语言和思维的神经基础 The Neural Basis of Thought and Language.ppt

《语言和思维的神经基础 The Neural Basis of Thought and Language.ppt》由会员分享，可在线阅读，更多相关《语言和思维的神经基础 The Neural Basis of Thought and Language.ppt（53页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、The Neural Basis ofThought and LanguageFinalReview SessionAdministriviaFinal in class next Tuesday,May 9thBe there on time!Format:closed books,closed notesshort answers,no blue booksAnd then youre done with the course!The Second HalfCognition and LanguageComputationStructured ConnectionismComputatio

2、nal NeurobiologyBiologyMidtermFinalabstractionMotor ControlMetaphorGrammarBailey ModelKARMAECGSHRUTIBayesian Model of HSPBayes NetsOverviewBailey Modelfeature structuresBayesian model mergingrecruitment learningKARMAX-schema,framesaspectevent-structure metaphorinferenceGrammar Learningparsingconstru

3、ction grammarlearning algorithmSHRUTIFrameNetBayesian Model of Human Sentence ProcessingFull CircleNeural System&DevelopmentMotor Control&Visual System Spatial RelationPsycholinguistics ExperimentsMetaphorGrammarVerbs&Spatial RelationEmbodied RepresentationStructuredConnectionismProbabilistic algori

4、thmsConvergingConstraintsQ&AHow can we capture the difference between“Harry walked into the cafe.”“Harry is walking into the cafe.”“Harry walked into the wall.”Analysis ProcessUtteranceSimulationBelief StateGeneral KnowledgeConstructionsSemanticSpecification“Harry walked into the caf.”The INTO const

5、ructionconstruction INTO subcase of Spatial-Relationform selff.orth “into”meaning:Trajector-Landmarkevokes Container as contevokes Source-Path-Goal as spgtrajector landmark contcont.interior cont.exterior The Spatial-Phrase constructionconstruction SPATIAL-PHRASEconstructionalconstituentssr:Spatial-

6、Relationlm:Ref-Exprformsrf before lmfmeaningsrm.landmark lmmThe Directed-Motion constructionconstruction DIRECTED-MOTIONconstructionalconstituentsa:Ref-Expm:Motion-Verbp:Spatial-Phraseform af before mf mf before pfmeaningevokes Directed-Motion as dmselfm.scene dmdm.agent amdm.motion mmdm.path pmsche

7、ma Directed-Motionrolesagent:Entitymotion:Motionpath:SPGWhat exactly is simulation?Belief update plus X-schema executionhungrymeetingcafetime of dayreadystartongoingfinishdoneiterateWALKat goal“Harry walked into the caf.”readywalkdonewalker=Harrygoal=cafeAnalysis ProcessUtteranceSimulationBelief Sta

8、teGeneral KnowledgeConstructionsSemanticSpecification“Harry is walking to the caf.”“Harry is walking to the caf.”readystartongoingfinishdoneiterateabortcancelledinterruptresumesuspendedWALKwalker=Harrygoal=cafeAnalysis ProcessUtteranceSimulationBelief StateGeneral KnowledgeConstructionsSemanticSpeci

9、fication“Harry has walked into the wall.”Perhaps a different sense of INTO?construction INTO subcase of spatial-prepform selff.orth “into”meaningevokes Trajector-Landmark as tlevokes Container as contevokes Source-Path-Goal as spg contcont.interior cont.exterior construction INTO subcase of spatial-

10、prepform selff.orth “into”meaningevokes Trajector-Landmark as tlevokes Impact as imevokes Source-Path-Goal as spg im.obj1 im.obj2 “Harry has walked into the wall.”readystartongoingfinishdoneiterateabortcancelledinterruptresumesuspendedWALKwalker=Harrygoal=wallMap down to timelineS R E readystartongo

11、ingfinishdoneconsequencefurther questions?What about“Harry walked into trouble”or for stronger emphasis,“Harry walked into trouble,eyes wide open.”Metaphorsmetaphors are mappings from a source domain to a target domainmetaphor maps specify the correlation between source domain entities/relation and

12、target domain entities/relationthey also allow inference to transfer from source domain to target domain(possibly,but less frequently,vice versa)is Event Structure MetaphorTarget Domain:event structureSource Domain:physical spaceStates are Locations Changes are Movements Causes are Forces Causation

13、is Forced Movement Actions are Self-propelled Movements Purposes are Destinations Means are Paths Difficulties are Impediments to Motion External Events are Large,Moving Objects Long-term Purposeful Activities are JourneysKARMADBN to represent target domain knowledgeMetaphor maps link target and sou

14、rce domain X-schema to represent source domain knowledgeMetaphor Maps1.map entities and objects between embodied and abstract domains2.invariantly map the aspect of the embodied domain event onto the target domain 1.by setting the evidence for the status variable based on controller state(event stru

15、cture metaphor)3.project x-schema parameters onto the target domainfurther questions?How do you learnthe meanings of spatial relations,the meanings of verbs,the metaphors,andthe constructions?How do you learnthe meanings of spatial relations,the meanings of verbs,the metaphors,andthe constructions?T

16、hats the Regier model.(first half of semester)How do you learnthe meanings of spatial relations,the meanings of verbs,the metaphors,andthe constructions?VerbLearnschemaelbow jntpostureaccelslideextendpalm6schemaelbow jntpostureaccelslideextendpalm8schemaelbow jntpostureaccelslide 0.9extend 0.9palm 0

17、.96data#1data#2data#3data#4schemaelbow jntpostureacceldepress 0.9fixed 0.9index 0.92schemaelbow jntpostureaccelslide 0.9extend 0.9palm 0.96-8schemaelbow jntpostureslide 0.9extend 0.9palm 0.7grasp 0.3schemaelbow jntpostureacceldepressfixedindex2schemaelbow jntpostureaccelslideextendgrasp2Computationa

18、l Detailscomplexity of model+ability to explain datamaximum a posteriori(MAP)hypothesishow likely is the data given this model?penalize complex models those with too many word senseswants the best model given dataHow do you learnthe meanings of spatial relations,the meanings of verbs,the metaphors,a

19、ndthe constructions?conflation hypothesis(primary metaphors)How do you learnthe meanings of spatial relations,the meanings of verbs,the metaphors,andthe constructions?construction learningAcquisitionReorganizeHypothesizeProductionUtterance(Comm.Intent,Situation)GenerateConstructions(Utterance,Situat

20、ion)AnalysisComprehensionAnalyzePartialUsage-based Language LearningMain Learning Loopwhile available and cost stoppingCriterionanalysis=analyzeAndResolve(utterance,situation,currentGrammar);newCxns=hypothesize(analysis);if cost(currentGrammar+newCxns)cost(currentGrammar)addNewCxns(newCxns);if(re-og

21、anize=true)/frequency depends on learning parameterreorganizeCxns();Three ways to get new constructionsRelational mappingthrow the ballMergingthrow the blockthrowing the ballComposingthrow the ballball offyou throw the ball offTHROW BALL OFFTHROW OBJECTTHROW BALLMinimum Description LengthChoose gram

22、mar G to minimize cost(G|D):cost(G|D)=size(G)+complexity(D|G)Approximates Bayesian learning;cost(G|D)posterior probability P(G|D)Size of grammar=size(G)1/prior P(G)favor fewer/smaller constructions/roles;isomorphic mappingsComplexity of data given grammar 1/likelihood P(D|G)favor simpler analyses(fe

23、wer,more likely constructions)based on derivation length+score of derivationfurther questions?Connectionist RepresentationHow can entities and relations be represented at the structured connectionist level?orHow can we represent Harry walked to the cafin a connectionist model?SHRUTIentity,type,and p

24、redicate focal clustersAn“entity”is a phase in the rhythmic activity.Bindings are synchronous firings of role and entity cellsRules are interconnection patterns mediated by coincidence detector circuits that allow selective propagation of activityAn episode of reflexive processing is a transient pro

25、pagation of rhythmic activityasserting that walk(Harry,caf)Harry fires in phase with agent rolecafe fires in phase with goal role+-?agtgoal+e+v?e?v+?walkcafe Harry typeentitypredicate“Harry walked to the caf.”asserting that walk(Harry,caf)Harry fires in phase with agent rolecafe fires in phase with

26、goal role+-?agtgoal+e+v?e?v+?walkcafe Harry typeentitypredicate“Harry walked to the caf.”Activation Trace for walk(Harry,caf)+:Harry+:walk+e:cafewalk-agtwalk-goal1234further questions?Human Sentence ProcessingCan we use any of the mechanisms we just discussed to predict reaction time/behavior when h

27、uman subjects read sentences?Good and Bad NewsBad news:No,not as it is.ECG,the analysis process and simulation process are represented at a higher computational level of abstraction than human sentence processing(lacks timing information,requirement on cognitive capacity,etc)Good news:we can constru

28、ct bayesian model of human sentence processing behavior borrowing the same insightsBayesian Model of Sentence ProcessingDo you wait for sentence boundaries to interpret the meaning of a sentence?No!As words come in,we constructpartial meaning representationsome candidate interpretations if ambiguous

29、expectation for the next wordsModelProbability of each interpretation given words seenStochastic CFGs,N-Grams,Lexical valence probabilitiesSCFG+N-gramReduced RelativeMain VerbSNPVPDNVBNThecoparrestedthe detectiveSNPVPNPVPDNVBDPPThecoparrestedbyStochastic CFGSCFG+N-gramMain VerbReduced RelativeSNPVPD

30、NVBNThecoparrestedthe detectiveSNPVPNPVPDNVBDPPThecoparrestedbyN-Gram SCFG+N-gramMain VerbReduced RelativeSNPVPDNVBNThecoparrestedthe detectiveSNPVPNPVPDNVBDPPThecoparrestedbyDifferent Interpretations Predicting effects on reading timeProbability predicts human disambiguationIncrease in reading time because of.Limited ParallelismMemory limitations cause correct interpretation to be prunedThe horse raced past the barn fellAttentionDemotion of interpretation in attentional focusExpectationUnexpected wordsOpen for questions