Age of learning affects the authenticity of voice-onsettime (VOT) in

stop consonants produced in a second language

James Emil Flege
Department
ofBiocommunication,
University
ofAlabamaat Birmingham.
Birmingham,
Alabama35294
(Received1January1990;acceptedfor publication23 August1990)
This studyexaminedwhetherSpanish-English bilingualsare ableto fully differentiateSpanish
andEnglish/t/according to voice-onsettime (VOT) if theylearnEnglishasa second
language (L2) in earlychildhood.
In experiment1,VOT wasmeasured in Spanishwords
spoken by Spanish monolinguals, in Englishwordsspoken by Englishmonolinguals, andin
Spanish andEnglishwordsspokenbybilinguals wholearnedEnglisheitherasyoungchildren
or asadults.As expected, theSpanish monolinguals produced/t/with considerably shorter
VOT valuesthantheEnglishmonolinguals. Alsoasexpected, the lateL2 learnersproduced
English/t/with "compromise" VOT valuesthatwereintermediate to theshort-lagvalues
observed for Spanishmonolinguals andthelong-lagvaluesobserved for Englishmonolinguals.
The earlylearners'VOT valuesfor English/t/, on theotherhand,did notdifferfromEnglish
monolinguals' VOT. Thesamepatternof resultswasobtained for stopsJinutterance-medial
positionandin absolute utterance-initialposition.The resultsof experiment 1 werereplicated
in experiment 2, wherebilingualsubjects wererequired to produce Spanish andEnglish
utterances (sentences,
phrases,words)in alternation. Theresultsareinterpreted to meanthat
individualswholearnan L2 in earlychildhood,but not thosewholearnan L2 laterin life, are
ableto establishphoneticcategories for soundsin the L2 that differacousticallyfrom
corresponding sounds in thenativelanguage. It ishypothesized thatthelateL2 learners
produced/t/withslightlylongerVOT valuesin EnglishthanSpanish by applyingdifferent
realizationrulesto a singlephoneticcategory.
PACS numbers:43.70.Ep,43.70.Kv

INTRODUCTION Hammond, 1982;Flege, 1984;Flege, 1990c),listenersseem

to classifyrealizationsof/t/in Spanishand Englishas the
The nativelanguage(L1) onelearnsin early childhood "same" a•I a phonologicallevel. For example,Born and
and a secondlanguage(L2) learnedlater in life ofteninflu- Flege (1990) foundthat Spanishmonolingualsconsistently
enceoneanother.Thishasbeenshownto betruefor process- identified long-lag English[th] tokens as/t/in a two-alter-
ing in the semanticdomain(e.g., Lambertand Rawlings, nativeforced-choice test. Englishmonolinguals identified
1969;OblerandAlbert, 1978;Magiste,1979;Mack, 1986), Spanishshort-lag[t] tokensas/t/in the majority of in-
in the syntactic domain (Blair and Harris, 1981; Mack stanceseventhoughtheyhad VOT valuesthat, in an experi-
1986), and in the phonologicaldomain (Altenbergand mentwith syntheticstimuli,wouldbeexpectedto giveriseto
Cairns, 1983). It is still uncertain whether such mutual in- the perceptionof/d/(Williams, 1977b;Flegeand Eefting,
flueneeholdstrue in the phoneticdomain,at leastfor indi- 1986;seealsoForrestand Rockman,1988). It thusappears
vidualswho learn two languagesin early childhood.The likely that Spanish-Englishbilingualsfilter out at leastsome
presentstudyexploredthedegreeolindependence ofL 1and of the audibleacousticphoneticdifferences betweenrealiza-
L2 phoneticsystemsby examiningin detail the production tionsof Spanishand English/t/because suchdifferencesare
of/t/in Spanishand Englishby two groupsof Spanish- not usedto contrastmeaningin eitherlanguage(Trubetz-
Englishbilinguals:"earlylearners"firstexposedto English koy, 1939;Weinre,ich, 1953;Morosanand Jamieson,1989).
at the ageof 5-6 yearsand"latelearners"whobeganlearn- Grosjean(1982, 1985,1989;GrosjeanandSoares,1986;
ing Englishas adults. seealso @bler and Albert, 1978) claimed that the two lan-
The/t/of SpanishandEnglishdifferin twomajorways. guagesystemsof bilingualscan neveroperatecompletely
The Spanish/t/is formedwith the tonguetip and blade independently of oneanotherbecause bothsystems areacti-
againstthe teeth, whereasthe English/t/is formedwith vatedat all times,at leastto somedegree.Thisviewsuggests
contactagainstthealveolarridge(Dalbor, 1980). Voiceless that it m•tybe impossiblefor anyonewho learnsan L2 to
stopsin thetwolanguages alsodifferin glottal-supraglottal preventpronunciation characteristicsof the L1 from influ-
timing.The/t/of Spanishis a voiceless unaspirated stop encingthe,ir pronunciation of theL2, evenif theylearnedthe
with short-lagvoice-onset time (VOT) values,whereasthe L2 asyoungchildren(seeAsherandGarcia, 1969;Thomp-
/t/of Englishisa voiceless aspiratedstopwithlong-lagVOT son, 1984). Flege (1988a, 1990a), on the otherhand,hy-
(Abraham and Lisker, 1973; Williams, 1977a; Flege and pothesizedthat completeseparationof soundsin the L 1 and
Eftting, 1986). L2 phoneticinventoriesis possible,at leastfor early learners.
Even thoughthe acousticdifferencesresultingfrom According to Flege's speechlearning model (hence-
thesearticulatorydifferences
maybedetectable
(Flegeand forth, SLM), a native Spanish-speaking
child who is first

395 d. Acoust.Sec. Am. 89 (1), January1991 0001-4966/91/010395-17500.80 © 1990 AcousticalSocietyof America 395
exposedto Englishby the ageof 5 or 6 yearswill establisha L2 stops.Contraryto the "upperlimit" hypothesisof Flege
separatephoneticcategoryfor English/t/in additionto the and Hillenbrand(1984), a few subjectsin previousstudies
one alreadyestablished for Spanish/t/. Spanishspeakers havebeenobserved to produceEnglish/p,t,k/with VOT
who learn Englishlater in life, on the other hand, are not valuesthat equaledor evenovershot 2 valuesfor native
expectedto addphoneticcategories for "similar"L2 sounds speakers of English(e.g.,Suomi,1980;Major, 1987;Flege
suchasEnglish/t/. By hypothesis, themechanism of equiv- and Eefting, 1987b).
alenceclassification
preventsthem from noting acoustic The majorityof subjectsin previousstudieswho have
phoneticdifferences between Spanish[t] andEnglish[th] produced/p,t,k/with longerVOT valuesin Englishthanin
phones,therebypreventing themfromestablishing a pho- theirL1 mustsurelyhavenotedat leastsomeof theacoustic
neticcategoryfor English/t/. Alsobyhypothesis, a phonet- differencesdistinguishing L 1andL2 stops.It isuncertainat
ic categoryis neededfor the accurateproduction andopti- present,however,whethertheydifferedfromnativespeak-
mallyefficientperception of speechsounds. ersofEnglishbecause theirperceptual knowledge ofEnglish
SinceSpanish/t/andEnglish/t/are aptto beidentified /p,t,k/was inaccurate, theirabilityto reproducewhatthey
withoneanother,anexamination ofhowbilinguals produce heardwasimperfect,or somecombinationof both.It is also
thesesounds will providean evenstrongertestof the "pho- uncertainfrompreviousresearch whetherearlylearnersare
neticindependence" hypothesis derivedfrom Flege'sSLM betterablethanlatelearnersto produceEnglish/p,t,k/with
than wouldan examination of Englishsoundsthat might VOT valuesresemblingthoseof nativespeakers.
potentiallyevadeequivalenceclassification
by virtueof dif- The resultsof severalstudiessuggest thatearlylearners
feringgreatlyfromanysoundin theSpanish phonetic inven- may fully differentiate/p,t,k/ in L1 and L2. Williams
tory (seeFlege, 1987a). Grosjean's"constantdual activa- (1977b) reportedthat adultswholearnedbothEnglishand
tion" hypothesiswould be supportedfor the phonetic Spanish by theageof sixyearsdid notdifferfromEnglish
domainif it wereshownthat bothlateand earlyL2 learners monolinguals in producingEnglishstops,nor differ from
wereunableto fully differentiate
the/t/sounds in Spanish Spanishmonolinguals for Spanishstops.Mack (1989)
andEnglish(thatis,if bothbilingualgroups produced Eng- foundthatadultswhohadlearnedbothFrenchandEnglish
lish/t/with significantly
shorterVOT valuesthana group by the ageof 7 yearsdid not differfromEnglishmonolin-
of Englishmonolinguals). Supportfor a phonetic indepen- gualsin producingEnglish/t/. Fokeset al. (1985) exam-
dencehypothesis (Flege,1988a,1990a)wouldbeprovided inedEnglishstops spoken by 12nativeArabicchildrenrang-
by thefindingthat lateL2 learnersbut notearlyL2 learners ing from 2-11 yearsof age.All but one seemedto have
differedsignificantly
fromEnglishmonolinguals in produc- producedEnglish/p/and/t/with VOT valuesthatwereas
ing English/t/. long(or longer)thanthosetypicalfor nativeEnglishchil-
L2 speechproductionresearchhasshownthat few late dren.
learnersfully differentiate/p,t,k/in their two languagesif Otherresearchindirectlysupports the viewthat early
voicelessstopsin the L1 are realizedwith short-lagVOT learnersmayfully differentiate the/p,t,k/of their L1 and
valuesandvoiceless stopsin theL2 arerealizedwith long-lag L2. NativeChinesesubjects whobeganlearningEnglishat
VOT values.Previousstudieshave shown that many adult an averageageof 7.6 yearswerefoundto produceEnglish
L2 learners produce English /p,t,k/ with significantly sentences witha detectable foreignaccent,whereas Spanish
shorterVOT valuesthan Englishmonolinguals,but with subjectswhobeganlearningEnglishbytheageof 5 to 6 years
significantlylonger ¾OT valuesthan monolingualnative producedthe samesentences without an accent(Flege,
speakers of thelearners'L 1 (e.g.,FlegeandPort, 1981;Port 1988b,1990b).Strengthof foreignaccentin sentences is
andMitleb, 1980,1983;Nathan, 1987;Flege,1987a;Major, knownto be inversely relatedto VOT in English/p,t,k/
1987. Lowie, 1988). When late learners' VOT values for (Flegeand Eefting,1987b;Major, 1987).Thusindividuals
English/p,t,k/are intermediateto the valuesobservedfor whobeginlearningEnglishin earlychildhood, butnotthose
monolingualspeakersof the L 1 and L2 they are saidto have whobeginlearningEnglishin laterchildhoodor asadults,
been produced with "compromise" values (Williams, may produceEnglish/p,t,k/with authenticVOT values.
1980). The seeminglimitation on how accuratelyVOT in The resultsof otherL2 productionstudies,ontheother
English/p,t,k/is producedalsoseemsto applyto adoles- hand,suggest that evenearlylearnersmayfail to produce
centsandolderchildren(Williams, 19791980.Suomi,1980; 'English/p,t,k/ authentically.Caramazzaet al. (1973)
FlegeandEefting,1987b;Schmidt,1988).• FlegeandHil- foundthatnativeFrenchspeakers whobeganlearningEng-
lenbrand(1984) hypothesized that an upperlimit existson lishbytheageof 7 yearsproduced English/p,t,k/withsig-
the extent to which late L2 learners can approximate the nificantlyshorterVOT valuesthan nativespeakers of Eng-
phoneticnorm of Englishfor/p,t,k/based on the observa- lish.FlegeandEefting(1987a)alsofoundthatnativeSpan-
tion that compromiseVOT valuesare commonfor late L2 ish adultsand childrenwho beganlearningEnglishL2 by
learners.
theageof 5 to 6 yearsproduced English/p,t,k/withsignifi-
Not all latelearnersproduceEnglish/p,t,k/with com- cantlyshorterVOT valuesthanage-matched groupsof na-
promisevalues,however.Some have producedEnglish tiveEnglishsubjects. Thesestudiessuggest thatearlylearn-
/p,t,k/with short-lagVOT valuesresembling thosetypical ersmay be unableto fully differentiate/p,t,k/in Ll and L2,
for/p,t,k/in theL 1,suggesting theysimplyproduced Eng- andthussupportthe viewthat boththe LI andL2 phonetic
lishwordswith L1 sounds.It is possible that suchindividu- systems remainactivatedto somedegree.
als fail to detect VOT differences between voicelessL 1 and There is reasonto think, however,that the two studies

396 J.Acoust.
Soc.Am.,Vol.89,No.1,January
1991 James
EmilFlege:
VOTina second
language 396
just citeddo not indicateaccuratelyhowwell earlylearners TABLE I. Characteriistics of monolingual nativespeakersof Spanishand
mayproduceL2 stops.It isnot certainwhatkindof English- Englishwhoparticipated in experiment 1."POB"indicates placeof birth.
languageinputsubjects in theCaramazzaetal. (1973) study Spanish English
received, norhowwelltheyspokeEnglish.SinceFrenchwas Subject Age Sex POB Age Sex POB
the dominantlanguagespokenin the city wherethe study
was carriedout (Montreal), the possibilityexiststhat the I 40 M MexicoCity 41 F Bethesda,MD
2 23 F Monterey 22 F Tampa,FL
earlylearnersexaminedby Caramazzaet al. did not receive 3 46 M Veracruz 20 F Ft. Benning, GA
sufficientnative speakerinput to enablethem to produce 4 26 MEn Escinapa 21 M Houston,TX
English/p,t,k? authentically.VOT valuesin Englishstops 5 48 M Mexico City 22 M Farmers Branch,TX
6 17 M San Luis Potosi 44 M Port Arthur, TX
spokenwith a FrenchaccentaretypicallyshorterthanVOT
7 17 M Mexico City 20 M Akron, OH
valuesin Englishstopsspokenby Englishmonolinguals 8 31 F Frontera 21 M New London,CT
(Flegeand Hillenbrand,1984;Flege, 1987a).The French- 9 20 F San LuisPotosi 26 F Hondo, TX
Englishbilingualsexaminedby Caramazzaetal. maysimply 10 32 F Frontera 21 M Cleburn, TX
M 30 26
haveproducedEnglishstopswith VOT valuesresembling
thosein the Englishstopsthey had heard.The samemight
alsobe true for the PuertoRicansubjectsexaminedby Flege
and Eefting (1987a), who were living in a predominantly
Spanish-speaking community (Mayaguez, Puerto Rico) at when they startedschoolin Texasat the ageof 5-6 years.
the time of the study. Four of the early learnerswere born in Mexico, the rest in
In summary,previousresearchhasestablished that late Texasbordertowns.Only individualswho had nativeEng-
learnersareapt to produceEnglish/p,t,k/with VOT values lish teachersin the firstthreeprimary grades,and/or had a
that are too short for English.It remainsuncertainas to majorityof nativeEnglishclassmates in thosegrades,were
whetherearlylearnerswill alsodifferfrom nativespeakersof included in the early L2 group. The early learnersreported
English,or if they will fully differentiatecorresponding L1 being unable to speak English when they beganschool,a
andL2 stops.Few previousstudieshaveexaminedwhether claim that seems reasonable in light of demographic data
learningan L2 affectshowbilingualsproducestopsin their (see Fernandez and Molinet-iMolina, 1988). A study by
L 1. It appearsthat no previousstudyhasdirectlycompared Flege (1990b) showed that, even though English was their
the productionof L2 stopsby early and late learners.Thus secondlanguage,the earlylearnersspokeit withoutaccent.
thepurpose of thepresentstudywasto determinehowclose- The Spanish-speaking researchassistantwho recordedthe
ly early and late learnerswould resemblenativespeakersof early learners in Austin indicatedthat their Spanishwasalso
English in producingEnglish /t?, and whether learning unaccented, but this was not testedformally.
Englishwouldaffecttheir productionof Spanish/t/. The late learners did not beginlearningEnglishuntil
ExperimentI examinedVOT valuesin utterance-initial they were adults. Four of them wererecordedin Austin,the
and utterance-medial stopsin Spanishand Englishwords remaining; six in Birmingham. The late learnersdiffered
that were read from lists.Experiment2 replicatedand ex- from the early learners principally accordingto theageof L2
tendedexperiment1. Its aim wasto determineif the same learning. However, the two groups of bilingualsdifferedin
SpanishversusEnglishVOT differences seenin experiment other ways, as summarized in Table II. Comparedto the
1 wouldbeobservedwhensubjectswererequiredto produce early learners, the late learners were somewhat older (34 vs
Spanishand Englishutterancesin alternation.In Sec.III, 29 years), had less formal education in English (6 vs 13
possibleunderlyingbasesfor the differences betweenearly years), aud spoke English somewhat lesson a daily basis
and late learnersobservedin experimentsI and 2 are dis- according:
to self-report
(66% vs82%).3Thelatelearners
cussedin thecontextofFlege's( 1988a,1990a)speechlearn- had arrivedin the U.S. at a much later averageage (20 vs 2
ing model. years), and so had lived there for a shortertotal periodof
time than the early learners( 14 vs 21 years).
I. EXPERIMENT 1

A. Methods
•. Materials and procedures
1. Subjects Owingtophonological differences betweenSpanish and
Two groupsof monolingualsand two groupsof bilin- English,it wasnotpossible to findlistsof matchedEnglish
guals(six malesand four femalesper group) participatedas and Spanishwords.The wordschosen,however,wereall
paid subjects.As summarizedin Table I, the Spanishand disyllabic
andhadvowelsof approximately thesamequality
Englishmonolinguals differedlittle in meanage (30 vs 26 followingthe word-initial/t/. The nativeSpanishsubjects
years).The Englishmonolinguals werestudentsat the Uni- readSpanish wordsat theendof theSpanish carrierphrase
versityof Texas.The Spanishmonolingualswere recruited "Tengoun ." H.'dfof the Spanishwordswerefollowedby
at a refugeecenterin Austin,TX. Most hadlivedin theU.S. /i/ (tigre',
•ipo,tiro,timbre),
theotherfourby/•/ (tema,
for lessthan 3 monthsat the time of testing.Only a few of termo,templo,texto).4 The Englishmonolinguals
andthe
them reported having studiedEnglish in school. Spanish-Englishbilingualsread Englishwordswith/i/or
Subjectsin the othertwo groupswerenativespeakersof /el (t-bone,teapot,teabag,t-shirt,teller,temple,textbook,
Spanishwho learnedEnglishas an L2. The early learners tempo)at theendof theEnglishcarrierphrase"Takea ."
indicatedthat they were first exposedmassivelyto English TheSpanish
andEnglish
materials
were
elicited
in-•he
397 J. Acoust.Soc.Am.,Vol.89, No. 1, January1991 Jame.,;EmilFlege:VOT ina secondlanguage 397
TABLE II. Characteristics
oftheearlyandlateL2 learners
whoparticipat- carrierphrases)and the otherin utterance-medial position
edin experiment
1.The 14subjectswholaterparticipatedin experiment
2 (in the/t/initiating test words at the end of the carrier
are markedby an asterisk.
phrases).VeT wasmeasured to thenearest0.1 msfromthe
Early L2 Learners screenof a graphicsterminalfrom the beginningof the re-
Subject Age Sex POB
a EDUbAOA'LORa PER' leaseburstto thefirstpositivepeakin theperiodicportionof
the waveform.
1' 23 F Edinburgh,TX 12 0 23 90 A total of four mean VeT values were calculated for
2* 26 M Mexico City 12 6 20 70
3* 19 M Guadalajara 13 5 14 eachsubject.Meanvaluesfor/t/in thetestwordsbeginning
4* 20 F Farr, TX 13 0 20 80 with/i/and/e/were calculated, most basedon sevenobser-
5 23 F Dallas, TX 12 0 23 90 vations.The fewmissingdatapointsweredueto theabsence
6* 20 F Obregon 11 3 15 80
7* 23 M unknown 13 I 22 75 of a visiblereleaseburst, which made it impossibleto mea-
8 21 F erie 13 0 21 90 sure VeT. Two mean valueswere calculatedfor stopsin
9 24 M Laredo, TX 16 0 24 80 utterance-initialposition.One wasfor utterancesendingin
10' 26 M Taft, TX 13 0 26 85 test words with/i/, the other for utterances with/e/test
words.
M 29 13 2 21 82
Late L2 Learners Measurement reliabilitywasassessedusingthe test-re-
Subject Age Sex POB' EDUbAOAcLOR• PERe test approach.The assistantwho measuredVeT remea-
sured 20 randomly selectedutterance-initialand 20 utter-
1' 28 M MexicoCity, Mex. 0 18 10 70
2* 23 M Juarez,Mex. 6 13 10 50 ance-medial/t/ tokensseveralweekslater. The average
3* 37 F Montetrey,Mex. 10 12 25 50 (unsigned)differencebetween thetwosetsof measurements
4* 19 F Matamoros, Mex. 7 11 8 70 was 1.5 mm for both the utterance-initial and utterance-me-
5* 41 M Chile 6 24 17 73
dialstops.The largestdifference notedwasonly5.3msso,of
6 38 F Ecquador 12 14 24 90
7 53 F Panama 0 35 18 95 course,thefirstandsecondsetsof measurements werehigh-
8 28 M Nicaragua 7 18 10 50 ly correlated(r ----0.997 for utterance-initialstops,r = 0.999
9* 33 M Guatemala 5 26 7 75
for utteranee-medial steps).
10' 40 F Chile 4 26 14 35

M 34 6 20 14 66
B. Results
Place of birth.
1. Utterance roedial stops
Yearsof formalinstruction
in English.
Age of arrival in the U.S., in years. Figure 1 (bottom) showsthe mean VeT valuesfor
Lengthof residence
in theU.S.,in yearn SpanishandEnglish/t/tokens that wereproducedin utter-
Self-estimated
percentagedailyuseof English.
ance-medial position.The valuesshownherehavebeenaver-
agedacrossthe/i/and/e/contexts, the effectof whichwill
be discussed below.As expected,the monolingualEnglish
speakers'/t/had substantially longerVeT valuesthanthat
appropriate languagesbybilingual research assistants?
The of the Spanish monolinguals (64 vs22 ms). Also asexpect-
monolingual subjects
producedonlytheSpanishor English ed, both the early and the late learners produced/t/with
materials,whereasthe bilingualsproducedbothin counter- longer VeT values in English than Spanish. The late learn-
balancedorder.Instructionsweregivento themonolinguals ers produced English/t/with shorter VeT values (40 ms)
in Spanishor English,asappropriate.The bilingualsubjects than the English monolinguals, whereas the early learners'
assignedto producethe Spanishmaterialsfirst heard the English/t! had the samemeanvalues(viz. 64 ms) as the
Spanishinstructions,and viceversa.The subjectsweretold Englishmonolinguals'. The meanVeT valuesobtainedfor
that theexperimentexaminedspeech,but not that theirpro- each of the 30 subjects who producedthe English speech
ductionof/t/would be assessed. They were instructedto material were submitted to a (3) groupX (2) vowelcontext
read eachsentence"as if talking to...friends"at a constant ANOVA, which yielded a significantgroup main effect
speakingrateandloudness level,andto repeatanyutterance [F(2,27) ----
14.0, p < 0.05 ]. Newman-Keulsposthoctests
with whichtheywereunsatisfied. The subjectssaidthenum- revealed that the English monolinguals and early learners
berof eachutterance(in theappropriate language), paused, produced English/t! with significantly longer VeT values
then produced the utterance. The reading task was modeled than the late learners, but did not differ significantlyfrom
at a moderatespeakingrate on the instructiontapeusinga one another (p < 0.05).
list of utterancesresembling thoseon the randomizedlists. The early and late learnersdid not differin producing
stopsin their LI. The subjectsin bothbilingualgroupspro-
ducedSpanish/t?with anaverageVeT of 20 ms,whichwas
3. Measurements
slightlyshorterthan the meanvalueobservedfor Spanish
Eachof the testwordsoccurredthreetimeson the Span- monolinguals(viz. 22 ms). The mean VeT valuesobtained
ishandEnglishlists.A total of 14 utterancesfrom the mid- for the 30 subjectswhoproducedSpanishwordsin the con-
dle of each list were digitizedat 10 kHz. Each utterance textof/i/and/e/were submittedto a ( 3) groupX (2) vow-
contained two word-initial/t/tokens, one in absoluteutter- el context ANOVA. The group factor was nonsignificant
ance-initialposition(i.e., the/t/in take and tengoin the [F(2,27) = 0.447]. This suggests that havinglearnedEng-

398 J. Acoust.Sec. Am., Vol. 89, No. 1, January 1991 James Emil Fiego: YeT in a second language 398
 VOTin utterance-initial/t/ /t/tokens [F(2,:!7) = 15.2,p <0.05]. A posthot'test re-
80
Spanish vealedthat theEnglishmonolinguals andtheearlylearners
English had significantly longerVOT valuesthan the late learners
60 but did not differfrom oneanother(p < 0.05).
The early anti late learnersproducedSpanish/t/with
VOT valuesthat wereaboutthesameon theaverageasthe
40
Spanishmonolinguals'meanvalues( 19, 20 vs 18 ms). Not
surprisingly,the groupfactorwasnonsignificant in the AN-
20
OVA examiningSpanishstops[F(2,27) = 0.28].
A (3) group)((2) languageANOVA wascarriedoutto
examinethebilingualsubjects' productionof absoluteutter-
o ance-initi,'tl
stopsin SpanishandEnglish.Thisanalysisyield-
80
VOTin utterance-medial /t/ eda significantinteraction[ F( 1,18) = 18.2,œ< 0.05]. As in
theanalysisof utterance-medial stops,boththeearlyandthe
• Spanish late learnersproduced/t/with significantlylonger VOT
• English
•' 60 values in Englisla than Spanish [F(1,9)= 225.3, 8.75;
E
p < 0.05]. The interactionwasprobablydueto the fact that
o the earlylearnersproduceda largerSpanishversusEnglish
B 40 VOT differencetl•an the late learners (34 vs 13 ms).

-• 20 $. VOT variability
To determineif the bilingualsubjectswere more vari-
able in producing:English/t/than the Englishmonolin-
Spanish Late L2 EarlyL2 English guals,intersubject variabilitywasexamined.Figure2 shows
monaling. Learners Learners monaling.
the meanVOT valuesobtainedfor individualsubjects in ut-
FIG. 1. (top) VOT in the stopsinitiating Spanishand English carrier terance-medialand utterance-initialEnglish/t/tokens in
phrases.
(bottom)ThemeanVOT valuefortheword-initial
stopsin Span- thetwovowelcontexts. Visualinspection
of thisfiguresug-
ish and Englishtestwordsbeginningin/t/that werespokenat the end of geststhat intersubjectvariabilitymay have beengreater
carrier phrases,in ms. Each mean VOT value is basedon ten individual
amongthe'.nonnativethannativesubjects. However,anF•
subjectmeanswhich,in turn, werebasedonup to sevenobservationsin each
of two vowel contexts (/i/de/). The error bars enclose q- / - one stan- testshowedthat the assumption of homogeneityof variance
dard deviation. for subjec[sin the threegroupscouldnot be rejectedeither
for utterance-medialstops (F•,• = 1.63 with three var-
iancesandninedfper variance)or for stopsin absoluteut-
lishdid not influence
howthe bilingualsproducedstopsin terance-initialposition(Fm• = 2.38)
their LI. Mostof theearlylearnersproducedEnglish/t/in utter-
A (2) groupX ( 2 ) languageANOVA wascarriedoutto ance-medialpositi.on with meanVOT valuesthat fell within
determineif thetwo groupsof bilingualsdifferedsignificant- or exceeded
therangeof valuesobserved
fornativespeakers
ly in termsof howwell theydifferentiated the/t/of Spanish of English(viz. 51-76 ms), whereasmostof the latelearners
had mean VOT values that were less than the smallest value
and English.The analysisyieldeda significantinteraction
[F(1,18) = 12.9,p < 0.05], which wasfollowedup by testsobserved for any of the iEnglishmonolinguals.
Eachof the
of the simplemaineffectof group.Thesetestsshowedthat early learnersproducedmeanvaluesfor/t/in
absoluteut-
boththeearlyandthelatelearnersproduced/t/with signifi- terance-initial position thatfellwithinor exceeded
theEng-
cantly longer VOT values in English than Spanish lishrange(viz. 34--63ms) whereas onlyabouthalfofthelate
IF(1,9) = 138.1,14.3;p < 0.05]. The interactionwasprob-learners'•neanvaluesfell within the Englishrange.Some
late learnersproducedEnglish/t/in bothutterance-medial
ably due, therefore,to the fact that the early learnerspro-
duceda substantially largerVOT differencebetweenSpan- and utterance-initialposition with mean VOT values of
ishand English/t/than the late learners(44 vs20 ms). about20ms,suggesting thattheyuseda Spanish/t/inEng-
lish. All of the early learner'smean valuesexceeded30 ms
andcouldtherefo•re betermed"long-lag"stops.
2. Utterance-inltial stops A secondmethodwasusedto determineif thebilingual
Figure1 (top) showsthemeanVOT valuesof Spanish subjectsweremore variablein producingEnglish/t/than
and English/t/tokens spokenin absoluteutterance-initial the Englishmonolinguals.The standarddeviations(s.d.)
position.The monolingualEnglishspeakersproduced/t/ associated with each of the four mean VOT values obtained
withsubstantially longerVOT valuesthantheSpanishmon- for eachsubjectwascalculated.Overall, the nativeEnglish
olinguals(51 vs 18ms). The earlylearnersproducedEng- speakers's.d.valuesdifferedlittle from thoseof the early and
lish/t/with about the same mean VOT value ( 53 ms) as the late learners (8.9 vs 9.8, 8.3 ). The s.d. values associated with
Englishmonolinguals, whereasthe late learnersproduced the production of English /t/ were submittedto a (3)
English/t/with a shortermeanvalue(33 ms). The group group• (2) vowel contextX (2) utterance position
factorwassignificant
in theANOVAexaminingtheEnglish ANOVA, which yieldeda significantthree-wayinteraction

399 J. Acoust.Soc.Am.,Vol.89, No.1, January1991 JamesEmilFlege:VOT ina secondlanguage 399

 Stops in Utteronce-mediol position werenonsignificant in the analysesof bothSpanish/t/and
9O English/t/.
The vowelcontexteffectsprobablyhad an aerodynamic
8O
basis.It appearsthat the time neededfor the transglottal
70
pressuredrop neededfor spontaneous voicing may take
E longerto reestablishitselfafterthe releaseof stopsthat pre-
6O cedevowelsformedwith tight lingual-palatalconstriction
than vowels without such a constriction.7 Vowel context ef-
50
8 8 o fectssimilarto the onesreportedherefor Spanishand Eng-
lishstopshavebeennotedin previousstudiesof stopproduc-
40 o
c tion by Englishmonolinguals(Port and Rotunno, 1979;
o
30
Weismer, 1979). The vowel context effect will not be dis-
cussedfurther sinceit doesnot appearto haveresultedfrom
o 2O language-specificphoneticlearning.

10
C. Discussion

0 Spanish/t/was found to have VOT valuesof 22 ms in

utterance-initialposition,and 18 ms in absoluteutterance-
Stops in Utteronce-initiol position
90 initial position.Thesevaluesareslightlylongerthanvalues
reported by Lisker and Abramson (1964) and Williams
80 (1977a) for Spanish.English/t/was foundto have ¾OT
valuesof 64 ms in utterance-initial
position,and 51 msin
7O
absoluteutterance-initialposition.These valuesare also
o o somewhatlongerthan valuesreportedin someprevious
60
studies.
For example, LiskerandAbramson (1967)report-
50
ed a mean value of 48 ms for word-initial/t/tokens in utter-
ance-medial position,and 45 ms for/t/tokens initiating
40 wordsspokenin isolation(andthusin absolute initialposi-
tion). Thesedifferencesbetweenstudies canprobablybeat-
30

20
o tributedto differences
speakingrates.
in vowelcontext,degreeof stress,and

Theprimarypurpose of experiment I wasto determine

10 if Spanish-Englishbilingualswouldbeableto fully differen-
tiate/t/in their two languages, that is, to producea VOT
0 differenceequalto thedifference between Spanish andEng-
i eh i eh i eh lish monolinguals. The resultssuggested that at leastsome
LATE L2 EARLY L2 NAT. ENG. bilingualsareableto fully differentiatetheirtwo languages
at a phoneticlevel.Early learnerswho learnedEnglishas
FIG. 2. (top) The meanVOT valuesobtainedfor/t/as producedin the youngchildrenproducedSpanish/t/with meanVOT val-
word-initialpositionof testwordswith/i/and/e/("eh") foundat the end
of a carrierphrase,in ms; (bottom) the meansfor stopsin the absolute
uesthat did not differsignificantlyfrom thoseof Spanish
initialposition
ofthecarrierphrase
"Takea _ fortheutterances
with/i/ monolinguals, and they producedEnglish/t/with mean
and/•/("eh") test words. VOT valuesthat did not differsignificantly from thoseof
English monolinguals.The same pattern of differences
betweengroupswasobtainedfor stopsproducedin utter-
[F(2,27) = 3.67,p< 0.05]. The interactionwasexploredby ance-medial positionandin absolute utterance-initial posi-
examiningthe simplemain effectof group at all four tion.
vowelX utterancepositioncombinations. The groupfactor Latelearners
whowere firstmassively exposedtoEng-
wasnonsignificant in everyinstance(p > 0.05), suggesting lishasadults,ontheotherhand,onlypartiallydifferentiated
that the nonnativespeakers wereno morevariablein pro- Englishand Spanish/t/. ThesesubjectsproducedSpanish
ducingEnglish/t/than thenativespeakers. 6 /t/with valuesmuchlike thoseof Spanishmonolinguals.
AlthoughtheyproducedEnglish/t/with significantly long-
4. Effect of vowel context
er VOT valuesthanwereobservedfor the Spanishmonolin-
For word-initialEnglishstopsin utterance-medialposi- guals'/t/ tokens,they producedEnglish/t/with signifi-
tion (that is, in the test words at the end of the carrier cantly shorterVOT valuesthan Englishmonolinguals.
phrase),VOT wassignificantly longerin thecontextof/i/ Previousstudiesof L2 productionof late learnershave also
than /e/ (60 vs 52 ms) [F(1,27) = 26.8, p<0.05]. For shownsuchcompromiseVOT values(e.g., Nathan, 1987;
Spanish stopsin the sameposition, VOT wasalsosignifi- Major, 1987).The findingthat fewlatelearners'VOT values
cantlylongerin the contextof/i/than/e/(23 vs 18 ms) forEnglish/t/closelyresembled thoseof thenativespeakers
[F(1,27) = 41.4,p < 0.05]. The group• vowelinteractions despiteexposureto native-produced English stopsover

400 J. Acoust.Soc.Am.,Vol.89, No.1, January1991 JamesEmilFlege:VOT ina secondlanguage 400

manyyearsagrees withtheobservation thatadults'L2 pron- more years (Williams, 1979, 1980; Flege and Port, 1981;
unciation tendsto "fossilize"(Selinker,1972;Scovel,1988). Port and Mitleb, 1983).
It is alsoconsistentwiththehypothesis (FlegeandHillen- A studyof intrasubjectvariabilityshowedthat neither
brand,1984) that an upperlimit existson howcloselylate the early nor the late learnershad greatertoken-to-token
learnerscan approximatethe phoneticnormsof an L2 for variability in producing/t/than the Englishmonolingual
soundsin the L2 that differacousticallyfrom corresponding subjects. Therewassomewhat moreintrasubject variability
sounds in the LI. amongthe late leamen than amongthe earlylearnersand
The resultsobtainedhere for early learnerssupportan Englishmonolinguals, but the differenceswerenonsignifi-
inferencedrawn from an imitation studyby Flegeand Eelt- cant.An inspection of individualsubjectdatasuggested that
ing (1988}. In that study, native speakersof English and themeanvaluereportedfor the latelearners'productionof
earlyL2 learnersproduced stopswithshort-lagandlong-lag English/t/did notadequately representall subjectsin that
VOT valueswhen imitating the short-lagand long-lag group.Althoughmostsubjects producedEnglish/t/with
membersof a syntheticVOT continuumrangingfrom/da/ theexpected compromise values,someseemed to havepro-
to/ta/. Late learners,on the otherhand,seldomproduced ducedEnglishwordswith a "Spanish"/t/(i.e., with short-
long-lagVOT valueswhenimitatingstimulifrom the long- lagVOT valuesof about20 ms) anda fewmanagedto pro-
lag end of the continuum.This findingsuggested that the duceEnglish/t/with long-lagVOT that fell within the
earlyL2 learnershadestablished a phoneticcategoryfor the rangeof valuesobserved forthenativeEnglishsubjects. It is
long-lag/t/realizations of Englisheventhoughthey pro- uncertain whether these individual differences were due to
ducedEnglish/p,t,k/with compromiseVOT valueswhen differencesin underlyingphoneticorganization,or in the
speakingspontaneously.The authors concludedthat the modulationof phoneticparameters.
earlylearnersmayhaveproducedEnglish/p,t,k/with "ac- The resultsof previousresearchwith late learnersre-
cented"VOT valuesbecausethey had receivedforeign-ac- viewedin the Introductionsuggested the possibilitythat the
centedinput as youngchildren. abilityto accurately produce voicelessaspiratedstops in an
Theresultspresented heremustbeinterpreted cautious- L2 may be normally distributed.This observation must be
ly for a numberof reasons. First, the studydid not examine considered tentative for several reasons,however. First, as
conversational speech.Second,only eight Englishwords alluded to above, most previous studies have examined
wereexamined,and thesewerenot evaluatedfor degreeof speechthat wasread rather than spokenconversationally.
subjective familiarity.If the effectsof phoneticlearningdif- Suchan elicitationprocedureis likely to increasethe likeli-
fusegraduallyacrossthe lexicon,as for childrenlearning hood that articulationstrategiesmay obscurenormal pat-
Englishas an LI (Ferguson,1986), thenthe accuracyof ternsof production.Second,relativelyfew previousstudies
VOT in an Englishstopspokenby L2 learnersmaydepend havereprotedindividualsubjectdata. Most L2 production
on when the word containingthat stop was first encoun- studieshavesimplypresentedmeanvaluesfor groupsof sub-
tered. jects.
Another reasonfor caution is that the two bilingual
groupscomparedin experiment1 differedaccordingto fac- II. EXPERIMENT 2

torsin additionto ageof L2 learning.The earlylearnerswere The purposeof thisexperimentwasto replicateandex-

likely to havereceivedmuchmorenative-speaker phonetic tendexperiment1. In experiment1,Spanish-Englishbilin-
input than the late learners(seeTable II). It is unlikely, gualsubjects
readlistsof EnglishandSpanishutterances in
however,that a lack of phoneticinputper secouldaccount counterbalanced order. The research assistants who elicited
for why the late learnersproducedEnglish/t/with signifi- the data switchedbetweenthe two languages, as appropri-
cantlyshorterVOT valuesthan the Englishmonolinguals. ate,to reducetheartificialityof thespeakingsituation.How-
Each of them had lived in the U.S. for at least7 years,and all ever, bilingualsseldomconfinethemselves to speakingL!
but oneof themreportedusingEnglishat leasthalf of the and L2 in distinct,nonoverlapping blocks.They typically
time on a daily basis. switchbetweentheirtwo languages, at leastwith interlocu-
It is alsounlikelythat differencesin amountof L2 input torsfamiliarwithbothlanguages, asweretheassistants who
could account for the observed VOT differences between the
elicitedthedata (Grosjean,1982}. Therefore,in thepresent
twobilingualgroups.Previousresearchhasshownthat age experiment, someof thebilingualsubjects fromthefirstex-
of learningisthesinglemostimportantdeterminant
of how perimentproducedSpanishand Englishin alternation.
wellanL2 will bepronounced.Lengthof residence,
whichis
oftenusedasa grossestimateof amountL2 phoneticinput, A. Methods
has often been shown to exert a much smaller, and usually
1. Subjects
nonsignificant, effecton globalforeignaccent(Tahta et el.,
1981;Seligeret al., 1975;Oyama, 1976;Purcelland Surer, Of the bilingual subjectsin experiment1, all but two
1980;Thompson,1984;Flege, 1990b). Global foreignac- early learnersparticipatedin this experiment.Data for a
cent, in turn, is known to be significantlycorrelatedwith third early learnerwere not usableowing to an equipment
VOT in stopsspokenby nonnativespeakers(Flege and Eelt- malfunction. To ensure an even number of subjectsin the
ing, 1987a;Major, 1987). A numberof studieshaveshown two bilingualgroups,threelate learnerswereeliminatedby
little or no differencefor VOT in stopsspokenby L2 learners randomselection.The sevenearlyandsevenlatelearnersfor
whohadjust arrivedin the U.S. or hadresidedtherefor 3 or whom data will be reportedare indicatedin Table II. This

401 J. Acoust.Soc. Am.,Vol.89, No. 1, January1991 JamesEmilFlege:VOT in a secondlanguage 401

subsetof subjectsdifferedin muchthe sameway described none was based on fewer than three tokens. The assistant
earlier in terms of chronologicalage, formal educationin listenedfor pausesbetweenutterancesbefore making the
English,self-estimated daily usageof Englishand--most VaT measurements. The/t/'s were not measured in the
importantly--ageof L2 learning.The early learnersin ex- 1.6% of sentence-condition
utterances
judgedto havebeen
periment2 beganlearningEnglishat theageof 5 to 6 years, precededby a pause;the smallpercentageof utterances
in
whereasthe late learnersbeganlearningEnglishwhenthey the phraseand word conditionsprecededby a perceived
arrivedin the U.S. at an averageageof 19 years. pause(3.1% in both instances)were alsoexcludedfrom
analysis.
2. Speech materials
B. Results
VOT was measuredin the sameEnglishand Spanish
wordsthat wereexaminedin experiment1. The wordswere 1. VOT
producedin three consecutiveconditionsdesignatedthe The meanVaT valuesobtainedfor the two bilingual
"sentence,""phrase,"and "word" conditions.In the sen- groups are shown in Fig. 3. Averaged acrossthe three
tencecondition,disyllabictest wordswere producedat the switching conditionsand two vowel contexts, the early
endof an Englishor Spanishcarrierphrase,asappropriate learnersproduceda substantially
greaterdifference
between
("Takeanother
wordsuchas_";"Tengopalabras
como_"). English/t/and Spanish/t/ (60 vs 21 ms) than the late
In the phrasecondition,the testsamewordswereproduced learners (38 vs 22 ms). The mean VaT valueswere submit-
at the end of an English or Spanish phrase ("Take ted.to a (2) groupX(3) switchingconditionž(2) lan-
a_";"Tengo un_").In thethirdcondition, Spanish guage• (2) vowelANOVA with repeatedmeasureson the
isolated
and Englishtest wordswereproducedin alternation.The last threefactors.This analysisyieldeda significantgroup
samerandomorder (that of experiment1) wasusedin all • languageinteraction[F(1,12) = 12.0,p < 0.05], which
three conditions. wasexploredby testsof simplemain effects.Both the early
andthe late learnerswerefoundto produce/t/with signifi-
3. Procedures cantly longer VaT values in English than Spanish
[F(1,6) = 163.1,8.72,p < 0.05]. The sourceof the interac-
The experiment wascarriedout by the samebilingual tionwastherefore likelytohavebeenthegreater
magnitude
researchassistants as in experiment1. Half of the subjects
of the SpanishversusEnglishdifferencefor the early than
heardrecordedinstructions in English,half in Spanish.The the late learners (41 vs 16 ms).
subjectswere told that they wouldproduceEnglishand
Spanish sentences, phrases, andisolatedwordsin alterna-
tion.To distinguish betweenlanguages, theSpanish materi- 80 Late L2 Learners • ENG
alswerehighlighted onthewrittenlistsusedto elicitproduc- • SPAN
70
tion. In the isolatedword condition,languageidentity was
60
redundantly specified by placingthe letters"E" or "S" in
front of each word. 50

The onset of each utterance in the three conditions was 40

regulatedby a light-flashingdevice.The subjectswere in- 30

structed to time the onset of successive utterances to coin-
20
cidewith the light flashes.The intervalbetweenflasheswas
10
3.2 sin thesentence condition,1.8sin thephrasecondition,
and 1.1 s in the word condition. Pilot tests revealed that these 0

werethe shortestintervalsthat somesubjectscouldaccom-

modate. The subjectswere permittedto practicewith the 80 Early L2 Learners • ENG
ISS3 SPAN
sentencematerialbeforedatacollectionbegan.The subjects 70

were told to skip an utterancealtogetherif they lost the 60

rhythm they had established.In several instancesthe sub- 5O
jectscameto a completehalt. The conditionsin which this
4O
happenedwerererun.
30

4. Measurement 20

10
The VaT of word-initial/t/'s in Spanishand English
test words were measured to the nearest 0.1 ms from oscillo- 0

graphicdisplays.A total of 20 word-initial/t/tokens from Sentence Phrase Word

the middleof thelistsweremeasuredfor eachsubjectin each SWITCHING CONDITION

condition.From these,four meanVaT valuesper condition
were calculated:two for English/t/tokens (one for words FIG. 3. The meanVaT (in ms) of SpanishandEnglish/t/s spokenbyearly
andlatelearnersin threeconditions(sentence,
phrase,word) in whichEng-
with/i/, one for wordswith/e/) and two for Spanish/t/ lish and Spanishutteranceswere producedin alternation.Each mean is
tokens (one for words with/i/, one for words with/e/). basedon measuresmadefor sevensubjectsin two vowelcontexts.The error
Most of the mean valueswere basedon five observations; bars enclose + / - one standard deviation.

402 J. Acoust.Sac. Am., Vol. 89, No. 1, January1991 James EmilFlege:VaT in a secondlanguage 402
 Averaged acrossthe twogroupsandtwo vowelcon- The measuredduration of the intervals between/t/'s
texts,themagnitude of theSpanishvsEnglishdifference initiatingsuccessiveSpanishand Englishtestwordswas
decreased astherateofcross-language
switching
increased.somewhat shorteron theaveragethanthe actualintervals
It averaged32msin thesentencecondition,
wheretheinter- specified bythelightflashes
in thesentence andphrase con-
valbetween successiveSpanishandEnglish/t/'swasthe ditions(2.9and1.6),butslightlylonger( 1.2s) in theword
greatest.
It averaged26 msin thephrasecondition,
and25 condition.The meanintervalsfor the earlyand the late
ms in the word condition.Thesedifferencesresultedin a learnersareshownin Fig. 4 asa functionof thedirectionof
significantlanguageX switchingconditioninteraction the switches.For the E-S switches,the/t/-to-/t/intervals
[F(2,24)= 4.44,p < 0.05],whichwasexploredbytestsof decreasedastherateof switching became morerapid(sen-
simplemaineffects.
Theconditionfactorwassignificant
for tence:2.7, phrase:1.7, word: 1.2 s). The sameheld true for
boththeEnglish/t/andtheSpanish/t/[F(2,26)= 5.03, the S-E switches(sentence: 3.0, phrase:1.5,word:1.2s).
6.02,p < 0.05]. Posthoctestsrevealedthat VOT valueswere Averagedacrossthe two groups,the /t/-to-/t/ intervals
significantly
greaterfor Englishit/in the sentencecondi- averaged62 ms longer (i.e., 1.909 vs 1.847 s) for the S-E
tionthanin eitherthephraseor thewordconditions (53 vs thanE-S switches. The/t/-to-/t/intervalsaveraged312ms
46, 48 ms). VOT for Spanish/t/,on theotherhand,was longer(i.e.,2.034vs1.722s) forthelatethanearlylearners.
significantly
shorterin thesentence andphrase conditions The difference betweenthe two groupsdecreased from the
thanin theisolatedwordcondition (21, 20vs24ms). sentence to thephraseto the wordcondition(497, 341,100
Asinexperiment 1,theVOTfor/t/waslonger forstops ms).
preceding/i/than/e/.Thevowel contextfactorwassignifi- The84mean/t/-to-/t/valuesobtained forsubjects
in
cantforbothEnglish/t/andSpanish/t/[F(1,13)= 30.8, the two bilingual groups were submittedto a (2)
21.0;p<0.05]. It is likelythat the significant
langua- group• (3) conditionX (2) direction-of-switch
ANOVA,
geX vowelcontextinteraction obtained[F(1,12) = 6.03, withrepeated
measures
onthelasttwofactors.
A significant
p < 0.05] wasdueto thefactthat,asin experiment1,the groupX condition interaction was obtained
vowelcontext effectwasgreater
in English
words(55 vs44 [F(2,24):--6.24,p < 0.05],whichwasexplored bytesting
ms) thanin Spanishwords(24 vs 19ms). thesimplemaineffectofgroupforthethreeswitching condi-
tions.The earlylearners'/t/-to-/t/intervalsweresignifi-
2. Switching time cantly shorter than the late learners' in the sentenceand
phrase
conditions
[F( l, 12) --- 10.18,9.82;p•O.05] butnot
Theearlyandlatelearners
weregiventhesameinstruc-
in thewordcondition[F(1,12) ----2.66].
tions,andtoldto timesuccessive
utterances
in thethreecon-
ditionsin synchrony
withlightflashes
emittedbya timer. 3600
Lore L2 Learners [•] ENG
The timebetweensuccessive
flashesdecreased
acrosscondi- N-X-'xlSPAN
• 3200
tions,whichmeantthatthesubjects
hadto switchevermore
rapidlybetween
languages.
Thelackof a significant
grou- • 2800
pX language
X condition
interaction
suggested
thattheneed ?400
toswitchmorerapidly between
L 1andL2 affectedspeech o 2000
productionby thetwobilingualgroups in thesameway. ß

However,it appearsthatthesubjects
in thetwogroups did • 1600
notperform thetaskinthesamewayeventhough theywere • 1200
giventhe sameinstructions.
800
Although
thesubjects
developed
a rhythmbased
onthe
Sertence Phrase Word
lightflashes,
theonsets
of theirutterances
sometimes
pre-
cededthelightflashes,andsomeutterances encroachedonto 3600
Early L2 Learners r---I ENG
the next interval.As an estimateof the time availablefor • SPAN

planning
theproduction
of eachSpanish
andEnglish/t/, • 3200
"/t/-to-/t/" intervalsweremeasuredto the nearest0.1 ms • 2800
:ineachconditionfrom the releaseburstof the/t/in one
(Spanishor English)testwordto thereleaseburstof/t/in •2400
o 2000
the followingtestword. Two meanvalueswere calculated ß
for eachsubjectin thethreeconditions.Onemeanvaluewas • 1600
based onmeasuresofthetimebetween the/t/in eachEng- • 1200
lishtestwordandthe/t/in thefollowing
Spanishtestword;
theothermeanvaluewasbasedonthetimebetween
the/t/ 800

ineachSpanishtestwordandthe/t/in thefollowing
Eng- Serl•ence Phrose Word

lishtestword.Theseintervals
weredesignatedthe"E-S" SWITCHING CONDITION
and"S-E" switches, respectively.MeanE-S andS-E values FIG. 4. The meandurationof the intervalbetweenthe releaseburstsof
werecalculatedfor eachsubjectfor eachof the threecondi- successive
SpanishandEnglish/t/'s inthreeconditions
(sentence,phrase,
tions.Eachmeanwasbasedon two observations, onein word)inwhichsubjectswererequired toswitchbetween
Spanish andEng-
which the vowel of the test word was/i/and one in which lish."SPAN"indicates
switchesfromSpanish toEnglish,
"ENG"theEng-
lishto Spanish
switches.
Theerrorbarsenclose+ / - onestandard
devi-
the vowel was/•/. ation.

403 J.Acoust.
Sac.Am.,
Vol.
89,No.1,January
1991 James
Emil
Flege:
VOTinasecond
language 403
 The direction-of-switch factor was marginallysignifi- voweland a consonant(viz. /e'l,lt/) that only partially
cant [F(1,12) = 4.61,p = 0.053]. A three-wayinteraction resembleanySpanishvowelor consonant.It containedthree
involving the direction factor was obtained vowelsand one consonant(viz./ae//$'/,/n/,/O/) without a
[F(2,24) = 10.4,p < 0.05]. The simplemaineffectof group directcounterpartin Spanish;andit containedthreeconson-
was testedfor all eight directionX conditioncombinations. ants(/z/,/•/,/d/) that arenot foundin a comparable sylla-
Only onebetweengroupdifferencewassignificantat a per- ble positionin Spanish.
experiment errorrateof 0.05 [F(1,12) = 16.0,p = 0.0018]. It is temptingto speculatethat the late learnersmay
For S-E switches in the sentencecondition, the/t/-to-/t/ haveprolongedsoundsor pausedbetweenwordsin anticipa-
intervalsweresignificantlylonger(by 772 ms) for the late tion of the needto producethe non-Spanish sounds.They
than early leamen. The two groupsshowedno significant may haveneededadditionaltime to formulateplansfor im-
differences whenswitchingfrom Englishto Spanish. plementingEnglishsoundsthat arenot foundin theSpanish
phoneticinventory.The addedtimemay havebeenexpend-
C. Discussion ed on producingsoundswhosemodeof motorieimplemen-
This experimentyielded resultsthat were much the tationhadnot yet beenfully automaticized,or to makemod-
sameas thoseobtainedin experiment1. Both early and late ifications of previously established "programs" for
implementingsoundsin Spanish.One resultobtainedin ex-
learnersproduced/t/with significantlylongerVOT values
in Englishthan Spanish,but the magnitudeof the Spanish periment2 is consistent with eitherinterpretation.The/t/-
versusEnglishVOT differencewassubstantially greaterfor to-/t/intervals were62 ms longeron the averagewhenthe
the early than the late learners(41 vs 16 ms). bilingual subjects
switched fromEnglishto Spanishthanthe
reverse.
The bilingualsubjectsproduceda largerVOT difference
One possibilityis that the "switchingtime" difference
in the sentence condition,whereSpanishand Englishsen-
betweenearly and late learnersderived from the method
tenceswereproducedin alternation,thanin thewordcondi-
usedto elicit productionin Spanishand English(viz. read-
tion, whereisolatedSpanishand Englishwordswere pro-
ing). However,Macnamara(1969) reportedthat, although
ducedin alternation.The differingsizeof the Spanishversus
readingspeedis a strongpredictorof relativelyproficiency
Englishdifferenceoccurredbecausethe VOT of English/t/
in two languages, the speedof switchingbetweenlanguages
decreased, whereasthe VOT of Spanish/t/increasedsome-
is not. Moreover, Macnamara et al. (1968) found no differ-
what, as the switchingrate increased from the sentence
to
encein switchingtimebetweentwo groupsof bilingualswho
the word condition.Given the oppositedirectionof the
apparentlydifferedin muchthe sameway as the earlyand
changesin VOT for theSpanish/t/and the English/t/, the
late learnersof the presentstudy.
changeswere unlikely to have been causedsimply by a
Finally, it is worth notingthat experiment2 examined
changein speakingrate.
codeswitchingratherthan codeborrotoing, definedby Gros-
The group• condition• languageinteractionwasnon-
jean and Soares(1986) asthe productionin a hostlanguage
significant.This mightbe takento meanthat the early and
of a word/phrasefroma donorlanguageusingthe "phonol-
late learnersswitchedbetweenEnglishand Spanishin the
ogy" of the host language.Their spectrographicdata sug-
sameway, but sucha conclusionwouldbe misleading.Even
gested,for example,that a nativeFrenchspeakerwho inserts
thoughVOT changedin much the sameway for the two
an Englishwordinto a Frenchsentence will saythe English
groups,andeventhoughthe samenominalprocedures were
word with Frenchacousticphoneticcharacteristics. The re-
followedfor bothgroups,therewasevidencethat earlyand
sultspresented heresuggest that the earlyand!atelearners
latelearnersdid not performthe switchingtaskin the same
may have codeswitchedin the samemanner, at leastin re-
way. The late learners'/t/-to-/t/intervals averaged312 ms
gardsto speechproduction.The possibilityexists,however,
longerthan the early learners'.
Previous research has shown that a measurable amount that the early and late learnerswould have been found to
differhad their productionof Englishwordsinsertedinto a
of time is neededto switchbetweentwo languages(Kolers,
1966). Maenamara et al. (1968) found that it took French-
Spanishconversation(or the reverse)beenexamined.
Englishbilinguals210 ms longerto rapidly name lists of
III. GENERAL DISCUSSION
digitsin Frenchand Englishthan to namedigitsin just one
language.However,the durationof the/t/-to-/t/intervals The resultsobtainedin experiments1 and 2 confirmeda
shouldprobablynot be regardedas a measureof the time resultobtainedfor late L2 learnersin manypreviousstudies
neededto switchbetweenlanguages,that is, to turn onelan- of L2 speechproduction.Native Spanishspeakerswho had
guagesystem"on" andtheother"off" ( Kolers,1966). This learnedEnglishas adults producedEnglish/t/with com-
is because,apart from the isolatedwords,the time from the promiseVOT valuesintermediateto the valuesobservedfor
onset of successiveutterancesand the (utterance-medial) Spanishand English monolinguals.The two experiments
/t/'s that weremeasured
mayhavevaried.8 showedthat nativespeakersof Spanishwho learnedEnglish
The differencebetweenthe early and late learners'/t/- asyoungchildren,on theother hand,fully differentiatedthe
to-/t/intervals impliesthat the late learnersprolongedcer- /t/of English and Spanish.Neither the early nor the late
tain sounds,or pausedto a greater extent, than the early learners'VOT valuesdifferedfrom Spanishmonolinguals'
learners. Some of the English soundsdiffered from any in the productionof Spanish/t/. The early learnerspro-
soundin the Spanishphoneticinventory.The Englishcar- ducedEnglish/t/with VOT valuesthat were significantly
rier sentence "Take another word such as "contained a longer---and therefore more English-like--than the late

404 J. Acoust.Soc.Am., VoL89, No. 1, January1991 JamesEmilFlege:VOT in a secondlanguage 404

learners'.
In fact,theearlylearnersdid notdifferfromEng- English/t/differ acoustically,they seemto sharecertain
lishmonolinguals in producing
English/t/. properties, suchasa lack of voicingimmediatelyfollowing
The presentstudyis apparentlythe first to havecom- stoprelease(Williams, 1977a), which causesbilingualsto
paredthe performanceof early and late learners.The early identifythem with oneanother.
learnerswereknownto havereceivedinputfromnativeEng- An importantissuefor L2 researchiswhethertheageat
lish speakerswhenthey first beganto learn English.Two which L2 learningcommences will affecthow much acous-
previousstudiesshowedthat earlylearnersproducedEng- tic phoneticinformationin L2 soundsis filteredout. Burn-
lish/p,t,k/with "compromise"VOT values(Caramazzaet ham (1986) suggestedthat certain phonemicallynonrele-
al., 1973;Flegeand Eelting, 1987a). The natureof the L2 vant acousticdimensionsare more easily perceivedby
phoneticinputreceived by theseearlylearnersisnotknown. listenersof all agesbecause theyare salientauditorily(see
The resultsof thepresentstudysuggest that theymayhave alsoBestet al., 1988). The acousticphoneticcontrastbe-
receivedaccentedL2 input. tween/p,t,k/in Spanishand Englishmightbe auditorily
How canweaccountforthedifference between theearly salientfor Spanishlearnersof English,sothat the acoustic
and late learnersin the productionof English/t/?. Three phoneticdifferences betweenSpanishandEnglish/t/can be
generaltypesof explanationare possible.(1) The early detectedreadily. The resultsobtainedin a foreignaccent
learnersmay have perceivedthe acousticphoneticcharac- mimicry experimentby Flegeand Hammond (1982) sug-
teristicsof Englishvoiceless stopsmoreaccuratelythan the gestedthat native speakersof Englishcan detect acoustic
late learners.This, in turn, might have preventedthe late differencesbetweenthe English/t/'s producedby native
learnersfrom producingEnglish/t/with native-likeVOT andSpanishspeakers of English(seealsoFlege,1984).VOT
values.(2) The early and late learnersmay haveperceived valueswere significantlylongerin/t/'s spokenin normal
English/t/in the samemanner,but the late learnersmay Englishutterances thanin utterances producedwith a mim-
havebeenrelativelylessablethan the early learnersto mo- ickedSpanishaccent.
torically output what they representedpereeptually.(3) The differencesbetweenphoneticand phonemicpro-
The early and late learnersmay havebeenequallyable to cessing suggeststhat the consciousperceptionof sound-
perceiveEnglish/t/accurately and to developmeansfor sized units occursprimarilyat the endof severalprocessing
producingvoiceless stopswith long-lagVOT values.How- stages. "Within-category" acousticphoneticdifferences be-
ever, the late learnersmay havebeenunableto fully utilize tween English/p,t,k/and Spanish/p,t,k/may normally go
their sensorimotor capabilitiesowingto the stateof develop- unnoticed at a conscious
level during the on-linecomprehen-
ment of their phoneticsystemwhen they beganlearning sionof spokenlanguage,but listenersmay be able to gain
Englishas an L2. Thesethreetypesof explanationwill be accessto phoneticinformation,or to exploit it in certain
considered in turn in the followingsections. auditory processingtasks.That is, even though Spanish
speakers may regardEnglish/p,t,k/as the "same"asSpan-
ish/p,t,k/at a phonemiclevel,they may treat the realiza-
A. Perception of L2 sounds tionsof thesephonemes asdifferentat a phoneticor an audi-
tory levelof processing.
As discussed in the Introduction,Spanishand English
Many late learnersexaminedin the presentstudy ap-
/t/differ in termsof glottal-supraglottal timingandplaceof
proximatedthe Englishphoneticnormfor/t/without actu-
linguapalatalconstriction.Cross-language differencesin
ally achievingit. The basisfor the apparentlimitation on
speechproductionoftencoincidewith differences in phonet-
how closelyL2 phoneticnorms were approximatedmay
ic perception.Experimentswith syntheticVOT continua,
havebeenperceptualin nature.Accordingto Flege( 1988a,
forexample,haveshownthatlongerVOT valuesareneeded
1990a;Flegeand Eefting, 1987al, "similar" soundssuchas
for nativeEnglishthan Spanishlistenersto judgestopsas
the/t/of Spanishand Englishwill be equatedat a phonetic
voiceless(Abramson and Lisker, 1973; Williams, 1977a;
categorylevelilL2 learningbeginsafterabouttheageof 5 or
Flegeand Eftting, 1986). This perceptualdifferencecorre-
6 years,sothata distinctperceptual representation for Eng-
sponds to thelongerVOT valuesobserved in theproduction
lish/t/will not be developed.A numberof "languageset"
of Englishthan Spanish/p,t,k/. Similarly,Elman et al.
experimentsareconsistentwith theview that latelearnersof
(1977) showedthat Spanishmonolinguals judgednatural
EnglishL2 do not establishdistinctperceptualrepresenta-
short-lagstopsas/p/, whereasnativeEnglishmonolinguals
tionsfor Englishvoiceless stops.Elmanet al. (1977) found
judgedthemas/b/.
that Spanish-Englishbilingualswho pronouncedEnglish
BohnandFlege(1990), on theotherhand,foundthat
with a foreignaccentdid not labelshort-lagstopdifferently
nativespeakers of Englishoftenjudgedshort-lagSpanish[ t ]
(i.e., as/p/vs/b/) whenthey wereprocessing the stopsin
tokensas/t/. Native speakers of Spanish---even thosewith
Spanishand English perceptual sets. Flege and Eelting
little previousexposureto English--consistently classified
(1987b) examined Dutch subjects'identificationof the
long-lagrealizationsof English/t/as voiceless(see also
membersofa VOT continuumrangingfrom/d/-/t/. Phon-
Munro, 1987;Yeni-Komshianet al., 1968). Thesefindings
emeboundariesobtainedin Dutch andEnglishlanguagesets
suggestthat VOT maybea lessimportantcueto word-initial
differed significantly,but the size of the phonemeboundary
stopvoicingcontraststhan is commonlysupposed. 9 More shifts were much smaller (3 ms) than the difference one
importantlyfor thecurrentdiscussion, it suggests that Span-
would expectbetweenDutch and Englishmonolinguals.
ish and English/p,t,k/are regardedasphonologically the There is someindirect evidence,on the other hand, that
samedespitedifferencesin VOT. Although Spanishand

405 J. Acoust.
Soc.Am..Vol.89, No.1, January1991 JamesEmilFlege:VOTina secondlanguage 405
earlylearnersmayestablish
a perceptual
representation
for /d/ distinction in a two-alternative forced-choice test.
English/t/. The identificationfunctionsobtainedfor early The resultsobtainedto date do not make it possibleto
learnersby Caramazzaet al. (1973) were nonmonotonic, choosebetweenthesetwo hypotheses. However,the finding
showingshiftsat two pointsalongtheVOT continuaexam- of the presentstudy that early but not late learnersfully
ined.A partialshiftoccurredat a pointalongthe continua differentiatedthe/t/'s of Spanishand Englishis morecon-
whereFrenchphoneme boundaries wouldbeexpected; com- sistentwith the hypothesis that early learnersdo establisha
pleteshiftswere observedat pointsnearerto the English perceptualphoneticrepresentation for English/t/.
phonemeboundary.Somesubjects in theElmanet al. study
showeda languageset effect.Thesewere the subjectswho B. Speech learning ability
spokeEnglishwith little accent,sotheymayhavebeenearly Even if one assumed that the late learners examined in
learners (DieM, 1988). the presentstudy did have perceptualrepresentations for
Otherevidence suggests,however, thatearlylearners do
English/t/, and thoserepresentations were as accurateas
notestablish a separateperceptualrepresentation for Eng-
thoseof the early learners,the late learnersmighthavedif-
lish/t/, or do so at the expenseof their previouslyestab-
fered from native speakersof English becausethey had
lishedperceptual representationfor Spanish/t/. The mean
passed a criticalperiodfor learningnewformsof pronuncia-
phoneme boundaries obtainedforearlylearners in a forced-
tion. Somehavesupposed(e.g., Sapon,1952) that pronun-
choicetestby Caramazzaet al. (1973) occurredat values ciationabilitydeclines withage.Lenneberg(1967) conclud-
intermediateto thoseobservedfor Frenchand Englishmon- edthata foreignaccentin anL2 is"inevitable"if it islearned
olinguals.
Someof theSpanish-English bilinguals
examined after pubertybecausebrain developmentand lateralization
byWilliams(1977b)hadphoneme boundaries
nearthoseof for languagefunction have reachedcompletionby that
Spanishmonolinguals; othershad boundariesnearer to time)øManyothers havealsohypothesized theexistenceof
thoseof Englishmonolinguais. a criticalperiodfor humanspeechlearningthatderivesfrom
Two experiments
usingsyntheticstimuliwith early brain maturation (Penfieldand Roberts, 1959;Lamendella,
learnersfailedto showsignificantlanguageseteffectswith 1977;Scovel,1988;but seeFlege, 1987b).
early learners(Caramazzaet al., 1974;Williams, 1977b). One might hypothesize,therefore,that the late learners
BohnandFlege(1990) foundthesame,smalllanguage set hadlessabilitythantheearlylearnersto motoricallyimple-
effectfor naturallyproducedshort-lagstopsfor bothearly menttheir perceptualrepresentations
for sounds.Thereis,
and late learners. Williams' (1979, 1980) research with
however,no direct evidencefor age-relatedatrophy or
Spanish-English bilingualchildrensuggested thatalthough changein thosecentersof thehumanbrainthatdirectspeech
theymayinitiallydivideVOT continualikeSpanishmono- movementsor regulateauditory processing.Moreover,the
linguals,theirphoneme boundaries will shifttowardEnglish neural maturationhypothesiscan be questionedon both
valuesastheygainexperience in English.Also,sensitivityto neurologicaland empiricalgrounds(seeSnow, 1987;Kins-
the (Spanish)distinction between stopswithleadandshort- bourne, 1981;Whitakeretal., 1981;Krashen, 1973). Even if
lag VOT in discrimination testsseemed to diminish,espe- this werenot so,an importantproblemexistsfor a critical
ciallythatof thechildrenwholearnedEnglishin earlychild- periodaccount.
hood.
A criticalperiodaccountprovidesno insightinto which
The resultsobtainedin previousspeechperceptionre- specificaspects of thephoneticlearningprocessmaychange
searchwith earlyand late learnerssuggest two hypotheses with age.One wonders,for example,why the late learners
concerning theperception of similarsounds in LI andL2. showedcompromise VOT values.Thereisnoa•oriorireason
The firstis that neitherearlynor late learnersestablishper- to think that it issomehoweasierfor latelearnersto produce
ceptualrepresentations for English/p,t,k/that aredistinct a partialmodificationof previouslyestablished articulatory
from representations establishedpreviouslyfor the LI patternsthanto producea complete modificationthat would
/p,t,k/. On thisview,the existingL1 categories eithergo enablethemto matchnativespeakers of English.In fact,the
unmodified,or elseevolveso as to reflectthe acousticprop- comparativerarity of stopswith VOT valuesin the "com-
ertiesof voicelessstopsin both L 1 and L2. promise"rangeobservedherefor late L2 learnersis prob-
The secondhypothesis is that the resultsobtainedin ablydisfavoredfor articulatoryor perceptualreasons.Lan-
previousexperiments employing a two-alternative forced- guagestendto haveeithershort-lagstopsor aspiratedstops
choicetask,especially thoseemploying synthetic stimuli,do like thoseof English.Few languages,
it seems,havestops
notprovideinsightintohowstopsarespecified at aphonetic with VOT valuesin betweenthesetwo "modal" categories
categorylevel. The existenceof phonetic categoryrepresen- (Lisker and Abramson, 1964). But this isjust what wasob-
tationsmaybe difficultto demonstratein a taskthat encour- served here for late L2 learners.
ageslistenersto access
thefinalstageof auditoryperceptual
processing,thatis,thephonemic stage.It hasbeensuggested C. Phonetic system depelopment
that the acousticphoneticcontrastbetween/t/and/d/in The speechlearningmodel (SLM) describedby Flege
EnglishismoresalientauditorilythantheSpanish phonetic (e.g., 1988a,1990a)canbe usedto accountfor why theearly
contrastbetween/t/and/d/(Williams, 1980). If so,then if and late learners differed. As alluded to earlier, the SLM
bilingualsubjectsdid have distinctperceptualrepresenta- positsthat auditory processingoccursat distinct auditory,
tions for short-lagand long-lagvoicelessstops,they might phonetic,and phonemiclevels(seealsoWerker and Logan,
havepreferredto usetheonethat underliesthe English/t/- 1985; Burnham, 1986). The auditory level makes no refer-

406 d. Acoust.Soc. Am., Vol. 89, No. 1, January1091 James Emil Flege:VOT in a secondlanguage 406
enceto meaning or phoneticfunction.At thephoneticlevel, succeeded in fully differentiating
Spanish/t/and English
classesof phonesare contrasted acoustically
by properties /t/because they usedtwo differentphoneticcategoriesto
thataresufficientlyrobustthattheymightbeusedto signal implementthephoneme/t/in Spanish andEnglish.Thelate
meaningcontrasts in somelanguage, but notnecessarily
the learnerswereunableto fully differentiate the/t/'s of Span-
listener'snativelanguage(e.g., the difference
betweenre- ishandEnglishbecause theydid not havea separatephonet-
leasedand unreleased word-finalstopsin English).At the ic categorywith whichto implement/t/in English.If the
mostabstract,phonemic,levelof processing, phonetically late learnersdid not possess a phoneticcategoryfor English
relevantclassesof phones("sounds")are groupedtogether /t/, howthendid mostof themproduceit with longerVOT
in functionalunitsthroughtherulesthatcomprisea learned, valuesthan Spanish/t/?
phonological system. The SLM positsthat, whenlate learnersidentifycorre-
When processed at a phonemiclevel,soundsthat may spondingL1 and L2 soundsin termsof a singlephonetic
be distinctauditorily are treatedas realizationsof a single category,but auditorilydetectacousticdifferencesbetween
category.Fodor (1983) proposedthat an importantcharac- them,theymay producethe LI andL2 soundsdifferentlyby
teristicof input systemsthat makeuseof highlyspecialized, applyingdifferentphoneticrealizationrules.Accordingto
encapsulated systemsis that the perceiverwill have con- Port and Mitleb (1983), realizationrulesdeterminethe "de-
sciousawareness onlyof representations that arederivedin tailsof speechtimingandcoordinatethe commands to the
the final stagesof processing. However,accordingto the speecharticulators"(p. 220). The notionof "realization
SLM, listenersremainableto access a phoneticlevelof rep- rule" is well establishedin the literature, but far more atten-
resentation,whichenablesthemto learnto distinguishnovel tion has beenpaid to temporalthan spatialaspectsof the
phoneticcontrasts. If humansdidnotpossess suchanability gestures usedto formspeechsounds. Thismaybedueto the
it wouldbeimpossible, for example,for studentsto learnto fact that speechtiming is often measuredmoreeasilythan
transcribe foreignsounds in phoneticclasses or learnto note spatialpropertiessuchasthe placeof tongue-palatecontact
differences betweenallophonesof a phonemein their L1. in stopconsonants.
The SLM positsthat speechproductionis organizedat Most investigatorshaveconsideredrealizationrulesto
phonemiccategory,phoneticcategory,and sensory-motor belanguagespecific(see,e.g.,Liberman,1970;Nooteboom,
levels.The phonemiccategoriesspecifiedin lexicalentries 1973;Klatt, 1976;Kent and Minifie, 1977;but cf. Stevens
are outputusinga finitenumberof universalphoneticcate- andHouse,1972)perhapsowingto thesmallbutsystematic
goriesthat, in turn, are motoricallyoutputusingphonetic timing differencesthat have beenobservedbetweencorre-
realization rules. The model builds on work by Keating sponding soundsin differentlanguages (e.g.,Lehiste,1970;
(1984), who describedhow phoneticand phonemiccatego- Ladefoged,1980;Port et al., 1980). Within a singlelan-
riesmightbeinterfaced.Keatingconcludedthat phonologi- guage,phoneticrealizationrulesare neededto accountfor
callyvoicedand voiceless stopsare implemented by oneof how speakerssystematically modifytheir productionof a
threeuniversalphoneticcategories, corresponding roughly phoneticcategoryasa functionof, for example,socialcon-
to Liskerand Abramsoh'sthreemodalVOT categories.So, text (e.g., Labov, 1981). It is uncertainwhetherrealization
for example,the voicedphoneme/d/may be implemented rulesare distinctfrom the parametermanipulations usedto
usingshort-lagor leadcategories, andthevoiceless phoneme effectchangesin speakingrate or emphasis.
/t/may beimplemented usinga short-lagor a long-lagpho- In fact,relativelylittle is knownconcerning the neural
neticcategory. controlmechanisms for phoneticrealizationrules.Lofqvist
One way in whichEnglishdiffersphonologically from andhiscolleagues haveexaminedin detailtheproduction of
Spanishin thatthelong-lagphoneticcategoryis usedto im- stopconsonants (Lofqvist, 1980;Lofqvistand Yoshioka,
plement/t/rather than the short-lagphoneticcategory. 1980, 1981; Lofqvist, 1980; Yoshiokaet al., 1981). This
Language-specific realizationrulesare usedto motorically bodyof work indicatesthat thestereotypiclaryngealgesture
outputphoneticcategories(Lieberman,1970).The rulesof usedto ensurean intervalof voicelessness in/p,t,k/is effect-
one language,when applied to a long-lag stop category, edby thecoordinated innerrationof intrinsiclaryngealmus-
might result in VOT that were slightly,but significantly clesthat rapidly abduct,then adduet,the vocal folds. In
longer,than the VOT valuesof anotherlanguage. agreementwith the earlier VOT research(Lisker and
The SLM positsthatafterphoneticcategories havebren Abramson, 1964, 1967; Abramson, 1977), Lofqvist con-
establishedfor LI soundsin early childhood,listenersare cludedthat it is principallythe timingof the laryngealde-
increasinglylikely to identify L2 soundsthat partially re- voicinggesturewith respectto supraglottal gestures which
semblecor.responding soundsin the L 1 (referredto as"simi- givesriseto a rich complexof acousticfeaturesthat includes
lar" sounds)as beingrealizationsof an L1 category.Late VOT.
learnerswill persistin identifyingsimilarL2 soundssuchas It is uncertainif the sizeof the glottalapertureor the
Spanishand English/t/, whereasearly learnerswill even- rate of opening-closingcan be regulatedvolitionallyby
tually notethe acousticphoneticdifferencesbetweenthem. talkers.It is alsouncertainwhetherthe timingof the devoic-
As a result,earlybut not latelearnerswill establish phonetic inggestures canbesoregulated.Shaimanetal. (1985) found
categories forsimilarL2 sounds, andearlybut notlatelearn- that. when the lip closinggesturesfor/p/were delayedby an
ers will producethem authentically (i.e., like native speak- unanticipatedperturbation,the laryngealderciting gestures
ers). were delayedproportionally.This suggests that the tempo-
On this account,the early learnersin the presentstudy ral coordinationneededto specifylanguage-specific VOT

407 J. Acoust.Soc. Am.. Vol. 89, No. 1, January 1991 James Emil Flege: VOT in a second language 407
valuesmay derivefrom rapid sensorimotor reflexinterac- In experiment2,thebilingualsubjects
wererequiredto
tionsbetweenlaryngealmusclessuchas the PCA and mus- switchwithincreasing rapiditybetweenSpanishandEng-
clesusedin formingsupraglottalconstrictions.
The authors lish.Undersufficienttimepressure,thelatelearners
might
concluded that, althoughpatternsof laryngeal-supralaryn- havebeenexpected toabandon thelater-acquired
(English)
gealcoordination (and,ultimately,VOT) mayrepresent a realization
ruleinfavoroftheearly-acquired(Spanish)
real-
predeterminedphonetic"goal," the timing pattern itself izationrule(seeKewley-Port andPreston, 1974).Experi-
maynot be"explicitlyprogrammedbut implemented down- ment2 did not,however,providesupportfor thedistinction
stream,by sensorimotor actions"(p. 185). between phonetic implementation
andrealization.Themag-
nitudeof theSpanish versusEnglishVOT difference didde-
D. Critique of the three-level model
creaseslightlyastherateofswitchingincreased,butthesize
Speechproductionhas historicallybeenviewedas a of the decrease wasnot significantlygreaterfor the late
two-stageprocessin which abstractsound units (pho- learnersthanfor theearlylearners.
neroes)arefirstselected at higherlevelsof a message-gener- Thisfindingdid not providecounterevidence against
atingsystem,then relatedto a lowerlevelsystemfor trans- the implementation-realization distinction,however.The
formationinto a codesuitablefor the generationandcontrol intervalbetween successive EnglishandSpanish/t/'swas
of artieulatorymovement(e.g., Perkell, 1980;MacNeilage, over300mslongerfor thelatethantheearlylearners. It is
1980;MaeNeilageet al., 1981;Kent and McNeil, 1987;but conceivablethat this extra time was due to the late learners'
seeLieherman,1970). The speechproductionmodel pro- needto selectbetweencompetingrealizationrules.Addi-
posedto accountfor differences betweenearlyandlatelearn- tionalresearchisneeded to testthishypothesis, forit isalso
ers differsfrom previousmodelsin that it proposesthree possible thattheaddedtimeneeded bythelateleamenwas
distinctlevelsof organization(Flege, 1988a,1990a). That duetoa needforaccessing relativelyunfamiliar L2 phonetic
is, it distinguishes
betweena levelof representation at which categories.
"universal"characteristics of phoneticsegments are speci- Finally,the presentstudydid not providedirectevi-
fiedand a levelat whichfine-grained, language-specificde- dencethatthelatelearners did nothaoedistinctcategories
tail isprovided.The formerisdesignated thelevelofphonetic for the/t/'s of Spanish andEnglish.As notedearlier,one
implementation, thelatterasthelevelofphoneticrealization. mightargue thatmostofthelatelearners produced English
At theimplementation level,a phonemesuchasEnglish /t/with onlyslightlylongerVOT valuesin Englishthan
/t/would berepresented ashavingtongue-tipconstriction Spanish--rather thanthesubstantially longerVOT values
and a patternof laryngealtiming that resultsin long-lag seen fortheearlylearners•beeause theirphonetic catego-
VOT values.The realizationlevelwouldspecify,amongoth- riesfor English/t/wereinaccurate. Or, onemightargue
er things,a constriction of thetonguetip anddorsumagainst thattheywereless skillfulthanearlylearners inimplement-
thealveolarridgeanda laryngealtimingpatternthatresults ingEnglish/t/asa long-lag stop.
in VOT of approximately 65 msratherthan,say,45 ms.This
approachis consistentwith thebeliefin the existenceof uni- E. Summary and conclusions
versalsoundtypesthat are modifiedthroughlanguage-spe- The presentstudyprovidedevidencethat Spanish-Eng-
cificlearning(Chomskyand Halle, 1968). lish bilingualscan fully differentiateSpanishand English
One problemfor this approachis that it restson an un- /t/, at leastin termsof VOT, if theylearnEnglishasan L2 in
testedassumption.With respectto the data of the present early childhoodbut not if they begin learning English as
study, it is assumedthat large VOT differencesbetween adults.A reviewof theliteratureprovidednosupportfor the
Spanish [t] andEnglish[th] phones canonlyarisethrough viewthat latelearnersareinherentlylesscapableof learning
the implementationof/t/using differentphoneticcatego- newformsof pronunciation.The literaturereviewsuggested
ries. It is further assumedthat relativelysmall VOT differ- that early learnersmay be more apt than late learnersto
enees, such as the difference observed in the late learners' developa centralperceptualrepresentation for the long-lag
productionof Spanishand Englishstops,will arisethrough stopsusedto implementEnglish/t/, althoughthis conclu-
the applicationof differentrealizationrulesto a singlepho- sion is by no meanscertainand no perceptiondata were
neticcategory. providedfor the subjectsin the presentstudy.
Experiment2 affordedthe opportunityfor testingthe The VOT differencebetweenearly and late learnersfor
distinctionbetweenphoneticimplementationand realiza- English/t/was interpretedto reflecta difference in phonetic
tion.The needto choosebetweencompetingstructuresadds organization.Specifically,it was claimed that the early
a finite amount of processing time in motor tasks (e.g., learners fully differentiatedEnglish/t/from •pani•h/t/
Sternberg, 1969). If one assumesthat lexical items are becausethey, unlikethe late learners,had distinctphonetic
markedfor languageidentity (Macnamara, 1969;Neufeld, categories for the two/t/'s. The late learnerswerehypoth-
1976), andthat phoneticimplementation processesare rea- esizedto producetheir relativelysmallSpanishversusEng-
diedwhena wordislexicallyaccessed for production(Flege, lishVOT differenceby usingtwo differentphoneticrealiza-
1990c), then one might expectlate learnersto take slightly tion rules to output a single phonetic category. This
longer to motoricallyoutput L2 words containingsimilar interpretationofthe data presentedheremustbe considered
soundsthan early learners.This is becauselate but not the tentative.No direct evidencewas providedfor the distinc-
earlylearnerswouldneedto choosebetweentwocompleting tion drawn betweenphonetic implementationand realiza-
realizationrulesto motoricallyoutputSpanishand English tion. Moreover,althoughit appearedthat the early learners
words with/t/. were better able to preventthe L1 phoneticsystemfrom
408 d. Acoust.Soc. Am., Vol. 89, No. 1, January1991 James EmilFlege:VOT in a secondlanguage 408
influencingtheir L2 speechproduction,factorsthat may in- haturally
between
theSpanish
andEnglish
portions
oftheexperiment,
as
flueneethedegreeof independence of L 1andL2 systems are requiredby theprotocol.
(q•hestandarddeviations
associated
with utterance-medial
stopswere
poorlyunderstood and thereforemay not havebeencon- slightlygreaterthanthoseassociated
withutterance-initial
stops( 10.1vs
trolledfor adequately in the presentstudy(Obler and Al- 7.9), perhapsbecausethe utterance-medicistopsoccurredin eightdiffer-
bert, 1978;Grosjean,1985). ent wordswhereastheutterance-initial stopsoccurredin a singleword
(viz. take). The three-wayinteractionwasprobablysignificantbecause
However,if theinterpretationofferedhereis supported
the vowelX utterancepositioninteractionwassignificant for the early
by additionalresearch, it wouldsuggest two importantcon- learners[ F( 1,9) = 15.l, p < 0.05] butnotfor theEnglishmonolinguals or
clusionsabout the nature of bilingualism:(1) Early L2 the late learners.Further analysisrevealedthat, for the earlylearners,s.d.
leamenmayhaveanenriched phoneticsystem thatincludes valuesweregreaterfor utterance-medial /t/ tokenswhichpreceded/i/
than/e/( 13.5vs7.8}. It cameasnosurprisethat thevowelcontexteffect
allphonetic
categories possessedbynativespeakers
oftheL 1 wasnonsignificant Jbrutterance-initial/t/(9.3 vs 8.6), for thesestops
andL2; and (2) a bilingual'stwo languages
arerepresented were more distant from the differential vowel context in the test words at
by a uniquesystemthat doesnot represent the sumof the theendof thecarrierphrase.
competences oftwomonolingual speakers.
Thissecondcon- ?Therewasnovowelcontexteffectforthetwosetsof utterance-initial
stops
that weremeasured.The VOT of the utterance-initialSpanishstopsin
clusion is basedon the inference that the late learnersdevel-
tengohad the samemeanVOT value ( 19 ms) in utterances endingin
opeda phonetic realization
rulethatneitherSpanishmono- wordswith/i/and/e/. This wasexpectedbecause the differentialvowel
lingualsnorEnglishmonolinguals possess,
namely,oneused contextwas too distantto exert an effecton the utterance-initialstops.
to slightlyincrease
VOT in Englishasopposed to Spanish Similarly,the Englishstopsin takehad aboutthe samemeanVOT values
in the utterances
endingin testwordswith/i/and/œ/(45 vs47 ms).
It/.
SAmoredirectmeasare
of actuallanguage-to-language"switching
time"
mighthavebeena measureof utteranceonsetto utteranceonset,This
ACKNOWLEDGMENTS intervalwasnotmeasured, however,owingto theproblemof howto inter-
pret inter- and intra-utterancepauses.
This researchwas supportedby NIDCD Grant 9Othercuessuchasburstandaspiration
intensity,
andF• onset
frequency
DC00257. Thanks are extended to R. Diehi and P. MacNei- andtransitionduratton,areknownto playa rolein theperception of stop
lagefor helplocatingsubjects in Austin;L. CuevaandC. voicing( see,e.g.,Stevens
andKlatt, 1974;Lisketeta!., 1977). Forrestand
Rockman(1988) showedrecentlythatEnglishadultsmaycorrectlyiden-
Menafor helpgatheringdata;G. Allen, M. Munro,andtwo tify at leastsomeof misarticulatingchildren'sshort-lagrealizations of
anonymous reviewersfor comments on a previousdraftof English/p,t,k/asvoiceless.Conversely, somechildrenperceived to neu-
thisarticle;and L. Skeltonfor helpwith all phasesof the tralize the distinctionbetweeninitial voicedand voiceless
stopsmay nev-
research. erthelessproducea significantVOT contrastbetweenthem (see,e.g.,
Carts and Jensen, 1983).
•Fhereisnothinginherently
iraplausible
w•tha criticalperiodhypothesis.
Forexample,Nottebohm (1989)showed thattheseasonal patternofsong
'Suomi(1980) foundthatFinnishhighschoolstudents
oftenproduced learning,loss,andrelearningin themalecanaryis relatedto changesin
English/p,t,k/with Finnish-likeshort-lagVOT values.Schmidt(1988) thesizeof two neuralsongcontrolcentersin thecanaries'
forebrain(see
examinedthe Englishspokenby threenativeSpanishL2 learnerswith a also Nottebohm, 1969).
demonstrated "superior"pronunciationof Englishwho had begunlearn-
ingEnglishbetweentheagesof 10-13years.Thesesubjects producedEng- Abramson,A. (1977l. "Laryngeal timing in consonantdistinctions,"
lish/p/with shorterVOT valuesthana groupof nativeEnglishspeakers Phoneflea 34, 295-303.
(33, 40, and 51 vs 72 ms). Flegeand Eefting (1987b) foundthat most Abramson,A., andLisker,L. (1973). "Voicetimingperceptionin Spanish
DutchadultswhobeganlearningEnglishat theageof 12yearstendedto word-initialstops,"I. Phon.1, 1-8.
produceEnglish/t/with short-lagVOT valuesif theyhadlittle English- AItenberg,E., and C:firns,H. (1983). "The effectsof phonotactic
con-
language experience.
Thosewhowereexperienced tendedto produceEng- straintson lexicalprocessing
in bilingualandmonolingualsubjects,"
J.
lish/t/with compromise VOT values.Williams( 1979,1980) foundthat Verb. Learn. Verb. [Sehav.22, 174-188.
evennativeSpanishchildrenwhobeganlearningEnglishbetweentheages Arsenian,S. (1937). "Bilingualismand Mental Development:A studyof
of 8-10 yearsproducedEnglish/p/with shorterVOT valuesthannative theintelligence
andthe socialbackground of bilingualchildrenin New
Englishchildren. York City," TeachersCollege,ColumbiaUniv.
2Mack(1990)observed
"overshooting"
oftheEnglishphonetic
normbya Asher,J., andGarcia,R. (1969). "Theoptimalageto learna foreignlan-
10yearold whohadlearnedbothFrenchandEnglishin earlychildhood. guage,"Mod. Lang.J. 53, 334-341.
The child producedEnglish/p,t,k/with longerVOT valuesthan native Best,C., Mc Roberts,G., andSithole,N. (1988). "Examinationof percep-
Englishchildrenof thesameage.He producedEnglish/p,t,k/with longer tual reorganizationfornonnative speech contrasts:
Zuluclickdiscrimi-
VOT valuesthan French/p,t,k/which, in turn, had longerVOT values nationby English-speaking adultsandinfants,"JEP:HPP14,345-360.
than French/b,d,g/, which were unexpectedlyproducedwith English- Blair,D., andHarris,R. (1981). "A testof interlingual
interaction
in com-
likeshort-lagVOT values.It appears
thatthechildwasattemptingto dif- prehension by bilinguals,"
J. Psycholing. Res.10,457-467.
ferentiate
LI andL2 stopsthatdifferedphonetically
butnotphonological- Bohn,O.-S.,andFlege,J. (1990). "Perceptual switchingin Spanish/Eng-
ly. lishbilinguals:
Evidence for universal
factorsin voicing judgments,"J.
•Arsenian( 1937;citedby Macnamara,1969)estimated thatthereliability Phon. (submitted).
of subjects'self-reportsconcerninglanguageusagewasof the order of Burnham,D. (1986). "Developmentallossof speechperception:
Exposure
r = 0.800 or better. toandexperience
w•tha firstlanguage,"
Appl.Psycholing.
7, 207-240.
4TheSpanish/e/phoneme
issymbolized
hereas/r/becauseit isusually Caramazza,A., Yeni-Komshian,G., Zurif, E., and Carbone,E. (1973).
realizedwith an [ • ] variantin closedsyllables(Dalbor, 1980}. The vowel "Theacquisition
ofa newphnnological contrast:
Thecaseofstopconson-
in temamayactuallyhavebeenan [el-quality vowelbecause it occurred antsin French-English
bilinguals,"J. Acoust.Soc.Am. 54, 421-428.
in an opensyllable. Caramazza, A., Yeni-Komshian, G., and Zurif, E. (1974). "Bilingual
STheresearch
assistant
whoeliciteddatafromthenativeEnglishandSpan- switching: The phonologieallevel,"Can.J. Psychol.28, 310-318.
ishsubjects
in Austinwasa nativespeakerof Spanishwho had learned Carts,H., andJensen,P. (1983). "Speechtimingof phonologicallydisor-
Englishat aboutthe ageof 5 years.Sheseemedto the author to speak deredchildren:Voicingcontrastof initial and finalstopconsonants,"
J.
English without an accent, and reported that this was also true of her SpeechHear. Res. 26, 501-510.
Spanish.The researchassistantwho eliciteddatafrom sixlate learnersin Chomsky,N., andHalle, M. (1968). TheSoundPatternofEnglish(Harper
Birmingham,Alabamawasa Honduranwho hadbegunlearningEnglish & Row, New York)
at abouttheageof 12yearsin a bilingualschool.ShespokeEnglishwith a Dalbor, J. (1980). SpanishPronunciatton
(Holt, Rinehart,and Winston,
slightaccentin the author'sopinion.Bothassistantsswitchedeasilyand New York).

409 J.Acou'st.
Soc.Am.,Vol.89,No.1,January
1991 James
EmilFlege:
VOTinasecond
language 409
Diehi, R. (1988). Personalcommunication. Kinsboume,
M. (1981)."Neuropsychological
aspects
of bilingualism,"
in
Elman, J., Diehi, R., and Buchwald,S. (1977)."Perceptualswitchingin NatioeLanguageand ForeignLanguageAcquisition,
editedby H. Winitz
bilinguals,"J. Acoust.Soc.Am. 62, 971-974. (Annalsof the New York Academyof Sciences,
New York), Vol. 379,
Ferguson,C. (1986). "Discoveringsoundunitsand constructing sound pp. 50-58.
systems: It's child play," in Inoarianceand Variabilityin SpeechPro- Klatt,D. (1976)."Linguistic
uses
ofsegmental
durationin English:
acous-
cesses,
editedby J. Perkeliand D. K!att (Earlbaum,Hillsdale,NJ), pp. tic and perceptualevidence,"J. Acoust.SOc.Am. 59, 1208-1221.
36-57. Kolers,P. (1965}. "Bilingualism
and bicodalism,"
Lang. Speech8,
Fernandez,M., andMolinet-Molina,M. (1988). "El espafiolde losnifios 122-126.
pequenosenMiami," Paperpresentedat the9th Int. Conf.onSpanish in Kolers,P. (1966). "Readingandtalkingbilingually,"Am. J. Psychol.79,
the U.S., Miami, October 1988. 357-376.
Flege,J. (1984). "The detectionof Frenchaccentby American listeners," Krashen,
S. (1973)."Lateralization,
language
learning,
andthecriticalpe-
J. Aconst. SOc.Am. 76, 692-707. riod:Somenewevidence,"[ang. Learn.23. 63-74.
Flege,J. (1987a). "The productionof'new' and 'similar'phonesin a for- Labor, W. (1981). "Resolvingthe Neogrammarian
controversy,"
Lan-
eign language:Evidencefor the effectof equivalenceclassification,"
J. guage87, 267-309.
Phon. 15, 47-65. Ladefoged,
P. (1980). "Whatarelinguistic
sounds
madeof?," Language
Flege,J. (19871})."A criticalperiodfor learningto pronounceforeignlan- 56, 485-502.
guages7,"Appl. Ling. 8, 162-177. Lambert,W., and Rawlings,C. (1969}. "Bilingualprocessing
of mixed-
Flege,J. (1988a). "The productionand pemeptionof speechsoundsin a languageassociative
networks,"J. Verb. Learn.Verb. Behar.8, 604-609.
foreignlanguage,"in Human Communication and Its Disorders,A Re- Lamendella,J. (1977). "Generalprinciples
of neurofunctional
organiza-
uiew1988,editedby H. Winitz (Ablex, Norwood,NJ), pp. 224-401. tionandtheirmanifestation
inprimaryandnon-primary languageacqui-
Flege,J. (1988b). "Factorsaffectingdegreeof perceived foreignaccentin sition," Lang. Learn. 27, i 55-196.
Englishsentences," J. Acoust.Soc.Am. 84, 70-79. Lehiste,I. (1970}. Suprasegmentals
(MIT, Cambridge,MA).
Flege,J. (1990a}. "Ih½ Intelligibilityof Englishvowelsspokenby British Lenneberg,
E. (1967). Biological
Foundations
of Language
(Wiley,New
andDutch talkers,"in Intelligibilityin SpeechDisorders: Theory,Mea- York).
surement,and Management,edited by R. Kent (Benjamins,Amster- Liberman,A. (1970). "The grammarsof speechand language,"Cognit.
dam) (in press). Psychol.1, 301-323.
Flege,J. (1990b). "At what ageof learningis a foreignaccentfirst evi- Lieberman,
P. (1970). "Towardsa unifiedphonetictheory,"Ling.Inquiry
dent.9,"submitted to J. Acoust. Soc. Am. 3, 307-321.
Flege, J. (1990e). "The productionof cognateEnglish words by native Lisker,L., andAbramson,A. (1964). "A cross-language
studyof voicingin
speakers of Spanish:More evidencefor the distinctionbetweenphonetic initialstops:Acousticalmeasurements,"Word 20, 384-423.
implementation and realization,"submittedto J. Phon. Lisker, L., and Abramson, A. (1967). "Some effectsof contexl on voice
Flege,J., andEelting,W. (1986)."Linguisticanddevelopmental
effectson onsettime," Lang.Speech10, 1-28.
the productionand perceptionof stop consonants," Phonetica43, Lisker, L., Liberman, A., Erickson, D., Dechovitz, D., and Mandiet, R.
155-171. (1977). "On pushingthe voice-onset-time (VOT) boundaryabout,"
Flege,J., andEelting,W. (1987a). "The productionandperception of Eng- Lang.Speech20, 209-216.
lish stopsby Spanishspeakers of English,"J. Phon. 15, 67-83. Lofqvist,A. (1980}. "Interarticulatorprogramming
in stopproduction,"J.
Flege,J., and Eefting,W. (1987b) "Cross-language switchingin stopcon- Phon. g, 475-490.
sonantproduction andperception byDutchspeakers ofEnglish,"Speech Lofqvist,A., and Yoshioka,H. (1980). "LaryngeaiactivitySwedishob-
Commun. 6, 185-202. struent clusters," J. Acoust. Soc. Am. 68, 792-801.
Flege,J., andEefting,W. (1988). "Imitationofa VOT continuumby native Lofqvist,A., andYoshioka,H. (1981). "Interarticulatorprogramming
in
speakers of Englishand Spanish:Evidencefor phoneticcategoryforma- obstruentproduction," Phonetica38, 21-34.
tion," J. Acoust. Soc. Am. 83, 729-740. Lowie,W. (1988). "Age and foreignlanguagepronunciationin the class-
Flege,J., and Hammond,R. (1982). "Mimicry of non-distinctive
phonetic room,"unpublishedthesis,Universityof Amsterdam.
differencesbetweenlanguagevarieties,"Stud.Sec.Lung.Acquis.8, i- 18. Nfack,M. (1986)."A studyofsemantic
andsyntactic
processing
inmono-
Flege,J.,andHillenbrand,J. (1984). "Limitsonpronunciation
accuracy in lingualsand fluentearlybilinguals,"J. Psycholing.Res.15, 463--488.
adult foreignlanguagespeechproduction,"J. Acoust.Soc. Am. 76, Mack,M. (1989). "Consonant andvowelperception andproduction: Early
708-72 I. English-French bilingualsand Englishmonolinguais," Percept.Psy-
Flege,J.,andPort,R. (1981). "Cross-language
phonetic
interference:
Ara- ohophys.46, 187-200.
bicto English,"Lang.Speech24, 125-146. Mack,M. ( 1990}."Phonetic transferin a French-Englishbilingualchild."
Fodor, J. ( 1983). The Modularity of Mind (MII, Cambridge,MA). in Languages,4ttitudesand LanguageConflict, edited by P. Nelde
Fokes,J., Bond,Z., andSteinberg,M. (1985). "Acquisitionof the English (Dummler, Bonn), pp. 107-124.
voicingcontrastby Arab children,"Lang.Speech28, 81-92. Macnamara,J. (1969). "How canonemeasuretheextentof a person's bi-
Forrest,K., and Rockman,B. (1988). "Acousticand perceptualanalysisof lingualproficiency?,"in Description
and Measurement of Bilingualism,
word-initialstopconsonants in phonologically disorderedchildren,"J. edited by L. Kelley (University of Toronto, Toronto, Canada}, pp.
SpeechHear. Res. 31, 449-459. 80-98.
Grosjean,F. (1982). Life toithTwo Languages, An Introductionto Bilin- Maenamara,J., Krauthammer,M., and Bolgar,M. (1968). "Language
gualism(CambridgeU. P., Cambridge). switchingin bilingualsasa functionof stimulusand response
uncertain-
Grosjean,F. (1985). "The bilingualis a competentbut specificspeaker- ty," J. Exp. Psychol.78, 208-215.
hearer,"I. Multiling. MulticulturalDer. 6, 467-477. MacNeilage,P. (1980}. "Speechproduction,"Lang.Speech23, 3-23.
Grosjean,F. (1989}. "Nenrolinguists,Beware!The bilingualis not two MacNeilage,P., Hutchinson,J., andLasater,S. (1981). "The productionof
monolinguals in oneperson,"BrainLung.36, 3-15. speech:Developmentand dissolution of motoricand premotoricpro-
Grosjean,F., and Soares,C. (1986). "Processing mixedlanguages:
Some cesses,"
inAttentionandPerformance, IX, editedby$. LongandA. Bad-
preliminaryfindings,"in LanguageProcessing
in Bilinguals:Psycholœn- deley (Earlbaum, Hillsdale, NJ), pp. 503-519.
guisticandNeurophysiological
Perspectioe&
editedby J. Vaid (Erlbaum, Magiate,E. (1979). "The competinglanguagesystemsof the multilingual:
Hillsdale, NJ). A developmental studyof decodingand encodingprocesses," J. Verb.
Keating,P. (1984)."Phoneticandphonological representationof stopcon- Learn. Verb. Behav. 18, 79-89.
sonantvoicing,"Language60, 286-319. Major, R. (1987). "Englishvoiceless
stopproductionbyspeakers of Brazi-
Kent, R., and McNeil, M. (1987). "Relativetimingof sentencerepetition lian Portuguese," J. Phon.15, 197-202.
in apraxisof speech andconduction aphasia,"in Phonetic Approaches to Morosan,D., and Jamieson,D. (1989). "Evaluationof a techniquefor
SpeechProductionin Aphasiaand RelatedDisorders, editedby J. Ryalls training new speechcontrasts:Generalizationacrossvoices,but not
(CollegeHill, SanDiego,CA), pp. 181-220. word-positionor task,"J. SpeechHear. Res.32, 501-511.
Kent, R., and Minifie, F. (1977). "Coarticulationin recentspeechproduc- Munro, M. (1987). "Voicing Contrastsin French and EnglishBilabial
tion models,"J. Phon. 5, 115-134. Stops,"unpubl.M.Se. thesis,Univ. of Alberta, Edmonton,Alberta.
Kewley-Port,D., andPreston,M. (1974). "Early apicalstopproduction:
A Nathan, G. (1987). "On second-language acquisitionof voicedstops,"J.
voiceonsettime analysis,"J. Phon. 2, 195-210. Phon. 15, 313-322.

410 J. Acoust.Soc. Am., Vol. 89, No. 1, January 1991 James Emil Flege: VOT in a second language 410
Neufeld,G. (1976). "The bilingual'slexicalstore,"Int. Rev. Appl. Lin- acquisition,"in Sensitioe
PeriodsinDeeelopment. editedbyM. Bornstein
guist.14, 15-35. (Erlbaum,Hillsdale,NJ), pp. 183-210.
Nooteboom, S. (1973). "Perceptual
realityofprosodic durations,"J. Phon. Sternberg,S. (1969). "Memory-scanning: Mentalprocessesrevealed
byre-
1, 25-45. actiontimeexperiments," Am. Sci.57, 421-457.
Nottebohm,F. (1969). "The 'criticalperiod'for songlearning,"Ibis 111, Stevens,K., and Klatt, D. (1974). "The role of formant transitions in the
386-387. voiced-voiceless
distinctionfor stops,"J. Acoust.Soc.Am. 55, 653-659.
Nottebohm,F. ( 1989). "From birdsongto neurogenesis,"
Sei.Am., Febru- Stevens,
K., andHome,A. (1972)."Speechperception," in Foundationsof
ary 1989,74-79. Modernduditoty Theory,Vol.2, editedby J. Tobias(Academic,New
Obler,L., andAlbert,M. (1978). "A monitorsystemfor bilinguallanguage York ).
processing,"in Aspectsof Bilingualism,editedby M. Paradis(Horn- Suomi,K. (1980). "Voicingin EnglishandFinnishstops,"Publ. 10of the
beam,Columbia,SC), pp. 156-164. Dept. of EnglishandGeneralLinguistics,Turku Univ., Finland.
Oyama,S. (1976). "The sensitive periodfortheacquisitionof a nonnative Tahta,S.,Wood,M., andLowenthal,K. (1981). "Foreignaccents: Factors
phonoiogicalsystem,"J. of Psycholing.Res.5, 261-285. relatingto transferof accentfrom the first languageto the secondlan-
Penfield,W., and Roberts, L. (1959). Speechand Brain Mechanisms guage,"Long.Speech24, 265-272.
(Princeton U.P., Princeton, NJ). Thompson,I. (1984). "Experimentalstudyof foreignaccents,"unpubl.
Perkell,1. (1980). "Phoneticfeaturesandthephysiologyof speechproduc- Ph.D. thesis,GeorgeWashington Univ., Washington, DC.
tion," in LanguageProduction,Vol. 1, editedby B. Butterworth(Aca- Trubetzkoy,N. (1939). Grundzugeder Phonologie (Universityof C.alifor-
demic,New York), pp. 337-372. nia, Berkeley);Englishtranslationby C. Baltaxe,1969.
Port,R., A!-ani,S.,andMaeda,S. (1981})."Temporalcompensation and Weinreich,LI. (1953). Languagesin Contact,Findingsand Problems
universalphonetics," Phonetica37, 235-252. (Mouton,The Hague).
Port, R., and Mitleb, F. (1980). "Phoneticand phonologicalmanifesta- Weisruer,G. (1979). "Sensitivityof voice-onset time (VOT} measures to
tionsof thevoicingcontrastin Arabic-accented English,"Res.Phon.1, certainsegmental featuresin speechproduction,"J. Phon.7, 197-204.
137-165(Dept. of Linguistics, IndianUniversity). Werker,J., and Logan,J. (1985). "Cross-language evidencefor threefac-
Port, R., and Mitleb, F. (1983). "Segmentalfeaturesand implementation torsin speechperception,"Percept.Psychophys. 37, 35-44.
in acquisition
of Englishby Arabicspeakers,"
J. Phon,11,219-231. Whitaker,H., Bub,D., andLeventer,S. (1981). "Neurolinguistic aspects
Port, R., and Rotunno,R. ( 1979)."Relationbetweenvoice-onsettime and of languageacquisilionand bilingualism,"in flatJoeand ForeignLan-
vowel duration," J. Acoust. Soc. Am. 66, 654-662. guagedcquisition, editedby H. Winitz (New York Academyof Sciences,
Purcell,E., andSuter,R. (1980). "Predictors
of pronunciationaccuracy:
A New York), pp. 59-74.
reexamination," Long.Learn.30, 271-287. Williams, L. (1977a). "The voicingcontrastin Spanish,"J. Phon. 5,
Sapon,S. (1982). "An application ofpsychologicaltheorytopronunciation 169-184.
problemsin secondlanguagelearning,"MOd.Long.1.36, ! I 1-114. Williams,L. { 1977b}."The perceptionof stopconsonant voicingby Span-
Schmidt,A. (1988). "The acquisitionof someAmericanEnglishduration ish-Englishbilinguals,"Percept.Psychophys. 21, 289-297.
parameters by normatirespeakers of English,"unpubl.Ph.D. disserta- Williams,L. (1979). "The modificationof speechperceptionand produc-
tion, Univ. of Florida. tion in second-languagelearning,"Percept.Psychophys. 26, 95-104.
Scovel,T. ( 1988)..4 Timeto$pealc-
d Psycholinguistic
InquiryintotheCriti- Williams,L. (1980). "Phoneticvariationasa functionof second-language
cal Periodfor HumanSpeech(Harper& Row, New York). learning,"in ChildPhonology, Vol.2 Perception,
editedbyG. Yeni-Kom-
Seliger,H., Krashen,S., and Ladefoged,P. (1975). "Maturationalcon- shian, J. Kavanagh,and C. FergusonIAcademic, New York), pp.
straintsin the acquisitionof secondlanguageaccent,"Lang. Sci. 36, 185-216.
20-22. Yeni-Komshian,G., Preston,M., and Kewley-Port,D. (1968). "A cross-
Sclinker,L. (1972). "Interlanguage,"
Int. Rev.Appl. Ling. 10, 209-23I. languagestudy of the perceptionof stop consonantsin American and
Shaiman,S., Abbs,J., and Gracco, V. (1985). "Sensorimotorcontributions Lebanese children,"Neurocom•nun. Lab.,JohnsHopkinsUniv. 3, 1-15.
to oral-laryngeal
coordination
for speech,"
Soc.for Neurosei.Abst.11, Yoshioka,H., Lofqvist,A., andHirose(1981). "Laryngealadjustments in
76. the productionof consonantclustersand geminatesin AmericanEng-
Snow,C. (1987). "Rdevanceof the notionof a criticalperiodto language lish," 1. Acoust. Soc. Am. 70, 1615-1632.

411 J. Acoust.Soc.Am.,Vol.89, No.1, January1991 JamesEmilFlege:VOT !na secondlanguage 411