• Keine Ergebnisse gefunden

Pronoun ambiguity resolution in Greek : Evidence from monolingual adults and children

N/A
N/A
Protected

Academic year: 2022

Aktie "Pronoun ambiguity resolution in Greek : Evidence from monolingual adults and children"

Copied!
23
0
0

Wird geladen.... (Jetzt Volltext ansehen)

Volltext

(1)

Pronoun ambiguity resolution in Greek:

Evidence from monolingual adults and children

Despina Papadopoulou

a,

* , Eleni Peristeri

a

, Evagelia Plemenou

b

, Theodoros Marinis

c

, Ianthi Tsimpli

a

aAristotleUniversityofThessaloniki,Greece

bPanteionUniversity,Greece

cUniversityofReading,England,UnitedKingdom

Abstract

Alarge bodyofpsycholinguistic research hasrevealedthat during sentenceinterpretationadultscoordinate multiplesourcesof information.Particularly,theydrawbothonlinguisticpropertiesofthemessageandoninformationfromthecontexttoconstraintheir interpretations.Relativelylittlehoweverisknownabouthowthisintegrativeprocessordevelopsthroughlanguageacquisitionandabouthow children processlanguage.Inthisstudy,twoon-linepictureverificationtaskswereusedtoexaminehow1st,2nd and4th/5thgrade monolingualGreekchildrenresolvepronounambiguitiesduringsentenceinterpretationandhowtheirperformancecomparestothatofadults onthesametasks.Specifically,wemanipulatedthetypeofsubjectpronoun,i.e.nullorovert,andexaminedhowthisaffectedparticipants’ preferencesforcompetingantecedents,i.e.inthesubjectorobjectposition.Theresultsrevealedbothsimilaritiesanddifferencesinhow adultsandthevariouschildgroupscomprehendedambiguouspronominalforms.Particularly,althoughadultsandchildrenalikeshowed sensitivitytothedistributionofovertandnullsubjectpronouns,thisdidnotalwaysleadtoconvergentinterpretationpreferences.

Keywords: Pronounresolution;Nullandovertpronominals;Greek;Languagedevelopment

1. Introduction

Overthepast20yearsanaccumulatingbodyofevidenceonsentenceprocessinghasshownthatawiderangeof syntacticandnon-syntacticsourcesofinformationmediatessentenceinterpretation.Forexample,semanticplausibility, syntacticinformation,relevantcontextualinformation,statisticalregularities,andfrequencyoflexicalco-occurrencehaveall been foundtoinformadults’ parsingcommitments(AltmannandSteedman,1988;Pearlmutter andMacDonald,1995;

Taraban and McClelland, 1988; Tanenhaus and Trueswell, 1995; Trueswell, 1996; Trueswell and Tanenhaus, 1994;

Trueswelletal.,1994).

Althoughthereisplentyofevidencethatadultsarecapableofcoordinatingthesetypesofinformation,itisunclearwhether thesameappliestochildren.Recentresearchonthenatureofchildren’ssentenceprocessingabilitieshasattemptedtounify theoriesoflanguageprocessingandlanguagedevelopmentandtoexaminehowtheprocessingsystemdevelopsinchildren (e.g.,ClahsenandFelser,2006;Felseretal.,2003b;Goodluck,1990;GoodluckandTavakolian,1982;McKeeetal.,1993;

PapadopoulouandTsimpli,2005;Trueswelletal.,1999;TylerandMarslen-Wilson,1981).

*Correspondingauthorat:DepartmentofLinguistics,SchoolofPhilology,AristotleUniversityofThessaloniki,Thessaloniki54124,Greece.

E-mailaddress:depapa@lit.auth.gr(D.Papadopoulou).

Konstanzer Online-Publikations-System (KOPS)

URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-2-1n7vyotknc3s35 https://dx.doi.org/10.1016/j.lingua.2014.09.006

(2)

Onephenomenonthatrequiresthecoordinationofdifferenttypesofinformationispronounresolutioninnullsubject languages,wherethechoiceofsubjectpronounform(overtornull)isregulatedbydiscoursefactors.Inthispaper,we present the results of two experiments that are designed to examine how Greek children and adults coordinate grammaticalanddiscoursepragmaticinformationinordertoresolvepronounambiguitiesduringsentenceinterpretation.

ThefocusisontheresolutionofovertandnullsubjectpronounswhosedistributioninGreekisregulatedbydiscourse factors.Ifchildrenhaveacquiredthegrammaticalanddiscoursepragmaticprinciplesofpronoundistribution,thenwe wouldexpecttheirinterpretationpreferencestobethesameasthoseofadultsandtovaryasafunctionofpronounform.

Toexplorethisresearchquestionweemployedaself-pacedlisteningpictureverificationtaskinwhichweinvestigated both the participants’ pronoun resolution preferences and processing patterns as revealed from listening times on particular words/zones of the critical stimuli. Our findings indicate that, whereas adults have an established subject preferencefornullpronominalformsandanobjectpreferenceforovertpronouns,childrenshowadistinctdevelopmental patternfornullandovertpronouns.Inparticular,theresolutionofovertpronounsisadult-likein10-to-11-year-oldchildren, butnotsteadyinyoungerchildgroups,althoughtheytooseemtobesensitivetotheappropriatediscoursefeatureduring sentenceprocessing.Ontheotherhand,nullpronounsrevealaU-shapeddevelopmentwithnoclearresolutionpattern evenattheageof10-to-11years.

Insection2,wefirstgiveabriefoverviewofpronominalsubjectsinGreek,whilecurrentaccountsaswellasempirical data on theinterpretationofnulland overt pronounsareoutlinedin section 3.Insection 4wereview studieson the acquisition of null and overt subjects in null-subject languages. In section 5 we describe the experiments that we conductedandreportontheirresults;first,wedescribetheexperimentinvestigatingnull(section5.1)andsecondlythe one testingovertsubject pronouns(section5.2).Insection6wediscusstheresultsfromboth experimentsand their implicationsforcurrenttheoriesoflanguageprocessinginchildren.Insection7wedrawanumberofconclusionsfromthe evidence provided.

2. DistributionofnullandovertsubjectpronounsinGreek

Greekisanull-subjectlanguage,thatis,itallowsnullsubjectsinfiniteclausesasshown bytheexamplein(1):

(1) Xtes pro ída to Jáni.

yesterday pro saw-PERF-1SG the-ACC Jani-ACC

‘‘YesterdayIsawJohn.’’

However,therepertoireofpronominalformsinGreekconsistsofbothnullandovertpronouns.Thenullpronominaloption isaweakpronounandisconsideredtobethedefaultform,whereastheovertpronominalformisastrongpronounand constitutes the ‘‘marked’’ option (Cardinaletti and Starke, 2001; Montalbetti, 1984). For example, the null and overt pronounsin(2a)and(2b)respectively,thoughambiguous,exhibitspecificpreferenceswithrespecttotheirreference.

Namely,thenullpronounin(2a),beingthedefaultpronominalforminGreek,ispreferablyanchoredtothemostsalient/

prominentreferent,i.e.thesententialsubject/topic,andhenceleadstoanon-shiftedinterpretationforthesubject.Onthe otherhand,theovertpronounin(2b),whenpronouncedasunstressed,markstopicshiftandispreferablyassociatedwith lesssalient/prominententitiesofthediscourse,i.e.non-topicreferents.Noticethatsuchareferentcouldbeeitherthe complement ofthemainclause,tin kopéla,orathirdreferentnotmentionedinthemainclause1:

(2a) O papús milúse dinatá ston egono´ tu o´tan pro djávaze éna vivlío.

the old-man spoke-IMP-3SG loudly to-the grandson his when pro read-PAST-IMP-3SG a book

‘‘Theold-manwasspeakingloudlyto hisgrandsonwhenhewasreadingabook.’’

(2b) I jajá xerétise tin kopéla o´tan aftí pernúse to dro´mo.

the old-lady greeted-PERF-3SG the girl when she crossed-IMP-3SG the street

‘‘Theold-ladygreetedthegirlwhenSHEwascrossingthestreet.’’

Considerfurthertheinterpretationofnullandovertsubjectpronounsin(3).Thenullpronounin(3b)isstraightforwardly anchored to the topic of the previous utterance, o Pétros, introduced in (3a). Notice, however, that the use of an unstressed overt pronoun in (3c)to refer to the previous topic makes the sentence sound awkward. However,(3c) becomes easilyacceptableandacquires additionalreadings,ifthe overtpronounisstressedas in(3d).Namely, the presenceofastressedovertpronounwithahighpitchaccentassociatedwithcontrastivefocusin(3d)impliesthatPeter

1Thethird referentoptionis indeedhighlymarked, asananonymousreviewermentions,and requirescontextual(discourse orvisual) supportingcues.Inourexperimentaltasksweincludedpictureswithathirdreferentasapossibleantecedent.

(3)

greetedMaryandnooneelsedid.If,ontheotherhandthestressedpronouncarriesanaccentlinkedwithcontrastivetopic the speakerassertsthatPetergreetedMarybut someoneelsedidnot.2

(3a) Xerétise o Pétros ti María;

greeted-3SG the-NOM Petros-NOM the-ACC Maria-ACC

‘‘DidPetrosgreet Maria?’’

(3b) Ne, pro tixerétise.

yes pro her-greeted-3SG

‘‘Yes,hegreetedher.’’

(3c) #Ne, afto´s tixerétise.

yes he her-greeted-3SG

‘‘Yes,hegreetedher.’’

(3d) Ne, AFTOS F/CTtixerétise.

yes he her-greeted-3SG

‘‘Yes,HEgreetedher.’’

In thisstudy, wefocus ourdiscussiononnull andunstressedovertsubjectpronounsandouraimistoexperimentally investigate in adults and three childgroups whether theirinterpretationis affected by the salience/prominenceof the antecedent.Giventhecomplexmappingsbetweenreferringexpressionsandantecedents,theinterpretationofsubject pronounsinGreekrequiresthecoordinationofgrammaticalandpragmaticknowledge.Knowledgeofgrammarprovides languageuserswiththeoptionofanovertornullpronounasthesubjectofafiniteclause.Knowledgeofpragmaticsguides thechoiceofpronounformgiventhediscoursecontext.ThemostunderspecifiedformofreferenceavailableinGreek,pro, shouldpickasubjectantecedent,whereasafullyspecifiedformshouldpickalesssalient,non-subjectantecedent.

3. Accountsofpronounambiguityresolution

Pronounresolutioninanullsubjectlanguageisacomplextaskinsofarasitdrawsnotonlyonspecificgrammaticaland pragmaticknowledge,butalsoonpropertiesofageneraland/orlanguage-specificprocessingmechanism.Atheoretical accountthatmakesexplicitpredictionsregardingpronounresolutionpatternsisCenteringTheory(Brennanetal.,1987;

Grosz et al., 1995). According to Centering Theory, pronouns prefer to have antecedents in the subject position.

Furthermore,prominenceofanantecedentisdefinedasalistof‘‘forwardlookingcenters’’,i.e.availablereferents,where thehighestmemberofthelististhemostcompatibleantecedent.Compatibilityisdefinedstructurally,e.g.intermsofw- features,andthematicallyintermsofagentivity(Groszetal.,1995).Itshouldbenotedherethatthediscoursenotionsof salienceandprominencehavebeenconsideredasthekeydeterminantsofanaphoraresolutionbyseveralresearchers despite differencesin theway theyhavebeen defined,e.g. TopicalitybyGivon (1983),Accessibilityby Ariel(1990), Givennessby Gundeletal.(1993),orInformativenessby Allen(1997).3

Anothertheoreticalaccount,whichhasbeenproposedmorerecently,isthePositionofAntecedentStrategy(PAS) postulatedbyCarminati(2002).PASdiffersfromCenteringTheoryinthatitassociatesaparticularstructuralposition,i.e.

Spec IP, with specific discourse properties, as explicitly exemplified below. Carminati (2002) experimentally tested pronounresolutioninanull-subjectlanguage,i.e.Italian,byconductingon-lineandoff-lineexperiments.Thefindingsof theseexperimentsrevealedthatnullandovertsubjectpronounshavedistinctbiasestowardthesyntacticpositionoftheir antecedent.Namely,shefoundthatnullpronounsmanifestapreferencetobelinkedwithsubjectantecedents,whileovert pronounsarebiasedtowardnon-subjectantecedents.TheseresultshavebeenreplicatedmorerecentlybyFiliaci(2010).

Basedonthisevidence,CarminatiproposedthePASinresolvingintra-sententialanaphora,accordingtowhichthenull pronounprefersanantecedentwhichisinthecanonicalsubjectposition,i.e.intheSpecIPposition,whereastheovert pronounprefersanantecedentwhichisnotintheSpecIPposition.Thisstructuralpreferenceisinlinewithadiscourse principlewherebythemoresaliententitiesarereferencedwiththemostreducedform,alongthelinesoftheaccessibility hierarchyproposedbyAriel(1990).Giventhattheentityintroducedbythesubjectisnormallyconsideredthedefaulttopic ofthesentence,hencehighlyaccessibleinmemory,itwillbepickedasthereferentforanullratherthanforanovert pronoun, aslessinformative,lessrigidand moreattenuatedmarkers arelinked withprominententities.Theopposite

2Wethankananonymousreviewerforpointingouttheseinterpretativedifferencesbetweenstressedandunstressedovertsubjectpronouns.

3AnotheraccountthatattemptstocapturethepronounresolutionpatternsistheParallelPreferenceAccount(ChambersandSmyth,1998),which predictsthatpronounspreferantecedentsinaparallelpositionwhencertainconditionsaremet.Thus,subjectpronounsshouldprefersubject antecedentsandobjectpronounsshouldpreferobjectantecedents.However,causalinferencebasedaccounts(Ehrlich,1980;Kehler,2002)argue thatcausalinferencesratherthantheparallelsyntacticpositionofthepronounsandtheirantecedentsbetterexplainpronounresolutionstrategies.

Theexperimentalitemsofthepresentstudydonotmanipulatesuchfactorsand,thus,theseproposalswillnotbefurtherdiscussed.

(4)

patternisexpectedforovertpronouns:beingmoreinformative,morerigidandlessattenuatedmarkers,overtpronouns preferlessprominentdiscourseentitiesasreferents.

The PAS proposed by Carminati (2002) has been further investigated in several Spanish studies on pronoun resolution.Alonso-Ovalleetal.(2002)conductedanoff-linequestionnairestudyinwhichtheparticipantswereaskedto denotethereferentof nulland overtpronominals, whenthesewereprecededbya sentencecontaining twopossible referents,oneinthesubjectandoneintheobjectposition.Thefindingsindicatedthatthenullpronounwaspreferably anchoredwiththesubjectofthepreviousutterance,whilethelinkingoftheovertpronountothesubjectantecedentwas attestedapproximately50%.ThesefindingshavebeenrecentlyreplicatedbyJegerskietal.(2011)aswellasbyKeating etal.(2011).Moreover,Filiaci(2010)conductedtwoself-pacedreadingtasks,oneinItalianandoneinSpanish,using Carminati’sexperimental materials.TheItalianexperimentreproducedthefindings reportedinCarminati (2002).The Spanishresults,however,weresimilartothosereportedin theaforementionedquestionnairestudies.Namely, Filiaci foundprocessingcostassociatedwiththelinkingofnullpronominalstonon-salient/prominentantecedents,whileovert pronounsdidnotexhibitprocessingcostwhenreferringtosalientantecedents.Therefore,theSpanishstudiessuggest thatcross-linguisticdifferencesexistevenamongnullsubjectlanguages.Inparticular,resolutionofovertpronominals displaysaclearbiastowardnon-salientantecedentsinItalian(Carminati,2002;Tsimplietal.,2004),whileinSpanishthey donotseemtocomplywiththediscoursefeatureofsalience/prominence.InastudyonCatalan(MayolandClark,2010), however,thePAShasbeensupportedbybothofflineandonlinedata,eventhoughthedispreferenceofovertpronominals torefertotopicantecedentswasnotasstrong asinthe Italian(Carminati,2002;Filiaci,2010)study.

Tsimplietal.(2004)alsoexploredtheresolutionpreferencesofnullandovertpronounsinItalian.Theirfindingsprovide supportforthedifferentialstatusofthetwopronominalforms,assuggestedbytheaforementionedaccounts(Ariel,1990;

Carminati,2002).Inparticular,Tsimplietal.(2004)investigatedforwardandbackwardanaphoraresolutioninadultnative speakersofItalianbymeansofasentence--picturematchingtask.Participantsheardsentencesconsistingofamainand asubordinateclause;themainclauseintroducedtworeferents,oneinthesubjectandoneintheobjectposition.The subjectofthesubordinateclausewaseitheranulloranovertpronounandparticipantshadtoselectthereferentforthe nullorovertpronounbychoosingoneofthreepictures.Onepicturecorrespondedtothesubjectandonetotheobjectof themainclause.Thethirdpicturedepictedanewreferentwhowasnotintroducedinthesentence.Thefindingsshowed thatin sentenceswithforwardanaphora,whenthesubjectofthesubordinateclausewasanullpronoun, participants preferredtointerpretitasthesubjectofthemainclause.Incontrast,whenthesubjectofthesubordinateclausewasan overtpronoun,participantsstronglypreferredtointerpretitasreferringtothenewreferent.Insentenceswithbackward anaphora,whenthesubjectofthesubordinateclausewasanullpronoun,participantsallowedforeitherthesubjectorthe objectofthemainclauseasapossiblereferent.However,whentheembeddedsubjectwasanovertpronoun,participants showedastrongpreferenceforthematrixobjecttobethereferentofthepronoun.Thesefindingsareinlinewithproposals putforwardbyAriel(1990)andCarminati(2002)onanaphoraresolutioninnullsubjectlanguages,accordingtowhichthe pronoun formandthe prominenceofthecompatible antecedentshaveanimpacton referenceoptions.Furthermore, Tsimplietal.’sresultsresemblethoseattestedinCarminati(2002)andreplicatedbyFiliaci(2010),inthatthetwopronoun formsdisplaydistinctreferencechoices.Nevertheless,inTsimplietal.’sstudythenullsubjectpronounswerenotfoundto berestrictedtothesubjectreferentinbackwardanaphora,incontrasttoCarminati’sresults.Notice,however,thatthis differencebetweenTsimplietal.ontheonehandandCarminati(2002)andFiliaci(2010)ontheothermaybeattributedto thedifferentnatureofthetasksemployedinthetwostudies,eventhoughwecannotprovideastraightforwardexplanation ofthisdiscrepancy.Furthermore,thestudybyTsimplietal.reportsdivergentresultsforforwardandbackwardanaphora, indicatingthattheorderingofthepronounsandtheantecedentsmayhaveanimpactonpronounresolution.Interestingly, the overt pronoun in sentences involvingforward anaphora preferablyrefers to the third referent, rather than to the complementofthemainclause,asisthecaseinbackwardanaphora.Noticethatthethirdreferenthasbeenchosenasa possibleantecedentinveryfew(lessthan10%)instancesofallotherexperimentalconditions;thisisexpected,sincethe activationofthethirdreferent,notbeingpresentinthesententialcontextgiventotheparticipants,iscostly.Itseems, however,thatthe presenceofanovertpronominalinforwardanaphorasentencesmakesthisoptionpreferable.

TurningtostudiesinvestigatingpronounresolutionstrategiesinGreek,Dimitriadis(1996),basedonthetheoretical assumptionsofCenteringtheoryaswellasonGreekcorpus-data,arguesthatthereferenceroutinesattestedwiththetwo typesofpronominals,nullandovert,areinherentintheirlexicalentries.Morespecifically,overtsubjectpronounscannot be construed with the most prominent antecedent of the previous utterance. On the other hand, null pronouns are construedwiththehighestcompatibleantecedentfromtheavailablelistofreferents.This‘‘distributionoflabor’’between null and overt pronounsisnot, however, without exceptions.Althoughit partly relies on the lexicalpropertiesof the pronounform,pragmaticconsiderationsmaycancelthedictatedoption,thus,accountingforthenon-categoricalnatureof anaphoraresolution.HefurtherarguesthattheovertpronominalforminGreekexhibitsastrongbias,inthatitsanchoring withsalient/prominentantecedentsisalmostalwaysexcluded,whilethepreferenceofnullpronounstorefertosalient/

prominent antecedents is not that strong. According to Dimitriadis, the reference of overt pronouns is part of their conventionalizedmeaningandisconstrainedbytheOvertPronounRule,whichdisallowsthedependencybetweenan

(5)

overt pronoun and asalient/prominent antecedent.Thesetheoretical assumptionsare supportedby thecorpus data reportedin Dimitriadis(1996).

Miltsakaki(2001,2007),alsoworkingwithintheframeworkofCenteringtheoryinGreek,providescorpusdataand makessimilarsuggestionsfornullandovertpronouns.Inparticular,shearguesthatnullpronounstendtorefertothe discoursetopic,whileovertpronounsdispreferreferencetothediscoursetopic,assuming,however,thatthesebiasesare equally strong for bothpronominal types.Furthermore, Miltsakaki’s corpusdata (2007)reveal apreference for overt pronominals torefertopatientargumentsappearingbothin preverbalandpostvebral positions.

TheGreekcorpusdataprovidedbyDimitriadis(1996)andMiltsakaki(2001,2007)areinlinewithfindingsfromother nullsubjectlanguagesinthatthetwopronominalformsexhibitdifferentresolutionroutines(Alonso-Ovalleetal.,2002;

Carminati,2002;Filiaci,2010;Jegerskietal.,2011;Keatingetal.,2011;MayolandClark,2010;Tsimplietal.,2004).

Additionally, thepreferenceofovertpronounstobelinkedwithnon-salientantecedentsobservedinItalian(Carminati, 2002;Filiaci,2010;Tsimplietal.,2004)andCatalan(MayolandClark,2010)issupportedbytheGreekdata.However, theovertpronounresolutionpreferencesinGreekdifferfromthoseattestedinSpanish(Alonso-Ovalleetal.,2002;Filiaci, 2010;Jegerskietal.,2011;Keatingetal.,2011),whereovertpronominalsdonotdemonstrateastrongpreferencefornon- salientantecedents.Regardingtheresolutionpatternsfornullpronouns,Miltsakaki’scorpusstudies(2001,2007)indicate aclearpreferenceforsalientantecedent;thisfindingisconvergentwiththeSpanish(Alonso-Ovalleetal.,2002;Filiaci, 2010;Jegerskietal.,2011;Keatingetal.,2011),theCatalan(MayolandClark,2010)andtheItaliandataprovidedby Carminati(2002)andFiliaci(2010).Ontheotherhand,thecorpusdatainDimitriadis’study(1996)denotethatthenull pronominalsinGreekarecompatiblewithbothsalientandnon-salientantecedents;thisresultagreeswiththeItaliandata fromnullpronominalformspresentedinTsimplietal.(2004).

It followsfrom the above that pronoun interpretationin a null-subjectlanguage isdriven by a mechanismthat is constrainedbythespecificgrammaticalanddiscourseprinciplesofthelanguage.Morespecifically,theaccountsandthe studiesreviewedinthissectionindicatethattheanaphorresolutionstrategyemployedbyspeakersintegratesboththe grammaticalpropertiesofthenullandovertpronounsaswellasthediscourse-relatedfeaturesassociatedwiththetwo forms(Carminati,2002;Dimitriadis,1996;Filiaci,2010;Jegerskietal.,2011;Keatingetal.,2011).Furthermore,thedata sofarindicatethattherearecross-linguisticdifferences(theItalian,CatalanandGreekdatavs.theSpanishdataonovert subjectpronounresolution)aswellasintra-languagedifferences(Dimitriadis’svs.Miltsakaki’sstudyandCarminati’sdata vs.Tsimplietal.’sdataonnullsubjectpronounresolutioninGreekandItalianrespectively)inthewaythepronominal formsarepreferablyinterpreted.Inthepresentstudy,weprovide,apartfromthechilddata,Greekadultdataonpronoun resolution,which willaddressthequestion ofwhattypesofinformation,i.e. grammatical,discourse-level,adultnative speakersemploytointerpretambiguouspronominals.Moreover,thepresentdatawillbediscussedinrelationtoprevious data andtheoreticalaccounts.

4. Nullandovertsubjectpronounsinchildlanguage

A growing body of research on how children produce and comprehend subject pronouns investigates whether differencesattestedbetweenadultsandchildrenareattributedtoadelayeduseofdiscoursefactorsinon-linesentence comprehensionandproduction,as opposedtothe earlyuseofpurelygrammaticalcues.

Thelanguagedevelopmentliteratureoverthelasttenyearsfocusedontheinvestigationofchildsentenceprocessing bymeansofvariousmethodologiesandaddressedthequestionofhowsyntactic,semanticanddiscourseinformationis employedbychildrenwhileprocessingtemporarilysyntacticambiguities(self-pacedreadingandlisteningstudies:Felser etal.,2003a,b;PapadopoulouandTsimpli,2005;Traxler,2002;eye-trackingstudies:Trueswelletal.,1999;cross-modal primingtasks:McKeeetal.,1993;word-monitoringtasks:TylerandMarslen-Wilson,1981; Tyler,1983;event-related brain potentials:HahneandFriederici,2001).Whatemergesfromsuchstudiesisthatchildsentenceprocessinghas essentiallytwosidestoit.First,fromalanguagecompetencestandpoint,thequestioniswhetherchildrenhaveinfact acquiredthesemantic,grammatical,andpragmaticknowledgethatunderliescomprehensionandproduction.Second, fromalanguageperformancestandpoint,thequestioniswhethertheL1grammarautomaticallygivesrisetolanguage- specificortodefaultprocessingstrategiesthatareavailablefromtheearlieststagesoflanguageacquisition.

Inpronounresolution,forexample,knowledgeofgrammarwoulddeterminewhetherthelanguageallowsnullsubjects;

knowledge ofpragmaticscould compriseprinciplessuchas thatof‘‘informativeness’’4 (Greenfield andSmith, 1976), wherebyinformationislinguisticallyencodedwhenitpertainstoaspectsofaneventthatcannotbeeasilyrecoveredbyan

4Informativenessisadiscoursepragmaticprinciplewhichprovidesanaccountofargumentdrop(Clancy,1993,1997;GreenfieldandSmith, 1976).Accordingtoinformativeness,argumentrealizationdependsuponanumberofpragmaticfeaturessuchasperson,absence,contrast, query,differentiationindiscourseetc.(Allen,2000,2001).Thehighertheinformativevaluesofthepragmaticfeaturesetassignedtoanargument thelesslikelyitisfortheargumenttobedropped.Incontrastanargumentwithonlyoneinformativefeatureassignedsuchasa3rdperson referentwhoisphysicallypresentandnotnewinthediscourseislikelytobedropped(Serratriceetal.,2004).

(6)

interlocutor,butisomittedwhenthisishighlyaccessible.Inthecaseofnull-subjectlanguages,thisisinstantiatedina positivecorrelationbetweenhighinformativenessandnullsubjectsontheonehand,andbetweenlowinformativeness and overtsubjectsonthe other.

Althoughthefindingsfromthechildsentenceprocessingliteraturedonotclearlydistinguishbetween‘competence’

and‘performance’accountsofthedata,theydosuggestthatitisthelatterratherthantheformerwhichgivesrisetochild-

adultdifferences. Specifically, itappearsthat whenweaker orno effectsof referential factorsarefound in children’s performance,thisisnotdue toalackofknowledgeofpragmatics,butrathertochildren’sinability torapidlyintegrate semantic and pragmaticcues in specific taskswhen interpretingsentences in real-time. Previous workon syntactic disambiguationinchildrenhasdemonstratedthatchildrenareabletorapidlyemploymorphologicalandphrase-structure information online(Papadopoulouand Tsimpli,2005; Papangeli,2010; Papangeliand Marinis,2009; Robertset al., 2007)andexhibitapreferenceforthesimplestsyntacticanalysisdisregardingplausibilityconsiderations(Traxler,2002).

Furthermore,Trueswelletal.(1999)showedthatchildrenwereunabletousecontextualinformationintheresolutionof structuralambiguitiessuchas‘‘Putthefrogonthenapkininthebox’’andratherreliedsolelyonphrasestructurerules.

Morespecifically,childrenanalyzedthepost-verbalPPastheDestinationargumentoftheverbeveninanNP-modifier context,unlikeadultswhowereinfluencedbythetypeofcontextinaccordancewiththeprincipleofReferentialSupport5 (AltmannandSteedman,1988).Moreover,childrenareslowerinparsingthesententialinput(seeClahsenandFelser, 2006 forareview)thanadultsandtheirprocessingstrategiesaremoreaffectedby workingmemorythantheparsing routinesemployedby adults(Felser etal.,2003a,b).Insum,inthestudiespresentedabovechildrenshowedgreater relianceonstructuralpropertiesoftheinputandlessornorelianceondiscoursecues.Itisstillamatterofdebatewhether such findings signal immature pragmatics ora problem withthe integration of lexicaland pragmaticcues in on-line sentence processing(Trueswelletal.,1999; ClahsenandFelser, 2006andreferencestherein).

However,otherstudieshaveillustratedthatchildrenmakeuseofthereferentialcontextinoff-linetasks.Crainand colleagues,forexample,showedthatchildrenevenbytheageoffiveyearsweremorelikelytounderstandandproduce relativeclausemodifierswhenthevisualcontextreferentiallysupportedmodification(Crainetal.,1990;Hamburgerand Crain,1982)orin off-line taskswhentwoequally salientpotentialreferents wereinvolved(Meroni and Crain,2003).

Overall, some studiesindicated that children’s processingrelies more ongrammatical (syntactic and morphological) ratherthansemanticandpragmaticinformation(Felseretal.,2003a,b;Trueswelletal.,1999),althoughtheredoesexist someevidencetothecontrary (e.g.Crainetal., 1990;HamburgerandCrain,1982;MeroniandCrain,2003).

Incontrasttothesefindingsforcomprehension,studiesonproductionhaveshownthatpragmaticcuesareemployed evenbyveryyoungchildren.Studiesontheproductionofsubjectsinnullsubjectlanguagesrevealthattheuseofnullor overtsubjectsisclearlyregulatedbypragmatics.Forexample,inananalysisofspontaneouslyproducedutterancesfrom twoGreekchildrenaged1;9--2;2years,Tsimpli(2005)found thattheformofthesubject producedbythechildrenis affectedbydiscourse-levelinformation,suchasinformationstructure.Namely,nullsubjectswereusedinalmosthalfof the children’s utterances inwhich the referent waseasily recoverable by the context,showing that nullsubjectsare establishedasthedefaultforminthechildgrammar,whileovertsubjectswereregulatedbydiscoursefactors,e.g.focus.

Inthisrespect,afurtherdistinctionbetweenpreverbalandpostverbalsubjectsissuggested:preverbalsubjects,nominal or pronominal,wereusedwhenthe subject wasfocused whereasthe postverbalposition wasused fornon-focused subjects.

Discourse coreferenceseems to beanother factor affecting the formof the referringexpressions. Hickmann and Hendriks(1999)examinedthenarrativesofEnglish,French,ChineseandGermanchildrenagedfrom4to10yearsand foundthatpronominalsweremostfrequentlyusedwhenfunctioningascoreferentialsubjects.Moreover,thistendencyto linkpronounswithunambiguoussubjecttopicsbecamemoreprominentfromtheageof7yearsonwardsatleastfor Englishand Chinesechildren.

In other studies, the form of the subject has been related tothe informativeness features carried by itsreferent.

Serratriceetal.(2004)analyzedtwosetsofmonolingualEnglishandItaliancorporaandabilingualEnglish-Italiancorpus ofchildrenagedfrom1;7to4;7andfoundthatnullargumentsweresystematicallyassociatedwithuninformativerather than informative features (see footnote 4).Interestingly, thispattern was obtained fora null(Italian) and a non-null (English) subjectlanguagealike,indicatingthatindevelopingsystemsthepragmaticprincipleofinformativenesscan evenoverridetheovertsubjectsyntacticconstraint.Thus,inbothEnglishandItalian,thedistributionofovertandnull subjects wasregulated by discourse factors. Ina further studywhich assessed the useof wordorder and morpho- syntacticformtosignalthedistinctionbetweennewvs.giveninformation,Serratrice(2007)replicatedherearlierfindings whereby children use language-specific and pragmatically appropriate referring expressions for the realization of subjects.Furthermore,ArgyriandSorace(2007)investigatedtheproductionandoff-lineacceptabilityofnullandovert

5AccordingtotheprincipleofReferentialSupport,resolutionofambiguousphrasessuchas‘‘Putthefrogonthenapkin...’’willfavortheNP- modificationreadingwhenthecontextprovidesmorethanoneuniquereferent(AltmannandSteedman,1988).

(7)

subjectpronounsinGreekwhenthereisnotopicshiftand,thus,thenullpronounoptionispredictedtobefavoredoverthe overtone.Greekchildrenattheageof7;5--9;7yearsandadultsalwaysproducedanullpronominalinthenotopicshift conditions.In theacceptabilityjudgmenttask,however,eventhoughtheGreekadultsaccepted onlythenullsubject pronoun utterance,the Greekchildrendidnotdo soconsistently.Thisfinding,onthe onehand,indicatesdifferences betweenthetwomodalities,productionandjudgment,and,ontheother,itrevealsdevelopmentaldifferences,inthatthe children,evenattheageof9years,seemtoaccepttheovert subjectpronounforsalientreferents,i.e.topics.

Literature on the processing and interpretation of anaphors suggests that children may be sensitive to different informationcomparedtoadults.Inanearlyon-linestudy,Tyler(1983)examinedhow4-,5-,7-year-oldDutchchildrenand adultsprocessedanaphoricexpressions,fullDPsandpersonalpronouns,andconcludedthatattheageof7yearsthe mechanismsusedbychildrennolongerdifferfromthoseemployedbyadults.Shealsoclaimed thatuptotheageof 5 years the children experienced difficulties in rapidly integrating all necessary cues, i.e. grammatical, lexical and contextual, fortheanchoringoftheanaphoricexpressionstotheantecedent.

Laterstudiesinvestigatedtheimpactofdiscoursefactorsontheinterpretationofpronouns.Forexample,useofcues contributingtodiscourseprominenceofapronoun’spotentialreferent,likefirst-mentionandsubjecthood,hasbeenwell documentedinadults(CorbettandChang,1983;Arnoldetal.,2000),butcontroversialinchildren.Forinstance,Arnold et al.’s(2007)studyrevealedthat5year olds,althoughtheyrelyongrammaticalfeatures, suchasthegender ofthe pronoun,toanchorthepronounwiththeappropriateantecedent,showednofirst-mentionbiaseffectinbothoff-lineand on-line tasks. For example, when the children are orally presented with a sentence which contains two possible antecedents ofthesamegender,Puppy andPandaBear,theydonotshowareliablepreference tothefirst-mention antecedent,Puppy,intheinterpretationofthesubjectpronounhe.

(4) Puppy ishavinglunchwithPandaBear.Hewantssomemilk.

Ontheotherhand,otherstudieshavefoundfirst-mentioneffectsin2--3yearolds(SongandFisher,2005,2007)inoff-line storyinterpretationtasks.Morespecifically,the2--3-year-oldchildrenprefertointerpretthepronounheasreferringtothe first-mentionantecedent,thealligator,asdoolder childrenandadults:

(5) On asunnyday,thealligatorwentoutside.Hewenttothetiger’syard. Whatdid hefind?Hefoundabucket.

Noticethat,eventhoughbothstudiesfocusonthereferentialpropertiesoftheantecedent,i.e.first-mentionbias,they arrive atcontradictory results.Thisdiscrepancy maybe duetothe interminglingof agrammatical,i.e. gender,and a discoursefactorintheformerstudyaswellaswiththeamountandcomplexityofcontextualinformationgiveninthetwo studies.InSongandFisher’sstudy(2007)thesalienceoftheappropriateantecedent,thealligator,increaseswiththeuse oftheco-referentialpronoun,he,inmorethanoneutterance,whileinsentence(5)thedisambiguationofthepronoun reliesontheselectionoftheappropriateantecedentwhichoccursonlyonceintheprecedingsentence.Althoughthe sentencesinourstudycontainasubordinateandamainclauseandinthisrespecttheyaremorecomplexthantheones employedbyArnoldetal.(2007),thesalienceoftheantecedentsisparalleltothisstudyratherthantothestudybySong andFisher(2007),sincethepossibleantecedentsoftheambiguouspronounareintheimmediatelypreviousclauseand theyoccupythesubjectand theobjectpositionasin(4).

Furthermore Sekerinaet al. (2004) conducted eye-tracking experiments in order to investigatethe processing of referringexpressionsinEnglishadultsand4-to7-year-oldchildren.Theirfindingsindicatedthatchildrendifferedfrom adultsinthattheyweresignificantlylesslikelytochooseanextra-sentential,butpossible,antecedentforapronounthan theadults.Additionally,theanalysisofchildren’seyemovementsshowedthatchildrenaresignificantlyslowerthanthe adultsindetecting theambiguity,althoughtheyshowawarenessoftheambiguityinvolved.

Theresultsofthe previousstudiesindicatediscrepancies withrespecttotheearlyuseof discourseconstraints in pronounresolutionandtheyfurtherindicatedifferencesbetweenthemodalitiesofproductionandcomprehension,with spontaneousproductionbeingmoreintactthan comprehension.Moreover,thecomprehensionorinterpretationofthe ambiguouspronounseemstobeaffectedbycontextualfactorssupportingthefirst-mention/subjectbias(seeSongand Fisher, 2007vs.Arnoldetal.,2007).

ThepresentstudyaimsatexploringthedevelopmentoftheanaphorresolutionstrategyinGreekbysheddingmore light on speed and accuracy in which children resolve null and overt pronoun ambiguity with minimal contextual support andwithout anygrammaticalgender cue. In particular, weinvestigate theeffects of agrammatical factor, namely the pronominal form, overt vs. null pronouns, and an associated discourse property, topic shift vs. topic continuity, on pronoun resolution in Greek in three age groups of children. Additionally, we employed an on line interpretationtaskinordertoexplorewhetheranydiscrepanciesbetweentheadultandthechilddataareattributedto immature pragmatics or alternatively to the inefficiency of the child processing mechanism to integrate various information sources on line.

(8)

5. Thepresentstudy

ThedesignemployedtotestinterpretationofnullandovertpronounsinbackwardanaphorainGreekissimilartothat employed by Tsimpli etal. (2004),but isimplemented in twoon-lineexperiments.More specifically,weinvestigated pronounambiguityresolutionofnullandovertsubjectpronounsinsentenceswithtwocompatiblereferentswhichappear inthesubjectandobjectpositionofthepreviousclause.Bothreferentsarealwaysanimatewithagentivityassociatedwith thesubject.Thediscoursefeatureweexamineisapresencevs.absenceofashiftedtopicinterpretationintheuseofnull and overtpronouns.6

5.1. Experiment1: Resolutionofnullpronouns 5.1.1. Method

5.1.1.1. Participants. Thirty six monolingual Greek adults participated in the study (mean age: 30.4; SD: 5.85; 21 females).Moreover, wetestedthreegroupsofchildrenagedfrom6;0 to11;2:(a)twentymonolingualGreekchildren (10females)agedfrom6;0to6;3years(meanage:6.10;SD:0.24)attendingthe1stgradeofprimaryschool,(b)twenty twomonolingualGreekchildren(10females)agedfrom7;0to7;8years(meanage:7.35;SD:0.28)attendingthe2nd gradeofprimaryschooland(c)thirtyonemonolingualGreekchildren(16females)agedfrom10;0to11;2years(mean age:10.68;SD:0.35)beingatthe4thor5thgradeoftheprimaryschool.Thechoiceofthespecificagegroupswasbased ontherequirementsofthetaskaswellasonpreviousworktestingchildrenandadultsonpronouninterpretation.Thetask we employed was a demanding on-line task, in which the participants had to coordinate the oral presentation of a sentencewithapossiblevisualrepresentationofitand,thus,wedecidednottoruntheexperimentswithchildrenyounger than6years.Furthermore,studiestestingpronounresolutioninchildren(Arnoldetal.,2007;HickmannandHendriks, 1999; Tyler, 1983) showed insufficient integration of discourse-internal constraints, like first-mention biases, during pronouninterpretationin5-year-oldpreschoolers.Thisisconsistentwithclaimsintheliteraturethatchildrenarebeginning todeveloppragmaticabilitiesattheageof5years(VerbukandRoeper,2010).Onthecontrary,discourseprominence cues werefound to be more accessible to7 and 10 year old children whose pronoun resolution patterns appear to convergewiththoseofadults(ArgyriandSorace,2007;Tyler,1983).Thereisgrowingconsensusthatatthisagelevel childrenareabletoprocesson-linedifferenttypesofcues:lexical,syntacticandsemantic,inordertoidentifythereferent ofananaphor.Thisbase-linedifferencebetween5yearoldsandolderchildren(i.e.7and10yearolds),aswellasadults, in pronoun resolution brings out questions surrounding the robustness of discourse effects in 6 year old children.

Furthermore,studiesinGreek(ArgyriandSorace,2007)revealedthat,eventhoughtheproductionofnullpronominalsis adult-likeattheagefrom7;5to9;7years,thejudgmentofthesameconstructionsdiffersfromthatofadults(seesection4).

Takingalltheseconsiderationsinmind,wefocusedonthedevelopmentofthepronounresolutionmechanisminGreek from6to11yearsandwedirectlycomparedthe childtothe adultdata.

5.1.1.2. Design and procedure. Weemployed anon-linesentence picturematchingmethodology based onMarinis (2007,2008)andMarinisandSaddy(2013).Inthistask,apicturewaspresentedonthecomputerscreenfortwoseconds, afterwhichtheparticipantsstartedlisteningtoasentenceinaphrase-by-phrasefashion.Theparticipantscontrolledthe presentationofthesegmentsbypressingabuttononapushbutton box.Therefore,thepresentationofthesentential segments was self-paced. During the temporal unfolding of each sentence, the corresponding picture-probe was continuouslydisplayedatthecenterofthescreen,andremainedthereuntiltheparticipanthadheardtheentiresentence.

Assuch,displayofthevisualsceneandlisteningtoeachexperimentalsentencetookplaceinaconcurrentmanner.Once thelastsegmentoftherecordedsentencewaslistenedto,thepicturedisappearedandaquestionmark‘?’appearedon thecenterofthescreenforaslongastheparticipantneededinordertodecideifthesentencematchedthevisualsceneor not. Participantsrespondedby pressingoneoftwopre-specifiedbuttonson thepushbutton box.

Participantsweretaught howto pressthe buttonto hearthe sentencesand weregiven10practice sentencesto familiarize themselves withthe task. E-prime(Schneideret al., 2002) wasused topresentthe stimuli and to record Listening Times(LTs)fromtheonsetof eachsegmentuntilthe buttonpress.

Wemeasuredtheparticipants’responsestothematchingtask,inordertoexplorethepronounresolutionpreferences attheendofthesentence.Moreover,RTsonthematchingtaskhavealsobeenrecordedinordertoexaminewhetherthe participants’matchingpreferencesarereflectedintheresponsetimes.ListeningTimesoneachregionofthesentence (see example(6)below)havealsobeenmeasuredin ordertoinvestigatewhethertheresolutionpreferencesemerge rapidly,duringon-linesentencecomprehension.

6Weacknowledgethattherearemanymorediscoursefeatureswhichregulateanaphoraresolution(e.g.contrastivefocus,logophoricity;see Dimitriadis(1996)amongothers);however,inourstudywerestrictourexperimentalconditionstothepresencevs.absenceoftopic-shift.Forthis reasontheprosodiccontouroftheexperimentalsentencesisneutral.

(9)

5.1.1.3. Materials. The material for this task comprised 10 practice sentences, 10 experimental sentences, 30 experimentalpictures,30fillersentences,and30fillerpictures.Theexperimentalsentencesconsistedofamainclause, inwhichtworeferentswereintroduced,oneinthesubjectandoneintheobjectposition,followedbyanadverbialclause the subjectofwhichwasanullpronoun (see(6)).

(6) Opapús/milúse/dinatá/stonegono´ tu/o´tan /prodjávaze/enavivlío.

‘‘Theold-man/spoke-3SG/loudly/tohisgrand-child/when/proread-PAST.3SG/abook.’’

Theslashesin(6)indicatethewaythesentencewasdividedintosegments.Thesentenceswererecordedbyafemale native speakerof Greekata normalspeaking rate alwaysusinga defaultintonationalpattern to avoidanyprosodic information effect on the participants’ preferred interpretations.The recorded sentenceswereafterwards segmented usingCoolEdit.

Eachexperimentalsentencewasmatchedtothreedifferentpictures.Thepicturesdepictedtheactionpresentedinthe adverbialclauseandeachpictureprovidedadifferentinterpretation:inonepicturetheactorwasthesubjectofthemain clause(theoldman,for(6)),inthesecondtheactorwastheobjectofthemainclause(thegrandchild,for(6)),andinthe third the actorwasa thirdreferent (forexample,ayoung manfor(6)). Anexample ofthe experimentalconditionsis provided in Appendix A.Notice that allthree reference interpretations are possiblefor a sentence suchas (6). The anchoring of thenullpronoun to the subjector the objectora third antecedentreflects apreference of the pronoun resolutionmechanismratherthanagrammaticalprinciple(seealsoDimitriadis(1996)aswellCarminati(2002)forsimilar arguments). Therefore,therewerenopre-specifiedcorrect orwrongresponsestoeithercondition.7

Thecritical segmentinthisexperimentisSegment 6,i.e.the segmentcontainingthe verboftheadverbial clause (djávazefor(6)),becauseatthispointtheactionoftheadverbialclauseisevidentandparticipantshavetoassignthe thematicroleofthesubject.Hence,atthispointitbecomesclearwhetherornotthepicturematchesthesentence,i.e.it has the preferred interpretation of the sentence. If the picture does not match the participants’ interpretation of the sentence, thisshouldcauseincreasedLTsindicatingprocessingdifficulty.

Threelistswerecreated,eachcontainingthe10experimentalandthe30fillersentences.Ineachlist,theexperimental sentenceswerematchedtooneofthe threepictures.Each participantencounteredonlyoneofthe threelists.

Toensurethattheparticipantslistenedtothesentencesforcomprehension,attheendofeachsentencetheyhadto judgewhetherornotthesentencematchedthepicturebypressingoneoftwobuttons.Thebuttonpressrecordedthe participants’judgmentandalsotheirRTforthisdecision.Theexperimentlastedapproximately30min.

5.1.1.4. Predictions. Basedontheoreticalaccountsofthedistributionofnullpronounsinnull-subjectlanguages(see Ariel,1990;Dimitriadis,1996;Carminati,2002)aswellasonempiricalevidenceinGreek(Dimitriadis,1996;Miltsakaki, 2001),we predictthat Greekadults will exhibit a preference to interpret the nullpronoun as referring to the subject antecedent,sinceintheexperimentalsentencesitisthemostsalientreferent,i.e.thetopicofthesentence.Thismeans thatthereshouldbemorematchesforthesubjectreferentthantheobjectandthethirdreferentandthedecisiontimesat theendofthesentenceshouldbeshorterforthesubjectthanfortheobjectandthethirdreferent.Moreover,assumingthat pronoun resolutionstrategiesrapidlyevolveduringon-linesentencecomprehension,weexpectthatadultswillexhibit shorterlisteningtimesonthecriticalsegment,i.e.theverb,forthesubjectratherthantheobjectandthethirdreferent conditions.

Turningtothedevelopmentaldata,ifchildrenhavenotyetacquiredtherelevantdiscoursefeature,i.e.prominenceof the antecedent,forpronounresolution, weexpect nodifferencesbetweenthe subjectand theobjectconditionin the children’smatches,decisiontimesandlisteningtimes.BasedonpreviousfindingsfromGreek(ArgyriandSorace,2007), weexpectthatnullpronounswillbeinterpretedinanadult-likewayfromtheageof7yearsonwards.However,ifthe children, at all age groups, have difficulties with the integration of the relevant discourse cue in on-line sentence processing,weexpectnodifferencesintheirlisteningtimesbetweenthesubjectandtheobjectconditionsbutasignificant preference forthesubjectantecedentintheirmatchesanddecisiontimes.

Notice,however,thatifthebiasforthenullpronominaltoselectasalientantecedentisnotstrong,assuggestedby Dimitriadis(1996),weexpecttofindnodifferencesbetweenthesubjectandtheobjectantecedentineithertheadultorthe childdata.

7Sincetherewerenopre-specified(in)correctresponsestoourexperimentalitems,wedecidedtoanalyzematchandmismatchresponses togetherintheRTsonthedecisiontaskaswellasinthelisteningtimesonthesegments.Ananonymousreviewercommentsthatthisdecision mighthaveresultedindistortionsoftheresults.Inordertoaddressthispoint,weconductedtheWilcoxonSignedRanksnon-parametrictestto comparetheRTsfromthematchesandthemismatchesonthedecisiontaskforthethirdreferentconditionacrossthetwoexperimentsandfor eachexperimentalgroup.Noneofthesedifferencesturnedouttobesignificant(p>.1).Nevertheless,itremainstobeseenwhetherthepresent findingswillbereplicatedifadifferentexperimentaldesignisadopted.

(10)

Regardingthethirdreferentcondition,wepredictthatbothadultsandchildrenwillexhibitweakerpreference,slower decisionand listening timeson the conditionin which thenull pronounisdepicted to refertoa thirdreferent viathe accompaniedpicture.ThispredictionisbasedonpreviousfindingsbyTsimplietal.(2004)onbackwardanaphoraaswell asonthatthethirdreferentisnotexplicitlymentionedinthesententialcontext,i.e.themainclause,whilethesubjectand the object referents are mentionedin thesentence. Hence, we expect thatreference tothe third referent, an extra- sententialone,willinflictextraprocessingcostrelativetothesubjectandtheobjectreferents.However,wepredictthat thisdifferencewillbemorepronouncedinthechilddata,sincechildrenaremorelimitedintheirprocessingresourcesthan the adultsand, inparticular,theydisprefer referencetoextra-sentential antecedents(Felseretal., 2003a,b;Sekerina etal., 2004).

5.1.2. Results

Theexperimentresultedinthreetypesofdata:(1)off-linepicturematchingjudgment,(2)RTsonthepicturematching judgments, and(3)on-linelistening times.Alltypesofdata hadone variable,the typeof reference,withthree levels (subject,object,andthirdreference).WeanalyzedeachtypeofdatabyusingrepeatedmeasuresANOVAswithreference (subjectvs.objectvs.third)asthewithinsubjectsvariableandgroup(adultsvs.childgroups)asthebetweensubjects variable.

5.1.2.1. Off-line picture matching judgment and RTs on the picture matching judgments. Graph 1 shows the participants’matchespergroupandconditionin percentages.

Ascoreof5wasgivenwhentheparticipantrepliedthatthepicturematchedthesentenceandascoreof1whenthe participantrepliedthatthepicturedoesnotmatchthesentence.Thestatisticalanalyseshavebeenconductedonthese raw scores. There was a significant main effect of reference (F1(2,210)=199.838, p<0.001, h2=0.656, F2(2,72)

=76.847,p<0.001,h2=0.681),group(F1(3,105)=7.156,p<0.001,h2=0.170,F2(3,36)=3.218,p<0.05,h2=0.176), and asignificantinteractionbetweenreferenceandgroup(F1(6,210)=4.592, p<0.001,h2=0.116,F2(6,72)=3.007, p=0.011,h2=0.200).

Inordertoinvestigategroupeffectsineachexperimentalcondition,weperformedplannedcomparisonsbymeansof independentsamplest-tests.Thestatisticaltestsshowedthattheadultdatadifferedfromthe1stchilddatainboththe object(t(54)=3.209,p=0.002)andthethirdreference(t(54)=2.505,p=0.015)condition.Moreover,theadultdatawere found todifferfromthe4th/5thchild dataonlyintheobjectreferencecondition (t(65)=4.540,p<0.001). Finally,2nd gradechildrenwerefoundtoprefertheobjectantecedentsignificantlylessfrequentlythan4th/5thgradechildren(t(51)

=2.795,p=.007)andsimilarlypickedthethirdreferentsignificantlylessfrequentlythanboth4th/5thgrade(t(51)=2.495, p=.016)and1stgradechildren(t(40)=2.881,p=.006).Ontheotherhand,theresultsshowedthatadultsand7year oldsshowedasimilarpatternofperformanceacrossconditions.

To further explore the interaction between reference and group we performed planned comparisons with the Bonferronitestineachgroup.Theadultdatashowedthatthesubjectreferencereceivedsignificantlymorematchesthan the object (p<0.001) and the third reference (p<0.001) and that the object reference received significantly more

adults 4th/5th grade

2nd grade 1st grade

100,00

80,00

60,00

40,00

20,00

0,00

35,66 43,22

25,03

53,09 57,42

84,64

65,86 79,01

89,21 91,43

95,26 95,04

other object subject

Graph1. Adultandchildmatchingdecisionpreferences(%)onnullpronouns.ErrorbarsrepresentSDs.

(11)

matchesthanthethirdreference(p<0.001).Likewise,the1stand2ndgradechilddatashowedthatthesubjectreference receivedsignificantlymorematchesrelativetoboththeobject(p=0.009andp<0.001for1stand2ndgradechildren, respectively)andthethirdreference(p<0.001forboth1stand2ndchildgroups),whiletheobjectreceivedconsiderably morematchesrelativetothethirdreference(p<0.001forboth1stand2ndchildgroups).Ontheotherhand,the4th/5th grade childdatashowed thatboththesubject(p<0.001)andtheobjectreference(p<0.001)hadsignificantlymore matches than the third reference, whereas there was no significant difference between the subject and the object reference (p>0.1).

Turningtothedecisiontimesonthematchingtask,thedatainTable1indicateshorterRTsonthesubjectthanthe othertwoconditionsinallgroups.

Therewasamaineffectofgroup(F1(3,105)=10.303,p<0.001,h2=0.227,F2(3,36)=9.721,p<0.001,h2=0.448) and a main effect of reference (F1(2,210)=8.375, p<0.001, h2=0.074, F2(2,72)=5.611, p=0.005, h2=0.135).

PlannedcomparisonswiththeBonferronitestforthebetween-groupdifferencesshowedthatthe1stgradechildrenwere significantly slower than the 2nd and 4th/5th grade children as well as adults (p=0.004, p<0.001, and p<0.001, respectively). Toexplorethe reference effect obtained,weconducted plannedcomparisonswiththe Bonferroni test, which revealedthat the subjectcondition elicited significantlyshorter RTsthan boththe objectand the thirdreferent (p=0.006andp=0.001, respectively).

Even thoughwedid notobtainan interactionbetween groupand reference, weranposthocrepeated measures analysesineachgroupasanexploratoryanalysis,inordertoexaminewhetherthedifferentreferencepreferencesamong thechildgroupsevidencedonthematches,alsoappearontheirdecisiontimes.Theseanalysesrevealedasignificant reference effect onlyfor the 2nd grade children (F1(2,42)=4.420, p=0.018, h2=0.174, F2(2,18)=3.849, p=0.41, h2=0.300)andtheadultsF1(2,70)=4.516,p=0.014,h2=0.114,F2(2,18)=6.074,p=0.010,h2=0.403).Pairedt-tests showedthatthiseffectwasduetotheconsiderablyshorterRTsforthesubjectreferentrelativetotheobjectreferent(t(21)

=2.610,p=0.016 and t(35)=3.139, p=0.003 for2nd gradechildren andadults,respectively)and the thirdreferent (t(21)=2.882,p=0.009 forthe2nd gradechildrenonly).

5.1.2.2. On-linelisteningtimes. LTsthatwere2SDaboveorbelowthemeanperconditionandsubjectwereexcluded fromanyfurtheranalyses.Thisaffectedlessthan4.0%ofthedata.AllstatisticalanalyseswereperformedonrawLTs.

Wemeasuredlisteningtimesforthefirstsixsegmentsofourexperimentalitems(seeexample6).8Thecriticalsegment isthesixthone,i.e.thesubordinateverb,onwhichdifferencesamongthegroupsandtheexperimentalconditionsare expected.Onthesubordinateverb(Graph2),therewasahighlysignificantmaineffectofreference(F1(2,210)=6.118, p=0.003, h2=0.055, F2(2,72)=3.731, p=0.029, h2=0.094), but no main effect of group or interaction between referenceandgroup.PlannedcomparisonswiththeBonferronitestrevealedthatthethirdreferentco-indexationcondition elicitedsignificantlylongerLTsrelativetothesubjectreferentcondition(p=0.009).Exploratoryposthocanalyseshave alsobeenconductedpergroup,despitethefactthattheinteractionbetweengroupandreferenceswasnotsignificant,in order toinvestigate possiblereference preference inon-linesentence comprehension.Separate repeated measures ANOVAsconductedforeachgrouprevealedasignificantreferenceeffectonthesubordinateverbforthe1st(F1(2,38)

=5.542,p=0.008,h2=0.226,F2(2,18)=7.146,p=0.005,h2=0.443)andthe2ndgrade(F1(2,42)=4.566,p=0.016, h2=0.179;F2(2,18)=2.845,p=0.069,h2=0.240).Pairedt-testsshowedthatthiseffectwasduetoconsiderablylonger LTsforthethirdreferentrelativetothesubject(t(19)=2.947,p=0.008andt(21)=2.341,p=0.029for1stand2ndgrade, respectively) and the object referent (t(19)=1.954, p=0.066 and t(21)=2.129, p=0.045 for 1st and 2nd grade, respectively).

Thedatafromthesegmentsbeforethecriticalone(seeTableB1ofAppendixB)werealsostatisticallyanalyzedin order to ensure that no differences with respect to reference preferences emerge prior to encountering the critical

Table1

Adultandchilddata:meanRTsfromthematchingtaskwithSDsinparentheses.

Nullpronoun

Subject Object Third

Adults 1136(514) 1522(725) 1259(701)

1stgradechildren 1832(594) 2441(887) 2684(773)

2ndgradechildren 1270(495) 1643(606) 1842(986)

4th/5thgradechildren 1365(835) 1431(712) 1474(692)

8Listeningtimesonthefinalsegment(7th)ofthesentencewerenotrecorded,sinceaquestionmarkautomaticallyappearedafterthefinal segmentwasorallypresentedtotheparticipants.ParticipantsmeanLTsandSDspersentencesegmentarepresentedinTable1ofAppendixB.

(12)

segment.Onlygroupeffectsareexpectedonthesesegments,sincethechildlisteningtimesmaybeslowerthantheadult ones.Indeed,theonlysignificanteffectobtainedwasthatofgroup(Segment1:F1(3,105)=8.049,p<0.001,h2=0.187, F2(3,36)=3.533, p=0.024, h2=0.227; Segment 2: F1(3,105)=10.877, p<0.001, h2=0.237, F2(3,36)=6.961, p=0.001, h2=0.367; Segment 3: F1(3,105)=8.259, p<0.001, h2=0.191, F2(3,36)=6.115, p=0.002, h2=0.338;

Segment 4: F1(3,105)=6.719, p<0.001, h2=0.161, F2(3,36)=6.669, p=0.001, h2=0.357; Segment 5:F1(3,105)

=11.226,p<0.001,h2=0.243,F2(3,36)=7.372,p=0.001,h2=0.381).PlannedcomparisonswiththeBonferronitest showedthattheadultsweresignificantlyfasterthanboththe1standthe2ndgradechildreninthefirst(p<0.001and p=0.002for1stand2ndgrade,respectively),second(p<0.001andp=0.001for1stand2ndgrade,respectively),third (p<0.001andp=0.002 for1stand2nd grade,respectively),fourth(p=0.001andp=0.014for1stand2nd grade, respectively),andfifth (p<0.001forbothgroups)segment.

5.2. Experiment2: Resolutionofovert pronouns 5.2.1. Method

Theparticipantsaswellthedesignandtheprocedureofthesecondexperimentwereexactlythesameasthoseofthe first experiment(see section5.1.1).

5.2.1.1. Materials. Theexperimentalitemswerecreatedexactlyastheonesofthefirstexperiment.Theonlydifference wasthattheadverbialclauseincludedanovertsubjectpronoun(see(7)).Thetworeferentsinthesubjectandtheobject positionofthemainclausealwayshadthesamegendersothattheovertpronounwascompatiblewithbothpossible antecedents intermsofitsgendermarking.

(7) Ijajá/xerétise/tinkopéla/o´tan /aftí/pernúse/todro´mo.

‘‘Theold-lady/greeted-3SG /thegirl/when/she/crossed-3SG/thestreet.’’

Theslashesindicatethewaythesentencesweredividedintosegments.Thecriticalsegmentsweretheovertsubject pronounandtheverboftheadverbialclause.Thesecondexperimentincludedthirtyfilleritems,asthefirstexperiment.

Asinthenullpronounexperiment,threereferenceconditions(subject,objectandthirdreferent)wereincludedinthis experiment too. All three reference options are considered as grammatical, for the reasons already outlined in section5.1.1.

5.2.1.2. Predictions. Basedonprevioustheoreticalwork(Ariel,1990;Dimitriadis,1996;Carminati,2002)andempirical research in Greek(Dimitriadis, 1996; Miltsakaki,2001),we predictthat the Greekadults will exhibit a preference to interprettheovertsubjectpronounasreferringtotheobjectantecedent,sinceitisnotthetopicoftheclauseprecedingthe oneincludingthepronoun.Thismeansthatwewillhavemorematchesforthepicturedepictingtheobjectthanthesubject referentaswellasshorterRTsfortheobjectthanthesubjectreferentinthedecisiontask.Iftheseresolutionpreferences emergeduring on-linecomprehension,shorterlisteningtimesfortheobjectreferentoverthesubjectreferentarealso expectedonthecriticalsegments.AlsonoticethataccordingtoDimitriadis’(1996)OvertPronounRule(section3),we predictthatthebiasoftheovertpronoun towardnon-topicantecedentswillbestrong.

Graph2. Adultandchildmeanlisteningtimes(inms)onthesubordinateverbforthenullpronounexperiment.

(13)

Regardingthedevelopmentaldata,weexpectnoadvantagefortheobjectreferentifthechildrenhavenotyetacquired thediscoursefeature,topicshift,associatedwithovertpronouns.Thiswillbereflectedintheirmatches,decisiontimes andlisteningtimes.However,basedonpreviousfindings(ArgyriandSorace,2007),weexpectthechildrentoemploy adult-likeresolutionstrategies,i.e.preferencefortheobjectantecedent,fromtheageof7yearsonwards.If,ontheother hand, childrenhaveacquiredtherelevantdiscoursefeatureoftheovertpronouns,butinsteadfacedifficultieswithits incorporationinsentence interpretation,weexpectnodifferencesbetweenthesubject andtheobjectcondition inthe listeningtimesofthecriticalsegments,butanadvantagefortheobjectoverthesubjectreferentinthematchesandthe decisiontimes.

Regardingthethirdreferentcondition,wemakesimilarpredictionsastheonesforthetaskonnullpronouns(section 5.1.1).Namely,wepredictadisadvantageforthethirdreferentinboththeadultandchilddata,thoughweexpectthis disadvantagetobeincreasedin thechilddata.

5.2.2. Results

Inthesecondexperiment,asinthefirstexperiment,therewasonevariable,thetypeofreference(subject,object,and third).Ineachdataset,weperformedrepeatedmeasuresANOVAswithreference(subjectvs.objectvs.third)asthe within subjectsvariableandgroup(adultsvs.childgroups) asthebetweensubjectsvariable.

5.2.2.1. Off-linepicturematchingjudgmentandRTsonthepicturematchingjudgments. Firstwepresentthedatafrom the sentence picturematching task performed at the end of each sentence as well as on the decision times. The participants’matchingdecisionsinpercentagespergroupandcondition areillustratedinGraph3.

As in the experiment on null pronominals, a score of 5 was given when theparticipant repliedthat the picture matchedthesentenceandascoreof1whentheparticipantrepliedthatthepicturedoesnotmatchthesentence.The statistical analyseshavebeenrun onthese rawscores.Therewasamaineffect ofreference(F1(2,210)=61.453, p<0.001, h2=0.369, F2(2,72)=17.010, p<0.001, h2=0.321), group (F1(3,105)=17.870, p<0.001, h2=0.338, F2(3,36)=5.530, p=0.003, h2=0.315), and a significant interaction between reference and group (F1(6,210)

=6.266, p<0.001, h2=0.152, F2(6,72)=4.602, p=0.001, h2=0.277). To explore the groupeffect we performed planned comparisonsby means ofindependentsamples t-tests comparing theperformancesof the four groupsin each condition.These revealed thatboth 1st and 2ndgrade children provided significantly morematches for the subject reference condition relative to the 4th/5th children (t(49)=2.583, p=0.013, and t(51)=3.758, p<0.001, respectively). Moreover, 1st and 2ndgrade children provided considerably more matches than adults in boththe subject (t(54)=6.950,p<0.001and t(56)=9.751, p<0.001, respectively)andthethird referencecondition(t(54)

=3.193, p=0.002 and t(56)=2.092, p=0.041, respectively). The 2nd grade children were also found to provide considerablymorematchesfortheobjectantecedentrelativetotheadults(t(56)=2.099,p=0.040).Finally,4th/5th childrenprovidedsignificantlymorematchesthanadultsinboththesubject(t(65)=4.379,p<0.001)andtheobject (t(65)=2.105,p=0.039)referencecondition.

Graph3. Adultandchildmatchingdecisionpreferences(%)onovertpronouns.ErrorbarsrepresentSDs.

(14)

Furthermore,duetoasignificantinteractionbetweengroupandreferenceweperformedplannedcomparisonswiththe Bonferronitestineachgroup.Theseanalysesshowedthatboth1stand2ndgradechildrenprovidedsignificantlyless matchesforthethirdreferentrelativetotheobject(p=0.05andp<0.001for1stand2ndgradechildren,respectively) andthat2ndgradechildrenprovidedsignificantlylessmatchesforthethirdreferentrelativetothesubject(p<0.001).In addition,therewasnosignificantdifferencebetweenthesubjectandtheobjectreferenceineitherthe1stgradeorthe second grade group (p>0.1 for both groups). The same tests showed that both the adultand the 4th/5th children providedmorematchesfortheobjectthanthesubject(p<0.001)andthethird(p<0.001)reference.

Thedecisiontimesforthematchingtaskaredisplayedin Table2.

Therewasamaineffectofreference(F1(2,210)=8.945,p<0.001,h2=.062,F2(2,72)=7.132,p=0.001,h2=.165), group (F1(3,105)=12.255, p<0.001, h2=.259, F2(3,36)=5.687, p=0.003, h2=.322), and a significant interaction betweenreferenceandgroup(F1(6,210)=5.733,p<0.001,h2=.141,F2(6,72)=3.053,p=0.010,h2=.203).Planned comparisons withthe Bonferronitest forthe factor reference revealedthat RTson matchingdecisions forthe extra- sentential referentweresignificantly slowerthan thedecisions forthesubject and theobjectreferent (p=0.009 and p<0.001,respectively).PlannedcomparisonswiththeBonferronitestforthegroupeffectshowedthatthe1stgraders weresignificantlyslowerthanthe4th/5thgradersandtheadults(p<0.001forbothcomparisons),whilethe2ndgraders weresignificantlyslowerthantheadults(p=0.008).

Tofurtherexplorethedirectionofthesignificantinteractionbetweengroupandreference,weranseparaterepeated measures posthocanalyses foreach group.Theseanalyses revealedasignificant referenceeffect forallgroups;

(F1(2,38)=3.122, p=0.05, h2=.141, F2(2,18)=3.836, p=0.41, h2=.299) for the 1st grade, (F1(2,42)=23.025, p<0.001,h2=.523,F2(2,18)=11.096,p=0.001,h2=.552)forthe2ndgrade,F1(2,60)=3.652,p=0.032,h2=.109, F2(2,18)=5.092,p=0.018,h2=.361)forthe4th/5thgrade,andamarginallysignificantreferenceeffectfortheadults;

F1(2,70)=2.626,p=0.080, h2=.070;(no such effectwas observedon theF2analysis).Posthocpaired-samples t-testsshowedthatthiseffectwasduetoconsiderablyshorterRTsforthesubjectrelativetothe thirdreferent(t(19)

=2.502,p=0.022)forthe1stgrade,significantlylongerRTsforthethirdrelativetothesubjectandtheobjectreferent (t(21)=5.825,p<0.001andt(21)=4.830,p<0.001,respectively)forthe2ndgrade,considerablyshorterRTsforthe objectrelativetothesubjectandthethirdreferent(t(30)=3.063,p=0.005andt(30)=1.876,p=0.070,respectively)for the 4th/5th grade,and, finally, significantly shorter RTsfor the object relativeto thesubject referent t(35)=1.975, p=0.05)fortheadults.

5.2.2.2. On-linelisteningtimes. LTsthatwere2SDaboveorbelowthemeanperconditionandsubjectwereexcluded fromanyfurtheranalyses.Thisaffectedlessthan3.6%ofthedata.AllstatisticalanalyseswereperformedonrawLTs.

Asintheexperimentonnullpronominalsweanalyzedthelisteningtimesfromthefirstsixsegments(seeexample5).9 Thecriticalsegmentinthistaskisthesubordinateverb(6thsegment).However,listeningtimesontheovertpronoun(5th segment)mayalsobeinformativeabouttheprocessingofreferringexpressions.Ontheovertpronoun(Graph4),there was amaineffect ofgroup (F1(3,105)=7.907, p<0.001, h2=0.184, F2(2,72)=3.731,p=0.029, h2=0.094),but no maineffectofreferenceoraninteractionbetweenreferenceandgroup.Thegroupeffectwasfurtherexaminedbymeans ofplannedcomparisonswiththeBonferronitest,whichrevealedthatthegroupeffectwasduetotheconsiderablyfaster responsesoftheadultgrouprelativetothe1st(p=0.001)and2ndgradechildren(p=0.001).Eventhoughwedidnot obtainasignificantinteractionbetweengroupandreference,weranposthocanalysesasexploratorymeasuresinorder toexaminehowreferencepropertiesemergeon-lineineachgroup.Separaterepeatedmeasuresanalysesconductedfor eachgrouprevealedasignificantreferenceeffectontheovertpronoun(F1(2,42)=5.170,p=0.010,h2=0.198,F2(2,18)

=6.479, p=0.008, h2=0.419) for the 2nd grade children only. Paired t-tests showed that this effect was due to significantlylongerLTsforthesubjectreferentrelativetotheobjectandthethirdreferent(t(21)=2.290,p=0.032andt (21)=2.259,p=0.035, respectively).

Table2

Adultandchilddata:meandecisiontimesformthetaskonovertpronounswithSDsinparentheses.

Subject Object Third

Adults 1544(710) 1272(621) 1286(559)

1stgradechildren 1916(845) 2411(973) 2703(994)

2ndgradechildren 1718(547) 1570(438) 2467(931)

4th/5thgradechildren 1732(889) 1320(667) 1615(710)

9ParticipantsmeanLTsandSDspersegmentarepresentedinTable2ofAppendixB.

Referenzen

ÄHNLICHE DOKUMENTE

Presence experience evoked by a virtual roller coaster scenario is associated with an increase in activation in a distributed network, which includes the dorsal

• East as compared to West German mothers placed more importance on obedience and reported more controlling parenting. • → Results largely in line

Our empirical strategy is based on a lab-in-the-field experiment that uses a modified version of the solidarity game by Selten and Ockenfels (1998). In this three-person

The present study compared production and on-line comprehension of definite articles and third person direct object clitic pronouns in Greek-speaking typically developing,

The present study addresses whether L2 children’s difficulties with definite articles are post-syntactic, as the MSIH and the PTH would suggest, and whether L2 children can

However, this trend to increased parallel processing compared to the serial instruction in the control experiments underpins our previous observation that, in dual tasks,

In addition, the presently observed dissociation between reading and spelling deficits has not been reported before for the Arabic Language and calls into question that

International Labour Organization (ILO) and was one of a series of ten national studies conducted around the world, to measure the incidence of forced labour. The survey was