52] = 28.83, p < o.oo1; F,[2, 37) = 56.46, p < o.oo1).6 In order to study the effect of the two extensions separately, pair-wise comparisons of the means were performed. These were all significant: EXT1 vs. INC: Fs(1,52] = 9.61, p <.01; F,[1,37] = 19.62, p <0.000 1; EXT2 vs. INC: Fs(1,52] = 57.04, p < 0.001; F,[1,J7] = 112.01, p < 0.001; EXT! VS. EXT2: Fs[1, 52] = 19.83, p < o.oo1; F,(I,37) = 28.41, p < o.ooi. These figures reflect two fmdings: (i) responses to items in both EXT categories were significantly longer than those to INC items and (ii) EXT2 caused significantly longer RTs than EXTI. Error analyses. A one-way ANOVA (EXT vs. INC) was performed on the error rates of subjects and items. Significantly more errors (difference: 2 5% )
Steven Frisson, Dominiek Sandra, Frank Brisard, Gert Van Rillaer, Hubert Cuyckeru 201
2.2.2
NAAST
Table 4 presents the mean reaction times, standard deviations and error percentages for the different categories of naast. Reaction timt analyses. As in the case of aan, only RTs for correct responses and below 4000 ms were analysed. Twenty per cent of the responses in the EXT and INC categories were errors, and an additional 7% was removed because the RT exceeded the time limit. A one-way ANOVA (EXT vs. INC) was performed on the mean subject and item reaction times. The higher RT that was found in the EXT category differed significantly from the RT in the INC category (difference: 622 ms):Fs[I,28] = 76.68, p
RT
SD
Errors
COR EXT EXT I EXTz INC
IISS 1904 1 7 24 2024 1282
3 10 43 1 349 538 324
4 38 39 36
(rn.s)
(%)
2.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
were made on EXT items than on INC items {Fs[I,29] = 26.6o, p
202
Flexible Semantic Processing of Spatial Preposinons
2. 3
Discussion
The major outcome of this experiment is that language users find it difficult to reject word usages which do not occur in the language but are nevertheless perceived as possible. In a speeded decision task (where subjects were confronted with the question 'Is this a possible Dutch sentence?'), both decision latencies and error rates were significantly higher for items of this category than for overtly incorrect usages. The effects were very robust and were obtained for two similar but distinct Dutch prepositions. This difference between EXT items and INC items reveals a decision uncertainty for the 'extensions' used in the experiment. Moreover, this effect was obtained in a speeded-decision task, which means that the conflict between what is sanctioned by the language and the novelty of the extensions occurs very fast. This process seems to be outside the domain of subjects' conscious controL If subjects had been able to inhibit the process of flexible semantic processing, they would have done so, as this factor considerably interfered with task performance. For both prepositions in the experiment, the extension indicating large landmarks turned out to be more difficult than the other (as evidenced by a larger effect on decision latencies and/or error rates), which reflects the fact that the extension in question is perceived as more plausible. In other
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Fs(I,46) = 27.64,p
Steven
Fnsson,
Dominiek Sandra, Frank Brisard, Gert Van Rillaer, Hubert Cuyckeru
203
words, when the prepositions were followed by a noun referring to a large entity (e.g., 'He was standing by the Sahara', 'He was standing next to the prairie'), longer decision times were obtained than when the prepositional object referred to a small entity (e.g., 'He was standing by the peach', 'He was standing next to the bracelet'). Apparently, in the configuration where a human trajector is positioned relative to a landmark, more variation is allowed towards the upper end of the landmark's magnitudinal scale than
the target NP would be expected to lead to the acceptance of the sentence and hence to a response error. Moreover, the effects of construing an alternative scenario, in which typically small objects would somehow be inflated and geographic regions would be projected onto some more spatially manageable mode of representation, would also have shown in the reported pre-test. This was clearly not the case: in an off-line task, where subjects had ample opportunity to construe any number of possible scenarios, the majority of subjects did classify the sentences of both EXT categories as incorrect, even though the above hypothesis would predict a high number of corrtct-responses. Obviously, since these
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
towards its lower end. It could be argued that subjects might try to reinterpret the referent of the noun in order to make it fit the aan or naast scheme. If, for example, subjects are confronted with a sentence like John is standing next to the bottle of wine', they might interpret the landmark as a giant model of a bottle, say outside a liquor store, or they might adopt any other strategy which would help them in making sense of the utterance (think of the context of a game manual, where a player has to move (small or inflated) pieces on a games' board). The higher error rate (i.e. many yes responses) might then be due to the fact that subjects succeed in making the referent fit the described scene and thus consider the sentence to be correct. The same line of reasoning might hold, it should be said, for the EXT category featuring large landmarks as well, as, to offer but one scenario, geographic regions could easily be projected onto maps or other types of representa tions next to which physical location is quite conceivable. The point is that the longer reaction times for the EXT category could be attributed to the processing time taken up in the attempt to reinterpret the target noun, whichever route might be taken in this process of reinterpretation. The high error percentages that were observed would also follow from the adoption of such an interpretation strategy. If this alternative hypothesis is true, however, it would be difficult to explain the significant difference between the EXT and INC categories in the reaction time analysis. This analysis only involved the data for correct responses (i.e., rejections of the EXT items), which could not be affected by the reinterpretation strategy described above. Indeed, a reinterpretation of
204
Aexible Semantic Processing of Spatial Prepostrioru
theoretical arguments do not completely rule out the possibility of subjects reinterpreting, through some alternative strategy, the scene presented to them by the stimuli (in order to either reject or accept the utterance), a second experiment is needed in which the referent of the prepositional object is fixed in such a way that it explicitly prevents subjects from making up their own idiosyncratic interpretations. If the same results are found in the second experiment, in which the alternative hypothesis presented here cannot be held to account for the data, we may conclude that flexible semantic processing is responsible for them. One aspect of the data needs more comment. The error percentage for the extension categories of each preposition in the Hrst experiment ranges
39%). This
means that subjects
decided, in a considerable number of cases, that the presented word usages
are actually sanctioned by the language. One should not conclude from this
that the items in the extension conditions were inadequately selected (i.e. that many of them did not really belong in a condition where a no response was expected, so that high error rates would be the result of unsuccessful stimulus selection). On the pre-test, items of the EXT category were classified as incorrect Dutch sentences in the large majority of cases. The differences between judgements from an on-line task, where relatively many category mistakes are made, and those from an off-line task, where subjects' judgements conform to the expected responses for Dutch prepositional usage, can more readily be related to the difference between speeded and non-speeded responding. Speeded responding will not cause particular problems for items in the COR and INC categories. Indeed, those items clearly conform to or deviate from the conventionalized usage norms in the language and, for that reason, can easily be accepted or rejected. However, items in the EXT category will be more difficult to reject (or to accept, for that matter). As they are not completely implausible, subjects can come up with an interpretation (e.g. a person can physically stand next to a needle).
This
interpretability
will
create a bias towards a yes response.
However, at the same time, subjects will notice that the semantic representation of the sentence does not completely match the stored semantic information for the preposition. As this mismatch might lead them to reject the sentence, they will proceed with a more detailed check on the acceptability of the sentence. If the decision conflict is not resolved by the time the subject's imposed response deadline is exceeded, she is likely to be influenced by the response bias created earlier, i.e., to provide a � response in the absence of clear negative evidence. In other words, the high error percentage for items of the
EXT
category follows from the same
decision conflict that has been proposed to account for the longer response times in that category (for correct responses). It is obvious that fewer errors
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
from high to very high (varying from 21 to
Steven Fruson, Dominiek Sandra, funk Brisard, Gen Van Rillaer, Hubert Cuyckens 205 will be observed in an off-line task, which gives subjects ample time to perform the check on semantic acceptability. The contrast between the error percentages in the pre-test and in the experiment can thus be explained in terms of the difference in response pressure for both tasks. Note, incidentally, that dus account of the high error percentage for items in the reaction time task implies that semantic extensions are processed very rapidly and that the detection of their error status necessarily
EXT
follows
this processing
stage. This underscores the naturalness of flexibility
in lexical-semantic processing. Apparently, errors are not detected before processing the extension but after such processing has been completed. As a matter of fact, if errors were detected straight away, the effects for the EXT A second experiment was run for two reasons. First, we wanted to demonstrate unambiguously that the effects we observed in Experiment I were due to the flexible processing of the meaning of the preposition rather than the meaning of the noun. Second, we wanted to minimize the error percentages for the
EXT category.
In order to achieve these goals, context
sentences were used to determine the referent of the noun designating the landmark in a non-ambiguous way. Thus, scenarios were excluded in which the landmark would refer to atypical instances of its category (in terms ofsize).
3 E X PE R I M E N T
2
3.1 Method 3.1.1 Materials and design In the second experiment, only the
naast
stimuli were used. The intro
duction of preceding context sentences made the experimental task (the
same as in Experiment I ) quite long, and using both aan and
naast
might
have resulted in a tedious experiment leading to serious concentration problems. Furthermore, Experiment I ,
this
if the experiment yields the same results for naast as
is sufficient to reject the alternative explanation of the
results suggested in the discussion above. The preceding context consisted of a single sentence. This was followed by a matrix sentence of the same type as in Experiment I ('X is/was standing next to Y'). In the
matrix
sentence, the noun designating the
landmark was replaced with three dots:
Dt btwOnn- ���tn tku woon wagm l1tt zijn slaapp•l ���tlkn tom hij «n luhu duiztlmg votldt. Jan stond naast d� . . 'The occupant of the tniler dropped his sleeping pill when he felt a slight dizzy spell. John was standing next to this . . . .
'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
category would not tum up in the data at all.
206
Flex�ble Semantic Processing of Spatial Prepostnoru
the target noun (LM) is already introduced in the context sentence and is referred to a second time through the use of a demonstrative in the matrix clause, its referent is kept constant and any possibility of assigning a non canorucal referent to the landmark will be maximally reduced. The stimuli were selected on the basis of the same pre-test as in Experiment I in order to test their acceptability intersubjectively. This time, the matrix sentences were shown together with their corresponding context sentences. Subjects were told that students learning Dutch as a foreign language were presented with the context sentences and had to produce sentences of the kind 'X VERB next to this Y'. They were also told that some of the matrix sentences were not produced by learners of Dutch but were constructed by the experimenters, sometimes containing deliber ate errors. Subjects were then asked to indicate, for each sentence, whether the sentence was correct (i.e. 'whether a native speaker of Dutch would use that sentence') or whether it was mcorrect. For each sentence classified as incorrect, subjects had to state (i) whether they thought it likely that this sentence had been produced by a learner of Dutch as a foreign language, or (ii) whether they thought this unlikely, i.e. whether they suspected that the sentence had been made up by the experimenters. The stimuli used in the experiment were selected from the sentences in the pre-test on the following basis: As
Forty-five first-year undergraduate language students volunteered for this pre-test. They were divided into three groups, so that each item was presented to I 5 subjects. There were 30 sentences in each category. The EXT category comprised I 2 items where small landmarks were used (EXT1) and I 8 items where large landmarks were used {EXT2). Subjects considered both extension types and the corresponding sets of incorrect items, matched for length and frequency (henceforth: INCI and INC2), as incorrect (INCI: I 2.o (out of 1 2), EXTI : 7.9; INC2: I 7.8 (out of I 8), EXT2: I 7. 1). Furthermore, the difference between INC and EXT in the number of li/uly-responses was significant for EXTI (INCI : 3 I%, EXTI: 64%, x" = 3 1.57, p < o.oooi) and for EXT2 (INC2: 24%, EXT2: 63%, F[ I , I4] = 2J.73. p < o.ooi).9 As the target is the same as one of the NPs in the context sentence, subjects might try to predict which noun in the context sentence will
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(i) sentences predormnantly classified as com�ct in the pretest made up the COR category; (ii) sentences predominantly classified as incorrect and as unlikely mistakes by the learner of Dutch made up the INC category; (iii) sentences predominantly classified as incorrect but as likely mistakes by the learner of Dutch made up the EXT category.
Steven Frisson, Dominiek Sandra,
Frank Brisard, Gcrt Van Rillac:r, Hubert Cuyckens
207
experimental lists were used, the three target types of each sentence being assigned to the different lists. Thus, each subject saw each context sentence only once. In each experimental list, there were 10 sentences for each of the three target types. Since both the EXT and the INC category were expected to yield no answers, 10 sentences with target items of the COR category were added as fillers to each list in order to balance the amount of ytS and no answers. Each of the three lists was randomized in three different ways, with each randomization being assigned to the same number of subjects (see above). Six filler sentences were added to the beginning of the experiment to make subjects familiar with the experimental procedure. Filler items were not analysed. As there were three NP positions in the context sentence (apart from the subject NP), each target type was made to occur equally often in each of the three NP positions. Thus it was possible to control for possible recency effects in the data. The number of words intervening between the first and second occurrence of the NP did not significantly differ between the three target types (EXT1 : F < 1 , EXT2: F < 1). All nouns in the context sentence were of the same gender to make sure that the demonstrative could not act as the cue in a prediction strategy (dat versus die in Dutch). Context sentences conveying a scene where another noun would be a more obvious candidate for localizing the trajector were avoided (for example, The lady put the ball-point on the chair and started to sing. John was standing next to this pen.'). In this sentence, 'chair' would be a more obvious and therefore pragmatically preferable candidate for localizing John. Subjects might reject the sentence purely for this reason. It was also taken care of that the critical noun remained present in the described scene, that is, scenes where the referent (the landmark) was 'lost'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
appear in the naast-construction. The experiment was designed to avoid this. Each context sentence contained three NPs (apart from the subject NP, which was never used as a target in the matrix sentence): one of the COR category (e.g. 'trailer'), one of the INC category (e.g. 'dizzy spell'), and one of the EXT category (e.g. 'sleeping pill'). Subjects could not predict which NP would be presented as a target, since the three target types occurred equally often in the experimental list. Each context sentence thus gave rise to three stimuli, one of each category. In order to make the comparison between the INC and the EXT categories possible, the items of each extension type were matched for word length and frequency to the INC items based on the same sentences (means for word length: EXT1 = 7.3, INC1 = 8.o, F(1 , 1 1] = 1 .03, p > 0.1 0; EXT2 = 8.o, INC2 = 84, F < 1 ; means for frequency: EXT1 = 105, INC I = 94. F [1, I I ] = 2.12, p > .10; EXT2 = 8 8, INC2 = 82, F < I ). Since each context sentence defined three target sentences, three
208 Flexible Semantic Processing of Spattal Prepositions
or 'destroyed' or even 'eaten' were omitted, since th� would obviously limit its potential for serving as a reference point to localize the trajector. 3 . 1 .2
Procedure
3.1.3
Subjects
Forty-five students and staff members participated in this experiment. They were native speakers of Dutch.
3 .2 32 .
.
1
Results
Small landmarks (EXT I)
Mean reaction times and error rates for EXT1 and INC1 are presented in Table 5· In the analysis of reaction times, responses with a reaction time above 4000 rns (time-outs) or below JOO ms were removed from the data set
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Instructions and experimental conditions were the same as those reported for the first experiment (c£ section 2.1 ). Each trial started with the context sentence and the matrix sentence appearing (in lowercase letters) simultaneously at the top of the screen. After having read these sentences, subjects had to push the middle key on the button box to clear the screen. Next, a fixation point (an asterisk) appeared for 500 ms in the centre of the screen. When this fixation point disappeared, the target NP appeared in the same position (in uppercase). Subjects had to strike the yes or no key, depending on whether they thought the sentence could be used in Dutch or not. Reaction times were measured from the onset of the target NP to the moment of pressing the response key. If subjects did not respond in time (within 4000 ms), they heard a beep and the experiment moved on. After making a decision, either the message 'Push middle key to proceed' appeared on the screen, or a question was asked about either the context sentence or the matrix sentence, which subjects had to answer with yes or no. These questions (fifteen in each expenmenul list) allowed us to establish whether subjects had actually read the context and matrix sentences. Subjects who did not correctly answer at least r o out of 1 5 questions were excluded from the analysis. After the response, a message appeared on the screen informing subjects that the next trial would start. The following context sentence then appeared. Halfway through the experiment, there was a pause of ten seconds.
Steven Frisson, Dominiek Sandra, Frank Bnsard, Gcrt Van Rillaer, Hubert Cuyckens 209 Table 5 Mean reacnon times (RT), standard devtations {SD) and error percentages for the two extenston �
Category
RT (nu)
SD
Errors
COR I EXT I INC I COR2 EXT2 INC2
1 135 1 276 1 125 1 1 83 1276 1 1 10
273 273 193 246 217 225
s .6 7·8 0.6 12.2 8.1 1.9
(%)
3 .2.2
Large landmarks (EXT2)
The data for EXT2 are also presented in Table S· As in the analysis of the EXTx data, outliers (values above 4000 ms or below 300 ms) were not included in the reaction time analysis. These accounted for two percent of the data. Responses which did not correspond to the expected response type were also excluded (s.I %). A one-way ANOVA (EXT2 vs. INC2) was performed on the mean subject and item reaction times. Significantly higher reaction times were obtained for items in the EXT2 category than for items in the INC2 category (difference: I66 ms; Fs (1 , 42) = 2J.09,p
Discussion
A major motivation behind Experiment 2 was to demonstrate the effect of flexible semantic processing in an environment where subjects made
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(1.7%). Answers which did not correspond to the expected response type were also excluded (4-2%). A one-way ANOVA (EXTI vs. INCI ) was performed on the mean subject and item reaction times. Reaction times in the EXTI category were significantly higher than those in the INCI category (difference: Is I ms; F1{1, 42} = 14.65, p
210 Flexible Semmtic Processmg of Spatial Prepositions
4 GE NERAL D I S CUSSION The major outcome of the two experiments is that language users find it difficult to reject prepositional usages which do not occur in the language, i.e. which are wrong usages of the word in question, but which are nevertheless perceived as cognitively plausible. In a speeded decision task, decision latencies and/or error rates were significantly higher for items of this category than for overtly incorrect usages. In Experiment I, these effects were obtained for two similar but distinct Dutch prepositions (aan and naast), both describing a trajector in a proximity relationship to a landmark. The sentences in this experiment were presented out of context. In Experiment 2, these effects were replicated when the sentences were preceded by a short context sentence. The introduction of this context removed two disadvantages of the first experiment. First, in contrast to what was the case in the previous experiment, the observed semantic flexibility
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
substantially fewer errors than in Experiment I. This purpose was clearly achieved. There was a considerable drop in the error rates between experiments for each extension type of naast {EXTI: 39% errors in Experiment I versus 7.8% in Experiment 2, EXT2: 36% in Experiment I versus 8.I% in Experiment 2� This is probably due to the preceding context sentence substantially restricting the number of possible scenes or interpretations subjects are able to conjure up during the experiment. Apart from a reduction in the error rates, we also witnessed a reduction in the magnitude of the observed effect on reaction time (i.e. the difference between INC and EXT items� For items in the EXTI category, the effect dropped from 442 ms in Experiment I to I s I ms and for items in the EXT2 category, it dropped from 742 ms to I66 ms. Together with the low error rates, these figures seem to warrant the hypothesis that subjects, when confronted with the extended prepositional usages in the context of an explicit scene (described in an immediately preceding sentence), feel more confident in their judgements and reject these extensions far more rapidly (and correctly, this time). Crucially, however, the effect remained significant in the context of the items used in Experiment 2 and hence is not restricted to isolated sentences of the type presented in Experiment 1 . In other words, Experiment 2 basically replicates the effects measured in the previous expenment and extends them to a stimulus environment which seems to be more natural (due to the presence of context) and which, at the same time, gives rise to considerably fewer errors. Thus, it counters the criticism that might be raised with respect to Experiment I.
Steven Frisson, Domiruek Sandra, Frank Brisard, Gen Van Rillaer, Hubert Cuyckens 2 1 1
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
could only be localized i n the meaning of the preposition (and not in that of the prepositional object) due to fixing the referent. Second, the presentation in context caused a considerable drop in error rates for the items in the extension categories, which made it possible to use a larger data set for the reaction time analysis. The difference in outcome for the EXT and INC conditions reveals a decision uncertainty for the extensions used in the experiment. This uncertainty results from the tension between a possible semantic inter pretation which is output by the processor and the subject's stored knowledge that this particular usage does not occur in the language. The resolution of this conflict will obviously take some processing time. The obtained effects support the hypothesis that language users are unable to inhibit flexible lexical-semantic processing. If they were able to bypass this procedure, it seems they would certainly have done so in an experimental situation where this would have facilitated their compliance with task demands. In other words, the effects and their recurrence across" different prepositions, extensions, and experimental methods suggest that stretching word meanings happens quite automatically. This is, indeed, what would be expected if extension principles constitute a natural part of the language processor. The decision stage is the locus where the effects on decision latencies and errors originate. However, the processes leading to a mental representation of the linguistic stimulus may also contribute to the observed latency difference between extensions and plainly incorrect usages. Since the semantic overlap with established usages is higher for extensions than for incorrect usages, the processor may persist longer in trying to generate a mental representation. It seems a plausible assumption that smaller degrees of mismatch between the linguistic stimulus and stored semantic information would look more promising to the processor. The notion of flexibility in on-line lexical-semantic processing is hardly compatible with the conception of word meanings as discrete, rigidly defined entities in the mind of the language user. Rather, it would seem that meanings are essentially malleable, and that the picture of lexical meaning emerging from these experiments is much in line with the type of lexical models as advocated in the field of cognitive linguistics, and more specifically within various versions of cognitive semantics.'0 Thus, if communicative strategies employed by language users to cope with an infinite array of expressible situations will theoretically allow two possibilities, ie. inventing a new word (and explicitly defining it) for every new referent or making flexible use of a familiar word, the latter option seems the more natural one and the least damaging to the communication process. If stretching word usages amounts to a natural
212
Flexible Semantic Processing of Spatial Preposinons
process, a clear dichotomy between correct and incorrect usages cannot be maintained. One must accept an in-between region of non-occurring yet quite conceivable usages of words, which can be introduced into a
Acknowledgements was supponed by the following grants to the second author: a Belgian government grant (IUAP-ll, contract ## .17� a grant of the Bljzotukr Orukrzotlufonds of the Univenity ofAntwerp {BOF-project # D'IL-Go3) and a grant of the
The research for this publication
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
conversation whenever a language user needs them, or which can even become part of the lexicon when they are adopted by other speakers as welL This region can be conceived of as a continuum, ranging from usages that are very 'close' to an existing one to usages that are hardly acceptable. The experimental results are compatible with this notion of a continuum. As mentioned earlier, tinkering with the meanings of words brings the language user into a state of uncertainty as long as an extended usage is not plainly incorrect. The data show that different language users deal with this uncertainty in different ways. For both extensions of each preposition, some subjects gave a majority of yes responses (i.e. they considered the extended usages to be acceptable). The fact that certain subjects predominantly decided in favour of a yes answer and others in favour of a no answer suggests that language users adopt a certain tolerance threshold when dealing with extended word usages and that this threshold is exceeded earlier for some subjects than for others. The possibility, especially in Experiment 1, that subjects confronted with EXT -items try to reinterpret the scenes presented to them in terms of a coherent mental picture might ultimately lead to an acceptance or rejection of the constructed inter pretation, depending on whether the threshold has been surpassed or not. In the end, however, what concerns us primarily in the light of the experimental evidence proposed is what precedes such a reinterpretation, viz. the (automatic) realization that the extensions discussed here are semantically close enough to actual instances of these prepositions encoun tered in Dutch. This semantic similarity triggers an equally automatic mechanism that attempts to link conventionalized usages to the extensions subjects are confronted with. Such a mechanism cannot be suppressed, since it is part of the normal workings of the language processor and is responsible, in cases where a functional link can indeed be found, for diachronic changes in a word's meaning range as they occur in the language community. Experiment 2 unambiguously conft.rrns this scenario, as subjects conceptualizing a referentially plausible and coherent scene which has been worked out for them still experience difficulties in sanctioning the linguistic expression of that scene.
Steven Frisson, Dominick: Sandra, Frank Brisard, Gert Van Rllker, Hubert Cuyckem 2 1 3 Fonds wor Wtlnlschapptlijk Ond�k (contnct # G.o246.97� The first author also work:s at
the University of Glasgow, Departtnent of Psychology, Human Communicanon Research Centre. The authors would like to thank Stamper, Simon Garrod, Alasdair Robertson for hU help with programming the experiment, and two anonymous reviewers for their crincal suggestions. STEVEN FRISSON, DOMINIEK. SAANDRA, FRANK BRISARD, GERT VAN Ril.L\ER, HUBERT CUYCKENS Received: 03.02.98 Uni��US�ty of Antwnp (UF.SIA} Pmwtraat IJ Fmal version received: 12.07.98 B-2000 An� &lg.um t-marl: JUwtt.fnsson®ufoa.ac.�
1
The tenru 'trajector' and 1andmark'
function a.s notional equtvalents for what is ttadtnonally labelled the pre positional object (landmark) and the sentential subject (trajector). The nonons are tak:en from Langack:er's (e.g. 1987) work: m cogmnve grammar. 2 Prepositions such a.s aan and naast clearly differ &om those prepositions that indicate a defirute relation neces sarily holding between TR and LM m that the former class displays relatively vague spectfications a.s to the orientation TR bears to LM. The only thing speci fied for aan and naast is a domiruon or search domam (correspondirtg to the traditional notion of 'proxirnity1 within which TR is to be situated. For preposi tions like '" or on, it suffices to say that their use is sanctioned a.s soon a.s TR conforrru wholly or partially to the relation expressed, and size is at most derivative of the type of relation indi cated (john li11t1 in China andJohn II11t1 1t1 a ttttt are both acceptable sentences, and TR can even exceed the boundaries of LM, provided there remains some perception of inclusion in a functional sense, a.s in John is standing in tk uuur� Thus, for specific locative prepositions such a.s these, size is an entatlment of the
relanon expressed and therefore undoubtedly semantic. At the same time, we see that the pragmatic constramt on size is dismbuted dtf ferenrially over vanous types of loca tive expres.nons, such that it is not completely predictable for each. 3 Sentences such as Dt bibiiotktkbtdlnUU stQat aan tk balu 'The hbrarian is standing at the counter' were not included m the set of expenmental sentences because the spatial relation m these sentences u clearly comple mented by a functional one. That 15 to say that aan, in contrast with naast, is primarily used m contexts where TR is both sparially dose to LM and neces sarily mvolved in an actton or state that is typically associated with the physical object that constitutes LM. Thus, aan tttt burtau ZJtlnt 'to sit at a desk' conveys a strong sense of a person working at that desk (tf this is perceived a.s the prototypical actton hnked with this piece of furniture), which is not at all implied if the same expression is used with naa.st. 4 To run the experiment, we made use of Psyscope (Cohen tt a/. 1993� 5 !u subjects made more errors in EXT2, the overall RT mean for the whole of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
N OTES
� 1 4 Flexible Semannc Processing of Spatial Prepositions
the EXT category is skewed towards EXT1. In Table 4, as there are more erron for EXTI , the mean is skewed toward! EXT2.. 6 There were missing values for three subjects in the subject analysis. because all of theJI responses to EXTx-items were erron. 7 There were mming data for one sub 8
RE FERENCES Baayen.
R
H.,
L
Plepenbrock,
R
�
(1995� '1"k CELEX LLxual Databast (CD-ROM� Phila Gulikers,
delphia, Lingwstic Data Consortium, Uruvenity of Penruylvania. Cohen, J. D., MacWhinney, B., Flatt, M & Provost, J. (1993 � 'PsyScope: a new graphic interacnve environment for designing psychology experiments', &havWral �arch Mdlwds, Instruments
& Omputm, 2.5, �57-71. Gluchberg, S ., Gildea, P. � Book.in,
H.
(1982.� speech: phor?',
'On
understanding non-literal
can people tgnore meta Journal of Vtrbal uanung and Vtrbal &h4vior, 2.1, 85--98. Langacker, R W. (1987� Foutulatwns of Cognitllh! Grammar, VoL I: lko�tical Prmquintfi, Stanford Uruvenity Press,
Stanford. Vandeloise, C
(1991), Spatial Prqxmtwns: A Cast Study from Fmuh, University of
Chicago Press, Chtcago.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
9
ject in the subject amlysis, because all responses to EXT -items were erron. There were missing data for six subJects in the subject analysis, becawe all of their responses to EXTI -items were erron. ru the INC1 items were perceived to be more incorrect than the EXTI itenu, an an.a.lym of variance would be inap propriate, the m:oomu m number of
10
'likely' - responses being much smaller for EXT1 than for INC1. For this reawn a chi-square test was used. It should be noted that we do not necessarily wish to imply the actual representation of extended usages in the mental loucon. Instead, this can be said to be dependent upon the degree of entrenchment (or, in other word!, the relative frequency of the attested usage) attributed to the extension in question. As this u mostly an empirical matter, we choose to stress the potential for such extensions figuring m the lexicon and to descnbe the nature of the possible mechanisms which are involved in these instances of semantic flexibility, wtthout makmg any hard dainu with respect to their representational status.
'Quantifying' Superlatives and Homo Sapiens IOANNIS VELOUDIS
Aristotk Univmity, 'Ikssalonilei
Abstract
I THE PHENOMENA In a series of papers since I97S. Gilles Fauconnier has tried to account for an any- like use of superlatives, grammatical (e.g. the most difficult, the simplest, etc.) or 'pragmatic' (e.g. Einstnn, Onassis, etc.). I will argue below that his pragmatic account is not as 'uniform' as he thinks, or at least not as uniform as it might be; and I will try to provide an alternative, logical, account Simplifying Fauconnier (I975a, b, I979, I98o) we can say (a) that superlative noun phrases can perform a semantic function very similar to the (universal/existential) quantification of any; c£ the parentheses in ( 1 ) below ('+'/'-' shows the presence/absence of a 'quantificational' reading): (I)
a.
John can solve the most difficult problem. (+ john can solve any problem) b. John cannot solve the simplest problem (+ john cannot solve any problem) c. Einstein couldn't solve this problem. (+ 'nobody could solve this problem)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The hypothesis that two logical schemes are, more or less dtrect:ly, involved in the so-called 'quannficational' readings of superlanves LS defended m the present paper. It u argued, m particular, that sentences like, e.g.john can solvt tk most difficult problnn and John cannot solvt tk ta=st probkm can be associated With therr corresponding 'quantificational' interpreta tions, i.e. John can solve any problem' and John cannot solve any problem', only in the contexts in which they are uttered, and undentood, as the key prenuses, A and not B, of modus pontndo pontns and modus tollmdo tolkns, respectively. ThU hypoth.em, it u also argued, gams some generalizations missed m Fauconnier's (1975a. b, 1979, 1980) well-known analysu of the relevant phenomena m tenru of 'pragmatic scalanty'. In particular, (1) 1t can clearly diStinguish between contexts in which these 'quantificational' readings are welcome and contexts in which they are not, (ii) it can naturally account for the alleged 'similarity' between (existential/universal) any and 'quannfying' superlaoves Wee tk most difficult, tht tOSitSt, etc., and, wkt u important, (iii) 1t can uniformly characterize the mosaic of the envtronments in which 'scalarity' phenomena occur, allowing \lS to explain wkt 1t LS about exactly these environments (and not others) that makes them licensers of the 'quantificaoonal' readings.
.2. 1 6
'Qwnnfymg' Superlatives and Homo Sap1ens
d.
Onassis couldn't afford her expenses. (+ 'nobody could afford her expenses');
and (b) that this 'quantificational behaviour' is licensed in some (syntactic)' environments and is forbidden in others; c£, respectively, the parentheses in (2), as well as (1) above, and in ( 3 ): a.
negativ� matrix don't think John can solve the simplest problem. (+ 'I don't think John can solve any problem') b. if clauses I'll be surprised if John can solve the simplest problem. (+ 'I will be surprised if John can solve any problem'} c. too . . . for . . . to . . . John is too stupid to solve the simplest problem. (+ John is too stupid to solve any problem') d. comparativts It is harder to solve the NPI's puzzle than (it is) to solve the most difficult mathematical problem. (+ 'it is harder to solve the NPI's puzzle than (it is) to solve any mathematical problem'} e. univmal statmml ts. Anybody who can solve the simplest problem is fit for this job. (+ 'anybody who can solve any problem is fit for this job') £ only, first John was the only one to make the faintest attempt to solve the problem. (+ John was the only one to make any attempt to solve the problem') g. bifor� John solved the most difficult problem before his audience could show the faintest sign of fatigue. (+ John solved the most difficult problem before his audience could show any sign of fatigue') h. q�tions Can John solve the simplest problem? (+ 'can John solve any problem?'Y (3) a. John solved the most difficult problem. ( - John solved any problem') b. John is solving the simplest problem. (John is solving 'any problem') etc. (2}
I
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Ioannis
Veloudis
2 17
(Only the 'literal' reading of the superlative noun phrases survives in (3): they are understood as definite descriptions with specific reference.) Fauconnier argues that this any- like aspect of superlatives is explicable in pragmatic terms. It is commonly assumed, he says, that the ability to solve a difficult problem, for example, implies (pragmatically, not logically) the ability to solve a simpler one; and, conversely, the inability to solve a simple problem (pragmatically) implies the inability to solve a more difficult one. That is, a pragmatic scale of increasing simplicity is associated with the propositional schema john can solve x'; and, conversely, a pragmatic scale of decreasing simplicity is associated with the propositional schema john cannot solve x':
x. simplicity x� m
If we assume that the extremes of such scales are denoted by superlatives, 1.e. M = 'the simplest problem' and m = 'the most difficult problem', then, Fauconnier argues, we can easily account for the similarities between superlatives and any: G1ven the type of implication which characterizes such scales, if m is the minimum point on the scale, then for any x: R(m)
=>
R(x)
and therefore R{m)
=>
Tlx R{x)
(Fauconnier 1 979: 290)
(c£ (xa) above); and, conversely, if M is the then for any x: "' R(M)
=>
maxim um
point on the scale,
rvR(x)
and therefore "' R(M)
=>
Vx "' R (x)
(c£ (xb) above� That is to say, �arion reverses implication: if R (x . ) => R(x� ) , then "' R(x� ) => rvR(x . ).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
M
+
2. 1 8 'Quantifying' Superlatives :and Homo Sapiem This 'scale reversing' is the key notion in Fauconnier's account. Compare the examples in (1) with those in ( r ' ) below: (r')
a.
John cannot solve the most difficult problem. ( - john cannot solve any problem') b. John can solve the simplest problem. (- john can solve any problem')
etc.
(2' )
a.
b.
I don't think John can solve the most difficult problem. (- 'I don't think John can solve any problem') I'll be surprised if John can solve the most difficult problem. {- 'I'll be surprised if John can solve any problem')
etc. We can conclude the presentation of the main points of this analysis by quoting the following clarification with respect to reversing implication:
negation plays no speoal role in explaining these phenomexu: it i! simply one of the nuny scale reversing environments (perhaps a statistically dominant one) [fn. omitted]. (Fauconnier 1979= 2.9 5) 0
0
0
In my view, Fauconnier's papers raise at least which he does not address:
three serious questions,
Question (i) It can easily be seen that environments like those in ( 1 ) and (2) above constitute a necessary but not sufficient condition for the occurrence of the 'quantificational' readings; sentence ( 1 a), for example, can have a second,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It is obvious that in (� a-b), for example, both the lower and the upper end of the relevant scale can give rise to (pragmatic) implications; the environment of negation simply reverses their flow: it can change a sterile end (c£ 'the simplest problem' in (1'b) ) into a fertile one (c£ (rb) }; and, conversely, it can change a fertile end (c£ 'the most difficult problem' in (1a) ) into a sterile one (c£ {I'a)). If we now add (a) that a number of (quite varied) environments (c£ the cases in (2) } share with negation this property of 'scale reversing', i.e. can give rise to 'quantificational' readings for superlatives when such readings are absent in the simple declarative counterparts, and (b) that these environments are 'the very same which reverse standard (grammatical) polarity' (Fauconnier 1 979: 292), then Fauconnier's account spreads over the whole of our data so far, and the facts in (2) and ( 3 ) above, as well as in (2') below, become naturally predictable:
Ioannis
Veloudis
2 19
non-'quantificational', reading as well: john can solve the most difficult of the (specific) problems he has in front of him' (c£ the 'literal' readings of the superlatives in ( 3) ) Is there any sufficient condition for the any- like aspect of superlatives, or should we simply be content with the observation that in some (syntactic, see note I) environments sentences with superlatives become ambiguous between a 'literal' and a 'quantificational' reading? Fauconnier's papers presumably say 'yes' to the second of these questions, considering the ambiguity between a literal and a 'quantificational' reading simply as an instance of ambiguity resolved in context. I will argue below that the answer to the first question, 'is there any sufficient condition for the any- like aspect of superlatives?', can (non-trivially) be 'yes'. .
We have seen that in their 'quantificational' reading superlatives belong to the same end of a pragiDatic scale as, and become paraphrasable by, the gramiDatical quantifier any. What does this mean as far as their semantics is concerned? Can we really assume that the expression tlu simpkst in {I b), for instance, is (pragmatically?) 'synonymous' with any? In any case, what is it with any that IDakes it pattern here differently from other universal quantifiers, e.g. �? ' Fauconnier's answer is that any is not a quantifier but rather indicates the extremity of a scale' (I 979: 297). (Actually, it may indicate either end point of a scale, being thus 'synonymous' with superlative expressions like tlu simplest or tlu most difficult.) I will argue below that this quasi-synonymity between any and the 'quantificational' aspect of superlatives can be accounted for in logical terms. In particular, I will argue that any and the corresponding 'extremity of a scale' are real alternatives in either of the key premises A and not B of two well-known logical schemes, namely, modus ponendo ponms and modus tollendo tollens, respectively; and that in being real alternatives in either of these premises they logically guarantee the truth of the same conclusion.
Question (iii) The 'quantificational' readings under discussion are welcome only in environments like those in {I) and (2). What are the grounds for this restriction? To say simply that the 'scale reversing' environments, i.e. negation (c£ {Ilrd)), comparison, if-clauses, etc. (c£ (2a-h) ), are, or are among. the environments that license negative polarity items (NPI's) would hardly be a satisfactory answer. As Haspelmath points out in his
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Question (ii)
220
'Quantifying' Superlatives and Homo Sapiens
illuminating typological study of indefinites, 'one must of course ask further what it is about j ust these contexts that gives them their scale reversing property' (I993: 1 1 2, fn. I 9).
This is not the only problem here, though: we must ask further what it is
about all these environments and that in (�a) that makes them 'licensers' of the any- like aspect of superlatives. Is it simply a matter of coincidence that they share this property? of imperative.
This
Any, for instance, is welcome in the environment
environment, though, cannot function as a 'licenser' of
that any- like aspect of superlatives. What can this be due to? Obviously, to
answer, one should of course ask further what it is about imperatives that precludes them from contexts of pragmatic scalarity.
In what follows I will try to provide an explanation for the phenomena examined by Fauconnier, 3 modifying his pragmatic account so that it can meet questions (i)-(iii) above. In the first part of the discussion I will restrict myself to the data in (I). 2 A HYP O T HE S I S I n my opinion. the 'quantificational' readings of (I a-d) are explicable o n the basis of the hypothesis that the propositions expressed by the 'quantifying' superlative- versions of examples like (I a-d) above directly correspond to the minor premises of modus to premise
(ponmdo) ponms and modus (tollendo) tollms, i.e.
A (c£ (� a)),
If A. then B4
A Therefore, B or to premise
not B (c£ (I b-d) ),
If A. then B
not B
Therefore, not A functioning thus as symbolic, or, better, metonymic, representatives of �e whole modus. (Note that in a sense A and ponens and modus tollens, respectively.)
not B are the key premises of modus
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
say that imperatives, i.e. instances of ordering someone to do x, do not quantify over scales, because ordering someone to do x just doesn't contain the order to also do x'(if x and x'are on some scale), would hardly explain anything; the original problem still remains, to the extent that the content of that 'because-' clause simply paraphrases the restriction it is supposed to account for. If one wants to avoid such a, more or less, question-begging
Ioannis
Veloudis
.2.21
This hypothesis seems to be encouraged by two facts: (i) the major premise, ifA1 then B, in both modi is easily recoverable, as it is pan of the assumed common ground between speaker and hearer (c£ Fauconnier's 'pragmatic scalarity); and (ii) what the speaker wants to communicate to his/her hearer in the case of a 'quantificational' reading corresponds in fact to the conclusion of either modus. Suppose that speaker S , cannot solve a particular mathematical problem, say problem a, and discusses with her hearer, S�, candidates for solving it: How about John? John cannot solve the simplest problem.
S2's (unambiguously 'quantificati�nal') contribution in this context probably imphes that john cannot solve problem a'. This implication (not A}, however, is exactly what S, would logtcally conclude on the grounds of what S2 has said, i.e. john cannot solve the simplest problem' (not B), taken together with the pragmatic implication 'If John can solve problem a, he can solve the simplest problem' (ifA1 thm B). And the sources of the latter implication are obvious: S:;r.'s utterance, as far as the apodosis B is concerned, the context of S�'s utterance, as far as the protasis A is concerned, their common knowledge, as far as the (pragmatic) connection if A1 thm B is concerned. Suppose now that S:;r."s answer in the context above is as follows: S,: S:;r.:
How about John? John can solve the most difficult problem.
S:;r.'s (unambiguously 'quantificational') contribution this time probably implies that john can solve problem a'. This implication (B), however, is exactly what S, would logically conclude on the grounds of what S:;r. has said, i.e. john can solve the most difficult problem' (A), taken together with the pragmatic implication 'if John can solve the most difficult problem, he can solve problem a' (if A1 thm B). And the sources of the latter implication are again obvious: S./s very utterance, as far as the protasis A is concerned, the context of S:;r.'s utterance, as far as the apodosis is concerned, their common knowledge, as far as the (pragmatic) connection if A1 thm B is concerned. s What can this hypothesis predict concerning our data in (1)? Probably, it can predict that (1a) has a 'quantificational' reading simply because/when it is underscmdable as the second premise of the logical sequence (modus ponen.s}:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
S,: S�:
222
'Quantifying' Superlatives and Homo Sapiens If John
can solve {the most difficult/any}6 problem, then he problem(s} a (, b, etc.}.
can
solve
John can solve the (most dijfteult/any} problmr. Therefore, John can solve problem(s} a (, b, etc.). (where John can solve the most difficult/any problem' = A , John can solve problem(s) a (, b, etc.)' = B, and 'problern(s) a (, b, etc.)' is (are) somehow present in the universe of discourse� Similarly, our hypothesis predicts that (1b) has a 'quantificational' reading simply because/when it is understandable as the second premise of the logical sequence (modus to/lens): solve problern(s} a (, b, etc.), then he can solve {the simplest/ some} problem. John cannot solve {the simpkst/anyy problem. Therefore, John cannot solve problem(s} a ( b, etc.) ,
(where John can solve problem(s} a (, b, etc.)' = A, John can solve the simplest/some problem' = B, and 'problem(s) a (, b, etc.}' is (are) somehow present in the universe of discourse). In the same manner, our hypothesis predicts that (1c) has a 'quantifica tional' reading because/when it is understandable as a second premise of the logical sequence (modus to/lens): If a (, b, etc.) could solve this problem, {Einstein/somebody} could solve it. {Einstrin couldn't/Nobody could} solve this problem.
Therefore, a (,b, etc.) couldn't solve this problem. (where 'a (, b, etc.) could solve this problem' = A, 'Einstein/somebody could solve this problem' = B, and 'person(s) a, (b, etc.)' is (are) somehow present in the universe of discourse�
3
E XPLANATO RY VALUE
Let me examine to what extent these predictions answer questions (i}-(iii) above: Answer
We
(i)
now say that the 'quantificational' reading of (�a), of (1b), etc., is allowed on condition that (1a), (1b), etc., can be understood as the second can
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
IfJohn can
loannis
Veloudis 2.2.3
IfJohn can solve problem(s) a (, b, etc.), then he can solve {the simplest/ some/*every} problem.8 John cannot solvt {tk simpkst/any/*every} probkm. John cannot solve problem(s) a (, b, etc.). It is worth pointing out in this connection that the fact that the superlative
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
premise of either modus, as they have been presented above; if this is not possible, i.e. if the proposition expressed cannot be understood as the second premiss of either modus, simply for logico-conversational reasons, the 'quantificational' reading is precluded. Suppose, for example, that John, an intelligent schoolboy, is in a puzzle about the easiest problem of his homework, although he has quite easily solved the other, more difficult ones. Referring to this rather unexpected situation his father can naturally utter something like ( 1 b): John cannot solve tht simpkst probmn; probably what he is 'saying' in this case has no place in the sequence of modus tollms (and, of course, it has no place in the sequence of modus potwtS, either); and, predictably, what he is 'saying' in this case can by no means have a 'quantificational' reading. More specifically, what the father is saying in this context is acceptable only as a description of the (unexpected) situation; that is, his utterance of (1b) can by no means be understood as a premise, namely not B, of modus tollms, for conversational, as well as logical reasons: obviously, what the father 'knows' in this case, i.e. (i) his (pragmatic) knowledge of the fact that 'ifJohn cannot solve the easiest problem, he cannot solve any problem', on the one hand, and of the fact that 'ifJohn can solve the most difficult problem, he can solve any problem', on the other hand, and (ii) his experience of the (unexpected) fact that john cannot solve the easiest problem', on the one hand, and john can solve the most difficult problem', on the other hand, would logically (by modus ponens) lead him simultaneously to two contradictory conclusions. The environment of imperative (see above) also creates an analogous situation: Don't solvt the simpkst probkm, John!, for example, can by no means be understood as the minor premise, not B, of modus to/lens; and, predictably, it can by no means have a 'quantificational' reading. An imperative, that is to say, cannot quantify over scales, simply because the utterance in which it occurs does not qualify as the second premise, not B, of modus tollms. (In fact it does not qualify as a premise at all.) What is more important, however, is that we can now readily explain the asymmetry between the any- and the evtry- versions of quantification in cases where, according to the hypothesis defended here, modus tollens is involved. Evtry patterns differently, we can now say, simply because it has no place in the relevant logical sequence. C£ the anomaly in, e.g.,
224
'Quantifying' Superlatives and Homo Sapiens
simpkst corresponds to a negated existential (not universal} any (c£ Fauconnier I97Sa and references therein) becomes now readily predictable, and is naturally explicable: due to the particular version of Qogical) implication, A --+ B (: 'if John can solve problem(s} a (, b, etc.}, then he can solve some problem'), that fills in the first premise slot in the sequence above, for example, any in the second premise, not B (: john cannot solve any problem'), invariably corresponds to the existential quantifier so� it can thus be understood only as an existential any. In addition, we can now readily explain another asymmetry, pointed out in Fauconnier ( I 97sa: 3 7 I}: 'Superlatives, but not the quantifier any, can occur in subject position of a negated sentence.' C£ the following examples (his (1 5 7) and ( I s 8), respectively):
the
It is obvious that only the second of these two sentences (Fauconnier's (I s 8 ) ) can b e uttered, and understood, as the second premise, not B, of modus
toliens: If girl(s) a (, b, etc.) could seduce John, then the most beautiful girl could seduce John. The most btautiful girl could not seduce John. Therefore, girl(s} a (, b, etc.} could not seduce John. It is, predictably, then acceptable. On the other hand, the first sentence (Fauconnier's (I s 7) ) cannot be uttered, and understood, as the second premise, not B, of this logical sequence: If girl(s) a (, b, etc.) could seduce John, then some girl could seduce John.
*Any girl could not seduce John. ??? It is, predictably, then anomalous, since what it 'says' differs considerably from 'it is not the case that some girl could seduce John', i.e. from not B. We can legitimately maintain, therefore, that in terms of the hypothesis argued for here a common characteristic is uniformly ascribable to the contexts in which the 'quantificational' reading of a superlative is welcome: these contexts constitute a subset of the set of contexts in which modus potwtS and modus tolkns are involved,9 being metonymically represented by their key premises, A and not B, respectively. (Needless to say, in terms of Fauconnier's account the ambiguity between the literal and the 'quantifi cational' readings of a superlative is considered as no different from any other kind of ambiguity resolved in context.}
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
*Any girl could not seduce John. The most beautiful girl could not seduce John.
Ioannis Vdoudis Answer
225
(ii)
We can now characterize the semantic relationship between the super lative- and the corresponding any- versions of our data: what makes them look 'synonymous' is the fact that they share the same premise slot in modus ponens or modus tollms. In particular, if a superlative occurs in the A- slot of
·10
�
respectively). That is, the 'quantificational' versions of the examples
in (I) cannot be differentiated from their any- counterparts, as far as their contributions to the logical sequence of either modus are concerned: this is
what their alleged 'synonymity' rests with, and is confined to. We can conclude, therefore, that the hypothesis presented above
allows us to naturally explain the similarity in function between superlatives and any in cases involving pragmatic scalarity, avoiding the (otherwise unwarranted) assumption that 'any is not a quantifier but rather indicates the extremity of a scale' (see question
(ii) in the preceding
section).
Answer
(iii)
We can now explain the so-called 'scale reversing' property of negation, i.e., why the presence of negation can give rise to a 'quanrificational' reading in some cases (c£ (I b-d) ) and obliterate it in other cases (c£ (I 1a) ), on the one hand; and why the absence of negation can give rise to a 'quanrificational' reading in some cases
(c£
(�a) ) and obliterate it in other
cases (c£ (I 'b)), on the other. In particular, the second premise of
tollens
modus
This obviously means that a sentence with a not, conforming thus to not B, if what it 'says' is
is negative in form.
superlative must contain
expected, according to the logical flow of the sequence, to occur in that premise slot; otherwise, my hypothesis predicts, the 'quantificational' reading of that sentence will be cancelled. C£ (I'b): it cannot possibly
be
incorporated in
modus tolkns:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
modus ponens, then there is no logical harm in substituting this premise by A' , provided that A' differs from A only in that it contains the any counterpart of the superlative (c£ (�a): John can solve the most difficult problem can naturally be substituted by John can solve any probkm); and if a superlative occurs in the not B- slot of modus tollens, then there is no logical harm in substituting this premise by not H, provided that H differs from B only in that it contains the any- counterpart of the superlative (c£ ( Ib-d}: John cannot solvt the simpkrt problem, Eins�n couldn't solvt this probkm, and Onassis couldn't alford her oprnses can naturally be substituted byJohn cannot solve any probkm, Nobody could solvt this probkm, and Nobody could alford her
2.2.6 'Quantifying' Superlatives and Homo Sapiens IfJohn can solve problem(s) a (, b, etc.), then he can solve {the simplest/ some} problem. John can solve {the simplest/some} problem. m (the logical sequence breaks down, since the form of distorted:
modus tolkns
1S
If A, then B
B
???);
If John can solve {the most difficult/any} problem, he can solve problem(s) a (, b, etc.). John cannot solve {the most diffu:ult/any} problem.
??? (the logical sequence breaks down, since the form of modus potm!S is distorted: If A, then B not A
???); the non-'quantificational'reading is, then, the only one left for (11a). We can conclude, therefore, that the hypothesis defended here accounts for the 'quantificational' reading of superlatives on the grounds of logical necessity, in particular of logical necessity dependent on the idiosyncrasies of material implication (: A -+ B is false if (and only if) A is true and B is false� In Fauconnier's proposal, on the other hand, this reading is accounted for in terms of 'pragmatic scalarity' and the logical/semantic propenies of the expressions involved. This difference will be made more evident if we pass to the second part of our discussion. It is obvious that the preceding paragraphs meet our questions (i)-{iii) in part only: in fact the mosaic of environments in (2) has hardly been touched
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the non-'quantificational' reading is, then, the only one left for (1 'b). Similarly, the second premise of modus ponens is affirmative. This obviously means that a sentence with a superlative must not contain not, conforming thus to A, if what it 'says' is expected, according to the logical flow of the sequence, to occur in that premise slot; otherwise, my hypothesis predicts, ' the 'quantificational' reading of that sentence will be cancelled. C£ ( I a): it cannot possibly be incorporated in modus ponms:
Joannis Veloudis
1.27
(2a) I don't think John can solve {the simplest/any} problem.1 1 as expressing a subjectively modified paraphrasis of the 'quantificational' reading of ( 1 b) (rb) John cannot solve {the simplest/any} problem. i.e. as a (moderate) manifestation of the key premise not B of modus tolkns. Similarly, will bt surprised if and is too stupid to can be taken as analogous, though less moderate, variants of not; thus the 'quantificational' readings of (2b) and (2e), respectively, (2b)
111
be surprised if John can solve {the simplest/any} problem.
(2e) John is too stupid to solve {the simplest/any} problem.
can intuitively be understood as expressing manifestations of not B.14 On the other hand, the 'quantificational' reading of (2d) (2d) It is harder to solve the NPI's puzzle than (it is) to solve {the most difficult/any} mathematical problem. does not seem to be readily reducible to not B. How can this be accounted for? Should we say that modus ponms, instead of modus tolkns, is involved here? (2d) could of course be said to directly correspond to the minor
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
on so far. What is it that makes them 1icensers' of the 'quantificational' reading of superlatives? And what is their relationship, if any, with the 1icensers' of the 'quantificational' readings already discussed (c£ (�a-d))? Moreover, how can the hypothesis presented above be extended so that it can be applicable to our data as a whole? I will argue below that the 'quantificational' readings in (2) are more or less directly reducible to the second premise, not B, of modus tollens, that the environments in (2) function as 'licensers' for the same reason as the environments in (rb-d);1 1 and that our data on 'quanrificational' super latives are in general explicable in terms of their association with modus potm�S and modus tollens, the only requirement for this being that my hypothesis above be slightly modified: The propositions expressed by the 'quantifying' superlative versions of (�a-d) and (2a-h) (as well as by their any- counterparts, it can now be added) more or less directly correspond to the minor premises of modus ponms and modus tollens, i.e. to A (c£ (�a) ), or to not B (c£ (rb-d) and (2a-h) ), functioning thus as metonymic representatives of, or in some sense reflecting, the whole modus. Let me in tum discuss (2a-h) in the light of this modification. Considering not think as a moderate variant of not/"' we can intuitively understand (2a)
2.2.8 'Quantifying' Superlatives and Homo Sapiens
premise A of modus ponms. This would be of no following sequence:
use,
though. Consider the
If it is harder to solve the NPI's puzzle than it is to solve {the most difficult/any} mathematical problem, then it is harder to solve the NPI's puzzle than it is to solve problem(s) a (, b, etc.). It is harder to solvt tht NPfs puzzlt than it is . . . (=ul)
Therefore, it is harder to solve the NPI's puzzle than it is to solve problem(s) a (, b, etc.) Is this
S 1 : I'm going to solve the NPI's puzzle. S�: It is harder to solve the NPI's puzzle than it is . . . (=2d) It is obvious that (2d) in this context is intuitively understood as strongly implying 'you cannot solve the NPI's puzzle', 'forget itl', 'you can't be serious!', and so on. These implications, however, have nothing to do with the conclusion 'it is harder to solve the NPI's puzzle than to solve problem(s) a (, b, etc.)' in the sequence above. How then can this conversational contribution of (2d) to the contexts of its use be accounted for? People working on grammaticalization have emphasized the role of speakers and hearers negotiating meaning in communicative situations. It has been maintained, in particular (c£ Hopper & Traugott 1993: 86-7), that expressivity may motivate metonymic inferencing, more specifically, that 'the search for ways to regulate communication and negotiate speaker and hearer interaction' underlies metonymic processes such as 'specifying one meaning in terms of another that is present, even if covertly, in the context' (1993: 87), or better, 'indexing or pointing to meanings that might otherwise be only covert, but are a natural part of conversational practice' (1993: 86-7). In my opinion (ul), as well as the whole of our examples here, under their 'quantificational' readings, constitute instances of such a metonynic 'pointing to meanings'. In this light, the peculiarity of (ul) can be said to be simply due to the fact that what (ul) 'says' in a sense comments on a (moderate) variant of, rather than, more or less, directly corresponds to, the key premise not B of modus tolltns. Consider (4) below: (4) It is hard
to
solve {the most difficult/any} mathematical problem
the content of which is comparable to the key premise of a (moderate) variant of modus tolltns:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
sequence involved in the contexts in which the 'quantificational' reading of (2d) is welcome? Suppose that S 1 has decided to deal with the NPI's puzzle and that she wants to discuss her decision with S�:
loannis Veloudis
2.2.9
If one can solve the NPI's puzzle, one can solve {the most difficult/any} mathematical problem. o� can hardly solvt {the most difficult/any) mathematical probkm. Therefore, one can hardly solve the NPI's puzzle.
(2e)
Anybody who can solve {the simplest/any} problem is fit for
this job.
The 'quantificational' reading here involves the sequence If somebody can solve some specific problem(s) a (, b, etc.), then somebody can solve the simplest problem.
Nobody can solvt t� simplest problem. Therefore, nobody can solve some specific problem(s) a (, b, etc.) in a very interesting way. What (2e) communicates is something like 'everybody is fit for this job'. What is it then that makes this reading possible? Intuitively, (2e) defines a very low standard ofjudgement; and this definition is made on the basis of the following (minimal) exception: if there is somebody who falsifies 'nobody can solve the simplest problem'; i.e. it is made on the basis of the falsification of the key premise above. If this is the case, however, the 'exceptional' cases implied here are directly dependent on the key premise of modus tollms.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If we take (2d) as expressing a comment on the grade denoted by (4) (= 'the degree of difficulty denoted by (4) is the starting point of the gradation of the difficulty of NPI's'), we can intuitively understand what it is that makes the 'quanti£cational' reading of (ul) imply a much stronger variant of the conclusion 'one can hardly solve the NPI's puzzle' above. By asserting that the difficulty of NPI's is of a higher grade than the difficulty of the most stubborn mathematical problems, (2d) not only anticipates the truth of the conclusion 'one can hardly solve the NPI's puzzle' above, but also makes it comparable to the truth of the key premise 'one can hardly solve the most difficult/any mathematical problems' of our variant of modus tollms. Considering now this comment on the key premise as a reinforced variant of that key premiss, we expect our modified sequence of modus tollms to lead to a much stronger conclusion, i.e. to 'the NPI's puzzle can by no means be solved', more or less; this is, however, exactly what our (2d) implies. The 'quantificational' reading of (ul), therefore, is after all somehow reducible to the key premise of modus to/lens. A second instance of indirect reducibility to not B, and another characteristic piece of meaning negotiation, is exemplified by (2e):
�30 'Quantifying' Superlatives and Homo Sapiens An analogous exception involving 'quantificational' reading of (2f):
not B
is
implied
in the
(2f) John was the only one to make {the faintest/any} attempt to solve the problem. John is described here as the only person that falsifies the key premise, and consequently the necessity of the conclusion, of the sequence If one
made attempt(s) a (, b, etc.) to solve the problem, one made {the faintest/some} attempt to solve the problem.
No otU nuuk tM faintest/any attmtpt to solve the problem. That is, the 'quantificational' reading of (2f} is characteristically dependent on modus tolkns. In the same vein, the 'quantificational' reading of (2g) (2g) John solved the most difficult problem before his audience could show {the faintest/any} sign of fatigue.
expresses a comment on the validity of the sequence If somebody solved the most difficult problem before his audience could show sign(s} a (, b, etc.} of fatigue, then somebody solved the most difficult problem before his audience could show {the faintest/any} sign of fatigue.
Nobody sol�d the most difficult problmt btfore his audience could show the fointfit/any sign offatigue. Therefore, nobody solved the most difficult problem before his audience could show sign(s} a (, b, etc.) of fatigue. '
'
Intuitively, this reading of (2g) says that the modus is inapplicable as far as John is concerned. What is it that makes John exceptional? John is said to be the only person that falsified the key premise Nobody solved tk most difficult
problm� befor� his audienu could show {the faintest/any) sign of fotig�. The relationship between not B and the 'quantificational' reading is again obvious. The question in (2h), finally, (2h) Can John solve {the simplest/any} problem? can be said to point to this relationship in a very interesting way. Questions with 'quantifying'superlatives are normally given a rhetorical interpretation.' � Needless to say, this peculiarity can hardly be a matter of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Therefore, no one made attempt(s) a (, b, etc.) to solve the problem.
loannu Vdoudis
3J I
4
EVIDENCE
The intuitive predictions, as well as the linguistic generalizations, in the preceding section can only show the explanatory value and the 'attractive ness', if any, of the hypothesis I have been arguing for. They can hardly be considered as constituting direct evidence for the involvement of modus ponens and modus tollms in the 'quantificational' readings. In what follows I will attempt to provide some pieces of this evidence by distinguishing three semantic characteristics of our data that have been passed over in the relevant literature, to the best of my knowledge. First, I have been arguing that the 'quantificational' readings of the negative examples in ( 1) directly correspond to the key premise not B of modus tolkns. However, not in this premise unexceptionally exemplifies the contradictory aspect of negation (: 'it is not the case that . . .'). The implications this restriction has for my hypothesis are obvious: if the 'quantificational' readings of ( 1 b-d) are actually related to not B, their negative particles must necessarily have a contradictory interpretation only.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
coincidence. Has it anything to do with the hypothesis that the 'quantificational' reading of (2h) is reducible to the key premise of modus tolkns? We can intuitively understand (2h) as an emphatic alternative to its probable answer. More clearly, (2h) can be considered as an emphatic assertion of the proposition that the answerJohn cannot solve {the simpkst/any} problem (c£ (1b)) normally expresses: the speaker feels confident that the truth of the latter is self-evident, well-known, etc.; and to show this confidence (s)he &eely takes the 'risk' of the corresponding question, implying thus that the difference between 'questioning' and 'answering' is neutralized in this case: the question can safely be used to denote this particular answer, since any other answer is (considered to be) precluded! (The same intuition, of course, extends to rhetorical questions in general.) If we take this point of view, the presence of (2h) among the data involving modus tollens is no more curious: if its 'quantillcational' reading is in fact an alternative, and emphatic, way of saying john cannot solve the simplest/any problem', it is simply an alternative, and emphatic, way of saying not B (c£ my discussion of (1b) above). We can, therefore, conclude that our data in (2), as well, are explicable in terms of the (modified) hypothesis presented above: the 'quantificational' readings of (2a-h) echo, in a more or less indirect way, modus tolkns, 'presupposing' in a sense, rather than posing (c£ (1b-d)), the key premise of this logical sequence; hence their metalinguistic, so to speak, flavour.
232 'Quantifying' Superlaoves and Homo Sapiens
(4)
a John can solve problem(s) a·(, b, etc.). b. John cannot solve problem(s) a ( b, etc.). ,
I can equally well cooperate, however, by responding with either of the indirect answers in (s) (c£ also (�a-b) above): (s)
a John can solve {the most difficult/any} problem. b. John cannot solve {the easiest/any} problem.
According to the hypothesis I have been defending, this is probably due to the tautological character of the two logical modi, modus ponens and modus tollens, symbolically represented here by their key premises A, i.e. (sa) (c£ also (�a)) and not B, i.e. (sb) (c£ also {Ib) ), respectively. This being the case, the hypothesis that modus ponnu and modus tolkns are involved in the 'quantificational' readings is in accordance with the conversational characteristics of the latter. To restrict myself to (4) and (s) above, (�) and (4b), the conversational variants of (sa) and (sb), respectively, are exactly the conclusions that the relevant logical sequences (modus ponms
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We can easily see that this is in fact the case: the 'quantifi.cational' versions of (I b-d) can be associated with the contradictory interpretation ('it is not the case that . . . can/could solve/afford . . .'), and by no means with a contrary interpretation ('it is the case that . . . cannot/could not solve/ afford . . .') of not; under its 'quantifi.cational' interpretation (Ib), for instance, can never be associated with a paraphrasis like 'it is the case that John leaves the simplest (: any) problem unsolved'. Their corresponding literal versions, on the other hand, do not have this restriction; in the context described in the preceding section (Answer (i) ), for example, what the father says in uttering {Ib) is indeed paraphrasable as 'it is the case that John leaves the simplest problem unsolved'. Second, it has been pointed out that rhetorical questions are usually asked if the speaker considers it obvious that the relevant answer is negative (see note I S; see also Haspelrnath I 993: 48, and references cited there). I consider this remark as a piece of (independent) evidence for the treatment of (2h) above: the fact that rhetorical questions like Who wants any b�ans? (from Lakoff I 969: 609) are normally understood as anticipating a negative answer is in accordance with, if it is not explicable in terms of. my suggestion that these rhetorical questions are in a �ogico-conversational) way associated with the key premise not B. Third, surprisingly it has not been stressed in the relevant literature that the 'quantificational' readings we have been discussing are indirect ways of 'saying' something. Suppose, for example, that I am asked whether John can solve (specific) problem(s) a (, b, etc.). I can cooperate by responding with either of the 'literal' answers in (4):
Ioann.is
VeloudiJ
233
and modus tollens, respectively) lead to. That is, the hypothesis I have been arguing for gains additional support &om the fact that the uses of sentences it (independently) brings together; i.e. the key premises and the conclusions in the analyses of the logical sequences presented above, are exactly uses of sentences which are conversationally very close to each other. Needless to say, this conversational interchangeability will simply be considered as a matter of coincidence by any account in which modus ponms and modus tollens play no role.
The hypothesis defended above, if correct, has some obvious theoretical implications. It would seem interesting, for instance, to those interested in grarnmaticalization phenomena to examine to what extent modus tollens has been conversationally exploited as a mechanism for the reinforcement of negative constructions. It is not probably a matter of coincidence that not ( < na wiht 'no thing') in English. (�) . . . pas ( < � va pas) in French, (ov)S€v) (< otiS€ lv 'not even one {thing)') in Greek, nemo (< � hemo [homo < hmw] 'nobody') in Latin. etc., historically come &om expressions meaning 'minimal quantities','6 i.e. &om 'quantificational' expressions like those dealt with in Fauconnier's papers. More generally, the preceding analysis of 'quantificational' readings as involving logical modi can be naturally accommodated in any theory that proposes a systematic analysis of language as rooted in general human cognitive abilities. Such a proposal is argued for in Sweetser (1 990), for example. She maintains in particular that a cognitively based theory is 'the right way to tackle the issues of multiple form-to-function mappings' (1990: 12� Considering our data as a particular instance of such 'multiple form-to-function mappings', we can hardly be surprised from the following correspondences. Sweetser ( 1 990: 64- s) argues:
Any sentence can be viewed under two :upects: :u a description of a real-world situation or event, and :u a self-contained part of our belief system (e.g. a conclwion or a premise� & descriptions, sentences describe real-world events and the causal forces leading up to those events; :u conclusums they are themselves undentood :u bemg the result of the episternic forces which cause the tram of reasoning leading to a concluston.
This general statement is in a sense exemplified by the data we have been
analysing: the non-'quantificational' and the 'quantificational' readings in (r)-(2) can be said to correspond to the real world aspect and the episternic world aspect, respectively. In addition. what is more important, the 'quantificational' readings can be understood as premises {or better: as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
5 THE O RE T I CAL CO N S I D E RAT I O N S
234 'Quantifying' Superlatives and Homo Sapiens
.
.
Now what we mean by a mathematical symbol is a set of mathemancal equations and relanonships. We are able, therefore, in mathematical tenru, to find the optimum solution. (S.6.!c.26) can
would not be rmpossible implications of actuality.
m
these examples, but � abk to is preferred because of the
Rather curiously, if the present positive form can is used there may be indication of actuality, but in the future rather than the present . . . That is to say, while IS abk to says 'can and does', can says 'can and wiU. do'. (Palmer 1979: 77) If this is the case, i.e. ifcan in fact means that 'there is ability now which may be actualized in the future' (Palmer 1 979: Ss ), then can is distantly related to the actual world, to say the least; and this is probably in harmony with the epistemic atmosphere of our 'quantifying' superlative noun phrases.
6
E P I L O GUE
Obviously, the analysis I have been arguing for, if acceptable, has considerable consequences for a number of relevant questions. It would be interesting to examine, for instance, to what extent the alleged 'genericity' of 'free-choice' any in sentences like Any owl hunts mice (from
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
key premisses-substitutes for conclusions) of modus potmtS and modus to/lens, i.e. of two well-known examples of 'epistemic forces which cause the train of reasoning leading to a conclusion'. Besides these remarkable (descriptive) correspondences, we have, I think, good reasons for arguing that our data belong to the (speaker's) epistemic world. First, as it has been pointed out independently, in their 'quantifica tional' interpretation superlative noun phrases 'need not have any reference or even any possible reference: their function is [ . . . ] not referential' (Fauconnier 1 980: 6o).'7 That is, the simplest problem, t� most difficult problem, etc. in (1}-2) do not refer to, or even entail the existence of, a particular problem which is such-and-such; the latter has no place in the real world: it probably belongs to the speaker's episternic world. Second, the various verbal modifications of the infinitival pattern (to) solve x in (1}-(2), i.e. can, cannot, I do not think . . . , it is harder . . . , etc. are of a particular character: they imply, if they do not denote, the speaker's beliefs, judgements, calculations, etc. as far as the truth of the infinitival achievement is concerned. That is, can, cannot, I don't think . , etc. introduce us to the speaker's episternic world, rather than describe real-world situations. Third, it is interesting to note the special character of can in the 'quantificational' readings we have been discussing. It has been pointed out in particular that can implies non-actuality:
Joannis Velou
Acknowledgements I would like to thank two anonymow referees for helpful comments on an earlier version of thU paper.
JOANNIS VELOUDIS
Dtpartmnrt of Lmguistics Faculty of PhilMophy
Anstotlt Univmity T1msaIon ilti
540 o6 GTU« t-mail: �
[email protected] gr
received: 03.02.98
Final version received: 1 7.08.98
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
K.admon & Landman 1 993; see also Haspelmath 1 993: s4), the 'law-like' character of what these sentences 'say', the fact that 'free-choice' any is parallel to any in conditionals, as far as the licensing of any is concerned, etc. (c£ Kadmon & Landman 1 993), are explicable in terms of the key premise A of modus po� or, to what extent rhetorical questions like Cannot John solw {the most d!ffuult/any} problem? (= john can solve {the most difficult/any} problem1 and Can John solve {the simplest/any) problem? (= John cannot solve {the simplest/any} problem') are in general reducible to the key premises A and not B of modus ponms and modus tolkns, respectively (c£ our (1 a-b) above). In my epilogue, though, I prefer to comment on the conversational 'practice' I have pointed out above (see section J, in particular), rather than try to present a complete list of such 'relevant questions'. I have argued that, instead of being asserted directly, the conclusion of either modus is asserted indirectly in the case of the so-called 'quantifica tional' readings (as well as in the case of rhetorical questions). The question now is what is behind, and guarantees, this substitution of the indirect (i.e. of the key premises of either modus) for the direct (i.e. for the conclusion of the logical sequence): pragmatic probability (necessity, etc.), logical neces sity, or both? In incorporating a version of Fauconnier's 'pragmatic scales' (c£ the (pragmatic) implications in the first premises of either modus in the analyses above) into purely logical sequences of reasoning., i.e. modus ponens and modus tolkns, the explanation I have tried to defend seems to be confused: world experience (pragmatic probability, necessity, etc.) is, apparently, mixed up with mathematical logic Qogical necessity) in the context of my discussion. However, if we (naturally) assume that the thinking speaker is the 'subject' of both (i) the world experience and (ii) the calculations the two logical modi exemplify, this apparent discrepancy disappears. To put it differently: one cannot speak of a discrepancy here, unless one (unnaturally) assumes that (i) and (ii) above are not reducible to a common 'source', namely, the thinking speaker!
236 'Quantifymg' Superlatives and Homo Sapiens
NOTES See below; m my view, these environ menu are semantic in the first place; nevertheless, I w:tll continue to use fauconnier's characterization. 2 For a paralle l series of examples, see fauconnier {I979� 3 See alw the upward entailmg! downward-entailing environments in Ladusaw (1980� 4 The implication here may be pragmatic, not logtcal (c£ fauconnier's 'pragmatic
I
12
I3 I4
Is
'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
unphcations').
s '1f A, then B' IS, of course, equivalent With 'if not B, then not A'. We cannot, however, consider modUJ ponms as the only logical modw mvolved in the 'quanti6cational' readings of super latives for reasons that will be obviow below, where we extend our discussion to our data in (2� 6 any here (cf also sonu, sonubody below) IS meant to 5how that the hypothesis under corurderanon is apphcable to the any- venions of our data as well; see the diSCWSion m section 2. 7 some is transformed into any as 1t falls withm the scope of not (see section 3 below). 8 The asterisk on � is meant to indi cate that the universal quantificaoon it expresses cannot occur in the apodosis for logical re2SOns, as John can solve problern(s) a (, b, etc.)' can be true, John can solve every problem' bemg at the same time false. 9 More accurately, a proper subset. I will not, however, pursue this matter here. I o nobody is considered here as the any venion of Einstrin and Onassu in ( I c-d), ie. of animate (pragrnanc) superlatives occurring in neganve sentences. C£ the extended variant of {Ic) in the end of the precedmg section. I r C£ Haspdmath's intuinon about the scale-reversing property of negative, conditional, and interrogative sentences: '. . . it could perhaps turn out that
somehow the presence of negation is of central importance after all' (I 99 3: I I 2, fn.I9� C£ also not conVtna, not urtam, doubt, etc. in Fauconnier (1979: 29I ); they can all be understood as expressing different degrees of 'not'. AJ it is noted in Fauconruer (1 980: 6sf), 'a sentence like [I wonda if thu lmifo can cut twn tht most tmd" mtat) will easily implicate that the knife cannot cut any meat'; and in general, 'to quesnon the extreme propo smon on an �mplicanonal scale IS to express that no proposrnon on that scale is beheved'. The 'quannficational' venions of the examples in (2) w:tll be repeated here together With their any- counterparts. These intmtions are further encouraged by the fact that any of (za-c) can be substituted for, and be treated as, S,•s onginal contribution in the context described m secnon I above: S,: How about John? S,: I don't think John can/I wrll be surpn.sed ifJohn can I John is too stupid to solve the simplest problem. Taken together with 'If John can solve problem(s) a (, b, etc.), he can rolve {the simplest/some} problem', they can equally well lead to the Qogical) con clwion John cannot solve problem(s) a (, b, etc.)'. C£ Haspelrnath (1993: uo): 'However, questions with scalar-endpoint Indefi nites are somewhat odd, and make best sense as rhetorical questioiU. A quesnon hke Did you kar tht sl1gh�t no�.St? could hardly be meant as an information question, because it is very unlikely that the speaker should be mterested m 1nformation about an extreme value.' C£ also Fauconnier (I 980: 66): . . . so called "negative polarity items" could appear in questions and give rise to an expected negative answer or even carry declararive negative force (rhetorical
Ioannu
quesnon.sr And (ibid.): '. . . questioning at scale extremities favors negative irnphcarures; [Dul you lift a fin� whm I T!ffd� It?), taken hterally, entaili "I have no evidence tlut you dtd even the slightest thing when I needed it" and consequently has the same negative unplicatures [fit. oiDitted].'
Veloudis 237
16 For the exploitation of'rnirumal expres Sloru' in the remforccment of negatton, see Hopper & Traugott ( 199 3: 1 1 5 ff). 17 This aspect of induced quantification and tts impona.nce have been pomted out to f:luconnier by A. Lentin and mdependently by B. de Cornulier; see f:luconnier (1980: note 19).
R E FE RE NCES
study of indefirute pronouns', Ph.D. dissenation, Freie Universitat Berlin. Hopper, Paul J. & Traugott, Elisabeth d03S ( 199 3 ), Grammatrcallzatum, Cambridge Uruversity Press, Cambridge. Kadmon, Nrrit & Landman, Fred ( 1 99 3� 'Any', LmgwtstiCS and Plulosophy, 16, 3 5 3-42.2. Ladwaw, Willtam (1980), Polanty &nsitmty as Inhemtt &ope R.tlat1on.s, Garland, New Yor� Lakoff, Robm (r¢9), 'Some reasoru why there can't be any some-any rule', Langwag�. 45, 6o8- 1 5 . Palmer, Frank R. (1979), Modal1ty and th� Engltsh Modals, Longman, London. Sweetser, Eve (1990� From Etymology to Pragmatus: M�taphorical and Cultural Asp«ts of &mantiC Structur�. Cambndge Uruversity Press, Cambridge.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
f:luconruer, Gules (1975a), 'Pragmatic scales and logical structure', Lmgwtstrc Inqu1ry, 6, J 53-75· Fauconnier, Gilles (197 5b� 'Polanty and the scale principle', Chrcago Lmgwistrc Somty, 1 1, 1 8 8-99. f:luconnier, Gilles (1979� 'Irnplicatton reversal in a natural language', m F. Guenthner & S. J. Schrrudt (eds), Formal &mantles and PragmatiCS for Natural Languages, Retdel, Dordrecht, 289-301. Fauconmer, Gilles (1980), 'Pragmatic entail ment and quesnon.s', in John R. Searle, Ferenc Kiefer & Manfred Bterwisch (eds), Spmh Act 17uory and PragmatiCS, Syntw I.mgtuge Library, 1 o, Reidel, Dordrecht, 5 7-69 Haspelmath, Martin (199 3), 'A typologtcal
1998
The Semantics and Pragmatics of Presupposition N I C HOLA S A S HER University of To:as, Austin ALEX LA SCAR IDE S
Uni�ity of Edinbu rgh
Abstract this paper, we offer a novel analysis of presuppositioru, paying particular attention to the mteraction between the lmowledge resources that are required to interpret them. The analysis has two main fearures. first, we caprure an analogy between presuppositioru, anaphora and scope ambigwty (c£ van der Sandt 1992), by utilizing semantic under spectfication (c£ Reyle 1993� Second, resolving this underspecification requires reasorung about how the presupposttion is rhetoncally connected to the diSCourse context. Tlrn has several consequences. first, since pragmatic information plays a role m computrng the rhetorical relation, it also constrains the mterpretation of presuppositions. Our account therefore provides a formal framework for analysing problematic data, which requrre pragmatic reasoning. Second, bindmg presuppositions to the context vra rhetorical links replaces accommodatrng them, in the sense of adding them to the context (c£ Lewis 1979). The treaonent of presupposition is thus generalJ.Zed and integrated into the discourse update procedure. We formalize this approach in soRT (Asher 1 993; Lascarides & &her 1993), and demonstrate that it provides a rich framework for interpreting presuppositions, where semantic and pragmatic constraints are integrated.
I I N T R O D U CT I O N The interpretation of a presupposition typically depends on the context in which it is made. Consider, for instance, sentences {I) vs. (2), adapted &om van der Sandt ( I992); the presupposition triggered byJack's son (that Jack has a son) is implied by {I), but not by {2) .
{I) If baldness is hereditary, then Jack's son is bald (z) If Jack has a son, then Jack's son is bald. The challenge for a formal semantic theory of presuppositions is to capture contextual effects such as these in an adequate manner. In particular, such a theory must account for why the presupposition in (1) projects &om an embedded context, while the presupposition in (2) does not. This is a special case of the Projection Problem: If a compound sentence S is made up of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In
240 The Semantics and Pragmancs of Presupposition
constituent sentences sl ' . . . ' Sn , each with presuppositions P I . . . ' Pn . then what are the presuppositions of S? Many recent accounts of presupposition that offer solutions to the Projection Problem have exploited the dynamics in dynamic semantics (e.g. Beaver 1 996; Geurts 1 996; Heim 1982; van der Sandt 1 992). In these frameworks, assertional meaning is a relation between an input context (or information state) and an output context. Presuppositions impose tests on the input context, which researchers have analysed in two ways: either the context must satisfy the presuppositions of the clause being interpreted (e.g. Beaver 1996; Heim 1982) or the presuppositions are anaphoric (e.g. van der Sandt 1 992) and so must be bound to elements in the context. But clauses carrying presuppositions can be felicitous even when the context fails these tests (e.g. ( 1 )). A special purpose procedure known as accommodation is used to account for this (c£ Lewis 1 979): if the context fails the presupposition test, then the presupposition is accommodated or addtd to it, provided vanous constraints are met (e.g. the result must be satisfiable). Tills combination of test and accommodation determines the projection of a presupposition. For example, in ( 1 ), the antecedent produces a context which fails the test imposed by the presupposition in the consequent (either satisfaction or binding). So it is accommodated. Since it can be added to the context outside the scope of the conditional, it can project out from its embedding. In contrast, the antecedent in (2) ensures that the input context passes the presupposition test. So the presupposition is not accommodated, the input context is not changed, and the presupposition is not projected out from the conditional. Despite these successes, this approach has trouble with some simple pred1ctions. Compare the following two dialogues (3abc) and (3abd): '
a.
A: Did you hear about John?
b. B: No, what? c. A: He had an accident. A car hit him. d A: He had an accident. ??The car hit rum. The classic approach we just outlined would predict no difference between these two discourses and would find them both acceptable. But (3 abd) is unacceptable. As it stands it lacks discourse coherence, while (3abc) does not; the presupposition of t� car cannot be accommodated in (3abd). We will argue that the proper treatment of presuppositions in discourse, like a proper treatment of assertions, requires a notion of discourse coherence and must take into account the rhetorical function of both presupposed and asserted information. We will provide a formal account of presuppositions, which integrates constraints from compositional semantics and pragmatics in the required manner.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(3)
Nicholas Asher and Alex Lascarides
24 1
show that presupposed information is sensitive to rhetorical function and that the notion of accommodation should be replaced with a more constrained notion of discourse update. The third consequence concerns the compositional treatment of presupposition. Our approach affords that one could call a compositional treatment of presuppositions. The discourse semantics of soRT is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We will start by examining van der Sandt's theory of presupposition satisfaction, since he offers the most detailed proposal concerning accommodation. We will highlight some difficulties, and offer a new proposal which attempts to overcome them. We will adopt van der Sandt's view that presuppositions are anaphoric, but give it some new twists. First, like other anaphoric expressions (e.g. anaphoric pronouns), presuppositions have an underspuified semantic content. Interpreting them in context involves resolving the underspecification. The second distinctive feature is the way we resolve underspecification. We assume a formal model of discourse semantics known as soRT (e.g. Asher 1993; !..ascarides & Asher 1993), where semantic underspecification in a proposition is resolved by reasoning about the way that proposition rhetorically connects to the discourse context. Thus, interpreting presuppositions becomes a part of discourse update in SORT. This has three important consequences. The first concerns pragmatics. soRT provides an expliat formal account of how semantic and pragmatic information interact when computing a rhetorical link between a proposi tion and its discourse context. This interaction will define the interpretation of presuppositions, and thus provide a richer source of constraints on presuppositions than standard accounts. This account of presuppositions will exploit pragmatic information over and above the clausal irnplicatures of the kind used in Gazdar's (1979} theory of presuppositions. We'll argue in section 2 that going beyond these implicatures is necessary to account for some of the data. The second consequence of interpreting presuppositions is SORT concerns accommodation. In all prevtous dynamic theories of presupposi tion, accommodation amounts to adding, but not relating, the presupposed content to some accessible part of the context. This mechanism is peculiar to presuppositions; it does not feature in accounts of any other phenomena, including other anaphoric phenomena. In contrast, we model presupposi tions entirely in terms of the soRT discourse update procedure. We replace the notion that presuppositions are added to the discourse context with the notion that they are rhetorically linked to it. Given that the theory of rhetorical structure in soRT is used to model a wide range of linguistic phenomena when applied to assertions, it would be odd if presupposed information were to be entirely insensitive to rhetorical function. We will
242 The Semantics and Pragmatics of �upposttion compositional upon discourse structure: the meaning of a discourse is a function of the meaning of its parts and how they are related to each other. In soRT presuppositions, like assertions, generate underspecified but interpretable logical forms. The procedure for constructing the semantic representation of discourse takes these underspecified logical forms, resolves some of the underspecifications and relates them together by means of discourse relations representing their rhetorical function in the discourse.
So
presuppositions have a content that contributes to the content of the discourse as a whole. Indeed, presuppositions have no less a compositional
treatment than assertions. Our discourse-based approach affords a wider perspective on presuppo
amounts to an important special case, which applies to single sentence
discourses, of the more general 'discourse' problem: how do presuppositions triggered by elements of a multi-sentence discourse affect its structure and content? We aim to tackle this question here. And we claim that a rich
notion of discourse structure, which utilizes rhetorical relations, is needed. While we believe that our discourse based theory of presupposition is novel, we hasten to add that many authors on presupposition like Beaver (1996) and van der Sandt (1992) would agree with us that the treatment of presupposition must be integrated with a richer notion of discourse structure and discourse update than is available in standard dynamic semantics (e.g. K.amp & Reyle's DRT, Dynamic Predicate Logic or Update Semantics), because they believe that pragmatic information constrains the interpretation of presuppositions. We wish to extend their theories with this requisite notion of discourse structure.
2
VAN DER SANDT'S DYNAMIC A C C OUNT AND ITS PROBLEMS
Van der Sandt
(1992)
views presuppositions as anaphors with semantic
content. He develops this view within the framework of DRT (K.amp & Reyle
1993),
in order to exploit its constraints on anaphoric antecedents.
A presupposition can bind to an antecedent only if there is the same content in either an accessible part of the DRS which represents the discourse context, or an accessible part of the DRS which represents the current clause (i.e. the clause that introduced the presupposition trigger� In
(2),
for example, the antecedent of the conditional is accessible to the
consequent, and it contains the same content as the presupposition that's triggered there.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
sitions. Present dynamic accounts of presupposition have concentrated on phenomena like the Projection Problem. For us the Projection Problem
Nicholas Asher and Alex Lascarides 243
(I)
(2)
If baldness is hereditary, then Jack's son is bald. If Jack has a son, then Jack's son is bald.
So this
presupposition binds to it.
which can be paraphrased as
IfJack
intuitions.
This
provides a representation of (z)
has a son, then he is bald,
which matches
In contrast, the presupposition in (I) cannot be bound, because the context lacks the required content. Following Karttunen (I974) and Hetm
(I982),
(I992) resorts to accommodation: he context (c£ Stalnaker I974; Lewis I979).
van der Sandt
presupposition to the
adds the
Van der Sandt provides an algorithm which specifies how binding and supposed material is separated from the asserted material in the DRS which represents the current sentence (which may be complex in that it contains several clauses), in order to allow them to be processed differently. One handles the presupposed material first. If it can be bound in the manner specified above, then it is. Otherwise, it is added to an accessible site. One then adds the DRS which represents the asserted material of the current sentence to the DRS representing the previous sentences in the discourse, or some subDRS of it, via DRT's notion of update (note that one of these DRSS may have been modified with the addition of the presupposition). Essentially, DRT's notion of update is set union on both the discourse referents and the DRS conditions. When the contexts are structurally complex (i.e. contain subDRSS and complex conditions), different possibilities for accommodation arise. Van der Sandt distinguishes between global accommodation (as in (Ia)), in�diate accommodation (as in (Ib)) and local accommodation (as in (Ic)).
j,x ( a)
son(x)
b
has( j, x)
lm-tditary(b)
{ Ib )
=>
I �Li(x) l
b,j,x has( j, x) lm-tditary( b) son(x)
=?
I �Li(x) l
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
accommodation jointly model presupposition satisfaction. First, the pre
2« The Semanria and Pngtnatia of Prerupposition
( 1c)
b hn-tditary( b)
=>
x, j son(x) has( j, x) bald(x)
(4) Either John didn't solve the problem or else Mary realizes that the problem has been solved. The second disjunct presupposes that the problem has been solved. There's no suitable accessible antecedent, and so this information has to be accommodated. (4) is reminiscent of one discussed by van der Sandt (1992}.'
( s ) Either John has no donkey or his donkey is eating quietly in the stable.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
These capture the different possible inferences that presuppositions can give rise to. (1a) means: Jack has a son, and if baldness is hereditary, then Jack's son is bald. (1b) means: if baldness is hereditary and Jack has a son, then Jack's son is bald. (1c) means: if baldness is hereditary, then Jack has a son and he is bald. Only (1a) reflects the case where the presupposition projects from the embedding. Accommodation is subject to certain constraints: the result of the addition should be logically consistent and should not render any part of the asserted content uninformative. Furthermore, van der Sandt argues that if these constraints yield a choice as to where to accommodate the presupposition, then one prefers to add it to the most superordinate or highest DRS context in which the constraints are satisfied. So in ( 1 ), global accommodation (i.e. (�a)) is predicted. Van der Sandt's account of presupposition is compelling, because he offers a precise solution to the Projection Problem. A presupposition projects from an embedded context only if the above algorithm predicts that it is accommodated at a superordinate site in the DRS (e.g. (�a)). There are, however, a number of difficulties with the theory's predictions, some of which are particular to van der Sandt's formulation of accommodation, others of which are endemic to the semantic notion of accommodation itself-which we will take henceforth to mean the addition of presupposed information to the context. The first problem, which is particular to van der Sandt's account, is that local accommodation isn't predicted in certain cases when it should be. Consider (4) (modified from Beaver 1 997):
Nicholas Asher and Alex Lascarides .2.4 5
(6)
a.
A:.
The Problem Solving Group is given a problem each day, and the group leader Mary has to assign it to someone in her group. John is the best problem solver. But when he solves a problem, he always boasts about it. This annoys Mary, and so if she thinks that the day's problem is an unsolved one, she gives it to him, to test him. Otherwise, she gives it to someone else. b. B: John's being very quiet just now. Did she give him today's problem?
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The presupposition in ( s ) that John has a donkey cannot be accommodated globally in his theory, because this would render the first disjunct of ( s ) uninformative: assuming ¢ is the proposition that John has a donkey, then accommodating globally would produce a semantic representation of the form ¢ A (..., ¢ V 1/J), which is equivalent to ¢ A 1/J. So by appealing to the info�tiveness constraint we mentioned above, van der Sandt predicts that global accommodation is not possible for (s):• We cannot block global accommodation in (4) with the informative ness constraint, however. Adding the presupposition that someone solved the problem does not render the first disjunct, that john did not solve the problem. uninformative. Since this global accommodation also results in a consistent discourse, van der Sandt's representation of (4) amounts to: the problem has been solved, and either John did not solve it or Mary realizes that it has been solved. As Beaver points out, this is contrary to intuitions, which favor local accommodation: either John did not solve the problem. or it has been solved and Mary realizes it has been solved. To capture the intuitive semantics of sentence (1), Van der Sandt relies on the preference for global accommodation. But in (4) this preference causes problems, because local accommodation should be preferred, and yet the constraints on accommodation are satisfied at the superordinate site. We shall see in the course of this section that this puzzle arises because of a general observation about presuppositions that DRT alone cannot capture: the preferred site for accommodation depends on a variety of pragmatic factors (c£ Beaver 1996) that are not represented in DRSS. Nevertheless, (4) is in fact a bit odd unless it is uttered in a particular context; we need to know which problem is being talked about. Van der Sandt's theory fails to capture this as well, because it predicts that the presupposition that there is a problem, which is triggered by tk problem, can be globally accommodated. But discourse contexts which improve the acceptability of (4) strongly suggest that the way presuppositions are satisfied depends on the rhetorical structure of the discourse as a whole. Dialogue (6) illustrates this:
246 The Semantics and Pragmatics of Presuppos1tion
(6)
c.
A:
Well I'm not sure she did. Either John didn't solve the problem or else Mary realized that the problem's been solved.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
B's question identifies the problem being talked about, and this improves the acceptability of (4). In addition, the rhetorical links between the (6a), (6b) and (6c) help one interpret the presupposition that the problem has been solved. But our more basic problem remains: to explain the local accommoda tion of the presupposition of realize in (6c). We could with Geurts (1995) integrate in Van der Sandt's theory further constraints on accommodation such as those in Gazdar's theory, which are based on Grice's notion of conversational implicature. Gazdar's theory predicts local accommodation in any context where by uttering the sentence, the speaker implicates that the presupposition is not taken for granted (i.e. the speaker implicates that he believes that it is possible that the presupposition is false). Would such a constraint block global accommodation in (6)? B's question about John's behavior in (6b) serves as an elaboration to the generalization about who gets problems in (6a), and so the answer to the question hangs on whether Mary believes that the problem is solved or not. A implies by his utterance (6c) that he doesn't know whether Mary gave John the problem. But what implicatures does this raise? Given what A said in (6a), A and B would know the answer to the question as to whether John got today's problem, if they knew if Mary believed it was solved. So this context implicates that A and B don't know if Mary believes the problems been solved. However, the context does not implicate that A and B themselves believe the presupposi tion-that the problem has been solved-is in doubt; the context is mute on this issue. So Gazdar's constraints do not predict local accommodation in this case. Something more is needed to explain why local accommodation occurs here. To predict local accommodation in (6), we analyze presuppositions in terms of the rhetorical structure of the discourse context. Rhetorical structure is something about which DRT has little to say. We claim that rhetorical structure explains why local accommodation is preferred in some cases where global accommodation would be informative and consistent, and furthermore would meet Gazdar's constraints on how Gricean-style conversational implicatures determine global vs. local accommodation. The place where presuppositions get accommodated depends on the rhetorical links between propositions in discourse as well as their content. In �cular, presuppositions are interpreted so that the rhetorical links are as strong as possible. We will return to the detailed analysis of (4) in section 5·3 and demonstrate that the Contrast relation between the disjuncts in (4) is at its strongest in (6) if the presupposition is accommodated locally.
Nicholas Asher and Alex Uscarides
247
The difference in acceptability between (Je-d) also demonstrates that the constraints of informativeness and consistency are too weak: (3)
a. A: b. B: c. A: d. A:
Did you hear about John? No, what? He had an accident. A car hit him. He had an accident. ??The car hit him.
(7) ?? I don't know whether the Pope has measles. But every Catholic realizes the Pope has measles. (8) I don't know whether the Pope has measles. But for every person, if he's a Catholic and the Pope has measles, then he realizes the Pope has measles. Van der Sandt's theory predicts that presupposition that the Pope has measles is globally accommodated in (8). But intuitively, global accommo dation shouldn't be possible, because a statement akin to Moore's paradox ( 1 9 1 2) would be asserted ('Ik Pope has measks but I don't know that tk P� has measks). Van der Sandt's constraints on accommodation do not block this. But even if they did, the resulting account would still allow intermediate accommodation. But this incorrectly predicts that (7) is felicitous. In contrast, the presupposition in (8) is bound to this intermediate position. But the difference in acceptability between (7) and (8) demon strates that an explicit linguistic antecedent is required, and accommo dation at the same site as this antecedent is odd. Van der Sandt fails to capture this. There are certain presupposition triggers, such as too, which are like (7) in that their presuppositions require a linguistic antecedent They pose similar challenges. Consider (9}: (9) John lived in New York too. Various people have supposed that (9) presupposes that there is someone other than John who lived in New York. This content is already common knowledge to most speakers. However, (9) is odd in a context where the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
While (3c) is perfectly felicitous, (3d) is not Most theories would predict that the presupposition in (3d) is accommodated globally (i.e. added to the main context), because the result is informative and consistent. So these constraints predict coherence where there is none. These constraints on accommodation also cause difficulties for examples where the presupposition seems to require an explicit linguistic antecedent to be felicitous. Consider, for instance, the contrast between (7) and (8), taken from Beaver (1997):
248 The Semantics and Pngmatics of Presupposition
presupposition or something simibr isn't already explicitly introduced into the discourse context. In fact, we think that the received wisdom about the presuppositional content of too is wrong: (9) is perfectly felicitous if uttered in a context in which other places John has lived have already been mentioned. The presupposition of too is rather that it requires that there be some proposition in the context that hears the rhetorical relation Parallel to the content of the sentence in which too occurs.3 In any case the presupposition triggered by too must be introduced explicitly into the discourse context, and cannot be accommodated. Another difficulty for this theory is that it on occasion fails to specify the appropriate content of the presupposition itsel£ Beaver (I997) argues that the following minimal pairs show that the content of the presupposition cannot depend simply on constraints like consistency and informativity: ·
If David wrote the article, then the knowledge that no good logician was involved will confound the editors. (I I) If David wrote the article, then the knowledge that David IS a computer program running on a PC will confound the editors.
The presupposition of (10) is (12), whereas for ( u ) it is (I J): ( 12) If David wrote the article, then no good logician was involved. (I J) David is a computer program running on a PC. The factive nominal knowledge triggers the presupposition in both (10) and (I I ). But whether conditional presuppositions arise or not cannot be a structural matter, since the same structure gives rise to a conditional presupposition in (10) but not in ( u ). So the standard theory of accommodation cannot account for the intuitions about these examples; it does not explain the presence of conditional presuppositions in some cases but not in others. Beaver (I997) argues that presupposition satisfaction must also depend on some world knowledge dependent notion of plausibility, but he too does not give an explicit or detailed account that would make the correct predictions in ( 10) and ( 1 1 ). Van der Sandt's conjecture that presuppositions are anaphors is a compelling idea. But it has shortcomings because accommodation is not constrained enough. Background knowledge such as information about the domain can block accommodation (Lewis I 979). and the rhetorical function of the utterance in the text can influence the projection of its presupposi tions, as shown with (6). Though Van der Sandt and Beaver would both acknowledge that pragmatic constraints on accommodation are important, these constraints have not been integrated into the formal theory of presuppositions or made sufficiently precise to have any predictive force. For instance, Van der Sandt's (I992) explicit theory ignores them.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(Io)
N1cholas Asher and Alex Uscarides 249 Consequently, presuppositions are sometimes accommodated in the wrong to produce the wrong readings, or they are accommodated when they should not be. We will approach these problems by tackling presupposi DRS
tions in a framework of discourse representation that formalizes the semantics pragmatics interface.
3 T H E B A S I C P I CT U RE Our framework of discourse interpretation, soRT, extends DRT in two fundamental ways. First, discourse contexts are represented as recursive,
representing the rhetorical functions of these discourse constituents in the context. Second, these discourse relations can affect the content of the clauses they relate, and hence of the discourse in general. To build such representations of contexts, SORT proceeds incrementally, interpreting each bit of new information as yielding a change in the discourse context. Like other dynamic theories of interpretation (e.g. and Groenendijk & Stokhof's 1 99 1
)
DPL ,
SORT
DRT
must specify how new
information can change an existing discourse context-i.e. it must give an account of the context change potential (ccP) of an utterance.
SORT
does so
as follows. First, one uses the grammar to build up compositionally the oRSs for each clause (c£ Muskens 1 995; Fernando 1 994; Asher 1993 ). The soRS is
then built from these oass dynamically. In the simple cases examined here and in Lascarides & Asher ( 1 99 3), we update clause by clause.• When updating the SDRS built so far with a new DRS, one uses the glue logic to decide where to attach
this DRS and to infer one or more rhetorical relations
to attach it. This last function of the glue logic was spelled out in Lascarides
& Asher ( 1993 ) and elsewhere in what we have called DICE (Discourse in Commonsense Entailment), but there the glue logic was restricted to this
task. Hence in this paper we will speak more generically and more generally
of a glue logic. The rhetorical relations may trigger modifications to the
DRSS
that were
produced by the grammar. For example, in ( 1 4), the glue logic determines
that Narration connects the clauses. Spatia-temporal constraints on Narration
are then used to add to the representation of ( 1 4) that the boxcar is in Dansville:
( 1 4)
a.
John took an engine from Avon to Dansville.
b. He picked up a boxcar, c.
and took it to Broxburn.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
relational structures (soass), involving oass representing the contents of clauses, and discourse relations like Parallel, Narration and Background
250
The Semantics and Pragmatia of Presupposition
{ I s) The King of France is bald. We take a different view. {I S) means: that there is a King of France is Background to his being bald. Much of this paper focuses on spelling out this idea in detail Our 'rhetorical' approach to presuppositions has at least three advantages. First, as we mentioned in section I, loosening binding in this way eliminates the need for semantic accommodation altogether, where by semantic accommodation we mean that the presupposition is added to the context, but not related to any element of it Alternatively, accommodation can be viewed as constrained by much more than informativeness and consistency in this picture, because it is licensed only when the presupposition can be bound to the context with a rhetorical relation (c£ Lascarides & Oberlander I99J).s Either way, the accommodation mechanism where information is simply added to the context is removed. This is a very desirable feature of the theory, since this 'addition' mechanism is only used to account for presuppositions. We are replacing it with a general procedure of inter pretation. which works for assertions too and which is used to model a wide range of linguistic phenomena, particularly semantic ambiguity resolution. The second advantage of this approach is that it records the influence of pragmatic information on presupposition satisfaction. Pragmatic constraints on computing rhetorical relations, which are provided by the glue logic
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The rhetorical relations, their semantic effects and the new information are now integrated together with the given SDRS representing the context to form a new soRS. Thus, SDRS construction proceeds dynamically, and each given SDRS and piece of new information produces a new SDRS, which can serve as a representation of a new discourse context Suppose one were to analyse presuppositions from the perspective of this kind of discourse semantics. Then standard binding and accommodation both seem unsatisfactory. First, standard binding is too restrictive: there are many rhetorical relations that bind propositions together-not just identity, but also relations like Paralld and Background. Binding presuppositions only with identity wouldn't reflect these alternatives. Second, accommodation where it is viewed as addition alone-is unsatisfactory, because in soRT the representation of discourse is well defined or coherent only if each DRS is related to another with a rhetorical relation. There is no scope for just adding a proposition to the context; it has to be related to it as well. This perspective leads us to loosen van der Sandt's notion of binding. Instead of a presupposition binding with identity to an accessible site in the context, it will bind with a rhetorical relation. For example, in {I s), van der Sandt would accommodate the presupposition that there is a King of France, because binding isn't possible.
Nicholas Asher
and Alex I...ascandes 25 1
(Lascarides & Asher 1 993), provide a much more constrained mechanism for dealing with presupposition than informativeness and consistency alone. As a result this theory will overcome some of the problems discussed in section 2. Third, it maintains a close relationship between presupposition satisfaction and discourse coherence. The presupposition failure in (3a, b, d) will be predicted by the fact that the presupposition that there is a car cannot be rhetorically bound to the propositions in the context. This is roughly analogous to the discourses given in (x 6a-c) being incoherent; we cannot attach the presupposition anywhere in the context and get a coherent discourse.
(3 )
a.
car.
.
Our account of presupposition will buttress and develop van der Sandt's view that presupposition satisfaction and discourse coherence are closely related, and we will exploit soRT's rich notion of discourse coherence to do
this. In
soRT, assertions are linked to the context by computing a rhetorical relation. We claim that presuppositions are handled this way too. But many linguists have argued that assertions and presuppositions are different. Presuppositions project from embeddings and assertions don't. And Clark ( 1977) has intuitively described presupposed information as givtn, and asserted information as �.6 We need to account for this intuitive difference. We capture the difference between presuppositions and assertions in three ways. First, the rules for updating discourse ensure that the rhetorical role of presuppositions is in general different &om that for asse rted information. To capture the intuition that by and large, presuppositions are given, the rules for computing rhetorical relations ensure that pre suppositions in general bind with Background (i.e. the presupposition provides background or sets the stage for the main story line). For example,
the representation of ( 1 s ) will include: (a) the proposition a that x is bald, (b) the proposition {3 that there is a King of France and that is x/ and (c) the condition Background(a , /3) (so the presupposition provides background to the main point, which is that x is bald� Presuppositions may also bind with Dtf-Co�qucut in some cases. This relation means that the presupposition is a defeasible consequence of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
A:. Did you hear about John? b. B: No, what? d. A:. He had an accident. ?'?The car hit him. (16) a. A:. Did you hear about John? ??There was a b. B: No, what? A:. ??There was a car. c. A:. John had an accident. ??There was a car
252 The Semantics and Pragmatics of Presupposition proposition it attaches to. If the consequence relation is idempotent (a standard assumption), then binding in van der Sandt's sense turns out to be a species of Def-Consequena. So binding in van der Sandt's sense will in general correspond in this theory to a
case
where the presupposition is
Def-Co�qumu (because it is defeasibly entailed by the attaches to) and Background (because in our theory pre
related by both proposition it
suppositions usually bind this way). Presuppositions can be attached with relations other than Background and Def-Consequence, but in general tlus arises because the presupposition trigger itself specifies some other rhetorical relation. For example, the presupposition trigger too entails Parallel, and because entails Explanation. In contrast to presuppositions,
Background-e.g. Narration, Result, because. Further Def-Consequenu, unless it is explicitly
Explanation-even in the absence of words like too and
more, assertions do not attach with signalled by a discourse particle. A second difference between presuppositions and asserted information is that there is an important distinction between their compositional semantics. The grammar produces a representation of presupposed infor mation which is always incomplete or umkrspecified. This is the technical way of ensuring that presuppositions are always anaphoric, and must be
bound. Because of this, asserted information and presupposed information behave differently at the start of a discourse. Asserted information is simply introduced into the null context. But this is not possible for presupposed information. It is
always
bound to some antecedent with a rhetorical
relation, even if that antecedent is the asserted information in the clause
which introduced the presupposition-as in e.g. (I 5).
Finally, there will be differenC:es in the preferences for which part of the
discourse structure presuppositions vs. assertions attach to. Presuppositions are freer in their attachment possibilities than assertions; those clausal
constituents of the assertion that are combined with logical operators have their attachment sites determined by the grammar, whereas presuppositions
are always free in principle to attach outside the SDRS constituent in which they were generated. The preferences for attachment predict that pre suppositions
can
project from embeddings in a way that assertions cannot.
The underspecification in presuppositions is used to record the influence of pragmatics on presupposition satisfaction. A distinguishing feature of SDRT
is that computing a rhetorical connection between propositions in the
glue logic can result in those propositions receiving additional content. Update in SDRT is defined so that if this additional content can resolve underspecified conditions that arose in the grammar, then it does.
This
captures the intuition that when processing discourse, people are expected
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
asserted information may be attached to the discourse context with a variety of rhetorical relations other than
Ntcholas &her and Alex Lascarides 2 5 3
4
PRESU PP O S I T I O NS I N M O RE DETAIL
Having given the basic picture, we will now turn to the details of the theory. First, we will give an overview of the soRT compositional semantics of presupposition triggers. Processing the presupposition relative to the context amounts to resolving its underspecified elements, and this is handled via the soRT update function. We will illustrate how this works, by analysing some simple sentences (i.e. sentences that do not contain logical operators such as if or not). We will then examine the interpretation of presuppositions in more complex sentences (e.g. (1o) and ( u )), and compare our treatment to that of other dynamic accounts, such as Beaver's (1996) and van der Sandt's (1992). Using SDRT makes comparisons and reformulations of van der Sandt's DRT-based theory of presuppositions relatively straightforward. We will demonstrate that at least some of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
to and often do fill in gaps in what's been made lmguistically explicit. This feature of soRT provides a formal model for how pragmatics (wluch influences the rhetorical relation) affects semantics (i.e. the resolution of underspecification). We will use tlus feature to record the influence of pragmatics on the semantics of presuppositions. The grammar typically introduces at least two underspecified elements in a presupposition: one is the rhetorical relation used to bind the presupposed mformation, and the other is the other term of this relation, which is known as the attachment point. The latter underspecificatton ensures that the structural position in the discourse context in wluch the presupposed information is entered is not determined by the syntactic position of the presupposition trigger that generates it, because the attachment point that is pan of the representation of the discourse context could be a proposition that was introduced by an earlier sentence (this 1s explained in more detail in section 4-2). This captures the intuition that that presuppositions may project from the embedded contexts m which they are introduced. Underspecification for presuppositions actually comes in a vanety of types, which depend on the presupposition trigger. We wtll examine some of these in section 4-1. In all cases, however, in processing a presupposition one must resolve the underspecified elements with respect to the discourse context. So presupposition projection, binding and accommodation all occur as a byproduct of the SDRT update procedure. They occur by computing how the presupposition rhetorically connects to the context. Since tlus update procedure is determined by the pragmatic mformation spectfied in the glue logic, presuppositions are mfluenced by pragmatics as well as compositional semantics.8
254
The Semantics
and Pragmattcs of Presupposition
problems concerning conditional presuppositions can be overcome, because of the role of pragmatics in soRT. Finally, we will examine presuppositions in multi-sentence discourse, and then return to some cases that one might think are difficult to handle in a proposal where semantic accommodation is excluded. 4. 1
Semantics of clauses
( I 7)
I:, K,l
Updating the discourse with information such as that in (17) will involve computing a rhetorical relation R between 1f' and some accessible speech act discourse referent 1f' 1 in the discourse context Computing this rhetorical relation may include resolving underspecified conditions that are in K1r, as we shall see in section 4-2. Note that R's arguments are the labels, and not the DRSs themselves. We explain why in section 4-2. A presupposition that is introduced by a clause can receive different scope &om the assertioiL For example, in (I), the presupposed information has wider scope than the conditional, but the asserted information does not9 In section 2, we showed that the relative scopes of the asserted and presupposed content of the clause is not syntactically determined; and so, in line with the general strategy for representing ambiguity, the grammar
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We first examine some important aspects of the compositional semantics of the clauses. Following Reyle (1993) and Asher & Fernando (1997), the grammar utilizes underspecification to represent semantic ambiguity. The grammar produces a set of labelled pieces of information and a set of constraints about how these labels can combme. Semantic ambiguity occurs when the constraints underdetennine the combination of these labels. Since the combination of labels is underspecified, so is the representation of the meaning of the clause. Resolving the relative scopes of the labels is then done by updating the discourse context with this (underspecified) meaning. Discourse update in SORT is specifically designed to resolve underspecified elements when it calL To avoid clutter in notation, semantic information that must be grouped together according to syntax receives just one label. Moreover, as in HPSG (Pollard & Sag 1 994), these labels may appear in conditions. A special case of this labelling procedure is the following: if the grammar produces the DRS Kn for some clause, then the SDRS for the discourse will include (17) where the label 1f' is a discourse referent, which we call a 'speech act discourse referent' since it labels a DRS (we will always assume a notational convention that 1f' labels Kn):
Nicholas
Asher and Alex Lascandes 2 5 5
groups asserted information under one label, and the presupposed informa tion under another. Consequently, the grammar produces an SDRS like that in ( 1 7) for the asserted content of a clause, and it will produce another SDRS ' ' with speech act discourse referent rr , and condition rr : K1r' for the presupposed content of this clause, where K1r' will be the DRS discourse
constituent that represents the presupposttion (e.g.Jack has a son for { I )). The grammar does not determine the scope of 7r and 71" 1• Rather, all of these
soRSs must be attached to the representation of the discourse achieved so far through the soRT discourse update procedure: that is, they must be attached to the context with a rhetorical relation. Rhetorical relations produce ' hierarchical structures for the labels 1r, rr etc, since some rhetorical relations
way and providing them with different labels, the grammar distinguishes between presuppositions and asserttons in one other important respect. Presuppositions are exphcitly encoded as anaphoric, since the grammar invokes for each presupposition an underspecified attachment point and underspeetfied rhetorical relation. These are gtven respectively by and
R = ? in the , 7r , U
(r8)
7r
u
=
?
canorucal representation of presuppositions, given m (I 8):
' . K' •
R(u,
7r
rr
'
)
R=? u=? This distinguishes this treatment of presuppositions from van der Sandt's, because this underspecification means that the presuppositions must be rhetorically bound to the context, rather than added.
always
The Projection Problem is challenging, because the scope of presupposed information is contextually rather than grammatically determined. Our semantic representation of presuppositions allows for this through the use is determined by resolving u
=
=
u ?. The scope of a presupposition ?. Sine� this is done via the discourse update
of the underspecified attachment site
procedure, and hence via the glue logic, pragmatics will influence the scope of presuppositions. Moreover, for presupposition triggers that are intro duced in an embedded context such as a conditional, soRT allows the presupposition to access attachment points that are outside the scope of the embedding. If discourse update resolves the antecedent attachment point
u
to one of these, then the presupposition has essentially projected out from
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
are subordinating and some are coordinating. So this discourse structure will resolve the relative scope of the asserted and presupposed content. As well as separatmg asserted content and presupposed content in this
256 The
Semantics and Pragmatics of Presupposltlon
its embedded context. This contrasts with the assertions in the conditional, which cannot project in this way. To illustrate the ideas, let us examine the compositional semanttcs of some example clauses that contain presupposition triggers. Presuppositions of abstract type-i.e. propositional or factual presuppositions-are triggered by
stop/start verbs, rtgret, realiZe, and
expressions such as wh-questions, focus, comparatives,
mana� and
succeed
verbs, factive verbs like know,
presuppos1tional adverbs like
regret. containing regret
agam.
Let us consider, for instance, a factive
verb like
structures, and therefore act as guides for putting the SDRS together; they will disappear after SDRS update. This will involve binding
11', 11' 1
etc to available
referents with rhetorical relations. Note also that the DRS representingjohn is
sick forms part of the representation of the asserted content, and also part of the representation of the presupposition. We use the DRS in a rather than the label because the intensional operator works over a DRS and not a term.
11' 1 , R, u 11' 11' :
11'
s regret(s,
p:
'
:
[2]
R(u, 11' 1 ) R=?
[ I ] , 11 [2] )
u=? (2o) John regretted that he was sick. 11'
11' 1 , R u j, s", t"
j, s, t, n I
11':
regret s, ),· II
sick(s', j ) -<
n
11'' :
'
p:
hold.s(s', t') t'
hold.s(s, t ) t -< n
\
s', t'
I
sick(s" , j ) hold.s(s'' ' t")
t" R(u, 11' 1 ) R=? u=?
-<
n
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The grammar will produce two SDRSs for a sentence as shown m ( I 9): a is the asserted information, and p is the presupposed information; [ I ] is ftlled in with the subject NP and [2) with the sentential complement of regret. So the grammar produces the representation (2o' ) for sentence (2o). a and p below each label bits of structure that are produced by the grammar, but they are not discourse referents themselves. They merely provide ways of talking about the whole
Ntchol:as Asher and Alex Lascarides 257 The SDRS p contains the underspecified conditions R = ? and u = ? , which must be resolved by updating the context With p, i.e. by computing a rhetorical
tr 1 and an antecedent site in the context. Resolving attachment site u will determine the scope of the presupposition. relanon between
the
As we have mentioned before, some cue phrases are presupposition triggers which actually specify the rhetorical relation (e.g.
too specifies
the The presupposed information tnggered by too and for the
SDRS
Such cue phrases presuppose that an antecedent of an appropriate sort exists. This is represented by the anaphoric cond1tton u = ?. Other cue phrases may introduce different discourse relations-e.g. because introduces
Explanation; although mtroduces Contrast. All of these will generate the same sort of conditions as (2 1 ). Such conditions impose constraints on what u can be, because each rhetorical relation specifies constraints on the relationship
between the semantic content of the propositions it connects. These constraints are descnbed in section 4-2.
Let us consider now the presuppositions of definites. In line with Chierchia
(I 995) and Asher & Lascarides (in press), we assume that defirutes introduce an underspecified bridging relation, which connects the object denoted by the definite to some antecedent in the context. In Asher & Lascandes (in press), we
demonstrated that this bridging relanon could he computed as a byproduct of soRT Update, which is how we also treat presuppositions. ' 0 Definites introduce not only underspecified 'bridging' relations at the 'micro-structure' level to connect the denoted objects together, but in line
with other presupposition triggers, they also introduce underspecified 'rhetorical' relations at the 'macro-structure' level of the from
R=?
and
u = ?,
which connect the propositions together:
rr', R, v x, u, B, t, t [t](x) B(�, x, u) 1f ': holds(�. t), B=? u=? R(v, rr ' ) R=? v=?
R( u, 1r ) ,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
relation to be Parallel). also is represented in (2 I ), where [I] is to be filled in by a label over which too (or also) 1s assigned scope by the grammar: u ( 2 I ) Parallel( u , [I]) u=?
2 5 8 The
Semantics
and
Pragmancs of Presupposltlon
The definite NP furnishes argument [ I), while the VP contributes to argument [2) (depending on whether the definite occurs in subject or object position). As an illustrative example, sentence (23) has the compositional semantics given in (23 1 ): (2 3) The man walked 1r 1 , R, v x, u, t I ' t I ' B e, t, n 7r :
walk(e, x) holsd(e , t) t -< n
7r l :
p:
man(x) B(e 1 , x, u) holds( e 1 , t 1 )
B=? u=? R(v, 1r 1 ) R=? v= ?
The presupposltlon must be bound to the content via a rhetorical relation. In addition, the man denoted by the definite must be (bridging) related to an antecedent object (so (23) couldn't be uttered in a null context). Our treatment of definites extends to quantified NPS, which also convey presuppositions (c£ von Fintel I 994). The restrictor includes underspecified bridgmg conditions, which require the objects denoted by the NP to be related to antecedent objects. This is intended to capture the bridging that occurs in discourses such as (24): the strikers mentioned in (24b) played in the football match mentioned in (24a): (24)
a.
b.
I went to the Scotland vs. Latvia football match yesterday. Every striker attempted to score a goal.
We reflect this in the semantics of quantified NPS by invoking under specified conditions of the type found in definites, which are placed in the restrictor of the DRS duplex condition. For example, the compositional semantics of evtry is specified in (25), where [I] is derived in the grammar from the NP that contains the determiner every, and [2] is derived from whatever the NP combines with to form a clause:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
1f'
Nicholas
1r R, u B, x , �, t , y [ 1] {x) B ( e, x, y) 7r : holds ( e, t) B=?
1f' ::::}
Asher and Alex Lascandes
259
"
" 1f' :
]2] (x)
In (24) for example, [ 1 ] {x) is replaced with stnker(x); [2] {x) is replaced with the VP at�mpted to score a goal, and B, y and u will be computed through soRT Update, so that the quantified NP connects with (24a) in the right way. That is, B will be resolved to play, y identified with a discourse referent denoting the football match, u identified with the label for the DRS which represents (24a), and R resolved to Background {as we will see shortly). Note that unlike definites, the presupposition associated with every is restncted in scope to being in the restrictor; it isn't free to take wider (or in DRT, 'higher') scope. This restriction ensures that evtry stnlw- V-�d does not entail that there are any strikers. Of course, these observations hold provided that the discourse referents within the constituent labelled by 1r get the appropriate quantificational force from ::::} . This is assured if R is a veridical relation (c£ section 4.3), because then any embedding of the right hand side of the conditional must perforce be a proper embedding of the 1r constituent. These examples illustrate the variety of underspecifications that pre supposition triggers generate. The update task involves resolving the underspecified conditions where possible. This includes scope resolution. So using underspecification to represent the compositional semantics of presuppositions allows us to assimilate presupposition projection to a more general problem-that of assigning a scope to various informational units. We now turn to the problem of defining discourse update, and the way discourse update resolves underspecification. 4.2
From clauses
to
discourse
The soRSs produced by the grammar must be dynamically integrated together, into a representation of the semantics of the discourse. In soRT,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
y=? R(1r, u) u=? R=?
1 l
26o The Se!ll2lti l cs
and Pragmatics of Presupposiuon
one updates the discourse context T that has been built so far wtth some new information /3, by attaching /3 �o a with a rhetorical relation, where is an available attachment site (i.e. speech act discourse referent) in T.
a
But which a's are available attachment sites? Rhetorical relations proVIde a richer discourse structure than ORT alone gives: Elaboratton, .JJ. (a .JJ. /3 can be glossed as 'a is a topic for /3') and Explanation are subordinating relations; the rest are not Therefore, the notion of accessibility in soRT (which is termed avatlability to distinguish it from
)
ORT
is more constrained than in
ORT.
Available labels m an soRS are those on its right frontier: that is, the previously attached label, and ones which elaborate or explain it, or are a topic to it So, for example, new information could attach to 1r 1 in (26) but not to 1r:
1r : K1r , 1r 1 : K1r' Narration(1r, 1r 1 )
And hence the antecedents to anaphora must be picked up from K1r', but not from K1r. In general, tlus offers a much smaller set of possibilities than
ORT
itself would offer. This is one way m which interpreting
presuppositions will be more constrained
m
SORT
than m
ORT.
To compute which rhetorical relation to use, it is essential to use a nonrnonotonic, 'glue' log1c, as we have argued in many places (e.g. Lascarides & Asher 1993). It is also essential to make use of nonmonotinicity to place constraints on where. to attach new information, especially when
this information to be attached is presupposed. The glue logic is expressed in a quantifier-free language augmented with a weak conditional con nective: A > B means 'If A, then normally B'. This glue logic for constructing logical forms for discourse is quite simple, even though the content of the discourse may be very complex. To build logical forms, the glue logic must have access to the contents of the discourse, but to keep the logic s1mple it has only limited access to the fuJI contents of the propositions-namely the soRSs. More specifically, we suppose that the glue logic has access to the form of the soRSs, but not what they denote (Asher & Fernando
1997).
To this end, we suppose that there is an information
transfer function p. from the language of information content to the glue language, which takes conditions inside a given soRS and turns them into predicates on that soRS's label. So propositional variable
a,
J.L(Ka) (a), where
¢ is a predicate on the
and 1t corresponds to an soRS condition in
Ko. .
The glue language uses various nonlogical predicates in describing rules which constrain where new information should attach. and what rhetorical relation should be used to attach it We exploit the labels of soRSs in the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(26)
Nicholas
Asher
and Alex �...ascarides 26 I
glue language, as arguments to our nonlogical predicates {Asher 1 996). One of these is ( r, a, {3), which can be glossed as 'the constituent labelled by {3 is to be attached to the constituent labelled by a with a rhetorical relation,
where a is an available label in the SDRS labelled T that's built so far'. There are also: relation symbols for the discourse relations, whose arguments are the labels (e.g. Explanation( a, /3)); predicates on labels, which allow us to specify whether the main eventuality described in the SDRS with that label is
an event or a state (e.g. �ffl t ( a ) and state( a )); and predicates on the labels, which describe conditions within the SDRSS (e.g. overlap(�, e' ) ( a)). For
stative, then by Background, the rhetorical relation is normally • •
Narration: {( r,
a,
Background: (( T,
Background:
{3) 1\ �ent ( a ) 1\ evmt(/3)) > Narration( a, f3) /3) 1\state(/3)) > Background( a, /3)
a,
For example, Background ensures that Background relates the contents of the first and second sentences in
{27)
(27):
Jack has a son. He is bald.
The language also includes a predicate l and a function symbol Update. Upda�( r, a, /3) returns the SDRS constructed by the glue logic given an input SDRS T, an attachment point a and T, and a new labelled SDRS
Upda�( r, a, /3)
means that updating T with
{3. l
{3 by attaching it to a produces ?, and every
a well-defined SDRS (i.e. there are no unresolved conditions x
=
constituent is rhetorically attached to another). We will define SDRS Update
precisely in a minute and use this to specify constraints on which site a new material
{3 should
attach to. Once again, it is important to stress that while
the glue logic has access to the structure of soRSs-e.g. the conditions within them-and indeed reasons about SDRSS, it does not have access to their denotations. The glue logic reasons about logical forms of discourses, not (except in the very limited way we have mentioned) about their content Because the majority of the glue logic axioms feature the connective >, and > does not support a modus ponens-hke rule, computing rhetorical relations is nonmonotonic. The nonmonotonic notion of validity
�
that
underlies the SDRT glue logic has several desirable properties, which we have discussed in detail
elsewhere (Lascarides & Asher
that are relevant for our purposes. First, Ponens
�
1993).
There are three
validates Defeasible Modus
(oMP): when the default laws whose antecedents are verified all have
consequents that are consistent with the monotonic information and with each other, then all the consequents are nonmonotonically inferred. Second,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
example, the Narrat 1on axiom of our glue logic below states: If the constituent labelled by {3 is to be attached to the constituent labelled by a with a rhetorical relation, and those constituents describe events, then normally, the rhetorical relation is Narration. On the other hand, if {3 is
262
The Semantio and Pragmatio of Presupposition
� validates Specificity: when conflicting default rules apply, the consequent
(if there is one) is inferred. And fmally, � is ¢, then r u 1/J � ¢, where 1/J is logically independent
of the most specific default rule robust m that if r � information.
The glue logic also contains (monotonic) axioms concerning the semantic effects of rhetorical relations. For example, Axiom on Narration stipulates that any items that are connected with Narration must describe eventualities where the Hrst precedes the second (ta -< ef3 means a's main eventuality precedes {J's, and it's a gloss for a formula of the propositional glue language). The Axiom on Background stipulates that the eventualities overlap. And the Spat ial Consequence of Narrat ion is derived from
• • •
Axiom o n Narrat�on: Narration( a, /3) � ea -< e{3 AXlom on Background: Background( a , /3) � overlap( ea , e{3) Spat ial Consequence of Narrat ion: (Narration( a, /3) 1\ actor(x, a) 1\ actor(x, /3) ) �
loc(x, source(e{3))
=
loc(x,goal(ea) )
These rhetorical relations and their inferred semantic effects are used to define the appropriate update function in
SORT.
As well as axioms for determining a rhetorical relation, we also need to encode constraints on where to attach, because in general more than one attachment site is available in the SORS T that's been built so far (e.g. if the previous constituent a.:�. was attached to a 1 with Explanation, then both a 1 and a2 are available, and new information /3 could potentially bind t o either one of these). We consider this issue m the particular case when the information to be attached is a presupposition. As many authors have noted, there is the intuition that presupposed information conventionally portrays given information in some sense (Clark 1 977): assigning the presupposed information wide scope in soRT may very well place it in an earlier part of the discourse context than the one containing the presupposition trigger. To reflect this, we will encode a default version of van der Sandt's preference for global accommodation (c£ Mercer 1988). This is an important distinction. In van der Sandt's account, the preference for global accommodation always wins, unless the constraints of informative ness and consistency block it or the information can be bound at some lower site. In our theory, the exceptions to the default preference aren't enumerable. Any conflicting monotonic rule or more specific default will override it. We'll see examples of such rules shortly, and also in section 5.2.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
certam commonplace assumptions about eventualities (Asher et al. 1 996): if Narration( a, /3) holds and a and {3 share an actor x then the location of x is the same at the end of the event ea as it is at the beginning of the event e13:"
Nicholas Asher and Alex Lasc:uides 2.6 3
Furthermore, the default preference can be encoded declaratively in soRT, in contrast to van der Sandt where the way presuppositions are processed isn't part of the declarative definition of DRT Update. The declarative rule, which captures a default preference for global accommodation, is given below. First some notation: If {3 labels asserted content, then 8({3) is the label for the corresponding presupposed content So Prefer Global Attachment says: if o: is the most superordinate node in T to which the presupposition 8({3) can attach, then normally, it attaches there:u •
Prefer Global Attachment
This default will ensure that global attachment occurs in ( 1 ) and {I I). However, as we just mentioned, making this rule default allows other constraints on soRT update to ovemde it One particular rule that can do this is a constraint called Maximize Discourse Coherence, which intuitively captures the idea that people tend to interpret text in the way that makes 'most sense'. We have argued elsewhere that such a rule is needed to explain data where ambiguity resolution and bridging inferences are influenced by a complex interaction between semantics and pragmatics (e.g. Lascarides, Copestake & Briscoe 1 996; Asher & Lascarides, in press). The rule we gtve below is a generalization of the rules in those papers. We assume the following notational convention: {3 -+ • {3, means that K[3, is a DRS which represents one way of resolving the underspecification in K13. As background to our rule, we assume that in some contexts, attaching with one kind of rhetorical relation produces a 'closer connection' or 'better coherence' than attaching with another kind (see Asher & Lascarides, in press). To express this, we mtroduce a strict partial order on rhetorical relations, relative to the content of the context: Explanation >-r Background means that it would be preferable to interpret the new information that is to be attached as an explanation in T, rather than as background information. Both alternatives may be coherent, but one is better than the other, and this is partly because of the content of T. But it will not do just to look at the strength of the relation whtch attaches new information, perhaps resolved or not, to some site in T. We have to compare the strength of all the discourse relations that are affected by the update. This means that we have to compare various updates-viz. those using different attachment points or different resolutions of underspecified elements in the informa tion {3 to be attached to the discourse context So we will introduce a predicate that says that a particular update is (coherence) maximal with respect to the new information and the context, which, for a given SDRS T and new information {3, we write as T, {3-maximal. The semantics of this
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If o: is the most superordinate node in T, then normally: ( T, o:, 8({3) )
264
The SemantiCS and Pragmatics of Presupposition
•
Maximize Discourse Coherence:
If (a) (3 --t. {31 ; and (b) Update(r, a, {31 ) is T, {3-maximal; and (c) a' is the most superordinate node in Then normally: (c) ( r, o: , {31 ) First, note that this rule has a more specific antecedent than Prefer Global Attachment. So it will override that rule, if it conflicts with it But what circumstances will produce conflict between these two rules? Suppose that {3 labels a presupposition, and therefore there is a condition u = ?, which encapsulates that the attachment point to which {3 should be rhetorically linked is underdetermined. Suppose furthermore that there is a Qocal) attachment site 0:1 which yields a discourse relation R1 between 0:1 and {3, whereas the global attachment site O:g results in the rhetorical relation R1 for linking {3 to O:g. Suppose furthermore that R. is maximal in the partial order. Then Maximize Discourse Coherence favours resolving the under specification in {3 so that u = ? resolves to u = Ctf, and R = ? resolves to R = R1 • This means that the attachment point 0:1 is preferred over the global s1te o:g according to this rule. However, Prefer Global Attachment applies where O:g replaces o:. And so Prefer Global Attachment yields a conflicting default preference, that the presupposition {3 attach to O:g rather than O:f. Given that Maximize Discourse Coherence is more specific, the local attachment will be chosen over the global one in these circumstances. This interaction between Prefer Global Attachment and Maximize Discourse Coherence will explain why the presupposition that the problem has been solved in (6) is bound locally, in spite of the default preference for 'presupposition projection'. (6) A The Problem Solvmg Group is given a problem each day, and the group leader Mary has to assign it to someone in her group. John T
a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
notion is given in terms of the above partial order on relations, and comparing pairs of relations in any two updates under a partial, structural isomorphism '19 whose kernel on all updates is r.'3 We say that an update, Update(r, a1 , {31 ), is T, {3- maximal iff (3 --t. {31 and for any update Update(r, a1 , {31) such that {3 --t• {32: (i) there is a relation R in Update(r, a 1 , {31 ) and '19(R) in Update(r, a1, {31 ) such that R >-r '!9(R) and (ii) for every relation R in Update(r, a1 , {31 ) and 19(R) in Update(r, a1 , {3z}, R �r 19(R). Smce --t• resolves all underspecifications present in the soas (an idealization), maximal updates are well defined. The rule below then captures the following: the underspecified elements in K13 are normally resolved so as to maximize discourse coherence, regardless of which attachment Site is the most superordinate one:
Nicholas Asher and Alex Lascandes 265 is the best problem solver. But when he solves a problem, he always boasts about it. T1us annoys Mary, and so if she thinks that the day's problem is an unsolved one, she gives it to him, to test him. Otherwise, she gives it to someone else. b. B: John's being very quiet just now. Did she give him today's problem? c. A; Well, I'm not sure she did. Either John didn't solve the problem or else Mary realized that the problem's been solved. We shall gtve a detailed analysiS of this in section S·3· We'll also see how Maximize Discourse Coherence accounts for the conditional presupposition in sentence ( 1 0) in section 5.2. was involved will confound the editors.
Arguably, the most important feature of SORT update is that it models the way the content of a discourse is more than what's explicitly stated linguistically. It does this as follows: if the DICE axioms yield a nonmono tonic conclusion that the rhetorical relation between a and /3 is R, and information that's necessary for the coherence of R is not already m Ka or K13 (e.g. Background(a, /3) is nonmonotically inferred, but ovtrlap ( ea , e{J ) is not m Ka 'or in K13), then this content is added to the constituents in a constrained manner through the soas update process. Hence the result of an updating procedure can be not only the addition of a rhetorically connected constituent, but also the addition of semantic content to that constituent, that the grammar did not provide. We're now in a position to defme the update procedure in SORT. Update m ORT (glossed as Updateon) is used to define update in SORT (glossed as Updatemn)· Updateon uses set union:'4 •
DRT ' s Update Funct l.on:
Update0n(K. , K�) = [UK,
U
UKJ [CK,
U
CKJ
When the discourse context is 'empty' (i.e. when we are just starting a diScourse), Upda�son is just Update0rr But m general, Update50n is much more complex. In particular, when the discourse context T is a non-empty soas, Upda�son requires an available attachment point a in T to be identified, for attaching the new information /3 to with a rhetorical relation. In general, there is some indeterminacy in this choice of a in spite of the default preferences given in DICE, which constrain but may not fully determine which available site to attach new information to. So to specify the update of old information with new, we need to specify three arguments: Kr for the old information Ka (which is on the right frontier of Kr) for the attachment point, and K13 for the new information. We write
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(1 0) If David wrote the article, then the knowledge that no good logictan
266 The SemantiCS
wd
Pragmatics of Presupposition
this as Upda�sorrr (Kn Ka , Kp). Moreover, we do not perform set union on
Ka and Kp. Rather, they are brought together by computing a rhetorical relation between a and {3 in DICE.
Were we to ignore the effects of discourse structure on semantic content, and in particular on the resolution of underspecified conditions, we could leave the matter here: a non-empty discourse context with an attachment point a, Upda�sorrr (Kn Ka , Kp) would be a new SDRS KT'• which IS just like Kn except that it also includes: (a) the new information K13; and (b) an attachment of {3 to a with a rhetorical relation R that is computed via DICE.
•
SDilT 1 S
Update Funct�on:
I. Upda�sorrr ( 0, 0, [{3] [{3 : Kp] ) = Upda �orrr ( 0, [f3] [{3 : Kp] )
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
However, it is a central point of our theory that the choice of a rhetorical relation does affect semantic content, and the resolution of underspecified conditions. The update of a discourse context should therefore ensure that the content of Kn Ka and Kp are modified, so that the coherence constraints on R are met (where R(o:, !3) is inferred via DICE), and underspecified conditions are resolved where appropriate. In fact, SORT Update is defined so that additional semantic content cp, that's inferrable in DICE from the discourse context and Kp, IS added to Kp in the update. In addition, cp can be used to resolve underspecified conditions m Kp, Ka. and any Ka' such that o:' is attached to a with a rhetorical relation. A final complexity comes with the need to distinguish several cases. First, we need to specify how updates of a null context with new asserted information proceed (clause 1 below). This case is distinct from the others, because the information is added to the null context, rather than linked to it with a rhetorical relation. We then need to specify updates of a non null context with asserted information (clause 2), and with presupposed information (clause 3 below). These are different, because of the under specified rhetorical conditions in presuppositions, which are absent from asserted information. We now define Upda tesorrr formally. As we have just seen, there are three cases: one when the discourse context is empty, one when asserted information like ( 17) is attached to a non-empty discourse context, and one when presupposed information (with underspecified rhetorical condi tions) is attached to a non-empty discourse context Let � be the (nonmonotonic) proof theoretic counterpart of the consequence relation for the glue logic. Let Pred1T be the label of the SDRS constituent in which 11" is declared, or (equivalently) in which a condition of the form 11" : K occurs. And let a [f3/ 1"] be the result of replacing 1" in a with {3. Then Upda tesorer is defined as follows:
Nicholas Asher
and Alex Lascandes 267
(c)
K + = Updateon(K;,.td(a) ' [,BJ [,B : Kp (cp) , Ra = ?, u = ? , . . . ] ) [Ro = R/Ro = ?] fu = a/u = ?], where Kp(cp) is specified
as
above, and (d) K�(a ' ) is specified as above.
In words, clause I deals with the case when we start a new discourse. It applies only to the asserted information of a clause, for the new informatiOn does not contain underspecified rhetorical conditions R = ? (in contrast to clause 3). It applies only to asserted content, because clause 1 simply adds this to the SDRS for the discourse. Doing this with a presupposition would produce an SDRS which contains the unresolved underspecified conditions R = ? and u = ?, resulting in a representation that's not well-defined. Consequently, at the start of a discourse, one must attach the asserted information befort one updates this with the presupposed information (using clause 3). However, when continuing a discourse rather than starting it, the relative order for attaching the asserted or the presupposed content of a sentence is not specified. We'll say more on this shortly. Clauses 2 and 3 deal with the case when we continue with a discourse (because Kr is non-empty). Clause 2 deals with asserted information (because the SDRS ,B does not contain conditions of the form R = ? and u = ?), and clause 3 deals with presupposed information (because these conditions are present in ,B). Let us look at clause 2 first. Clause 2 demands that you rhetorically connect ,B to a. Clause (2a) eruures that the rhetorical relation R that is chosen is the one that results from the DICE axioms. As can be seen from clauses (2b) and (2c),
R ( a, ,B)
gets added to the update,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
2. For Kr =/:- 0, Updateson(Kn Ka , [,BJ [,B : Kp] ) is the SDRS Kr' such that: (a) (( r , a , ,B ), J.L (Ka ) ( a), J.L(Kp ) (,B) ) r-- (R ( a , ,B) 1\ cp ); and {b) Kr' = Kr [K + /Kfud(a)]. where: {c) K+ = Updateon(K;,.td(a} 1 [,B] [,8 : Kp ( cp) , R ( a , ,B)]), and Kp ( cp) = Kp together wtth those conditions specified by cp (where cp is r-- provable) that are needed to satisfy the coherence constraints on R; where these conditions cp are also used in Kp ( cp) to resolve underspecified conditions in Kp (e.g. x = ?); and: (d) K�(a) is just KPrtd(a) except that it may contain specifications of underspecified conditions in Ka,, such that a' = a, or for some rhetorical relation R, R ( a' , a) or R ( a , a') are conditions in Kfud(a)• if those specifications are r-- provable from ( J.L(Ka ) ( a), J.L(Kp ) (,B)) and R ( a , ,B) . 3 · For Kr =/:- 0, Update50n(Kr , Ka , (,B] [,B : K,e, Ra = ? , U = ? , . . . ] ) lS the SDRS Kr' such that: (a) ( ( r, a , ,B ) , J.L(Ka) ( a) , J.L (Kp ) (,B) ) r-- (R ( a , ,B) 1\ cp ); and {b) Kr' = Kr [K + /Kfud(a)], where:
268 The Semanncs :md Pragnuttcs of Presupposltlon
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
specifically to the constituent that introduced the attachment point a as a speech act discourse referent Furthermore, a in clause (2a) stands for information that follows from DICE when reasoning about how {3 attaches to a. Thus cp will include information additional to that gtven by the grammar, that follows from R's coherence constraints (e.g. if R is Background, cp includes owrlap(ec0 �{3}). Clauses (2b) and (2c} guarantee that this additional semantic content that's inferred via DICE also gets added, this time to the constituent Kf3 that was derived via the grammar. Moreover, (2c) stipulates that cp replaces underspecified conditions in Kf3 with specified ones (in Kf3( cp )), if it can (we shall see many examples where this happens in section s). Clauses (2c) and (2d) also ensure that information cp that is inferred in DICE resolves as many underspectfied conditions as it can in Kfud(a) · More precisely, underspecified conditions in the attachment Ka and things that are rhetorically linked to it can be resolved by cp. Clauses (3a-d) stipulate that updating a discourse with a presupposition is just like doing it with assertions (i.e. clauses (2a-d)): that is, a condition R(a, {3) is added; Kf3 gets modified with additional information c.p; and Kf3. Ka and Ka' (where a' is attached to a) can have underspecified conditions resolved by cp. But there's one more thing: the underspecified conditions Ro = ? and u = ? are replaced with Ro = R and u = a. In other words, the rhetorical anaphoric elements in the presupposition reflect the rhetorical connection and attachment point that's given by DICE. Note that Update50a:r has been defmed for attaching only one packet of information to the discourse structure. But the grammar produces more than one packet of information for each clause: there's one for the asserted content, and one for the presupposition. So defining how to update a discourse with the content of a clause involves refining the definition of soRT update, to allow (perhaps) several attachments and their consequent modifications to the content For the purposes of this paper, we shall assume that updating the discourse with a clause is defmed with respect to a set of attachment points-one for the asserted information and one for each distmct presupposition. This more complex function is simply the result of sequentially carrying out the procedure of Update50aT as defined above, for the asserted information and each presupposition according to some sequence. In contrast to van der Sandt, we remain agnostic as to whether updating with presupposed information or asserted information is done first (except for the first clause in the discourse). The result of any sequence of updates will be viewed as an admissible mtegration of the new infonnation in the discourse. Overall, the d�finition of soRT update guarantees that the way a presupposition is interpreted in discourse is dependent on a complex
Nicholas Asher and Alex Lascarides 269 mixture of compositional semantic and pragmatic information. This is because the DICE axioms that are used to infer
R (in clause
3) are based on
linguistic and non-linguistic knowledge. Moreover, the interpretation of presuppositions is constrained by a richer notion of accessibility than that
given by DRT, since presuppositions can attach to what is SDRT-available, which in general is a smaller set than what is DRT-accessible. We shall demonstrate in our analysis of examples that these two features overcome some of the problems with traditional dynamic approaches to presuppositions. Further more, the above axioms in the glue logic, which constrain which attachment site to use, will ensure that updating the context with a presupposition is in general different from doing it with an assertion, in the manner we described
structure thati where the corresponding assertion attaches. In other words, presuppositions will tend to have wider scope than assertions. The definition of Update models the resolution of ambiguity, because it constrains the updated SDRS to be one where inferences in DICE resolve underspecified conditions whenever possible (clauses (2c-d) and (3c-d)). In fact, clauses
(w)
and (3d) mean that underspecified conditions can
be otlm- constituents update the discourse (e.g. Ka, way K13 attaches to Ka. where R(a ' , a) holds for
resolved by the way certain can be resolved by the some
R). This
is what happens in (28):
(28) a. John took an engine from Avon to Dansville. b. He picked up the boxcar. c. and took it to Broxburn.
the boxcar in (28b) means that the DRS Kh, • which appears in the SDRS that represents the presupposed content of (28b), contains the under
The definite
B = ? to some underspeci£ed antecedent object ) v = ?. As in ( 14 , Narration connects the asserted content of (28b), which we label b.,, to the content a of (28a). And as before, this imposes a spatia specified bridging relation
temporal constraint, that the boxcar is in Dansville . Thanks to clause (3d),
this additional information, that is added to. Kb. via updatesoan resolves and v = ? in Kb to B = in and v = Dansville respectively. Thus in , this example, the resolution of content in the presupposition bp is determined by the way the ass�d content ba attaches to the context a.
B=? We'll
see
the formal details of this example in section S·3·
4- 3
Some remarks about background
We proposed in section 3 that the principal relation that binds presupposi tions to the context is
Background. This
captures the intuition that they
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
in section 3 · In particular, these axioms will determine which attachment sites presuppositions should attach to, and normally, they will be higher in the
270 The Semantics and Pngmatics of Presupposition
that the two propositions. There is a King ofFrance and he is bald in the discourse (29), which features Background, are both true: {29) There is a King of France. He is bald. Second. both Background and Narration require a common topic. This explains why discourses like (3o) sound odd without any special prior context. (3o) a. ??Max walked in. Mary dyed her hair black. b. ??Max smoked a cigarette. Mary had black hair. The final constraint is designed to account for the behaviour of pronouns in discourses that feature Background. Compare the text (J I)-where {J ib) connects to (3 Ia) with Background-with (3 2), where the relation is Narration. (3 1) A burglar broke into Mary's apartment. b. Mary was asleep. c. He stole the silver. (32) a. A burglar broke into Mary's apartment. b. Mary discovered the break-in the next day. c. ??He stole the silver. Asher et aL {I 996) use {3 I) vs. {32) to argue that if the discourse features a condition of the form Background( a, {3), then objects introduced in a are available for pronominal reference in the proposition "Y that comes after {3. This is not the case for Narration( a., {3). But Background is a coordinating relation, and therefore a. itself is not available. So why are its objects possible antecedent's to ""('s pronouns? This pronominal behaviour is captured in soRT via the different kinds of topics that are constructed for Narrations and Backgrounds (Asher et aL This guarantees
a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
provide given information. And the DICE rules predict that Background is the principal relation, because generally, the constituent in the context or the presupposition itself is stative (c£ the rule Background). Since Background plays such an important role in this theory, it is important to define the constraints it places on the contents ofthe propositions it connects. This is because these constraints affect the interpretation of presuppositions through Updates oJCr Ultimately, Backgrqund ensures that presuppositions are constrained by more than informativeness and consistency. We have already encountered the temporal constraint on Bacleground, supplied by the Axiom on Background {Lascarides & Asher I993; Asher et al. I996}. There are three other constraints. First, all the so-called simple rhetorical relations {Asher I 993)-such as Background, Narration, Elaboration, Paralkl and others-are veridicial: that is, the DRSS associated with the terms they relate are true: • Veridicality: If R is simple, then R(a, {3) --+ (Ka 1\ K13)
Nicholas Asher and Alex Lascarides
271
19¢). We concentrate here on the topics for Backgrounds. When Background( 1T" 1r� ) is inferred, this induces a structural modification to the soRS. It creates what is known as a Foreground-Background Pair, which contains (i) a constituent 1T1 in which the information of both the background constituent 7T� and the foreground 7T1 is 'repeated' (1r3 is the topic and its content amounts to UpdateoRT(K'Ir, , K1r, )), (ii) the SDRS which contains the constituents 1T 1 , 1T� and the condition Background (1T 1 , 1T� ) forms a constituent, which gets labelled (with, say, 7r.). and (iii) a link between 1r1 and 7T• is introduced, known as FBP (standing for Foreground-Background Pair): i.e. FBP( 1T1 , 1T. ) is introduced into the SDRS. Thus whenever an SDRS contains the conditions given in (3 3), it also contains those given in (34):
( 34)
7T I : K7T, , 7T� : K'Tr, Background(1T 1 , 7r� )
7T3 : Upda�oRT(K7r, , K7rJ FBP(1r1 , 1r• ) 7T. :
7T I : K7r , , 7T� : K?r, Background( 7T 1 , 1r� )
Intuitively, the topic 1r3 ensures that the 'main story line' content in K1r, is still access1ble. FBP is a subordinating relation. And so the available constituents to which future representations of clauses can attach are: 7Tl' 7T• and 7T�. Since there are referents in the topic 1r3 which denote objects that were introduced in 7T1 (since 1r1 'repeats' the content of 7T1), this makes these objects available for future anaphoric reference. This predicts, correctly, that the burglar is available for anaphoric reference in (3 I ). In contrast, Narrations like that gtven m (32) do not have this topic structure. Moreover, the constituent 7T 1 which represents the first sentence is not an available attachment site. So the pronoun he in (32c) cannot refer to the burglar, as required. To handle examples like (3ob� we impose the constraint on FBP formation that an FBP can be constructed only if the backgrounding constituent sets the stage for the foregrounded constituent. That is, there must be some thematic link between the two constituents that can either be determined already from the discourse context or from world knowledge. So
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 3 3)
'
272 The Semantics and Pragmatics of Presupposition
in (3ob), we cannot form an FBP because neither the context nor conventional knowledge have made clear any link between Mary's hair being black and Max's smoking a cigarette. Though we would like to have more to say about thematic continuity, at present we must leave the matter here. We shall see in section s. 3 that the FBP topic structUre plays an important role in constraining the interpretation of presuppositions. However, in many cases, it is not essential to the analysis. So on occasion, we will omit it for the sake of simplicity.
s THE PRAGMAT ICS O F PRESUPPOSITIONS
5.1
Consider (I s )
Presuppositions in simp le sentences
again
(I s) The King of France is bald. We must build the soRS conditions from the grammar, and then use SORT update to build the final representation of the sentence. We will give an informal synopsis first. We saw in section 4-I that the grammar will produce two soRSs for (Is ): one representing the presupposed content that there is a King of France, and the other representing the asserted content that he is bald. Since this is the first sentence in the discourse, the SDRS for the whole sentence must be updated with the asserted content first (by adding it to the empty set), since the presupposed content demands a (rhetorical) anaphoric link. Now the presupposed content must be attached to this asserted content with a rhetorical relation. DICE is used to compute this. The presupposed content is stative, and so the rule Background given in section 4-2 will mean that the presupposed content is attached with Background. That is, the soRS for ( I s) will amount to: that there is a King of France is Background to his being bald. &c/eground's topic StructUre is thus invoked in the analysis of (I 5), just as it is in any soRS containing Background. This topic structUre does not have a crucial effect here, largely because (I S) is a simple sentence and does not feature the kind of discourse phenomena that are central to our concerns in this paper. However, we spell out the topic structUre in this simple example, for the sake of completeness. In future analyses, in order to simplify matters,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We now analyse some examples of how discourse is updated with presuppositions. We will start with simple sentences (ie. sentences without scope bearing elements), and then move on to complex sentences and multi-sentence discourse.
Nicholas Asher and Alex Lascarides
273
we may omit the topic structure, unless it plays an important role in producing the right semantic effects. Let us now go into more detail First, we construct the two soass that the grammar produces for the presupposed content and the asserted content. The former includes a bridging relation B between the object denoted by the King of France and some (anaphoric) antecedent, that must be resolved. As argued in Asher & Lascarides (in press), the definite the King ofFrana is a special case, where the grammar itself ensures that Frana and of serve to fill the underspecified antecedent object and bridging relation respectively.' s Therefore, the presupposed content of (I S } that is produced by the grammar is the SDRS p below; its asserted content is a:
x, u , ( ' , t ' rr, (1
( I S) a :
rr, :
t, n
bald(e, x) holds(e, t) t�n
rr1.:
p:
king(x) of(e ' , x, u) Frana(u) holds(( ' , t ' )
R(v, rr') R=? v=?
Now the task is to create an soRS from these two constituents, using the definition of soRT Update. (I s ) is the first clause in a discourse. So by clause I of the Update definition, the soRS when the assertion a is processed is just a itsel£ We must then update this with p. According to clause 3 of soRT Update, this involves using DICE to compute a rhetorical relation between rr1. and an available propositional discourse referent. In this case, there's only one: rr, . Both rr1 and rr1.'s main eventualities are states, and therefore Background applies. So by Defeasible Modus Ponens, Background( rr. , rrJ is inferred. That is, that there is a King of France is background to his being bald. And according to clause 3 of the definition of soRT Update, the underspecified conditions R = ? and v = ? need to be replaced with R = Background and v = rr, in the updated SDRS. The condition Background( rr 1 , rr1. ) imposes coherence constraints, described in section 4-3· First, by Axiom on Background, ovtrlap(e, e' ) holds. So, by clause 3 of the soRT Update definition, this condition must be added to K1r,: that is, to the DRS that is labelled rr1. in p above. Second, the above topic structure is created, since the thematic link constraint is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
rr1. , R, v
274
The Semantics and Pragmatia of Presupposition
satisfied by the backgrounding constituent is giving a property of the subject of the assertion. So the final representation of (I s) is (Is a): 7Tt , 7T4 7T3 :
t, t, n, x, u, e I , t I bald(t, x} holds(e, t) t £; n, ovalap(e, e1) king(x) of(t1, x, u ) Frana(u) holds(t1, t1)
FBP( 7rl , 7T4 ) 7T1 1T, , R
V
( I s a}
t, t, n
1T I : 7T4 :
bald(e, x) holds(t, t) t £; n
7T, :
king(x) of(t1, x, u) France(u) holds( e1, t1) ovtrlap(t, e1)
R (v, 1r1) R Background V = 1T I =
The SDRS (I s a) entails that there is a King of France, because the axiom Verid icality applies to Background (and to FBP). Moreover, the King of France could be referred to with a pronoun in a subsequent sentence to (I s}.
even though there being a King of France is background information. This is because 1T3 and 1T, are available constituents. This is a much more complex than the analysis given by van der Sandt (I 992). Van der Sandt represents (I s) as (I sb), and this is identical to his analysis of (J s):
x, u, t, t, t I , t I , n king-of-France (x) ( I sh ) bald(e, x) holds(t, t) t £; n
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
x, u, t I , t I
Nicholas Asher and Alex Lascarides
275
(3 S) A King of France is bald. fact, it is also (almost} identical to the topic 1r3 in our representation. So we make the same predictions about which antecedents are available for future pronominal reference, in continuations of the discourse (I S}.'6 Our analysis of (3 s) is (I sh). This is different from the representation (I sa) for (I s). The difference arises because definites introduce anaphoric elements, including, for instance, the underspecified rhetorical relation R, which must be resolved by Upda�sorrr- Indefinites are not anaphoric in this way. More importantly, however, our analysis of definites in discourse is more constrained than van der Sandt's. Van der Sandt imposes constraints such as informativeness, consistency, and accessibility. We strengthen these with pragmatics. To illustrate this, consider (36}: In
According to our theory, the underspecified bridging relation B and the rhetorical relation R that is triggered by the content ofthe definite in (3 6) must be resolved through soRT Update. ButB cannot be resolved in this case, because reasoning about the rhetorical link fails to produce a plausible connection between the man and an antecedent object (the woman or the house) (see Asher & Lascarides, in press, for details). In contrast, van der Sandt does not explain why (36) is odd: since there is no antecedent man to bind the presupposition to, one attempts to accommodate it, and since the result is consistent and informative, the accommodation is successful So (36) is predicted on van der Sandt's account to be acceptable, when it is not. 5 .2
Presuppositions in conditional sentences
Conditional sentences such as (I) and (2) give rise to choices about where to attach presupposed content, because the Condition relation, that is introduced by if, is a subordinating relation. (I) If baldness is hereditary, then Jack's son is bald. (2) If Jack has a son, then Jack's son is bald. Like van der Sandt, the presupposed content in the consequent can attach in the global position (i.e. to the conditional as a whole), the intermediate position (i.e. to the antecedent) or the local position (i.e. to the asserted content of the consequent). Our glue logic must therefore yield inferences about which attachment site is chosen. Recall Beaver's (1996) idea that the interpretation of presuppositions is influenced by which reading is most plausible. One could imagine invoking here a general Gricean constraint of the form: make all of your attachments or presupposed and asserted information so as to make the speaker's
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(36) ??A woman lives in that house. The man is bald.
276 The Semantics and Pragmatics of Presupposition contribution as plausible as possible. We think, however, that the choice between attachment points that is governed by plausibility is specifically linked to conditional contexts and is a matter of discourse coherence. We claim that there is a linguistic convention about the strength of a Defeasible Consequence (or conditional) relation vis a vis a higher attachment with &ckground that is affected by plausibility. And it is just such a conventional comparison between the strength of discourse relations and attachments that Maximize Discourse Coherence can exploit. Indeed, the theory predicts differences between (10) and {1 I ), because for {10) Maximize Discourse Coherence conflicts with the default Prefer Global Attachment and will therefore override it, whereas in ( I I ) there is no conflict
soRT alone has little to say about plausibility itsel£ In general, this is a matter of extra-linguistic knowledge, and even the task of comparing the plausibility of one scenario over another requires reasoning with arbitrary domain knowledge. Defining a logic of plausibility would therefore take us too far from our main concerns here. So, we'll simply assume that we have some partial, qualitative or modal notion of comparative plausibility in DICE: PL ( a) >- PL ( f3) means a is more plausible than {3.'7 Let us now look a t conditional contexts in more detail. Given an attachment of 8( C ) to B in the context A > B, where > is some standard conditional operator, we would expect A > (B 1\ 8( C)) to be equivalent to A > B and A > 8( C). This equivalence holds in many conditional logics (for instance, Stalnaker's conditional logic, Lewis's ( 1 973) counterfactual logics, or the normality conditional of Asher & Morreau I 99 1 which underlies DICE). One consequence of this is that the attachment of the presupposition 8( C) to the consequent B of a conditional whose antecedent is A, is equivalent to attaching A > 8( C) to the constituent in which the conditional A > B occurs. So local attachments in this case are equivalent to conditional presuppositions. But since these conditional presuppositions seem to be preferred in at least some cases, this would go against the van der Sandt proposal of accommodating presuppositions globally whenever possible. Our approach to these conditional presuppositions will be to treat them as cases of binding with Dif-ConMquffl££. If we assume that > is the operator version of the relation Def-Co�qunue, then such attachments in conditional contexts amount to attaching the presupposed information with
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(I o) If David wrote the article, then the knowledge that no good logician was involved will confound the editors. ( I I ) If DaVid wrote the article, then the knowledge that David is a computer program running on a PC will confound the editors.
Nicholas Asher and Alex Lasc:uides
277
Dif-Constqumce. We will stqumct relation can be
suppose in addition that this Dtftasible Con demonstrated by a deduction, and that the information at the attachment point plays an essential role in the derivation of the presupposed information, in the sense that one could not make the derivation without it. In some cases, it is clear that commonly supposed background knowledge is enough to link presupposition with an appropriate premise via Deftasible Constqumce. For example, consider the pair (37) (due to M K.rifka, p.c.): (37)
a. If David b. If David
is is
going diving, he11 bring going diving, he'll bring
his wetsuit. his dog.
(3 8) If David goes diving, then he has a wetsuit. On the other hand (3 7b) does not. It generates the simple presupposition: (39) David has a dog. In (3 7a), general world knowledge allows us to derive defeasibly the fact that David has a wetsuit from the fact that David goes diving. But from the fact that David goes diving we can't derive that he has a dog. Given a preference for binding with Def-Constquena, we will predict this binding for (37a) but not for (3 7b). Conditional presuppositions need not be generated only from the consequents of conditionals, however. We can attach presuppositions generated in the antecedent of a conditional to that antecedent with Dif Constqumu as well, as in (4oa) due to Beaver (p.c.): (4o)
a.
If David is a shepherd and has a good rapport with his dog, he'll get a job on the farm. b. David is a shepherd and he has a good rapport with his dog.
Once again it appears plausible that we can already defeasibly derive David's having a dog from the fact that he is a shepherd. Note, howeve�, that it is the conditional that licenses this conditional presupposition. The conditional presupposition vanishes when there is no conditional in context as in (4ob). Further, the presupposition changes depending on the modal context. For example in (41) we only get the counterfactual presupposition that were David a shepherd, he would have a dog: (4 1) If David were a shepherd and had a good rapport with would get a job on the farm.
his
dog, he
We claim that conditional or counterfactual presuppositions surface only in conditional or counterfactual contexts. In (4oa) the presupposition is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(37a) generates the conditional presupposition:
278
The Semantics and Pragmatics of Presupposition
attached with Dq-Consequmce, but because the premises are taken to be asserted as true the defeasible consequence is itself drawn and assumed to hold in the context. In cases where the attachment point (and essential premise) falls under a modal or conditional context, the defeasible context cannot be drawn, and the presupposition that is bound by Defeasible Consequence inherits the modal force of its attachment point. This inheritance is predicted if we think of the discourse function of these antecedents of conditionals as conditional or counterfactual assumptions and we take these conditionals to be closed under the consequence relation'8 •
Let's turn now to the analysis ( 1 0) and (u). While ( 1 0) generates a conditional presupposition, it is not plausible to claim that general world knowledge and the discourse context make available the relevant premises for deriving the presupposition from the antecedent. Such a conclusion would follow if the context supported particular facts about David, like the fact that he's not a good logician. But the interesting and puzzling fact is that the conditional presupposition seems present even when the context doesn't support such facts. We claim that people add these distinct facts 1/J as implicatures in order to bind the presupposition with Dtftasible Co�quence, as long as the assumption of tf; is more plausible than the original presupposition. We'll say that one update of r with {3 is strictly preferred to another when of the pair it is r, {3-maximal, which was defined in § 4-2. •
Preference for Binding:
Suppose
Update(r, a, 8(/3) ) links a and 8({3) with Dif-Co�uence, assuming some extra information tf; distinct from Ka(/3) that is attached to r via Background and 2. Updak(r, a, , 8(f3)) links a, and 8(/3) with Background but not 1.
Dif-Consequence
Then:
1
Updak( r, a, 8(/3)) is strictly preferred to Update( r, a , 8(/3)) {normally if) and only if PL(J.L(KI/J) (x)) is greater than PL(J.L(Kacf3))(x))) Note of course that tf; could simply be a tautology, in which case the update with Def-Consequmce will be strictly preferred to the update with Background without restriction. This statement on preference of updates will play an important role in
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Closure of Conditionals under Consequence: Suppose DefCons( a, /3) and Ka. => K-y hold, where => represents either a normal or counterfactual conditional Then: Ka. => K13.
Nicholas Asher and Alex Lascarides
279
detennining whether the presupposition should be locally or globally bound in conditional sentences such as ( I), (2), ( 1 0) and (I I). Note that the above axiom on its own does not guarantee that a unique attachment site is chosen, even for conditional sentences. Heirn ( I 98 3 , I 992) and van der Sandt ( I 992) note that presuppositions have a conventional property: they project out from embeddings to have the widest scope possible. In fact, in certain cases, this tendency to project actually seems to win over what's plausible. Consider sentence ( I I), for example. According to common sense, it is more plausible that sentence (42) is true than (43):
(42)
(42)
is the reading one gets of ( I I ) if the presupposition attaches in the intermediate position. (43) is the result of global attachment, and corres ponds to the intuitive interpretation of ( u ) (since it entails David is a computer program). So global attachment seems to be favoured here, even though the result is the less plausible scenario. These examples show that, in contrast to Beaver (I 996), a preference for attaching presuppositions that is based solely on plausibility cannot be right. It is important to stress that (42) being dispreferred is not a counter example to the Preference f or Binding constraint. That axiom together with Maximize Discourse Coherence only rules out a conditional attach ment, unless it is most plausible. But the presupposition in (42) is not attached with a conditional or Dif-Co�qumce, and so the constraint does not apply. There seems to be some tension between deriving a plausible reading, and the conventional property of presuppositions that they project to the widest scope. The constraint Preference for Binding and the axioms of Maximize Discourse Coherence and Prefer Global Attachment reflect this tension. Prefer Global Attachment will determine that global attachment occurs in (I) and ( u ). But Preference for Binding with Maximize Discourse Coherence will override this default to produce local attachment in ( 1 0). With these constraints on conditional attachments in place, let us analyse (I) and (2). Let (I)'s antecedent that baldness is hereditary be represented by (so Krr, supplies the information that baldness is hereditary):
a (a ) � �
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If David wrote the article and he's a computer program, then the knowledge that David is a computer program will confound the editors. (43) David is a computer program, and if he wrote the article, then the . knowledge that David is a computer program running on a PC will confound the editors.
280 The Semantics
and Pragmatics of Presupposition
The consequent (that Jack's son is bald), contains a presupposition, and so it is represented by two constituents b4 and bp given below (as before, we have
glossed over the presupposition triggered by the proper name jack, by making the simplifying assumption that Jack is represented by the
constant j}.
7T� t, t 7T�:
7T3 x, j 7T3 :
son(x) of(x , J )
The soRS after updating with the first clause is simply a itsel£ Moreover, because of the cue word if, the grammar produces a Condition relation between the speech act discourse referent 1r 1 in a and another discourse referent, which must be equal to or include within its scope the asse rted information b4 of the consequent. We will label the result-which includes the
Condition
relation, a and b4-with the speech act discourse referent
1r4•
Now SDRS Update must be used to attach the rest of the discourse information, namely bp, which is labeled 1r3• Condition is a subordinating relation, and so there are three available attachment sites for
1r3• Condition is
a subordinating relation, and so there are three available attachment sites
for 1r3 : either (a) 1r3 binds 'globally', to the speech act discourse referent 1r4; or (b) 1r3 binds in the 'intermediate' position, to 7r 1 ; or (c) 1r3 attaches in the 'local' position, to 1r �· One must check which of these choices produces a
coherent result, and then choose among these coherent alternatives, using the axioms we specified earlier.
Attaching 1r3 Background(1r1 , 1T3 )
to
7r1
is
not
possible:.
Background
applies,
but
cannot be inferred, because a decent topic cannot be
constructed; there is no thematic link between background and foreground, just as there is no thematic link in the incoherent (44).
(44)
Baldness is hereditary. Jack has a son.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
bald(t, x) lwlds(t, t) t�n
Nicholas Asher and Alex Lascarides 28 1
7r. , 7rl 7r[ ) 7r1. 7r. :
7r 1 : K1r, , 1r1. : K1r, Condition ( 1r 1 , 7r ) 1.
7rl : K1f1 Background(1r. , 1r3 ) Now consider the analysis of (2). As before, the grammar produces a constituent a which contains the information in the antecedent clause, which we label 7r1 again. And it induces a Condition relation between this and (at least) the asserted information in the consequent So just as with ( I ), there are three available attachment points for binding 1r3 in the presupposed information bp : 1r•, 1r" and 7r:�.· In ( I ), 1r3 could not attach to 7r1 with a rhetorical relation. But now the situation is different, because the content of 1r 1 is different As we suggested in section J, if the content of the new information is a defeasible consequence of the content of the proposition to which it is to be attached, then normally, the rhetorical relation Dej-Cons�quenu is inferred. This is encapsulated in Defeas ible Consequence: •
Defeasible Consequence:
(( r, a, {J) 1\ ((J1(Kr) (r) I\ J1(Ka) (a)) > J1(Kf3 ) (f3))) > Dif-Co�quenu(a, fJ) The constraints on Dif-Cons�quma are compatible with those on Background, and so it is possible for both Def -consequence and Background
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Neither can we infer Dif-Cons�quenu(7r " 1r3), for the auxilliary information needed to make the derivation is surely less plausible than the presupposi tion 1tsel£ Thus, Prifernru for Binding does not apply. 1r3 can attach to 1r• or 1r1. coherently, both with Background (since Background applies in each case). We claimed earlier that the local attachment corresponds to the reading: If baldness is hereditary, then Jack has a son and he is bald. We can now explain why. The Condition relation holds between 7r1 and the topic that's introduced by the Background relation between 7r2 and 7rp and this topic contains a 'repeat' of the content given in bd and br Thus part of the meaning of this SDRS is that if baldness is hereditary, then Jack has a son. However, the default Prefer Global Attachment applies and so by Defeasible Modus Ponens, global binding is chosen in this case. The final ' analysis is given in ( I ) below (in simplified form, since we have not included the FBP structure that's introduced by Background):
282 The Semantics and Pragmatics of Presupposttion to
fire,
yielding
Background( a:, {3)
the
result
that
hold. Note also that
both
4J > 4J
Dif-Constqunta( a, {3)
4J. So Defeasible Consequence will apply whenever one
attempts to attach
a presupposition to a constituent of the same content. So
Binding
and
is valid for any proposition
Prefuence for
holds without restriction, since no extra information is needed to
establish it. Our constraints Priferrnce for
way van der Sandt's principle that binding is preferred over accommoda tion. And for (2), these axioms predict, correctly, that the presupposition does not project out &om the conditional. In general, whether a presupposition projects from an embedding or not depends on several things. First, it depends on pragmatic and semantic content of the prior discourse and the presupposition, because this is used to reason about which available attachment sites the presupposition can coherently bind to, and which rhetorical relation to use. Second, the projection depends on the relative plausibility of the various choices of attachment (that are coherent� Third, it depends on the relative strength of the rhetorical connections provided by various choices of attachment. Finally, it depends on the default which favours attaching presuppositions high up in the soas structUre. Some general results concerning projection follow from this. Some of these are given below (for single sentences in the null context):
I . Projection does not occur in a sentence of the fonn. IfA tkn B, when the presupposition triggered by B is a default consequence of A. 2. Projection does occur in a sentence of the form. If A tkn B, when the presupposition triggered by B is logically independent of that given by A and plausibilities reflect logical independence in the way that probabilities do (note that it can always attach coherently with
Background
to the global site, because
B
will contain sufficient content
that's similar to its presuppositions so that the topic required by
Background can be constructed). 3· In a sentence of the form A and not B, the presuppositions triggered by B will project from the embedding so long as they can coherently attach to A with a rhetorical relation. To say more in general is tricky,
because
projection is determined by a
complex interaction between semantic and pragmatic knowledge resources.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Binding and HaximJ.Ze Discourse Coherence force us in the analysis of (2) to bind 1r3 to 11"1 (i.e. to bind the presupposition to the antecedent of the conditional) with Dif-Constqutn«. We therefore infer a binding to the (intermediate) attachment site with Background and Dif-Constqutnce, even though the presupposition could coherently bind at the global level with Background. Thus Maximize Discourse Coherence and Prtjuma for Binding capture in a more general
Nicholas Asher and Alex Lascarides .2.83
(7) ?? I don't know whether the Pope realizes the Pope has measles.
has
measles. But every Catholic
We now examine these examples in more detail First, note that intermediate accommodation isn't always blocked:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Furthermore, we view these solutions to the Projection Problem as a side issue; we are more concerned with addressing the problem of interpreting presuppositions in multi-sentence discourse. We now consider the analyses of (zo) and (1 1� Van der Sandt's analysis was inadequate, because it failed to distinguish their presupposed contents. In our theory, the constraint Priferencefor Binding accounts for this variation. Consider (10) first. The pre-supposition trigger knowledge that ¢ generates a propositional presupposition ¢ that must be attached to the context. We will defeasibly infer ¢ if we assume that David is not a good logician. But this assumption about one individual seems more plausible than the general assertion that no good logician was involved.'9 So local binding (to the asserted information in the consequent) is inferred for (10) via Prejtrt:na for Binding and Maximize Discourse Coherence in the now familiar way. In contrast, the information needed to establish an inference &om the fact that David wrote the article to the fact that he's a computer program is no more plausible than David's being a computer program tout court. So the truth functional 'only if' part of Conditional Presupposit �ons forces one to eschew the conditional attachment and to bind the presupposition in (I I) non-locally (Le. as Background to the antecedent, or to the whole conditional statement). Prefer Global Attachment leads us to infer global attachment, because of the default preference for it. Note that global attachment occurs even though the intermediate attachment describes a more plausible scenario, as we explained earlier. Our theory predicts this, because plaustbility is a deciding factor only in quite specific circumstances. Thus the axioms which constrain the update of discourse with presuppositions predict the difference between ( 1 o} and (I I). It is interesting to compare this approach to conditional presuppositions to that offered by satisfaction theorists (e.g. Beaver 1997). Satisfaction accounts always generate conditional presuppositions in contexts such as (1o). But such conditional presuppositions do not always occur (c£ (n )). So they have tried to provide accounts for how the unconditional presupposi tions arise &om the conditional ones. Geurts (1996) argues convincingly that these attempts are flawed. This approach is different: the conditional presuppositions occur only when Dif-Consequence can be inferred. In section 2, we observed examples where intermediate accommodation is predicted in standard theories of presupposition (e.g. van der Sandt I992), even though intuitively it should be blocked:
284 The Semantics and Pragmatics of Presupposition
(45) Nobody regrets leaving school . (46) Everybody takes their children to school (47) Most people in the neighbourhood take their children to school cases apparently all involve information in the presupposition that to resolve the underspecified elements like the B relation of section 4- 1 (example 24) generated by the treatment of the generalized quantifiers. Further, it is just these resolutions that will allow us to infer an appropriate rhetorical link between the presupposition generated in the nuclear scope of the quantification and the restrictor. This predicts that intermediate attachment is allowed in (45-47). The presupposed material in these cases describe a property of the 'quantified' discourse referent. For instance in {46), the presupposed material is that x has children. where x is the discourse referent introduced by �uybody. So we can se t B to the 'has' relation, and so relate x to y where y is a discourse referent representing a child of x, and at the same time establish the relevant thematic link between presupposed information arising &om the nuclear scope and asse rted information in the restrictor so as to establish a coherent Background relation (and FBP which we omit below for simplicity). So the representation of {46) is {46'):
These serves
u, 11'3 x, y, B, �, t person(x)
7r l :
B(e, x, z) holds(�, t) B = has z=y
Background ( 1r, u) u = 11'3 y 11' : 3
child(y) has(x, y)
7rl =>
7rl :
I
I
I take-ttrschool(x, y) I
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
7rl '
Nicholas Asher and Alex Lasca.rides 285 In the cases where intermediate accommodation is blocked, the presup
posed information is not of the right type to be identified with the contextually specified element in the restrictor of the quantifier. More specifically, it is not a property of the discourse referent that's being
So it doesn't give rise to an appropriate instantiation for the B, and thus doesn't provide the appropriate thematic Background. So some and perhaps all of the underspecified
quantified over.
bridging relation continuity for
conditions in the restrictor fail to receive a resolution. Consider for instance
(7). The presupposed material that the Pope has measles is not a property of every Catholic. More precisely, it does not involve the variable introduced by every Catholic. So the presupposed information cannot help resolve the
we explained in section 2, the presupposition should not bind globally because of Moore's paradox. Assuming that the constraints on rhetorical relations capture this constraint on assertability, global attachment will
be
blocked in SORT. In fact, local attachment will also be blocked in this way; note that (48), which is the reading given by local attachment, is also unassertable: (48) I don't know whether the Pope has measles. But for every person, if he's a Catholic, then the Pope has measles and he realizes the Pope has measles.
So
SORT
predicts that
(7)
is odd.
5·3
Multi-sentence discourse
In this section, we examine presuppositions in multi-sentence discourse. Consider (49): (49) a. John failed at school years ago. b. He now regrets that. The pronoun that in (49b) introduces a propositional anaphoric discourse referent One must resolve this to an available discourse referent of the same type.
So,
This
the pronoun
gives only one choice: the proposition expressed by (49a�
that
is identified with the content of (49a), regardless of
how the constituents attach together with rhetorical relations. Therefore,
the representation of (49a) is a, and the representation (49b) is bd for the as.Serted content and b1 for the presupposed content (in slightly simplified form):
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
bridging relation B or its other term. And we cannot get a thematic link so as to validate a Background relation between the presupposition and the restrictor; so the presupposition fails to bind to the intermediate position. As
286 The Semantics and Pragmatics of Presupposition
11"1 j, x
(a)
Jail(j , x)
1T I:
school(x) 11"� j
' 'T'
(
j , ' ja1IU, x) school(x)
)
We now attempt to update a with
b,.
1r3 , R, v ;, x (b,)
11"3 :
fail( j, x) school(x)
R=? v=?
R(v, 1rJ Defeasible Consequence applies to
1r 1 and 1r3 • And the temporal and topical constraints on Background are met, smce 1r 1 and 1r 3 specify the same content So, Updateson yields
Background( 1r 1 , 1r3 ) and Def- Constqumu( 1r 1 , 1r3 ) via DICE, and the condi tion R = ? and v = ? in bp are replaced with R = Background 1\ Def Constqutttce and v = 1r . . Furthermore, a topic 1r is built on top of this soas rr' which contains a, b, and the Background relation. The content in the topic 1r is a repeat of the content o: and ba, which is just a. And FBP( 1r, 1r 1 ) holds. This topic structUre
is important in this case, because it means that the asserted content ba can attach to the content of a, which is 'repeated' in 1r-for note that a itself is blocked from attachment by the Background relation with the presupposi tion. Intuitively, we want the asserted content of (49b) to be connected to the content (49a), because there is a causal relation between the action of failing in school and regretting it. The topic 1r lets us do this. In fact, since the presupposed content in this case is exactly that of (4�). attaching 7r1 to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
;, x
11"� :
Nicholal Asher and Alex Lascarides 287 the presupposition 1r3 would have been adequate. But this would not be the case in the slightly modified example, given in ( s o):
(so)
a.
John failed at school years ago.
b. He now regrets that awful mistake.
(49)
a. John failed in school. b. He now regrets that. ,
7r , 7r , 7rl j, s
j, x 7r :
•do well(}, x) school(x)
7rl :
FBP( 1r, 1r1)
j, x
regre{, J,'
fail(}, x) school(x)
)
Result(1r, 7r1)
7r, ' 7r3 7r' :
j, x 7r, :
fail(}, x) school(x)
j, x 7r3 :
fail(j, x) school(x)
Background( 7r1 , 1r3) Dif-Co�quena(7r1 1 1r3 )
(49) this
to van der Sandt's binding of presupposed content. In the presupposed information is a defeasible consequence of
is analogous case
the constituent it attaches to, and it attaches with
Background
and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Suppose Background did not produce a topic that 'repeated' the content of ( soa) and the presupposition (that there is an awful mistake). Then the regret would have to attach to the awful mistake with a rhetorical relation. So the regret would be caused by the awful mistake, rather than the action of failing in school and the fact that this was a mistake. By repeating all the content in the topic, new elements can attach to both the foreground and background information as required. In (49), therefore, in order to capture this intuition that the regret is caused by the action described in (49a), we attach 1r1 to the topic 1r with R�lt. Therefore, the final representation of (49) is (49' ):
288 The Semantics and Pragmatics of J>re,upposition
Dej-Cons�quenu.
In contrast,
(s I}
is an example where the presupposition
would traditionally be accommodated:
(S I }
a.
The US bombing of Baghdad during the Gulf War was much more indiscriminate and brutal than the American public was led to believe.
b. But
military
commanders
will
not
acknowledge
that their
campaign resulted in massive civilian casualttes.
(s I b)-which ledgt,
we label 7rb-contains the presupposition trigger
which introduces the presupposition
a( 1rb)
acknow
that their campaign
but
indicates that
the asserted content (that the military commanders will not acknow ledge a( 7rb}) must attach with Contrast to the discourse context (5 I a) which we label 1!"0• This Contrast imposes coherence constraints: the two propositions 1!"0 and 7rb must have contrasting thetnfi (Asher I993). In parncular, the truth of one should lead to an expectation of the negation of the other. We must check this is the case. We must also compute
a
rhetorical
attachment
between
a( 1rb) and an available a( rrb} to rra (i.e., attach
attachment site. Suppose we attempt to attach 'globally'). Then given that world
knowledge supports a defeasible
consequence relation between 1!"0 and a( 7rb) , they are connected with and Difeasible Conseq uence via Defeasible Modus Ponens on
Backgrou nd
the rules Background and Defeasible Consequence. If a( 7rb ) attaches
this
way, then the coherence constraints on Contrast between 1rb and the (modified) discourse context-which contains 1ra, a( 1rb) and a Background between them-are verified: there is a contrasting theme between a( 1rb) (which is in the discourse context} and alternattve to itsel£ But
this
this
not acknowkdg� that
a( 1rb). The
is to attach a ( 1rb) 'locally' instead of 'globally': i.e. to 1!"6
will not produce as strong a contrasting theme. Therefore,
the default preference for global attachment wins out by Maximize Discourse Coherence, and ensures that a( 1rb) attaches to 1ra rather than 1rb, because this produces the better discourse. Consider now an example which involves bridging as well as presupposition satisfaction: (28)
a.
John took an engine from Avon co Dansville.
b. He picked up the boxcar c.
and took it to Broxburn.
(28a) is represented as a, while the asserted and presupposed components of (28b) are
ba
and
bp :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
resulted in massive civilian casualties. The cue phrase
Nicholas Asher and Alex Lascarides 289 11'
j, EI , a, d, t, t, n (a)
1TJ :
John( } ) tngmt(EI) A110n(a) DaiiSVllk(d)
tJJ /u(t, ) , EI) from(t1 , a) to(t. ' d) holds(t. ' '· )
'· -< "
'IT]. e]. , t]., n pick-up (e]. , j, y) holds(e]., t1.) t]. -< n
7r3 , v, R B , u , e3 , t3 , y 'IT] :
B=? u=?
B(e3 , y, u) holsd(e3 , t3 )
boxcar( y) R(v, 7r3 ) R=? v=? According to Upda�soan we should check whether there is sufficient information in 'IT 1 , 'IT1. and 'IT3 that we can attach them together with rhetorical relations, which lead to the underspecified elements in 'IT3 being resolved. Suppose we were to try and attach 'IT 3 to 'IT 1 (or 'IT 1.) first: 'IT 3 cannot attach to 'IT 1 (or 'IT1.) with any relation-in particular Narration or Backgrou nd because there is insufficient information in 1r3 to trigger the relevant DICE axioms. So let us ignore 'IT 3 for now, and try to attach 'IT1. to 'IT 1 , to see if this produces possible resolutions ofunderspecified elements in 'IT3 .:w And indeed, it does. 'IT 1 and 'ITJ. both describe events, and so Defeasible Modus Ponens on the axiom Narra tion yields Narration(1r1 , 1r1.). We must then accommodate content arising &om Narration's coherence constraints. First, trr. occurs before t1r. : that is, the taking of the engine &om Avon to Dansville occurs before y (which is the boxcar according to Krr) is picked up. Second, by modus -
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'IT]. :
290 The Senunncs and Pragmatics of Presupposttion p<>nens on Spat ial Consequence on Narration, the actorJohn is in the same place at the end of taking the engine to Dansville as when he starts to pick up y. By the prepositional phrase in (28a), this is Dansville. And by the lexical semantic content of the phrase picking up, this event starts in the same location as it finishes, and the object y that's picked up is also at this
11', 11'1 , 11''
11'1 :
j, EI , a , d, t 1 , t1 , n
y, t1. , � , t3 , t3 , n
John ( j) tngint(EI ) Avon (a ) Dansville(d ) t, -< n
puk-up (t1. , j, y) holds (t l , t, )
talu(t, j , EI ) .from(t. , a ) to (t, d ) holds( t. , t, )
1!':
boxcar (y) in (t3 , y, d ) ovtrlap (t1. , tl )
Narration (11'. , 1r) 1!'1.1 1!'3 , v, R
11'':
pick-up (tJ. , j, y) holds (t, t, )
R(v, 11'1 ) R = Badeground V = 1!'1
-<
n
holds(tl , tl )
FBP(11', 11' 1) B, u , t3 , t3 1 y
t1. , t1. 1 n 11'1. :
t,
'lrl : t, -<
n
boxcar (y) B = in u=d
B(t3 , y , u) holds(tl , tl )
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
location. Therefore, y is in Dansville. Thus, the coherence constraints on Narration provide a particular way of resolving some of the underspecified conditions in the presupposed informa tion-viz. B resolves to in, and u to d or Dansville. So the definition ofUpdatemn ensures that B and u are resolved this way. So the content of bp now contains sufficient information for us to compute a rhetorical relation between it and the discourse context: since B has resolved to in, e3 is a state. Moreover, attaching rr1. to 7r1 with Narration has made only 1r1. available for attachment. Therefore Defeasible Modus Ponens on Background yields Background(rr1 , rrJ The temporal constraint on Background means that err, and e1r1 overlap (ie. picking up the boxcar and the boxcar being in Dansville temporally overlap). This is as required. Moreover, a topic is computed by creating a new constituent rr with all the content ofba and bp repeated in it, and this is related via FBP to the SDRS rr', which contains ba, bp and the condition Background(rr1., rrJ Because this new topic rr gets built over the soRS rr', the original constituent rr1 that was attached with Narration moves up to be attached to 1r (for details, see Asher 1993 and Asher et al. 1 996). The final result is given in (281):
Nicholas Asher and Alex Lascarides
2.9 1
(4) Either John didn't solve the problem or else Mary realizes that the problem has been solved.
The presupposition that the problem has been solved cannot bind anywhere coherently, because unresolved bridging relations would remain, which are triggered by the definite the problnn. So sentence (4) is predicted to be odd in this null context. However, in the appropriate discourse context, (4) is coherent: (6)
a.
A: The Problem Solving Group is given a problem each day, and the group leader Mary has to assign it to someone in her group.
John is the best problem solver. But when he solves a problem, he always boasts about it. This annoys Mary, and so if she thinks that the day's problem is an unsolved one, she gives it to him, to test him. Otherwise, she gives it to another member of her group. b. B: John's being very quiet just now. Did she give him today's problem? c. A: Well, I'm not sure she did. Either John didn't solve the problem or else Mary realized that the problem's been solved. The overall structUre of this example is that (6b) is a response to (6a), perhaps some sort of Com�t4ry, while (6c) is a response to the question in (6b) (we'll ignore the internal structUre of (6b) here). The disjunction in (6c) offers two, contrasting Explanations to the response fm not sure. The explanations explain why John is behaving quietly (i.e. why (6b) is a fact). How do the presuppositions fit into this StructUre? A sets up a relationship between Mary and John in (6a); Mary gives problems that she believes are unsolved to John, and problems that she believes are solved to the others. (6b) refers to a problem (today's problem), to which the presuppositions triggered by the definite � problma in (6c) can bind. Note that (6b) is available to these for binding, because (6c) is a response to (6b).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Note that we predicted in this analysis that the presupposition is locally accommodated to 1r�: 1r� must be attached first because 1r3 is not informative enough, and since the rhetorical relation is Narration, 1r, is blocked from future attachments. This predicts correctly that the engine in (28a) cannot be an antecedent to a pronoun in a continuation of (28): e.g. ?John drow it quickly. However, John drove the wgine quickly is a much more acceptable continuation. And our theory reflects this: t� engint requires a bridging relation, and computing the bridging relation is possible, by adding to 7r� (and hence to the topic 1r) the content that the original engine in (28a) (to which 1r� is linked) is used to pick up the boxcar. Let us now return to an example which was problematic for van der Sandt's account:
1.92 1b.e Semantics and Pragmatics of Presupposttion So
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the bridging relation and antecedent are resolved to identity and today's problem respectively. The second sentence in (6c} triggers two further presuppositions. John didn't solvt the problnn presupposes he got the problem, and Mary rtaliztd that tk problnn's bun solved presupposes that the problem has been solved. Note that given the content of (6a), this second disjunct also implies, at least defeasibly, that John did not get the problem. Let us look at the binding possibilities for these presuppositions. Binding John got tk probkm locally with Background allows the coherence constraints on Contrast to be satisfied: the first disjunct now implies that John got the problem, and the second implies that he did not. It also ensures that the second sentence in (6c} does provide two contrasting Explanations to the answer in the first sentence of (6c): either (i) John did get the problem because Mary believed it was unsolved, but he could not solve it, or (ii) he did not get the problem because Mary believed the problem already had a solution (note that rtalizt entails bel�ts). In effect both (i) and (ii) allow us to compute answers to B 's question as to whether Mary gave John the problem: (i} implies that he got the problem (and presumably that she gave it to him); (ii) implies that she did not. They both also offer contrasting explanations for what happened yesterday, and their disjunction explains why A is not sure about how to answer B's questiorL Now consider what happens if we attempt to bind the presupposition that John got the problem globally. In that case, the explanation for the first sentence to (6c) becomes: John got the problem, and either he did not solve it, or Mary realizes that the problem has been solved. This utterance implicates that John got the problem, but from someone other than Mary, because the second disjunct implies that Mary didn't give the problem to JohlL This utterance would form a coherent explanation for John's quiet behavior only if it is interpreted a5 follows: John got the problem from someone other than Mary, and either he did not solve it, or he did, but because Mary realized the problem had been solved and so John did not get the problem from her, he didn't boast that he has solved it. Thus, there are more assumptions required to support this explanation than for the local binding case. Intuitively, the explanation for the local binding case is stronger than this one. But then our principle of Maximize Discourse Coherence predicts that the presupposition ofJohn didn't solve tk probkm is bound locally. Now consider the presupposition that the problem has been solved. According to the discourse structure, tk problnn 's bten solvtd could bind locally to Mary reali.us that the problem's bun solvtd, to the first sentence fm not surt sk did in (6c), or to the question to which (6c) is a response, namely the utterance (6b� In this case again, the constraints on discourse relations
Nicholas Asher and Alex Lascaricles
293
locally in these theories in virtUe of the conversational implicatures supported by the context, at least in any straightforward interpretation of Gazdar's views. Given what A said in (6a), A and B would know the answer to the question as to whether John got today's problem, if they knew if Mary believed it was solved. This context implicates that A and B do not know if Mary believes the problems been solved. However, as we explained earlier, the context does not implicate whether or not A and B themselves believe that the problem has been solved. And so these implications on their own do not determine whether the presupposition should be bound globally or locally. Hence, the coherence constraints on Contrast, and the preference for interpreting discourse in a way that strengthens rhetorical connections, both play a crucial role in interpreting the presuppositions in this example. This demonstrates that constraints imposed by discourse structure are in general stronger than informativeness and consistency,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
and discourse update in soRT determine a preferred attachment site, and once again Maximize Discourse Coherence plays a crucial role. The Contrast between options (i) and (ii) is strengthened if the presupposition of reali.u binds locally to Mary realizes tM problml's been solved with Background, rather than being bound globally. To see this, consider the interpretation of the discourse under local binding, vs. that given by global binding. The 'local' interpretation is: Either (i) John got the problem because Mary believes it was unsolved, but John didn't solve it, or (ii) John didn't get the problem because the problem has been solved and Mary realized this. The 'global' interpretation is: the problem has been solved, and either (i) John got the problem because Mary thought it had not been solved, and he failed to solve it, or (ii) John did not get the problem because Mary realized that the problem has been solved. In the local case, Contrast relates two propositions where one contains the content that John did not solve the problem and the other contains the content that the problem has been solved. But the 'global' case does not relate two propositions with this content. Rather, the Contrast in the global case is supported only by the contrast that is there for the local case too, between John getting the problem and not getting it. So, the Contrast relation is better in the local case than in the global one. Therefore, Maximize Discourse Coherence yields the prediction that the presupposition is bound locally in this discourse, rather than globally. In contrast to van der Sandt ( 1992) and to Gazdar ( 1979), we have predicted local attachment in (6), because of soaT's constraints on discourse relations and discourse update. The context in (6) does not implicate that the problem has been solved; nor does it implicate that it has not been. And therefore one cannot determine that the presupposition should be bound
294
The Semantics and Pragmatics of Presupposinon
and indeed go beyond constraints provided by conversational implicature alone.
5 ·4
Presuppositions, cognitive states and global accommodation
We have sketched already in section 5.1 how the constraints on Background rules out many cases of infelicitous accommodation: the presupposition to be accommodated and the global attachment site do not have the requisite thematic continuity. However, what appears as global accommodations often seem felicitous. Consider the following from Vallduvi (I990F a. A:. I bought the President a tray for his Delft china set. b. B: That wasn't a good idea. The president (hates}F the Delft china set. (from Vallduvi 1 990)
We claim that in ( 52) the presupposition introduced by the focussed phrase haw is bound by means of Background to a belief of the interpreter about the speaker. That is, the presupposition binds the presuppositions to what one reconstructs about the speaker's cognitive states. To reconstruct the Elaboration or Explanation given by B 's response to A, we need to introduce a wide range of assumptions about what B thought was behind A's utterance. Ordinary theories of focus linked with a theory of presupposition (e.g. Rooth 1992) claim that some element of the alternative set for the focussed expression is presupposed. But the standard view of the pre suppositions of focus would in this case simply accommodate this element, when the presupposition is in fact bound to what B reconstructS about A's mental state from what A said.�' The possibility that presuppositions can bind via Background or even Dtj-Constqun�a to an interpreter's beliefs to be sure puts some slop into our discourse based account of presupposition. Nevertheless, in contrast to the standard account, we predict that such so called global accommodations obey more constraints than consistency and informativity. Our approach would in principle distinguish between ( 5 3) and (54) (Asher & Lascarides, in press):
( 53 ) The present King of France does not exist. (54) ?The man does not exist.
In contrast to ( 15 ), &clcground cannot hold between the asserted and presupposed content in ( 53 ), because the two states of the king existing and not existing cannot overlap. We claim that Background holds in (54) between the presupposed information and the cognitive state of the hearer, who is familiar with the concept, the King of France. But as (54) shows, this
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 52)
Nicholas Asher and Alex Lasca.rides
295
strategy is not available in all cases. With (S4) there simply isn't enough content associated with the description to provide the necessary thematic continuity and to bind the presupposition to any particular concept via
Background.
But even in cases where this strategy isn't available, it appears that sometimes presuppositions can be introduced felicitously in information that is 'discourse structure initial' (i.e. is simply added and not attached to an empty discourse context). Consider the first sentence from Austen's
and &nsibility:
&nst
The family of Dashwood had been long settled in Sussex. Their estate was large. and their
Austen knows that there is no Dashwood that the reader knows about, but she fills in the scene during the course of the novel. There is presupposition failure but the reader does not really care. He expects the author to tell him more later. Presupposition in such cases is not anaphoric but cataphoric upon what will come next. The licensing conditions for such cataphors is a matter for future research.
6 C O N C L US I O N We have offered a new picture of presuppositions: we have modified van der Sandt's view that they are anaphoric by loosening the notion of binding, thereby foregoing the need for accommodation. We have demonstrated that there are three main advantages to processing presuppositions through binding them to the context with a rhetorical relation. First, accommodation by addition as in van der Sandt ( I 992) fails to account for the interaction that many have observed between the interpretation of presuppositions, domain knowledge and the content of the discourse context (e.g. Beaver I 996, I 997; Lascarides & Oberlander I 993 ). In contrast, binding material together with rhetorical relations involves reasoning about domain knowledge and semantic content, and so in essence, we have provided pragmatic constraints on presupposition satisfaction that is lacking in current dynamic accounts of presuppositions. Second, through taking rhetorical constraints into account, we model how the semantic content itself of a presupposition is dependent on pragmatic information (e.g. ( 1 0) vs. ( I I )� and we model how the level of accommodation (i.e. global vs. intermediate vs. local) is also pragmatically determined. The result is a more fine-grained analysis of presuppositions
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
residence was at Norland Park, m the center of their property, where for many generanons they had liVed in so respectable a manner as to engage the general good opinion of their surrounding acquaintance.
296 The Semantics and Pragmattcs of Presupposttion been offered so far. Finally, and arguably most importantly, we have reduced the problem of presupposition satisfaction to the problem of resolving scope ambiguities-a phenomenon that arises in many other areas of linguistic study.
than h2s
Acknowledgements This work u sponsored by a granted funded by BSRC UK: {grant number Rooo2 36o52� and
by the I!Sll.c -funded Human Communication Research Centre, Univenity of Edinburgh. We would like to th2llk. Davul Beaver, Hans K.amp, Rob van der Sandt and an anonymow revtewer for helpful comments on previow drafts. Recetved: I 2.02.98 Final venion received: o8.02.99
ALEX LASCARIDES
Ontrt for Cognittw Selma and HCRC Univmity of Edinburgh l, Buccleuch Piau Edtnburgh EHB 9LW Scotland, UK t-matl: akx®cogsci.ac.ult N OTES
I We are endebted to a reviewer from the
Journal of&mantics for pointing thU out to us. 2 We could make this notton of informa tiveness more precise in a w.1y proposed by Michael Morreau: any accommoda tion of a presupposition if> m a context C of a sentence ,P mwt not make ,P in C logically equivalent to x. such that X I= if> but it's not the cue that ¢J I= X· 3 For more on too and Paralkl, 5ee Asher {I993) or Asher, Hardt & Busqueu
( I 997).
4 In
more complex examples, we may fonn local soJt.Ss that are then atuched
5
together. The idea of handling presupposition
5atis&ction through computing haw it
u rhetorically connected to the dis course context was used in Lascarides & Oberlander {I993) to process the content of wlwl -, b4Qr?- and a.Jtn clauses. This approach is similar m spirit, but more general We have replaced therr procedural approach with a declarative one; we have integrated sernannc and presupposi tiona! information into the constructton of an soilS in a compositional way by wing underspecification; and finally, we cover a wider variety of presupposition triggers. Furthermore, Lascarides and Oberlander ignored the way computing a rhetorical relation can lead to further inferences about semantic content, and so they did not model the w.1y the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
NICHOLAS ASHER Dq1artmnll of Philosophy Umvmity of Taas at Austin Austin, To:as 787Il USA �-mail: naslm@bm�.la.ut=s.edu
Nicholas Asher and Alex Lascarides 297
Narration because, as we have noted elsewhere (Asher t.c Lascarides I995� it does not apply for all discourse rela noru. FOr example Contrast and Paralkl have different spatial comtramts, as given in Asher t.c Lascarides (I995, in press� I 2 For sirnphcity, we have written this rule m words, although it can be encoded wing the modal connective > . I 3 This kernel ensures that only the rela tiom m the context change, and not the structure.
I 4 Note that this definition holds only for DRSS without cond!noru of the form :x = ? . We gloss over the complications that occur when we countenance such condinoru. They are not relevant for our purposes, since we resolve condl noru such as these in different ways &om those in on. 1 5 This is a slightly simplistic analysiS. For example, we have ignored the presup positional content of proper names such as Frana, and we have also �gnored any urnqueness conditions that might form part of the content of 1k Kmg ofFrana. 16 Of course Russell's an.alysis of (15) IS different from ours and van der Sandt's because ofh.IS uniqueness condition, but we won't go into that here. I 7 There are many ways one could develop this notion. For instance, one could tentatively identify the plausibility of a conditional plausibility of the conse quent upon the antecedent. If so, we'll also want to assume that propositions can have a relative or conditional degree of plausibility PL(IJ I a) , JUSt as they can have conditional probabilities (in fact plausibility could be interpreted as subjective probability, though we will remain agnostic on this wue� On the other hand, one might explore a modal approach to comparative plausibility; PL(a) >- PL(IJ) iff there are some a worlds that are closer to the expected worlds (given the content of the dis course so far) than any f3 worlds are. But
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
content of the presupposition iaelf is determined by context (e.g. (Io) vs (I I}). 6 One could take an even more extreme posinon, and suggest that presuppo sitiom are extra inferred knowledge, which strictly spe2king doesn't comprise part of the semantic content of the text at all. & Gazdar (I979) and Mercer (I985) have argued, they are irnpli catures of some kmd. We reject thiS extreme position. Treating presuppoSI tiom as part of the semannc content allows one to record how it affects the mterpretation of other expressions in the text, such as pronouns (van der Sandt I992). 7 We ignore here and below the con troverSial wue as to whether definites introduce a uniqueness requirement (c£ Russell I905)· 8 Indeed, the prinapal advantage of usmg son for mvestigatmg presupposinon over other theones ofdiscourse structure (Hobbs tt al. I993 ; Grosz & S1dner I986; Polanyi I985; and others) is that son has already explored extensively the mter action between compositional semantic mforrnation extracted from the grammar and p�tic inforrnanon. 9 Since our main concern is modelling presuppositions, we will ignore pro blems of quannfier scope and concen trate only on the different possible 'scopes' of presupposed and asserted information. IO Asher & Lascarides (in press), however, ignored spatio-temporal conditions on when the bridging relation holds. We recnfy this here by subsuming the bridging within our general treatment of presupposinoru. We demonstrate that resolving the rhetorical connection between the presupposed content and the discourse context introduces spano temporal constraints on that content, which in turn impose the spatio temporal conditions on the bridging relation B in the required way. I I This spatial constraint applies only to
298 The Semantics and Pragmatics of Presupposition agam to pursue this discussion would
lead us too far afield. I 8 In general we would like to thank David Beaver for the examples that led w to thmk about conditional presuppositions generated from the antecedents of conditionals. I 9 Given this :urumption, there is a logical dependence between the statements David wrote 1M arti& and no good Wglcian wrote 1M article. And so the general rules we've stated about projection, which
arise from Update..,... and the plau siliility axionu, predtcts that the pre supposition doesn't project from the embedding. 20 Note that clauses (2e-d) and (3c-d) in Update� allow content rp, that is inferred by attaching 1r2 to 1r, , to affect the resolution of undenpecifi cation in 1r3 to occur, because 1r3 will attach to 11'2 or 1r. . 2 I For some details as to h ow this story should go, see Asher (I995�
Asher, N. (I993 � &jtrtna To Abstract Thtory (SALT VII� Stanford, CA, March Objtcts in Ducou�. Kluwer Academic I997 . PublUhen, Dordrecht. Asher, N. & Lascarides, A. {I994� Asher, N. (I995), 'From discourse micro 'lntenttoru and information in dis structure to macro-structure and back course', Procttdmgs of tk JZM Annual again: the interpretation of focus', in Mutmg of 1M A.!soaabon of Computational Lmguistics, NM U5 Cruces, June I994. H Kamp & B. Partee (eds.� Ccnfmnus on &mantics and CcnUxt, Prague and 34-4 1. Bad Teinach, SFB Report, University of Asher, N. & Lascarides, A. {I995� 'Lexical Stuttgart, I995· disamlnguation in a discourse context', Asher, N. {I996� 'Mathematical treat journal of &manbcr, u, I, 69-- I08. ments of discourse contexts', in P. Asher, N. & Uscarides, A. {I998� Dekker & M Stokhof (eck� Procffding.s 'Questions in dialogue', Linguistics and of tht Ttnth Amsterdam Colloquium on Philosophy, 21, 23 7-309. Formal &mantus, IUC Publications, Asher, N. & Lascandes, A. {I 998� 'Bridging', Uruversity of Amsterdam, 2I-40. journal of&mantics, I S , 83-1 1 3 . Asher, N. (forthcoming), The logical foun Asher, N. & Morreau, M (1991� 'Common danoru of discourse interpretation', in sense entailment a modal theory of J. M Larrazabal {ed.), Logic Colloquium non-monotonic reasoning', Procffdmg.s 19¢, Springer Verlag. Berlin. to 1M uth International joint Ccnftmta Asher, N� Aurnague, M, Bras, M & on Artifiaal Intelligtna, Sydney, Australia, Vieu, L {I996), 'De l'espace-ternps August 1991. daru !'analyse du discoun', &mwtiqut: Beaver, D. {19¢� 'Presupposition', in J. Numiro Sp«Ud: 11tio,us simantiq� tt van Bentham & A. ter Meulen (eels.� modaliSIUUm , 9Handbook of Logic and Linguistics. .Asher, N. & Fernando, T. (I997� 'Effective Beaver, D. (1997), 'Presupposition in DRT', labeling for disambiguation', Procffding.s SALT VII, Stanford, CA. of 1M &cond lnttrnotionol Workshop in Chierchia, G. (1995� Dynamics of Mtaning: Computational Linguistics, TJlburg, The Anaphora, Presupposition, and 1M 'Ikory of Netherlands. Grammar, University of Chicago Press, Asher, N., Hardt, D. & Bwquets, J. (I997� Chicago. 'Discourse parallelism, scope and ellip Clark. H (1977� 'Bridging', in P. N. sis', Proatdings ofStmantics and linguistic Johmon-Laird & P. C. Wason (eels.�
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
RE F E R E N C E S
Nicholas Asher and Alex Lascarides 299
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Mmin, P. (1993� 'Interpretation as abducnon', Artificial Inulligma, 63, 69- I4J.. Kamp, H & Reyle, U. (1993� From Dis cou� to Logic Introduction to Model tluomic Snrulntics of Natural Langua�, Formal Logic and Discou� � 'Jkory, Kluwer Academic Publishers, Dordrecht. Karttunen, L (1974). 'Presupposition and linguistic context', 'JkomJaJJ Linguistics, I, 1 8 1-94Lascarides, A. & Asher, N. (1993� Temporal intetpretanon, discourse relations and commonsense entailment', LmguutiCS and Philosophy. r6, 437-93· Lascarides, A.., Copestake, A. & B nscoe, E. J. (19¢), 'Ambiguity and coherence', Journal of Snrulntics, 1 3, 1, 41-65. Lascarides, A. & Oberlander, J. (1993� Temporal connectives in a discourse context', Proadmgs of 1M Eu�n Chapkr of 1M Associa tion for Compu tatJonal Lmguistics (EACl-93� Utrecht, The Netherlands, 26o-8. Lewis, D. (1973� Countnjactuals, Blackwell, Oxford. Lewis, D. (1979� 'Scorekeeping in a lan guage game', journal of Philosophical Logic. 8, 3 39-59· Mercer, R E. (1985� 'A Default Logic Approach to the Derivation of Natural Amherst, MA. Heim, L (1983� 'On the projection prob Language Presuppositiom', Ph.D. disser lem of presuppositiom', Promdmgs of tation, University of British Columbia. tk West Coast Conf� on Formal Mercer, R E. (1988� 'Solving some persis Lmguistics, 2, 1 14-26. tent presupposition problem', Proatdings of COLING 11}88, 420-5· Heirn, L (1992� 'Presupposition projection and the semantics of attitude verbs', Moore, G. E. (1912� Ethics prm, London. journal of &mantics, 9, 3· Muskens, R (1995� 'Order independence Hein.am.ah, 0. (1972� 'Semantics of and underspecification', in Dyana-2 EngliSh temporal connectives', Ph.D. Dehverable R2.2c: EUipru, Undmp«i}lal dissertation, University of Texas at tion and Evmts in Dynamic Stmantics, Austin. available from lllC, University of Hobbs, J. (1985� 'On the coherence and Amste rdam. structure of discourse', Report No. Polanyi, L (1985� 'A theory of discourse CSU-85-37. Center for the Study of structure and discourse coherence', Prlpm fram 1M Gmn-al &ssion at 1M Language and Infornution, October Twmty-First &gional Mttung of tht 1985 . Chicago Linguistics Soddy, 25-7. Hobbs, J. R., Stickel, M E., Appelt, D. E. &
Thinking: &mLngs m Cogniuw &mla, Cambridge University Press, Cambridge, 41 1-20. Fernando, T. (1994� 'What is a DRS?', in R Cooper & J. Groenen.dtjk (eds.� In�ng &mantle 'IkorW II, Dyana fHu&JmJbk Rz.I.B, abo appeared in the Proceedin� of the First International Workshop on Computational Semantics, Tilburg 1994von fintel, K. (1994� 'Restricnom on quantifier domains', Ph.D. thesis, Uni versity of Massachwsetts, Amherst, MA. Gazdar, G. (1979), Pragmatus, Implicaturt, Prtsuppontwn, and Logical Form, Academic Press, London. Geurts, B. (1995� 'Presupposing', Ph.D. dissertation, Universitit Stuttgart. Geum, B. (19¢), 'Local satisfaction guaranteed', Lmguistrcs and Philosophy, 19, 25�3· Groenendijk, J. & Stokho� M (1991� 'Dynamic predicate logic', Linguutrcs and Philosophy, 14. 39-100. Grosz, B. & S1dner, C. (1986), 'Attention, intentiom and the structure of dis course', Comput.Jtional Linguutrcs, rz, 175-20.f.. Heim. L (1982� The semanncs of definite and indefinite noun phrases', Ph.D. thesis, University of Massachussetts,
300 The Semantics and Pragmatics of Presupposition
Pollard, C. & Sag. L A. (1994� Head Driwn Phrast Struawrt Grammar, University of Chicago Press, Chicago. Reyle, U. (1993� 'Dealing with ambigwnes by undenpecification: construction. represenation and deduction', journal of&mantics, 10, I, 123-79. Rooth, M (1991.� 'A Theory of Focus Inter pretation', Natural LAnguagt &mantics, I.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'On denoting', Mmd, 14. 479-93· van der Sandt, R (1991.� 'Presupposition projection as anaphon resolutton', Journal of &mant�a, 19, 4-
Rwsell, B. (1905�
Stalnaker, R (1974 'Pngnunc presuppo sitions', in M K. Murutz & P. K. Unger (eds.� &mantics and P#ulosophy, New York University Press, New York, 197-21 3· Stalnaker, R (1978), 'Assertion', in Peter Cole (ed.� Syni4X and &mant�a, VoL 9: Pragmatics, Academic Press, New York. Thompson. S. & Mann. W. (1987� 'Rhetorical structure theory: a frame work for the analys15 of texts', IPRA Papm in Pragmatics, I, 79- 105. Vallduvi, E. (1990� 'The infornutional component', Ph.D. thesis, Univemty of Pennsylvania, Philadelphia.
