1

DiscourseandLogicalForm:

Pronouns,AttentionandCoherence

1 Introduction

Themeaningofthepronoun‘he’doesnotseemambiguousinthewaythat,say,thenoun‘bank’is.Yetwhatweexpressinuttering‘Heishappy,’pointingatBill,differsfromwhatweexpresswhenuttering it,pointingatadifferentperson,Sam. ‘he’ inthefirstcasereferstoBill,inthesecondtoSam.Andfurtherwithanutteranceof‘Amanalwaysdrives thecarheowns,’or ‘Amancame in.Hesatdown,’withoutanaccompanyingpointinggesture, ‘he’refersnotatall; itsinterpretation,instead,co-varieswithpossibleinstancesintherangeofthequantifier ‘Aman’.Thoseusesinwhich the pronoun is interpreted referentially are usually called ‘demonstrative,’andthoseinwhichitisinterpretednon-referentiallyareusuallycalled‘bound’.Weimmediately face two questions: what is the semantic value of a pronoun in acontext,giventhatitcanhavebothboundanddemonstrativeuses?And,sincethisvaluecanvarywithcontext,howdoescontextfixit?Thefirstquestionconcernsthesemanticsofpronouns,andthesecondtheirmeta-semantics.

Thereceivedviewanswersthefirstquestionbypositinganambiguity:boundusesof‘he’functionasboundvariables;demonstrativeones,bycontrast,arereferential.Thereceivedview’sanswer to thesecondquestion is that thesemanticvalueofabounduse co-varieswith possible instances in the range of a binding expression,and for demonstrative uses, linguistic meaning constrains, but does not fullydetermine, its semantic value in context. For instance, in uttering ‘He is happy,’pointingatBill,itslinguisticmeaningissupplementedbyapointinggestureinfixingthe referent of ‘he’, and, in general, a surrounding contextual cue, either anaccompanying demonstration, or an appropriate and adequately transparentspeaker intention, or evenboth, is required to fix the referentof ademonstrativepronoun.1Suchsupplementationisunderstoodtobeextra-linguisticorpragmatic,and,mostcertainly,notlinguisticallyencoded.

This paper challenges the received view on both counts:we reject the ambiguitythesis and any meta-semantics that invokes extra-linguistic determinants ofsemantic values in context.We argue that the semantic value of a demonstrative1Proponentsofthereceivedviewaremany.NotableonesareKaplan(1989.,b),King(2014a,b),King(2014a, b),Neale (2004),Reimer (1992). There are important differences between thesedifferentversions,buttheydonotmatterforourpresentpurposes.Webelievetheseviewsaremisguided,butourfocusisprimarilyhereonourpositiveview.Forourcriticismsofsuchviews,see[Redacted].

2

pronoun,likethatofaboundvariable,dependson,andcanco-varywith,antecedentexpressions; and further, the linguisticmeaning of a pronoun, on both uses, fullydeterminesitssemanticvalueinacontext.Inparticular,‘he’islikethefirstpersonpronoun ‘I’ since for both context automatically fixes a semantic value on anyoccasionofuse.Justas‘I,’givenitscharacter,referstothespeakerofacontext,sotoo,contrathereceivedview,‘he’,givenitscharacter,determinesitssemanticvalueas a functionof a fixed featureof context. On first pass, its semantic value is themaleatthe“centerofattention”inacontext.(Wewillprecisify“centerofattention”below.) Incontrasttothereceivedview,whatliesatthecenterofattentionisnotgovernedbyextra-linguisticmechanisms,speakerintentionsordemonstrations,butentirely by linguistic rules. More precisely, linguistic mechanisms govern thedynamicsofprominenceinadiscourse,renderingcertainsemanticvaluespreferredforpronounresolutioninacontext.Themeaningofapronounfullydeterminesitsreferentinacontext—itpicksoutwhateverlinguisticrulesdeterminetobe“atthecenterofattention”atthatpointinadiscourse.Inwhatfollows,wespellouttheserulesandhowtheyaffectpronounresolution.

As a clarification, note that our claim that linguistic conventions play a role inpronoun resolution is not altogether novel. Most theories of pronoun resolutionmaintain linguistic conventions play some role in fixing semantic value. Manybelieveconventionsplayaroleinconstrainingwhatspeakerscanevenbetakentointend—e.g.,inanormalcontext,onecannotreasonablyintendtorefertoawomanusing‘he’.Webelieve,however,suchconstraintsunderminethemainingredientoftheseaccounts,namely,thatspeakerintentions,oradditionalepistemiccues,playacrucialroleinfixingthesemanticvalueofapronoun.Aswehavearguedelsewhere,2the meaning of ‘he' is determined in conformity with linguistic conventionsregardlessofthepresenceorabsenceofaspeakerintention,oranepistemiccue.

Briefly,toseewhatwemean,supposewhileSueismakingherselfhighlysalientbyjumpingupanddown,youhavebeenasked, ‘Whocametotheparty lastnight,’ towhichyourespond,‘Marycametotheparty.Shehadfun.’Inuttering‘shehadfun’,you make no overt gesture towards Sue, not even a glance towards her, butneverthelessyouintendedtorefertoSue.ItisourintuitionthatwhatyouhavesaidisMarycametothepartyandhadfun,despiteyourintentionorSue’ssalience. Inshort,contrarytothereceivedview,anintentionisneithernecessarynorsufficientto fix a referent on an occasion of use. To say in such cases that linguisticconventions constrain what can reasonably be intended is to agree with us thattheseconventionsfixareferentonanoccasionofuse.

We also note that various authors argue that linguistic rules do play a role inresolving bound uses of pronouns, andmore generally, so-called anaphoric uses,

2Wehaveprovidedextensivecriticismofsuchaccountselsewhere.[Referenceredacted].

3

wherepronounresolutionisdependentonapriorlinguisticexpression.3Ourviewgoesbeyond,aswellasdiverges from, theseauthors inseveral respects.4Forone,weargue for aunique linguisticmeaning (a character) forall usesofpronouns—demonstrative, bound and anaphoric alike—and that, for all these uses, theinterpretation of a pronoun is fully determined by linguistic rules. Secondly, weelucidate a wide range of linguistic mechanisms that affect the dynamics ofprominenceinadiscourse,whichhaveeithergoneunnoticedinthecurrentdebate,or otherwise are confused with pragmatic effects on interpretation. Given themechanisms that maintain the dynamics of prominence in a discourse, anyoccurrenceofpronounpicksout its semanticvalueautomatically, according to itscharacter—selectingwhateverismostprominent,or“atthecenterofattention”.

Finally, we want to register that our view has far-reaching philosophicalconsequencesbeyondthesemanticsandmeta-semanticsofpronouns.Philosophersregularly invoke context-sensitivity in analyzingphilosophically interesting terms,suchas‘know’,‘good’,‘might’and‘all’,wheretheirsemanticvaluesareallegedtobedeterminedbycontextualparameterssuchastheepistemicstandards,standardsofprecision, value commitments, implicit restrictions on quantifier domains, etc. Inbuilding theories about how the values of these parameters are fixed in context,philosophers are invariably guidedby the receivedviewof themeta-semantics ofpronouns: namely, that the values of these parameters are determined jointly bylinguistic meaning and non-linguistic cues.5 Often, in fact, both the linguistic

3Our distinction between bound and demonstrative uses of pronouns does not perfectly coincidewithonebetweendemonstrativeandanaphoricusesofpronouns,whereananaphoricuseisanysuethe interpretationofwhichdependsonanexpression introducedearlier in thediscourse.Namely,someanaphoricpronounsareclearlyreferential.Towit,consider:‘Billcamein.Hesatdown.’Sincewewillarguethatallusesofpronounshavethesamesemantics,andareresolvedbyappealtothesamelinguisticmechanism,alldemonstrative,boundandanaphoricusesare,accordingtous,alike.4TheoristswhostresslinguisticconstraintsontheinterpretationofananaphoricpronountovariousdegreesincludeChomsky’sBindingTheory(cf.Chomsky(1981)aswellasvariousformsofdynamicsemantics (e.g., Kamp (1981) and Heim (1982)). Almost everyone agrees there are linguisticconstraintsontheresolutionofanaphoricpronouns.FiengoandMay(1994,2006),forinstance,holdthat any two co-indexed expression-tokens are tokens of the same expression type, and are co-referential as amatter of grammar, andpoint to a number of linguistic constraints on co-indexed,including, but not limited to, those of Binding Theory. Though we agree with these authors thatanaphoricdependencesareamatteroflingusticconventions,webothgobeyond,anddevergefromthem,inseveralrespects.First,weargueallpronounresolution—whetherdemonstrative,boundoranaphoric—is determined by a lingusitc rules. Secondly, the theories described allow for a singlepronoun to be representd by different expressions in the logical form (corrseponding to differentindices).Thissuggestsanambiguityintheword‘he’like,e.g.,theword‘bank’.We,bycontrast,arguethat ‘he’,hasaunique,unambiguouslinguisticmeaning,andthatall itsoccurencesarerepresnetedbyatokenofthesameexpression-typeinlogicalform.Thirdly,andmostimportantly,wecharcterizeawiderspecturmofhitherounappreciatedlinguisticmechanismsthatgovernpronounresolution.5That the meta-semantics of context-sensitive expressions generally should be modeled on thereceivedviewofthemeta-semanticsofdemonstrativepronounsisofteneitherimplicitlyassumed,or

4

analysesoftheseterms,andthephilosophicalargumentsexploitingtheseanalyses,dependonthismodelofcontextdependence.Ifweareright,however,thestandardview ismisguided even for pronouns. This suggestswe should bewary ofmeta-semantical theories about philosophically interesting expressions built on thismodel.Infact,ourviewrecommendsaprojectofitsextensiontoothercases;ifso,philosopherswillhavetorethinktheroleofcontext-sensitivityintheirarguments.6

Inwhat follows,wedevelop an approach to the semantics andmeta-semantics ofpronounswecalltheAttention-CoherenceApproach;itcontainstwocomponents—arankingof candidate interpretations according to relativeprominence (attention),and implicit mechanisms affecting this ranking (coherence). The role of bothcomponentswillbemadepreciseinwhatfollows. Webegininthenextsectiontodeveloptheattentioncomponent;thetheorywillnotreceiveitsfullshapeuntilafterwepresentthecoherencecomponentinlatersections.

2 Attention

Onaroughfirstpass,ourtheoryisthatpronouns,asamatteroflinguisticmeaning,(character, ifyoulike),alwayspickoutwhateveris ‘inthecenterofattention’ inadiscourse.Tocapturethis,weborrowresources fromCenteringtheory,7accordingtowhich,candidateresolutionsofapronouninadiscoursearerankedaccordingtotheir relative prominence: those higher in the ranking are preferred asinterpretationsof pronounsover those lower in the ranking. Call this ranking theattentional state of an ongoing discourse.8 As a discourse progresses, utterancesbring new candidates into focus and adjust the prominence of old ones, therebychanging the attentional state. We shall argue that pronouns, as a matter ofmeaning,alwayspickoutthetop-rankedcandidateinthecurrentattentionalstate,andfurther,thattheattentionalstateitselfismaintainedthroughlinguisticrules.

Tomotivatethedynamicsofanattentionalstate,webeginwith(1):

1 Amanwalkedin.Hesatdown.

explicitlyendorsed.See.e.g.Dowell(2011),Cohen(1999),King(2014a,b),KingandStanley(2005),Neale(2004),StanleyandSzabóGendler(2000).6 A linguistically constrained account of context sensitivity that extends to context-sensitiveexpressionsgenerallyisdevelopedanddefendedin[Redacted].7SeeBittner(2014),Grosz,Joshi,andWeinstein(1995),andSidner(1983).8Centeringtheoryistraditionallyunderstoodaspragmatic;itspreferenceforhigherrankedentitiesisoneofmanynon-linguisticcuesthatcanpotentiallygiveevidenceabouttheintendedreferentofademonstrativepronoun. Ourview,bycontrast, issemantic:therankingismaintainedandupdatedbylinguisticrules.Thecharacterofapronoun,giventhisranking,determinesitssemanticvalue.

5

Onereadingisthat‘He’isinterpretedasco-varyingwith‘Aman’.Onthisconstrual,(1)isalltruejustincasesomemanwalkedinandsatdown.Onasecondreading,the speaker, perhaps uttering ‘He’ pointing at a small child, Bill, is interpreted asreferringtoBill.ThediscourseistruejustincaseamancameinandBillsatdown.

Speakerscanuse(1)indifferentwaysandhearerscanpointtoreasonswhycertaininterpretationsare recovered, andothersnot.Pronoun resolution is guidedbyanimplicitorganizationthatknitstogetherinformationindiscourse.Onitsanaphoricreading, the discourse begins with a description involving ‘A man’ and proceedsdirectly todevelopanarrative:accordingly, ‘He’ is interpretedasdependenton ‘Aman’. On itsdeicticreading,however, thespeakermarksa transition fromtalkofpast events to the present situation with a demonstrative gesture: on thisorganization, ‘He’isinterpretedasdependentontheindividualdemonstrated,andnoton‘Aman’.Weshallarguethatthedifferencebetweentheseinterpretationsof(1)isadifferenceinthedynamicsofattention,guidedbylinguisticmechanisms.Itisbecausethedescription‘Aman’makesitspotentialwitnessesprominent,andthediscourseproceedstodevelopanarrativeaboutwhomeverisawitnessfor‘Aman’,thuskeepingitinthecenterofattention,that‘He’isunderstoodasco-varyingwithpotentialwitnessesinthedomainoftheexistential‘Aman’.Similarly,itisbecausethepointinggestureshiftstheattentionfrompotentialwitnessesfor‘Aman’tothechildpointedat,Billy,that‘He’onthedeicticreadingisunderstoodtorefertoBilly.Thesechangesinattention,governingtheprominenceofpotentialresolutionsforapronoun,weshallargue,areguidedbylinguisticmechanisms.

To make our proposal clear and precise, we shall endorse a particular way offormalizing our key ideas. But our philosophical point—that pronominalinterpretationisfixedthroughlinguisticrules—isindependentofourformalization,and could be implemented in other frameworks. Accordingly, our preferredformalizationisnotessentialtoourargument.9

Sinceononeinterpretationof(1),‘He’isunderstoodas‘bound’bytheindefinite‘Aman’, we adopt a model for bound variables that allows us to capture relevantinterpretivedependenciesacrossdiscourses.OnesuchmodelcleavescloselytothefamiliarTarskianmachineryforvariablebinding,accordingtowhichthetruthofaformulaisspecifiedrelativetoanassignmentofvaluestovariables.Avariable,xi,isinterpretedby retrieving the ith elementof a currentassignment.Boundvariableshavedependent interpretationsbecausequantifiers vary the assignments in forcewithintheirscope.Oneshortcomingisthatinterpretationso-construedislimitedtothesyntacticscopeofaquantifier.Butquantifierscanintroducedependenciesthat9Foralternativeframeworks,seee.g.,Elbourne(2005).SeealsoFiengoandMay(1994,2006).

6

persistacrosssubsequentsentences,asintheanaphoricreadingof(1).Weneedaframeworkforextendingthefamiliarnotionsofavariableandbindingtothekindofdependentinterpretationsin(1).Afixisprovidedbydynamicsemantics.10

In dynamic semantics, just as in a Tarskian framework, variable binding createsdependent interpretations by varying assignment functions. However, unlike theTarskianapproach,dynamicsemanticsassignstruthconditionstoformulasrelativetoapairofassignmentfunctions,nottoasingleone.Thefirstassignmentaccountsfor interpretive dependencies potentially already available in the discourse forinterpretingaformula;thesecond,however,accountsforinterpretivedependenciestheformulaitselfmakesavailableforinterpretingsubsequentdiscourse.Aformulaisinterpretedasaninstructiontoupdatetheavailableinterpretivedependencies:itaugmentsthepotentialinterpretivedependenciesfrompriordiscourse,encodedinthe input assignment, with the additional interpretive dependencies a formulaintroduces for interpreting subsequent discourse, so as to create the outputassignment. Just as in Tarskian semantics, in dynamic semantics, we interpret aboundvariablexibyretrievingtheithelementoftheinputassignment,butnowwecan define the existential quantifier in a way that allows for alternative outputassignments,sothattheinterpretationofsubsequentvariablescancontinuetotakeabound interpretation throughout thesubsequentdiscourse.Overall, a formula istrueonagiveninputassignmentifanoutputassignmentmakesittrue.11

Duetotheexistentialquantifierin(1),theformulamodelingthefirstsentencein(1)is true on a given input assignment just in case there is an output assignmentaccording towhich some unspecifiedman—awitness for the existential—walkedin. 12 The second sentence is modeled with a formula that takes this outputassignment as a new input; ‘he’ can then be interpreted as a bound variable thattakesasitsvaluewhateverthewitnessfor‘aman’isinthisnewinputassignment,resulting in theanaphoricreadingof (1).Thediscourse is trueonthe initial inputassignmentjustincaseamanwalkedinandsatdown.

10SeeDekker(2011)foranaccessibleoverviewofdynamicsemantics.11In this framework, we interpret formulas via updates that relate an input assignment g to anoutputoneh.ThesimplestupdatecontributesinformationfixedbyaconditionC,writtenas‘[C]’andinterpretedasapartialidentityrelationbetweenassignments.Ifgisaninputassignment,andhanoutputone, [C](g,h) iffg=h andCholdson the interpretationofvariablesgivenbyg. Instandardfashion,asequenceofupdatesHandKisrepresentedbyasingleoneH;KthatperformsHfollowedbyK.Anupdate is true for an input assignment iff it is related to someoutput oneby theupdaterelation.Inadditiontothesimplestupdates,weneedupdatesthataffecttheprominencerankingofthe candidate referents.We introduce thesepresently. Asnotedearlier,weexploit thisparticularformal system just for the sake of concreteness and clarity; our philosophical ideas can beimplementedinotherframeworks.FordetailedexpositionoftheformalsystemseetheAppendix.12Any output assignment thatmakes it true that there’s amanwhowalked inwill do. This is thesenseinwhichthewitnessfortheexistentialisan“unspecified”man—anyonewilldo.

7

Buthowdoestheinterpretationof‘he’in(1)getboundbytheexistentialquantifier‘A man’? Both in Tarskian and dynamic semantics alike, variables behave liketemporarynames.Aquantifierlike∃xire-defineswhatxinameswithinitsscope;wecan then interpret occurrences of xias local names for potentialwitnesses of thequantifier.So,if‘Aman’istranslatedwith‘∃xi’,then‘he’getstranslatedwithan‘xi’,thuslookingfortheithelementoftheinputassignment;thiselement,giventhepriordiscourse,willberequiredtobeamanwhowalkedin.

By contrast, sincewewant tomodel prominence in a discourse,wewill organizeinterpretivedependenciesnotwithnames,butbyprominence.Toachievethis,wetreat assignments asstacks in the sense of theoretical computer science. Eachassignmentspecifiesasequenceofpossibleindividualsorderedbyprominence.Themostprominent individual is in initialposition—thetopof thestack—andones insubsequent positionsare deeperin the stack, thereby, receding in prominence.Quantifiersintroducenewpossibilitiesfordependentinterpretationsinsubsequentdiscourse bypushingvalues onto the stack: inserting themat a specifiedposition,forexample,at thetop, thereby,decreasingtheprominenceofothercandidates inthediscoursebyoneposition. In thisway, theyvaryoutputassignment functions.Wetreatavariablenotasatemporaryname,butasamarkerofprominence;wecandefineavariablethatpicksoutthetopofthecurrentstack,andthus,co-variesinitsinterpretationwithwhicheverquantifiermost recentlypushedanewvalue there.Weuse ‘@’asavariable, interpretedrelativetoanassignmentg,asspecifyingthetop-rankedelementofg.(Themnemonicis‘@’isatthecenterofattention.)

Themeaningof ‘@’ isafirstapproximationtothemeaningofanEnglishpronoun.To illustrate, let ‘[man(@)]’ be a condition that requires that whatever is at thecenterofattentionbeaman.Andlet‘⟨�⟩’indicateadynamicexistentialquantifier,whichchangesthe inputassignmentby introducinganewunspecifiedtop–rankedindividualatthecenterofattention,withallothercandidatesdemotedonepositionintheordering.13ThisupdateisafirstapproximationtothemeaningoftheEnglishindefinitearticle.Itaffectstheprominenceranking,changingwhatisatthecenterofattention.(1),onits‘anaphoric’interpretation,isrepresentedas(2):

2 ⟨α⟩;[man(@)];[walk.in(@)];[sit.down(@)]

(2)beginswithanexistentialquantifier,correspondingtotheindefinitearticle,thatintroduces itswitness as thenew top-ranked resolution for subsequentvariables;

13This is a non-deterministic update in the jargon of dynamic semantics. ⟨ α ⟩ relates an inputassignmentfunctiong toanoutputoneh iffhpotentiallydifferswithg inthetopposition,andforeverysubsequentpositioni,hi=gi-1.

8

thewitnessisconstrainedtobeaman,tohavewalkedin,and—oncepickedupbythe anaphoric pronoun—to have sat down. Thus, the formula is true just in casesomemanwalkedinandsatdown.14

Bycontrast,consideradeicticreadingof(1),utteredpointingatachild,asin(3):

3 Amanwalkedin.He[pointingtoachild,Bill]satdown.

WeknowitsindefiniteNPputsitswitnessatthetopofthestack,whilethepronoun,asamatterofmeaning,resolvestothetop-rankedcandidate.Yet,‘He’in(3)doesn’tresolvetothewitnessfor‘Aman’.Thisisbecauseofafurthershiftinattention—onetriggeredbyapointinggesture.Thisshiftupdatestheattentionalstatesothat theentityindicatedbythepointinggesturebecomestop-ranked.

Weintroduceafamilyofupdates,written‘⟨πc⟩’,where‘π’correspondstotheactofpointingand ‘c’ tosome individualbeingpointedat.15Thisupdatestorescas thetop-rankedentity,and(asalways)pushesallothersdownapositionintheordering.Ittherebyrepresentstheeffectofthepointingthataccompaniestheuseof‘he’.(3)isrepresentedformallyas(4):

4 ⟨α⟩;[man(@)];[walk.in(@)];⟨πb⟩;[sit.down(@)]

In (4), we formalize ‘He sat down’ with ‘[sit.down(@)]’, just as in (2). But in itscontext in (4), the condition is true just in case Bill sat down. The interveningexpression ‘⟨ πb ⟩’, corresponding to the act of pointing at Bill, updates theattentionalstate.Asaresult,whomeverthespeakerhaspointedat isatthetopofthe assignment when ‘[sit.down(@)]’ is interpreted. In other words, a pointinggestureupdatestheprominenceranking,makingBillthecenterofattention.

Asin(4),theshiftofattentionprovokedbyapointinggestureisreflectedinlogicalform.However,traditionally,theeffectofapointinggestureaccompanyingtheuseofademonstrativepronounhasbeentreatedastheresultofapragmaticprocessinwhichthehearercomestorecognizethespeaker’sreferentialintentionbyexploitingtheepistemiccueprovidedbythespeaker’spointinggesture.Theonlywaytomakesense of the speaker having chosen to accompany her use of the pronounwith apointinggestureisthatsheintendstorefertosomeobjectintheindicateddirection.Crucially,thepointinggesturedoesnotrenderthedemonstratedobjectprominent14Theformulaistrueforaninputassignmentgiffitistrueforsomeoutputassignment.Since(2)hasnofreeoccurrencesofavariable,itstruthdoesnotdependonaninitialinputassignmentatall.15Moregenerally,‘⟨α⟩’correspondstothecontributionofindefinites,and‘⟨πc⟩’todefinites,where‘c’isadenotingexpression.Wedonotofferhereanaccountofdefinitesotherthanpronouns.

9

as amatter of a convention. It serves as an epistemic cue that helpsmanifest thespeaker’sintentionbecauseitmakessensetopointatanintendedreferent.Werejectthisview.Thechangeofattentioncontributedbythepointinggesturein(3) is and should be reflected in logical form, we maintain, because the sorts ofgestures that secure pronoun resolution in these cases do so as a matter ofgrammar. Our claim is that rules of language dictate that a demonstrative act ofpointing introducesanewcandidatereferentat the topof thestack.16Wesubmitseveralreasonsfortreatingdeicticgesturesasgovernedbygrammar.

First, as argued by Kendon (2004), non-verbal means of indicating an entity aregoverned by rules sensitive to form, meaning, and the relationship with ongoingspeech. English speakers seem to count deixis as well-formed only when thepointingactionissynchronizedappropriatelywiththeprosodyoftheaccompanyingutterance.Theyoftenrepairutteranceswhentheirperformancefailstoalignspeechandgestureintime,asonewouldexpectiftherequirementforsynchronicityweredictatedbyanunderlyingconvention.17

In addition, Kendon observes that although English speakers can use a range ofhand shapes when they indicate an object, their particular choice affects thesemantic contribution they make. A gesture with the index finger and thumbextended,andtheotherfingerscurledclosed,usesthedirectionoftheindexfingertosingleoutanobjectasanindividualdistinctfromitsalternatives.Bycontrast,agesturewith a flat hand open toward the audience, its four fingers extended in atightline,usesthedirectionofthefingerstoexhibitanobjectasarepresentativeofabroader class.Moreover, therearemanypossiblegestures thatarenot typicallyusedasdeicticgesturesbyEnglishspeakers thoughtheycouldhavebeen.Englishspeakersuse the thumb in the“thumbsup”handshape,with the thumbextendedfroma tight fist, todemonstratemovement in thedirection that the thumbpoints,butnormallynottodemonstrateanobjectlocatedinthatdirection,thoughnothingin thegesture itselfprecludes it fromplaying this function.Thearbitrary fact thatsomespatiallydirectedactionsaretakentoindicateobjects—whileanalogousonesare not—reveals convention at work in English speakers’ demonstrations.Numerous languages, unlike English, allow speakers to indicate objects by deicticgestures of their lips, and not an extended index finger. Cuna, a language from

16Foragrammarintegratinggestureandspeech,seeAlahverdzhievaandLascarides(2011).17This is similar to how prosodic focus is grammatically constrained to appear at a particularpositionwithaparticularcontourtocontributeasemanticcontribution.

10

Panama, is one of many examples.18 Such conventionality and cross-linguisticvariabilityisahallmarkoflinguisticmeaning.19

Further, in typical cases, actssignalingshifts inattentionare indispensable for theappropriate interpretation. In (3), explicit signaling of a shift of attention isnecessarytoestablishadeicticreading,eventoaparticularlysalient individual inthesituationofutterance;otherwise,grammarseemstocommitthespeakertoananaphoricreadingofthepronoun.EvenwithBill jumpingupanddowninfrontofthe interlocutors, attracting their psychological attention, unless an overt signalestablishesBillasthereferentof‘he’,theaudiencewillpreferthedefaultanaphoricreading.Indeed,somemeansofexplicitsignalingofashiftofattentionisrequiredevenwhena linguistic antecedent isunavailable, suchaswhenapronoun isuseddeictically in an utterance that initiates a conversation. If deictic gestures wereepistemiccuesthatsuggestreal-worldsalience,thiswouldbequitemysterious.20

Moreover, the linguistic contribution of a pointing gesture clearly affects theinference patterns (1) licenses. The inference from (1), on its anaphoricinterpretation,where ‘He’ isunderstoodasco-varyingwith ‘Aman,’ to ‘Somemansatdown’isvalid;but,onitsdeicticreading,itisnot.Giventhat,aswehaveargued,thecontributionofapointinggestureisunderwrittenbylinguisticconventions,andthat this contribution affectswhich inference patterns (1) can license,we submitthatthecontributionofthepointinggestureshouldbereflectedinthelogicalformof(1).Indeed,(2)entailsthatsomemancameinandsatdown,asamatteroflogicalform;but(4),ontheotherhand,doesnot.21

Finally, we deny that pragmatic mechanisms determine the demonstratum of ademonstrativegesture.Werepresentthepointingin(3)asapointingatBill,notaspointingsimpliciter,withanintendeddemonstratumtobedeterminedbycontext. Inother words, we deny pointing is semantically interpreted as having a uniformcharacter that, given a context, determines a referent. Pointing at Bill is notinterpretedaspointingatx,wherex is contextuallydetermined tobeBill.Rather,

18SeeWilkins(2003).19WeunderstandconventionalaLewis(1969),thoughweneednotsubscribetoallofhisconditionstoregisterourpoint.20Grammar specifies a diverse set of resources for raising entities to prominence (e.g., deicticgesturesandindefiniteNPs).Ourformalismisexpressiveenoughtomakesuchresourcesavailableeven forutterances thatarenot followingupanongoingdiscourse.Ourprediction is thereforenotthatdeixiswithoutdemonstrationisimpossibleatthebeginningofadiscourse,butthatitsucceedsonly for utterances that recognizably accomplish acts that independently require construing thereferentasthecenterofattentioninthecurrentstateoftheongoingdiscourse.Weprovideamoredetailedaccountofdeixiswithoutovertaccompanyingdemonstrativegesturesin[Redacted].21AninferencefromaformulaKtoHisvalidiffforanyassignmentsg,andh,ifKrelatesgtoh,thenthereisanassignmentisuchthatHrelateshtoi.FormaldefinitionsareintheAppendix.

11

pointinggesturesareambiguousbetweenmultiplepossible forms,e.g.,pointingatBill,pointingathisshirt,pointingathisbutton.Actsofpointingare,onouraccount,ambiguous.Theinterpretiveworkin(3)comesinsettlingwhethertheformofthegestureispointingatBillorpointingatsomethingelse.Butdisambiguation ispre-semantics,inKaplan’ssense:22itinvolvestheinterpretiveworkneededtosettlethelinguisticformofanutterance.Thisshouldnotbeconfusedwithwhatisinvolvedinthe semantic interpretation of a given linguistic form. Though, according to us,epistemiccuesplaynoroleinsemanticinterpretation,theycanallowobserverstorecognizewhichformofapointinggestureisinplay.Thisistherolethey’dplayinhelpingtodisambiguateanyambiguity,e.g., indisambiguatingauseof ‘bank’.And,aswithotherambiguities, conventionsgoverningdemonstrativeactions constrainpossible disambiguations. As we have seen, a flat hand shape with the fingerstowards theaudienceallows foracertainrangeof interpretations,butnotothers;similarly,foranextendedindexfinger.

Tosumup:wehavearguedtherearereasonstotreattheeffectofanattentionshifttriggeredby an indefiniteNPor an act of demonstration as rule governed and asrepresented in logical form. The recognition of these systematic effects on theprominence ranking of candidate referents in a context allows us to assign onemeaning toeachpronoun that fullydetermines its resolution inagivencontext—roughly,pronounspickoutthemostprominentcandidate,theoneatthecenterofattention.Refinements,however,areobviouslyrequired;weturntothemdirectly.

3 Refinements

That ‘he’, as a matter of meaning, takes on the most prominent candidateinterpretationcannotbe thewhole story. ‘He’ in (5) cannotbe interpretedas co-varyingwith‘Agirl’.Itmustberesolvedtoamale.

5 Agirlcamein.Hesatdown.

‘She’ in (6) cannot be interpreted as co-varyingwith thewitness for ‘Some girls’.‘she’musthaveasingularinterpretation.

6 Somegirlscamein.Shesatdown.

22SeeKaplan(1989b.).

12

‘we’ and ‘they’ in (7) cannot be interpreted as co-varying since ‘they’ requires athird-personinterpretationdisjointfromspeakerandaddressee,and‘we’requiresafirst-personpronouninterpretationthatincludesthespeaker.23

7 Wecamein.Theysatdown.

Pronouns come with person, gender and number requirements that must besatisfied in pronoun resolution. These requirements are linguistic: it matterswhetheraspeakerutters‘he’,‘she’or‘they’becausetheydifferinmeaning.24

Further,in(8),‘him’cannotsemanticallyco-varywithPaul,eventhoughPaulwouldnormallybeaneligiblereferentfor‘him’.

8 Paulmethim.

The explanation is that reference is constrained by syntactic principles—in thiscase, Principle B of Chomsky’s Binding Theory, which requires non-reflexivepronouns to be free in their governing category. Roughly, this means the non-reflexivepronounmustnotbeboundbyanexpression in thesameclause, that is,that itcannotshareaclausewith itsownantecedent.25So, in(8),where ‘him’and‘Paul’areclausemates,‘him’cannotrefertoPaulonpainofviolatingPrincipleB.

Withthisinmind,wesayapronoundenotesthevaluestoredatthehighest-rankedposition on the stack that respects grammatical features and applicable syntacticprinciples.Tointerpretapronoun,weconsidercandidatesinorderofprominencewithinthecurrentattentionalstateofthediscourse,untilwefindonethatsatisfies

23Sometimesaspeakercanuse‘he’torefertohimselfor‘they’topickoutagroupincludinghimself.Witness a Perry-type scenario,where the speaker seeing, but not recognizing, himself in amirrorsays,“Heisthemessyshopper”(Perry,1979).Suchcasesarearguablymarkedinthesensethattheexploitationofanextantrulemakesforthesurprisingeffect.Otherssuggestthatthedisjointednessconditionispragmatically implicated,notgrammaticallydictated(seeFiengoandMay(2006)). Bycontrast, we treat number requirement as a part of the character of a pronoun. We agree thatsometimesthespeakercanmanageto‘speaker-refer’,inthesenseofKripke(1977),tohimselfwith‘he’,ortoagroupincludinghimselfwith‘they’.Also,sometimesanaudiencecanunderstandthatthespeakermeanttopickouthimselfwith‘he’,orsomegroupincludinghimselfwith‘they’,inthesamesenseaswecansometimesunderstandaspeakerwhomistakenlyattemptedtopickoutamalewith‘her’.Thisissoeventhoughthecharacterofapronounencodesnumber,genderandpersonfeatures.However,wecouldmodifyouraccount,ifoneinsistedonexcludingthesefeaturesfromthemeaningofapronoun.WhicheverpropertyPspecifiesthecharacterof‘he’,wemaintainitsreferentisthetop-rankedentitysatisfyingthatproperty.24Again, this is true regardlessofwhethera speaker canmanage to ‘speaker-refer’ in the senseofKripke,toawomanwiththepronoun‘he’(oradefinite‘thatman’).25More precisely, the antecedent of a non-reflexive anaphoric pronoun must not be local, or c-commandthepronoun.Forprecisedefinitions,seeChomsky(1981).

13

theoperativelinguisticconstraints.26(9)integratestheconstraintsofsemanticsandsyntaxforthepronoun‘he’.

9 Wheninterpretedrelativetoanassignmentgmodelingtheavailabledependent interpretationsonanoccasionofuse, ‘he’ denotesg’s highest-rankedentitythat issingular,masculine,anddisjoint fromthespeakerandaddresseeofthe utterance, and that yields an interpretation where the occurrence of thepronounisfreeinitsgoverningcategory.27

To incorporate (9) formally, let ‘he’ be a predicate representing the constraintsassociatedwith the English third person singularmale pronoun and ‘@he’ be anindividual expression denoting the highest-ranked entity from the currentassignmentthatsatisfiesthepropertydenotedby‘he’.(2)—(3)become(10)—(11).

10 ⟨α⟩;[man(@)];[walk.in(@)];[sit.down(@he)]11 ⟨α⟩;[man(@)];[walk.in(@)];⟨πb⟩;[sit.down(@he)]

However,evenwiththesequalifications,(12)and(13)pointtodeficiencies.

12 AmanmetSam.Hegreetedhim.13 JohnwasdisappointedwithTim.

i. Hefiredhim.ii. Hedidsloppywork.

In(12),Englishspeakerswouldnormallyresolve‘He’tothemanintroducedinthefirst sentence and ‘him’ to Sam.28Although the syntactic constraint on bindingexplainswhyEnglish speakersdonot treat thepronouns in (12)as co-referential(violationofPrincipleB), itwon’texplainwhythe firstpronoun isresolvedto themanintroducedinthefirstsentence,andnottoSam.Theexplanationcan’tbethatonly‘Aman’andnot‘Sam’updatesthetopofthestacktoallowforanewdependentinterpretation;anamecanalsoaffectattentionfocus.29So,whyistheformer,andnot the latter, the highest-ranked candidate for ‘He’? Similarly, given what we’ve26Theproposaltoresolvethepronountothetop-rankedelementonthestackinvolvessearchratherthanasimultaneousimpositionofpotentiallycompetingconstraints;thepronounpicksoutthevaluestoredatthehighestrankedpositiononthestackthatsatisfiestheconstraints.27Roberts (2002) proposes similar rules for demonstrative pronouns, butwith crucial differences.Instead of a linguistically governed notion of attentional prominence ranking, she uses a liberalnotion of contextual, “real-world” salience.Moreover, she treats semantic constraints on pronounresolutionaspresuppositionsofapronoun,butweunderstandthemaspartofcharacter.28Assume (13) is uttered without accompanying demonstrative gestures. That the NP in subjectpositionispreferredasanantecedentforthesubsequentanaphorovertheoneinobjectpositioniswell-documented by a number of corpus and psycholinguistic studies. See Kameyama (1999),Walker,Masayo,andCote(1994),andBittner(2014).29Witness:‘Billcamein.Hesatdown’.

14

saidsofar,(13)ispuzzling.In(13i),‘He’isnaturallyresolvedtoJohn,butin(13ii),toTim.Clearly,bothreferentscan’tbetop-rankedsimultaneously.WhatmakesJohnprominentin(13i)andTimin(13ii)?Wetakeupthesechallengesinturn.

For (12), an explanation is available. In English, attention ranking mirrors thegrammatical role inwhich noun phrases are realized; the noun phrase in subjectpositiontakesprecedenceovertheoneinobjectposition.30Thisiswhy,with(12),‘He’ is resolved to the candidate referent introduced by the previous subject, ‘Aman’, rather than one introduced by the previous object ‘Sam’. Crucially, thispreference for referents introduced by noun phrases in subject position is agrammatical featureofEnglish,andnotuniversally sharedacross languages.31Butwhat about ‘him’? The dependent interpretation linked to ‘A man’ remains mostprominent when we turn to interpreting ‘him’; the ranking remains constantthroughoutthesecondsentenceof(12).Butbecause‘him’cannotbesoresolved,onpainofviolatingPrincipleB,itmustresolvetothehighest-rankedcandidatelowerintherankingthatsatisfiestherelevantconstraints,i.e.,Sam.

A streamlined way of representing this aspect of English formally is to linkgrammaticalrolestospecificpositionsonthestackofcandidateinterpretations:thesubjectcorrespondstoposition0,thedirectobjectto1,theindirectobjectto2.Weuseexpressionsofforms‘⟨ αn⟩’and‘⟨ πnc⟩’toencodeupdatesthatpushreferentstopositionninthestack(arbitrarilylimitingourselveston=0,1or2andassumingthevalueofnfollowscompositionallyfromthegrammaticalstatusoftheexpressionbeingrepresented),pushingallotheronepositiondown.Onthisstrategy,thefirstsentencesof(1)and(12)areformalizedas(14)and(15)respectively.

14 ⟨α0⟩;[man(x0)];[walk.in(x0)];15 ⟨α0⟩;[man(x0)];⟨π1s⟩;[met(x0,x1)];[greeted(x0,x1)]

Given these LFs, in formalizing successive sentences across discourse, we shouldencode the pronoun as assigned a grammatical role in its own sentence, andrepresentthisbyanexplicitfurtherupdatetotheattentionalstateofthediscourse.(To represent the effect of Principle B in resolving the pronoun, we introduce‘@hex0’, an expression that denotes the most prominent referent whose value

30This observation is well documented. See e.g. Kameyama (1996), Kehler (2002), and Bittner(2014).31 Languages with a flexible syntax can exploit word order to indicate prominence. Somegrammaticize prominence with morphemes like topic markers that crosscut word order andgrammaticalrole.SeeKameyama(1986),Walker,Masayo,andCote(1994)andBittner(2014).

15

satisfies thepronounother thanx0—meaning the top-rankedmaleother than thesubject.)Weget(16)and(17):32

16 ⟨α0⟩;[man(x0)];[walk.in(x0)];⟨π0@he⟩;[sit.down(x0)]17 ⟨α0⟩;[man(x0)];⟨π1s⟩;[met(x0,x1)];

⟨π0@he⟩;⟨π1@hex0⟩;[greeted(x0,x1)]

This approach achieves themost uniformpossible syntax-semantics interface. Inparticular,ourformalrepresentationconsistsentirelyofupdatesreadabledirectlyoffthelexicalitemsthatcomprisetheutterance,alongwiththeirgrammaticalroles.

What about the second challenge created by (13)? While (13i) is explained inparallelfashionto(12),byappealingtosubjectpreferencefor‘he’takingJohn,andPrincipleBfor ‘him’takingTim,(13ii)seemstoruncounter, for,thoughthereisaprominent,accessibleantecedentinsubjectposition(‘John’)satisfyingthecharacterofthepronoun,thepreferredresolutionof‘he’istothereferentintroducedbytheantecedentinobjectposition(‘Tim’).Thisiswherecoherenceenters.

4 Coherence

Coherence theory starts from an obvious observation: a discourse ismore than asequence of grammatical sentences. Successive contributions must be linkedtogether by a recognizable flow of interpretive relationships. We see therequirementofcoherencein(18)and(19),fromHobbs(1979).

18 JohntookthetrainfromParistoIstanbul.Hehasfamilythere.19 JohntookthetrainfromParistoIstanbul.Helikesspinach.

(18) doesn’t just reveal two facts about John: it suggests the reason he went toIstanbulwastovisithisfamily.AcoherencerelationofExplanationlinksthesecondsentencetothefirst.33Coherencetheoryrecommendswerepresentthisconnectionexplicitlytocapturethecorrectinterpretationof(18).Conversely,therequirementthatadiscoursemustbecoherentisstrikinglyevidentintheinterpretiveeffort(19)elicits. Given apparently unrelated facts about John in (19), we search for aconnection. Is Istanbul known for its spinach? Is the train? Clearly, interlocutors

32Since the pronoun is in subject position, the update—‘⟨ π0@he ⟩’—that corresponds to thepronounpromotesthecandidatereferentassociatedwith‘he’asthetop-rankedreferent.Itdoesnotintroduceanewone;itratherre-stores(i.e.,pushesacopyof)anoldonetothedesignatedpositiononthelistofprominentcandidates.Iftheentityisalreadyatthetop-rankedposition,asin(16)-(17),theupdateassociatedwiththepronounwillnotchangepossibleinterpretationsofsubsequentones.33SeeAsherandLascarides(2003),Hobbs(1979)andKehler(2002).

16

must use the common ground to disambiguate between discourses harboringdifferentcoherencerelations, justastheymustusethecommongroundtoresolveother ambiguities. But just as clearly, a failure to acknowledge any of theseingredientsofinterpretationconstitutesafailuretounderstandthediscourse.34

Kehler advances the view that coherence relations cluster into (at least) threequalitatively different sorts, reflecting alternative strategies for organizingdiscourse,35asillustratedin(20)-(22).

20 Maxspiltabucketofwater.Hetrippedonhisshoelace.21 Maxspiltabucketofwater.Hespiltitallovertherug.22 Maxspiltabucketofwater.Johndroppedajarofcookies.36

(20) illustrates the kind of an explanatory discoursewe considered in (18). ForKehler,Explanationisaninstanceofabroaderclassofcause-effect(orevent-result)relationsthatspeakerscanutilizetoorganizediscourse.(21),meanwhile,givesanextendeddescriptionofunfoldingevents—thus, aNarrative connection,which forKehler epitomizes a broader class of Contiguity relations. (22) exemplifies whatKehlercallsResemblancerelations,organizingadiscoursetodrawcomparisonsandcontrasts.In(22),there’saParallelbetweenMax’sandJohn’srespectiveaccidents.Different coherence relations are alike in signaling relationships amongpropositionsindiscourse.However,asweshallsee,theserelationsalsoshapehowother material in a discourse is interpreted. This is particularly important forcontext-dependent elements. The best way to capture these interpretive effectsformally,weargue,istorepresentcoherencerelationsexplicitlyinlogicalform.

Coherence theorists view identifying coherence relations and resolving semanticambiguitiesasmutuallyconstraining.In(13ii),onlyresolving‘he’toTimallowsforaplausibleexplanationof John’sdisappointment. In(18),onlyresolving ‘there’ toIstanbulallowsforaplausibleexplanationofJohn’strip.37In(20)–(22),meanwhile,34Representingcoherencerelationsexplicitlydoesnotmeananexplanatoryrelationisapartofthetruth-conditional content of (18). Though we will argue that connections signaled by coherencerelations, like Explanation in (18), are not conveyed as a conversational implicature, this is not atoddswith this content being not at-issue, in the sense of Tonhauser, Beaver, Roberts, and Simons(2013); conventional encoding of non-asserted content is not unusual (semantic presuppositions,conventional implicatures, expressive content). For simplicity, our formalism includes only onedimension of semantic content, in the sense of Potts (2005), but we could easily modify ourframeworktocaptureconventionalizednotat-issuecontent.However,thistaskisforanotherpaper.35Kehler’stypologyisusefulinprovidinganintuitivepictureofthekindsofinterpretiveconnectionswe find in discourse. Other typologies might be better suited for other purposes. Cf. Mann andThompson(1988),andKnott(1996).36SeeKehler(2002).(20)and(21)areadaptedbyKehler(2002)fromAsherandLascarides(2003).37Note that changing (19) to ‘John tooka train fromParis to Istanbul.He likes spinach there,’ stillonly allows resolving ‘there’ to Istanbul for a plausible explanation of John’s trip. And since

17

we infera temporal relationbetween thespilldescribed initiallyand the tripping,spilling or dropping described next thatmatches the inferred coherence relation.Referenceandcoherencerelationsfittogetherinsuchcases.

A number of studies have experimentally confirmed the interdependence ofresolvingpronounsandestablishingcoherence.38 An illustrativeexample is (23),fromSmyth(1994):39

23 PhiltickledStanley,andLizpokedhim.

Speakerstendtointerpret(23)ineitheroftwoways.OnewayassumesLiz’sactionwasprompted byPhil’s. Liz isperhaps reactingwithdisapproval towhat Phil hasdone.Thiscause-effectinterpretationcomeswiththeunderstandingthat‘him’mustrefertoPhil.ThesecondreadingassumesLiz’sactionwassimilartoPhil’sincertainrespects.Thisparallelinterpretationcomeswiththeunderstandingthat‘him’mustrefertoStanley.Crucially,thechoiceofacoherencerelation(ResultorParallel)andthe pronoun resolution (to Phil or Stanley) go in tandem.40Pragmatic theories ofreference resolution—even on standard coherence approaches—take this asevidenceofan inferentialrelationshipbetweenaspeaker’s intention inorganizingthediscourseandherreferentialintentions.Accordingtous,thesestudiesconfirma

Explanationisnaturallysignaledinsuchexamples,asin(18)and(19),thepronounisunderstoodtorefertoIstanbul.Thisfurthersupportsourclaimthatthechoiceofrelationgovernstheresolutionofthepronoun.(Ofcourse,somodifiedthediscourseisnolongerasincoherentastheonesequencingrandominformationin(19),becauseitisclearhowthesecondsentenceprovidesanexplanationofthefirst,asisrequiredbyExplanation.Butthatisbesideourpoint.)38See,forinstance,Kaiser,E.(2011),“OntheRelationbetweenCoherenceRelationsandAnaphoricDemonstrativesinGerman,”InI.Reich,E.Horch,&D.Pauly(Eds.),ProceedingsofSinn&Bedeutung15,pp.337–351,Saarbrücken:SaarlandUniversityPress,andreferencestherein.39Weassumenodemonstrativegesturesareaccompanying(23).40When(23) isembeddedwithina largerdiscourse,wemightbeabletogetotherresolutions,butthisisbecauseembeddingscanlicensedifferentcoherencerelations.Similarly,prosodicmarkingcanaffect pronoun resolution. This is expected, since prosodicmarking can affect prominence just asdemonstrativegesturescan.Likewise,fillinginacontextmightchangetheinterpretivedependencesthataffectpronounresolutionbyvirtueofchangingthecoherencerelationsorganizingadiscourse,or by giving rise to other linguistic constraints on interpretation.Onemight imagine a scenario inwhichitisapartofthecommongroundthatLizalwayscopiesPhil,andtriestomimicwhateverhedoes. One could then say, “Phil tickled Stanley, and as a result, she poked him”, where ‘him’ isinterpretedasPhil.Twothingstonote:first,weshouldnotassumecoherencerelationsaremutuallyexclusive.Inmanycases,morethanoneisneededtocapturethestructureofadiscourse(seeAsherand Lascarides, 2003). Thismight give rise to amore complex pattern of promoting referents toprominence.And,second,thepresenceofanexplicitdescriptivesignalsuchas“asaresult”neednotautomaticallymean a particular relation—Result—is organizing the discourse (seeWebber, Stone,Joshi,andKnott(2003)).Coherencerelationsareamatteroflinguisticdependencesandnotjustanywayofdescribinghoweventsarerelatedgivesrisetoaparticularcoherencerelation.Similarly,notjust any way of describing cause-effect relations counts as Explanation. “as a result” might betargeting a background presupposition about Liz’s known pattern of behavior. There is reason tothinkpresuppositionsplacefurtherconstraintsonpossibleantecedents(seeHobbs,1990).

18

tighter connection between coherence relations and pronoun resolution thanCoherencetheoristshavebeeninclinedtoposit.

The contrast between these interpretations leads us to conclude it’s amistake totreat (23)asharboringseparateambiguitiesaudiencesmust resolve in turn—oneabout discourse coherence, another about pronoun resolution. The examplessuggestthat,onceacoherencerelationisestablished,acertainpronounresolutionis automatically set up. We hypothesize that pronoun resolution is settled bywhichevercoherencerelationsorganizeadiscourse.

Wearguecoherencerelationsareanothersortofmechanismthateffectschangesintheattentionalstateofadiscourse.Moreprecisely,weformallyrepresentcoherencerelations in the logical formof a discourse, but propose further that this requiresrepresentingnotonlyinferentialconnections,butalsoshiftsinattentionassociatedwithcoherencerelations.Putsimply,coherenceestablishesaninterpretiverelation,and moreover, (as we shall argue) as a matter of linguistic contribution, theypromote certain entities to prominence. We depart from standard Coherencetheorists inasmuch as we maintain that coherence relations come withgrammaticallyencodedshiftsinattention,muchlikeNPsorpointinggesturesdo.Itisonlyaftertheseshiftsareacknowledgedasapartofalinguisticcontributionthatwe can represent the intuitively correct interpretations of pronouns,while givingpronounstheuniformandunambiguousmeaningswehavealreadyproposed. Wefirstexplainhowthisaccountcapturesdatafrom(13i)and(13ii),andthen,presentseveral reasons in defense of the view that the attention-shifts associated withcoherencerelationsaregrammaticallyencoded,notpragmaticallyimplied.

Accordingtous,thedifferencebetweenresolutionsin(13i)and(13ii)isfixedbythecoherencerelationthatfiguresintheirrespectiverepresentations;thesecomewithdifferent attention shifting updates, which affects, semantically, the resolution ofsubsequentpronouns. In (13i), the coherence relation isNarration: the contentofthe second sentence followsup on the content of the first, providing an extendeddescriptionofunfoldingevents.In(13ii),thecoherencerelationisExplanation:thecontentofthesecondsentenceistakentoexplainthecontentofthefirst.Accordingto Coherence theory, one or the other of these relations surfaces in formalrepresentationsof (13i) and (13ii). Ourproposal is that,asamatterof language,these distinct coherence relations affect the attentional state of the discourse,promotingoneor theotherof the candidate referents to the top-rankedposition.When a formal representation featuresNarration, the attention-shifting operationraises John (the subject) to prominence; and when it features Explanation, the

19

attention-shifting operation raises Tim (the direct object) to prominence. 41Accordingly,werepresent(13i)as(24),and(13ii)as(25):

24 ⟨π0j⟩;⟨π1t⟩;[was.disappointed.with(x0,x1)];[Narration(x0)];⟨π0x0⟩;⟨π0@he⟩;⟨π1@hex0⟩;[fired(x0,x1)]

25 ⟨π0j⟩;⟨π1t⟩;[was.disappointed.with(x0,x1)];[Explanation(x0,x1)];⟨π0x1⟩;⟨π0@he⟩;⟨α1⟩;[work(x1)];[sloppy(x1)];[did(x0,x1)]

(24) introduces John into subject position, and Tim into direct object position,requiring that John was disappointed by Tim. The second sentence continues anarrativeaboutJohn;werepresentthecontributionofNarrationas:‘[Narration(x0)];⟨π0x0⟩’,anupdatethatrequiresthatthediscoursecontinuesthenarrativeaboutx0, and correspondingly, renders x0 prominent in the attentional state of thediscourse.42‘He’picksoutthemostprominentcandidateappropriateforit,whichisJohn,while‘him’picksoutthecurrentlymostprominentcandidateotherthanJohn,whichisTim.Further,itisrequiredthesubject(John)firedtheobject(Tim).

Thefirstlinein(25)isthesameasin(24).Thecrucialdifferencecomesnext.Thisdiscourse is modeled as harboring Explanation, which we formalize using theupdate ‘[Explanation(x0, x1) ] ; ⟨ π0x1 ⟩’. Thisupdate requires that the twobits ofdiscourse stand in an explanatory relation, and promotes the entity in objectpositiontoprominence.‘he’continuestopickoutthetop-rankedinterpretation,butdue to the update associated with Explanation, this is Tim. The formula thenproceedswithanadditionaldynamicexistentialquantifier, and further conditionsthatensureitswitnessissloppyworkdonebythesubjectTim.

Crucially, differences in pronoun resolution follow from specifications of thecoherence relations that organize the discourse and update the attentional state,41 As we shall see below, not all explanatory coherence relations raise the direct object toprominence. Coherence relations encode linguistic interdependences in discourse, not merecommon-sensedependencesbetweeneventsintheworld.Thus,whichentityisraisedtoprominencebywhich coherence relation is amatter of empirical inquiry, not apriori judgment, or reasoningaboutcommon-senserelationsbetweeneventsintheworld.42This representationofNarration suffices to capture theeffectsonattention that concernus.Forother purposes, we might want to refine the representation of Narration to account for spatio-temporal and causal links we find in narrative discourse. Hobbs (1979, 1990) formalizes theseinferential connections in terms of a relationship between eventualities described in successivesentences. Asher and Lascarides (2003)model it as a relationship between dynamic propositionsexpressedbysuccessivesentences.

20

and not from an open-ended process of pronoun resolution. It might seemsurprising todescribeattention-shiftsassociatedwithcoherenceasgrammaticallyencoded.Bycontrast,youmightholdthatattentionshiftscontributedbycoherencerelationsonlyreflectspeakerintentionsandthehearer’scommon-senseinference.Thisviewis,infact,standardamongCoherencetheorists.Soconstrued,anattentionshiftthatguaranteesthecorrectinterpretationofthesubsequentpronounisnotalinguistic effect of coherence, as we urge, but rather of pragmatic reasoning thatoccurs once a hearer has established that a particular coherence relation isstructuring the discourse. A hearer reasons that in (13i), sinceNarration holdsbetween the two sentences, the speaker must be intending to promote a certainreferent, in this case, John, to the center of attention. On this view, intentionrecognitionaffectsthere-rankingofthelistofprominentreferents.

It certainlymakessense forattention toshift in theways (24)and(25)suggest. Itwouldbeperversetoreversethesepreferences,sothatwhenwecametoconsideranarrative about Johnwho is disappointed inTim, the referent of ‘he’wasTim, orwhenwecametoexplainwhataboutTimmadeJohndisappointedinhim,suddenlythe referent for ‘he’was John. Toorganizediscourse in sucha confusingmannerwouldmakeitmuchhardertocommunicateourideasconcisely.

However, ask not where coherence relations shift attention, but rather in whatcircumstances they do so. This question gives us reason for maintaining thatattention-shiftingoperationsaregrammaticallyencoded.Toseewhy, firstobservethatspeakersandhearerstakearestrictedsetofcuesintoaccountin instancesofpronounresolution.Theyprivilege linguisticones,over thebroaderconstraintsofbackgroundknowledgeandrationalinferencethattheymightpotentiallyconsider.Foraperfectexampleofthisregularity,consider(26),fromKehler(2002):

26 *MargaretThatcheradmiresRonaldReagan,andGeorgeW.Bushabsolutelyworshipsher.

Kehlerreports(26)isgenerallyjudgedtobeinfelicitousbysubjects.Whatexplainsthis?Byvirtueoffollowing‘admires’with‘absolutelyworships’—astrongerterminan obvious scalar relationship—the speaker provides clear evidence (26) isorganized around a contrast betweenMargaret Thatcher’s and GeorgeW. Bush’scomparableattitudes.CoherencetheorypredictsthisparallelshouldmakeReagan(the object of the first clause) the preferred referent for the pronoun in objectpositioninthesecondclause.43And,indeed,reading(26),itseemsasifthespeaker

43Kehler is not committed to this preference being a result of a grammaticized contribution ofcoherence.Heuses(26)fordifferentpurposes,butitperfectlyillustratesourpoint.

21

haserred,inadvertentlyreferringtoReaganas‘her’. Ofcourse,sinceThatcherhasbeen evoked in the previous sentence, in subject position, and is a well-knownobject of Bush’s admiration, you’d expect it would be rather easy to refer toThatcherwith‘her’weretheeffectofcoherenceonprominencemerelyaby-productof the general pragmatic, common-sense reasoning that interpreters use torecognizeaplausibleinterpretation;yet,thisisnotwhatwefind.

This point is analogous to the conventionality of demonstration. Thatcher mayattract our visual attention with what she’s doing, but if the speaker is pointingelsewherewhile saying ‘her’—or continuing anongoingdiscourse about someoneelse—we do not take ‘her’ to refer to Thatcher. Likewise, the Parallel relationencodedin(26)accomplishesakindofinferreddemonstration,indicatingReaganinawaythat’sdifficultforacommon-senseinferencetooverride.44

Furthersupport forour thesiscomes fromthevariationwe findacross languages.Many have explicit operations for shifting attention, such as grammatical topicmarking, or a distinction between topic and non-topic pronouns. Some aremoreconstrainedthanEnglishinthesortsofshiftstheypermittotakeplaceimplicitly.AroughinspectionofSerbian,alanguagethatallows(sometimesrequires)pronounsto be “dropped”, i.e. remain unpronounced in certain grammatical positions,suggestsitisalanguageofthiskind.So,considertwopossibletranslationsof(13i):

27 DžonjebiorazočaranTimom.John-NOMis-PRS-3msbe-PPA-3msdisappointed-ADJ-3msTim-INS.i. Otpustiogaje.

Fired-PPA-3mshimis-PRS-3ms.ii. Ongajeotpustio.

Hehimis-PRS-3msfired-PPA-3ms.

When(13i) is translated intoSerbianwith its third-personmalesingularpronoun“dropped” (as in (27i)), the interpretation is that John fired Tim. Yet when thepronounisovertasin(27ii),theinterpretationisTimfiredJohn.Itseemsthattheovert pronoun signals a change in prominence the covert counterpart does not.English,lackingthisexplicitmeansofsignalingashiftinattention,ismoreflexiblewithimplicitshiftsinattention;thus,wewitnessanambiguityin(13i)-(13ii).

44Focal stress on ‘her’ makes (26) felicitous, with the referent of ‘her’ being Thatcher (as Kehlerreports). But this stress carries interpretive requirements, changing the point of the discourse: itexplainshowBushfollowsThatcher’sopinions,nothowconservativepoliticiansfeelaboutReagan.There is no readingwhere the relation isParallel and the pronoun resolves to Thatcher. But thisshouldbeachievableiftheattention-shiftingeffectofParallelwereapragmatic,defaultone.

22

TheGreenlandic languageKalaallisutprovidesanother clear case. Bittner (2014),contrasts(28i)inEnglishwith(28ii)inKalaallisut.

28 TravelingtoDenmarki. WhenIcametoDenmark,Iboughtmyticketsixmonthsin

advance.ii. Danmarkimuttikikkama,

Danmarki-muttikit-ga-ma,Denmark-sg.DATcome-FCT-1sWhenIcametoDenmark,…

qaammatitarvinillitsiuqqullugubillitsisivunga.qaammat-tarvinilli-tsiuqqut-llu-gubillitsi-si-pu-nga.month-plsix-plv.ahead-ELA-3sticket-get-IND.IV-1sIgotaticket(forsomeotherevent)sixmonthsahead(ofthatevent).

FollowingMoens,andSteedmanandWebber,wetakesentenceswithsubordinateclauses, such as ‘when’ clauses, to bemini-discourses, with coherent interpretiveconnections between the clauses that mirror interpretive connections foundbetween successive sentences in a discourse, or between conjuncts.45For (28i),English speakers find a natural interpretation where buying the ticket early is adescription thatelaborates how the speaker came toDenmark. For its translation(28ii),however,theanalogousinterpretationisunavailabletoKalaallisutspeakers.TheKalaallisutsentencerequiresaninterpretationwherethemainclausedescribeswhat happens after the speaker came toDenmark—which is compatiblewith thenarrative interpretation,butnotelaboration.WewitnessthisdiscrepancybecauseKalaallisutmarkstheresolutionoftemporalanaphora,sothataneventverbforcesthe temporal progression, making the time “right after the event described”prominent fortemporalanaphora. Sincesuchatemporalprogression,dictatedbyan eventive verb, is consistent with the narrative interpretation, we find thisinterpretationavailable.However,thoughthislanguagehasgrammaticalmeansforsignalingshifts inattentionfortemporalanaphora, it,unlikeEnglish,seemsnottoassociatecoherencerelations(viz.,Elaboration)withattentionshiftingupdates. InEnglish,Elaboration contributes an attention shifting update thatmakes the timeprior to coming to Denmark prominent for temporal anaphora. In Kalaallisut,Elaborationlacksthisattentionshiftingupdate,andso,absentothergrammaticallymarkedwaysofrenderingthetimebeforethetriptoDenmarkprominent,theonlyway to resolve temporal anaphora in (28ii) is for the trip toDenmark toprecede

45SeeMoensandSteedman(1988),Webber(1988),Webberetal.(2003).

23

buyingthe ticket.This interpretation isconsistentwithNarrative,notElaboration,andso,there’snoElaborationinterpretationof(28ii).

ThediscrepancybetweenEnglishandKalaallisutwouldbepuzzlingweretheshiftsinducedbyElaborationinEnglisharesultofpragmaticreasoning.Ifitwereso,wewouldexpecttofindthesamerangeofinterpretationsinKalaallisut.However,wedo not. Only if languages interpret analogous expressions differently as signals oftransitions indiscoursecanweaccommodate thedifferenceswe find.So,we inferthateachcoherencerelationinlogicalformcarriesarule-governedcontributiontoattentionindiscourse.Giventhestateofattention,pronounresolutionfollows.

Ourviewisconsistentwithgeneralreasoningbeingcrucialtointerpretation.Whenthegrammardeliversmultiplecandidatereadings,hearersneedtochooseonethatmakesthemostsenseonagivenoccasion.(23)isambiguousbetweenadiscoursecontaining Result and one featuring Parallel. Some general reasoning might beinvokedindisambiguatingbetweenthese,muchasitmightbeinvolvedinfiguringoutwhethera speakermeansa financial institutionora riverbank,withauseof‘bank’,orwhichquantifierscopeisintendedwithauseof‘Everyboykissedagirl’.Tointerpret,ahearermustfirstsettledisambiguations.Thismayinvolveassessingtheplausibilityofinferentiallinksconveyedbycoherencerelations.Itmayinvolveevaluatingwhethera readingengendersaplausible resolutionofpronouns. Suchreasoning constitutes an important principle of disambiguation, but it cannotcontributecontenttologicalform.Itservestoprivilegealogicalformofdiscourse,among available ones grammar delivers. But once a coherence relation isestablished,pronounresolutionisdeterminedbygrammar,notgeneralreasoning.

In sum, strong evidence favors treating attention-shifting effects of coherencerelationsasgovernedbylinguisticrules,notasabyproductofpragmaticreasoning.These attention-shifting updates change the attentional state of a discourse, thus,settingtheparametersofthecontextthatdeterminetheresolutionofapronoun.Apronoun,inturn,asafunctionoftheseparameters,automaticallyselectsareferent,according to linguistic meaning. Thus, not only is the resolution of a pronoundeterminedbyitslinguisticmeaningasafunctionofcontext,butmoreover,relevantfeatures of context that fix this reference are themselves orchestrated by rules oflanguage. So, pronominal resolution is linguistically determined—through andthrough. Linguisticmechanisms of attention-shifting updates and rules governingpronominalusagetogetherdeterminepronominalresolutiononanyoccasionofuse.

24

5 Conclusion

We defended a joint attention-coherence account for pronoun resolution thatassigns one linguistic meaning to each pronoun. Pronouns are variables withdependent interpretations, but are interpreted relative to a prominence ranking.Eachresolutionisrestrictedbyadditionalconstraints,includingperson,genderandnumberfeatures,andindependentlymotivatedsyntacticconstraints.Theytriggerasearch for a matching interpretation. Think of the character of a pronoun asincorporating these constraints; on our account, it determines the resolution of apronounautomaticallyas functionofcontext.Tomake thiswork,contextmustbeappropriately set up. This is achieved by the attentional state of a discourse, theresultofa seriesofattention-shiftingupdates,which intuitively re-rankcandidateresolutions for pronouns; formally, they are pushing new entities onto a stack ofvaluesforvariables,demotingothers.Theseupdatesarecontributedbythediversemechanisms for structuring discourse and shaping pronominal interpretation,including evoking discourse entities in specific grammatical roles, demonstratingentities with non-verbal (yet linguistic) actions, and signaling the direction ofdiscoursethroughvariousinterpretiveconnectionsbetweenclauses.Althoughthesemechanisms are heterogeneous, their contribution is governed by rules, notpragmaticreasoning,andassuch,shouldbeformallyrepresentedinlogicalform.

Ouraccountisprovisionalinseveralrespects.Wesaidlittleaboutthemodalprofileof context-sensitive utterances. 46 We remained silent about many ambiguitiesassociated with discourses containing pronouns—both within English and acrosslanguages.47Andwehavenotprovideda (exhaustive) list of coherence relations.Toseewhat’satstake,consider(29)fromWinograd(1972):

29 Thecitycouncildeniedthedemonstratorsapermit.i. Theyfearedviolence.ii. Theyadvocatedviolence.

Both (29i)and (29ii)exhibitanexplanatory relation,yet theoccurrencesof ‘they’are resolved differently. Our suggestion is that, though in (29i) and (29ii) thecontent of the second sentence is taken to explain the content of the first, theseexplanatory relations are qualitatively different. In (29ii), the council’s decisionaboutthedemonstratorscanbeexplainedonthebasisoftheformer’sbeliefsaboutthe latter: the relevant explanation being that because the demonstrators arepotentiallyviolent,orbelievedtobesobythecitycouncil,thecouncildeniedthema

46Still,ourformalsemanticshasresourcestocapturebasicfactsaboutthesemodalprofiles.47Forexample,wehavenotofferedatreatmentofdependentclauses.

25

permit.Meanwhile, in(29i), thecouncil’sdecisionaboutthedemonstratorscanbeexplained based on (other aspects of) the council’s attitudes: it is because thecouncilfearedviolencethattheydecidedtodenythedemonstratorsapermit.Thesequalitative differences suggest distinct coherence relations are at play. So, it’s nosurprise they come with different attention-shifting operations—when theexplanationgoesviasubject, thesubject ispromotedtoprominence;when itgoesvia object, the pattern is reversed. To flesh out this suggestion, we need tosubstantiate systematic differences between these kinds of explanations. Onlyfurtherempiricalresearchcanguideusclosertohavingafirmgrasponalloftheseissues,andwesubmitourproposalheretothescrutinyofthisfutureresearch.

We remain excited about the philosophical ramifications of the tools developedhere. Philosophers oftenuse linguistic examples to argue for context dependence.The kind of context dependence they have in mind, implicitly, is where they seecontext-dependent elements as freely selecting one interpretation from an open-ended array of candidates by unspecified, broadly pragmatic and open-endedmechanisms. Often, both the linguistic analyses and ensuing philosophicalargumentsdependonthismodelofcontextdependence.Butifweareright,contextis not as powerful as philosophers have presumed. In interpreting even astraightforwardcase like(1),ourdiscourseultimately lacksadependencyonnon-linguisticcontext.Thissuggestsaprojectofextendingouraccounttoothercasesofapparent contextual variability. If we can succeed in capturing theie apparentcontextual variability with uniform meanings and constrained variation,philosopherswillhavetogiveupmanycustomaryappealstocontextsensitivity.

6 Bibliography

Alahverdzhieva, K., & Lascarides, A. (2011). An HPSG Approach to SynchronousDeixis and Speech,. In S.Muller (Ed.),Proceedingsof the18th InternationalConferenceonHead-DrivenPhaseStructureGrammar(HPSG),6–24.Seattle.

Asher,N.,&Lascarides,A.(2003).LogicsofConversation.CambridgePress.Bittner ,M. (2012).PerspectivalDiscourseReferents for Indexicals.Proceedingsof

SULA7.Amhrest:GLSA,UMass.______(2007).OnlineUpdate:Temporal,Modal,andDeSeAnaphorainPolysynthetic

Discourse. In C. Barker, & P. Jacobson (Eds.),DirectCompositionality, 363–404.Oxford:OxfordPress.

Bittner,M.(2014).Temporality:UniversalsandVariation.Oxford:Wiley-Blackwell.Chomsky,N. (1981/1993).LecturesonGovernmentandBinding:ThePisaLectures.

MoutondeGruyter.Cohen, S. (1999). Contextualism, Skepticism, and The Structure of Reasons.

PhilosophicalPerspectives:Epistemology,13,57–89.

26

Dekker, P. (2011). Dynamic Semantics. In K. von Heusinger, C. Maienborn, & P.Portner (Eds.), Semantics: An International Handbook of Natural LanguageMeaning,Vol.1.(pp.923–945).Berlin:MoutonDeGruyter.

________(1994).PredicateLogicWithAnaphora.InL.Santelmann,&M.Harvey(Ed.),ProceedingsFromSALTIX,79–95.Ithaca,NY:CornellUniversity.

Dowell, J. L. (2011). A Flexible Contextualist Account of Epistemic Modals.Philosophers'Imprint,11(14),1–25.

Elbourne,P.(2005).SituationsandIndividuals.MITPress.Fiengo,R.,&May,R.(2006).DeLinguaBelief.Cambridge,MA:MITPress.______.(1994).IndicesandIdentity.Cambridge,MA:MITPress.Grosz,B.,&Sidner,C.(1986).Attention,IntentionsandTheStructureofDiscourse.

ComputationalLinguistics,12,175–204.Grosz,B.,Joshi,A.K.,&Weinstein,S.(1995).Centering:AFrameworkforModelling

theLocalCoherenceofDiscourse.ComputationalLinguistics,21,203–225.Heim, I. (1982).TheSemanticsofDefiniteandIndefiniteNounPhrases. PhDThesis,

UniversityofMassachusettesAmherst.Hobbs,J.(1979).CoherenceandCoreference.CognitiveScience,3,67–90._______.(1990).LiteratureandCognition(Vol.LectureNotes21).Stanford,California:

CenterfortheStudyofLanguageandInformation.Kameyama , M. (1996). Indefeasible Semantics and Defeasible Pragmatics. In M.

Kanazava,C.Pinon,&H.deSwart(Eds.),Quantifiers,DeductionandContext.111–138.CSLIPublications.

______. (1986). A Property-Sharing Constraint in Centering. Proceedings of 24thMeetingofAssoc.forComputationalLinguistics,200–206.NewYork.

______.(1999).StressedandUnstressedPronouns:ComplementaryPreferences.InP.Bosch, & R. van der Sandt (Ed.), Focus: Linguistic, Cognitive andComputationalPerspectives(pp.306–321).CambridgeUniversityPress.

Kamp,H.(1981).ATheoryofTruthandSemanticRepresentation.InJ.Groenendijk,T.Janssen,&M.Stokhof(Eds.),FormalMethodsintheStudyofLanguage(pp.277–322).Amsterdam:MathematischCentrum.

Kaplan, D. (1989b). Afterthoughts. In J. Almong, J. Perry, & H. Wettstein (Eds.),ThemesfromKaplan.OxfordUniversityPress.

Kaplan, D. (1989a). Demonstratives. In J. Almong, J. Perry, & H.Wettstein (Eds.),ThemesfromKaplan,481–565.OxfordUniversityPress.

Kehler,A.(2002).Coherence,Reference&theTheoryofGrammar.CSLIPublications.Kehler, A., Kertz, L., Rohde, H., & Elman, J. (2008). Coherence and Coreference

Revisited.JournalofSemantics,25,1–44.Keiser,E.(2009).EffectsofAnaphoricDependenciesandSemanticRepresentations

on Pronoun Interpretation. In S. L. Devi, A. Branco, & R. Mitkov (Eds.),AnaphoraProcessingandApplications(pp.121–130).Heidelberg:Springer.

Kendon,A.(2004).Gesture:VisibleActionasUtterance.NewYork:CambridgePress.King,J.C.(2014a.).SpeakerIntentionsinContext.Nous,48(2),219–237.______. (2014 b.). TheMetasemantics of Contextual Sensitivity. In A. Burgess, & B.

Sherman (Eds.),Metasemantics: new essays on the foundations of meaning,97–118.Oxford:OxfordUniversityPress.

27

King, J. C., & Stanley, J. (2005). Semantics, Pragmatics, and The Role of SemanticContent.InG.Zoltán,SemanticsvsPragmatics,111–164.OxfordPress.

Knott, A. (1996). A Data Driven Methodology for Motivating a Set of CoherenceRelations.DepartmentofArtificialIntelligence,UniversityofEdinburgh.

Kripke, S. (1977). Speaker's Reference and Semantic Reference. In P. French, T.Uehlig,&H.Wettstein(Eds.),MidwestStudiesinPhilosophy,Vo.II.255–276.

Lewis,D.(1969).Convention.Cambridge:HarvardUniversityPress.Mann, W. C., & Thompson, S. A. (1988). Rhetorical Structure Theory: Toward a

functionaltheoryoftextorganization.Text,8(3),243–281.Moens ,M., & Steedman,M. (1988). Temporal Ontology and Temporal Reference.

ComputationalLinguistics,14(2),15–28.Neale,S.(1990).Descriptions.MITPress.______. (2004). This, That, and the Other. In M. Reimer, & A. Bezuidenhout (Eds.),

DescriptionsandBeyond(pp.68–182).NewYork:OxfordUniversityPress.Perry,J.(1979).TheProblemoftheEssentialIndexicaal.Nous,13(1),3–21.Potts,C.(2005).TheLogicOfConventionalImplicatures.OxfordUniversityPress.Reimer,M.(1992).ThreeViewsofDemonstrativeReference.Synthese,93,373–402.Roberts, C. (2002). Demonstratives as Definites. In K. van Deemter, & R. Kibble

(Eds.), nformation Sharing: Reference and Presupposition in LanguageGenerationandInterpretation(pp.89–196).Stanford:CSLIPress.

Sidner,C.(1983).FocusingintheComprehensionofDefiniteAnaphora.InM.Brady,&R.Berwick(Eds.),ComputationalModelsofDiscourse,267–330.MITPress.

Smyth, R. (1994). Grammatical Determinants of Ambiguous Pronoun Resolution.JournalofPsycholinguisticResearch,23(3),197–229.

Stanley,J.,&SzabóGendler,Z.(2000).OnQuantifeierDomainRestriction.MindandLanguage,15(2-3),219–261.

Tonhauser, J.,Beaver,D.,Roberts,C.,&Simons,M. (2013).TowardaTaxonomyofProjectiveContent.Language,89,66–109.

Walker,M.A.,Masayo, I.,&Cote,S. (1994). JapaeaseDiscourseand theProcessofCentering.ComputationalLinguistics,20(2),193–232.

Webber,B. L. (1988).TenseandDiscourseAnaphor.ComputationalLinguistics,14(2),61–73.

Webber,B. L., Stone ,M., Joshi, A.K.,&Knott, A. (2003).Anaphora andDiscourseStructure.ComputationalLinguistics,29,545–587.

Wilkins, D. (2003). Why Pointing With the Index Finger is Not a Universal (inSocioculturalandSemioticTerms).InS.Kita(Ed.),Pointing:WhereLanguage,CultureandCognitionMeet171–215. Mahwah,NJ:LawrenceErlbaum.

Winograd,T.(1972).UnderstandingNaturalLanguage.NewYork:AcademicPress.Wolf, F., Gibson, E., & Desmet, T. (2004). Discourse Coherence and Pronoun

Resolution.LanguageandCognitiveProcesses,19,665–675.

Appendix: Formal definitions

We conclude with complete definitions for a logical language that formalizes our Attention–Coherenceapproach to pronouns. We assume a set of individual constants C ; a set of predicate symbols P ,each taking a specified number of arguments; and variables xi for each natural number i. Theinterpretation of constants is set up in terms of frames and models in the usual way.

• A Frame is a tuple F = hDw,R,De,Dti where Dw is a domain of possible worlds, R is a(transitive and reflexive) accessibility relation on Dw, De is a domain of individuals, and Dtis a domain of truth values (Dt = {0,1}). We require that the domains be disjoint: Dt \Dw =Dw \De = Dt \De = /0.

• A Model is a pair M = hF ,I i, where F is a frame and I is an interpretation function, whichassigns to each individual constant an element of De and each n-place predicate constant aset of pairs hw,si with w 2 Dw, and s an n-tuple of elements of De.

The Attention–Coherence approach captures interpretive dependencies across formulas using dy-namic semantics. Operations on sequences of individuals play a key role in the semantics. Giventhe individuals defined by a Frame F , we use the following notation to specify these operations.

• imIf m is an integer, then im is the mth element of i.

• im.nIf m and n are integers, then im.n is a sequence containing the subsequence of elements of iin order from element number m up through the element that precedes n (if any).

• im...

If m is an integer, then im... is the sequence containing the complete subsequence of elementsof i in order beginning from element number m.

• i+ jIf i is a sequence and j is a sequence, then i+ j is the sequence containing the elements of iin order followed by the elements of j in order.Note then that i = i0.k + ik...

• u.iIf u 2 Dt [Dw [De is an individual and i is a sequence, then u.i is the sequence that beginswith u and continues with the elements of i in order.

• w(i)If i is a sequence, then w(i) is the first element u of i such that u 2 Dw.

Now we can define the expressions of the language and their interpretations in a model M .

• Individual expressions

28

– if t is an individual constant, then t is an individual expression(represents the name of an individual)

– the variable xm is an individual expression(represents a discourse reference contributed by argument structure)

– if p is a unary predicate, then @p is an individual expression(represents a syntactically unconstrained anaphor)

– if p is a unary predicate and o is an individual expression, then @po is an individualexpression(represents a syntactically constrained anaphor)

The interpretation of individual expressions at a sequence i and world w:

• [[t]]i,w = I(t) for interpretation function I.(Access constants from model.)

• [[xm]]i,w = im.(Look up values of variables. We need variables to manage argument structure; otherwise,it will be very cumbersome to deal with the syntax–semantics interface for transitive andditransitive verbs; we need to potentially distinguish the order in which arguments are in-troduced, how salient they are after the utterance, and what role they play in the describedevent. Having variables clears this all up. Basically, x0 will correspond to the subject, x1 tothe direct object, x2 to the indirect object, and so forth.)

• [[@p]]i,w = i0 if hw, i0i 2 I(p).[[@p]]i,w = [[@p]]i1...,w otherwise.(Find most prominent referent that agrees with anaphor.)

• [[@pt ]]i,w = i0 if hw, i0i 2 I(p) and i0 6= [[t]]i,w.[[@pt ]]i,w = [[@pt ]]i1...,w otherwise.(Find most prominent free referent that agrees with anaphor.)

Conditions:

• If r is an n-place predicate symbol and t1 through tn are individual expressions, then r(t1, . . . , tn)is a condition.

The interpretation of conditions:

• [[r(t1, . . . , tn)]]i,w is true if and only if hw,h[[t1]]i,w, . . . , [[tn]]i,wii 2 I(r)(interpret atomic conditions by making sure the specified entities are in the specified relationat the world of evaluation)

Dynamic updates:

• haki is an update, for k 2N.(push new indefinite assignment for variable xk)

29

• hpkti is an update, where t is an individual expression, and k 2N.(push new assignment of t as a value of a variable xk)

• [j] is an update if j is a condition.(restrict the values of variables)

• H;K is an update if H and K are updates(composition—conjunction)

• ⇤K is an update if K is an update(metaphysical necessity)

The interpretation of dynamic updates: At each possible world w, the interpretation of a dynamicupdate is a relation on sequences:

• [[haki]](w, i, j) if and only if j = i0.k +o.ik... for some individual o 2 De.

• [[hpkti]](w, i, j) if and only if o = [[t]]i,w and j = i0.k +o.ik... for some individual o of type e.

• [[[j]]](w, i, j) if and only if j = i and [[j]]i,w is true.

• [[H;K]](w, i, j) if and only if there is some sequence h such that [[H]](w, i,h) and [[K]](w,h, j).

• [[⇤K]](w, i, j) if and only if j = i and for all worlds v accessible from w, there is some k suchthat [[K]](v, i,k).

Following Kaplan, we can define an initial context of a model M as any sequence (a,x,y) wherea 2 Dw (representing the actual world of the context), x 2 De (representing the speaker of thecontext) and y 2 De (representing the addressee of the context). (In a more general language,this could be extended by whatever parameters are appropriate for the interpretation of relevantindexical elements.) Then we can define truth in a context and a model as abstractions over thebasic dynamic updates:

• H is true in a model M and initial context J for M, if and only if there is some sequence isuch that [[H]](a,J, i).

• H is valid if and only if H is true in every model M for every initial context for M.

• H entailsS K if and only if for any model M and initial context J for M, if H is true at M andJ, then K is true at M and J.This is the “static” sense of entailment: K is a summary of H.

• H entailsD K if and only if for any model M and initial context J for M, and any assignmenti such that [[H]](a,J, i), there is an assignment k such that [[K]](a, i,k).This is the “dynamic” sense of entailment: K doesn’t add information to H. This versionof entailment is the one that’s usually given in treatments of dynamic semantics designed toaccommodate anaphora, because it allows anaphoric links not only between the premises butfrom the premises to the conclusion.

30