Representations and Processes in the Production of

Journal of Memory and Language41, 281–301 (1999)Article ID jmla.1999.2649, available online at http://www.idealibrary.com on

Representations and Processes in the Production of Pronouns:Some Perspectives from Dutch

Antje S. Meyer

Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands

and

Kathryn Bock

University of Illinois

The production and interpretation of pronouns involves the identification of a mental referent and,in connected speech or text, a discourse antecedent. One of the few overt signals of the relationshipbetween a pronoun and its antecedent is agreement in features such as number and grammaticalgender. To examine how speakers create these signals, two experiments tested conceptual, lexical,and morphophonological accounts of pronoun production in Dutch. The experiments employedsentence completion and continuation tasks with materials containing noun phrases that conflicted oragreed in grammatical gender. The noun phrases served as the antecedents for demonstrativepronouns (in Experiment 1) and relative pronouns (in Experiment 2) that required gender marking.Gender errors were used to assess the nature of the processes that established the link betweenpronouns and antecedents. There were more gender errors when candidate antecedents conflicted ingrammatical gender, counter to the predictions of a pure conceptual hypothesis. Gender marking oncandidate antecedents did not change the magnitude of this interference effect, counter to thepredictions of an overt-morphology hypothesis. Mirroring previous findings about pronoun compre-hension, the results suggest that speakers of gender-marking languages call on specific linguisticinformation about antecedents in order to select pronouns and that the information consists ofspecifications of grammatical gender associated with the lemmas of words.© 1999 Academic Press

Key Words:language production; pronouns; grammatical agreement.

ththeenbaar

nc-romis-oniveon,ayof

rk,

ter-ngtheaf-

ttleacy.set,theare

supnda01ax

andmaforthrf an

tjetbusail:te,1.

Pronouns can be found in all languages ofworld. In English and Dutch they are amongmost commonly used words in print (BaayPiepenbrock, & van Rijn, 1993) and are probly even more frequent in speech. They

Order of authorship is arbitrary. The research wasported in part by grants from the National Science Foution (SBR 94-11627), the National Institutes of Health (RHD21011), and visiting research fellowships from the MPlanck Society. We thank Femke van der MeulenMarianne Wissink for their assistance in preparing theterials and carrying out the experiments, Don Mitchelldiscussions about relative-clause comprehension, andanonymous reviewers for their constructive critiques oearlier version of the manuscript.

Address correspondence and reprint requests to AnMeyer, Max Planck Institute for Psycholinguistics, Pos310, 6500 AH Nijmegen, The Netherlands ([email protected]) or Kathryn Bock, Beckman InstituUniversity of Illinois, 405 N. Mathews, Urbana, IL 6180E-mail: [email protected].

281

e

,-e

fundamental to both the structure and the fution of language. They are nonetheless far fsimple in their conditions of use, either lingutically or cognitively, in either comprehensior production. One testimonial to their cognitcomplexity comes from language acquisitiwhere children’s problems with pronouns mpersist well past the point at which the usecommon nouns is firmly established (Cla1978; Clark & Sengul, 1978).

For adult listeners, the challenges of inpreting pronouns have to do with identifyipronominal referents and antecedents fromminimal feature specifications that pronounsford. In English, these features include libeyond number, natural gender, and animSome languages add more features to thenotably grammatical gender, but becausesparse meaning specifications of pronouns

--

-

ee

S.

0749-596X/99 $30.00Copyright © 1999 by Academic Press

All rights of reproduction in any form reserved.

hally

proowushapricsunry

rticak

awimes ihispro

pleto

peexriendta-lylopproage-nsbo

my tenreshiswieilkg.,as

p rre

s for-e.g.,nd

reun’sthattlyc-

h or

ionheffer-th an-orewithatesin-

ts

orypre-es

dyor,

t ofwhoege.f the

ation8)en-

rpre-ts are

of5).lesse-g athatticuny,

282 MEYER AND BOCK

what sets them apart and gives them so muctheir pragmatic value, they remain semanticimpoverished.

For speakers, the cognitive challenges ofnouns are markedly different. Speakers knthe referents of the personal pronouns theyso they have no uncertainty about who or wthe pronouns denote. Speakers are also thecipal beneficiaries of the cognitive economof pronouns. Relative to other words, pronoare efficient: Being very frequent and veshort, they are easily retrieved and easily aulated. The challenges they present to a spehave much more to do with deciding whenpronoun can successfully be used to alloparticular listener at a particular place and tto pick out the referent that the speaker hamind (Levelt, 1989). What remains once thurdle is crossed is merely to select the appriate pronoun from the lexicon.

Although pronoun selection is surely simin comparison to what speakers must dodetermine when a pronoun is felicitous, the scifics of the selection process are largely unplored. These specifics are integral to theoof lexical encoding in language production ainstructive for theories of pronoun interpretion in language comprehension. Accordingour focus in the present work was to deveand test alternative hypotheses about thecesses of pronoun selection during languproduction. We begin with an overview of rlated work on the interpretation of pronoubefore turning to the questions to be asked atheir production.

THE COGNITIVE UNDERPINNINGS OFPRONOUN INTERPRETATION

One consistent finding about pronoun coprehension is that listeners and readers tridentify candidate antecedents when theycounter pronouns or other anaphoric expsions. In one of the first efforts to explore tprocess, Chang (1980) presented readerssentences such asJohn and Mary went to thgrocery store and he bought a quart of mfollowed immediately by a probe word (e.John). The participants indicated as quickly

ossible whether the probe word had occu

of

-

e,tin-

s

-er

a

n

-

--s

,

-e

ut

-o--

th

d

in the sentence. Relative to control sentencewhich the probe wordJohn was not the antecedent of the pronoun in the second clause (John and Mary went to the grocery store ashe bought a quart of milk), responses wefaster when the probe represented the pronoantecedent. Subsequent work has shownantecedent reactivation can follow prompupon the occurrence of an anaphor (Dell, MKoon, & Ratcliff, 1983; McKoon, Gerrig, &Greene, 1996) without an intervening searcretrieval process.

Depending on the nature of the informatthat is reactivated in memory, facilitation of tresponse to a probe may be explained in dient ways. The probe might make contact wiphonologically intact, verbatim trace of the atecedent referring expression, or with a mabstract lexical–semantic representation, ora conceptual representation that incorporfeatures of the intended referent itself. Forstance, in the sentenceAfter the psycholinguissent their son to Princeton, they. . . . the oc-curence of the pronounthey might prompt thereactivation of a morphophonological memtrace of the phrase or a lexical–semantic resentation including information along the linof “noun [plural]: denoting those who stulanguage from a psychological perspective”if the listener grasps the real-world referenthe expression, a mental image of a coupleseem too young to be sending a son to coll

Some evidence that the representation oantecedent for a pronoun is fairly close toreferential, mental-model based representacomes from work by Cloitre and Bever (198and Gernsbacher (1991). This kind of represtation has been dubbedconceptual.We willassume that when comprehension and intetation are successful, conceptual antecedeneventually recovered during the resolutionpronominal anaphora (Bock & Brewer, 198However, conceptual antecedents may belikely to form the representation that is immdiately created upon first hearing or readinpronoun. A variety of evidence suggestsinitial processing makes use of the linguisfeatures of the context in which the pronoappears (Cowart & Cairns, 1987; Murph

o-

ms olatenIn

r oo

t ana mt -m theg napo eenz tw agSn ouni -i la-t hats ntc tivp nau

er-m icha ar ndq chT allya ickb tuaa ta thp tp res thr anp . As s oi thed abl

n,G ron uir

t redm thec con-tt cec Thew rm rals( fc in-g ion( s”t ry).E een,G c-t ro-n ons allyi oob unn thep

rmal,G elec-t ionc icalg nticp iono edi nticf n aj ands fer-e

thek lt ino heyw foru dis-c denti cifyt ablyb ualr le-v in-c con-

283PRODUCING PRONOUNS

1985; Nicol, 1988). This kind of pronoun prcessing is commonly said to involvesurfaceanaphora.

Languages with pronouns that mark gramatical gender allow strong competitive testsurface and conceptual processing. In suchguages, the grammatical features of nounsto be reflected in the forms of pronouns.Dutch, for example, the grammatical gendethe nounvrouwtje (little woman) is neuter, shat the pronouns for which it serves asntecedent are likely to be neuter. At the sa

ime, becausevrouwtje normally refers to feales, the natural gender is different fromrammatical gender, allowing conceptual ahors to differ from surface anaphors. InHetude vrouwtje dat door het bos liep droegware tas. Zij was. . . (The little old lady thaalked through the forest carried a heavy bhe was . . .) therelative pronoundat (that) iseuter, while the personal nominative pron

ntroducing the next sentence,zij (she), is femnine. Accordingly, the antecedent of the reive pronoun in this instance is a word thares its grammatical gender (a surface aedent), while the antecedent of the nominaronoun is a mental referent that shares itsral gender (a conceptual antecedent).Exploiting a similar property in other gendarking languages, Garnham, Oakhill, Ehrlnd Carreiras (1995; see also Cacciari, Ceiras, & Cionini, 1997) examined reading auestion-answering in Spanish and Frenheir findings suggested that readers initiccessed a surface antecedent, but fairly quegan to call on the properties of a concepntecedent that influenced the time takennswer a question about the contents ofassages that they had read. Garnham eroposed two alternative accounts of theseults. One preserves a distinction betweenepresentations that surface and conceptualhors consult, along the lines implied aboveecond, however, suggests that the two kindnformation are combined into a model ofiscourse and can be drawn on interchangeIn support of this second suggestioarnham et al. (1995) noted a feature of pominal reference in circumstances that req

-fn-d

f

e

-

.

e-et-

,r-

.

lyl

oeal.-ea-

f

y.

-e

he antecedent to be found from the shaental or environmental common ground of

onversation rather than from the discourseext. This kind of reference can be termeddeic-ic. The relevant property of deictic referenan be seen in an example from English.ord pants is grammatically plural, like otheembers of the class of summation plu

scissors, binoculars,etc.). Yet the piece olothing the word denotes is conceptually sular when there is only one item in questforcing the peculiar locution “a pair of panto denote a singleton member of the categoven in unheralded uses (in the sense of Grerrig, McKoon, & Ratcliff, 1994) and in dei

ic uses, with no linguistic antecedents, the poun may be plural. Imagine a man tryingome new trousers, inspecting himself criticn a mirror and saying to the clerk “They’re tig, aren’t they?” Clearly, the plural pronoeed not come from discourse reference tolural wordpants.Because these uses are natural and noarnham et al. (1995) suggested that the s

ion of pronouns during language productalls on a representation in which grammatender is intimately connected with semaroperties. Accordingly, the surface expressf pronouns in production may be determin

nterchangeably by grammatical and semaeatures. In turn, comprehension may call ooint representation of surface grammaticalemantic properties to resolve anaphoric rences.

SELECTING PRONOUNSFOR PRODUCTION

Our research was designed to exploreinds of representations that speakers consurder to determine the form of the pronoun till use. We will assume that the conditionssing a pronoun have been met within aourse context, where the pronoun’s anteces something previously mentioned. To spehe appropriate pronoun, speakers presumegin with information about the concepteferent that they intend to pick out. The reant features for such a specification maylude various pragmatic features as well as

turonlin

neimtuticoryapro

ouenedureof

theke

at-a

hiandor

ncek, aing

bte

amas

toe -

sinoun

heomeral

g -t ter-m g fi st tha ul gi-

n ter-m hewp dew

he-sph tivep Le-v ualh tifiest n oft di-r vantf ther n se-l im-p edr outt ati-c d toi s re-fl d al n-t het singc iona lts am in-c heirp ofe thee ion,w

erieso theg ing,p te-c onlyt om-p ter-m eo thattn is am ine

284 MEYER AND BOCK

ceptual features (such as natural gender, nanumber, and so on). In addition to these cceptual features, speakers may also call onguistic properties of thedefault categorizatioof the referent, as in thepantsexample abov(Bosch, 1986; Garnham et al., 1995). Thisplies that speakers determine the concepcategory of the referent, access the grammafeatures of the basic-level term for the categand employ those features in selecting anpropriate pronoun. Because this route to pnoun selection requires no information abwhether or how a referent was previously mtioned, it would serve for producing unheraldand deictic pronouns as well as offering a pconceptually driven route for the productionanaphoric pronouns.

Alternatively, when proceeding beyondconceptual representation of a referent, speamay call directly or indirectly on the grammical features of the referent’s categorizationestablished within the discourse context. Twould mean consulting a syntactically relevrepresentation of a word that has been usedenote the referent, relying on a memory recof the current discourse to do so. For instain a telephone conversation with a hotel clerguest might introduce the topic of some misssuitcases. Subsequently the objects couldreferred to asthey,a plural pronoun appropriato the number of the nounsuitcases.On theother hand, if the same set of objects in the sconversation had initially been introducedbaggage,the guest might be more likely

mploy the singular pronounit in later references, using the pronoun appropriate to thegular grammatical number of the mass nbaggage.

Yet another possibility involves using tmorphological features of specific words frthe discourse. In Dutch, the definite determindeandhetmark nouns of different grammatic

enders (calledcommonand neuter, respecively). In consequence, any definitely deined singular noun phrase has a surface ta

ts gender. When such a noun phrase servehe antecedent of an anaphoric pronoun,ppropriate form of the pronoun can be form

ated from the surface information that ori

al--

-alal,--t-

rs

ssttod,

e

e

-

s

orase-

ally accompanied the antecedent: If the deiner wasde, the pronoun options include tord die, but if the determiner washet, theronominal options are different and incluords likedat.We will call these three respective hypot

es about pronoun selection theconceptualhy-othesis, thelexical hypothesis, and thetagypothesis. In terms of theories of the cognirocesses of language production (Bock &elt, 1994), the strong form of the conceptypothesis suggests that the speaker iden

he intended referent within a representatiohe to-be-conveyed message, directly (i.e.,ectly from the concept) accesses the releeatures of a word that accurately denoteseferent, and uses only these features wheecting a pronoun. The lexical hypothesislies that, in addition to identifying the intendeferent, the speaker consults information abhe prior discourse that incorporates grammally relevant features of specific words usentroduce discourse referents. These featureect the part of a lexical representation calleemma,which includes information about syactic properties like grammatical gender. Taghypothesis assumes that, in addition to uonceptual and abstract lexical informatbout the antecedent, the speaker consuemory record of the prior discourse that

ludes traces of words actually produced in thonologically encoded forms. Strong formsach of these hypotheses would argue forxclusive use of the corresponding informatith little influence from other sources.

THE GENDER SYSTEM OF DUTCH

To test these hypotheses, we devised a sf production experiments that exploitedender system in Dutch. Normatively speakronouns in Dutch agree with their noun anedents in grammatical gender. There arewo genders in regular use, one neuter (crising those nouns that take the definite deiner het, informally calledhet nouns) and thther common (comprising those nounsake the definite determinerde, called deouns). The class of common-gender nounserger of what were historically the mascul

letn i6)

ndeive

s tivct unwgd eg rsw unsi ishg est

,

areth

c o-d -n flect abs neB gend as othe

tiot icaa acht rdp ena

rtiveedRthdinrtin

thelic-ro-de-

la-

ofe inraloc-te-ame

ici-oseced-e)be

, aiatey

d theple,oc-

r ofan-

nosekersan-utch

eheerhiserdedea-

atedingeter-the

aultuc-ce


and feminine genders, which are all but obsoin standard contemporary Dutch, as spokemost of The Netherlands (van Berkum, 199The corresponding neuter- and common-gepronouns include the singular demonstratdat (for het nouns) anddie (for de nouns). Theame demonstratives serve to introduce relalauses, analogous to Englishthat, and whenhey do they carry the gender of the head nohich the relative clause modifies (e.g.,de jon-endie lang is[the boy who is tall] orhet meisjeat lang is [the girl who is tall]). To make thesender distinctions explicit for English readee will subscript the Dutch nouns and prono

n our examples, as well as their Engllosses, to indicate whether the word in qu

ion is a common-genderdeword (e.g.,jongenD/boyD//dieD) or a neuter-genderhet word (e.g.meisjeH/girl H//datH).

Although demonstratives and relativesthe only pronouns relevant to our research,definite determinersde [theD] and het [theH],along with the indefinite determinereen [a],play an important part. The indefiniteeencan beused with de and het words alike, with nohange in its form. However, if a noun intruced witheen is later referred to with a prooun, the pronoun may be expected to re

he noun’s grammatical gender, despite theence of gender marking on the determiecause most Dutch nouns do not indicateer overtly in their phonology, this offerstraightforward means of testing the tag hypsis.

AN OVERVIEW OF THE EXPERIMENTS

In this research we used sentence compleasks adapted from studies of grammatgreement in English (Bock, 1995). On e

rial in both experiments the participants heareamble sentence, e.g., “Kijk, daar ligt eardappelD bij een badpakH” (Look, there’s a

potatoD lying next to a swimsuitH). Shortly aftethe offset of the preamble, a printed adjecappeared on a computer screen and remainview until the end of the trial (e.g., GAA[cooked]). When the adjective appeared,participant reproduced the preamble, appena second sentence (in Experiment 1) or inse

en.r

s

e

,

,

-

e

t-

r.-

-

nl

a

in

egg

a relative clause (in Experiment 2) usingadjective displayed on the screen. This impitly demanded the use of gender-marking pnouns. In Experiment 1 the pronouns weremonstratives [e.g., “DieD is gaar” (ItD iscooked)] and in Experiment 2 they were retives [e.g., “. . . dieD gaar is . . .” (. . . thatD iscooked . . .)]. Although the pronominal formsdemonstratives and relatives are the samDutch (as they are in English), their structucontexts differ. Specifically, demonstrativescurred in a different sentence from their ancedents, whereas relatives occurred in the ssentence as their antecedents.

The main questions were whether partpants would normally use a pronoun whgender is appropriate for the intended anteent (thede-nounpotatoD, in the example abovand whether the choice of pronoun wouldinfluenced by the gender of an interlopernonantecedent noun phrase in the immedvicinity (thehet-nounswimsuitH). Speakers mabe more vulnerable togender interferenceef-fects when the genders of the antecedent aninterloper mismatch, as they do in the examthan when they match. Gender interferencecurs when a pronoun displays the gendethe interloper rather than the gender of thetecedent.

Gender interference can be used to diagthe nature of the processes by which speaaccess the relevant features of a pronoun’stecedent. Suppose a speaker says the Dequivalent ofLook, there’s a potatoD next to aswimsuitH. ItH is cooked.Assuming that thspeaker meansit H to refer to the potato (as tadjectivecookedstrongly implies), the gendof the pronoun is inappropriate. Should tkind of error occur reliably when the interlopin the preamble is the opposite of the intengender, the implication would be that the ftures for pronoun selection are sought fromrepresentation of lexical features associawith potential antecedents from the preceddiscourse. Were gender features instead dmined from a message representation ofantecedent supplemented with only a defcategorization of the antecedent during prodtion, it is less obvious why gender interferen

f thhe

trale

edipuniensineanceer-

ineede

isur

th, i

en-thenceon

ex-deneenham

er)tivdinndnte

pr

retenas

utcon

unin

unsers.nderap-ert-derortive

r-ey

rsitya-

al-

msre-pre-

al-elic-waswasu-t to

denthed

nte-ried

eseom-

om-

si-32ts.chere

washe

.n

286 MEYER AND BOCK

should arise more often when the genders oantecedent and interloper mismatch than wthey match. This therefore constitutes a conbetween the conceptual hypothesis and theical hypothesis.

In addition to varying the genders of antecents and interlopers, the experiments manlated the types of determiners that accompathem. All of the preambles had two versioone with the ungendered indefinite determeen(a) accompanying both the antecedentthe interloper and a second version that replathe indefinite determiners with the gendmarking definite determinersde (theD) and het(theH). Whether gender interference occursthe absence of the surface distinctions betwantecedents and interlopers provided by theinite determiners and whether interferenceselectively ameliorated or exacerbated byface markers constitutes a contrast betweenlexical and the tag hypotheses. Specificallygender is a property carried principally by gder tags (like determiners) rather than bylexical entries of nouns, gender interfereshould be most evident in the overt-gender cditions.

Both of the experiments reported belowamined the incidence and distribution of geninterference after preambles that containouns of uniform genders or contrasting gders and both compared preambles thatexplicit gender tagging (overt gender) to preables that lacked explicit tagging (covert gendThe first experiment used Dutch demonstrapronouns, which had antecedents in a precesentence. In the second experiment we extethe findings to relative pronouns, whose acedents were in the same sentence as thenouns.

EXPERIMENT 1

In Experiment 1, the participants had to cate sentences that served as sequels to senthey had just heard. The form of their task wsuch that the sequels naturally began with Ddemonstrative pronouns, either the commgenderdieD (it D) or the neuter-genderdatH (it H).For half of the participants the preamble nocarried gender overtly, using definite determ

enstx-

--d

,rdd

nf-s-e

f

-

rd-d-.eg

ed-o-

-ces

h-

s-

ers; for the other half the gender of the noremained covert, using indefinite determinThe nouns themselves carried no reliable gecues (except in one instance), so that thepearance of gender interference in the covgender conditions could arise only from geninformation retrieved from the lexical entry fa potential antecedent for the demonstrapronoun.1

Method

Participants.The main experiment was caried out with 48 native speakers of Dutch. Thwere undergraduate students at the Univeof Nijmegen and were paid for their particiption. Thirty-two other students took part in vidation studies carried out on the materials.

Materials.There were 96 experimental ite(see Appendix). Each item consisted of a pamble sentence and a single adjective. Theamble was always of the formKijk, daar ligteen/de aardappelD bij een/het badpakH (Look,there’s a potatoD lying next to a swimsuitH). Ofthe two nouns in each preamble, one wasways an expected antecedent for the to-be-ited demonstrative pronoun, and the otherthe interloper. For each item, an adjectiveselected [e.g.,gaar (cooked)] that applied plasibly to only the expected antecedent and nothe interloper. The genders of the anteceand the interloper, whether the genders matcor mismatched, and the positions of the acedent (first noun or second noun) were vaorthogonally.

Two versions were created for each of thitems. In one version, both nouns were accpanied by definite determiners (de or het),whereas in the other version, they were cbined with indefinite determiners.

To validate our judgments about the plaubility of the adjective–noun combinations,raters were given written lists of word tripleThe triplets included the two nouns from eapreamble and the adjective. The raters winstructed to indicate whether the adjectivemore likely to be a property of the first or t

1 The exception was the nounslipje (pair of panties), No52 in the Appendix. The diminutive suffix -je flags the nouas neuter.

ththsefeedtheneth6%biaun

ereerend

lesdan

ndfe-tioning55iseardeamlese oled

msin

acdi

onthe

inncen

d,

liseis00an

ra

ul-ic-ratedcon-al-tive,

96osi-m-an-ntededin-efi-

-n oftiveerecor-ex-the

ewounromthaterewereere

pro-ntalects

1.akesin

theer-

e anti fecto heng ert.

theg were noun


second noun. Half of the raters receivednouns in one order (the same order as inpreambles), and the other half in the reverorder. The raters showed a very strong preence (99.3% overall) for the intended antecent as the noun most likely to be modified byadjective. For 92 of the 96 items at most orater failed to select the expected noun. Ofremaining four items, three showed a 90.bias (29 of 32 raters) and one had an 84.4%(27 out of 32 raters) toward the expected no

In addition to the experimental items, thwere 72 fillers and 16 practice items. They wsimilar to the experimental items in length acomplexity and also began withkijk (look), butthey differed from the experimental preambin syntactic structure. All of the fillers includea transitive verb, a direct object, and oftenadverb.

Procedure.The experimental sentences afiller materials were digitally recorded by amale native speaker of Dutch for presentaduring the experimental sessions. Recordwere made in a quiet room using a Sony DTCdigital audio tape recorder and a SennheME40 microphone. Speech analysis softw(waves/ESPS, Entropics Inc.) was used totermine the beginnings and ends of the prebles and to write them to individual speech fiThe speech files were stored on the hard driva Hermac 386SX computer, which controlthe experiment.

Each session began with the practice itefollowed by the experimental and filler itemsrandom order, a different random order for eparticipant. The participants were tested invidually in a quiet room. They were told thateach trial they would hear a sentence andsee a word on the computer screen. Theirstructions were to listen to the entire senteand then repeat it, adding a continuation stence beginning with either “DieD is. . .” or“DatH is. . .” and using the adjective provideas in “DieD is mooi” (ThatD is beautiful).

Preambles were played at a comfortabletening amplitude over headphones (SennhMD211N). Each preamble was followed 2ms after its offset by the appearance ofadjective, in black lowercase letters on a g

eedr--

e

s.

s

re--

.f

,

h-

n-e-

-er

y

background, on a computer monitor (NEC Mtisync30) in front of the participant. The partipant repeated the preamble and then genea continuation sentence. The repetition andtinuation were digitally recorded for later anysis. Six seconds after the onset of the adjecthe next trial began.

Design. Each speaker was tested on allitems. Every combination of antecedent ption (first or second), grammatical gender (comon or neuter), and match vs mismatch oftecedent and interloper gender was represeby 12 items. Half of the participants receivthe versions of the items with definite determers, and half received the versions with indnite determiners.

Scoring and analyses.The speakers’ responses were evaluated for correct repetitiothe preamble and for the type of demonstrapronoun used. The data from one item wexcluded because one of its nouns was inrectly classified as neuter. On 3.4% of theperimental trials, speakers failed to repeatpreamble correctly or failed to provide a nsentence including a demonstrative pronand adjective. These trials were excluded fanalysis. For the remaining trials, pronounsagreed with the antecedent in gender wcounted as correct and those that disagreedcounted as incorrect. Analyses of variance wperformed on the percentages of incorrectnouns produced in each cell of the experimedesign. Except when noted, all reported effwere significant at or beyond the .05 level.

Results

The main findings are shown in Fig.Speakers made more pronoun-gender mistwhen antecedent and interloper differedgrammatical gender than when they hadsame grammatical gender [22.8% vs 16.9%ror; F1(1, 46) 5 31.7; F2(1, 91) 5 4.6). How-

ver, overt gender-marking had no significmpact on the error rate as either a main efr interaction. The error rate was 20.8% wender was covert and 19.0% when it was ovTable 1 shows the results broken down by

ender of the antecedent. There were ferrors when the antecedent was a common

1.rkedrkedin a

288 MEYER AND BOCK

FIG. 1. Errors in grammatical gender in the production of Dutch demonstrative pronouns in ExperimentThe error bars represent the standard errors of the means. The antecedent’s gender was not explicitly ma(covert antecedent gender) when an indefinite determiner accompanied the antecedent and was explicitly ma(overt antecedent gender) when a definite determiner accompanied the antecedent. Pronouns occurredsentence that followed their antecedents.

theunte

ctshenner

tiveou

r oh aionor-y temamheonse

s athetan

se,as-l oronlym-is

calex-

thatark-theto

eentchder-rfer-tlysults

l-ical

aseavying

ivepla-

ns,he

tionces-ne

isr-m-ncere-m-oci-larkch

cog-t ine ofan-

ouw ent(

e


than when it was a neuter noun. Overall,percentage of correct common-gender pronowas 86.6%, compared to 73.8% correct neugender pronouns [F1 (1, 46) 5 15.3; F2 (1,91)5 27.7]. In addition, the interference effein the mismatch conditions were stronger wthe interloper was a common noun than whewas a neuter noun [29.0% errors vs 16.7%rors; F1 (1, 46)5 31.3;F2 (1, 91)5 5.6].

Discussion

The errors in producing the demonstrapronouns suggest that the gender of the pronwas influenced by the grammatical gendethe interloper. This pattern is consistent witlexical hypothesis for pronoun-gender selectAccording to the lexical hypothesis, when fmulating an anaphoric pronoun, speakers traccess the grammatical features of lexical itthat were previously used to denote the sreferent in the active discourse context. Tresults are less compatible with a strong cceptual hypothesis, which attributes genderlection to a default categorization (such abasic-level categorization of the referent) rathan an active categorization (the lexical ins

TABLE 1

Percentages of Gender Errors on Demonstrative Pronith Gender-Unmarked and Gender-Marked Anteced

Experiment 1)

Genders of candidateantecedents (first andsecond noun phrases)

Location of expectedantecedent

First nounphrase

Secondnoun phras

Gender covert (indefinite determiners)

Common-common 9.2 10.2Common-neuter 24.3 29.6Neuter-neuter 25.5 22.9Neuter-common 29.8 14.6

Gender overt (definite determiners)


sr-

it-

nf

.

ose

--

r-

tiation within the current discourse). Of courthe lexical and conceptual hypotheses bothsume that speakers begin with a conceptuamessage representation of the referent, butthe lexical hypothesis is straightforwardly copatible with evidence that the pronoun’s forminfluenced by a previously instantiated lexispecification of a nonantecedent referringpression.

Experiment 1 also tested the hypothesisspeakers rely on an antecedent’s explicit mers of grammatical gender when selectinggender of an anaphoric pronoun. Accordingthis tag hypothesis, there should have bfewer gender errors on pronouns in the macondition when the antecedents carried genmarked determiners and more gender inteence in the mismatch condition from explicimarked, nonantecedent interlopers. The reoffered little support for these predictions. Athough there were nonsignificant numertrends in the predicted directions, there wnothing to suggest that speakers place hreliance on overt gender marking in formulatdemonstrative pronouns.

The overall superiority for the demonstratdieD was unexpected, but has a possible exnation in terms of the distributions ofdeandhetnouns in Dutch. Because of these distributiodieD is far more frequent. We return to this in tGeneral Discussion.

EXPERIMENT 2

The tag hypothesis assumes that informaabout the surface markings of gender is acsible at the time a pronoun is formulated. Owell-known limitation on the availability of thkind of information is the fast decay of infomation about specific wording in human meory (Sachs, 1967), especially when a senteboundary intervenes (Jarvella, 1971). Thetrievability of pronoun antecedents during coprehension also shows discontinuities assated with clause or sentence boundaries (C& Sengul, 1979; Ehrlich & Rayner, 1983). Sufacts about memory may help to shape thenitive mechanisms that carry out agreemengender-marking languages. Notably, the usgrammatical gender in gender-marking l

nss

mawes a, bica

i-icas iuts ttere-paex

dearenclu

toecihy

encraesdeecde

tivt 1

ance taseglyd abl

ece

(

e

harm

orein-in-

veas

in

ofkersallyuse.tiverase

he

be-ent.firstthend, theble

kersmeandhe

ded23be-

deromt 1)assi-or-hen

am-ameis-di-

the

290 MEYER AND BOCK

guages is sometimes seen to change systecally at sentence boundaries. German isknown for this phenomenon. If a pronoun hagrammatically neuter noun as an antecedentthe referent of the neuter noun has biolog(natural) gender [e.g.,das Madchen (girl) isgrammatically neuter but biologically femnine], the biological rather than the grammatgender may appear on the pronoun if it occura different sentence than the antecedent. Bthe antecedent is in the same sentence apronoun, the grammatical gender domina(Drosdowski, 1984, p. 664). With the closelationship between German and Dutch, comrable patterns of gender usage might bepected among Dutch speakers.

If sentential bounds on grammatical genagreement apply in Dutch and reflect in pthe restrictions on the accessibility of overt gder markers in immediate memory, the consions from Experiment 1 may not generalizewithin-sentence agreement processes. Spcally, the absence of evidence for the tagpothesis in Experiment 1 may be a consequof the restriction to cross-sentence anaphothe experimental materials. The tag hypothcould be more successful in predicting genagreement and gender interference effwithin sentences. Experiment 2 was run in orto find out.

Dutch speakers were asked to create relaclauses with the materials from ExperimenThey were presented with a preamble andadjective and asked to produce an utterawith a relative clause. The speakers were freadd the relative clause to either noun phralthough the meaning of the adjective stronconstrained which of the choices would yielsensible result. For example, given a preamsuch aseen slotH in een tuinD and the adjectivstuk,speakers would be most likely to produthe utterance “een slotH datH stuk is in een tuinD”literally, a lockH thatH broken is in a gardenD).

In contrast, giveneen papierH op een truiD andthe adjectivewarm, the more likely utterancwould be “een papierH op een truiD dieD warmis” (literally, a piece of paperH on a sweaterD

thatD warm is). The tag hypothesis predicts trelative pronouns (which are the same in fo

ti-ll

utl

lnifhes

--

rt--

fi--e

inisrtsr

e.neo,

e

t

as the demonstratives) should be marked mreliably for gender after definite than afterdefinite determiners and should suffer moreterference in the same circumstances.

Method

Participants.The participants were 48 natispeakers of Dutch from the same sourceExperiment 1.

Materials.The materials were the same asExperiment 1.

Procedure.The procedure duplicated thatprevious experiment, except that the speawere instructed by example to use the visupresented adjective to create a relative claThey were permitted to append the relaclause to either the first or second noun phof the preamble as they wished.

Design. The experimental design was tsame as in Experiment 1.

Results

Speakers usually minimized the distancetween the relative clause and its antecedConsequently, when the antecedent was thenoun, the relative clause was inserted intopreamble immediately after the first noun, awhen the antecedent was the second nounrelative clause was appended to the preamimmediately after the second noun. Speadeviated from this pattern only 2.3% of the tiwhen gender was covert in the preamble3.2% of the time when gender was overt. Tdeviant utterances (124 in total) were exclufrom the analyses below. An additional 2cases (4.8% of all responses) were missingcause of preamble-repetition failures.

The percentages of relative-pronoun generrors are shown in Table 2, with the data frone item (the same item as in Experimenexcluded because of an incorrect gender clfication. Figure 2 summarizes the most imptant effects. Speakers made more errors wthe antecedent and interloper differed in grmatical gender than when they had the sgrammatical gender [10.0% errors in the mmatch condition vs 5.5% in the match contion; F1(1, 46) 5 39.9; F2(1, 91) 5 9.0] andwhen the antecedent was the second of

mschs indeug

ouoran

errean

no ndw h

nifi

Fth

thaherec.4%rre

erethein-hener,on-m-dernt-ub-theed a

c-

yi derm

D

ritd fors edi-a pro-n oft ida-t te-c ts fort t 1w ced-e eri-m eak-e thep tivea

ash unp pro-n eced-e om-p unsw en-d t ex-p oft hani alc ro-m er-a to

wiG (Exp

e


candidate noun phrases [F1(1, 46) 5 4.7], al-though this effect was not significant for ite[F2(1, 91)5 1.1,p . .1]. Because the mismateffect was larger when the antecedent wasecond position, the interaction between gematch and position was also significant, thoagain only by subjects [F1(1, 46)5 10.1;F2(1,91) 5 2.0, p . .1].

Gender marking on the antecedent nphrases influenced the error rate, yielding merrors overall when marking was covert thwhen it was overt [12.1% vs 3.4%;F1(1, 46)517.1;F2(1, 91)5 60.2]. A mismatch in gendbetween the candidate antecedents had a geffect when gender was unmarked than whewas marked [F1(1, 46)5 8.0], and the positio

f the antecedent mattered less when geas unmarked [F1(1, 46) 5 10.14], althoug

neither of these interactions achieved sigcance by items [F2(1, 91) 5 3.1, p . .05 and

2(1, 91)5 2.0, p . .1, respectively].Once again, there were fewer errors when

antecedent was a common-gender nounwhen it was neuter. This was clearest wgender was covert: There were 93.6% corcommon-gender pronouns, compared to 82correct neuter-gender pronouns. The co

TABLE 2

Percentages of Gender Errors on Relative Pronounsender-Unmarked and Gender-Marked Antecedentseriment 2)

Genders of candidateantecedents (first andsecond noun phrases)

Location of expectedantecedent

First nounphrase

Secondnoun phras

Gender covert (indefinite determiners)


Gender overt (definite determiners)


nr

h

ne

terit

er

-

en

nt

-

sponding rates when gender was overt w97.0 and 96.1%. Likewise, interference inmismatch conditions was greater when theterloper was a common-gender noun than wthe interloper was neuter (with covert gendthere were 20.7% errors in the mismatch cdition after a common-gender interloper copared to 10.1% errors after a neuter-geninterloper; with overt gender, the error perceages were 4.7 and 5.1%, respectively). In ssidiary analyses including the gender ofantecedent as a factor, these trends yieldsignificant main effect of gender [F1(1, 46) 56.2; F2(1, 91)5 17.6] and a significant interation between gender and overt marking [F1(1,46) 5 4.7; F2(1, 91) 5 26.4]. The three-wanteraction among these factors and gen

atch was not significant (bothFs , 1).

iscussion

Two of the findings from Experiment 2 meiscussion. One was the strong tendencypeakers to place the relative clauses immtely after the selected antecedent for theoun. This helps to solidify the interpretation

he results by providing an independent valion of the normative designations of the anedents. Because the expected antecedenhe demonstrative pronouns in Experimenere exactly the same as the expected antents for the relative pronouns in this expent, we can be more confident that the sprs in both experiments generally intendedronouns to be coreferential with the normantecedents.The most notable result in Experiment 2 w

ow gender marking on antecedent nohrases affected the forms of coreferentialouns in the same sentences. When the antnt’s gender was explicitly tagged by an accanying article, erroneously gendered pronoere significantly less likely than when the ger of the antecedent noun phrase was nolicitly indicated. Moreover, the magnitude

his difference was greater in Experiment 2 tn Experiment 1. Although a direct statisticomparison of the two experiments is compised by differences in the forms of the uttnces that were produced, it is defensible

th-

orovertovertsame

292 MEYER AND BOCK

FIG. 2. Errors in grammatical gender in the production of Dutch relative pronouns in Experiment 2. The errbars represent the standard errors of the means. The antecedent’s gender was not explicitly marked (cantecedent gender) when an indefinite determiner accompanied the antecedent and was explicitly marked (antecedent gender) when a definite determiner accompanied the antecedent. Pronouns occurred in thesentences as their antecedents.

asthd-atleds itiv

s19..0%eeectnal

1ers

s ti-c ana errp nsf As form ingo thag oft thif una tioni data eai ata Thr ertc lenr e ths

ider hap enc amm ant Thm amm ant 4)a sent the

er-

oree-hy-tifi-ory

ificof

ert

th-r inultslec-

blerep-thecif-thelex-enticalouldby

rlier

derbyersark-car-negyr-on.

inse-

riedhention

iniedation

llto-


contrast the results when the antecedent wsecond position since, in both experiments,pronoun immediately followed a seconposition antecedent. An analysis of the dfrom the second-position conditions reveathat speakers made significantly fewer errorselecting relative pronouns than demonstrapronouns [8.5 vs 19.4% errors;F1(1, 92) 525.74;F2(1, 46) 5 60.18]. This reduction wamost pronounced when gender was overt (vs 5.0% when overt, compared to 19.3 vs 12when covert), although the interaction betwpronoun type and gender tagging that reflthis reduction was marginal in the subject aysis [F1(1, 92) 5 3.02, p , .10; F2(1, 46) 5

3.65].This difference in pronoun accuracy off

upport for the traditional view that grammaal gender is more reliably marked within thcross sentence boundaries. Even so, theattern is not consistent with the predictio

rom a strong form of the tag hypothesis.trong tag hypothesis implies that gender ination is proprietarily conveyed by the markf the determiner. The argument would beender information resides in the overt form

he determiner and that in the absence oform gender must be inferred from the nolone. Although this indeed predicts a reduc

n pronoun errors when the genders of candintecedents match, it also predicts an incr

n errors when the genders of the candidntecedents mismatch. This did not occur:eduction in errors when gender was ovompared to when it was covert, was equivaegardless of whether the antecedents werame or different in gender.This suggests that gender tagging prov

edundant information about gender rather trivileged information: When an antecedarries a gender-explicit determiner, the gratical gender is carried by both the noun

he determiner rather than either one alone.emory representation of a determiner’s gratical information may be more fragile th

hat of the noun (Koriat & Greenberg, 199nd hence more vulnerable to disruption at

ence boundaries. This would account for

ine

a

ne

4

ns-

or

-

t

s

eseee,te

snt-de-

-

nonsignificant impact of overt gender in Expiment 1.

In short, the results of Experiment 2 are mcompatible with the lexical hypothesis of immdiate antecedent recovery than with the tagpothesis. The findings indicate that the idencation of an antecedent takes place in a memrepresentation that contains lexically specinformation about the grammatical privilegeswords, but not information about their ovmorphological forms.

GENERAL DISCUSSION

Our results lend support to a lexical hypoesis for the selection of grammatical gendepronoun production. To recapitulate, the resof both experiments indicated that gender setion calls on more information than is availasolely from a conceptual or message-levelresentation of the intended referent or fromreferent’s default lexical categorization. Speically, pronoun gender selection reflectedgrammatical features of discourse-activatedical information. This was shown in consistinterference from the mismatching grammatgender of a nonantecedent noun, which shoccur only if pronoun gender is influencedmemory representations that include eamentions of the intended referent.

The experiments further showed that genselection is not controlled solely or stronglyovert markers of gender on definite determinthat accompanied the antecedents: Overt mers appeared to supplement the informationried by the noun but did not override it, as owould expect if gender-specific morpholowere the only information or the primary infomation consulted during pronoun selectiThis disconfirmed the tag hypothesis.

Consistent with the lexical hypothesis,both experiments the accuracy of pronounlection decreased when an interloper carconflicting gender and decreased further wthe interloper occupied the preferred posifor the antecedent (sentence-initial positionExperiment 1) or when the interloper occupa place in the current sentence represent(sentence-initial position in Experiment 2). Aof these gender-interference effects point

edorm.heathear

skss o

oousnteamecente

anforinaat

usepr

ed)thati-a

ro-a

hate

ingreliex

inerk-enorre

thetalenths

tenaytio

ss toa-7;

n-de-re-

caloreed

e thear-6;alre-en-

esare

t theiatenlyro-is ahisly)ies.s beep-stillis,

tlero-m-berthisa-chn-ate

entent

ourthe

andon-ter-rall,oun

294 MEYER AND BOCK

ward a process involving a lexically specifirepresentation of candidate antecedents fpronoun in order to select the pronoun’s for

The results accord well with others in tpsycholinguistic literature in indicating thspecific linguistic information is used in tprocessing of pronouns in languages that mgrammatical gender. In comprehension taGarnham et al. (1995) found that speakerFrench and Spanish read sentences mquickly when pronouns carried unambigucues to the grammatical genders of their acedents and that this effect was of the smagnitude for pronouns with inanimate-objantecedents (which have only grammatical gder) as it was for pronouns with human ancedents (which had matching grammaticalbiological genders). This points to a rolegrammatical gender in establishing pronomantecedents. However, the effect of grammcal gender diminished rapidly: When a claseparated the pronoun from its antecedent,nouns with human (i.e., biologically genderantecedents were understood more readilypronouns with inanimate (i.e., only grammacally gendered) antecedents. Cacciari et(1997) found similar tendencies in Italian: Pnouns whose antecedents bore overt grammcal gender were understood more readily tpronouns without an explicitly gendered ancedent.

Of course, Romance languages (includItalian, Spanish, and French) tend to haveable gender cues in their word forms. Forample, Spanish nouns ending in -o are typicallymasculine, while those ending in -a are typi-cally feminine. This makes it hard to determwhether it is overt morphophonological maing that mediates the effect of grammatical gder on pronoun resolution rather than a mabstract lexical classification of gender. Oursults strongly suggest that in Dutch it islatter, and it may not be entirely coincidenthat Dutch lacks phonological correlates of gder. But we doubt that such cues obviateneed for or the use of abstract representationgrammatical gender in either speaking or lising. For one thing, speakers of Italian msometimes have access to gender informa

a

k,fre

-et--d

li-

o-

n

l.

ti-n-

--

-e-

-eof-

n

about words even when they have no accesound information (Badecker, Miozzo, & Znuttini, 1995; Miozzo & Caramazza, 199Vigliocco, Antonini, & Garrett, 1997). For aother, the syntactic planning processesmanded by sentence production appear toquire that structurally relevant grammatiinformation (like gender) be accessible befphonological information is normally accessin the course of spontaneous speaking. Sincscope of phonological preparation is fairly nrow (Dell & O’Seaghdha, 1992; Meyer, 199Schmitt, Meyer, & Levelt, 1999), grammaticinformation may become active before andmain active after sound specifications arecoded.

Still, in principle there may be differencamong languages in the ways that pronounsprocessed, and these differences may refleckinds of information that are needed to medcoreference. In English, for example, the ogrammatically relevant information that pnouns reliably share with their antecedentsnumber feature (singular or plural), and tnumber feature is typically (albeit imperfectcorrelated with conceptual number propertThis means that pronouns can almost alwayformulated directly from a message or conctual representation of the antecedent andreflect the relevant number. In line with thBock, Nicol, and Cutting (in press) found litinfluence of grammatical number on the pduction of pronouns in American English copared to the influence of grammatical numon the production of verbs. What may makepossible is the reliability of conceptual informtion for marking grammatical number, whicontrasts sharply with the unreliability of coceptual information (especially about inanimobjects) for marking grammatical gender.

Even within a single language, agreemfeatures as well as different values of agreemfeatures need not behave identically. Indata, there were unexpected differences inmagnitude of interference effects for neutercommon gender. In both experiments, commgender pronouns were more likely than neugender pronouns to be used correctly (ove85.1% vs 75.8%) and common-gender n

nte.1heThpearr atenuc7)reeerou

elyo

inaranseevuhe

theth

ithre

ducnter

outharefprocetheseanb

f thf th

re-t bprefor

eratemaysultnte-al),in-mesup-ob-lly

hy-on-x-

dentselyonan)idepre-therental

s ofible,ndromwn

urereddedand

theor-theced-nci-t de-on

ltsre-

oryeechnci-ed

er-notigh


antecedents were more potent sources of iference than neuter antecedents (overall, 27vs 18.0% error in the mismatch conditions). Treasons for this can only be guessed at.simplest possibility has to do with the tyfrequency of the two noun classes: Thereroughly three times as many common-gendeneuter-gender nouns in Dutch, despite adency for neuter-gender nouns to be mhigher in frequency (van Berkum, 1996, 199The net result is that about two of every thnouns in running text will bear common gendIf speakers tend to resolve uncertainty abwhich pronoun to use in terms of the most likgender, they should use common gender moften than neuter gender, and they do.

In other respects, the overall patterns ofterference in pronoun selection were compble for neuter- and common-gender pronouBoth suffered from gender mismatches betwpotential antecedents, and neither was morenerable to the effects of gender mismatch wgender was overtly tagged.

Apart from these points, the findings frompresent experiments bear on two issues inpsycholinguistic literature having to do wlanguage comprehension. One involves thelationship between comprehension and protion in the cognitive processing of agreemeand the other has to do with how Dutch readinterpret relative clauses.

Comprehending and Producing Pronouns

The most salient difference between pronproduction and pronoun comprehension isspeakers normally know the conceptual orerential antecedents of the pronouns theyduce, whereas listeners must infer the anteent from their interlocutors’ speech or fromcurrent context. The tasks used in the preexperiments demanded both productioncomprehension: comprehension of the preamsentences, from which the representations oantecedents were derived, and production opronouns.

This complicates the interpretation of thesults in at least three ways. First, one cannocertain whether antecedent gender is resented similarly for understanding and

r-%

e

es-

h.

.t

re

--.nl-n

e

--,s

nt--

d-

ntdleee

e-

speaking. For example, when speakers genantecedents directly from messages, theybe less likely than our speakers were to conany sort of surface representation of the acedent (either grammatical or morphologicinstead relying more heavily on conceptualformation alone. Although plausible, this sascenario might lead one to expect strongerport for the tag hypothesis than what wetained. In fact, none of our results were fuconsistent with the predictions of the tagpothesis, and our only clear evidence for a ctribution of overt gender marking came in Eperiment 2. There the pronoun and antecewere in the same sentence, which is preciwhat traditional accounts (based primarilyobservational evidence from written Germwould predict. Moreover, our results provevidence about what kind of antecedent resentation is preferentially consulted duringgeneration of pronouns when several diffetypes of information are available. In normepisodes of pronoun use, all of the sourceinformation that we investigated are accessin principle. For example, in conversation aother dialogues, speakers employ material ftheir interlocutors’ speech, as well as their ospeech, in formulating pronouns.

A related complication in interpreting oresults is that the speakers may have sufffrom problems in understanding the intenrelationship between potential antecedentsmodifiers or from problems in rememberingpreambles. This could serve to disrupt the nmal pronoun selection process. In fact, withincreasing determinacy of the intended anteents between Experiments 1 and 2, the idence of gender interference decreased. Buspite this, the theoretically telling configuratiof effects remained the same.

A third source of ambiguity in the resustems from the requirement for speakers topeat the preambles, imposing explicit memdemands that are absent from normal spsituations. This could have increased the idence of errors or, more worryingly, changtheir distribution. The incidence of pronounrors in spoken Dutch is unknown, so we cantell whether the task elicited an unusually h

diskshey dhis

ndsen&

heranin

turses

venncwhan

ther

dehed tun

thenertmowapop

theond8).hatsioro-

recIn

ialscli-tur

s indente%,

ounallyys-ro-in

an-rth,ec-rela-red:tivetingailedWeinedTheeivefol-re-

y-nt

eren ant ofself,ent-

writ-pro-henof

wountsof

thefi rvesa h.

296 MEYER AND BOCK

number of errors. However, regarding thetributions of speech errors in laboratory tasthere is considerable evidence that even wtask demands increase the rates of error, theso without changing the error patterns. Tseems to hold for speech errors of all ki(Stemberger, 1992) as well as for agreemerrors elicited in tasks similar to ours (BockMiller, 1991).

Interpreting Relative Clauses

The foregoing discussion assumes that tare deep parallels between comprehensionproduction processes. But arguments agasuch parallels have been raised in the literamost notably with regard to how relative clauare understood during reading or listening.

The analysis or interpretation of relaticlauses has been a long-standing concerpsycholinguistic theories of parsing. A sentelike Someone shot the servant of the actresswas on the balconycan be understood to methat either the servant or the actress was onbalcony. Recent studies suggest that readinclinations about this may vary, sometimespending on their language. Cuetos and Mitc(1988) found that Spanish readers tendefavor the reading in which the relative pronoin Spanish would be taken to refer toservantinthe sentence above (so-calledhigh attachment),whereas readers of British English showedreverse tendency (though see Gilboy, SopClifton, & Frazier, 1995). For Dutch, Brysbaand Mitchell (1996) reported a preference coparable to that in Spanish. Subsequently, hever, they reported an analysis of a newspcorpus which appeared to reveal a stronglyposing distribution: In these written texts,relative clauses reliably modified the secnoun phrase (Mitchell & Brysbaert, 199Mitchell and Brysbaert concluded from this tDutch readers display biases in comprehenthat diverge from the patterns normally pduced by Dutch writers.

We suspect that there may be a simpleronciliation of these conflicting outcomes.preliminary norming studies on the materfor Experiment 1, we observed a strong innation for readers to take the first (and struc

-,no

t

ed

ste,

ofeo

es’-llo

ea,

--

er-

n

-

-

ally more prominent) of the two noun phrasethe experimental preambles as the antecefor a demonstrative pronoun.2 The preferencfor the first noun phrase was roughly 68including even the cases in which the first nphrase was both semantically and syntacticunacceptable. This finding accords with Brbaert and Mitchell’s (1996) inasmuch as pnoun resolution may be intimately involvedthe interpretation of relative clauses in lguages like Dutch and German (HemfoKonieczny, & Scheepers, 1999). But in our sond experiment, when speakers producedtive clauses, an interesting tendency appeaSpeakers almost always placed the relaclause immediately after the noun representhe intended antecedent, even when this entinterrupting the complex noun phrase.found the same tendency when we examone issue of a daily national newspaper inNetherlands,De Volkskrant.The complete issucontained 65,101 words and 312 relatclauses. In the relative clauses, over 86%lowed their antecedents immediately. As asult, there were few cases such as . . .gebelddoor iemandD van de BVDD dieD van alles wildeweten . . . (. . . called by someoneD from theFBID whoD wanted to know about everthing. . .), where a structurally more prominebut distant noun [iemandD (someoneD)] servedas the antecedent of the relative pronoun [dieD

(whoD)].Consequently, when relative clauses w

used immediately after the second noun icomplex noun phrase, the usual antecedethe relative pronoun was the second noun itand not the earlier, structurally more prominfirst noun. If proximity is the single most important means used by Dutch speakers anders to indicate the antecedents of relativenouns, corpus counts can be misleading wthey focus exclusively on the antecedentsrelative clauses that follow the second of tnouns in a complex noun phrase. Such conaturally overlook the overwhelming weight

2 Structural prominence has to do with the fact thatrst noun of complex subject noun phrases normally ses what is informally called the subject noun in Englis

tele,la-nd

ardutclcuwoentth

na

uen

itieulthichou

od-.di-obn

mbfied

eis apobilbyri-

uc-tive3),ive, atunnds,

ubt-rlyter-

s of

ofr is

ino-97;s dol oringcep-u-theyticthern,

lec-ed-en-rallr er-wererpris-ci-derfectsghtbutder

condunsd

entstive

AA 75)

sese


instances in which a relative clause immediafollows the initial noun of the complex phrascounting equally all instances in which a retive clause immediately follows the seconoun phrase.

To roughly estimate the tendencies towalternative patterns of usage among the Dspeakers we tested in Experiment 2, we calated the probabilities of employing each of trelative clause sites for each of two differintended antecedents in our materials. Forcomplex noun phrases that we employed (aogous tothe servant of the actress), the twosites are after one or the other of the constitnoun phrases (eitherthe servantor the actress).These two phrases are likewise the possibilfor the intended antecedents. For these calctions, we also included the 124 responseswere not analyzed in Experiment 2, in whspeakers placed a relative clause after a nphrase that it was semantically unlikely to mify. Table 3 shows the relevant probabilities

The overwhelming tendency toward immeate placement is evident, with an overall prability of .98. Beyond that, after initial nouphrases our speakers produced a small nuof relative clauses that more sensibly modithe second noun phrase (along the lines oftheservant1 who2 married Prince Rainier of thactress2), and these may be errors. If thereroughly equivalent tendency to err in the opsite direction, a better estimate of the probaity of a distant antecedent may be gottensubtracting the probability of the clearly spu

TABLE 3

Probabilities of Alternative Relative Clause Placemfor Each of Two Intended Antecedents of the RelaPronouns in Experiment 2

Relative clause site

Intended antecedent

First nounSecondnoun

fter first noun phrase .98 (2057) .02 (46)fter second noun phrase .03 (78) .97 (21

Note.Numbers in parentheses represent the responach cell.

y

h-

el-

t

sa-at

n

-

er

--

ous cases (.02) from the probability of proding after the second noun phrase a relaclause that is intended to modify the first (.0yielding .01. Extrapolating, among the relatclauses that follow complex noun phrasesmost one in one hundred will take the first nophrase as the antecedent. In light of these trecoupled with our corpus data, it appears doful that Dutch speakers (and writers) regulaproduce relative clauses whose intended inpretations are at odds with the predilectiontheir readers and listeners.

CONCLUSION

In this work we contrasted three accountshow information about grammatical gendeselected for the production of pronounsDutch. In keeping with previous work on prnoun comprehension (Cacciari et al., 19Garnham et al., 1995) we found that speakernot select pronouns on the basis of referentiaconceptual information alone. Of course, bespeakers, they must use referential and contual information from the outset of the formlation process. But our results suggest thatgo beyond this information to call on linguisfeatures of the pronoun’s antecedent. To furspecify the nature of the linguistic informatiowe examined the contribution to pronoun setion of explicit gender marking on the antecent. There were two notable effects. First, gder marking was associated with an ovedecrease in the number of pronoun-genderors when the pronoun and the antecedentin the same sentence. Second, and more suingly, overt marking did not increase the indence of interference from a different-gencandidate antecedent. Together, these efargue that grammatical gender is not soufrom an overt morphological representation,rather from abstract specifications of genassociated with the lemmas of words.

APPENDIX

Materials

Items are grouped by the genders of the first and senoun (e.g.,de–deindicates that the first and second nowere common gender;de–het means that the first an

in

spee.gcear

tok,thassial

IG

r

KR-

D

L)

NT

ND)

LM-

RIG

D

298 MEYER AND BOCK

second nouns were common and neuter gender, retively) and by the location of the expected antecedent (first indicates that the first noun was the expected anteent). Only the definite-determiner versions of the itemsshown. In the indefinite versions, the determinersdeandhetwere replaced with the indefiniteeen.When presentedparticipants, all items began with “Kijk, daar. . .” (“Loothere. . .”). The English versions are glosses ratherliteral translations. Note that item 35 contains a misclafied noun (biefstuk) and was omitted from the statisticanalyses.de–de first

01 . . .staat de eend op de pannekoek. LELIJK( . . . is theduck standing on the pancake. UGLY)

02 . . .staat de ezel naast de fiets. DOM( . . . is thedon-key next to the bicycle. STUPID)

03 . . . zit destrik op de olifant. ELASTISCH( . . . is thebow on the elephant. ELASTIC)

04 . . .ligt de mat onder de populier. VUIL( . . . is themat lying under the poplar tree. DIRTY)

05 . . .ligt de rugzak op de punaise. ZWAAR( . . . is thebackpack lying on the thumbtack. HEAVY)

06 . . .ligt de vesting bij de rivier. OMMUURD( . . . isthe fortress by the river. WALLED)

07 . . .staat de hond voor de deur. TROUW( . . . is thedog standing in front of the door. FAITHFUL)

08 . . .hangt de sluier over de kat. DOORZICHT( . . . is theveil draped over the cat. SHEER)

09 . . .staat de motor voor de flat. SNEL( . . . is themotorcycle in front of the apartment building. FAST)

10 . . .staat de eik bij de vijver. VERTAKT( . . . is theoak by the pond. BRANCHING)

11 . . .ligt de crepe in de pan. DUN( . . . is thecrepe inthe frying pan. THIN)

12 . . . zit despijker in de tas. KROM( . . . is thenail inthe bag. BENT)de–de second

13 . . .staat de auto op de panty. DUN( . . . is the castanding on the pair of panty hose. THIN)

14 . . .staat de koelkast achter de streep. DIK( . . . is therefrigerator standing behind the stripe. THICK)

15 . . .ligt de atlas in de kelder. DONKER( . . . is theatlas in the cellar. DARK)

16 . . .staat de stoel naast de kleuter. ACTIEF( . . . is thechair next to the toddler. ACTIVE)

17 . . .ligt de steen onder de distel. STEKELIG( . . . isthe stone lying under the thistle. THORNY)

18 . . .ligt de roos op de vrieskist. KOUD( . . . is theroselying on the freezer. COLD)

19 . . .staat de poedel in de bus. VERROEST( . . . is thepoodle in the bus. RUSTY)

20 . . .ligt de worm in de flat. MODERN( . . . is theworm in the apartment building. MODERN)

21 . . .ligt de maillot naast de slang. GEVAARLIJ( . . . is thepair of tights lying next to the snake. DANGEOUS)

c-.,d-e

n-

22 . . .staat de lijn op de telefoongids. VEROUDER( . . . is thestripe on the telephone book. OUTDATED)

23 . . . zit desnor boven de lip. ONTSTOKEN( . . . is themustache above the lip. INFLAMED)

24 . . .ligt de baby bij de bouvier. WAAKS( . . . is thebaby lying close to the Belgian sheepdog. WATCHFUde–het first

25 . . . is degootsteen in het cafe. GEBARSTEN( . . . isthe sink in the bar. CRACKED)

26 . . . zit dehaas naast het brood. ANGSTIG( . . . is thehare sitting next to the loaf of bread. FRIGHTENED)

27 . . . ligt de fietsband naast het konijn. LEK( . . . isthe bicycle tire next to the rabbit. FLAT)

28 . . .ligt de boterham in het kanaal. BELEGD( . . . isthe sandwich lying in the canal. BUTTERED)

29 . . . is dedeur achter het scherm. OPEN( . . . is thedoor behind the screen. OPEN)

30 . . . is dedisco bij het kerkhof. LAWAAIERIG( . . . isthe discotheque near the cemetery. NOISY)

31 . . .ligt de creamcracker op het luchtbed. KROKA( . . . is thecracker lying on the air mattress. CRISPY)

32 . . .ligt de zee achter het pad. DIEP( . . . is theoceanbeyond the path. DEEP)

33 . . .ligt de bikini op het kussen. SEXY( . . . is thebikini lying on the cushion. SEXY)

34 . . .ligt de aardappel bij het badpak. GAAR( . . . is thepotato lying next to the swimsuit. COOKED)

35 . . .ligt de sloop over het biefstuk. GESTREKE( . . . is thepillowcase lying on top of the steak. IRONE

36 . . .ligt de stift in het toilet. HARD( . . . is the peg inthe restroom. HARD)de–het second

37 . . .ligt de schaar op het fauteuil. COMFORTABE( . . . is thepair of scissors lying on the easy chair. COFORTABLE)

38 . . .ligt de bumper in het ravijn. DIEP( . . . is thebumper lying in the ravine. DEEP)

39 . . .ligt de sloep in het lokaal. GESLOTEN( . . . is thedory in the community center. CLOSED)

40 . . .ligt de zakdoek op het schip. STABIEL( . . . is thehandkerchief lying on the ship. STABLE)

41 . . .staat de stoel naast het spatbord. MODDE( . . . is thechair sitting next to the mudflap. MUDDY)

42 . . .ligt de broek in het fort. VERSTERKT( . . . is thepair of trousers lying in the fortress. FORTIFIED)

43 . . . zit de pen in hetschort. GEBLOEMD( . . . is thepen in the apron. FLOWERED)

44 . . .ligt de bloem naast het mes. BOT( . . . is theflower lying next to the knife. BLUNT)

45 . . . is dekuil in het weiland. GEMAAID( . . . is thepit in the pasture. MOWN)

46 . . .ligt de bunker onder het zeil. GESCHEUR( . . . is thebunker lying beneath the sail. TORN)

47 . . .ligt de veer onder het juk. ZWAAR( . . . is thefeather lying under the yoke. HEAVY)

48 . . .ligt de ketting in het riool. STINKEND( . . . is thechain in the sewer. STINKY)

LI-T)

D)R

)N

N

RDR)KT-

Y)

M

ND

D

EN

D

IG)

T

ND

--

a

hea

am-

ure-


het–de first49 . . .ligt het lint in de berm. ROSE( . . . is theribbon

lying on the curb lawn. PINK)50 . . .staat het paard achter de schutting. INTEL

GENT ( . . . is thehorse behind the fence. INTELLIGEN51 . . . is hetportaal in de straat. OVERDEKT( . . . is the

porch on the street. COVERED)52 . . .ligt het slipje op de kist. PIKANT( . . . is thepair

of panties lying on the dresser. RACY)53 . . .hangt het schort over de ezel. GEBLOEM

( . . . is theapron draped over the donkey. FLOWERED54 . . .ligt het landgoed achter de struik. OPENBAA

( . . . is thecountry house beyond the bush. PUBLIC)55 . . .ligt het slot in de tuin. STUK( . . . is thelock in the

garden. BROKEN)56 . . . zit hetvest in de doos. GEBREID( . . . is thevest

in the box. KNITTED)57 . . .ligt het boeket in de trein. VERWELKT( . . . is

the bouquet lying in the train. WILTED)58 . . .ligt het hakmes op de broek. GESLEPEN( . . . is

the cleaver lying on the pair of trousers. SHARPENED59 . . .ligt het perron achter de boom. VERLATE

( . . . is theplatform beyond the tree. DESERTED)60 . . .ligt het gazon achter de flat. GEMAAID( . . . is

the lawn behind the apartment building. MOWN)het–de second

61 . . .staat het aanrecht in de folder. DUN( . . . is thecounter shown in the brochure. THIN)

62 . . . is hetbloedvat rond de ruggegraat. LENIG( . . . isthe blood vessel surrounding the spine. SUPPLE)

63 . . .ligt het papier op de trui. WARM( . . . is thepieceof paper lying on the sweater. WARM)

64 . . . ligt het pistool naast de mok. GEBARSTE( . . . is thepistole beside the mug. CRACKED)

65 . . .ligt het visnet in de straat. BETEGELD( . . . is thefishnet lying on the street. PAVED)

66 . . .hangt het kleed over de schouder. GESPIE( . . . is thedress draped over the shoulder. MUSCULA

67 . . .ligt het boek naast de koekepan. AANGEKOE( . . . is the book lying next to the frying pan. BLACKENED)

68 . . .ligt het album naast de bril. BESLAGEN( . . . isthe scrapbook lying next to the pair of glasses. STEAM

69 . . .ligt het wiel naast de hengel. BUIGZAAM( . . . isthe wheel lying next to the fishing rod. FLEXIBLE)

70 . . .staat het glas op de postzegel. ZELDZAA( . . . is theglass standing on the stamp. RARE)

71 . . .ligt het touw naast de ezel. DOM( . . . is theropelying next to the donkey. STUPID)

72 . . .ligt het vest over de diamant. RUW( . . . is thevestlying over the diamond. RAW)het–het first

73 . . .staat het veulen in het huis. DARTEL( . . . is thefoal standing in the house. PLAYFUL)

74 . . .ligt het eiland in het fjord. ONBEWOOND( . . . isthe island in the fjord. UNINHABITED)

75 . . .ligt het juweel op het papier. KOSTBAAR( . . . isthe jewel lying on the piece of paper. PRECIOUS)

76 . . .staat het beeld achter het hek. GELIJKE( . . . is thestatue behind the fence. LIFELIKE)

77 . . .ligt het schrift in het nest. VERKREUKEL( . . . is thenotebook lying in the nest. CRUMPLED)

78 . . .staat het gehucht in het boek. UITGESTORV( . . . is thehamlet shown in the book. DESERTED)

79 . . .staat het varken in het kasteel. VET ( . . . is the pigstanding in the castle. FAT)

80 . . .ligt het tapijt op het plein. GESTOOMD( . . . isthe carpet lying on the town square. DRYCLEANED)

81 . . .staat het kuiken op het parket. KLEIN( . . . is thechicken standing on the parquet floor. SMALL)

82 . . .staat het huis in het dal. VERVALLEN( . . . is thehouse in the valley. DILAPIDATED)

83 . . . is hetwegdek langs het kanaal. GEASFALTEER( . . . is theroadway along the canal. ASPHALT)

84 . . .staat het paleis in het hof. SPROOKJESACHT( . . . is thepalace standing in the yard. FAIRYTALE-LIKEhet–het second

85 . . .ligt het potlood op het telegram. BEKNOP( . . . is thepencil lying on the telegram. CONCISE)

86 . . .staat het hek rond het zwijn. AGRESSIEF( . . . isthe fence surrounding the hog. AGGRESSIVE)

87 . . .staat het varken in het klooster. ONBEWOO( . . . is the pig in themonastery. UNINHABITED)

88 . . .ligt het geweer op het trottoir. BETEGELD( . . . isthe rifle lying on the sidewalk. PAVED)

89 . . .ligt het kompas in het kozijn. GEVERFD( . . . isthe compass lying on the window sill. PAINTED)

90 . . .ligt het artikel in het gebouw. HOOG( . . . is thearticle lying in the building. HIGH)

91 . . . ligt het tijdschrift in het stadhuis. GERENOVEERD ( . . . is the journal lying in city hall. RENOVATED)

92 . . . zit het dakboven het matras. ZACHT( . . . is theroof above the mattress. SOFT)

93 . . .ligt het ei op het boek. SAAI( . . . is the egglyingon the book. BORING)

94 . . .ligt het laken op het terras. POPULAIR( . . . is thesheet lying on the terrace. POPULAR)

95 . . .ligt het papier naast het horloge. PRECIES( . . . isthe piece of paper lying next to the watch. PRECISE)

96 . . .ligt het eikeblad naast het pistool. ZWAAR( . . . isthe oak leaf lying next to the pistol. HEAVY)

REFERENCES

Baayen, R. H., Piepenbrock, R., & van Rijn, H. (1993).TheCELEX lexical database.Philadelphia: Linguistic DatConsortium.

Badecker, W., Miozzo, M., & Zanuttini, R. (1995). Ttwo-stage model of lexical retrieval: Evidence fromcase of anomia with selective preservation of grmatical gender.Cognition,57, 193–216.

Bock, J. K. (1995). Producing agreement.Current Direc-tions in Psychological Science,8, 56–61.

Bock, J. K., & Brewer, W. F. (1985). Discourse structand mental models. In T. H. Carr (Ed.),The develop

o:

uc-her.

B nt.

B hat

B anes,

B h-

C enliand

C suaefer

C uraS.l-

C the

nts

rs,d

na-omecon

if-Lat

ti-singl

ica

enr-

andent

l

M.rpre-and.ncesationcher.

ref-

tual

R.t bye,

L.fer-

Thee ofo-

yn-s ofer-

a--

ctedv-

oftter

c-

ouns of-

od-lim-

tenceex-

M al–

300 MEYER AND BOCK

ment of reading skills(pp. 55–75). San FranciscJossey–Bass.

Bock, J. K., & Levelt, W. J. M. (1994). Language prodtion: Grammatical encoding. In M. A. Gernsbac(Ed.), Handbook of psycholinguistics(pp. 945–984)San Diego: Academic Press.

ock, J. K., & Miller, C. A. (1991). Broken agreemeCognitive Psychology,23, 45–93.

ock, J. K., Nicol, J., & Cutting, J. C. (1999). The ties tbind: Creating number agreement in speech.Journal ofMemory and Language,40, 330–346.

osch, P. (1986). Pronouns under control: A reply to LiliTasmowski and Paul Verluyten.Journal of Semantic5, 65–78.

rysbaert, M., & Mitchell, D. C. (1996). Modifier attacment in sentence parsing: Evidence from Dutch.Quar-terly Journal of Experimental Psychology,49A, 664–695.

acciari, C., Carreiras, M., & Cionini, C. B. (1997). Whwords have two genders: Anaphor resolution for Itafunctionally ambiguous words.Journal of Memory anLanguage,37, 517–532.

hang, F. R. (1980). Active memory processes in vicomprehension: Clause effects and pronominal rence.Memory & Cognition,8, 58–64.

lark, E. V. (1978). From gesture to word: On the nathistory of deixis in language acquisition. In J.Bruner & A. Garton (Eds.),Human growth and deveopment: Wolfson College lectures 1976.Oxford, UK:Oxford Univ. Press.

lark, E. V., & Sengul, C. J. (1978). Strategies inacquisition of deixis.Journal of Child Language,5,457–475.

Clark, H. H., & Sengul, C. J. (1979). In search of referefor nouns and pronouns.Memory & Cognition, 7,35–41.

Cloitre, M., & Bever, T. G. (1988). Linguistic anapholevels of representations, and discourse.Language anCognitive Processes,3, 293–322.

Cowart, W., & Cairns, H. S. (1987). Evidence for an aphoric mechanism within syntactic processing: Sreference relations defy semantic and pragmaticstraints.Memory & Cognition,15, 318–331.

Cuetos, F., & Mitchell, D. C. (1988). Cross-linguistic dferences in parsing: Restrictions on the use of theClosure strategy in Spanish.Cognition,30, 73–105.

Dell, G. S., McKoon, G., & Ratcliff, R. (1983). The acvation of antecedent information during the procesof anaphoric reference in reading.Journal of VerbaLearning and Verbal Behavior,22, 121–132.

Dell, G. S., & O’Seaghdha, P. G. (1992). Stages of lexaccess in language production.Cognition, 42, 287–314.

Drosdowski, G. (1984).Duden Grammatik der deutschGegenwartssprache(Vol. 4). Mannheim: Dudenvelag.

Ehrlich, K., & Rayner, K. (1983). Pronoun assignmentsemantic integration during reading: Eye movem

l-

l

-

e

l

s

and immediacy of processing.Journal of VerbaLearning and Verbal Behavior,22, 75–87.

Garnham, A., Oakhill, J., Ehrlich, M.-F., & Carreiras,(1995). Representations and processes in the intetation of pronouns: New evidence from SpanishFrench.Journal of Memory and Language,34,41–62

Garrod, S. C., & Sanford, A. J. (1994). Resolving sentein a discourse context: How discourse representaffects language understanding. In M. A. Gernsba(Ed.), Handbook of psycholinguistics(pp. 675–698)San Diego: Academic Press.

Gernsbacher, M. A. (1989). Mechanisms that improveerential access.Cognition,32, 99–156.

Gernsbacher, M. A. (1991). Comprehending concepanaphors.Language and Cognitive Processes,6, 81–105.

Greene, S. B., Gerrig, R. J., McKoon, G., & Ratcliff,(1994). Unheralded pronouns and managemencommon ground.Journal of Memory and Languag33, 511–526.

Gilboy, E., Sopena, J.-M., Clifton, C., Jr., & Frazier,(1995). Argument structure and association preences in Spanish and English complex NPs.Cognition,54, 131–167.

Hagoort, P., Brown, C., & Groothusen, J. (1993).syntactic positive shift (SPS) as an ERP measursyntactic processing.Language and Cognitive Prcesses,8, 439–483.

Hemforth, B., Konieczny, L., & Scheepers, C. (1999). Stactic attachment and anaphor resolution: Two siderelative clause attachment. In M. Crocker, M. Picking, & C. Clifton (Eds.), Architectures and mechnisms for language processingCambridge, UK: Cambridge Univ. Press.

Jarvella, R. J. (1971). Syntactic processing of connespeech.Journal of Verbal Learning and Verbal Behaior, 10, 409–416.

Koriat, A., & Greenberg, S. N. (1994). The extractionphrase structure during reading: Evidence from ledetection errors.Psychonomic Bulletin & Review,1,345–356.

Levelt, W. J. M. (1989).Speaking: From intention to artiulation. Cambridge, MA: MIT Press.

McKoon, G., Gerrig, R. J., & Greene, S. B. (1996). Pronresolution without pronouns: Some consequencememory-based text processing.Journal of Experimental Psychology: Leaning, Memory, and Cognition,22,919–932.

Meyer, A. S. (1992). Investigation of phonological encing through speech error analyses: Achievements,itations, and alternatives.Cognition,42, 181–211.

Meyer, A. S. (1996). Lexical access in phrase and senproduction: Results from picture–word interferenceperiments.Journal of Memory and Language,35,477–496.

iozzo, M., & Caramazza, A. (1997). Retreival of lexicsyntactic features in tip-of-the-tongue states.Journal

nd

toing:dpea-

hor

ncen,

an

S ).

S y ofex-

ar-:

v -en

v eech

V --

((


of Experimental Psychology: Learning, Memory aCognition,23, 1410–1423.

Mitchell, D. C., & Brysbaert, M. (1998). Challengesrecent theories of crosslinguistic variation in parsEvidence from Dutch. In D. Hillert (Ed.),Syntax ansemantics: Sentence processing: A crosslinguisticspective(Vol. 31, pp. 313–335). Orlando, FL: Acdemic Press.

Murphy, G. L. (1985). Processes of understanding anapJournal of Memory and Language,24, 290–303.

Nicol, J. (1988).Coreference processing during sentecomprehension.Unpublished doctoral dissertatioMassachusetts Institute of Technology.

Sachs, J. S. (1967). Recognition memory for syntacticsemantic aspects of connected discourse.Perception &Psychophysics,2, 437–442.

chmitt, B. M., Meyer, A. S., & Levelt, W. J. M. (1999Lexical access in the production of pronouns.Cogni-tion, 69, 313–335.

r-

a.

d

temberger, J. P. (1992). The reliability and replicabilitnaturalistic speech error data: A comparison withperimentally induced errors. In B. J. Baars (Ed.),Ex-perimental slips and human error: Exploring thechitecture of volition (pp. 195–215). New YorkPlenum.

an Berkum, J. J. A. (1996).The psycholinguistics of grammatical gender.Nijmegen, The Netherlands: NijmegUniv. Press.

an Berkum, J. J. A. (1997). Syntactic processes in spproduction: The retrieval of grammatical gender.Cog-nition, 64, 115–152.

igliocco, G., Antonini, T., & Garrett, M. F. (1997). Grammatical gender is on the tip of Italian tongues.Psychological Science,8, 314–317.

Received July 10, 1998)Revision received February 16, 1999)

Documents

Representations and Processes in the Production of