Analog Magnitude Representations

download Analog Magnitude Representations

of 39

Transcript of Analog Magnitude Representations

  • 7/30/2019 Analog Magnitude Representations

    1/39

    AnalogueMagnitudeRepresentations:APhilosophicalIntroduction

    JacobBeck

    YorkUniversity,Toronto

    [email protected]

    Empiricaldiscussionsofmentalrepresentationpositawidevarietyofrepresentational

    kinds.Someofthesekinds,suchasthesententialrepresentationsunderlyinglanguageuse

    andthepictorialrepresentationsofvisualimagery,arethoroughlyfamiliartophilosophers.

    Others,suchascognitivemaps,aresomewhatlessfamiliar.Stillothershavereceived

    almostnoattentionatall.Includedinthislattercategoryareanaloguemagnitude

    representations(AMRs).1

    AMRsareprimitiverepresentationsofspatial,temporal,numerical,andrelated

    magnitudes.Theyareprimitivebecausetheyrepresentmagnitudeswithoutpresupposing

    theabilitytorepresentanyunitsofmeasurementormathematicallydefinedsystemof

    numbers.AMRsarealsoprimitiveontogeneticallyandphylogenetically.Theyarepresent

    insix-month-oldhumaninfants,awidevarietyofmammals,manybirds,andatleastsome

    fish.Inhumanadultswithaformaleducation,AMRsexistalongsideculturallyacquired

    representationsofspace,time,andnumber.AMRsareanalogueinaspecialsensetobe

    explainedinSection2.

    1WhileAMRsarerarelymentionedinthephilosophicalliterature,thereareafewnotableexceptions.(LaurenceandMargolis[2005])discussesAMRsinthecontextofassessingtheinfluenceofnaturallanguageontheacquisitionofnaturalnumberconcepts;(Pietroskietal.[2009])appealstoAMRstoanalysethesemanticsofthewordmost;(Burge[2010])discusseshowAMRsofnumerosityarerepresentedinperception;and(Beck[2012])arguesthatAMRsserveasanexampleofcognitiverepresentationsthatarenotsystematicallyrecombinableandthushavenonconceptualcontent.SofarasIknow,however,thispapermarksthefirstattempttoprovideageneral,philosophicallyorientedanalysisofAMRs.

  • 7/30/2019 Analog Magnitude Representations

    2/39

    2

    Thenatureofarepresentationalkindisdeterminedbyitsformat,itscontent,and

    thecomputationsitsupports.Specifyingthenatureofanygivenrepresentationalkindis

    alwayscontroversial,butwehaveatleastaroughgraspofthenaturesofsome

    representationalkinds.Thus,toafirstapproximation:therepresentationsunderlying

    languageusehaveasentence-likeformat,topic-neutralcontents,andsupportsyntacticand

    logicalcomputations;therepresentationsunderlyingvisualimageryhaveapicture-like

    format,representvisiblepropertiesaslaidoutinegocentricspace,andsupport

    computationssuchasrotationandscanning;andcognitivemapshaveageometricformat,

    representthelocationsofentitiesinallocentricspace,andsupportcomputationsrelatedto

    navigationsuchaslocalizationandpath-planning.2AnanalysisofthenatureofAMRs

    shouldlikewiseprovideinsightintotheirformat,content,andcomputations.

    AMRsareoverdueforanalysis.Thereisnowconsiderableevidencethattheyexist

    andplayanimportantroleinthecognitivelivesofawiderangeofanimals,including

    humans.Giventheirrelevancetotherepresentationofcategoriessuchasspace,time,and

    numberthathavelongfeaturedcentrallyinphilosophicaldiscussionsofcognition,AMRs

    oughttobeofintrinsicinteresttophilosophers.Theyshouldalsointerestphilosophers

    becausetheybearuponarangeoffoundationalissuesinthephilosophyofmind,including

    howmentalrepresentationsarerealizedinthebrain,theproperanalysisofanalogue

    representation,therequirementsonrepresentationalcontent,theexistenceof

    nonconceptualcontent,andthenatureofanimalcognition.Thispaperaimstointroduce

    AMRstoaphilosophicalaudiencebyreviewingevidencefortheirexistence(1)and

    2(Fodor[1975])and(Block[1983])containclassicphilosophicaldiscussionsofthenatureoflinguisticandimagisticrepresentations,respectively.(Rescorla[2009])containsanexcellentphilosophicalanalysisofcognitivemaps.

  • 7/30/2019 Analog Magnitude Representations

    3/39

    3

    carefullyanalysingtheirformat(2),theircontent(3),andthecomputationstheysupport

    (4).

    1.Background

    1.1EvidenceofAMRs

    Oneinterestinglessontoemergefromthestudyofcognitionoverthepastfewdecadesis

    thateducatedhumanadultsarenotaloneinrepresentingmagnitudessuchasdistance,

    area,duration,numerosity,3andrate.Awiderangeoforganisms,includingfish,birds,

    lowermammalssuchasrats,non-humanprimates,humanchildrenincludingpre-

    linguisticinfants,andhumanadultsfromallculturesandeducationalbackgrounds

    representsuchmagnitudesaswell(forreviewsseeGallistel[1990],Dehaene[2011],and

    Walsh[2003]).Ibeginbyrehearsingafewhighlightsfromthisliterature.

    WhenGodinandKeenleyside([1984])fedaschoolofcichlidfishatdifferingrates

    fromthreeseparatetubes,thefishquicklyapportionedthemselvesinratiosthatmatched

    thefeedingrates,evenbeforemanyofthemhadhadachancetoberewardedbytasting

    thefood.Thus,merelyseeingonetubereleasefoodmorselsattwicetherateofanother

    wassufficienttoleadtwiceasmanyfishtocongregateinfrontofthefirsttube.Anatural

    explanationisthatthefishrepresenttherateatwhichfoodisdispersedfromeachtube

    andadjusttheirpositionsaccordingly.Fromabiologicalperspective,thisprobability

    matchingbehaviourmakessensesinceitisevolutionarilystable.Iffishgenerallyfedonly

    atthesourcewiththegreatestpayoffthenanonconformistfishthatinsteadfedatasource

    3Numerosityisanumber-likeproperty.Idiscussthedifferencebetweennumberandnumerosityin3.2.

  • 7/30/2019 Analog Magnitude Representations

    4/39

    4

    withaslightlylowerpayoutbutnocompetitorspresentwouldrecovermorefood.Natural

    selectionthusencouragesgroupforagerstoadoptaprobability-matchingstrategy.

    Probabilitymatchingiswidelyobservedthroughouttheanimalkingdom.Ina

    similarexperimentthatwasperformedonducksbyHarper([1982]),twoexperimenters

    tossedmorselsofbreadtotheducksattwodifferentrates.Withinaminute,theducks

    dividedthemselvesinproportiontothoserates.Butwhenoneexperimentertossed

    morselsthatweretwicethesizeofthemorselstossedbytheotherexperimenter,theducks

    alteredtheirstrategy,andwithinfiveminutesrepositionedthemselvesinproportiontothe

    productofthemorselsizeandfeedingrate.Gallistel([1990],p.358)summarizesthis

    findingasfollows.

    Thisresultsuggeststhatbirdsaccuratelyrepresentrates,thattheyaccuratelyrepresent

    morselmagnitudes,andthattheycanmultiplytherepresentationofmorselsperunittime

    bytherepresentationofmorselmagnitudetocomputetheinternalvariablesthatdetermine

    therelativelikelihoodoftheirchoosingoneforagingpatchovertheother.

    Inotherwords,Gallistelinterpretstheseresultsasshowingthatducksnotonlyrepresent

    magnitudessuchasrateandphysicalsize,butalsoperformcomputationssuchas

    multiplicationoverthoserepresentations.

    Laboratorystudiesofratspaintasimilarpicture.ChurchandMeck([1984])trained

    ratstopressoneleverinresponsetoatwo-secondsequenceoftwotonesandasecond

    leverinresponsetoaneight-secondsequenceofeighttones.Theythenvariedeitherthe

    durationofthetoneswhileholdingnumberconstant,orthenumberofthetoneswhile

    holdingdurationconstant.Inbothcases,theratsgeneralizedfromthetrainingexperiment.

    Forexample,whennumberwasheldconstantatfourtones,theypressedthefirstleverin

  • 7/30/2019 Analog Magnitude Representations

    5/39

    5

    responsetoatwo-orthree-secondsequenceandthesecondleverinresponsetoafive-,

    six-,oreight-secondsequence,thusshowingthattheyrepresentedthetotaldurationof

    eachtonesequence.Butwhendurationwasheldconstantatfourseconds,theypressedthe

    firstleverinresponsetotwoorthreetonesandthesecondleverinresponsetofive,six,or

    eighttones,thusshowingthattheyalsorepresentedhowmanytoneswereineach

    sequence.Theratsalsoimmediatelygeneralizedtheirlearningtonewstimulipresentedto

    newsensorymodalities.Forexample,whenpresentedwithatwo-oreight-secondlight

    sequence,theypressedthesamelevertheyweretrainedtopressinresponsetoatwo-or

    eight-secondtone;andwhenpresentedwithtwooreightflashesoflight,theratspressed

    thesamelevertheyweretrainedtopressinresponsetotwooreighttones.Thesecross-

    modaltransferexperimentsprovideevidencethattheratsrepresentabstractdurations

    andnumericalinformationandnotjustmodality-specificstimuli.

    Whileitwillhardlybenewstothereaderthathumanadultskeeptrackofdurations,

    distances,numerosities,andothermagnitudes,itmaycomeasasurprisethathuman

    infantsdothesame,suggestingthatsuchabilitiesareinnate.XuandSpelke([2000])

    presentedsix-month-oldinfantswithadisplayof8or16dotsuntiltheyhabituatedtothe

    display.Theinfantswerethenpresentedwithatestdisplayofeither8or16dots.Infants

    whowerepresentedwithatestdisplaythathadanovelnumberofdotslooked

    significantlylongerthaninfantswhowerepresentedwithatestdisplaythathadthesame

    numberofdotsasthehabituationdisplay.Asinfantsareknowntoapportiontheir

    attentiontostimulitheydeemnovel,andasothervariableswerecontrolledfor(suchas

    totaldotareaanddotdensity),thisfindingsuggeststhatinfantscanmakediscriminations

    basedonnumericalinformation.Otherdishabituationstudiessuggestthatinfantsare

  • 7/30/2019 Analog Magnitude Representations

    6/39

    6

    likewisesensitivetoduration(VanMarleandWynn[2006])andarea(Brannonetal.

    [2006]).

    1.2WebersLaw

    Apeculiarfeatureofmagnitudediscriminationsdeservesemphasis:theyobeyWebers

    Law,whichholdsthattheabilitytodiscriminatetwomagnitudesisdeterminedbytheir

    ratio.Astheratiooftwomagnitudesapproaches1:1theybecomehardertodiscriminate,

    andbeyondacertainthresholddeterminedbythesubjectsWeberconstanttheycannot

    bediscriminatedatall.Asanillustrationofthisratiosensitivity,consideragainthe

    experimentsofChurchandMeck([1984])inwhichtheratsweretrainedtopressthefirst

    leverinresponsetoatwo-secondsequenceoftwotonesandthesecondleverinresponse

    toaneight-secondsequenceofeighttones.Whenpresentedwithafive-secondsequenceof

    fivetones,theratstendedtopressthesecondlever,foralthoughfiveisequidistant

    betweentwoandeight,theratioof5:2isgreaterthantheratioof8:5.Whentheywere

    presentedwithafour-secondsequenceoffourtones,however,theratsfavouredneither

    lever.Since8:4=4:2,fourwasthesubjectivemidpointfortheratsbetweeneightandtwo.

    Whenexplicitcountingisnotpossible,themagnitudediscriminationsofhuman

    adultsalsoexhibittheratiosensitivityassociatedwithWebersLaw.Forexample,iftwo

    displaysofdotsareflashedtooquicklyforyoutoseriallycountthedotsoneach,your

    probabilityofcorrectlyguessingwhichdisplayhasmoredotswilldecreaseastheratio

    approaches1:1,andbeyondacertainratio(roughly7:8),yourguesseswillbeatchance

    (Barthetal.[2003]).Similarly,incultureswhereexplicitcountingisabsentandthereisno

  • 7/30/2019 Analog Magnitude Representations

    7/39

    7

    culturallyacquiredintegerlist,magnitudediscriminationsarealmost4alwayssubjectto

    WebersLaw(Gordon[2005];Picaetal.[2005]).Theseresultsareconsistentwiththe

    hypothesisthatmosteducatedhumanadultshavetwosystemsforrepresenting

    magnitudes:aprimitiveanaloguemagnitudesystemthatissharedwithawiderangeof

    otherorganisms;andauniquelyhumanmagnitudesystemthatdependsuponthe

    acquisitionofconceptsofnaturalnumbersandunitsofmeasurement.

    1.3ScepticismaboutAMRs

    Itisnaturaltowonderwhetherthereisasimplerexplanationoftheresultsreviewedin

    thissectionthatdoesnotappealtoAMRsanexplanationthatappealstoless

    sophisticatedrepresentationsorperhapstonorepresentationsatall.Afterall,researchers

    havebeenleddownthegardenpathbefore.Intheearly20thCentury,manywere

    convincedthatthehorseCleverHanscouldperformarithmeticsincehewouldstomphis

    hooftheappropriatenumberoftimesinresponsetoproblemsposedbyonlookers.Butthe

    psychologistOskarPfungst([1965])demonstratedthattherewasasimplerexplanation:

    Hanswaspickinguponsmallunintentionalmovementsinhisaudiencethatindicated

    whentostartandstopstomping.Similarly,whilethewaggledanceofhoneybeesisoften

    takentorepresentthedistancetoafoodsource,thereisevidencethatthedancecorrelates

    lesswithdistancethanwithopticflowtheamountanimagemovesovertheretina

    suggestingthatbeesdontrepresentdistanceassuchafterall(Eschetal.[2001]).These

    4Whyalmostalways?Tworeasons:(1)discriminationerrorsarelesscommonthanWebersLawpredictsforsetswithfourorfewermembers(thisisonefindinginsupportoftheexistenceofaseparatesmallnumberorobjectfilesystem);and(2)thereissomeevidencethatwhenitemscanbelinedupandputinone-to-onecorrespondence,participantscanbeinducedtorelyonHumesprincipletodeterminewhethertwosetsareequinumerous.

  • 7/30/2019 Analog Magnitude Representations

    8/39

    8

    examplesshowthatcareneedstobetakenbeforerepresentationsofmagnitudesare

    posited.Alternativehypothesesneedtobeconsideredandruledout.

    InthecaseofthecontemporarystudyofAMRs,experimentershavetakengreatcare

    toavoidobserverexpectancyeffects.ManyofthetechniquespioneeredbyPfungsttotest

    Hansarenowroutinelyappliedincomparativepsychology.Forexample,experimental

    assistantsareoftenkeptblindaboutcrucialaspectsoftheexperimenttopreventthem

    frominadvertentlysignallingtheanimalswithwhichtheyinteract.StudiesofAMRsalso

    routinelycontrolforotheralternativehypotheses.Forinstance,ChurchandMeckscross-

    modaltransferexperimentswereinpartdesignedtotestwhetheranimalsrepresent

    simplesensoryproperties(suchasretinalareaortotalillumination)ratherthanabstract

    propertiessuchasnumerosityandduration.Thefactthatratsareabletotrackmagnitudes

    acrossdisparatetypesofstimulipresentedtodisparatesensorymodalitiessuggeststhat

    ratsrepresentmorethansensoryproperties.Theoverallbodyofempiricalevidencethus

    stronglysuggestsevenifitdoesnotapodicticallyprovethatmagnitudesare

    represented.Moreover,whiletherearecomprehensiveandmathematicallyrigorous

    explanationsofthefindingsdiscussedinthissectionthatappealtoAMRs(e.g.,Gallistel

    [1990]),tothebestofmyknowledgetherearenocorrespondingexplanationsofthese

    resultsinnon-representationalterms,orintermsofrepresentationsofnon-magnitudes.I

    thustaketheexistenceofAMRsasareasonable,yetdefeasible,empiricalhypothesis.5

    5SomephilosophersmaywonderhowtheoriesthatpositAMRsrelatetotheoriesofthemindthatarentsquarelyinthesymbolicmould,suchasbehaviourism,connectionism,anddynamicalsystemstheory.TheoriesthatpositAMRsareincompatiblewithbehaviourismsincetheypositinternalrepresentationsthatarenotgovernedbyprinciplesofassociationthatcharacterizeclassicalorinstrumentalconditioning.Infact,(Gallistel[1990])and(GallistelandGibbon[2002])arguethatmanyphenomenathathavetraditionallybeenexplainedthroughconditioningarebetterexplainedbyarithmeticcomputationsoverAMRs.Bycontrast,becauserepresentationscanberealizedinconnectionistnetworks(eveniftheyareoftendistributedacrossthosenetworks),thereisnoconflictbetweenconnectionismandtheexistenceofAMRs.Someresearchers

  • 7/30/2019 Analog Magnitude Representations

    9/39

    9

    2.Format

    2.1Careysanalogy

    MostresearchersagreethattheratiosensitivityassociatedwithWebersLawisevidence

    thatmagnituderepresentationshaveaspecificformat:theyinvolvesomeneuralentitythat

    isadirectanalogueofthemagnitudeitrepresents(henceAMR).Astheratiooftwo

    magnitudesapproaches1:1,theneuralentitiesbecomeincreasinglysimilarandthus,

    assumingthereissomenoiseinthesystem,hardertodiscriminate.Toexplainwhy

    researchersbelieveAMRshaveananalogueformat,SusanCarey([2009],p.118)provides

    ahelpfulanalogytothefollowingexternalsystemofanaloguenumberrepresentations.

    number symbol

    1: __

    2: ____

    3: ______

    4: ________

    7: ______________

    8: ________________

    Inthissystem,linelengthisadirectanalogueofnumber.Thegreaterthenumber

    represented,thelongertheline.Supposeourbrainsdeploymagnituderepresentationsthat

    arelikewiseanalogue.Thenjustasitisharderforustovisuallydiscriminate______________

    from________________than__from____,wewouldexpectourbrainstofinditharderto

    haveevendevelopedconnectionistmodelsfortheimplementationofAMRs(ChurchandBroadbent[1990];DehaeneandChangeux[1993]).Inprinciple,AMRsarealsocompatiblewithdynamicalsystemstheorysinceasystemsbeingdescribabledynamicallyneednotprecludeitsbeingdescribableintermsofrepresentationsandcomputations.Tothebestofmyknowledge,however,noonehasdevelopeddynamicalmodelsofthesortsofbehavioursthatAMRsarepositedtoexplain.

  • 7/30/2019 Analog Magnitude Representations

    10/39

  • 7/30/2019 Analog Magnitude Representations

    11/39

    11

    insidetheintraparietalsulcusthatcorrelateswithmagnitudediscriminationsinhumans

    andotherprimatesregardlessofthemodalitythroughwhichthemagnitudeisperceivedor

    represented(Dehaeneetal.[2003]).Moreover,theactivationinthisareaexhibitsitsown

    typeofratiosensitivity:astheratiooftwomagnitudesbeingcomparedapproaches1:1,the

    activationincreases(Pineletal.[2001]).Infact,researchershavefoundneuronsinthe

    intraparietalsulcusofmonkeysthataretunedtospecificmagnitudes,withbell-shaped

    activationfunctionsoffixedvariancewhenplottedonalogarithmicaxis(NiederandMiller

    [2003];NiederandMiller[2004]).Theseactivationfunctionsareexactlywhatonewould

    predictgivenWebersLaw;astheratiooftwomagnitudesapproaches1:1,theactivation

    patternsoftheneuronscorrespondingtothosemagnitudesbecomehardertotellapart.

    Theactivationpatternsoftheseneuronsthusmirrorthebehaviouralpatternsassociated

    withmagnitudediscriminations.Whilemanyquestionsstillexistabouthowtheseneurons

    arecoordinatedwithinlargerpopulations,functionalaccountsofAMRsareincreasingly

    beingintegratedwithneuroscience.AMRsshouldthusserveaswelcometargetsfor

    philosophersinterestingintheorizingabouttheneuralrealizationofmental

    representations.

    2.3Analoguerepresentation

    Second,thelinesegments(suchas______)thatCareyanalogizestoAMRsarecontinuous,

    andthusanalogueinGoodmans([1976])senseofbeingdense(betweenanytwo

    representedvaluesthereisalwaysathirdrepresentedvalue).Butitisanopenquestion

    whetherAMRsthemselvesarecontinuousordensesincediscretesymbolscouldlikewise

    giverisetoratiosensitivityinmagnitudediscriminations.Toseethis,noticethatthe

  • 7/30/2019 Analog Magnitude Representations

    12/39

    12

    followingexternalmagnitudesystemwouldhaveservedjustaswelltoillustrateCareys

    point.

    number symbol

    1: |

    2: ||

    3: |||

    4: ||||

    7: |||||||

    8: ||||||||

    Althoughtheserepresentationsarediscrete,theytooareadirectanalogueofthenumbers

    theyrepresent,andthusbecomehardertodiscriminateastheirratioapproaches1:1.

    Indeed,asCareyherselfobserves,AMRscouldberepresentedbythetotalnumberof

    neuronsfiringwithinagivenpopulation.Sinceactionpotentialsareallornoneneurons

    eitherfireortheydonttherewouldthenbenointermediaterepresentationbetweena

    networkofnneuronsfiringandanetworkofn+1neuronsfiring.

    How,then,shouldtherelevantsenseofanaloguerepresentationbespecifiedifnot

    intermsofcontinuity?Careyoffersthefollowingsuggestion.

    Iconicrepresentationsareanalogue;roughly,thepartsoftherepresentationcorrespondto

    thepartsoftheentitiesrepresented.Apictureofatigerisaniconicrepresentation;the

    wordtigerisnot.Theheadinthepicturerepresentstheheadofthetiger;thetailinthe

    picturerepresentsthetail.Thetintigerdoesnotrepresentanypartofthetiger.([2009],

    458)

    Careyssuggestion,inotherwords,isthaticonicrepresentationssuchaspicturesare

    analoguebecausetheyhavepartsthatcorrespondtothepartsoftheentitythatis

  • 7/30/2019 Analog Magnitude Representations

    13/39

    13

    represented,andthatrepresentationssuchasthewordtigerarenotanaloguebecause

    theydonothavepartsthatcorrespondtothepartsoftheentityrepresented.Careythus

    seemstoendorsewhatIwillcallthePartAnalogueThesis(PAT):Risananalogue

    representationofXifandonlyifthepartsofRrepresentthepartsofX.ApplyingPATto

    AMRs,Carey([2009],p.458)illustrateswhysheholdsthatAMRsareanalogue.

    Analoguerepresentationsofnumberrepresentaswouldanumberlinetherepresentation

    oftwo(____)isaquantitythatissmallerthanandiscontainedintherepresentationfor

    three(______).

    Thus,CareymaintainsthatAMRsareanaloguebecausetheyobeyPAT;thepartsofanAMR

    representpartofwhattheAMRasawholerepresents.

    Notice,however,thatPATismissingaquantifier.Musteverypart,oronlysome

    parts,ofRrepresentpartofX?Ifwechoosetheexistentialquantifier,PATbecomestoo

    permissive.ThesentenceBillistallrepresentsanindividual,Bill,andthepropertyof

    beingtall.SoBillrepresentspartofwhatthatsentenceasawholerepresents,andPAT

    failstodistinguishAMRsfromsentences.UsingauniversalquantifierdoesntmakePAT

    anymoreplausible.IfeverypartofRhastorepresentpartofX,PATbecomestoo

    restrictive.RecallCareyssuggestionthatAMRsmightberealizedbyapopulationof

    neuronsfiring.Certainlyapartofoneofthoseneuronsdoesntrepresentanythingallonits

    own.Onlytheadditionofeachneuron(orperhapstheadditionofafamilyornetworkof

    neurons)indicatesanewmagnitude.Moreover,thereisnothingincoherentintheideathatgreatermagnitudesmightberepresentedbysmallerpopulationsofneuronsorlower

    neuralfiringrates.Inthatcase,however,apartofarepresentationofamagnitudeof(say)

  • 7/30/2019 Analog Magnitude Representations

    14/39

    14

    twentywouldnotrepresentpartoftwenty;itwouldrepresentmorethantwenty.PATthus

    failstocapturethesenseinwhichAMRsareanalogue.

    Maley([2011])providesamorepromisingaccountofanaloguerepresentationthat

    canbeappliedtoAMRs.Modifyinghisanalysisonlyslightly,wecansaythata

    representationRofarepresentedmagnitudeMisanalogueifandonlyif:(1)thereissome

    propertyPofRsuchthatthequantityoramountofPdeterminesM;and(2)asMincreases

    ordecreasesbyanamountd,PincreasesordecreasesasamonotonicfunctionofM+dor

    Md.6Thus,inorderforanAMRthatrepresentsamagnitude,M,tobeanalogue,therehas

    tobesomeproperty,P,oftheAMRthatincreasesordecreasesmonotonicallywithM.For

    example,considerthreeAMRsthatrepresentfour,five,andsixflashesoflightcallthem

    AMR4,AMR5,andAMR6,respectively.AccordingtoMaleysaccount,iftheserepresentations

    aretocountasgenuinelyanalogue,thenthereneedstobesomepropertypresumably

    someneuralpropertythatbelongstoeachandthatmonotonicallyincreasesordecreases

    fromAMR4toAMR5toAMR6.Thatpropertycouldbetherateatwhichapopulationof

    neuronsfires,thenumberofneuronsthatfire,orsomethingelseentirely;butsomesuch

    propertyhastoexistthatdetermineswhichmagnitudeeachoftheAMRsrepresent.

    Noticethatthisaccountdoesnotrequireanaloguerepresentationstobecontinuous

    ordense.EvenifPturnedouttobeanon-continuouspropertysuchasthenumberof

    neuronsfiring,themagnituderepresentedbyRwouldbeamonotonicfunctionofthe

    6ThisanalysisdiffersfromMaleys([2011],p.123)intwosignificantrespects.First,Maleysaccountiscouchedintermsofrepresentationsofnumberswhereasthepresentaccountiscouchedintermsofrepresentationsofmagnitudes.ObviouslythisalterationisnecessaryifwewanttoapplytheaccounttoAMRssincetheyrepresentnon-numericalmagnitudessuchasdurationanddistance.Second,Maleysaccountappealstoalinearfunctionratherthanamonotonicfunction.SinceAMRsarestandardlyinterpretedaslogarithmicallycompressed(Dehaene[2003]),thisamendmentiscrucial.Tohiscredit,Maley([2011],p.123,n.2)anticipatestheamendment.

  • 7/30/2019 Analog Magnitude Representations

    15/39

    15

    quantityoramountofPandwouldthuscountasanalogue.Additionally,becauseMaleys

    accountavoidstheproblematicnotionofapart,itavoidsthedifficultiesthatbefallPAT.

    TheanaloguenatureofAMRsmakesthemunlikethesymbolsfamiliarfrommodern

    digitalcomputerssincethereisnopropertyofsuchsymbolsthatmonotonicallyincreases

    ordecreaseswiththemagnitudestheydetermine.Forexample,althoughthebinary

    symbols00,01,10,and11representmagnitudesthatcanbeorderedfromleastto

    greatest,thereisnopropertyofthesymbolsthemselvesthatdeterminesthosemagnitudes

    andincreasesordecreasesmonotonicallywiththem.

    TherelationshipbetweentheanaloguenatureofAMRsandthelanguageofthought

    hypothesisislessclear,primarilybecauseproponentsofthelanguageofthought

    hypothesisarerarelyexplicitaboutthepreciseformatthatrepresentationsinthelanguage

    ofthoughtmusthave.Whileeveryoneagreesthatrepresentationsinthelanguageof

    thoughtmustbelanguage-like,paradigmaticlanguageshavearangeofproperties,andit

    isunclearwhichofthesepropertiesrepresentationsinthelanguageofthoughtneedto

    possess.Iflanguage-likerepresentationsmerelyneedtobecompositionalandsupport

    accuracyconditions,thereisnothingstoppingthemfrombeinganalogue.However,itis

    alsoafamiliarpropertyofprototypicallanguagesthattheirrepresentationsarenot

    analogue.Likethebinarysymbolsofdigitalcomputers,naturallanguagewordssuchas

    threeandfourdonothavesomepropertythatmonotonicallyincreasesordecreases

    withwhattheyrepresent.WhethertheanalogueformatofAMRsputsthematoddswith

  • 7/30/2019 Analog Magnitude Representations

    16/39

    16

    thelanguageofthoughthypothesisthusultimatelydependsonthevexedquestionofhow

    language-likerepresentationsinthelanguageofthoughtmustbe.7

    2.4AMRcomponents

    ThereisathirdwayinwhichCareysanalysismaybemisleading.Thereisnothinginthe

    linelengths(suchas______)thatsheanalogizestoAMRstoindicatethattheycorrespond

    tonumberasopposedtodistance,duration,rate,oranyothermagnitude.Butthebrain

    clearlyhassomemeansofdistinguishinganAMRofsevenindividualsfromanAMRof

    sevenmetresorsevenseconds.Additionally,AMRsdifferintheirobjects.AnAMRofa

    seven-secondlightdiffersfromanAMRofaseven-secondtone.Thus,amoreperspicuous

    depictionofAMRswouldincludearepresentationnotjustofthemagnitudessize,butalso

    whatwemightcallitsmodeandobject.WecanthusdepicteachAMRasanorderedtriplet

    thatincludesasize,mode,andobjectcomponent.Forexample,wecoulddepictanAMRof

    fourfishas{________,NUMEROSITY,FISH},anAMRofathree-secondtoneas{______,DURATION,

    TONE},etc.NoticethatalthoughtheratiosensitivityassociatedwithWebersLawis

    evidencethatthesizecomponentofAMRsisanalogue,itdoesnotprovideevidencethat

    themodeorobjectcomponentsareanalogue.Infact,giventhattheanalysisofanalogue

    representationinspiredbyMaleyisdefinedtoapplyonlytorepresentationsofmagnitudes,

    themodeandobjectcomponentsconsideredontheirowncannotbeanalogueaccordingto

    thatanalysis.Magnitudesarethingsthatcanbequantitativelyrelated;yetmodesand

    objectsassuchcannotbequantitativelyrelated.ThequestionsIsdurationmoreorless

    thandistance?andArelightflashesgreaterorfewerthantones?donotmakesense.Itis

    7Inotherwork(Beck[2012],[forthcoming]),IhavearguedthatthereisafurtherrespectinwhichAMRsareunlikeparadigmaticlinguisticrepresentations:theylackthesystematicrecombinabilityofsuchrepresentations.Ireturntothisissuein3.3.

  • 7/30/2019 Analog Magnitude Representations

    17/39

    17

    onlythesizecomponentofAMRs,orAMRsconsideredasawhole,thatcanbe

    quantitativelyrelated,andthusanalogueaccordingtotheanalysisinspiredbyMaley.

    IsthesizecomponentofAMRstrulyindependentofthemodeandobject

    components?Orisitanabstraction,imposedbyusastheorists,onwhatinrealityisan

    undifferentiated,homogenousrepresentation?Althoughthesequestionshavenotbeen

    definitivelyresolved,theycanbe,andhavebeen,subjectedtoempiricalinvestigation.The

    basicstrategyistolookforevidenceofacommonneuralmechanismacrossAMRsthat

    differintheirmodeandobjectcomponents.Anysuchcommonmechanismwouldthen

    presumablybeattributabletothesizecomponent,suggestingthatthesizecomponenttruly

    isindependentofthemodeandobjectcomponents.Thereareseveralfindingsthathave

    beencitedinsupportofacommonmechanism.

    Althoughratiosensitivityvariesacrossspeciesandages,thereisevidencethatthedegreeofsensitivityisthesameforallmodeswithinagivenspeciesatagivenage.

    Thus,theratiosatwhichmatureratssuccessfullydiscriminatedurationsarethe

    sameastheratiosatwhichtheysuccessfullydiscriminatenumerosities(Meckand

    Church[1983]),andsix-month-oldinfantsshowthesameratiosensitivityforarea,

    numerosity,andduration(Feigenson[2007]).

    Attemptstointerveneontherepresentationofsizeinonemodetendtohavesimilareffectsontherepresentationofsizeinallmodes.Forexample,when

    methamphetamineisadministeredtorats,thereisanidenticalacceleratedshiftin

    therepresentationofbothnumerosityandduration(MeckandChurch[1983]).

    Behaviouraltasksthattapintoonemodetendtointerferewithperformanceontasksthattapintoothermodes.Forexample,adultsarefasteratcomparingtwo

  • 7/30/2019 Analog Magnitude Representations

    18/39

    18

    digitswhenthegreaterdigitisinalargerfontcomparedtothelesserdigitand

    slowerwhenthegreaterdigitisinasmallerfontcomparedtothelesserdigit(Henik

    andTzelgov[1982]).Likewise,adultsjudgethedurationofpresentationofonedigit

    tobelongerthanthe(equal)durationofpresentationofaseconddigitwhenthe

    firstdigitisgreaterthanthesecond(Oliverietal.[2008]).8

    Associationswithinonemodetendtogeneralizetoothermodes,eveninyounginfants.Forexample,if9-month-oldinfantsareshownthatblackstripedrectangles

    arelargerthanwhitedottedrectangles,theyexpecttheblackstripedrectanglesto

    bemorenumerousandtolastforalongerdurationaswell(LourencoandLongo

    [2010]).

    Evidencefrombothbraindeficitsandbrainimagingpointtooverlappinglociformagnituderepresentationsofvariousmodesintheinferiorparietalcortexof

    humansandotherprimates(Walsh[2003];Jacob,Vallentin,andNieder[2012]).

    Tobesure,thesevariouslinesofevidencearehardlyconclusive.Otherexplanationsofthe

    resultsarepossible.PerhapstheratiosensitivityofAMRsofdifferentmodesallimproveat

    thesamerateacrossontogenynotbecausetheyshareacommonsizemechanism,but

    becauseorganismstendtousetheirAMRsofdistance,duration,numerosity,etc.equally

    often,leadingthemalltobeindependentlyhonedatthesamerate.Similarly,otherresults

    mightbepickingupnotonacommonsizemechanism,butonhomogenousAMRsthatare

    instantiatedinoverlappingoradjacentneuralnetworks.Forexample,ifAMRsofduration

    andnumerositywereembeddedinsufficientlyproximalneuralnetworks,tasksthat8Interestingly,noteverymagnitudeshowsthesameinterferenceeffects.Forexample,althoughloudnessdiscriminationsaresubjecttoWebersLaw,loudnessdoesnotinterferewithdistancediscriminationsinthewaythatdurationdoes(SrinivasanandCarey[2010]),suggestingthatrepresentationsofloudnessdonotrecruitthesamesizecomponentthatAMRsofdistance,duration,andnumerosityseeminglydrawupon.

  • 7/30/2019 Analog Magnitude Representations

    19/39

  • 7/30/2019 Analog Magnitude Representations

    20/39

    20

    fish,birds,lowermammals,andevenhumaninfantsrepresentmagnitudes.Consequently,

    demandingtheoriessituneasilywiththeempiricalevidence,reviewedinSection1,that

    AMRsdorealworkexplainingthebehavioursoftheseanimals,suchaswhyaratwillpress

    leverAratherthanleverBfollowingthepresentationofaparticularstimulus.AlthoughI

    cannotdefendtheopinionhere,IalsoagreewithBurge([2010])thatthearguments

    advancedinfavourofdemandingtheoriesofrepresentationalcontentarenotcompelling.I

    amthusinclinedtojoinBurgeinrejectingsuchtheoriesasunmotivatedandempirically

    dubious.10

    Attheotherextremesitdeflationarytheoriesofrepresentation,whichseekto

    reducethenotionofrepresentationalcontenttoothernotions,suchasinformationand

    learning(Dretske[1981]),biologicalfunction(Millikan[1989]),orfunctioning

    isomorphism(Gallistel[1990]).Becausethesereductiveanalysesarerelativelyeasyto

    satisfy,representationsinthedeflationarysensearerelativelyeasytocomeby.Allofthese

    theoriesarethuslikelytocountAMRsashavingrepresentationalcontent,evenin

    relativelysimpleorganisms.Gallistel([1990])istheonedeflationarytheoristIamawareof

    whoexplicitlydiscussesAMRs.AccordingtoGallistel([1990],p.15),

    Thebrainissaidtorepresentanaspectoftheenvironmentwhenthereisafunctioning

    isomorphismbetweensomeaspectoftheenvironmentandabrainprocessthatadaptsthe

    animalsbehaviourtoit.

    10SomeproponentsofdemandingtheoriesofrepresentationmightallowthatAMRsarerepresentationaleveninloweranimalsprovidedthattheyaretreatedassub-individualstates.Some subsystemoftheratsrepresentsmagnitudesbut theratdoesnotrepresentsmagnitudes.Yetgiventhatitisthesurprisinglyintelligentbehaviouroftheratthatneedstobeexplained,includingtheratsabilitytolearn,thisreplystrikesmeasimplausible.Suchsophisticatedbehaviours,itseemstome,arebestexplainedintermsofmentalstatesthatbelongtorats.

  • 7/30/2019 Analog Magnitude Representations

    21/39

  • 7/30/2019 Analog Magnitude Representations

    22/39

  • 7/30/2019 Analog Magnitude Representations

    23/39

    23

    ThesedifficultiesalsospelltroubleforGallistelandGelmans([2000])proposalthat

    thecontentsofAMRsarebestcharacterizedbyrealnumbers.AsBurge([2010],p.481)

    observes,theintegersareasubsetoftherealnumbers,soifAMRscannotrepresent

    integersnorcantheyrepresenttherealnumbers.GallistelandGelmanreasonthatinsofar

    astheformatofthesizecomponentofAMRsisanalogue,itmustbecontinuous,andthus

    thatthesizecomponentofAMRsmustrepresentrealnumbers,whicharelikewise

    continuous.Butthislineofreasoningisproblematic.AswesawinSection2.3,itisa

    mistaketosupposethatAMRsmustbecontinuousjustbecausetheyareanalogue.Webers

    Lawcanbeexplainedbydiscretesymbolsthatareadirectanalogueofthemagnitudes

    represented.Moreover,asLaurenceandMargolis([2005])emphasize,GallistelandGelman

    seemtobeseducedbyaspuriousformat-contentconflation.Justasdiscrete

    representationalvehiclessuchasor2canstandforrealnumbers,continuous

    representationalvehiclescansurelystandforsomethingotherthanrealnumbers.Thus,

    evenifAMRsweretohaveacontinuousformat,itsimplywouldntfollowthattheymust

    representtherealnumbers.

    RatherthaninterpretingAMRsintermsofintegersorrealnumbers,Carey([2009],

    p.135)proposestointerpretthemasapproximatecardinalvalues.Forexample,anAMR

    mightrepresentapproximatelysevenlightflashesorapproximatelysevenseconds.Butof

    courseonecannotrepresentapproximatelysevenwithoutbeingabletorepresentseven,

    andsoCareyssuggestionfaresnobetterthanGallistelandGelmans.Moreover,Careys

    suggestionfailstospecifyhowapproximateAMRsare.Whileonemightreachforgreater

    specificitybyappealingtoWebersLaw,WebersLawisitselfdefinedintermsofratios

    amongintegers,whichweweretryingtoavoid.

  • 7/30/2019 Analog Magnitude Representations

    24/39

    24

    Burge([2010],pp.4823)recommendslookingtoEudoxustheoryofpure

    magnitudestoarticulatethecontentsofAMRs.12Eudoxusdevelopedhistheoryofpure

    magnitudesinordertohandleratiosthatbedevilledthePythagoreans.Becausethe

    Pythagoreansmaintainedthatwholenumbersarethebasisofallratios,theycouldnot

    expressincommensurableratiossuchasthatbetweenthesideofasquareanditsdiagonal.

    (Inourterminology,theycouldnotexpressirrationalnumberssuchas2.)Eudoxus

    respondedbydefiningratiosandproportionswithoutappealingtonumbers.Instead,he

    definedratiosintermsofsizerelationsamonghomogenousmagnitudesthemselves,and

    proportionsintermsofcomparativesizerelationsofratios.13Thus,twolinelengthscan

    enterintooneratio;twoweightscanenterintoanotherratio;andthenthosetworatios

    canbecomparedandfoundtobeproportionalornot.Discretemagnitudessuchaswhole

    numberscanenterintoratiosandproportionsaswell,andareunderstoodinthesameway

    asratiosandproportionsthatinvolvecontinuousmagnitudessuchaslinelengthorweight.

    Thus,thekeyfeatureofEudoxustheory,accordingtoBurge,isthatitsconceptofpure

    magnitudedoesnotdifferentiatebetweencontinuousanddiscretemagnitudes.Pure

    magnitudesarelikethesizecomponentofAMRsinthattheycannotbeexpressedusing

    numbers.ThisleadsBurgetohypothesizethatAMRsrefertopuremagnitudes.

    Burgeshypothesisissuggestive,andseemslikeastepintherightdirection.Itis

    plausiblethatthesizecomponentsofAMRsdonotrefertonumbers,whichmakespure

    magnitudesinvitingcandidatestoserveastheircontents.Nevertheless,Eudoxeanpure

    12EudoxustheoryofpuremagnitudesisdescribedbyEuclid([1956],book5).Iamindebtedto(Sutherland[2006])and(Stein[1990])forexposition.

    13Eudoxusprincipalinsightwasthatsuchcomparisonscouldbespelledoutintermsof equimultiples.See(Euclid[1956],book5,definition5;Sutherland[2006],pp.5367;andStein[1990],pp.1669).

  • 7/30/2019 Analog Magnitude Representations

    25/39

  • 7/30/2019 Analog Magnitude Representations

    26/39

  • 7/30/2019 Analog Magnitude Representations

    27/39

    27

    predictsthatwhenthemodelisrunninginitsnumerositymodethetimeittakesto

    representasetisproportionaltothesetssize(Carey[2009],pp.1323).Morerecent

    models,whichoperateinparallel,arenotobviouslycommittedtoanyunits.Forexample,

    accordingtoChurchandBroadbents([1990])model,durationsareassociatedwiththe

    phasesofasetofneuraloscillatorsofdifferingperiods.Thus,agivendurationwilltypically

    beassociatedwithseveraloscillatorswhoseindividualperiodstotaltothedurationbeing

    measured(ignoringerror).Asaresult,thereisnounitinthismodelbywhichatotal

    durationismeasured.Ofcourse,weastheoristsmightuseseconds,minutes,orsomeother

    unitoftimetodescribeatotalduration,butsuchunitsdonotplayafunctionalroleinthe

    modelitself.ToborrowahelpfulphrasefromChristopherPeacocke([1986]),AMRsmight

    thusbecharacterizedasunitfree.14

    TurningfinallytotheobjectcomponentofAMRs,itisnotablethatAMRsarenot

    restrictedtorepresentingobjectsfromonlyonemodality.Forexample,AMRscan

    representvisualobjectssuchaslightflashesordotsonascreenaswellasauditoryobjects

    suchastones.NorareAMRsrestrictedtoattributingmagnitudestospatiallywell-defined

    objects.AMRscanalsoassignmagnitudestoabstractevents,suchasasequenceofrabbit

    jumps(WoodandSpelke[2005]).Infact,thereisnoevidenceIamawareofsuggestingthat

    theobjectsofAMRshaveanyrestrictionsbeyondthoseimposedbythebroader

    representationalcapacitiesofthesubject.15

    14Peacockeintroducesthistermtocharacterizethecontentofperceptualexperiencesofmagnitudes,aswhenyouvisuallyexperiencethelengthofapianowithoutexperiencingitinfeetormetres.IhypothesizethatthephenomenonPeacockeisolatesisgroundedinAMRsi.e.,thatourconsciousexperienceshaveaunit-freecharacterbecausethoseexperiencesaregeneratedbyAMRswhicharethemselvesunitfree.

    15Burge([2010],p.472)suggeststhatwhenAMRsareusedinperceptiontheyarelimitedtorepresentingperceivableentities,buthedoesnotconsiderwhetherAMRsaresolimitedwhentheyareusedoutsideofperceptione.g.,whenapersonconsiderstheabstractquestionwhethertheunionoftwosetswith35and

  • 7/30/2019 Analog Magnitude Representations

    28/39

    28

    3.3WhatcontenttypesdoAMRshave?

    WehavejustbeenconsideringwhatAMRsrepresent.Thereisafurtherquestion,however,

    concerninghowAMRsrepresentwhattheydo.WhattypeofcontentsdoAMRshave?

    AsIunderstandthem,mentalcontentshavethefunctionofcapturinghowthings

    arefromtheperspectiveofthethinker.Theyfulfilthisfunction,inpart,bymarkinga

    thinkersmentalabilities.Whereamentalstateistheresultoftwoseparateabilities,many

    philosophershavethereforewantedtouseacontentthatisstructuredfromtwoelements

    tocaptureit.Forexample,ifapersonwhocanthinkthatAmydiedrichandthatAmydied

    poorwouldexerciseasingleabilitytheabilitytothinkaboutAmyinthecourseof

    thinkingeachthought,wecanmarkthatfactbyattributingacommonelementthe

    conceptAmytothecontentofeachofthosethoughts(Evans[1982],pp.100-5).Sincethe

    conceptualthoughtsofhumanbeingsseemtobestructuredfromdiscreteabilitiesinthis

    sense,manyphilosophershavewantedtousestructuredcontentstocapturethem.

    WehaveseenreasontothinkthatAMRsarealsostructuredfromdiscreteabilities.

    Forexample,theabilitytorepresentsevenflashesoflightseemstosharesomethingin

    commonwiththeabilitytorepresentasequenceofseventones.Moregenerally,theability

    todeployAMRsseemstodecomposeintoabilitiestorepresentasize,amode,andan

    object.Ifthatisright,thenwehavereasontoviewthecontentsofAMRsasstructured.

    Anotherimportantdimensionalongwhichmentalcontentscandifferistheir

    finenessofgrain.Anargumentoftenadvancedinfavouroffine-grainedcontentsisthat

    theycanaccommodatedistinctmodesofpresentationofthesameentity.Forexample,a

    24members,respectively,wouldhavegreaterorfewerthan90totalmembers.Thereisevidence,however,thatAMRsareinvolvedinsuchabstractcomparisonswhentheobjectsofcomparisoncannotbeperceived(Dehaene[2011],p.239ff.).IthusseenoreasontodenythatAMRscanrepresentnon-perceivableentitiesoutsideofperception.

  • 7/30/2019 Analog Magnitude Representations

    29/39

    29

    personmighthavetheabilitytothinkaboutVenusundertheHesperusmodeof

    presentationwhilelackingtheabilitytothinkaboutVenusunderthePhosphorusmodeof

    presentation.ThisconsiderationappliestoAMRs.Theabilitytothinkaboutagiven

    magnitudeusingnumbersandunitsdiffersinkindfromtheabilitytothinkaboutitusing

    AMRs.Forexample,theabilitytorepresentaten-seconddurationastensecondsisquite

    differentfromtheabilitytorepresentitintheunit-freemanorassociatedwithAMRs.

    ThereisthusreasontoviewAMRcontentsascomposedfrommodesofpresentation.

    Theideathatcontentsarestructuredfrommodesofpresentationislikelytobring

    tomindFregeanThoughts,whicharestructuredfromsenses.Infact,byclaimingthatAMR

    contentsarestructuredfrommodesofpresentationitmayseemthatIhaveidentifiedthe

    contentsofAMRswithFregeanThoughts.Itwouldbetoohastytodrawthatconclusion,

    however,andfortworeasons.

    First,ifpropertiesareindividuatedfinelyenough,itmaybepossibletoexplainthe

    differencebetweenrepresentingaten-seconddurationastensecondsandrepresentingit

    intheunit-freemanorassociatedwithAMRsintermsoftherepresentationofdistinct

    properties.Thus,justassomearguethatthedifferencebetweentheHesperusand

    Phosphorusmodesofpresentationboilsdowntothedifferencebetweenrepresentingthe

    propertiesvisibleintheeveningandvisibleinthemorning,itisopentosomeonetoargue

    thatthedifferencebetweenunit-ladenandunit-freemodesofpresentationboilsdownto

    whetherintegerandunitpropertiesarerepresented.Inthatcase,however,therewouldbe

    noneedtoappealtoFregeansenses,whicharesupposedtobedistinctfromrepresented

    properties.

  • 7/30/2019 Analog Magnitude Representations

    30/39

    30

    Second,atleastastypicallyconceived,FregeanThoughtsandtheircomponents,

    senses,aremeanttomarkaparticulartypeofabilitythatIllcallaconceptualability.One

    featureofconceptualabilitiesisthattheyaresystematicallyrecombinable.Theconceptual

    abilitytothinkthataisFandthatbisGentailstheconceptualabilitytothinkthataisG

    andthatbisF.Inotherwords,conceptualabilitiesobeywhatEvans([1982])callsthe

    GeneralityConstraint,whichholdsthattheconceptualthoughtsonecanthinkareclosed

    underallmeaningfulrecombinationsoftheconstituentsofthesentencesthatbestexpress

    them.Itisquestionable,however,whetherAMRsobeytheGeneralityConstraint.Ihave

    arguedelsewhere(Beck[2012],[forthcoming])thatbecauseofWebersLawathinker

    usingAMRscanhavetheabilitytorepresentthatamagnitudeof9islessthanamagnitude

    of18,andthatamagnitudeof10islessthanamagnitudeof20,butnotthatamagnitudeof

    9islessthanamagnitudeof10northatamagnitudeof18islessthanamagnitudeof20.

    Moreover,becauseWebersLawistraceabletotheanalogueformatofAMRsthemselves,I

    arguedthatthisfailureofrecombinabilityisafailureofrepresentationalcompetenceand

    notmerelyafailureofdiscriminativeperformance.ItisinthenatureofAMRsthattheyare

    unabletorepresentthatonemagnitudeislessthananotherwhentheirratioexceedsa

    certainthreshold.Ifthatisright,thesortsofrepresentationalabilitiesthatunderlieAMRs

    differinkindfromconceptualabilities.Theydonotexhibitthesameunfettered

    recombinability.WethushavereasontodistinguishthecontentsofAMRsfromFregean

    Thoughtsevenifbothtypesofcontentarestructuredfrommodesofpresentation.We

    shouldconcludethatAMRshaveasuigeneristypeofnonconceptualcontentinstead.

    4.Computations

  • 7/30/2019 Analog Magnitude Representations

    31/39

    31

    SupposingthatAMRsexist,thequestionarisesofwhatorganismscandowiththem.What

    sortsofcomputationsdotheysupport?

    4.1Arithmeticcomputation

    Aboveall,AMRsareassociatedwitharithmeticcomputations,includingcomparison,

    addition,subtraction,multiplication,anddivision.Recall,forexample,theabilityofducks

    topositionthemselvesinproportiontotherateatwhichexperimenterstossmorselsof

    bread.Thisabilityembodiesacapacityfordivisionsincetheabilitytorepresentrate

    plausiblyderivesfromtheabilitytodividerepresentationsofnumerositiesby

    representationsofdurations(Gallistel[1990],pp.351383).Italsoembodiesanabilityto

    comparetworatesandcalculatehowmuchgreateroneisthantheother.Recallaswellthat

    theducksalteredtheirstrategywhenoneexperimentertossedmorselsthatweretwicethe

    sizeofthosetossedbytheother,therebyexemplifyinganabilitytomultiplymorselsizeby

    feedingrate.

    IfAMRssupportarithmeticcomputationsandarepartoftheinnatecognitive

    hardwareofhumanbeings,thenhumanchildrenshouldexhibitaprimitiveabilityto

    engageinarithmeticpriortobeingtrainedinit.Barthetal.([2006])setouttotestthis

    prediction.Usinganimateddisplaysinwhichsetsofcoloureddotsmovebehindoremerge

    fromanoccluder,theytestedtheabilityofpre-schoolchildrentocompare,add,and

    subtractsetsofdots.Forexample,thechildrenmightseeasetof25bluedotsmovebehind

    anoccluder,thensee25morebluedotsjointhembehindtheoccluder,thensee30reddots

    andbeaskedwhethertherearemore(occluded)bluedotsor(unoccluded)reddots.

    Childrensucceededoncomparison,addition,andsubtractiontasksevenaftercontrolling

    fornon-numericalvariablessuchasdotcircumference,area,anddensity.Morerecently,

  • 7/30/2019 Analog Magnitude Representations

    32/39

    32

    McCrinkandSpelke([2010])usedasimilarparadigmtoshowthatchildrencouldsucceed

    onnon-symbolicmultiplicationtasksbeforebeingschooledinmultiplicationordivision.

    BothsetsofresultssuggestthatAMRssupportaprimitivetypeofarithmeticthatdoesnot

    dependuponformalmathematicaltraining.16

    4.2Practicaldeliberation

    TofullyappreciatethecomputationalpowerofAMRs,itisessentialtonoticethattheycan

    beusednotonlytorepresenthowtheworldis,butalsohowthecognizerwouldlikeitto

    be.Thatis,AMRscanplayadesire-likeroleinadditiontoabelief-likerole.Considerthe

    long-tailedhummingbird,whichforagesbyrecoveringnectarfromavarietyofwidely

    dispersedsources,oftenflyingatleasthalfakilometreforanyonefeeding.Becausethe

    birdistoosmalltostoremuchenergy,andconsumesenergyratherquickly,thereis

    considerablepressureforittooptimizeitsfrequentforagingrunstofindanintervalthat

    islongenoughfortheharvesttohavereplenishedsincethepreviousvisit,butnotsolong

    thatthebountyislikelytohavebeenpilferedbyacompetitor.Thatinturnrequiresthe

    birdtorepresenttherateatwhichvariousnectarsourcesreplenishafterdepletionandthe

    temporalintervalsbetweenvisits.Itcanthencomputeanestimationoftheamountof

    nectarthatiscurrentlyateachsource,andusethatvalueasaproxyfortheutilityof

    visitingeachsource.Incontrolledenvironmentsusingartificialflowersthatarefilledwith

    sugarwateratintervalssetbytheexperimenter,itcanbeshownthatbirdswill,infact,

    optimizetheirvisits(Gill[1988]).

    16ArithmeticcomputationsoverAMRsarediscussedatlengthin(Gallistel[1990]).Seealso(Brannonetal.[2001];Flombaumetal.[2005];McCrinkandWynn[2004];andBeranandBeran[2004]).

  • 7/30/2019 Analog Magnitude Representations

    33/39

    33

    OtherexamplesofhowAMRsmightencodedesiresorutilitiesarenothardto

    dreamup.ArobinmightuseitsAMRoftherateofreturnofwormsinagivenfieldasa

    proxyforthedesirabilityofforagingthere,amonkeymightuseitsAMRofthenumerosity

    ofpredatorsstalkingittoestimatetheutilityofretreat,andahumanchildmightuseits

    AMRofthesizeoftwopiecesofcakeasameasureofthedesirabilityofeach.Theabilityto

    useAMRsinthiswayisimportant,asitopensupthepossibilityofembeddinganentire

    processofpracticaldeliberationoverAMRswithinaformaldecision-theoreticframework,

    suchasexpectedutilitytheory,wherebyananimalcalculatestheexpectedutilityofeachof

    arangeofactionsandthenchoosestheactionwiththemaximumexpectedutility.

    Toseehowthismightworkforasimplifiedcase,supposethatarobinisdeciding

    betweentwoactions:foragingforwormsinthefield(A1)orforagingforberriesinthe

    forest(A2).Andsupposethattheworldcanbeinoneoftwopossiblestates:raining(S1)or

    notraining(S2),whererainincreasestheprevalenceofwormsinthefieldbuthasno

    immediateeffectonhowmanyberriesareavailableintheforest.Assumingthatrobins

    valuewormsandberriesonapar,theycouldcalculatetheexpectedutilityofeachactionas

    follows(whereu(A|S)isthedesirabilitythattherobinassignstoactionAgiventhatthe

    worldisinS,andprob(S|A)isthesubjectiveprobabilitythattheworldwillbeinSgiven

    thatactionAisperformed).

    EU(A1)=[u(A1|S1)xprob(S1|A1)]+[u(A1|S2)xprob(S2|A1)]

    EU(A2)=[u(A2|S1)xprob(S1|A2)]+[u(A2|S2)xprob(S2|A2)]

    Aswevealreadyseen,u(A|S)canbecalculatedbytherobinsAMRoftherateofreturnof

    wormsinthefieldandberriesintheforestduringrain,andnorain,inthepast.Buthow

    willtherobincalculatethevaluesofprob(S|A),i.e.theprobabilityofrain?Onepossibilityis

  • 7/30/2019 Analog Magnitude Representations

    34/39

  • 7/30/2019 Analog Magnitude Representations

    35/39

    35

    spiteoftherelevanceofAMRstoahostofcentralissuesinthephilosophyofmind

    concerningneuralrealization,analoguerepresentation,representationalcontent,

    nonconceptualcontent,andanimalcognition.InthispaperIhavesoughttotakesomefirst

    stepstowardsredressingthisimbalancebydrawingattentiontoAMRsandanalysingtheir

    format,theircontent,andthecomputationstheysupport.AlthoughmyanalysisofAMRs

    hasonlyscratchedthesurface,Ihopethatitwillassistphilosopherswhowanttogive

    AMRsamorecentralplacewhentheytheorizeaboutthemind.

    Acknowledgements

    IpresentedmaterialfromthispaperattheannualmeetingoftheSouthernSocietyfor

    PhilosophyandPsychologyinAustin,TexasinFebruary2013.Thankstothosewhobraved

    an8amstarttoprovidemewithfeedback.IamalsoindebtedtomycolleaguesBrianHuss,

    HenryJackman,JoshMugg,andespeciallyKristinAndrewsforvaluablecommentson

    drafts,andtotwoanonymousrefereesforthisjournalwhosehelpfulsuggestionsspawned

    substantialimprovements.

    References

    Barth,H.,Kanwisher,N.andSpelke,E.[2003]:TheConstructionofLargeNumberRepresentationsinAdults,Cognition,86,pp.20121.

    Barth,H.,LaMont,K.,Lipton,J.,Dehaene,S.,Kanwisher,N.andSpelke,E.[2006]:Non-

    SymbolicArithmeticinAdultsandYoungChildren,Cognition,98,pp.199222.

    Beck,J.[2012]:TheGeneralityConstraintandtheStructureofThought,Mind,121,pp.563600.

    Beck,J.[forthcoming]:AnalogueMagnitudes,theGeneralityConstraint,andNonconceptualThought,Mind.

  • 7/30/2019 Analog Magnitude Representations

    36/39

  • 7/30/2019 Analog Magnitude Representations

    37/39

  • 7/30/2019 Analog Magnitude Representations

    38/39

    38

    pp.11731.

    McCrink,K.andSpelke,E.S.[2010]:CoreMultiplicationinChildhood,Cognition,116,pp.20416.

    McDowell,J.[1994]:MindandWorld,Cambridge,MA:HarvardUniversityPress.

    Meck,W.H.andChurch,R.M.[1983]:AModeControlModelofCountingandTimingProcesses,JournalofExperimentalPsychology:AnimalBehaviourProcesses,9,pp.32034.

    McCrink,K.andWynn,K.[2004]:Large-NumberAdditionandSubtractioninInfants,PsychologicalScience,15,pp.77681.

    Millikan,R.[1989]:Biosemantics,JournalofPhilosophy,86,pp.28197.

    Nieder,A.andMiller,E.K.[2003]:CodingofCognitiveMagnitude:CompressedScalingof

    NumericalInformationinthePrimatePrefrontalCortex,Neuron,37,pp.14957.

    [2004]:AParieto-FrontalNetworkforVisualNumericalInformationintheMonkey,ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,101,pp.745762.

    Oliveri,M.,Vicario,C.M.,Salerno,S.,Koch,G.,Turriziani,P.,Mangano,R.,Chillemi,G.andCaltagirone,C.[2008]:PerceivingNumbersAltersTimePerception,NeuroscienceLetters,438,pp.30811.

    Peacocke,C.[1986]:AnalogueContent,ProceedingsoftheAristotelianSociety,

    SupplementaryVolume,60,pp.117.Pfungst,O.[1965]:CleverHans:TheHorseofMr.vonOsten,R.Rosenthal(ed.),NewYork:

    HenryHolt.

    Pica,P.,Lemer,C.,Izard,V.andDehaene,S.[2005]:ExactandApproximateArithmeticinanAmazonianIndigeneGroup,Science,306,pp.499503.

    Pietroski,P.,Lidz,J.,Hunter,T.andHalberda,J.[2009]:TheMeaningofMost:Semantics,Numerosity,andPsychology.MindandLanguage,24,pp.55485.

    Pinel,P.,Dehaene,S.,Riviere,D.andLeBihan,D.[2001]:ModulationofParietalActivation

    bySemanticDistanceinaNumberComparisonTask,Neuroimage,14,pp.101326.

    Rescorla,M.[2009]:CognitiveMapsandtheLanguageofThought,BritishJournalforthePhilosophyofScience,60,pp.377-407.

    Srinivasan,M.andCarey,S.[2010]:TheLongandtheShortofIt:OntheNatureandOriginofFunctionalOverlapbetweenRepresentationsofSpaceandTime,Cognition,116,pp.21741.

  • 7/30/2019 Analog Magnitude Representations

    39/39

    39

    Stein,H.[1990]:EudoxusandDedekind:OntheAncientGreekTheoryofRatiosandItsRelationtoModernMathematics,Synthese,84,pp.163211.

    Strawson,P.F.[1959]:Individuals:AnEssayinDescriptiveMetaphysics,NewYork:Methuen.

    Sutherland,D.[2006]:KantonArithmetic,Algebra,andtheTheoryofProportions,JournaloftheHistoryofPhilosophy,44,pp.53358.

    VanMarle,K.andWynn,K.[2006]:Six-Month-OldInfantsUseAnalogueMagnitudestoRepresentDuration,DevelopmentalScience,9,pp.F419.

    Walsh,V.[2003]:ATheoryofMagnitude:CommonCorticalMetricsofTime,SpaceandQuantity,TrendsinCognitiveSciences,7,pp.4838.

    Wood,J.andSpelke,E.[2005]:InfantsEnumerationofActions:NumericalDiscriminationandItsSignatureLimits,DevelopmentalScience,8,pp.17381.

    Xu,F.,andSpelke,E.[2000]:LargeNumberDiscriminationin6-Month-OldInfants,Cognition,74,pp.B111.