UvA-DARE is a service provided by the library of the University of Amsterdam (http://dare.uva.nl)

UvA-DARE (Digital Academic Repository)

An experimental study on the learning of arbitrary and non-arbitrary gender of pseudo Dutchnouns by nonnative and native speakers of Dutch

Zekhnini, A.; Hulstijn, J.H.

Published in:Toegepaste Taalwetenschap in Artikelen

DOI:10.1075/ttwia.53.11zek

Link to publication

Citation for published version (APA):Zekhnini, A., & Hulstijn, J. H. (1995). An experimental study on the learning of arbitrary and non-arbitrary genderof pseudo Dutch nouns by nonnative and native speakers of Dutch. Toegepaste Taalwetenschap in Artikelen,53, 121-136. https://doi.org/10.1075/ttwia.53.11zek

General rightsIt is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s),other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).

Disclaimer/Complaints regulationsIf you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, statingyour reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Askthe Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam,The Netherlands. You will be contacted as soon as possible.

Download date: 08 Mar 2021

AN EXPERIMENTA L STUD Y O N TH E LEARNING O F ARBITRARY AND NON-ARBITRAR Y GENDE R O F PSEUDO DUTCH NOUN S B Y NONNATIVE AN D NATIV E SPEAKER S O F DUTCH

Ahmed Zekhnin i & Jan H . Hulstij n Vakgroep Taalkund e Vrije Universitei t Amsterda m

Introduction

There ha s bee n som e debat e i n th e recen t literatur e concernin g th e questio n of whether , an d t o wha t extent , nou n gende r attributio n i n language s suc h a s French, Germa n an d Dutc h ca n b e formulate d i n term s o f semanti c o r morphonological rules , i n term s o f competin g semanti c an d morphonologica l cues, o r onl y i n term s o f arbitrar y idiosyncrati c features . Consequently , opinions diffe r o n th e questio n ho w th e proces s o f gende r learnin g b y nativ e and nonnativ e learner s o f suc h language s ha s t o b e explained . I n on e vie w (Carroll, 1989) , French L1 learner s encod e nou n gende r a s a n inheren t featur e of th e nou n i n a n obligator y fashion . Englis h learner s o f French , however , learn noun s a s words distinc t fro m thei r determiners .

Carroll (1989 ) claim s "tha t th e gende r distinctio n i s no t learne d durin g primary acquisition . Rather , i t i s simpl y availabl e t o b e use d i f a languag e requires it . Thi s clai m i s consisten t wit h th e genera l absenc e o f erro r i n gender attributio n durin g th e early stage s o f Frenc h acquisition. " (p.556/557 )

Furthermore, Carrol l claim s "tha t th e gende r featur e i s a defining featur e of th e nou n class . Whe n th e chil d recognize s tha t th e sam e phonologica l sequence i s showin g u p ove r an d ove r agai n a s th e initia l par t o f th e noun , he o r sh e wil l extrac t i t fro m th e underlyin g phonologica l representation . (... ) Once thes e lexeme s [un, une , le , la , mon , m a etc] ar e independentl y represented, th e universall y availabl e gende r featur e relevan t t o definin g th e noun clas s i n th e morphosyntacti c lexico n wil l simpl y b e triggered . A t tha t point, no nomina l expressio n ca n b e lexicall y encode d withou t a gende r feature. Thi s follow s becaus e th e gende r featur e i s par t o f th e specificatio n o f what a noun is. " (571/572)

In summar y (p.580/581) , Carroll make s th e followin g claims : Gender feature s ar e definin g feature s o f th e nomina l clas s an d henc e appear a s suc h i n ever y lexica l entr y o f a nomina l categor y (noun s an d nominal suffixes) . Children learnin g Frenc h a s thei r L1 therefor e d o no t mak e systemati c developmental gende r errors . Anglophone childre n acquirin g Frenc h throug h immersio n program s d o not demonstrat e anythin g resemblin g native-speake r competence . The y

Toegepaste Taalwetenschap in Artikelen 53 (1995), 121-136. DOI 10.1075/ttwia.53.11ze k ISSN 0169-7420 / E-ISSN 2213-4883 All rights reserved

122 Zekhnini & Hulstijn

transfer their nou n categor y - crucially withou t a n inheren t gende r featur e to th e tas k o f acquirin g ne w words . Carroll, however , doe s no t mentio n th e followin g point , which , i n ou r

view need s t o be mad e i n orde r t o rende r mor e complet e th e pictur e o f wha t gender learnin g is : Once learner s hav e learn t t o reserve a place fo r th e gende r feature, th e plac e mus t b e fille d wit h a value , whic h i n Frenc h i s eithe r mal e or female . Settin g th e gende r valu e ha s t o b e don e fo r eac h nou n individually. Th e valu e ca n b e predicte d o n th e basi s o f rule s o r cue s (nonarbitrary assignmen t o f th e value ) fo r a numbe r o f noun s bu t canno t b e predicted fo r othe r noun s (arbitrar y assignment) .

We argu e tha t th e learnin g o f arbitrar y value s i s equally difficul t (o r easy ) for youn g nativ e speakers , olde r nativ e speakers , an d nonnativ e speakers . However, th e learnin g o f nonarbitrar y value s i s easie r fo r olde r nativ e speakers, sinc e the y hav e alread y acquire d man y noun s fro m whic h rule s o r cues ca n b e derived .

We wil l repor t o n tw o learnin g experiment s i n which adul t nonnativ e an d native speaker s o f Dutc h ha d t o lear n pseud o Dutc h nouns . W e attempte d t o falsify on e o f Carroll' s claims , b y providin g evidenc e fo r th e clai m tha t nonnative an d nativ e speaker s alik e woul d hav e equa l difficult y i n learnin g arbitrary nou n gende r attribution s (Exp . 1) , bu t tha t nativ e speaker s woul d have les s difficult y tha n nonnativ e speaker s i n learnin g les s arbitrar y gende r attributions (Exp . 2).

First, however , le t u s briefl y giv e som e informatio n abou t Dutc h nou n gender. Dutc h originall y ha d thre e gende r classes : masculine , feminin e an d neuter. Moder n Dutch , however , ha s tw o classes : a combine d masculin e an d feminine clas s (thes e noun s tak e th e definit e articl e "de"), and a neute r clas s (nouns takin g th e definit e articl e "het"). From no w on , w e wil l refe r t o thes e two classe s a s DE-noun s an d HET-nouns .

Part 1. The acquisition of arbitrary gender specifications

In par t on e o f ou r stud y w e teste d th e hypothesi s tha t ther e ar e n o differences betwee n nativ e an d nonnativ e speaker s i n learnin g Dutc h pseud o nouns wit h arbitrar y gender . Th e pilo t stud y aime d a t selectin g 8 pseud o nouns wit h arbitrar y gender . Thes e noun s wer e the n use d i n learnin g experiment 1 .

Pilot 1 Elicitation of nativ e speake r intuition s o n th e gende r o f 2 5 pseudo nouns . Ou r expectation i s tha t nativ e speaker s wil l assig n gende r o n basi s o f th e fonologic cues whe n thes e ar e presen t i n a pseudo noun . Thes e cue s ar e t o b e gnerated fro m rhyme s an d onset s tracebal e i n th e Dutc h language .

An experimental study on the learning of arbitrary and non-arbitrary gender 12 3

Method

Subjects 28 native speaker s (non-languag e universit y students )

Materials and procedure Subjects receive d a lis t o f 7 5 nouns (5 0 existing noun s + 25 pseud o nouns , i n random order) . They wer e instructe d t o write DE or before eac h noun . Ten minute s wer e allowe d fo r th e completio n o f thi s task . W e selecte d 2 5 monosyllabic pseud o (bu t possible ) noun s wit h varyin g onset s an d rhymes . Some o f thes e word s rhyme d wit h mor e DE-word s tha n HET-words , som e with mor e HET-word s tha n DE-words , an d som e wit h abou t th e sam e number o f DE- a s HET-words. Fo r example , ther e ar e 8 noun s (i n Martin , 1978) rhymin g o n th e pseud o wor d "nerk" , fou r words an d fou r DE words. Ou r predictio n wa s tha t subject s woul d respon d wit h abou t 50 % DE and 50 % . Furthermore, ther e ar e 5 noun s (i n Martin , 1978 ) rhymin g o n the pseud o wor d "luif" , al l DE words. Ou r predictio n therefor e wa s tha t subjects giv e DE responses only . I n summary , w e predicte d tha t nativ e speaker intuition s woul d correspon d wit h tendencie s foun d amon g rhymin g existing nouns .

Results The Pearso n correlatio n betwee n expecte d an d obtaine d proportion s o f DE/ responses wa s .7 6 (d f = 23) . W e selecte d 8 noun s whos e obtaine d DE proportions wer e closes t t o th e 50 % mark (rangin g betwee n 3 3 and 67%) : nerk, flond, loor, spam, glak, zweil, floom, and speg, for us e i n Experimen t 1 .

Experiment 1 In thi s experimen t nativ e an d normativ e speaker s ha d t o lear n th e 8 monosyllabic pseud o noun s wit h arbitrar y gende r tha t ha d bee n selecte d i n pilot stud y 1 . W e hypothesize d tha t nativ e an d normatives will hav e equa l difficulty i n learnin g th e arbitrar y gende r assignment s o f thes e hithert o ne w nouns (pseud o nouns) .

Subjects Seventy individual s wh o wer e pai d fo r thei r participation . * 35 native speaker s (non-languag e universit y students) ; * 3 5 intermediat e an d advance d nonnativ e speaker s (preparin g fo r Dutc h language universit y entr y examinations) , mostl y wit h non-Europea n firs t languages.

Task and materials A concept-learnin g tas k wa s use d unde r join t visua l an d auditor y presenta -tion. Th e experimenta l sessio n consiste d o f 8 learnin g + testin g phases , followed b y tw o unexpecte d extr a test s (test s 9 an d 10) . Subject s ha d t o

124 Zekhnini & Hulstijn

discover how eigh t name s (pseud o nouns ) relate to eigh t visua l concepts . Four name s referre d t o a n object , fou r name s referre d t o a location .

The visua l stimul i wer e 1 6 line-drawings, photographe d an d mounte d o n slides. Th e fou r object s wer e crosse d wit h th e fou r locations , yieldin g 1 6 object-location instance s (se e Figur e 1) . Th e auditor y stimul i fo r al l 1 6 instances wer e constructe d a s follows :

DE/ + [objec t noun ] + "is op" + DE/ + [location noun ]

The apparatu s use d fo r th e join t exposur e o f th e visua l an d auditor y stimul i was a Bel l & Howel l RM 850 random-acces s tape-slid e projector , whic h incorporates a built-i n cassett e playe r an d a microprocessor , controllin g slid e selection.

Of th e eigh t noun s selecte d i n Pilo t 1 , fou r wer e use d a s referrin g t o a n object an d fou r a s referrin g t o a location . O f eac h o f thes e tw o set s o f fou r nouns, two wer e assigne d th e DE article an d tw o th e article:

Object nouns : de nerk Location nouns : de glak de flond de zweil het loor het floom het spam het speg

Figure 1: The four objects (left column) crossed with the four locations (top row), yielding 16 combinations (cells)

An experimental study on the learning of arbitrary and non-arbitrary gender 12 5

Het loor is op De nerk is op De nerk is op Deflond is op De flond is op Het spam is op Het spam is op Het loor is op

Procedure and instructions Subjects participate d i n smal l group s (betwee n 5 and 8 subjects a t a time) . They wer e tol d tha t the y ar e participatin g i n a concept-learnin g experimen t involving th e processin g o f unknow n language . The y wer e alo informed about th e learnin g an d testin g phase s i n eac h session . Th e experimen t consist s of eigh t sessions .

In th e 8 learnin g an d testin g phases , onl y 8 o f th e 1 6 object-locatio n instances wer e presente d an d tested . The y wer e chose n suc h tha t tw o consecutive instance s alway s share d eithe r thei r objec t o r thei r locatio n t o fcilitate th e leming task (number s 1 throug h 8 i n Figur e 1) . Th e eigh t learning-testing phase s eac h bega n wit h a differen t instanc e (wit h 8 , 3, 7, 2, 6, 5, 4 , 1 respectively).B y wa y o f example , her e ar e th e auditor y stimul i i n learning phas e 1 :

de glak. de glak. de zweil. de zweil het floom. het floom. het speg. het speg.

Immediately afte r testin g phas e 8 , subject s wer e give n tw o unexpecte d additional tests . Tes t 9 involve d th e 8 remainin g instances , no t presente d i n the precedin g 8 learning-testin g phase s (number s 9 throug h 1 6 i n Figur e 1) . The picture s wer e show n (th e orde r o f presentatio n wa s 11 , 15, 9, 12 , 14 , 16, 10, 13 ) an d subject s wer e require d t o writ e dow n th e correspondin g sentences. Thus , tes t 9 purporte d t o measur e t o wha t exten t subject s wer e able t o transfe r th e acquire d knowledg e (instance s 1-8 ) t o ne w instance s (9 -16). Finally , i n tes t 10 , subject s wer e show n th e picture s o f th e fou r object s and th e fou r location s separately . The y wer e require d t o writ e dow n th e name precede d b y th e articl e DE or .

It wa s i n tes t 1 0 only tha t subjects ' attentio n wa s draw n t o th e article . I n the 8 testing-learnin g phas e sessions , thei r attentio n ha d bee n draw n onl y t o the name s (nouns) . Thus , th e learnin g tas k wa s designe d a s a tas k whic h purported t o brin g abou t a n incidenta l learnin g behaviour , i n tha t subjects ' attention wa s focuse d o n th e name s (nouns ) themselve s rathe r tha n o n thei r accompanying articles .

Scoring procedures For each subjec t an d fo r eac h test , the response s wer e code d a s follows : - noun correc t - noun incorrec t o r absen t - noun correc t an d articl e correc t - noun correc t bu t articl e incorrec t o r absen t Since w e wer e no t intereste d i n spelling , th e followin g spelling s wer e code d as being correct :

126 Zekhnini & Hulstijn

- nerk, nerck - flond, flont, vlond, vlont - loor - spam - glak, glack - zwijl, zweïl, swijl, szveil - floom, vloom - speg, spech

Results We firs t present the result s o f th e fina l thre e tests (8-10), sinc e the y wer e administered afte r th e eigh t learnin g phases , an d thereb y represen t th e learning outcome .

Table 1: Correct noun forms with correct article: Means and standard deviations

Native speakers N = 35

Normative speakers N = 35

t-test (df = 68)

Test 8 J Ma x = 16

13.9 (2.5)

12.6 (3.9)

1.57 NS

Test 9 J Ma x = 16

13.4 (3.2)

11.8 (3.8)

1.82 NS

Test 10 Max = 8

6.7 (1.4)

6.4 (1.6)

0.60 NS

Test 8 elicited subjects ' bes t performanc e sinc e tha t tes t measure d th e exact instance s whic h ha d bee n presente d an d teste d i n th e previous eigh t learning /testing sessions .

Figure 2 presents th e learnin g curve s o f correc t noun s wit h correc t article s from Tes t 3 through Tes t 8 (performance i n Test s 1 and 2 was to o poo r t o b e informative).

An experimental study on the learning of arbitrary and non-arbitrary gender 12 7

Correct nou n forms wit h correct articl e (Exp . 1 ) maximum = 16

Test 2 Test 4 Tes t 5 Tes t 6 Tes t 7 _*_ Native s _*_ _ Non-native s

Test8

Figure 2

Error analyses Error analyse s (Chi-squar e tests ) o n th e responses i n test s 8 , 9, and 1 0 were conducted t o asses s

to wha t exten t th e eigh t nouns (withou t thei r articles ) wer e successfull y learned b y native s an d nornnatives; to what exten t gende r error s (wron g article s when noun s wer e correct ) differed amon g th e eight pseud o word s by native s an d nonnatives .

The result s o f th e erro r analyse s ca n be summarized a s follows : I n tes t 8 , the nouns 'floom ' an d 'flond ' ha d bee n learne d les s wel l tha n th e othe r nou -ns. This wa s tru e fo r native s an d nonnative s alike . Perhaps subject s wer e con -fused b y th e identica l onset s o f thes e noun s (althoug h th e noun s 'spam ' an d 'speg', also with identica l onsets , were no t mixe d up) . In test s 9 and 10 , ho-wever, n o significan t difference s wer e foun d amon g th e 8 nouns.

There wer e n o significan t difference s amon g th e 8 nouns i n term s of article error s i n tes t 8 . However, i n tes t 9 significantly fewe r articl e error s were obtaine d fo r th e nou n 'nerk ' tha n fo r th e othe r nouns ; i n tes t 10 , fewer article error s tha n averag e wer e obtaine d fo r 'nerk ' an d mor e tha n averag e for 'flond' . I n none o f th e thre e test s (8 , 9, and 10) , however, th e article-erro r pattern o f native s differe d fro m tha t o f nonnatives .

128 Zekhnini & Hulstijn

Summary and Conclusions In experimen t 1 , eight monosyllabi c pseud o noun s wit h arbitrar y gende r wer e taught t o native an d normativ e speakers . The hypothesis wa s tha t nativ e an d nonnatives woul d hav e equa l difficult y i n learnin g thes e arbitrar y gende r assignments. Thi s hypothesi s wa s supported : althoug h native s learne d th e nouns somewha t mor e quickl y tha n th e nonnatives , thei r ultimat e performan -ce did no t diffe r fro m tha t o f th e natives (i.e . in test s 8 , 9, and 10) : no signifi -cant difference s wer e foun d betwee n native s an d nonnative s - in th e mea n numbe r o f noun s learned , - i n th e mean numbe r o f gende r errors , - i n th e distributio n o f nou n for m error s amon g th e 8 target words ; - i n th e distributio n o f gende r error s amon g th e 8 target words .

We dra w th e followin g tw o conclusion s fro m Experimen t 1 : 1. Contrary t o Carroll' s claim , native speaker s d o sometime s mak e articl e errors, i.e . they d o no t alway s encod e th e articl e (gende r feature ) o f ne w nouns immediatel y an d flawlessly togethe r wit h th e noun form s themselves , at leas t no t i n th e case of Dutc h nativ e speaker s learnin g pseud o monosylla -bic nouns whos e gende r canno t be predicted o n th e basis o f morphological , phonological o r semanti c rule s o r statistica l cues . 2. With suc h nouns , gender acquisitio n i s equally difficul t (o r easy ) fo r nativ e and normativ e speakers .

Part 2. The acquisition of nonarbitrary and arbitrary gender specifications

In stud y 2 , we aime d a t testin g th e followin g hypotheses : HI Nativ e speaker s wil l lear n th e article s o f pseud o noun s whos e gende r ca n

be predicte d o n th e basi s o f morphologica l o r phonologica l statistica l cue s (i.e. nonarbitrary gender ) mor e quickl y tha n th e article s o f pseud o noun s whose gende r canno t be predicted o n th e basis o f an y rul e o r cu e (arbitra -ry gender) .

H2 Fo r normative speakers , however , th e differenc e i n learnin g difficult y between nonarbitrar y an d arbitrar y gende r wil l be les s pronounced tha n for nativ e speakers , since nornnativ e speaker s hav e take n i n smalle r amounts o f inpu t an d hav e acquire d smalle r vocabularie s an d ar e therefore les s susceptible t o statistica l tendencie s (cause d b y rule s o r cues ) than nativ e speakers .

Pilot 2

The pilo t stud y aime d a t selectin g 8 pseudo noun s wit h arbitrar y gender . These nouns wer e the n use d i s learning experimen t 2 . Elicitation of nativ e speake r intuition s o n th e gende r o f 1 8 pseudo noun s

An experimental study on the learning of arbitrary and non-arbitrary gender 12 9

Method

Subjects 41 native speaker s (non-languag e universit y students ) participate d i n thi s pilot study .

Materials and procedure List o f 6 8 nouns (5 0 existing noun s + 18 pseudo nouns , i n rando m order) . Subjects wer e instructe d t o write DE or before eac h noun . Te h minute s were allowe d fo r th e completio n o f thi s task . Eighteen pseud o noun s wer e used :

8 twosyllabi c noun s whos e gende r ca n be predicted o n th e basi s of morphological o r phonologica l cues ;

2 twosyllabic noun s endin g o n -ing 2 twosyllabic noun s endin g o n -de 2 twosyllabic noun s beginnin g wit h be-2 twosyllabic noun s beginnin g wit h ont-

10 monosyllabic nouns , whose gende r i s (4 nouns) o r i s not ( 6 nouns) cued (an d therefor e ca n o r canno t b e predicted ) b y rhymin g existin g words.

Results Our expectancie s wer e confirme d

in th e cas e o f 7 out o f th e 8 twosyllabic words ; in th e cas e o f 3 out o f th e 4 monosyllabic word s wit h a preponderance o f rhyming DE words; in th e cas e o f 5 out o f th e 6 words withou t a preponderance o f rhymin g DE or words.

From thes e 1 8 pseudo noun s th e best one s (i n terms o f native speakers ' intuitions confirmin g ou r predictions ) wer e selecte d fo r us e i n Experimen t 2 . See Table 2

130 Zekhnini & Hulstijn

Table 2: Target words selected for Experiment 2

Target Word Prediction Cue

zering DE Strong morphological/phonologica l cu e node DE Strong phonologica l cu e (5 3 out o f 5 3 two-

syllabic nouns o n -de have DE bevrog Strong morphological/phonologica l cu e

(45 out o f 49 twosyllabic noun s beginnin g with be- have )

ontgref Moderate morphologica l cu e ( 7 our o f 8 twosyllabic noun s beginnin g wit h ont-have )

doon DE Moderate phonologica l cu e ( 7 out o f 8 rhy-ming word s hav e DE)

wear DE Moderate phonologica l cu e ( 7 out o f 7 rhy-ming word s hav e DE)

schok DE/ No cu e (o f 1 4 rhyming word s 7 have DE and 7 have )

jeer DE/ No cu e (o f 1 4 rhyming word s 8 have DE and 6 have )

Discussion -ING: all nouns wit h th e affi x -ing have DE-gender . Ther e ar e innumerabl e nouns wit h thi s -ing affix, o f which 16 4 twosyllabic noun s endin g o n -ring, of which 7 on -ering. (Note. In th e case o f a handful o f -ing nouns, th e ending -ing is not a n affi x (boun d morpheme ) bu t par t o f th e noun root. )

-DE: all twosyllabi c noun s endin g o n th e unstresse d syllabl e -de are DE words. Thi s endin g i s not a n affi x (boun d morpheme) . Ther e ar e 53 twosyllabic noun s endin g o n -de (40 frequent an d 1 3 infrequent), 6 of whic h ending o n -ode (5 frequent noun s an d on e rar e noun) .

BE-: Ther e ar e 4 9 twosyllabic noun s beginning wit h th e unstresse d syllabl e be-, 45 nouns an d 4 DE nouns. These 4 DE nouns ar e no t relate d t o (derived o r converte d from ) correspondin g verbs . Most o f th e 45 nouns, however, hav e bee n derive d (o r converted ) fro m verb s beginning wit h be-. Not al l verbs beginnin g wit h be- have been derive d fro m verb s withou t be-. Hence th e statu s o f be- as a n affi x (derivationa l morpheme ) i s debatable .

ONT-: There ar e 8 twosyllabic noun s beginnin g wit h unstresse d ont-, 7 words an d on e DE word. Al l these 8 nouns hav e been derive d (o r converted ) from verb s beginnin g wit h ont-. All thes e verbs beginning wit h ont- have been derive d fro m verb s withou t ont-. Therefore, ont- can be considere d t o b e an affix .

An experimental study on the learning of arbitrary and non-arbitrary gender 13 1

Experiment 2 In thi s experimen t nativ e an d nonnativ e speaker s ha d t o learn th e 8 pseudo nouns tha t ha d bee n selecte d i n pilo t stud y 2 . We hypothesized tha t nativ e and nonnative s wil l hav e equa l difficult y i n learning th e arbitrar y gende r assignments o f thes e hithert o ne w noun s (pseud o nouns) , but tha t native s would hav e les s difficulty tha n nonnative s i n learnin g th e non-arbitrary ones . Subjects wer e taugh t fou r twosyllabi c an d fou r monosyllabi c pseud o noun s whose gende r

could b e predicted o n th e basis o f phonologica l o r morphologica l cue s (i n the cas e o f th e 4 twosyllabic nouns) ; could b e predicted o n th e basis o f phonologica l rhymin g cue s (i n the cas e of tw o monosyllabi c nouns) ; could no t b e predicted o n th e basis o f an y cue s (i n th e cas e o f tw o mono -syllabic nouns) .

Method

Subjects Fifty seve n individual s participate d i n thi s experiment : * 29 native speaker s (non-languag e universit y students) ; * 28 intermediate an d advance d nonnativ e speaker s (preparin g fo r Dutc h language universit y entr y examinations) , mostly wit h non-Europea n firs t languages. Subject s wer e paid fo r thei r participation .

Task an d procedure s wer e identica l t o the tas k an d th e procedure s i n experiment 1 .

Materials The sam e picture s wer e use d a s i n Experimen t 1 . However, a new recordin g was mad e o f th e audiotape , replacin g th e 8 target word s o f Experimen t 1 by the 8 target word s o f Experimen t 2 .

Results We presen t th e result s o f Experimen t 2 in term s o f proportion s o f error s made durin g th e entir e experimen t (test s 1 through 10) , as thes e dat a ar e most relevan t t o our hypotheses . Table s 3 and 4 give thes e proportions ; erro r proportions ar e graphicall y displaye d i n Figure 3 . Table 5 provides th e sam e data a s Table 3 but ran k ordere d i n terms o f increasin g erro r proportions .

132 Zekhnini & Hulstijn

Table 3: Mean proportions of gender errors during experiment 2 (tests 1 through 10)

Natives N - 2 9

Normatives N = 28

t-tests

Nonarbitrary gender

de zering 00 08 *

de node 01 05 NS het bevrog 03 18 **

het ontgref 03 20 **

de doon 03 06 NS |

de weur 11 18 NS 1

Arbitrary gende r

het schak 08 31 **

het jeer 23 25 NS

All targe t word s 07 16 **

Table 4: Mean proportions of gender errors during experiment 2 (tests 1 through 10)

Natives N = 29

Normatives N = 28

t-tests

All fou r DE words

04 09 NS

All fou r words

09 23 **

All fou r twosyllabic word s

02 13 **

All fou r mono -syllabic words

11 20 **

All eigh t targe t words

07 16 **

An experimental study on the learning of arbitrary and non-arbitrary gender 13 3

Mean proportion s o f gender error s (Exp . 2) (Test 1 trough 10)

Figure 3

£ 2 0

de zering d e node hetbevro g he t ontgref de doon d e weur b et schak b et jeer Target Words

. Natives . Non-natives

Table 5: Rank orders of difficulty in experiment 2 in mean percentages of gender errors*

Natives Nornnatives

00 zering 05 node

01 node 05 doon

08 zering

03 bevrog

03 doon 17 weur

03 ontgref 18 bevrog

19 ontgref

08 schak

25 jeer

11 weur

31 schak

23 jeer

*) Identica l ran k order s wer e obtaine d (a ) whe n absolut e number s o f gende r error s wer e calculated an d (b ) whe n th e numbe r o f subject s w h o committe d a t leas t on e gende r erro r o n th e word concerne d wer e calculated .

134 Zekhnini & Hulstijn

The followin g Manova's were conducte d 1. Manova on mea n percentages of gende r error s in al l 10 tests by Group (natives vs normatives), with repeate d measure s o n th e 8 target words . Results: a main Grou p effec t ( F = 14.83 ; df = 1,44 ; ρ <Ό01) , a main Targe t Word effec t ( F = 8.67; df = 7,308; ρ < .001) , but n o significan t Grou p χ Target Word interaction . 2. A Manova o n mea n percentage s o f gende r error s i n al l 1 0 tests by Grou p (natives v s nonnatives) , with repeate d measure s o n th e followin g fou r pair s of targe t words : 1 + 2, 3 + 4, 5 + 6, and 7 + 8. Results: a main Grou p effec t ( F = 14.83 ; df = 1,44 ; ρ <.001) , a main effec t fo r Word Pair s ( F = 13.79 ; df = 3,132; ρ < .001), but n o significan t Grou p χ Word Pair interaction . 3. A Manova o n mea n percentage s o f gende r error s i n al l 1 0 tests by grou p (natives v s nonnatives) , with repeate d measure s o n Gender (th e 4 DE words against th e fou r words). Results: a main Grou p effec t ( F = 14.83 ; df = 1,44 ; ρ <.001) , a main Gende r effect ( F = 20.70; df = 1,44; ρ < .001) , but n o significan t Grou p χ Gender inter -action. 4. A Manova o n mea n percentage s o f gende r error s i n al l 1 0 tests by Grou p (natives v s nonnatives) , with repeate d measure s o n Syllabicity (th e fou r two -syllabic agains t th e fou r monosyllabi c words) . Results: a main Grou p effec t ( F = 14.83 ; df = 1,44 ; ρ <Ό01) , a main Syllabicit y effect ( F = 15.14 ; df = 1,44 ; ρ < .001) , but n o significan t Grou p χ Syllabicity interaction.

Summary of the results of Experiment 2 The firs t hypothesi s (H1 ) was largel y supported :

Natives hardl y mad e an y gende r error s wit h 5 out o f 6 nouns wit h non -arbitrary gender ; Nonnatives mad e substantiall y an d significantl y mor e gende r error s wit h the 2 arbitrary noun s tha n wit h th e 6 nonarbitrary nouns ;

Within th e grou p o f 6 nouns wit h nonarbitrar y gende r cue s no clea r differences wer e foun d betwee n th e 3 nouns wit h strong gender cue s (stemming fro m gang s o f more tha n 4 5 similar nouns ) an d th e 3 nouns wit h moderate gender cue s (stemmin g fro m gang s o f only 7 or 8 similar nouns) .

The secon d hypothesi s (H2 ) was als o largely supported . Althoug h th e 2 nouns wit h arbitrar y gende r generate d mor e error s tha n th e 6 nouns wit h nonarbitrary gender , th e differences ar e les s pronounced a s i n th e cas e of th e Nonnatives, sinc e thre e o f th e six nonarbitrary noun s stil l generated a consi -derable numbe r o f errors . The error patter n canno t b e explained i n term s of the cu e typ e (gan g size , morphological o r phonologica l nature , numbe r o f syllables).

An experimental study on the learning of arbitrary and non-arbitrary gender 13 5

Conclusions

1. Contrary t o what on e migh t expec t o n th e basis o f Carrol l (1989) , native speakers o f Dutc h d o mak e gende r error s when learnin g ne w word s i n thei r native language . Hence i t i s not th e case that the y automaticall y an d withou t failure encod e th e gende r featur e upo n th e firs t fe w encounter s wit h ne w nouns. 2. When learnin g ne w noun s i n thei r nativ e language , native speaker s o f Dutch mak e fewe r gende r error s learnin g noun s wit h nonarbitrar y gende r than whe n learnin g noun s wit h arbitrar y gender . 3. Native speaker s o f Dutc h ar e superio r t o nonnative speaker s o f Dutc h onl y when learnin g th e gende r o f noun s wit h nonarbitrar y gender . The y ar e no t superior, however , whe n learnin g noun s wit h arbitrar y gender .

Speculations and questions for further research The result s o f thi s study sugges t tha t ther e doe s no t appea r t o be a funda -mental differenc e i n th e way nonnatives and native s proces s ne w linguisti c materials a s fa r a s noun gende r i s concerned. T o the exten t tha t w e foun d a difference betwee n nonnative s an d natives , namely i n th e error s mad e wit h nonarbitrary gender , th e differenc e i s not a fundamenta l one . I t only reflect s the fac t tha t adul t native s hav e ha d muc h mor e inpu t tha n nonnative s an d that therefor e the y hav e accumulate d stronge r gende r cues . Both.groups hav e passed th e momen t tha t the y discovere d tha t noun s i n th e Dutc h languag e must b e code d fo r gende r an d bot h group s ar e subjec t t o th e sam e cumulati -ve cues emanatin g fro m th e inpu t wit h whic h the y hav e bee n confronted . Natives hav e undergon e thi s influenc e fo r a longer perio d tha n nonnatives . To the exten t tha t gende r i s nonarbitrary then , gende r cue s ar e stronger fo r natives tha n fo r nonnative s (i.e . nonnatives a t a n intermediat e leve l o f L 2 proficiency). T o the exten t tha t gende r i s arbitrary, however , i t poses a n equally difficult (o r easy) learnin g proble m fo r native s a s fo r nonnatives .

How ca n th e fac t tha t native s di d no t perfor m errorles s b e reconcile d with th e commo n belief , a s expressed i n Carrol l (1989) , that smal l childre n learning thei r mothe r tongu e see m t o have no problem s a t al l learnin g gende r values, once the y hav e discovere d tha t noun s mus t b e obligator y specifie d fo r gender? O n firs t glance , i t seems tha t childre n indee d d o no t mak e man y gender errors . We wonder however , t o what exten t thi s i s true . Most word s are learne d betwee n th e age s o f si x and sixteen , when childre n g o t o schoo l and becom e acquainte d wit h larg e numbers o f ne w concept s an d thei r wor d forms. Fe w studie s giv e figure s o f gende r error s (error s o f incorrec t article s and othe r determiners ) elicite d o f childre n i n thi s ag e range . Furthermore , t o our knowledg e n o studie s hav e investigate d th e correctnes s o r incorrectnes s of articl e us e when nouns were produced for the first time. Thus, when a stud y found tha t childre n mad e ver y fe w nou n gende r error s (wron g articl e o r determiner use) , the questio n stil l remain s whether , fo r eac h chil d an d fo r each individua l wor d elicited , ther e ma y hav e been a n initia l perio d prio r t o the momen t o f elicitation, that th e gende r valu e ha d no t ye t wel l bee n encoded i n th e menta l lexico n (Corbett , 1991 ; de Houwer , 1987; Mills, 1986).

136 Zekhnini & Hulstijn

Let u s speculat e fo r a moment tha t L1 acquisition studie s woul d indispu -tably demonstrat e tha t smal l children lear n arbitrar y nou n gende r value s without difficult y (i.e . hardly makin g productio n error s wit h article s an d demonstratives), how woul d w e explai n th e differenc e betwee n thei r eas e of learning arbitrary , non-semantic , forma l feature s an d th e difficult y wit h which adult s (native s an d nonnatives alike) appea r t o lear n suc h feature s (cf. Andersson, 1992; MacWhinney, 1989 ; Magnan, 1983 ; Rogers, 1987 ; Sokolik & Smith, 1992 ; Taraban, McDonald & MacWhinney, 1989) ? What w e ha d t o explain the n i s a difference no t between firs t an d secon d languag e learnin g but a differenc e betwee n younge r an d olde r learner s (a s fa r a s arbitrary , non -semantic, forma l feature s ar e concerned) . Might tha t be a matte r o f mnemoni c rather tha n o f linguisti c abilities ? D o young childre n perceive , encode an d retain verbal , formal , non-semanti c informatio n i n a different wa y tha n adults?

Such speculation s an d question s clearl y cal l fo r furthe r researc h int o th e learning o f arbitrar y lexica l feature s b y younge r versu s olde r an d nativ e versus nonnativ e individuals . The result s o f th e presen t stud y sugges t tha t the difference s betwee n nativ e an d nonnativ e learner s ma y no t b e fundamen -tally different . Mor e researc h i s needed int o th e questio n whethe r ther e ar e fundamental learnin g difference s betwee n youn g childre n an d adults .

Bibliography

Andersson, A.-B . 1992. Second language learners' acquisition of grammatical gen-der in Swedish. University o f Göteborg, Sweden: Departmen t o f Linguis -tics.

Carroll, S . 1989. Second-language acquisitio n an d th e computationa l paradig -m. Language Learning, 39/4 , 535-594.

Corbett, G . 1991 . Gender. Cambridge , U.K. : Cambridge Universit y Press . de Houwer , Α. 1987. Nouns an d thei r companion s - or how a three-year-ol d

handles th e Dutc h gende r system . Belgian Journal of Linguistics, 2, 55-74. MacWhinney, B . 1989. Competition an d connectionism . In : B. MacWhinney &

E. Bates (eds.), The cross-linguistic study of sentence processing (pp . 422-457). Cambridge, U.K. : Cambridge Universit y Press .

Magnan, S.S . 1983 . Age an d sensitivit y t o gender i n French . Studies in Second Language Acquisition, 5/2, 194-212 .

Martin, W . 1971 . Inverte frequentielijst van het Nederlands. Universitei t van Leu-ven: Instituu t voo r Toegepast e Linguïstiek .

Mills, A.E. 1986. The acquisition of gender: A study of English and German. Berlin : Springer.

Rogers, M. 1987 . Learners' difficultie s wit h grammatica l gende r i n German a s a foreig n language . Applied Linguistics, 8, 48-74.

Sokolik, M.E . & Smith, M.E. 1992 . Assignment t o gende r o f Frenc h noun s i n primary an d secondar y language : A connectionis t model . Second Language Research, 8/1 , 39-58.

Taraban, R., McDonald, J.L. & MacWhinney, B . 1989. Category learnin g i n a connectionist model : Learning t o decline th e German definit e article . In: R.I. Corrigan, F. Eckman & M. Noonan (eds.). Linguistic Categorization (pp . 164-193). Amsterdam, Netherlands : Benjamins .