INFANT BEHAVIOR AND DEVELOPMENT 7, 49-63 (1984)

Cross-LanguageSpeechPerception:
Evidencefor Perceptual Reorganization
During the First Year of Life*
JANET F. WERKER AND RICHARD C. TEES
University of British Columbia

Previous work in which we compared English infants, English odults, and Hindi
odults on their obility to discriminate two pairs of Hindi (non-English) speech con-
trasts has indicated thot infants discriminate speech sounds according to phonetic
category without prior specific language experience (Werker, Gilbert, Humphrey,
8 Tees, 1981). whereas adults and children OS young OS age 4 (Werker 8 Tees, in
press), may lose this obility as a function of age and or linguistic experience. The
present work was designed to (0) determine the generolizability of such a decline
by comparing adult English, adult Salish, and English infant subjects on their per-
ception of a new non-English (Salish) speech contrast, and (b) delineate the time
course of the developmental decline in this ability. The results of these experi-
ments replicate our original findings by showing that infonts con discriminate
nonnative speech contrasts without relevont experience, and thot there is a de-
cline in this ability during ontogeny. Furthermore, dota from both cross-sectional
and longitudinal studies shows that this decline occurs within the first yeor of life,
and thot it is o function of specific longuage experience.

infants speech perception cross-longuage decline

While a large (but finite) number of sound segments occur in the languages of
the world, only a subset is used phonemically (to differentiate meaning) in any
particular language. Several researchers have predicted that human infants are
born with the ability to discriminate the universal set of phonetic contrasts
regardless of language experience, and that this ability declines as a function of
specific linguistic experience (Eimas, 1978; Morse, 1978; Werker et al., 1981).
Alternatively, it has been proposed that experience listening to a language may
be necessary to facilitate the perception of the phonetic distinctions used in
that language (Eilers, Gavin, & Wilson, 1979). Most relevant data support the
first of these predictions, suggesting that rather than having to learn to differ-
entiate phonetic features, young infants seem to respond to speech sounds ac-
cording to the categories that could serve as the basis for adult phonemic
l This work was jointly supported by grants to Richard C. Tees from the Social Sciences
and Humanities Research Council (410-81-0796), the National Research Council (PA0179) of
Canada, and the National Institute of Mental Health (lR03NH35829), and by NICHD Grant
HD12420 to Haskins Laboratories. We thank the infants and mothers who made this study possi-
ble. We also thank KathySearcy, Sue Tees, and Carole Bawden for their assistance. Special thanks
to AI Liberman for making us welcome at Haskins Laboratories. Requests for reprints should be
sent to Janet F. Werker, Department of Psychology, DaIhousie University, Halifax, Nova Scotia,
B3H 451, or to Richard C. Tees, Department of Psychology, University of British Columbia, Van-
couver, BC, V6T lY7, Canada.
49
50 WERKER AND TEES

categories (for a review, see Jusczyk, 1980). Specifically, evidence shows

young infants can discriminate speech sounds according to phonetic category,
even if those phonetic distinctions are not used in their native language. This
has been shown in the case of stop consonants differing in voice onset time
(VOT) (Aslin, Pisoni, Hennessy, & Perey, 1981; Syrdal-Lasky, & Klein, 1975;
Streeter, 1976); sibilants (Trehub, 1976); vowels (Trehub, 1976), and in the
case of liquids (Eilers, Oller, & Gavin, 1978). The one possible exception is the
lead boundary in VOT. Although this non-English boundary has been shown
to be discriminable to young English-learning infants (Aslin et al., 1981), there
is evidence to suggest that it may be more difficult to discriminate without ex-
perience than other phonetic distinctions (Eilers et al., 1979).
In contrast to the evidence pointing to the infant’s high level of com-
petence, research with adults has shown that they may easily perceive only
those sound differences which are used phonemically in their native language
and, in many cases, may no longer be able to identify or discriminate sound
distinctions used in nonnative language. This has been shown to be the case in
nonnative speechcontrasts involving differences in VOT (Lisker & Abramson,
1970; Singh & Black, 1966) and in place and manner of articulation (Goto,
1971; McKain, Best, & Strange, 1980; Miyawaki, Strange, Verbrugge, Liber-
man, Jenkins, & Fujimura, 1975; Tees & Werker, 1982; Trehub, 1976). If, as
suggested, infant speech perception is characterized by a high degree of initial l

ability and adult speech perception is more restricted, it becomes of interest to

ask how and when speech perception becomes modified (i.e., limited) to more
precisely match only those sound units which are used phonemically in the
learner’s native language.
In our previous work (Werker et al., 1981), we compared English adults,
English infants, and Hindi adults on their ability to discriminate two pairs of
Hindi speech contrasts that are not used in English. One of the Hindi contrasts
was the dental voiceless aspirated versus the voiced aspirated voicing distinction
/ tha/-/dha/ and the other, the voiceless, unaspirated retroflex versus dental
place of articulation distinction /!a/-/ta/. The results showed that infants
aged 6-8 months can discriminate these sounds as well as Hindi adults, but
that English adults cannot, particularly in the case of the place of articulation
contrast. Furthermore, English children at ages 12, 8, and 4 were as poor as
the English adults in terms of their discriminative performance (Werker &
Tees, 1983), showing the decline to be evident in children as young as 4 years of
age. The present research was designed to examine the generality of these
earlier results, and to identify the developmental time period within which the
decline in nonnative speech perception might occur.
EXPERIMENT 1

The first study attempted to determine whether the results obtained in this
earlier work were representative of developmental changes in cross-language
speech perception. To test this, an experiment similar to that reported by
Werker and colleagues (1981) was designed using a different non-English
 CROSS-LANGUAGE SPEECH PERCEPTION 51

(Thompson) place-of-articulation distinction. The Thompson language is an In-

terior Salish (Native Indian) language spoken in south central British Colum-
bia. The consonantal system of this language has two contrasting series of
back stops, including plain and glottalized versions of rounded and unrounded
sounds. These are variously called velars (/r’s) and uvulars (q’s) or pre- and
postvelar sounds (Mayes, 1979). In English, there is no distinction between
back consonants, in that only velar stops carry phonemic significance. The
Thompson pair chosen contrasts glottalized velar and glottalized uvular
sounds, /ii/-/$/. English infants, English adults, and Thompson adults
were compared on their ability to discriminate the non-English, Thompson
contrast, /ki/-/$I.

Method
Subjects. Twelve full-term infants (8 girls, 4 boys) ranging in age from 6
months, 4 days, to 7 months, 29 days, with an average age of 6 months, 29
days, were recruited by advertising in local newspapers. Infants were requested
to participate on days when they had no evidence of colds or ear infections.
Care was taken to ensure that each infant was comfortable in the experimental
room before testing began.
Ten English-speaking adults (6 males, 4 females), aged 22-35, were re-
cruited from the University of British Columbia campus. As it is difficult to
find adults with no second language training, notes were made on formal and
informal training. No English adults had exposure to a second language con-
taining the contrast being studied.
Five native Thompson-speaking adults (3 females, 2 males) ranging in
age from 30 to 65 were tested on their discrimination of the Thompson tokens.
Stimuli. Multiple natural exemplars of each sound were used in the dis-
crimination task, so that subjects would have to ignore within category acous-
tic variability and differentiate the sounds according to phonetic category,
much as is done in natural language processing. Care was taken to ensure that
the exemplars from the two categories were equated for intensity, fundamental
frequency, duration, and intonation contour. The English contrast used was
the place of articulation distinction, /h/-/da/, in which bilabial and alveolar
voiced stop consonants are differentiated. Four exemplars of /ba/ and four
exemplars of /da/ were used. The Thompson (non-English) contrast /kr/-
/ii/ involved two glottalized voiceless stop consonants where the uvular ver-
sus velar place of articulation distinction is the critical difference. These
sounds are produced by obstructing the air flow by raising the back of the
tongue either against the velum (velar) or behind the velum (uvular). Back con-
sonants are characteristic of North American Indian languages. English listeners
typically label both velar and uvular stops as velar, since uvular consonants are
not typically used in English.
In recording native Indians who are not accustomed to reading their lan-
guage, it was necessary to record whole words, and then ask the speaker to
52 WERKER AND TEES

repeat the first consonant-vowel (CV) sound. It was then possible to perform
acoustic analyses of words and CV repetitions to ensure that the CV syllables
contained the same consonant sounds as the words. The vowels in Interior
Salish languages vary (somewhat in free variation and in a somewhat systema-
tic fashion) between speakers and between consonants (Thompson & Kinkade,
in press). In over 100 recordings of k and q words and sounds from three dif-
ferent speakers, we were unable to find exemplars wherein a similar enough
vowel followed multiple CV only repetitions of k and q.
In the CV repetitions from the words kixm (to fry an egg) and &xm (to
make one see), however, there was one exemplar (or token) of /ki/ and one
token of /ii/ in one speaker’s recording in which the vowels sounded nearly
identical to one another and appeared similar in a wave form analysis. Since
there is a discontinuity in the wave form of glottalized stops, (a 0 amplitude
segment in the wave form) it was easy to use the /i/ periodic segment from a
single /I&/ and the /i/ periodic portion from a single / @/ to splice on to addi-
tional exemplars of the ejective portion taken from other k and of q repeti-
tions. This was done to yield three tokens of /iii/ with a single /i/ segment
and three different ejective portions, and three tokens of /&/ with a single
/i/ segment and three different ejective portions.
Classical spectrographic analysis has been shown to provide little infor-
mation as to the acoustic differences between velar and uvular sounds (Mayes, .
1979). In our spectrographic analysis, the only apparent differences between
typical spectograms from /ki/ and /&/ were in the third formant
transition, and possibly in the amplitude and duration of the burst. F, is flat
for /Qi/ at around 2300 kHz, whereas it rises for /ki/ from 2400 kHz to ap-
proximately 2900 kHz. The amplitude and duration of the /q/ burst are
greater than in the /k/ burst. Representative spectograms are shown in Fig. 1.
The average duration for each token was 400 ms with a 1500-ms silent interval
between tokens. Final tapes were prepared and set up with the use of ihe
PDP-224 computer at Haskins Laboratories in New Haven, CN. All tapes
were played on a Revox A-77 tape recorder at approximately 65 db SPL in a
tracoustics sound-attenuated test chamber. The entire operation was controlled
by a logic system (Werker et al., 1981).
Procedure. Infants were tested in a “head turn” (HT) paradigm (some-
times referred to as “visually reinforced infant speech discrimination” para-
digm) in which the infants were conditioned to turn their heads away from an
experimental assistant and toward a loud speaker within a specified time inter-
val (4 l/2 s) when there is a change in the speechsound category. Correct head-
turns are reinforced with the presentation, and illumination, of an electrically
activated toy animal inside a smoked plexiglass box while incorrect head-turns
(i.e., false positives) are not reinforced. Three exemplars of /ki/ were set up in
random order on Track 1 of a two-track tape, and 3 exemplars of /hi/ were set
up and aligned on Track 2. When changes from Track 1 to Track 2 occurred
during the testing, the subject’s task was to indicate when there was a change
 GLOTTALIZEO VELAR /ki/

. ...
v- 1
260 ’ 460 ’

GLOTTALIZEO UVULAR $i
//

Time (ms)
Figure 1. Spectogroms of typical exemplars of the Thompson glottalid velar/
uvular contrast (/k//-/4//).

53
54 WERKER AND TEES

in the phonetic category from /ki/ to /ii/. In this sense, it could be argued
that the HT procedure functioned as a categorizing discrimination task since
multiple exemplars were used. However, exemplars from a single category
were much more similar than those typically used in categorizing tasks (cf
Kuhl, 1979).
In the experimental setup the infant sat on its parents’ lap facing an ex-
perimental assistant (E2) across the table in a sound-attenuated chamber. The
speaker and the visual reinforcer were located at a 110O-angle,90 cm to the left
of the parent/infant. Both the parent and E2 wore headphones through which
music was played so they would not be able to influence the infant’s behavior.
The E2 kept the infant looking in his/her direction by manipulating small toys.
Another Experimenter (El) sat outside the chamber observing the infant
through a one-way observation window and monitored the logic system con-
sole (for details, see Werker et al., 1981).
In the conditioning phase of this procedure, the experimenter activated
the toy animal immediately following a sound change. Once the infant formed
the association between the sound change and activation of the toy animal
(usually within 2 to 10 trials), the infant, upon hearing the sound change,
turned its head to seethe toy animal perform, and activation of the reinforcer
became contingent on an appropriate head turn.
When conditioning was successful (i.e., three correct anticipatory head,
turns in a row) presentation of stimuli and activation of the visual reinforcer
became controlled by a logic system. Every time the infant turned its head, E2
pressed a button on the floor. All button presseswere recorded on a Grason-
Stadler event recorder. If the button press occurred within 4 l/2 s of the stim-
uli changing from one phoneme (i.e., ba) to another (i.e., da), the visual
reinforcer was activated by the logic system. A record of each was recorded.
The operation of the logic system also yielded a record of each time an infant
did not turn his/her head during a change trial (i.e., misses) and each incorrect
head turn (i.e., false positives).
A variate of this paradigm was used with adult English subjects where a
button press rather than a head-turn was the required behavioral response (see
Werker & Tees, 1983).
The criterion for successful discrimination was 8 out of 10 correct re-
sponsesto change trials with no more than two errors (i.e., two misses or two
false positives).’ The criterion for deciding an infant could not discriminate a
’ Typically, in the HT procedure, head-turns are only counted during demarcated observa-
tion intervals (e.g., Aslin et al., 1981; Kuhl, 1979; Werker et al., 1981). In this series of experi-
ments, we modified the procedure and controlled for bias by random manipulation of the timing
between experimental trials. To do this, we had to bring each infant under tight experimental con-
trol during conditioning. For example, if an infant was inclined to make frequent false positive
head-turns, we extinguished that response proclivity by lengthening the interval between sound
changes. Following conditioning, experimental trials occurred according to a random schedule
(every 4 to 15 trials) when the infant was continuously oriented toward the experimental assistant.
Since the timing of experimental trials varied, control trials were not used. Every head-turn during
this period was counted, yielding an overall probability of c .Ol for achieving an 8 out of 10 cor-
rect response to change trials.
 CROSS-LANGUAGE SPEECH PERCEPTION 55

contrast had two phases. First, the infant had to successfully discriminate /!~a/-
/da/ directly before and after failing to reach criterion on a nonnative contrast.
This was done to ensure that the failure of the infant was due to an inability to
readily perceive the sound difference, and was not due to nonspecific factors
such as boredom, dirty diapers, etc. Two infants (1 male, 1 female) were elimi-
nated from further analysis because they failed this phase. Second, the infant
was given 25 change trials on the nonnative contrast in their unsuccessful at-
tempt to reach criterion. Adults were also given 25 change trials in which to
reach criterion.
Results and Discussion
The portion of subjects that either reached or did not reach the 8 out of 10
criterion on the Thompson contrast is illustrated in Fig. 2. All 5 of the adult
Thompson speakers reached criterion, whereas only 3 out of 10 adult English
speakers did so. Of the English infants tested, 8 out of 10 reached criterion.

SUBJECTS REACHING CRITERION

ON THOMPSON CONTRASTS

100

+
s
: 50
2

0
Thompson En lish
Adults EA”3 it,” In 3ants
Figure 2. Proportion of Thompson-speaking adults, English-speaking adults, and
infants from English-speaking homes reaching criterion on the Thompson glottal-
ized velar/uvular contrast (/r)I / - / #/).
56 WERKER AND TEES

An analysis of proportions based on a chi-square analogue of the Scheffe

theorem (Marascuilo, 1966) was applied to these data. This yielded a signifi-
cant overall chi-square @ < .05; x2 = 8.94). Multiple comparisons between the
different groups were performed to determine which groups differed. The
results showed that performance of the Adult-English speakers was signifi-
cantly worse than performance of either the Adult-Thompson or the Infant-
English groups. The difference between the English infants and the Thompson
adults was not significant.
These results were similar to those obtained in our earlier work (Werker
et al., 1981). They showed that a decline in cross-language speech perception is
evident between infancy and adulthood. That is, young infants discriminate
the place of articulation contrasts according to linguistic category without spe-
cific linguistic experience, whereas adult speech perceptual ability is more
limited, reflecting discrimination of only those contrasts which are phonemic
in the listener’s native language.

EXPERIMENT 2
The second experiment was designed to establish the developmental time
period in which the decline in speech discriminative ability occurred. In this
endeavor, subjects were tested on both the Thompson /h/-/hi/ contrast, a,s
well as on one of the Hindi contrasts (/{a/-/ta/) employed in our earlier re-
search (Werker et al., 1981). Two contrasts were used to increase our con-
fidence in the generality of any results we might obtain.
Since we had already ascertained that by age 4, children appear to dis-
criminate nonnative contrasts as poorly as adults (Werker & Tees, 1983), we
decided to examine perception in children between the ages of 8 months and 4
years. After testing about 15 children of various ages, it became apparent that
important changes were occurring during the first year of life. At that time, we
narrowed our investigation to study cross-language perception in infants be-
tween 6 and 12 months of age. In addition to testing English infants, there was
an attempt to test infants being raised in homes in which either Hindi or
Thompson was primarily spoken. This was done to determine whether the
observed decline was a result of specific language experience, or whether it
could be explained by a general developmental decline in the ability to make
difficult perceptual distinctions.
Method
Subjects. In this study, data were collected from infants aged 8-10
months and 10-12 months, and were compared to the earlier data we had col-
lected on infants aged 6- to 8-months-old under identical testing conditions,
either in a previous study (Werker et al., 1981) in the caseof the Hindi contrast
or from Experiment 1 of the present study in the case of the Thompson con-
trast. One group of 8- to lo-month-old children (7 females and 5 males, rang-
ing in age from 8 months, 3 days, to 9 months, 10 days, with an average age of
 CROSS-LANGUAGE SPEECH PERCEPTION 57

8 months, 20 days) was tested on the Hindi contrast. A second group of 9

females and 5 males (ranging in age from 8 months to 9 months, 12 days, with
an average age of 8 months, 18 days) was tested on the Salish contrast. (An
additional 3 infants were dropped from further analysis for failing to reach
criterion on /ba/-/da/). One lo- to lZmonth-old group (5 females and 5
males, ranging in age from 10 months, 2 days, to 11 months, 15 days, with an
average age of 10 months, 20 days) was tested on the Hindi contrast. A sec-
ond group of 5 males and 5 females (ranging in age from 10 months, 2 days, to
12 months, 4 days, with an average of 10 months, 29 days) was tested on the
Salish contrast. (Data from 12 infants aged 10 to 12 months had to be dis-
carded because they failed to reach criterion on the /ba/-/da/ contrast). All
subjects were recruited from advertisements in the local newspapers.
Stimuli and Procedures. All infants were tested on the English labial/
alveolar contrast /ba/-/da/. The two non-English contrasts used were the
Thompson glottalized velar/uvular contrast /ki/-/&/ (as described in Ex-
periment 1) and the Hindi unvoiced, unaspirated retroflex/dental contrast
/{a/-/ ta/ . The Hindi language distinguishes four places of articulation in
contrast to the three used in English. Dental stops are produced by obtsructing
the air flow by placing the tongue behind the front teeth. Retroflex stops are
produced by curling the tongue back, and placing the top posterior to the al-
veolar ridge. The retroflex and dental Hindi stop consonants would both be
typically categorized as alveolar, [t], stops by an English listener.
Four exemplars of each sound were recorded by a native speaker. Final
exemplars were chosen so that variations in duration, fundamental frequency,
and intonation contour were randomized both within and between phonetic
categories. The average duration of a stimulus exemplar was 500 ms, with a
1500-ms interstimulus interval.
Acoustic analyses showed the main cues differentiating these naturally
produced speech sounds to be in amplitude of the burst, and in the slope of the
second and third formant transitions (see Fig. 3); for a full description of
those stimuli, seeWerker et al. (1981). Infants were tested in the HT procedure,
as described in Experiment 1. Requirements for concluding an infant could or
could not reach criterion on a contrast were identical to those of Experiment 1.
Results and Discussion
The number of subjects that either reached or did not reach criterion on the
two contrasts is shown in Table 1. As can be seen, most of the infants aged 6 to
8 months reached criterion on both contrasts, whereas by 10 to 12 months of
age, few infants reached criterion on either. An analysis of proportions was
performed on this data for each of the two contrasts. The overall x2 was sig-
nificant @< .05) for both contrasts (x*=21.67 for /k/-I@/; x2=24.59 for
/{a/-/&/). Planned multiple comparisons showed the significant differences
to be between the 6- to 8-month and the lo- to 12-month groups @ < MU), and
between the 8- to lo- and the lo- to 1Zmonth groups (PC .05). This suggests
RETROFLEX vs. DENTAL, STOP CONSONANT VOWEL SYLLABLES
!a/ ta

400 0 200 400

Time (ms)
Figure 3. Spectrograms of typical exemplars of the Hindi retroflex/dental contrast, / !a/ - / to/.
 CROSS-LANGUAGE SPEECH PERCEPTION 59

TABLE 1
Infant Discrimination Performance on Two Non-English.Speech Contrasts

(J) (2)
Reached Criterion 6-8 months tl- JO months JO- J2!!on+hs

The Retroflex/Dentol Contrast /to/-/ to/

Yes 11 0 2
No 1 4 0

The Velar/Uvular contrast / kl/ - / I$ /

Yes 0 0 1
No 2 6 9

that the lo- to 12-month-olds performed significantly less well than the two
groups of younger infants on both contrasts.
The number of trials to criterion was compared across contrasts and
across ages for those infants who could discriminate the sounds. It was
assumed that if the decline in the ability to discriminate nonnative contrasts ac-
cording to phonetic category occurred gradually between 6 and 12 months of
age, this gradual change would be evident in an increase in the number of trials
required to reach criterion. However a 3 x 3 repeated measures analysis of
variance showed there to be no significant differences between age groups,
F= 1.57, p< .05, or between sound contrasts F=2.78, p< .05, making it dif-
ficult to argue that there was a gradual increase across age in the number of
trials required to reach criterion.
To make sure that the decline around lo-12 months of age was not sim-
ply a function of a general performance decline for difficult perceptual tasks at
this age, a few same-aged babies being raised in homes in which Hindi or
Thompson are primarily spoken were also tested. To date, we have only been
able to find 5 infants (3 Hindi and 2 Thompson) between 11 and 12 months
who meet this criterion, and only 3 of these infants who would condition in the
HT procedure (i.e., reach criterion on the ha/da contrast). This drop-out rate
is similar to that found in the same-aged English infants. All three of these in-
fants reached discrimination criterion on their native contrast within 10 change
trials.
These findings show the decline in the ability to discriminate nonnative
phonetic contrasts occurs within the first year of life. That is, most of the
English infants tested could discriminate both non-English contrasts at 6 to 8
months of age. By 8 to 10 months a smaller percentage could discriminate the
contrasts, and by 10 to 12 months the infants were performing as poorly as the
young children and adults in Experiments 1 and 2. However, infants being
raised to speak Hindi or Thompson sounds could still discriminate the relevant
contrasts at 11 to 12 months of age. The results provide strong support for the
supposition that specific linguistic experience is necessaryto maintain phonetic
discrimination ability. Without such experience, there is a loss in this ability by
10 to 12 months of age.
60 WERKER AND TEES

EXPERIMENT3

The two-part criterion employed in Experiment 2 (i.e., reaching criterion

within 25 change trials and passing /k/-/da/ before and after failing a non-
English contrast) for concluding that an infant could or could not discriminate
a nonnative contrast resulted in an unequal drop-out rate which may have
biased the results obtained for infants in the lo- to 12-month range. (Comple-
tion rate was only 60% in the lo- to 12-month group compared to 85% in the
6-g,- and 8-10 month group. Older infants prefer to try to “visit” the toy
animal, rather than simply turn their head to seeit.) To control for this possi-
ble confound and to examine within subject developmental change, we decided
to attempt to replicate Experiment 2 using a longitudinal design.
Subjects
Six subjects, 3 males and 3 females, were tested successively at three ages. Sub-
jects were chosen who were particularly cooperative in the procedure at 6 to 8
months of age, in the hope that these same subjects would be relatively more
cooperative at 10 to 12 months of age. During Time 1 (6 to 8 months), infants
ranged in age from 6 months, 22 days, to 7 months, 29 days, with an average
age of 7 months, 15 days. At Time 2 (8 to 10 months), the infants ranged in age
from 8 months, 22 days, to 9 months, 25 days, with an average of 9 months, 2
days. At Time 3 (10 to 12 months), the infants ranged in age from 10 months, 2
days to 11 months, 11 days, with an average of 10 months, 22 days.
Stimuli and Procedure
Infants were tested on the Hindi contrast /{a/-/ ta/, the Salish contrast /iii/-
/&/, and the English contrast /ba/-/da/. In addition, the HT procedure was
used.
Results and Discussion
The results from this longitudinal study replicate those from the cross-sec-
tional study. In this experiment, all 6 subjects reached criterion on both non-
English contrasts when they were 6 to 8 months of age. When the subjects
reached 8 to 10 months, all 6 reached criterion on the Hindi contrast, and only
3 reached criterion on the Thompson contrast. By 10 to 12 months of age,
none of the 6 infants reached criterion on either contrast even though they
could reach it on the English /k/-/da/ both before and after failing the
Thompson sound.
In examining the data, it can be seenthat the results from the longitudinal
study closely match those from the cross-sectional study (see Fig. 4). The pat-
tern of change across infancy is precisely mirrored for the Thompson contrast.
The time course of the change was somewhat different in the case of the Hindi
contrast, with an apparent abrupt decline in discriminability occurring when
subjects reached 10 to 12 months of age.
 CROSS-LANGUAGE SPEECH PERCEPTION 61

INFANT SUBJECTS REACHING CRITERION

ON HINDI AND SALISH CONTRASTS

100
Cross-Sectional Data
80

5 60
u”
,”
0 40

20

100

Longitudinal Doto
80

c 60
Z
E
E 40
a.

20

6-8 months 8-10 months lo-12 months 11-12 months

Hindi /!CI/VS /to/

1 Solish /;i/vs /4 i/

Figure 4. Proportion of infant subiects from three ages and various backgrounds
reaching criterion on Hindi and Thompson (Salish) contrasts.

GENERAL DISCUSSION
In summary, these experiments provide strong support for the claim that
young infants can discriminate many of the phonetic distinctions used across
natural languages without relevant experience, and that there is a decline in
this ability as a function of specific language experience. Furthermore, these
experiments provide data showing that this decline may be evident by the end
of the first year of life.
62 WERKER AND TEES

It is easy to understand how an innate ability to perceive speech sounds

according to phonetic categories would ease the process of language learning,
and the eventual identification of meaningful units by predisposing the infant
to segment sounds according to functionally useful categories. In addition,
highly refined infant discriminative abilities followed by a selective loss
(and/or broadening) of category boundaries could facilitate the learning of
particular languages and dialects by allowing for the selective tuning of initial
sensitivities in accordance with a specific phonology. It is probably no accident
that this decline, or tuning, occurs at about the age that the child is beginning
to understand and possibly produce sounds appropriate to his/her native lan-
guage. It could be expected that this perceptual reorganization is closely
related to the acquisition of phonological contrasts. In future work we are in-
terested in investigating the relationship between cross-language speech
perception performance and both vocal output and language comprehension.
Finally, we are interested in ascertaining whether the changes identified in both
speech perception and production performance are mediated by general
changes in cognitive functioning, or whether they are more a function of
specific perceptual learning.
REFERENCES .
Aslin, R. N., Pisoni, D. B., Hennessy, B. L., & Perey, A. V. (1981). Discrimination of voice onset
time by human infants: New findings and implications for the effect of early experience.
Child Development. 52, 1135-l 145.
Eilers, R. E., Gavin, W., &Wilson, W. R. (1979). Linguistic experience and phonemic perception
in infancy: A cross-linguistic study. Child Development, 50, 14-18.
Eilers, R. E., Oiler, D. K., & Gavin, W. J. (1978). A cross-linguistic study ofin/ant speech per-
ception. Paper presented at the Southeastern Conference on Human Development. Atlanta,
Georgia, 1978.
Eimas, P. D. (1978). Developmental aspects of speech perception. In R. Held, H. W. Leibowitz,
& H. L. Teuber (Eds.), Handbook ofsensoryphysiology, (Vol. 8). Berlin: Springer-Verlag.
Goto, H. (1971). Auditory perception by normal Japanese adults of the sounds “L” and “R.”
Neuropsychologia, 9, 3 17-323.
Jusczyk, P. W. (1980, June). Auditory versus phonetic coding of speech signals during infancy.
Paper presented at the C.N.R.S. Conference, Paris.
Kuhl, P. K. (1979). Speech perception in early infancy: Perceptual constancy for spectrally dis-
similar vowel categories. Journal of the Acoustical Society of America, 66, 1668-1679.
Lasky, R. E., Syrdal-Lasky, A., &Klein, R. E. (1975). VOT discrimination by four and six and a
half month old infants from Spanish environments. Journal of Experimental Child Psy-
chology, 20, 215-225.
Lisker, L., & Abramson, A. S. (1970). The voicing dimensions: Some experiments in comparative
phonetics. In Proceedings of the Sixth International Congress of Phonetic Sciences.
Prague: Academia.
MacKain, K. W., Best, C. T., &Strange, W. (1980). Native language effects on the perception of
liquids. Journal of the Acoustical Society of America, 27.
Marascuilo, L. A. (1966). Large-scale multiple comparisons. Psychological Bulletin, 69, 280-290.
Mayes, S. V. (1979). An acoustic analysis of Thompson velar /k/ and uvular /q/. University of
Hawaii, Working Papers in Linguistics, I I, 11-22.
Miyawaki, K., Strange, W., Verbrugge, R. R., Liberman, A. M., Jenkins, J. J., & Fujimura, 0.
(1975). An effect of linguistic experience: The discrimination of(r) and (1) by native speak-
ers of Japanese and English. Perception and Psychophysics, 18, 331-340.
 CROSS-LANGUAGE SPEECH PERCEPTION 63

Morse, P. A. (1978). Infant speech perception: Origins, processes and alpha centauri. In F. D.
Minifie & L. L. Lloyd (Eds.), Communicative and cognitive abilities-Early behavioral
assessment. Baltimore, MD: University Park Press.
Singh, S., & Black, J. W. (1966). Study of twenty-six intervocal consonants as spoken and recog-
nized by four language groups. Journal of the Acoustical Society of America, 39, 371-387.
Streeter, L. A. (1976). Language perception of two-month-old infants shows effects of both in-
nate mechanisms and experience. Nature, 259, 39-41.
Tees, R. C., & Werker, J. F. (1982, June). Perceptualflexibility: Recording of the ability to dis-
criminate nonnative speech sound. Paper presented at the meeting of the Canadian Psycho-
logical Association, Montreal, Canada.
Thompson, L. C., & Kinkade, M. D. (in press). Linguistic relations and distributions. In W.
Sturtevant (Eds.), Handbook of North American Indians. Vol. 8, The northwest coast.
Trehub, S. (1976). The discrimination of foreign speech contrasts by infants and adults. Child
Development, 47, 466-472.
Werker, J. F., Gilbert, J. H. V., Humphrey, K., & Tees, R. C. (1981). Developmental aspects
of cross-language speech perception. Child Development, 52, 349-355.
Werker, J. F., & Tees, R. C. (1983). Developmental changes across childhood in the perception of
nonnative speech sounds. Canadian Journal of Psychology, 37, 278-286.

20 October 1982; Revised 11 January 1983 n