Comics

A plan-based model of misunderstandings in cooperative dialogue

Description
A plan-based model of misunderstandings in cooperative dialogue
Categories
Published
of 32
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Aplan-basedmodelofmisunderstandingsin   cooperativedialogue    LilianaArdissono,GuidoBoellaandRossanaDamiano  DipartimentodiInformatica,Universit adiTorino,CorsoSvizzeran.185,10149Torino, Italy.email:  f  liliana,guido  g  @di.unito.it  ThisworkisgoingtoappearintheInternationalJournalofHuman-ComputerStudies, SpecialIssueon  Detecting,Repairing,andPreventingHuman-MachineMiscommunica- tion.  Wedescribeaplan-basedagentarchitecturethatmodelsmisunderstandingsincooperativeNL agentcommunication;itexploitsanotionofcoherenceindialoguebasedontheideathatthe explicitandimplicitgoalswhichcanbeidentiedbyinterpretingaconversationalturncan berelatedwiththepreviousexplicit/implicitgoalsoftheinteractants.Misunderstandings arehypothesizedwhenthecoherenceoftheinteractionislost(i.e.anunrelatedutterance comes).Theprocessesofanalysis(andtreatment)ofamisunderstandingaremodeledasrational behaviorscausedbytheacquisitionofasupplementarygoal,whenanincoherentturncomes:theagentdetectingtheincoherencecommitstorestoretheintersubjectivityinthedialogue;so,herestructureshisowncontextualinterpretation,orheinducesthepartnertorestructurehis (accordingtowhoseemstohavemadethemistake).Thiscommitmentleadshimtoproducea repairturn,whichinitiatesasubdialogueaimedatrestoringthecommoninterpretationground.Sincewemodelspeechactsuniformlywithrespecttotheotheractions(thedomainlevelactions),ourmodelisgeneralandcoversmisunderstandingsoccurringatthelinguisticlevelaswellasat theunderlyingdomainactivitiesoftheinteractants. 1Thephenomenonofmisunderstandings  1.1INTRODUCTION  Inorderforadialoguetoproceedsuccessfully,itisimportantthatthespeakersmain- tainaconsistentviewoftheinteraction.Nevertheless,vaguenessandambiguitiescan interveneinaspeaker'sutteranceandinterferewiththeprocessofinterpretationof thatutterance.Manyresearchershaverecognizedtheroleofvariousambiguitytypesin generatingmisunderstandings,fromdierentperspectives(Zaeerer1977,Dascal1989, Blum-Kulka&Weizman1988,Vendler1994).Inthelinguisticandphilosophicalresearch,misunderstandingshavebeenanalyzed underdierentpointsofview:forexample,Zaeerer(1977)andDascalsupporttheidea thatthestudyofmisunderstandingsisrelatedwithcomprehensioninanindirect,external way,\muchinthesamewayaspathologicalbehaviorisoftensaidtobeabletoillumi- natethenatureof`normal'behavior"(Dascal1989).Morerecently,misunderstanding   hasbeenrecognizedbyBlum-Kulka&Weizman(1988)andWeigand(1997)asanormal phenomenonincommunication,tointroduceintoanharmonicalmodelofdialogue.Intheresearchonconversationalanalysis,Scheglo(1987),Scheglo(1992)hasana- lyzedmisunderstandingswithrespecttothesequencesofturnsinadialogue,inorderto identifywhichspecicmechanismsareusedbytheinteractantstodefendthecommonset ofbeliefsnecessaryfortheinteractiontogoonsuccessfully:hecalls  \intersubjectivity"  thiscommongroundindialogueandpointsoutthat,duringaninteraction,thespeakers monitortheirpartners'reactionsandinterpretthemasdisplaysofunderstanding/mis- understandingofthepreviousturns.\Havingregisteredtheobservationthat,through theirtalk,speakerscandisplayaspectsoftheirunderstandingofpriortalk,itremains tobenotedthat,indoingso,theycanrevealunderstandingsthatthespeakersofthat priortalkndproblematic-inotherwords-whattheytaketobe  mis  understanding" (Scheglo1992).Schegloconsidersvariouscaseswherethereareproblemsinmaintain- ingtheintersubjectivityamongtheinteractants;hestudiesdierenttypesofrepairs, whichcanbestartedbythemisunderstoodspeaker,orbytheagentwhohasmisunder- stoodapreviousturn,whentheyrealizethatsomethingiswrongwiththeinterpretation.Heexplainsthatrepairsareanimportantinstrumentforreconstructingthemutualityin theinteractants'beliefs.Inthispaper,wedescribeaplan-basedagentmodelthattakesintoaccounttheprob- lemofmisunderstandingsin(cooperative)agentinteraction.ThismodelsupportsNL communicationandinterpretsdialogueasarational,cooperativeformofinteraction amongagents.Thecoherenceofadialogueisassessedbyidentifyingtherelationamong eachturnandthepreviousinteractioncontext:wemodelagoal-basednotionofcoherence, whichbuildsontheideathattheexplicitandimplicitgoalsidentifyablebyinterpreting aconversationalturncanberelatedwiththepreviousexplicit/implicitgoalsofthe interactants(Allen1983,Litman&Allen1987,Carberry1990).Undertheassumption thattheinteractantscooperateandthateveryturnisperformedtocarryonsomeoftheir goalsjointly,amisunderstandingishypothesizedwhenthecoherenceofthedialogueis lost,i.e.anutterancecomeswhichisnotrelatedwiththecontextualgoals.Infact,the presenceofanutterancethatdoesnotcontributetothegoalspursuedinthereceiver's interpretationofthedialogueistakenasasignthathisinterpretationcontextisdierent fromthespeaker'sone(whocertainlyconsideredhisownturncoherent).Theprocessesofanalysisandresolutionofamisunderstandingarerationalbehaviors causedbytheacquisitionofasupplementarygoalwhenanincoherentturncomes:when anagent  A  interpretsaturnandndsitincoherentwithrespecttothepreviouscontext, hecanadopttheintentionoflookingforanalternativeinterpretationoftheinteraction.Thischoiceistheconsequenceofanothergoalwhichhedecidestopursue: A  wantsthat hisviewofdialogueconvergeswithhisinterlocutor'sone.Ifheconsidersitunlikely thatthespeaker  B  hasperformedatopicshift,orthatabreakdownincooperationhas occurred,helooksforanactiontomodifythewronginterpretationwhichcausedthe misunderstanding.Suchanactionmustleadoneoftheinteractants,possiblyhimself, tochangetheinterpretationcontext:so,  A  restructureshisowninterpretationofthe previousdialogue,orheinduces  B  torestructurehisdialoguecontext,dependingonwho  A  believestohavemadethemistake.Thiscommitmentleads  A  toproducearepairturn, whichinitiatesasubdialogueaimedatrestoringthecommoninterpretationground.Manyresearchers(e.g.considerPollack(1990),Perrault(1990),Hobbs  etal.  (1993),   Nagao(1993)andHirst  etal.  (1994))haveusedabductiveframeworksforcarryingonthe interpretationofadialogueandpossiblyrestructuringthedialoguecontext,whenafailure occursintheintegrationofnewutterances.Inourplan-basedmodel,thewholeinter- pretationprocessisrepresentedastheexecutionofinterpretation(andreinterpretation) actions,aswellasrestructuringactionswhichmodifytheagent'sdialogueinterpretation;moreover,themaintenanceofacorrectinterpretationcontextisamutualgoalofthe interactants(Cohen&Levesque1991).Inthissense,theactionsundertakenbyagentsto recoverfromamisunderstandingarenotaseparateactivitywithrespecttotheirnormal rationalbehavior:theyareadoptedascommitmentsbythem,whennecessary. y  Webe- lievethatthisisamajordierencewithrespecttotheothermodelsofdialogueprocessing, wherethetreatmentofmisunderstandings(andgeneralcommunicationproblems)isem- beddedinthenormalinterpretationprocess,orismanagedbyexternalrecoverystrategies (likethemetarulesusedinEller(1993)torestructurethedialoguecontext).Moreover, someofthesemodelsaremainlyfocusedonmisunderstandingsintheinterpretationof theillocutionaryforceofutterancesandareconstrainedbyarigidnotionofcoherence.Instead,weclaimthatthemisunderstandingphenomenongoesbeyonddialogueandre- gardswiderkindsofinteractions;inourmodel,weanalyzemisunderstandingsdueto ambiguitiesbothatthelevelofcommunicativeanddomainactivity.Incoherentturnsarenotalwaysduetomisinterpretations:alsotopicshiftsandinten- tionalbreakdownsincooperationshouldbeconsidered.Currently,wedon'tmodeltopic shiftsduetotheinitiationofnewdialogues;however,aspointedoutinotherworks,  z  focusandtopicshiftsareusuallymarkedbythepresenceof\cue"words.So,theycanbe distinguishedfrommisunderstandings.Inourmodel,thismeansthatwhenanunexpected changeinadialogueoccurs,thepresenceofacuewordshouldtriggerthehypothesisthat atopicshifthasoccurred,beforehypothesizingthatamisunderstandinghasoccurred.So, nomisunderstandinganalysiswouldstart.Onthecontrary,sincewemodelcooperative dialogues,weexcludethehypothesisthatabreakdownincooperationcanoccur.Thepaperisorganizedasfollows:afterabriefdescriptionofthephenomenonof misunderstandings(section1.2),wewillpresentournotionofcoherenceindialogueand ourapproachtotherepairofamisunderstanding(section2).Then,wewilldescribe ourcomputationalmodelofdialogue(section3):theagentmodel(section3.1),the interpretationprocessofutterances(section3.2)andtherecognitionandrecoveryfroma misalignmentindialogue(section3.3and3.4).Wewillthenprovideadetailedexample toshowhowourmodelworks(section4).Section5describestheevidenceaboutthe occurrenceofmisunderstandingsgatheredbystudyinganumberofdialogues.Finally, section6comparesourmodeltootherrelatedworksandsection7concludesthepaper. 1.2BACKGROUND  Scheglo(1992)describesthebehaviorofagentswhenamisunderstandingoccursand identiesdierenttypesofrepairs.Healsoimplicitlysupportstheideathatmisunder- standingscanbedistinguishedaccordingtowhichtypeofinformationisactuallymisinter- preted;forexample,anincorrectidenticationofthereferentofadescriptionrepresents  y  Theinterpretationandrestructuringactionsundergothesameplanningandexecutionprocessas theotheractions,likedomain-levelactionsandspeechacts. z  Forexample,considerCohen(1984),Cohen(1987),Litman(1986),Litman&Allen(1987),Grosz&Sidner(1986)andGrosz&Sidner(1990).  adierentlevelphenomenonwithrespecttoanincorrectidenticationoftheintended speechact,oroftheplanunderlyingtheinteraction.InScheglo'sdescription,thirdpositionrepaircorrespondstothecaseswherethemis- understoodagentrealizesthatthepartnerhasawronginterpretationandurgeshimtore- structureit.Forexample,considerturnT4inthefollowinginteraction(Scheglo(1992), page1317):Example1:T1:Dan: \...SeeAltends,itseems,topullinoneortwoindividualsonhisside ...]"  T2:Al: \W'l"  T3:Roger: \WellsodoI."  T4:Dan: \Yeah.I'mnotcriticizing,Imeanwe'lljust..."  T5:Roger: \Ohyouwannatalkabouthim"  T6:Dan: \look,let'sjusttalk"  Intheexample,DanrealizesthathehasbeenmisunderstoodbyRogerandprovides himwiththeintendedinterpretationofhisownprevioussentencebyperformingathird positionrepairinturnT4:inthisway,heurgesRogertochangehisinterpretationofthe dialogue. y  Infourthpositionrepair,  z  aspeaker  A  producesaturnthatismisunderstoodby  B  , whoreplieswithanotherturn,sequentiallyappropriatetohisown(wrong)understanding oftherstturn. A  doesn'trealizethatthetwointerpretationsarenotalignedanymore andrespondswithafurtherturn,coherentin  A  'sviewofdialogue.Atthispoint,  B  understandshisownmistake,restructureshisowninterpretationaccordinglyandinforms thepartnerabouttherealignment.Forexample(Scheglo(1992),page1321):Example2:T1:Marty: \Loes,doyouhaveacalendar,"  T2:Loes: \Yeah"  ((reachesforherdeskcalendar)) T3:Marty: \Doyouhaveonethathangsonthewall?"  T4:Loes: \Oh,youwantone."  T5:Marty: \Yeah"  Therecognitionthatamisunderstandinghasoccurredistriggeredbythelackof coherencebetweentheinterpretationofthemisunderstoodturnandthatofthesubsequent turns.Havingmisinterpreted,infact,thehearergiveshiscontributiontodialogueina waythatiscoherentinhiswronginterpretation,butnotfromthepointofviewofhis partner'sintendedmeaning. y  TurnT5lookslikeafourthpositionrepairbyRoger,butwewilldiscussthisaspectinsection2.2. z  Thirdandfourthpositionrepairsarenotnecessarilythethird/fourthturnsoftheinteraction,startingfromthemisunderstoodturns:severalturns(evenlongsubdialogues)cancome,beforeanagent realizesthataninterpretationproblemhasoccurred.Anyway,inthirdpositionrepairs,themisunderstood speakerperformstherepairturnaftera\sequentiallyinappropriate"turnfromhispartnercomes(so,the repaircomesinthirdposition,inanextendedway);instead,infourthpositionrepairthemisunderstander detectshisownmistakeafterthepartnerspeaksagain;so,therepaircomesinanextendedfourthposition.InScheglo(1992),theexpression  \sequentiallyappropriate"  hasbeenusedtoindicatethecoherence relationbetweenaparticipant'sunderstandingofaturnandhisreply.  2Coherenceandmisunderstandingsindialogue  2.1COHERENCEASARATIONALPHENOMENON  Weanalyzedialoguefromtheintentionrecognitionpointofview(Cohen  etal.  1981, Allen1983,Allen&Perrault1980,Cohen&Levesque1991):whenanagentacts,are- lationofhisactionwiththeinteractioncontextislookedfor,toseewhethertheaction representsanattempttosatisfyanyintentionexpressedexplicitlybythepartner,oran implicitgoalwhichcanonlybeinferredbyreasoningonthepartner'splans,orifitis afurtherstepinaplanthattheagenthasalreadystarted.Adierentapproachto theanalysisofcoherencewasadoptedbytheconversationalanalysts,whointroduced  adjacencypairs  (Sacks  etal.  1974)tomodeltheexpectedcontinuationsofaninterac- tion:adjacencypairsaresequencesofspeechacts(e.g.question-answerpairs)suchthat, aftertherstelementoccurs,thesecondoneisexpected.However,aspointedoutin Levinson(1981),agentbehaviorcannotdirectlybeexplainedbymeansofsuchtypeof strictinteractionalrules.Onthecontrary,theintentionalapproachtodialogueinterpre- tationmakesitpossibletoadoptaexiblenotionofcoherence:anewcontributionis consideredcoherentaslongasarelationcanbeidentiedamongtheintentionsunder- lyinganactionandthepreviouspendingintentionsoftheinteractants.Followingthe ideasofCastelfranchi&Parisi(1980),weconsideranutterancecoherentwiththeprevi- ouscontextifandonlyifitsreceivercaninterpretitasameansofthespeakertoachieve anunsatisedgoal  g  whichrealizesoneofthefollowingcoherencerelations:1. Goaladherence:  g  isoneofthegoalsaddressedexplicitlybythepartnerinaspeech act(notnecessarilyinthelastturnoftheinteraction).Goaladherencecoversthephenomenatreatedbyadjacencypairs,butitisamore generalapproachthanthat.Werepresentspeechactsasactions(Austin1962) havingtheeectthattheinteractantssharethebeliefthatthespeakerintendsthe partnertoexecuteacertain(domain-levelorlinguistic)action.Anillocutionary actandtheactionoccurringinitseectclearlycorrespondtotherstandsecond componentsofanadjacencypair(e.g.aquestionisperformedtoinducethehearer toanswerit).Inthesimplestcase,dialoguesarecomposedofsequencesofadjacency pairsbecausetheparticipantsrecognizeeachother'sintentionsandreacttothem asexpected.However,otherbehaviorsaremodeledinourintentionalapproach:e.g.insertionsequences(Scheglo1972),like\question-question-answer-answer" sequences(wherefurtherquestionsareaskedbeforetheanswertotheinitialquestion isprovided)canbemodeledasinvestigationsontheactionsaddressedintheinitial question.Askingquestionsaboutthepreconditionsofanactionisexplainedasa coherentcontinuation,wherethespeakertriestoexecutetheactionand,todothat, movestothesatisfactionofthesubgoalofcheckingwhethertheactionisexecutable ornot.E.g.considerthefollowingexcerpt,takenfromMerritt(1976):T1:A: \Sellmeabottleofwhisky,please."  T2:B: \Areyou18?"  Alsonoticationsof(un)successfulperformanceofactions(e.g.acknowledgements) followfromthesatisfactionofajointgoal(seeCohen&Levesque(1991)):T1:A: CouldyouregistermefortheArticialIntelligenceexam?  T2:B: Ok,youareregistered. 
Search
Similar documents
View more...
Tags
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks