The acquisition of a new phonologicalcontrast: The case of stop

consonantsin French-English bilinguals

A. Caramazza
Departmentof Psychology.The Johns Hopkins University,Baltimore. Maryland 21218

, G.H. Yeni-Komshian
Divisionof Laryngologyand Otolog7, The JohnsHopkins UniversitySchoolof Medicine.Baltimore,Maryland 21205

E. B. Zurif
AphasiaResearchCenter,Boston UniversitySchoolof Medicine,and Boston V. A. Hospital, Boston,Massachusetts

E. Carbone

Departmet?t
of Psychology,
Sir George
Williams
University,
Montreal,
Quebec,
Canada
(Received13 February1973; 28 March 1973)

Cross-language studieshave shownthat Voice OnsetTime (VOT) is a sufficientcue to separateinitial stop

consonants into phonemiccategories.The presentstudyusedVOT asa linguisticcuein examiningthe percep-
tion and productionof stop consonants in three groupsof subjects:unilingualCanadianFrench,unilingual
Canadian English,and bilingual French-Englishspeakers.Perceptionwas studied by having subjectslabel
syntheticallyproducedstop-vowelsyllableswhile productionwas as•ssed throughspectrographic measure-
menCsof VOT in word-initialstops.Six stopconsonants, commonto both languages,were usedfor thesetasks.
On the perceptiontask,the two groupsof unilingualsubjectsshoweddifferentperceptualcro•soverswith those
of the bilingualsoccupying an intermediate position. The production data indicate that VOT measurescan
separatevoicingcontrastsfor speakersof CanadianEnglish,but not for speakersof CanadianFrench.The data
also show that languageswitchingin bilingualsis well controlled for productionbut poorly controlledfor
perceptionat the phonologicallevel.

SubjectClassification:
9.7, 9.2, 9.5.

INTRODUCTION derivedfrom the singlearticulatoryvariable of voice

onsettime (VOT). VOT standsfor the temporalrela-
Bilingualspeakers appearto have little difficultyin tion existingbetweenchangesin the glottal aperture
keeping theirywolanguages distinct.Thisobservation,and the supraglottalgestures;and acoustically,it is
backedby the resultsof experimentalstudiesat the realizedas the timingdifferencebetweenthe releaseof
lexical,syntactic,and semanticlevels, •-2 impliesthat the stop occlusionand the onsetof quasi-periodical
bilingualshavedeveloped separatestrategies for inde- laryngealvibrations. s
pendentlyprocessing linguisticmaterial in each lan- The importanceof this temporalcue has now been
guage.Yet, little is knownabouthowbilingualsprocess well documented. Cross-language studiesof speechpro-
languageat the phonologicallevel. The experiment ductionhaveshownthat variationsin VOT distinguish
reportedhere was designedto study this question;our voicing contrastsnot only in English, but in many
specificintention was to investigatehow bilinguals other languages2Moreover, these same studieshave
perceiveand producethe voicedand voiceless formsof foundthat thedifferentlanguages usesomewhat similar
the stopconsonants/b-p/,/d-t/, and/g-k/. locationsalongthe VOT continuumto producevoicing
Phonetidanshave invokedvariousphoneticdimen- distinctions.In addition, experimental observations
sionsin order to capturethe voiced-voiceless distinc- have indicateda complementary use of VOT in per-
tion prevalent in most languages.In the particular ception--vadations alongthis dimension beinga suffi-
caseof Englishstopconsonants, for example,distinc- cientcuefor perceptually categodzing stopconsonants
tive featuressuchas voicing,aspiration,and articula- into phonemiccategories. TM
tory forcehave all beenimplicated,eitherindividually Given this evidence,we felt it wouldbe instructive
or in combination,as featuressufficientto generatethe to determine how individuals who have acquired two
phonemiccategories/p,t,k/ and/b,d,g/. a-4 However, languages deal with VOT as a phonological cue.What
the apparentindependence of thesefeatureshas been was done experimentally,then, was to test bilingual
questioned.Lisker and Abramson *-e have sought to subjectsundertwo separateconditions;whenthey were
demonstratethat these features can be completely set to speakandbe spokento in onelanguage,and then

The Journalof the AcousticalSocietyof America 421

 CARAMAZZA ET AL.

o /
FIO. 1. Percentageof /p/ responsesas a
g t I/ functionof VOT. E•ch data point in the UE and
UF functions is based on 50 observations. The
•o I // • dat• pointsfor the functionsof the bilingualsin
the French set (BF) and the bilingualsin the
Englishset (BE) are basedon 100 observations
each.

• 0 __••-=•==• - - -• ...... •--•--•--•-

VOICE O•SET TIUE (msec)

again, when they were in a set for the other language. were analyzed. Thus the number of words analyzed
Further, and in order to establishbaselinesagainst were: 180 for each unilingualgroupand 360 for each
whichto measurethe effectsof acquiringtwo languages of the two languageconditions of the bilingualgroup.
on the phonological system,we alsotestedunilingual
speakers of the languages involved.In eachcondition, B. Subjects
perceptionwas assessed by having subjectslabel syn-
thetic speechsoundsthat differed in VOT, while Forty subjects were paid to participate in this
experiment:10 unilingualCanadianEnglishspeakers
spectrographic analysesof spokenstop-initial words
were used to assessproduction.Our subjectswere (UE), 10 unilingualFrenchCanadianspeakers(UF),
either unilingualor bilingualwith respectto Canadian and 20 CanadianFrench-Englishbilinguals(BF and
Englishand CanadianFrench. BE). The subjectswere either high schoolor uni-
versityundergraduate studentswith agesrangingfrom
17 to 25 years. Of thesesubjects,19 were female and
I. METHOD 21 were male, and all had normalheatingas reported
A. Materials subjectively.
The unilingualsubjectswere all native speakersof
The experimentalstimuli in the perceptionpart of their respectivelanguages,either English or French.
the studywere threedifferentcontinuaof stop+vowel However, they all knew some words in the other
syllablessynthesizedat the HaskinsLaboratories.The language,which is unavoidablein a bilingual country.
basicpattern for each stimulusitem consistedof three The bilingnal subjects were all native speakers of
steady-state formantsfor the vowelIs] to whichwas French and had begun to acquireEnglish at no later
addedthe appropriatereleaseburstand formant transi- than their seventh birthday. Actually, 10 of these
tions to produceeither the bilabial, alveolar, or velar subjectswere classifiedas compoundbilingualsand 10
stops.Each of thesethree setscontained37 syllables as coordinatebilinguals,the formercategoryreferring
and variedonly on VOT. The duxationof eachsyllable to thosesubjectswho had learnedboth languagesin
was 350 msec.The VOT valuesrangedfrom 150 msec the same social context, the latter to those who had
preceding(--150) to 150 msecfollowingthe release lea.med their languagesin differentcontexts?We had
burstin stepsof 10 msec,exceptfor the rangeof -10 felt initially that there might be somedifferencesbe-
to 50 msec where steps of 5 msec were taken. Five tween these two groups. However, since no major
random sequences of each of the three basiccontinua differences arose,eitherin the perception
or production
were producedby splicing.These 15 sequenceswere part of the experiment,we have treated them as one
employedin the perceptiontest. group.
The wordsused to assessproductionwere common Bilingualproficiencywas determinedin two ways.
stop-initialEnglishand Frenchwords,typed on 3X5 Subjectswereinitially screened by havingthemproduce
white cards.With the exceptionof one English word, self-ratingsof their knowledgeof Englishon a 7-point
noneof the wordsusedin the experimentwereminimal scale,with valuesrangingfrom no knowledge (1) to
pairs. There were nine wordsfor eachof the six stop excellent knowledge(7). Only thosesubjectswhoachieved
consonantsfor each language,thus making a total of a score of at least 5 were admitted for further assess-
54 English and 54 French words. Only three words, ment. The final selectionwas then made by having
those with a stop+l-a-] initial, for each consonant, subjectsread aloud a passagefrom JaneAustin'sSense
422 Volume 54 Number 2 1973
 ACQUISITION OF A NEW PHONOLOGICAL CONTRAST

ioo

60
FIo. 2. Percentage
of/t/responses as a func-
tion of VOT. Each data point in the UE and
UF functions is based on 50 observations. The
data pointsfor the functionsof the bilingualsin 40
the French set (BF) and the bilingualsin the
English set (BE) are basedon 100 observations
each.

-50 -40 ~30 -20

VOICE ONSET TIME (msec)

and Sensibility.Readingspeedhas beenshownto be sisted of the five different random orders of the con-
an effectivemeasureof bilingualism • and so the first tinuum.This procedure, of assessing productionfirst
20 subjectsthat read aloud at a rate of at least 180 and thenperception,wasrepeatedfor eachof the three
words/min.weretestedin the presentexperiment. classes of stops,bilabial,alveolar,and vetar.-Theuni-
lingual subjectsthus read three sets of words and
C. Procedure labelledeachof the three stop continua.In contrast,
Subjectswere tested individually in acoustically
the bilingualsubjects,who paxticip•ted •bothin the
Englishand Frenchconditions, readsixsetsoœwords,
quiet rooms.The unilingualspeakerswere testedin a
threeset• in•Englishand threesetsin French;also,
singlesession eit_h_e_r•at_
an Englishuniversity or at a they labelledeach of the three stop continuatwice.
Frenchhighschool, depending on the language group The order of presentingthe three typesof stop con-
they belongedto. Two different experimenterswere
sonantpairs,i.e.,bilabial,alveolarorvelar,wasrandom.
employed:one, a native speakerof CanadianEnglish
Wide-band (and whenevernecessarynarrow-band)
who tested the Englishsubjects,the other, a native
spectrogramswere made of the recordedwords and_
speakerof Canadian French who tested the French
analyzedon a voiceprintsoundspectrograph.VOT
subjects.The bilingualgroup was tested twice, once
valueswereobtaineddirectlyfrom thesespectrograms
by the Englishexperimenter in the Englishuniversity
by measuring the distancebetweenthe onsetof energy
and onceby the Frenchexperimenter in the French
in theformantfrequency rangerepresenting therelease
highschool.The two testingsessions for the bilinguals
of air pressureand the first verticalstriationsrepre-
were betweentwo and three weeksapart, and were
sentingglottalpulsation.
counterbalanced in termsof whichlanguage wastested
first.In addition,precautions weretakenbothto main-
tain uniformitybetweenthe two testingsessions and H. RESULTS
to createa psychological set for one or the other lan-
A. Perception
guage.In the Englishtestingsituation,then,materials,
instructions,and the atmosphere,in general, were The averageidentificationfunctionsfor eachof the
maximallyaccentuatedas beingEnglish,while in the three groupsof subjectsare presentedseparatelyfor
Frenchconditionthesevariableswere emphasized as the bilabial,alveolar,and velar stops(Figs. 1, 2, and
beingFrench. 3). In each of thesethree figures,the percentageof
Each testingsessionbeganby having the subject "voiceless" responses (p, t, or k) isplottedasa function
readalouda setof Englishor Frenchstop-initialwords of the VOT continuum.The 50% crossover pointsfor
containingeitherof two homorganic consonants, e.g., the two unilingualgroupsare clearly different from
papillon-ballade,and by recordinghis responses on eachother.In all cases,theseperceptualcrossovers are
tape. The subject was then asked to label the VOT at lowerVOT valuesfor the UF subjectsthan for the
variantsfor the sameclassof stopsthat he had just UE subjects.This is especiallyso for the bilabial and
read. The responsemode in this task was to have the velar stops. In addition, the identification functions
subjectsmark on a printed form their choiceof either for the two groupsare noticeablydifferentin termsof
the voicelessor voicedform of the hornorganic pair their shapeand their rate of change.
beingtested.The stimulifor thispart of theexperiment If rate of changefrom onephonemecategoryto the
weredeliveredthroughbinauralheadphones, and con- other is taken as an indication of noise in the decision

The Journalof the AcousticalSocietyof America 423

 CARAMAZZA ET AL.

80 *-----* u• / I/ ,* I

' Fro. 3. Percentageof /k/ responsesas a

function of VOT. Each data point in the UE
and UF functions is based on 50 observations.
5,0 I ,'•' I I The data pointsfor the functionsof the bilin-
guals in the French set (BF) and the bilinguals
I ,if; I in the English set (BE) are basedon 100 ob-
I
/A ,,/ ; I servations eax:h.

o-- ............... --' I

-50 -40 -30 -20 -I0 0 I0 20 30 40 50

VOICE ONSET TIME (ms.ec)

process, then the rathersteepcurvesfor the UE group UE speakers,they utilize this acousticcue more than
suggest that VOT is a strongperceptualcuein Canadian UF speakersdo.
English.However,this doesnot appearto be the case Anotherpoint of interestin the bilingualdata is the
o•n•l•re•nch; on the contrary,speakersof similarity of the curvesin the two languagesets. In
CanadianFrench seemrelatively insensitiveto VOT neitherlanguagemodewasthereany switchingof the
as a categoricalphonological cue.This is suggested
by bilingual functionsto match the functionsof the uni-
noting tha•LheS•cation functionsshowa slower lingual subjects. Rather, bilinguals appear to have
r•(e of-changeand are lessmo•notonicthan thosefor based their perceptualderisionson the same criteria
theUE speakers.
Actually,theU•F•ur•ly &•oth t•heFrenchandEnglishsets.
ambiguouswith respectto how they divide the con- In our descriptionof--•immlions•plesen_t.e•
tinuum, but it should also be pointed out that the Figs. 1-3, we have emphasizedthree measures.One is
staggeredshapesof the UF group'sfunctionsdid not the crossoverpoint marking the VOT value at which
resultfromthe spurious averaging of individualcurves. 50% of the responses are for one phonemecategory
_ haspectionof individual identificatior•functionsre- and 50% for the other]This valuecan be taken as an
yealed that all the UF subjectshad nonmonotonic estimateof the categoryboundaryfor the two pho-
functions.Many of the "dips" in the individualcurves nemes.The other two measuresare rate of change
matchedthosepresentin the averagefunctions. (slope)fromonecategoryto thenextandtheõhapeof
The bilingualsubjects'50o/0crossovervalues,both the function(e.g.,monotonicity). While rateof change
in the Englishand Frenchmodes,occupyintermediate capturesthe degreeof dispersion,or variability of re-
positionsrelative to thoseof the two unilingualgroups. sponsebest, this characteristiccan be further defined
However, their identification functions have certain by the shape of the function. Thus far, however, we
characteristics in common with the UF as well as the have been estimatingthese three measuresdirectly
UE functions.On the onehand, the bilinguals,like the from the graphs;and althoughdifferencesand similari-
UF speakers,generatedcurvesthat are not especially ties amongthe group functionsare readily apparent
monotonic.On the other hand, the rates of change from the graphs,they do not permit descriptivesta-
shownby the bilingualcurvesapproachthosegenerated tistical analyses.
by the UE speakers.This suggeststhat, while bilin- Therefore,to measurethe degreeof dispersionof the
gualsare perhapslesssensitiveto VOT variationsthan variousperceptualfunctionswe have adopteda curve-
fitting proceduretermedProbit Analysis.tsThis analy-
TtmLs• I. Means and standard deviations (in milliseconds)of sis makes use only of the data lying between the
the perceptualfunctionsas estimatedby Probit Analysis.The asymptot(c
regions
of eachcurveandyieldsquantita-
meansare estimatesof the crossoverpoints, and the standard
deviationsare estimatesof the slopeof the perceptualfunctions. tive estimatesof both the 50% crossoverpoints
(means)and the slopesof the curves,the latter being
/•p/ /d-t/ /g-k/ inverselyproportionalto the standarddeviationsof
Mean SD Mean SD Mean SD
the distributionof data points. Parenthetically,it
UF 8 17 21 15 14 23 should be noted that if identification functions are to
UE 24 7 25 5 38 7
BF 19 8 24 6 28 8 be usedas evidencefor positingthat speechperception
BE 17 10 23 7 27 8 is categorical,then thesefunctionsshouldbe monotonic
in the crossover range.Grossviolationsof monotonicity

424 Volume54 Number2 1973

 ACQUISITION OF A NEW PHONOLOGICAL CONTRAST

4O

10-

-•50 -io0 -50 0 .50 IO0 i•0

VOICE ONSET TIME IN msec

FIG.4. VOT distributions

for theproduction
of stopinitialwordsby FrenchandEnglishunilingualspeakers.
The distributions
for
eachmemberof the pair of hornorganic
stopsis basedon 30 observations.

imply that the independent variable (i.e., VOT) is Figure 5 presentsthe bilingual productiondata
only weakly.correlatedwith somevariablewhich is separatelyfor the FrenchandEnglishmodes.Asshown,
influencingthe subject'sdecision. 'the bilingualsubjectsproducedvoicingdistinctions
Table I presentsthe outcomeof the Probit Analysis: whichwereclearlydifferentfor the twolanguages; and
meansand standarddeviationsfor the threegroupsof this disparitystandsin markedcontrastto the simi-
subjectsare shown separatelyfor the threestop- larity of theirperceptual
functionsin the two language
consonant pairs.In all casesth• quantitative results modes.When speakingFrench, they producedstops
are well alignedwith the estimatestaken from the with overlapped VOT valuessimilarto thoseof the UF
graphs.The variability of responseis lowestfor the group, whereaswhen speakingEnglish, their VOT
UE group,next lowestfor the bilingualgroup(in both values shifted toward the LTE subjects' distribution
languagemodes),and highestfor the IJF group.The rangeand showeda clearseparationfor eachphonetalc
largeamountof variabilityin the UF groupsupports contrast.However,asis apparentfrom the meanvalues
the suggestion that they do not useVOT as a major of VOT for the voicelessconsonants(Table II), the
cuein discriminating voicedfrom voicelessconsonants. bilingualsubjectswere morecloselyalignedwith the
The bilingualgroup,however,doesappear to utilize UF group,whenin the Frenchmode,than they were
this acoustic cue: the standard deviations for this with the UE subjects,whenin the Englishmode.This
group, while slightly greater than those for the UE may reflect the fact that, while our subjectswere
group,are muchsmallerthan thoseobtainedfor the bilingual,they had acquiredEnglishas their second
UF group.In addition,the consistency of the means language.
and standarddeviationsin the two languagemodes Parametricstatisticaltests could not be performed
suggests that bilingualsuse the samephoneticcriteria on all the productiondata becauseof the clearly
whenperceiving voicingdistinctionsin Frenchas they bimodal, in fact discontinuous,distributionsof the
do whenperceiving thesedistinctionsin English. voicedstops.A repeatedmeasures analysisof variance
compared
theperformance
of thebilingual
speakers
on
B. Production the threevoiceless stopconsonants in the twolanguage
modes.Performancein thesetwo modeswas signifi-
Histograms of VOT distributions for the two uni- cantly different[F(1,19)=81.08, P<0.001], with bi-
lingual groups are presentedin Fig. 4. As can be lingualsubjectsproducinglongervoidnglagswhenin
seenfrom the lowerpanelsof this figure,the distribu-
tionsof VOT valuesfor the UE subjectsdo not overlap
on any of the three phonetalccontrasts.For the UF voicelessTanrE II. Means of VOT values (in rnilliseconds)for the
consonants
in the productiontest.
group,however,the separationbetweenphonetalccate-
goriesis not so marked; there is, in fact, a substantial
/p/ /t/ /k/
degreeof overlapfor eachphonetalccontrast(Fig. 4,
UF 18 23 32
upperpanels).Theseresnitscomplement the findings UE 62 70 90
on the perceptualtask: VOT appearsto be an impor- BF 20 28 35
tant variable for voicing distinctionsin Canadian BE 39 48 67

Englishbut not in CanadianFrench.

The Journalof the AcousticalSocietyof America 425
 CARAMAZZA ET AL.

50- 50 50-

40

30

2O

IO h

20

IO' I0

-I•O -IO0 -50 0 50 I00 I.•) -I•0 -I00 -50 0 50 IO0 0 50 Ioo •0
voICE ONSET TIME IN msec VOICE ONSET TIME IN msec VOICE ONSET TIME IN msec

Fro.5. VOT distributions

fortheproduction
ofstopinitialwordsin Frenchandin English
bythebilingual
speakers.
Thedistributions
for eachmemberof the pair of homorganic
stopsis basedon 60 observations.

the English mode than in the French mode. In addi- ing thephonemic contrasts.Soit appearsthat VOT is
tion, the three stop consonants differedsignificantly phonemicin English, whether American or Canadian.
from one another[-F(2,38)=108.55, P<0.001], with Of course,this inferenceis basedon the assumption
with voicing lag being progressivelylonger for /p/, that steep monotonicfunctions in perception and
/t/, and/k/. A significanteffectwas obtainedfor the clearlyseparatedVOT distributionsin productionare
interaction of languagemode by type of consonant sufficientcriteria for assigningphonemicstatus to
[-F(2,38)= 15.04,P<0.001]. This latter resultindicates physicaldimensions.
that the bilingual subjectswere producingrelatively No comparisonscan be drawn for data obtained
longervoicinglagsfor/k/in English,as compared to from the UF subjects.However,their grosslynon-
French,than the languagedifferences seenfor/p/and monotonicidentificationfunctions,evincinga wide
/t/. rangeof perceptualuncertainty,stronglysuggestthat
Two other analysesof variance were carried out on VOT isnot a sufficientcuefor theperceptionof voicing
the productionof the voicelessstops.Thesecompared distinctions in CanadianFrench.Had we only investi-
bilingual productionsin the French mode with UF gatedspeechperception,then perhapsthe noisein the
productionsand bilingualproductions in the English UF perceptioncurvescould have been attributed to
modewith UE productions. The voicinglagsproduced the influenceof the vowelquality in the stimuli; but
by the bilingualsin the Frenchmodewerenot signifi- thispossibilityseemsruledout by the UF production
cantly different from those producedby the UF data. In fact, sincethe UF speakersshoweda sub-
speakers [F(1,84)=3.56,P>0.05], but in the English stantialamountof VOT overlapin their production
mode,voicinglagsfor the bilinguals weresignificantly of the hornorganic
stops,it seemsreasonableto suggest
shorterthan the lags for the UE subjects[F(1,84) that voicingandaspiration, themselves,
areonlymini-
=44.10, P<0.001]. Both analysesalso revealedsig- mally usefulas perceptualcuesin CanadianFrench.
nificant (P<0.001) consonant differences, but no It may be, then, that other phonetic dimensionssuch
significantlanguageby consonantinteractions. as articulatoryforce (fortis-lenis)
4 or perhapsrate of
formant transitionxeare the relevantphoneticdimen-
III. DISCUSSION sionsfor the UF speakers.In any event, the data ob-
A. Unilingual Speakers
tained from the UF group cast doubt on axty theory
assigningVOT a universalstatus in the total determi-
The findingswe have reportedfor the CanadianUE nationof the phoneticdimensions of voicing,aspiration,
group are entirely consistentwith Lisker and Abram- and articulatoryforce.
son'sdata on AmericanEnglish speakers2 ,7-8 Both
theirAmericansubjectsandour CanadianUE subjects B. Bilingual Speakers
generatedperceptualfunctionswith sharp monotonic
slopes;and both producedsimilar perceptual cross- While there are a numberof fairly evidentconclu-
overs at each place of articulation when labellingthe sionsto be drawnfromthe bilingualdata in thisstudy,
synthetic VOT variants. Further, the American and the generalizationof these conclusionsto other bilin-
Canadiansubjectsproducedsimilar and, in eachcase, gualpopulations mustbe limited.Specifically, having
clearly nonoverlapping VOT distributionswhen utter- suggested that VOT is phonemicin CanadianEnglish

426 Volume54 Number2 1973

 ACQUISITION OF A NEW PHONOLOGICAL CONTRAST

but not in Canadian French, our resultscover only are steepsuggests that the bilingualswere making
thosebilingualswho have acquireda secondlanguage greateruseof VOT as a phonetalc cuethan werethe
in whichphonemicdistinctionsare basedon an articu- UF subjects.The resultssuggestthat perhapsthe bi•
latory variable (e.g., VOT) not presentin the first- lingualsubjectshad to de•i with the test stimuli as if
learnedlanguage. Suchbilingualsareto bedistinguished they wereexemplaxs of Englishspeechsounds in order
from any who have acquiredtwo languages,both of to usethis VOT information.This is supportedby the
which basephonetalcdistinctionson the samearticu- fact that the identificationfunctionsof the bilingual
latory variable. subjectscloselymatch predictionswhichcan be made
In this context,considerfirst the productiondata. from their productionsof English words but not
When speakingin French, the bilingualsshoweda Frenchwords.The lack of monotonidtyin their identi-
markedoverlapin their VOT distributionsfor each ficationfunctions,however,suggests that the effectsof
phonetalccontrast.These distributionsvery closely interferencefrom their first-learnedlanguageare seen
resembledthoseshownby the [IF subjects,and, in in perception aswellasproduction.
fact, no statistical differencesemergedbetween the In general,it appearsreasonable to suggest that the
two groupson this measure.It seems,therefore,that phonological processors the bilingualacquiresfor his
the bilingual subjects had retained the Canadian secondlanguageare contaminatedby propertiesac-
Frenchmechanisms of encoding speech.In otherwords, cruingto hisfirst language.Further,unlikethe learning
there is no evidenceof phonologicalinterferencefrom of a secondvocabulary,the acquisitionof a second
their secondlanguage(English). phonologicalsystemdoes not appear to be quantal.
Yet the phonologicalsystemof the bilingual is not Rather, the processseemsto consistof a gradualand
completelyfreefrominterlanguage interference.Rather continuousprogressiontoward a target which may
the interference appearsto be unidirectional: from the neverbe attained:and this appearsto hold for per-
first,perhapsstrongerlanguageto the second, perhaps ceptionandproduction.
weakerlanguage.The bilingualsubjectswere capable However,whenthe similarityof the perceptualfunc-
of switchingencodingmechanisms at the phonological tionsin the two languagemodesis contrastedwith the
level,but they did so imperfectly.Thus, althoughthey differenceshown for the productiondistributionsin
showed both a shift in VOT to the UE distribution theselanguagemodes,the bilingualsappearbetter able
rangeand a clearVOT separationfor eachphonemic to adapt their productionmechanisms than their per-
contrastwhenspeakingin English,theydid not totally ceptual mechanismsto the secondlanguage.This
alignthemselves with the UE groupon thesemeasures. ability to switchmechanisms from one languageto
Analyses' of varianceperformedon the distributions of anotherhas been reportedin detail elsewhere, •7 but
VOT values for the voicelessconsonantsdid yield brieflystated,it seems that language switching is easier
significantdifferencesbetweenthesetwo groups. for production than for perception. In perception, the
Statisticalanalysescould only be carriedout on stimulusitself seemsto determinethe type of analysis
VOT distributionsfor the voiceless stops,but the VOT to be performed.
distributions producedby the bilingualsubjectsfor the
voicedstopsalso warrants consideration.Specifically, ACKNOWLEDGMENTS
it shouldbe noted that they showedno appreciable
We would like to expressour thanks to Dr. A. S.
differencesin the two languagemodesfor the voicing Abramson and Dr. L. Lisker of the Haskins Labora-
leadand shortvoidnglag regions.Thus, the difference
toriesfor their helpful commentson this researchand
in their productions of Frenchand Englishwordswas
in the amountof voicinglag producedin uttering the
for their loan of their synthesized stop continua.We
would also like to thank Dr. B. Green for suggesting
voiceless words.This suggests that in acquiringEnglish the rise of Probit Analysis and Mr. F. Sonsini for
the bilingualsubjectslearnedto controlVOT in their
writing part of the programfor the analysisof the
productions of the voiceless consonants and did not
modify their productionsof the voicedconsonax•ts. perceptual data.The assistance of Mrs. Bignellin data
collectionis appreciated.
Theseresultssuggestthat VOT controlis important
This research wassupported ia part by the National
only at phonemic boundaryregionsand relativelyun- Research Council of Canada, Grant number A0291,
importantat other points in the productiverange,
NINDS ResearchGrant number 09994, and NINDS
wherethe informationcarriedby VOT is phonemically Research Grant number 06209.
irrelevant.
Unlike the productionresults,the perceptualfunc-
tionsin the two languagesetswere not very different
from eax:hother: both curveshad similar shapes,steep lw. E. Lambert,"Psychological
Studiesof the
Interdependencies of the Bilingual'sTwo Languages,"in
yet nonomonotonic,
andbothshowed perceptual
cross- Substanceand Structure of Language, editedby Jaan
overpointsat positions
intermediate
to the UF and Puhvel (U. of California Press,Berkeley and Los Angeles,
UE functions. The fact that the identification functions 1969).

The Journalof the AcousticalSocietyof America 427

 CARAMAZZA ET AL.

2.LMacnamam,"The Bilingual'sLinguisticPerformance: a Some Experiments in Comparative Phonetics," Proc. 6th

PsychologicalOverview,"in Problemsof Bilingualism, Lang. lntn. Congr. Phon. Sc/., Prague563-567 (1970).
edited by i. Maenamara, $. SocialIssues 23, 58-77 (1967). UA. M. Liberman,P. C. Delattre,and F. S. Cooper,"Some
3R-M. S. Hefther,GeneralPhonetics (U. of WisconsinPress, Cues for the Distinction betweenVoiced and VoicelessStops
Madison, Wisc., 1949). in Initial Position," Languageand Speech 1, 153-167
4R.Jakobson,G. G. M. Fani, and M. Halle. "Preliminaries
to (195S).
SpeechAnalysis,"Aeoust.Lab., MIT, Cambridge,Mass., t3A.M. Liberman,
K. S. Harris,J. A. Kinney,andH. Lane,
Rep. No. 13 (1952) (Reprintedby MIT, Cambridge,Mass., "The Discrimination of Relative Onset-Time of the
1963). Componentsof Certain Speechand Non Speech Patterns," $.
SL. Lisker and A. S. Abramson,"A CrossLanguageStudy of Exp. Psychol. 61, 379-388 (1961).
Voicing in Initial Stops:AcousticalMeasurements," Word •S. M. ErvinandC. E. Osgood,"SecondLanguage
Learning
20, 384-422 (1964). and Bilingualism," .L Abnormal Social Psychol. (Suppl.)
SL.Liskerand A. S. //,btamson,
"DistinctiveFeaturesand 49, 139-146 0954).
LaryngealControl," Language47, 767-785 (1971). •4J.Macnamara,"How Can We Measurethe Extent of a
7A. S. Abramsonand L. Lisker, "Voice Onset Time in Stop Person'sBilingual Proficiency.
9" in Descriptionand
Consonants: Acoustic Analysis and Synthesis," 5th !ntn. Measurementof Bilingualism, edited by L. O. Kelly
Congr.Acoust.,Liege,1-4, 1965. (Toronto U. P., Toronto, 1969).
*A. S. Abramsonand L. Lisker, "DiscriminabilityAlong the •D. J. Finhey,ProbitAnalysis(Cambridge
U. P., Cambridge,
Voicing Continuum:CrossLanguageTests," Proc. 6th England, 1952).
Intn.Congr.Phon.Sci.,Prague,
569-573(1970). •6C.Summerfield
andM.P. Haggard,
"Articulatory
Rate
OA.S. Abramson
andL. Lisker,"VoiceTimingin Korean VersusAcoustic Invariantsin SpeechPerception,"J. Acoust.
Stops," Paper presentedat the 7th Intn. Congr.Phon. Sci., Soc. Am. 52, 113(A) (1972).
Montreal (1971). •?A.Caramazza,G. Yeni-Komshian,E. Zurif, and E. Carbone,
IøL.LiskerandA. S. Abramson,
"The VoicingDimension: "LanguageSwitchingin B'•mguals:the PhonologicalLevel,"
(in preparation).

428 Volume54 Number2 1973