You are on page 1of 33

8/20/2016

LanguageAcquisition

LanguageAcquisition
StevenPinker
MassachusettsInstituteofTechnology
ChaptertoappearinL.R.Gleitman,M.Liberman,andD.N.Osherson(Eds.),
AnInvitationtoCognitiveScience,2ndEd.Volume1:Language.Cambridge,MA:MITPress.
NONFINALVERSION:PLEASEDONOTEQUOTE.
PreparationofthechapterwassupportedbyNIHgrantHD18381andNSFgrantBNS9109766,andbythe
McDonnellPewCenterforCognitiveNeuroscienceatMIT.

1Introduction
Languageacquisitionisoneofthecentraltopicsincognitivescience.Everytheoryofcognitionhastriedto
explainitprobablynoothertopichasarousedsuchcontroversy.Possessingalanguageisthequintessentially
humantrait:allnormalhumansspeak,nononhumananimaldoes.Languageisthemainvehiclebywhichwe
knowaboutotherpeople'sthoughts,andthetwomustbeintimatelyrelated.Everytimewespeakweare
revealingsomethingaboutlanguage,sothefactsoflanguagestructureareeasytocomebythesedatahintat
asystemofextraordinarycomplexity.Nonetheless,learningafirstlanguageissomethingeverychilddoes
successfully,inamatterofafewyearsandwithouttheneedforformallessons.Withlanguagesoclosetothe
coreofwhatitmeanstobehuman,itisnotsurprisingthatchildren'sacquisitionoflanguagehasreceivedso
muchattention.Anyonewithstrongviewsaboutthehumanmindwouldliketoshowthatchildren'sfirstfew
stepsarestepsintherightdirection.
Languageacquisitionisnotonlyinherentlyinterestingstudyingitisonewaytolookforconcreteanswersto
questionsthatpermeatecognitivescience:
Modularity.Dochildrenlearnlanguageusinga"mentalorgan,"someofwhoseprinciplesoforganizationare
notsharedwithothercognitivesystemssuchasperception,motorcontrol,andreasoning(Chomsky,1975,
1991Fodor,1983)?Orislanguageacquisitionjustanotherproblemtobesolvedbygeneralintelligence,in
thiscase,theproblemofhowtocommunicatewithotherhumansovertheauditorychannel(Putnam,1971
Bates,1989)?
HumanUniqueness.Arelatedquestioniswhetherlanguageisuniquetohumans.Atfirstglancetheanswer
seemsobvious.Otheranimalscommunicationwithafixedrepertoireofsymbols,orwithanaloguevariation
likethemercuryinathermometer.Butnoneappearstohavethecombinatorialrulesystemofhuman
language,inwhichsymbolsarepermutedintoanunlimitedsetofcombinations,eachwithadeterminate
meaning.Ontheotherhand,manyotherclaimsabouthumanuniqueness,suchasthathumansweretheonly
animalstousetoolsortofabricatethem,haveturnedouttobefalse.Someresearchershavethoughtthatapes
havethecapacityforlanguagebutneverprofitedfromahumanlikeculturalmilieuinwhichlanguagewas
taught,andtheyhavethustriedtoteachapeslanguagelikesystems.Whethertheyhavesucceeded,and
whetherhumanchildrenarereally"taught"languagethemselves,arequestionswewillsooncometo.
LanguageandThought.Islanguagesimplygraftedontopofcognitionasawayofstickingcommunicable
labelsontothoughts(Fodor,1975Piaget,1926)?Ordoeslearningalanguagesomehowmeanlearningto
thinkinthatlanguage?Afamoushypothesis,outlinedbyBenjaminWhorf(1956),assertsthatthecategories
andrelationsthatweusetounderstandtheworldcomefromourparticularlanguage,sothatspeakersof
differentlanguagesconceptualizetheworldindifferentways.Languageacquisition,then,wouldbelearning
tothink,notjustlearningtotalk.
Thisisanintriguinghypothesis,butvirtuallyallmoderncognitivescientistsbelieveitisfalse(seePinker,
1994a).Babiescanthinkbeforetheycantalk(ChapterX).Cognitivepsychologyhasshownthatpeople
thinknotjustinwordsbutinimages(seeChapterX)andabstractlogicalpropositions(seethechapterby
Larson).Andlinguisticshasshownthathumanlanguagesaretooambiguousandschematictouseasa
mediumofinternalcomputation:whenpeoplethinkabout"spring,"surelytheyarenotconfusedasto
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

1/33

8/20/2016

LanguageAcquisition

whethertheyarethinkingaboutaseasonorsomethingthatgoes"boing"andifonewordcancorrespondto
twothoughts,thoughtscan'tbewords.
Butlanguageacquisitionhasauniquecontributiontomaketothisissue.Asweshallsee,itisvirtually
impossibletoshowhowchildrencouldlearnalanguageunlessyouassumetheyhaveaconsiderableamount
ofnonlinguisticcognitivemachineryinplacebeforetheystart.
LearningandInnateness.Allhumanstalkbutnohousepetsorhouseplantsdo,nomatterhowpampered,so
hereditymustbeinvolvedinlanguage.ButachildgrowingupinJapanspeaksJapanesewhereasthesame
childbroughtupinCaliforniawouldspeakEnglish,sotheenvironmentisalsocrucial.Thusthereisno
questionaboutwhetherheredityorenvironmentisinvolvedinlanguage,orevenwhetheroneortheotheris
"moreimportant."Instead,languageacquisitionmightbeourbesthopeoffindingouthowheredityand
environmentinteract.Weknowthatadultlanguageisintricatelycomplex,andweknowthatchildrenbecome
adults.Thereforesomethinginthechild'smindmustbecapableofattainingthatcomplexity.Anytheorythat
positstoolittleinnatestructure,sothatitshypotheticalchildendsupspeakingsomethinglessthanareal
language,mustbefalse.Thesameistrueforanytheorythatpositstoomuchinnatestructure,sothatthe
hypotheticalchildcanacquireEnglishbutnot,say,BantuorVietnamese.
Andnotonlydoweknowabouttheoutputoflanguageacquisition,weknowafairamountabouttheinputto
it,namely,parent'sspeechtotheirchildren.Soeveniflanguageacquisition,likeallcognitiveprocesses,is
essentiallya"blackbox,"weknowenoughaboutitsinputandoutputtobeabletomakepreciseguesses
aboutitscontents.
Thescientificstudyoflanguageacquisitionbeganaroundthesametimeasthebirthofcognitivescience,in
thelate1950's.Wecanseenowwhythatisnotacoincidence.ThehistoricalcatalystwasNoamChomsky's
reviewofSkinner'sVerbalBehavior(Chomsky,1959).Atthattime,AngloAmericannaturalscience,social
science,andphilosophyhadcometoavirtualconsensusabouttheanswerstothequestionslistedabove.The
mindconsistedofsensorimotorabilitiesplusafewsimplelawsoflearninggoverninggradualchangesinan
organism'sbehavioralrepertoire.Thereforelanguagemustbelearned,itcannotbeamodule,andthinking
mustbeaformofverbalbehavior,sinceverbalbehavioristheprimemanifestationof"thought"thatcanbe
observedexternally.Chomskyarguedthatlanguageacquisitionfalsifiedthesebeliefsinasinglestroke:
childrenlearnlanguagesthataregovernedbyhighlysubtleandabstractprinciples,andtheydosowithout
explicitinstructionoranyotherenvironmentalcluestothenatureofsuchprinciples.Hencelanguage
acquisitiondependsonaninnate,speciesspecificmodulethatisdistinctfromgeneralintelligence.Muchof
thedebateinlanguageacquisitionhasattemptedtotestthisoncerevolutionary,andstillcontroversial,
collectionofideas.Theimplicationsextendtotherestofhumancognition.

2TheBiologyofLanguageAcquisition
Humanlanguageismadepossiblebyspecialadaptationsofthehumanmindandbodythatoccurredinthe
courseofhumanevolution,andwhichareputtousebychildreninacquiringtheirmothertongue.

2.1EvolutionofLanguage
Mostobviously,theshapeofthehumanvocaltractseemstohavebeenmodifiedinevolutionforthe
demandsofspeech.Ourlarynxesarelowinourthroats,andourvocaltractshaveasharprightanglebend
thatcreatestwoindependentlymodifiableresonantcavities(themouthandthepharynxorthroat)that
definesalargetwodimensionalrangeofvowelsounds(seethechapterbyLiberman).Butitcomesata
sacrificeofefficiencyforbreathing,swallowing,andchewing(Lieberman,1984).Beforetheinventionofthe
Heimlichmaneuver,chokingonfoodwasacommoncauseofaccidentaldeathinhumans,causing6,000
deathsayearintheUnitedStates.Theevolutionaryselectiveadvantagesforlanguagemusthavebeenvery
largetooutweighsuchadisadvantage.
ItistemptingtothinkthatiflanguageevolvedbygradualDarwiniannaturalselection,wemustbeableto
findsomeprecursorofitinourclosestrelatives,thechimpanzees.Inseveralfamousandcontroversial
demonstrations,chimpanzeeshavebeentaughtsomehandsignsbasedonAmericanSignLanguage,to
manipulatecoloredswitchesortokens,andtounderstandsomespokencommands(Gardner&Gardner,
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

2/33

8/20/2016

LanguageAcquisition

1969Premack&Premack,1983SavageRumbaugh,1991).Whetheronewantstocalltheirabilities
"language"isnotreallyascientificquestion,butamatterofdefinition:howfarwearewillingtostretchthe
meaningoftheword"language".
Thescientificquestioniswhetherthechimps'abilitiesarehomologoustohumanlanguagethatis,whether
thetwosystemsshowthesamebasicorganizationowingtodescentfromasinglesystemintheircommon
ancestor.Forexample,biologistsdon'tdebatewhetherthewinglikestructuresofglidingrodentsmaybe
called"genuinewings"orsomethingelse(aboringquestionofdefinitions).It'sclearthatthesestructuresare
nothomologoustothewingsofbats,becausetheyhaveafundamentallydifferentanatomicalplan,reflecting
adifferentevolutionaryhistory.Bats'wingsaremodificationsofthehandsofthecommonmammalian
ancestorflyingsquirrels'wingsaremodificationsofitsribcage.Thetwostructuresaremerelyanalogous:
similarinfunction.
Thoughartificialchimpsignalingsystemshavesomeanalogiestohumanlanguage(e.g.,usein
communication,combinationsofmorebasicsignals),itseemsunlikelythattheyarehomologous.
Chimpanzeesrequiremassiveregimentedteachingsequencescontrivedbyhumanstoacquirequite
rudimentaryabilities,mostlylimitedtoasmallnumberofsigns,strungtogetherinrepetitive,quasirandom
sequences,usedwiththeintentofrequestingfoodortickling(Terrace,Petitto,Sanders,&Bever,1979
Seidenberg&Petitto,1979,1987Seidenberg,1986Wallman,1992Pinker,1994a).Thiscontrastssharply
withhumanchildren,whopickupthousandsofwordsspontaneously,combinetheminstructuredsequences
whereeverywordhasadeterminaterole,respectthewordorderoftheadultlanguage,andusesentencesfor
avarietyofpurposessuchascommentingoninterestingobjects.
Thislackofhomologydoesnot,bytheway,castdoubtonagradualisticDarwinianaccountoflanguage
evolution.Humansdidnotevolvedirectlyfromchimpanzees.Bothderivedfromcommonancestor,probably
around67millionyearsago.Thisleavesabout300,000generationsinwhichlanguagecouldhaveevolved
graduallyinthelineageleadingtohumans,afteritsplitofffromthelineageleadingtochimpanzees.
Presumablylanguageevolvedinthehumanlineagefortworeasons:ourancestorsdevelopedtechnologyand
knowledgeofthelocalenvironmentintheirlifetimes,andwereinvolvedinextensivereciprocalcooperation.
Thisallowedthemtobenefitbysharinghardwonknowledgewiththeirkinandexchangingitwiththeir
neighbors(Pinker&Bloom,1990).

2.2DissociationsbetweenLanguageandGeneralIntelligence
Humansevolvedbraincircuitry,mostlyinthelefthemispheresurroundingthesylvianfissure,thatappearsto
bedesignedforlanguage,thoughhowexactlytheirinternalwiringgivesrisetorulesoflanguageisunknown
(seetheChapterbyZurif).Thebrainmechanismsunderlyinglanguagearenotjustthoseallowingustobe
smartingeneral.Strokesoftenleaveadultswithcatastrophiclossesinlanguage(seetheChapterbyZurif,
andPinker,1994a),thoughnotnecessarilyimpairedinotheraspectsofintelligence,suchasthosemeasured
onthenonverbalpartsofIQtests.Similarly,thereisaninheritedsetofsyndromescalledSpecificLanguage
Impairment(GopnikandCrago,1993Tallal,Ross,&Curtiss,1989)whichismarkedbydelayedonsetof
language,difficultiesinarticulationinchildhood,andlastingdifficultiesinunderstanding,producing,and
judginggrammaticalsentences.Bydefinition,SpecificallyLanguageImpairedpeopleshowsuchdeficits
despitetheabsenceofcognitiveproblemslikeretardation,sensoryproblemslikehearingloss,orsocial
problemslikeautism.
Moreinterestingly,therearesyndromesshowingtheoppositedissociation,whereintactlanguagecoexists
withsevereretardation.Thesecasesshowthatlanguagedevelopmentdoesnotdependonfullyfunctioning
generalintelligence.OneexamplecomesfromchildrenwithSpinaBifida,amalformationofthevertebrae
thatleavesthespinalcordunprotected,oftenresultinginhydrocephalus,anincreaseinpressureinthe
cerebrospinalfluidfillingtheventricles(largecavities)ofthebrain,distendingthebrainfromwithin.
Hydrocephalicchildrenoccasionallyendupsignificantlyretardedbutcancarryonlong,articulate,andfully
grammaticalconversations,inwhichtheyearnestlyrecountvivideventsthatare,infact,productsoftheir
imaginations(Cromer,1992Curtiss,1989Pinker,1994a).AnotherexampleisWilliamsSyndrome,an
inheritedconditioninvolvingphysicalabnormalities,significantretardation(theaverageIQisabout50),
incompetenceatsimpleeverydaytasks(tyingshoelaces,findingone'sway,addingtwonumbers,and
retrievingitemsfromacupboard),socialwarmthandgregariousness,andfluent,articulatelanguageabilities
(Bellugi,etal.,1990).
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

3/33

8/20/2016

LanguageAcquisition

2.3MaturationoftheLanguageSystem
AsthechapterbyNewportandGleitmansuggests,thematurationoflanguagecircuitsduringachild'searly
yearsmaybeadrivingforceunderlyingthecourseoflanguageacquisition(Pinker,1994,Chapter9Bates,
Thal,&Janowsky,1992Locke,1992Huttenlocher,1990).Beforebirth,virtuallyalltheneurons(nerve
cells)areformed,andtheymigrateintotheirproperlocationsinthebrain.Butheadsize,brainweight,and
thicknessofthecerebralcortex(graymatter),wherethesynapses(junctions)subservingmentalcomputation
takeplace,continuetoincreaserapidlyintheyearafterbirth.Longdistanceconnections(whitematter)are
notcompleteuntilninemonths,andtheycontinuetogrowtheirspeedinducingmyelininsulationthroughout
childhood.Synapsescontinuetodevelop,peakinginnumberbetweenninemonthsandtwoyears(depending
onthebrainregion),atwhichpointthechildhas50%moresynapsesthantheadult.Metabolicactivityinthe
brainreachesadultlevelsbyninetotenmonths,andsoonexceedsit,peakingaroundtheageoffour.In
addition,hugenumbersofneuronsdieinutero,andthedyingcontinuesduringthefirsttwoyearsbefore
levelingoffatageseven.Synapseswitherfromtheageoftwothroughtherestofchildhoodandinto
adolescence,whenthebrain'smetabolicratefallsbacktoadultlevels.Perhapslinguisticmilestoneslike
babbling,firstwords,andgrammarrequireminimumlevelsofbrainsize,longdistanceconnections,orextra
synapses,particularlyinthelanguagecentersofthebrain.
Similarly,onecanconjecturethatthesechangesareresponsibleforthedeclineintheabilitytolearna
languageoverthelifespan.Thelanguagelearningcircuitryofthebrainismoreplasticinchildhoodchildren
learnorrecoverlanguagewhenthelefthemisphereofthebrainisdamagedorevensurgicallyremoved
(thoughnotquiteatnormallevels),butcomparabledamageinanadultusuallyleadstopermanentaphasia
(Curtiss,1989Lenneberg,1967).Mostadultsnevermasteraforeignlanguage,especiallythephonology,
givingrisetowhatwecalla"foreignaccent."Theirdevelopmentoftenfossilizesintopermanenterror
patternsthatnoteachingorcorrectioncanundo.Therearegreatindividualdifferences,whichdependon
effort,attitudes,amountofexposure,qualityofteaching,andplaintalent.
Manyexplanationshavebeenadvancedforchildren'ssuperiority:theycanexploitthespecialwaysthattheir
motherstalkthem,theymakeerrorsunselfconsciously,theyaremoremotivatedtocommunicate,theylike
toconform,theyarenotxenophobicorsetintheirways,andtheyhavenofirstlanguagetointerfere.But
someoftheseaccountsareunlikely,basedonwhatwelearnabouthowlanguageacquisitionworkslaterin
thischapter.Forexample,childrencanlearnalanguagewithoutthespecialindulgentspeechfromtheir
motherstheymakefewerrorsandtheygetnofeedbackfortheerrorstheydomake.Anditcan'tbean
acrosstheboarddeclineinlearning.Thereisnoevidence,forexample,thatlearningwords(asopposedto
phonologyorgrammar)declinesinadulthood.
ThechapterbyNewportandGleitmanshowshowsheerageseemstoplayanimportantrole.Successful
acquisitionoflanguagetypicallyhappensby4(asweshallseeinthenextsection),isguaranteedforchildren
uptotheageofsix,issteadilycompromisedfromthenuntilshortlyafterpuberty,andisrarethereafter.
Maturationalchangesinthebrain,suchasthedeclineinmetabolicrateandnumberofneuronsduringthe
earlyschoolageyears,andthebottomingoutofthenumberofsynapsesandmetabolicratearoundpuberty,
areplausiblecauses.Thus,theremaybeaneurologicallydetermined"criticalperiod"forsuccessful
languageacquisition,analogoustothecriticalperiodsdocumentedinvisualdevelopmentinmammalsandin
theacquisitionofsongsbysomebirds.

3TheCourseofLanguageAcquisition
Althoughscholarshavekeptdiariesoftheirchildren'sspeechforoveracentury(CharlesDarwinwasoneof
thefirst),itwasonlyafterportabletaperecordersbecameavailableinthelate1950'sthatchildren's
spontaneousspeechbegantobeanalyzedsystematicallywithindevelopmentalpsychology.Thesenaturalistic
studiesofchildren'sspontaneousspeechhavebecomeevenmoreaccessiblenowthattheycanbeputinto
computerfilesandcanbedisseminatedandanalyzedautomatically(MacWhinney&Snow,1985,1990
MacWhinney,1991).Theyarecomplementedbyexperimentalmethods.Inproductiontasks,childrenutter
sentencestodescribepicturesorscenes,inresponsetoquestions,ortoimitatetargetsentences.In
comprehensiontasks,theylistentosentencesandthenpointtopicturesoractouteventswithtoys.In
judgementtasks,theyindicatewhetherorwhichsentencesprovidedbyanexperimentersound"silly"to
them.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

4/33

8/20/2016

LanguageAcquisition

AsthechapterbyWerkershows,languageacquisitionbeginsveryearlyinthehumanlifespan,andbegins,
logicallyenough,withtheacquisitionofalanguage'ssoundpatterns.Themainlinguisticaccomplishments
duringthefirstyearoflifearecontrolofthespeechmusculatureandsensitivitytothephoneticdistinctions
usedintheparents'language.Interestingly,babiesachievethesefeatsbeforetheyproduceorunderstand
words,sotheirlearningcannotdependoncorrelatingsoundwithmeaning.Thatis,theycannotbelistening
forthedifferenceinsoundbetweenawordtheythinkmeansbitandawordtheythinkmeansbeet,because
theyhavelearnedneitherword.Theymustbesortingthesoundsdirectly,somehowtuningtheirspeech
analysismoduletodeliverthephonemesusedintheirlanguage(Kuhl,etal.,1992).Themodulecanthen
serveasthefrontendofthesystemthatlearnswordsandgrammar.
Shortlybeforetheirfirstbirthday,babiesbegintounderstandwords,andaroundthatbirthday,theystartto
producethem(seeClark,1993Ingram,1989).Wordsareusuallyproducedinisolationthisonewordstage
canlastfromtwomonthstoayear.Children'sfirstwordsaresimilarallovertheplanet.Abouthalfthewords
areforobjects:food(juice,cookie,bodyparts(eye,nose),clothing(diaper,sock),vehicles(car,boat),toys
(doll,block),householditems(bottle,light,animals(dog,kitty),andpeople(dada,baby).Therearewords
foractions,motions,androutines,like(up,off,open,peekaboo,eat,andgo,andmodifiers,likehot,allgone,
more,dirty,andcold.Finally,thereareroutinesusedinsocialinteraction,likeyes,no,want,byebye,andhi
afewofwhich,likelookatthatandwhatisthat,arewordsinthesenseofmemorizedchunks,thoughthey
arenotsinglewordsfortheadult.Childrendifferinhowmuchtheynameobjectsorengageinsocial
interactionusingmemorizedroutines,thoughallchildrendoboth.
Around18months,languagechangesintwoways.Vocabularygrowthincreasesthechildbeginstolearn
wordsatarateofoneeverytwowakinghours,andwillkeeplearningthatrateorfasterthroughadolescence
(Clark,1993Pinker,1994).Andprimitivesyntaxbegins,withtwowordstringslikethefollowing:
Alldry.Allmessy.Allwet.
Isit.Ishut.Nobed.
Nopee.Seebaby.Seepretty.
Morecereal.Morehot.HiCalico.
Otherpocket.Bootoff.Sirenby.
Mailcome.Airplaneallgone.Bybebyecar.
Ourcar.Papaaway.Drypants.

Ourcar.Papaaway.Drypants.Children'stwowordcombinationsarehighlysimilaracrosscultures.
Everywhere,childrenannouncewhenobjectsappear,disappear,andmoveabout,pointouttheirproperties
andowners,commentonpeopledoingthingsandseeingthings,rejectandrequestobjectsandactivities,and
askaboutwho,what,andwhere.Thesesequencesalreadyreflectthelanguagebeingacquired:in95%of
them,thewordsareproperlyordered(Braine,1976Brown,1973Pinker,1984Ingram,1989).
Evenbeforetheyputwordstogether,babiescancomprehendasentenceusingitssyntax.Forexample,inone
experiment,babieswhospokeonlyinsinglewordswereseatedinfrontoftwotelevisionscreens,eachof
whichfeaturedapairofadultsdressedupasCookieMonsterandBigBirdfromSesameStreet.Onescreen
showedCookieMonsterticklingBigBirdtheothershowedBigBirdticklingCookieMonster.Avoiceover
said,"OHLOOK!!!BIGBIRDISTICKLINGCOOKIEMONSTER!!FINDBIGBIRDTICKLING
COOKIEMONSTER!!"(Orviceversa.)Thechildrenmusthaveunderstoodthemeaningoftheorderingof
subject,verb,andobject,becausetheylookedmoreatthescreenthatdepictedthesentenceinthevoiceover
(HirshPasek&Golinkoff,1991).
Children'soutputseemstomeetupwithabottleneckattheoutputend(Brown,1973Bloom,1970Pinker,
1984).Theirtwoandthreewordutteranceslooklikesamplesdrawnfromlongerpotentialsentences
expressingacompleteandmorecomplicatedidea.RogerBrown,oneofthefoundersofthemodernstudyof
languagedevelopment,notedthatalthoughthethreechildrenhestudiedintensivelyneverproduceda
sentenceascomplicatedasMothergaveJohnlunchinthekitchen,theydidproducestringscontainingallof
itscomponents,andinthecorrectorder:(Brown,1973,p.205):
AgentActionRecipientObjectLocation
(MothergaveJohnlunchinthekitchen.)
Mommyfix.
Mommypumpkin.
Babytable.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

5/33

8/20/2016

LanguageAcquisition

Givedoggie.
Putlight.
Putfloor.
Iridehorsie.
Tractorgofloor.
Givedoggiepaper.
Puttruckwindow.
Adamputitbox.

Betweenthelatetwo'sandmidthree's,children'slanguagebloomsintofluentgrammaticalconversationso
rapidlythatitoverwhelmstheresearcherswhostudyit,andnoonehasworkedouttheexactsequence.
Sentencelengthincreasessteadily,andbecausegrammarisacombinatorialsystem,thenumberofsyntactic
typesincreasesexponentially,doublingeverymonth,reachingthethousandsbeforethethirdbirthday
(Ingram,1989,p.235Brown,1973Limber,1973Pinker,1984).Forexample,herearesnapshotsofthe
developmentofoneofBrown'slongitudinalsubjects,Adam,intheyearfollowinghisfirstword
combinationsattheageof2yearsand3months(Pinker,1994a):
23:Playcheckers.Bigdrum.Igothorn.
24:Seemarchingbeargo?Screwpartmachine.
25:Nowputbootson.Wherewrenchgo?Whatthatpaperclipdoing?
26:Writeapieceapaper.Whatthateggdoing?No,Idon'twanttositseat.
27:Wherepieceapapergo?Droppedarubberband.Rintintindon'tfly,Mommy.
28:Letmegetdownwiththebootson.Howtigerbesohealthyand
flylikekite?Joshuathrowlikeapenguin.
29:WhereMommykeepherpocketbook?Showyousomethingfunny.
210:LookatthattrainUrsulabrought.Youdon'thavepaper.Doyouwantlittlebit,Cromer?
211:Dowantsomepieonyourface?Whyyoumixingbabychocolate?Isaidwhynotyou
comingin?Wegoingturnlightonsoyoucan'tsee.
30:Igoingcomeinfourteenminutes.Igoingwearthattowedding.Thosearenotstrongmens.
Youdressmeuplikeababyelephant.
31:Iliketoplaywithsomethingelse.Youknowhowtoputitbacktogether.Igon'makeitlike
arockettoblastoffwith.Youwanttogivemesomecarrotsandsomebeans?Pressthebutton
andcatchit,sir.Whyyouputthepacifierinhismouth?
32:Soitcan'tbecleaned?Ibrokemyracingcar.Doyouknowthelightwentsoff?Whenit's
gotaflattireit'sneedagotothestation.I'mgoingtomailthissothelettercan'tcomeoff.I
wanttohavesomeespresso.CanIputmyheadinthemailboxsothemailmancanknowwhere
Iareandputmeinthemailbox?CanIkeepthescrewdriverjustlikeacarpenterkeepthe
screwdriver?
Normalchildrencandifferbyayearormoreintheirrateoflanguagedevelopment,thoughthestagesthey
passthrougharegenerallythesameregardlessofhowstretchedoutorcompressed.Adam'slanguage
development,forexample,wasrelativelyleisurelymanychildrenspeakincomplexsentencesbeforethey
turntwo.
Duringthegrammarexplosion,children'ssentencesaregettingnotonlylongerbutmorecomplex,withfuller
trees,becausethechildrencanembedoneconstituentinsideanother.Whereasbeforetheymighthavesaid
Givedoggiepaper(athreebranchVerbPhrase)andBigdoggie(atwobranchNounPhrase),theynowsay
Givebigdoggiepaper,withthetwobranchNPembeddedinsidethethreebranchVP.Theearliersentences
resembledtelegrams,missingunstressedfunctionwordslikeof,the,on,anddoes,aswellasinflectionslike
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

6/33

8/20/2016

LanguageAcquisition

ed,ing,ands.Bythe3's,childrenareusingthesefunctionwordsmoreoftenthantheyareomittingthem,
manyinmorethan90%ofthesentencesthatrequirethem.Afullrangeofsentencetypesflowerquestions
withwordslikewho,whatandwhere,relativeclauses,comparatives,negations,complements,conjunctions,
andpassives.Theseconstructionsappeartodisplaythemost,perhapsevenall,ofthegrammaticalmachinery
neededtoaccountforadultgrammar.
Thoughmanyoftheyoung3yearold'ssentencesareungrammaticalforonereasonoranother,itisbecause
therearemanythingsthatcangowronginanysinglesentence.Whenresearchersfocusonasingle
grammaticalruleandcounthowoftenachildobeysitandhowoftenheorsheversusfloutsit,theresultsare
veryimpressive:forjustabouteveryrulethathasbeenlookedat,threeyearoldsobeyitamajorityofthe
time(Stromswold,1990Pinker,1984,1989Crain,1992Marcus,etal.,1992).Aswehaveseen,children
rarelyscramblewordordersand,bytheageofthree,cometosupplymostinflectionsandfunctionwordsin
sentencesthatrequirethem.Thoughourearsperkupwhenwehearerrorslikemens,wents,Canyoubroke
those?,Whathecanridein?,That'safurniture,Buttonmetherest,andGoingtoseekitten,theerrorsoccur
inanywherefrom0.1%to8%oftheopportunitiesformakingthemmorethan90%ofthetime,thechildis
ontarget.Thenextchapterfollowsoneofthoseerrorsindetail.
Childrendonotseemtofavoranyparticularkindoflanguage(indeed,itwouldbepuzzlinghowanykindof
languagecouldsurviveifchildrendidnoteasilylearnit!).Theyswiftlyacquirefreewordorder,SOVand
VSOorders,richsystemsofcaseandagreement,stringsofagglutinatedsuffixes,ergativecasemarking,and
whateverelsetheirlanguagethrowsatthem,withnolagrelativetotheirEnglishspeakingcounterparts.
Evengrammaticalgender,whichmanyadultslearningasecondlanguagefindmystifying,presentsno
problem:childrenacquiringlanguagelikeFrench,German,andHebrewacquiregendermarkingquickly,
makefewerrors,andneverusetheassociationwithmalenessandfemalenessasafalsecriterion(Levy,
1983).Itissafetosaythatexceptforconstructionsthatarerare,predominantlyusedinwrittenlanguage,or
mentallytaxingeventoanadult(likeThehorsethattheelephanttickledkissedthepig),allpartsofall
languagesareacquiredbeforethechildturnsfour(Slobin,1985/1992).

4ExplainingLanguageAcquisition
Howdoweexplainchildren'scourseoflanguageacquisitionmostimportantly,theirinevitableandearly
mastery?Severalkindsofmechanismsareatwork.Aswesawinsection(),thebrainchangesafterbirth,and
thesematurationalchangesmaygoverntheonset,rate,andadultdeclineoflanguageacquisitioncapacity.
Generalchangesinthechild'sinformationprocessingabilities(attention,memory,shorttermbuffersfor
acousticinputandarticulatoryoutput)couldleavetheirmarkaswell.Inthenextchapter,Ishowhowa
memoryretrievallimitationchildrenarelessreliableatrecallingthatbrokeisthepasttenseofbreakcan
accountforaconspicuousanduniversalerrorpattern,overregularizationslikebreaked(seealsoMarcus,et
al.,1992).
Manyothersmalleffectshavebeendocumentedwherechangesininformationprocessingabilitiesaffect
languagedevelopment.Forexample,childrenselectivelypickupinformationattheendsofwords(Slobin,
1973),andatthebeginningsandendsofsentences(Newport,etal,1977),presumablybecausethesearethe
partsofstringsthatarebestretainedinshorttermmemory.Similarly,theprogressivelywideningbottleneck
forearlywordcombinationspresumablyreflectsageneralincreaseinmotorplanningcapacity.Conceptual
development(seeChapterX),too,mightaffectlanguagedevelopment:ifachildhasnotyetmastereda
difficultsemanticdistinction,suchasthecomplextemporalrelationsinvolvedinJohnwillhavegone,heor
shemaybeunabletomasterthesyntaxoftheconstructiondedicatedtoexpressingit.
Thecomplexityofagrammaticalformhasademonstrableroleindevelopment:simplerrulesandforms
appearinspeechbeforemorecomplexones,allotherthingsbeingequal.Forexample,thepluralmarkersin
English(e.g.cats),whichrequiresknowingonlywhetherthenumberofreferentsissingularorplural,isused
consistentlybeforethepresenttensemarkers(hewalks),whichrequiresknowingwhetherthesubjectis
singularorpluralandwhetheritisafirst,second,orthirdpersonandwhethertheeventisinthepresent
tense(Brown,1973).Similarly,complexformsaresometimesfirstusedinsimplerapproximations.Russian
containsonecasemarkerformasculinenominative(i.e.,asuffixonamasculinenounindicatingthatitisthe
subjectofthesentence),oneforfemininenominative,oneformasculineaccusative(usedtoindicatethata
nounisadirectobject),andoneforfeminineaccusative.Childrenoftenuseeachmarkerwiththecorrect
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

7/33

8/20/2016

LanguageAcquisition

case,neverusinganominativemarkerforaccusativenounsorviceversa,butdon'tproperlyusethe
masculineandfemininevariantswithmasculineandfemininenouns(Slobin,1985).
Buttheseglobaltrendsdonotexplainthemainevent:howchildrensucceed.Languageacquisitionisso
complexthatoneneedsapreciseframeworkforunderstandingwhatitinvolvesindeed,whatlearningin
generalinvolves.

4.1LearnabilityTheory
Whatislanguageacquisition,inprinciple?AbranchoftheoreticalcomputersciencecalledLearnability
Theoryattemptstoanswerthisquestion(Gold,1967Osherson,Stob,&Weinstein,1985Pinker,1979).
Learnabilitytheoryhasdefinedlearningasascenarioinvolvingfourparts(thetheoryembracesallformsof
learning,butIwilluselanguageastheexample):
1.Aclassoflanguages.Oneofthemisthe"target"language,tobeattainedbythelearner,butthe
learnerdoesnot,ofcourse,knowwhichitis.Inthecaseofchildren,theclassoflanguageswould
consistoftheexistingandpossiblehumanlanguagesthetargetlanguageistheonespokenintheir
community.
2.Anenvironment.Thisistheinformationintheworldthatthelearnerhastogoonintryingtoacquire
thelanguage.Inthecaseofchildren,itmightincludethesentencesparentsutter,thecontextinwhich
theyutterthem,feedbacktothechild(verbalornonverbal)inresponsetothechild'sownspeech,and
soon.Parentalutterancescanbearandomsampleofthelanguage,ortheymighthavesomespecial
properties:theymightbeorderedincertainways,sentencesmightberepeatedoronlyutteredonce,
andsoon.
3.Alearningstrategy.Thelearner,usinginformationintheenvironment,triesout"hypotheses"aboutthe
targetlanguage.Thelearningstrategyisthealgorithmthatcreatesthehypothesesanddetermines
whethertheyareconsistentwiththeinputinformationfromtheenvironment.Forchildren,itisthe
"grammarforming"mechanismintheirbrainstheir"languageacquisitiondevice."
4.Asuccesscriterion.Ifwewanttosaythat"learning"occurs,presumablyitisbecausethelearners'
hypothesesarenotrandom,butthatbysometimethehypothesesarerelatedinsomesystematic
waytothetargetlanguage.Learnersmayarriveatahypothesisidenticaltothetargetlanguageafter
somefixedperiodoftimetheymayarriveatanapproximationtoittheymaywaiveramongaset
ofhypothesesoneofwhichiscorrect.
Theoremsinlearnabilitytheoryshowhowassumptionsaboutanyofthethreecomponentsimposeslogical
constraintsonthefourth.Itisnothardtoshowwhylearningalanguage,onlogicalgroundsalone,issohard.
Likeall"inductionproblems"(uncertaingeneralizationsfrominstances),thereareaninfinitenumberof
hypothesesconsistentwithanyfinitesampleofenvironmentalinformation.Learnabilitytheoryshowswhich
inductionproblemsaresolvableandwhicharenot.
Akeyfactoristheroleofnegativeevidence,orinformationaboutwhichstringsofwordsarenotsentences
inthelanguagetobeacquired.Humanchildrenmightgetsuchinformationbybeingcorrectedeverytime
theyspeakungrammatically.Iftheyaren'tandasweshallsee,theyprobablyaren'ttheacquisition
problemisalltheharder.ConsiderFigure1,wherelanguagesaredepictedascirclescorrespondingtosetsof
wordstrings,andallthelogicalpossibilitiesforhowthechild'slanguagecoulddifferfromtheadultlanguage
aredepicted.Therearefourpossibilities.(a)Thechild'shypothesislanguage(H)isdisjointfromthe
languagetobeacquired(the"targetlanguage,"T).Thatwouldcorrespondtothestateofchildlearning
EnglishwhocannotsayasinglewellformedEnglishsentence.Forexample,thechildmightbeableonlyto
saythingslikewebreakedit,andwegoed,neverwebrokeitorwewent.(b)Thechild'shypothesisandthe
targetlanguageintersect.HerethechildwouldbeabletouttersomeEnglishsentences,likehewent.
However,heorshealsousesstringsofwordsthatarenotEnglish,suchaswebreakeditandsomesentences
ofEnglish,suchaswebrokeit,wouldstillbeoutsidetheirabilities.(c)Thechild'shypothesislanguageisa
subsetofthetargetlanguage.ThatwouldmeanthatthechildwouldhavemasteredsomeofEnglish,butnot
allofit,butthateverythingthechildhadmasteredwouldbepartofEnglish.Thechildmightnotbeableto
saywebrokeit,butheorshewouldbeabletosaysomegrammaticalsentences,suchaswewentnoerrors
suchasshebreakeditorwegoedwouldoccur.Thefinallogicalpossibilityis(d),whereThechild's
hypothesislanguageisasupersetofthetargetlanguage.Thatwouldoccur,forexample,ifthechildcould
saywebrokeit,wewent,webreakeditandwegoed.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

8/33

8/20/2016

LanguageAcquisition

Incases(ac),thechildcanrealizethatthehypothesisisincorrectbyhearingsentencesfromparental
"positiveevidence,"(indicatedbythe"+"symbol)thatareinthetargetlanguagebutnotthehypothesized
one:sentencessuchaswebrokeit.Thisisimpossibleincase(d)negativeevidence(suchascorrectionsof
thechild'sungrammaticalsentencesbyhisorherparents)wouldbeneeded.Inotherwords,withoutnegative
evidence,ifachildguessestoolargealanguage,theworldcannevertellhimhe'swrong.
Thishasseveralconsequences.Foronething,themostgenerallearningalgorithmonemightconceiveof
onethatiscapableofhypothesizinganygrammar,oranycomputerprogramcapableofgeneratinga
languageisintrouble.Withoutnegativeevidence(andeveninmanycaseswithit),thereisnogeneral
purpose,allpowerfullearningmachineamachinemustinsomesense"know"somethingaboutthe
constraintsinthedomaininwhichitislearning.
Moreconcretely,ifchildrendon'treceivenegativeevidence(seeSection)wehavealotofexplainingtodo,
becauseoverlylargehypothesesareveryeasyforthechildtomake.Forexample,childrenactuallydogo
throughstagesinwhichtheyusetwoormorepasttenseformsforagivenverb,suchasbrokeandbreaked
thiscaseisdiscussedindetailinmyotherchapterinthisvolume.Theyderivetransitiveverbsfrom
intransitivestoofreely:whereanadultmightsaybothTheicemeltedandImeltedtheice,childrenalsocan
sayThegirlgiggledandDon'tgiggleme!(Bowerman,1982bPinker,1989).Ineachcasetheyarein
situation(d)inFigure1,andunlesstheirparentsslipthemsomesignalineverycasethatletsthemknow
theyarenotspeakingproperly,itispuzzlingthattheyeventuallystop.Thatis,wewouldneedtoexplainhow
theygrowintoadultswhoaremorerestrictiveintheirspeechoranotherwayofputtingisthatit'spuzzling
thattheEnglishlanguagedoesn'tallowdon'tgigglemeandsheeatedgiventhatchildrenaretemptedtogrow
uptalkingthatway.Iftheworldisn'ttellingchildrentostop,somethingintheirbrainsis,andwehavetofind
outwhoorwhatiscausingthechange.
Let'snowexaminelanguageacquisitioninthehumanspeciesbybreakingitdownintothefourelementsthat
giveaprecisedefinitiontolearning:thetargetoflearning,theinput,thedegreeofsuccess,andthelearning
strategy.

5WhatisLearned
TounderstandhowXislearned,youfirsthavetounderstandwhatXis.Linguistictheoryisthusanessential
partofthestudyoflanguageacquisition(seetheChapterbyLasnik).Linguisticresearchtriesdothree
things.First,itmustcharacterizethefactsofEnglish,andalltheotherlanguageswhoseacquisitionweare
interestedinexplaining.Second,sincechildrenarenotpredisposedtolearnEnglishoranyotherlanguage,
linguisticshastoexaminethestructureofotherlanguages.Inparticular,linguistscharacterizewhichaspects
ofgrammarareuniversal,prevalent,rare,andnonexistentacrosslanguages.Contrarytoearlysuspicions,
languagesdonotvaryarbitrarilyandwithoutlimitthereisbynowalargecatalogueoflanguageuniversals,
propertiessharedexactly,orinasmallnumberofvariations,byalllanguages(seeComrie,1981Greenberg,
1978Shopen,1985).Thisobviouslybearsonwhatchildren'slanguageacquisitionmechanismsfindeasyor
hardtolearn.
Andonemustgobeyondamerelistofuniversals.Manyuniversalpropertiesoflanguagearenotspecificto
languagebutaresimplyreflectionsofuniversalsofhumanexperience.Alllanguageshavewordsfor"water"
and"foot"becauseallpeopleneedtorefertowaterandfeetnolanguagehasawordamillionsyllableslong
becausenopersonwouldhavetimetosayit.Butothersmightbespecifictotheinnatedesignoflanguage
itself.Forexample,ifalanguagehasbothderivationalsuffixes(whichcreatenewwordsfromoldones,like
ism)andinflectionalsuffixes(whichmodifyawordtofititsroleinthesentence,likeplurals),thenthe
derivationalsuffixesarealwaysclosertothewordstemthantheinflectionalones.Forexample,inEnglish
onecansayDarwinisms(derivationalismclosertothestemthaninflectionals)butnotDarwinsism.Itis
hardtothinkofareasonhowthislawwouldfitintoanyuniversallawofthoughtormemory:whywould
theconceptoftwoideologiesbasedononeDarwinshouldbethinkable,buttheconceptofoneideology
basedontwoDarwins(say,CharlesandErasmus)notbethinkable(unlessonereasonsinacircleand
declaresthatthemindmustfindismtobemorecognitivelybasicthantheplural,becausethat'stheorderwe
seeinlanguage).Universalslikethis,thatarespecificallylinguistic,shouldbecapturedinatheoryof
UniversalGrammar(UG)(Chomsky,1965,1981,1991).UGspecifiestheallowablementalrepresentations
andoperationsthatalllanguagesareconfinedtouse.Thetheoryofuniversalgrammariscloselytiedtothe
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

9/33

8/20/2016

LanguageAcquisition

theoryofthementalmechanismschildrenuseinacquiringlanguagetheirhypothesesaboutlanguagemust
becouchedinstructuressanctionedbyUG.
Toseehowlinguisticresearchcan'tbeignoredinunderstandinglanguageacquisition,considerthesentences
below.Ineachoftheexamples,alearnerwhoheardthe(a)and(b)sentencescouldquitesensiblyextracta
generalrulethat,whenappliedtothe(c)sentence,yieldversion(d).Yettheresultisanoddsentencethatno
onewouldsay:
1.(a)JohnsawMarywithherbestfriend'shusband.
(b)WhodidJohnseeMarywith?
(c)JohnsawMaryandherbestfriend'shusband.
(d)*WhodidJohnseeMaryand?
2.(a)Irvdrovethecarintothegarage.
(b)Irvdrovethecar.
(c)Irvputthecarintothegarage.
(d)*Irvputthecar.
3.(a)Iexpectthefurtofly.
(b)Iexpectthefurwillfly.
(c)Thefurisexpectedtofly.
(d)*Thefurisexpectedwillfly.
4.(a)Thebabyseemstobeasleep.
(b)Thebabyseemsasleep.
(c)Thebabyseemstobesleeping.
(d)*Thebabyseemssleeping.
5.(a)JohnlikedthepicturesofBillthatMarytook.
(b)JohnlikedMary'spicturesofBill.
(c)JohnlikedthepicturesofhimselfthatMarytook.
(d)*JohnlikedMary'spicturesofhimself.
Thesolutiontotheproblemmustbethatchildren'slearningmechanismsultimatelydon'tallowthemtomake
whatwouldotherwisebeatemptinggeneralization.Forexample,in(1),constraintsthatpreventextractionof
asinglephraseoutofacoordinatestructure(phrasesjoinedbyawordlikeandoror)wouldblockwould
whatotherwisebeanaturalgeneralizationfromotherexamplesofextraction,suchas1(ab).Theother
examplespresentsotherpuzzlesthatthetheoryofuniversalgrammar,aspartofatheoryoflanguage
acquisition,mustsolve.Itisbecauseofthesubtletyoftheseexamples,andtheabstractnessoftheprinciples
ofuniversalgrammarthatmustbepositedtoexplainthem,thatChomskyhasclaimedthattheoverall
structureoflanguagemustbeinnate,basedonhispaperandpencilexaminationofthefactsoflanguage
alone.

6Input
Tounderstandhowchildrenlearnlanguage,wehavetoknowwhataspectsoflanguage(fromtheirparentsor
peers)theyhaveaccessto.

6.1PositiveEvidence
Childrenclearlyneedsomekindoflinguisticinputtoacquirealanguage.Therehavebeenoccasionalcases
inhistorywhereabandonedchildrenhavesomehowsurvivedinforests,suchasVictor,theWildBoyof
Aveyron(subjectofafilmbyFrancoisTruffaut).Occasionallyothermodernchildrenhavegrownupwild
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

10/33

8/20/2016

LanguageAcquisition

becausedepravedparentshaveraisedthemsilentlyindarkroomsandatticsthechapterbyNewportand
Gleitmandiscusssomeofthosecases.Theoutcomeisalwaysthesame:thechildren,whenfound,aremute.
Whateverinnategrammaticalabilitiesthereare,theyaretooschematictogenerateconcretespeech,words,
andgrammaticalconstructionsontheirown.
Childrendonot,however,needtohearafullfledgedlanguageaslongastheyareinacommunitywithother
children,andhavesomesourceforindividualwords,theywillinventoneontheirown,ofteninasingle
generation.Childrenwhogrewupinplantationsandslavecolonieswereoftenexposedtoacrudepidginthat
servedasthelinguafrancaintheseBabelsoflaborers.Buttheygrewuptospeakgenuinelynewlanguages,
expressive"creoles"withtheirowncomplexgrammars(Bickerton,1984seealsotheChapterbyNewport
andGleitman).Thesignlanguagesofthedeafaroseinsimilarways.Indeed,theyarisespontaneouslyand
quicklywhereverthereisacommunityofdeafchildren(Senghas,1994Kegl,1994).
Childrenmostdefinitelydoneedtohearanexistinglanguagetolearnthatlanguage,ofcourse.Childrenwith
JapanesegenesdonotfindJapaneseanyeasierthanEnglish,orviceversatheylearnwhicheverlanguage
theyareexposedto.Theterm"positiveevidence"referstotheinformationavailabletothechildaboutwhich
stringsofwordsaregrammaticalsentencesofthetargetlanguage.
By"grammatical,"incidentally,linguistsandpsycholinguistsmeanonlythosesentencesthatsoundnaturalin
colloquialspeech,notnecessarilythosethatwouldbedeemed"properEnglish"informalwrittenprose.Thus
splitinfinitives,danglingparticiples,slang,andsoon,are"grammatical"inthissense(andindeed,areas
logical,systematic,expressive,andpreciseas"correct"writtenEnglish,oftenmoresoseePinker,1994a).
Similarly,ellipticalutterances,suchaswhenthequestionWhereareyougoing?isansweredwithTothe
store),countasgrammatical.Ellipsisisnotjustrandomsnippingfromsentences,butisgovernedbyrules
thatarepartofthegrammarofone'slanguageordialect.Forexample,thegrammarofcasualBritishEnglish
allowsyoutoanswerthequestionWillhego?bysayingHemightdo,whereasthegrammarofAmerican
Englishdoesn'tallowit.
Giventhisscientificdefinitionof"grammatical,"dowefindthatparents'speechcountsas"positive
evidence"?Thatis,whenaparentusesasentence,canthechildassumethatitispartofthelanguagetobe
learned,ordoparentsusesomanyungrammaticalsentencesrandomfragments,slipsofthetongue,
hesitations,andfalsestartsthatthechildwouldhavetotakemuchofitwithagrainofsalt?Fortunatelyfor
thechild,thevastmajorityofthespeechtheyhearduringthelanguagelearningyearsisfluent,complete,
andgrammaticallywellformed:99.93%,accordingtooneestimate(Newport,Gleitman,&Gleitman,1977).
Indeed,thisistrueofconversationamongadultsingeneral(Labov,1969).
Thuslanguageacquisitionisordinarilydrivenbyagrammaticalsampleofthetargetlanguage.Notethathis
istrueevenforformsofEnglishthatpeopleunthinkinglycall"ungrammatical,""fractured,"or"bad
English,"suchasruralAmericanEnglish(e.g.,thembookshedon'tweain'ttheydrughimaway)and
urbanblackEnglish(e.g.,ShewalkingHebeworkingseetheChapterbyLabov).Thesearenotcorrupted
versionsofstandardEnglishtoalinguisttheylookjustlikedifferentdialects,asrulegovernedasthe
southernEnglanddialectofEnglishthat,forhistoricalreasons,becamethestandardseveralcenturiesago.
Scientificallyspeaking,thegrammarofworkingclassspeechindeed,everyhumanlanguagesystemthat
hasbeenstudiedisintricatelycomplex,thoughdifferentlanguagesarecomplexindifferentways.

6.2NegativeEvidence
Negativeevidencereferstoinformationaboutwhichstringsofwordsarenotgrammaticalsentencesinthe
language,suchascorrectionsorotherformsoffeedbackfromaparentthattellthechildthatoneofhisorher
utterancesisungrammatical.AsmentionedinSection),it'sveryimportantforustoknowwhetherchildren
getandneednegative,becauseintheabsenceofnegativeevidence,anychildwhohypothesizesarulethat
generatesasupersetofthelanguagewillhavenowayofknowingthatheorsheiswrongGold,1967Pinker,
1979,1989).Ifchildrendon'tget,ordon'tuse,negativeevidence,theymusthavesomemechanismthat
eitheravoidsgeneratingtoolargealanguagethechildwouldbeconservativeorthatcanrecoverfromsuch
overgeneration.
RogerBrownandCamilleHanlon(1970)attemptedtotestB.F.Skinner'sbehavioristclaimthatlanguage
learningdependsonparents'reinforcementofchildren'sgrammaticalbehaviors.Usingtranscriptsof
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

11/33

8/20/2016

LanguageAcquisition

naturalisticparentchilddialogue,theydividedchildren'ssentencesintoonesthatweregrammaticallywell
formedandonesthatcontainedgrammaticalerrors.Theythendividedadults'responsestothosesentences
intoonesthatexpressedsomekindofapproval(e.g.,"yes,that'sgood")andthosethatexpressedsomekind
ofdisapproval.Theylookedforacorrelation,butfailedtofindone:parentsdidnotdifferentiallyexpress
approvalordisapprovaltotheirchildrencontingentonwhetherthechild'spriorutterancewaswellformedor
not(approvaldepends,instead,onwhetherthechild'sutterancewastrue).BrownandHanlonalsolookedat
children'swellformedandbadlyformedquestions,andwhetherparentsseemedtoanswerthem
appropriately,asiftheyunderstoodthem,orwithnonsequiturs.Theyfoundparentsdonotunderstandtheir
children'swellformedquestionsbetterthantheirbadlyformedones.
Otherstudies(e.g.HirshPasek,Treiman,andSchneiderman,1984Demetras,Post,andSnow,1986Penner,
1987Bohannon&Stanowicz,1988)havereplicatedthatresult,butwithatwist.Somehavefoundsmall
statisticalcontingenciesbetweenthegrammaticalityofsomechildren'ssentenceandthekindoffollowup
givenbytheirparentsforexample,whethertheparentrepeatsthesentenceverbatim,asksafollowup
question,orchangesthetopic.ButMarcus(1993)hasfoundthatthesepatternsfallfarshortofnegative
evidence(reliableinformationaboutthegrammaticalstatusofanywordstring).Differentparentsreactin
oppositewaystotheirchildren'sungrammaticalsentences,andmanyformsofungrammaticalityarenot
reactedtoatallleavingagivenchildunabletoknowwhattomakeofanyparentalreaction.Evenwhena
parentdoesreactdifferentially,achildwouldhavetorepeataparticularerror,verbatim,hundredsoftimesto
eliminatetheerror,becausetheparent'sreactionisonlystatistical:thefeedbacksignalsgivento
ungrammaticalsignalsarealsogivennearlyasoftentogrammaticalsentences.
Stromswold(1994)hasanevenmoredramaticdemonstrationthatparentalfeedbackcannotbecrucial.She
studiedachildwho,forunknownneurologicalreasons,wascongenitallyunabletotalk.Hewasagood
listener,though,andwhentestedhewasabletounderstandcomplicatedsentencesperfectly,andtojudge
accuratelywhetherasentencewasgrammaticalorungrammatical.Theboy'sabilitiesshowthatchildren
certainlydonotneednegativeevidencetolearngrammaticalrulesproperly,evenintheunlikelyeventthat
theirparentsprovidedit.
Theseresults,thoughofprofoundimportance,shouldnotbetoosurprising.EveryspeakerofEnglishjudges
sentencessuchasIdribbledthefloorwithpaintandTenpoundswasweighedbytheboyandWhodoyou
believetheclaimthatJohnsaw?andJohnaskedMarytolookathimselftobeungrammatical.Butitis
unlikelythateverysuchspeakerhasatsomepointutteredthesesentencesandbenefitedfromnegative
feedback.Thechildmusthavesomementalmechanismsthatruleoutvastnumbersof"reasonable"stringsof
wordswithoutanyoutsideintervention.

6.3Motherese
Parentsandcaretakersinmostpartsoftheworldmodifytheirspeechwhentalkingtoyoungchildren,one
exampleofhowpeopleingeneraluseseveral"registers"indifferentsocialsettings.Speechtochildrenis
slower,shorter,insomeways(butnotall)simpler,higherpitched,moreexaggeratedinintonation,more
fluentandgrammaticallywellformed,andmoredirectedincontenttothepresentsituation,comparedto
speechamongadults(Snow&Ferguson,1977).Manyparentsalsoexpandtheirchildren'sutterancesinto
fullsentences,oroffersequencesofparaphrasesofagivensentence.
Oneshouldnot,though,considerthisspeechregister,sometimescalled"Motherese,"tobeasetof"language
lessons."Thoughmother'sspeechmayseemsimpleatfirstglance,inmanywaysitisnot.Forexample,
speechtochildrenisfullofquestionssometimesamajorityofthesentences.Ifyouthinkquestionsare
simple,justtrytowriteasetofrulesthataccountsforthefollowingsentencesandnonsentences:
1.Hecangosomewhere.
Wherecanhego?
*Wherecanhegosomewhere?
*Wherehecango?
*Wheredidhecango?
2.Hewentsomewhere.
Wheredidhego?
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

12/33

8/20/2016

LanguageAcquisition

HewentWHERE?
*Wherewenthe?
*Wheredidhewent?
*Wherehewent?
*HedidgoWHERE?
3.Hewenthome.
Whydidhegohome?
Howcomehewenthome?
*Whyhewenthome?
*Howcomedidhegohome?
Linguistsstruggleoverthesefacts(seetheChaptersbyLasnikandLarson),someofthemostpuzzlinginthe
Englishlanguage.Butthesearetheconstructionsthatinfantsarebombardedwithandthattheymasterin
theirpreschoolyears.
ThechapterbyNewportandGleitmangivesanotherreasonfordoubtingthatMothereseisasetoflanguage
lessons.ChildrenwhosemothersuseMotheresemoreconsistentlydon'tpassthroughthemilestonesof
languagedevelopmentanyfaster(Newport,etal,1977).Furthermore,therearesomecommunitieswith
radicallydifferentideasaboutchildren'sproperplaceinsociety.Insomesocieties,forexample,people
tacitlyassumethatthatchildrenaren'tworthspeakingto,anddon'thaveanythingtosaythatisworth
listeningto.Suchchildrenlearntospeakbyoverhearingstreamsofadulttoadultspeech(Heath,1983).In
somecommunitiesinNewGuinea,mothersconsciouslytrytoteachtheirchildrenlanguage,butnotinthe
stylefamiliartous,oftalkingtothemindulgently.Rather,theywaituntilathirdpartyispresent,andcoach
thechildastotheproper,adultlikesentencestheyshoulduse(seeSchieffelin&Eisenberg,1981).
Nonetheless,thosechildren,likeallchildren,growuptobefluentlanguagespeakers.Itsurelymusthelp
childrenwhentheirparentsspeakslowly,clearly,andsuccinctlytothem,buttheirsuccessatlearningcan'tbe
explainedbyanyspecialgrammarunveilingpropertiesofparentalbabytalk.

6.4Prosody
Parentalspeechisnotastringofprintedwordsonatickertape,norisitinamonotonelikesciencefiction
robots.Normalhumanspeechhasapatternofmelody,timing,andstresscalledprosody.Andmotherese
directedtoyounginfantshasacharacteristic,exaggeratedprosodyofitsown:ariseandfallcontourfor
approving,asetofsharpstaccatoburstsforprohibiting,arisepatternfordirectingattention,andsmooth,low
legatomurmursforcomforting.Fernald(1992)hasshownthatthesepatternsareverywidespreadacross
languagecommunities,andmaybeuniversal.Themelodiesseemtoattractthechild'sattention,markthe
soundsasspeechasopposedtostomachgrowlingsorothernoises,andmightdistinguishstatements,
questions,andimperatives,delineatemajorsentenceboundaries,andhighlightnewwords.Whengivena
choice,babiesprefertolistentospeechwiththesepropertiesthantospeechintendedforadults(Fernald,
1984,1992HirshPasek,Nelson,Jusczyk,Cassidy,Druss,&Kennedy,1987).
Inallspeech,anumberofprosodicpropertiesofthespeechwave,suchaslengthening,intonation,and
pausing,areinfluencedbythesyntacticstructureofthesentence(Cooper&PacciaCooper,1980).Justlisten
tohowyouwouldsaythewordlikeinthesentenceTheboyIlikesleptcomparedtoTheboyIsawlikes
sleds.Inthefirstsentence,thewordlikeisattheboundaryofarelativeclauseandisdrawnout,exaggerated
inintonation,andfollowedbyapauseinthesecond,itisinthemiddleofaverbphraseandispronounced
morequickly,uniformlyinintonation,andisruntogetherwiththefollowingword.Somepsychologists(e.g.,
Gleitman&Wanner,1984Gleitman,1990)havesuggestedthatchildrenusethisinformationinthereverse
direction,andreadthesyntacticstructureofasentencedirectlyoffitsmelodyandtiming.Wewillexamine
thehypothesisinSection.

6.5Context
Childrendonothearsentencesinisolation,butinacontext.Nochildhaslearnedlanguagefromtheradio
indeed,childrenrarelyifeverlearnlanguagefromtelevision.ErvinTripp(1973)studiedhearingchildrenof
deafparentswhoseonlyaccesstoEnglishwasfromradioortelevisionbroadcasts.Thechildrendidnotlearn
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

13/33

8/20/2016

LanguageAcquisition

anyspeechfromthatinput.Onereasonisthatwithoutalreadyknowingthelanguage,itwouldbedifficultfor
achildtofigureoutwhatthecharactersintheunresponsivetelevisedworldsaretalkingabout.Ininteracting
withlivehumanspeakers,whotendtotalkaboutthehereandnowinthepresenceofchildren,thechildcan
bemoreofamindreader,guessingwhatthespeakermighthavemeant(Macnamara,1972,1982
Schlesinger,1971).Thatis,beforechildrenhavelearnedsyntax,theyknowthemeaningofmanywords,and
theymightbeabletomakegoodguessesastowhattheirparentsaresayingbasedontheirknowledgeofhow
thereferentsofthesewordstypicallyact(forexample,peopletendtoeatapples,butnotviceversa).Infact,
parentalspeechtoyoungchildrenissoredundantwithitscontextthatapersonwithnoknowledgeofthe
orderinwhichparents'wordsarespoken,onlythewordsthemselves,caninferfromtranscripts,withhigh
accuracy,whatwasbeingsaid(Slobin,1977).
Manymodelsoflanguageacquisitionassumethattheinputtothechildconsistsofasentenceanda
representationofthemeaningofthatsentence,inferredfromcontextandfromthechild'sknowledgeofthe
meaningsofthewords(e.g.Anderson,1977Berwick,1986Pinker,1982,1984Wexler&Culicover,
1980).Ofcourse,thiscan'tliterallybetruechildrendon'theareverywordofeverysentence,andsurely
don't,tobeginwith,perceivetheentiremeaningofasentencefromcontext.Blindchildren,whoseaccessto
thenonlinguisticworldisobviouslyseverelylimited,learnlanguagewithoutmanyproblems(Landau&
Gleitman,1985).Andwhenchildrendosucceedinguessingaparent'smeaning,itcan'tbebysimple
temporalcontiguity.Forexample,Gleitman(1990)pointsoutthatwhenamotherarrivinghomefromwork
opensthedoor,sheislikelytosay,"Whatdidyoudotoday?,"notI'mopeningthedoor.Similarly,sheis
likelytosay"Eatyourpeas"whenherchildis,say,lookingatthedog,andcertainlynotwhenthechildis
alreadyeatingpeas.
Still,theassumptionofcontextderivedsemanticinputisareasonableidealization,ifoneconsidersthe
abilitiesofthewholechild.Thechildmustkeepanupdatedmentalmodelofthecurrentsituation,createdby
mentalfacultiesforperceivingobjectsandeventsandthestatesofmindandcommunicativeintentionsof
otherhumans.Thechildcanusethisknowledge,plusthemeaningsofanyfamiliarwordsinthesentence,to
inferwhattheparentprobablymeant.InSectionwewilldiscusshowchildrenmightfilltheimportantgaps
inwhattheycaninferfromcontext.

7WhatandWhenChildrenLearn
Peopledonotreproducetheirparents'languageexactly.Iftheydid,wewouldallstillbespeakinglike
Chaucer.Butinanygeneration,inmosttimes,thedifferencesbetweenparents'languageandtheonetheir
childrenultimatelyacquireissmall.Andrememberthat,judgingbytheirspontaneousspeech,wecan
concludethatmostchildrenhavemasteredtheirmothertongue(allowingforperformanceerrorsdueto
complexityorrarityofaconstruction)sometimeintheirthrees.Itseemsthatthesuccesscriterionforhuman
languageissomethingclosetofullmastery,andinashortperiodoftime.
Toshowthatyoungchildrenreallyhavegraspedthedesignplanoflanguage,ratherthanmerely
approximatingitwithoutwardlyconvincingroutinesorrulesofthumbwhichwouldhavetobesupplanted
laterinlife,wecan'tjustrelyonwhattheysayweneedtousecleverexperimentaltechniques.Let'slookat
twoexamplesthatillustratehowevenveryyoungchildrenseemtoobeytheinnatecomplexdesignof
UniversalGrammar.
EarlierImentionedthatinalllanguages,iftherearederivationalaffixesthatbuildnewwordsoutofold
ones,likeism,er,andable,andinflectionalaffixesthatmodifyawordaccordingtoitsroleinthesentence,
likes,ed,anding,thenthederivationalaffixappearsinsidetheinflectionalone:Darwinismsispossible,
Darwinsismisnot.Thisandmanyothergrammaticalquirkswerenicelyexplainedinatheoryofword
structureproposedbyPaulKiparsky(1982).
Kiparskyshowedthatwordsarebuiltinlayersor"levels."Tobuildaword,youcanstartwitharoot(like
Darwin).Thenyoucanrulesofacertainkindtoit,called"Level1Rules,"toyieldamorecomplexword.
Forexample,thereisaruleaddingthesuffixian,turningthewordintoDarwinian.Level1Rules,according
tothetheory,canaffectthesoundofthesteminthiscase,thesyllablecarryingthestressshiftsfromDarto
win.Level2rulesapplytoawordafteranyLevel1ruleshavebeenapplied.AnexampleofaLevel2ruleis
theonethataddsthesuffixism,yielding,forexample,Darwinism.Level2rulesgenerallydonotaffectthe
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

14/33

8/20/2016

LanguageAcquisition

pronunciationofthewordstheyapplytotheyjustaddmaterialontotheword,leavingthepronunciation
intact.(ThestressinDarwinismisthesameasitwasinDarwin.)Finally,Level3rulesapplytoawordafter
anyLevel2ruleshavebeenapplied.TheregularrulesofinflectionalmorphologyareexamplesofLevel3
rules.Anexampleistherulethataddsanstotheendofanountoformitspluralforexample,Darwinians
orDarwinisms.
Crucially,therulescannotapplyoutoforder.TheinputtoaLevel1rulesmustbeawordroot.Theinputtoa
level2rulemustbeeitherarootortheoutputofLevel1rules.TheinputtoaLevel3rulemustbearoot,the
outputofLevel1rules,ortheoutputofLevel2rules.Thatconstraintyieldspredictionsaboutwhatkindsof
wordsarepossibleandwhichareimpossible.Forexample,theorderingmakesitimpossibletoderive
DarwinianismandDarwinianisms,butnotDarwinsian,Darwinsism,andDarwinismian.
Now,irregularinflection,suchasthepairingofmousewithmice,belongstoLevel1,whereasregular
inflectionalrules,suchastheonethatrelatesrattorats,belongstoLevel3.Compounding,therulethat
wouldproduceDarwinloverandmousetrap,isaLevel2rule,inbetween.Thiscorrectlypredictsthatan
irregularpluralcaneasilyappearinsideacompound,butaregularpluralcannot.Comparethefollowing:
iceinfested(OK)ratsinfested(bad)
menbashing(OK)guysbashing(bad)
teethmarks(OK)clawsmarks(bad)
feetwarmer(OK)handwarmer(bad)
purplepeopleeater(OK)purplebabieseater(bad)
Miceinfestedisapossibleword,becausetheprocessconnectingmousewithmicecomesbeforetherule
combiningthenounwithinfested.However,ratsinfested,eventhoughitiscognitivelyquitesimilartomice
infested,soundsstrangewecansayonlyratinfested(eventhoughbydefinitiononeratdoesnotmakean
infestation).
PeterGordon(1986)hadchildrenbetweentheagesof3and5participateinanelicitedproduction
experimentinwhichhewouldsay,"Hereisapuppetwholikestoeat_____.Whatwouldyoucallhim?"He
providedaresponseforseveralsingularmassnouns,likemud,beforehand,sothatthechildrenwereaware
oftheexistenceofthe"xeater"compoundform.Childrenbehavedjustlikeadults:apuppetwholikestoeat
amousewascalledamouseeater,apuppetwholikestoeataratwascalledarateater,apuppetwholikesto
eatmicewascalledeitheramouseeateroramiceeaterbutapuppetwholikestoeatratswascalleda
rateater,neveraratseater.Interestingly,childrentreatedtheirownoverregularizations,suchasmouses,
exactlyastheytreatedlegitimateregularplurals:theywouldnevercallthepuppetamouseseater,evenif
theyusedmousesintheirownspeech.
Evenmoreinterestingly,Gordonexaminedhowchildrencouldhaveacquiredtheconstraint.Perhaps,he
reasoned,theyhadlearnedthefactthatcompoundscancontaineithersingularsorirregularplurals,never
regularplurals,bypayingkeepingtrackofallthekindsofcompoundsthatdoanddon'toccurintheir
parents'speech.Itturnsoutthattheywouldhavenowayoflearningthatfact.Althoughthereisno
grammaticalreasonwhycompoundswouldnotcontainirregularplurals,thespeechthatmostchildrenhear
doesnotcontainany.Compoundsliketoothbrushaboundcompoundscontainingirregularpluralslike
teethmarks,peopleeater,andmenbashing,thoughgrammaticallypossible,arestatisticallyrare,accordingto
thestandardizedfrequencydatathatGordonexamined,andhefoundnonethatwaslikelytoappearinthe
speechchildrenhear.Thereforechildrenwerewillingtosaymiceeaterandunwillingtosayratseaterwith
nogoodevidencefromtheinputthatthatisthepatternrequiredinEnglish.Gordonsuggeststhatthisshows
thattheconstraintsonlevelorderingmaybeinnate.
Let'snowgofromwordstosentences.Sentenceareorderedstringsofwords.Nochildcouldfailtonotice
wordorderinlearningandunderstandinglanguage.Butmostregularitiesoflanguagegovernhierarchically
organizedstructureswordsgroupedintophrases,phrasesgroupedintoclauses,clausesgroupedinto
sentences(seetheChaptersbyLasnik,byLarson,andbyNewport&Gleitman).Ifthestructuresoflinguistic
theorycorrespondtothehypothesesthatchildrenformulatewhentheyanalyzeparentalspeechandform
rules,childrenshouldcreaterulesdefinedoverhierarchicalstructures,notsimplepropertiesoflinearorder
suchaswhichwordcomesbeforewhichotherwordorhowclosetwowordsareinasentence.Thechapter
byGleitmanandNewportdiscussesonenicedemonstrationofhowadults(whoare,afterall,justgrownup
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

15/33

8/20/2016

LanguageAcquisition

children)respectconstituentstructure,notsimplewordorder,whenformingquestions.Hereisanexample
makingasimilarpointthathasbeentriedoutwithchildren.
Languagesoftenhaveembeddedclausesmissingasubject,suchasJohntoldMarytoleave,wherethe
embedded"downstairs"clausetoleavehasnosubject.Thephenomenonofcontrolgovernshowthemissing
subjectisinterpreted.InthissentenceitisMarywhoisunderstoodashavingtheembeddedsubject'srole,
thatis,thepersondoingtheleaving.WesaythatthephraseMary"controls"themissingsubjectpositionof
thelowerclause.Formostverbs,thereisasimpleprincipledefiningcontrol.Iftheupstairsverbhasno
object,thenthesubjectoftheupstairsverbcontrolsthemissingsubjectofthedownstairsverb.Forexample,
inJohntriedtoleave,Johnisinterpretedasthesubjectofbothtryandleave.Iftheupstairsverbhasasubject
andanobject,thenitistheobjectthatcontrolsthemissingsubjectofthedownstairsverb,aswesawinJohn
toldMarytoleave.
In1969,CarolChomskypublishedasetofclassicexperimentsindevelopmentalpsycholinguistics.She
showedthatchildrenapplythisprinciplequiteextensively,evenforthehandfulofverbsthatareexceptions
toit.Inactoutcomprehensionexperimentsonchildrenbetweentheagesof5and10,sheshowedthateven
relativelyoldchildrenwerepronetothiskindofmistake.Whentold"MickeypromisedDonaldtojump
Makehimjump,"thechildrenmadeDonald,theobjectofthefirstverb,dothejumping,inaccordwiththe
generalprinciple.The"rightanswer"inthiscasewouldhavebeenMickey,becausepromiseisanexception
totheprinciple,callingforanunusualkindofcontrolwherethesubjectoftheupstairsverb,nottheobjectof
theupstairsverb,shouldactascontroller.
Butwhat,exactly,istheprinciplethatchildrenareoverapplying?OnepossibilitycanbecalledtheMinimal
DistancePrinciple:thecontrollerofthedownstairsverbisthenounphrasenearesttoitinthelinearstringof
wordsinthesentence.Ifchildrenanalyzesentencesintermsoflinearorder,thisshouldbeanatural
generalization.However,itisn'trightfortheadultlanguage.ConsiderthepassivesentenceMarywastoldby
Johntoleave.ThephraseJohnisclosesttothesubjectpositionforleave,butadultEnglishspeakers
understandthesentenceasmeaningthatMaryistheoneleaving.TheMinimalDistancePrinciplegivesthe
wronganswerhere.Instead,fortheadultlanguage,weneedaprinciplesensitivetogrammaticalstructure,
suchasthe"ccontrol"structuralrelationdiscussedintheChapterbyLasnik[?].Let'sconsiderasimplified
version,whichwecancalltheStructuralPrinciple.Itmightsaythatthecontrollerofamissingsubjectisthe
grammaticalobjectoftheupstairsverbifithasoneotherwiseitisthegrammaticalsubjectoftheupstairs
verb(bothofthemccommandthemissingsubject).Theobjectofaprepositioninthehigherclause,
however,isneverallowedtobeacontroller,basicallybecauseitisembedded"toodeeply"inthesentence's
treestructuretoccommandthemissingsubject.That'swhyMarywastoldbyJohntoleavehasMaryasthe
controller.(Itisalsowhy,incidentally,thesentenceMarywaspromisedbyJohntoleaveisunintelligibleit
wouldrequireaprepositionalphrasetobethecontroller,whichisruledoutbytheStructuralPrinciple.)
ItwouldcertainlybeunderstandableifchildrenweretofollowtheMinimalDistancePrinciple.Notonlyisit
easilystatedintermsofsurfacepropertiesthatchildrencaneasilyperceive,butsentencesthatwould
disconfirmitlikeMarywastoldbyJohntoleaveareextremelyrareinparents'speech.MichaelMaratsos
(1974)didthecrucialexperiment.Hegavechildrensuchsentencesandaskedthemwhowasleaving.Of
course,oneitheraccountchildrenwouldhavetobeabletounderstandthepassiveconstructiontointerpret
thesesentences,andMaratsosgavethemaseparatetestofcomprehensionofsimplepassivesentencesto
selectoutonlythosechildrenwhocoulddoso.Andindeed,hefoundthatthosechildreninterpretedpassive
sentenceswithmissingembeddedsubjectsjustasadultswould.Thatis,inaccordwiththeStructural
PrincipleandinviolationoftheMinimalDistancePrinciple,theyinterpretedMarywastoldbyJohntoleave
ashavingthesubject,Mary,dotheleavingthatis,asthecontroller.Theexperimentshowshowyoung
childrenhavegraspedtheabstractstructuralrelationsinsentences,andhaveacquiredagrammarofthesame
designasthatspokenbytheirparents.

8TheChild'sLanguageLearningAlgorithm
Hereisthemostbasicprobleminunderstandinghowchildrenlearnalanguage:Theinputtolanguage
acquisitionconsistsofsoundsandsituationstheoutputisagrammarspecifying,forthatlanguage,theorder
andarrangementofabstractentitieslikenouns,verbs,subjects,phrasestructures,control,andccommand
(seetheChaptersbyLasnikandLarson,andthedemonstrationsinthischapterandtheonebyGleitmanand
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

16/33

8/20/2016

LanguageAcquisition

Newport).Somehowthechildmustdiscovertheseentitiestolearnthelanguage.Weknowthateven
preschoolchildrenhaveanextensiveunconsciousgraspofgrammaticalstructure,totheexperimentson
discussedintheprevioussection,buthowhasthechildmanagedtogofromsoundsandsituationsto
syntacticstructure?
Innateknowledgeofgrammaritselfisnotsufficient.Itdoesnogoodforthechildtohavewrittendowninhis
brain"Thereexistnouns"childrenneedsomewayoffindingtheminparents'speech,sothattheycan
determine,amongotherthings,whetherthenounscomebeforetheverb,asinEnglish,orafter,asinIrish.
Oncethechildfindsnounsandverbs,anyinnateknowledgewouldimmediatelybehelpful,becausethechild
couldthendeduceallkindsofimplicationsabouthowtheycanbeused.Butfindingthemisthecrucialfirst
step,anditisnotaneasyone.
InEnglish,nounscanbeidentifiedasthosethingsthatcomeafterarticles,getsuffixedwithsintheplural,
andsoon.Buttheinfantobviouslydoesn'tknowthatyet.Nounsdon'toccurinanyconstantpositionina
sentenceacrossthelanguagesoftheworld,andtheyaren'tsaidwithanyparticulartoneofvoice.Nordo
nounshaveaconstantmeaningtheyoftenrefertophysicalthings,likedogs,butdon'thaveto,asinThe
daysofourlivesandThewarmthofthesun.Thesameistrueforotherlinguisticentities,suchasverbs,
subjects,objects,auxiliaries,andtense.Sincethechildmustsomehow"lifthimselfupbyhisbootstraps"to
getstartedinformulatingagrammarforthelanguage,thisiscalledthe"bootstrappingproblem"(seePinker,
1982,1984,1987b,1989,1994Morgan,1986Gleitman,1990andthecontributorstoMorganandDemuth,
1995).Severalsolutionscanbeenvisioned.

8.1ExtractingSimpleCorrelations
Onepossibilityisthatthechildsetsupamassivecorrelationmatrix,andtallieswhichwordsappearinwhich
positions,whichwordsappearnexttowhichotherwords,whichwordsgetwhichprefixesandsuffixesin
whichcircumstances,andsoon.Syntacticcategorieswouldariseimplicitlyasthechilddiscoveredthat
certainsetsofpropertiesaremutuallyintercorrelatedinlargesetsofwords.Forexample,manywordstend
tooccurbetweenasubjectandanobject,areinflectedwithswhenthesubjectissingularandinthethird
personandthetenseispresent,andoftenappearafterthewordto.Thissetofwordswouldbegrouped
togetherastheequivalentofthe"verb"category(Maratsos&Chalkley,1981).
Therearetwoproblemswiththisproposal.Themainoneisthatthefeaturesthattheprelinguisticchildis
supposedtobecrossreferencingarenotaudiblymarkedinparentalspeech.Rather,theyareperceptibleonly
tochildwhohasalreadyanalyzedthegrammarofthelanguagejustwhattheproposalistryingtoexplain
inthefirstplace!Howisaprelinguisticchildsupposedtofindthe"subject"ofthesentenceinorderto
correlateitwiththeendingonthewordsheorsheisfocusingon?Asubjectisnotthesamethingasthefirst
wordortwoofthesentence(e.g.,Thebigbadwolfhuffedandpuffed)oreventhefirstphrase(e.g.,Whatdid
thebigbadwolfdo?).Wehaveadilemma.Ifthefeaturesdefiningtherowsandcolumnsofthecorrelation
matrixarethingsthatareperceptibletothechild,like"firstwordinasentence,"thengrammaticalcategories
willneveremerge,becausetheyhavenoconsistentcorrelationwiththesefeatures.Butifthefeaturesarethe
thingsthatdodefinegrammaticalcategories,likeagreementandphrasestructureposition,theproposal
assumesjustwhatitsetsouttoexplain,namelythatthechildhasanalyzedtheinputintoitscorrect
grammaticalstructures.Somehow,thechildmustbreakintothiscircle.Itisageneraldangerthatpopsupin
cognitivepsychologywheneveranyoneproposesamodelthatdependsoncorrelationsamongfeatures:there
isalwaysatemptationtogliblyendowthefeatureswiththecomplex,abstractrepresentationswhose
acquisitiononeistryingtoexplain.
Thesecondproblemisthat,withoutpriorconstraintsonthedesignofthefeaturecorrelator,therearean
astronomicalnumberofpossibleintercorrelationsamonglinguisticpropertiesforthechildtotest.Totake
justtwo,thechildwouldhavetodeterminewhetherasentencecontainingthewordcatinthirdpositionmust
haveapluralwordattheend,andwhethersentencesendinginwordsendingindareinvariablyprecededby
wordsreferringtopluralentities.Mostofthesecorrelationsneveroccurinanynaturallanguage.Itwouldbe
mystery,then,whychildrenarebuiltwithcomplexmachinerydesignedtotestforthemthoughanother
wayofputtingitisthatitwouldbeamysterywhytherearenolanguagesexhibitingcertainkindsof
correlationsgiventhatchildrenarecapableoffindingthem.

http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

17/33

8/20/2016

LanguageAcquisition

8.2UsingProsody
Asecondwayinwhichthechildcouldbeginsyntaxlearningwouldbetoattendtotheprosodyofsentences,
andtopositphraseboundariesatpointsintheacousticstreammarkedbylengthening,pausing,anddropsin
fundamentalfrequency.Theproposalseemsattractive,becauseprosodicpropertiesareperceptiblein
advanceofknowinganysyntax,soatfirstglanceprosodyseemslikeastraightforwardwayforachildto
breakintothelanguagesystem.
Butoncloserexamination,theproposaldoesnotseemtowork(Pinker,1987,1994bFernaldand
McRoberts,inpressSteedman,inpress).Justasgoldglitters,butallthatglittersisnotgold,syntactic
structureaffectsaspectsofprosody,butaspectsofprosodyareaffectedbymanythingsbesidessyntax.The
effectsofemotionalstateofthespeaker,intentofthespeaker,wordfrequency,contrastivestress,and
syllabicstructureofindividualwords,areallmixedtogether,andthereisnowayforachildtodisentangle
themfromthesoundwavealone.Forexample,inthesentenceThebabyatetheslug,themainpause
coincideswiththemajorsyntacticboundarybetweenthesubjectandthepredicate.Butachildcannotwork
backwardsandassumethatthemainpauseinaninputsentencemarkstheboundarybetweenthesubjectand
thepredicate.InthesimilarsentenceHeatetheslug,themainpauseisatthemoreembeddedboundary
betweentheverbanditsobject.
Worse,themappingbetweensyntaxandprosody,evenwhenitisconsistent,isconsistentindifferentwaysin
differentlanguages.Soayoungchildcannotuseanysuchconsistency,atleastnotattheverybeginningof
languageacquisition,todecipherthesyntaxofthesentence,becauseititselfisoneofthethingsthathasto
belearned.

8.3UsingContextandSemantics
Athirdpossibility(seePinker,1982,1984,1989Macnamara,1982Grimshaw1981Wexler&Culicover,
1980Bloom,inpress)exploitsthefactthatthereisaonewaycontingencybetweensyntaxandsemanticsin
thebasicsentencesofmostoftheworld'slanguages.Thoughnotallnounsarephysicalobjects,allphysical
objectsarenamedbynouns.Similarly,ifaverbhasanargumentplayingthesemanticroleof'agent',then
thatargumentwillbeexpressedasthesubjectofbasicsentencesinlanguageafterlanguage.(Again,this
doesnotworkinreverse:thesubjectisnotnecessarilyanagent.InJohnlikedMarythesubjectisan
"experiencer"inJohnpleasedMaryitisanobjectofexperienceinJohnreceivedapackageitisagoalor
recipientinJohnunderwentanoperationitisapatient.)Similarly,entitiesdirectlyaffectedbyanactionare
expressedasobjects(butnotallobjectsareentitiesaffectedbyanaction)actionsthemselvesareexpressed
asverbs(thoughnotallverbsexpressactions).Evenphrasestructureconfigurationshavesemantic
correlates:argumentsofverbsreliablyappearas"sisters"totheminsidetheverbphraseinphrasestructure
trees(seethechapterbyLasnik).
Ifchildrenassumethatsemanticandsyntacticcategoriesarerelatedinrestrictedwaysintheearlyinput,they
couldusesemanticpropertiesofwordsandphrases(inferredfromcontextseeSection)asevidencethat
theybelongtocertainsyntacticcategories.Forexample,achildcaninferthatawordthatdesignateda
person,placeorthingisanoun,thataworddesignatinganactionisaverb,thatawordexpressingtheagent
argumentofanactionpredicateisthesubjectofitssentence,andsoon.Forexample,uponhearingthe
sentenceThecatchasedtherat,thechildcandeducethatinEnglishthesubjectcomesbeforetheverb,that
theobjectcomesaftertheverb,andsoon.Thiswouldgivethechildthebasisforcreatingthephrase
structuretreesthatallowhimorhertoanalyzetherulesofthelanguage.
Ofcourse,achildcannotliterallycreateagrammarthatcontainsruleslike"Agentwordscomebeforeaction
words."ThiswouldleavethechildnowayofknowinghowtoorderthewordsinsentencessuchasApples
appealtoMaryorJohnreceivedapackage.Butonceaninitialsetofrulesislearned,itemsthataremore
abstractorthatdon'tfollowtheusualpatternsrelatingsyntaxandsemanticcouldbelearnedthroughtheir
distributioninalreadylearnedstructures.Thatis,thechildcouldnowinferthatApplesisthesubjectof
appeal,andthatJohnisthesubjectofreceive,becausetheyareinsubjectposition,afactthechildnow
knowsthankstotheearliercatchasedratsentences.Similarly,thechildcouldinferthatappealisaverbto
beginwithbecauseitisinthe"verb"position.

http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

18/33

8/20/2016

LanguageAcquisition

9AcquisitioninAction
Whatdoalltheseargumentsmeanforwhatgoesoninachild'smindmomentbymomentasheorsheis
acquiringrulesfromparentalspeech?Let'slookattheprocessasconcretelyaspossible.

9.1BootstrappingtheFirstRules
Firstimagineahypotheticalchildtryingtoextractpatternsfromthefollowingsentences,withoutanyinnate
guidanceastohowhumangrammarworks.
Myroneatslamb.
Myroneatsfish.
Myronlikesfish.
Atfirstglance,onemightthinkthatthechildcouldanalyzetheinputasfollows.Sentencesconsistofthree
words:thefirstmustbeMyron,thesecondeithereatsorlikes,thethirdlamborfish.Withthesemicrorules,
thechildcanalreadygeneralizebeyondtheinput,tothebrandnewsentenceMyronlikeschicken.
Butlet'ssaythenexttwosentencesare
Myroneatsloudly.
Myronmightfish.
Thewordmightgetsaddedtothelistofwordsthatcanappearinsecondposition,andthewordloudlyis
addedtothelistthatcanappearinthirdposition.Butlookatthegeneralizationsthiswouldallow:
Myronmightloudly.
Myronlikesloudly.
Myronmightlamb.
Thisisnotworking.Thechildmustcouchrulesingrammaticalcategorieslikenoun,verb,andauxiliary,not
inactualwords.Thatway,fishasanounandfishasaverbcanbekeptseparate,andthechildwouldnot
adulteratethenounrulewithinstancesofverbsandviceversa.Ifchildrenarewillingtoguessthatwordsfor
objectsarenouns,wordsforactionsareverbs,andsoon,theywouldhavealegupontherulelearning
problem.
Butwordsarenotenoughtheymustbeordered.Imaginethechildtryingtofigureoutwhatkindofwordcan
occurbeforetheverbbother.Itcan'tbedone:
Thatdogbothersme.[dog,anoun]
Whatshewearsbothersme.[wears,averb]
Musicthatistooloudbothersme.[loud,anadjective]
Cheeringtooloudlybothersme.[loudly,anadverb]
Theguyshehangsoutwithbothersme.[with,apreposition]
Theproblemisobvious.Thereisacertainsomethingthatmustcomebeforetheverbbother,butthat
somethingisnotakindofworditisakindofphrase,anounphrase.Anounphrasealwayscontainsahead
noun,butthatnouncanbefollowedbymanyotherphrases.Soitisuselessoftrytolearnalanguageby
analyzingsentenceswordbyword.Thechildmustlookforphrasesandtheexperimentsongrammatical
controldiscussedearliershowsthattheydo.
Whatdoesitmeantolookforphrases?Aphraseisagroupofwords.Mostofthelogicallypossiblegroups
ofwordsinasentenceareuselessforconstructingnewsentences,suchaswearsbothersandcheeringtoo,
butthechild,unabletorelyonparentalfeedback,hasnowayofknowingthis.Soonceagain,children
cannotattackthelanguagelearningtasklikesomelogicianfreeofpreconceptionstheyneedprior
constraints.Wehavealreadyseenwheresuchconstraintscouldcome.First,thechildcouldassumethat
parents'speechrespectsthebasicdesignofhumanphrasestructure:phrasescontainheads(e.g.,anoun
phraseisbuiltaroundaheadnoun)argumentsaregroupedwithheadsinsmallphrases,sometimescalledX
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

19/33

8/20/2016

LanguageAcquisition

bars(seethechapterbyLasnik)Xbarsaregroupedwiththeirmodifiersinsidelargephrases(NounPhrase,
VerbPhrase,andsoon)phrasescanhavesubjects.Second,sincethemeaningsofparents'sentencesare
guessableincontext,thechildcouldusethemeaningstohelpsetuptherightphrasestructure.Imaginethata
parentsaysThebigdogateicecream.Ifthechildalreadyknowsthewordsbig,dog,ate,andicecream,he
orshecanguesstheircategoriesandgrowthefirstbranchesofatree:Inturn,nounsandverbsmustbelong
tonounphrasesandverbphrases,sothechildcanpositoneforeachofthesewords.Andifthereisabigdog
around,thechildcanguessthattheandbigmodifydog,andconnectthemproperlyinsidethenounphrase:If
thechildknowsthatthedogjustateicecream,heorshecanalsoguessthaticecreamanddogarearguments
oftheverbeat.Dogisaspecialkindofargument,becauseitisthecausalagentoftheactionandthetopicof
thesentence,andhenceitislikelytobethesubjectofthesentence,andthereforeattachestothe"S."Atree
forthesentencehasbeencompleted:Therulesanddictionaryentriescanbepeeledoffthetree:
S>NPVP
NP>(det)(A)N
VP>VNP
dog:N
icecream:N
ate:Veater=subject,thingeaten=object
the:det
big:A
Thishypotheticalexampleshowshowachild,ifsuitablyequipped,couldlearnthreerulesandfivewords
fromasinglesentenceincontext.
Theuseofpartofspeechcategories,phrasestructure,andmeaningguessedfromcontextarepowerfultools
thatcanhelpthechildinthedauntingtaskoflearninggrammarquicklyandwithoutsystematicparental
feedback(Pinker,1984).Inparticular,therearemanybenefitstousingasmallnumberofcategorieslikeN
andVtoorganizeincomingspeech.Bycallingboththesubjectandobjectphrases"NP,"ratherthan,say
Phrase#1andPhrase#2,thechildautomaticallycanapplyknowledgeaboutnounsinsubjectpositionto
nounsinobjectposition,andviceversa.Forexample,ourmodelchildcanalreadygeneralize,andusedogas
anobject,withouthavingheardanadultdoso,andthechildtacitlyknowsthatadjectivesprecedenounsnot
justinsubjectsbutinobjects,againwithoutdirectevidence.Thechildknowsthatifmorethanonedogis
dogsinsubjectposition,morethanonedogisdogsinobjectposition.
Moregenerally,Englishallowsatleasteightpossiblephrasematesofaheadnouninsideanounphrase,such
asJohn'sdogdogsintheparkbigdogsdogsthatIlike,andsoon.Inturn,thereareabouteightplacesina
sentencewherethewholenounphrasecango,suchasDogbitesmanManbitesdogAdog'slifeGivethe
boyadogTalktothedogandsoon.Therearethreewaystoinflectanoun:dog,dogs,dog's.Andatypical
childbythetimeheorsheisinhighschoolhaslearnedsomethinglike20,000differentnouns(Miller,1991
Pinker,1994a).Ifchildrenhadtolearnallthecombinationsseparately,theywouldneedtolistentoabout
140milliondifferentsentences.Atarateofasentenceeverytenseconds,tenhoursaday,itwouldtakeover
acentury.Butbyunconsciouslylabelingallnounsas"N"andallnounphrasesas"NP,"thechildhasonlyto
hearabouttwentyfivedifferentkindsofnounphraseandlearnthenounsonebyone,andthemillionsof
possiblecombinationsfalloutautomatically.
Indeed,ifchildrenareconstrainedtolookforonlyasmallnumberofphrasetypes,theyautomaticallygain
theabilitytoproduceaninfinitenumberofsentences,oneofthehallmarksofhumanlanguage.Takethe
phrasethetreeinthepark.IfthechildmentallylabelstheparkasanNP,andalsolabelsthetreeinthepark
asanNP,theresultingrulesgenerateanNPinsideaPPinsideanNPaloopthatcanbeiterated
indefinitely,asinthetreeneartheledgebythelakeintheparkinthecityintheeastofthestate....In
contrast,achildwhowasfreetotolabelintheparkasonekindofphrase,andthetreeintheparkanother,
wouldbedeprivedoftheinsightthatthephrasecontainsanexampleofitself.Thechildwouldbelimitedto
reproducingthatphrasestructurealone.
Witharudimentarybutroughlyaccurateanalysisofsentencestructuresetup,theotherpartsoflanguagecan
beacquiredsystematically.Abstractwords,suchasnounsthatdonotrefertoobjectsandpeople,canbe
learnedbypayingattentiontowheretheysitinsideasentence.SincesituationinThesituationjustifies
drasticmeasuresoccursinsideaphraseinNPposition,itmustbeanoun.Ifthelanguageallowsphrasesto
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

20/33

8/20/2016

LanguageAcquisition

bescrambledaroundthesentence,likeLatinortheAustralianaboriginallanguageWarlpiri,thechildcan
discoverthisfeatureuponcomingacrossawordthatcannotbeconnectedtoatreeintheexpectedplace
withoutcrossingbranches(inSection,wewillseethatchildrendoseemtoproceedinthisorder).Thechild's
mindcanalsoknowwhattofocusonindecodingcaseandagreementinflections:anoun'sinflectioncanbe
checkedtoseeifitappearswheneverthenounappearsinsubjectposition,inobjectposition,andsoona
verb'sinflectionmightcanbecheckedfortense,aspect,andthenumber,person,andgenderofitssubject
andobject.Thechildneednotbothercheckingwhetherthethirdwordinthesentencereferredtoareddishor
abluishobject,whetherthelastwordwaslongorshort,whetherthesentencewasbeingutteredindoorsor
outdoors,andbillionsofotherfruitlesspossibilitiesthatapurelycorrelationallearnerwouldhavetocheck.

9.2TheOrganizationofGrammarasaGuidetoAcquisition
Agrammarisnotabagofrulesthereareprinciplesthatlinkthevariouspartstogetherintoafunctioning
whole.ThechildcanusesuchprinciplesofUniversalGrammartoallowonebitofknowledgeabout
languagetoaffectanother.Thishelpssolvetheproblemofhowthechildcanavoidgeneralizingtotoolarge
alanguage,whichintheabsenceofnegativeevidencewouldbeincorrigible.Incaseswerechildrendo
overgeneralize,theseprinciplescanhelpthechildrecover:ifthereisaprinciplethatsaysthatAandB
cannotcoexistinalanguage,achildacquiringBcanuseittocatapultAoutofthegrammar.
9.2.1BlockingandInflectionalOverregularization
Thenextchapterpresentsagoodexample.TheBlockingprincipleinmorphologydictatesthatanirregular
formlistedinthementaldictionaryascorrespondingtoaparticularinflectionalcategory(say,pasttense),
blockstheapplicationofthecorrespondinggeneralrule.Forexample,adultsknowtheirregularformbroke,
andthatpreventsthemfromapplyingtheregular"added"ruletobreakandsayingbreaked.Children,who
havenotheardbrokeenoughtimestorememberitreliablyondemand,thusfailtoblocktheruleand
occasionallysaybreaked.Astheyhearbrokeenoughtimestorecallitreliably,Blockingwouldsuppressthe
regularrule,andtheywouldgraduallyrecoverfromtheseovergeneralizationerrors(Marcus,etal.,1992).
9.2.2InteractionsbetweenWordMeaningandSyntax
Hereisanotherexampleinwhichageneralprinciplerulesoutaformintheadultgrammar,butinthechild's
grammar,thecrucialinformationallowingtheprincipletoapplyismissing.Asthechild'sknowledge
increases,therelevanceoftheprincipletotheerrantformmanifestsitself,andtheformcanberuledoutso
astomakethegrammarasawholeconsistentwiththeprinciple.
Everyverbhasan"argumentstructure":aspecificationofwhatkindsofphrasesitcanappearwith(Pinker,
1989).Afamiliarexampleisthedistinctionbetweenatransitiveverblikedevour,whichrequiresadirect
object(youcansayHedevouredthesteakbutnotjustHedevoured)andanintransitiveverblikedine,which
doesnot(youcansayHedinedbutnotHedinedthesteak).Childrensometimesmakeerrorswiththe
argumentstructuresofverbsthatrefertotheactofmovingsomethingtoaspecifiedlocation(Bowerman,
1982bGropen,Pinker,Hollander,andGoldberg,1991a):
Ididn'tfillwateruptodrinkitIfilleditupfortheflowerstodrinkit.
CanIfillsomesaltintothebear?[abearshapedsaltshaker]
I'mgoingtocoverascreenoverme.
Feelyourhandtothat.
Terrisaidifthis[arhinestoneonashirt]wereadiamondthenpeoplewouldbetryingtorobthe
shirt.
Ageneralprincipleofargumentstructureisthattheargumentthatisaffectedinsomewayspecifiedbythe
verbgetsmappedontothesyntacticobject.Thisisanexampleofa"linkingrule,"whichlinkssemantics
withsyntax(andwhichisanexampleofthecontingencyayoungchildwouldhaveemployedtouse
semanticinformationtobootstrapintothesyntax).Forexample,foradults,the"container"argument(where
thewatergoes)isthedirectobjectoffillfilltheglasswithwater,notfillwaterintotheglassbecausethe
mentaldefinitionoftheverbfillsaysthattheglassbecomesfull,butsaysnothingabouthowthathappens
(onecanfillaglassbypouringwaterintoit,bydrippingwaterintoit,bydippingitintoapond,andsoon).
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

21/33

8/20/2016

LanguageAcquisition

Incontrast,foraverblikepour,itisthe"content"argument(thewater)thatistheobjectpourwaterinto
theglass,notpourtheglasswithwaterbecausethementaldefinitionoftheverbpoursaysthatthewater
mustmoveinacertainmanner(downward,inastream)butdoesnotspecifywhathappenstothecontainer
(thewatermightfilltheglass,merelywetit,endupbesideit,andsoon).Inbothcases,theentityspecified
as"affected"endsupastheobject,butforfill,itistheobjectwhosestateisaffected(goingfromnotfullto
full),whereasforpour,itistheobjectwhoselocationisaffected(goingfromoneplacetoalowerone).
Now,let'ssaychildrenmistakenlythinkthatfillreferstoamannerofmotion(presumably,somekindof
tippingorpouring),insteadofanendstateoffullness.(Childrencommonlyuseendstateverbsasmanner
verbs:forexample,theythinkthatmixjustmeansstir,regardlessofwhetherthestirredingredientsendup
mixedtogetherGentner,1978).Ifso,thelinkingrulefordirectobjectswouldcausethemtomaketheerror
weobserve:fillxintoy.Howcouldtheyrecover?Whenchildrenobservetheverbfillinenoughcontextsto
realizethatitactuallyencodestheendstateoffullness,notamannerofpouringoranyotherparticular
manner(forexampleeventuallytheymayhearsomeonetalkingaboutfillingaglassbyleavingitona
windowsillduringastorm),theycanchangetheirmentaldictionaryentryforfill.Asaresult,theywould
withdrawitfromeligibilitytotaketheargumentstructurewiththecontentsasdirectobject,onthegrounds
thatitviolatestheconstraintthat"directobject=specificallyaffectedentity."Theprinciplecouldhave
existedallalong,butonlybeendeemedrelevanttotheverbfillwhenmoreinformationaboutitsdefinition
hadbeenaccumulated(Gropen,etal.,1991a,bPinker,1989).
Thereisevidencethattheprocessworksinjustthatway.Gropenetal.(1991a)askedpreschoolchildrento
selectwhichpicturecorrespondedtothesentenceShefilledtheglasswithwater.Mostchildren
indiscriminatelychoseanypictureshowingwaterpouringtheydidnotcarewhethertheglassendedupfull.
Thisshowsthattheydomisconstruethemeaningoffill.Inaseparatetask,thechildrenwereaskedto
describeintheirownwordswhatwashappeninginapictureshowingaglassbeingfilled.Manyofthese
childrenusedincorrectsentenceslikeHe'sfillingwaterintotheglass.Olderchildrentendedtomakefewer
errorsofbothverbmeaningandverbsyntax,andchildrenwhogottheverbmeaningrightwerelesslikelyto
makesyntaxerrorsandviceversa.Inanevenmoredirectdemonstration,Gropen,etal.(1991b)taught
childrennewverbsliketopilk,referringtoactionslikemovingaspongeovertoacloth.Forsomechildren,
themotionhadadistinctivezigzagmanner,buttheclothremainedunchanged.Forothers,themotionwas
nondescript,buttheclothchangedcolorinalitmuslikereactionwhenthespongeendeduponit.Though
noneofthechildrenheardtheverbusedinasentence,whenaskedtodescribetheevent,thefirstgroupsaid
thattheexperimenterwaspilkingthesponge,whereasthesecondgroupsaidthathewaspilkingthecloth.
Thisisjustthekindofinferencethatwouldcauseachildwhofinallyfiguredoutwhatfillmeanstostop
usingitwiththewrongdirectobject.
Interestingly,theconnectionsbetweenverbs'syntaxandsemanticsgobothways.Gleitman(1990)pointsout
thattherearesomeaspectsofaverb'smeaningthataredifficult,ifnotimpossible,forachildtolearnby
observingonlythesituationsinwhichtheverbisused.Forexample,verbpairslikepushandmove,giveand
receive,winandbeat,buyandsell,chaseandflee,anddropandfalloftencanbeusedtodescribethesame
eventonlytheperspectiveassumedbytheverbdiffers.Also,mentalverbslikesee,know,andwant,are
difficulttoinferbymerelyobservingtheircontexts.Gleitmansuggeststhatthecrucialmissinginformation
comesfromthesyntaxofthesentence.Forexample,fallisintransitive(itfell,notJohnfelltheball)drop
canbetransitive(Hedroppedtheball).Thisreflectsthefactthatthemeaningoffallinvolvesthemereactof
plummeting,independentofwhoifanyonecausedit,whereastheextraargumentofdropreferstoanagent
whoiscausingthedescent.Achildcouldfigureoutthemeaningdifferencebetweenthetwobypaying
attentiontothetransitiveandintransitivesyntaxanexampleofusingsyntaxtolearnsemantics,ratherthan
viceversa.(Ofcourse,itcanonlyworkifthechildhasacquiredsomesyntaxtobeginwith.)Similarly,a
verbthatappearswithaclauseasitscomplement(asinIthinkthat...)mustrefertoastateinvolvinga
proposition,andnot,say,ofmotion(thereisnoverblikeHejumpedthathewasintheroom).Thereforea
childhearingaverbappearingwithaclausalcomplementcaninferthatitmightbeamentalverb.
Naigles(1990)conductedanexperimentthatsuggestthatchildrenindeedcanlearnsomeofaverb'smeaning
fromthesyntaxofasentenceitisusedin.Twentyfourmontholdsfirstsawavideoofarabbitpushinga
duckupanddown,whilebothmadelargecircleswithonearm.Onegroupofchildrenheardavoicesaying
"Therabbitisgorpingtheduck"anotherheard"Therabbitandtheduckaregorping."Thenbothgroupssaw
apairofscreens,oneshowingtherabbitpushingtheduckupanddown,neithermakingarmcircles,the
othershowingthetwocharactersmakingarmcircles,neitherpushingdowntheother.Inresponsetothe
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

22/33

8/20/2016

LanguageAcquisition

command"Where'sgorpingnow?Findgorping!",thechildrenwhoheardthetransitivesentencelookedat
thescreenshowingtheupanddownaction,andthechildrenwhoheardtheintransitivesentencelookedat
thescreenshowingthemakingcirclesaction.Forageneraldiscussionofhowchildrencoulduseverbsyntax
tolearnverbsemantics,andviceversa,seePinker(1994b).

9.3ParameterSettingandtheSubsetPrinciple
Astrikingdiscoveryofmoderngenerativegrammaristhatnaturallanguagesseemtobebuiltonthesame
basicplan.Manydifferencesamonglanguagesrepresentnotseparatedesignsbutdifferentsettingsofafew
"parameters"thatallowlanguagestovary,ordifferentchoicesofruletypesfromafairlysmallinventoryof
possibilities.Thenotionofa"parameter"isborrowedfrommathematics.Forexample,alloftheequationsof
theform"y=3x+b,"whengraphed,correspondtoafamilyofparallellineswithaslopeof3theparameter
btakesonadifferentvalueforeachline,andcorrespondstohowhighorlowitisonthegraph.Similarly,
languagesmayhaveparameters(seethechapterbyLasnik).
Forexample,alllanguagesinsomesensehavesubjects,butthereisaparametercorrespondingtowhethera
languageallowsthespeakertoomitthesubjectinatensedsentencewithaninflectedverb.This"null
subject"parameter(sometimescalled"PROdrop")issetto"off"inEnglishand"on"inSpanishandItalian
(Chomsky,1981).InEnglish,onecan'tsayGoestothestore,butinSpanish,onecansaytheequivalent.The
reasonthisdifferenceisa"parameter"ratherthananisolatedfactisthatitpredictsavarietyofmoresubtle
linguisticfacts.Forexample,innullsubjectlanguages,onecanalsousesentenceslikeWhodoyouthink
thatleft?andAteJohntheapple,whichareungrammaticalinEnglish.Thisisbecausetherulesofagrammar
interacttightlyifonethingchanges,itwillhaveseriesofcascadingeffectsthroughoutthegrammar.(For
example,Whodoyouthinkthatleft?isungrammaticalinEnglishbecausethesurfacesubjectofleftisan
inaudible"trace"leftbehindwhentheunderlyingsubject,who,wasmovedtothefrontofthesentence.For
reasonsweneednotcoverhere,atracecannotappearafterawordlikethat,soitspresencetaintsthe
sentence.RecallthatinSpanish,onecandeletesubjects.Therefore,onecandeletethetracesubjectofleft,
justlikeanyothersubject(yes,onecan"delete"amentalsymbolevenitwouldhavemadenosoundtobegin
with).Theistracenolongerthere,sotheprinciplethatdisallowsatraceinthatpositionisnolongerviolated,
andthesentencesoundsfineinSpanish.
Onthisview,thechildwouldsetparametersonthebasisofafewexamplesfromtheparentalinput,andthe
fullcomplexityofalanguagewillensuewhenthoseparameterizedrulesinteractwithoneanotherandwith
universalprinciples.Theparametersettingviewcanhelpexplaintheuniversalityandrapidityofthe
acquisitionoflanguage,despitethearcanecomplexityofwhatisandisnotgrammatical(e.g.,the
ungrammaticalityofWhodoyouthinkthatleft?).Whenchildrenlearnonefactaboutalanguage,theycan
deducethatotherfactsarealsotrueofitwithouthavingtolearnthemonebyone.
Thisraisesthequestionofhowthechildsetstheparameters.Onesuggestionisthatparametersettingsare
ordered,withchildrenassumingaparticularsettingasthedefaultcase,movingtoothersettingsastheinput
evidenceforcesthemto(Chomsky,1981).Buthowwouldtheparametersettingsbeordered?Onevery
generalrationalecomesfromthefactthatchildrenhavenosystematicaccesstonegativeevidence.Thusfor
everycaseinwhichparametersettingAgeneratesasubsetofthesentencesgeneratedbysettingB(asin
diagrams(c)and(d)ofFigure1),thechildmustfirsthypothesizeA,thenabandonitforBonlyifasentence
generatedbyBbutnotbyAwasencounteredintheinput(Pinker,1984Berwick,1985Osherson,etal,
1985).Thechildwouldthenhavenoneedfornegativeevidenceheorshewouldneverguesstoolargea
language.(Forsettingsthatgeneratelanguagesthatintersectoraredisjoint,asindiagrams(a)and(b)of
Figure1,eithersettingcanbediscardedifincorrect,becausethechildwilleventuallyencounterasentence
thatonegrammargeneratesbuttheotherdoesnot).
Muchinterestingresearchinlanguageacquisitionhingesonwhetherchildren'sfirstguessfromamongaset
ofnestedpossiblelanguagesreallyisthesmallestsubset.Forexample,somelanguages,likeEnglish,
mandatestrictwordordersothers,suchasRussianorJapanese,listasmallsetofadmissibleordersstill
others,suchastheAustralianaboriginelanguageWarlpiri,allowalmosttotalscramblingofwordorder
withinaclause.Wordorderfreedomthusseemstobeaparameterofvariation,andthesettinggeneratingthe
smallestlanguagewouldobviouslybetheonedictatingfixedwordorder.IfchildrenfollowtheSubset
Principle,theyshouldassume,bydefault,thatlanguageshaveafixedconstituentorder.Theywouldbackoff
fromthatpredictionifandonlyiftheyhearalternativewordorders,whichindicatethatthelanguagedoes
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

23/33

8/20/2016

LanguageAcquisition

permitconstituentorderfreedom.Thealternativeisthatthechildcouldassumethatthedefaultcasewas
constituentorderfreedom.
Iffixedorderisindeedthedefault,childrenshouldmakefewwordordererrorsforafixedorderlanguage
likeEnglish,andmightbeconservativeinlearningfreerwordorderlanguages,stickingwithasubsetofthe
sanctionedorders(whethertheyinfactareconservativewoulddependonhowmuchevidenceofmultiple
orderstheyneedbeforeleapingtotheconclusionthatmultipleordersarepermissible,andonhowfrequentin
parentalspeechthevariousordersare).If,ontheotherhand,freeorderisthedefault,childrenacquiring
fixedwordorderlanguagesmightgothroughastageofovergenerating(saying,givedoggiepapergive
paperdoggie,paperdoggiegivedoggiepapergive,andsoon),whilechildrenacquiringfreewordorder
languageswouldimmediatelybeabletousealltheorders.Infact,asIhavementioned,childrenlearning
Englishneverleaptotheconclusionthatitisafreewordorderlanguageandspeakinallorders(Brown,
1973Braine,1976Pinker,1984Bloom,Lightbown,&Hood,1975).Logicallyspeaking,though,that
wouldbeconsistentwithwhattheyheariftheywerewillingtoentertainthepossibilitythattheirparents
werejustconservativespeakersofKorean,RussianorSwedish,whereseveralordersarepossible.But
childrenlearningKorean,Russian,andSwedishdosometimes(thoughnotalways)erronthesideofcaution,
anduseonlyoneoftheordersallowedinthelanguage,pendingfurtherevidence(Brown,1973).Itlookslike
fixedorderisthedefault,justastheSubsetPrinciplewouldpredict.
Wexler&Manzini(1987)presentaparticularlyniceexampleconcerningthedifferencebetween"anaphors"
likeherselfand"pronouns"likeher.Ananaphorhastobehaveitsantecedentlieasmalldistanceaway
(measuredintermsofphrasesize,ofcourse,notnumberofwords)theantecedentissaidtobeinsidethe
anaphor's"governingcategory."ThatiswhythesentenceJohnlikedhimselfisfine,butJohnthoughtthat
Marylikedhimselfisungrammatical:himselfneedsanantecedent(likeJohn)withinthesameclauseas
itself,whichithasinthefirstexamplebutnotthesecond.Differentlanguagespermitdifferentsize
governingcategoriesfortheequivalentsofanaphorslikehimselfinsomelanguages,thetranslationsofboth
sentencesaregrammatical.TheSubsetPrinciplepredictsthatchildrenshouldstartoffassumingthattheir
languagerequiresthetiniestpossiblegoverningcategoryforanaphors,andthentoexpandthepossibilities
outwardastheyhearthetelltalesentences.Interestingly,forpronounslike"her,"theorderingispredictedto
betheopposite.Pronounsmaynothaveanantecedentwithintheirgoverningcategories:Johnlikedhim
(meaningJohnlikedhimself]isungrammatical,becausetheantecedentofhimistooclose,butJohnthought
thatMarylikedhimisfine.Setsoflanguageswithbiggerandbiggergoverningcategoriesforpronouns
allowfewerandfewergrammaticalpossibilities,becausetheydefinelargerrangesinwhichapronoun
prohibitsitsantecedentfromappearinganeffectofcategorysizeonlanguagesizethatisintheopposite
directiontowhathappensforanaphors.WexlerandManzinithuspredictthatforpronouns,childrenshould
startoffassumingthattheirlanguagerequiresthelargestpossiblegoverningcategory,andthentoshrinkthe
possibilitiesinwardastheyhearthetelltalesentences.Theyreviewexperimentsandspontaneousspeech
studiesthatprovidesomesupportforthissubtlepatternofpredictions.

10Conclusion
Thetopicoflanguageacquisitionimplicatethemostprofoundquestionsaboutourunderstandingofthe
humanmind,anditssubjectmatter,thespeechofchildren,isendlesslyfascinating.Buttheattemptto
understanditscientificallyisguaranteedtobringonacertaindegreeoffrustration.Languagesarecomplex
combinationsofelegantprinciplesandhistoricalaccidents.Wecannotdesignnewoneswithindependent
propertieswearestuckwiththeconfoundedonesentrenchedincommunities.Children,too,werenot
designedforthebenefitofpsychologists:theircognitive,social,perceptual,andmotorskillsareall
developingatthesametimeastheirlinguisticsystemsarematuringandtheirknowledgeofaparticular
languageisincreasing,andnoneoftheirbehaviorreflectsoneofthesecomponentsactinginisolation.
Giventheseproblems,itmaybesurprisingthatwehavelearnedanythingaboutlanguageacquisitionatall,
butwehave.Whenwehave,Ibelieve,itisonlybecauseadiversesetofconceptualandmethodologicaltools
hasbeenusedtotraptheelusiveanswerstoourquestions:neurobiology,ethology,linguistictheory,
naturalisticandexperimentalchildpsychology,cognitivepsychology,philosophyofinduction,theoretical
andappliedcomputerscience.Languageacquisition,then,isoneofthebestexamplesoftheindispensability
ofthemultidisciplinaryapproachcalledcognitivescience.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

24/33

8/20/2016

LanguageAcquisition

11FurtherReading
AgeneralintroductiontolanguagecanbefoundinmybookTheLanguageInstinct(Pinker,1994),from
whichseveralportionsofthischapterwereadapted.Thereisachapteronlanguageacquisition,andchapters
onsyntacticstructure,wordstructure,universalsandchange,prescriptivegrammar,neurologyandgenetics,
andothertopics.
ThelogicalproblemoflanguageacquisitionisdiscussedindetailbyWexlerandCulicover(1980),Pinker
(1979,1984,1987,1989),Osherson,Stob,&Weinstein(1985),Berwick(1985),andMorgan(1986).Pinker
(1979)isanontechnicalintroduction.Thestudyoflearnabilitywithintheoreticalcomputersciencehas
recentlytakenoninterestingnewturns,reviewedinKearns&Vazirani(1994),thoughwithlittlediscussion
ofthespecialcaseweareinterestedin,languageacquisition.Brent(1995)containsstateoftheartworkon
computermodelsoflanguageacquisition.
ThemostcomprehensiverecenttextbookonlanguagedevelopmentisIngram(1989).Amongotherrecent
textbooks,Gleason(1993)hasafocusonchildren'sandmothers'behavior,whereasAtkinson(1992),
Goodluck(1991),andCrainandLilloMartin(inpress),havemoreofafocusonlinguistictheory.Bloom
(1993)isanexcellentcollectionofreprintedarticles,organizedaroundtheacquisitionofwordsand
grammar.HoekstraandSchwartz(1994)isacollectionofrecentpapersmorecloselytiedtotheoriesof
generativegrammar.Fletcher&MacWhinney'sTheHandbookofChildLanguage(1995),hasmanyuseful
surveychaptersseealsothesurveysbyPaulBloominGernsbacher'sHandbookofPsycholinguistics(1994)
andbyMichaelMaratsosinMussen'sCarmichael'sManualofChildPsychology(4thedition19835th
editioninpreparationatthetimeofthiswriting).
EarliercollectionsofimportantarticlesincludeKrasnegor,etal.,(1991),MacWhinney(1987),Roeper&
Williams(1987),Wanner&Gleitman(1982),Baker&McCarthy(1981),FletcherandGarman(1979),
Ferguson&Slobin(1973),Hayes(1970),Brown&Bellugi(1964),andLenneberg(1964).Slobin
(1985a/1993)isalargecollectionofmajorreviewsontheacquisitionofparticularlanguages.
Themostambitiousattemptstosynthesizelargeamountsofdataonlanguagedevelopmentintoacohesive
frameworkareBrown(1973),Pinker(1984),andSlobin(1985b).Clark(1993)reviewstheacquisitionof
words.Locke(1993)coverstheearlieststagesofacquisition,withafocusonspeechinputandoutput.
Morgan&Demuth(inpress)containspapersonchildren'sperceptionofinputspeechanditsinteractionwith
theirlanguagedevelopment.

12Problems
1."Negativeevidence"isreliableinformationavailabletoalanguagelearneraboutwhichstringsof
wordsareungrammaticalinthelanguagetobeacquired.Whichofthefollowingwould,andwould
not,countasnegativeevidence.Justifyyouranswers.
a.MotherexpressesdisapprovaleverytimeJuniorspeaksungrammatically.
b.FatheroftenrewardsJuniorwhenhespeaksgrammatically,andoftenpunisheshimwhenhespeaks
ungrammatically.
c.MotherwrinkleshernoseeverytimeJuniorspeaksungrammatically,andneverwrinkleshernose
anyothertime.
d.FatherrepeatsallofJunior'sgrammaticalsentencesverbatim,andconvertsallofhisungrammatical
sentencesintogrammaticalones.
e.Motherblathersincessantly,utteringallthegrammaticalsentencesofEnglishinorderoflengthall
thetwowordsentences,thenallthethreewordsentences,andsoon.
f.FathercorrectsJuniorwheneverheproducesanoverregularizationlikebreaked,butnevercorrects
himwhenheproducesacorrectpasttenseformlikebroke.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

25/33

8/20/2016

LanguageAcquisition

g.WheneverJuniorspeaksungrammatically,Motherrespondsbycorrectingthesentencetothe
grammaticalversion.Whenhespeaksgrammatically,Motherrespondswithafollowupthatmerely
recaststhesentenceindifferentwords.
h.WheneverJuniorspeaksungrammatically,Fatherchangesthesubject.
i.MotherneverrepeatsJunior'sungrammaticalsentencesverbatim,butsometimesrepeatshis
grammaticalsentencesverbatim.
j.Fatherblathersincessantly,producingallpossiblestringsofEnglishwords,furrowinghisbrows
aftereveryungrammaticalstringandpursinghislipsaftereverygrammaticalsentence.
2.Considerthreelanguages.LanguageAisisEnglish,inwhichsentencemustcontainagrammatical
subject:HeatetheappleisgoodAtetheappleisungrammatical.InLanguageB,thesubjectis
optional,buttheverbalwayshasasuffixwhichagreeswiththesubject(whetheritispresentor
absent)inperson,number,andgender.ThusHeate3MStheappleisgood(assumethat"3MS"isa
suffix,likeoorik,thatisusedonlywhenthesubjectis3rdpersonmasculinesingular),asisAte
3MStheapple.(ThoseofyouwhospeakSpanishorItalianwillseethatthishypotheticallanguageis
similartothem.)LanguageChasnoinflectionontheverb,butallowsthesubjecttobeomitted:Heate
theappleandAtetheapplearebothgood.Assumingachildhasnoaccesstonegativeevidence,but
knowsthatthelanguagetobelearnedisoneofthesethree.Doesthechildhavetoentertainthese
hypothesesinanyfixedorder?Ifso,whatisit?Whatlearningstrategywouldguaranteethatthechild
wouldarriveatthecorrectlanguage?Showwhy.
3.Imagineaverbpilkthatmeans"tohavebothofone'selbowsgrabbedbysomeoneelse,"soJohn
pilkedBillmeantthatBillgrabbedJohn'selbows.
a.WhyisthisverbunlikelytooccurinEnglish?
b.Ifchildrenusesemanticcontextandsemanticsyntaxlinkingrulestobootstraptheirwayintoa
language,whatwouldalanguagelesschildinferaboutEnglishuponhearing"Thisispilking"and
seeingBillgrabJohn'selbows?
c.Ifchildrenusesemanticcontextandsemanticssyntaxlinkingrulestobootstraptheirwayintoa
language,whatwouldalanguagelesschildinferaboutEnglishuponhearing"JohnpilkedBill"and
seeingBillgrabJohn'selbows?
d.Ifchildrenusesemanticcontextandsemanticssyntaxlinkingrulestobootstraptheirwayintoa
language,whatwouldachildhavetoexperienceinordertolearnEnglishsyntaxandthecorrectuseof
thewordpilk?

13AnswerstoProblems
1.a.No.PresumablyMotheralsoexpressesdisapprovalforotherreasons,suchasJuniorutteringarude
orfalsebutgrammaticalsentence.IfJuniorweretoassumethatdisapprovedofsentenceswere
ungrammatical,hewouldspuriouslyeliminatemanygrammaticalsentencesfromhislanguage.
b.No,becauseFathermayalsorewardhimwhenhespeaksungrammaticallyandpunishhimwhenhe
speaksgrammatically.
c.Yes,becauseJuniorcandeducethatanynosewrinkleelicitingsentenceisgrammatical.
d.Yes,becauseJuniorcandeducethatanysentencethatisnotrepeatedverbatimisungrammatical.
e.Yes,becauseforanysentencethatJuniorisunsureabout,hecankeeplisteningtomotheruntilshe
beginstouttersentenceslongerthanthatone.If,bythattime,Motherhasutteredhissentence,itis
grammaticalifshehasn't,it'sungrammatical.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

26/33

8/20/2016

LanguageAcquisition

f.No,becausewedon'tknowwhatFatherdoesfortherestofthelanguage.
g.No,becausewhileweknowwhetherthechangeoverinJunior'ssentenceisa"correction"ora
"recasting,"becauseweknowwhat'sungrammatical(hencecorrected)orgrammatical(hencerecast),
Juniorhasnowayofknowingthatfromhispointofview,Motherjustchangeseverythinghesaysinto
differentwords.
h.No,becausepresumablyFatherchangesthesubjectonsomeoccasionswhenJunior'ssentencewas
grammaticalbutFatherwasjustgettingboredwiththetopic.
i.No,becausemanyofhisgrammaticalsentencesmightneverberepeatedverbatim,either.
j.Yes,becausesoonerorlaterFatherwillutterJunior'slastwordstring,andJuniorcanseewhether
Father'sbrowwasfurrowed.
2.English(LanguageA)hastobehypothesizedbeforeLanguageC,andrejectedonlyifasubjectlessand
suffixlesssentenceturnsupintheinput.ThatisbecauseLanguageCisasupersetofEnglishifthe
learnertriesCfirst,nothingintheinputwillevertellhimhe'swrong.LanguageBcanbehypothesized
atanypoint,andconfirmedwheneverthechildhearsasentencewithanagreementinitor
disconfirmedwhenthechildhearsasentencewithoutagreement.
3.a.InEnglish(andalmosteveryotherlanguage),theagentoftheactionisthesubjectofanactive
sentence,andtheentityaffectedbytheactionistheobject.
b.Hewouldinfer,incorrectly,thatpilkmeans"toholdsomeone'selbows."
c.Hewouldinfer,incorrectly,thatEnglishwordorderwasObjectVerb
Subject.Thatwouldcausehimsubsequentlytoapplyuniversalsaboutsubjectstoobjects,andvice
versa.
d.HewouldhavetohaveheardenoughordinaryEnglishverbs(withagentsassubjectsandaffected
entitiesasobjects)tohaveinferredthatthesubjectcomesbeforetheverb,whichinturncomesbefore
theobject.ThenhewouldhavetohearJohnpilkedBillandseeBillgrabJohn'selbows,andusethe
verb'ssyntaxtoinferitsunusualsemantics.

References
Anderson,J.(1977)Inductionofaugmentedtransitionnetworks.CognitiveScience,1,125157.
Bates,E.(1989)Functionalismandthecompetitionmodel.InB.MacWhinneyandE.Bates(Eds.),The
crosslinguisticstudyofsentenceprocessing.NewYork:CambridgeUniversityPress.
Bates,E.,Thal,D.,&Janowsky,J.S.(1992)Earlylanguagedevelopmentanditsneuralcorrelates.InI.
Rapin&S.Segalowitz(Eds.),HandbookofNeuropsychology.Vol.6,ChildNeurology.Amsterdam:
Elsevier.
Bellugi,U.,Bihrle,A.,Jernigan,T.,Trauner,D.,&Doherty,S.(1990)Neuropsychological,neurological,and
neuroanatomicalprofileofWilliamsSyndrome.AmericanJournalofMedicalGeneticsSupplement,6,115
125.
Berwick,R.C.(1985)Theacquisitionofsyntacticknowledge.Cambridge,MA:MITPress.
Bickerton,D.(1984)Thelanguagebioprogramhypothesis.BehavioralandBrainSciences,7,173221.
Bloom,L.(1970)LanguageDevelopment:FormandFunctioninEmergingGrammars.
Cambridge,MA:MITPress.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

27/33

8/20/2016

LanguageAcquisition

Bloom,L.,Lightbown,P.,&Hood,M.(1975)Structureandvariationinchildlanguage.Monographsofthe
SocietyforResearchinChildDevelopment,vol.40.
Bloom,P.(inpress)Bohannon,J.N.,andStanowicz,L.(1988)Theissueofnegativeevidence:Adult
responsestochildren'slanguageerrors.DevelopmentalPsychology24:684689.
Bowerman,M.(1982).Evaluatingcompetinglinguisticmodelswithlanguageacquisitiondata:Implications
ofdevelopmentalerrorswithcausativeverbs.QuadernidiSemantica3:566.
Bowerman,M.(1982b)Reorganizationalprocessesinlexicalandsyntacticdevelopment.InE.Wanner&L.
Gleitman(eds.),LanguageAcquisition:TheStateoftheArt.NewYork:CambridgeUniversityPress.
Braine,M.D.S.(1976)Children'sfirstwordcombinations.MonographsoftheSocietyforResearchinChild
Development,41.
Brent,M.(Ed.)(inpress)SpecialissueofCognitiononComputationalModelsofLanguageAcquisition.
Brown,R.&Bellugi,U.(Eds.)(1964)SpecialissueofHarvardEducationalReview.
Brown,R.,&Hanlon,C.(1970).Derivationalcomplexityandorderofacquisitioninchildspeech.InJ.R.
Hayes(Ed.),CognitionandtheDevelopmentofLanguage.NewYork:Wiley.
Brown,R.(1973)AFirstLanguage:theEarlyStages.Cambridge,Mass.:HarvardUniversityPress.
Chomsky,C.(1969)AcquisitionofSyntaxinChildrenfrom510.Cambridge,MA:MITPress.
Chomsky,N.(1959)AReviewofB.F.Skinner's"VerbalBehavior."Language,35,2658.
Chomsky,N,(1975)ReflectionsonLanguage.NewYork:RandomHouse.
Chomsky,N.(1981)LecturesonGovernmentandBinding.Dordrecht,Netherlands:ForisPublications.
Chomsky,N.(1991)Linguisticsandcognitivescience:Problemsandmysteries.InA.Kasher,(Ed.),The
Chomskyanturn.Cambridge,MA:Blackwell.
Clark,E.V.(1993)Thelexiconinacquisition.NewYork:CambridgeUniversityPress.
Comrie,B.(1981)Languageuniversalsandlinguistictypology.Chicago:UniversityofChicagoPress.
Cooper,W.E.andPacciaCooper,J.(1980)SyntaxandSpeech.Cambridge,Mass.:HarvardUniversity
Press.
Crain,S.(1992)Languageacquisitionintheabsenceofexperience.BehavioralandBrainSciences.
Crain,S.,&LilloMartin,D.(inpress)Languageacquisition.Cambridge,MA:Blackwell.
Cromer,R.F.(1992)Languageandthoughtinnormalandhandicappedchildren.
Cambridge,MA:Blackwell.
Curtiss,S.(1989)Theindependenceandtaskspecificityoflanguage.InA.Bornstein&J.Bruner(Eds.),
Interactioninhumandevelopment.Hillsdale,NJ:Erlbaum.
Demetras,M.J.,Post,K.N,&Snow,C.E.(1986)Feedbacktofirstlanguagelearners:Theroleof
repetitionsandclarificationquestions.JournalofChildLanguage,13,275292.
ErvinTripp,S.(1973)Somestrategiesforthefirsttwoyears.InT.InT.E.Moore.(Ed.),Cognitive
DevelopmentandtheAcquisitionofLanguage.NewYork:AcademicPress.

http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

28/33

8/20/2016

LanguageAcquisition

Ferguson,C.andSlobin,D.I.,(Eds.)(1973),StudiesofChildLanguageDevelopment.NewYork:Holt,
RinehartandWinston.
Fernald,A.(1984)theperceptualandaffectivesalienceofmothers'speechtoinfants.InL.Feagans,C.
Garvey,&R.Golinkoff(Eds.),Theoriginsandgrowthofcommunication.Norwood,NJ:Ablex.
Fernald,A.(1992)Humanmaternalvocalizationstoinfantsasbiologicallyrelevantsignals:Anevolutionary
perspective.InBarkow,etal.InJ.H.Barkow,L.Cosmides,&J.Tooby(Eds.),TheAdaptedMind:
EvolutionaryPsychologyandtheGenerationofCulture.NewYork:OxfordUniversityPress.
Fernald,A.,&McRoberts,G.(inpress)Prosodicbootstrapping:Acriticalanalysisoftheargumentandthe
evidence.InJ.L.Morgan&K.Demuth,Eds.Signaltosyntax.Hillsdale,NJ:Erlbaum.
Fisher,C.,Hall,G.,Rakowitz,S.,&Gleitman,L.R.(1991)Whenitisbettertoreceivethantogive:
Syntacticandconceptualconstraintsonvocabularygrowth.Unpublishedmanuscript,Departmentof
Psychology,UniversityofIllinois,Urbana.
FletcherandM.Garman,(Eds.)(1979)LanguageAcquisition.NewYork:CambridgeUniversityPress.
Fletcher,P.andMacWhinney,B.(Eds.)(1995)Thehandbookofchildlanguage.Cambridge,MA:Blackwell.
Fodor,J.A.(1975)TheLanguageofThought.NewYork:T.Y.Crowell.
Fodor,J.A.(1983)Modularityofmind.Cambridge,MA:BradfordBooks/MITPress.
Gardner,R.A.&Gardner,B.T.(1969)Teachingsignlanguagetoachimpanzee.Science,165,664672.
Gentner,D.(1978)Onrelationalmeaning:Theacquisitionofverbmeaning.ChildDevelopment49:988
998.
Gernsbacher,M.A.(Ed.)(1994)Handbookofpsycholinguistics.SanDiego:AcademicPress.
Gleason,J.Berko(Ed.)(1993)Thedevelopmentoflanguage,3rdedition.NewYork:Macmillan.
Gleitman,L.R.andWanner,E.(1984)Richlyspecifiedinputtolanguagelearning.InO.Selfridge,E.L.
Rissland,&M.Arbib(Eds.),Adaptivecontrolofilldefinedsystems.NewYork:Plenum.
Gold,E.(1967)Languageidentificationinthelimit.InformationandControl,10,447474.
Goodluck,H.(1991)LanguageAcquisition:alinguisticintroduction.Cambridge,MA:Blackwell.
Greenberg,J.(Ed.)(1978)Universalsofhumanlanguage.Vol.4:Syntax.Stanford,CA:StanfordUniversity
Press.
Grimshaw,J.(1981)Form,function,andthelanguageacquisitiondevice.InC.L.BakerandJ.McCarthy
(eds.),TheLogicalProblemofLanguageAcquisition.Cambridge,MA:MITPress.
Gropen,J.Pinker,S.,Hollander,M.,&Goldberg,R.(1991a)Syntaxandsemanticsintheacquisitionof
locativeverbs.JournalofChildLanguage,18,115151.
Gropen,J.,Pinker,S.,Hollander,M.,&Goldberg,R.(1991b)Affectednessanddirectobjects:Theroleof
lexicalsemanticsintheacquisitionofverbargumentstructure.Cognition,41,153195.
Hayes,J.R.(Ed.),CognitionandtheDevelopmentofLanguage.NewYork:Wiley.
Heath,S.B.(1983)Wayswithwords:Language,life,andworkincommunitiesandclassrooms.NewYork:
CambridgeUniversityPress.
HirshPasek,K.,Nelson,D.G.N.,Jusczyk,P.W.,Cassidy,K.W.,Druss,B.,&Kennedy,L.(1987)Clauses
areperceptualunitsforyounginfants.Cognition,26,269286.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

29/33

8/20/2016

LanguageAcquisition

HirshPasek,K.,Treiman,R.,&Schneiderman,M.(1984).BrownandHanlonrevisited:Mothers'sensitivity
toungrammaticalforms.JournalofChildLanguage,11,8188.
HirshPasek,K.,&Golinkoff,R.M.(1991)Languagecomprehension:Anewlookatsomeoldthemes.InN.
Krasnegor,D.M.Rumbaugh,R.L.Schiefelbusch,&M.StuddertKennedy(Eds.)(1991)Biologicaland
behavioraldeterminantsoflanguagedevelopment.HIllsdale,NJ:Erlbaum.
Hoekstra,T.&Schwartz,B.(Eds.)(1994)Languageacquisitionstudiesingenerativegrammar.Philadelphia:
JohnBenjamins.
Huttenlocher,P.R.(1990)Morphometricstudyofhumancerebralcortexdevelopment.Neuropsychologia,
28,517527.
Ingram.,D.(1989)Firstlanguageacquisition:Method,description,andexplanation.NewYork:Cambridge
UniversityPress.
Kearns,M.J.,&Vazirani,U.V.(1994).AnIntroductiontoComputationalLearningTheory.Cambridge,
MA:MITPress.
Kegl,J.(1994)TheNicaraguanSignLanguageProject:anoverview.Signpost,7,3239.
Kiparsky,P.(1982)Lexicalphonologyandmorphology.InI.S.Yang(Ed.),Linguisticsinthemorningcalm.
Seoul:Hansin,pp.391.
Krasnegor,N.A.,Rumbaugh,D.M.,Schiefelbusch,R.L.,&StuddertKennedy,M.(eds.)(1991)Biological
andbehavioraldeterminantsoflanguagedevelopment.HIllsdale,NJ:Erlbaum.
Kuhl,P.,Williams,K.A.,Lacerda,F.,Stevens,K.N.,&Lindblom,B.(1992)Linguisticexperiencealters
phoneticperceptionininfantsbysixmonthsofage.Science,255,606608.
Labov,W.(1969)ThelogicofnonstandardEnglish.GeorgetownMonographsonLanguageandLinguistics,
22,131.
Landau,B.&Gleitman,L.R.(1985)Languageandexperience.Cambridge,MA:HarvardUniversityPress.
Lenneberg,E.H.(Ed.)(1984)Newdirectionsinthestudyoflanguage.Cambridge,MA:MITPress.
Lenneberg,E.H.(1967)BiologicalFoundationsofLanguage.NewYork:Wiley.
Levy,Y.(1983)It'sfrogsallthewaydown.Cognition,15,7593.
Lieberman,P.(1984)Thebiologyandevolutionoflanguage.Cambridge,MA:HarvardUniversityPress.
Limber,J.(1973)Thegenesisofcomplexsentences.InT.E.Moore.,CognitiveDevelopmentandthe
AcquisitionofLanguage.NewYork:AcademicPress.
Locke,J.L.(1992)Structureandstimulationintheontogenyofspokenlanguage.Developmental
Psychobiology,28,430440.
Locke,J.L.(1993)Thechild'spathtospokenlanguage.Cambridge,MA:HarvardUniversityPress.
MacWhinney,B.&Snow,C.(1985)TheChildLanguageDataExchangeSystem.JournalofChild
Language,12,271296.
MacWhinney,B.&Snow,C.(1990)TheChildLanguageDataExchangeSystem:Anupdate.Journalof
ChildLanguage,17,457472.
MacWhinney,B.(Ed.)(1987)Mechanismsoflanguageacquisition.Hillsdale,NJ:Erlbaum.

http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

30/33

8/20/2016

LanguageAcquisition

MacWhinney,B.(1991)TheCHILDESProject:Computationaltoolsforanalyzingtalk.Hillsdale,NJ:
Erlbaum.
Macnamara,J.(1972)Cognitivebasisoflanguagelearningininfants.PsychologicalReview79,113.
Macnamara,J.(1982)NamesforThings:aStudyofChildLanguage.Cambridge,Mass.:Bradford
Books/MITPress.
Maratsos,M.P.(1974a)Howpreschoolchildrenunderstandmissingcomplementsubjects.Child
Development,45,700706.
Maratsos,M.P.,andChalkley,M.(1981)Theinternallanguageofchildren'ssyntax:theontogenesisand
representationofsyntacticcategories.InK.Nelson,ed.,Children'sLanguagevol.2.NewYork:Gardner
Press.
Marcus,G.F.(1993)Negativeevidenceinlanguageacquisition.Cognition,46],5385.
Marcus,G.F.,Pinker,S.,Ullman,M.,Hollander,M.,Rosen,T.J.,&Xu,F.(1992)Overregularizationin
languageacquisition.MonographsoftheSocietyforResearchinChildDevelopment,57.
Miller,G.A.(1991)Thescienceofwords.NewYork:W.H.Freeman.
Morgan,J.L.(1986)Fromsimpleinputtocomplexgrammar.Cambridge,MA:BradfordBooks/MITPress.
Morgan,J.L.,&Demuth,K.,(Eds.)(inpress).Signaltosyntax.Hillsdale,NJ:Erlbaum.
Mussen,P.ed.,Carmichael'sManualofChildPsychology,4thed.NewYork:Wiley.
Newport,E.,Gleitman,H.&Gleitman,E.(1977)MotherI'dratherdoitmyself:Someeffectsandnon
effectsofmaternalspeechstyle.InC.E.SnowandC.A.Ferguson(eds.),TalkingtoChildren:Language
InputandAcquisition.Cambridge:CambridgeUniversityPress.
Osherson,D.N.,Stob,M.&Weinstein,S.(1985).Systemsthatlearn.Cambridge,MA:BradfordBooks/MIT
Press.
Penner,S.(1987)Parentalresponsestogrammaticalandungrammaticalchildutterances.ChildDevelopment
58:376384.
Piaget,J.(1926)Thelanguageandthoughtofthechild.NewYork:Routledge&KeganPaul.
Pinker,S.(1979)Formalmodelsoflanguagelearning.Cognition,7,217283.
Pinker,S.(1984).Languagelearnabilityandlanguagedevelopment.Cambridge,MA:HarvardUniversity
Press.(a)
Pinker,S.(1987)Thebootstrappingprobleminlanguageacquisition.InB.MacWhinney(Ed.),Mechanisms
oflanguageacquisition.Hillsdale,NJ:Erlbaum.
Pinker,S.(1989)Learnabilityandcognition:Theacquisitionofargumentstructure.Cambridge,MA:MIT
Press.
Pinker,S.(1994a)Thelanguageinstinct.NewYork:Morrow.
Pinker,S.(1994b)Howcouldachilduseverbsyntaxtolearnverbsemantics?Lingua,92,377410.Tobe
reprintedinL.GleitmanandB.Landau(Eds.),Lexicalacquisition.Cambridge,MA:MITPress.
Pinker,S.,&Bloom,P.(1990)Naturallanguageandnaturalselection.BehavioralandBrainSciences,13,
707784.
Premack,D.&Premack,A.J.(1983)Themindofanape.NewYork:Norton.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

31/33

8/20/2016

LanguageAcquisition

Putnam,H.(1971)The"innatenesshypothesis"andexplanatorymodelsinlinguistics.InJ.Searle(Ed.),The
philosophyoflanguage.NewYork:OxfordUniversityPress.
T.Roeper&E.Williams(Eds.),Parametersettingandlanguageacquisition(Reidel,1987).
SavageRumbaugh,E.S.(1991)Languagelearninginthebonobo:Howandwhytheylearn.InN.A.
Krasnegor,D.M.Rumbaugh,R.L.Schiefelbusch,R.L.,&M.StuddertKennedy,(Eds.)Biologicaland
behavioraldeterminantsoflanguagedevelopment.HIllsdale,NJ:Erlbaum.
Schlesinger,I.M.(1971)Productionofutterancesandlanguageacquisition.InD.I.Slobin,ed.,The
OntogenesisofGrammar.NewYork:AcademicPress.
Schieffelin,B.&Eisenberg,A.R.(1981)Culturalvariationinchildren'sconversations.InR.L.
Schiefelbusch&D.D.Bricker(Eds.),Earlylanguage:acquisitionandintervention.Baltimore:University
ParkPress.
Seidenberg,M.S.(1986)Evidencefromgreatapesconcerningthebiologicalbasesoflanguage.InW.
Demopoulos&A.Marras(Eds.),Languagelearningandconceptacquisition.Norwood,NJ:Ablex.
Seidenberg,M.S.&Petitto,L.A.(1979)Signingbehaviorinapes:Acriticalreview.Cognition,7,177215.
Seidenberg,M.S.&Petitto,L.A.(1987)Communication,symboliccommunication,andlanguage:
CommentonSavageRumbaugh,McDonald,Sevcik,Hopkins,andRupert(1986)JournalofExperimental
Psychology:General,116,279287.
Senghas,A.(1994).Nicaragua'slessonsforlanguageacquisition.Signpost,7,3239.
Shopen,T.(Ed.)(1985)Languagetypologyandsyntacticdescription.Vol.II:ComplexConstructions.New
York:CambridgeUniversityPress.
Slobin,D.(1973)Cognitiveprerequisitesforthedevelopmentofgrammar.In.C.FergusonandD.I.Slobin
(ed.),StudiesinChildLanguageDevelopment.
NewYork:Holt,RinehartandWinston.
Slobin,D.I.(1977)Languagechangeinchildhoodandinhistory.InJ.Macnamara,ed.,LanguageLearning
andThought.NewYork:AcademicPress.
Slobin,D.I.(Ed.)(1985a/1992)Thecrosslinguisticstudyoflanguageacquisition.Vols.1&2,1985Vol.3,
1992}.Hillsdale,NJ:Erlbaum.
Slobin,D.I.(1985b)Crosslinguisticevidenceforthelanguagemakingcapacity.InD.I.Slobin(Ed.),The
crosslinguisticstudyoflanguageacquisition.Vol.II:Theoreticalissues.Hillsdale,NJ:Erlbaum.
Snow,C.E.,andFerguson,C.A.(1977)TalkingtoChildren:LanguageInputandAcquisition.Cambridge:
CambridgeUniversityPress.
Steedman,M.(inpress)Phrasalintonationandtheacquisitionofsyntax.InJ.Morgan&K.Demuth,Eds.
Signaltosyntax.Hillsdale,NJ:Erlbaum.
Stromswold,K.(1994)Whatamutechildtellsusaboutlanguage.Unpublishedmanuscript,Rutgers
University.
Tallal,P.,Ross,R.,Curtiss,S..JournalofSpeechandHearingDisorders,[54,167.
Terrace,H.,Petitto,L.A.,Sanders,R.J.&Bever,T.G.(1979)Cananapecreateasentence?Science,206,
891902.
Wanner,E.&GleitmanL.(Eds.),LanguageAcquisition:TheStateoftheArt.NewYork:Cambridge
UniversityPress.
http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

32/33

8/20/2016

LanguageAcquisition

Wexler,K.&Culicover,P.(1980)FormalPrinciplesofLanguageAcquisition.Cambridge,MA:MITPress.
Wexler,K.&Manzini,R.(1987)Parametersandlearnabilityinbindingtheory.InT.Roeper&E.Williams
(Eds.),Parametersandlinguistictheory.Dordrecht:Reidel.
Whorf,B.(1956)Language,thought,andreality.Cambridge,MA:MITPress.
FigureCaption
Foursituationsthatachildcouldbeinwhilelearningalanguage.Eachcirclerepresentsthesetofsentences
constitutingalanguage."H"standsfor"hypothesizedlanguage""T"standsfor"targetlanguage.""+"
indicatesagrammaticalsentenceinthelanguage""indicatesanungrammaticalsentence.

http://users.ecs.soton.ac.uk/harnad/Papers/Py104/pinker.langacq.html

33/33

You might also like