Jouna~ of’ Pragmatks l1 (1 HO) I- 13 0 North-HokmlPlrblishing Company
PRAGMATICS
AND LQGfC’r\i,
FO
SLIPCC the ;nru’dIr:tit’ the lc)~~Oslinguists have recc~n.izJ that ,in important goal 4 t iltf Jls~iplrnc is the provisi an 13f a comprehen:~ve 3.plicit theijry of meaning for I1Jt lml Isr&!gu :rgcL Just ia the syllti ctic componel.7 7. -he overall linguistic theory wr~lcf recursi+zly enumerate the we’Mormcd striirl.. 8x,u.sentences) of the language, SLIfhr. nlea~~i~qzj 20mp0nent w~iu!~Ir.xursivcly 3ssqr’, r.lcanings to the sentenres of t hl: language u Within m:&ernatical logic, theories of mea@ iw the formal languages emplq~d by logicians had been around tire many years: tru &functional set rantics for the ~mposjt~md cakulus had been aI.8ailablcess( ntiall> since Boole, referential semantics tk first 0rJer prcdxate c&uIus since FI :g6, an\’ possible world semant~s ILr mod 11,CY&sinse the wu~k of’ Krr#e and &ers in rhc 1950s. What these semanti: theories have in colnmon is a ~trcern wit ‘7 VZ.U~CCr&tiorzs: they assign recursMy t1.jeach sentence LIP’tke logicall language the conditi*,nls under whizl: the senterrcl: would be true. Art~ur~_IIWO lhi: rued fjr a theory ot’ meanings,within IinguMcs liaid become impctutivz. ‘i’trconly theory that had been dcveLcrpisc.within the tra,l.formationalist paradigm (Katr. and Fodor i9363) W;;LS no longer !ak :n srtric usly, and \-et issues cru* I am grateful to Steve Lcvinson for some help. A Germa: transhltion of this pa:jer appeared under the titk Une pragmatisch-serwntische Mischttieorie c:er Bedeutung’ as Cardx” (1978). 1
C. Cordar / Pragmal’icsand logical form
2
ciaIl,q concerning meaning were among tt.e most divisive and conzentious of the pcr!vd. At the same t ,me, influential philo:;ophets of language (e.g. Harman, Davidser.) and philosophical logicians Ct.g. Mont;lgue) were claiming tha: there ws “no ‘~21 difference between natural language and tl:e artificial Ian. important theoretl,,, gL,agesof logicians” (Montague l(~70: 222). if this claim co&l be maintained then the moral was obvious: the thwries of meaning developed by logicians for their formal langclages could be appropriately applied to natural lallguage;. III the 1970s there has been the most remarkdblt: confluence between, on rhe *,>nehand, those linguists who had been propoGng increasingly “logic-!ike” un&tlying (or semantic) representatifons for natud language sentences, and, on :hc other, the philosophers of language and philosophical lo@cians mentioned abow. Tlie semantic representations (often identified with the h,#cal fht?ls) uf nzlt ural language sentences are now widely held to he sentences 01’a f~~rnlalIog~c for 4uch a recursive t rutli-conciitiollnl theory of nicaniq: is &-iincd. A natural cot\~llq. cl!’ this view of semantic represenPation is thy need to idcrrtilk particular weds cd ttrc natural 1:mguage wit)1 certain of the logical constants of the formal langurrgc. t+r example: and and or are treated as !k?c~lea~~ truth-functions, some and all arc giwn a hegem semantics, and posibly is iden!itlt?d wIti the possibility oper;ttnr of Kri;)ke’s modal logic. l_er me suniniarlze the position m6M ained by many scmantic’;llly ofiWed linguists and philoql lers in t 1~ I VOs:
2. Semantic problems
( :) (2)
John stole the money ant went to the bsnk John went to the bank ar:d stole the money
intultiv4r (1) and (2) do not rnt)m the same thinp: the order of she conjuncts rti3kes ;i diff~rcnze to Ge way we un&:rstand the sent me. And yet thI: counterpart of and in iogi~~the tmth-function n. gill not cq.~ture the difference: in logic P A Q means the :LU~IC z, Q n P, PI different pMlem mses wfh or as tlw follm?ng cxatnplcs hmonstrate:
G. Gazdar / Pmgnratics and logical form
4
(14) I believe John met a wornar. last night One interpretation
(the so-called sgecific interpretation)
can be pdraphrascd as
(IS):
(15) There is a woman such that 1 believe of that woman that John met her last night ‘The other interpretation phrased as (16):
(the socailei! r~nspccfic
interpretation)
clan Fe pqra-
(16) 1 bebevz ihat t.,ere is a woman such that John met her last night Sentence (I 4). of itseif, poses no problem for the position WCarc cunsiderinl; sinct .A,” linguisr can proGde for the ambiguity by reversing the [incar artier (and hcnc,‘c P-C the scope) of the quantified phrase (Lc. a woman) atx-! the vccb krf prqx‘rsitlc~ll~i attitude (i.e. belkve) in the two semant:c reprcsentntluns assigned to (14). ill the .L.._Ll,_.__.l.. ..Y ._..... _L bh.n palcaprrraJsb W-r-..L..,.nnn[I/I JJc\ LlllU ..-A /r L\ Tl”.s ~..n..nnln?4 _#-b..“.3_ 1110111151 3U~~GJrcuLb., uy LIIC \ 1 uj. 111c: ~JIUlJlCIII CJlily ClllC!&;cb when we consider sentences like ( 17): (I 7) John met a worllan last night This sentence ap;xrs to manifest exactly the ambiguity of (14): the spe;&er my have a specific wLbmanin mind or he may mer4.0 bc ckkning that Johir had unspcuiW filed female co.nqmy. But the ambiguity, it’that is what it is, ot’( 17) r’;u:not tll*C;II~tured in terms ol’ a difference in scope irl tl~ scllurrtic xprCsCnt;rtiun situ! I~C:UC I:nothing irr rhat repltisii;;Jtion which would allow the’ q~;ultitkd phi XX to h.r~ two scope possibilities. ?Vorse still, the ambiguity tends to dis~ppci~r whr‘si ~vt* change the subject of the sentence from third persc!ll t 1) first ~LVWI~.Thus I~C WImarked interpretation \1i’ ( 18) is the specific oni‘ : ( 18) I met a woman last night
(14’1 John regrets having failed the c.. ~1 (20) John doesn’t regret having faile~l tht- t’MII~ (2 1) John has f&d the ex.:m
tic representation d the sentence, taken :ogether with the theory 01 truth for the language 0f semr~~tii: representation, would make s-xqht forward predictions respecting the prcserxe of those presuppositions. This go3.l turns out to be inipossibk to achxuc for LInumber of reasons. An initial dif&xlty concerns negation in natural language: in order to capture the relation between (19) and (20) negation has to be defined cm way, but in order to xlOl.mt for thtl acceptability and nonconliradictor=~ rmtrm of( 22). rqytiurr has to be defined in a q&c: different way:
G. Gazdar / Pragnwtics ad !ogicallform
6
Romantic representations but tire The grammar will map (29) and (30:~into dhsc! rheor-y of’ t mt.1 will predict that they have th,e same meaning; sixe both are tautol= ogies. both have the same truthconditions, namely they are both true under an) ~ir~ums~~~~es whatsoever. But intuitively (29) and (30) mean different thjn:;s. So muzh lhe worse, apparenrly, for the position sketched at the end of the first seLii4Jll.
1~ i ‘10i the philosopher Paul &ice gave the Wilkm James lectures at Harvard JJll tfitl t3piC hLOj$C and conversation’. Tfiese kctures, especially the one subs+ qtc;-d~ pubii~kd (under the same title) ASGrice (197.55 have provided the ba.ck of tf;~ i;G’or)i :ha; haj beers done on ljragr,atics k,y ling-uists during ihis !c;r a 1:;: ufx~cle, Much of this work. ilke Grice’s own, has been geared to providing so!uGons trj probknss Ike those nokd in the previous section. The central theme of this 2.i,l k i'J ?ikhZ cme cm sdvage the logician’s truth-conditis.;al account of meaning f~ I-Q).UU~ language if one augments it with a theov abou; those aspects of meanring{c&d ‘Pri~gInatiCS’)which arise through the actual us? of laquage for communrca~ucrn The ‘Jnitary hnonYAithicmeaning component of the grammar assumed by irnguists in the 1960s has been replaced by a highly restrictive semantic (i.e. truthccx~Monal) wmponent operating in tandem with a pragmatic component which ~prcifies *Ahat 3 sentence will Convey on an occasion of utterance. Crrice proposed that conversationalists standardly conform to certain maxims c~Jlrci-fr,rrr~ the conduct of discourse. Here are some of the maxims he proposed: (3 I 1 & wderly!
(32 I Make your ct2ntribution as informative as is required! !‘,73j Bz re?ev;nt! CJ1 I uan be understood to stipulate that if certain Items or events have a spatial, temporai or c~.~l order, then in describing those items or events o3e should reproduce !hit order in the sequence of one’s description. Thus if I wish to report the sslen:s t.exrIbed Oy (34) and (35) and if l know that the event described by (34) ~rsce&~I the event described by 135) then 1 sholtild utter (36) or (37) rather than IMJ w (39) if I l am to obey (3 1). And conversely.
G. Gazdur / Pragmatics and logical form
7
(37) Jcehn stole the money. He went to ths bank (38) Jchn went to the bank and stole tlx money [= (:!)I (39) John went to the bank. He stole the money We now have a pragmatic explanation for the difference in meaning between (36j and (38), one which equally explains thle parallel meaning difference between (37) and (3sj and one which is quite in lepe!ldent of the word and (sines alnddoes not occw in (37) and (39)). This explanaticn thus leaves us free to trea’t arld semantically as a truth-function. Some words are inherently more informative than other words. That is, if, in constructing a sentence, 1 chcose to use (3nl:rather than another, I may thus convey to my addressee more about the wc~ld. For example, and is more informative than or. Suppose my addressee has been inqurring into John’s character, then Bcan reply with (40) Ur (41): (40) John is patriotic and quixotic (41) John is either patriotic: or quixotic [= (3)] ‘2 _; 1’ ti 40es. Clearly t40) tellls my addressee more about John than (-I _ In such insit u7,tii,n, if I know (40) to be true but only utter (41) in response io my addressee’s inquiries then 1 shall be violating Grice’s maxim (32): ‘Make your contribution as inforrnative as is required. 1’ So if my addressee hears me reply with (41.), and if he ;Lssumcs that I am conforming t:ll the maxims whicll govern conversation:;, then he can infer at least :!lat i do not krlow (40) to be true, or even, more strongly, that 1 blelieve 149) to be false. And if ?~einfers the latter thell we havt: an explanation for why disjunctive sentences like (41) are often heard 10 mean thcl same as (12): (42) John is either patriotic or quixotic, bu!. not bcrI or I= (4)] Thus the linguist ctilr treat or within his sem;mtic theory as unambiguously i& ntical to the logicitin’s inch&e disjunction., and explain the exclusive sense it often seems io have by reference t 9 the Gricean maxim within his pragmatic theory. Dther pairs of words related in a similar way to and and or include necessarily and possibly, must .znd may, and all and some. In each case the first of these Fairs is more informative than the second. Consider the following exampks: (43) (44) (45 j (46) (47) (48)
All students came to the meeting Some students came to the meeting [= (S)] Necessarily John will come Possibly John will come I:=(lO)] John must come John may come [= (12)]
Clearly (43) teCs more about the world than (44) does: thus if we know that Bill is L dent tlten we car-!infer from (43). but not from (44), that Bill came to the meeting. (45) t&s us more about the world than (46): (45) entitles us to be sure of Sohn”s coming, ivhereas (46) leaves it open. (47) and (48) are ambiguous between episremi~ and dmntic senses: on the epistemic sense they are synonymous with (-IS; artd (46) respectively, and on the deontic sense they mean something like (49) anJ (SO) respectively:
Cmdwt I.dz types of interpretation the first sentence 011the pair is the more inforrna?rv~~i-49) tcUs us th;). Joi>n is not only permitted but #alsorequired to come, w91~re;t;~ i50) at+y tells k s :Ibout the permission. Given these observa?ions about the rr’]grj’;Cjrlform&eness uf these pairs or words, the addressee who hears the less I cpntpnrtv 2s (#), (A/;1 QTVI id521 ia ir? o nncitinn to lnll_l~-~~~i3fE~~.,: _ ‘.C;?. . . I _WI _ Iav-4 -C-w _ 11) cllrh -. I_.. “W...W..“VY TV] U..U \-TV, .U I.. 64 yvhwclvll rnfer t9~t rile speaker, zlsumed TV be following Grice’s maxim (32) does not ‘MJWZ ttrc n‘r~rt!informative sentences to bc true (since, if he did believe them to R:c lrl>e I.kr: he h~uld have uttered {hem instead). thus from (44:) he can infer PA q
ax! frsns (4-Y)he can infer (54;: ! 54 j It fcj not
the case that Jol?n m.-lst come
G. Gazdar / &at matics and lo&al form
9
Again, there is no need for the linguist’s semantic theory to account for these inferences since they are susceptible to a natural explanation within the prirgmatic component. Consider now the following pair of examples: (56) .lohn met Mary last night (57) John met a woman last night [= (1711 Clearly (56) is more informative than (57): (56) tells us everything that (57,) does, and then some. In rrany conrexts therefore : he speaker of (57) will be heard to in+ ply that he doesn’t know ang particuiar sentence of the form John met X last night (where x is a proper name (JT an identifying description) to b: true, e.g. h(, doesn’t know (55) to be true, since, if he did, then he could ‘havebeen more informative by uttering that sentence insr.ead. But in other contexts the addressee may know pt;rf2ctly well that the speaker knows whom John met, but that what is relevant to tht: conversation in progress is just that John met ““II&Y”.L” cnn-finnp .“.I.ul”. ffim&b Tn I v iAnn+ihr IUVILLII) tb,e individual would simply be to introduce distracting material mto the discourse. In these contexts the spcake:r can be seen to be being ‘zlsinformative as is requi; zd’ as well as obeying the maxim ‘Be relevant”. But now we ha;e a pragmatic acco jnt of the so-called specific/rxmpecific ambiguity associated with sentences like (57). Note that this account provides obvious grounds for the unmar-ked interpretat on of (58): (58) I met a woma,n last night [= <18)) If the speaker himself met [he woman then it is very ukjlikely t+at he is unable to supply a proper name or identifyjng description shouI4 that be relevant. So it is unsurprising tPiat this sentence tends to appear to t‘i iirnambiguclusly specific. Hzre we have a pragmatic explanation for data which is quite bafflirhg if seen purely from the point of view of a truth-conditional semantic theory [I!;. In the second section we listed a whole set of problems that arise when attern) ts are made to handle the phenomenon of presuppositr’aIr within the semantic compo~.t of the grammar. None of these problems arise ii’we remove presupposition in toto from the domain of semantics and provide a pragmatic thleory for it. Let us sketch here, very informally, how such a theory works (a lengthy expasilm.ionan:! a full formalization can be fcbllnd in Gazdar 1979): (i)
specific words (e.g. regret, know) and syntactic const ructiurz are associated with the potential presuppositions they give rise to.
[ 2 J This explanation and thir examples are adapted from Klein (1978). I am g.rateful to Barbara Abbozt, Hartmut Haberland and John Lyons; i’or convincing me that ther: are circumstances in which (58) would be assigned a no~n-spetific interpretatirn.
ssn:prle sentences potcnriaily pr~upposc all the iiotential presuppositions associ;;ited with the words and c~>ns::uctlr)ns ihat make them up. compound scntcnccs potcntiatly presuppose #II1 the potential presuppositions ~Aed by their cornpnent simple sentences. senyT’ces, simple or compound, actuul[v prestippose all and only the potent i;h plesupyositions that are csnsistent with the other itnplications carried by sentcslce in tl:e lcontext of utterance. Scfrrcnucs t:G) tc (64) all potentially pre9qpose d!Xlrb’e:
(65) in view of clauses (i)-(iii)
John regrets having faited the ex:im [= (19)j John d~~sn? regret having failed the exam [= (X)] PossibJy JCI~AI regrets having faiied the exam [= (23 j] John ntay regret having failed the exam [= (24)] Sofhnd~sn’~ know thh! !le failed tht: exam [= (291 !t’s p;.;sibie that John wants to be an accountant and it’s possibie that he regrets ‘ailing the exam 1-z(26)] Speakers knows that John failed the exam Ihcir: senter:ce:; also acmdly p r~suppo!le (65) since nona of them carry imphAifjn: that XLSimJnsi!mnt with (65). Compare now the following examples:
‘fi-~cg sentezces. like (5Y)-(641, alzo potentially presuppose (65) in Gew of (i)I M) H~i~~ever,none of them CM ever mtzdly presuppose (65) since they al! give 1~132 to implications that are incor,sistenl wit:] (65): t&C)) S?eJker tziows that John passed the e?.am (‘0) Spea;:er doesn’t know that John fz.i:ed the exam t”B J Speal:er knows that it’s po:;sibIe that .lo~hn didn’t fail the exam utterance of (66) implies (69) and this i2 inconsisten: with (65). Uttera:lce of (67) asserts (70) and this contradicts r’65). Uttcralce of (613)implies (7 1) (cf. the discusr:on of (46) and (53) above) ancl this tclc is inconsistent with (65). The pragmatic t.hersry skeiched in fi).-(iv) thu:; accounts s~raightfc~wardly for all the data that ~zt~~.:~ semantic theories ~:t’presupposition so much c!mbarrassLnent. Notie ihat (72) and (73) both typically imply (71) and (75):
il (72) (73) (74) (75)
Either John has failed the c:xam or he has lost his job If John has failed the exam the% be has lost his job it is compatible with all tht: spzrker knows, that John has failed the exam It is compatibb! with all thl: speaker knows, that John hae lost his job
These implicaCuns d3 not follow from the semantic peoperties of disjunctions and conditionals, but a natural exphlnation for them does follow from &ice’s maxim (32) concerning infCrrnativeness . If the speaker of (72) knew that John hadn’t failed the exam, ther*, in utterin; (72) he would be being less informative than he could have been since he could just as well have uttered (76) which would have convcyed much more information to hi; addressee: (76) John hasn’t failed the exam but he has lost his job So the addressee can infer from (72), together with the assumption that the speaker is conforming to Grice’s maxim, that it is not the case that the speaker knows John not to have failed the exam. And that is equivalent to (74). Hence the implication relation that holds between (72) and (74). Exactly analogous chains of reasoning explain the implication relations that hold between (72) and (75) and between (73) and (74) and (75). It is implications like these which explain the nonsynsnyrny of the tautologies with which the seond sect ion oi this paper closed : (77) (78) (79) (SO)
Either McCarth_~is a communist or he isn’t [= (29)] If thlere’s going to be war then there’s going to be war [= 1[30)1 It is compatible Gth all the speaker knows, that McCarthy is a communist It is compatible with all the spreaker knows, that there will be war
Although (77) and (78) have the same truth coildil a(!-~, they ilre not hearl to bc syrronymous because they give rise to different prageiiatic implications: fo: elr.am?!r:, (77) implies (79) but (78) does not) and (78) implies (80) but (77) does 1101.
4. Conchlsions The proqect raised by these consider;ltions is rather different from t radltional views of meaning. On those views, a theory of meaning v’as going to assign some Utary object (i.e. a irteurring) to each and every senter ce in the language. But the present prospect is rather that what we intujtIvcly think af as ‘nnean.ing’is a compoGte and hybrid affair, ccnmpounded from a restricted semantics on the one hand, irnd a wide-ranging pr;lgmatics on the other. To some this prospect is, no doubt, disappoint.ing. Instead of an elegant and Itinitary account of the significance of utterances, we are now offered. a complex account compoundeil from a theory of truth conditions and a theory df communi-
cation. Indee b3atters are more complex than the text of this paper has so far admitted. 7Ilere is increasing evidence that the semantic component of the theory must sometimes have xws to the pragmatic properties of constituent claus(:s when assigning the truth conditions of compound sentences (see Gazdar 1979: ch. 7). This &den.:e indktes that the semantic component is not autonomous with respect to the pragm; tic component. There is also d lot of evidence to show that pragmatic factors play a :;i,@fica.nt role in the syntax and phonology of natural ianguages (see Gazdar (forth,:oming) for a survey). The question now is Aether this hybrid theory accurately reflects the complexities elf the facts, GI whether it is begiming to look suspiciously epicyclic - 3 fird and ingenious attempt to salvage a doomed semantic theory by the creation uf a Suff’er pragmatic component. At the present time, given the v.:ry considerabie suc:XSSof the hybrid theory, and given the complete absence of any alternative nonhybrid theory whkh can toddle the same range of phenomena, it seeems to me that framework within which to locate their research.
References Atkts,
I.D.
1975.
Presuppositions:
a semanticc -pragmatic
account.
Pragmaticv
Microfick
: .4,
1)13-G14. c+ 184?. 2he mathemb:kal analysis of I:@c, being an essay towards a ;a!culus of dcducWe rs~~fiing. Lundon: George EM. cd., Linguaggi nella hp5vICsont D. 1970. Semantics for natural :arquagcs’. In: B. Vknrini, *w:iet.a c nelh tecnicn. War : Edizioni di Conlmunita. oc, 177- 186. I qe C. 1879. i@riffsf;hrift, eme der arithmrtrischen nachgebildere Formelsprache des reincn hnkenb. flalle ;I. S : Louis Kebert. Gazdz:r. W.M. 1978. Eine pragrmatisch~manti!;che Mischtheorie der Bt.deutung. Liligtiisiische Be&ate 58 : 5 - 17. Gazdlt, G.J .!+f. 1979. Prsgmatics. New York: Academic Press. Gazdz, G.J.M. Forthcoming. ‘Pragmatic constraints on linguistic prcnduction’. In: B. Butter~otih, sd., Language production. London: Academic Press. Grice, H.P. 1975. ‘Logic an,d conversation’. in: P. Cole L d J. Morgan eds. Syntax a rd semantics 3: Speech acts. New York: Academic Press. pp. 41-58. Hazur~ul, G. 1972. ‘Deep structure as logical form’. In: D. Davidson and G. Haxmnn,eds.,SemanIICSof natural larq-uge. Dordrecht: Reidel. pp, 25-47. Horn, LX. 1973. ‘Greek Grice: a brief survey of protoconversational rules in the history of iog.K’. in: Papers from the Ninth Regional Meeting, Chicago Linguistics Society, pp. 205214. KXZ, SJ. and J.A. Fgdor. 1963. The structure of a semantic theory, Languag,o 39: 17C-210. Ken;pj.~: R.V. 1975. PreFuppositititi and the delimitation of semantics. Cambridge: Cambridge Cniverr,ity Press. Klein, E.H. 1978. On sentences which report beliefs, &sires and othe.: mental attitulles. Cambridge: Ph.D, disertation. Woke, S.A. 1963. Semantical considerations on modal logic. Acta PUosophica Felmica 16: 83 -94. bJk.
13 Martin, J. 1975.Ksrttunen on possibility. Linguistic Inquiry 6: 339-21. Montague, R, 1970, ‘UniversaJ grammar’. In: R.H. Thomason, ed., Formal philosophy: selected papers of Richard Montague. New Haven, CT: Yale University Prtm. pp. 222-246. Wilson, D.M. 1975. Presuppositions and non-truth-conriitional semantics. London: Academic Press. Gerald Gazdar is Lecturer in Linguistics at the University of Sussex. t!e studied at the Universities of Eaet Angha, Cambridge, London, and Reading, from which he reccivcd his Ph.D. i,i 1976. His pubbcations include &~g~ks (1979), with E.H. Klein and G.K. Pullurn: A bibliog raphy 0-f contempwary linguistic research (1978), and papers in Linguistics and Philosophy, Linguistkcs, Chkago 1 inguistic Srrciety pubiicaQons, the Syntax and smantics series, and various other edited volumes. He was a four ding editor of I+agmatics Microfiche (with Stephen C. LevinEon), and is currently principal investigator of an SSRC-thnd,cd project on the syntax, scrrurntics. and p;agmatics of the comparative construction.