volume12,no.14 september2012
The Open Instruction
Theory of Attitude
Reports and the
Pragmatics of Answers
Philipp KoralusPrinceton University
© 2012 PhilippKoralusThis work is licensed under a Creative Commons
Attribution-NonCommercial-NoDerivatives 3.0 License. <www.philosophersimprint.org/012014/>
1. Introduction1
Itisaquotidianthoughextremelyremarkablefactthatbyassertivelyutteringsentenceswecanmakeclaimsabouttheworld.Whydowetake anutteranceof a sentence to conveyone claim rather than an-other?Ourunderstandingofcontextandintentionsplaysanimportantrole.Butthemostimportantfactorarguablyisourgraspofthelinguistic meaningassociatedwiththesentence.Somemaythinkofthisasthese-mantic contributionofthesentence.Onstandardtheoriesinphilosophyandlinguistics,weunderstandanovelutteranceinlargepartbyderiv-ingitslinguisticmeaningfromthelinguisticmeaningsofitspartsanditsgrammaticalstructure.
According to one commonsensical view, the linguistic meaningofpropernames isunproblematic:names juststand for things.Thelinguisticmeaningof‘Otto’justistheindividualOtto,etc.Wecouldaddthatthelinguisticmeaningofasentencelike‘Ottosmokes’justisthesingularpropositionthattheindividualOttohasthepropertyofsmoking.Letting‘<P<Y>>’standforthepropositionthatthepropertyPappliestotheindividualY,wecanregimentthisviewasfollows:Thelinguisticmeaningof‘Ottosmokes’is<smoke<Otto>>.Thisnowclas-sicalviewfamouslyrunsintotroublewhenweconsidersentencesthatreportonbeliefs,desires,andotherattitudes.
1.1 Complex interpretations and Kripke’s PuzzleImaginethescenarioofSupermancomics.Intuitively,anassertionof(1)inthescenariowouldbetrue,whileanassertionof(2)wouldbe
1. Thispaperhasbeenlonginthemaking,benefitingfromdiscussionswithseveralaudiences,beginning in2008with thePhilosophySociety,Austra-lianNational University, and followed by the Philosophy Program at theGraduateCenter,CityUniversityofNewYork,andtheDepartmentsofPhi-losophy andPsychology atPrincetonUniversity. I amgreatly indebted toGideonRosen,GilbertHarman,PhilipJohnson-Laird,andtwoanonymousreferees.Forhelpfuldiscussionsandcommentsovertheyears,IalsothankAngela Mendelovici, as well as Sarah-Jane Leslie, Paul Benacerraf, DeliaGraffFara,andWillStarr.IoweaspecialdebtofgratitudetoJayAtlas,whoseseminalworkonsemanticnonspecificityandneo-Griceanpragmaticspro-videdthecrucialstartingpointfortheviewsIdevelopinthispaper,andwhoprovidedinsightfulcommentsandcriticismsthroughseveraldrafts.
ImprintPhilosophers’
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –2– vol.12,no.14(september2012)
ourpurposes,complexinterpretationscorrespondtopropositionslike(3)and(4)thatincludepropertiesofrepresentations,whichIwillcall“PoRs”forshort.5
Intheaboveexample,distinctpropertiesofrepresentationsareas-sociatedwithdifferentnames,butthisisinessential.Thesamenamecan be associated with distinct properties on different occasions.6 ImaginePeterdoesnotknowthatPaderewskithe pianististhesameperson as Paderewski the former Prime Minister of Poland. Peter hasheardPaderewskigiveastunningrecital,andasaresult:
(5) PeterthinksthatPaderewskiistalented.
However,Peterdoesnotthinkmuchofpoliticians,andhence:
(6) PeterdoesnotthinkthatPaderewskiistalented.
Paradoxically,thereisatrueutteranceof ‘PeterdoesnotthinkthatPaderewskiistalented’eventhoughPeterbelievesthatPaderewskiistalented.Theapparentparadox,duetoKripke,isresolvedifonegives(5)aninterpretationveryroughlyglossedas‘PeterthinksthatPaderewskirepresented in a pianist-wayistalented’and(6)onerough-ly glossed as ‘Peterdoesnot think thatPaderewski represented in a politician-wayistalented’.
1.2 Default singular interpretationsNotall attitude report interpretations seem to involvewhat I calledcomplexinterpretationsthatwouldincludeacharacterizationofhowindividuals are represented. Imagine three of your colleagues arenamedNancy,Ruth,andTerence.SupposeNancyreports,
(7) TerencebelievesthatRuthsmokes.
likelytomisleadwhenappliedtotheregimentationsuggestedhere.
5. Idonotuse themore familiar term ‘MoP’ (ModeofPresentation), since itseemstohavebecomestronglyassociatedwiththe“token”viewmentionedinfootnote2.
6. Kripke(1979)
false.Atthesametime,‘ClarkKent’and‘Superman’standforthesameindividuals, sopurelysingularpropositionsarenotsufficienttodistin-guishwhatisintuitivelyconveyedby(1)and(2).
(1) LoisbelievesthatSupermanflies.
(2) LoisbelievesthatClarkKentflies.
Anattractiveresponseis tosaythatwemustdistinguishwhat isbe-ingrepresentedandhowitisbeingrepresented.(1)and(2)arepartlyinterpreted asmaking different claims about the properties of cer-tain representations that Lois has of s. It seemsnatural to say thatweinterpret(1)asconveyingthatLoisbelievesthatscanfly,whereLois’srelevantrepresentationofshasthepropertyofpresentings asasuperhero.Bycontrast,weinterpret(2)asconveyingthatthatLoisbelieves thatscanfly,whereLois’srelevantrepresentationofshasthepropertyofpresenting s asanerdy reporter.Putdifferently, (1)and(2)areinterpretedasmakingdifferentclaimsaboutthetypeofguiseunderwhichLoisrepresentss ashavingcertainproperties.2Thetruthconditionsof these interpretationsare independent.This sug-geststhefollowingregimentationofwhat(1)and(2)convey,where‘Superhero’and‘NerdyReporter’denotedifferentpropertiesofrepresen-tations,tobedistinguishedfrommoreordinaryproperties:
(3) <belief<Lois,<fly<s, Superhero>>>>
(4) <belief<Lois,<fly<s, NerdyReporter>>>>
Schiffer has recently argued for a regimentation along those lines,originallydue toRécanati.3 Iwill call interpretationsof this sort, in-cludingpropertiesofrepresentations,“complexinterpretations”.4For2. Contrast a view likeCrimmins’ (1992), onwhichwe specify tokenwaysof
representing,whichSchiffer(2000)andothershavecriticized.
3. Schiffer(2000);Récanati(1993).Intheinterestsoffocusingthediscussion,Iamsettingasidethequestionofwhetherwesometimesincludepropertiesofrepresentationsalongwithapropertylikesmokingaswell.
4. Iadoptnovelterminologyheresinceexperiencehasshownthatexistingter-minology like “Fregean interpretations” and “MoP-laden interpretations” is
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –3– vol.12,no.14(september2012)
anybodybelieved tobeofgoodcharacterbyacertainMs.Smith isguilty.Onedetectiveasserts(9)tohiscolleagues.
(9) SmithbelievesthatSuspectAisanupstandingcitizen.
Itseemstomethatwedonotinterpret(9)asmakinganyparticularclaimabouthowSmithrepresentsSuspectA,letalonethatshethinksofthesuspectas a suspect.For(9),asintheeverydaycaseexemplifiedby(7),thenaturalinterpretationissingular.
Now,theviewduetoSchifferdiscussedabovetakesthesemanticcontributionofattitudereportsentencestobepropositionsincludingpropertiesofrepresentations,asin(3)and(4),whereattitudereportsentencesaretakentobeindexicalwithrespecttothoseproperties.Incaseslike(7)and(9),itseemsmorenaturaltosaythatnoinformationisconveyedaboutwaysofrepresenting.AsIseetheissue,thiscreateswhatSchiffercallsthe“meaningintentionproblem”fortheindexicalview:Often,ifnotmostofthetime,wedonottakeattitudereportsen-tencestobeutteredwithanyintentiontoconveyinformationaboutwaysof representing,nordowehaveany inclination todistinguishwaysofrepresentinginourinterpretation,butthepresenceofanin-dexicalforpropertiesofwaysofrepresentingmakesithardtorespectthat in interpretation.10 Ifanattitude report sentence includesan in-dexicalforaPoR,thensomethingmustbeassignedtothatindexical.
1.3 The task ahead Iproposethatitwouldbeattractivetohaveatheoryofpropernamesandattitudereportsentencesthatallowsustosaythatthelattercanbeliterallyinterpretedasconveyingcomplexpropositionslike(3)and(4)aswellassingularpropositionslike(8),dependingoncontext.11
10. Schiffer (2000).AsSchiffermakesclear, thisproblem isnotdirectlyaboutcapturingtruth conditionsofintuitiveinterpretations.
11. Idonothavethespacetoconsiderotherproposalsthatbeginwithadifferentanalysisofwhat isconveyedbyattitudereportsentences.Foraprominentexample,seeRichard(1990),whoincludesexpressionslike‘Paderewski’inthesemantic contributionofanattitudereportandarguesthatcontextsupplies
Does the report convey informationabouthow Terence representsRuth? Intuitively, theanswer is “No”.Everydayattitudereports like(7) do not seem to be interpreted as conveying such information.What(7)conveysseemstobeadequatelyrepresentedbythesingu-larproposition(8):
(8) <belief<Terence,<smoke<Ruth>>>>
Thesesortsofcasesgiveinitialplausibilitytowhatissometimescalledthe“naïveneo-Russellian”viewthatthesemanticcontributionofanattitude report sentence just is a singular proposition after all.7 Onsuchaview,acomplex interpretationmightbetakentoresult fromapragmatic implicature,added toa semanticallydeterminedpurelysingularinterpretation.Fewpeoplenowaccepttheviewthatattitudereportsentenceslike(1)and(2)havethesametruthconditions,bar-ring“bribery,threats,hypnosis,orthelike”.8Butthen,whyhasnaïveneo-Russellianismbeensoinfluential?Ibelieveoneimportantreasonisthatsingularinterpretationsare,infact,thedefaultinterpretationofattitudereports,assuggestedbyexampleslike(7).9Whatthismeansisthatsingularinterpretationsarechosenintheabsenceofaspecialreasontopromptthehearertodootherwise.Inmyview,thisexplainswhy,tomanypeople,includingthepresentauthor,somethingaboutthenaïveneo-Russellianviewsomehow“seemsright”,evenafteronemeditatesoncaseslike(5)and(6),etc.
Evennamesthatarestronglyassociatedwithacertainwayofrep-resentinganindividualwillnotalwaysleadtocomplexinterpretations.Imaginedetectives inacovert investigationsosecret thatwitnessesarequestionedundera ruse.Let’s further imagine that theyaredis-cussingwhatvariouswitnesseshavesaid,togetapreliminarynotionofwhoislikelytobetheculprit.Allbelievethatitisveryunlikelythat
7. Forexample:Soames(1987);BraunandSaul(2002)
8. To use Richard’s phrase. Richard (1990), p. 125. Soames himself has aban-donedtheviewinSoames(2004).
9. Jaszczolt(2000)seemstosharethisview.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –4– vol.12,no.14(september2012)
todifferentwaysofexecutingtheinstructionsthatarethelinguisticmeaningofattitudereportsentences.Myaccountiscalledthe“OpenInstruction Theory” (OIT) because linguistic meaning leaves open which of these interpretations is the correct one. These sentencesarejust silentontherelevantdifferencesininterpretation.Inparticu-lar,therelevantdifferencesleaveno traceinthelinguisticmeaningofattitude report sentences andhavenothing to dowith syntax. Theframework of linguistic meanings as instructions to create mentalrepresentationsmakesitpossibletogiveclearcontenttothedistinc-tionthathasbeendrawnbytheorists likeAtlas,Bach,andSoamesbetweenambiguity,indexicality,andnonspecificity.16Intheproposedframework,wecanconceptualizeambiguityasacaseofmultipledis-tinct instructions that correspond to the same surface form,whichwould be independently lexicalized or correspond to different syn-tacticstructures.Wecanconceptualizeindexicalityasacaseofasetofinstructionsincludingacontextuallyfilled-invariable.Finally,wecanconceptualizenonspecificitywithrespecttoarangeofinterpreta-tionsasacaseofstaticinstructionsthatleaveopenarangeofpossiblewaysofexecutingthem,leadingtoarangeofmentalrepresentationsthatcorrespondstotherangeofinterpretations.17
Ofcourse,wedonot intendattitude reportutterances toassertinstructions: weintendthemtoassert truth-evaluableclaims.Wesim-ply do not assert the linguisticmeanings of the sentenceswe use.Linguistic meaning is one particularly important part of what sys-tematicallyaccountsforourintuitivejudgmentsofthe“meaning”ofsentencesweencounterineverydaylife.Linguistic meaning isnottobeconfusedwiththepre-theoreticalnotionof“meaning”thataccom-paniesthesejudgments.
16. Atlas(1977;2005);Soames(2004);Bach(2000).Thisdistinctionisalsodis-cussedindetailinKoralus(2011;2010).
17. Contrasttheoriespostulating“weak”existentiallyquantifiedlinguisticmean-ingsordisjunctivelinguisticmeanings,criticizedasinadequateanalysesofnonspecificitybyAtlas(2005).
Asnoted,oneproposaltomakeattitudereportsentencesrelevant-lycontext-sensitiveistosaythattheyinvolveindexicality,though,asnoted, this routemakes ithard to capturebothpurely singular andcomplexinterpretations.12Anotheroptionthathasbeenexploredbyvariousauthorswouldbetoletattitudereportsentencessemanticallyencode singular propositions, which can be supplemented by prag-maticimplicatures.13
Takingamoreunusualstartingpoint,Iproposeaviewonwhichpurelysingular interpretationsofattitude reportsandwhat I calledcomplex interpretations, involving a characterization of how indi-viduals are represented, are both literal interpretations, while thelinguisticmeaningofattitudereportsentencesisneitherambiguousnorindexicalwithrespecttothem.Attitudereportsentencesarese-mantically nonspecific with respect to the interpretations discussedabove.ThisputsmeinatleastpartialagreementwithBachandwithmorerecentworksofSoames,whohaverecentlyarguedthatattitudereport sentenceswithpropernames in the that-clauseare semanti-callynonspecific.14Whatseemstobemissingsofarisacleartheoryofwhatexactlycompositionalsemantics doesspecifyonsuchaviewandwhatpragmaticprinciplescanyield thevarious interpretationsincontext.Inthefollowing,Iwillexpoundsuchatheory.Iwillthendiscusssomeofitsadvantages.
2. The Open Instruction Theory of attitude reports
IproposetofollowChomskyinsupposingthatthe linguistic meaning ofanexpressionconsistsinasetofinstructionstoconceptualsystemsforconstructingmentalrepresentations.15Iaddthatthevarietyofin-terpretationsofattitudereportsdiscussedintheintroductionreduces
atranslationmanualthatdetermineswhichexpressionscanbeusedtoreportonaperson’sbelief.Forsomecriticismsofthisapproach,seeSoames(2002).
12. Schiffer(1977;2000)
13. Soames(1987);McGlone(2007)
14. Soames(2004);Bach(2000)
15. Chomsky(1965;2000)
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –5– vol.12,no.14(september2012)
powersimilartothosethatmotivateChomsky’sminimalistprogramin syntax, which attempts to do awaywithmultiple levels of syn-tax-internal representations (Chomsky 1995; 2000).19 I am highlysympathetictothespirit ofStanley’s(2002;2005)principleinsistingon linguistic constraint in semantics that these concernsmotivate.However, I think this principle is ultimately too strong.20 I believethatwherewehavereasontobelievethatlinguisticmeaningissilent,wejusthavetofindotherwaysofconstrainingourtheoriesbeyondlinguisticmeaning.Sincewecharacteristically reasonwithwhat isconveyedinconversation,itseemsnaturaltolooktothepsychologyof reasoning tomotivate asmany aspects of theOpen InstructionTheoryaspossible.Evenifwecannotimmediatelyfindindependentsupportforeverymovingpartofthetheory,whatcountsisthatweopenourselvestoanewrangeofindependentconsiderations,wheresyntactictheorynolongerprovidesany.
According to Johnson-Laird’s influential theory,peopleordinarilyreasonbybuildingmental modelsthatstandincertainrelationshipstoeachother.21Thetheory is intendedtoexplainourcompetenceandperformanceinreasoningevenwithnonverbalinputs,andneuroim-agingevidenceisbeginningtobeavailabletoshowthattherelevantoperationsareprocessedinareasofthebrainthatarelanguage-inde-pendent.22Thoughreasoningandlinguisticinterpretationarecloselyrelated,thestudyofreasoninghasitsownsubject,whichallowsustolooktothetheoryofreasoningforindependentsupportintheorizingabout semantics andpragmatics, just aswemight look to syntactictheoryforindependentsupport.Nearlyallaspectsofmytheorythatgobeyondwhatissyntacticallydeterminedwillbedescribedinterms
19. Thesesortsofconsiderationscreatesomeappealforformalreconstructionsof themachineryofDRT in terms thatdonot requirea fully independentlevel of representationof discourse, as inGroenendijk andStokhof (1991)andMuskens(1996).
20.Foracritique,seeRécanati(2002).
21. Johnson-Laird(1983;2008);Barrouilletetal.(2000);Byrne(2005)
22. BauerandJohnson-Laird(1993);Knauffetal.(2001);Krogeretal.(2008)
OIT shareswithKamp’s (1981)DiscourseRepresentationTheory(DRT)andHeim’s (1982)file change semantics (FCS) theview thatliteral utterance interpretation involves a level of representation be-yondwhatisdirectlyconstrainedbysyntax;theyarerepresentational-isttheoriesofinterpretation.However,adoptingarepresentationalisttheoryofinterpretationdoesnotyetguaranteethatthenonspecificityofattitudereportsentenceswillbeanalyzedintherightsortofway,asdistinctfrom,forexample,ambiguityoranaphora.18ThesedistinctionshaveimportantempiricalconsequencesthatIwilldiscussinsections5and6.Nordoesadoptingarepresentationalisttheoryguaranteethatwecancapturedefaultinterpretationsofattitudereportsentencesaswell as departures from these defaults. Ifwe assign nonspecific lin-guisticmeanings,westillneedapragmatictheorythatworkswithin-puts thatdonothavedeterminate truthconditions,whichrulesoutclassical Gricean proposals. As Levinson (2000) observes, the con-nectionbetweenpragmaticsandrepresentationalisttheoriesofinter-pretationisunder-theorized.ThismaybepartlyduetothefactthatrepresentationalisttheorieshavereceivedsomewhatlessattentioninmainstreamphilosophyoflanguagethanIbelievetheymerit.
Hereiswhatmay,inpart,beresponsibleforthiscomparativelackofattention:manyhaveworriedthatrepresentationalisttheoriesofinterpretation are generally less attractive, because they can helpthemselves to linguistically unconstrained postulates. On the onehand,thisraisesthethreatofhavingthesystemovergenerateinter-pretations.Ontheotherhand,appealtoalevelofrepresentationthatisnotsyntacticallyconstrainedmightmakeproposalsseemlessex-planatorilypowerful,duetoworriesaboutunconstrainedexpressive
18. So-called “underspecification” in discourse representation theory, as dis-cussedbyAsher andLascarides (2003), is taken to encompass lexical am-biguityandquantifierscopeambiguity,aswellasanaphora.Thisnotionof“underspecification”,unlikethatputforthbyAtlas,Bach,andSoames,seemstoo broad. Atlas (1977; 2005), Koralus (2011; 2010), and Zwicky and Sa-dock (1975)discuss someof theempirical consequencesof thedistinctionbetweennonspecificity, lexicalandquantifierscopeambiguity, indexicality,andanaphora,suchasdifferentconstraintsoninterpretationimposedbyVPellipsis,whichIwillturntolater.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –6– vol.12,no.14(september2012)
2.1 Capturing the interpretations of attitude report sentencesSupposeyouaregivenaninstructiontocreateamodelrepresentingSupermanatacertainplace.Youcouldjustfixapinonanappropriatemap,asin(10).
(10)
It is not conceptuallymysterious that such instructionsmight leaveopenhow individualsshouldberepresented.YoucouldinsteaduseadetailedSupermanfigurethatpresentstheindividualinamoreelabo-rateway:
(11)
ofoperationsandrepresentationsindependentlyrequiredbymentalmodel theory. For now,what Iwill take frommentalmodel theoryis that individuals are represented by individual “tokens” in amen-talmodel,andpropertiesandrelationsarerepresentedby“features”bound to those tokens,where featuresand tokensaremental repre-sentationsandwherebindingisapsychologicalrelationshipbetweenmentalrepresentations.Wewillsaythat,foreachpropositionwecanentertain, we can build amentalmodel that represents that propo-sition.Singularpropositionsdecompose into individuals,properties,relations, and propositions.Mentalmodels correspondingly decom-pose intomentalmodel tokens, features, and othermentalmodels.Forexample,MaryisrepresentedbythementalmodeltokenMARY.Theproposition<smoke<Mary>>isrepresentedbythementalmodelSMOKE(MARY).Theproposition<love<Mary,John>>isrepresentedbythementalmodelLOVE(MARY,JOHN),etc.
Mentalmodeltheoryrequiresanoperationthatassemblesmentalmodels,whichIwillcall the“modelbuilder”(MB).OntheOpenIn-structionTheory,utteranceinterpretationproceedsasfollows:Givenanutteranceof a sentenceS, the language faculty computes the lin-guistic meaningofS,whichIwillwriteas ‘||S||’,nottobeconfusedwiththedenotationofS.Asnoted,thelinguisticmeaningofSisasetofin-structionstobuildmentalmodels.Themodelbuilderthenconstructsamentalmodel,byexecuting ||S||.Theresultingmodelexpressesorrepresentsaproposition.Theclaimwetaketobemadebyanassertive utteranceofasentenceSisthepropositionrepresentedbythementalmodelMthatisobtainedastheresultofapplyingthemodelbuilderto||S||.WejudgetheutteranceofS tobetrueifandonlyifwejudgethepropositionrepresentedbyMtobetrue.
Tofleshoutthispicture,weneedanaccountofthosementalmod-els thatamount to the interpretationsdiscussed in the introduction.We thenneed recursive rules allowingus to compute the linguisticmeaning of attitude report sentences and arrive at correspondingmentalmodels.Finally,weneedapragmaticcomponentofOITthatcanmakesenseofbothsingularandcomplexinterpretations.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –7– vol.12,no.14(september2012)
Wetake(14)torepresentapropositionthatincludesapropertyofrep-resentations,viz.thepropertyofpresentinganindividualasafamouspianist.Forexpositoryconvenience,Iwillusecombinationsofcapitallettersinitalicstodenotepropertiesofthissort.
(15) <belief<Terence,<smoke<Ruth,FAMOUSPIANIST>>>>
(15) is apropositionof the sort suggestedbyRécanati andSchiffer,discussed in the introduction, as corresponding to what I called a“complexinterpretation”.Asnoted,thetruthconditionsofaproposi-tionlike(15)areindependentfromapropositionlike(16):
(16) <belief<Terence,<smoke<Ruth,FAMOUSPOLITICIAN>>>>
Whatabouttherelationshipbetween(16)and(13)?Itseemshardtoavoidtheviewthat(16)hastoentail(13).Intuitively,wewant(13)tobeverifiedbyanysituation inwhichTerence representsRuthassmokingandtakesthatrepresentationtobetrue,regardlessofhowherepresentsRuth.Thishastheconsequencethat(16)hastoentail(13).This,inturn,quicklyleadstotheconsequencethat(17)isconsistentanddoesnotentailthatTerenceisirrational:
(17) <belief<Terence,<smoke<Ruth>>>>
&<belief<Terence,<¬smoke<Ruth>>>>
Similarly:
(18) <belief<Terence,<¬smoke<Ruth>>>>
⊬ <¬belief<Terence,<smoke<Ruth>>>>
Giventheresourceswehaveappealedto inordertoaccountforvarious interpretations of attitude reports, the question arises as towhattomakeofamentalmodellike(19)thatinvolvesnorepresenta-tionofattituderelationsandrepresentsapropositionlike(20).
(19) SMOKE(TERENCEFAMOUSPIANIST)
Youcanrepresentanindividualinyourmodelwithmoreorlesselabo-ratedplaceholders.Inadditiontowhatisrepresented,thereisawayinwhichitisrepresented.Theinstructionsleft it openhowSupermanwastoberepresented.
Justasinstructionsintheabovetoyanalogycanleaveopenhowanindividualistoberepresented,theinstructionsthatamounttothelinguisticmeaningof an attitude report like (7) leaveopenhowanindividual,e. g.,Ruth,istoberepresented.
(7) TerencebelievesthatRuthsmokes.
Thesimplestwaytorepresentanindividual inamentalmodel istouseamentalmodeltokenthatsimplystandsfortheindividual(likeasimplepininthetoyanalogy).For(7),thiswouldgiveusthefollowingmentalmodel:
(12) BELIEF(TERENCE,SMOKE(RUTH))
Wewilltake(12)torepresentthesingularproposition(13).
(13) <belief<Terence,<smoke<Ruth>>>>
Othermentalmodelscouldserveasaninterpretationof(7).RuthmightbelikePaderewski,knownbysomeasafamouspianistandbyothersasafamouspolitician.PerhapsTerencebelievesthatonlyRuthrepresented in a pianist guise smokes.Inthatcase,ourinterpretivemodelmightinvolvearepresentationofRuthwiththepropertyofpresentingRuthasapianist.Iwilltaketherelevantinterpretivemodeltoincludea complex mental model token, comprising a mental model tokenrepresentingRuthbound toamentalmodel feature that representsthe property of presenting Ruth as a pianist.Iwillwritethesymbolforamentalmodel featureof this sortasa superscript to the symbol forthementalmodeltokentowhichitisbound.Weobtainthefollowinginterpretivemodel:
(14) BELIEF(TERENCE,SMOKE(RUTHFAMOUSPIANIST))
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –8– vol.12,no.14(september2012)
(R1)
τ, where τ represents i ifXisapropername oftheindividualiMB(||X||)è ϕ, where ϕ represents PifXisaverbthatdenotes thepropertyorrelationP
Thekeytocapturingthefullrangeofattitudereportinterpretationswastoleaveopen how individualsarerepresentedintheinterpretivemodel.Welett, above,rangeoverbothsimpleandcomplexmentalmodeltokens.IwillthinkofMB(||X||)asamappingthattakesusfromthelinguisticmeaningofXtomentalmodelconstituentsthatarecorrectexecutionsofit.Forexample,MB(||Superman||)canmap||Superman||toanyofthementalmodeltokensin(23).Whichmen-talmodeltokenwill,infact,enterintotheinterpretivemodelwilldependonpragmaticprinciplesdescribedinthenextsection.Sometokensarecomplex,inthattheyincludefeaturesrepresentingprop-ertiesofrepresentations:
(23) MB(||Superman||) SUPERMAN Simple token for singular interpretations
SUPERMANS Complex token with feature S representing the PoR of pre- senting as a superhero
SUPERMANR Complex token with feature R representing the PoR of pre- senting as a reporter
SUPERMANX Complex token with feature X representing some other PoR
(20)<smoke<Terence,FAMOUSPIANIST>>
Intuitively, it is truth-conditionally irrelevant how we represent Ter-ence,ifwewanttoaskifwearerightabouttheworldinrepresentingthathesmokes.(20)hasthesametruthconditionsas(21).
(21) smoke<Terence>
Thisdoesnotmeanthattherearenogenuinedifferencesininterpreta-tionhere.SupposeTerencealsogoesbythederogatorynickname‘Nit-wit’.Aninterpretationof‘Nitwitsmokes’mayleadtoamentalmodelinwhichTerenceisrepresentedinaderogatoryway.Yet, intuitively,NitwitsmokesifandonlyifTerencesmokes,becausethepropertyofsmoking has nothing to dowithhow a smoking individual is repre-sented.Ordinarypropertiesandrelationsdifferfromattituderelationsinthisregard.
Thenextstepwillbetogiveanaccountofthelinguisticmeaningof attitude report sentences that allowsus to generate the rangeofinterpretationsjustdiscussed.
2.2 Primitive instructions for proper names and verbsWe need recursive rules that generate the linguisticmeaning of at-titudereportsentences,basedonprimitive instructions.Where ‘||X||’denotesthelinguisticmeaningofX,and‘PN’and‘V’arevariablesforpropernamesandverbs:
(22) ||PN||=“createamentalmodeltokenrepresentingthebear-erofPN”.
||V|| = “create a mental model feature representing thepropertydenotedbyV”.
(22)isaninformalgloss.Withthemodel-buildingoperationMB,wecanimplicitlydefineinstructionstocreatementalmodelconstituentsintermsoftheresultofapplyingMBtothoseinstructions.Lettingtbeamentalmodeltoken,jamentalmodelfeature,thefirstruleis:
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –9– vol.12,no.14(september2012)
thefollowingtwosorts:BELIEF(TERENCE,SMOKE(RUTH)),fora sin-gularinterpretation,orBELIEF(TERENCE,SMOKE(RUTHF)),whereFisa featurerepresentingapropertyofrepresentation,asneededforwhatIcalled“complexinterpretations”.
(7) TerencebelievesthatRuthsmokes.
Asimplifiedsyntacticanalysisof(7)willsufficeforourpurposes:
(24)
Howdoweletmentalmodelfeaturesbindtotherightmentalmodelconstituents? For simplicity, I assume a syntactic theory thatmarksargumentrolesinthesyntacticstructure.Weneedanoperationthatbindsmentalmodelfeaturestomentalmodelconstituentsindifferentargument roles.Call thatbindingoperation ‘B’. For amentalmodelfeatureϕandamentalmodelconstituent£,where £couldbeeithera tokenτora (sub-)mentalmodelM, B can, inprinciple,bindϕ to£ agent-wiseorobject-wise.Syntaxdetermineswhetherbinding istobeagent-wiseorobject-wise.23ThebindingoperationB has twoinputs:a set, {ϕ, £},andanargument roleparameter,θ,whichcan
23. Anotheroptionwouldbe tomathematicallymodelmentalmodel featuresaslambdaabstractions,mimickingmoreclassicalapproaches.Wewouldget
λ£.λτ.BELIEVE(τ,£) forBELIEVE.On thisoption,mentalmodelbindingismodeledasfunction-application.Argumentrolemarkinginsyntaxwouldeffectivelybemaderedundant.
Informalterms,(R1)isaproductionrulethatallowsustotransitionfromMB(||PN||)toanyavailablementalmodeltokenthatrepresentsthebearerofPN (moreonconstraintsontherangeoftokensinsec-tionstofollow).
2.3 Complex instructions and binding
Mentalmodel features bind to oneormorementalmodel constitu-ents.Forexample,amentalmodelfeatureSMOKEthatrepresentsthepropertyofsmokingbindstoonementalmodeltokenrepresentinganindividual.AmentalmodelinwhichSMOKEisboundtoJOHNrep-resentsthepropositionthatJohnsmokes.WeshouldthinkofSMOKEashavinganattachmentsiteforan“agent”ofthepropertyrepresentedby the feature. The feature representing smoking can bewritten as‘SMOKE(△)’,where△ isaplaceholderforamentalmodeltokentowhichthefeatureSMOKEisbound.
Mentalmodelfeaturesrepresentingrelations,suchasthementalmodelfeatureLOVE,canbindtotwomentalmodeltokens.Depend-ingonhowLOVEisboundtoamentalmodeltoken,thattokenrep-resentseithertheagentortheobjectoftheloverelation.Hence,thementalmodelfeatureLOVE(△, £),‘LOVE’forshort,hastwositesforattachingtomentalmodeltokens:onefortheloverandoneforthebeloved.LOVEisatwo-placefeature.Iadopttheconventionthattheleft-boundmentalmodelconstituent in thesymbolic representationofatwo-placementalmodelfeatureistherepresentationoftheagent,andtheright-boundconstituenttherepresentationoftheobject.At-titudefeatureslikeBELIEF(△, £)bindtotwomentalmodelconstitu-ents,likeLOVE.BELIEFbindstoa(sub-)mentalmodelM,astheob-ject (therepresentationofwhatisbeingbelieved),aswellasamentalmodeltokenτ,astheagent(therepresentationofthebeliever).Wewill shortlyhave tospecifyhow languagecontrolshowfeaturesgetbound.
Now,thetaskistogiveanaccountonwhich(7)determinesanin-structiontocreateamentalmodelthatMBmapstomentalmodelsof
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –10– vol.12,no.14(september2012)
MB(||CP||)è B(θ,{MB(||NP||),MB(||VP||)})[R4, R2]
è B(subject,{MB(||Ruth||),MB(||smokes||)})
So,
MB(||IP||) è B(subject,{MB(||Terence||),B(complement,{MB (||believe||),B(subject,{MB(||Ruth||),MB(||smokes||)})})})
MB(||IP||)è B(subject,{MB(||Terence||),B(complement, {BELIEF,B(subject,{MB(||Ruth||),SMOKE})})})
By(R1),wecaninserteitherasimpleoracomplexmentalmodelto-keninplaceof‘MB(||Ruth||)’.Forasimpletoken,weget:
MB(||IP||)èBELIEF(TERENCE,SMOKE(RUTH))
ForacomplextokenwithaPoRfeatureF,weget:
MB(||IP||)èBELIEF(TERENCE,SMOKE(RUTHF))
Hence,(R1)–(R4)allowustogeneratebothsingularandcomplexin-terpretationsofattitudereportsentences.Ofcourse, therulesallowustogenerateevenmoreinterpretations,suchas(26),withsomePoRfeatureXforTerence:
(26)BELIEF(TERENCEX,SMOKE(RUTH))
EventhoughaddingPoRfeaturestoTERENCEdoesnotleadtodifferenttruthconditions,asdiscussed,wewilleventuallyneedanexplanationofwhyweordinarilydon’tgetsuchinterpretationsinpractice(settingasidecaseslike‘NitwitbelievesthatRuthsmokes’,discussedabove).
Somemayfinditobjectionablethatonthisaccountofthelinguis-ticmeaningofattitudereportsentences,asmanyinterpretationsarepossibleasthehearercan,inprinciple,distinguishpropertiesofwaysofrepresenting.Butthisishowitshouldbe!AsBachhasconvincinglyargued,foranytwosuchdistinguishableproperties,wecanconstruct
takethevalues“subject”or“complement”,wheretheformerspecifiestheagentrole andthelattertheobjectrole.Let‘__’standforbindingsitesthatBignores.
(25)
ϕ{£} if ϕ is one-place and θ = subject B(θ,{ϕ, £}) è ϕ{£, _} if ϕ is two-place and θ = subject ϕ{_, £} if ϕ is two-place and θ = complement
Now,wedonothavetoexplicitlyformulatetheinstructionsassociat-edwithcomplexexpressions,aslongaswecandefinetheapplicationofMBtocomplexinstructionsintermsoftheprimitivesjustdiscussedtogetherwiththebindingoperation.
ThemodelbuilderMB appliestocomplexexpressionsasfollows.With(R4),Iassume,forsimplicity,that‘that’isnotdirectlycontribut-ingtothementalmodel.
(R2)MB(||IP||)è B(θ,{MB(||NP||),MB(||VP||)})
(R3)MB(||VP||)è B(θ,{MB(||V||),MB(||X||)}),X ∈{NP, CP} for VP ≠ V
(R4)MB(||CP||)èMB(||IP||)
Let’sapplytheserulesto(7).
MB(||IP||)è B(θ,{MB(||NP||),MB(||VP||)}) [R2]
è B(subject,{MB(||Terence||),MB(||VP||)})
MB(||VP||)è B(θ,{MB(||V||),MB(||CP||)}), [R3]
è B(complement,{MB(||believe||),MB(||CP||)})
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –11– vol.12,no.14(september2012)
beinapplicabletoattitudereportsentencesandtocreatedifficultiesIwouldliketoavoid.AtlasandLevinsonreduce“inferencetothebestinterpretation”to“inferencetothemostinformativeinterpretation”.28 Morespecifically,theyproposethattheinformationcontentofagiveninterpretationis tobeunderstoodinsomecasesasthesetof itspo-tentialfalsifiers,followingPopper,orinothercasesasitssetoflogicalconsequences,followingCarnap.Theproposalhasconsequencesthataresomewhatpuzzling,inmyview.29
IsuggestthatweshouldsaveAtlasandLevinson’scrucialinsightthatpragmaticprinciplesbridging thegapbetween linguisticmean-ingandparticularinterpretationsamounttoprinciplesofinferencetothebestinterpretationandlookforanewsetofpragmaticprinciplesthatcanbeappliedtoattitudereportsentencesinthesemanticframe-workIproposed.Thekeyideabehindinferencetothebestexplanation isthatwebeginwithasetofhypothesescompatiblewiththedata.30 Wewouldthenchoosethebesthypothesis,“basedonconsiderationssuchaswhichhypothesisissimpler,whichismoreplausible,whichexplainsmore,whichislessad hoc, andsoforth”.31Now,onthepropos-aloflinguisticmeaningdevelopedintheprevioussection,asentence
28.AtlasandLevinson(1981);Atlas(2005)
29.The proposal is operationalized via the idea that normalized statementsfrontedbymoreexistentialquantifiersare less informativeby thePopperi-anmeasure.Yetempiricalexistentialstatementswithnon-finitedomainsofdiscoursejustdonothave(finite)falsifiers.Itispuzzlingwhyaddingmoreexistentialquantifierswould,inanysense,furtherdecreaseinformativeness.Moreover,consider:‘(A)Thereisapinkelephantinyourroom.’‘(B)Thereisapinkelephantinyourroom,andthereisabluerhinocerosinyourcar.’Regimentingthepredicateswithsomeharmlesssimplifications,weget:‘(A')∃x(Ex&Px).’‘(B')∃x∃y((Ex&Px)&(Ry&Cy)).’BythePopperianexistentialquantifiermeasure,(B)islessinformativethan(A),eventhough(B)entails(A).Thisseemsodd.Anyfalsifierof(A)isalsoafalsifierof(B),soitjustcan-notbethecasethat(B)hasfewerfalsifiers.Atlas(p.c.)holdsthatweshouldsaythat(A)isintuitivelymoreinformativebecauseitismore“specific”.Ipre-fertoavoidthecomplicationsjustsketched.
30.Harman(1965)contraststhiswithenumerativeinduction,whereonewoulddeduceahypothesisfromalistofobservationsviaaninductionprinciple.
31. Harman(1965)
aPaderewski-stylescenariothatrequiresdistinguishingthemtomakesenseofanattitudereport.24Thequestionofwhywedonot,inprac-tice,havetoconsideranarbitrarilylargesetofpossibleinterpretationstomakesenseofanattitudereportwillbeansweredbythepragmaticcomponentofmytheory,suppliedbelow.
Takingstock,bothsingularandcomplexinterpretationsofattitudereportsentencesareaccommodatedbytheproposedcharacterizationoflinguisticmeaning,withoutappealtoambiguityorindexicality.Ontheaboveproposal,attitudereportsentencesareindeednonspecificwithrespect to the interpretationsat issue.Thedistinctionbetweentheseinterpretationsisnotmarkedinthegrammar.Atthesametime,wehavegivenaprecisecharacterizationofhowthenonspecific lin-guisticmeaningofattitudereportsentenceswithpropernamesinthethat-clauseissystematicallygeneratedfromsyntaxandthelexicon.25
Whatremainsistogiveanaccountofhowwesettleonaparticularmentalmodelfromtherangeofmentalmodelsthatwouldcountascorrectexecutionsoftheinstructionsthatarethelinguisticmeaningofattitudereportsentences.Thistheoryhastoexplainwhysingularinterpretationsarethedefault.Italsohastoexplainwhenthemodelbuilderdepartsfromthedefaultinterpretation.
2.4 From inference to the best interpretation to inference to the most respon-sive interpretationIwilladopt the following insightdue toAtlasandLevinson:We in-terpretanutteranceby tacitlymakingan inference to thebest inter-pretation.26Here, inference to the best interpretation is roughly anal-ogousto inferencetothebest explanation inscience.27Unfortunately,AtlasandLevinson’sparticularinterpretiveprinciplesseemtometo
24. Bach(2000)
25. Bycontrast,fromSoames(2004)andBach’s(2000)discussions,itisnotclearhowtoprovideacompositionalaccountofthelinguisticmeaningofattitudereportsentencesthatreflectstheirnonspecificity.
26.AtlasandLevinson(1981);Atlas(2005);Levinson(2000)
27. Theterm“inferencetothebestexplanation”isduetoHarman(1965).
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –12– vol.12,no.14(september2012)
role toquestions, Iwill call theproposal “Inference to theMostRe-sponsiveInterpretation”(IMRI):
(IMRI)GivenanutteranceofasentenceS, amentalmodelofbackgroundbeliefsBG,andamentalmodelofbackgroundquestionsQ, themodel builderwill produce thementalmodelMthatrepresentsthebestinterpretationofS.Misthementalmodelthat satisfiesthefollowingconstraints:34
(Literality)Letthesetofmentalmodelsthatcanbegeneratedbyexecuting||S||beIS.M isamemberofIS.
(Responsiveness)IfamodelM1conjoinedwiththementalmodelBG(representingbackgroundbeliefs)representsmorean-swers toquestions (representedby thementalmodelQ)thatthehearerwantstoanswerbyengagingindiscoursethanM2 similarly combined, thenM1 ismore responsivethanM2 relative toBG andQ (see (E1)–(E3) below). LetthesetofallmentalmodelsinIS suchthatnoothermentalmodelinIS ismoreresponsivewhenconjoinedwithQ andBGbeJBG,S.M isamemberofJBG,S.
(Simplicity)LetKBG,SbethesetofmentalmodelsinJBG,S suchthat,foreachmodelconjoinedwithBG, theresultingmodelissuchthatnoothermentalmodelinJBG,S conjoinedwithBG hasfewerconstituents.35 M isamemberofKBG,S.
IwillassumethatBGismadeupofthesamesortsofconstituentsasinterpretivemodels.Wenowneedaworkingaccountofhowback-groundquestions are represented inmentalmodels, andweneed
34. For now, I will postpone a discussion of what happenswhen there is nouniquementalmodel.Moreover,Iwillnarrowmyviewexclusivelytoliteral utteranceinterpretation.
35.Wewillcountconstituentsinourregimentedrepresentationofmentalmod-elsthewaywecountfreevariables.Forexample,‘xlovesxandwishesxtosucceed’hasonefreevariable;‘xlovesy’hastwofreevariables.Forexample,onthiswayofcounting,thementalmodelLOVE(JOHN,JOHN)hasonecon-stituentlessthanLOVE(JOHN,MARY).
encodes instructions that allowvariousmentalmodels to be gener-ated.Wewillsaythatthesetofthesementalmodelscorrespondstowhatwouldbethesetofalternativehypothesesforinferencetothebestexplanation.Thebestinterpretationthencorrespondstothebestmentalmodel.Toobtainsubstantivepredictions,weneedaproposalaboutthefactorsthatdeterminewhichmentalmodelisthebest.Isug-gestthatonefactorissimplicity,inthesenseofthenumberofmentalmodelconstituents.Morewillbeneeded.32Itisafamiliarmoveinthe-oreticaldebatestosayofasetofcompetingtheories(nonefalsifiedbyextantdata) that thebest theorywouldanswersuch-and-suchques-tionswhencombinedwithourbackgroundbeliefs.ThismaybepartofwhattheIBEtheoristwantstocapturewiththenotionofacertaintheorybeingmoreexplanatorythanothers.Isuggestthatliteralutter-anceinterpretationinvolvessomethingsimilar,proceedingrelative to a background question.Onemightplausiblysupposethat,bydefault,thebackgroundquestion is something like, “Whataresome factsabouttheenvironmentIdon’tknowyet?”,whichisnarrowedtosomethingmorespecificinalmostanyactualdialogueandmaybeshiftedto,e. g.,“Whataresome facts Idon’tknowyet thathold in theworldof theIliad?”,say,whenwelistentoastory.Iwillcallthevirtueofaninter-pretationof answering relativelymorequestions “responsiveness”. Iproposethat,forahearerH,inputstoliteralinterpretationareasen-tence,backgroundbeliefsofH,andquestionsH wouldliketoanswerbyengagingindiscourse.Moreover,Iproposetodefinetheprinciplesofinferencetothebestinterpretationusingresourcesindependentlyrequiredbythementalmodeltheoryofreasoning.33Toavoidconfu-sionwithAtlasandLevinson’sproposalsthatdonotassignaspecial
32.Needlesstosay,itisaseparatequestionwhetherinferencetothebestexpla-nationisthenormativelycorrectaccountofhowweshoulddrawconclusionsabouttheoriesinscience.Totheextentthatwehavereasontobelievethatit iscorrect, there is thepotential toarguethat theproposedviewofutter-anceinterpretationisarational modelinafairlystrongsense.ComparetheGriceanprojectofaccountingforimplicaturesintermsofarationaltheoryofcommunication.
33. Johnson-Laird(2008)
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –13– vol.12,no.14(september2012)
moreor less responsive than its alternatives,with the consequencethatJBG,S = IS.Finally,toforestallapossiblemisunderstanding,notethattheprinciplesof(IMRI)applyonlytoliteralutteranceinterpretation;theydonot,intheirpresentform,yieldpredictionsaboutmoregen-eral “conversational implicatures” that are not literal interpretations.Bydesign,(IMRI)onlygeneratespredictionsabouthowsentencesareinterpretedrelativetoaparticularanalysisoftheirlinguisticmeaning,whichhastoleaveopenarangeofliteralinterpretationsinorderforthepredictions tobenontrivial. (IMRI) isconcernedwithobtainingliteralinterpretationsinthefaceofsemanticnonspecificity.ClassicalGriceanpragmaticsbeginsonlyoncewehavealiteralinterpretation,anditcould,inprinciple,becombinedwith(IMRI).MovingbeyondGrice,a revieweralertedmeto theneedtodistinguish(IMRI) fromHobbsetal.’stheoryof“interpretationasabduction”.40Ontheirview,“theprocessofinterpretingsentencesindiscoursecanbeviewedastheprocessofprovidingthebestexplanationofwhythesentenceswouldbetrue”.41Ontheirtheory,theinputtotheirpragmaticprocessisase-manticallydeterminedlogicalformin“first-orderpredicatecalculus”.42 Thepragmaticcomponentthenproducesan“elaboratedlogicalform”,which entails the semantically determined logical form. The addedmaterial amounts to the explanation ofwhy the semantically deter-mined logical form is true.Bycontrast,onmyaccount, the input tothepragmaticcomponentdoesnothavedeterminatetruthconditions,whichastatementinpredicatecalculuswouldhave,merelydetermin-inga rangeofpossible interpretations.Theresultofpragmaticprin-ciplesisaninterpretationwithtruthconditions.Inaddition,myview,unlikeHobbsetal.’stheory,isdefinedintermsofresponsivenesstoabackgroundquestionthatthehearerseekstoanswer.Finally,abduc-tiveprinciplesthatwouldyieldanexplanationofwhyanutteranceis
question:theinterpretationcouldsomehowraiseaquestion.
40.Hobbsetal.(1993)
41. Hobbsetal.(1993),p.69
42. Ibid.,p.75
anaccountofhowmentalmodelsareconjoined.Iwillprovisionallyadopttheviewthatabackground-questionmentalmodelhasthesetofitsexclusivealternativeanswersasitsrepresentationalcontent.36 A background question is then represented in the mental modelframeworkasasetofmentalmodels,eachrepresentinganalterna-tive answer to the question.37 Johnson-Laird’s theory of reasoningwithmentalmodelsindependentlyrequiresmentalmodelsconsist-ingof furthermentalmodels representing alternatives.38 Iwill use‘[M1/M2/…/Mn]’tostandforamentalmodelthatconsistsofmentalmodelsM1M2,…,Mn,whereeachmentalmodelhasanalternativepropositionasitsrepresentationalcontent.
Ihaveencounteredtwoobjectionstotheviewjustpresented.Oneobjectioncentersonanallegedviciouscircularity.Theworry is thattheproposedtheoryof interpretationpresupposesaquestion that it-self needs tobe interpreted.But, of course, thementalmodel repre-sentingourbackgroundquestiondoesnothavetoenterourcognitiveeconomybyverbalmeans,anymorethanourbackgroundbeliefshavetocometousthroughaverbalbriefingbeforethebeginningofacon-versation.Anotherobjectionisthatitdoesnotseemintuitivelyimpos-sibletoapproachanutterancewithoutaparticularquestioninmind.Nothingin(IMRI)forcesustodenythisintuition.Ifwehavenoback-groundquestion—ifweareinastatethatwecanthinkofasanemptyquestionmodel—nointerpretationcan(withanimportantcaveat39)be
36.Note that, at least inprinciple, theviewaboutbackground questions for thepurposesof (IMRI)doesnot commitone to theview that this is the rightaccountofthesemanticsofinterrogativesentencesinnaturallanguage.SeeMascarenhas(2009).SeeHamblin(1973)fortheoriginalnotionofquestionsdenotingsetsofalternatives.
37. Since this yieldsmentalmodels with infinitelymany compartments for aquestionlike“Whatisyourfavoritenumber?”,wewillneedtosaysomethingaboutthepossibilityofrepresentingalternativesimplicitlyinmentalmodels,which Johnson-Laird’s (2008) theory of reasoning independently requires.Moreonthisbelow.
38. Johnson-Laird(2008)
39.Thereis,infact,awayinwhichwecouldusefullythinkofaninterpretationaslessresponsivethanitsalternatives,evenintheabsenceofabackground
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –14– vol.12,no.14(september2012)
Besidesmentalmodelswithmultipledistinctsub-models,wewillneedmentalmodelsofalternatives.Wewill takeamentalmodelofalternativestobeasetofsets—namelyasetofmentalmodels,whichare,inturn,setsofmentalmodelfeaturesboundtomentalmodelto-kens.Foreaseofexposition,wewilluse‘[’and‘]’insteadof‘{’and‘}’and ‘/’ insteadof ‘,’ torepresentsetsofmentalmodels.Forexample,‘[{GUILTY(SUSPECTA), ¬BELIEF(SMITH,UPSTANDING(SUSPECTA))}/{¬GUILTY(SUSPECTA)}]’denotesthesetofalternativementalmod-els comprising {GUILTY(SUSPECTA), ¬BELIEF(SMITH,UPSTAND- ING(SUSPECTA))}and{¬GUILTY(SUSPECTA)}.Moregenerally,ifM1 andM2 arementalmodels,then[M1 / M2] isamentalmodelrepresent-ingthealternativesrepresentedbyM1 andM2.Finally,wewilladdtoourinventoryaspecialprimitive“null”mentalmodelM∅ thatrepre-sentscontradiction.
WecannowgiveasimplifieddefinitionofJohnson-Laird’smainpro-cedureforconjoiningmentalmodels,whichappearstoworkasfollows:
(Conjoin) Given two mental models M1 and M2 , CONJOIN(M1 , M2) è MX.
If M2isamodelofalternatives[M2/M3/…],then
MX = [CONJOIN(M1 , M2)/CONJOIN(M1 , M3)/…].
If M2 is drawn froma set of alternativementalmodels[M2 /M3 /…]andasub-modelM ofM1isalsoasub-modelof at leastoneof thealternativemodels in [M2 /M3 /…] butnotasub-modelofM2,thenMX = M∅.
If M1 , M2arementalmodelsandthereisanMsuchthatoneofM1 , M2hasM asasub-modelandtheotherhas¬M asasub-model,thenMX = M∅.
truearefarmorepowerfulthanwhatIpropose.Inmyview,AsherandLascaridescorrectlycriticizeHobbsetal.forfailingtodistinguishlit-eralutteranceinterpretationfromageneralaccountofhowweupdateourbeliefs.43
Movingon,wewillalsoneedamentalmodelconstituentthatrep-resents negation,which is also already a component of thementalmodel theoryofdeduction.44 Iwilluse thesymbol ‘¬’ todenote thementalmodelconstituentrepresentingnegation,whenitoccursinthecontextofmentalmodels,andtodenotenegation,whenitoccursinthecontextofpropositions.
Next,weneed an account of howmentalmodels are conjoined.Johnson-Laird is committed to a certain procedure governing howmentalmodelsareconjoined.Thisprocedurecanberegimentedfor-mally,thoughasketchwillsufficeforthepurposesofthispaper,withmoretechnicaldetailgivenelsewhere.45
So far, we have talked only about mental models representingsimplepropositions. Forwhat follows,wewillneedmore structure.We will say that mental models are sets of mental model featuresboundtomentalmodel tokens.So,wherewepreviously just talkedabout amentalmodel SMOKE(JOHN) as representing the proposi-tionthatJohnsmokes,wewillnowtalkaboutthatmentalmodelasa set {SMOKE(JOHN)} that has, as its unique element, the featureSMOKEboundtothetokenJOHN.Torepresentboththepropositionthat Johnsmokesand theproposition thatMarysmokes inasinglementalmodel,wetake{SMOKE(JOHN),SMOKE(MARY)},whichhastwoelements,etc.Wewillsaythat{SMOKE(JOHN),SMOKE(MARY)}has {SMOKE(JOHN)}asasub-model.As inprevioussections, ifap-plicationsofmentalmodeloperationsarenotdirectlyatissueandwearedealingwithmentalmodelsthathaveuniquesub-models, Iwillsuppresssetnotationinexposition.
43. AsherandLascarides(2003)
44. Johnson-Laird(2008)
45. Koralus(in preparation)
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –15– vol.12,no.14(september2012)
inwhichnoalternativesinQ havebeeneliminated(i. e.,noreductioninoccurrencesof‘/’intheregimentation).
(E3)FortwointerpretationsI1andI2 thatruleoutsomebutnotallalternativeswhenconjoinedwithBG andQ, I1ismoreresponsive to Q than I2 iff more (explicit48) alternativemodels are ruled out byCONJOIN((CONJOIN(I1, BG)),Q) than byCONJOIN((CONJOIN(I2, BG)),Q) (i. e., oneeliminates more occurrences of ‘/’ in the regimentationthantheother).
Bothmentalmodelconjoiningandtheabilitytocomparenumbersofalternativesarecognitiveresourcesthatareindependentlyrequiredbythementalmodeltheoryofreasoning.49Thus,theproposedtheoryofinferencetothebestinterpretationcanbedefinedintermsofcogni-tiveresourcesthatareindependentlymotivated.Itmustbenotedthatthenumberofmentalmodelsthatcanbeentertainedatatimeisfinite.Yetpotentiallyopen-endedquestionslike‘Whowillcometotheparty?’maycorrespondtoanunboundednumberofalternativesthathavetoberepresentedbymentalmodels.Wecanonlyimplicitlyrepresentthementalmodels that represent those alternatives.Thementalmodel
48. ThoughIhavenospaceforaformaltreatmentinthispaper,onceweaddJohnson-Laird’s distinction between explicit and implicitly representedmentalmodels,we can account for the fact that, even given backgroundquestionswithinfinitelymanyalternatives,someincompleteanswersseembetterthanothersbecausetheymayeliminatemoreexplicitly representedal-ternatives.Forexample,suppose thatsomeonewhoknows thatMikehasagirlfriendaskshim,‘Whowillcometoyourparty?’IfMikeisanordinaryperson,itwouldseemmuchmoreresponsiveforMiketosay,‘Well,mygirl-friendwon’tcome’thantosay,‘Well,BillClintonwon’tcome.’TheaccountarguedforbyJohnson-Laird,Legrenzi,etal. (1999)suggests thatordinaryreasonerstendtorepresenttherelativeprobabilityofaneventviatherela-tivenumberofexplicitalternativementalmodelsthatrepresenttheeventasoccurring.Ifthisiscorrect,thenapartialanswerthatrulesoutaneventwe considered relatively likelywill be relativelymore responsive by (E3),because iteliminatesmoreexplicitly representedalternatives.There isnoneed to stipulate an independent preference for information-theoreticallymoreinformativeinterpretations.
49. Johnson-Laird(2008);Johnson-Lairdetal.(1999)
If M1 , M2 arementalmodelsandoneofM1 , M2 = M∅, thenMX = M∅.
Otherwise, MX = M1 ∪ M2.
A fullaccountof thementalmodel theoryof reasoningrequires fur-therprinciples,andsomerevisionsmaybenecessary.Yettheabovesufficestodescribethecoreof(IMRI)intermsofmentalmodelpro-cedures.Background-questionansweringcorresponds toconjoiningquestionmodelsandmodelsrepresentingputativeanswers.Wecanthendefineprinciplesgoverningrelativeresponsivenessasfollows:
(E1) An interpretation I is maximally responsive to Q ifCONJOIN((CONJOIN(I, BG)),Q) yieldsamentalmodelinwhich thealternatives inQ havebeenreduced toone(i. e., no occurrences of ‘/’ in the regimentation), distinctfromM∅. If the one remaining alternative isM∅,we saythatI rejectsthequestion.46
Itmaybehardforaninterpretationtobeless responsivethananinter-pretationthat leavesthebackgroundquestionas it isorthatrejectsthequestion.47
(E2)An interpretation I has zero responsiveness to Q ifCONJOIN((CONJOIN(I, BG)),Q) yieldsamentalmodel
46. Suitablyanalyzed,thismightbeobservedincaseslike ‘A:WhomdidJohnkiss?B:Johndidnotkissanyone.’
47. Forthepurposesofthispaper,I’msettingasidethepossibilityofaninterpre-tationgeneratingadditionalquestions.However,wecouldaddaprinciple(E4)along the following lines:An interpretation I isnegatively responsive(“inquisitive”)ifconjoiningIwithBGandQ yieldsamentalmodelwithmore alternatives than inQ. Such anotionmay capture the intuitive sense thatsome interpretations “raisemore answers than theyprovide”. It shouldbeexploredwhether there are language-independent reasons to believe thathighly controversial or unlikely propositions tend to independently raisequestionsinbeingcombinedwithbackgroundbeliefsandwhetherthisac-countsforwhywemayprefer,allotherthingsbeingequal,uncontroversialandlikelyinterpretations.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –16– vol.12,no.14(september2012)
correspondtointerpretationswithsomePoRfeatureonmentalmodeltokensoutsideoftheattitudecontext.Asnotedearlier,interpretationsofthissortarepossiblebutnonstandard.
2.5.1 Default interpretations in unembedded sentencesConsideranutteranceof(31)withnorelevantbackgroundbeliefsandadefaultbackgroundquestionalongthelinesof“WhataresomefactsabouttheworldIdon’tknowabout?”
(31) Johnsmokes.
As far as OIT semantics is concerned, MB could generateSMOKE(JOHN)oranynumberofmodels likeSMOKE(JOHNPOR) in-cludingsomePoRfeatureorother.Asdefinedearlier,thepropositionsrepresentedby thosemodelsare true inall thesamecircumstances,corresponding to the intuition that the truth conditionsof literal in-terpretationsof(31)arebasicallycontext-invariant.AsIsuggested,bydefault, thebackgroundquestion is about (non-mental) facts abouttheworld.Thismeansthat,givenadefaultbackgroundquestion,theliteralinterpretations(inthesenseof(IMRI))of||Johnsmokes||wouldtendtoruleoutthesamesetofalternatives.IfJOHNPORisrecognizedasacorrectexecutionof||John||,thenitmustbeabackgroundbeliefofthehearerthatanyordinaryfeatureapplyingtoJOHNPORalsoappliestoJOHNandviceversa.Asaresult,SMOKE(JOHNPOR)combinedwithbackgroundbeliefswould,atmost,beasresponsive inthesenseof(IMRI)asSMOKE(JOHN).However,SMOKE(JOHN)is thesimplestmodel,so(IMRI)correctlypredictsthataPoR-freesingularinterpreta-tionisthedefaultinterpretationof(31).
2.5.2 Default interpretations with attitude verbsWhyarewegenerallysogoodatmakingsenseofotherpeople?Ona veryplausible view,wehave considerable success at figuringoutotherpeople’smentallivesbecausewetendtoattributeourownback-groundbeliefsandsimpleinferencepatternstothepeoplewearetry-ingtounderstand,exceptforthosebeliefsandinferencepatternswe
theoryofreasoningitselffundamentallyrequiresapreciseaccountofimplicitmentalmodels,butextendingmyformaltreatmenttocapturethisfactis,asareviewerhasconvincedme,beyondthescopeofthispaperandistobediscussedelsewhere.50Wecannowconsiderwhatpredictionsthetheorymakesabouttheinterpretationofsimplesen-tencesaswellasaboutattitudereportsentences.
2.5 IMRI and the interpretation of attitude report sentencesOn the view presented so far, syntax and lexical items generate in-structionstocreatementalmodels.Dependingonthesentence,theseinstructionscanallowforarangeofmentalmodels.Giventhisrangeofmentalmodels, (IMRI)determines thementalmodel thatwill, infact,begeneratedandwhoserepresentationalcontentcorrespondstotheintuitivecontentoftheutterance.Theinstructionsencodedby(7)(seederivationattheendofsection2.3)allowforinterpretivementalmodelsofthefollowingsort:
(7) TerencebelievesthatRuthsmokes.
(27)BELIEF(TERENCE,SMOKE(RUTH))
(28)BELIEF(TERENCEX,SMOKE(RUTH))
…
(29)BELIEF(TERENCE,SMOKE(RUTHY))
…
(30)BELIEF(TERENCEX,SMOKE(RUTHY))
…
Thementalmodelsofparticular concernare (27), corresponding toasingular interpretation,andinterpretationsof theformof(29)cor-respondingtocomplexinterpretationsinvolvingPoRfeaturesofthesort involved in ‘Superman’ and ‘Paderewski’ cases. (28) and (30)
50.Koralus(in preparation)
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –17– vol.12,no.14(september2012)
Intheintroduction,Inotedthatwedonotintuitivelyinterpret(9)asmakinganyparticularclaimabouthowSmithrepresentsSuspectA,letalonethatshethinksofthesuspectas a suspect.For(9),thenaturalinterpretationissimple.
ThebackgroundquestioniswhetherSuspectAisguilty.Conjoin-ing thisquestionwith thebackgroundbelief thatSuspectA’sbeingguiltyis(simplifyingabit)incompatiblewithSmithbelievingofthatindividualthatheisupstanding,wegetamentalmodelofalternativesofthefollowingsort:
[{GUILTY(SUSPECTA),¬BELIEF(SMITH, UPSTANDING (SUSPECTA))}/{¬GUILTY(SUSPECTA)}]
Aninterpretivemodelofthefollowingsortwouldthenbemaximallyresponsive:
(32)BELIEF(SMITH,UPSTANDING(SUSPECTA))
Thekeyideaisthatifourbackgroundbeliefsandbackgroundques-tiondonot already involvedistinctionsofwaysof representinganindividual that bear on how the alternatives in a question can bereduced,an interpretivementalmodelofanattitude report that in-cludesPoRfeaturescannotbemoreresponsivethanasimpledefaultinterpretation. Unless our background beliefs conjoined with ourbackgroundquestionindependentlyleadtoasetofalternativeswheresomeofthosealternativesaredistinguishedbydifferentpropertiesofwaysofrepresenting,thennoPoR-ladeninterpretationisgoingtobemoreresponsivethanasimpleone.Ontheassumptionthatthealter-nativesdeterminedbybackgroundquestionsandbackgroundbeliefsnormallydonothingeondistinguishingdifferentPoRs,because,bydefault,wetendtobeinterestedinwhattheenvironmentislikeandwhatfactsabouttheenvironmentarerepresentedbypeople,(IMRI)predicts that simple interpretationsofattitude report sentences,cor-responding to singular propositions, are the default.One upshot isthatprocessingthelinguisticmeaningofanattitudereportsentence
havespecialreasontothinkarenotshared.51Onewaytofollowthisstrategywouldbetoexplicitlyrepresentthatweandotherpeoplerep-resentalike.However,thisseemsneedlesslycomplicated.Unlesswehaveaspecialreasontothinkthatourwayofrepresentingdivergesfromothers’ in respects thatmake a difference for our purposes, itseemseasiestsimplynottomarkanydifferences.Let’stakeacasewithadefaultbackgroundquestion(e. g.,“WhataresomefactsIdon’tknowyet?”)andabackgroundbeliefmodelthatincludesonlytheinforma-tionthatNancy,Ruth,andTerencearecolleaguesofthehearer.Now,Nancyutters,
(7) TerencebelievesthatRuthsmokes.
IfthesimplestwayforustorepresentthefactthatRuthsmokesistoentertain themodelSMOKE(RUTH), thenunlesswehavea specialreasontocontemplatethepossibilitythatTerencerepresentsthisfactinsomedifferentway,wecanjustinterpret(7)asBELIEF(TERENCE,SMOKE(RUTH)).Ifweassumethatwedonotdistinguishourwayofrepresentingfromthatofotherpeopleunlesswehaveaspecialreasontodoso,andweassumethatwenormallyrepresentfactsinthesim-plestwayavailabletous,thennocomplexinterpretationinacaselikethatof(7)isgoingtobebetterthanasimpleinterpretation.
Now, let’s return to theexample involvingdetectives ina covertinvestigation so secret that witnesses are questioned under a ruse.Again,imaginethattheyarediscussingwhatvariouswitnesseshavesaid,togetapreliminarynotionofwhoislikelytobeguilty.AllbelievethatitisveryunlikelythatanybodybelievedtobeofgoodcharacterbyacertainSmithisguilty.52Onedetectiveasserts(9)tohiscolleagues.
(9) SmithbelievesthatSuspectAisanupstandingcitizen.
51. NicholsandStich(2003),p.65
52. The fact that Smithmay not arrive at the right judgment if, say, Jack theRipperwereintroducedtoherinasheepcostume,andinnumerableotherpossiblethingsthatcouldcauseabreakdownofSmithasareliablejudge(drugs,sleepdeprivation,etc.),maysimplybefoldedintoageneralmeasureofuncertaintyofSmith’sreliabilityinpractice.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –18– vol.12,no.14(september2012)
CONJOIN({¬BELIEF(JOHN,SMOKE(RUTH)),BELIEF(JOHN,SMOKE(RUTH)),…},Q)
This reduces toCONJOIN(M∅, Q) and further reduces to [M∅] , re-gardlessofthecontentofQ. By(E1),thismeansthatthisinterpretationamountstorejectingthebackgroundquestion,changingthesubject.Theconversationcannotproceedinanormalincrementalfashion.
IfweinsteaduseaninterpretivemodelforthesecondsentencethatincludesaPoRfeature,thenthecollapseintoM∅ isblocked.However,nothingincontexttellsuswhatsortofPoRfeatureweshoulduse.Asaresult,neitherstayingwithadefaultinterpretationnordepartingfromitintheabsenceoffurthercontextualinformationmakesitpossibletoproceednormally.Thisseemstocorrespondwelltotheintuitivereac-tionthatwouldbeproducedby(33)intheabsenceofaspecialcontext.
In contrast to the case just discussed, the Paderewski examples,asnormallydescribed, includecontextual information thatsuggestswhatsortofPoRfeaturesweshouldincludeinourinterpretations.Pe-terdoesnotknowthatthefamouspianistbythenameofPaderewskiandthepoliticianbythenameofPaderewskiarethesameindividual.PeterfollowsanannouncementthatPaderewskiisplayingthepianoontheradioandcomestothinkthatPaderewski’sperformanceisrath-ergood.Someonereports:
(34)PeterbelievesthatPaderewskiistalented.
Inthiscase,weassumethatourmentalmodelofPeter’sbeliefsalreadyincludestwomentalmodeltokenswithdifferentPoRfeatures,wherebothof those tokens representPaderewski.Oneof the tokenshasa“pianist”PoRfeature,whiletheotherhasa“politician”PoRfeature.Thequestion,then,iswhichofthosetokenswillbeincludedininterpreta-tion,incasethisbecomesnecessaryforaresponsiveinterpretation.Itseems tome thatany successful theorywouldhave to invokeano-tionofarelativedegreeofsalienceofinformationestablishedintheinterpreter’s background in such cases. On the present theory, thiswouldmeana relativedegreeof salienceofdifferentmentalmodel
togetherwith(IMRI)neverrequiresconsideringmoredifferentPoRfeatures thanarealready included in thebackgroundquestioncon-joinedwith background beliefs, so the set of alternative interpreta-tionstoconsiderisnaturallylimited.53
2.5.3 Non-default interpretations involving attitude verbsSupposeTerencetellshiscolleagueNancy,
(33) JohnbelievesthatRuthsmokes.JohndoesnotbelievethatRuthsmokes.
Intuitively,wewould expect these utterances to give rise to a reac-tionlike“Hey,waitaminute!Whatisgoingonhere?”Facedwith(33)withoutspecialbackgroundinformation,itishardtofigureouthowtoproceedwithinterpretation.
Ontheproposedtheory,thisisnotsurprising.Supposewegiveadefaultinterpretationtothefirstsentencein(33).Wegettheinterpre-tivemodel:
BELIEF(JOHN,SMOKE(RUTH))
SupposethisthenbecomespartofourbackgroundbeliefmodelBG forthepurposesofinterpretingthenextutterance.Ifwenowweretogivethesecondsentenceadefaultinterpretationaswell,wewouldgetthefollowinginterpretivemodel:
¬BELIEF(JOHN,SMOKE(RUTH))
Now,toassesstheresponsivenessaccordingto(IMRI)ofthesecondinterpretation relative to abackgroundquestionQ andbackgroundbeliefsBG,wehavetolookatthenumberofalternativesin
53. Thisisnottosaythatthetheorypresentedsofarrequiresthatpeopleactu-allyconsiderafullrangeofexplicitlyrepresentedalternativeinterpretations.(IMRI)canbeseenasacompetencetheoryofutteranceinterpretationinaChomskyansense.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –19– vol.12,no.14(september2012)
becomesassociatedwiththatname.Thismayexplainwhychoosinganame that is sharedbywell-knownpeoplewithsociallydesirabletraitsratherthanundesirableonesisconsideredimportantbymanypeoplewhoarechoosinganameforachild.There isawidespreadintuitionthatcommonassociationswithanameinfluencewhattraitsareperceivedas salient,particularly in theabsenceofbetterknowl-edgeoftheperson.55Toaccountfortheexampleathand,wesaythat‘Superman’makesa“superhero”PoRfeaturemoresalient,while‘ClarkKent’makesa“reporter”PoRfeaturemoresalient.
Intherestofthispaper,IwillbrieflydiscusssomeoftherelativeadvantagesoftheOpenInstructionTheory.
3. Modeling other minds and interpreting attitude reports
Intuitively,(36)feelsatleastasproblematicas(33):
(36)Johnbelieves thatRuth smokes. Johnbelieves thatRuthdoesnotsmoke.
Thisintuitionisinteresting,sinceitwouldbehardtoarguethat(37)isinconsistent:
(37) <bel ief<John,<smoke<Ruth>>>>&<belief<John,<¬smoke<Ruth>>>>
Believingof Ruththatshesmokesseemsindisputablycompatiblewithbelievingof Ruththatshedoesn’tsmoke.Onemaysimplynotrealizethatone’sbeliefsareofthesameperson,so(37)canbetrueofJohnevenifJohnisn’tirrational.Ifthisiscorrect,thenweshouldbeabletohavedefaultinterpretationsforthetwosentencesin(36)withoutissue.Butthen,whywould(36)seematleastasbadas(33)?
55. Oneadvicecolumnnotes,“Youwanttoavoidthebaggageattachedto in-famouspeopleorplaces”(http://baby-name-generator.com/BabyNaming4.html,accessedon1/13/11).Abest-sellingbookonbabynamesnotes,“Onthefirstdayinschool,theteacherdoesarollcall.Theonlythingtheotherstudentsknowaboutachildiswhatheorshelookslikeandwhathisorhernameis.Kidsare likelyto formaquickopinionfromjust thosefacts”(Lansky1999).
constituents.54Wewouldsaythat,withinthebackgroundbeliefmodel,somementalmodeltokensaremoresalientthanothersandthatthebestinterpretationwilluseamentalmodeltokenwhoseoccurrencesalreadyinBGaremoresalientthanalternatives.Thisavenuedoesnotseemaltogetherunattractive.Thismeansweneedtoaddthefollow-ingprincipleto(IMRI)fromsection2.4:
(Salience) If there isnouniquememberofKBG,S, thenM is themodelwhoseconstituentsarethemostsalienttothehear-eratthemomentofinterpretation.
Finally,consideracaseinwhichweknowfromcontextthattherearemultiplewaysof representingan individualbutwhere the localcontextdoesnot telluswhichtouse foraparticularutterance.Sup-posesomebodyreports,
(35) LoisbelievesthatSupermanistalented.Loisdoesnotbe-lievethatClarkKentistalented.
Asbefore,withoutaddingaPoRfeaturetoourinterpretationof(35),wewill again be rejecting the question, as in the case of (33), dis-cussedabove.However,ourbackgroundbeliefmodelofLois’sbeliefswill include at least two sorts ofmentalmodel tokens for the indi-vidualSuperman:onewitha“superhero”PoR-featureandonewitha“reporter”PoRfeature.However,nothingincontextistellinguswhichofthesetokensshouldbepreferred.
Obviously,whatbreaksthesymmetryarethenames.Itshouldbefairlyuncontroversialthatpropernamesareassociatedwithvariousproperties,inawaythatnobodywouldtaketobepartoftheirlinguis-ticmeaning.Theseassociationsplausiblyinfluencewhatinformationis salient to us. Presumably, themore a proper nameoccurs in thecompanyofcertaindescriptiveinformation,themorethatinformation
54. ThiseffectivelymeansadoptinganideathatwassuggestedbyLewis,whowrotethattomakesenseofhowcertainexpressionsareinterpreted,itmaybenecessary“toappealtoasaliencerankingnotofindividualsbutratherofindividuals-in-guises”(Lewis1979).
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –20– vol.12,no.14(september2012)
right.Itmighthaveseemed,atfirst,thatbuildinganintegratedrepre-sentationofsomeone’sbeliefsbasedonsuccessivebeliefreports,justaswemightintegrateourownmentalmodels,isinnocuous.However,onacloser look, itamounts toasubstantive furtherassumption. Ingeneral,thebeliefrelationisnotclosedunderCONJOIN.
Inthecaseathand,beliefconsolidationtogetherwiththenatureofmentalmodelsguaranteesthatfurtherbeliefreportscan’tgenuinelyaddtoourmodelofJohn’sbeliefs,leavingusstuckwithBELIEF(JOHN,M∅).Toacertainextent,therenolongerremainsasubstantiveques-tionastowhatJohnbelievesonceweadmitBELIEF(JOHN,M∅),soitpracticallywouldnotmakesensetocontinuethediscourseiftheback-groundquestionisconcernedwithJohn’sbeliefs.Ifwedobelievethatthere is a substantivequestionofwhat Johnbelieves, thenwemayhaveamongourbackgroundbeliefs¬BELIEF(JOHN,M∅).Inthiscase,defaultinterpretationsof(37)wouldleadtoarejectionofthequestion,inthesenseof(E1)and(Responsiveness).
Ifweinsteaduseinterpretivemodelsfor(37)thatincludePoRfea-tures, then the collapse intoM∅ is blocked.However, as in earlierexamples, nothing in context tells uswhat sort of PoR featureweshoulduse.Theprinciplesofinferencetothebestinterpretationcan-notsettleonauniquebestinterpretation.Asaresult,itappearsthattheoddnessof(37)outofcontextisexplained.Adefaultinterpreta-tionwouldbasicallyforceustochangethesubjectofconversation,while there isnodeterminatenon-default interpretationthatcouldavoidthis.Insum,oneadvantageofOITisthatitcanbereadilyin-tegratedwith independentlymotivatedviews abouthowwemakesenseofotherminds,whichcanhelpexplaincertainintuitionsthatarepuzzlingifweexclusivelyfocusonpropositionalcontent.
4. Interpretations that diverge from recognized communicative inten-tion without appeal to charity
(IMRI)predictsthatthebestinterpretationofcertainattitudereportsincontextissuchthatthehearerknowsthatthatparticularinterpreta-tionisnot intendedbythespeaker.Onthispoint,(IMRI)divergesfrom
Ipropose thatwhatexplains the fact that (33),withouta specialcontext,seemsbadhastodowithafurtherbackgroundassumptionabout other people’s beliefs that we naturallymake in interpretingmultiple attitude reports. There seem to be good empirical reasonsto think thatwhenweprocess informationabout aperson’sbeliefs,weusethisinformationtobuildanintegratedrepresentationofthatperson’sbeliefs,anditseemsveryplausiblethatweordinarilyusethesamereasoningproceduresappliedtoourownmodeloftheworldtogaininsightintothebeliefsofotherpeople.56IntheframeworkofOIT,Iproposethatthisisreflectedinabackgroundassumptionthatwecanconsolidateourmentalmodelsofotherpeople’sbeliefcontentsinthesamewayinwhichwecanconjointhementalmodelsthatcorrespondtoourownbeliefs.Thiswouldamounttotheassumptionthatthebe-lowconsolidationruleisfreelyapplicable.
(E4)CONSOLIDATE(BELIEF(T, M1),BELIEF(T, M2))èBELIEF (T, CONJOIN(M1, M2)
Withthisconsolidationassumption,ifwegivethesecondsentencein(37)adefault interpretation,afteradefault interpretationofthefirstsentencehasalreadypassedintoourbackgroundbeliefs,weget
BEL IEF ( JOHN,C O N J O I N ( {¬SMOKE(RUTH) } ,{SMOKE(RUTH)}))
whichreducesto
BELIEF(JOHN,M∅)
Thismeansthat,givenourbackgroundassumptionofbeliefconsoli-dation, (36)uttered in theabsenceof a special context seems to at-tribute irrationality to John. Togetherwith a further plausible back-groundassumptionthatpeopleareminimallyrational,wecanexplaintheeffect that suchanutteranceof (36) seems like it can’tbequite
56.NicholsandStich’s(2003)discussionofourmentalizingcapabilitiescontainsmanyobservationsthatsupportthisview.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –21– vol.12,no.14(september2012)
FromtheperspectiveofJohn,thereshouldbenointentiontoconveyaPoR.Fromhisperspective,thereisnomoreneedforaPoRtodistin-guishwhathetakestobetwoPaderewskisthanthereisaneedforPoRtodistinguishthetwoHarmansfromourperspective.IfJohneverhasanintentiontoconveyPoR-free,singularattitudereports,heshouldhavethatintentionwhenheutters(38)and(39).
WenaturallyinterpretthesesentencesinawaythatincludesPoRstocompartmentalizeMary’sbeliefsaboutPaderewski-the-pianistandherbeliefsaboutPaderewski-the-statesman.Butagain,thisinterpreta-tioncannotplausiblybewhat John intends toconvey.AnadvantageoftheviewIhaveproposedisthattheintuitiveinterpretationispre-dictedbythesamepragmaticrulesthatwouldpredictinterpretationsinordinarycases(thecasehereisparalleltothecaseintheprevioussection).Unlessthebackgroundquestionofthehearerspecificallyisconcernedwithhowthespeakerrepresentstheworld,ourknowledgeofwhat the speakerdoesordoesnot likely intend tocommunicatedoesnotinfluencehow(IMRI)determinesthebestliteralinterpreta-tion.57Fromtheperspectiveof(IMRI),theexamplediscussedinthissectionisjustafurthermanifestationofthefactthatourdefaultback-groundquestionsareprimarilyaimedattheenvironment.Bycontrast,Asher,whodiscussesasimilarexample,hastosaythat“defaultrulescanbeoverriddenbyaconstraint[hecalls]‘charity’”.58Onmyaccount,theintuitiveinterpretationsaretheresultofadirectapplicationoftherules,notanexception.
It is very plausible that we do often employ some kind of char-ity principle in interpretation. However, to deal with the case de-scribed, quite a strong principle would be needed. Recall that <belief<i,<P<x>>>>&<belief<i,<¬P<x>>>>isnotacontradiction,nordoesitentailanattributionofirrationality.Infact,therelevantinstan-tiationofthispatternintheexampleistrue!Thismeansthat,forexam-ple,ifweweretopreferaweakprincipleofcharityalongthelinesof
57. Throughout,wehavebeenconcernedexclusivelywithliteralinterpretation,notwithageneralclassofconversationalimplicatures.
58.Asher(1986),p.146
amoretraditionalviewofutteranceinterpretationaccordingtowhichthehearer,asareceiver,triestorecoverthemessagethatthespeaker,asthesender,intendstotransmit.
You and I know that not everybody knows that Paderewski-the-statesmanisthesamepersonasPaderewski-the-pianist.JohnandMaryaretwopeoplewhoareignorantinthisway.ImagineJohnseesMaryhandingPaderewskiacigarette.HethinksheislookingatPaderewski-the-pianist.Johncommentstous,
(38)MarybelievesthatPaderewskismokes.
Later, John sees a similar scene.About togive a speechas a states-man,PaderewskiisofferedacigarettebyMary.Thistime,Paderewskirefuses. Johnthinkshe isseeingPaderewski-the-politiciananddoesnotrealizethatheisseeingthesamepersonasbefore.Marysimilarlyremainsignorant.Johncommentstous,
(39)MarybelievesthatPaderewskidoesnotsmoke.
Inthecontextdescribed,weintuitivelyinterpret(38)asconveyingabeliefaboutPaderewskirepresentedinonesortofwayandinterpret(39)asconveyingadifferentbeliefaboutPaderewskirepresentedinanothersortofway.However,wecannotobtaintheseinterpretationsifutterance interpretationconsists in recovering thecommunicativeintentionofthepersonmakingtheutterance,asIwillnowargue.
Asnotedbefore,interpretationsofattitudereportsthatconveyPoRsareconfinedtocasesinwhichtherearespecialfeaturesintheback-groundthatsuggesttousthatwehavetodistinguishdifferentsortsofrepresentationsof the same individual.Yet, from John’s perspective,(38)and(39)requirenomorespecialeffortstomakefine-graineddis-tinctionsaboutrepresentationsthan(40)and(41)requireforus:
(40)Marybelievesthat[Elizabeth]Harmanisatthedepartmentreception.
(41) Mary believes that [Gilbert]Harman is not at the depart-mentreception.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –22– vol.12,no.14(september2012)
dependentonguessing intentions.62 Ihavepresentedsucha theorywiththeOpenInstructionTheorycombinedwith(IMRI).
5. Contrasting Discourse Representation Theory and “hidden anaphora”
The framework of Discourse Representation Theory (DRT) shareswiththeOpenInstructionTheorytheviewthatinordertomakesenseofutteranceinterpretation,weneedalevelofrepresentationbeyondwhatisgeneratedbysyntaxandlexicalitems.Now,AsherandKamp’sDRTaccountof attitude report sentences couldbe characterizedasa“hiddenanaphora”theory.63Thestartingpointforthistheoryisthenotion that the use of a proper namepresupposes that a referencemarkerstandingforthereferentofthenamealreadyexistsintherep-resentationofthediscourse,wherethisantecedentreferencemarkermustbelinkedtoasetofconditionsfordeterminingthereferent.Now,ifapropernameoccurswithinthethat-clauseofabeliefreport,theuseofthepropernamepresupposesthatthehearerhasanantecedentreferencemaker standing for the referent of the proper namewith-in a dedicated representation of the believer’s total cognitive state.64Asherholdsthat,“fromaDRtheoreticperspective,thegoalofthespeakeristogettherecipienttoapproximatethetruepicture(insofarasitisknowntothespeaker)ofthetargetBeliefanditsinternalconnectionstoothercomponentsofthebeliever’scognitivestate”.65Ifthereisnoantecedent representation representing referencemarkers in thebe-liever’scognitivestate,wehavetoaccommodatetheexistenceofsucharepresentation,including,onAsher’sview,connectionsto“schematic”
62.Atlas(2005)
63.Asher(1986;1993);Kamp(1988;1990)
64.Asher(1986),p.144.AlsoseeKamp(1990),p.41–87.Asher’spositionseemstobethatapropernameinanattitudereportcontextshouldintroducetwopresuppositions:oneforthehearerandoneforthetargetofthebeliefreport.IamnotsurewhetherKampwouldsaythatthebeliefreport“shifts”thepre-suppositionofthepropernametoapresuppositionaboutthebelieverorthatthebeliefreportgeneratesanadditionalpresupposition.
65. Ibid.
“maximizethetruthorrationalityinthesubject’ssayings”,59wewouldhavenoreasontodivergefromwhatwetakethespeakertohavehadinmindintheexampleatissue.(IMRI)allowsustoavoidhavingtopostulateaverystrongprincipleofcharityjusttodealwiththecaseunderconsideration.
OntheviewIhaveproposed,thepragmaticprinciplesthatarein-volved in arriving at literal utterance interpretations arenot directlyconcernedwith speaker intention, andmaking an appeal to a prin-cipleof”charity”isunnecessarytodealwiththeexamplediscussed.60 The speaker’s primary opportunity atmaking the hearer consider aparticularpropositionisinpickingsentenceswithacertainlinguisticmeaningthatconstrainstherangeofavailableliteralinterpretations.Whatquestionsthehearerwantstoanswerbyengagingindiscourseandwhatbackgroundbeliefsshehasdeterminewhethershewillin-cludeaPoRfeatureinherinterpretativementalmodelofanattitudereportsentence.Thequestion“Didthespeakerintendtoconveythiswayofrepresentingso-and-so?”arisesonlyifthehearerisespeciallyconcernedwithwhatexactlyisonthespeaker’smind.NothingIhavesaidrulesoutthatweoftendo particularlycareaboutrecoveringthepropositionthatthespeakerspecificallyintendedtobringacross.Theclaimissimplythatreflectionsaboutaspeaker’sintentionsarenotby defaultdrivingliteralutteranceinterpretation.
General utterance interpretation as the recognition of particularcommunicativeintentionsisnot“cognitivelyencapsulated”inFodor’ssense,soastrongpredictivetheoryishighlyunlikelyinthisarea.61Yet,asAtlashas argued,we can give a theoryof default literal interpre-tationswithsomepredictivesuccess, if sucha theory isnotdirectly
59.Blackburn(1994),p.62
60.Itisaseparatequestionwhatprinciplesgovernconversationalimplicaturesbeyondwhat is involved in literal interpretationandhow thoseprinciplesrelatetoperceivedintention.
61. Fodor(1983)
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –23– vol.12,no.14(september2012)
cognitivestates.As Iarguedwithrespect tovariousexamples,evenifwe do have additional information about peculiarities of people’scognitivestates,westillgo forsingular interpretationsbydefault. IfwehadthepresuppositionthatAsherandKamppostulate,itwouldbesurprisingifwegotdefaultsingularinterpretationsevenwhenweare inapositiontotakeintoaccounttheinternalconnectionsofthebeliefsofthetargetoftheattitudereport.Toinvolvearepresentationofanythingclose tosomeone’s “totalcognitivestate” in interpretinganattitudereportseemstobetheexceptionratherthanthenorm,soitseemswrongtobuildapresuppositionofsucharepresentationintoattitudereports.
Evenifwegrantthat,withasuitablyrevisedversionofAsherandKamp’sviewthat relies lessonconditions fordeterminingreferentsinsteadofPoRs,acorrectlyaccommodatedpresuppositioncan fullymimicadefaultsingularinterpretationintermsoftheresultingcon-tentof theutterance incontext, there isstilladifferencebetweenatheory thatgets the result viaaccommodation anda theory thatgetsitviadefault interpretation.Sincepsycholinguisticshasbeguntoshedsome lighton theprocessingcostsofpresupposition failureandac-commodation,therelevantdifferencebetweenmyviewandtheDRTtheory of attitude reports is amenable to experimental inquiry. Vio-lationsofpresuppositions followedbyaccommodationyield longerreadingtimesonthecriticalwordthatengendersthepresupposition,asempiricallydemonstratedbyTiemannetal.(2011).Myviewpredictsthesamedefaultinterpretationsinvariouscasesindependentlyoftheavailabilityofbackground informationabout the cognitive statesofthetarget.AsherandKamp’sviewpredictspresuppositionfailureandsubsequent accommodation for cases without relevant backgroundinformation,butnotforcaseswiththerightbackgroundinformation.AssumingTiemannetal.’sresults,thismeansthatAsherandKamp’sviewpredictsadifferenceinreadingtimesbetweencertaincasesthatone could construct, where my theory predicts no such difference.Designingacarefullymatchedsetofcases foranactualexperiment
representations of the believer’s conditions for determining the ref-erent.66Differentwaysof representing an individualwould roughlycorrespondtodifferentinternalconnectionstoothercomponentsofthebeliever’s cognitive state.Oncewehave at least accommodated arepresentationaltokeninarepresentationofsomeone’stotalcognitivestate,wegetaninterpretationofapropernameinanattitudereportbyresolvingtheanaphorapostulatedinthepropernametothattoken.
This view faces a versionof themeaning-intentionproblemdis-cussedintheintroduction.Whilethereisasenseinwhichtheuseofpropernames tends topresuppose familiaritywith the referent,67 itjustdoesnotseemrighttosaythatpropernamesinthethat-clausesof attitude reports carry thepresuppositionof anantecedent repre-sentationof the cognitive stateof thebearerof a reportedattitude.SupposeMike seesananonymousvalentine in John’smailboxandtellshisneighbor,
(42)SomeonelikesJohn.68
ItisimplausiblethatthisutteranceinanywaypresupposesthatJohn’sneighbor has an antecedent representation of “someone’s” mentalstate,anymorethanMike’sutteranceconveysacertaintypeofwayofrepresentingJohn.OntheDRTaccount,therearenodefaultsingularinterpretations,becausethecoreideaisthatwehavetotieourinter-pretationstoanexistingrepresentationofthetarget’stotalcognitivestate. Insteadofdefault singular interpretations,youhavea formofaccommodationifthepresuppositionofanexistingrepresentationofthetarget’s totalcognitivestate fails.However,whatIarguedintheintroduction, inseemingagreementwithSchifferandJascszolt,wasthatunlesstherearespecialcontextualfeaturesthatpromptustodootherwise,ourinterpretationsofattitudereportsareconcernedwiththeenvironmentratherthandirectlywiththepeculiaritiesofpeople’s
66.Asher(1986),p.145;Kamp(1987),p.172
67. SeeHeim(1982).
68.I take forgranted that therelevantsenseof ‘likes’behaves likeanattituderelation.Cf.‘LoislikesSuperman,butsheonlytoleratesClarkKent.’
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –24– vol.12,no.14(september2012)
respecttoanindividualreferent.Aninterestingfeatureofindexicalexpressionsisthatonecannotshifttheirinterpretationacrossvariousellipsisconstructions.Considerthefollowingsentenceswithuncon-troversialindexicalexpressions:
(43)Loisownsthat,andsodoesMary.
(44)[LoishatesonlyScorsese,andMaryhatesonlyLukas]Loishithim,andsodidMary.
(45)Loiscameyesterday,butMarydidnot.
TheindexicalintheelidedVPcannotindependentlycontributeanov-elreferentininterpretation.Wehavetointerpret(43)asclaimingthatLoisandMaryownthesamething(orkindofthing),(44)asclaimingthattheyhitthesameperson,and(45)asclaimingthatMarycameonadifferentday.Thattheseconstraintsexistisnotparticularlycontro-versial,thoughthereisanopenquestionastowhatgeneralprincipleaccountsforthem.AtleastpartoftheconstraintseemstobeduetoagrammaticalrequirementthattheelidedVPhastobesyntactically“parallel” to its antecedentVP in certainways.70On a classical viewonwhichsyntaxencodes indices for indexicals (or some functionalequivalent),aversionofthisprinciplecanexplaintheconstraintsoninterpretationintheexamples.Incontrasttoindexicals,certainotherkindsofexpressionsthatalsoinvolveinterpretationsthatcanvaryincontextdonotseemtoincurthisconstraintoninterpretationunderellipsis.
Anon-indexicalkindofcontextsensitivityisplausiblyinvolvedinthefollowingexample:
(46)[MikeisthrowingapartyinNewYork;Bill isthrowingapartyinBoston.]
70.FiengoandMay(1994)
wouldbeadifficulttask,butthepointisthatthedebateisamenabletoexperimentalinquiryinprinciple.
AgeneralworryaboutDRTapproaches,raisedbyLevinson(2000),isthatthereisalackofpredictivepoweraboutwhichinterpretationswillbedominant invarious typesof contexts.69Different interpreta-tionsofattitudereportsentencesaretreatedasdifferentwaysofresolv-inganaphora(hencethecommitmenttoapresuppositionofasuitablerangeofantecedents),withoutafullysystematicdecisionprocedurebetweencompetinginterpretations.Bycontrast,oneofthekeymoti-vationsofthetheoryIhavepresentedwastoaccountforwhatseemedtobestrikingregularitiesinhowcontextinfluencestheinterpretationof attitude report sentencesand in the sortsof intuitionsgeneratedby“null”ordefaultcontexts.Moreover,everyeffortwasmadetobuildthosepartsofOITthatarenotsyntacticallyconstrainedoutofcompo-nentsthathavebeenindependentlypostulatedinthepsychologyofreasoning,while littleeffortseemstohavebeenmadetoconcretelymotivatethecomplexmachineryofDRTindependently.Thatsaid,iftheworries justdescribedareproperly takenintoaccount, thereap-pearstobenoin-principlebarriertointegratingvariousinsightsDRThasyieldedintovariousotherfragmentsoflanguagewiththeOpenInstructionTheoryofattitudereports.
6. Interpretation shifts across VP ellipsis
Asnotedintheintroduction,therearedifferentwaysinwhichthelin-guisticmeaningofattitudereportsentencescouldrelatetothevarioustypesof interpretations thatwecanobtain indifferentcontexts.OntheOpenInstructionTheory,asontheviewsproposedinrecentworkbySoamesandBach,thelinguisticmeaningissilentonthedifferencebetweentheseinterpretations;itisnonspecificwithrespecttothem.
Nonspecificity theories contrast with views on which attitudereport sentences includea constituent that is indexical with respecttotypesofwaysofrepresenting, justas,say, ‘that’ is indexicalwith
69.Levinson(2000),p.248
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –25– vol.12,no.14(september2012)
(48)SpielbergbelievesthatHathawayistalented,butCoppoladoesnot<believethatHathawayistalented>.
Inthecontextprovided,(48)seemstrueandintuitivelyconveysthatCoppolaandSpielbergrepresentHathawaydifferently:Spielbergbe-lievesthatHathaway,presented as Viola,istalented,whileCoppoladoesnotbelievethatHathaway,presented as Sebastian,istalented.Ofcourse,theybothbelieveofHathawaythatsheistalented.
Consideranotherscenario:LikemostordinaryGothamCitycrimi-nals,mafiabossZuccoisterriblyafraidtomeetBatman.However,heisverykeentomeetBruceWayne,becausehewantstoconvincehimtojoinhisPonzischeme.HedoesnotknowBruceWayneisBatman.Bycontrast,TheRiddlerhasnointerestinWaynebutwantstochallengeBatmantoafight.Bruce’sbutler,whoknowsallofthis,sumsitup:
(49)The Riddler hopes to meet MasterWayne, and so doesZucco.
Intuitively, (49) can convey in context thatTheRiddlerbears an at-titude towardWaynerepresented inoneway,whileZuccobearsanattitudetowardWaynerepresentedinanotherway.
Examples like (48) and (49) suggest that the relevantdifferencesinhowwecaninterpret‘believesthatF(x)’incontextismoresimilartothesortofcontextsensitivitywefindin‘come’thantomoregram-maticallyconstrainedcontextsensitivityofindexicals.Theavailabilityof interpretations like theonewecanobtain from(48) incontext is,attheveryleast,surprisingonthehypothesisthatattitudereportsen-tencesareindexicalwithrespecttotypesofwaysofrepresenting.OntheOpenInstructionTheory,thelinguisticmeaningissilentontypesofwaysofrepresenting,soitisnotsurprisingthatwedonotfindrel-evantgrammaticalconstraintsoninterpretation.Notethattheissueisnotwhethercontextscanshiftmid-sentence;theyclearlycanbeshifted.Theissueisthatsomechangesininterpretationseemtobegrammati-callyblockedregardlessofcontextwhileothersarenot.Attitudereport
(47)MikehopesMiawillcome,andsodoesBill.Oneofthemwillbedisappointed.71
Here,MikeandBillarecharacterizedashopingthatMiawillarriveatdifferentlocations,eventhough‘hopesMiawillcome’iselided.Whileit is possible to shift relevant aspects of context across ellipsis con-structions,indexicalexpressionsdonotseemtobeabletoshifttheircontribution across ellipsis, in contrast to cases like (47). There aregrammaticalconstraintsonhowindexicalscanbeinterpretedunderellipsis,regardlessofwhetherwehaveashiftofcontextmid-sentence.
A theory like theOpen InstructionTheory thatcharacterizesatti-tudereportsentencesasnonspecificwithrespecttotypesofwaysofrepresentingpredictsthatitshouldbepossibletofindcasesinwhichweswitchour interpretationofaVPlike ‘believesthatF(x)’withre-specttotypesofwaysofrepresentingacrossellipsis,givenasuitablecontext.Inotherwords,weexpectthat,withrespecttotypesofwaysofrepresenting,attitudereportsentencesshouldpatternwith‘come’,ratherthanwith‘that’or‘him’.
Considerthefollowingscenario:SpielbergandCoppolahaveanunshakablebeliefthatHathawayisatalentedShakespeareanactress,andbothinvitehertostarintheirrespectiveupcomingrivalproduc-tionsofShakespeare’sTwelfth Night.Asapersonalchallenge,Hatha-way decides to audition incognito for both Spielberg andCoppola,appearingasViola forSpielbergandasSebastian forCoppola.Weknowaboutallofthisandhaveaninformantwhospiesonbothaudi-tions.HathawayhasabaddayanddoeswellforSpielbergbutnotforCoppola.Ourinformantreports,
71. Ifoneisworriedaboutthepossibilityofareflexiveinterpretationof‘come’,onecouldchangethescenarioto:‘MikeisthrowingapartyinNewYork;Billis inTimbuktu,buthissoniscelebratinghis21stbirthdayinBoston.MikehopesMiawill come,andsodoesBill.Oneof themwillbedisappointed.’Contrastthiswith‘MikeisthrowingapartyinNewYork;BillisinTimbuktu,buthissoniscelebratinghis21stbirthdayinBoston.#MikehopesMiawillbehisguest,andsodoesBill[understoodinacrossedway].’
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –26– vol.12,no.14(september2012)
OITdirectlystartswiththeproblemofgivingasystematicaccountoflinguisticmeaning.Thelinguisticmeaningofasentenceisfullycom-positionallydeterminedbyrulescontinuouswithgenerativegrammar,andtheinterpretationsatissuearestillliteral interpretations.Therangeof literal interpretations is systematically determined by linguisticmeaning,notjustassumed;thereisnorelianceonageneral-purposeenrichmentprocess.Thepragmaticcomponentof thetheorymakesclearpredictionsabouthowparticular featuresof contextdrivepar-ticularinterpretationsandappearstopredictcorrectlythedefaultin-terpretationsofattitudereportsentencesaswellastherangeofnon-defaultinterpretations.
8. Mental models, content, and truth conditions
OntheOpenInstructionTheory,attitudereportsentences,takenbythemselves, do not have truth conditions. Their contributions arepartiallyopeninstructionstoconstructmentalmodels,whichconsti-tutepossibleinterpretationsofutterancesofattitudereportsentences.Mentalmodels expresspropositions thathave truth conditions.Wemay identify thetruthconditionsof themodelwiththetruthcondi-tionsofthepropositionitexpresses.Whenanutteranceofanattitudereportsentencecanbeassociatedwithacorrectinterpretation—thatis,acorrectkindofmentalmodel—(orifallcorrectmodelshavethesametruthconditions),weidentifythetruthconditionsofthatutter-ancewiththetruthconditionsofitscorrectmentalmodel(s).
Not every sentence in natural language can be associated withunique truth conditions. On this point, theOpen InstructionTheo-ryisinagreementwithearlytheoristslikeAtlasandChomskywhoheld that linguisticmeaningsareoftensemanticallynonspecific,aswellaswiththeoristslikeBachandSoames.75Atthesametime,wegivespecificinterpretationstoutterancesofsemanticallynon-specificsentences, because (IMRI) assigns interpretivementalmodelswithspecifictruthconditions.Bydefault,theinterpretation(IMRI)assigns
75. Atlas(1977;2005);Chomsky(2000);Soames(2004);Bach(2000)
sentencesdonotseemtorelevantlypatternwithindexicalexpressionsinthisregard.
Schifferhasexploredother syntacticargumentsagainst indexical-ismaboutattitudereports,promptingvariousrevisionstohisview.72 Iargueelsewhere that the intuitionsdiscussedabovepresentachal-lengetovariousversionsofindexicalismthatseemssurprisinglyhardtoavoidbytechnicalmodifications,butIdonothavethespacetopur-suethisissuehere.73Forthepurposesofthispaper,IsimplywanttoobservethatitisavirtueoftheOpenInstructionTheorythatitmakesitunsurprisingthatthewayattitudereportsbehaveunderVPellipsispat-ternswithcaseslike‘come’ratherthanwithcaseslike‘that’,‘him’,etc.
7. The overgeneration objection
Onegeneralobjection thathasbeenraisedagainst theories thatdonotderivealldifferencesininterpretationfromassociateddifferencesin semantics is that those theoriesovergenerate interpretations.Forexample,Stanleyhasattacked so-called “freepragmaticenrichment”theories on these grounds.74 Unlike the “free enrichment” tradition,
72. Schiffer(1992);Ludlow(1995)
73. SeeKoralus(2010).Iwilljustbrieflynotethatitwillnotdo,forexample,topostulateareflexiveconstituentratherthananindexical,ontheanalogyof‘hismother’. It isclearlypossibleto interpret ‘Johnlikeshismother,andsodoesJack’assayingthatJohnlikesJohn’smotherwhileJacklikesJack’smoth-er.Onethenmighttrytocircumventtheellipsisproblembyabandoningin-dexicalisminfavorof“reflexical-ism”aboutattitudereportsentences,whichwouldroughlyinvolveparaphraseslike‘Peter1believesthatPaderewskicanfly,involvinghis1PoR.’Figuringoutthereferentof‘hisPoR’isratherdifferentfromfig-uringoutthereferentof,e.g.,‘John’smother’.Johnpresumablyhasonlyonemother.Bycontrast,Peterhasatleasttwotypesofwaysofrepresent-ingPaderewskifortheindividualPaderewski,andhecouldhavearbitrarilymany.WhichPoRentersintoaninterpretationofanattitudereportsentencedependsoncontext.Giventhedialecticalsituation,itwillnotdotosaythat‘hisPoR’isambiguousorindexicalaswellasre-flexive.ThiswouldbringusbacktotheVPellipsisproblem.AfullerdiscussioncanbefoundinKoralus(2010),whereIalsoconsiderandrejecttheproposalthattheinterpretationatissuecouldbecapturediftheindexicalresidesonwhatsyntacticianscall“little”vP.
74. Stanley(2005);Hall(2008);CarstonandPowell(2006)
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –27– vol.12,no.14(september2012)
thetheoryofutteranceinterpretationbeyondsyntaxandthelexiconisapathtowardtheunityofcognitivescience.
References
Asher,N.(1986).“Beliefindiscourserepresentationtheory”.Journal of Philosophical Logic15(2),pp.127–189.
Asher,N. (1993).Reference to Abstract Objects in Discourse.Dordrecht:Kluwer.
Asher,N.,andLascarides,A.(2003).Logics of Conversation.Cambridge:CambridgeUniversityPress.
Atlas,J.(1977).“Negation,ambiguity,andpresupposition”.Linguistics and Philosophy1(3),pp.321–336.
Atlas, J. (2005). Logic, Meaning, and Conversation: Semantical Underde-terminacy, Implicature, and Their Interface.Oxford:OxfordUniversityPress.
Atlas,J.,andLevinson,S.(1981).“It-Clefts,Informativeness,andLogi-calForm:RadicalPragmatics(RevisedStandardVersion)”.In:PeterCole(ed.),Radical Pragmatics.NewYork:AcademicPress.
Bach, K. (2000). “Do Belief Reports Report Beliefs?” In: K. M. Jaszczolt (ed.), The Pragmatics of Propositional Attitude Reports. Oxford: Oxford University Press.
Barrouillet, P., Grosset, N., and Lecas, J. (2000). “Conditional reason-ing by mental models: chronometric and developmental evidence”. Cognition, 75(3), pp. 237–266.
Bauer, M. I. and Johnson-Laird, P. N. (1993). “How Diagrams Can Im-prove Reasoning”. Psychological Science 4(6), pp. 372–378.
Blackburn, S. (1994). The Oxford Dictionary of Philosophy. Oxford: Oxford University Press.
Braun, D., and Saul, J. (2002). “Simple Sentences, Substitutions, and Mistaken Evaluations”. Philosophical Studies 111(1), pp. 1–41.
Byrne,R.(2005).The Rational Imagination: How People Create Alternatives to Reality.Cambridge:MITPress.
isthebestinterpretationrelativetoourbackgroundbeliefsandthequestionwearetryingtoanswerbyengagingindiscourse,whichmaynotbewhatwehavebestreasontothinkisintended bythespeaker.Bydefault,ourbackgroundquestionstendtobedirectedtowardtheen-vironmentandtowardwhatpeopletaketheirenvironmenttobelike,ratherthandirectedtowardthementalstatesbymeansofwhichtheyrepresenttheenvironment.However,ifitispartofourbackgroundquestionthatwewanttoknowspecificallywhatthespeakerthinks,then(IMRI)allows for thepossibilityofdifferent results.Moreover,nothing I have said rules out that oncewehave a literal utteranceinterpretation,traditionalGriceanpragmaticsmayberesponsibleforgeneratingconversationalimplicatures.
9. The unity of cognitive science: from the principle of linguistic con-straint to the principle of psychological constraint
Oneofthemainattractionsofaccountsofliteralutteranceinterpreta-tionpurelyintermsofthesemanticsoflexicalitemsandtheirmodeofcombinationisthattheirinteractionwithsyntacticconstraintscanoftenyieldsurprisingpredictions.Purelysyntacticobservationsareanindependent sourceofdata that constrain suchaccounts.As Ihaveargued,alongwithAtlas,Bach,Chomsky,andothers,notallaspectsofinterpretation,even“literalinterpretation”,aregoingtoyieldtoanaccountpurelyintheseterms.
Isuggestthatthelessontobedrawnisnotthatlegitimatetheoriz-ing is possibleonly ifwe stipulate that everydifference in interpre-tationhastobetraceabletoadifferenceinsyntaxandlexical items.Rather,thelessonshouldbethatweshouldproceedinawaythatputsus in contactwith independent sourcesofdata.Thispaper showedhow cognitive psychology can be used to independently motivateprinciplesofatheoryofutteranceinterpretationbeyondsyntaxandlexicalitems.Ifwearediscussingdifferencesininterpretationthatarenotreflectedinsyntax,weshouldfinddifferentsourcesofevidencefor themechanismsweinvoke.Cognitivepsychology isagoodcan-didatesourceforindependentevidence.Themostpromisingpathfor
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –28– vol.12,no.14(september2012)
Johnson-Laird, P. N. (1983). Mental models: Towards a Cognitive Science of Language, Inference, and Consciousness. Cambridge, MA: Harvard University Press.
Johnson-Laird, P. N. (2008) “Mental models and deductive reasoning”. In: L. Rips and J. Adler (eds.), Reasoning: Studies of Human Inference and Its Foundations. Cambridge: Cambridge University Press.
Johnson-Laird, P. N., Legrenzi, P., Girotto, V., Legrenzi, M., and Caverni, J. (1999). “Naïve Probability: A Mental Model Theory of Extensional Reasoning”. Psychological Review 106(1), pp. 62–88.
Kamp,H.(1981).“Atheoryoftruthandsemanticrepresentation”.In:J.A.G. Groenendijk, T.M.V. Janssen, and M.B.J. Stokhof (eds.),Formal Methods in the Study of Language.Amsterdam:MathematicalCentreTracts135,pp.277–322.
Kamp,H.(1987).“CommentsonStalnaker,BeliefAttributionandCon-text”.In:R.H.GrimmandD.D.Merrill(eds.),Contents of Thought.Tucson:UniversityofArizonaPress.
Kamp,H.(1990).“Prolegomenatoastructuraltheoryofbeliefandoth-er attitudes”. In:C.A.Andersonand J.Owens (eds.),Propositional Attitudes: The Role of Content in Language, Logic, and Mind.Stanford:CSLIPublications.
Kaplan,D.(1989).“Demonstratives”.In:J.Almog,J.Perry,andH.Wet-tstein(eds.),Themes from Kaplan.Oxford:OxfordUniversityPress.
Knauff,M.,Mulack,T.,Kassubek,J.,Salih,H.R.,andGreenlee,M.W.(2002).“Spatialimageryindeductivereasoning:afunctionalMRIstudy”.Cognitive Brain Research13(2),pp.203–212.
Koralus, P. (2010).Semantics in Philosophy and Cognitive Neuroscience: The Open Instruction Theory of Attitude Report Sentences, Descriptions, and the Necker Cube. Doctoral dissertation, Princeton University.
Koralus, P. (2011). “Descriptions, Ambiguity, and Representational-istTheoriesofInterpretation”.Philosophical Studies[Epubaheadofprintdoi:10.1007/s11098–011–9759–5].
Koralus,P.(in preparation).“AxiomaticMentalModelTheory”.Kripke,S.(1979).“A Puzzle about Belief”. In: A. Margalit (ed.), Meaning
and Use. Dordrecht: D. Reidel.
Carston, R., and Powell, G. (2006). “Relevance Theory—New Directions and Developments”. In: Lepore, E., and Smith, B. (eds.), The Oxford Handbook of Philosophy of Language. Oxford: Oxford University Press.
Chomsky,N. (1965).Aspects of the Theory of Syntax.Cambridge:MITPress.
Chomsky,N.(1986).Knowledge of Language: Its Nature, Origin, and Use.NewYork:Praeger.
Chomsky,N.(1995).The Minimalist Program.Cambridge:MITPress.Chomsky,N.(2000).New Horizons in the Study of Language and Mind.
Cambridge:CambridgeUniversityPress.Crimmins,M.(1992).“Contextintheattitudes”.Linguistics and Philoso-
phy15(2),pp.185–198.Fiengo, R., and May, R. (1994). Indices and Identity. Cambridge: MIT
Press.Fodor,J.(1983).The Modularity of Mind.Cambridge:MITPress.Groenendijk, J., and Stokhof, M. (1991). “Dynamic Predicate Logic”.
Linguistics and Philosophy 14 (1), pp. 39–100.Hall, A. (2008). “Free enrichment or hidden indexicals?”. Mind and Lan-
guage 23(4), pp. 426–456.Hamblin,C. (1973). “Questions inMontagueEnglish”.Foundations of
Language10(1),pp.41–53.Harman, G. (1965). “The Inference to the Best Explanation”. The Philo-
sophical Review 74(1), pp. 88–95.Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases.
Ph.D.thesis,UniversityofMassachusettsAmherst.Heim,I.,andKratzer,A.(1998).Semantics in Generative Grammar.Mal-
den,MA:Blackwell. Hintikka, J. (1969). “Semantics for propositional attitudes”. In: J. Davis
et al. (eds.), Philosophical Logic. Dordrecht: D. Reidel.Hobbs,J.R.,Stickel,M.,Appelt,D.,andMartin,P.(1993).“Interpreta-
tionasabduction”.Artificial Intelligence63(1–2),pp.69–142.Jaszczolt,K.M.(2000).“Thedefault-basedcontext-dependenceofbe-
liefreports”.In:K.M.Jaszczolt(ed.),The Pragmatics of Propositional Attitude Reports. NewYork:Elsevier.
philippkoralus The Open Instruction Theory of Attitude Reports
philosophers’imprint –29– vol.12,no.14(september2012)
Schiffer,S.(2000).“Propositionalattitudesindirect-referenceseman-tics”.In:K.M.Jaszczolt(ed.),The Pragmatics of Propositional Attitude Reports.Oxford:OxfordUniversityPress.
Soames,S.(1987).“Substitutivity”.In:J.J.Thomson(ed.),On Being and Saying: Essays for Richard Cartwright.Cambridge,MA:MITPress.
Soames.S. (2002).Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. Oxford:OxfordUniversityPress.
Soames,S.(2004).“NamingandAsserting.”In:ZoltanSzabo(ed.),Se-mantics versus Pragmatics.Oxford:OxfordUniversityPress.
Stanley,J.(2002).“Makingitarticulated”.Mind & Language17(1–2),pp.149–168.
Stanley,J.(2005).“SemanticsinContext”.In:G.PreyerandG.Peter(eds.),Contextualism in Philosophy: Knowledge, Meaning, and Truth.Oxford:OxfordUniversityPress.
Tiemann, S., Schmid, M., Bade, N., Rolke, B., Hertrich, I., Ackermann, H., Knapp, J., and Beck, S. (2011). “Psycholinguistic Evidence for Presuppositions: On-line and Off-line Data”. In: I. Reich et al. (eds.), Proceedings of Sinn & Bedeutung 15. Saarbrücken: Saarland University Press.
Zwicky,A.M.,andSadock, J.M.(1975). “Ambiguity testsandhowtofailthem”.In:J.P.Kimball(ed.),Syntax and Semantics 4.NewYork:AcademicPress.
Kroger, J. K., Nystrom, L. E., Cohen, J. D., and Johnson-Laird, P. N. (2008). “Distinct neural substrates for deductive and mathematical processing”. Brain Research 1243, pp. 86–103.
Lansky,B.(1999).Baby Names Around the World.Minnetonka:Meadow-brookPress.
Lewis,D.(1979).“ScorekeepinginaLanguageGame”.Journal of Philo-sophical Logic 8(1),pp.339–359.
Levinson, S. (2000).Presumptive Meanings: The Theory of Generalized Conversational Implicature.Cambridge:MITPress.
Ludlow, P. (1995). “Logical form and the hidden-indexical theory: areplytoSchiffer”.The Journal of Philosophy92(2),pp.102–107.
Mascarenhas, S. (2009). Inquisitive Semantics and Logic. M.Sc. thesis.Amsterdam:InstituteforLogic,LanguageandComputation.
McGlone,M.(2007).Doctoraldissertation,PrincetonUniversity.Muskens, R. (1996). “Combining Montague Semantics and Discourse
Representation”. Linguistics and Philosophy (19), pp. 143–186.Nichols, S., and Stich, S. (2003).Mindreading: An Integrated Account
of Pretence, Self-Awareness, and Understanding Other Minds.Oxford:ClarendonPress.
Pylkkänen,L.,andMcElree,B.(2007).“AnMEGstudyofsilentmean-ing”.Journal of Cognitive Neuroscience19(11),pp.1905–1921.
Récanati, F. (1993). Direct Reference: From Language to Thought. Oxford: Blackwell.
Récanati, F. (2002). “Unarticulated constituents”. Linguistics and Philoso-phy 25(3), pp. 299–345.
Richard,M. (1990).Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them. Cambridge:CambridgeUniversityPress.
Rips, L.J. (1983). “Cognitive Processes in Propositional Reasoning”.Psychological Review90(1),pp.38–71.
Schiffer,S.(1977). “NamingandKnowing”.Midwest Studies in Philoso-phy2(1),pp.28–41.
Schiffer,S.(1992).“BeliefAscription”.The Journal of Philosophy89(10),pp.499–521.