gender agreement and multiple...

Genderagreementandmultiplereferents

ChiaraFinocchiaro,BradfordZ.Mahon&AlfonsoCaramazza

We report a new pattern of usage in current, spoken Italian that hasimplicationsforbothpsycholinguisticmodelsoflanguageproductionandlin-guistictheoriesof languagechange.InItalian,genderagreementismanda-toryforbothsingularandpluralnouns.However,whentwoormorenounsofdifferentgrammaticalgenderappearinaconjoinednounphrase(NP),mas-culinepluralagreementisrequired.Inthisstudy,wecombinedon-lineandoff-linemethodologiesinordertoassessthemechanismsinvolvedingendermarkinginthecontextofmultiplereferents.Theresultsoftwopronounpro-ductiontasksshowedthatpluralfeminineagreementwassignificantlymoredifficult thanpluralmasculineagreement. Inaseparatestudyusingofflinejudgements of acceptability, we found that agreement violations in Italianare tolerated more readily in the case of feminine conjoined noun phrases(e.g.,la mela e la banana‘the:femapple:femandthe: fembanana: fem’)thanmasculineconjoinednounphrases(e.g.,il fiore e il libro ‘the:mas flower: masandthe: masbook:mas’).Implicationsoftheseresultsarediscussedbothattheleveloffunctionalarchitecturewithinthelanguageproductionsystemandatthelevelofchangesinlanguageuse.*

1. Introduction

Ingender-markedlanguages,grammaticalgenderisanintrinsicpropertyof lexical items.Whenagiven lexical itemappears in con-text,itsgenderfeatureisinheritedbyalltheelementswithwhichitisinanagreementrelationshipandthatrequirethegenderfeature.For instance, in the Italian sentence Il cappello è rosso ‘The:mas.sg hat:mas.sg is red:mas.sg’, both the determiner and the adjectiveare marked with the masculine ending, as inherited from the wordcappello, which is masculine. If the noun cappello has already beenintroducedinacommoncontext,thespeakermayuseapronoun,asinlo prendo‘Itakeit:mas.sg’,where,again,thepronounlo ismarkedformasculinegender.Therules forgender-markinghold forpluralsaswell.ThepluralequivalentforbothsentenceswouldbeI cappelli sono rossi‘The:mas.plhats:mas.plarered:mas.pl’.

In the examples above, the gender feature is inherited from alexicalnodebytheelementsinthesentencethatrequirethatfeature:critically, in those examples, there is always one linguistic unit – a

Rivista di Linguistica19.2(2007),pp.285-307 (ricevuto nel marzo 2008)

Chiara Finocchiaro, Bradford Z. Mahon & Alfonso Caramazza

286

singular or a plural noun – triggering agreement. In this study weexploredthesituationinwhichagreementistriggeredbythecombi-nationofmorethanonelinguisticunit.ThisisthecaseforconjoinedNPs(e.g.,“thehatandthetie”).InItalian,theagreementresolutionrule foraconjunctionofnouns isas follows (seeDardano&Trifone1997;Serianni&Castelvecchi2004;Salvi&Vanelli2004):whenthenounsareall feminine (e.g., la mela e la banana ‘theappleandthebanana’) they invoke feminineagreement (e.g.,sono buone ‘theyaregood:fem.pl’);whenthenounsareallmasculine(e.g.,il fiore e il libro‘the flower and the book’), they require masculine agreement (e.g.,sono belli ‘they are beautiful:mas.pl’’); when the nouns have differ-entgender features (e.g., il fiore e la pianta ‘the flower:mas and theplant:fem’),nomatterhowmanyfeminineandmasculinenounsthereare,masculinepluralagreementisrequired(e.g.,sono belli‘theyarebeautiful:mas.pl’’).

While the existence of theabovedescribedgrammatical rule in

Italian is not in dispute, informal observation indicates that nativeItalianspeakerssometimesusemasculineagreementwhentherulewouldhaverequiredfeminineagreement.Such ‘errors’ intheuseofgender resolution rules may reveal the complex interplay of factorsthatarenotconsideredintheruleasitiscommonlystated(Corbett1991). To our knowledge, there have been no systematic investiga-tionsofgenderagreementformultiplereferents inItalian,andspe-cifically,ofthosesituationsthatgenerateviolationsoftheruleinthespontaneousspeechofnativeItalianspeakers.Thatisthesubjectofthepresentinvestigation.

One issue concerns the reasons why native Italian speakers,in some situations, ‘violate’ gender agreement rules in spontaneousspeech.Theavailablehypothesescanberoughlydivided intoclaimsaboutperformance factorsandclaimsaboutcompetence factors.Forinstance,onepossibility is thatover-generalizationof themasculineformisduetocontextualreasonsintheconcreteuseof language.Ifonewantstorefertoacollectionofobjects,andtherearenoparticu-larstylisticdemands,heorshemaywanttousetheformthatappliestothemajorityofcases–themasculineform–eventhoughprescrip-tivelyincorrect.Ifsuchwerethecase,thenobservationsthatnativeItalian speakers over-generalize masculine agreement would notnecessarily indicate that they represent grammatical rules that aredifferent from thosepresumed togoverngrammaticalgenderagree-ment in Italian. Another possibility, however, is that the masculine

Gender agreement and multiple referents

287

generalizationisindeedduetocompetencefactors.Inotherwords,itmaybethatnativespeakersofItaliandoinfactrepresentgrammati-calrulesthatpermitover-generalizationofmasculineagreement.

There is no direct measurement of speakers’ competence.Traditionally,however, speakers’ intuitionsareassumed to reflect,moreorlessdirectly,languageknowledge.Thetaskthatistypicallyused for such purposes requires participants to judge the gram-maticalityofagivensentenceinagivenlanguage.Therationaleisthatparticipants rely onwhat theyknowof the language in orderto accept or reject, as well formed, the sentence to which they areexposed.

Incontrast,performanceisusuallyassessedtroughon-linemeas-urementsoflanguageuse–typicallyspeedofresponseorerrorrates.

Inthisstudy,weuseacombinationofon-lineandoff-linetests.Specifically,weusedbothchronometricproductionofpronounproduc-tionaswellasparticipants’grammaticalityjudgmentsgenderagree-mentviolationsforconjoinednounphrasesinItalian.Inthefirsttwoexperimentsweusedonlinetasksundertimeconstraint.Participantsinthoseexperimentswereaskedtorefertopicturesbyusingsingularorpluralpronouns.Inthethirdexperiment,weaskedparticipantstorateonaseven-point scale theacceptabilityof sentences containinggender-agreementviolationsonverbsorpronounsforconjunctionsofnouns.Nounswereconjoinedtroughtheconjunctione‘and’,sinceitistheonlywaytoformallyexpresspositivecoordinationinItalian.Thistaskwasperformedwithouttimeconstraints.

Different predictions follow from the competence and the per-formancehypotheses.

Accordingtotheperformancehypothesis,thefemininedisadvan-tageshouldemergeintheon-linetasks,butnotintheoff-linetasks.The competence hypothesis predicts instead a consistent patternin both on-line and off-line tasks. This is because, according to thehypothesis,participants’performanceisdrivendirectlybyknowledgeoftherulesofthelanguage.

We acknowledge that introducing multiple referents deictically(Experiments1-2)andintroducingthemtroughacoordinationofNPs(Experiment3)maydiffer in other respects than theon-line/off-linecontrast. However, the focus here is not on noun conjunctions per se, but on the production/judgement of forms that need to agree ingenderwithapairofreferents.Tothatend,weassumethatthetwomethodologiesmaybeequivalent.


288

2. Experiment 1

Inthisexperimentweinvestigatedtheproductionofverbphras-es [V-accusative pronoun], where the verb form was kept constant(imperative, II sing, e.g., porta- ‘bring [you:sg]’). In contrast, theaccusative formof thepronounvarieddependingon thegenderandthenumberof thepicturespresented.Thepronouns thatwereusedbelonged to the group of clitics. Italian clitics are phonologicallyunstressedpronounsthatcanbeplacedonlyadjacenttotheverbonwhichtheydepend.1

Oneverytrial,eitherasingleor twopictureswouldappear. Inonecase,participantswererequiredtoproducethesingularpronounform(lo ‘it:mas’orla ‘it:fem’dependingonthegenderofthepicturednoun).Intheothercase,theywererequiredtoproduceapluralpro-nouncorrespondingtotheintegratedvalueofthegendersofthetwopictured nouns. The correct plural forms were le ‘them:fem’ for twofeminine gender nouns, and li ‘them:mas’ for two masculine gendernouns,aswellasforsituationsinwhichonefeminineandonemascu-linepicturewerepresented.

Ofcriticalimportanceforaddressingthecausesofover-generali-zationofmasculineagreementarepluraltrials:areparticipantsslow-er to respond when they are presented with two feminine picturescomparedtotheotherexperimentalconditions(congruent-masculineandmixed-gendertrials)?

2.1. Method

2.1.1. Participants. Twelve native Italian speakers, students atPisaUniversity, tookpart intheexperiment forpayment.Theyhadnormalorcorrected-to-normalvisionandreportedneverhavinghadanytypeoflanguagedisorder.

2.1.2. Materials and design. Forty-eight pictures were selected.They represented familiarandconcreteobjects,halfwithmasculinegendernamesandhalfwith femininegendernames.Masculineandfemininepicturesdidnotdifferwithrespecttolength(meanlength,mas: 6.8; fem: 6.3) or frequency of use as reported in Bortolini etal. (1971. Mean freq., mas: 19.2; fem: 16.7). Pictures could appearalone(i.e.,withablankframe:e.g.,gonna‘skirt:fem’–[blankframe];guanto ‘glove:mas’–[blankframe])orinpairs.Picturepairscouldbecongruentingender(i.e.bothmasculineorbothfeminine:e.g.,gonna‘skirt:fem’ – cravatta ‘tie:fem’;guanto ‘glove:mas’ – cappello ‘hat:mas’)


289

or incongruent in gender (i.e., one masculine and one feminine: e.g.gonna ‘skirt:fem’ – cappello ‘hat:mas’). The complete set of picturepairs are reported in Appendix A. Pictures were divided in threeblocks.Agivenpictureappearedonceperblock.Therewereanequalnumberofcongruent-genderconditions,mixed-genderconditionsandsingular(i.e.,onepicturepresented)conditionsineachblock.Inthesingularcondition,halfof thepicturesappearedontherightsideofthe computer screen, the other half on the left. Stimuli were rand-omized within blocks with the following constraints: (i) there was amaximum of two consecutive trials with the same number values;(ii)thetwotrialsatthebeginningandattheendofeachblockweresingularconditiontrials.Sixdifferentinter-blockrandomizationsandtwodifferentintra-blockrandomizationswereused.Anadditionalsetof24picturepairswereselected.Theyservedaspracticetrialsbeforetheexperimentproper.

2.1.3. Procedure.Participantsweretestedindividuallyinatest-ingroomandseatedatadistanceofabout60cm.fromthecomputerscreen.Theexperimentstartedwithanamingtaskaimedatfamiliar-izing participants with the pictures and their names. When partici-pantsproducedanameotherthanthatexpectedbytheexperimenter,theywerecorrected.Suchinstanceswereveryrare.Participantsweretheninstructedtoorallyproducethedefaultverbportare‘tobring’inthesecondsingular,imperativeform.Inaddition,theyhadto‘attach’to the verb the clitic pronoun corresponding to the picture(s) thattheywouldsee(e.g.,portalo‘bringit:mas’).Thus,theresponsesetwasconstitutedbythefollowingcombinations:portala (‘bringit:fem’;sin-gular condition, feminine), portalo (‘bring it:mas’; singular condition,masculine), portale (‘bring them:fem’; plural condition, two femininepictures), portali (‘bring them:mas’; plural condition, two masculinepictures or two mixed-gender pictures). Instructions emphasizedresponsespeedandaccuracy. Participantsthenperformedapracticeblock,afterwhichtheexperimentproperbegan.

2.1.4. Trial Structure.Atthebeginningofeachtrial,aquestionmarkappearedinthecenterofthecomputerscreen.Assoonaspar-ticipantspressedthespacebarthequestionmarkdisappearedanditwasimmediatelyreplacedbyafixationpoint(plussign)for500ms.Ablankscreenof200msfollowed,afterwhichthepicture(s)appeared.Pictures remained on the screen until participants responded oruntil2000mshadelapsed–whichevercamefirst.Stimuluspresen-tationwas controlledby theprogramPsyscope (Cohenetal., 1993).Response latencies were measured by means of a voice key, fromthe onset of the picture stimuli(us). The experimenter was present


290

throughout the testing session and recorded participants’ responsesmanually.

2.1.5. Analysis. Verbal dysfluencies, responses differing fromthose intendedby the experimenter,no-responses, failures to recordand outliers (RTs less than 300 ms. or greater than 3 StandardDeviations(s.d.)fromeachparticipant’smean)werescoredaserrors.Errorswere removed from theanalyses of response timesandweresubmitted to separate analyses. Separate ANOVAs were conductedforthepluralconditionsandforthesingularcondition,treatingsub-jects as a random variable (Raaijmakers et al., 1999). This choiceisalsomotivatedby the fact that thenumberof items,especially inExperiment 2, is not large enough to motivate generalization overitems. In the singular condition, there was one within-participantfactor, TargeT gender, with two levels (feminine vs. masculine). Inthepluralcondition,therewasonewithin-participantfactor, genderCongruenCy,withthreelevels(congruent-femininevs.congruent-mas-culinevs.mixed-gender).Sincetheobservationsforthemixed-genderconditionweretwiceasmanyastheobservationsforeachoftheotherconditions,onlyhalfof the trials for that conditionwere included inthestatisticalanalysis.Thiswasdoneinordertokeeptheerrortermconstantacrossconditionsandtofulfiltherequirementsofhomogene-ityofvariance.However,inordertonotbiastheresultsbyarbitrarilyexcludinghalfofthetrialsforthemixed-gendercondition,theanaly-siswascarriedoutusingasoftwarethatrandomlyselectedhalfofthetrials fromtherelevant conditions foreach iterationof theanalysis.In thisway,weautomaticallygeneratedone-thousand randomsam-plesof50%ofthetrialsforthemixed-genderconditionandcomputedthe statistical differences among the three experimental conditionsone-thousand times, once for each randomsamplegenerated for themixed-gender condition. We will report the range, the mean, andthe standard deviation of F values, and the percentages of cases inwhichpvaluesweresignificant(p<.05)acrosstheiterationsofeachanalysis.Acontrastwillbeconsideredsignificantwhenthepvalueissignificanton95%oftheanalyses.Incaseofsignificance,two-tailedt-testcomparisonswereconducted.Ontheotherhand,meanRTsandpercentageoferrorrates,whenreported,refertothewholedataset.

2.2. Results

2.2.1. Singular Condition. A total of 3.6% of the trials in theexperiment across all participants were errors and were discarded.


291

Errors were equally distributed between feminine (3.5%) and mas-culinepictures(3.8%).Responselatenciesdidnotsignificantlydifferaccordingthegenderofthepicturenames(fem:604,mas:617,p>.8).

2.2.2. Plural Condition. A total of 21.4% of the trials in theexperimentacrossallparticipantswereerrorsandwerediscarded.Intheanalysisoferrors,therewasamaineffectofGenderCongruency(F(2,22)=14.07to32.3,meanF=21.40(±3.2),p< .05in100%ofcases).Post-hocpaired-samplet-testsrevealedthatparticipantsmademoreerrorsonfeminine-thanonmasculine-genderpairs(t(1,11)=6,p<.0001),oncongruent-femininegenderpairsthanonmixed-genderpairs (t (1,11)=1.3 to5.3,meant=2.9 (±0.6),p< .05 in88.3%ofcases),andonmixed-genderpairsthanoncongruent-masculinegen-derpairs(t(1,11)=-2.3to-11,meant=-5(±1.3),p<.05in100%ofcases).

In theanalysis of latencies, therewasa significantmain effectofGenderCongruency(F(2,22)=3.2to6.8,meanF=5.1(±0.5),p<.05in99.7%ofcases).Paired-samplet-testsshowedthatparticipantsweresloweronfeminine-thanonmasculine-genderpairs (t (1,11)=2.5,p= .03).Ontheotherhand,themixed-genderconditiondidnotsignificantly differ from either of the gender congruent conditions.Indeed,pvaluesturnedouttobesignificantonaminorityoftheanal-ysesperformedaccordingtotheprocedureexplainedabove(contrastmixed-gendervs. congruentmasculine-gender: (t (1,11)=0.7to -3.1,meant=-1.7(±0.5),p<.05in14.7%ofcases;contrastmixed-gendervs. congruent feminine-gender: (t (1,11) = 0.1 to 3.2, mean t = 2 (±0.5),p<.05in40.3%ofcases).ResultsaresummarizedinTable1.

Table 1. Mean RTs (with SE in parentheses) and error percentage observed inExperiments1-2fordifferentexperimentalcondition.

singular CondiTions plural CondiTions

F_SING(LA) M_SING(LO) F-F(LE) M-M(LI) MIXED(LI)

EXP1(RTs) 604(32.7) 617(38.1) 846(93) 754(61) 788(72)

EXP1(ERR) 3.5% 3.8% 38.1% 5.6% 21%

EXP2(RTs) 644(18.3) 677(24.7) 854(41) 827(47) 819(42)

EXP2(ERR) 4.1% 4.8% 30.6% 4.2% 17.3%


292

2.3. Discussion

The results of Experiment 1 demonstrate that participants areslowerandmorepronetoerrorwhenproducingfemininegrammati-cal gender, but only for multiple referents. Indeed, the congruent-feminineconditionwasthemost ‘difficult’conditionforparticipants,and themixed-gender conditionwasmoredifficult than the congru-ent-masculine condition. Importantly, thispatternwaspresentbothin the error and RT analyses. One objection that may be raised isthat, in Experiment 1, participants produced masculine-markedcliticsmoreoftenthanfemininemarkedclitics (i.e.,li-response> le-response).This isbecausetherewerethesamenumberoftrialspercondition (congruent-gender feminine, congruent-gender masculine,mixed-gender),andsotheresponseliwasrequiredon75%oftheplu-ral-picturetrials.Itmaybearguedthatthisdifferenceinthepropor-tionofresponsesthatweremarkedformasculineversusfeminineiswhatunderlies thedifferencesobserved inRTsanderrors. Inordertoaddressthisconcern,weranExperiment2,inwhichwebalancedtheproportionoftrialsinwhichparticipantsproducedmasculineandfemininemarkedclitics.

3. Experiment 2

ThisexperimentsoughttoreplicatethefindingsofExperiment1whilemaintainingthesameproportionofliandleresponses.

3.1. Method3.1.1. Participants. Twelve Italian speakers, recruited from the

samepopulationasinExperiment1,tookpartintheexperiment.3.1.2. Materials, Design, Procedure. Twenty-four pictures (half

masculine, half feminine) were selected and divided in pairs. Somepairs were the same as in the previous experiment, while someformed new pairs (see Appendix B). The name of the pictures didnot significantly differ with respect to length (mean length, m: 6.5;f:5.8)or frequencyasreported inBortoliniet al. (1971.Mean freq.,m: 23.3; f: 20.8). Pictures appeared alone (singular CondiTion), witha same-gender pictured noun (CongruenT-gender CondiTion), or witha different-gender pictured noun (mixed-gender CondiTion) in sepa-rate blocks. In addition, we included 12 filler trials, in which twofemininepicturesappeared.Thiswasdoneinordertoyieldthesameproportionofliandle responses.Therewerethusatotalof60trials,36 of which elicited plural responses. Li responses were elicited by


293

the masculine-gender condition (N = 6) and by the mixed condition(N = 12). Le responses were elicited by the feminine-gender condi-tion (N=6)andby the feminine-feminine filler trials (N=12).Theremainingtrials(N=24)elicitedsingularresponses,equallydividedbetweenmasculineandfeminine(i.e.,loandla).Allotheraspectsoftheexperimentaldesign,procedure,andstatisticalanalysiswerethesameasinExperiment1.Fillertrialswereexcludedfromthestatisti-calanalysis.

3.2. Results3.2.1. Singular condition.Following thesamecriteria fordeter-

mining errorsas inExperiment1, 4.5%of the trialswere removed.Errorswereequallydistributedbetweenfeminine(4.1%)andmascu-linepictures(4.8%;p≥.7).

In theanalysisof response latencies therewasamaineffectofPictureGender (F (1,11)=13,p< .005), showing that femininepic-tures(644ms)wererespondedtofasterthanmasculinepictures(677ms).

3.2.2. Plural condition.Discardeddataaccountedfor17.4%ofthedata.ThemaineffectofGenderCongruencywassignificant(F(2,22)=between5.6to15,meanF=9.4(±1.6),p<.05in100%ofcases).Asrevealed by post-hoc paired-sample comparisons, participants mademore errors on congruent-feminine than on congruent-masculinepairs(t(1,11)=4.4,p=.001).Therealsowasatrendformixed-gen-derpairstoelicitmoreerrorsthanmasculine-genderpairst(1,11)=-1to-4,meant=-2.5(±0.5),p<.05in69.4%ofcases).Thedifferencebetween mixed-gender pairs and feminine-gender pairs reached thespecifiedalphalevelonaminorityofcases(t(1,11)=0.7to3.9,meant=1.9(±0.6),p<.05in27%ofcases).Intheanalysisofnaminglaten-cies,theeffectofGenderCongruencywassignificantonlyon5.3%oftheiterations(F(2,22)=0.4to6.2,meanF=1.8(±0.9),p<.05in5.3%ofcases).Thus,wedidnotperformpost-hocanalysesonthecontrastsbetweenthemixed-genderconditionandeachcongruentcondition.

Since we were particularly interested in the contrast betweenthetwocongruentconditions(masculinevs. feminine),weconductedaplannedcomparisonbetweencongruent-gendermasculinepairsandcongruent-genderfemininepairs.Thedifferencewasmarginallysig-nificant(t(1,11)=2,p=.07).ResultsaresummarizedinTable1.

3.3. DiscussionTheresultsobservedinExperiment2convergewiththefindings

in Experiment 1. This means that the pattern initially observed inExperiment1 isnotduetothefactthatparticipants inthatexperi-


294

ment produced masculine-marked responses more often than femi-nine-marked responses. In both experiments, the congruent-genderfeminineandmasculineconditionswerethehardestandtheeasiestconditions respectively, with the mixed-gender condition in betweenthe two. Moreover, the same pattern of results was observed in theanalysesoferrorsandlatencies–thoughmarginally,forlatencies,inExperiment2.

It could be argued that the reported pattern of response timeeffectsisduetotheadoptionofaresponsestrategyonthepartofsub-jects.Specifically,itcouldbethatassoonasparticipantsrecognizedan object with a masculine name, they responded without furtherprocessingofthesecondobject.Thus,assoonastheyperceivedthattwo(insteadofone)objectswerepresented,andthatoneofthemwasmasculine, they interrupted further processing of the second objectand respondedwith themasculinepluralpronoun.On thisaccount,the difference between the masculine-congruent vs. mixed-genderconditions could be explained by assuming that participants ran-domlychosewhichobjecttolookatfirst.However,thisargumenthasdifficulty explaining the pattern of findings in the error rates. Thatargumentwouldpredictthataslongastwofeminineobjectsareiden-tified, the accuracy level would not differ with respect to the otherconditions.Contraryto thisprediction,errorrateswereconsistentlyhigher in the congruent-gender feminine condition compared to theotherconditionsinbothexperiments.

Inthelightoftheseconsiderations,weassumethattheobservedpattern of results does indeed reflect how agreement is realized innatural language situations. Do these findings reveal aspects ofcompetence or aspects of performance? Given the massive numberoferrorsproducedinthesimpletaskusedhere,itwouldseemmorereasonable to assume that they reflect competence and not merelyperformance factors. Indeed, the number of errors was surprisinglyhighinbothexperiments,andtheerrordistributioncloselymirroredthepatternofRTs.Thus,thefemininedisadvantagemostlikelyindi-catesthatspeakers’competenceonthisaspectofagreementdoesnotexactlyreflecttheprescriptiverulesofagreement.

Another finding that deserves attention is the result in theincongruent-gendercondition,roughlyintermediatebetweenthecon-gruent-masculine and the congruent-feminine conditions. This find-ing may be due to the experimental procedure: in order to producethe correct agreement form, people have to integrate the gender oftwoobjects,oneofwhichismasculineandoneisfeminine.Oneway


295

in which this integration may be thought of, is as a two-step proc-ess:peoplefirstsuppressthefemininegendervalueinordertothencorrectlyproducethemasculineform.Thesuppressionofthegendervalue would take time, thus slowing down the response and givingrise tomoreerrors incomparison to thecongruent-masculinecondi-tion.

Alternatively,itmayreflectthefactthatspeakers’competenceissuchthatitisrelativelypronetoacceptthewrongagreementmark-inginacondition–suchasthemixed-gendercondition-inwhichthewrongformrepresentsoneofthetwoobjectsinacombination.

Inordertobetterassesswhetherspeakers’behaviourwithcon-gruent-feminine and mixed-gender combinations is due to perform-ance or competence factors we carried out an off-line acceptabilityjudgementsstudy.

4. Experiment 3

Inthisexperimentwesetouttoexplorethemechanismsinvolvedin gender-marking by asking participants to judge the grammaticalacceptability of sentences violating gender agreement in differentconditions.Therationaleisthatagreementviolationsshouldreceiveahigheracceptabilityratingthemorethewrongagreementisconsid-eredacceptable ineachcondition.Conditionsvariedasafunctionofthegender(femininevs.masculine)andnumber(onevs. two)ofthereferents.Wealso includedasimplepluralcondition inwhichtherewasonlyonepluralreferent (feminineormasculine).Thisconditionallows us to directly compare agreement violations with conjoinedNPs(e.g.,il bicchiere e il cappello ‘theglassandthehat’)andagree-mentviolationswithpluralnouns(e.g.,i bicchieri‘theglasses’).Bothconditions require plural agreement, but only the first involves theintegration of different gender features across nouns. On the otherhand, agreement with a plural noun – as is the case for agreementwith a singular noun – involves only one lexical node. In this way,it is possible to verify whether the single plural lexical entries aretreated as the combination of two singular lexical entries, as far astheagreementprocessisconcerned.

If the difference between congruent-masculine, congruent-femi-nine and mixed-gender conditions observed in the previous experi-ments results from the application of an experimental strategy dueto time pressure, no differences among the conditions are predictedhere. In other words, in the measure to which such differences are


296

observed, we may conclude that they index speakers’ (competent)knowledgeaboutgenderagreement.

Twopossibilitiesareconsideredconcerningfeminineagreement.Accordingtothefirstone,theagreementmarkingoffemininepluralisintrinsicallymoredifficultcomparedtothepluralmasculineagree-ment marking. Therefore, agreement violations with conjunctionsof femininenouns (e.g., la bottiglia e la cravatta ‘thebottle:fem andthetie:fem’)andpluralfemininenouns(e.g.,le bottiglie ‘thebottles’)should pattern the same; namely, they should be tolerated morereadily than agreement violations for the corresponding masculineconditions.Accordingtothesecondpossibility,theproblemwiththefemininegenderliesintheintegrationprocess.Thatis,whentwodif-ferent femininereferentshave tobe integrated, the integrationruleforfeminineagreementislessstrongthanisdictatedbyprescriptiveItaliangrammar.Ifthiswerethecase,agreementviolationswithcon-junctionsoffemininenouns(e.g.,la bottiglia e la cravatta‘thebottle:femandthetie:fem’)shouldbetoleratedmorereadilythanagreementviolations with conjunctions of masculine nouns (e.g., il bicchiere e il cappello ‘the glass:mas and the hat:mas’). However, no differencein agreement violation ratings is predicted between plural feminine(e.g., le bottiglie ‘the bottles’) and masculine nouns (e.g., i bicchieri‘theglasses’).

4.1. Method

4.1.1. Participants. Forty-three Italianspeakersvolunteered fortheexperiment.Thirty-eightofthemwerehigh-schoolstudents(age:18-20)fromthePisaarea.

4.1.2. Materials. Fourteen correct sentences were created. Halfofthemexploredgenderagreementonverbs,theotherhalfexploredgender agreement on clitics.2 For four sentences there was singularagreement,while for tensentences therewaspluralagreement.Foreach number agreement, agreement on both verbs and clitics wasexplored.Fourdifferentagreementconditionswereused:1.singular CondiTion(infoursentences);2.simple plural CondiTion(infoursen-tences);3.Complex plural CongruenT CondiTion(infoursentences);4.Complex plural mixed CondiTion(intwosentences).Thefirstconditionexploresagreementonverbsorcliticswithasingularfeminine(e.g.,una bottiglia ‘abottle’)ormasculinenoun(e.g.,un quaderno ‘acopy-book’). The second condition explores agreement on verbs or cliticswithaplural feminine (e.g.,due bottiglie ‘twobottles’) ormasculine


297

noun(e.g.,due quaderni‘twocopybooks’).ThethirdconditionexploresagreementonverbsorcliticswithaconjunctionoftwosingularNPsof the same gender (e.g., una bottiglia e una forchetta ‘a bottle andafork’,bothfeminine,orun libro e un quaderno ‘abookandacopy-book’, both masculine). The fourth condition explores agreement onverbs or clitics with a conjunction of two singular NPs of differentgender(e.g.,una bottigliae un libro‘abottle:fem andabook:mas’).

Using the same syntactic structures we generated fourteen‘incorrect’sentencesforwhichtherewereerrorsingenderagreement.Number agreement was always preserved. Incorrect sentences usedthe same lexical items as correct sentences but those lexical itemswere arranged differently. This was done in order to use the samematerialsacrosscorrectandincorrectsentences;theotheradvantageof this approach is that participants cannot use strategies to judgethe validity of a presented sentence based on the mere presence ofa given word. For instance, the incorrect version of the sentence Il pinguino è balzato in mare ‘Thepenguin:mas jumped:mas in the sea’was Il pesce è balzata in mare‘Thefish:masjumped:feminthesea’,inwhichthesubjectnounismasculine,butthepastparticipleisincor-rectlymarkedwiththefeminineending.AlloftheincorrectsentencesarereportedinAppendixC.Theorderofthe28sentenceswasrand-omizedforeachparticipantwiththefollowingconstraints:1.Nomorethan three consecutive sentences corresponded to the same experi-mentalcondition;2.Therewerenomorethanthreeconsecutivecor-rectorincorrectsentences.

4.1.3. Procedure. Participants were presented with the 28 sen-tencesonaprintedsheet.

Theywereasked to rate thegrammaticalacceptabilityofeachsentenceonaseven-pointscale (from1: completelyunacceptable, to7: completely acceptable). Participants were encouraged to use theentire scale. They were also explicitly instructed to focus on gram-matical acceptability while neglecting any personal preference onsemanticgrounds.

4.1.4. Analysis.Thepreconditionforaparticipanttobeincludedin the analysis was that the mean rating for correct sentences behigherthanthemeanratingforincorrectsentences.Allparticipantssatisfiedthiscriterion.Overall,themeanratingforcorrectandincor-rect sentences was 6.1 [±1.2] and 2.2 [±1.2] respectively. Ratingsfor correct sentences were not considered any further. Scores forincorrect sentenceswereanalysedbyperforminganANOVAon thesubjects’meanratings for thesentenceswheretheverbor theclitichad to agree with a singular noun or two different nouns sharing


298

theirgenderfeature.Twovariableswereconsidered:CondiTion(withthreelevels:Singular;SimplePlural;Complexpluralcongruent)andnoun(s) gender(withtwolevels:Femininevs.Masculine).Two-tailedt-testcomparisonswerealsoperformedbetweenthemeanratingsforthecomplexagreementconditions:Complex plural mixed CondiTionvs.Complex plural feminineCondiTionvs.Complex plural masCuline Con-diTion.

4.2. Results

Arepeated-measuresANOVAshowedasignificantmaineffectofCondition(F(2,41)=15.8,p< .00001).T-Testcomparisonsrevealedthat the Complex plural congruent condition (2.84) was rated, onaverage,asmoreacceptablethantheSingular(2.23;t(1,85)=3,p<.003)andtheSimplepluralcondition(2.07;t(1,85)=5.2,p<.00001).On the other hand, the Singular and the Simple plural conditionswere not significantly different (t = -1.18, p > .1. See Table 2 andFigure1).TherealsowasamaineffectofNoun(s)Gender(F(1,42)=10.3,p<.003),showingthatincorrectagreementonfemininenouns(2.59) was rated as more acceptable than incorrect agreement onmasculinenouns(2.18).Thesefindingsarequalifiedbyasignificanttwo-way interactionCondition*Noun(s)Gender (F (2,41)=13,F<.00001).Thisinteractionshowsthatthedifferencebetweenratingsonfeminineandmasculinenounsismodulatedbycondition.T-testcom-parisons revealed that, in the Complex plural congruent condition,incorrectagreementonfemininenounswasratedasmoreacceptablethanincorrectagreementonmasculinenouns(3.53vs.2.15;t(1,42)= 5.9, p < .00001). On the other hand, no difference between incor-rectagreementonfeminineandmasculinenounswasobservedintheSingular(2.06vs.2.4;t(1,42)=-1.5,p>.1)orintheSimplepluralConditions(2.16vs.1.98;t(1,42)=1.1,p>.1).TheComplexpluralmixedconditionwasthencomparedwiththeothertwoComplexplu-ralconditions(feminineandmasculine)bymeansofseparatepairedt-Test. Results showed that incorrect agreement on mixed-gendernouns was rated as more acceptable than incorrect agreement ongender congruent masculine nouns (2.71 vs. 2.15; t (1, 42) = -2.8, p= .007), but as less acceptable than incorrect agreement on gender-congruentfemininenouns(2.71vs.3.53;t(1,42)=3.7,p=.0007.SeeFigure2).


299

Table 2.Participants’meanratings(withS.E.inparentheses)forsentencesviola-tingagreementacrossdifferentexperimentalconditions(Experiment3).

Sing 2.06(0.20)Feminine Simple pl 2.16(0.20)

Complex pl 3.53(0.24)Sing 2.41(0.23)

MasCuline Simple pl 1.98(0.19)Complex pl 2.15(0.19)

Mixed gender Complex plural 2.71(0.22)

Figure 1. Participants’ mean ratings for sentences violating gender agreement(Experiment3).ViolationsoffeminineagreementaretoleratedsignificantlymorereadilythanviolationsofmasculineagreementonlyinthecaseofconjoinedNPs(complexplurals).Themixed-condition isnotconsidered.Errorbarsdepictstan-darderrorsofthemeans.

Figure 2.Participants’meanratingsforincorrectsentencesrequiringagreementwith complex plurals. (Experiment 3). Mean acceptability ratings for incorrectagreementinthemixedconditionaresignificantlyhigherthaninthecongruent-gender-masculinecondition,andsignificantlylowerthaninthecongruent-gender-femininecondition.Errorbarsdepictstandarderrorsofthemeans.


300

4.3. Discussion

Themainfindingsofthepresentexperimentcanbesummarizedasfollows:

a. Agreementviolationswith conjunctionof feminineNPsweretolerated significantly more readily than agreement violations withconjunctionofmasculineNPsorconjunctionofNPsofdifferentgen-der. On the other hand, no difference between simple feminine andsimplemasculinepluralswasobserved.

b. AgreementviolationswithconjunctionofNPsofdifferentgen-der were tolerated significantly more readily than agreement viola-tionswithconjunctionofmasculineNPs.

Themerefactthatacceptabilityratingsvariedacrossconditionsis at odds with the view that the results from the previous experi-mentsareexclusivelyduetoanexperimentalstrategyrelatedtotimepressure.

Both findings (a) and (b) may be assumed to reflect speakers’competence on aspects of gender agreement. As to finding (a), thecontrast between complex and simple feminine plurals should beparticularly emphasized. This finding rules out the hypothesis ofan intrinsic greater difficulty for the plural feminine marking. Ontheotherhand,itsupportsthehypothesisthatthegreaterdifficultyassociated with processing feminine gender lies in the integrationprocess (i.e., in the process that integrates information from morethanonelexicalnode).Specifically,thesefindingssuggestthatwhentwodifferentfemininereferentshavetobeintegrated,theresolutionruleforfeminineagreementisnotsostrongastocompletelyexcludethepossibilityforthemasculineagreementmarking.Withrespecttofinding(b),itisworthmentioningthattherelativeorderoffeminineandmasculinenounsinmixedconjunctionsmayaffectparticipants’acceptabilityandproductionofnoun-verbagreement.Thereisinfactindependentevidencesuggestingthatthecloser(withinasentence)that a noun appears next to a verb, the higher is the probabilitythat the noun evokes incorrect agreement on the verb (Haskell &MacDonald(2005;butseeVigliocco&Nicol(1998)andFrancketal.(2002)forevidenceagainstaroleofnoun-verbproximityininfluenc-ingagreement).Ourresultsdonotdirectlybearontheissueofnoun-verb proximity, as the feminine noun was always the first noun inmixed conjunctions– that is, thenounmoredistant from theverb.However, if anything, we may have underestimated the extent towhichItalianspeakerstoleratefeminineagreementwithaconjunc-tionofnounshavingdifferentgenders.Inotherwords,inthemeas-


301

uretowhichproximitymayinfluencejudgmentsofacceptability,onewouldexpectthatincorrectfeminineagreementwouldberatedevenmore acceptable if the feminine noun were the noun nearer to theverbinaconjunctionofmixed-gendernouns.

5. General Discussion

Theobjectiveofthisstudywastoexplorethemechanismofgen-deragreementwithpairsofpicturedobjectsorconjunctionsofwrit-tennouns; inparticular,weaskedwhethertheoccasionalreportsoftheover-generalizationofthemasculineformevenwhenthefeminineformwouldhavebeenrequiredbyrule,isduetocompetence-orper-formance-related reasons. The results from the on-line experimentsand the off-line task reported in this study converge to indicate acompetence-basedprocess.

Themain findings canbesummarizedas follows:1.Peoplearelessaccurateandslowertoproduce,aswellasmoreinclinedtotoler-ateviolationsincomprehension,forconjunctionsoffemininereferents(e.g.,la bottiglia e la tazza‘thebottle:femandthecup:fem’)comparedto conjunctions of masculine or mixed-gender referents. 2. Peopleare lessaccurateand slower toproduce (inExp. 1) aswell asmoreinclined to tolerate violations in comprehension, for conjunctions ofmixed-genderreferents (e.g., la bottiglia e il bicchiere ‘thebottle:femandtheglass:mas’)thanforconjunctionsofmasculinereferents(e.g.,il libro e il bicchiere‘thebook:masandtheglass:mas’).Additionally,asisevidencedbytheacceptabilityratings,thedifferencebetweenfemi-nineandmasculinegenderisrestrictedtothecaseofconjoinedNPs:There isnodifference in theacceptabilityof incorrectgenderagree-mentbetweenmasculineandfeminineplurals(e.g.,due bottiglie‘twobottles:fem’vs.due bicchieri‘twoglasses:mas’).

Wehavearguedthattheconvergenceofon-lineandoff-linedataconstitutes support for the competence hypothesis outlined in theIntroduction. Alternatively, one may argue that the masculine gen-derisover-extendedjustbecauseitisthedefaultvalueintheItaliangender system. This may be easily appreciated if we consider thosenounsthatallowforbothgendervalues.Sincethosenounstypicallyrefertonaturallysexedentities,thechoiceofthespecificvalueusu-ally depends on the semantic gender of the referent. For instance,the Italianword for ‘friend’ isamico oramica dependingon thesex– male or female, of the referent. Importantly, whenever the sex ofthe referent is left unspecified, the masculine form is used. Note,


302

however, that the notion of default has to be applied on a case-by-casebasis (Corbett,p.c.).That is, the fact thataparticularmorpho-logicalrealizationisthedefaultinagivencase,doesnotentailthatitisadefaultinallcases.Thisthusreducestosayingthatthemerefact that themasculinegendermayworkas thedefault in the spe-cificcaseofamico/amica‘friend:mas/fem’,doesnotentailthatitmustworkasthedefaultgenderinalltheothercases.

For the agreement phenomena considered here, the use of themasculineagreementwithconjunctionsoffemininenounswasneverpermitted,atleastaccordingtothegrammarofStandardItalian(cf.Dardano&Trifone (1997:201), in relation toAdjective-Nounagree-ment (free translation): “When it refers to a conjunction of nounshaving the same gender, all singular, all plural, or some singularsandsomeplurals,theadjectivetakesthegenderofthenounsandisusually plural”). Since gender agreement is always mandatory, theover-extension of the masculine gender cannot be attributed to theuseofadefaultvalue.This,however, cannotbe thewhole story,asthedataalsodocumentan inclinationtowardstherepresentationofthegenderofeachreferent in theresponse (relativeacceptabilityofthewrongagreementwithmixedreferents).Thisisalsoatoddswiththedictataofthegrammar:cf.againDardano&Trifone(1997:201),in relation to Adjective-Noun agreement (free translation): “Whenitreferstoaconjunctionofnounshavingdifferentgenderandnum-bers, theadjective isusuallymasculineplural”.SeealsoSerianni&Castelvecchi(1988:169;355)onthispoint).Thismeansthattherulefor gender agreement in its complex cannot be reduced to the newcreationofadefault.

OurresultsmayreflectthebeginningoftheerosionofthesystemforgenderagreementinItalian.Thishypothesiswouldfitnicelywiththecommonlyacceptedviewthatlanguagechangesstartfromperiph-eralcases.Thereasoningwouldgoasfollows.Ourresultsdonotalignwithrespect to theapplicationofaruleonly inthecaseofmultiplereferents(e.g.,themisuseofthemasculineagreementwithfemininenounsisrestrictedtothecaseofconjunctionsoffemininenouns).Onthereasonableassumptionthatgenderisapropertyoflexicalnodes,in the case of agreement with a single noun the gender feature isinheritedfromasinglelexicalnode.Ontheotherhand,inthecaseofagreementwithaconjnctionofnouns,thegenderfeatureisinheritedfromthevaluethatarisesduetotheapplicationofaruleregardingtheintegrationofallthenounsintheconjunction.Whereasthefirstcaseisassumedtobethecentralorprototypicalcaseofgenderagree-ment–asitislexically-dependent,thesecondcaseisassumedtobe


303

theperiphericalcaseofgenderagreement–asitcannotberesolvedwithinasinglelexicalnode,butrequirestheconsiderationsoftwoormorelexicalnodes.Thus,ifaprocessoferosionofthegenderagree-mentsystemisinfactoccurringinItalian,onereasonablehypothesisisthatitstartedfromoutsideofthelexicalboundaries,thatis,fromthe‘peripheral’casewherethefeaturesofdifferentlexicalnodeshaveto be integrated into the value-triggering agreement. One way totacklethehypothesisofgendererosioncouldbetotestpeopleofdif-ferentageranges: ifolderpeopletoleratedthemasculineagreementwith multiple feminine referents, and the feminine agreement withmixed-gender referents less readily than young people, this wouldspeak in favorof thehypothesisofa languagechange intheItaliangendersystem.

An alternative hypothesis would be that the language changespecifically concerns the functionorusage of Italiangender-markedpronouns, and not the general underlying system of gender agree-ment.However,werethisthecase,wewouldhaveexpectedadiffer-enceintheacceptabilityofviolationsofagreementdependingontheword (verb or pronoun) on which gender is marked. The results ofExperiment3donotsupportthisprediction.

Leavingasidethehypothesisofgendererosion,agreementwithmultiplereferentsappeartobesensitivetotwodistinctkindsofpres-sure:ontheonehand,abiastowardsthemasculineresponse,andontheotherhand,aninclinationtowardstherepresentationofthegen-derofeachreferentintheresponse.Thismeans,inotherwords,thatthere is confusion regarding the agreement rules that operate overconjunctionsofnouns,asisrevealedbythefactthatItalianspeakersdonotapplytheruleconsistentlyexceptwhenthenounsinthecon-junctionsarebothmasculine.

Itmightbeachallengingissueforfutureresearchtobetterchar-acterizethenatureoftheinconsistencyintheapplicationofaruleasafunctionofcontext.Forinstance,itwouldbeinterestingtoexplorethe limits of the generalization of the masculine gender to multiplefemininereferents.Cantheon-lineactivationof incorrectmasculineagreementforms(orthepossibilityforincorrectmasculineagreementinanoff-linetask)bemodulatedbytheabsolutenumberoffemininereferents,orbythecombinationofnaturalandgrammaticalgender?Forinstance,onecouldmakethehypothesisthatwhenthereferentshavebothgrammaticalandnaturalfemininegender(e.g.,la nonna e la zia‘thegrandmotherandtheaunt’),peoplearelesspronetoacceptthemasculineagreementthanwhenthereferentsareonlygrammati-callymarked(e.g.,la bottiglia e la tazza ‘thebottle:femandthecup:


304

fem’).ThispredictioniscorroboratedbydatafromSerbian/Croatian/Bosnian (e.g., Wechsler & Zlatić, 2000; 2003). In this language, thegenderresolutionrulewouldbesimilartotheItalianone,exceptthatSerbian/Croatian/Bosnian has a three-way gender system (mascu-line/feminine/neuter).Thefeminineisrestrictedtothecaseinwhichallofthenounsinaconjunctionarefeminine,whereasthemasculinehas tobeused inall other cases.Amore carefulanalysis,however,hasrevealedthatthemasculineformmaybeusedevenwhenallthenounsinaconjunctionarefeminine,unlesstheyarealsomarkedfornaturalgender(Corbett,1991:299-303).Whennounsarebothgram-maticallyandsemanticallymarkedfor feminine, themasculinecan-notbeusedasadefault.

Similarquestionscouldalsobeaddressedwithmixed-genderref-erents.Doesthepossibilityforfeminineagreementvaryasafunctionoftheabsolutenumberoffemininereferentsinaconjunction,ortherelativeproportionofmasculineandfemininereferentsinaconjunc-tion,or,even,therelativeorderofmasculineandfemininereferents(e.g.,la bottiglia e il bicchiere ‘thebottle:femandtheglass:mas’vs.il bicchiere e la bottiglia‘theglass:mas andthebottle:fem’)?Onereason-able hypothesis could be that the probability for the misselection ofthefemininegenderincreaseswiththe(absoluteorrelative)numberoffemininereferentsinamixed-genderconjunction.

Toconclude,ourfindingsmaysetthestageforfutureresearchinmultipledirections:thefirst,morediacronically-oriented,focusesonabettercharacterizationofinstabilityinthegenderagreementsystemin Italian; the second, more synchronically-oriented, focuses on theinterplaybetweengenderagreementandothercontextualfactors.

Address of the first Author:

Chiara Finocchiaro, Center for Mind/Brain Sciences - CIMeC,UniversityofTrento,PolodiRovereto,CorsoBettini,31,38068Rovereto(TN),Italy,[email protected]


305

Appendix A. Materials used in Experiment 1. The pictures in thefirsttwocolumnswerepairedwiththepicturesintheothertwocolumnsofthesamerowondifferenttrials.

FEMPICs(1) MASPICs(1) FEMPICs(2) MASPICs(2)banana‘banana’ fungo‘mushroom’ zucca‘pumpkin’ carciofo‘artichoke’bilancia‘balance’ vaso‘vase’ pistola‘gun’ piatto‘plate’foca‘seal’ topo‘mouse’ giraffa‘giraffe’ pinguino‘penguin’foglia‘leaf’ fiore‘flower’ carota‘carrot’ rastrello‘rake’gonna‘skirt’ guanto‘glove’ cravatta‘tie’ cappello‘hat’mucca‘cow’ serpente‘snake’ farfalla‘butterfly’ canguro‘kangaroo’pipa‘pipe’ fiocco‘bow’ bottiglia‘bottle’ coltello‘knife’ruota‘wheel’ ombrello‘umbrella’ campana‘bell’ ventaglio‘fan’scarpa‘shoe’ pettine‘comb’ chitarra‘guitar’ lucchetto‘padlock’sedia‘chair’ tavolo‘table’ chiave‘key’ casco‘helmet’tenda‘tent’ quadro‘painting’ valigia‘luggage’ orologio‘watch’tromba‘trumpet’ tamburo‘drum’ racchetta‘racket’ libro‘book’

Appendix B. Materials used in Experiment 2. The pictures in thefirsttwocolumnswerepairedwiththepicturesintheothertwocolumnsofthesamerowondifferenttrials.

FEMPICs(1) MASPICs(1) FEMPICs(2) MASPICs(2)banana‘banana’ fungo‘mushroom’ zucca‘pumpkin’ carciofo‘artichoke’bottiglia‘bottle’ coltello‘knife’ pipa‘pipe’ fiocco‘bow’giraffa‘giraffe’ pinguino‘penguin’ foca‘seal’ topo‘mouse’foglia‘leaf’ fiore‘flower’ carota‘carrot’ rastrello‘rake’valigia‘luggage’ orologio‘watch’ tenda‘tent’ quadro‘painting’sedia‘chair’ tavolo‘table’ chiave‘key’ casco‘helmet’


306

Appendix C. Sentences violating gender agreement used inExperiment3.

C’èunabottiglia.Portalo. ‘Thereisabottle:fem.Bringit:mas.’Lapanteraèfuggitodaicacciatori. ‘Thepanther:fem escaped:masfromthehunters.’C’èunquaderno.Prendila. ‘Thereisacopybook:mas.Takeit:fem.’Ilpesceèbalzatainmare. ‘Thefish:masjumped:fem inthesea.’Cisonodueforchette.Portali. ‘Therearetwoforks:fem. Bringthem:mas.’Legiraffesonofuggitidaicacciatori.

‘Thegiraffes:fem escaped:masfromthehunters.’

Cisonoduequaderni.Prendile. ‘Therearetwocopybooks:mas.Takethem:fem.’Ipinguinisonobalzateinmare. ‘Thepenguins:masjumped:fem inthesea.’Cisonounaforchettaeunabottiglia.Portali.

‘Thereareafork:femandabottle:fem.Bringthem:mas.’

Lagiraffaelapanterasonofuggitidaicacciatori.

‘Thegiraffe:fem andthepanther:fem escaped:masfromthehunters.’

Cisonounlibroeunquaderno.Prendile.

‘Thereareacopybook:masandabook:mas.Takethem: fem.’

Ilpinguinoeilpescesonobalzateinmare.

‘Thepenguin:masandthefish:masjumped:fem inthesea.’

Cisonounagommaeunquaderno.Prendile.

‘Therearearubber:femandacopybook:mas.Takethem: fem.’

Lalontraeilpinguinosonobalzateinmare.

‘Theotter:femandthepenguin:masjumped: feminthesea’.

Notes

* We would like to thank Luigi Lombardi and Francesco Vespignani for theirassistancewiththestatisticalanalysis.B.Z.MahonwassupportedinpartbyanNSFGraduateResearchFellowship.Theresearchwassupported inpartby theFondazioneCassadiRisparmiodiTrentoeRovereto.1 Italiancliticscanprecede(i.e.,theyareproclitics)orfollow(i.e.,theyareencli-tics)theverb(cf.lo porto‘Ibringit:mas’vs.portalo‘bringit:mas’,dependingonthegivenverbform(finitetensesyieldproclisis,whereasnon-finitetensesandaffir-mative imperatives yield enclisis). In this study, only enclitic object forms wereused(e.g.,portalo‘bringit:mas’).2 VerbsarenotusuallymarkedforgenderinItalian.Therearethreeexceptions:1)verbsrequiringtheauxiliaryverbessere‘tobe’whenusedinthepresentper-fect.Inthiscase,thepastparticipleofthemainverbhastoagreewiththesubject(e.g., la mia amica è partita ‘my friend:fem left:fem’); 2) all verbs when used inpassiveanalyticforms(e.g.,la porta è aperta dalla segretaria ‘thedoorisopenedbythesecretary’).Asinthepreviouscase,agreementiswiththesubject;3)tran-sitiveverbsrequiringtheauxiliaryverbavere ‘tohave’whenusedinthepresentperfectwithdirectobjectcliticpronouns.Inthiscase,themainverbhastoagreewith the direct object (e.g., le ho prese ‘I took [fem] them [fem]’). In this experi-ment,onlythefirstconditionofverbagreementwasconsidered.


307

Bibliographical References

BorTolini Umberto, Carlo Tagliavini & Antonio Zampolli 1971. Lessico di Frequenza della Lingua Italiana Contemporanea. Milano:Garzanti.

CohenJonathanD.,BrianmaCWhinney,MatthewflaTT&JeffersonprovosT1993. PsyScope: A new graphic interactive environment for designingpsychologyexperiments.Behavioral ResearchMethods, Instruments and Computers 25. 257-271.

CorBeTTGrevilleG.1991.Gender.Cambridge:CambridgeUniversityPress.CorBeTT Greville G. 2006. Agreement. Cambridge: Cambridge University

Press.dardanoMaurizio&PietroTrifone1997.La nuova grammatica della lingua

italiana.Milano:Zanichelli.franCkJulie,GabriellaviglioCCo&JanetL.niCol2002.Theroleofsyntactic

treestructureandcomplexityinsubject-verbagreement.Language and Cognitive Processes17.371-404.

haskell Todd R. & Maryellen maCdonald 2005. Constituent structure andlinearorder in languageproduction:Evidencefromsubjectverbagree-ment. Journal of Experimental Psychology: Learning, Memory and Cognition 35.891-904.

raaijmakers Jeroen G.W., Joseph M.C. sChrijnemakers & Frans gremmen1999. How to deal with ‘the language-as-a-fixed-effect fallacy’: com-monmisconceptionsandalternative solutions.Journal of Memory and Language 41. 416-426.

salviGiampaolo&Lauravanelli2004.Nuova grammatica italiana.Bologna:IlMulino.

serianni Luca & Alberto CasTelveCChi 1988. Grammatica italiana. Italiano comune e lingua letteraria.Torino:Utet.

viglioCCoGabriella&JanetL.niCol1998.Separatinghierarchicalrelationsandwordorder in languageproduction. Isproximity concord syntacticorlinear?Cognition 68.13-29.

WeChslerStephen&LarisaZlaTić2000.Atheoryofagreementanditsappli-cationtoCroatian.Language 76.799-832.

WeChsler Stephen & Larisa ZlaTić 2003. The many faces of Agreement,Stanford:CSIPublications.

gender agreement and multiple...

Documents