{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
1PhilosophicalPsychology
Vol.19,No.3,June2006,1–17
MaterialSymbols
5
AndyClark
Whatistherelationbetweenthematerial,conventionalsymbolstructuresthatweencounterinthespokenandwrittenword,andhumanthought?Acommonassumption,thatstructuresawidevarietyofotherwisecompetingviews,isthatthewayinwhich10
thesematerial,conventionalsymbol-structuresdotheirworkisbybeingtranslatedintosomekindofcontent-matchinginnercode.Onealternativetothisviewisthetemptingbutthoroughlyelusiveideathatwesomehowthinkinsomenaturallanguage(suchasEnglish).InthepresenttreatmentIexploreathirdoption,whichIshallcallthe‘‘complementarity’’viewoflanguage.Accordingtothisthirdviewtheactualsymbol15
structuresofagivenlanguageaddcognitivevaluebycomplementing(withoutbeingreplicatedby)themorebasicmodesofoperationandrepresentationendemictothebiologicalbrain.The‘‘cognitivebonus’’thatlanguagebringsis,onthismodel,nottobecashedouteitherviatheultimatelymysteriousnotionof‘‘thinkinginagivennaturallanguage’’orviasomeprocessofexhaustivetranslationintoanotherinnercode.Instead,20
weshouldtrytothinkintermsofakindofcoordinationdynamicsinwhichtheformsandstructuresofalanguagequamaterialsymbolsystemplayakeyandirreduciblerole.Understandinglanguageasacomplementarycognitiveresourceis,Iargue,animportantpartofthemuchlargerproject(sometimesglossedintermsofthe‘‘extendedmind’’)ofunderstandinghumancognitionasessentiallyandmultiplyhybrid:asinvolving25
acomplexinterplaybetweeninternalbiologicalresourcesandexternalnon-biologicalresources.Keywords:222
1.TranslationModelsofLanguage
30
JerryFodorfamouslyholdsthat‘‘knowinganaturallanguageisknowinghowtopairitsexpressionswithMentaleseexpressions’’(Fodor,1998,p.67).Tohaveacertain
Correspondenceto:AndyClark,DepartmentofPhilosophy,GeorgeSquare,EdinburghEH89JX,Scotland,UK.Email:andy.clark@ed.ac.uk
ISSN0951-5089(print)/ISSN1465-394X(online)/06/030001-17ß2006Taylor&FrancisDOI:10.1080/09515080600689872
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
35
40
45
50
55
60
65
70
75
2A.Clark
thought,onthisview,istotokenacertainmentalesesentence.Languageimpactsthought,onsuchaccounts,invirtueofaprocessoftranslationthattransformsthepublicsentenceintoacontent-capturinginnercode.Thisistheprimeexampleofwhatmightbedubbeda‘‘translationviewoflanguage.’’Encounteredlanguage(beitspeechorthewrittenword),ifthisviewiscorrect,merelyservestoactivatecomplexesofinternalstatesorrepresentationsthataretherealcognitiveworkhorses.Itturnsuptoo,thoughwitharadicallydifferenttwist,inPaulChurchland’sconnectionist-inspiredvisionofhumancognition.ForChurchland(1989,p.18;1996,p.107)publiclanguageoffersonlywhatmightbedubbed‘‘thintranslations’’ofthemuchrichermeaningsmadeavailablebyvectorcodingsandhigh-dimensionalstatespaces.Publicwordsandsentences,Churchlandsuggests,offeratbestashallowor‘‘one-dimensional’’(1989,p.18)echooftherichandsupra-linguisticmeaningsencodedusingtheformidableresourcesofthesehigh-dimensionalstate-spaceencodings.Butdespitedisagreeingovertheprecise‘‘fit’’(excellentversusdisappointinglysparse)betweenthepublicstructuresandtheinnerrealm,Churchland,likeFodor,retainswhatisessentiallyatranslationviewofhowpubliclanguageworks.Itisjustthatthecontentsoftheinternaltranslations(thecontentspropertotherealcognitiveworkhorses)forChurchlandtypicallyexceed,ratherthansimplyreplicate,thoseofthepubliclanguagestructuresthemselves.Thusinsofaraspubliclanguageisausefultoolatall,itworks,accordingtoChurchland,byactivatingoneormoresuitesofrichinternal(connectionist)representations—onemightdubthem‘‘neuralese’’—thatthenencodethemeaning.Theactualpubliclanguageitemsarethusonceagainmerescaffoldingtobekickedawayoncecontenthas,howeverimperfectly,beentransmittedfrompersontoperson.
Accordingtothetranslationpicture,then,languageworksitsmagicbybeingunderstood,andunderstandingisinturnconceivedasconsistingwhollyinsomethingliketranslationintosomeothercontent-matching(orcontent-exceeding)format.Suchaviewdepictslanguageasakindofhigh-levelcodethatneedstobecompiledorinterpreted(inthecomputersciencesense)todoitswork.Asaresult,thematerialformscanthenbethrownawayastheessence—themeaningscarried,conveyed,implied—hasbeenfullyextractedandrenderedinsomealternativeinnerformat.
Comparenowtheuseofastandardtool.WhenIuseaspadetodigthegarden,thespademakesanongoingandcomplementarycontributiontothatmadebymybiologicalbody.Thereis,insuchacase,noobvioussenseinwhichIbiologicallyreplicatetheessenceofthespade’sactivity.Instead,thediggingpowerresidesinthelargercoupledsystem.
Thealternativetothetranslationpicture,thatIwishtopursuehere,makestheroleofpubliclanguagemorelikethatofthespade.Ontheviewtobeexplored,language(andmaterialsymbolsmoregenerally)playadoublerole.Ontheonehandtheydo(crucially,always)activateotherkindsofcognitiveresource,bitsofmentaleseorneuraleseasyouprefer.Buttheyalsoplayanirreducibleroleasthematerialsymbolstheyare.Forpartoftherole(andthepower)ofsuchitems(spokenorwrittenwordsandsentences)istocomplementthebasicmodesofoperationandrepresentation
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
80
85
90
95
100
105
110
115
PhilosophicalPsychology3
endemictothebiologicalbrain.Understandinglanguage,ifthisviewiscorrect,involvesgettingtogripswithaspecialkindofcoordinationdynamics:oneinwhichtheactualmaterialstructuresofpubliclanguage(orsometimestheirshallow‘‘imagistic’’internalrepresentations)playakeyandirreduciblerole.Thisview,asIshalldevelopit,isrelatedto,butIthinkremainsdistinctfrom,Dennett’sfamous(1991)accountoflanguageasinstallinganewserialvirtualmachineinthehead.ForwhereasDennettdepictsexperiencewithlanguageasessentiallytransformative,aschangingthefundamentalnatureof(partof)thein-headmachinery,Ishallattempttodepictpubliclanguageasacomplementaryresourcethatworkswiththemorebasicmachinerywithoutinstallinganyfundamentallynewstylesofrepresentationorprocessingwithinthatmachinery.Thisviewoflanguage,Ishallfinallysuggest,canusefullybeseenaspartofthemuchlargerproject(sometimesglossedintermsofthe‘‘extendedmind’’—seeClark,1997;Clark&Chalmers,1998)ofunderstandinghumancognitionasessentiallyandmultiplyhybrid:asinvolvingacomplexinterplaybetweeninternalbiologicalresourcesandexternalnon-biologicalresources.Language,however,occupiesawonderfullyambiguouspositiononanyhybridcognitivestage,sinceitseemstostraddletheinternal-externalborderlineitself,lookingonemomentlikeanyotherpieceofthebiologicalequipment,andatthenextlikeapeculiarlypotentpieceofexternalcognitivescaffolding.2.SomeTrialCases
Itwillbehelpfultoputarangeofconcretecasesonthetableasakindof(somewhathopeful)anchorforthesubsequentdiscussion.Theexamplesthatfollowmaybefamiliar,butIaskthereader’spatience.Itisnotthecasesthemselvesthatmatter,somuchasthegeneralpattern,displayingsomeoftheinterlinkedvarietyofwaysthattheactualmaterialformsoflanguagemayimpactcognition.Thecasesthatfollowarearrangedin(whatseemstometobe)ascendingorderofcognitiveimpact.2.1.FirstGradeofCognitiveInvolvement:LanguageasaSourceofAdditionalTargetsforAttentionandLearning
Therearethreeexamplesfallingintothiscategory.Thefirst,andbyfarthesimplest,isthewell-knowncaseofShebaandthetreats,asrecountedinBoysen,Bernston,HannanandCacioppo(1996).Sheba(anadultfemalechimpanzee)hashadsymbolandnumeraltraining:sheknowsaboutnumerals.ShebasitswithSarah(anotherchimp),andtwoplatesoftreatsareshown.WhatShebapointsto,Sarahgets.Shebapointstothegreaterpile,thusgettingless.Shevisiblyhatesthisresult,but(unlesstherewardmatrixisgreatlyexaggerated)can’tseemtoimprove.However,whenthetreatsarriveincontainerswithacoverbearingnumeralsontop,thespellisbrokenandShebapointstothelessernumber,thusgainingmoretreats.
Whatseemstobegoingonhere,accordingtoBoysenetal.,isthatthematerialsymbols,bybeingsimpleandstrippedofmosttreat-signifyingphysicalcues,allowthechimpstosidestepthecaptureoftheirownbehaviorbyecologically-specific
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
120
125
130
135
140
145
150
155
4A.Clark
fast-and-frugalsubroutines.Thesymbolloosensthebondbetweenagentandworld,andbetweenperceptionandaction,anditdoessonotinvirtueofbeingthekeytoarichmentalrepresentation(thoughitmaybethattoo)butratherbyitself,quamaterialsymbol,providinganewtargetforselectiveattentionandanewfulcrumforthecontrolofaction.
Inmuchthesamewaytheactoflabelingcreatesanewrealmofperceptibleobjectsuponwhichtotargetbasiccapacitiesofstatisticalandassociativelearning.Theactoflabelingthusaltersthecomputationalburdensimposedbycertainkindsofproblem.Ihavewrittenquiteabitonthiselsewhere,soI’llkeepthisbrief.Myfavoriteexample
(Clark,1998)beginswiththeuse,byotherwiselanguage-naı
¨vechimpanzees,ofconcretetags(simpleanddistinctplasticshapes)forrelationssuchassamenessanddifference.Thus,apairsuchascup–cupmightbeassociatedwitharedtriangle(sameness)andcup–shoewithabluecircle(difference).Thisisnotinitselfsurprising.Whatismoreinterestingisthatafterthistraining,thetag-trainedchimps(andonlytag-trainedchimps)proveabletolearnabouttheabstractpropertiesofhigher-ordersameness,i.e.theyareabletolearntojudgeoftwopresentedpairs(suchascup–cupandcup–shoe)thattherelationbetweentherelationsisoneofhigherorderdifference(orbetter,lackofhigher-ordersameness)sincethefirstpairexhibitsthesamenessrelationandthesecondpairthedifferencerelation(Thompson,Oden,&Boysen,1997).Thereasonthetag-trainedchimpscanperformthissurprisingfeatis,sotheauthorssuggest,becausebymentallyrecallingthetagsthechimpscanreducethehigher-orderproblemtoalower-orderone:alltheyhavetodoisspotthattherelationofdifferencedescribesthepairingofthetworecalledtags(redtriangleandbluecircle).Thelearningmadepossiblethroughtheinitialloopintotheworldofstable,perceptibleplastictokenshasallowedthebraintobuildcircuitsthat,perhapsbysimplyimagingthetokensthemselvesatappropriatemoments,reducethehigher-orderproblemtoalower-orderoneofakindtheirbrainsarealreadycapableofsolving.Experiencewithexternaltagsandlabelsthusenablesthebrainitself—byshallowlyrepresentingthosetagsandlabels—tosolveproblemswhoselevelofcomplexityandabstractionwouldotherwiseleaveusbaffled.1Arelatedeffectmayalsobeobserved(andthisisourthirdandfinalcaseinthiscategory)inrecentconnectionistworkonlanguagelearning.Thusinarecentreview,SmithandGasser(2005)askaverynicequestion.Why,giventhathumanbeingsaresuchexpertsatgrounded,concrete,sensorimotordrivenformsoflearning,dothesymbolsystemsofpubliclanguagetakethespecialandratherrarifiedformsthattheydo?
Onemightexpectthatamultimodal,grounded,sensorimotorsortoflearningwouldfavoramoreiconic,pantomime-likelanguageinwhichsymbolsweresimilartoreferents.Butlanguageisdecidedlynotlikethis...thereisnointrinsicsimilaritybetweenthesoundsofmostwordsandtheirreferents:theformoftheworddoggivesusnohintsaboutthekindofthingtowhichitrefers.Andnothinginthesimilarityoftheformsofdiganddogconveysasimilarityinmeaning.(Smith&Gasser,2005,p.22)
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
PhilosophicalPsychology
160
5
165
Thequestion,inshort,is‘‘Whyinasoprofoundlymultimodalsensorimotoragentsuchasourselvesislanguageanarbitrarysymbolsystem?’’(p.24).
Onepossibleanswer,ofcourse,isthatlanguageislikethatbecause(biologicallybasic)thoughtislikethat,andtheformsandstructuresoflanguagereflectthisfact.Butanotheranswer,andtheoneIwanttopursue,saysjusttheopposite.Languageislikethat,itmightbesuggested,becausethought(orrather,biologicallybasic170
175
180
185
190
195
200
thought)isnotlikethat.Thecomputationalvalueofapublicsystemofessentiallycontext-free,arbitrarysymbols,lies,accordingtothisopposingview,inthewaysuchasystemcanpush,pull,tweak,cajoleandeventuallycooperatewithvariousnon-arbitrary,modality-rich,context-sensitiveformsofbiologicallybasicencoding.Consider,totakethemaincasepresentedbyGasserandSmith,thedevelopmentofone-trialwordlearning.Thispowerfulcapacitymaybemultiplydependent,SmithandGassersuggest,onthepresenceofapubliccodecomprisingarbitrarylabels.Earlywordlearning,theysuggest,isallaboutbuildingupmultimodalclustersofassociatedproperties.Butlateron,asiswellknown,childrenbecomerapidwordlearners,addingfourtoninenewwordsaday,andgeneralizingthenewwordsinwaysappropriatetotheirdistinctcategories.Suchrapid-firelearninglookstorequirethedeploymentofwhatSmithandGasserdescribeas‘‘second-order,rule-likegeneralizations.’’Suchgeneralizations,theyargue,aredrivenbypropertiesofarbitrarypublicsymbolsystems.
Forexample,anewwordforanartifactwillprobablyapplytosimilarlyshapedthings(thinkoftractors,fryingpans,toothbrushes).Whereasanewwordforasubstancewillapplytootherthingsmadeofthesamematerial(woolyhats,woolyjumpers,woolymittensetc.).Rapidwordlearninglookstoinvolvejustsuchabilitiesofhigher-ordergeneralization.NeuralnetworksimulationsbyElianaColunga(Colunga&Smith,2005)suggestthattheformationofsuchsecond-ordergeneralizationsdependsonthearbitrarinessandorthogonalityofthelinguisticlabelsprovided.Makethelabelsnon-orthogonal,andthesecond-orderknowledgeisnotacquired(non-arbitrarylabelsmusttendtowardsnon-orthogonalityduetopropertyoverlapsintheobjectsandeventslabeled).
Itisnotfullyclearwhythisshouldbeso,butitseemslikelythatexperiencewithconcreteorthogonallabelshelpsthesystemtopullperceptuallysimilarcategoriesapart,andthussupportsnewkindsofgroupingthatmakevisibledeepercommonalitiesanddifferences,yieldingthekindsofimplicitknowledge(e.g.,concerningthetypicalkindsoffeaturethatindividuateartifactsratherthansubstances)thatunderpinrapid-firelearningandthatwouldotherwisebeburiedtoodeepinthesearchspaceforbasicsensorimotorformsofintelligencelikeourselves.
2.2.SecondGradeofCognitiveInvolvement:LanguageasaResourceforDirectingandMaintainingAttentiononComplexConjoinedCues
Thekeycaseinthiscategoryconcernsspatialreasoningininfantsandadults.InafamousstudybyHermer-Vazquez,Spelke,andKatsnelson(1999),pre-linguistic
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
205
210
215
220
225
230
235
240
6A.Clark
infantswereshownthelocationofatoyorfoodinaroom,thenwerespunaroundorotherwisedisorientedandrequiredtotrytofindthedesireditem.Thelocationwasuniquelydeterminableonlybyrememberingconjoinedcuesconcerningthecolorofthewallanditsgeometry(thetoymightbehiddeninthecornerbetweenthelongwallandtheshortbluewall).Theroomsweredesignedsothatthegeometricorcolorcueswereindividuallyinsufficient,andwouldyieldanunambiguousresultonlywhencombinedtogether.Pre-linguisticinfants,thoughperfectlyabletodetectandusebothkindsofcue,wereshowntoexploitonlythegeometricinformation,searchingrandomlyineachofthetwogeometricallyindistinguishablesites.Yetadultsandolderchildrenwereeasilycapableofcombiningthegeometricandnon-geometriccuestosolvetheproblem.Importantly,successatcombiningthecueswasnotpredictedbyanymeasureofthechildren’sintelligenceordevelopmentalstageexceptforthechild’suseoflanguage.Onlychildrenwhowereabletospontaneouslyconjoinspatialand(e.g.)colortermsintheirfreespeech(whowoulddescribesomethingas,say,totherightofthelonggreenwall)wereabletosolvetheproblem.
Hermer-Vazquezetal.(1999)thenprobedtheroleoflanguageinthistaskbyaskingsubjectstosolveproblemsrequiringtheintegrationofgeometricandnon-geometricinformationwhileperformingoneoftwoothertasks.Thefirsttaskinvolvedshadowing(repeatingback)speechplayedoverheadphones.Theotherinvolvedshadowing,withtheirhands,arhythmplayedovertheheadphones.Theworkingmemorydemandsofthelattertaskwereatleastasheavyasthoseoftheformer.Yetsubjectsengagedinspeechshadowingwereunabletosolvetheintegration-demandingproblem,whilethoseshadowingrhythmwereunaffected.Anagent’slinguisticabilities,theresearchersconcluded,areindeedactivelyinvolvedintheirabilitytosolveproblemsrequiringtheintegrationofgeometricandnon-geometricinformation.
Therearecurrentlyvariouscompetingmodelsofjusthowthisinvolvementisbestunpacked(seeespeciallyCarruthers,2002).Butprobablythesimpleststoryisthatheretoolinguisticresourcesprovideaconvenientfulcrumforthecomplexdistributionofattention.Theyenableusbettertocontrolthedispositionofselectiveattentiontoever-morecomplexfeaturecombinations.Theshadowingresultisthenexplainedbytheideathatactiveattentiontoacomplexconjoinedcuerequiresthe(possiblyunconscious)retrievalofatleastsomeoftherelevantlexicalitems.Layingtheemphasisonattentionaleffectsthusallowsustoaccommodatethiscaseinawaythatdovetailswiththeearlierones.Ineachcase,linguisticactivity(somekindofconsciousorunconsciousaccesstorepresentationsoflanguage-specificlexicalitems)helpsustotargetourattentionalresourcesoncomplex,conjunctive,orotherwiseelusive,elementsoftheencounteredscene.
2.3.ThirdGradeofCognitiveInvolvement:LanguageasProvidingSomeoftheProperPartsofHybridThoughts
Atlast,then,wearriveatthehighestgradeofcognitiveinvolvementIwanttoscout,andsurelythemostcontentious.Thisistheidea(tobeexplainedshortly)oflanguage
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
245
250
255
260
265
270
275
280
285
PhilosophicalPsychology
7
asprovidingsomeoftheproperpartsofhybridthoughts.Thekeyexamplehereconcernstheroleofnumberwordsinmathematicalreason.
Whatisgoingonwhenyouthinkthethoughtthat‘‘98isonemorethan97’’?Accordingtothetranslation-basedmodel,tothinkthatthoughtistotranslatetheEnglishsentenceintosomethingelse,wherethatsomethingelsemightbeasentenceofmentalese(forFodor)orapointinsomeexotichigh-dimensionalstatespace(forChurchland).
ButconsiderarecentaccountduetoStanislasDehaeneandcolleagues(seeDehaene,1997;Dehaene,Spelke,Pinel,Stanescu,&Tviskin,1999).Dehaenedepictsthiskindofprecisemathematicalthoughtasemergingattheproductiveintersectionofthreedistinctcognitivecontributions.Thefirstinvolvesabasicbiologicalcapacitytoindividuatesmallquantities:1-ness,2-ness,3-nessandmore-then-that-ness,totakethestandardset.Thesecondinvolvesanotherbiologicallybasiccapacity,thistimeforapproximatereasoningconcerningmagnitudes(discriminating,say,arraysof8dotsfromarraysof16,butnotmorecloselymatchedarrays).Thethird,notbiologicallybasicbutarguablytransformative,isthelearntcapacitytousethespecificnumberwordsofalanguage,andtheeventualappreciationthateachsuchnumberwordnamesadistinctquantity.Noticethatthisisnotthesameasappreciating,inatleastoneimportantsense,justwhatthatquantityis.Mostofuscan’tformanyclearimageof,e.g.,of98-ness(unlike,say,2-ness).Butweappreciatenonethelessthatthenumberword‘98’namesauniquequantityinbetween97and99.
Whenweaddtheuseofnumberwordstothemorebasicbiologicalnexus,Dehaeneargues,weacquireanevolutionarilynovelcapacitytothinkaboutanunlimitedsetofexactquantities.Wegainthiscapacitynotbecausewenowhaveanencodingof98-nessjustlikeourencodingof2-ness.Rather,thenewthoughtsdependdirectly(butnotexhaustively)uponourtokeningthenumericalexpressionsthemselves,assymbolstringsofourownpubliclanguage.Theactualnumericalthought,onthismodel,ishadcourtesyofthecombinationofthistokening(ofthesymbolstringofagivenlanguage)andtheappropriateactivationofthemorebiologicallybasicresourcesmentionedearlier.
Hereissomeevidenceforthisview,aspresentedinDehaeneetal.(1999).First,therearetheresultsofstudiesofRussian-Englishbilinguals.Inthesestudies,Russian-Englishbilingualsweretrained(quiteextensively)on12casesinvolvingexactandapproximatesumsof(thesame)pairsoftwo-digitnumbers,presentedaswordsinoneorotherlanguage.Forexample,(inEnglish),asubjectmightbetrainedonthequestion‘‘FourþFive’’andaskedtoselecttheiranswerfrom‘‘Nine’’and‘‘Seven’’.Thisiscalledtheexactcondition,asitrequiresexactreasoningsincethetwocandidatenumbersareclosetoeachother.Bycontrast,aquestionlike‘‘‘FourþFive’,selectanswerfrom‘Eight’and‘Three’’’belongstotheapproximatecondition,asitrequiresonlyroughreasoningasthecandidatesarenowquitefarapart.
Afterextensivetrainingonthepairs,subjectswerelatertestedontheverysamesumsineithertheoriginalortheother(non-trained)language.Aftertraining,performanceintheapproximationconditionwasshowntobeunaffectedbyswitchingthelanguage,whereasintheexactcondition,languageswitchingresulted
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
290
295
300
305
310
315
320
325
8A.Clark
inasymmetricperformance,withsubjectsrespondingmuchfasterifthetest-languagecorrespondedtothetraining-language.Crucially,then,therewerenoswitchingcostsatallfortrainedapproximatesums.Performancewasthesameregardlessoflanguageswitching.Training-basedspeedupisthusnon-languageswitchablefortheexactsumsandfullyswitchablefortheinexactones.Suchstudies,Dehaeneetal.concluded,provide:
evidencethatthearithmeticknowledgeacquiredduringtrainingwithexactproblemswasstoredinalanguage-specificformat....Forapproximateaddition,incontrast,performancewasequivalentinthetwolanguagesprovidingevidencethattheknowledgewasstoredinalanguage-independentform.(1999,p.973)
Asecondlineofevidencedrawsonlesionstudiesinwhich(totakeoneexample)apatientwithsevereleft-hemispheredamagecannotdeterminewhether2þ2is3or4,butreliablychooses3or4over9,indicatingasparingoftheapproximationsystem.Finally,Dehaeneetal.(1999)presentneuroimagingdatafromsubjectsengagedinexactandapproximatenumericaltasks.Theexacttasksshowsignificantactivityinthespeed-relatedareasoftheleftfrontallobe,whiletheapproximatetasksrecruitbilateralareasoftheparietallobesimplicatedinvisuo-spatialreasoning.Theseresultsarepresentedasademonstration‘‘thatexactcalculationislanguagedependent,whereasapproximationreliesonnonverbalvisuo-spatialcerebralnetworks’’(p.970)andthat‘‘evenwithinthesmalldomainofelementaryarithmetic,multiplementalrepresentationsareusedfordifferenttasks’’(p.973).
Dehaene(1997)alsomakessomenicepointsabouttheneedtosomehowestablishlinksbetweenthelinguisticlabelsandourinnatesenseofsimplequantities.Atfirst,itseems,childrenlearnlanguage-basednumericalfactswithoutsuchappreciation.AccordingtoDehaene,‘‘forawholeyear,childrenrealizethattheword‘three’isanumberwithoutknowingtheprecisevalueitrefersto’’(1997,p.107).Butoncethelabelgetsattachedtothesimpleinnatenumberline,thedoorisopentounderstandingthatallnumbersrefertoprecisequantities,evenwhenwelacktheintuitivesenseofwhatthequantityis(e.g.myownintuitivesenseof53-nessisnotdistinctfrommyintuitivesenseof52-ness,thoughallsuchresultsarevariableaccordingtothelevelofmathematicalexpertiseofthesubject).
Typicalhumanmathematicalcompetence,allthissuggest,isplausiblyseenasakindofhybrid,whoseelementsinclude:
(i)Imagesorencodingsofactualwordsinaspecificlanguage;
(ii)anappreciationofthefactthateachdistinctnumberwordnamesaspecificand
distinctquantity;and
(iii)aroughappreciationofwherethatquantityliesonakindofapproximate,
analognumberline(e.g.98isjustlessthanhalfwaybetween1and200).Inacertainsensethen,werelyonthecoordinatedactionofvariousresources.Onthisview,thereis(atleast)aninternalrepresentationofthenumeral,oftheword-form,andofthephonetics,alongwithotherresources(suchastheanalognumberline)towhichthesebecome(withlearning)roughlykeyedviasomesense
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
330
335
340
345
350
355
360
365
370
PhilosophicalPsychology
9
ofrelativelocation.Whatmattersforpresentpurposes(forwhatIamcallingthethirdgradeofcognitiveinvolvement)isthattheremaybenoneedtoposit(fortheaverageagent),inadditiontothiscoordinatedmedley,anyfurthercontent-matchinginternalrepresentationof,say,98-ness.Instead,thepresenceofactualnumberwordsinapubliccode(andofshallow,imagistic,internalrepresentationsofthoseverypublicitems)isitselfpartofthecoordinatedrepresentationalmedleythatconstitutesmanykindsofarithmeticalknowing.
Thusconsiderthethoughtthatthereare98toysonthetable.Accordingtothetranslationview,tothinkthethoughtthatthereare98toysonthetableyoumusthavesucceededintranslatingtheEnglishsentenceintoafullycontent-providing‘‘somethingelse.’’The‘‘somethingelse’’mightbeanatomorsentenceofmentalese(forFodor)orapointinsomeexoticstatespace(forChurchland).Bycontrast,accordingtothisquiteradicalalternative,thethoughtthatthereare98toysonthetableis(formostofus)dependentuponthepresenceofahybridrepresentationalvehicle.2Thisisavehiclethatincludes,asexpected,theactivationofavarietyofcontent-relevantinternalrepresentations(inneuraleseormentalese,let’sassume).Butitalsoincludesasaco-optedproperpart,atoken(let’sthinkofitasanimage,verybroadlyconstrued)ofaconventionalpubliclanguageencoding(‘‘ninety-eight’’)appropriatelylinkedtovariousotherresources(suchassomeroughpositiononananalognumberline).
Thishalf-glimpsedpossibilityis,Isuspect,actuallythemostimportantwaythatlanguage(andindeedallkindsofculturalpropsandartifacts)mayimpactthought:byactuallybecomingpartsofthethinkingsthemselves.Thisisnot,asyouwillhavenoticed,themosttransparentofideas,andIdoubtIhaveitevenhalfwayright.Butthescopeissatisfyinglylarge.Inthecaseathand,thevehicleorprocess,thougharguablygenuinelyhybrid,isfullyinternaltothebiologicalagent.Butinothercases,thereseemsnoreasontoinsistthatthismatters.Perhapssomeofourrepresentationalvehiclesandprocesses(theactualmechanisticunderpinningsofourthinkingsnoless)maygetspreadoutacrossbiologicalbrainsandallsortsofsocio-culturalartifacts,includinggestures,diagrams,externaltext,softwareapplications,andmore.
Thisviewoflanguageisaperfectfitwith(thoughnot,Isuppose,essentialfor)averybigpictureaccordingtowhichhumancognitiongainsmuchofitsdistinctiveforceandpowerfromits(biologically-based)abilitytobuildandmaintainnewformsofexternalrepresentationalstructure,thatarethenaptfornon-fully-replicateduse,inotherwordsforcognitiveincorporation.3Thatistosay,wemakeourselvesintonewkindsofcognitiveengineby(amongstotherthings)annexingandco-optingelementsofexternalcognitivescaffoldingasproperpartsofhybridcomputationalroutines.Inthiscontext,itisworthobserving(thoughonlyasakindofcodatothemainstory)thatthe(putative)abilityofmaterialsymbolstoparticipateincognitiveprocesseshelpsshowthewayoutofadilemmaoftenurgeduponthefriendsofextendedcognition.Forourverybestcognitiveartifacts,iftheyaresometimestoplayaroleasproperpartsofcognitiveprocesses,needtobeassimilatedwithinbutnottotallyswallowedupbytheworkingsofthepotentbasic
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
375
380
385
390
395
400
405
410
10A.Clark
biologicalcognitiveengineitself.Thusitisverytempting,whenconfrontedwithargumentsthatwouldgiveastrongcognitiveroletoartifacts(or,inthiscase,topublic,conventional,symboliccodes)torespondwithakindofdilemma.Eithertheartifact/publiccodeisnotplayingatrulycognitiverole(itismerelyinput,notpartoftheprocessing)or(insofarasitseemstobeplayingsucharole)itdoessoonlybecauseithasbeentranslatedintosomethingelse,somequitedifferentinnerthing,thatreallyissuitedtoplaysucharole.Eitherway,itseems,thebenefitsthataccruecanbefullyexplained,atleastasfarashereandnowthinkingisconcerned,withoutcontinuedreferencetothefeaturesandpropertiesoftheartifact/publiccode.4(Forsomeversionsofthisdilemmaintheliteratureopposingthe‘‘extendedmind,’’seeAdams&Aizawa,2001;Rupert,2004).
Thewayaroundthedilemmashouldnowbeclear.Bystressingcoordinationdynamicsandhybridrepresentationalforms,weleaveroomforgenuinecomplementaritybetweenthebiologicalandartifactualcognitivecontributions.Wethusbegintoseehowartifactualresourcesmaysometimesbeco-optedwithoutbeingfullyrecapitulatedbythebiologicalelements.Thisiswhatthenotionofhybriditywasalwaysmeanttosuggest,anditavoidsbothhornsofthedilemma.Onebadreasonwhythiscanseemimpossibleinthecaseoflanguageis,ofcourse,ifwestillthinkthatunderstanding‘‘obviously’’alwaysrequirestranslationintosomeothercontent-matching(orbetter)innercode.Butitisprettyclearthatthiscannotbethecaseallthewaydown,onpain(seeFodor,1975)ofanendlessregressofsuchcodes.Sooncetherightcoordinationdynamicsareinplace,thereisnoreasonwhysomehybridwholecouldnotitselfbethephysicalvehicle,appropriatelypoisedtocontrolactionandchoice,oftherelevantunderstanding.Indeed,thewholeofartificialintelligenceissurelyitselftestimonytothepoweroftheideathatwell-poisedphysicallyinstantiatedrepresentationscansometimesconstituteunder-standingwithoutneedingtobe(inanyfurtherway)understoodthemselves.
3.HybridThoughts?
Theideaonofferthenisthatthesymbolicenvironment(verybroadlyconstrued)cansometimesimpactthoughtandlearningnotbysomeprocessoffull-translation,inwhichthemeaningsofsymbolicobjectsareexhaustivelytranslatedintoaninnercode,amentalese,orevenaChurchland-styleneuralese,butbysomethingclosertocoordination.Onthecoordinationmodel,thesymbolicenvironmentimpactsthoughtbothbyactivatingsuchotherresources(theusualsuspects)andbyusingeitherthesymbolicobjectsthemselves(orinnerimage-likeinternalrepresentationsoftheobjects)asadditionalfulcrumsofattention,memoryandcontrol.Inthemaximumstrengthversion,thesesymbolicobjectsquiteliterallyappearaselementsinrepresentationallyhybridthoughts.5Nowforaconfession.Forquiteafewyears,Ithoughtthiswasaradicalideathatfansof(totakethemostextremeexample)thelanguageofthought(LOT)hypothesiswouldsurelyrejectoutofhand.Theiridea,afterall,wasthatwordsmeanwhatthey
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
415
420
425
430
435
440
445
450
455
PhilosophicalPsychology
11
doinvirtueofbeingpairedwithexpressivelyparallelsnippetsofmentalese.Imaginemysurprisethen,whenIfoundthislittlesnippethiddenawayinthat1998reviewofCarruthersbyJerryFodor:
Idon’tthinkthattherearedecisiveargumentsforthetheorythatallthoughtisinMentalese.Infact,Idon’tthinkit’seventrue,inanydetail....Iwouldn’tbeintheleastsurprised,forexample,ifitturnedoutthatsomearithmeticthinkingiscarriedoutbyexecutingpreviouslymemorizedalgorithmsthataredefinedoverpubliclanguagesymbolsfornumbers(‘‘nowcarrythe‘2’’’andsoforth).It’squitelikelythatMentaleseco-optsbitsofnaturallanguageinallsortsofways;quitelikelythestoryabouthowitdoessowillbeverycomplicatedindeedbythetimethatthepsychologistsgetfinishedtellingit.(1998,p.72,italicsinoriginal)
Fodorheregestures,itseemstome,atanincrediblypotentmechanismofcognitiveexpansion.Prettyclearlythough,Fodorhimselfattacheslittleimportancetotheconcession,quicklyaddingthat‘‘Forallourphilosophicalpurposes(e.g.forpurposesofunderstandingwhatthoughtcontentis,andwhatconceptpossessionis,andsoforth)nothingessentialislostifyouassumethatallthoughtisinMentalese’’(1998,p.72,italicsadded).
Bycontrast,Iaminclinedtoseethepotentialforrepresentationalhybridityasmassivelyimportanttounderstandingthenatureandpowerofmuchdistinctivelyhumancognition.OneobviousreasonforthisdifferenceinassessmentisthatFodorhastheLOTalreadyinplace.Sothebasicbiologicalengine,onhisaccount,comesfactory-primedwithinnovationsfavoringstructure,integration,generalityandcompositionality.If,however,yourvisionofthebasicbiologicalengineisnotonethatsocloselyechoesthepropertiesandfeaturesofsentencesandpropositionalattitudes(if,forexample,itisclosertoChurchland’svisionofacomplexbutthoroughlyconnectionistdevice,ortoBarsalou’s,1999,visionofa‘‘perceptualsymbolsystem’’)thenthepotentialcognitiveimpactofalittlehybridityandco-optingmaybemuchgreaterthanFodorconcedes.Itmaybeessentialtosuchasystem’sabilitytothinkratherawidevarietyofthoughtsthattheinnergoings-oninvolve,asgenuinelyconstitutiveelements,somethinglikeimagesortracesofthepubliclanguagesymbols(words)themselves.Wordsandsentences,onthisview,maybepotentstructuresmanyofwhosefeaturesandproperties(arbitraryamodalnature,extremecompactnessandabstraction,compositionalstructure,andsoon)deeplycomplementthecontributionsofbasicbiologicalcognition.Insuchacase,itwouldhardlyberighttotreattheco-optingstrategiesasmarginalfortheunderstandingofthoughtandconcepts.6Thisvisionofmind-expansionbytheuseofhybridrepresentationalformsremainsvisiblyclosetothatofDennett(1991,1996).ButDennett,asmentionedearlier,placesmostofhisbetsontheradicallytransformativepowerofourencounterswithlanguage,andthusendsupwithastorythatseemsmoredevelopmentalthangenuinelyhybrid.Admittedly,drawingtheselinesisadelicatetask(Densmore&Dennett,1999).ButwhereDennettdepictsexposuretolanguageasinstallinganewvirtualserialmachineviaaffecting‘‘myriadmicrosettingsintheplasticityofthebrain’’(1991,p.219),onthehybridmodelwordsandsentences
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
460
465
470
475
480
485
490
495
12A.Clark
remainpotentreal-worldstructuresencounteredandusedbyabasically(thoughthisisobviouslytoocrude)pattern-completingbrain.Ofcourse,evenonthisaccountthebrainsometimesrepresents(shallowly,imagistically)thesestructures.Butlanguageneednotprofoundlyreorganize7theshapeandtextureoftheneuralcodingroutinesthemselves.84.WorkingModels?
Theideaoftrulyhybrid,bio-artifactuallydistributedcognitionis,Ihopetohaveshown,atleastintelligible.Moreover,theexamplesarrayedinx2aremeanttosuggestthatitisalsoactual.Buthow,indetail,mightthewholethingwork?Dowehaveevenasingleexistenceproof,intheformofanupandrunningsimulation,thatshowshowsuchhybriditymightbemechanicallyimplemented?
ThenearestIhavesofarfoundisasmallbutsuggestivesetofsimulationsreportedinClowes&Morse(2005).Thesimulationsinvestigatewaysinwhichtheinternalre-useofapublicsymbolsystemmightaidcognition.Internalre-usewasenabledbytheprovision,insomeagents,ofadedicatedre-entrantloopabletorecycle‘‘heard’’linguisticinputsduringprocessing.Inthesimulations,simpleagentswereevolvedtofindandmovegeometricfiguresinresponsetocommandscouchedina‘‘public’’code.Thecommandstelltheagent’s(simplerecurrentneuralnetswith‘‘visual’’and‘‘word’’inputs)whichoffourtaskstoperformonobjectsinanon-screenarena.Thetasksaretomovetheobjectstothetop(‘‘up’’),tomovetheobjectstothebottom(‘‘down’’),tomovetheobjectstotheright(‘‘right’’)ortomovetheobjectstotheleft(‘‘left’’).
Groupsofagentswereevolvedunderthreeconditions:
1.Acontrolcondition,withnodedicatedwordre-entranceloop.Inthisconditiontheagent‘‘hears’’wordsascommandsandmustactonthatbasisalone(butthearchitectureisstillthatofasimplerecurrentneuralnet(SRNN),sothereismemoryavailableastheoutputlayercyclesbacktotheinputlayeralongsidenewinputsatthenexttimestep).
2.PermanentWordRe-entrance:Inthiscondition,the‘‘heard’’commandwordsarecycledbackviaadedicatedpartofarecurrentloopwhileproblemsolvingcontinues.
3.Self-controlledRe-entrance:Thisisas(2)exceptthenethasanadditionaloutputunitthatcangatethededicatedwordre-entrancelooponandoff.‘‘Heard’’wordscanthusberecycledduringprocessingattheagent’sdiscretion.ClowesandMorsefoundthatunderthecontrolcondition(nodedicatedwordre-entrance)theagentstakelongertolearntosucceedatanyofthetasks,andseemunabletolearntosucceedatallfour.Thisisbecauseimprovementsinonetaskseemedtoalwaysresultinimpairmenttoperformanceononeormoreoftheothers.Thenetswithpermanentwordre-entrance(condition2)faredbetter.Goodperformancewasquiterapidlyevolved,andtypicallydisplayedinatleastthree
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
500
505
510
515
520
525
530
535
PhilosophicalPsychology
13
andoftenallfourtasks.Mostimpressiveofall,however,werethe(condition3)netswithself-gateablewordre-entrance.Theseagentsproducedthebestperformance,onalltasks,andwiththeleastevolutionarycosts(intermsofnumbersofgenerationsrequiredforcompetence).Overall,theauthorsconclude,‘‘[the]resultsclearlydemonstrateaqualitativedifferencebetweenthecontrolgroupandthe[wordre-entrant]conditions,despitetheinternalre-entranceofSRNNarchitecturespresentinallthreeconditions’’(Clowes&Morse,2005,p.104).
Underlyingthisresult,Iwouldfinallyconjecture,maybesomethingquitefundamental.Perhaps(butbeware:thisisnowpurespeculation)theroleofre-presentations(imagisticrecyclings)ofwordsherecanbeunderstoodasanexampleofthepoweroflooselycoupleddistinctprocesses.Thisisaneffectalreadyobservedinworkonso-calledGasNetsinwhichthecombinationof(asimulationof)freelydiffusinggaseousneurotransmittersandofamorestandard‘‘electrical’’artificialneuralnetworklearningresourcelikewiseimprovesperformanceandspeedsevolvability.Toexplainthisresult,Philipides,Husbands,Smith,andO’Shea(2005)suggestthatwhenanorganismmustaccommodateconflictingpressures(justasinthefour‘‘contradictory’’tasksconfrontingtheClowes-Morsenet)thepresenceofdistinctbutlooselycoupledprocesses‘‘allowsthepossibilityoftuningoneprocessagainsttheotherwithoutdestructiveinterference’’(p.154).Perhapsthenpartoftheroleofrehearsedwordsinaidingcognition,evenontheveryshorttime-scalesofongoingepisodesofthinking,mightonedaybeseenasanotherinstanceofthemoregeneralpowerofloosecouplingsbetweendynamicallydistinctprocesses.Perhaps,thatistosay,wordsarejustanespeciallypotentresourceabletoenterintolooselycoupledformsofonlineactivity,allowingthesystemtofindvaluabletrajectoriesthroughsearchspacethatmightotherwisebeblockedbydestructiveinterferencebetweensuperficiallyconflictingcurrentideas,goals,orcontexts.Forthistooccur,ongoingcontroloverthecurrentdegreeofcoupling,asinthe‘‘gated’’self-cueingnet,maywellbecrucial(again,seePhilippides,Husbands,Smith,O’Shea,2005,p.158).Allthisis,torepeat,purespeculation.ButIdosuspectthattheseverygeneralkindsofconsideration,concerningsearch,dynamicsandcomplexsystems,willeventuallyproveverygermanetothegeneralprojectoftryingtounderstandtheadvantagesconferredbyvariousformsofhybridcognition.
5.Conclusions:LeapsandBoundaries
Sowhatisthe‘‘cognitivebonus’’thatlanguagebrings?Inthistreatment,Ihavebeguntoexploreoneoftheless-visitedregionsofthissurprisinglymysteriouslandscape:theregioninwhichthematerialstructuresoflanguageplayacognitiverolethatinsomewayactuallydependson,andexploits,thatverymateriality.Toevenglimpsethisregionweneedtolookbeyondaseeminglyinescapablemodelofhowlanguagemustdoitswork,themodelaccordingtowhichencounteredlinguistictokensactsolelyinvirtueofaprocessofexhaustivetranslationintosomeothercontent-matching(orexceeding)internalrepresentationalformat.Thiswaswhat
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
540
545
550
555
560
565
570
575
580
14A.Clark
wedubbedthe‘‘translationview’’oflanguage.Onapuretranslationview,itishardtoseehowourlinguisticencounterscandoanythingmorethaninculcateakindofusefulshorthandforideaswhoseverythinkabilityrequiresonlyonthemorefundamentaltokenings(inmentaleseorneuralese)withwhichtheyhavecometobeassociated.Thealternativeonofferisa‘‘hybridmodel’’accordingtowhichsomeofthecognitivebenefitsthatlanguagebringsdependonthecomplementaryactionofactualmaterialsymbols(andimage-likeinnerencodingsofsuchsymbols)andmorebiologicallybasicmodesofinternalrepresentation.
Effectstentativelyexploredunderthisumbrellaincludedtheideasthat:1.Otherwiseinacessiblecontentscanbelearntandgraspedbyagentsskilledintheuseofperceptuallysimpletokensthatreifycomplexideas.
2.Thepresenceofmaterialsymbols(orimagesthereof)canproductivelyalterthefulcrumsofattention,perceptionandaction.Andmostcontentiouslyofall:
3.Materialsymbols(ortheirshallowimagisticencodings)cancoordinatewithmorebasicrepresentationalresourcestoyieldnewformsofhybridthought.Ifthiskindofstoryisevenhalfwaycorrect,thenmindslikeoursareindeedtransformedbythewebofmaterialsymbolsandepistemicartifacts.Butthattransformationmayneitherrequirenorresultintheinstallationofbrandnewinternalrepresentationalforms.Instead,theremaybemuchunder-exploredmeritinthecannyuseoftheexternalforms(andinternalimagesofthoseveryforms)themselves.Suchformsmayhelpsculptandmodifyprocessesofselectiveattention,andactaselementswithinhybridrepresentationalwholes.
Oneimmediatemeritofsuchaviewisamorenuancedattitudetothevexedquestionofevolutionarycognitivecontinuity.JessePrinz(2004)makesthepointwell:
Researcherswhopresumethatwethinkinamodalsymbolsfaceadilemma.Iftheyarguethatnonhumananimalslacksuchamodalsymbols,theymustpostulatearadicalleapinevolution.Iftheysupposethatanimalshaveamodalthoughts,theymustexplainwhyhumanthoughtissomuchmorepowerful.Empiricism[Prinz’sfavorite,thoughnotobligatoryinthepresentcontext!]whencoupledwiththeassumptionthatwecanthinkinpubliclanguage,explainsthediscrepancyincognitivecapacitieswithoutpostulatingamajordiscontinuityinevolution.(p.427)
Needlesstosay,muchremainstobedone.Itwouldbegoodtohaveaclearaccountofjustwhatattention,thatcrucialvariablethatlinguisticscaffoldingseemssopotentlytoadjust,actuallyis.Itwouldbegoodtohavemuchmoreinthewayofgenuine,implementable,fullymechanisticmodelsofthevariouswaysthatinternalizedlanguagemightenhancethought.Anditwouldbegoodtoknowjustwhatitisabouthumanbrainsand/orhumanhistorythathasenabledstructuredpubliclanguagetogetsuchacomprehensivegriponmindslikeours.Butshortfallsaside,Ihopetohaveatleastbroughttheartifactmodelintoclearerview,andtohave
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
585
590
595
600
605
610
615
620
PhilosophicalPsychology15
shownwhyitmightbeattractivetoanyonewhothinksthatlanguagemakesatrulydeepcontributiontohumanthoughtandreason.
Acknowledgements
ThispapergrewoutofmaterialproducedfortheworkshopsonMemory,MindandMediaorganizedbyJohnSuttonatMacquarieUniversity,Sydney,AustraliainDecember2004.ThankstoJohnSutton,RobWilson,MarkRowlands,andallthespeakersandparticipantsatthosemeetingsfortheirinvaluableinputandcriticism.Thanksalsototwoanonymousrefereesforimportantandthought-provokingcomments.ThisprojectwascompletedthankstoteachingreliefprovidedbyEdinburghUniversityandbymatchingleaveprovidedundertheAHRCResearchLeaveScheme.
Notes
[1]
Notethatthesuggestionhereisnotthatprocessesofabstractionalwaysoreventypicallyrequiretheloopthroughpublictokensorsymbols.Ratheritisthatsuchloops,whenpresent,canplayadistinctivecognitionenhancingrole.Forsomeimportantexplorationsofthenature,scopeandpossiblelimitsofsuchroles,seeSchwartzandBlack(1996),andSchyns,Goldstone,andThibaut(1998).
[2]
Apossibleworry(thankstoananonymousrefereeforraisingthisissue)isthatthekindsofrichinteractionbetweendifferentresourcespositedbyhybridaccountsmayfirstrequirethetranslationofthevariousdifferentelementsintoa‘‘commoncode,’’thusundermininganyclaimofgenuinehybridity.Apossibleanalogyhereiswithcasesofintermodalinteraction,alsosometimessaidtorequiretheexistenceofacommoncode.Butinbothcasesapossibleresponse,itseemstome,issimplytodenytherequirement.Potentcoordinatedinteractionneednotrequireacommoncode.Considerthecaseofcodinginthedorsalandventralvisualstreams.Thetwostreams(seeMilner&Goodale,1995)looktotradeinhighlydistinctrepresentationalforms,yetindailylife(inuncompromisedsubjects)theyworktogetherseamlesslyintheserviceofgoaldirectedbehaviour.
[3]
Seeworkon‘‘toolsforthought,’’the‘‘extendedmind,’’‘‘widecomputation,’’‘‘vehicleexternalism’’:Clark(1997,2003);Clark&Chalmers(1998);Dennett(1991,1996);Hurley(1998);Rowlands(1999);Wilson(1994,2004).
[4]
Perhapsthereareeffectsonlearningtrajectories(seethegradeoneexamples)thatresistthedilemmabutforhereandnowthinking(sotheargumentgoes)theoptionsareasstated.[5]
Fromthispointon,wheneverIspeakof‘hybridrepresentationalforms’Ishallmeanformsthatincludebothstandardkindsofinternalrepresentation(mentalese,neuralese,perceptualsymbolsystems,...)and,asproperpartsofakindofdistributedencoding,eitherthematerialsymbolsofsomepubliclanguage,orshallowimagisticencodingsofthoseveryforms.
[6]
Asecondreason(forFodor’sdownplayingthepowerofhybridity)flowsfromhis(in)famousviewsconcerningconceptlearning.Forgiventhoseviews,themeaningofhybridrepresentationalformscouldnotbelearntunlessthelearneralreadyhadtheresourcestorepresentthatverymeaningusingmorebiologicallybasic(indeed,innate)resources.This,however,isnotthetimeorplacetoengageinthisimportantdiscussion(forsomecountervailingthoughts,seePrinz&Clark,2004).
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
216A.Clark
625
[7]
Itisamootpointexactlywhatconstitutes‘‘profound’’reorganization.Butinessence,themostradicalversionoftheviewIamdefendingholdsthatalthoughthebrainmustlearntodealwiththespecialclassoflinguisticstructures,insoitneednotreorganizeitsneuralcodingroutinesinanywaythatisdeeperormoreprofoundthanmightoccur,say,when630
wefirstlearntoswim,ortoplayvolleyball.
[8]
AfurtherquestionisexactlyhowthehybridviewdefendedinthispaperrelatestothatofCarruthers(2002).Therelationhereishardtodetermine,asthestartingpointsofthetwoaccountsareverydifferent.Carruthersbuysintolarge-scalementalmodularityandseesnaturallanguageascognitionenhancinginvirtueofbeingthesolemediumofallmodule-635
integratingthoughts.Thenotionofhybridcognitivevehiclesdefendedhereseemstometobeattractivelyweakerthanthis.Itisindifferenttothetruthorfalsityofmodularity.
References
Adams,F.,&Aizawa,K.(2001).Theboundsofcognition.PhilosophicalPsychology,14,43–64.Barsalou,L.W.(1999).Perceptualsymbolsystems.BehavioralandBrainSciences,22,577–609.640
Boysen,S.T.,Bernston,G.,Hannan,M.,&Cacioppo,J.(1996).Quantity-basedinferenceand
symbolicrepresentationinchimpanzees(Pantroglodytes).JournalofExperimentalPsychology:AnimalBehaviorProcesses,22,76–86.
Carruthers,P.(2002).Thecognitivefunctionsoflanguage.BehavioralandBrainSciences,25,
657–726.
645
Churchland,P.M.(1989).Aneurocomputationalperspective.Cambridge,MA:MITPress.
Churchland,P.M.(1995).Theengineofreason,theseatofthesoul.Cambridge,MA:MITPress.Churchland,P.M.(1996).Theneuralrepresentationofthesocialworld.InL.May,M.Friedman&
A.Clark(Eds.),Mindsandmorals(pp.91–108).Cambridge,MA:MITPress.
Clark,A.(1997).Beingthere:Puttingbrain,bodyandworldtogetheragain.Cambridge,MA:MIT
650
Press.
Clark,A.(1998).Magicwords:Howlanguageaugmentshumancomputation.InP.Carruthers&
J.Boucher(Eds.),Languageandthought:Interdisciplinarythemes(pp.162–183).Cambridge,England:CambridgeUniversityPress.
Clark,A.(2003).Natural-borncyborgs:Minds,technologies,andthefutureofhumanintelligence.
655
NewYork:OxfordUniversityPress.
Clark,A.,&Chalmers,D.(1998).Theextendedmind.Analysis,58,7–19.
Clowes,R.W.,&Morse,A.F.(2005).Scaffoldingcognitionwithwords.InL.Berthouze,F.Kaplan,
H.Kozima,Y.Yano,J.Konczak,G.Metta,J.Nadel,G.Sandini,G.Stojanov&C.Balkenius(Eds.),Proceedingsof5thInternationalWorkshoponEpigeneticRobotics:ModelingCognitive660
DevelopmentinRoboticSystems(LundUniversityCognitiveStudies,Vol.123,pp.101–105).Lund,Sweden:LundUniversityCognitiveStudies.
Colunga,E.,&Smith,L.B.(2005).Fromthelexicontoexpectationsaboutkinds:Arolefor
associativelearning.PsychologicalReview,112,347–382.
Dehaene,S.(1997).Thenumbersense.Oxford,England:OxfordUniversityPress.
665
Dehaene,S.,Spelke,E.,Pinel,P.,Stanescu,R.,&Tviskin,S.(1999).Sourcesofmathematical
thinking:Behavioralandbrainimagingevidence.Science,284,970–974.Dennett,D.C.(1991).Consciousnessexplained.NewYork:LittleBrown.Dennett,D.C.(1996).Kindsofminds.NewYork:BasicBooks.
Densmore,S.,&Dennett,D.C.(1999).Thevirtuesofvirtualmachines.Philosophyand
670
PhenomenologicalResearch,59,747–767.
Fodor,J.A.(1975).Thelanguageofthought.NewYork:Crowell.
Fodor,J.A.(1998).Dowethinkinmentalese:RemarksonsomeargumentsofPeterCarruthers.In
criticalcondition:Polemicalessaysoncognitivescienceandthephilosophyofmind(pp.63–74).Cambridge,MA:MITPress.
NewXMLTemplate(2006)
{TandF}CPHP/CPHP_A_168956.3d
[27.3.2006–8:06pm](CPHP)[1–17]CPHP_A_168956
2675
680
685
690
695
700
PhilosophicalPsychology17
Fodor,J.A.(2004).Havingconcepts:Abriefrefutationofthe20thcentury.Mind&Language,19,
29–47.
Hermer-Vazquez,L.,Spelke,E.,&Katsnelson,A.(1999).Sourcesofflexibilityinhumancognition:
Dual-taskstudiesofspaceandlanguage.CognitivePsychology,39,3–36.
Hurley,S.(1998).Consciousnessinaction.Cambridge,MA:HarvardUniversityPress.
Milner,A.,&Goodale,M.(1995).Thevisualbraininaction.Oxford,England:OxfordUniversity
Press.
Philippides,A.,Husbands,P.,Smith,T.,&O’Shea,M.(2005).Flexiblecouplings:Diffusing
neuromodulatorsandadaptiverobotics.ArtificialLife,11,139–160.
Prinz,J.(2004).Sensibleideas:AreplytoSarneckiandMarkmanandStilwell.Philosophical
Psychology,17,419–430.
Prinz,J.,&Clark,A.(2004).Puttingconceptstowork:Somethoughtsforthe21stCentury.Mind
&Language,19,57–69.
Rowlands,M.(1999).Thebodyinmind:Understandingcognitiveprocesses.Cambridge,England:
CambridgeUniversityPress.
Rupert,R.(2004).Challengestothehypothesisofextendedcognition.JournalofPhilosophy,101,
389–428.
Schwartz,D.L.,&Black,J.B.(1996).Shuttlingbetweendepictivemodelsandabstractrules:
Inductionandfallback.CognitiveScience,20,457–498.
Schyns,P.G.,Goldstone,R.L.,&Thibaut,J.-P.(1998).Thedevelopmentoffeaturesinobject
concepts.BehavioralandBrainSciences,21,1–54.
Smith,L.,&Gasser,M.(2005).Thedevelopmentofembodiedcognition:Sixlessonsfrombabies.
ArtificialLife,11,13–30.
Thompson,R.K.R.,Oden,D.L.,&Boysen,S.T.(1997).Language-naivechimpanzees
(Pantroglodytes)judgerelationsbetweenrelationsinaconceptualmatching-to-sampletask.JournalofExperimentalPsychology:AnimalBehaviorProcesses,23,31–43.Wilson,R.A.(1994).Widecomputationalism.Mind,103,351–372.
Wilson,R.A.(2004).Boundariesofthemind:Theindividualinthefragilesciences—Cognition.
Cambridge,England:CambridgeUniversityPress.
因篇幅问题不能全部显示，请点此查看更多更全内容