Linguistics in the context of the computational an

LinguisticsinthecontextofthecomputationalandcognitivesciencesWinter2000JohnGoldsmithMindLogicComputationLanguageMachinesIstheoriginofknowledgeinthesenses,orinthemind?WorldrepresentationRepresentation=F(P1,P2)P1P2Whatlessonscanwelearnfromthenaturalsciencesindevelopingthehumansciences?InformationSciencesTheIllusionofReality,HowardL.Resnikoff(SpringerVerlag,1989)“Thehistoryofinformationscienceconsistsoffourinterwovenstrands:thestudyofthermodynamicsandthetheoryofmeasurementinphysics;thestudyofsensoryinformationprocessingandknowledgerepresentationinbiologicalsystems;theriseofelectricalcommunicationengineering;andthestudyofcomputabilityandthedevelopmentofcomputingmachines.Inrecentyears,thecommonfeaturesofthesesuperficiallydisparatethemeshavebeguntocoalescetoformanintegratedintellectualdisciplinewhoseprincipalproblemsblendelementsofallfourconstituentfields.”(p.3)TheCognitiveSciencesGoal:tostudyintelligence;intelligentsystems;intelligencebehavior;(Lookingahead:islinguistic“behavior”or“knowledge”aformofintelligence?)Goal:thestudyofintelligenceanditscomputationalprocesses.WhatcontributestoCogSci?Psychology:–Cognitivepsychology(GeorgeMiller,J.J.Gibson,U.Neisser)–Psycholinguistics–Useof“informationprocessing”asapointofviewwithoutactualcomputermodeling(simulations).DisciplinesincogsciArtificialintelligence(asapartofcomputerscience--computerscienceasapartofelectricalengineering!)SeymourPapert,MarvinMinskyatMIT;NewellandSimon;LinguisticsFromthemid-1950s,stakingaclaimtopsychologicalreality.Chomskianprogram;Computationallinguistics;Psycholinguistics.(whatiscomputationallinguistics?)SyntacticparsingMappingsurfacestructures(linearwordorder)tosemanticslotsfordatabaseaccess.AutomaticspeechrecognitionWhatistherelationshipoflanguagetolinguisticswithincogsci?TheotherdisciplinesinthecognitivesciencesdoNOTcedetheproblemoflanguagetolinguistics.Inmanycases,theyalmostentirelyignorethecurrentlinguistictakeonlanguageproblems.ContributerstoCogSciPhilosophy:epistemology(thestudyoftheoriginandlogicalfoundationsofknowledge)logic;studyofprobability;?Wheretoplace:thestudyofBayesiandecisiontheory.ContributerstoCogSciNeuroscience:especiallyneuroanatomyOthers:Economics(studyofrationaldecision-takers);Cognitiveanthropology,Cognitivesociology.Whatisananalysisofacognitivesystem?Atitsmostabstract:Ananalysisofthecontrolandflowofinformation.I.Memory:howisinformationretainedinapermanentfashion?(long-term);2.Howisnewinformationadded(learning)?3.Howistheenvironmentperceived?4.Howisaproblemdefinedandsolved?SometypicalproblemsforcognitivescienceVision:perceptionofobjects(in3Dspace)Memory:ofevents;ofinformation;oflanguage;Categorization/classificationPlanning:howtoachievegoalsfrominitialstate;TypesofmodelsSimpleprosemodelswithaveneerofinformationprocessing.Implementedmodels(=programwritteninaprogramminglanguage)MathematicalmodelsConnectionist/neuralnetworkmodelsTraditionalsymbolicmodelsusingvonNeumann-stylecomputermodelsMemoryaccessedthroughaddress-labeling;Memorycanstorebothdataandprogram;Inputandoutputdevices;Serialcomputationprimarily;Structuredlinearcomputationallogic.Heavyuseoflinkedlistsandassociativelists(hashes).NeuralnetworkstylemodelsLargenumbersofsimple,identicalobjects;Eachobjecthasanactivationlevel(arealnumber);Distributedconnectionsamongunits;Learningisinherenttothesystem;Nocentralizedcontrol.Comingtogripswiththeanalysisofintelligence:2modelsInferenceversusSearchInference:philosophy.Linguisticsandpsycholinguistics,bayesiandecisions,computerscience;Search:traditionalAI,muchpsychology:studyofheuristics(=rulesofthumbthatworkmostofthetimeprettywell.)Issuesinserialism:The100-steplimit:Manyinterestingcognitivecomputationstakeonesecondorless.Eachmachinecycleisabout10msec;hencenomorethan100cyclesarerequiredtoaccomplishmostinterestingcognitivetasks.That’saseriouslimit.Somecommonassumptionsindealingwithintelligence:Alargenumberofrulesoftheformofproductions:Rule#N:UnderconditionsC,doP.Whatdoyoudowhenmorethanoneruleissatisfiedatagiventime?Thecontrolstructure.Linguistsknowaboutthatproblem!Butlinguistshavebeenremarkablyuncreativeinconsideringalternatives.NetworkmodelsWilliamJamesDonaldHebbWarrenMcCulloch(andPitts)FrankRosenblattRossQuillian(spreadingactivationmodels)PDP:McClellandandRumelhart,back-propagation.JohnHopfieldLanguage--finally.Islanguage:a.uniqueb.whatpullsallofcognitiontogether?c.bothaandb(NO!)d.neitheranorb.Doeswhatweknowaboutcategorizationandmemoryfromgeneralfunctionstellusanythingaboutlanguage?Couldlanguage’sapproachtotheproblemsofcategorizationandmemorybedistinctfromotherproblems?Doeswhatweknowaboutlanguageinformusaboutothercognitivefunctions?Islanguageunique?(1)Islanguageinnate?(2)Doesthatquestionbearonwhetherlanguageisisolatedfromotherfunctions/deeplylinkedtootherfunctions?Doesourgrowingknowledgeoflocalizationoffunctionsinthebrainhelpanswerquestions(1)or(2)?LearningSoabigpairofquestionsis:Whetherlanguageisinnate;Whetherlanguageisintegratedvslanguageisisolatedfrom(other)cognitivefunct