Contentslistsavailableatjournalhomepagewwwelseviercomlocatecognit - PDF document

1 Contentslistsavailableatjournalhomepage:
Contentslistsavailableatjournalhomepage:www.elsevier.com/locate/cognit OriginalArticlessmoretothanmeetstheeye:KnowledgeofvisionandlightverbsamongcongenitallyblindandsightedindividualsMarinaBedny,JorieKoster-Hale,GiuliaElli,LindsayYazzolino Correspondingauthorat:DepartmentofPsychologicalandBrainSciences,JohnsHopkinsUniversity,3400N.CharlesStreet,Baltimore,MD21218,UnitedStates.E-mailaddress:(M.Bedny). sightednativeItalianspeakers.Again,theblindindividualsgeneratedreasonablemeaning-featuresforalloftheseverbs.Forinstance,topeepwasassociatedwiththefeaturestowatch,nottobeseen,whereastospotwasassociatedwiththefeaturestoseefarawaybybothsightedandblindpeople.Thedisadvantageoffreeresponses,however,isthatthesedataareverysparse,asfeaturesmayberarelymentioned,ordescribedusinghomonyms.Thus,itremainshardtoquantitativelytestwhetherblindindividualsmeaningsoftopeeptospotaredierentfromthoseofsightedindividuals.Moregenerally,peoplemayhaveveryrichanddetailedknowledgeofthemeaningsofwords,butnotrevealthatknowledgeintheirdenitions.Thepragmaticsofthetaskmaycausepeopletolimitthefea-turestheygeneratetorelativelydistinctivepropertieswithinanimpliedcontext.Forexample,peoplemaybemorelikelytovolunteerthatze-brashavestripesthanthattheyhavemouths.ShepardandChipmanarguedthatpeopleseemunabletotellusanythingsigniaboutthestructureofanindividualmental[representation]assuch.Whattheycan,however,tellusaboutistherelationsbetweenthatinternalrepresentationandotherinternalrepresentations.Thatisapractical,albeitincomplete,waytoelicitrichinformationaboutthemeaningofawordistoaskpeopletoproducenotthemeaningitselfbutjudgmentsofhowitrelatestothemeaningsofotherwords.Thus,weeasilyreportthatorangeismoresimilartoredthantobluewithoutbeingabletosayanythingsignicant[]abouttheuniquesubjectiveexperienceofthecolororangeitselfShepard&Chipman,1970Estimatesofthesimilaritybetweenpairsofword-meaningsareeasytoelicitfromawiderangeofdomainsandarenaturallyquantitative.Peoplemakehighlystable,reliablejudgmentsofthesimilaritiesofpairsofmammals(Rumelhart&Abrahamson,1973),birds(Rips,Shoben,&Smith,1973),fruits(Hutchinson&Lockhead,1977),foods(Ross&Murphy,1999),numbers(Shepard,Kilpatric,&Cunningham,1975colors(Shepard&Cooper,1992),emotions(&Wedell,1994andpersonalitytypes(Bimler&Kirkland,2007),amongmanyotherexamples.Thesemanticsimilarityspacesderivedfromsuchjudgmentsdonotprovideacompletemeasureofwhatpeopleknowaboutado-main.Forexample,thesimilarityspaceofanimalsmaynotresknowledgeabouttheirdiets,orhowtheanimalsareusedbyhumansformakingfoodorclothing,unlesspeopleareaskedexplicitlytojudgetheseparticularfeatuers(e.g.Medinetal.,2002,Tenenbaum&ths,2001,Murphy,2004).Nevertheless,semanticsimilarityjudgmentscapturealargeamountofinformationquicklyandquanti-tativelyandpredictperformanceonmoreimplicittasksforthesamewords,suchasmemoryconfusionsandprimingeects,suggestingthatsimilarityjudgmentscapturesomestablesemanticpropertiesofwordsHutchinson&Lockhead,1977Criticalforthepresentpurposes,semanticsimilarityjudgmentsaresensitivetobetween-groupdierencesinsemanticknowledge.Forex-ample,changesinsimilarityjudgmentsprovideanearlysignalofcognitivedeteriorationinpatientswithAlzheimersDisease(AD).Disruptionofsimilarityjudgmentspredictstherateofpatientslossofcognitivefunctionoverthefollowingyear(Chanetal.,1993).Semanticsimilarityjudgmentsarealsosensitivetochangesincultureandex-perience.Forinstance,theperceivedsimilarityofmammalsrevealsbothsubstantialagreement,andsignicantdeviations,betweenAmericancollegestudentsandItzajMayanadults(Lopez,Atran,Coley,Medin,&Smith,1997Thus,semanticsimilarityratingsofpairsofwords,whileonlyapartialmeasureofwhatpeopleknow,oerawaytoquantitativelycomparethemeaningsofvision-relatedwordsamongblindandsightedindividuals.Indeed,thislogichasbeenusedbypreviousstudiestotestblindindividualsknowledgeofcolor.Onaveragebothsightedandblindgroupsshowasystematicpattern,resemblingacolorwheel:redissimilartoorangewhichissimilartoyellow,andsoon,untilviolet,whichissimilartoblueandred(Shepard&Cooper,1992).Thereismorevariationacrossblindthansightedadults,however,someblindadultsreproducethecolorwheel,whileothersmakeidiosyncraticjudgmentswithlargedeviationsfromthetypicalpattern(1978;Saysani,Corballis,&Corballis,2018;Shepard&Cooper,1992Theseresultssuggest(i)thatitispossibletoacquiretypicalknowledgeofcolorsimilaritywithoutdirectrst-personexperience,butalso(ii)rst-personexperienceisaparticularlyecientwayofdoingsoatleastforcolor.Giventhesemixedpriorresults,iti

2 sanopenquestionhowgen-erally,andhowprofo
sanopenquestionhowgen-erally,andhowprofoundly,blindindividualsknowledgeofvisualwordsdiersfromthatofsightedpeople.Inparticular,itisuncertainwhetherblindandsightedindividualssharedetailedknowledgeofvi-sualverbmeanings.Toaddressthisquestion,weacquiredthelargestsampletodateofsimilarityratingsforvisualverbsfromcongenitallyblindandsightedEnglishspeakingadults(seedata).Participantsjudgedthesemanticsimilarityofvisualverbsin-cludingverbsofvisualperception(e.g.topeek,topeer)andlightemission(e.g.sparkle,toshine).Wechosefteenverbsfromeachcategory,thusincludingnearlyallfrequentlyusedvisualverbsintheEnglishlanguage(Levin,1993).Englishhasafairlylargevocabularyofsuchwordsrelativetootherlanguages(Majidetal.,2018;Winteretal.,).Knowledgeofvisualperceptionverbswascomparedtoknowl-edgeoftactileperception(e.g.totouch,tofeel)andamodalknowledgeacquisition(e.g.toperceive,toexamine,todiscover).Lightemissionverbswerecomparedtoverbsofsoundemission,bothnon-agentive(e.g.boom,toclank)andagentive(e.g.togrunt,toshout).Intotal,eachblindandsightedcontrolparticipantmade2041judgments.Inaddition,wecollectedasecondsampleofsimilarityjudgmentsfromworkersonAmazonMechanicalTurk.Thissecondsampleofsighteddataenabledustogetabenchmarkoflexicalvariabilityacrosssightedparticipants.Wereasonedthatjudgmentswoulddieracrosspeopleduetomea-surementnoiseaswellasblindness-unrelatedindividualdi(e.g.education,memorycapacity).Ifblindnesssystematicallyaknowledgeofvisualverbmeanings,thenthesemanticsimilarityjudgmentsofasampleofblindindividualsshoulddiermorefromasightedsamplethantworandomlysampledgroupsofsightedspeakersdofromeachother.Ifso,sensoryexperiencemayhavespecialeonthelexicon,apartfromotherindividualvariation.Insum,thedataenableustomeasurehowrst-personsensoryexperienceinuencesthemeaningsofwordswhosereferentsaresensory.2.Methods2.1.Participantsvecongenitallyblind(20female)andtwenty-twosighted(11female)participantstookpartintheexperiment.Allparticipantswentthroughadetailedscreeninginterviewoverthephoneandre-portedhavingnocognitiveorneurologicaldisabilitiesandbeingEnglishnativespeakers(learnedEnglishbeforeage5).Blindpartici-pantsweretotallyblindfrombirth(hadatmostminimallightper-ception)andhadlosttheirvisionduetoabnormalitiesoftheeyesortheopticnerve(notduetobraindamage)(Table1).Sightedandblindparticipantswerematchedtoeachotherinage(blind:=44.86,=14,missingageinformationfor3participants;sighted:=50.64,=8.51)andlevelofeducation(blind:rangingfromsomecollege(nodegree)toDoctoralDegree,Mode=MastersDegree;sighted:rangingfromHighSchoolDiplomatoDoctoralDegree,Mode=BachelorsDegree).Threeparticipantsdidnotprovidesimi-larityjudgmentsforonewholesemanticcategory.Thus,weobtainedsimilarityjudgmentsontheperceptionverbsby22sightedand24blindparticipants,andontheemissionandmannerofmotionverbsby21sightedand25blindparticipants.Inaddition,weobtaineddataonAmazonMechanicalTurk(AMTurk)fromasightedreferencegroup(=303,henceforthsightedreferencegroup)thatwasthencomparedtotheratingsoftheblindadultsandsightedcontrols.MechanicalTurkparticipantswereallM.Bedny,etal. EnglishnativespeakersfromtheUnitedStates,accordingtoself-reportandAMTurkdata.Nootherdemographicdatawereavailablefortheseparticipants.Weexcluded37participantsbecausetheyeithergavethesameresponsetoallitems,oransweredthesurveyin4min,leaving266participantsintheanalyses.Nofurtherdemographicinformationwascollectedfromtheparticipants.2.2.StimuliThestimuliconsistedofthreebroadcategoriesofverbs(Table2VerbfrequencieswereobtainedfromtheSubtlexUSdatabaseBrysbaert&New,2009).WeselectedverbsfromtheLevin(1993)choosingthosethatwerefrequentandlikelytobefamiliartomostspeakers.Therstcategoryincludedverbsreferringtoagentiveex-periencesthatwereeithervisual(e.g.toglance,tostare,N=15),tactiletotouchtofeel=15)oramodal(e.g.toinvestigatetonotice=15)(log10(freq.)visual:=2.77,=1.22;tactile:=2.65,=0.8;amodal:=2.79,=0.9).Wehenceforthrefertothisclassastheperceptionverbssincealloftheminvolvedactsofknowl-edgeacquisition.Onevisualperceptionverb,toogle,wasreportedasunfamiliarbymostofbothsightedandblindparticipants,andsowasexcludedfromallanalyses,leaving44perceptionverbs(notin-cludedinfrequencycalculationsabove).Thesecondclassconsistedofverbsthatrefertoeventsintheenvironmentthatareperceptibleeitherthroughvisiononly(i.e.lightemission,e.g.tosparkletoshine,N=15)orhearingonly(soundemissione.g.tobuzz,tobang,N=30).Amongthesoundemissionverbs,halfreferredtosoundsgeneratedbyanimateagents(e.g.tobark)andhalfbyinanimateobjects(e.

3 g.toclang(log10(freq.)light:=2.02,=0.54;
g.toclang(log10(freq.)light:=2.02,=0.54;animatesound:=1.82,=0.56;inanimatesound:=1.88,=0.65).Thethirdcategoryincludedmannerofmotionverbs(e.g.tohobble,toroll,N=15;=2.11,=0.78).Inadditiontotheabovedescribedverbs,participantsalsojudged15mentalverbs(e.g.toenjoytotoleratebutthesewerenotrelevanttothehypothesesofthecurrentstudyandwerethusnotincludedinthereportedanalyses.Allverbswerepre-sentedasin(i.e.precededby).Notethatalthoughnoattemptwasmadetoincludeallpossibleverbs,thestimuliincludemostofthefrequentvisualverbswithintheEnglishlanguage.Verb-pairswereconstructedbymakingallpossiblepairingswithinthebroadsemanticcategories.Thesepairingsavoidedputtingtogetherverbsthatwerehighlydissimilarinmeaningandthusrequiredparti-cipantstomakemorene-grainedjudgments.Perceptionverbswerepairedwitheachotherwithinandacrossmodalities(topeektostaretosee,total946pairs).Amongverbsthatdescribedperceptibleevents,allemissionverbswerepairedwitheachotherbothwithinmodality(e.g.tobuzztoring)andacrossmodalities,(tobuzz,total990pairs).Motionverbswerepairedonlyamongstthemselves(105pairs.)2.3.TaskBlindandsightedparticipantscompletedanonlinesurveyinwhichtheyratedthesemanticsimilarityofverbpairs(e.g.toseetotouch)ascalefrom1(notatallsimilar)to7(verysimilar).Blindparticipantslistenedtothestimuliusingcommerciallyavailablescreen-readers;sightedparticipantsreadwrittenwordsonacomputerscreen.Eachparticipantwasaskedtorateallpossiblepairingsofverbswithineachbroadsemanticclassforatotalof2041wordpairsperparticipant(includingpairswithtoogleandthe15mentalverbs,whichwerelaterdroppedfromanalysis).Thesurveytookatotalof68h.Participantscompetedthesurveyfromhomeoverthespanof24weeks.Eachparticipantwasgivenausernameandpassword.Theywouldlogintotheiraccount,completeaportionofthesurvey,andthenreturntoitattheirconvenience.Allparticipantswerewalkedthroughthesurveyinstructionseitheroverthephone(blindparticipants)orinperson(sightedcontrols)(seeSupplementaryInformationforinstructions).Participantswereaskedtousethefullscaleof17.Theyweretoldtherewerenocorrectanswersandremindedtoratehowsimilartheverbswereinmeaning,notinsoundorspelling.Beforestartingthemainsurvey,participantscom-pletedapracticesessionwith100pairsofanimalnouns(e.g.thebearthetigerThesurveywasdividedintothreesections:perceptualexperienceverbs,emissionverbs,andmannerofmotionverbs.Theorderofthe Table1Participantsdemographicinformation.CauseofblindnessLightperceptionEducationlevelRetinopathyofPrematurityBAinprogressRetinopathyofPrematurityRetinopathyofPrematurityBAinprogressRetinopathyofPrematurityRetinopathyofPrematurityHereditaryeyecondition(unspeciBAinprogressNorrie'sSyndromeRetinopathyofPrematurityBAinprogressRetinopathyofPrematurityRetinopathyofPrematurityCongenitaldeformationofrodsandconesRetinopathyofPrematurityRetinopathyofPrematurityBornwithunderdevelopedeyes(unspeciBAinprogressRetinopathyofPrematurityRetinopathyofPrematurityRetinopathyofPrematurityRetinopathyofPrematurityRetinopathyofPrematurityRetinopathyofPrematurityProfessionalDegreeRetinopathyofPrematurityRetinopathyofPrematurityBAinprogressLeber'sCongenitalAmaurosisBlind(=25)Sighted(=22)M.Bedny,etal. sectionswascounterbalancedacrossparticipants,aswastheorderofwordswithinapair(i.e.halfoftheparticipantswerepresentedwithtostareandtheotherhalftostaretolook).Theorderofverbpairswithineachsectionwasrandomizedacrossparticipants.Wein-cludedcatchtrialstoensureparticipantswereattendingtothetask.Vegetablenamesdisguisedasverbs(e.g.tocarrottopotato)appeared10%ofthetime.ParticipantswereinstructedtoenterVforthesepairs,ratherthanasimilarityrating.Participantsinthesightedreferencegroupratedsubsetsof100wordpairs.Eachpairwasratedby20participants.Theorderofpairsandwordswascounterbalancedacrossparticipants.TheAMTurkrat-ingswerecombinedtogenerateasinglecompletedatasetandwereneveranalyzedatthesingle-subjectlevel,sincenoparticipantgener-atedafulldataset.2.4.Analyses2.4.1.GeneratingsemanticdissimilaritymatricesWecreatedsemanticdissimilaritymatricesbothontheparticipantsrawsimilarityjudgmentsandonthenormalizedscores.Thenormalizedscoreswereobtainedbyrstz-scoring(=0,=1)thesimilarityjudgmentswithinparticipantstoaccountforindividualdierencesinLikertscaleuse,thennormalizingwithinthethreemainsemanticcategories(perception,emission,mannerofmotion)suchthateachverbpairhadasimilaritydistancewithin[0,1]range.Wegeneratedindividualsubjectssimilaritymatricesseparatelyforthethreemainsemanticcategories(i.e.perception,emissionandmannerofmotionverb

4 s),aswellasforeachsemanticsubcategory(i.
s),aswellasforeachsemanticsubcategory(i.e.sight,touchandamodalperception;lightandanimate/inanimatesoundemission).Groupsimilaritymatriceswerethencreatedbyaveragingindividualsubjectmatricesacrossparticipantsforeachpairofverbs.Dissimilaritymatriceswerenallyobtainedasthemaximumvalueofthescale(7fortherawdata,1forthenormalizedscore)minusthesimilaritymatrices(Fig.1).Dataarepubliclyavailableat:2.4.2.Acrossandwithin-GroupagreementWeusedblindandsightedparticipantsnormalizedsimilarityrat-ingstomeasurewithinandacrossgroupcoherence.Toquantifythesimilarityoflexicalknowledgebetweenblindandsightedparticipants,rstcomputedtheSpearmansrhorankcorrelationbetweenthesightedreferencegroupmatrixandtheaverageblindandsightedgroupdissimilaritymatrices,respectively.Asthegroupaveragematricesarelessnoisythanthesingle-subjectones,group-levelcorrelationsprovidethebestbetween-groupsimilarityestimates.However,theydonottakeintoaccountwithin-groupvariance.Totestwhetherthebetween-groupcorrelationsarereliableacrossindividuals,wecorrelated(Spearmanrhorankcorrelation)thesightedreferencegroupmatriceswiththeblindandsightedindividualsingle-subjectmatrices.WereporttheFisher-Ztransformedaveragecorrelationsofeachsubjecttothesightedreference,whichgivesameasureofhowcorrelatedeachindividualblindorsightedtestsubjectistoarandomlysampledgroupofsightedsubjects.Thesignicanceofthesecorrelationswasthentestedusingst-testsacrosstheFisher-Ztransformedsingle-subjectcorrela-tions.CorrelationswerecomputedonthenormalizeddissimilaritymatricesusingtheHmiscpackageinR(Harrell&Dupont,2014Finally,wemeasuredwithin-groupcoherencetocomparethe Table2Completelistofverbstimulibysemanticsubcategory.Notethatallverbswereprecededbytheininitivemarkerintheactualexperiment.VisualTouchAmodalgawkcaresscharacterizegazedabclassifyglancefeeldiscoverglimpsegripexamineleernudgeidentifylookpatinvestigatepeekpetlearnpeerpinchnotescanprodnoticeseepubperceivespotscrapequestionstarestrokerecognizeviewtapscrutinizewatchticklesearchMannerofMotionAnimateSoundInanimateSound*Excludedfromanalysesastoounfamiliartoparticipants.M.Bedny,etal. degreetowhichblindandsightedspeakersagreeamongstthemselvesonthemeaningsofthetestedverbs.FirstwecalculatedtheKendallcientofConcordanceforeachverbtypeandgroup.TheKendallWisanestimateofthecorrelationbetweenallpairsofparticipantswithinagroupandhaspreviouslybeenusedtomeasuresubjectagreementonsemanticsimilaritymeasures(Barsalou&Sewell,1984;Barsalou,1987,1993).SinceKendallsWisagroup-wise-metricwithnovariance,weusedaleave-one-subject-outproceduretotestfordif-ferencesinwithin-groupcoherenceacrossgroups.Thisanalysiscorre-latedresponsesofeachparticipanttotheirowngroupholdingtheirowndataout(n-1).TheKendallWandleave-one-subjectoutprocedureproducednearlyidenticalmeasuresofwithin-groupcoherence.2.4.3.MultidimensionalscalingMultidimensionalscaling(MDS)wasusedtovisualizethenaturalclusteringofthegroupdissimilaritymatricesusingthebasicMDSap-proachasimplementedintheSMACOFpackageinR(Mair,deLeeuw,&Groenen,2015).Specically,wecomputedtwo-wayintervalMDSmodelsusingtheStressMajorizationofaComplicatedFunction(SMACOF)approach,whichminimizesthestress-functionbymeansofaniterativemajorizationprocess(i.e.theSMACOFalgorithm;DeLeeuw&Heiser,1977,DeLeeuw&Mair,2011).Wedeterminedthedi-mensionalityofthemodelsbycreatingscreeplots,whichshowthestressvaluesasafunctionofdimensions(rangingbetween2and10,seeSupplementaryFig.1),andselectingthevalueatwhichaddingdi-mensionsnolongerimprovesthemodeltsubstantially(i.e.theelbowofthecurve).Basedonthisheuristic,wereporttheresultsofbasicMDStwith4dimensions.ForeachMDS,thegoodness-of-tisgivenastheKruskalsnormalizedstress-1value.Thesmallerthestressvalue,thebetterthetofthesolution.Thetsignicanceisevaluatedwithapermutationtestwhichprovidesanulldistributionofstressvaluesbasedontherandompermutation(=1000)ofthedissimilarityma-trices(Mair,deLeeuw&Groenen,20152.4.4.HierarchicalagglomerativeclusteringanddendrogramsDatawerehierarchicallyclusteredbeginningwitheachobservationasitsown,single-itemclusterandprogressivelymergingclustersupthehierarchy(Murtagh&Legendre,2014).WeusedWard'sminimumvariancecriteriontocreateclusters,whichmergesclustersminimizingthetotalwithin-clustervariance(i.e.theweightedsquareddistancebetweenclustercenters)relativetootherpossiblemerges(Murtagh&Legendre,2014).Thus,thedendrogramisbuiltfromthebottomup:eachverbisoriginallyassignedtoitsownclusterand,ateachstep,thetwoclosestclusters(i.e.thosewhosemergedresulthasthelea

5 stvar-iance)aremergedintoanew,largerclus
stvar-iance)aremergedintoanew,largercluster,eventuallyconvergingattheoriginofthebranching.DistancesbetweenclustercenterswererecomputedbytheLanceWilliamsdissimilarityupdateformulaMurtagh&Legendre,2014).Thisproceduretendstoleadtocompactandsphericalclusters.AnalysesweredoneusingthepvclustpackageinSuzuki&Shimodaira,2006),whichassessestheclustersusingmultiscalebootstrapresampling.Thisprocedurecomputessev-eralclustersbyresamplingoverspecicverbpairs.Then,thereliabilityacrossresamplingiscalculatedforeachbranchingandusedtogeneratenaldendrogram.Thisalgorithmisacomputationallyfastwayofimplementingstandarddouble-bootstrap,inwhichthestandarderroreachbootstrapreplicationisestimatedusingbootstrapresamplingwithintheresampledreplication.3.Results3.1.BlindindividualsdistinguishvisualverbsfromverbsinothermodalitiesandamodalverbsForsightedandblindparticipants,group-wiseMDSrevealedana-logoussemanticstructuresacrossgroups.Fig.2showstherst2di-mensionsthatemergefortheblindandsightedgroups.Perceptionverbsseparateintothreemajorclustersbymodality(sight,touch,amodal).Amongtheseclusters,visualverbs(e.g.topeek)andamodalverbs(e.g.toinvestigate)arecloser(moresimilar)toeachotherthantotouchverbs(e.g.tofeel).Likewise,theemissionverbsseparateac-cordingtolightversussoundverbsand,amongsoundverbs,intoagentiveandnon-agentiveverbs(Fig.2).Similarresultswereobtainedwhenrawsimilarityscoreswerecomparedacrossgroupsusingstan-dardparametricstatistics(seeSupplementaryInformationandSup-plementaryFig.2fordetails).Forbothblindandsightedgroups,thebasicMDStsweregood,i.e.stressmeasureswerelowandcomparableacrossgroupsforallverbcategories(perceptionverbsstress:sighted0.12,blind0.12;emissionverbsstress:sighted0.14,blind0.13;within-groupgoodness-of-tsignis0.0001).3.2.Preservedsemanticsimilaritystructureofvisualverbsinblindpeople3.2.1.Agreementofsemanticsimilarityratingswithinandacrossgroups.Doblindindividualsmakesimilardistinctionsamongvisualverbsassightedindividuals?Toaddressthisquestion,werstaskedwhetherthesemanticsimilarityratingsofblindadultswereascorrelatedwiththoseofagroupofsightedparticipantsastwoindependentgroupsofsightedparticipantsaretoeachother.Atthegrouplevel,theaverageratingsofblindindividualsforvisualverbswereashighlycorrelatedwiththoseofthesightedreferencegroupasthetwosightedgroupsweretoeachother(visualperceptionverbs:sightedtosightedreferencegroup(89)=0.84,blindto Fig.1.Groupaveragesemanticdissimilaritymatricesforperceptionandemissionverbs,blind,sightedandsightedreferencegroups(toptobottom).Darkerredcorrespondstomoresimilar.M.Bedny,etal. sightedreferencegroup(89)=0.81;lightemissionverbs:sightedtosightedreferencegroup(103)=0.91,blindtosightedreference(103)=0.93,Fig.3A).Totakevariabilityacrossindividualsintoaccount,weiterativelycorrelatedtheratingsofindividualblindandsightedparticipantstothemeanratingsofthesightedreferencegroup.Forvisualperceptionverbs,theratingsofblindandsightedindividualswereequallywell-correlatedwiththoseofthesightedre-ferencegroup(sightedtosightedreferenceMFisher-z=0.51,=0.12;blindtosightedreferenceMFisher-z=0.55,=0.15;-testacrossgroups:(41.55)=-0.84,Weobservedthesamepatternforverbsoflightemission.Thein-dividualratingsfromblindandsightedparticipantswereequallywellcorrelatedwiththoseofthesightedreferencegroup(sightedtosightedMFisher-z=0.68,=0.16;blindtosightedreference,=0.7,=0.17;two-sample-testacrossgroups:(43.82)=0.25,Wenextaskedwhethervisualverbswerelesssimilarbetweenblindandsightedparticipants,ascomparedtoamodalverbs.Onthecontrary,relativetothesightedtestgroup,theratingsofblindparticipantswereslightly,butnotsignicantlymoresimilartothoseofthesightedre-ferencegroupforvisualverbsascomparedtoamodalverbs(2Group(blind,sighted)×2Modality(visual,amodal)repeatedmeasuresANOVAsonblindtosightedreferenceandsightedtosightedreference;visualperceptionvs.amodalperceptionverbs:maineectofGroup(1,44)=0,0.9,maineectofModality(1,44)=5.39,Group×Modalityinteraction(1,44)=3.14,0.08;lightemissionvs.motionverbs:maineectofGroup(1,43)=0.03,0.88,mainectofModality(1,43)=22.89,0.0001,Group×Modality(1,43)=0.06,Finally,weconductedagroupcoherenceanalysistodeterminewhetherblindandsightedparticipantsdieredwithregardtoagree-mentwithintheirowngroup.First,wecomputedtheKendallcient,whichisanestimateoftheagreement/cor-relationbetweeneverypairofsubjectswithinagroup,foreachverb-typeandsubjectgroup(SupplementaryFig.3,SupplementaryTable).Blindandsightedspeakerswereequallycoherentintheirjudgmentsforvisualverbs(visu

6 alperceptionverbs:sighted=0.31,blind=0.3
alperceptionverbs:sighted=0.31,blind=0.33;lightemisionverbs:sighted=0.38,blind=0.39,allcantlygreaterthanzero,0.0001).BecauseKendallisagroup-wisemeasure,weusedaleave-one-participant-outproceduretoestimatevariabilityacrossparticipants,correlatingeachblindandsightedparticipanttotheirowngroup,holdingtheirdataout(Fig.3Theseanalysesrevelaedthatblindandsightedparticipantswereequallycoherentamongstthemselvesforverbsofvisualperceptionandlightemission(two-sample-testblindvs.sightedvisualperception(42.86)=1.3,0.2,lightemissionverbs(43.87)=0.48,=0.63;visualvs.amodaperception.Blindparticipantswerealsonolesscoherentforvisualverbsthanforamodalverbs,relativetothesightedeitherforperceptionoremissionverbs(perceptionverbs:Group(blind,sighted)×VerbModality(visual,amodal)repeatedmeasuresANOVAsmaineectofGroup(1,44)=0.69,0.4,maineectof(1,44)=14.72,0.0005,Group×Modalityinteraction(1,44)=1.81,0.18;emissionverbs(lightemissionvs.motionverbs:maineectofGroup(1,43)=0.74,0.4,maineectof(1,43)=18.51,0.0005,Group×Modalityinteraction(1,43)=0,0.97).Rather,bothsightedandblindgroupsshowedhighercoherenceforthecategoryies,hencethesigniFig.3).Theseresultsindicatethatthewithin-categorysimilaritystructureofvisualverbsispreservedamongblindindividuals.Tworandomsamplesofsightedparticipantsagreestothesameextentasblindandsightedsamples.Furthermore,blindindividualsshowthesamedegreeofwithin-groupcoherenceforvisualgroupsassighted3.2.2.Whatdoblindindividualsknowaboutvisualverbs?:MDSandclusteringanalysisWeusedMDSaswellashierarchicalclusteringanalyseswithinthevisualverbclassestogaininsightintothecontentofblindandsightedsknowledgeaboutthesewords.Forsightperceptionverbs,MDSproducedgoodandcomparabletsacrossgroupsbutdidnotyieldinterpretabledimensions(MDSstressvaluesighted0.08,blind0.07,teachgroup0.0001).Hierarchicalclusteringana-lysesrevealedthatforbothblindandsightedgroups,visualperceptionverbsclusteredintointense,prolongedactsofseeing(e.g.toleer,to Fig.2.Blind(top)andsighted(bottom)groupMDSresultsforperception(left)andemission(right)verbs.Firsttwodimensionsshown.M.Bedny,etal. gawk,tostare),briefactsofseeing(e.g.topeek,toglance,toglimpse)andactsofgenericlooking(i.e.tolook,tosee,toviewFig.4Forlightemissionverbs,previouslinguisticanalysesidentiedin-tensityandperiodicityastwocentraldimensionsofmeaning(Faber&Usón,1999).WethereforeaskedwhetherMDSwouldorganizelightemissionverbsalongthesedimensionsforblindandsightedgroups.Theverbtoblinkwasnotincludedintheoriginallinguisticwork,possiblybecauseithasacommonmeaningunrelatedtolightemissiontobrieyshuttheeyes;WordNet,2010)(Faber&Usón,1999thereforeexcludedtoblinkfromthisanalysis.Consistentwithlinguisticanalyses,thedimensionsofintensityandperiodicityemergedasthetoptwoforboththesightedandtheblindgroups.Forinstance,verbssuchtoblazeclusteredtogetheralongtheintensitydimensionandseparatedfromtotwinkletoglow.Ontheperiodicitydimension,totwinkleclusteredtogetherandseparatedfromtoglowFig.4B).TheMDStforthelightemissionverbswasgood(lowstressvalues)andcomparableacrossblindandsightedgroups(stressvalue:sighted0.08,blind0.07,goodnessofteachgroup0.0001).3.3.ThesemanticsimilarityratingsfortouchperceptionandsoundemissionverbsaremoreconsistentamongcongenitallyblindindividualsAsforvisualverbs,weusedKendallsWandleave-one-subjectoutanalysistomeasurewithin-groupcoherence.BlindpeoplessemanticsimilarityratingsforsoundandtouchverbsweremoreconsistentacrosssubjectswithintheirowngroupthanthoseofthesightedFig.4FigureS1,TableS1).Thiswasindependentlytrueforanimatesoundemission(sightedtosighted=0.59,=0.16;blindto=0.73,=0.13;two-sample-testacrossgroups(41.53)=2.33,0.02),inanimatesoundemission(sightedto=0.16;blindtoblind=0.52,=0.11;twosample-testacrossgroups(36.19)=2.59,andtouchperceptionverbs(sightedtosighted=0.58,SD=0.17;blindtoblind=0.71,=0.1;two-sampleacrossgroups(36.93)=2.17,0.04,allwithin-groupe0.0001).Blindparticipantsratingsfortheseverbcategorieswerealsomarginallymorecorrelatedwiththesightedreferencegroupthanwerethoseofthesightedparticipants(seeSupplementaryInformationfordetails).Thispatternwasspecictosoundemissionandtouchverbsandwasnotobservedforeitherofthecontrolverbclasses(amodalperceptionverbs:sightedtosightedreference=0.5,=0.13;blindtosightedreference=0.45,=0.14;two-sample-testacross(43.56)=-0.96,0.34;mannerofmotionverbs:sightedtosightedreference=0.47,=0.12;blindtosightedreference=0.47,=0.15;two-sample-testacrossgroups(43.99)=0.04,Whencomparingtouchperceptiontotheamodalcontrolverbs,blindsu

7 bjectsshowedhigherwithin-groupcoherenceo
bjectsshowedhigherwithin-groupcoherenceonthetouchverbsrelativetothesighted(i.e.weremorecorrelatedwiththeirowngroup(within-groupcoherence2Group(blind,sighted)×2Modality(touch,amodal)repeatedmeasuresANOVAs;maineectofGroup(1,44)=1.94,0.17,maineectofModality(1,44)=30.31, Fig.3.A.Correlationsofblind(lightcolors)andsighted(darkcolors)groupstothesightedreferencegroup,computerongroupsaveragenormalizedsi-milarityratings.Animateagentiveandnon-agentive(inanimateobject)soundverbsareshownseparatelyindierentshadesofblue.B.Within-groupco-herencemeasuredassinglesubjectscorrelationstotheirowngroup(blindtoblind:light;sightedtosighted:dark),usingaleave-one-subject-outproce-dure(SeeSupplementalMaterialsforKendallCoherence).Errorbars:±standarderrorofthemean.Signicantgroupdierencesaremarkedwithp-value.Animateagentiveandnon-agentive(in-animateobject)soundverbsshownseparatelyinerentshadesofblue.M.Bedny,etal. 0.0001,Group×Modalityinteraction(1,44)=4.93,andweremorecorrelatedtothesightedreferencegroup.Thegroup-by-modalityinteractiondidnotreachsignicanceforsoundemissionverbs(within-groupcoherence2Group(blind,sighted)×2Modality(sound,motion)repeatedmeasuresANOVAs;maineectofGroup(1,43)=5.05,0.03,maineectofModality(1,43)=2.49,0.12,Group×Modalityinteraction(1,43)=2.79,Despitethesesubtleincreasesinwithin-groupcoherenceoftactileandsoundemissionverbsforblindspeakers,hierarchicalclusteringandMDSanalysesrevealedaqualitiativelysimilarstructureandquanti-tivelysimilartsfortheseverbsacrossblindandsightedgroups(Fig.5touchverbsstressvalue:sighted0.11,blind0.09,group-wisegoodness-s0.0001;soundverbsanimate:sighted0.09,blind0.08;soundverbsinanimate:sighted0.13,blind0.11;group-wisegoodness-s0.0001).Forinstance,forbothgroupstouchverbssepa-ratedbetweenwhole-handsmovements(e.g.tostroke,torub)gertipsactions(e.g.topinch,toprod).Takentogether,theseresultssuggestthatblindandsightedindividualsrelyonsharedknowledgewhenmakingsimilarityjudgmentsoftouchandsoundemissionverb,butthesemeaningsaresomewhatmorelikelytobeconsistentacrossblindthansightedparticipants.4.Discussion4.1.PreservedrepresentationsofvisualverbsinblindnessThepresentndingsrevealsimilaritiesbetweenvisualverbknowledgeamongcongenitallyblindandsightedpeople.TheseminalworkofLandauandGleitman(1985)showedthatchildrenwhoareblindbegintoproduceandunderstandtheverbsthesameageassightedchildren.LandauandGleitman(1985)posedthatblind(andsighted)childrenacquirethesemeaningspartlybyrelyingonlanguageitselfi.e.occurindierentsyntacticframes(lookatbutnotseeatIseethatbutnotIlookthat).Thepresentresultsextendthesepreviousndingsbyrevealingfurtherrichnessintheknowledgethatblindandsightedpeopleshareaboutvisualverbs.First,wethatblindadultstreatthemodalityofperceptualaccessasacentraldiagnosticfeatureofperceptionverbs;theydonotconateverbsdescribingvisualaccesstotheworld(e.g.tolook)witheithertactile(e.g.tofeeltotouch)oramodalverbstoinvestigatetodiscover).AkeyobservationofLandau&Gleit-soriginalexperimentwasthattheblind4-year-old,Kelly,raisedherhandsexpectingtoexaminesomethingwheninstructedtoKellyinterpretedtheinstructiontoobservewiththehands',referredtoherself.UnliketheexperimentwithKelly,thecurrentstudypresentedverbswithoutspecifyingwhethertheagentwassightedorblind.Inthisunspeciedcontext,blindandsightedadultsalikedistinguishverbssuchaslook,starefromamodalandtactileverbs.Usingfeaturegeneration,Lencietal.(2013)foundthatblindpeoplegeneratevisualfeatureswhenaskedthemeaningofvisualverbs.Thus,althoughspeakersareabletoandappropriatelyapplyverbssuchastoblindagents,thepresentdatasuggestthatwhennoagentisspecied,visualverbsimplyasightedagentevenforblindspeakers.Wefurtherndthatthewithin-categorysimilaritystructureofvi-sualverbsispreservedinblindness.Blindpeoplesratingsofvisualverbswerenotmorenoisy,orheterogeneousthanthoseofthesighted,unlikethecaseofcolorsimilarityjudgments,whicharemorevariableacrossblindindividuals(Saysanietal.,2018;Marmor,1978;Saysanietal.,1992).Blindindividualsrepresentthetemporalstructureofvi-sualperceptionverbs(e.g.distinguishing)anddistin-guishlighteventsalongdimensionsoftemporalfrequency()andintensity().Insum,blindandsightedpeoplesharedetailedknowledgepertainingtoactsofvisualperceptionandeventsoflightemission.Onequestion,however,iswhetherthereportedpair-wisesimilarityjudgmentsprovideanaccuratemeasureofwordmeaningsorconcepts.Perhaps,blindparticipantsperformedthetaskbasedonjustthe Fig.4.A.Hierarchicalclusteringdendrogramsforsightperceptionverbsacrossgroups.B.Lig

8 htemissionverbsMDSanalysis,rsttwodimensi
htemissionverbsMDSanalysis,rsttwodimensionsshown.M.Bedny,etal. distributionofco-occurrenceofwordsinothersspeech,notare-presentationofthewordsmeaning.Wethinkthisexplanationisun-likelyforanumberofreasons.First,semanticsimilarityjudgmentsofthekindusedinthecurrentstudyarestronglycorrelatedwithperfor-manceonotherconceptualtasks,suchascategorization(isarobinabird?)(e.g.Ripsetal.,1973).Second,asnotedabove,smallerscalestudiesusinglanguageproduction,andexplicitdenitions,havealsofoundthatblindindividualsusevisualwordsinmeaningfulwaysduringcommunication(Landau&Gleitman,1985;Lencietal.,2013Third,blindandsightedpeoplerecruitthesamebrainregionswhenmakinginferencesbasedonsentenceswithvisualperceptionverbsBedny,Pascual-Leone,&Saxe,2009;Koster-Hale,Bedny,&Saxe,Thepresentdataalsoprovidesevidencethatthesemanticsimilaritytaskusedhereissensitivetosubtlebetween-groupdierencesinmeaningknowledgeasevidencedbyblindindividualssomewhatmoreconsistentjudgmentsoftactileperceptionandsoundemissionverbs.Asagroup,blindadultsratingsfortheseverbsareindistinguishablefromthoseofthesighted(i.e.blindandsightedspeakersusesimilarcriteriatodistinguish)butblindpeople'sresponsesarelessheterogeneousacrossparticipants.Whymightthisbethecase?Onepossibleinterpretationisthatsuperiorabilitiesintouchandhearingamongblindindividualstranslatetomoreprecisemeaningsfortouchandsoundverbs.Indeed,blindindividualsoutperformthesightedonsometactileandauditorytasks(Gougouxetal.,2004;VanBoven,Hamilton,Kauman,Keenan,&Pascual-Leone,2000).Suchaninter-pretationofthecurrentdataseemsunlikelytous,however,formultiplereasons.First,blindindividualsonlyoutperformsightedindividualsonasubsetoftactileandauditoryperceptualtasksandthebenetsaresubtle.Forexample,blindindividualsarebetteratperceivingBraille-likepatternsbutnotothertypesoftactilestimuli(Grant,Thiagarajah,&Sathian,2000).Thereareevensomeauditorytasks,suchaslocalization Fig.5.MDSresultsfortouch,animate/agentivesoundandinanimate-objectsoundemissionverbsMDSforblindandsightedgroups.Toptwodimensionsshown.Notethatdimension2wasrotatedfortouchverbs,sightedgroupandinanimatesoundverbs,sightedgroup.M.Bedny,etal. intheverticalplane,onwhichsightedindividualsperformbetterZwiers,VanOpstal,&Cruysberg,2001).Itisunlikelythatsuchsubtleandunevensensoryenhancementsinblindnesscauselexicaldiences.Furthermore,iftotalabsenceofrst-personvisualexperiencedoesnotmakevisualverbmeaningsmorenoisyinblindadults,itisunlikelythatsubtleimprovementsintactileorauditoryexperiencewouldmakethemlessnoisy.Wehypothesizeinsteadthatdierencesincommunicativere-levanceaccountfordierencesinsemanticsimilarityjudgmentcon-sistencyacrossblindandsightedspeakers.Sinceblindindividualsliveamongasightedmajority,beingblindoneselfdoesnotsubstantiallyreducethefrequencyofencounteringvisualverbs.Bycontrast,beingblindmightmakeonemorelikelytouseverbsofsoundemissionandtouchperceptionduringcommunication.Blindpeople,forexample,maybemorelikelythanthesightedtoasktotouchorholdsomething,ortodescribeaneventsauditoryproperties.Thisaccountisspeculativeandrequiresfuturetesting.Measuringthefrequencyofsoundandtouchverbsinnaturalisticspeechandwritingofindividualswhoareblindcouldproviderelevantevidence.Regardlessoftheunderlyingcause,thepresentndingssuggestthatsemanticsimilarityjudgmentscaninprinciplemeasurebetween-groupdierencesinmeaningthatresultfromblindness,makingtheobservationthatthesemanticsimi-larityjudgmentsforvisualverbsarenotalteredbycongenitalblindnessallthemorecompelling.Atthesametime,thereportedresultsareonlyapartialmeasureofwhatblindandsightedspeakersknowabouteventsoflightandvisualperceptionandthereforearenotthenalwordonwhetherthereareknowledgedierencesacrossthesepopulations.Thereareanumberofreasonswhysemanticsimilarityjudgmentsofthekindusedinthecurrentstudydonotprovidedirectaccesstopsychologicalspaceofconceptualrepresentations(Medin,Goldstone,&Gentner,1993;Shepard,1987;Tenenbaum&Griths,2001).Forone,semanticsimilarityjudgmentsarehighlysensitivetocontext(e.g.Gauker,1994;Goodman,1972;Goldstone,1994).Peoplessimilarityjudgmentsshiftrapidlyandexibly.Twoobjectsthatseeminitiallywildlydissimilar(e.g.childrenandjewelry)caneasilybejudgedassimilarwhengiventherightframe(objectstorescuefromaburningBarsalou,1983).Grayismoresimilartowhitethanblack,whenthecontextishaircolor,butmoresimilartoblackthanwhite,whenthecontextisclouds(Goldstone,1994;Medin&Shoben,1988).Indeed,thecontextmaybejusttheorderofpresentationofthepairofwords:saythatsurgeonsarelikebutchersmea

9 nssomethingdierentthantosaybutchersareli
nssomethingdierentthantosaybutchersarelikesurgeonsMedinetal.,1993).Similarly,RossandMurphy(1999)foundthatcollegestudentsrecognizetwoorthogonalwaystoorganizefoodcategories:bytaxonomy(e.g.milkandicecreamarebothdairyfoods)andbysocialcontext(e.g.milkandbagelsarebothbreakfastfoods).Whensimplyaskedaboutthetwofoods,participantstendedtoprioritizetaxonomicalcategories,judgingmilkmoresimilartoicecream;butwhenaskedtomakeinferencesaboutsocialbehaviours(e.g.inclusioninanovelritual),participantsmadepredictionsbasedonsocialscripts.Insum,thereisclearevidencethatpeopleknowmoreaboutconceptsandwordsthanwhatiscapturedbyaparticularsemanticsimilarityjudgmenttask.Wethereforedonotinterpretthepresentsimilarityjudgmentsasdirectlyrevealingtherepresentationalspaceofthemeaningofvisualverbs,ineithersightedorblindindividuals.Instead,similarityjudg-mentsprovideasensitivebutincompleteestimateofpeoplesknowl-edgeofadomain.Thus,thebestexplanationofourresultsisthatsightedandblindindividualsshareboth(i)relevantknowledgeofthemeaningsofvisualverbs,and(ii)commonpragmatics,thatleadthemtointerprettherequestforsimilarityjudgmentsintermsoftherelevantrespectsforthisdomain(e.g.modality,temporalduration,etc).4.2.SomeopenquestionsThepresentresultsdemonstratethatblindindividualsknowimportantaspectsofvisualverbmeanings,butdonotspeaktohowtheseareacquired.Howdoblindindividualslearnthemeaningsofvisualverbs?Languageitselfislikelyarichsourceofinformation.Themeaningsofvisualverbsmaybepartlyinferedfromthemeaningsofthephrasesinwhichtheyoccur.Landau&Gleitmanarguedthatblindchildrenusesentenceframestodistinguishbetween(1985).Analogously,hearingthelightashedonandoasopposedtothelightglowedmightprovidecluestothetemporalstructureoflightevents.Wordsthatoccurinsimilarlinguisticenvironmentshavemoresimilarmeanings(Landauer&Dumais,1997).Blindlearnerscouldusethemeaningsofwordstheyalreadyknow,theirinterpreta-tionofthediscourseaswellassocialandpragmaticcuestoconstrainhypothesesaboutvisualwords(Bloom,2002;Clark,1987;Frank&Goodman,2014;Markman&Wachtel,1988;Ouyang,Boroditsky,&Frank,2016;Tomasello&Barton,1994).Forexample,whenablindindividualhearsasightedspeakercommentonstarinthenightsky,shemightinferthatforsightedpeopleoccursatadistanceandthatcanbeascertainedthroughvisionbutnotthroughaudition,sincesheherselfcannotobservetheglowingofthestar.Whenshehearssomeonecomplainofbeingstaredatlunch,shemightinferthatstaringissomethingthatcanlastalllunchlong.Inthiswaylinguistic,socialandpragmaticinformation,togetherwithasharedinnateendowmentforprocessingit,acttoalignthemindsofsightedandblindspeakers.Understandingexactlyhowthisoccursisanimportantgoalforfutureresearchthatcouldbeattained,inpart,bystudyinglanguageacquisitioninblindchildren.Afurtheropenquestionconcernshowknowledgeofvisualverbsamongblindindividualscomparestoknowledgeaboutothervisualdomains.Priorstudiessuggestthatvisualverbsarenottheonlypartofvisualknowledgethatispreservedinblindness.Asnotedinthein-troduction,evenblindchildrenknowthatcolorsarephysicalpropertiesthatareperceptibleonlywiththeeyes(Landau&Gleitman,1985Blindadultssimilarityjudgmentsoncolorsproducedacolorwheelqualitativelysimilartothatofsightedadults(e.g.blueismoresimilartogreenthanred).However,relativetothesighted,thereishighervariabilityincolorsimilarityknowledgeacrossblindindividualsMarmor,1978;Saysanietal.,2018;Shepard&Cooper,1992).More-over,blindindividualsarelesslikelythanthesightedtousecolorduringsemanticsimilarityjudgments(Connolly,Gleitman&Thompson-Schill,2007).Whyarethemeaningsofvisualverbslessvariablethanthesimilaritiesamongcolors?Onepossibilityisthatsomesuchasbetweencolorsimilarity,islessinferentiallyre-levantandthereforeblindindividualsarelesslikelytolearnit.Connollyetal.(2007)suggestthatforblindindividualsareknowntobered,[but]nothingfollowsintermsoftheusefulnessofthisfactinreasoningaboutstrawberries.Whetheranagentismightlicensemoreinferences(e.g.aboutwhattheagentknows),thanknowingthecolorofanobject.Anotherpossibilityisthatsomevisualinformationismoreeasilyaccessiblethroughlanguage.Infuturework,computationalmodelscouldbeusedtoaskwhichvision-relatedinformationismostavailableintext.Wewouldthenbeinapositiontocompareandcontrastwhatisavailableandwhatblindspeakersactuallylearn.Furtherworkisneededtofullycharacterizeblindandsightedsknowledgeaboutvisionandlight.Thestimuliusedhereweresinglewordsandverbsinparticular.Themeaningsofsinglewordsarenecessarilygeneralandexible,therefore,perhapsmostlikelytoberobusttochangesinour

10 idiosyncraticlifehistories.Whenwordsarec
idiosyncraticlifehistories.Whenwordsarecombinedintophrases,thegeneratedmeaningsaremorethanthesumoftheirparts,andtheinferencesthatfollowadditionallydependonreal-worldknowledge.Forexample,theinferencesonemakesbasedonthephraseAbigailglancedatLeosfaceacrosstheroomdependonanunderstandingofvisionthatgoesbeyondthemeaningof.DoesAbigailknowthecolorofLeoseyes?Doesshehaveinformationabouthismood?Whetherheshungry?DoessheknowwhetherLeoiswearingahat?Thecolorofhisshoes?FutureworkcomparingblindandsightedM.Bedny,etal. sinferencesaboutthevisualexperiencesofotherswouldrevealfurtherinsightintothecontributionofvisiontoknowledgeacquisition.5.ConclusionsThepresentndingsrevealarichsetofknowledgeaboutvisionandlightthatissharedamongsightedandblindindividuals.Theseresultsprovideacompellingillustrationofthesharednatureofmeaninganditsresiliencetodramaticchangeinrst-personsensoryhistories.Wewouldliketothankmembersoftheblindcommunityforgen-eroslygivingoftheirtimetomakethisprojectpossible.WethankWilliamJohnsonforhisassistanceindatacollectionandgreatfullyacknowledgetheDavidandLucilePackardFoundationforfundingthisAppendixA.SupplementarymaterialSupplementarydatatothisarticlecanbefoundonlineatBarsalou,L.W.(1983).Adhoccategories.Memory&Cognition,11(3),211Barsalou,L.W.(1987).Theinstabilityofgradedstructure:ImplicationsforthenatureofConceptsandconceptualdevelopment:Ecologicalandintellectualfactorsincategorization,10139Barsalou,L.W.(1993).Flexibility,structure,andlinguisticvagaryinconcepts:Manifestationsofacompositionalsystemofperceptualsymbols.TheoriesofMemory,,29Barsalou,L.W.,&Sewell,D.R.(1984).Constructingrepresentationsofcategoriesfromerentpointsofview.Bedny,M.,Pascual-Leone,A.,&Saxe,R.R.(2009).GrowingupblinddoesnotchangetheneuralbasesofTheoryofMind.ProceedingsoftheNationalAcademyofSciences,(27),11312Berkeley,G.(1948).Anessaytowardanewtheoryofvision,1709.InW.Dennis(Ed.).Centurypsychologyseries.Readingsinthehistoryofpsychology(pp.6980).EastNorwalk,CT,US:Appleton-Century-CroftsBerkeley,G.(1732).AnEssaytowardsaNewTheoryofVision.(Basedonthefourthedition,London,1732).EditedbyDavidR.Wilkins,Dublin,December2002.Bimler,D.,&Kirkland,J.(2007).Constructingpersonalitymaps,mappingpersonalityconstructs:Multidimensionalscalingrecoversthebigvefactorsfrominternalandexternalstructure.TheSpanishJournalofPsychology,10(1),68Bloom,P.(2002).Mindreading,communicationandthelearningofnamesforthings.MindandLanguage,17,37Brysbaert,M.,&New,B.(2009).MovingbeyondKueraandFrancis:AcriticalevaluationofcurrentwordfrequencynormsandtheintroductionofanewandimprovedwordfrequencymeasureforAmericanEnglish.BehaviorResearchMethods,41(4),977Chan,A.S.,Butters,N.,Paulsen,J.S.,Salmon,D.P.,Swenson,M.R.,&Maloney,L.T.(1993).AnassessmentofthesemanticnetworkinpatientswithAlzheimer'sdisease.JournalofCognitiveNeuroscience,5(2),254Clark,J.M.(1987).Understandingpicturesandwords:CommentonPotter,Kroll,Yachzel,Carpenter,andSherman(1986).JournalofExperimentalPsychology:General,,307Connolly,A.C.,Gleitman,L.R.,&Thompson-Schill,S.L.(2007).Eectofcongenitalblindnessonthesemanticrepresentationofsomeeverydayconcepts.ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,104,8241Cutsforth,T.D.(1932).Theunrealityofwordstotheblind.TeachersForum,4,86Cutsforth,T.D.(1951).Theblindinschoolandsociety.NewYork:AmericanFoundationfortheBlindLeeuw,J.,&Mair,P.(2011).Multidimensionalscalingusingmajorization:SMACOFinR.DeLeeuw,J.,&Heiser,W.J.(1977).Convergenceofcorrectionmatrixalgorithmsformultidimensionalscaling.GeometricRepresentationsofRelationalData,Faber,P.B.,&Usón,R.M.(1999).ConstructingalexiconofEnglishverbs(Vol.23).WalterdeGruyter.Frank,M.C.,&Goodman,N.D.(2014).Inferringwordmeaningsbyassumingthatspeakersareinformative.CognitivePsychology,75,80Gleitman,L.(1990).Thestructuralsourcesofverbmeanings.LanguageAcquisition:AJournalofDevelopmentalLinguistics,1(1),3Goldstone,R.L.(1994).Theroleofsimilarityincategorization:Providingagroundwork.Cognition,52(2),125Goodman,N.(1972).Sevenstricturesonsimilarity.InN.Goodman(Ed.).Problemsand.NewYork:Bobbs-MerrillGougoux,F.,Lepore,F.,Lassonde,M.,Voss,P.,Zatorre,R.J.,&Belin,P.(2004).Neuropsychology:Pitchdiscriminationintheearlyblind.Nature,430(6997),309Grant,A.C.,Thiagarajah,M.C.,&Sathian,K.(2000).Tactileperceptioninblindbraillereaders:ApsychophysicalstudyofacuityandhyperacuityusinggratingsanddotPerception&Psychophysics,62,01HarrellJr.,F.E.,&Dupont,C.(2014).Hmisc:HarrellMiscellaneous,2011.URLCRAN.R-project.org/package=H

11 misc.Rpackageversion,3-9.Hobbes,T.(1641/
misc.Rpackageversion,3-9.Hobbes,T.(1641/1984).Meditationsonrstphilosophy,objectionswithreplies.InRobertStootbo,&DouglasMurdoch(Eds.).ThePhilosophicalWritingsofDescartesCambridge,UK:CambridgeUniversityPressHume,D.(1739/1978).Atreatiseofhumannature.Oxford,UK:OxfordUniversityPressHutchinson,J.,&Lockhead,G.R.(1977).Similarityasdistance:Astructuralprincipleforsemanticmemory.JournalofExperimentalPsychology:HumanLearningandMemory,(6),660Koster-Hale,J.,Bedny,M.,&Saxe,R.(2014).Thinkingaboutseeing:PerceptualsourcesofknowledgeareencodedinthetheoryofmindbrainregionsofsightedandblindCognition,133(1),65Landau,B.,&Gleitman,L.(1985).Languageandexperience:Evidencefromtheblindchild.Cambridge,MA:HarvardUniversityPressLandauer,T.K.,&Dumais,S.T.(1997).AsolutiontoPlato'sproblem:Thelatentsemanticanalysistheoryofacquisition,induction,andrepresentationofknowledge.PsychologicalReview,104(2),211Lenci,Baroni,M.,Cazzolli,G.,&Marotta,G.(2013).BLIND:Asetofsemanticfeaturenormsfromthecongenitallyblind.BehavioralResearchMethods,45,1218Levin,B.(1993).Englishverbclassesandalternations:Apreliminaryinvestigation.ofChicagopressJ.,&Nidditch,P.H.(2011).Anessayconcerninghumanunderstanding.ClarendonPressLopez,A.,Atran,S.,Coley,J.D.,Medin,D.L.,&Smith,E.E.(1997).Thetreeoflife:Universalandculturalfeaturesoffolkbiologicaltaxonomiesandinductions.Psychology,32(3),251Mair,P.,deLeeuw,J.,&Groenen,P.J.(2015).MultidimensionalscalinginR:smacof.https://cran.r-project.org/web/packages/smacof/vignettes/smacof.pdfMajid,A.,Roberts,S.G.,Cilissen,L.,Emmorey,K.,Nicodemus,B.,OGrady,L.,...Shayan,S.(2018).Dierentialcodingofperceptionintheworldslanguages.ProceedingsoftheNationalAcademyofSciences,115(45),11369Markman,E.M.,&Wachtel,G.F.(1988).Childrensuseofmutualexclusivitytoconstrainthemeaningsofwords.CognitivePsychology,20,121Marmor,G.S.(1978).Ageatonsetofblindnessandthedevelopmentofthesemanticsofcolornames.JournalofExperimentalChildPsychology,25(2),267Medin,D.L.,Goldstone,R.L.,&Gentner,D.(1993).Respectsforsimilarity.Review,100(2),254Medin,D.L.,&Shoben,E.J.(1988).Contextandstructureinconceptualcombination.CognitivePsychology,20(2),158Murphy,G.(2004).Thebigbookofconcepts.MITpressMurtagh,Fionn,&Legendre(2014).Ward'shierarchicalagglomerativeclusteringmethod:whichalgorithmsimplementWard'scriterion?JournalofClassication,31Ouyang,L.,Boroditsky,L.,&Frank,M.C.(2017).Semanticcoherencefacilitatesdis-tributionallearning.CognitiveScience,41,855Rips,L.J.,Shoben,E.J.,&Smith,E.E.(1973).Semanticdistanceandthevericationofsemanticrelations.JournalofVerbalLearningandVerbalBehavior,12(1),1Roberts,J.S.,&Wedell,D.H.(1994).Contexteectsonsimilarityjudgmentsofmulti-dimensionalstimuli:Inferringthestructureoftheemotionspace.JournalofExperimentalSocialPsychology,30(1),1Rosel,J.,Caballer,A.,Jara,P.,&Oliver,J.C.(2005).Verbalisminthenarrativelanguageofchildrenwhoareblindandsighted.ofVisualImpairment&Blindness,99Ross,B.H.,&Murphy,G.L.(1999).Foodforthought:Cross-classicationandcategoryorganizationinacomplexreal-worlddomain.CognitivePsychology.38,495Rumelhart,D.E.,&Abrahamson,A.A.(1973).Amodelforanalogicalreasoning.CognitivePsychology,5(1),1Saysani,A.,Corballis,M.C.,&Corballis,P.M.(2018).Colorenvisioned:Conceptsofcolorintheblindandsighted.VisualCognition,26(5),382Shepard,R.N.(1987).Towardauniversallawofgeneralizationforpsychologicalscience.Science,237(4820),1317Shepard,R.N.,&Chipman,S.(1970).Second-orderisomorphismofinternalre-presentations:Shapesofstates.CognitivePsychology,1(1),1Shepard,R.N.,&Cooper,L.A.(1992).Representationofcolorsintheblind,color-blind,andnormallysighted.PsychologicalScience,3(2),97Shepard,R.N.,Kilpatric,D.W.,&Cunningham,J.P.(1975).Theinternalrepresentationofnumbers.CognitivePsychology,7(1),82Suzuki,R.,&Shimodaira,H.(2006).Pvclust:AnRpackageforassessingtheuncertaintyinhierarchicalclustering.Bioinformatics,22(12),1540Tenenbaum,J.B.,&Griths,T.L.(2001).Generalization,similarity,andBayesianin-BehavioralandBrainSciences,24(4),629Tomasello,M.,&Barton,M.E.(1994).Learningwordsinnonostensivecontexts.DevelopmentalPsychology,30(5),639VanBoven,R.W.,Hamilton,R.H.,Kauman,T.,Keenan,J.P.,&Pascual-Leone,A.(2000).TactilespatialresolutioninblindBraillereaders.Neurology,54Winter,B.,Perlman,M.,&Majid,A.(2018).Visiondominatesinperceptuallanguage:Englishsensoryvocabularyisoptimizedforusage.Cognition,179,213Zwiers,M.P.,VanOpstal,A.J.,&Cruysberg,J.R.M.(2001).Aspatialhearingdecitinearly-blindhumans.JournalofNeuroscience,21(9),RC142RC142-RC142M.Bedny,etal.