Pallier

et

al.

Language-specific

listening

Language-specific listening
Christophe Pallier, Anne Christophe and Jacques Mehler

of their native

language.

As a

as native speaken which may a-ally YlFa dko.ptefont

do; worse, they apply their native interfere with successfd pr

data Srom studies on i&m&s

with

peakers of American English, Japanese or Igbo begin life identical language processing systems: were any of linguistic envi-

and grammatical main difficult.

structures of a second language, so that

perhaps they can read it with relative ease, listening can reFor instance, there may be a striking deafness to native speakers. between dental with phonemic contrasts: thus or between uvular to contrasts that are obvious immediately This is easy to demonstrate and retroflex

them to be adopted at birth into a different the adoptive However,

ronment, they would grow up as perfect native speakers of language, not of the parental language. All that, as monolingual adults, strucbabies are born equipped with the same processing abilities. it is also obvious these speakers cannot understand each others languages as they do not know the sounds, words or grammatical must, in part, be different guage may use information rures of languages other than their own. Thus, processing for each language; a given lanthat is not relevant in another -

English listeners have trouble distinguishing stop consonants of Hindi,

and velar stop consonants of rhe North American language Salish, although native speakers of those languages perceive the distinctions with ease2,3.The locus of this failure to disphonetic contrasts has been the target between criminate non-native

of years of research4,5. These efforts have revealed that adults have not necessarily lost the ability to discriminate crimination contrasting may occur without pairs of sounds that do not belong to their language: disacoustic experience if the Zulu phonemes cannot be subsumed by any native

Igbo distinguishes between words just by changing the tone with which they are spoken, English allows a vowel to occur in a full or a reduced form, Japanese places verbs at the end of a sentence, and so on. To what extent does the understanding of spoken language involve universal characteristics, which are fundamental ing dependent on particular to the cognitive architecture of the human language system, and to what extent is processfeatures of specific languages? considerIn this These are questions that recently have prompted able research efforts in the area of psycholinguistics. paper we focus on evidence that perception the language learned by the listener. Language-specificity in adult processing to a particular

categories - thus English listeners can discriminate

clicks. If, however, the foreign sounds are both similar to one phoneme of the native language, then typically discrimination is difficult. Not only do listeners of different languages use differabout the by ent phonemic categories to represent the speech signal, but their perceptual system also exploits knowledge constraints on the co-occurrence phonotactics). of these phonemes (i.e. the

is dependent on

In Japanese, a nasal consonant followed

a stop consonant will always have the same place of articulation as the stop; thus constraint

That adult language processing is tailored second language. The flexibility ers never heard previously, noisy background,

tomboand kinko are words, but tonbo

the same

language becomes apparent when adults attempt to learn a that human listeners are acunderstanding speakspeech against a customed to displaying - immediately

and kimko could not be. In English and Dutch,

holds in general, but there are many exceptions, Japanese

ml: +33 119 i4 22 76 fkc33145449R35
e-mail: paliier@Glscp.

especially in prefixed and compound words (unbearable and tomcat in English; renbaan and imker in Dutch). listeners can exploit place of articulation match to speed de-

understanding

and so on - disappears. Even for those

who have attained a high level of competence in the lexical

tection of a stop consonant preceded by a nasal consonant,

ehess.fr

Copyright

8 1997. Elsewer

Science
Trends

Ltd. All rights in

reserved.

1364-6613/97/$17.00 Sciences Vol.

PII. 51364.6613(97)0104&9 1, No. 4. July 1997

Cognitive

. QD

Pallier

et al.

- Language-specific

listening

whether

the last of three nonsense items (pronounced

by

native speakers of Dutch)

most resembled the first or the worse than

second item, French listeners performed significantly (bopelo, bojoelo, bopelo), but structure and required significantly better

than Spanish listeners when the decision was based on stress Spanish listeners when the decision was based on segmental that stress variation be ignored (rope%, bojelo, contrast bopelo; see Fig. 1). French listeners deafwith the acoustic in accent differences

ness to stress is not due to unfamiliarity per se, since inter-syllable do occur in French. However, never distinguish listeners may ignore them.

in French, such differences

one word from another; in consequence,

Speakers of Spanish and French show similar sensitivity

French

Spanish

to the syllabic structure of utterances in various psycholinguistic tasks 10-3 , but speakers ofJapanese are sensitive to another unit: they automatically group phonemes into morae - subsyllabic units consisting of a vowel, a CV or a syllablefinal consonant6,4,5. Importantly, these studies showed that listeners parse foreign language input using their native units. For example, French listeners segment Japanese in terms of syllables5, while Japanese listeners impose a moraic structure on English, French and Spanish words1*,i6. Another dimension in which languages differ concerns speakers do not pause between cues to word boundof words. The way have the impression cues to word boundaries: aries, yet listeners that this efficient with

words or otherwise provide definitive nevertheless segmentation

hearing speech as a sequence of individual

occurs also differs across

languages. Thus, in English and Dutch, most words begin vowel),* strong syllables (syllables containing an unreduced and, indeed, listeners treat strong syllables as and weak syllables as likely to be distinction is not used. Such a strategy is simply not available in

French
Fig. ments 1 Reaction in times and (in grey) Spanish and error subjects. rates (A)

Spanish
(in black) in ABX discrimination judg-

likely to be word-initial word-internal-*.

languages where the strong-weak Similarly,

Discrimination based only on accent, phonemes fixed (e g. VAsuma. vaSUma, VAsuma: correct response: first item). (6) Discrimlnation based only on phonemes, with irrelevant variations in accent (e.g. VAsuma, faSUma. vaSUma correct response: first item). These data show that Spanish subjects are more sensltlve to accent variations than French subjects.
French

Finnish listeners exploit vowel harmony in speech segmentation *a, but obviously this is not possible in lanvowel harmony. We expect that whatever in phonoltheir language, listeners exploit the regularities ogy and lexicon in order to help segmentation. Finally, the evidence combines to show that listening Partly, this is inevitable,

guages without

but Dutch listeners do not use this information. can speakers use phonotactics phoneme; to predict

Not only

itself is highly language-specific. that they provide. With employ non-native

the upcoming

simply because languages differ in the type of information But the effects are more far-reaching. languages, foreign accent in production equivalent: listeners processing procedures, for the structure of the

in their attempts to build up a representation phonemes when

of the speech signal that follows the patterns of the native language, speakers can even insert illusory a stimulus does not conform to this pattern. Thus, speakers of Japenese (a language that does not allows word-internal obstruent sequenceP. Some levels of phonological organization such as stress require longer stretches of speech in order to be extracted. In English, words such as insight and incite, or, in Spanish, bebe and bebe, contrast only in stress. Native speakers of these languages have no difficulty apart; but Dupoux telling such stress pairs that speakers of In an et al. demonstrated clusters) have a lot of trouble and VCVCV (V, vowel; discriminating C, consonant) between VCCV

appears to have a direct perceptual their native phonological and when these are inappropriate foreign language, listening is d&cult. The development started with quirements

of language specificity processing abilities, so exquisitely end up to the re-

How does it come about that adult language users, who the same infant with processing routines tailored

of the native language that they actually intertongues? Assuming

fere with the processing of non-native

French, a language that does not have stress contrasts between words, may ignore stress contrasts entirely. ABX discrimination paradigm, in which subjects judged

that babies are born equipped with constraints on what a human language can bez3,**, one has to explain how they learn their mother tongue. The current view is that when listening to speech signals, be they native or foreign, infants

Trends

in Cognitive

Sclencer

- Vol.

1, No.

4.

July

1997

Pallier

et

al.

Language-specific

listening

12 1

70

0 Experimental I Control

65 60 t

Rhythmi,.

2 vs 3 -2-

2 vs 3

Syllables

Morae

30 25

-41
Fig. items. rate morae This morae number 2 Discrimination trisyllabic experimental than significant but not rhythmic Discrimination in the number) is statistically of syllables, (the Japanese by French items, and group in the control in the unit, IS attested newborn lists (change group by a larger babies increase stimuli (change differing and stimuli differing in number of lists trimoraic in sucking syllable/ only). in of Fig. of di- versus of di- versus

20 1 BL

I -5

1 -4

I -3

I -2

I II--. -1 cs +l

~.
+2 +3

t4

Minutes
3 Mean and from sucking four minutes rate in a non-nutritive were the made taken two one and during from after change sucking the two in stimulation syllable-tlmed stress-timed Dutch). than Infants infants experiment baseline stress-timed languages but one the from and from period (CS). The with rhythmc languages (Spanish in each rhythmic the syllable-ttmen group non-rhythnlc phase 32 French aqwup ,Dutch and ~a( and newborn before, switched English) periment babies. Measurements (BL), ii ie minutes

case of stimuli see text).

for stimuli

a mixture

of sentences from group from Italian

to a mixture there and more were

of sentences sentences then change

Itallanj. language reacted group

or vice versa.

The non-rhythmic English, to the

also changed

languages,

of the ex

represent all the features necessary to process any of the worlds language (e.g. stress, vowel length, moraic structure, complex syllabic structure, tone and so on). During the first year, when infants are exposed to their mother tongue, they will stop using fearures thar are not relevant to this language. This has been amply documented for the perception sal phonetic phonetic testeti. of phonemes: babies start off with a univerinventory that allows them to perceive any pet-

(e.g. Spanish significantly

of stimulation

hypothesis: they have shown that newborn infants tend to neglect the difference rhythmic properties. between two languages with similar Thus, French newborns fail to dis-

criminate low-pas5 filtered English sentences from low-pass filtered Durch sentences while they are perfectly able to discriminate between English and Japanese filtered sentencea. when newborns are habituated from rhythmically wirh a set of close languages (e.g. family (e.g. Italian and they do not react to a to a mixture of Furthermore. Dutch

contrast from any of the worlds language so far Between six and 12 months, their phonetic becomes similar to that of the adults have explored another they compared

ception increasingly Bertoncini

from their linguistic environment5-,. and her colleaguesaspect of speech perception newborn babies perception newborn in newborns:

sentences drawn

and English),

they notice the change to new sen-

tences drawn from another rhythmic Spanish sentences). In contrast, (two languages with different

of items that varied in number between lists of phoneti-

of syllables and/or morae. They have shown that French babies can discriminate cally varied words on the basis of the number of syllables~ (two versus three; see Fig. 2), but not on the basis of the number of morae (either two as in ig~, or rhree as in iNga or iiga). As mentioned interpretations above, French adults rely on the syllable while the mora is more salient for the Japanese. Two are available: the first one is that babies learn of their mother tongue very rapidly (had
results

change from, say, a mixture of Dutch and Italian sentences rhythms) English and Spanish sentences (see Fig. 3). Once babies have established what features are relevant to represent speech, they can start using this representation to discover regularities about their native language. In paradults

ticular, we have seen that literature

exploit

language-specific

strategies to segment continuous

speech into words. The

this characteristic

suggests rhat by the age of nine months, babies nine-month-old

we tested Japanese infants the ferent); the second possibility universally

might have been difunits are em-

have already discovered at least some of the regularities that form the basis of these straregies. Ihus, American strong-weak babies were shown to listen longer to lists of words such as beaver (the most frequent patwords such as

is that syllable-like

more salient at birth. This is an important

pirical question for future research. On the basis of this result, among others, Meh!er etal. have proposed that initially with different rhythmic babies pay attention mostly to on the sequence of vowels in the speech stream. Languages properties can be distinguished such a representation. guages differing Thus, languages that share rhythmic Recent work by Nazzi confirmation for this

tern in English) than to lists of weak-strong abeam. adults, English-speaking

use

This implies that from rhe age of nine months, babies may, just like English-speaking of English to hypothesize speech stream. about the co-occurrence of cues to the word this regularity knowledge

boundaries in the continuous Similarly, phonemes (phonotactics)

properties may more readily be confused by babies than lanin this dimension. and his colleaguesJ provides initial

may provide powerful

presence of word boundaries

(e.g. there has to be a word

Trends

Cognltlve

Sciences

Vol

1,

NO

4.

July

1997

Pallier

et

al.

Language-specific

listening

boundary Dutch

between the d and the J in dstr, as in bad string). Dutch babies prefer to listen to lists of of Dutch

J. Acoust. 7 Mehler, Cognitive Roberts, 8 Kakehi, perceptlo Structure and Cutler, 9 Dupoux,

Sot. Am. 100, 3831-3842 1.. Pallier. and C. and Christophe, Aspech (in press) and Kashino, structure M. (1996) Phoneme/syllable in Phonological Studies (Otake. T. A. Language (Sabourin, M.. and cognition, Craik. F.M.I. in and

Nine-month-old

syllables that respect the phonotactics (e.g. febv, rtz~rn)~. When Dutch babies are played

Biological

M., eds), Erlbaum K., Kato, and the K.

(e.g. bref murt) rather than to lists of impossible syllables in Dutch and American and they nine-month-old lists of Dutch

temporal

of speech,

and Language

Processing:

Cross-linguistic de Gruyter deafness

American words that differ only in their phonotactics, (e.g. Dutch but not English allows VI word-initial

A., eds), pp. 125-143. E. et al. (1997)

Mouton

prefer to listen to the words from their native language clusters such as in ulammend; English but not Dutch allows a wordfinal voiced consonant such as in hubbmP3). American ninemonth-old babies also prefer to listen to lists of English that contain frequent rather than infrequent Most of these findings do not hold true babies are tested, indicating that this at some point between six and nine monosyllables

A destressing

in French?

1. Mem.

Lang. 36,406421 10 Cutler, A. et al. (1992) The monolingual nature of speech

segmentation 11 Kolinsky, R.,

by bilinguals Morais,

Cogn. Psycho/. J. and word Cluytens,

24, 381410 M. (1995) evidence Intermediate from word

representations illusions 12 Pallier, 1. Mem.

in spoken

recognition:

Lang. 34, 19-40 allocation within syllabic structure

C. et al. (1993) Attentional words J. Me/. N. eta/. J. Mem.

phonetic patterns. when six-month-old learning occurred months of age. Conclusions

of spoken

Lang. 32, 373-389 (1992) Contrasting syllabic effects in Catalan

13 Sebastian-Gall&, and Spanish 14 Cutler,

Lang. 31, 18-32 T. (1994) Mora 1. Mem. Mom or phoneme? Furtherevidencefor

A. and Qtake,

language-specific 15 Otake, Japanese 16 Otake, by T. et al. 1. Mem. T., Hatano,

listening (1993)

Lang. 33, 824-844 Speech segmentation in

or syllable?

Lang. 32,25&278 G. and Yoneyama, in Phonological Studies (Otake, K. (1996) Speech segmentation Structure T. and and Cutler, Language A., eds),

So far, we have only been able to present studies relevant to sound patterns. We anticipate that similar studies will appear on other aspects of language processing such as morphology, syntax and possibly even semantics. We have reviewed a number of studies that illustrate the importance language-specific acquired. A number of important issues remain to be explored by we are investigating future research. For instance, currently cessing routines that correspond procedures and representations. also shown when some of these language-specific of We have devices are

Japanese

listeners, Cross-linguistic Mouton

Processing: pp. 183-201, 17 McQueen,

de Gruyter Models of continuous Lang. Cognit. Prefix speech recognition

J.M. et al. (1995)

and the contents 18 Schreuder, 1. Mem. 19 Cutler,

of the vocabluary Baayen,

Procexs. 10.309-331 stripping w-revisited

R. and

R.H. (1994)

Lang. 33,357-375 A. and Butterfield, evidence from S. (1992) juncture Rhythmic misperception cues to speech Lang.

segmentatlo: 31,218-236 20 Norris, D.G.,

J. Mem.

McQuee, in spoken

J.M. word

and

Cutler,

A. (1995)

Competition

and Mem.

whether bilinguals can master equally well the specific proto the two languages. to one of the languages). exposure to Earlier researchj indicates that bilinguals have a dominant processing routine (corresponding two languages can produce Also, we are exploring We do not know whether early and equivalent

segmentation cognit 21 Vroomen, segmentation: spotting 22 Suomi, Mem.

recognition

1. Exp. Psycho/. Learn.

10, 309-33 1 J., van ian, M. and de Gelder, from juncture 13. (1996) misperceptions Cues to speech and word

evidence Cognit

24. 744-755 J.M. and Cutler, in Finnish J. Mem. Instinct, What A. (1997) Vowel Lang. Morrow Infants Know, Blackwell contrasts by 36.422-144 harmony and

K., McQueen.

two routines (one for each lan-

speech segmentation 23 Pinker, 24 Mehler, 25 Trehub. infants 26 Kuhl. P. Infants

guage mastered), each similar to that used by monolinguals. whether the cortical zones that meare the same as diate language processing in monolinguals of a bilingual. standing

5. (1994) The Language J. and Dupoux, E. (1994)

S.E. (1976) The discrimination and adults Child Dev. 47, 466472 Linguistic experience

of foreign

speech

those involved when processing either one of the languages These and other issues are crucial to underinvolved in language usage and the constraints

eta/.(1992)
by 6 months

alters phonetic

perception

in

of age Science 255,606-608 Morse and syllables: rhythmical basis of

27 Bertoncini,

J. et al. (1995)

language acquisition.

speech representations 28 Bijeljac-Babic, old infants

in neonates

Lang. Speech 38, 311-329 1. (1993) How do four-dayDev. Psycho/. 29,

R., Bertoncini, categorize

J. and Mehler.

multisyllablc

utterances?

.... ........ ........

References 1 Takata, in noise 1. Acoust. 2 Polka, Y. and and Nab&k.

........ ........ ........ ........ ........ ........ .... ..

A.K. (1990) English consonant and American recognition listeners

711-721 29 Mehler. viewpoint, Grammar J. et al. (1996) I Lgnal Coping to with linguistic Bootstrapping J.L. and diversity: from Demuth, the infants to

Syntax:

Speech

in reverberation

by Japanese

in Early

Acquisition Erlbaum 1. and a

(Morgan,

K.. eds),

Sot. Am. 88, 663-666 L. (1992) on adult J.F. and initial 24, 672-683 The emergence a perceptual Perception: (Goodman, of native-language assimilation the Transition Nusbaum, phonological model, from I The Characterizing the influence Percept. (1988) of native language 52,37-52 speech Dev.

pp. 101-116, 30 Nazzi, Psychophys. Cross-language change

Lawrence

T., Bertoncini, towards Hum. Percept

Mehler,

1. Language of the

discrimination

by 1. Exp.

experience 3 Werker, perception: Psycho/.

speech perception Lalonde, capabilities C.E. and

newborns: Psycho/. 31 Jusczyk, the

understanding Perform.

role of rhythm

(in press) N.J. (1993) Infants of English words preference Child Dev. for 64.

developmental

P.W., Cutler,

A. and Redanz. stress patterns

predominant

4 Rest, C.T. (1994) influence Development Sounds in

675-687 32 Friederici, word A.D. and Wessels, and J.M.I. (1993) Phonotactic knowledge of

infants: of

Speech Words

Speech

boundaries

its use in infant

speech-perception

Percept.

to Spoken

J.C. and

H.C., eds).

Psychophys. 33 Jusczyk, P.W.

54, 287-295

pp. 167-224, 5 Werker, change Perception: (Goodman, 6 Otake,

MIT Press speech perception: in The Development Sounds developmental of Speech Words

et al.

(1993)

Infants

sensltwty

to the sound

pattern

of

J.F. (1994) Cross-language does the not involve loss, from

native 34 Jusczyk,

language

words

1. Mem.

Lang. 32,402-420 J (1994) language Infants 1. Mem. sensitivity Lang. 33,

P.W., Lute,

P.A. and Charles-Lute. in the native

Transition

Speech

to Spoken MIT Press moraic

to phonotactic 630-645

patterns

J.C. and Nusbaum,

H.C., eds), pp. 93-120,

T. et al. (1996)

The representation

of Japanese

nasals

35 Cutler,

A. et al. (1989) Limits on bilinguism

Nature

340, 229-230

Trends

in

Cognitive

Sciences

Vol

1,

No.

4,

July

1997