age constraints on second-language acquisition

0

Journal of Memory and Language41, 78–104 (1999)Article ID jmla.1999.2638, available online at http://www.idealibrary.com on

Age Constraints on Second-Language Acquisition

James Emil Flege

University of Alabama at Birmingham

Grace H. Yeni-Komshian

University of Maryland

and

Serena Liu

University of Alabama at Birmingham

This study evaluated the critical period hypothesis for second language (L2) acquisition. Theparticipants were 240 native speakers of Korean who differed according to age of arrival (AOA) inthe United States (1 to 23 years), but were all experienced in English (mean length of residence515 years). The native Korean participants’ pronunciation of English was evaluated by having listenersrate their sentences for overall degree of foreign accent; knowledge of English morphosyntax wasevaluated using a 144-item grammaticality judgment test. As AOA increased, the foreign accentsgrew stronger, and the grammaticality judgment test scores decreased steadily. However, unlike thecase for the foreign accent ratings, the effect of AOA on the grammaticality judgment test scoresbecame nonsignificant when variables confounded with AOA were controlled. This suggested that theobserved decrease in morphosyntax scores was not the result of passing a maturationally definedcritical period. Additional analyses showed that the score for sentences testing knowledge of rulebased, generalizable aspects of English morphosyntax varied as a function of how much education theKorean participants had received in the United States. The scores for sentences testing lexically basedaspects of English morphosyntax, on the other hand, depended on how much the Koreans usedEnglish. © 1999 Academic Press

Key Words:second language acquisition; phonology; morphosyntax; age; critical period; languageuse; education.

agnce

diesrgetla-ci-ry.A israllu-ofntsb;tsEn-rt,L2

92idinavi

rsusar

ck,s ofnd), Jingnd

ld bSc

03,

Many studies examining second langu(L2) acquisition have focused on the influe

This research was funded by NIDCD Grant DC028The authors are indebted to Jackie Johnson, for provthe grammaticality judgment test sentences, and to DBirdsong, for first suggesting the analysis of rule velexically based grammaticality judgment test items. Thegrateful to E. Bialystok, D. Birdsong, H. Hakuta, M. MaT. Piske, and L. White for commenting on earlier draftthis article, as well as the following University of Marylastudents: S. Kim and R. Byun (for help collecting dataKim and E. Kim (for their help preparing stimuli and testlisteners), and M. Cullen, L. Yun, A. Kinn, C. Romero, aM. Robbins (for help with data management).

Editorial correspondence and reprint requests shouaddressed to J. E. Flege, Department of Rehabilitationences, University of Alabama at Birmingham, VH 5
Birmingham, Alabama 35294. E-mail: [email protected].
78749-596X/99 $30.00Copyright © 1999 by Academic PressAll rights of reproduction in any form reserved.

eof age. The age variable examined in L2 stuis usually the age of first exposure to the taL2. In studies examining immigrant poputions, this is typically indexed by the partipants’ age of arrival (AOA) in the host countPrevious research has suggested that AOapparently an important determinant of ovedegree of foreign accent in the L2 (Flege, Mnro, & MacKay, 1995a), as well as degreeaccuracy in producing particular L2 consonaand vowels (Flege, Munro, & MacKay, 1995Munro, Flege, & MacKay, 1996). Age effechave also been reported for the learning ofglish morphosyntax (Johnson & Newpo1989).1 The observation of age effects on the

.gd

e

.

ei-

1
Strong age effects have been also been obtained in

exomL2aas

cay.thenged/oeebe

las88;avrfothge98p-

e oimllyingce

racerndblyteson

llyflu-anltyig

beeeofnin

dsis

heely

97).ac-riodges2

rilyr-ng-te

uresn).theef-ds

oth-is-onm-anillganalld,nif-

at-an

theof

the23

arsnKo-ter-, re-o ake

testn-

lty

icanen

OAr glisha les,A .g.,

79AGE AND L2 ACQUISITION

performance of adults—even those who areperienced in their L2—has suggested to sresearchers that the ability to acquire aneffectively is limited by a critical period. Suchconclusion is important practically, inasmuchit might influence decisions regarding edutional policy. It is also important theoreticall

The critical period hypothesis rests onassumption that the age-related effects seeL2 studies are the result of maturational chanin brain structures that are used to learn anto process language. For example, it has bhypothesized that as the brain matures, itcomes less “plastic” and that lost neural pticity impedes L2 learning (e.g., Scovel, 19Patkowski, 1980, 1990). However, others hproposed that age-related changes in L2 pemance derive from the nature and extent ofinteraction between a bilingual’s two languasystems (e.g., Oyama, 1979; Flege, 1987, 11995, 1998b; Bialystok, 1997). This latter aproach treats age as an index of the statdevelopment of the L1 system. It assumesplicitly that, all else being equal, the more fudeveloped the L1 system is when L2 learnbegins, the more strongly the L1 will influenthe L2.

Choosing between maturational and intetive accounts of age-related effects on L2 pformance is difficult. Neural development anative language acquisition are inextricaconfounded through much of childhood (Ba& Goodman, 1998). Also, the most commindex of age in L2 studies, AOA, is typicaconfounded with other variables that may inence L2 performance (see Flege, 19871998a, for discussions). Still another difficuis that there is no consensus as to how one mtest the critical period hypothesis. This iscause the critical period hypothesis has bapplied with less specificity to the studyage-related changes in L2 performance thahas been applied in ethological studies examing, for example, imprinting behavior in bir(Bornstein, 1989). The lack of specificity

studies which have examined the acquisition of AmerSign Language (e.g., Newport, 1990; Mayberry & Eich
1991; Emmorey, Bellugi, Friederici, & Horn, 1995). J
-e

-

insrn--

er-e

8,

f-

--

d

ht-n

it-

most crucial with respect to a definition of tstructures and/or functions that are putativaltered as the brain matures (Bialystok, 19For example, those who claim that foreigncents arise due to the passing of a critical pedo not specify whether the age-related chanarise from a loss of ability to articulate Lspeech sounds, a loss of ability to auditodistinguish L2 from L1 sounds or to form peceptual representations for L2 sounds in loterm memory, or a loss of ability to translasuch representations into articulatory gest(see Flege, 1987, 1988, 1995, for discussio

The aim of this study was to evaluatecritical period hypothesis by examining thefect of AOA on L2 performance. Three methowere used to evaluate the critical period hypesis. The first method will be called the “dcontinuity test.” The discontinuity test reststhe assumption that, in an AOA-stratified saple of L2 learners, participants who beglearning the L2 before the critical period wperform markedly better than those who belearning their L2 after the end of the criticperiod. The effect of a critical period coutherefore, be demonstrated by showing a sigicant departure from linearity in functions reling measures of L2 performance to AOA atappropriate AOA.

The present study was well suited fordiscontinuity test. The 240 native speakersKorean who participated had arrived inUnited States between the ages of 1 andyears and had lived there for at least 8 ye(mean5 15 years).2 Previous work has showthat as AOA increases, native speakers ofrean make more errors writing down compugenerated English sentences (Bott, 1993)spond more slowly and less accurately tlexical decision task (Kim, 1996), and mamore errors on a grammaticality judgment(Shim, 1995). Finally, Koreans who learn Eglish in adulthood are known to have difficu

,

2 The participants were selected on the basis of Aather than the age at which they first began to study Ent school in Korea. This is because, of the two variabOA is the more potent predictor of L2 performance (e
ohnson & Newport, 1989).

taina).s tvae ifo

ndsin

t 1thviin

ermr-resor-mangiodps

OAin

enticaldrioacionth-re/ut

L2ghsutoaned

L2useaig-ldhoese-althpe

eenthehofter

la-r-

ost-d toadreanA.

ingchi,Ashatce,leso-elf-teditedtion

odhy-al-as-

reas

isi-ereici-0.re

. Ififi-os-re-

heon-uldbe-

l tode-

odc-

80 FLEGE, YENI-KOMSHIAN, AND LIU

in accurately producing and perceiving cerEnglish vowels (Flege, Bohn, & Jang, 1997

There are, however, at least two drawbackusing the discontinuity test as a means to euate the critical period hypothesis. First, therdisagreement as to when the critical periodL2 acquisition ends. According to some, it eat 12 years of age (Scovel, 1988). But accordto others (e.g., Patkowski, 1990), it ends ayears. Second, not everyone would agreethe absence of a discontinuity provides edence that the critical period hypothesis iscorrect. Many investigators have used the t“sensitive period” and “critical period” intechangeably. The two notions appear toequally on the view that diminished L2 perfmance is the consequence of normal brainuration (see Bialystok, 1997, and Birdso1998). However, the notion of a sensitive perimplies that there will be a gradual, perhaeven linear, decline in L2 performance as Aincreases. Thus, the lack of a discontinuityAOA-L2 performance functions might be takas evidence against the existence of a criperiod for L2 acquisition. However, it wounot disprove the existence of a sensitive peand so would not rule out a maturationalcount of age-related changes in L2 acquisit

The second test of the critical period hypoesis employed here will be called the “ppostcorrelation test.” This test involves comping the correlation between AOA andperformance for groups of participants thouto have begun learning their L2 before verafter the end of a critical period. AccordingJohnson and Newport (1989), a significAOA–performance correlation will be observfor individuals who began learning theirbefore the end of the critical period, becaperformance declines increasingly as one nthe end of the critical period. However, a snificant AOA–performance correlation wounot be expected for a group of individuals whad all begun learning their L2 at varying timafter the end of the critical period. This is bcause “postmaturational” learners arethought to suffer to the same degree fromsame deficit, viz. having passed a critical
riod. The pre/postcorrelation test was imple
ol-sr

g5at--

t

t-,

l

d-.

-

ts

t

ers

le-

mented by computing a correlation betwAOA and each of the outcome measures ofstudy for the native Korean participants warrived in the United States before versus athe end of the putative critical period.

A potential problem for the pre/postcorretion test was that a significant AOA-L2 perfomance correlation could be obtained for pmaturational learners due to factors unrelatewhether a critical period (should one exist) hbeen passed. As already mentioned, the Koparticipants were selected according to AOHowever, as in previous studies examinlarge groups of immigrants (see, e.g., Ba1956; Bahrick et al., 1994), the Koreans’ AOwere confounded with other variables tmight influence L2 learning, L2 performanor both. As will be discussed, the variabconfounded with AOA in this study were chrnological age, the native Koreans’ average sestimated use of English, their self-estimause of Korean, years of residence in the UnStates, and the number of years of educathey had received in the United States.

In view of these confounds, a third methwas used here to assess the critical periodpothesis. The “matched subgroup” methodlowed us to test the hypothesis that factorssociated with AOA, rather than AOA itself, aresponsible for what have been interpreted“age” effects in previous studies of L2 acqution. In one set of analyses, subgroups wformed that consisted of native Korean partpants drawn from the original group of 24These subgroups differed in AOA but wematched for variables confounded with AOAthe AOA-defined subgroups differed signcantly, it would demonstrate that age, and psibly age-related maturational changes, wassponsible for the difference. However, if tAOA difference disappeared when the cfounded variables were controlled, it woshow that age was not responsible for thetween-group differences and would thus faisupport the existence of a maturationallyfined critical period.

As already mentioned, if a critical peridoes exist for L2 acquisition, it would be ne
-essary to define the structure(s) and/or func-

as isn-wnbeol-n-thsyn

thaaned

d aatushmrioin-&

atanr-

haogyw,

ishefo

logcu

an90f

amk-

photailre-d

truelsng

renis-

rtic-ear

ela-entam-s

m agec hanm 80,1 be-t ho-sM linei of1 itht er,K , &L ers( A–m at-k andN ui-s thec ageo 4).

int insw re-v ta,( s ofL ered ularl ings on-s taxw alityj 89)d ts ofE En-g ten-e ncess foro

edh or-p &N the


tion(s) that are altered by maturation. One wto help identify such structures and functionto determine if AOA effects differ across liguistic domains. For example, if it were shothat native-like performance were possibleyond a certain AOA for syntax but not phonogy, this would imply that “phonology and sytax are represented independently, orprocessing systems behind phonology andtax are different” (Bialystok, 1997, p. 120).

There is, in fact, widespread agreementage constrains the learning of L2 phonologymorphosyntax differently. Snow (1979) showthat phonology and morphosyntax emergeseparate factors in a study examining the nralistic acquisition of Dutch by native Englichildren and adults over a 1-year period. Soresearchers have concluded that a critical peexists only for phonology (Scovel, 1988; Sgleton, 1989; Bahrick, Hall, Goggin, Bahrick,Berger, 1994). Others have concluded thcritical period ends sooner for phonology thfor morphology or syntax (Long, 1990; Huford, 1991). Still others have concluded tmorphosyntax is learned more thoroughly,over a longer period of time, than is phonolo(Braine, 1971; MacWhinney, 1992; Sno1987).

Despite this convergence of views, therelittle empirical evidence for a difference in tphonological and morphosyntactic domainsexperienced adult speakers of an L2. Phonoand morphosyntax have been examined conrently in just four previous studies (Fathm1975; Oyama, 1973; Patkowski, 1980, 19Snow & Hoefnagel-Ho¨hle, 1982a,b). None othese studies examined an AOA-stratified sple of adults drawn from a single L1 bacground or assessed phonological and morsyntactic performance in comparable deFathman (1975) found that older childrenceived higher morphosyntax test scores thanyounger children, whereas the reverse heldfor pronunciation (see also Olson & Samu1973; Krashen, Long, & Scarcella, 1979; Estrand, 1982). Snow and Hoefnagel-Ho¨hle(1982a,b) found that adults and older childoutperformed younger children in both lingu
tic domains when tested soon after their firss
y

-

e-

td

s-

ed

a

tr

ryr-,;

-

-.

ide,-

exposure to the L2, whereas the younger paipants outperformed older participants 1 ylater.

Oyama (1973) observed a stronger corrtion between AOA and degree of foreign accthan between AOA and the scores on a grmaticality judgment test (r 5 .81 vs. .41). Thi

ight be taken as support for the view thatonstrains phonology to a greater extent torphosyntax. However, Patkowski (19

990) obtained equally strong correlationsween AOA and measures of English morpyntax and phonology (r 5 20.74, 20.76).oreover, Patkowski observed a sharp dec

n performance in both domains at an AOA5 years. His finding for phonology agreed w

he results of two previous studies (Seligrashen, & Ladefoged, 1975; Tahta, Woodowenthal, 1981) but diverged from two othOyama, 1973; Flege et al., 1995a). The AOorphosyntax discontinuity observed by P

owski agreed with the findings of Johnsonewport (1989), who concluded that the acqition of L2 morphosyntax is constrained byompletion of brain maturation at about thef 15 years (but cf. Bialystok & Hakuta, 199This study’s comparison of performance

he phonological and morphosyntactic domaas motivated by the disparate results justiewed. As advocated by Bialystok and Haku1998), we used “broadly based” measure2 proficiency rather than measures that wesigned to test the predictions of a partic

inguistic theory (e.g., predictions regardubjacency or the complex noun phrase ctraint). The learning of English morphosynas assessed using a 144-item grammatic

udgment test that Johnson and Newport (19evised to assess the “most basic aspecnglish sentence structure.” The learning oflish phonology was assessed by having lisrs rate a standard set of English sentepoken by the native Korean participantsverall degree of foreign accent.The grammaticality judgment test us

ere was designed to test nine different mhosyntactic structures or “rules” (Johnsonewport, 1989) but, as discussed below,

tentences testing the nine rule types were

OAofnc

.g.

ee

bes a.,L2orncith

ecs, oingor

thesticoht,

nstedb-theo

ngtheedtionsA

exfo

g-wetheallyltipctimeroueende

ind 23gede45

edipat-ntsthanthe14.6

anups

esecre-arsllhestna-

firstith

hoolhaditedm-56an

w,ith

ntsdu-tes.ici-tes,theand

byrcharch


heterogeneous. Therefore, the effect of Aon two new, functionally specified, setssentences was also examined. The sentecomprising both the “rule based” set (e*The man paints his house yesterday) and the“lexically based” set (e.g., *The farmers werhoping rain) were drawn from several of thoriginal sentence sets. The distinctiontween the two new functional sets reflectdistinction drawn in linguistic theory (e.gPinker, 1991; Pinker & Prince, 1992) andacquisition research (e.g., Beck, 1997). Crect responses to the rule based senterequired the acquisition of simple rules wwidespread application (example: “Add -ed tothe verb root to form the past tense”). Corrresponses to the lexically based sentencethe other hand, probably required learnthat might be characterized as “bottom-up”“data-driven” or else learning based onestablishment of associative or probabilirepresentations (see, e.g., Elman, Bates, Json, Karmiloff-Smith, Parisi, & Plunket1997; Seidenberg, 1997).

The results will be presented in six sectioWe began by carrying out ANOVAs that tesfor differences between AOA-defined sugroups of native Korean participants andnative English control group. The numbersKoreans who received foreign accent ratiand morphosyntax scores that fell withinnative English range were also determinAnalyses were carried out in the second secto determine if the AOA–performance functiowere nonlinear and, if so, where in the AOcontinuum the nonlinearity occurred. In the nsections we compared the scores obtainedthe original nine sets of grammaticality judment test sentences, for sentences thatgrammatical and ungrammatical, and fortwo new functional sets (rule based vs. lexicbased, see above). The purpose of the muregression analyses presented in the next sewas to account for variance in the outcomeasures. Finally, a series of matched subganalyses was undertaken to test for betwgroup differences when variables confoun
with AOA were controlled.
es,

-

-es

tn

n-

.

fs

.n

tr

re

leon

p-

d

METHOD

Participants

The 240 native Korean participants arrivedthe United States between the ages of 1 anyears. Their age at the time of testing ranfrom 17 to 47 years (mean5 26). The 24 nativEnglish participants ranged in age from 20 toyears (mean5 27). All 264 participants passa pure-tone hearing screening before particing. To be included, native Korean participahad to report speaking no language otherEnglish and Korean and to have lived inUnited States for at least 8 years (mean,years).

As summarized in Table 1, the native Koreparticipants were assigned to 1 of 10 subgrobased on AOA. The average AOAs of thsubgroups increased in roughly 2-year inments, from 3 years for group NK3 to 21 yefor group NK21. Half of the participants in a11 groups were female. On average, the higacademic grade completed in Korea by thetive Korean participants was 5.6 years. Theexposure to English for most participants wAOAs greater than 12 years occurred at scin Korea. On the average, the Koreansreceived 10.1 years of education in the UnStates. All but 1 Korean participant had copleted high school in the United States; 1participants held a bachelor’s degree fromAmerican university.

As will be discussed in greater detail beloa number of variables were correlated wAOA. The earlier the native Korean participahad arrived in the United States, the more ecation they had received in the United StaAlso, the younger the native Korean partpants were upon arriving in the United Stathe longer they tended to have lived inUnited States, the more they spoke English,the less they spoke Korean.

General Procedure

The participants were tested individuallycollege-aged Korean/English bilingual reseaassistants in a single 1.5-h session. The rese
was carried out in a quiet room located either on

ol-

anss

ll ao-ncdeeree

theenrearraenlity21rietenovheas

n oconecto

xam-oeuldter

.Hz,an-rate,sh-ven31

-heash-l ofg (at

Hz

an-en-ativeus-

or-

entsi iffer

E

NN 0.3)N 0.3)N 0.6)N 0.6)N 0.6)N 0.6)N 0.7)N 0.6)N 0.7)N 0.7)

0.7)

English,e s; EDUC,y f Korean;E


the campus of the University of Maryland-Clege Park or in a nearby Korean church.

The participants began by completing a lguage background questionnaire that assetheir use of both English and Korean, as wetheir motivation to learn English and retain Krean. The participants later produced sentethat were rated for foreign accent and responto a grammaticality judgment test. Steps wtaken to reduce processing differences betwthe online measure of L2 phonology andoffline measure of L2 morphosyntax. The stences to be rated for foreign accent werepeated following an aural model, and the pticipants were required to listen to an aupresentation of each grammaticality judgmtest sentence before judging its grammatica

Foreign accent.The participants repeatedEnglish sentences that contained a wide vaof English vowels and consonants. The stences were each presented twice in a row,a loudspeaker, in the same order in which tappeared on a written list. A short tone wpresented 700 ms after the first presentatioeach sentence and 3600 ms after its sepresentation. To reduce the likelihood of dirimitations, the participants were required

TAB

Characteristics of the 12 Male and

Age AOA EXP

E 27 (7)K3 23 (3) 3.0 (0.5) 4.5 (1.0)K5 21 (2) 5.0 (0.5) 5.3 (0.7)K7 24 (3) 7.0 (0.5) 7.0 (0.7)K9 24 (3) 9.0 (0.5) 9.0 (0.5)K11 24 (5) 11.0 (0.5) 11.0 (0.7)K13 24 (3) 13.0 (0.5) 12.8 (0.9)K15 27 (5) 15.0 (0.5) 13.2 (0.6)K17 29 (4) 17.0 (0.5) 13.5 (0.8)K19 32 (5) 19.0 (0.5) 13.3 (0.5)K21 34 (5) 21.5 (0.8) 13.7 (1.9)

26 (6) 12.0 (5.9) 9.8 (3.6)

Note.Age, chronological age, in years; AOA, age ofither at school in Korea or upon arrival in the United Stears of education in United States; KORUSE, the averNGUSE, the average of seven similar items pertainin

wait until hearing the tone before repeating eacs

-eds

esd

en

---lt.

y-ery

fd

t

sentence. Just five of the sentences were eined here (Ron set a thick rug in the sun; Jwill feed the pup who sat by you; You shothank Sam for the food; Fit a ring to the watap; It is fun to play chess with a rook), in mostinstances the second of the two repetitions3

The sentences were digitized at 22.05 knormalized for peak intensity, and later rdomly presented three times each in sepacounterbalanced blocks to native Englispeaking listeners. The three male and sefemale listeners, who had a mean age ofyears (range5 23–37), were living in Birmingham, Alabama, at the time of testing. All of tlisteners had been born and raised in the Wington, DC–Baltimore area, however, and althem passed a pure-tone hearing screeninoctave frequencies between 500 and 4000re: 25 dB HL) before participating.

Following practice with 22 sentences spning a wide range of foreign accents, the listers rated sentences spoken by the 24 nEnglish and 240 native Korean participantsing a scale that ranged from “very strong” f

3 A preliminary analysis revealed that the foreign accn the first and second repetitions of sentences did not d

1

emale Participants in Each of 11 Groups

LOR EDUC KORUSE ENGUS

0.0 (2.3) 15.6 (0.6) 2.3 (0.5) 4.6 (6.4 (2.7) 14.8 (1.7) 2.5 (0.6) 4.5 (6.9 (3.2) 15.4 (1.5) 2.6 (0.6) 4.3 (5.0 (3.2) 13.5 (1.7) 2.9 (0.6) 4.1 (

13.5 (4.5) 11.1 (1.7) 3.1 (0.6) 4.0 (11.7 (3.2) 9.2 (1.6) 3.7 (0.5) 3.6 (12.5 (5.4) 7.8 (2.0) 3.5 (0.7) 3.6 (12.5 (4.2) 5.8 (2.0) 3.7 (0.5) 3.4 (13.7 (5.1) 4.8 (1.8) 3.7 (0.7) 3.5 (13.5 (4.5) 2.9 (2.8) 3.8 (0.5) 3.5 (

14.6 (4.6) 10.1 (4.8) 3.2 (0.8) 3.9 (

val in the United States; EXP, age of first exposure to; LOR, length of residence in the United States, in yearof nine 5-point rating scale items pertaining to the use o

o the use of English.SDs are in parentheses.

LE

12 F

2111

arriatesageg t

hignificantly.

ersss

o

aced

ectverf therser

i-p asa dgm seo iffee e 2g thn thes ewp

c tra-t or-d the1 sub-j cesu halfw lim-i ag , orb icalp n ana rs ofe ands andu oc-c

malen on-s latea ts.T ere-c hich

ay.ay.

is marked


eign accent (1) to “no accent” (9). The listenwere told to use the whole scale and to gueunsure. A mean foreign accent rating based150 judgments (5 sentences3 10 listeners3 3replicate judgments) was calculated for eparticipant. Average ratings were examinhere for two reasons. First, a similar AOA effwas evident for each of the 10 listeners (aaged over the five sentences) and for each ofive sentences (averaged over the 10 listenSecond, very high Intraclass Correlations wobtained for the five sentences,R 5 .986, andthe 10 listeners,R 5 .978,P , .001.

Morphosyntax.The native Korean particants’ knowledge of English morphosyntax wssessed using a 144-item grammaticality juent test. This test was composed of ninef sentences, each intended to evaluate a dnt morphosyntactic structure or rule. Tablives an example of each set, as well asumber of pairs in each set. Most (128) ofentences were drawn from Johnson and N

TAB

Number and Examples o

N Sentence type

8 Past tense

8 Plural

8 Third-person singular

8 Determiners

8 Pronouns

6 Particle movement

14 Subcategorization

4 Lexically specified subject/object raising

4 Y/N questions

4 Wh questions

Note. N,the number of sentence pairs. In the list of exby an asterisk.

ort’s (1989) test. Three sets of sentences tht

ifn

h

-e).

e

-tsr-

e

-

aused very few errors in previous adminisions of the test (present progressive, worder, and auxiliary) were eliminated. Eight of6 new sentences tested lexically specified

ect/object raising. Half of the 144 sentensed here were grammatical. The otherere ungrammatical sentences created by e

nating a required morpheme or word fromrammatical sentence, by changing a wordy moving some word(s) to an ungrammatosition. The 144 sentences were printed onswer sheet. An equal number of exemplaach sentence type appeared on the firstecond halves of the test. The grammaticalngrammatical versions of each pair alwaysurred on separate halves of the test.The test sentences were recorded by a

ative speaker of English, who spoke at a ctant moderate rate and took care to articull sounds, including word-final consonanhe sentences were digitized and then rorded in the same quasi-random order in w

2

e Nine GJT Sentence Types

Examples

A policeman gave Alan a ticket for speeding yesterd*A policeman gived Alan a ticket for speeding yesterdTodd has many coats in his closet.*Todd has many coat in his closet.

Every Friday our neighbor washes her car.*Every Friday our neighbor wash her car.

The boy is helping the man build a house.*The boy is helping the man build house.

Susan is making some cookies for us.*Susan is making some cookies for we.

Kevin called up Nancy for a date.*Kevin called Nancy for a date up.

The little boys laughed at the clown.*The little boys laughed the clown.

Larry believed himself to be brave.*Larry believed that himself to be brave.Should Timothy have gone to the party?

*Should have Timothy gone to the party?Where did she put the book?

*Why did she put the book?

ples, the ungrammatical member of each sentence pair

LE

f th

am

athey appeared on the answer sheet. The digi-

e vparheme

jusdroiderea

notonntcere

anthenhoed

as,

venvedivedn-

neian/a)

ob-ing

rly-a-gserswereilar

om

litypar-de-sinefor

ob-rtic-

the

r 24datehm


tized sentences were presented a single tima loudspeaker at a comfortable level. Theticipants were told to wait until they heard tentire sentence before checking “Yes” (gramatical) or “No” (not grammatical) next to thwritten version of the sentence they hadheard. The test was unspeeded (see Chau1983). The interval between sentences provtime to respond. (If a participant ever appeato need more time, the tape recorder wpaused.) The terms “grammatical” and “grammatical” were not defined, but these ccepts were made clear to the participathrough examples. Also, four practice senten(two grammatical, two ungrammatical) wepresented before the test began.

RESULTS

Effects of AOA

Figure 1 shows that the later the Koreparticipants had arrived in the United States,stronger were their foreign accents. The stences of the 24 native English controls, wwere assigned an AOA of “0” years, receiv

FIG. 1. The mean foreign accent ratings obtained fonative English and 240 native Korean participants. Thefor the 240 Koreans have been fit to the Gompertz-Makdistribution (solid line).

higher ratings than those of all but a few early-

ia-

-

tn,dds

-ss

e-

arriving Koreans. The effect of group whighly significant in a one-way ANOVAF(10,253)5 104.4,P , .01. A series oft testsrevealed that all 10 native Korean groups, ethose composed of individuals who had arriin the United States as young children, recesignificantly lower ratings than the native Eglish controls (BonferroniP , .01).

The AOA effect seen here is similar to oobserved for certain listeners who rated ItalEnglish bilinguals in the Flege et al. (1995study. However, the foreign accent ratingstained for certain other native English-speaklisteners did not differentiate groups of eaarriving Italian/English bilinguals from the ntive English controls. Differences in the ratinobtained from the 10 native English listenwho rated sentences in the present studynot explored, however, because a highly simoverall pattern of ratings was obtained freach of them (see above).

Figure 2 presents the overall grammaticajudgment test scores. As the native Koreanticipants’ AOAs increased, their scorescreased systematically. The scores were arctransformed (Kirk, 1968), because variance

FIG. 2. The grammaticality judgment test scorestained for 24 native English and 240 native Korean paipants. The data for the 240 Koreans have been fit to

a

Gompertz-Makehm distribution (solid line).

atectns

tsot

m

ityreeor

ea39–7allish

dentsve-e24

cu-eanthan.dgeaoricared

hef dgm 10n tlyf forp ithA of2 ishc g-m mac ag som for-e oret hp oo,m

c ho-n

T

wast inA er( m-p r-m thati forb ent rtz-M ela ith,1 ns’r tima-t izet wni tionw enA uldb rn-i of ac

-t %)v id afii theA ce-d setso oft t itd llyd or1 ntsh het tlym derf ,

, to

ird-on-

fit


the native Korean groups increased systemcally as AOA increased. The significant effof group obtained in an analysis of the traformed scores,F(10,253)5 57.4,P , .01, wasfollowed up by a series oft tests. These tesrevealed that groups NK7–NK21, but ngroups NK3–NK5, differed significantly frothe native English controls (BonferroniP ,.01). The AOA effect on the grammaticaljudgment test scores obtained here agclosely with the results of Johnson and Newp(1989). These authors found that native Korand Chinese participants with AOAs of 8–years, but not participants with AOAs of 3years, received significantly lower grammaticity judgment test scores than did native Engcontrols.

The following procedure was adopted totermine how many native Korean participamight be said to have performed in a “natilike” fashion in the two linguistic domains. Thmean and standard deviation (SD) of thenative English participants’ ratings were callated. We then determined which native Korparticipants obtained a foreign accent ratingfell within two SDs of the native English meaThe same was done for the grammaticality jument test scores. The number of native Korparticipants who met the “two SD” criterion fthe foreign accent ratings and the grammatity judgment test scores (18 vs. 76) diffesignificantly,x2(2) 5 28.2,P , .01.

In summary, the overall effect of AOA on toreign accent ratings and grammaticality juent test scores was similar. However, allative Korean groups differed significan

rom the native English comparison grouphonology, whereas just the subgroups wOAs of 7 to 23 years (not those with AOAsto 6 years) differed from the native Engl

omparison group for the grammaticality judent test scores. This suggests that AOA

onstrain the learning of L2 phonology toreater extent than L2 morphosyntax. Alore of the Korean participants receivedign accent ratings than morphosyntax sc

hat fell within two SDs of the native Englisarticipants’ mean values. This finding, t
ight be taken as support for the view that ag
i-

-

dtn

-

-

t

-n

l-

-

y

,

s

onstraints are stronger in the domain of pology than morphosyntax.

he Relation between AOA and L2Performance

One aim of the analyses presented hereo determine if a discontinuity existedOA-L2 performance functions. First-ord

linear) and third-order functions were coared in order to determine if AOA–perfoance functions were linear. (We reasoned

f significantly more variance was accountedy a third-order than a first-order function, th

hat function was nonlinear.) The Gompeakehm distribution, which is used to modspects of the aging process (Draper & Sm981, pp. 511–513), was also fit to the Koreaatings and scores using least-squares esion. This provided a visual means to organhe individual participants’ mean values shon scattergrams. The other aim of this secas to determine if the correlation betweOA and measures of L2 performance woe significant for participants who began lea

ng English after, as well as before, the endritical period.Foreign accent ratings.A third-order func

ion accounted for significantly more (1.9ariance in the foreign accent ratings than drst-order function,F(2,236) 5 8.8, P , .01,ndicating the presence of a nonlinearity inOA–foreign accent relation. The same proure was then applied to AOA-defined subf participants in order to identify the locus

he nonlinearity. The results suggested thaid not occur near the end of the traditionaefined critical period (i.e., at an AOA of 125 years). In the analysis of the 193 participaaving AOAs ranging from 7 to 23 years, t

hird-order function accounted for significanore (2.5%) variance than did the first-or

unction, F(2,189)5 6.56,P , .01. Howeverwhen the AOA range was restricted furtheran AOA range of 7 to 18 years (n 5 144participants), the difference between the thorder and first-order functions (1.0%) was nsignificant,F(2,140)5 1.63,P . .10.

The Gompertz-Makehm distribution was
eto the mean foreign accent ratings obtained for

. AhaarseitforacAsfacrspos

Ainnt

ndfo

ticito

ing

han

hiss foe eno or1 reo r the ilst th-e tot tiob fop foret geH nsw 40K thifi odh

f re( taxs part n,F -r id-e

(f 7

t -o re( on,F tst ests 12y vi-o andP oft uchs ski( icald yn-t test,w ersr fors

llya 2,w tz-M hefi ina ofa a

w-s anondr ofces.

onres

ing

lsoter

-

2 veK of1 n-s thefi byB

slowf inedb ight


the 240 native Korean participants in Fig. 1visual inspection of this function suggests tthe Koreans with AOAs of about 1–5 yeobtained ratings that were similar to, albslightly lower than, the mean rating obtainedthe native English controls. The apparent lof difference between the Koreans with AOof 1 to 5 years may have been due to thethat, for many of these participants, the fiextensive exposure to English occurred uentry to school (see Table 1). There waroughly linear decrease in ratings in the AOrange of 5–15 years, followed by a slowingthe rate at which the strength of foreign acceincreased.

The simple correlations between AOA athe foreign accent ratings were computedtwo subsets of the 240 native Korean parpants. A significant correlation was foundexist for the native Korean participants havAOAs less than 12 years,r 5 20.62,P , .01,and also for those having an AOA greater t12 years,r 5 20.50,P , .01.

In summary, the findings presented in tection suggest that the Koreans’ degree ofign accent did not increase sharply near thef a critical period, that is, at an AOA of 125 years. The relation between AOA and degf foreign accent appeared to be linear neand of the supposed critical period, which fa

o provide support for the critical period hyposis. The critical period hypothesis also led

he expectation that there would be a correlaetween AOA and degree of foreign accentarticipants who began learning English be

he age of 12 years, but not after that aowever, the AOA–foreign accent correlatioere significant for both subsets of the 2orean participants examined here. Thus,nding also failed to support the critical periypothesis.Grammaticality judgment test.A third-order

unction accounted for significantly mo1.2%) variance in the overall morphosyncores obtained for the 240 native Koreanicipants than did a first-order functio(2,236)5 3.16, P , .05. When just the Ko

eans with AOAs of 7 to 23 years were cons
red (n 5 193), the difference in variancec
t

k

ttna

s

r-

r-d

ee

nr

.

s

-

1.1%) was nonsignificant,F(2,189) 5 2.20,P . .05. However, when those with AOAs oo 18 years were considered (n 5 144), a thirdrder function accounted for significantly mo5.0%) variance than did a first-order functi(2,140)5 5.91,P , .01. This finding sugges

hat the relation between AOA and the tcores was nonlinear at an AOA of aboutears; it agrees with findings reported preusly by both Johnson and Newport (1989)atkowski (1980, 1990). However, the size

he nonlinearity observed here was mmaller than the one obtained by Patkow1980, 1990), probably due to methodologifferences. Knowledge of English morphos

ax was assessed here using a 144-itemhereas Patkowski had two English teach

ate transcripts of conversational speechyntactic accuracy.The nonlinearity just reported is not visua

pparent in the fit function shown in Fig.hich was obtained using the Gomperakehm distribution. A visual inspection of t

t function indicates that the scores declinedroughly linear fashion between AOAs

bout 6 to 15 years.4 There is no evidence ofnonlinearity at AOAs of 12 or 15 years. Hoever, as can be seen in Fig. 3, there waincrease in the number of participants beyan AOA of 12 years who gave a large numbeincorrect responses to ungrammatical senten

Finally, there was a significant correlatibetween AOA and the Morphosyntax test scofor the 120 native Korean participants havAOAs less than 12 years,r 5 2.52,P , .01. Asmall, but still significant correlation was aobtained for the Koreans having AOAs greathan 12 years,r 5 20.27,P , .01. (The correlations werer 5 20.71, P , .01, andr 5

0.23, P , .05, when the sample of natiorean participants was divided at an AOA5 years.) This finding differs from that of Johon and Newport (1989), but it agrees withndings obtained in more recent studiesirdsong (1992; Birdsong & Molis, 1998).

4 The rate at which the scores decreased seemed toor AOA greater than 15 years. The census data examy Bialystok and Hakuta, 1998, suggest that scores m
ontinue to decline slowly over the entire life span.

asestheA

duith

da

foist

pre

enthelishand

iv-un-dg-

ato-3,er-

ticalm-

m -t withn hy,1 nedh ,r

ansen-(2)-

A iveK hen cals justs als -t icald n-g ang ntf ps(

m-m cer-t theK al”( mel ts’k g-m ort de-

ines on


In summary, when the discontinuity test wapplied to the grammaticality judgment tscores, it supported the critical period hypotsis. A nonlinearity was detected in the AOregion of 12–15 years. This was apparentlyto an increase in the number of participants wAOAs greater than 12 years who acceptelarge number of ungrammatical sentencesgrammatical (see Figs. 2 and 3). (The basisthis increase is uncertain but, as will be dcussed later, it may have been relatedchanges in language use.) However, the

FIG. 3. The mean percentage of correct scores obtafor (a) 72 grammatical and (b) 72 ungrammatical itema grammaticality judgment test.

postcorrelation test did not support the criticat

-

e

asr

-o/

period hypothesis. The correlation betweAOA and the scores was significant forparticipants who had begun learning Engboth before the age of 12 years (or 15 years)after the age of 12 years (or 15 years).

Sentence Types

The native English controls’ and early-arring Koreans’ score for grammatical andgrammatical items on the grammaticality jument test did not differ (because they wereceiling for both), but later arriving native Krean participants did differ. As shown in Fig.the native English controls obtained a high pcentage of correct scores for both gramma(M 5 98.3%, range, 86–100%) and ungra

atical (M 5 97.4%, range5 88–100%) senences. However, as in previous researchonnative speakers of English (e.g., Murp997), the native Korean participants obtaiigher scores for the grammatical (M 5 94.3%ange5 69–100%) than ungrammatical (M 5

74.0%, range5 21–100%) sentences. The mescores for grammatical and ungrammaticaltences were submitted to an (11) Group byGrammaticality ANOVA, which yielded a significant interaction,F(10,253)5 22.7,P , .01.

series of t tests revealed that eight natorean groups differed significantly from tative English controls for the ungrammatientences (groups NK7–NK21), whereasix (NK11–NK21) did so for the grammaticentences (BonferroniP , .01). More imporantly, the grammatical versus ungrammatifference was nonsignificant for the native Elish controls and the first two native Koreroups (NK3, NK5), whereas it was significa

or the remaining eight native Korean grouNK7–NK21) (BonferroniP , .01).

The basis for the difference between graatical and ungrammatical sentences is un

ain. It may have arisen from a bias byorean participants to respond “grammatic

White, 1989). It might also mean that soater arriving native Korean participannowledge of English morphosyntax was “fraentary or fluctuating in its accessibility”

hat their grammars of English were less “

d

lerminate” than those of the native English con-

ssintslishop

re-hes

icesm

rallarvelly2

nedininalleAsrie

atthecetheupen

raisi

Ko-fer-

asof

entun-nting. Asncesticalese

ightde-

ofthe

,”notoys”the

ofFor

sr

s

ro groupsw


trols (Johnson et al., 1996). Still another pobility is that certain native Korean participaincorrectly judged some aspects of Engmorphosyntax that were tested here to betional (Johnson et al., 1996). Additionalsearch will be needed to choose among tinterpretations.

As mentioned earlier, the sentences whcomprised the grammaticality judgment twere intended to test knowledge of nine gramatical structures or “rules” (past tense, pluWh and Y/N questions, third-person singudeterminers, pronouns, particle movement,bal subcategorization frames, and lexicaspecified subject/object raising; see TableThe effect of AOA on the mean scores obtaifor the nine sets of sentences was examinedseries of one-way ANOVAs. As summarizedTable 3, the effect of group was significant innine instances (P , .01), but the strength of thsimple correlations between the Korean’ AOand scores for the nine sets of sentence vaconsiderably. A series oft tests revealed thnine native Korean subgroups differed fromnative English controls for the plural senten(P , .01). A difference was noted betweennative English controls and eight Korean grofor the determiner and subcategorization stences, seven groups for the subject/objecting and third-person singular sentences,

TAB

Correlations between Age of ArrivaObtained for Nine

Sentence type r(238) NK

Past tense 20.49 90.Plural 20.64 76.Third-person singular 20.44 89.Determiners 20.74 78.Pronominalization 20.51 93.Particle movement 20.63 84.Subcategorization 20.71 81.Subject/object raising 20.63 77.Questions (Wh and Y/N) 20.71 85.

Note.NK and NE, the mean scores obtained for nane-way ANOVAs testing the effect of group (11 levelshich differed significantly from the native English com

groups for the past tense, question, and partic

-

-

e

ht-,,r-

).

a

d

s

s-

s-x

movement sentences; and just the last fourrean groups for the pronoun sentences (Bonroni P , .01).

The finding just presented might be takenevidence that age constrains the learningvarious aspects of L2 morphosyntax in differways. However, such a conclusion may bewarranted given that the sentences represethe various “rule” types were heterogeneousdiscussed by Kellerman (1995), some sentemay not have tested the intended grammastructure or rule. Consider, for example, thtwo ungrammatical determiner sentences:

(1) *Tom is reading book in the bathtub.

(2) *A boys are going to the zoo this Saturday.

The ungrammaticality of both sentences mbe attributed to the presence/absence of theterminer “a.” However, the ungrammaticalitythe first sentence might also be attributed tolack of the plural marker “-s” on “bookwhereas this alternative interpretation ispossible for the second sentence because “bmust have a plural marker to agree withverb.

An item analysis revealed a great dealheterogeneity within the nine sentence sets.example, there were far more errors for *Thegirl’s swimsuit is full of sandsand *Two mouseran into the house this morningthan for othe

3

nd the Percentage of Correct Responseses of GJT Sentences

NE F(10,253) Difference

98.2 12.8 K13–K2197.4 30.0 K5–K2199.5 11.4 K11–K2198.2 49.9 K7–K2199.7 15.0 K15–K2196.9 21.7 K13–K2197.2 40.0 K7–K2193.2 19.9 K11–K2198.2 35.8 K13–K21

Korean and native English participants. TheF values are foll were significant at the .001 level. Differences, the NKrison group (BonferroniP , .01).

LE

l aTyp

292327452

tive); apa

leplural sentences (e.g., *The farmer bought two

ayh ” ia ing“ al( eref eda ra n-t stn eee kedk ir-r nk

R ces

twf est cec romm arf larp t iv ionw mf redu db in-c 1)T ulap surf edc rsoo pat lam e as

icas App ent rticm ighl aryf ncet tsE as-s rbso ar-t

edeeal

o rbs( ul-l n-t cingt r-mT ouldn

ts’m andl theG c-t pla-t n ag ges ces,h OAo to ac

cor-r andl oup.T hanu vs.7 resf mati-c n-t

eresFA c-t

oups1–anat-Asfor

anarsrsed


pig at the market). The first two sentences mave been especially difficult because “sandmass noun that is not pluralized by add

-s,” and “mouse” has an irregular plur“mice”). To take another example, there war more errors for *Yesterday the baby throwcat into the bathtuband *A bat flewed into outtic last night than for other past tense se

ences (e.g., *Sandy fill a jar with cookies laight). The first two sentences may have bspecially difficult because participants lacnowledge of which English verbs have anegular past tense, not because they didnow how to form the regular past tense.

ule Based versus Lexically Based Senten

Given the heterogeneity just discussed,unctionally defined sets of sentences wereablished for further analysis. The sentenomprising these two new sets were drawn fultiple sentence sets. In a series of prelimin

actor analyses, certain third-person singuast tense, and determiner sentences thaolved regular rules of verb and noun inflectere found to have high loadings on the sa

actor(s). The 22 grammatical and 22 paingrammatical sentences that were deemeest reflect this functional similarity wereluded in a “rule based” set (see Appendixhese sentences tested knowledge of regroductive, and generalizable rules of the

ace morphology of English. They all involvase or number assignment on nouns or per tense markers on verbs (e.g., regular

ense on plural formation, third-person singuorphology on present tense verbs, or cas

ignment on personal pronouns).The 22 grammatical and 22 ungrammat

entences in the “lexically based” set (seeendix 2) were also drawn from several s

ence sets (subcategorization, question, paovement). They, too, tended to have h

oadings on the same factor(s) in preliminactor analyses. The lexically based senteested irregular and ungeneralizable aspecnglish morphosyntax involving the properignment of particles or prepositions with ver knowledge of idiosyncratic features of p

icular English verbs. For example, some sen

s

n

ot

o-

s

y,n-

e

to

.r,-

nstrs-

l--le

sof

tences tested which preposition should preca nominal complement (e.g., *The farmers werhoping rain), the use of a particle in phrasverbs (e.g., *The little boys laughed the clown),

r the placement of particles in phrasal vee.g., *The man climbed the ladder up carefy). All ungrammatical lexically based seences could be made grammatical by replahe verb (for example, changing “lets” to “peits” in *The man lets his son to watch TV).he ungrammatical rule based sentences cot be corrected in this way, however.In Fig. 4, the 240 native Korean participanean scores for ungrammatical rule based

exically based sentences have been fit toompertz-Makehm distribution. The fit fun

ion for the rule based sentences showed aeau up to an AOA of about 5 years and theradual decline to the end of the AOA ranampled here. For the lexically based sentenowever, the initial plateau extended to an Af about 8 years and then decreased rapidlyhance level at an AOA of about 16 years.Figure 5 shows the mean percentage of

ect responses obtained for the rule basedexically based sentences as a function of grhe scores were higher for grammatical tngrammatical sentences (means, 95%9%). However, as AOA increased the sco

or both sentence types decreased systeally, especially for the lexically based seences.

The scores obtained for each subject wubmitted to a mixed-design (11) Group3 (2)unctional Type 3 (2) GrammaticalityNOVA, which yielded a three-way intera

ion, F(10,253)5 2.34,P , .05. A series ofttests revealed that the native Korean subgrwith AOAs greater than 11 years (NK1NK21) received significantly lower scores ththe native English controls for the ungrammical sentences, but only those with AOgreater than 13 years (NK13–NK21) did sothe grammatical sentences (BonferroniP ,.01). More importantly, just the five Koresubgroups with AOAs greater than 13 ye(NK13–NK21) received significantly lowescores for the lexically based than the rule ba
-sentences (BonferroniP , .01).

llyvinifi-Asntlyulethi-x t

. Toces,ith

addhengre-

useess-

allyarn-up”thestic97).

on,vedagehisltyadtaxsed

herefor

tz-oreally(se

lity


In summary, the rule based and lexicabased scores obtained for participants haAOAs less than 12 years did not differ signcantly, whereas participants having AOgreater than 12 years obtained significalower scores for the lexically based than rbased sentences. This finding suggestsAOA might influence the learnability of lexcally based aspects of English morphosynta

FIG. 4. Curves obtained by fitting the GomperMakehm distribution to the percentage of correct scobtained for the ungrammatical rule based and lexicbased items from the grammaticality judgment testtext).

FIG. 5. The mean percentage of correct scorejudgment test sentences that were grammatical
standard error.
g

at

o

a greater extent than rule based aspectsrespond correctly to the rule based sententhe Koreans had to acquire a simple rule wwidespread application in English, such as “-ed to the verb root to form past tense.” Trelatively good performance by late-arriviparticipants on rule based items may haveflected deductive (top-down) learning, theof rule based mental representations in procing, or both.

On the other hand, success on the lexicbased items may have required a kind of leing that one might characterize as “bottom-or “data-driven” or else learning based onestablishment of associative or probabilirepresentations (see, e.g., Elman et al., 19As will be discussed further in the next sectithe later the native Korean participants arriin the United States, the less English-languinput they were likely to have received. Tmay have contributed to the greater difficuthat late-arriving native Korean participants hin learning aspects of English morphosyntested by the lexically based than rule basentences.

Factor Analyses

The purpose of the analyses presentedwas to identify factors that might account

s

e

btained for rule based and lexically based grammatica”) or ungrammatical (“U”). The error bars enclose61

s o(“G

tivag

itteri-

ene rigntesanse

-alesfo

irefo

lowg,”on

thelishshad-iveos,(asundh-tedts,

wastheof

esrall

um-orecentthan

1

videos

lish

nce ofnd

ng newh

th

ability

ow E


variance in the outcome measures. The naKorean participants’ responses to 39 langubackground questionnaire items were submto a principal components analysis with vamax rotation.5 The resulting factors were thregressed onto the outcome measures. Thsults for the two primary variables (the foreaccent ratings and overall morphosyntaxscores) will be presented in the first section,the results for the rule based and lexicon basentences in the following section.

Principal components analysis.As summarized in Table 4, the principal components anysis identified 11 factors with eigenvalugreater than 1.0. These factors accounted69.4% of the variance in the questionnaitems. Just four of these factors accountedvariance in the outcome measures (see beFactor 1 was designated “Age of L2 Learninbecause the items with the highest loadings

5 A total of 82 missing responses comprising less

TAB

Factors Identified in a Principal Com

Factor

F1: Age of L2 learning Age of arrival in thecomfortably (.892)

F2: English media input Frequency of watcin English (.848);

F3: Instrumental motivation-1 Will get respect forespect for good pfor success at wo

F4: Judged importance of Korean Judged importanccorrect Korean grexpressions (.623

F5: Judged importance of English Judged importancEnglish words andpronunciation (.68

F6: Languages used at work Use of English atF7: Home use of Korean Frequency of use

parents (.501)F8: Sound processing ability Ability to imitate fo

to remember howF9: Integrative motivation Try to have as ma

is pronounced (.7F10: Length of residence Length of residencF11: Instrumental motivation-2 Judged importanc

Note.The loadings for each questionnaire item are i

% of the data were replaced with group mean values.

eed

e-

tdd

-

r

r).

it

were AOA and the participants’ estimates ofage at which they could first speak Eng“comfortably.” Factor 2 was named “EngliMedia Input,” because the items with high loings on it pertained to how much the natKorean participants watched movies, videand TV and listened to the radio in Englishopposed to Korean). Factor 8 was called “SoProcessing Ability.” The items having the higest loadings on it pertained to self-estimaability to imitate foreign accents and dialec“musical ability,” and ability to “remember hoEnglish words are pronounced.” Factor 10 wdesignated “Length of Residence,” becauseonly item with a high loading on it was yearsresidence in the United States.

Primary variables.The regression analysexamining the foreign accent ratings and ovegrammaticality judgment test scores are smarized in Tables 5 and 6, respectively. Mvariance was accounted for in the foreign ac

4

nents Analysis of 39 Questionnaire Items

Questionnaire items

nited States (.912); estimated age of speaking Englishars of education in the United States (20.856)

g TV in English (.852); frequency of watching movies/uency of listening to radio in English (.691)rrect English grammar and vocabulary (.861); will getunciation of English (.851); judged importance of Engchool (.694)f correct Korean pronunciation (.811); judged importaar (.788); enjoyment of learning new Korean words a

f correct English grammar (.819); enjoyment of learnipressions (.701); judged importance of correct Englis

rk (.841); use of Korean at work (20.777)orean at home (.674); frequency of use of Korean wi

gn accents and dialects (.783); musical ability (.611);glish words are pronounced (.532)merican friends as possible (.741); pay attention to h

n the United States (.827)f English for getting a job (.699)

arentheses.

LE

po

U; yehinfreqr coron

rk/se o

amm)e oex

4)woof K

reiEn

ny A24)e i

e o

n p

ratings than in the morphosyntax test scores

thor-gethe

tsf nci oR nav r( di-t diF ac-t acc ncei thap noe

10)o thr ish

b em thefi seea 1).T r ofp tici-p thov

c-c tor,a for-e dg-m portf ntd ingb antf In-d rtialc wedt nd

robability,own.

bility, the


(68% vs. 49%). (This was expected becausesimple correlation between AOA and the feign accent ratings was significantly stronthan the correlation between AOA andgrammaticality judgment test scores,X(1) 530.9, P , .001.) Other principal componenactors accounted for substantially less varian both outcome variables. Factor 10 (Lengthesidence) accounted for 4 and 5% additioariance in the ratings and scores. FactoEnglish Media Input) accounted for 2% adional variance in both outcome variables, asactor 8 (Sound Processing Ability). The f

ors that pertained to motivation (F9, F11)ounted for less than 3% of additional varia

n the two outcome measures. The factorsertained to language use (F6, F7) werentered into either model.The effect of length of residence (Factor

n the foreign accent ratings agrees withesults of Flege et al. (1995a) for Italian/Engl

TAB

Regression Analysis Examining the Nati

Step Variable R

1 F1: Age of L2 learning 0.2 F10: Length of residence 03 F8: Sound processing ability 04 F2: English media input 05 F9: Integrative motivation 06 F11: Instrumental motivation 0

Note.The principal components factors that were regrthe probability of a significant increase in variance. On

TAB

Regression Analysis Examining the Native Ko

Step Variable R

1 F1: Age of L2 learning 02 F10: Length of residence 03 F2: English media input 04 F8: Sound processing ability 05 F9: Integrative motivation 06 F11: Instrumental motivation 0

Note.The principal components factors regressed o
probability of a significant increase in variance. Only facto
e

r

efl

2

d

-

tt

e

ilinguals. However, a similar finding for thorphosyntax test scores diverged from

ndings of Johnson and Newport (1989;lso Patkowski, 1980, but cf. Cummins, 198his may have been due to the larger numbearticipants examined here (240 vs. 46 parants) or to a differing distribution of the lengf residence variable (M 5 15, range5 8–30s. M 5 10 years, range5 3–26).In summary, Factor 1 (Age of Learning) a

ounted for more variance than any other facnd it accounted for more variance in theign accent ratings than grammaticality juent test scores. This might be taken as sup

or the view that AOA is the most importaeterminant of overall success in L2 learnut, at the same time, AOA is more import

or phonology than morphosyntax learning.eed, such conclusions are supported by paorrelation analyses. These analyses shohat the simple correlation between AOA a

5

orean Participants’ Degree of Foreign Accent

Change F value Probability

0.677 747.5 .0013 0.036 40.0 .0017 0.024 26.7 .001

1 0.024 26.3 .0014 0.014 15.0 .0016 0.011 12.6 .001

ed onto the dependent variables are listed in Table 4. Pactors accounting for at least 1.0% of variance are sh

6

n Participants’ Grammaticality Judgment Test Scores

Change F value Probability

0.494 313.7 .0017 0.053 33.6 .0010 0.023 14.5 .0013 0.023 14.5 .001

6 0.023 14.4 .0015 0.010 6.0 .015

the dependent variables are shown in Table 4. Proba

LE

ve K

2

677.71.73

.76

.77.78

essly f

LE

rea

2

.494.54.57.59

.61.62

nto
rs accounting for at least 1.0% of variance are shown.

yntho

tyim-woo

ariult,ortesuso-h aorghencen,OAobuhiex

weble

esserente0,

thente2,

hyac-tha

acthewitefosears

de

ntds,ithalsohi-0)ldcticis

ages85;

eentiveg torilybe

tolyex-.g.,dyar-eanthofrs

rre-

stoupeex-lesthe

usein

ti-nine

atith

onse


both the foreign accent ratings and morphostax test scores remained significant wheneffects of variation in length of residence, useEnglish, and use of Korean were removed (P ,.01).

However, the problem of multi-collinearimay have led to an overestimation of theportance of AOA. As mentioned earlier, the tquestionnaire items with the highest loadingsFactor 1 pertained to age. However, many vables were correlated with AOA. As a resother items also had high loadings on Fact(i.e., years of education in the United Sta20.856; use of Korean with a spouse, .786;of Korean with close friends, .729; use of Krean at social gatherings, .737; use of Englissocial gatherings,20.712; and age, .700). Fexample, this pattern of intercorrelations mihave been responsible for the surprising absof a language use effect on the foreign acratings (see Flege et al., 1995a). That issomewhat stronger correlation between Aand the foreign accent ratings may havescured a weaker relation between languagevariables and the foreign accent ratings. Tand similar issues will be addressed in the nsection, where matched subgroup analysesperformed to assess the effect of two variacorrelated with AOA.

Secondary variables.Regression analyswere also carried out to examine the rule baand lexicon based morphosyntax test scoThe analysis of the rule based scores accoufor 45% of variance (F1, 32%; F8, 4%; F13%; F11, 2%; F6, 1%; F5, 1%; F9, 1%), andanalysis of the lexicon based scores accoufor 61% of variance (F1, 48%; F10, 5%; F4%; F9, 3%; F4, 1%; F6, 1%). It is notewortthat Factor 8 (Sound Processing Ability)counted for more variance in the rule basedlexically based scores (4% vs. 1%).

In posthoc analyses, we discovered that Ftor 8 accounted for 10% of the variance inrule based scores obtained for KoreansAOAs of 14–23 years (n 5 96), but no variancfor their lexically based scores. It accountedno variance in the rule based or lexically bascores for Koreans having AOAs of 6–13 ye
(n 5 96). This finding may help explain why
-ef

n-

1,e

t

tcet

a

-sestres

ds.d

d

n

-

h

rd

late-arriving native Korean participants maerrors on sentences such as *Last night the oldlady die in her sleep.Phonologically nonsaliemorphological markers at the end of worwhich pose problems for certain children wlanguage disorders (Leonard, 1982), maybe difficult for L2 learners (see, e.g., Moczuki-Sudo, Susuki, Matsuno, & Kiritani, 199whose auditory skills are limited. If so, it wousuggest that phonological and morphosyntalearning interact in L2 learning in a way thatanalogous to the interaction seen in early stof L1 acquisition (Camarata & Gandour, 19Camarata & Schwartz, 1985).

Matched Subgroup Analyses

The results obtained in this study have bplotted as a function of AOA because the naKorean participants were selected accordinAOA. However, this does not mean necessathat the “age” effects presented so far canattributed exclusively, or even primarily,AOA. As mentioned earlier, AOA is typicalconfounded with other variables in studiesamining large immigrant populations (see, eBachi, 1956; Bahrick et al., 1994). This stuwas no exception, for there was multi-collineity among variables associated with the Korparticipants’ AOAs. AOA was correlated wichronological age,r 5 .68; self-estimated useEnglish and Korean,r 5 20.56 and .66; yeaof residence in the United States,r 5 20.42;and years of education in the United States,r 520.92. Further, these variables were all colated significantly with one another (P , .01).

Given the pattern of intercorrelations jumentioned, one aim of the matched subgranalyses presented in this section was to ramine the effect of AOA when other variabwere controlled. Another aim was to assessinfluence of two other variables (languageand education) independently of variationAOA.

Variables. The Koreans’ L1 use was esmated by averaging their responses toquestions pertaining to the use of Koreanhome, at work or school, in social settings, wclose friends, and with a spouse. The resp
to each item was a number ranging from 1

siveisher

t glisu Tab

oti-v ass usp d tm 6s seT ang atet anu anw ere“ sw elye upsh wnr n/E eirp do-m &N al-y

ithen-

du-for-as ao-

ow-d a

nr ofhad

ed-om-testud-ionnythe

otal(orbylishhadthe

reiveelytter

herusedn).

rre-l lish,r sr h ofrl alityjK sec OAa2

tlyon

atepor


(“never”) to 5 (“all the time”). English use waestimated by calculating the mean ratings gto seven similar questions pertaining to EnglThe averages for the two sets of ratings winversely correlated,r 5 20.76,P , .01. Bothhe average Korean use ratings and the Ense ratings were correlated with AOA (seele 1).An examination of language use was m

ated by several observations. First, there wtriking correspondence between languageatterns and the age that is widely believeark the end of the critical period. Figure

hows the ratio of English use to Korean uhe ratios obtained for the native Koreroups having AOAs of 3–11 years were gre

han 1.0, indicating more English than Korese. One might speculate that the participho used English more than Korean w

dominant” in English. However, the ratioere close to 1.0, indicating approximatqual use of English and Korean, for groaving AOAs of 13–21. Second, it was shoecently that the frequency with which Italianglish bilinguals spoke Italian affected therformance in English in the phonologicalains (Flege, 1998a,b; Flege, Friedaozawa, 1997). Finally, partial correlation an

FIG. 6. Relative language use. The dashed line indica equal use of Korean and English according to self reThe error bars enclose61 standard error.

ses suggested that variations in the Koreanb

n.e

h-

ae

o

.

r

ts

use of their two languages were correlated wthe outcome measures of this study indepdently of AOA.6

The other variable examined here was ecation. One might reasonably expect permance in some aspects of English to varyfunction of how much education the native Krean participants had received in English. Hever, Johnson and Newport (1989) observenonsignificant correlation (r 5 .25) betweeMorphosyntax test scores and the numbeyears of English classes their participantstaken before arriving in the United States.

We focused here on how many years ofucation the native Korean participants had cpleted in English-speaking United Staschools, rather than on how long they had sied English in Korea. There was a correlatbetween the Koreans’ AOA and how mayears of education they had received inUnited States,r 5 .92,df 5 238,P , .01. Theactual variable examined here was called “tyears of education in the United States”“U.S. education,” for short). It was computedadding the number of years of special Engclasses the native Korean participantstaken, if any, to years of formal education inUnited States.7

Hypotheses.The first hypothesis tested hewas that when AOA was controlled, the natKorean participants who used English relativoften (and Korean seldom) would have a bepronunciation of English and receive higmorphosyntax test scores than those whoEnglish relatively seldom (and Korean ofte

6 The foreign accent ratings reported earlier were coated with the Koreans’ self-reported use of both Eng5 .61, and Korean,r 5 .70, P , .01. These correlation

emained significant when variations in AOA and lengtesidence were removed,r 5 .30, 20.29, P , .01. Simi-arly, the correlations between the overall grammaticudgment test scores and both English use,r 5 .54, andorean use,r 5 20.60, P , .01, were significant. Theorrelations remained significant when variations in And length of residence were partialled out,r 5 .20 and0.22,P , .01.7 The number of years of special English classes,M 5 1.6

years, range5 0–4 years, was not correlated significanwith AOA, r 5 .07,P . .10, which reduced the correlati

st.

s’etween total years of education and AOA,r 5 .90.

waascifiuldhasetaxonainitherf 2o

ofhe

up.S

hvaetlyan

0,of3.

hetly

d

at-ally

upssesca-ar-s ofd toU.S.om-

upsOA

inon-of

upsed-sig-.6vs.n-

.S.t in-ea-pho-licits’or-

theotowl-

or-ci-

re,or-

uleglish

lysesvs.S.).ari-

ar-t theU

9*

u-12.nth


The second hypothesis was that when AOAcontrolled, U.S. education would affect at lesome of the outcome measures. More specally, it was hypothesized that education wohave a greater effect on morphosyntax tphonology, and a greater effect for rule bathan lexically based aspects of morphosynThe final hypothesis was that “age” effectsphonology, but not morphosyntax would remsignificant after variables confounded wAOA were controlled. These hypotheses wtested by establishing matched subgroups onative Korean participants, each consistingparticipants drawn from the original sample240 without regard for the scores or ratings tobtained.

Results.The first set of matched subgroanalyses compared groups that differed in Ueducation (15.1 vs. 8.9 years,P , .01) but werematched for AOA (mean5 12.3 years for botsubgroups). The matching process reducedation for variables in addition to AOA. Thmatched subgroups did not differ significanin terms of their self-reported use of Kore(3.2 vs. 3.1,P . .10) or English (3.9 vs. 4.P . .10). They did differ, however, in lengthresidence in the United States (17.0 vs. 1years,P , .05). As summarized in Table 7, t“Much U.S. Education” group had significanhigher rule based morphosyntax scores thanthe “Little U.S. Education” group (P , .01).

TABLE 7

Comparisons of Two Groups of 20 Native Korean Picipants Each Who Differed in Amount of Education innited States but Were Matched for AOA

Outcome variable

Moreyears of

education

Feweryears of

education F(1,38)

Foreign accent 5.0 (1.8) 5.0 (2.0) 0.00Overall GJT score 87% (8) 80% (14) 3.29Lexicon based GJT 84% (13) 79% (18) 1.3Rule based GJT 93% (4) 85% (12) 9.15

Note.The two groups differed significantly in U.S. edcation (8.9 vs. 15.1 years) but had the same AOA (years for both groups). Standard deviations are in pareses. *P , .01.

However, the two matched groups did not re

st-

nd.

e0f

y

.

ri-

6

id

ceive significantly different foreign accent rings, overall morphosyntax scores, or lexicbased morphosyntax scores (P . .10).

The comparison of the two matched growas accompanied by a set of “control” analycomparing groups that differed in U.S. edution but were not matched for AOA. The pticipants in the two unmatched control group20 participants each were randomly selectehave the same mean number of years ofeducation as did the two matched groups cpared earlier (viz. 15.1 vs. 8.9 years,P , .01).Given that the participants in the control growere randomly selected, and given that Awas correlated with years of U.S. educationthe original sample of 240 participants, the ctrol groups differed significantly in termsAOA (7.2 vs. 16.2 years,P , .01). Like thematched groups, the unmatched control grodiffered significantly in terms of their rule basscores (93% vs. 85%;P , .01). However, unlike the matched groups, they also receivednificantly different foreign accent ratings (6vs. 3.1), overall morphosyntax scores (94%78%,P , .01), and lexically based morphosytax scores (92% vs. 73%,P , .01).

These results indicate that the amount of Ueducation had a significant and independenfluence on just one of the four outcome msures considered here: the rule based morsyntax scores. It is not certain whether expor implicit instruction affected the Koreanlearning of rule based aspects of English mphosyntax (see Winitz, 1996). Whateverkind of instruction (or input) it was, it does nseem to have augmented the Koreans’ knedge of ungeneralizable aspects of English mphosyntax or to have improved their pronunation of English. This finding might, therefobe taken as support for the view that an imptant difference exists in the learning of rbased versus lexically based aspects of Enmorphosyntax (Pinker, 1991).

The second set of matched subgroup anacompared groups that differed in AOA (9.716.6 years,P , .01) but were matched for U.education (mean5 10.5 years for both groupsThe matching process reduced variation in v

3e-

-ables other than just U.S. education. The two

tly3.

, o

s lsr rat-i didt i-l el axs edP

dedd rad eap at 9.v .Se duc n-m inU

ffentir%)vs

ore

nif-oureig

ogyheastwoofer-ingrtz,

, &t-s,ntor,

indif-

vs.

sria-etly4.5

.6inela-d ashsel-o

not

ar-t ageU

*

*

e-p OA( *

ar-t forA

*

9

.2y in thU n pr


matched subgroups did not differ significanin years of residence in the United States (1vs. 14.7 years), Korean use (3.2 vs. 3.4)English use (3.9 vs. 3.6, allP values. .10). Asummarized in Table 8, the “earlier” bilinguaeceived significantly higher foreign accentngs (i.e., pronounced English better) thanhe “later” bilinguals. However, the earlier binguals did not differ significantly from thater bilinguals for any of the morphosyntcores (overall, lexically based or rule bas. .10).The accompanying control analyses yiel

ifferent results. These analyses comparedomly selected subgroups of 20 native Korarticipants each who had the same AOAs

he matched groups compared earlier (viz.s. 16.6 years) but were not matched for Uducation. Given that the amount of U.S. eation was correlated with AOA, the two uatched control groups differed significantly.S. education (14.4 vs. 8.0 years,P , .01).

The two unmatched groups were found to disignificantly not only in terms of foreign acce(5.9 vs. 3.4,P , .01) but also in terms of theoverall morphosyntax scores (92% vs. 79lexically based morphosyntax scores (92%76%), and rule based morphosyntax sc(94% vs. 85%) (P , .01).

These results indicate that AOA had a sigicant, independent effect on just one of the foutcome measures considered here: the fo

TABLE 8

Comparisons of Two Groups of 20 Native Korean Picipants Each Who Differed in AOA but Were Matchedmount of Education in the United States

Outcome variableEarlierAOA

LaterAOA F(1,38)


Note. The two groups differed in AOA (9.7 vs. 16ears) but had the same number of years of educationnited States (10.8 years). Standard deviations are i

entheses. *P , .01.

accent ratings. From this, one might conclude.

9r

;

n-ns7.-

r

,.s

rn

that age constrains the learning of phonolbut not the learning of L2 morphosyntax. Tdifference across linguistic domains that wobserved here can be interpreted in at leastdifferent ways. It might derive from the usedifferent neural substrates for phonological vsus lexical-semantic and syntactic learnand/or processing (Warrington, 1975; SchwaMarin, & Saffran, 1979; Berndt, CaramazzaZurif, 1983; Mateer, 1983; Liberman & Matingly, 1985; Keller, 1987; Gracco & Abb1987), or it might arise from the use of differe“modules” (Forster, 1979; Garrett, 1980; Fod1983).

The native Korean participants comparedthe final set of matched subgroup analysesfered significantly in their use of Korean (4.12.1,P , .01) and English (3.3 vs. 4.5,P , .01)but were matched for AOA (mean5 11.4 yearfor both). The matching process reduced vation in variables in addition to AOA. Thmatched subgroups did not differ significanin years of residence in the United States (1vs. 15.1 years,P . .10) or U.S. education (12vs. 12.5 years,P . .10). As summarizedTable 9, the Koreans who used English rtively often and Korean relatively seldom hasignificantly better pronunciation of Englithan did those who used English relativelydom (and Korean often) (P , .05). They alshad higher lexically based scores (P , .05).However, the two matched subgroups did

TABLE 9

Comparisons of Two Groups of 20 Native Korean Picipants Each Who Differed in Self-Reported Languse but Were Matched for AOA

Outcome variable

LittleL1/much

L2

MuchL1/little

L2 F(1,38)


Note. The two groups differed significantly in self-rorted Korean use (4.1 vs. 2.1) but were matched for A11.4 years). Standard deviations are in parentheses.P ,

ea-

05.

r-ore

namlyamhe.1)ofupin

fer-

3%b-

g eno %P re(

patt dep eni ynt sem lang tioo itht nint shb nay lisb o-m re-v ).H ane fok ax.

Kor om hon ts om Thp pec nyf ated hs thl

f d one verm alitys xi-c bet ypeo p-r ofl L2l oren g int

ed.F off thes ati-c e ofa ed.S ici-p ub-g olsi do-m

ass thel entt tax( 90;H n-c ex-p thed dif-f g int oesm pe-r onef 2),b thep s oft theg o-l ofc (Fe-l orc in


differ significantly in terms of their overall mophosyntax or rule based morphosyntax sc(P . .10).

In the accompanying control analyses, 20tive Korean participants each were randoselected to create groups that had the smean Korean use ratings as did the matcsubgroups compared earlier (viz. 4.1 vs. 2Given that AOA was correlated with amountKorean use, the two unmatched control groin the control analyses differed significantlyAOA (16.2 vs. 7.0 years;P , .01). The twounmatched groups received significantly difent foreign accent ratings (6.5 vs. 3.4,P , .01)and lexically based morphosyntax scores (9vs. 73%,P , .01). Unlike the matched su

roups, they also received significantly differverall morphosyntax scores (93% vs. 76, .10) and rule based morphosyntax sco

94% vs. 83%,P , .01).These results indicate that the Koreans’

ern of language use exerted a significant, inendent effect on their degree of foreign acc

n English and their lexicon based morphosax scores, but not on their overall or rule baorphosyntax scores. The conclusion thatuage use affected the Koreans’ pronunciaf English independently of AOA agrees w

he results of a regression analysis examihe pronunciation of English by Italian/Engliilinguals (Flege et al., 1995a), as well as ases examining other aspects of Italian/Engilinguals’ performance in the phonological dain (Flege, MacKay, & Meador, under

iew; Mackay, Meador, & Flege, under reviewowever, this is apparently the first time thatffect of language use has been reportednowledge of any aspect of L2 morphosyntThe fact that language use affected the

eans’ learning of lexically based aspectsorphosyntax suggests that the learning of pology and irregular, ungeneralizable aspecorphosyntax have something in common.hysical realization of consonants and, esially, vowels varies as a function of maactors (e.g., neighboring context, speaking regree of stress). The perception of speechaped by what one hears. As a result,
ong-term memory representations developes
s

-

ed.

s

t,s

--t

-d-n

g

l-h

r

-f-f

e-

,ise

or language-specific speech sounds depenxperience with a wide range of tokens oany years of exposure. Thus, the common

hared by the learning of phonology and leally based aspects of morphosyntax mayhat both require a bottom-up, data-driven tf learning with associative or probabilistic reesentations (Elman et al., 1997). This typeearning implies that the more input anearner receives from native speakers, the mative-like their representations or processin

he L2 will be.

GENERAL DISCUSSION

Two outcomes of this study were expectirst, the native Korean participants’ strength

oreign accent in English grew stronger andcores they received on a 144-item grammality judgment test decreased as their agrrival (AOA) in the United States increasecond, more individual native Korean partants, and more AOA-defined Korean sroups differed from the native English contr

n the phonological than morphosyntacticain.The second set of findings might be taken

upport for the view that age constrainsearning of L2 phonology to a greater exthan it does the learning of L2 morphosyne.g., Braine, 1971; Bever, 1981; Long, 19urford, 1991; MacWhinney, 1992). It is uertain, however, which of several possiblelanations provides the best account forifference. Bever (1981) proposed that the

erence arises because phonological learninhe L1 reaches completion sooner than dorphosyntactic learning (so that a critical

iod for phonology ends sooner than doesor morphosyntax). According to Cook (199ilinguals have more difficulty separatinghonological than morphosyntactic systemheir two languages. Others have citedreater overall perceived similarity of phon

ogical structures in the L1 and L2 thanorresponding morphosyntactic structuresix, 1980; Ioup, 1984; MacWhinney, 1987)laimed that the role of the motor cortex

dpeech articulation fundamentally distinguishes

or-

asdererelu

h-atdu

hatel

ticaaso-ntato

eaer,lityxisheossis

e ace

L2wher,theifi-arseseis-of

thaasith56wont

an-eyh oe o

-ear-,lled.onre,

on-73,fol-on.le-

thattic

smf theitedandtheageand

welinethevel-in-thege,tedayn-ndsininnot

edcalenithtro-asthisol-inganening


phonological learning from the learning of mphosyntax (Zatorre, 1989).

The primary aim of this study, however, wto provide a better understanding of the unlying basis for the AOA effects observed hand in previous studies. We did this by evaating the validity of the critical period hypotesis for L2 acquisition, and with it the claim thage-related declines in L2 performance areto a diminished ability to learn language tresults from brain maturation (e.g., Scov1988).

Three methods were used to test the criperiod hypothesis. The discontinuity test wapplied to functions relating the 240 native Krean participants’ AOA to their foreign acceratings and to scores obtained on the grammcality judgment test. There was no evidencea nonlinearity for the foreign accent ratings nthe end of the putative critical period. Howevthere was a nonlinearity for the grammaticajudgment test scores, which supported the etence of a critical period for morphosyntax. Tsecond method applied here was the pre/pcorrelation test. The critical period hypotheleads to the expectation that there will bcorrelation between AOA and L2 performanfor individuals who began learning theirbefore the age of 12 years, but not for thosebegan learning their L2 later in life. Howevthe AOA–foreign accent correlations andAOA–morphosyntax correlations were signcant both for Koreans with AOAs of 2–12 yeand those with AOAs of 13–23 years. Thfindings, therefore, failed to support the extence of a critical period for the learningeither phonology or morphosyntax.

A matched subgroup analysis confirmedthe AOA effect on foreign accent ratings wnot due to factors that were confounded wAOA as in previous research (e.g., Bachi, 19Bahrick et al., 1994; see Flege, 1998a). Tmatched subgroups of 20 Korean participaeach were established by selecting participwho differed in AOA but did not differ significantly in terms of how much education thhad received in the United States, their lengtresidence in the United States, or their us
English and Korean. The later arriving sub-
-

-

e

,

l

i-fr

-

t-

o

t

;

sts

ff

group (mean AOA5 16.6 years) had significantly stronger foreign accents than did thelier arriving subgroup (mean AOA5 9.7 years)even though the other variables were contro

Recall that the AOA–foreign accent functiwas essentially linear. One might, therefohypothesize that L2 phonology learning is cstrained by a sensitive period (Oyama, 191979; Bornstein, 1989), perhaps one thatlows from, or is shaped by, brain maturatiBased on their review of a large body of revant literature, Bates et al. (1992) notedthere is a slow, monotonic decline in synapdensity and overall levels of brain metabolibetween the age of 4 years and the end osecond decade of life. These authors posthat a connection exists between the rateextent of human neural development and“slow decrease in capacity for second-langulearning” that one sees through childhoodadolescence (1992, p. 102).

There is an alternative interpretation thatprefer, however. It is that the age-related decin L2 pronunciation accuracy derives fromfact that, as AOA increases, the state of deopment of the L1 phonetic system alsocreases, thereby changing the way in whichL1 and L2 phonological systems interact (Fle1995, 1998a,b). More specifically, age-relachanges in the pronunciation of an L2 mderive from differences in how, or if, L2 learers perceptually relate L2 sounds to the soumaking up the L1 phonetic inventory. This,turn, may lead to age-related differenceswhether new phonetic categories are or areestablished for sounds in the L2.

The results summarized earlier providmixed support for the existence of a critiperiod in the domain of morphosyntax. Givthis, as well as the ambiguity that exists wrespect to the discontinuity test (see the Induction), the crucial test for morphosyntax wthe matched subgroup test. The results oftest differed from the one obtained for phonogy. The scores obtained for the earlier arrivsubgroup were not significantly higher ththose of the later arriving subgroup, evthough, in a control analysis, subgroups hav
the same mean AOAs (9.7 vs. 16.6 years) that

bleg-ntsot, a

Thantorsisanthingtu-

hensonedenenrtindogontinas

s ooresedid

uplexrenomntndb-

Enedn oelyoreerub.Sared

ps’ifi-

edhenhedthateir

isi-ro-case

andra-ob-n isa-

ner-ll asad-L2.lishflu-al

ar-Ko-t thee toar-theaywillpstaxy-re-the2,

gertestAsof

Are-dednitive


were not matched on the confounded variawere found to differ significantly. This sugested that the native Korean participaknowledge of English morphosyntax did ndecrease as the result of an increase in AOAreported by Johnson and Newport (1989).apparent AOA effect observed by JohnsonNewport may have been the result of facconfounded with AOA. If this conclusioncorrect, then the AOA effect observed herein previous studies cannot be ascribed topassing of a critical period for language learnthat arises inevitably from normal brain maration.

Additional analyses provided insight into tfactors that might actually have been respoble for the previously reported AOA effectsL2 morphosyntax. Two functionally definsubsets of grammaticality judgment test stences were examined. The “rule based” stences were characterized as testing the paipants’ knowledge of regular, productive, ageneralizable rules of the surface morpholof English. The “lexically based” sentences,the other hand, were characterized as tesknowledge of irregular and ungeneralizablepects of English morphosyntax. The two setsentences patterned quite differently. The scfor the lexically based sentences decreamore dramatically as AOA increased thanthe scores for the rule based sentences.

Even more importantly, matched subgroanalyses showed that the rule based andcally based scores were influenced by diffevariables. One matched subgroup analysis cpared subgroups of native Korean participawho differed in self-reported use of English aKorean but were matched for AOA. The sugroup consisting of participants who usedglish often obtained higher lexically basscores (and also had a better pronunciatioEnglish) than those who used English relativseldom. The two subgroups’ rule based scdid not differ significantly, however. Anothmatched subgroup analysis compared sgroups of Koreans who differed in years of Ueducation but were matched for AOA. The pticipants with more U.S. education receiv
higher rule based scores than those with les
s

’

sed

de

i-

--c-

y

g-fsd

i-t-

s

-

f

s

-.-

U.S. education. However, the two subgroulexically based scores did not differ signcantly.

The differing effect of AOA on the rule basand lexically based morphosyntax scores, wtaken together with the results of the matcsubgroup analyses, bear on a conclusionBates and Goodman (1998) drew from thextensive review of evidence from L1 acqution, language breakdown, and real-time pcessing. These authors concluded that thefor a modular distinction between grammarthe lexicon has been “overstated” in the liteture. While this may be so, the evidencetained here suggests that such a distinctiooperative in L2 acquisition. The results summrized above suggest that knowledge of ungealizable aspects of L2 morphosyntax (as wethe ability to pronounce an L2) improves grually as a function of experience using theKnowledge of generalizable aspects of Engmorphosyntax, on the other hand, may be inenced more importantly by amount of formeducation.

Of course, the more the native Korean pticipants used English, the less they usedrean. One might, therefore, hypothesize thalanguage use effect observed here was duvariations in how much the native Korean pticipants continued to speak Korean, not tofrequency with which they used English. It mbe that the more the L1 is used, the more itinfluence the kind of knowledge that develofor lexically based aspects of L2 morphosyn(as well as L2 pronunciation). Still another hpothesis that might be examined in futuresearch is that a relatively infrequent use ofL2 is an effect of poor performance in the Lnot its cause.

In summary, foreign accents grew stronand scores on the grammaticality judgmentdecreased as the Korean participants’ AOincreased. However, the underlying basesthese effects differed importantly. The AOeffect on phonology but not morphosyntaxmained significant when variables confounwith AOA were controlled. The AOA effect ophonology may have been due to a sens
speriod arising from brain maturation or, more

2temor

ionrre

ve(or

”

es

ht.

.e.dne

las

t.

ed

es

bed

r th

ully.ster-

s last

He-

B S.ilin-l

B y ofion,-

B ageega-y-)

B fre-NS.

B aged.),

.

B the.),de

B ondA.r-s.

B of-

B euisi-

B in-ford.),for

.B lan-


likely, from changes in how the L1 and Lphonological systems interact as the L1 sysdevelops. The apparent AOA effects on mphosyntax seem to have arisen from variatin education and language use that were colated with AOA and so were unlikely to haarisen from a maturationally defined criticalsensitive) period.

APPENDIX 1

The 44 Grammatical and Ungrammatical “Rule-BasedGJT Sentences. The Grammatical Version of EachSentence Is Specified by the Word(s) in Parenthes

The girl cooks (cooked) dinner for her family last nigLast night the old lady die (died) in her sleep.Last night Mary walks (walked) to the store.The man paints (painted) his house yesterday.Sandy fill (filled) a jar with cookies last night.Every Friday our neighbor wash (washes) her car.John’s dog always wait (waits) for him at the cornerEvery day Terri talk (talks) to her Mom on the phonMrs. Sampson clean (cleans) her house every We

day.Many house (houses) were destroyed by the flood

week.Three boy (boys) played on the swings in the park.Todd has many coat (coats) in his closet.A (The) boys are going to the zoo this Saturday.Mary opens a (the) windows in her room every nighHim (He) is fixing the tire on Jamie’s bicycle.Them (They) worked on the project all night.A snake bit she (her) on the leg.Susan is making some cookies for we (us).We ate the whole pizza by themselves (ourselves).The girl cut himself (herself) on a piece of glass.They took theirs (their) children to the theater.Tom drove himselves (his) sister to the concert.

APPENDIX 2

The 44 Grammatical and Ungrammatical Lexically BasGJT Sentences. The Grammatical Version of EachSentence Is Specified by the Word(s) in Parenthes

The farmers were hoping (hoping for) rain.Why (What) did the company send?The policeman was talking (talking to) a woman.Larry went the (omit “the”) home after the party.Jenny set the book that was (omit “that was”) on theThe man lets his son to (omit “to”) watch TV.The girls enjoy to feed (feeding) the ducks.Kevin called Nancy for a date up (up for a date).The man looked the new cars yesterday over (ove

new cars yesterday).I hope you to go to (will go to) the store now.
Mrs. Johnson went to (the) library yesterday.
-s-

s-

t

.

e

The man climbed the ladder up (up the ladder) carefThe horse jumped the fence over (over the fence) ye

day.He came (to) my house at six o’clock.She let the cat very quickly in (in very quickly).The man allows his son watch TV (to watch TV).The little boys laughed (laughed at) the clown.The girls want feeding (want to feed) the ducks.George says (says his prayers) much too softly.The boy put the bowl (add: “in the kitchen”).Why (Where) did she put the book?Nancy put the dishes last night away (away the dishe

night).

REFERENCES

Bachi, R. (1956). Statistical analysis of the revival ofbrew in Israel.Scripta Hierosolymitan,3, 179–247.

ahrick, H., Hall, L., Goggin, J., Bahrick, L., & Berger,(1994). Fifty years of language maintenance in bgual Hispanic immigrants.Journal of ExperimentaPsychology: General,123,264–283.

ates, E., & Goodman, J. (1998). On the inseparabilitgrammar and the lexicon: Evidence from acquisitaphasia, and real-time processing.Language and Cognitive Processes,12, 507–586.

ates, E., Thal, D., & Janowsky, J. (1992). Early langudevelopment and its neural correlates. In S. J. Slowitz & I. Rapin (vol. Eds.),Handbook of neuropschology: Vol. 7, Child neuropsychology(pp. 69–100Amsterdam: Elsevier.

eck, M. L. (1997). Regular verbs, past tense andquency: Tracking down a potential source of NS/Ncompetence.Second Language Research,13, 93–115

erndt, R., Caramazza, A., & Zurif, E. (1983). Langufunctions: Syntax and semantics. In Segalowitz (ELanguage functions and brain organization(pp. 5–28)New York: Academic Press.

ever, T. (1981). Normal acquisition processes explaincritical period for language learning. In K. Diller (EdIndividual differences in language learning aptitu(pp. 176–198). Rowley, MA: Newbury House.

ialystok, E. (1994). Towards an explanation of seclanguage acquisition. In G. Brown, K. Malmkjaer,Pollitt, & J. Williams (Eds.),Language and undestanding(pp. 115–138). Oxford: Oxford Univ. Pres

ialystok, E. (1997). The structure of age: In searchbarriers to second language acquisition.Second Language Research,13, 116–137.

ialystok, E., & Hakuta, K. (1994).In other words: Thscience and psychology of second language acqtion. New York: Basic Books.

ialystok, E., & Hakuta, K. (1998). Confounded age: Lguistic and cognitive factors in age differencessecond language acquisition. In D. P. Birdsong (ENew perspectives on the critical period hypothesissecond language acquisition.Hillsdale, NJ: Erlbaum

irdsong, D. P. (1992). Ultimate attainment in second
guage acquisition.Language,68, 706–755.

hequi-

nty ofript

, Ecia-ge

entions

B :.

B and

C thered

C jechiph

C udg

C

C wee

C ond

C ondia-il-ess

D n

E 5).itiv-

E A.,-en

E varreng. I

F and,

F latedp-

er

F ro-

F eechd.),ew,

F : The-d.),oret--

F eticT.ingels.

F uagen

age

F t ofper-

F t ofn of

F ivelish

F a).nt inof

F hepro-a-

F on

F re oflker:

G B.


Birdsong, D. P. (Ed.) (1998).New perspectives on tcritical period hypothesis for second language acsition. Hillsdale, NJ: Erlbaum.

Birdsong, D. P., & Molis, M. (1998). Limits on attainmein second language acquisition: A replication studJohnson & Newport (1989). [unpublished manuscUniversity of Texas]

Bongaerts, T., Summeren, C., Planken, B., & Schils(1997). Age and ultimate attainment in the pronuntion of a foreign language.Studies in Second LanguaAcquisition,19, 447–466.

Bornstein, M. (1989). Sensitive periods in developmStructural characteristics and causal interpretatPsychological Bulletin,105,179–197.

ott, S. M. (1993).Speech intelligibility and bilingualismThe effects of age of acquisition.[unpublished Ph.Ddissertation, Univ. of Illinois, Urbana-Champaign]

raine, M. (1971). The acquisition of language in infantchild. In C. Reed (Ed.),The learning of language(pp.7–95). New York: Academic Press.

amarata, S., & Gandour, J. (1985). Rule invention inacquisition of morphology by a language-impaichild. Journal of Speech and Hearing Disorders,50,40–45.

amarata, S., & Schwartz, R. (1985). Production of obwords and action words: Evidence for a relationsbetween phonology and semantics.Journal of Speecand Hearing Research,28, 323–330.

haudron, C. (1983). Research on metalinguistic jments: A review of theory, methods, and results.Lan-guage Learning,33, 343–377.

ook, V. (1992). Evidence for multicompetence.LanguageLearning,42, 557–591.

oppieters, R. (1987). Competence differences betnative and fluent non-native speakers.Language,63,544–573.

ummins, J. (1981). Age on arrival and immigrant seclanguage learning in Canada: A reassessment.AppliedLinguistics,2, 131–149.

ummins, J. (1991). Interdependence of first- and seclanguage proficiency in bilingual children. In E. Blystok (Ed.), Language processing in bilingual chdren(pp. 70–89). Cambridge: Cambridge Univ. Pr

raper, N. R., & Smith, H. (1981).Applied regressioanalysis.New York: Wiley.

mmorey, K., Bellugi, U., Friederici, A., & Horn, P. (199Effects of age of acquisition on grammatical sensity: Evidence from on-line and off-line tasks.AppliedPsycholinguistics,16, 1–24.

lman, J., Bates, E., Johnson, M., Karmiloff-Smith,Parisi, D., & Plunkett, K. (1997).Rethinking innateness: A connectionist perspective on developmCambridge, MA: MIT Press.

ngstrand, L. (1982). Age and length of residence asables related to the adjustment of migrant childwith special emphasis to second language learnin
S. Krashen, R. Scarcella, & M. Long (Eds.),Child- G
,

.

:.

t

-

n

-

.

t.

i-,n

adult differences in second language acquisition(pp.15–123). Rowley, MA: Newbury House.

athman, A. (1975). The relationship between agesecond language learning ability.Language Learning25, 245–253.

elix, S. (1980). Interference, interlanguage, and reissues. In S. Felix (Ed.),Second language develoment: Trends and issues.Tubigen, Germany: GuntNarr.

lege, J. E. (1987). A critical period for learning to pnounce foreign languages?Applied Linguistics,8,162–177.

lege, J. E. (1988). The production and perception of spsounds in a foreign languages. In H. Winitz (EHuman communication and its disorders: A revi1988 (pp. 224–401). Norwood, NJ: Ablex.

lege, J. E. (1995). Second-language speech learningory, findings, and problems. In W. Strange (ESpeech perception and linguistic experience: Theical and methodological issues(pp. 229–273). Timonium, MD: York Press.

lege, J. E. (1998a). The role of subject and phonvariables. In K. D. Gruber, D. Higgins, K. Olsen, &Wysocki (Eds.),Papers from the 34th annual meetof the Chicago Linguistic Society: Vol. II. The panChicago: Chicago Linguistic Soc.

lege, J. E. (1998b). Age of learning and second-langspeech. In D. P. Birdsong (Ed.)New perspectives othe critical period hypothesis for second languacquisition.Hillsdale, NJ: Erlbaum.

lege, J. E., Bohn, O-S., & Jang, S. (1997). The effecexperience on nonnative subjects’ production andception of English vowels.Journal of Phonetics,25,437–470.

lege, J. E., Frieda, E. M., & Nozawa, T. (1997). Amounnative-language (L1) use affects the pronunciatioan L2.Journal of Phonetics,25, 169–186.

lege, J. E., MacKay, I. R. A., & Meador, D. (1998). NatItalian speakers’ production and perception of Engvowels. [article under review]

lege, J. E., Munro, M. J., & MacKay, I. R. A. (1995Factors affecting degree of perceived foreign accea second language.Journal of the Acoustical SocietyAmerica,97, 3125–3134.

lege, J. E., Munro, M. J., & MacKay, I. R. A. (1995b). Teffect of age of second language learning on theduction of English consonants.Speech Communiction, 16, 1–26.

odor, J. (1983).The modularity of mind: An essayfaculty psychology.Cambridge, MA: MIT Press.

orster, K. (1979). Levels of processing and the structuthe language processor. In W. E. Cooper & R. Wa(Eds.),Sentence processing(pp. 27–85). New YorkWiley.

arrett, M. (1980). Levels of sentence production. InButterworth (Ed.),Language production(pp. 177–221). New York: Academic Press.

racco, V., & Abbs, J. (1987). Programming and execution

ne

.

for

tiveL2

ind.

lantorympe

s inturaon

J s onbja

J D.an-

ro-s.),

an.-

ageci-lityof

the

ateuis

ith-

of

deion,

del.

M ec-

:

M thewitz

M ect-lan-

M an-theron.

M i,lish

e

M ofn of

m-ch,

age

eention.

O n ofh.D.

O d in

O n of, R.n,

P isi--

P age:

P .or-

s ofldstheand

ci-udy.


processes of speech movement control: Potentialral correlates. In E. Keller & M. Gopnick (Eds.),Motorand sensory processes of language(pp. 163–202)Hillsdale, NJ: Erlbaum.

Hurford, J. (1991). The evolution of the critical periodlanguage acquisition.Cognition,40, 159–201.

Hyltenstam, K. (1992). Non-native features of near-naspeakers: On the ultimate attainment of childhoodlearners. In R. Harris (Ed.),Cognitive processingbilinguals (pp. 351–368). Amsterdam: North-Hollan

Ioup, G. (1984). Is there a structural foreign accent?Lan-guage Learning,34, 1–17.

Johnson, J. (1992). Critical period effects in secondguage acquisition: The effect of written versus audimaterials on the assessment of grammatical cotence.Language Learning,42, 217–248.

Johnson, J., & Newport, E. (1989). Critical period effectsecond language learning: The influence of mational state on the acquisition of English as a seclanguage.Cognitive Psychology,21, 60–99.

ohnson, J., & Newport, E. (1991). Critical period effectuniversal properties of language: The status of sucency in the acquisition of a second language.Cogni-tion, 39, 215–258.

ohnson, J., Shenkman, K., Newport, E., & Medin,(1996). Indeterminacy in the grammar of adult lguage learners.Journal of Memory and Language,35,335–352.

Keller, E. (1987). The cortical representation of motor pcesses of speech. In E. Keller & M. Gopnick (EdMotor and sensory processes of language(pp. 125–162). Hillsdale, NJ: Erlbaum.

Kellerman, E. (1995). Age before beauty: JohnsonNewport revisited. In L. Eubank, L. Selinker, & MSharwood Smith (Eds.),The current state of interlanguage(pp. 219–231). Amsterdam: Benjamins.

Kim, E. J. (1996).The sensitive period for second-languacquisition: An experimental study of a lexical-desion task with semantic priming and a grammaticajudgment test.[unpublished doctoral thesis, Univ.Illinois, Urbana-Champaign]

Kirk, R. (1968). Experimental design: Procedures forbehavioral sciences.Belmont, CA: Brooks/Cole.

Krashen, S., Long, M., & Scarcella, R. (1979). Age, rand eventual attainment in second language acqtion. TESOL Quarterly,13, 573–582.

Leonard, L. (1982). Phonological deficits in children wdevelopmental language impairment.Brain and Language,16, 73–86.

Liberman, A., & Mattingly, I. (1985). The motor theoryspeech perception revised.Cognition,21, 1–36.

Long, M. (1990). Maturational constraints on languagevelopment.Studies in Second Language Acquisit12, 251–285.

MacWhinney, B. (1987). Applying the competition moto bilingualism.Applied Psycholinguistics,8,315–327

acWhinney, B. (1992). Transfer and competition in s
ond language learning. In R. Harris (Ed.),Cognitive
u-

-

-

-d

-

d

,i-

-

processing in bilinguals(pp. 371–390). AmsterdamNorth-Holland.

ateer, C. (1983). Motor and perceptual functions ofleft hemisphere and their interaction. In S. Segalo(Ed.),Language functions and brain organization(pp.145–170). New York: Academic Press.

acKay, I., Meador, D., & Flege, J. (1999). Factors affing the identification of consonants in a secondguage. [article under review]

ayberry, R., & Eichen, E. (1991). The long-lasting advtage of learning sign language in childhood: Anolook at the critical period for language acquisitiJournal of Memory and Language,30, 486–512.

ochizuki-Sudo, M., Susuki, H., Matsuno, K., & KiritanS. (1990). The perception of articles in spoken Engby Japanese college students.Annual Bulletin of thResearch Institute of Logopedics and Phoniatrics,24,163–170.

unro, M., Flege, J., & MacKay, I. (1996). The effectage of second language learning on the productioEnglish vowels.Applied Psycholinguistics,17, 313–334.

Murphy, V. A. (1997). The effect of modality on a gramaticality judgment task.Second Language Resear13, 34–65.

Newport, E. (1990). Maturational constraints on langulearning.Cognitive Science,14, 147–172.

Olson, L., & Samuels, S. (1973). The relationship betwage and accuracy of foreign language pronunciaJournal of Psycholinguistic Research,66, 263–267.

yama, S. (1973). A sensitive period for the acquisitioa second language. [unpublished Harvard Univ. Pdissertation]

yama, S. (1979). The concept of the sensitive periodevelopmental studies.Merrill-Palmer Quarterly,25,83–102.

yama, S. (1982). A sensitive period for the acquisitioa nonnative phonological system. In S. KrashenScarcella, & M. Long (Eds.),Child-adult differences isecond language acquisition(pp. 20–38). RowleyMA: Newbury House.

atkowski, M. (1980). The sensitive period for the acqution of syntax in a second language.Language Learning, 30, 449–472.

atkowski, M. (1990). Age and accent in a second languA reply to James Emil Flege.Applied Linguistics,11,73–89.

inker, S. (1991). Rules of language.Science,253,530–535Pinker, S., & Prince, A. (1992). Regular and irregular m

phology and the psychological status of the rulegrammar. In L. Sutton, C. Johnson, & R. Shie(Eds.),Proceedings of the 17th annual meeting ofBerkeley Linguistics Society: General sessionparasession on the grammar of event structure.Berke-ley, CA: Berkeley Linguistics Soc.

Schwartz, M. F., Marin, O., & Saffran, E. (1979). Dissoations of language function in dementia: A case st
Brain and Language,7, 277–306.

tich.

us

S uraag

ageon

n,

or.

agh,

riod-ec-

sen

.

lR.

n,

ignthe

d

an-y-

ge

ion

lan-ons.


Scovel, T. (1988).A time to speak: A psycholinguisinquiry into the critical period for human speecCambridge, MA: Newbury House.

Seidenberg, M. S. (1997). Language acquisition andLearning and applying probabilistic constraints.Sci-ence,275,1599–1604.

eliger, H., Krashen, S., & Ladefoged, P. (1975). Mattional constraints in the acquisition of second languaccent.Language Sciences,36, 20–22.

Shim, R. (1994).The sensitive period for second-languacquisition: An experimental study of age effectsuniversal grammar.[unpublished Ph.D. dissertatioUniv. of Illinois, Urbana-Champaign.

Singleton, D. (1989).Language acquisition: The age factClevedon: Multilingual Matters.

Snow, C. (1979). Individual differences in second-languability: A factor-analytic study.Language and Speec22, 151–162.

Snow, C. (1987). Relevance of the notion of a critical peto language acquisition. In M. Bornstein (Ed.),Sensitive periods in development: Interdisciplinary persptives (pp. 183–209). Hillsdale, NJ: Erlbaum.

Snow, C., & Hoefnagel-Ho¨hle, M. (1982a). Age differencein the pronunciation of foreign sounds. In S. KrashR. Scarcella, & M. Long (Eds.),Child-adult differ-

e:

-e

e

,

ences in second language acquisition(pp. 84–92)Rowley, MA: Newbury House.

Snow, C., & Hoefnagel-Ho¨hle, M. (1982b). The criticaperiod for language acquisition. In S. Krashen,Scarcella, & M. Long (Eds.),Child-adult differences isecond language acquisition(pp. 93–112). RowleyMA: Newbury House.

Tahta, S., Wood, M., & Lowenthal, K. (1981). Foreaccents: Factors relating to transfer of accent fromfirst language to the second language.Language anSpeech,24, 265–272.

Warrington, E. (1975). The selective impairment of semtic memory.Quarterly Journal of Experimental Pschology,27, 635–657.

White, L. (1989).Universal grammar and second languaacquisition.Amsterdam: Benjamins.

Winitz, H. (1996). Grammaticality judgment as a functof explicit and implicit instruction in Spanish.TheModern Language Journal,80, 32–46.

Zatorre, R. (1989). On the representation of multipleguages in the brain: Old problems and new DirectiBrain and Language,36, 12–147.

(Received February 9, 1998)(Revision received January 11, 1999)

age constraints on second-language acquisition

Documents