A Mental Model for a Rhetorical Arguer Floriana Grasso - PDF document

“believability”characteristicsofagenthood(Ca˜namero,1998,forexample).Thismeansthatimportantques-tions,forthemodelofarhetoricalarguer,like:“whatitisliketo`remember'something?”,or“whatistheexactdifferencebetween`assuming',`believing'and`suppos-ing'?”,or“howtorepresent`prejudices'asopposedto`beliefs'?”havenotbeenansweredsatisfactorily.Ifthemodelofthesingleagent'smentalstateisimpor-tant,inrhetoricalargumentationitisevenmorefunda-mentalforanagenttorepresenttheinterlocutor'sbeliefs,as“Inargumentation,theimportantthingisnotknowingwhatthespeakerregardsastrueorimportant,butknow-ingtheviewsofthoseheisaddressing”(PerelmanandOlbrechts-Tyteca,1969,x5).Moreover,intheGriceanviewofcommunicationthatarguesitisnecessarythattheinterlocutorsshareacommongroundofbeliefs,andknowthattheysharethem,twoagents,AandB,shouldalsobeabletoreasonaboutmutualbeliefs.Thisiswhatisbelievedatdifferent“nestinglevels”,thatiswhat:1.Abelieves2.Bbelieves3.AbelievesthatBbelieves4.BbelievesthatAbelieves5.AbelievesthatBbelievesthatAbelieves6.BbelievesthatAbelievesthatBbelievesetc.adinnitum(ClarkandCarlson,1982).InAIsys-tems,theproblemofinniterecursionhasbeeneithersolveddirectly,bygenerating“beliefnestings”asneeded(BallimandWilks,1991,forexample,adoptthisview)oraddressedbyusinganintuition,comingfromthephi-losophyofmind,thatagentshavea“mutualbeliefinduc-tionschema”,asortofpre-compiledattitudethatcanbeactivatedwheneveragentsassumetheysharesomecom-monground.Computationallythismeansthat,assoonasagreementisassumedonaparticularlevelofbeliefnesting,allsubsequentlevelsareconsideredequivalent.Thisisparticularlyappropriatewhencooperationisas-sumedamongagents:ithasbeennoted(Tayloretal.,1996)the“apparentirrelevance”ofrepresentingdeeplynestedbeliefs,forifnodeceptionisexpectedorallowed,onlyafewlevelsofnesting(three,infact,theypropose)isneededtocompletelymodeladialogue.Whateverthesolution,itisclearthatmutualityisacrucialconceptforthemodelofaconversationalagent.A“Rhetorical”MentalModelWedescribehereaverygeneralmodelforanagentabletoundertakerhetoricalargumentation.Themodelismeanttobeasnoncommittingaspossible,aswediscusswhatwebelieveisimportanttoinclude,ratherthantryandsolveissuesofrepresentationandreasoning.Wedis-cussthemodelbyconsideringthethreeissueswemen-tionedaboveasfundamentaltothetask:howtomodelthementalstate,howtoaccountformultipleattitudes,andhowtorepresentmutuality.RepresentingMentalAttitudesWewouldlikeouragentstobeabletoreasonaboutanymentalattitude.Wedonotwishtostateinadvancethetypesofmentalstatesanagentcanbein,like,forin-stance,intheBDIframework,butwewillsimplywrite:a(x)tomeanthatagentaisinthepropositionalattitudeoftowardsthestateofaffairx.Forinstance,toexpressthatAgentAbelievesXwewillwrite:BELA(X).Weexpectouragentstoshowinformational,motiva-tional,socialandemotionalattitudesasneededbythear-gumentation,andindeedweexpectrhetoricalargumen-tationtomakeusetoagreatextentofavarietyofatti-tudenuances.However,wedonotpropose,inthiswork,atheoryofmentalattitudes:howtheycometobe,howtheychange,howtheyinuenceeachother.Whilethismayappeardisappointing,itisalsotruethattheonlyrea-sonablewaytoapproachthisproblemisbyinvestigatingeachandanyofthementalattitudeswithasmucheffortastherehasbeenintheepistemological/informationalat-titudesanalysis,andthisisdenitelyoutsidethescopeofourwork.Wepropose,however,ageneralframeworktorepre-sentmentalattitudes,whichisinspiredbyworkoncon-textualreasoning(Benerecettietal.,2000).Theintuitionofcontextualreasoningisthatagentsreasonlocally:theyarenotverylikelytoneedtheircompletementalstatetosolveaparticularreasoningtaskathand,buttheyisolatea“box”,acontext,that,whilecontainingallnecessaryinformation,ismoremanageable.OfparticularinteresttousisthenotionofMultiContextsystems(GiunchigliaandBouquet,1998):thisadds,totheabovementionedlocalityassumption,thecompatibilityassumptionthatallcontextshavetoberelatedinsomewayintheagent'smind.Therefore,inadditiontotheruleswhicharevalidwithineachcontext,andwhichmightbeofdifferentna-ture,bridgerulescanbeusedtoreasonacrosscontexts.Moreformally,acontextisatriple:hL; ;
iwhereLisalanguage(e.g.rstorderlogic), isthesetofaxiomsofthecontextand
isthesetofinferencerulesthatap-pliesinthecontext.AMultiContextsystemsisapair:hC;BiwhereCisthesetofallcontexts,andBthesetofbridgerules.Abridgeruleisinturnaruleofthetype:c1:1:::cn:n!cn+1:n+1whereci:imeansthatformulaiistrueincontextci.Weseetheagent'smentalstateasasetof“attitudecontexts”,wheredifferentcontextsareusedtorepresentdifferentmentalattitudes.Anattitudecontextcontains,forthesingleagent,notonlytheobjectsofthatparticularattitude,butalsothereasoningrulesthattheagentusestomanipulatethepropositionalattitudesinthe“box”.Thisallowstopostulatethatdifferentattitudescanhavedif-ferentreasoningstyles.Principlestocharacteriseanddiversifyattitudes(BarwiseandPerry,1983,forexam-ple,propose8principlestothisaim)couldbeanexam-pleofrulesinternaltoacontext.Inadditiontointernalrules,wealsoassumethattheagenthasasetofbridgerules,thatallowtoascribenewpropositionalattitudesofasortfromothersofdifferentsorts.Similartowhat see,inotherwords,thethirdbeliefnestingmentionedby(Tayloretal.,1996;Kobsa,1990)asinfacta“double”environment,whichdependsontheinterlocutorwhosestandpointisconsidered.Thisallowstorepresentsituationsinwhichanagentrealisesthatwhattheotherhasperceivedfromthecon-versationisnotexactlywhatisinfactbelieved,becauseofamisconception,orbecauseofastatementofa“beliefforthesakeofconversation”.AnagentAsaying“eat-ingfruithelpsslimming”mayassumetobeascribedbyanotheragentBthebeliefthat“slimmingisimportant”,evenifthisisnotthecase:1.AbelievesthatnotX;2.AbelievesthatBbelievesthatitismutuallybelievedthatAbelievesthatX;Symmetrically,agentAcanascribea“beliefforthesakeofconversation”toagentB(includingitinthemutualbeliefsection)whilenotreallyascribingthebelieftoB:1.AbelievesthatBbelievesthatnotX;2.AbelievesthatBbelievesthatitismutuallybelievedthatBbelievesthatX.Thereforethesituationof“mutualbeliefthatX”,asenvisagedbefore(ClarkandCarlson,1982),correspondstothesetofbeliefs(fromagentA'spointofview):1.AbelievesthatX;2.AbelievesthatBbelievesthatX;3.AbelievesthatBbelievesthatitismutuallybelievedthatAbelievesthatX;4.AbelievesthatBbelievesthatitismutuallybelievedthatBbelievesthatX.MutualityacrossAttitudesWeassumethatmutualitycanbeexpressedforseveralattitudes(thoughwearepreparedtoacceptthatforsomeattitudesitwillhavelesssense).Forinstance,onecanimaginetodenemutualdesire,asaparaphraseofthedenitionofmutualbelief,as:1.Adesiresthatp.2.Bdesiresthatp.3.AdesiresthatBdesiresthatp.4.BdesiresthatAdesiresthatp.5.AdesiresthatBdesiresthatAdesiresthatp.6.BdesiresthatAdesiresthatBdesiresthatp:::Thisdenitionusefullyimplementsco-ordinationofac-tionamongagents(ClarkandCarlson,1982,givetheexampleoftheviolinduet,wherethetwoplayersmuststartsimultaneously)inBDIcontextswheredesireleadstointention(andthereforeaction).Weassume,however,thatmutualattitudesareex-pressedtowardsagent'sattitudes,andnotmerefacts,thatiswewanttorepresentthatitis“mutuallybelievedthatAbelieveX”,orthatisit“mutuallydesiredthatAbe-lievesX”,andsoon.Weassume,inotherwords,agenericmultiattitudescenario,whereinnerattitudes,andmutualattitudes,canbedenedasneeded.Figure1showsagraphicalrepresentationofnestinglevelsinagenericmultiattitudescenario,where: A AB ABmA ABmB Figure1:ModelofBeliefNestingsAisagentA'sownmentalstate(AbelievesX,orwantsYordoubtsthatZ,etc.)ABareattitudesascribedbyAtoagentB(AbelievesthatBbelievesX,orAbelievesthatBwantsY,etc).ABmArepresentsA'sviewonB'sviewonwhatmutuallyhappens“asregardtoA”(e.g.AbelievesthatBwantsthatitismutuallybelievedthatAbe-lievesX).ABmBrepresentsA'sviewonB'sviewonwhatmutuallyhappens“asregardtoB”(e.g.AbelievesthatBwantsthatitismutuallybelievedthatBbe-lievesX).MutualityasAttitudeContextsGoingbacktotheproblemofrepresentingthementalmodel,weconsiderthelevelsofnestingasiftheyweredifferentproposi-tionalattitudes(Thomason,2000).Inotherwords,weassumethat,forinstance,“believing”and“believingthatagentXbelieves”aretwodifferentattitudecontexts,andweassignthemdifferent“boxes”.Thismakessenseifitisnotedthatagentsmayusereasoningrules,whentryingtoputthemselvesinotheragents'shoes,whichmaywellbedifferentthanthoseusedwhenreasoningontheirownmentalstate.Moreover,wecanassumethesereasoningrulesaredependentontheparticularagenttheyreasonupon,so“believingthatagentXbelieves”mayworkdifferentlythan“believingthatagentYbelieves”.Asdescribedabove,bridgerulesareusedtoreasonacrosscontexts,allowing,forexample,toconcludesomethingonwhatanotheragentbelievesonthebasisonone'sownbeliefs.Inthishypothesis,thenumberofattitudecontextsthatthereareinanagent'smentalstatedependsonthenum-berofagentstoreasonabout,andthenumberofproposi-tionalattitudestoconsider.Forinstance,iftwoattitudesonlyareconsidered(say“belief”and“want”)inatwoagent(AandB)scenario,agentAhasacontextforeachprivatementalattitude:1.whatAbelieves;2.whatAwants.Then,agentAmightbeinoneofthetwoattitudesto-wardsanyattitudeofagentB: ConclusionsWehaveexplored,inthispaper,theproblemofrepre-sentingthementalstateofarhetoricalarguer,inthespiritofproducinganidealframework,ratherthanaspecic,implementationorienteddesignfortheagent'smentalstate.Bydrawinguponfundamentalnotionsofphilos-ophyofmind,andbyanalysinghowthesenotionshavebeenimplemented,wehaveproposedaclearpictureofwhatisneededtorepresentanagentabletoarguerhetor-ically.Fromthis,thereaderinterestedinactuallyim-plementingtheagent,canbenetfromoff-the-shelffor-malmodels,whetherfromlogic,probabilityorconstraintreasoning,bothtoreasonwithinthecontextsandforim-plementingbridgerules.Wewillnotchooseapreferredone,andweleavetheevaluationofwhichisthebestap-proachoutofthescopeofthiswork.ReferencesBallim,A.andWilks,Y.(1991).ArticialBelievers.LawrenceErlbaumAssociates.Barwise,J.andPerry,J.(1983).SituationsandAttitudes.MITPress,Cambridge,MA.Benerecetti,M.,Bouquet,P.,andGhidini,C.(2000).Contextualreasoningdistilled.JournalofEx-perimentalandTheoreticalArticialIntelligence,12(3):279–305.Ca˜namero,D.,editor(1998).EmotionalandIntelligent:TheTangledKnotofCognition.Papersfromthe1998AAAIFallSymposium.AAAIPress.Casta˜neda,H.-N.(1975).ThinkingandDoing.Kluwer.Clark,H.andCarlson,T.(1982).SpeechActsandHearer'sBeliefs.InSmith,N.,editor,MutualKnowledge,chapter1,pages1–36.AcademicPress.Cohen,P.andLevesque,H.(1990).IntentionisChoicewithCommitment.ArticialIntelligence,42:213–261.Fagin,R.,Halpern,J.,Moses,Y.,andVardi,M.(1996).ReasoningaboutKnowledge.MITPress,London.Galliers,J.(1992).AutonomousBeliefRevisionandCommunication.InG¨ardenfors,P.,editor,Be-liefRevision,pages220–246.CambridgeUniversityPress.Gernsbacher,M.andDerry,S.,editors(1998).Proceed-ingsofthe20thAnnualMeetingoftheCognitiveScienceSociety.LawrenceErlbaumAssociates.Giunchiglia,F.andBouquet,P.(1998).Acontextbasedframeworkformentalrepresentation.In(Gerns-bacherandDerry,1998),pages392–397.Grasso,F.(1998).Excitingavocadosanddullpears:Combiningbehaviouralandargumentativetheoryforproducingeffectiveadvice.In(GernsbacherandDerry,1998),pages436–441.Grasso,F.(2002).Towardsaframeworkforrhetoricalargumentation.InBos,J.,Foster,M.,andMath-eson,C.,editors,EDILOG'02:Proceedingsofthe6thworkshoponthesemanticsandpragmaticsofdialogue,pages53–60,Edinburgh.Grasso,F.,Cawsey,A.,andJones,R.(2000).DialecticalArgumentationtoSolveConictsinAdviceGiving:acasestudyinthepromotionofhealthynutrition.InternationalJournalofHuman-ComputerStudies,53(6):1077–1115.Kobsa,A.(1990).ModelingtheUser'sConceptualKnowledgeinBGP-MS,aUserModelingShellSystem.ComputationalIntelligence,6:193–208.Lee,M.andWilks,Y.(1997).EliminatingDeceptionsandMistakenBelieftoInferConversationalImpli-cature.InJokinen,K.,Sadek,D.,andTraum,D.,editors,IJCAI1997WorkshoponCollaboration,CooperationandConictinDialogueSystems.Perelman,C.andOlbrechts-Tyteca,L.(1969).TheNewRhetoric:atreatiseonargumentation.UniversityofNotreDamePress,NotreDame,Indiana.Rao,A.andGeorgeff,M.(1991).ModelingRationalAgentswithinaBDI-Architecture.InAllen,J.,Fikes,R.,andSandewall,E.,editors,Proceedingsofthe2ndInternationalConferenceonPrinciplesofKnowledgeRepresentationandReasoning(KR'91),pages473–484.MorganKaufmannPublishers.Russell,S.andNorvig,P.(2003).ArticialIntelligence:AModernApproach.PrenticeHall,2ndedition.Shoham,Y.andCousins,S.(1994).LogicsofMen-talAttitudesinAI,aVeryPreliminarySurvey.InLakemeyer,G.andNebel,B.,editors,FoundationsofKnowledgeRepresentationandReasoning,vol-ume810ofLNAI.Springer-Verlag.Stalnaker,R.(1974).Pragmaticpresuppositions.InMunitz,M.andUnger,P.,editors,SemanticsandPhilosophy,pages197–213.NewYorkUniversityPress,NewYork,NY.Taylor,J.,Carletta,J.,andMellish,C.(1996).Re-quirementsforBeliefModelsinCooperativeDia-logue.UserModelingandUser-AdaptedInterac-tion,6(1):23–68.Thomason,R.(2000).Modelingthebeliefsofotheragents.InMinker,J.,editor,Logic-BasedArti-cialIntelligence,pages375–473.KluwerAcademicPublishers,Dordrecht.Wooldridge,M.(2000).ReasoningAboutRationalAgents.MITPress,Cambridge,MA.