Abstraction and the Origin of General Ideas

volume12,no.19 december2012

Abstraction and the

Origin of General Ideas

Stephen LaurenceUniversity of Sheffield

Eric MargolisUniversity of British Columbia

© 2012 StephenLaurence&EricMargolisThis work is licensed under a Creative Commons

Attribution-NonCommercial-NoDerivatives 3.0 License. <www.philosophersimprint.org/012019/>

1. Introduction1

Giventheiroppositiontoinnateideas,philosophersintheempiricisttradition have sought to explain how the rich and multifariousrepresentational capacities that human beings possess derive fromexperience. A key explanatory strategy in this tradition, tracingback at least as far as John Locke’s An Essay Concerning Human Understanding, is tomaintain that the acquisition ofmany of thesecapacities can be accounted for by a process of abstraction. In fact,LockehimselfclaimsintheEssaythatabstractionisthesourceofall general ideas (1690/1975, II, xii, §1).AlthoughBerkeley andHumewerehighlycriticalofLocke,abstractionasasourceofgeneralityhasbeenalastingthemeinempiricistthought.NearlyacenturyafterthepublicationofLocke’sEssay,forexample,ThomasReid,inhisEssays on the Intellectual Powers of Man, claims that “we cannot generalizewithoutsomedegreeofabstraction…”(Reid1785/2002,p.365).Andmorethanacenturylater,BertrandRussellremarksinThe Problems of Philosophy:“Whenweseeawhitepatch,weareacquainted,inthefirst instance,with the particular patch; but by seeingmanywhitepatches, we easily learn to abstract the whiteness which they allhave in common, and in learning to do thiswe are learning to beacquaintedwithwhiteness”(Russell1912,p.101).

Despite the importance of abstraction as a central empiriciststrategyforexplainingtheoriginofgeneralideas, ithasneverbeenclear exactly how the process of abstraction is supposed to work.Thereareanumberofreasonsforthis.Oneisthatmanyphilosopherswhohavewrittenaboutabstractionhavebeenmoreconcernedwiththeroleofabstractioninsupportingametaphysicalagendathanwiththe psychological details of the process of abstraction. Interestingly,philosophershaveappealedtoabstractionintheserviceofopposingmetaphysicalpositions. Some (e. g., LockeandReid)have calledonit as a means for explaining generality in a way that is consistent

1. Thisarticlewas fullycollaborative; theorderof theauthors’names isarbi-trary.WewouldbothliketothanktherefereesforPhilosophers’ Imprint.EricMargoliswouldalsoliketothankCanada’sSocialSciencesandHumanitiesResearchCouncilforsupportingthisresearch.

ImprintPhilosophers’

stephenlaurence&ericmargolis Abstraction and the Origin of General Ideas

philosophers’imprint –2– vol.12,no.19(december2012)

consideringofaMan’sself,orothers,andtheordinaryproceedingsoftheirMindsinKnowledge…”(1690/1975,III,iii,§9).

Perhaps it is not surprising, then, that we know so little aboutabstraction. But given the recurring interest in abstraction, andgiven the importance of general ideas in thought, philosophersclearly need an explicit framework for understanding abstractionthat isn’tbeholdento introspectionandthat isopento thefindingsofperceptualanddevelopmentalpsychologyandrelatedfields.Ouraiminthispaperistoprovideageneralframeworkthatfillsthisgapand to explore some of its philosophical implications. One of ourmotivationsisto identify theextenttowhichaprocessthatisbroadlylike the one invoked by Locke and other philosophers can explainthe acquisition of general representations.3 We should note at theoutset, though, thatwhile thispaper takes its inspiration fromearlyphilosophicaldiscussionsofabstraction,ourfocusistheoreticalratherthanhistorical.Weareprimarilyinterestedintheexplanatorybenefitsthatcanbeobtainedbysomethingakin to the traditionalnotionofabstraction,notwiththehistoricalcontroversiesregardinghowLockeandotherphilosophersinthemoderneraarebestinterpreted.We’llseethattherearegoodreasonstoabandonsomeofthefeaturesthatfiguredprominentlyintraditionalaccountsofabstraction—includingthe link between abstraction and anti-nativist views of cognitivedevelopment.Nonetheless,webelieve thatphilosophers likeLockewererighttoemphasizethesignificanceofabstractionasameansofacquiring generalmental representations. Even if theywerewrong

3. Inwhat follows,wewilloccasionallymakereference to theacquisitionofconcepts,whereaconceptisunderstoodasatypeofmentalrepresentation.However, nothing essential turnson thiswayof thinking about concepts.Onviewsthattakeconceptstobeatypeofabstractobject,abstractionmaystillbeimportanttotheacquisitionofgeneralconceptsbywayofmediatingaccesstotheseabstracta.Onsuchaview,ourtalkofacquiringconceptsviaabstraction should be understood in terms of acquiring general represen-tationsthathaveconceptsastheirsemanticvalues. Inanycase,ourfocusinthispaperisonthequestionofhowgeneralmentalrepresentationsareacquired;ouruseofthetermconceptcanbereadasstipulativelyreferringtogeneralmentalrepresentations.

withbroadlynominalistic scruples,whileothers (e. g.,Russell)haveunderstoodittobeanessentialingredientformakingsenseofrealismaboutuniversals.

Another reason why the psychological details of the process ofabstractionhavebeensounclearisthatphilosophershavereliedonintrospectionastheprincipalsourceofinformationabouttheprocess.Conflicting opinions regarding abstraction consequently turn ondivergent claims about what introspection uncovers. While Locketakesittobeevidentthatintrospectionshowsthatgeneralideaslikeman or horse2 are acquired through abstraction, others, includingBerkeleyandHume, claim that theydon’t see this at allwhen theylookintotheirownminds.Butevenifeveryoneweretoagreeaboutthedeliveranceofintrospection,thatwouldstillleaveuslargelyinthedarkabouttheprocess.Fromacontemporaryvantagepoint,itiswellestablished thatmuch of themind isn’t accessible to introspectionandthatintrospectivereportsofpsychologicalprocessesaren’talwaystrustworthy.Thereislittlereasontothinkthattheprocessesinvolvedinabstractionshouldbeanexception.

We suspect, however, that the most important reason why thepsychological details of the process of abstraction have remainedobscureisthatitsadherentshavenotappreciatedtheneedtoprovideasubstantiveexplanationofhowitworks.AsChomskyhasemphasized,this is often the casewhen it comes to themind. “Onedifficulty inthe psychological sciences lies in the familiarity of the phenomenawithwhich theydeal.A certain intellectual effort is required to seehowsuchphenomenacanposeseriousproblemsorcallforintricateexplanatory theories. One is inclined to take them for granted asnecessaryorsomehow‘natural’”(Chomsky2006,p.21).ConsiderhowLockepeppershisdiscussionwithphrasesthataremeanttohighlighttheobviousnessofhissubjectmatter.Forexample:“Thatthisistheway, whereby Men first formed general Ideas,and general Names to them, I think, is so evident, that thereneedsnootherproof of it, but the

2. WetakeLocke’sIdeastobementalrepresentationsandwilluseexpressionsinsmallcapstorefertomentalrepresentations.



onlyleaveoutofthecomplexIdeatheyhadofPeterandJames, MaryandJane,thatwhichispeculiartoeach,andretainonlywhatiscommontothemall.(III,iii,§7)

Locke scholars have debated how to interpret Locke’s remarksabout the nature of abstraction and even whether he has a singleaccount.This isunderstandable,sincethereissomeunclarityaboutwhether Lockean general ideas are formed by retaining the fullrepresentationsassociatedwiththeparticularsthatanagentperceives.To some readers, it sounds like the full representationsare retainedandthatabstractioninvolvesattendingtocertainfeaturesasopposedtoothers.However,toothersreaders,thereisthesuggestionthatanabstract ideamay involve the constructionof a new representation,onethattakessomefeaturesfromtherepresentationsofexperiencedparticularswhileomittingothers.4RegardlessofwhattherightstoryisaboutLocke,it isclearthatheviewsabstractionasaprocessthatisgroundedinperceptionandthatoperationsontherepresentationsresultingfromcontactwithparticularsarethesourceoftheabilitytorepresentfarmorethantheitemsthatwereoriginallyperceived—notjust thiswhitepaperbutallwhiteobjects,not just thismanbutallhumanbeings,andsoon.

Buthowexactlycanabstractionbethesourceofallgeneralideas?Toseetheforceofthisquestion,weneedtostepbackandconsidermore carefully what input gets the process going. If abstraction isto explain the origins of all general representations, what kindsof representations can it draw upon, and how do they depict theparticulars that an agent perceives? We will argue that there arefourmodelsoftherepresentationalinputthatareavailabletoLockebut thatnoneof thesemodelscanprovideasatisfactoryaccountofthe origins of all general representations. The result,wewill argue,

4. Thedifferencebetweenthesetwoapproachesisnicelysummedupbythecontrast between J.L.Mackie’s description of abstraction as selective atten-tionandJonathanDancy’ssloganthatabstraction is subtraction(Mackie1976,Dancy1987).

aboutsignificantdetailsabouthowtheprocessofabstractionworks,abstractiondoesplayan important role inexplaining theoriginsofgeneralrepresentations.

2. Some General Representations Are Innate

In Book II of the Essay, Locke describes the process of abstraction,claiming that abstraction is the source of all of the mind’s generalrepresentations.AccordingtoLocke,abstractionisthepowerofmindthatinvolves“separating[Ideas]fromallotherIdeasthataccompanythemintheirrealexistence;thisiscalledAbstraction.AndthusallitsGeneral Ideas aremade” (1690/1975, II, xii, §1). Locke gives severalexamples that are meant to illustrate the workings of abstraction.Regardingtheoriginsofthegeneralrepresentationwhite,wearetold:

… the sameColour being observed to day in Chalk orSnow,which theMindyesterday received fromMilk, itconsidersthatAppearancealone,makesitrepresentativeofallofthatkind;andhavinggivenitthenameWhiteness, itby thatsoundsignifies thesamequalitywheresoevertobeimagin’dormetwith;andthusUniversals,whetherIdeasorTerms,aremade.(II,xi,§9)

The claim is that a general representation for a simple quality isformedby (in somesense) leavingout specificdetails aboutwhereandwhenitoriginated,aswellasotherideasthatmayhaveinitiallyaccompaniedit.Later,inBookIII,Lockediscussesadifferentkindofexample—theformationofacomplexidea.Hesuggeststhatchildrenmayacquiremanbyfirstattendingtoparticularindividuals,suchastheirnurseormother,andlaterobservingthatotherthingsresemblethoseindividuals.Thisleadschildrento:

…frameanIdea,whichtheyfindthosemanyParticularsdopartakein;andtothattheygive,withothers,thenameMan,forExample.Andthus they come to have a general Name, andageneralIdea.Whereintheymakenothingnew,but



thatwouldcomprisetheinputtotheacquisitionprocess,anditwouldpresumably be the perception of its color that would support theacquisitionofwhite.Butthentheprocessofacquiringwhitewoulddepend upon prior representations that include, among others, therepresentationwhite.Themodelisplainlycircular.Itendsupsayingthatwhiteistheproductofaprocessthattakeswhiteasitsinput.

E.J. Lowe has made a related point in a criticism of Lockeanabstraction (Lowe 1995, pp. 161–2), but there is an importantdifferencebetweenLowe’s criticismandourown.Loweclaims thatabstractioncan’tgetoff theground if theagentdoesn’thaveawayto single out particulars in perception prior to abstraction takingplace,andheclaims that this requiresbeingable to representeachparticularunderasortalthatprovidesaprincipleofindividuationforthings of the same type. Then theproblem is that abstraction can’taccountforwherethesesortalrepresentationscomefrom,sincetheyareanecessaryprecursorforabstractiontotakeplace.Lowegivestheexampleofseeingananimal.Hesaysthatyoumaynothavetoknowwhattypeofanimalitis,butyouhavetoatleastrepresentitunderthesortalanimalinordertosingleitoutfromotherobjects.6

WeagreewithLowethatgeneralrepresentationsarerequiredtoget theprocessof abstractiongoing,butnot for the reason thathecites. The problem isn’t limited to sortal representations and isn’tprimarilygeneratedbytheneedtorepresentparticulars.Rather,theproblem arises for any of the salient features of a perceived objectthat,byhypothesis,arepartoftheinputtotheprocessofabstraction.Whether the representations of these features provide principles ofindividuationisirrelevant.Nowweourselveshaven’tyetarguedthatgeneral representationsmustfigure in the input to theprocess. Forthemoment,itissimplyanimmediateconsequenceofthefirstmodelthatweareconsideringthattheydo.Ourownargumentfortheneedforgeneralrepresentationswillemergethroughconsiderationofthe

6. Thoughitdoesnotaffectourpoint,Lowewouldn’tputthingsexactlyaswedointhetext,sinceheisagnosticaboutmentalrepresentationsandpreferstocouchtheissueintermsofrepresentationalabilities.

is that abstraction cannot plausibly be the source of all generalrepresentationsandthatitishighlyunlikelythatanylearningprocesscouldbethesourceofallgeneralrepresentations.Ifanorganismhasanygeneralrepresentationsatall,then,inalllikelihood,someofthesemustbeinnate.

Weshouldnoteattheoutsetthatthisargumentisintendedasaninferencetothebestexplanation,notaproof.Wedonotclaimthatitislogicallyimpossibleforallgeneralrepresentationstobeacquiredwithouttherebeingsomeinnategeneralrepresentations.Rather,ourpointisthatnon-nativistmodelsincurprohibitiveexplanatorycosts.Also, to simplify the discussion, we will suppose that the generalrepresentation that we are trying to understand is white and thatthe experience fromwhich it is abstracted is the visual perceptionofasnowball(oranumberofsnowballs).Wecannowrephrasetheissueasidentifyinghowthesnowballisinitiallyrepresentedsothatwhitecanbeabstractedfromtheexperience.Therearefourpotentialmodelstoconsider.

Model 1: Individual-representations and feature-representations. The firstmodeltakesasinputacombinationofindividual-representations(i. e.,representations which function like names or demonstratives andrepresentindividualsquaindividuals)andrepresentationsforeachofthesalientfeaturesoftheexperiencedparticular.Thusthesnowballmightinitiallyberepresentedwithsuchrepresentationsasthat, cold, spherical,andsolid.

Thismodelfacesanumberofproblems,butthemostseriousisthatit simplypresupposes that theprocessof abstraction takesas inputgeneral representations.5Thisclearlywon’tdoifthegoalisforabstractiontoexplaintheacquisitionofallgeneralrepresentations,astheappealtopriorgeneralrepresentationswillleadtoaregress.Moreover,color willundoubtedlybeamongthesalientgeneralfeaturesofthesnowball

5. Therepresentationsofshape,temperature,etc.intheinputmightbenoncon-ceptualrepresentations,asopposedtoconceptualones.Buttheywouldbegeneralrepresentationsallthesame.



individualobjectsassuch,theagentiseffectivelyrepresentationallycutofffromallthefeaturesoftheobjects.

Suppose,however,thatweoverlookthequestionofwhydifferentindividual-representations are grouped together and simply allowthat they are. Then a number of individual-representations couldbe combined, yieldinga representation like this and this and this (each ‘this’ referring to one of three different white snowballs).Still, the resulting representationwouldn’t do, since (1) it lacks therepresentationalbreadthofwhite(white isprojectible,whereastheconjoined individual representations only pick out the particularsthathavebeenencountered)and(2)itfailstosingleouttherelevantfeaturethattheseobjectshaveincommon(whiteness,asopposedto,forexample,sphericality,coldness,snowballness,etc.).It’sonethingtorepresentanumberofperceivedobjectsthathappentobewhiteandquiteanothertorepresentwhiteness(ortorepresentwhitethingsin general). No finite conjunction of individual-representations ofwhitethingswouldconstituteageneralrepresentationofwhiteness.

Model 3: Trope-representations. We are asking what the input to theprocess of abstraction might look like on the Lockean assumptionthat abstraction is the sourceof all general representations.A thirdpossibility, which is seen in thework of Thomas Reid, is that it isparticularized properties or abstract individuals, also known as tropes, that the input representations represent as such.7 A trope isproperty-likeinthatitconstitutesafeatureofaparticular,butunlike

7. Reidremarksthat“thewhitenessofthesheetofpaperuponwhichIwritecannotbethewhitenessofanothersheet,thoughbotharecalledwhite”,andhegoesontoaddthat“thewhitenessofthissheetisonething,whitenessanother” (Reid 1785/2002,p. 367). ForReid, there isno such thingas theuniversalwhiteness. There areonly the individual color tropes that are in-herent ineachpieceofpaper, each snowball, etc. Still, theappearanceofgeneralityandtheprevalenceofgeneraltermsinnaturallanguagearebothtobeexplainedbyreferenceto“generalconceptions”.ThoughReid’sgeneralconceptionsareverydifferentfromLocke’sgeneralideas,andReidhimselfwasatrenchantcriticofLockeanideas,ourcriticismsofthetropeviewdonotpresupposethatgeneralrepresentationsareakintoLockeanideasandapplyequallytoReid’sgeneralconceptions.

various options regarding the input and through highlighting thenecessityofexplaininghowlearnerscanselectivelyattendtofeaturesofstimuli.ButeventhenthereisnoreasontosupposewithLowethatsortals are required to isolate objects for further attention.There isgoodempiricalevidence foramechanismofvisualattention that isable to track objects by focusing on their spatial-temporal features,not their kind-individuating features, and that this mechanism ispresentearlyincognitivedevelopment(Scholl2001).Sowhilewe’llsee that Lowe is right to questionwhether abstraction can accountforallgeneral representations,his focusonsortals is too restrictive.Thefundamentalproblemis just thattheindividual-representations-and-feature-representationsmodelassumes thattherearefeaturesofaparticularthatinitiallyneedtoberepresentedassuch;whateverthesefeaturesare,therepresentationsofthesefeaturescannotthemselvesbeacquiredviaabstractiononthismodel.

Model 2: Individual-representations only.Inordertoaddresstheproblemwith thepreviousmodel, onemight suppose instead thatparticularsareinitiallyrepresentedonlybyindividual-representationswithoutanygeneralrepresentationscomingintoituntilabstractionhastakenplace.

Wedon’tknowofanytraditionalempiricistswhohaveproposeda model of this kind, however, and for good reason. Individual-representations alone don’t provide enough information to get theprocess of abstraction going. If particulars are represented simplyas objects, without representing any of their features, then theinput just isn’t rich enough.After all,with the canonical individual-representations—demonstratives—the whole idea is that theyrepresent their referents directly, conveying no information aboutwhat the representedobjects are like. But if all themindhas to gooninrepresentingtwowhiteobjectsisthisandthat,itwouldhavenobasis forcognitivelygroupingthetwotogether,andcertainlynobasisforbringingthemunderaspecificgeneralrepresentationsuchaswhite.Bylimitingtheinitialrepresentationstorepresentationsofthe



a process that appeals to an existing general representation (i. e.,similar)inexplainingtheacquisitionofanewgeneralrepresentationis prohibited.Moreover, a completely general concept of similaritywouldbeoflittleuseanyway.Supposethattheagentdeemsthatthereferentsofwhite1 andwhite2fallunderthefullygeneralsimilar(insomerespectorother).Sinceanytwoobjectsaresimilarininfinitelymanyways (Goodman1972), thisdoesnotbringusanycloser toageneral representationofwhite, and it leaves the learnerunable torepresent the specific respect inwhich these two tropesare similar.Indeed,itseemsthatnothingshortofageneralrepresentationintermsofsimilarwithrespecttowhitenesswilldothetrick,sinceanytwocolortropeswillbecolor-similarinindefinitelymanyrespectsaswell(corresponding to indefinitely many ways of partitioning the colorspacethatincludebothtropes).Butifweneedtoappealtoageneralrepresentationalongthelinesofsimilarwithrespecttowhiteness, wemightaswelladmitthatthelearnermustalreadyhavethegeneralrepresentation white. We are driven back to the problem we sawearlier.Theinputtoabstractionwouldpresupposetherepresentationswhoseacquisitionabstractionissupposedtoexplain.

There is, however, another option for explaining how differenttropesmightbedeemedsimilarandconsequentlywhytheindividual-representations for these tropes should be grouped togethercognitively. This is that the perceived similarity between thesetropesisitselfexplainedintermsofarepresentedtrope(namely,thetropeof the similarity between white1andwhite2), so thatallof therepresentationsinplayarerepresentationsoftropesassuch.Inthiscase, the agentwould represent the referentsofwhite1 andwhite2 asbeingrelatedviaarelationalsimilarity-tropethatisuniquetothemand that no other individuals can participate in. Let’s suppose thatthesimilarityinsuchcasesispickedoutbysimilar1.

8Considernow

8. Thisisanoversimplification,sinceanytwocolortropeswillstandinindefi-nitelymanysimilarityrelations(justastheywouldhaveindefinitelymanyfeaturesincommon).Butwewillgrantthissimplificationforthepurposesofargument.

auniversal,itcanbepresentonlyinoneparticular.Thisisnotmerelybecausenootherparticularshappentohavethatfeaturebutbecause,by itsmetaphysical nature, a given trope canbe possessed only byasingleindividual—tropesaren’tmultiplyinstantiable.Returningtothe snowball example, theproposal is that the input to theprocessofabstraction includesa representationof thesnowball’swhiteness,wherethisistakentobeatropethatisinherenttothesnowball;nootherparticularcanparticipateinthisverywhiteness.Inotherwords,themodelrestrictstheinputtorepresentationsofindividuals(tropes,asabstractparticulars)butoffersthehopethattheagentisnolongercutofffromrepresentingthefeaturesoftheparticularssheperceives(tropes, as particularized properties). Features can be representedwithoutanygeneralrepresentationsbeingillicitlysmuggledintothefoundationsoftheacquisitionprocess.

Unfortunately, appealing to tropes doesn’t help. In representingthewhitenessof twowhiteobjects, anagentwouldhave todeploytwo distinct representations, white1 and white2, to represent eachwhitenesstropeassuch.Becausetheserepresentationsareessentiallyof individuals (namely, the two tropes), this gives rise tomuch thesamesortofdifficultiesthataroseforthepreviousmodel.Thereisaquestionaboutwhytheseindividualsaretobegroupedtogetherandhowrepresentingthemtogetheryieldsafullygeneralrepresentationasopposedtoonethatmerelypicksouttheindividualsthathavebeenencounteredthusfar.

Onemightthinkthatsomeheadwaycanbemadeonthequestionof why tropes are grouped together by saying that the agent alsorepresentsthesimilaritybetweenthetropes.Intheend,thissuggestiondoesn’thelp,butitturnsouttobesomewhatcomplicatedtoseewhy.Thisisbecausetherearedifferentwaysinwhichthesimilaritymightitselfberepresented.

Thesimplestwaywouldbetouseageneralconceptofsimilarity,one that quantifies over the respects in which similar things aresimilartooneanother.However,ifwearelookingforaprocessthatwouldallowustoexplaintheacquisitionofallgeneralrepresentations,



Bertrand Russell uses a related argument in the context of thepurely metaphysical dispute about the status of universals. Heargues that a nominalistic metaphysics that relies on resemblancebetween particulars isn’t viable, since it would require “that theseresemblancesresembleeachother,andthusatlastweshallbeforcedtoadmitresemblanceasauniversal”(Russell1912,p.96).Russell isright that there isaneedforahigher-orderrelationofresemblance,butinprincipleanominalistcouldshuntheuniversalresemblancebyappealingtoaninfinitehierarchyoftropes,wheretheresemblanceatanygivenleveliscapturedbyatropethatisuniquetotheresemblancetropes at the previous level (Campbell 1990).We aren’t saying thisisanespeciallyappealingmetaphysics(seeDaly1994forcriticisms),but so far as we can see, trope theorists are free to postulate aninfinite hierarchy of similarity or resemblance tropes in this way ifthey like. Incontrast, theargument thatbearson thepsychologyofabstractionismuchstronger.Thereasonisthataninfinitehierarchyofrepresentationsoftropeshasnopsychologicalcredibilitywhatsoever.Finite creatures like ourselves can’t actually entertain an infinitenumberof representations.Yet that is exactlywhatwewouldhavetodotoappreciatewhitenessingeneraliftheinputtoabstractionisrestrictedtorepresentationsoftropesassuch.Onceagain,itlooksasifweneedarichersourceofinputifwearegoingtoexplainhowgeneralrepresentationsareacquired.

Model 4: Generality without discrete representations. We have beenlookingatthevariousoptionsregardingtheinputtotheabstractionprocess,keepinginmindthegoalofmakingabstractionthesourceofallgeneralrepresentations.Wehaveruledoutarangeofapproachesthat take some combination of representations of individuals assuchandrepresentationsoffeaturesassuch(models1and2),andalsoapproachesthattakeasinputrepresentationsofparticularizedproperties (tropes) as such (model 3). These come close toexhausting theoptions thatought tobeconsidered.However,onefurtherpossibility is thatmorecomplexmetaphysicalentities than

whathappenswhen the learner representsa thirdwhiteobject, say,awhitesheetofpaper.Thelearnerwillrepresentthepaperasbeingwhite3 (with white3). Shemight then come to compare the paper’swhitenesstotheothertwoobjectsandnoticethesimilaritybetweenthereferentsofwhite1andwhite3andbetweenwhite2andwhite3.Torepresentthesesimilarities,shecouldemployrepresentationsoftherelational tropes involved—similar2 and similar3. Now the learnerhasthreesimilarityrepresentations,buthowcansherecognizethatthesesimilarityrelationshaveanythingincommon?Ifwefollowtheprescription that perceived similarity requires representing a trope,thenwe’dhavetosaythat, foreachofthesepairs, thereisahigher-level similarity-trope representationof the similarity between theserelations (similar4, similar5, similar6), and presumably therewouldhavetobeyetanotherlevelrepresentingthetropesthatexplainthesimilarityamongtheserelations,andsoon(seefigure1).Amodelthatappeals exclusively to represented tropes endsupwith a regress inwhicheachfurtherlevelofrepresentedtropesfailstogetusanyclosertoafullygeneralrepresentationofwhiteness.

Figure 1.Representationsof tropesofwhiteness canbecompared using similarity-trope representations, butfurther similarity-trope representations are required tocomparethese,generatingarepresentationalregress.



have to be 2100 distinct representations that could serve as inputto the abstraction process. That’s about 500 trillion times morerepresentations than there have been seconds in the history of theuniverse(ontheestimatethattheuniverseis20billionyearsoldorroughly6.3x1017seconds).Inourview,thetrulystaggeringnumberofprimitiverepresentationsatplayisenoughtoundermineamodelthatrelieswhollyonunstructuredrepresentations.

But the problemwith thismodel isn’t just the sheer number ofprimitive representations that it would require. The real problemis with how it could account for our ability to acquire white fromsuch representations as this-is-cold-spherical-white without arepresentational basis for homing in on just the whiteness in theexperience. To mentally focus on whiteness itself would seem torequire the prior ability to represent whiteness as such, but thisamounts to helping ourselves to the general representation white.Onceagain, theaccount inquestion seems tobe circular: it cannotexplainhowthesystemcouldderivearepresentationofwhitefromtheinputwithoutpresupposingthatthesystemalreadyhastheabilitytorepresentwhiteness.

It is easy to suppose that abstraction explains the origins of all

generalrepresentationsifyoudon’tthinkthroughthepsychologicaldetails.Butwhatthefailureofthesefourmodelsshowsisthatthereis a substantial burden for theorists who want to maintain thisposition.Theprincipaloptionsforgettingtheprocessofabstractiongoingareallproblematic.Theyeitherpresupposeacertainamountof general representation or are unable to support the acquisitionof the target general representation. Of course, there is always thepossibilitythattheremightbesomefurthermodelofhowabstractiongets started that we have not considered, one that can (somehow)account for the origins of all general representations. For example,itmightbesaidthatabstractionisn’tarepresentationalprocessandhencethattheinputneedn’tincludeanyrepresentationsatall,muchless general ones. All that is required are causal interactions with

individuals and features are represented in the input—somethingakin to events or states of affairs. In this case, the initialrepresentations formingthe input to theabstractionprocessmightbeunstructuredrepresentationsthatmanagetopickoutthesemorecomplexentitieswithoutanycomponentsrepresentingtheobjects,properties, or tropes that are present in the event. For example, asnowballmight be represented as being cold, spherical, and white butwithoutseparaterepresentationscorrespondingtoeachofthesefeatures.Thesnowball’sbeingcold,spherical,andwhitewouldberepresented by a single unstructured representation (this-is-cold-spherical-white), not by a structured representation composed ofdistinct representations capableof independently representing theobjectandtheseseveralfeatures(this,cold,spherical,andwhite).In this way, whitewouldn’t have to be a precursor to abstraction,nor would there have to be prior access to any other generalrepresentationscorrespondingtoaparticular’sfeatures.

Once again, however, psychological considerations need to betaken into account. And from a psychological perspective, sucha model is not at all promising. One important feature of oursystemsof representation isproductivity.Themindcanrepresentanindefinite number of distinct combinations of features, for whichthe best explanation is that discrete representations are combinedand recombined in accordance with a compositional semantics.However,themodelunderconsideration(generalitywithoutdiscreterepresentations)isbuiltontheassumptionthattherepresentationalsystemdoesn’thavethecompositionalstructurethatthisexplanationrequires.Instead,foreachnewcombinationoffeaturesattributedtoanobject, therewouldhavetobeacorrespondingnewanduniqueprimitive representation. Unfortunately, this would require us topossessanastronomicalnumberofprimitiverepresentationstoserveas inputtotheprocessofabstraction.Sinceforanyn featuresthereare2npossiblecombinationsof these features, thismeans thatwithonly a single object and 100 basic features and their combinationsto represent—an absurdly conservative assumption—there would



First consider Berkeley’s criticisms, which primarily focus onLocke’sconstrualofideasasmentalimagesandtheviewthattheseimagescanrepresentonlywhattheyresemble(Berkeley1710/1975).Amongotherthings,Berkeleypointsoutthatimagesaredeterminatein ways that bar them from achieving the generality that Lockerequires. For example, you can’t have an image of a generic manthatrepresentsmeningeneral.Toberecognizableasanimageofaman,itwouldhavetoincludespecificdetails(e. g.,size,shape,color)thatmight be trueof somemenbut not of others.While thismaybe a trenchant criticismofLocke, given theLockeanviewof ideas,proponents of abstraction needn’t be committed to the view thatideasarementalimagesortotheresemblancetheoryofcontent,notevenfortherepresentationsthatsubserveperceptualprocesses.So,forcontemporarytheorists, thesecriticismsdon’treally identify thefundamentalproblemswithabstraction.

Now consider Berkeley’s own theory of the origins of generalrepresentations. According to Berkeley, a general representationarisesasanimagebecomesusedtorepresentarangeofparticularsthat are similar to theone that the image initiallypicksout. In thisway, a representation that is initiallyparticular canbecomegeneral.Berkeley gives the analogy of a drawing of a line in a geometricalproof.Althoughthelinemaybeoneinchlong,itcomestorepresentalllines,notjustone-inchlines,becausetheproofdoesn’tturnonitsparticularlength:

And, as that particular line becomes general by beingmadeasign,sothenameline,whichtakenabsolutelyisparticular,bybeingasign, ismadegeneral.Andas theformerowesitsgenerality,nottoitsbeingthesignofanabstractorgeneralline,butofallparticularrightlinesthatmaypossiblyexist,sothelattermustbethoughttoderiveits generality from the samecause,namely, thevariousparticular lineswhichit indifferentlydenotes.(Berkeley1710/1975,introduction,§12)

property instances. We grant that models of this sort aren’t ruledoutby anythingwehave said.But they aredecidedlyunattractive.9 They effectively postulate mysterious neurological processes thatinexplicablyyieldcontent-appropriategeneralrepresentationssimplyonthebasisofcausalcontactwiththeworld.Indeed,withoutawell-developedaccountofhowtheprocessworks,itishardtoseehowanon-representationalmodelofthiskindissubstantiallydifferentfromamodel that takes certaingeneral representations tobe innateandtriggeredbyappropriatecausalinteractions.Inanycase,aswenoted,we intendourargument tobean inference to thebest explanation.The burden is on theorists who think that abstraction can accountfortheacquisitionofallgeneralrepresentationstoproduceamodelofabstractionthatcanplausiblymeetthisdesideratum.Absentsuchanaccount,weconcludeongroundsofexplanatoryplausibilitythatabstractioncannotexplain theoriginsofallgeneral representationsandthatsomegeneralrepresentationsareinnate.

Lockewasnotaloneinfailingtoappreciatethesortsofdifficultieswehavebeenpointingtoandtheneedtoattendtothepsychologicalquestionofhowabstractionworks.Locke’saccountofabstractionwasfamouslyrejectedbyBerkeleyandbyHumeaswell(largelybasedonBerkeley’s vigorous criticismof the account). Froma contemporaryperspective,however,Berkeley’scriticismsdon’tcutverydeep,sinceanadvocateofabstractioncandroptheassumptionsthatBerkeley’scriticisms turn on. And despite their spirited rejection of Lockeanabstraction,thealternativestoabstractionembracedbyBerkeleyandHumefacemuchthesamesortsofproblemsregardingtheinputtotheprocessthatwehavebeenarguingLocke’saccountfaces.

9. MuchthesamemightbesaidforanAristotelianmodelwheresensibleformsare taken to be literally transmitted from an object, through a perceiver’ssenseorgans,intothemind.Adams(1975)succinctlydescribessuchaviewasfollows:“Perceptionwasinterpretedasatransactioninwhichaform(thesensible form) is transmitted fromtheperceivedobject to theperceiver.…Thereissomething(thesensibleform)whichliterallycomesintothemindfrom theobject.This theoryof perception is thebasis for theAristotelianempiricistanswertothequestion,howwegetourideas”(p.73).



generalrepresentationwhiteandhencereintroducetheproblemofcircularity.11 And general representations aren’t really an option forBerkeleyanyway,sincethewholepointofhistreatmentofgeneralityisthatitissupposedtodoawaywithfullyabstractgeneralideas.

ThesituationforBerkeleyisn’tallthatdifferentfromthesituationforLocke,andit’sthesameforanytheoryofabstraction,orsubstituteprocess, once the need to specify the input is taken seriously. Atleastsomegeneral representationshave tobeavailable togetsucha process going. Some general representations have to be innate. Themoralwedraw fromthese reflections is that thehopeofprovidinga comprehensive theory of the origins of general representationsshouldbeabandoned.Still,aprocessworthyofthenameabstraction might explain the origins of many general representations andthus be an important part of howhuman representational systemsdevelop.Inparticular,theprocessofabstractionmightprofitablybeseen as starting with relatively specific general representations asinput(e. g.,arepresentationforagivenshadeofcolororanarrowlycircumscribed shape) and delivering another type of generalrepresentationasoutput(e. g.,broadercolororshaperepresentations,suchasredortriangular).Theinputrepresentationswouldcapturetheparticularityoftherepresentedqualitiesinexperience—whatisoften called the fine-grainedness of perceptual experience. But theoutput representationswouldbecomparativelymore general in thatthey“abstract”fromtheparticularitiesoftheindividuallyexperiencedcolors, shapes, and soon.This is the idea thatweplan todevelopin the sections to come. It is a major departure from Locke andfrom other traditional accounts, but aswe’ve seen, these accountsfaceinsuperabledifficultiesinexplaininghowatheoryofcognitivedevelopmentcangetbywithanythingless.

11. Hume’saccountofgeneralityisnobetter.Humepresupposesthatpeoplecanrecognizethatdifferentobjectsresembleoneanother.It’sonthebasisoftheresemblance that thecorrespondingparticular ideasbecomeassociatedbyacommonword,suchas‘triangle’.ButHumedoesn’tconsiderthequestionofhowtheresemblanceisregisteredpsychologically.Hetoowouldconfrontthesamesetofproblematicoptions.

HumedescribedBerkeley’streatmentofgeneralrepresentationas“oneofthegreatestandmostvaluablediscoveriesthathavebeenmadeoflateyearsintherepublicofletters…”(1739/1978,I,i,§7).10Butdespitethishighpraise,it’shardtoseewhyBerkeley’saccountisanimprove-mentoverLocke’s.Basically,wearetoldthatanimageachievesgener-alitybecauseitisusedasageneralrepresentation.Anagentstartsoutwithanimageofaparticularbutthenenlistsittoreasonaboutotherthingsbyignoringirrelevantaspectsoftheimageandfocusingonjusttherelevantones.Theproblemwiththisaccountbecomesapparentwhenweaskhowthemindmanagestoachievethisfeat.

Suppose the image isofa specific snowball thata childhas justseen and that she ignores the depicted shape and texture, amongotherthings,intheserviceofthinkingaboutwhitethingsingeneral.Todothis,sheneedstoselectivelyattendtothecolorintheimage.YetBerkeley tellsusnothingabouthowheproposes toaccount for theabilitytoselectivelyattendtocertainaspectsofanimagewhileignoringothers.Inordertopsychologicallyfocusone’sattentiononwhiteness,onemust, ineffect,representwhiteness.But inordertodothis, theoptions are essentially thosewe considered above for the Lockeanaccount.Representingonlyparticulars,whetherconcreteparticularsortropes,doesn’tallowonetoattendtowhitenessassuch.Employingageneralrepresentationofwhitenesswould,ofcourse,allowonetoattend towhiteness, but thatwould requireprior possessionof the

10.Hume’sowntreatmentofgeneralideashasastrongaffinitywithBerkeley’s,thoughthedifferencesbetweenthemareworthnoting.Humedoesn’tfollowBerkeleyinclaimingthatwesimplyattendtorelevantaspectsofanideaandignoreothers.Hesays,instead,thataswenoticetheresemblancebetweendifferentobjects,wegive them thesamename,and then laterusesof thewordcalluprelatedideas.Forexample,theword‘triangle’mayinitiallybringtomindanisoscelestriangle,but,becauseoftheassociationwithothertri-angles, itmayalsobringtomindideasofequilaterals.Reasoningabouttri-anglesingeneralwouldthenamounttoreasoningwithanideaofaparticular(say,justoneisoscelestriangle)andforthistobeaccompaniedbymuchthesamereasoningwithotherrelatedideas(otherisoscelestriangles,equilater-als,etc).Inotherwords,youstartwithanimageofanindividualandconsiderthe situationwith respect toother imagesofother individuals thatbear aresemblancetothefirstimage.



The second component ofQuine’s account is a similaritymetric.Quine assumes that the fine-grained discriminatory capacitiesare innately ordered in terms of similarity (an innate “spacing ofqualities”), which he interprets behavioristically. “A standard ofsimilarityisinsomesenseinnate.Thispointisnotagainstempiricism;itisacommonplaceofbehavioralpsychology”(1969,p.123).Quine’sinnate similarity metric incorporates a further element of innategenerality,butitalsofacilitateslearning,allowingtheaccounttoavoidthedifficulties that earlier empiricist accountsof abstractionhad incapturingthesimilarityintheinputwithoutgeneralrepresentations.

The third andfinal component ofQuine’s account is a selectionprocess. Quine assumes that learners engage in hypothesis testing,whereovertbehaviors(e. g.,callingacolorsample‘white’)areselectedthrough positive and negative feedback in accordance with theprinciplesofconditioning.Theselectionprocessoperatesintandemwith the innate quality space to isolate a region within that spacecorresponding toaconventional term(e. g., thewhite regionwithintheinnatesimilarityspace).Inthisway,theinnatesimilarityspacecancometobepartitionedinculture-specificways.13

The structural features of Quine’s basic account—innate fine-grainedgenerality,aninnatesimilarityspace,andaselectionprocesstoisolateregionswithinthatsimilarityspace—providethefoundationwehavebeenlookingfortodevelopaworkabletheoryofabstraction.However,thedetailsofQuine’saccountareproblematicinvariousways.ThemostseriousdifficultiesstemfromQuine’sbehaviorism.Considerhisexplanationoftheinnatesimilarityspace.Quine’saccountofwhatitistohaveaninnatesimilarityspaceisessentiallythatweareinnatelydisposedtorespondtocertainstimuliinasimilarmanner.“Aresponsetoaredcircle,ifitisrewarded,willbeelicitedagainbyapinkellipsemorereadilythanbyabluetriangle”(1969,p.123).Thisexplanation,

13. Quinealsoenvisionsmore radicalchanges to thesimilarityspace throughfurtherlanguagelearning,formaleducation,andtheimpactofscience.Onewaytothinkaboutsomeofthesemoreradicalchangesisthattheyalterthecharacterofthesimilarityspaceby,forexample,introducingnewdimensions.

3. A Neo-Quinean Framework

In this section, we present a general framework for understandingabstraction.Aswillbecomeclear,wethinkthatthereisalargefamilyofrelatedacquisitionmodelsthatshareimportantsimilaritiesandareequally deserving of the label abstraction. Since what is interestingfromaphilosophicalpointofviewarethecontoursoftheframeworkrather than the details of any particular model, our aimwill be tosketch thebroadoutlinesof thegeneral framework.Wetakeasourstarting point W.V.O. Quine’s treatment of learning in his paper“NaturalKinds”(Quine1969).WhileQuine’saccountfacessignificantdifficulties,itcanbeadaptedandexpandedinvariouswaystoprovideapromisingbasisforunderstandingabstraction.12Theresultingneo-Quineanframeworkmakesitpossibletoexplainhowabstractioncanaccountfortheoriginsofmanygeneralrepresentationswithoutfallingprey to thedifficulties thatwepresented for traditional accountsofabstractioninsection2.

Quine’sdiscussioniscouchedintermsofanaccountofwordlearning.Hisaccounthasthreemaincomponents.First,Quineassumesthatthelearnercaninnatelydiscriminatearangeoffine-grainedpropertiesinthelearningdomain,forexample,differentshadesofcolorinlearningcolorwordslike‘white’and‘green’.Thesefine-graineddiscriminatorycapacitiesprovidetheinputtotheprocessofabstraction.Bybuildinggenerality(albeitfine-grainedgenerality)infromtheoutsetintheformof general capacities for discriminating shades of color,Quine doeslimit thescopeofabstraction.Hedoesn’t takeabstractiontoexplainthe origin of all general discriminatory capacities. Nonetheless, forQuine,abstractioncanexplainhowageneralwordlike‘white’couldbe learnedon thebasisof thefine-graineddiscriminatorycapacitiesassociatedwithparticularshadesofcolor.

12. WeshouldnotethatQuinedoesn’tdescribehimselfasofferingatheoryofabstraction.Nonetheless,wewilldiscussQuineintheseterms,sincewetakethelearningprocessthatQuinedescribestobeagoodstartingpointforun-derstandingabstraction.



AnotheraspectofQuine’saccountthatshouldbeaddressedisthecharacteroftheselectionprocess.Quinenarrowlyfocusesonasingletypeofselectionprocess(hypothesistestingdrivenbyconditioning).Though a representational version of this suggestion can accountfor the acquisition of general perceptual representations, there arenumerous possibilities for how a selection processmight function,andtheneo-Quineanframeworkshouldbetakentoencompassthefullrangeofsuchpossibilities.Notallmodelswillinvolvehypothesistesting, and among those that do, there will be differences in theassumptionstheymake.Theprocessesinvolvedinisolatingaregionin the innatequalityspacecanrange fromrelativelyunconstrainedprocesses (e. g., summation of positive instances, or hypothesizingsimple regularly shaped regions containing positive instances andexcluding negative instances) to highly constrained processes (e. g.,where hypotheses are drawn from a highly circumscribed set orwherethehypothesisspaceevolvesinaninnatelyspecifiedmanner).Wewillofferexamplesalongtheselinesbelow.Theimportantpointforpresentpurposes is that awidevarietyofoptions are availablefor the selection process, each ofwhich, in its ownway, isolates aregion of the innate quality space in response to the fine-grainedrepresentationsthataretakenasinput.

There are also anumberof other important sourcesof potentialvariation that Quine himself does not discuss but which ought tobe included in the neo-Quinean framework. For example, the fine-grained representations that form the basis of abstraction needn’talwaysbeinnate.Insomecases,theymightbelearned.Likewise,theinnate quality space might not be developmentally fixed. The sizeordimensionsof thisspacemightbealtered.Relationalparameterswithin a quality space might also be altered, or new relationssuperimposedonto the space.There couldalsobemultipledistinctqualityspacesandqualityspaces thatstand indifferent relationsofpsychologicalaccessibilitytooneanother.Takentogether,theseandthepreviouslymentionedsourcesofvariationintroduceconsiderable

however,islittlemorethanarestatementofthephenomenontobeexplained.Itisnobetterthansayingthatwetendtorespondtocertainstimulisimilarly(explanandum)becauseweareinnatelydisposedtorespondtothosestimulisimilarly(explanans).Trueenough,butwhatweneedtoknowiswhypeoplehavethesameresponsetothestimuli.Thisrequires,attheveryleast,theoutlinesofasynchronicmechanism.For this reason, a better account would be one that explains theinnatesenseofsimilarityintermsofaninnatecomputationalprocessoperatingoveraninnateclassoffine-grainedrepresentations,wherefeaturesoftherepresentationsandthecomputationalprocessresultin representations being ordered so as to produce the similarityeffects. Many computational-representational systems are possiblehere,andsothedetailsarebestlefttoempiricalpsychology.Butwewillassumethatsomesuchaccountofsimilarityistherightwaytoproceed,asanaccount thatstickspurely tobehavioraldispositionsisn’tsubstantive.Thisisthefirststepindevelopingtheneo-Quineanframeworkforunderstandingabstraction.Andonceacomputational-representationalsystemisusedtoexplainthesimilarityspace,it’sonlynaturaltoadoptrepresentationalversionsoftheothercomponentsofQuine’saccount—thefine-graineddiscriminatorycapacitiesandtheselectionprocess.Soourneo-Quinean frameworkwillalso includeinnate fine-grained representations and a selection process that isacomputationalprocess—onethatoperatesoveraqualityspaceofrepresentationalstates,notafieldofbehavioraldispositions.14

14. Withoutarepresentationalaccountoftheselectionprocess,wewouldneedanexplanationofwhyreinforcementhasitseffectsonovertbehavior,andwewouldfacedifficultiesarisingfromthefactthattheprinciplesofconditioningdon’tapplytomanyinstancesoflearning,includingwordlearning(Chomsky1959).Citingonlyexternal factors(the impingementofstimuli, the imposi-tionofrewards,etc.)isinadequate,sincetheseclearlydon’thavethesameeffectsoneveryphysicalsystem.Therehastobesomethingabouttheintrin-siccharacterof thelearningsystemthatexplainswhyconditioningshapesitsresponses.Thebestaccountthatpsychologyhastoofferisthat,inmanycases,themechanismisdeeplycognitive.It’sbecauseofthewaythatthecon-tingenciesofrewardsandpunishmentsarerepresentedthattheprinciplesofconditioninghaveanypurchaseonchangesinbehavioralregularities(Gal-listel1990;Gallistel&Gibbon2002).



This is just one example, but notice that such a model avoidsthedifficulties thatwe raised in theprevious section forLockeandothers,anddoessospecificallybyabandoningtheLockeanambitionof trying to explain the origins of all general representations viaabstraction. Instead, the model works by supposing that somegeneral representations are innate (e. g., the fine-grained but stillgeneral representations of particular shades of white). Abstraction,accordingtotheneo-Quineanframework,can’taccountforallgeneralrepresentations, but that is of no matter, since no framework canaccount for all general representations. What this new frameworkdoes do, however, is verymuch in the spirit of traditional theoriesofabstraction,inthatitexplainshowgeneralrepresentationscanbelearnedonthebasisoffine-grainedperceptualexperience.

4. Implications of the Neo-Quinean Framework

We’ve sketched the general outlines of a workable framework forunderstandingabstraction,butmanyquestionsremainregardinghowthe frameworkshouldbedevelopedandregarding the implicationsit has for philosophical theories of the mind. In this section, wemake some programmatic suggestions.We offer these in the spiritofan initialexplorationofapoorlyunderstoodarea that is ripe forphilosophicalattention.Butevenatthisearlystageofinquiry,wethinkthere are some important and perhaps surprising conclusions thatcanbedrawn.Weorganizeourremarksaroundthreegeneralissues:(1) theempiricism-nativismdebate, (2) theoutputof theprocessofabstraction,and(3)humanuniqueness.

(1) Abstraction and the Empiricism-Nativism DebateAbstraction has historically been seen as a distinctively empiricistacquisition process. However, we will argue that there is nothingabout abstraction per se that limits it to an empiricist psychology;abstractionisequallycompatiblewithnativistviewsofthemind.Toseewhy,weneedtostepbackandconsiderthecharacteristicfeaturesofnativismandempiricism.

flexibilitywithin the neo-Quinean framework.15Whilewewon’t beable to systematically explore all these different possibilities, somewillbediscussedbelow.

Insum,theneo-Quineanframeworkthatweareproposingtakesthe following form:Abstraction is a computational-representationallearning process that operates over a quality space of fine-grainedgeneral representations that are ordered by a similarity metric.Abstraction involves a selectionprocess that isolates regions of thequalityspace.Thesimilaritymetricneedn’tbesimple.Infact,itmightbe quite complex andmultifaceted. Likewise, the selection processcantakemanydifferentforms.Butonethingthatallvariationsonthisbasicmodelhaveincommonisthat,bybuildinginenoughstructureright from the outset (some general representations and a suitablesimilaritymetric),thecriticismsthatweresodamagingtotraditionaltheoriesofabstractionareavoided.

If we return to the example of the general representation white, therearenumerousalternativemodelsforhowsucharepresentationmight be acquired in the neo-Quinean framework. One possibility,justtogetthefeeloftheframework,wouldbeamodelmuchlikethecomputational-representationalanalogofQuine’sownaccountofcolorwords.Inthiscase,alearnercomesequippedforthetaskwithgeneralrepresentationsfordifferentshadesofwhite(amongothercolors),aswellasaninnatesimilaritymetricthatorganizeshercolorspace.Then,uponencounteringdifferentinstancesofwhitethings(snowballs,paper,milk,etc.),shewouldrepresentthoseparticularshadesand,throughaprocessofpositiveandnegativefeedback,developarepresentationthatincorporatesalloftheshadesthatreceivedapositivesignalandnoneoftheshadesthatthatreceivedanegativesignal.

15. Alsoopen to investigation is theclassof representations thatmightbeac-quiredbysuchaprocess.Thisislikelytoincludestandardperceptualrepre-sentations(e. g.,representationsforcolors,textures,andodors).Butitmightalso include representations involved in bodily sensations (pleasure, pain,heat, etc.) and representations of cross-modal and amodal categories (e. g.,shapeandspatialrelations),amongothers.



regarding the disagreement between empiricists and nativists, wereturntothecaseofcolor.

Currently, there is a lively debate regarding the extent towhichtheacquisitionofgeneralcolorrepresentations—conceptslikewhite,blue,andgreen—isinnatelyconstrained.Someresearchersviewthelearningofcolorcategoriesinstronglyempiricistterms.Forexample,inarecentreviewoftheliteratureoncolorcategorization,Regier&Kay(2009)provideadescriptionofaviewthatshouldsoundfamiliar:

DebiRoberson and colleagues… concluded that therearenouniversalfoci,thatcategoriesthereforecannotbeorganized around them, and that ‘‘color categories areformedfromboundarydemarcationbasedpredominantlyonlanguage’’…subjecttotheconstraintof‘groupingbysimilarity’:namely,thatcategoriesmustformcontiguousregionsofcolorspace.Theimplicationisthatapartfromthat rather loose constraint, category boundaries aredetermined exclusively by local linguistic convention.(Regier&Kay2009,p.442)

Put in these terms, Roberson et al.’s position bears a strikingresemblancetoQuine’s(minusthebehaviorism).Insupportoftheirview,Robersonetal.point tocross-culturalevidencedemonstratingsignificant variation in color representations. For example, in animportantstudy,Davidoff,Davies,&Roberson(1999)reportthattheBerinmoofPapuaNewGuineausefivebasiccolortermsthatcrosscutthe basic color terms in English; one Berinmo term covers bothyellow (i. e.,what’s called ‘yellow’ inEnglish)andnumerous shadesthatEnglishspeakersthinkofasgreen.OnRobersonetal.’saccount,color representations are learned by identifying different culturallysalientregionswithinacommoninitialsimilarityspace.Sincethereare only weak internal constraints on the learning process, colorrepresentationswillvarysignificantlycross-culturally.

Empiricistsandnativistsdisagreeaboutthewaythatpsychologicaltraits(psychologicalfaculties,states,dispositions,etc.)areacquired.16 Empiricistsmaintain thatmost psychological traits are acquired onthebasisofasmallnumberofgeneral-purposepsychologicalsystems,while nativists maintain that numerous specialized systems areneededaswell.Althoughcommentatorssometimeslosesightofthepoint, both nativists and empiricists appeal to innate psychologicaltraitsinaccountingfortheacquisitionoffurtherpsychologicaltraits.For example, empiricists who are opposed to innate knowledgenonethelesssupposethatbasicpsychologicalfacultiesforperceptionandmemory are innate.Another commonmisunderstanding is thesuppositionthatempiricistsarealoneingivingalargeroletolearning.But nativists aren’t opposed to learning. They just disagree withempiricists about how learning takes place and about the systemsinvolved. Empiricists only invoke general-purpose learning systems(e. g.,principlesofassociation),whilenativistsalsoinvokespecializedlearningsystems(e. g.,aninnatelanguage-acquisitiondevice).

Farmorecouldbesaidabouttheempiricism-nativismdispute,butevenwiththisbriefoutline,itoughttobeclearthatabstractionisn’tintrinsicallyempiricist;nativistversionsofabstractionarealsopossible.Whetheragivenoccurrenceofabstractionshouldcountasempiricistornativistdependsonhowthedetailsarefilledin.Thecrucialfactorshave todowith thecharacterof the innatesimilarityspaceand thetypes of selection processes that are invoked. For instance, wherethe selection process is domain-general and subject to few if anyinnateconstraints,theresultwillbeanempiricistmodel.Butwhereitisdomain-specificandsubjecttosignificantinnateconstraints,theresultwillbeanativistmodel.Toillustratethatabstractionisneutral16. Historically, concerns about the nature and origins of psychological traits

were often conflated with epistemological questions about justification(Cowie1999).Fromacontemporaryperspective,however,itisclearthatjus-tificationisonethingandpsychologyanother.Inprinciple,abeliefthatre-quiresempiricaljustificationcouldbeinnate(e. g.,thebeliefthatspidersaredangerous),whileabeliefthatisjustifiedapriorimightnotbe(e. g.,thebeliefthatarithmeticisincomplete).



colors?Moreover,Regieretal.also foundthat thebestexamplesofcolors across these languagesweremore closely clustered than thecenterpointsofthecolorfieldsassociatedwitheachlanguage’scolorterms. This suggests that the best examples are not simply derivedfrom the color fields associated with the terms but rather that thebestexamplesareprimaryandthecolorfieldsformaroundthem.Anaturalmodelthattakesaccountofthisfactwouldbetohaveinnatefocalcolorsaroundwhichcolorfieldsarebuiltthroughaprocessoflearning.Oranotherpossibilityistohaveinnate focal color fields,wherethebestexamplesofcolorsmustliewithinthesefieldsbutthepreciselocationsareopentolinguisticinfluenceandconsequentlysubjecttocross-culturalvariation.

Ourpurposehereisnottosettletheissueofwhetheranativistoranempiricistmodelprovidesthebestmodelofcolorconceptacquisition.Rather, the illustrations are intended to show that, despite thehistorical affiliation between abstraction and empiricist approachesto representational-conceptualdevelopment, there isnothing in theprocessofabstractionthatexclusivelytiesittoanempiricistpsychology. Empiricistsandnativistsalikecanhelpthemselvestotheprocessofabstraction.Theorists can evenmix andmatch the twoapproachesbyadoptingempiricistprocessesforsomedomainsofabstractionandnativistprocessesforothers.Whatwilldeterminewhethertheprocessisanempiricistornativistoneisn’tmerelywhetherabstractiontakesplacebutratherthecharacteroftheinnatestructureofthesimilarityspacesandtheinnateconstraintsthatguidetheprocessasitunfolds.

(2) The Output of the Process of Abstraction Most of the mind’s representations are complex representations.They have constituent structure in accordance with the principlesof compositional semantics.Theconceptwhite circle, forexample,is composed of the simpler concepts white and circle. Primitiverepresentations, on the other hand, do not have compositionalsemanticstructure.Theyarethesemanticatomsfromwhichcomplexrepresentationsarebuilt.ForLocke,manyoftheproductsofabstraction

However, other evidence suggests that the acquisition of colorrepresentations is guided by strong innate constraints. In animportantearlystudy,Bornstein,Kessen,&Weiskopf(1976)showed4-month-old infants examples of a primary hue until the infantsbegan to lose interest and then showed them novel instances ofthesamehueaswellasequallynovel instances thatcrossedahueboundary.Forexample,infantswerefamiliarizedwithashadeofblue(480-nm light; nm = nanometer) and subsequently shown a novelshade of blue (450-nm light) and an equally novel shade of green(510-nm light). The resultwas that the infants looked significantlylongeratthenovelshadeofthenewhue(green)butnotatthenovelshadeoftheoldhue(blue).Franklin&Davies(2004)haverecentlyreplicatedthesefindingsusingamorerigorousmetricformeasuringdistancesbetweenstimuli.Theyfoundboundariesnotonlybetweenprimary color categories (blue-green) but also between secondarycolor categories (blue-purple). Togetherwith the evidenceof adultvariability from Roberson et al., this evidence suggests that theabstractionprocessmaybeginnotwithanequipotentinnatesimilarityspacewithnocategoryboundariesbutwitha similarity space thatcomeswith its own innatelybounded regions that aremodified inlightof laterexperience.Suchamodel isstill fullycompatiblewiththeneo-Quineanframeworkforabstraction.It’sjustamodelinwhichtheselectionprocessisanativistone,involvingadjustmentsaroundinnatelyspecifiedboundariesinthesimilarityspace.

Otherevidence that suggests that the learningprocess isguidedby nativist constraints points in the direction of a different sort ofnativistmodel. For example,TerryRegier and colleagues examinedcolor naming in 110 languages from nonindustrialized societiesaround theworld (Regier,Kay,&Cook2005).They found that thebest examples of color terms across this diverse sample tended toclusteraroundthebestexamplesoftheEnglishterms‘black’, ‘white’,‘red’,‘yellow’,‘green’,and‘blue’.Onanempiricistmodel,thisishighlysurprising—iftherearenobuilt-inwaystogroupcolors,whyshouldpeopleineveryculturewindupwithhighlysimilarbestexamplesof



we can take the input to the process of abstraction to be a set ofrepresentations of various specific shades within a similarity space(several particular shades of white, corresponding to the colors ofseveralexperiencedwhiteobjects).Aselectionprocessoperatingonthis input results in thedemarcationofafieldwithin this similarityspace (a region in the color space corresponding to whiteness isdelimited). Let’s suppose that this process also generates a newrepresentation,white,thatislinkedtoeachoftherepresentationsintheselectedfieldsuchthattheactivationofanyelementinthefieldbrings about the activation of this new higher-level representation.Nowthesemanticsofthishigher-levelrepresentationcouldbehandledinanumberofdifferentways.Onewaywouldbeforthecontentoftherepresentationtobedeterminedbyitscausaldependenceontheenvironmental conditions that it has the function of responding to(Dretske1995);theinternalrepresentationsforspecificshadeswouldsimply mediate this mind-world link between external conditions(whiteness)andtherepresentationwhite.Elsewhere,wehavecalledsuchmediatingfactorssustaining mechanisms (Margolis1998;Laurence&Margolis2002).Asustainingmechanismdoesn’tdirectlydeterminea representation’s content but indirectly makes its contributionby establishing the mind-world relation that does constitute therepresentation’s content. On such an account, the products of theprocessofabstraction—representationslikewhite, circular,smooth,etc.—wouldhave theircontentdeterminednotcompositionallybutrather by the mind-world relations established by the sustainingmechanisms.18Henceabstractionwouldresultinnewprimitives.Soitlookslikeitispossibletolearnnewprimitiveconceptsviaabstractionontheneo-Quineanframework.18. Noticethat,onthistreatment,therepresentationsofthevariousfine-grained

shades aren’t constituents of the general representation, unlike white andcircleinwhitecircle.Theoristswhooptforsustainingmechanismsratherthanconstituencyrelationsoftendosobecauseitweakenstherelationshipbetweentherepresentations inthesustainingmechanismandtheconceptwhosecontentisindirectlyestablished,thusallowingforthepossessionofagivenconceptacrossagreatdealofperceptualandcognitivevariability(seeLaurence&Margolis1999,Dretske1981,Fodor1987).

seemtohavebeenprimitiverepresentationsofthissort(e. g.,white).This raises the interesting question ofwhether representations thatarelearnedviaabstractionwithintheneo-Quineanframeworkcouldbeprimitive,sinceitiswidelyassumedthatprimitiverepresentationscannotbelearned.AsStevenPinkerdescribestheconsensus:

Onthenurtureside,empiriciststendtomakedowiththeabstemiousinventoryofsensori-motorfeatures,invokingonly the process of association to buildmore complexones.Onthenatureside,nativistsarguethatalargerandmoreabstractsetofconcepts,suchas“cause,”“number,”“livingthing,”“exchange,”“kin,”and“danger,”comeready-maderatherthanbeingassembledonsite.

Bothsides, ifpressed,have toagree that thesimplebuilding blocks of cognition—like the keys on apiano, thealphabet ina typewriter,or thecrayons inabox—must themselves be innate. Type on a standardtypewriter all youwant; though you canbangout anynumber of English words and sentences, you’ll neversee a single character ofHebreworTamil or Japanese.(Pinker2007,p.93)

Accordingtothisbuilding blocks model ofrepresentational-conceptualdevelopment, theprimitive representationsmustbe innate,and therest of our representations and concepts are assembled from theseprimitives.However,ifabstractionoffersawaytolearnnewprimitiverepresentations, then itarguesagainst thebuildingblocksmodel. Itwouldshowthatacompellingandextremelyinfluentialviewabouttheoriginsofconceptsismisguided.17

Oneofthebenefitsofhavinganexplicitframeworkforunderstandingabstractionisthatitrenderssuchquestionstractable.Consideragainthecaseofcolorrepresentations.Giventheneo-Quineanframework,

17. Ifnewprimitivescanbelearnedviaabstraction,thiswouldstrengthenthecasethatwe’vemadeelsewhereagainstthebuildingblocksmodel(Laurence&Margolis2002);seealsoCarey(2009).



whichpreciseshadeisatissue,it’simportantnottofixatetoostronglyonanyparticularshade.

Forpurposesofthispaper,wedon’tneedtosettlethequestionofwhetherabstractedrepresentationsare,infact,primitive.Wesimplywant to call attention to the fact that the neo-Quinean frameworkallowsforthepossibilitythatnewprimitivescanbelearned.Sincestandard theories of development so often suppose that newprimitives must be innate, this is a possibility of considerablephilosophical interest. On the model we have sketched, the neo-Quinean framework would allow us to acquire new primitiveconcepts,therebyincreasingthecombinatorialexpressivepowerofourrepresentational-conceptualsystem.

(3) Is Abstraction Uniquely Human?AsLockeseesthings,theabilitytoformabstract ideasisauniquelyhuman capacity, one that is associated with our linguistic abilities.Locketakesthefactthatanimalsdon’tusepublicsignstobeagoodindicationthattheyaren’tabletohaveanygeneralideasatall:

…thepowerofAbstractingisnotatallinthem;andthatthehavingofgeneralIdeas, isthatwhichputsaperfectdistinctionbetweenManandBrutes;andisanExcellen-cywhichtheFacultiesofBrutesdobynomeansattainto. For it is evident,we observe no foot-steps in them,ofmakinguseofgeneralsignsforuniversalIdeas;fromwhich we have reason to imagine, that they have notthefacultyofabstracting,ormakinggeneralIdeas,sincetheyhavenouseofWords,oranyothergeneralSigns.(1690/1975,II,xi,§10)

Lockeisnotaloneintheseviews.ThomasReid,forone,wholeheartedlyagrees that animals “have not the powers of abstracting andgeneralizing;and that in thisparticular,Naturehasmadea specificdifferencebetweenthemandthehumanspecies”(Reid1785/2002,p.

This sort of model isn’t mandatory, however, and other modelsthatareconsistentwiththeneo-Quineanframeworkwouldhavetheoutputoftheprocessofabstractionbeacomplexrepresentation,nota primitive.Once again, consider the case of color representations.Asbefore,wecantakethe input toabstractionontheneo-Quineanframework to be a set of representations of various specific shadeswithin a similarity space, and a selection process will result in thedemarcationofafieldwithinthesimilarityspace.Thistime,though,wewillsupposethatthisprocessalsogeneratesanewrepresentationthatisadisjunctiverepresentationwhosemanydisjunctsarejusttherepresentations that appear in the demarcatedfield.On thismodel,thesemanticsoftheabstractedrepresentationisplainlycompositional.Thecontentofwhiteisafunctionofthecontentsofitsconstituentsandthecompositionalstructureinwhichtheyinhere.

Both the compositional model and the sustaining mechanismmodelarecompatiblewiththeneo-Quineanframework.Abstractioncanproducecomplexrepresentationsthatincorporatethefine-grainedrepresentations that are the input to the process, or it can producesimplerepresentationsthatareactivated by sustaining mechanismsthatincorporate the fine-grained representations. Nonetheless, severalconsiderationssuggestthatthesustainingmechanismmodelmaybepreferable.Oneisthecomputationalloadforprocessesthatoccurattheleveloftheabstractedrepresentation.If theseprocesseshavetooperateonahighlystructuredrepresentationanddealwitheachofitsnumerous constituents, this is likely toplaceaheavyprocessingburdenonthesystem.Ontheotherhand, if theprocessescanstickto an unstructured representation and ignore all of the structurethat is inherent in its sustaining mechanism, the computationalload would be considerably eased. There may also be advantagesin the informational loss that is inherent to the employment of anunstructuredrepresentation.Forexample,ifwhatmattersinapplyingalearnedruleisthemoregeneralcategorywhite,thenarepresentationthatfocusesattentiononjustthatcategory(andnotonsomeparticularshade)putstheemphasisjustwhereitshouldbe.Ifitdoesn’tmatter



just to the items in thetrainingregimenbutalso tonovel instancesof the category they exemplify. For instance, in a representativeandnowclassic study,Herrnsteinandhis colleagues (1976) trainedpigeonstodistinguishpicturesoftrees.Thesubtletyinvolvedinthesediscriminations is impressive, since the training set is very diverseand the contrasting stimuli are, inmany respects, highly similar toexemplarsofthetargetcategory(e. g.,whileHerrnsteinetal.’spigeonshadtogiveapositiveresponsetoapicturethatshowedjustthetopcornerofatreeinthebackgroundofascene,theyhadtogivenegativeresponsetoapicturethatshowedacelerystockfrontandcenterwithitsleavesintact).Thoughitispossiblethatthegeneralrepresentationtree is innate in pigeons, otherwork leaves nodoubt that pigeonsare capable of learning new general representations. Pigeons havebeen trained to selectively discriminate such artificial categories asautomobiles and chairs (Lazareva, Freiburger,&Wasserman 2004).Theyhave evenbeen trained todiscriminateMonets fromPicassos,and Stravinsky from Bach (Watanabe, Sakamoto, & Waikta 1995;Porter&Neuringer 1984).Our neo-Quinean framework provides aplausibleaccountofhowtheunderlying generalrepresentationsarelearned.Accordingtothisframework,theanimalsinitially representfine-grained(yetfullygeneral)perceptualpropertiesofthestimuliand,throughtraining,cometorepresentbroadercategoriesinapreviouslyestablishedsimilarityspace.

Quinehimself,weshouldpointout,doesrecognizethatnonhumananimals are capable of generalizing. Unfortunately, he draws thewrong moral from this similarity between humans and animals,suggestingthatourapparentlysophisticatedinductiveabilitiesshouldbedowngradedto“ananimal”model.

[O]ther animals are likeman.Their expectations, ifwechooseso toconceptualize theiravoidancemovementsand salivation and pressing of levers and the like, areclearly dependent on their appreciation of similarity.Or to put matters in their methodological order, these

388).Andanumberofcontemporaryphilosophershavepickeduponat least the strandofLocke’s view that ties thenotionof a conceptto language, claiming that animals lackbonafideconceptsbecausethey lack thenecessaryparticipation ina linguisticcommunity (e. g.,Davidson1975,Dummett1994,McDowell1994).19Thoughwecan’tgointotheissuesregardingconceptpossession,wedowanttoaddressthequestionofwhetheranimalsmustlackgeneralrepresentationsand,moresignificantly,whetherthecapacityforabstractionasunderstoodintheneo-Quineanframeworksetsusapartasaspecies.

Tobegin,weshouldnotethatitisbyapplyinggeneralrepresentationstoindividuals,andbyrelatingonegeneralrepresentationtoanother,thatagentsareabletodrawinferences,formexpectations,andlearnfromexperience.Mostanimalswouldnotsurviveverylongwithoutthem.Adeermightmanagetoquenchitsthirstwhendrinkingfroma pool of water, but no matter howmany pools it drinks from, itwouldn’t be able to infer that thenextpool is also able toquenchitsthirst.Similarly,awildebeest thatescapesa lion’sattackorevenmultiple attacks wouldn’t have the wherewithal to infer that thenext lionought tobe avoidedbecause it too is dangerous. So it isunsurprisingthenthatpsychologistshavedocumentedthatgeneralrepresentations are widely distributed in the animal kingdom. AsRichardHerrnsteinnotesinareviewandanalysisofworkonanimals,categorization and hence general representation “has turned up atevery level of the animal kingdomwhere it has been competentlysought”(Herrnstein1990,p.138).

Infact,oneofthecentralprojectsinanimalpsychologyhasbeentodeterminewhether,andtowhatextent,differentspeciesarecapableof discriminating sundry categories. Researchers routinely trainanimalsonnaturalandartificialstimulitoseeiftheycanrespondnot

19. Interestingly,Locke’sclaimthatanimalsdonotuseanypublicsignsturnsouttobefalse.Thoughanimalsdon’tpossessanythingasrichashumannaturallanguage, thereare specieswhose systemsof communication includepub-lic signs thatareunderananimal’s control, includingnonhumanprimates(Cheney&Seyfarth1990),meerkats (Manser2001),andeven thehumblechicken(Evans,Evans,&Marler1993).



A natural question, at this point, is to ask what other types ofprocesses, besides neo-Quinean abstraction, might support theacquisition of general representations. This question is closelyconnectedwithanumberofimportantphilosophicalissues,includingtheinfluenceoflanguageonthought,theinnatestructureofthemind,theoriginsofhumancreativity, and thenatureof theory change inscience.Howeverthesearetobesettled,inourview,thereisnoonekey acquisition system responsible for representational-conceptualdevelopment;humanrepresentationalandconceptualsystemsstemfromahighlyvariedcollectionofsystemsofacquisition.Likewise,thedifferencebetweenhumanandanimalmindsdoesnotdependonasinglepowerfulsourcefromwhichalluniquelyhumanrepresentationsderivebutinsteaddependsonaneclecticpotpourriofsources.

Justtogiveaflavorofthisdiversity,wewillmentiontwoproposalsabouthowhumans are able to acquire certain representations viacognitive resources that animals lack. The first is Susan Carey’sproposalthatmanyconceptscanonlybelearnedviaaprocesssherefers toasbootstrapping (Carey2009).Bootstrappingoccurswhenanagentreliesonanuninterpretedorpartially interpretedsymbolsystemwhosesymbolsactasplaceholdersfortherepresentationstobelearned.InterpretingthesystemisachievedthroughwhatCareycallsmodeling processes.Thesetypically involvedrawingananalogybetween two systems of representation, the uninterpreted systemandasystemthatalreadyhassomemeaningforthelearner.Carey’sflagship example of bootstrapping is an account of how childrenlearn the positive integers.On this account, children first have tolearnthecountingprocedureasameaninglessroutineandalsohavetodirectlypickuponthemeaningsofthefirstfewcountterms.Then,afteraprotractedperiodandmucheffort, childrencome toseeananalogybetweenthecognitivemodelstheyuseinconnectionwiththefirstfewcounttermsandwhathappenswiththesequenceinthecountlist.Theideaofnext word ismappedontotheideaofadding a single individual to a set.Careysuggeststhatanimalsaren’tcapableof learning in this way, since they lack the ability to work with

avoidance movements and salivation and pressing oflevers and the like are typical of what we have to goon in mapping the animals’ appreciation of similarity,their spacing of qualities. Induction itself is essentiallyonly more of the same: animal expectation and habitformation.(Quine1969,pp.124–5)

Quine gets things exactly backwards here, attempting to reduce asophisticated representationalability inhumans tosomethingmorebrute in the form of an unexplicated notion of animal expectation.Contrary to what Quine suggests, inductive inference in humansrequiresasubstantiveexplanation,onethatimplicatesrepresentationalstatesandprocesses.And,forthemostpart,animalexpectationmustbeunderstoodonthehumanmodelintermsofrepresentationalstatesandprocesses.Quineseemstobesuccumbingtothetendency,notedabove,tobecontentwithasuperficialtreatmentofordinarymentalphenomena.Butordinarymentalphenomena,whetherinhumansorinanimals,maskagreatdealofcomplexitythatourexplanationsneedtoregisteranddojusticeto.

Inanyevent,humansarebynomeansspecial in theirability torepresentgeneralcategories,nor, inall likelihood, toarriveat themviaabstraction.Ofcourse,thisdoesn’tmeanthatanimalsarecapableof developing the very same general representations as humans. Itoughttobeclearenoughthathumanscandevelopalargeassortmentof representations that are unavailable to other animals. In somecases, thesemay be representations that do indeed require naturallanguage, since they depend upon culturally acquired informationthatcannotbeconveyedinanyotherway.Inothercases,theymayberepresentationsthataregroundedindomain-specificrepresentationalsystemsthatarethemselvesuniquetothehumanlineage.Regardless,itshouldn’tbecontroversialthatgeneralrepresentationsaren’tallonapar. It’sonethingtohaveageneralrepresentationlikewhiteandquiteanothertohaveageneralrepresentationlikeproton.



hasimportantphilosophicalimplications.Itspeakstosuchquestionsashowtounderstandtheempiricism-nativismdispute,whatkindofstructureconceptshave,andwhetherhumanshaveuniqueconcept-formingabilities.Butmostimportantly,theneo-Quineanframeworkoffersanexplicit treatmentofanotherwisemysteriousprocess,andbecauseofthisexplicitness,itallowsfortheformulationofarangeofrealisticpossibilitiesregardingconceptacquisition.

5. Conclusion

Oneofthecentralprojectsinthephilosophyofmindistoexplaintheoriginsofourrepresentationalcapacities.Theaimof thispaperhasbeentoclarifyoneimportantpartoftheexplanation—theprocessofabstraction—by providing an explicit framework for understandinghowitworks.JustasLockesupposed,generalrepresentationscanbelearned via a process that beginswith fine-grained experience thatarises throughcontactwithparticulars.Abstractioncanstill explaintheacquisitionofrepresentationswithgreatergeneralityfrommorefine-grainedones,and it canstill explain theacquisitionofabroadrange of different kinds of general representations. However, ontheneo-Quinean framework thatwehavepresented, theprocessofabstractiondiffersfromthetraditionalempiricistpictureinanumberofimportantrespects.Unlikethetraditionalnotionofabstraction,itisperfectlyconsistentwithanativistpsychologyandappliestohumansandanimalsalike.Butthesedeparturesfromthetraditionalempiricistpicture are innowaydeficits of theneo-Quinean framework.Theyareadvantages,givingtheframeworkgreaterflexibilityandbroaderapplicability. Perhaps the most significant departure from thetraditional empiricist picture is that the neo-Quinean frameworkrequires a certain amount of general representation to be presentfrom the start, so it cannot explain the acquisition of all generalrepresentations.Butthisisnotadeficitoftheframeworkeither,sinceno account cando that.We conclude thatwhile abstraction cannotbe thewhole storyabout theoriginofgeneral representations, it isnonethelessonecentralandimportantpartofthestory.

uninterpretedsymbolsystemsandengageinthemodelingprocessesthat render themmeaningful. If she is right,bootstrappingmaybeanimportantpartoftheexplanationofwhywehumanbeingshavea conceptual systemwhose expressive power far exceeds what isfoundelsewhere.

The other proposal we wish to mention is one that we havedeveloped in previous work (Margolis 1998, Laurence & Margolis2002). On this approach, some concepts depend upon an innatetemplatethatunderliestheacquisitionofarangeofconceptsinagivendomain.Onemodel that illustrates thisapproachhas it thathumanbeingshaveatemplateforanimalsorlivingkindsthatcontainsslotsforinformationregardingpropertiesthatarehighlyindicativeofkind-membership—shape,colormarkings,characteristicmotion,etc.Whenalearnerconfrontsanewtypeofanimal,theinformationrequiredbythetemplateisassociatedwithanewrepresentationwhoseprocessingisconstrainedbyadispositiontotreatkind-membershipasamatterofhavinganunderlyingnaturethatisresponsibleforthekind’smoreaccessibleproperties.We’vearguedthat togetherthesecomponentscan establish themind-world causal relation that is constitutive ofa concept’s content according to an information-based semanticsapproach.Asimilaraccountcanbedevelopedforartifactconcepts.Inthiscase,theconstraintonprocessingisperhapsadispositiontodeferto the creator’s intent regarding issues of kind-membership (Bloom1996).Soanotherwaythatageneralconceptmightbeacquiredisforthistypeofcognitivemachinerytobeengagedwhenalearnerseesanewitemthatfallsunderthepurviewofaninnatetemplate.Andwhileanimals may share some of the cognitive machinery that supportsconceptacquisitionviainnatetemplates,itisdoubtfulthattheyhavetheverysametemplatesorallofthecognitivedispositionsthatturnour templates into themanynaturalkindandartifact concepts thatoccupymuchofhumanthought.

Alotmorecouldbesaidabouttheneo-Quineanframework,butwehopethatthesebriefremarksindicatethatitstreatmentofabstraction



Dummett,M.(1994).Origins of Analytical Philosophy.Cambridge,MA:HarvardUniversityPress.

Evans,C.S.,Evans,L.,&Marler,P.(1993).Onthemeaningofalarmcalls: functional reference in an avian vocal system. Animal Behaviour,46.1,23–38.

Fodor, J.A. (1987). Psychosemantics: The Problem of Meaning in the Philosophy of Mind.Cambridge,MA:MITPress.

Franklin,A.,&Davies,I.R.L.(2004).Newevidenceforinfantcolourcategories.British Journal of Developmental Psychology,22.3,349–377.

Gallistel, C.R. (1990).The Organization of Learning. Cambridge,MA:MITPress.

Gallistel, C.R., & Gibbon, J. (2002). The Symbolic Foundations of Conditioned Behavior.Hillsdale,NJ: LawrenceErlbaumAssociatesInc.

Goodman,N. (1972). Seven strictures on similarity. InN.Goodman,Problems and Projects.NewYork:Bobbs-Merrill.

Herrnstein,R.J.,Loveland,D.H.,&Cable,C.(1976).Naturalconceptsin pigeons. Journal of Experimental Psychology: Animal Behavior Processes,2.4,285–302.

Herrnstein, R.J. (1990). Levels of stimulus control: A functionalapproach.Cognition,37.1–2,133–66.

Hume, D. (1739/1978). A Treatise of Human Nature. Oxford: OxfordUniversityPress.

Laurence,S.,&Margolis,E.(1999).Conceptsandcognitivescience.InE.Margolis&S.Laurence(eds.),Concepts: Core Readings.Cambridge,MA:BradfordBooks/MITPress.

Laurence,S.,&Margolis,E.(2002).Radicalconceptnativism.Cognition, 86.1,25–55.

Lazareva, O.F., Freiburger, K. L, & Wasserman, E.A. (2004).Pigeons concurrently categorize photographs at both basic andsuperordinatelevels.Psychonomic Bulletin & Review,11.6,1111–7.

Locke,J.(1690/1975).An Essay Concerning Human Understanding.EditedbyP.H.Nidditch.Oxford:OxfordUniversityPress.

Lowe,E.J.(1995).Locke on Human Understanding.London:Routledge.

Bibliography

Adams,R.M.(1975).Wheredoourideascomefrom?—Descartesvs.Locke. In S. Stich (ed.), Innate Ideas. Berkeley, CA: University ofCaliforniaPress.

Berkeley,G. (1710/1975).A Treatise Concerning the Principles of Human Knowledge. In M.R. Ayers (ed.), G. Berkeley, Philosophical Works.Totowa,NJ:Rowman&Littlefield.

Bloom, P. (1996). Intention, history, and artifact concepts.Cognition, 60.1,1–29.

Bornstein,M.H.,Kessen,W.,&Weiskopf,S.(1976).Colorvisionandhuecategorizationinyounghumaninfants.Journal of Experimental Psychology: Human Perception and Performance,2.1,115–129.

Campbell,K.(1990).Abstract Particulars.Oxford:BlackwellPublishers.Carey, S. (2009).The Origins of Concepts. Oxford:OxfordUniversity

Press.Cheney,D.L.,& Seyfarth, R.M. (1990).How Monkeys See the World:

Inside the Mind of Another Species.Chicago:University ofChicagoPress.

Chomsky,N.(1959).AreviewofB.F.Skinner’sVerbal Behavior.Language, 35.1,26–58.

Chomsky, N. (2006). Language and Mind, 3rd edition. New York:CambridgeUniversityPress.

Cowie, F. (1999). What’s Within?: Nativism Reconsidered. New York:OxfordUniversityPress.

Daly,C.(1994).Tropes.Proceedings of the Aristotelian Society,94,253–61.Dancy,J.(1987).Berkeley: An Introduction.Oxford:BlackwellPublishers.Davidoff,J.,Davies,I.,&Roberson,D.(1999).Colourcategoriesina

stone-agetribe.Nature,398.6724,203–4.Davidson,D. (1975).Thoughtand talk. Inhis Inquiries into Truth and

Interpretation.Oxford:OxfordUniversityPress.Dretske, F. (1981).Knowledge and the Flow of Information. Cambridge,

MA:MITPress.Dretske,F.(1995).Naturalizing the Mind.Cambridge,MA:MITPress.



Mackie, J.L. (1976). Problems from Locke. Oxford: Oxford UniversityPress.

Manser, M.B. (2001). The acoustic structure of suricates’ alarmcallsvarieswithpredatortypeandthe levelofresponseurgency.Proceedings of the Royal Society B,268.1483,2315–24.

Margolis,E.(1998).Howtoacquireaconcept.Mind & Language,13.3,347–69.

McDowell, J. (1994). Mind and World. Cambridge, MA: HarvardUniversityPress.

Pinker,S.(2007).The Stuff of Thought: Language as a Window into Human Nature.NewYork:AllenLane.

Porter,D.,&Neuringer,A. (1984).Musicdiscriminationbypigeons.Journal of Experimental Psychology: Animal Behavior Processes, 10.2,138–48.

Quine, W.V.O. (1969). Natural kinds. In his Ontological Relativity & Other Essays.NewYork:ColumbiaUniversityPress.

Regier,T.,Kay,P.,&Cook,R.S.(2005).Focalcolorsareuniversalafterall.Proceedings of the National Academy of Sciences of the United States of America,102.23,8386–91.

Regier,T.,&Kay,P.(2009).Language,thought,andcolor:Whorfwashalfright.Trends in Cognitive Sciences,13.10,439–46.

Reid,T.(1785/2002).Essays on the Intellectual Powers of Man.EditedbyD. Brookes & K. Haakonssen. Edinburgh: Edinburgh UniversityPress.

Russell,B.(1912).The Problems of Philosophy. Oxford:OxfordUniversityPress.

Scholl,B.J.(2001).Objectsandattention:Thestateoftheart.Cognition, 80.1–2,1–46.

Watanabe,S.,Sakamoto,J.,&Wakita,M.(1995).Pigeons’discriminationof paintings by Monet and Picasso. Journal of the Experimental Analysis of Behavior,63.2,165–74.

Documents

Abstraction and the Origin of General Ideas