Common Neural Basis for Phoneme Processing in Infants ......Common Neural Basis for Phoneme Processing in Infants and Adults G. Dehaene-Lambertz 1,2 and T. Gliga 1 Abstract & Investigating

Common Neural Basis for Phoneme Processingin Infants and Adults

G Dehaene-Lambertz12 and T Gliga1

Abstract

amp Investigating the degree of similarity between infantsrsquo andadultsrsquo representation of speech is critical to our understandingof infantsrsquo ability to acquire language Phoneme perceptionplays a crucial role in language processing and numerousbehavioral studies have demonstrated similar capacities in in-fants and adults but are these subserved by the same neuralsubstrates or networks In this article we review event-relatedpotential (ERP) results obtained in infants during phoneme

discrimination tasks and compare them to results from the adultliterature The striking similarities observed both in behaviorand ERPs between initial and mature stages suggest a continuityin processing and neural structure We argue that infants haveaccess at the beginning of life to phonemic representationswhich are modified without training or implicit instruction butby the statistical distributions of speech input in order toconverge to the native phonemic categories amp

INTRODUCTION

Cognitive capacities such as language mathematics ormusic are highly developed in humans as compared toanimals Numerous studies have found precursors ofthese capacities in infants For example infants are ableto discriminate sentences in different languages (Mehleret al 1988) distinguish sets of objects based on theirnumerosity (Feigenson Carey amp Spelke 2002) orrecognize known faces (Bushnell 1982) These abilitiesare not so different from those of other animals Tam-arin monkeys are also able to discriminate two humanlanguages (Ramus Hauser Miller Morris amp Mehler2000) and two quantities of items (Hauser DehaeneDehaene-Lambertz amp Patalano 2002) and rhesusmonkeys respond to particular faces (Parr WinslowHopkins amp de Waal 2000) In a few years howeverchildren surpass these animals The challenge that de-velopmental psychology faces is to explain how theinfant brain becomes able to assume complex cognitivefunctions and why these functions rely on specializedneuronal networks that are reproducibly located acrosshuman adults Because the human genome cannot codedirectly for human complex cognitive capacities expo-sure to the human environment may take advantageof preexisting biases in human brain functional archi-tecture Therefore we need first to determine whatthe essential properties of a given processing systemare in adults and how similar are the capacities presentin infants Second we have to examine whether similarcapacities are subserved by analogous neural organiza-tion and to understand how initial biases in brainorganization could be shaped by the human envi-

ronment to give rise to the mature state In this articlewe will consider the case of phoneme perception1 Bystudying the cerebral bases of phoneme perception atthe initial and mature stages we can examine whetherthere is continuity or discontinuity between infants andadults and whether the linguistic environment shapes anexisting network or creates a new functional organization

Phoneme Perception in Adults

One main characteristic of phoneme perception is thecapacity to identify the same phoneme across differentacoustical realizations These differences are related tospeakersrsquo vocal tract speech rate speech mode or to thesurrounding phonemes Although many acoustical varia-tions are not phonetically distinctive sharp transitions(which are very robust across subjects) change thephonemic category Normalization and categorical per-ception are two important properties essential forspeech comprehension Duplex perception and integra-tion of visual information leading to MacGurk2 or ven-triloquy effects are also properties associated withphoneme perception None of these characteristics isstrictly specific to phoneme perception and most can beobserved in the perception of other sounds For exam-ple the perception of tones differing on the duration oftheir initial frequency transition is categorical (Pisoni1977) and duplex perception is described for nonspeechsounds (Fowler amp Rosenblum 1990) However no othertype of auditory processing relies so heavily and sys-tematically on these properties than speech processing

Another important characteristic of phoneme per-ception in adults is that it depends on the subjectsrsquonative language Adults have difficulties in discrimi-1INSERM-CEA 2Centre Hospitalier Universitaire Bicetre

D 2004 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 168 pp 1375ndash1387

nating foreign phonemic contrasts that are not presentin their native language such as rndashl for Japanesespeakers (Goto 1971) endashgt for Spanish speakers (Pal-lier Bosch amp Sebastian 1997) retroflex versus dental dfor English speakers (Werker amp Tees 1984b) amongmany examples The influence of native language onspeech perception is not limited to the phoneme reper-toire of the language but concerns all aspects of phonol-ogy Acoustic cues coding suprasegmental informationsuch as tones or stress and phonotactics the set of rulesthat govern how phonemes are combined within wordsare processed differently by subjects from different lan-guages For example Japanese adults have difficulties indiscriminating two pseudowords like ebzo and ebuzoThey perceive an illegal word like ebzo as ebuzoinserting an illusory vowel between consonants becauseJapanese does not allow for consonant clusters (DupouxKakehi Hirose Pallier amp Mehler 1999)

Phoneme Perception in Infants

Behavioral studies have demonstrated striking similari-ties between the phoneme perception capacities ofinfants and adults Like adults infants perceive pho-nemes categorically along acoustic dimensions such asvoice onset time (VOT) (Eimas Siqueland Jusczyk ampVigorito 1971) place of articulation (Werker GilbertHumphrey amp Tees 1981) and others They are able tonormalize across irrelevant acoustic variations such asthose related to different voices or pitches (JusczykPisoni amp Mullennix 1992 Kuhl 1983) or coarticulationcontext (Bertoncini Bijeljac-Babic Jusczyk Kennedy ampMehler 1988) They display duplex perception (Eimas ampMiller 1992) They are able to detect a match betweena vowel sound and the correct corresponding face move-ments (Patterson amp Werker 2003 Kuhl amp Meltzoff 1984)and are sensitive to the MacGurk effect (RosenblumSchmuckler amp Johnson 1997) The linguistic environ-ment rapidly modifies these initial capacities Around6 months infants show greater sensitivity to vowel cate-gories of their native language (Polka amp Werker 1994Kuhl Williams Lacerda Stevens amp Lindblom 1992) andaround 10 months to consonants (Kuhl Tsao amp Liu2003 Werker amp Tees 1984a) Although no experimenthas directly demonstrated a decrease of discriminationcapacities for other aspects of phonology than thoserelated to the repertoire of phonemes several experi-ments have shown that around 9ndash10 months of age in-fants use their knowledge of phonotactics to segmentwords (Mattys amp Jusczyk 2001) or to distinguish list ofwords differing on the probability of sequences of pho-nemes that constitute them (Jusczyk Luce amp Charles-Luce 1994) Thus at the end of the first year of life infantscertainly have access to phonological representationssimilar to those present in adults Are the same behav-ioral capacities in infants and adults sustained by thesame brain organization What is the initial state of or-

ganization of the human brain and how can it facilitatelanguage acquisition

METHODOLOGICAL ISSUES MISMATCHRESPONSES AS A TOOL TO EXPLOREAUDITORY REPRESENTATIONS

Event-related potentials (ERPs) are easily recorded evenin young children and elicitation of mismatch responsesin oddball paradigms is a powerful tool to access theproperties of the representations computed by thecortex from a sound It has been observed in single-cellrecordings in the visual (Miller Li amp Desimone 1991)and auditory cortex (Ulanovsky Las amp Nelken 2003)that the response of neurons decreases with repetitiveexposure to the same stimulus When a new or deviantstimulus is presented it elicits a response of a new set ofneurons Using this property called repetition suppres-sion it is possible to isolate cells that code differentproperties of the stimulus such as object identityindependently of its size or location (Miller et al1991) Of course the spatial resolution of brain imagingtechniques is too low to distinguish activity at the scaleof single neurons or columns of neurons that are thecoding elements of the nervous system However it ispossible to access this code at a macroscopic level bycontrasting the decreased activity recorded when thestimulus is repeated with the renewed activity due to theresponse of even a close set of neurons in the samebrain volume (see Naccache amp Dehaene 2001 for acomplete discussion of the priming method in brainimagery) This difference in activity is certainly thesource of mismatch responses recorded in ERP oddballparadigms By manipulating what elicits a mismatchresponse that is what counts or not as a repetitionfor a given neural network it is possible to infer theformat of representation that is computed by the net-work For example a network computing a phonemicrepresentation should habituate to the repetition of thesame phoneme independently of the speaker and re-cover when the phoneme is changed whereas a net-work coding for speaker should be affected by a voicechange but not by a phoneme change Although mis-match negativities (MMNs) observed in auditory oddballparadigms have been interpreted as a specific auditoryphenomenon related to a trace of the repeated stimulusin echoic memory (Naatanen 2001) these responses arebetter understood in this more general context ofrepetition suppression The experimental conditionsby selecting the level of representation that is targetedby the repetition context generate different MMNsrecorded with different topographies and latencies be-cause of the activation of different networks within thetemporal lobes Some MMNs are thus related to thecomputation of simple acoustic cues (eg durationintensity of the stimulus Giard et al 1995) but othersimply that multiple cues such as conjunction of features

1376 Journal of Cognitive Neuroscience Volume 16 Number 8

(Takegata Paavilainen Naatanen amp Winkler 1999)arbitrary rules (Horvath Czigler Sussman amp Winkler2001) lexical and grammatical status (Shtyrov amp Pulver-muller 2002 Pulvermuller et al 2001) have been inte-grated into more complex representations Beyond theidentification of the code computed by the networkwhich is obtained by manipulating the properties of therepeated stimulus crucial information can be obtainedabout the speed of the computation the brain areasinvolved and its temporal duration by measuring thelatency and the topography of the mismatch responsesas well as the maximal temporal spacing between stimulithat still permits the recording of a mismatch response

If similar representations are computed in infants andadults mismatch responses in both populations shouldpresent the same functional properties However weshould not expect to record the mismatch response atthe same latency and with the same topography ininfants than in adults As can be seen in Figure 1 thebrain response changes rapidly in morphology topog-raphy and latency during the first months of life (Kush-nerenko Ceponiene Balan Fellman Huotilaine et al2002 Novak Kurtzberg Kreuzer amp Vaughan 1989Barnet Ohlrich Weiss amp Shanks 1975 for a descriptionof auditory ERP maturation) because of changes withinauditory cortices such as myelinization differential mat-uration of the cortical layers folding of the corticalsurface and modifications of intracortical connectionsbut also to more general changes that may affect elec-trical transmission such as differential expansion of brainareas closure of the fontanel ossification of the skulland others For example N1 in adults is described as abroad negativity recorded 100 msec after a sound at thevertex and surrounding electrodes It has been relatedto the activation of numerous cerebral generators mainlyin the planum temporale (Eggermont amp Ponton 2002)If such a component is not recorded during the firstmonths of life it does not mean that sounds cannotaccess an immature auditory cortex Functional MRIstudies found activations to sound in cortical regionsthat are grossly similar in infants and in adults (Dehaene-Lambertz Dehaene amp Hertz-Pannier 2002 Altman ampBernal 2001 Anderson et al 2001) Therefore descrip-tive properties of ERP components such as latenciesand topographies determined from a few electrodes areoften not relevant to compare the cortical areas activat-ed and their computational properties at different agesWe should rather rely on an experimental design thattargets a precise computation on the stimulus and ondipole modeling of activated brain areas using high-density coverage of the scalp

Cerebral Bases of PhonemePerception in Adults

What are the cerebral bases of phoneme perception inadults Among the representations of sound that may be

computed and are accessible to mismatch paradigmscan we identify a phonemic representation A networkcoding such a representation should display the sameproperties as those described in behavioral paradigmsfor phoneme perception for example categorical per-ception Indeed a change of consonant occurring withina phonemic category elicits a significantly smaller MMNor no MMN than does a similar change that crossesa phonemic boundary (Sharma amp Dorman 1999 De-haene-Lambertz 1997) The MMN is also sensitive to thesubjectsrsquo native language The same change in a series ofsyllables induces an MMN in subjects for whom thechange is linguistically pertinent in their language butnot (or a significantly weaker one) in subjects for whomthe change is not linguistically pertinent (Sharma ampDorman 2000 Winkler et al 1999 Dehaene-Lambertz1997 Naatanen et al 1997) This effect of native lan-guage is not limited to the repertoire of phonemes butencompasses the entire phonology of the language Forexample phonotactics influence MMN responses Achange in pseudowords (eg ebuzo to ebzo) inducesan MMN in French subjects but not in Japanese subjects(Dehaene-Lambertz Dupoux amp Gout 2000) Finally thevisual integration of incongruent articulatory move-ments during phoneme perception induces a mismatchresponse even when the auditory stimulus itself doesnot change (Colin et al 2002 Sams et al 1991) Theseresults demonstrate that oddball paradigms are able toreveal a level of representation that displays propertiesrepresentative of phoneme perception namely categor-ical perception speaker normalization (Dehaene-Lam-bertz et al 2000) influence of the native language andvisuoauditory integration

By virtue of the temporal resolution provided by ERPswe can investigate whether acoustic and linguistic rep-resentations of a syllable are computed sequentially or inparallel When an MMN is present for an acoustic changein speech stimuli (eg a within-category change) itnever precedes a phonemic MMN (Phillips et al 2000Rivera-Gaxiola Csibra Johnson amp Karmiloff-Smith2000 Sharma Marsh amp Dorman 2000 Winkler et al1999 Dehaene-Lambertz 1997 Naatanen et al 1997)These results confirm Whalen and Libermanrsquos (1987)hypothesis that the representations computed fromspeech are immediately phonemic with no intermediateacoustic representation It is important to keep in mindthat syllables are not only linguistic but also acousticstimuli The very same dimensions for example VOT orformant transition shape are integrated by a phonemicnetwork into a phonemic representation but can alsobe computed by acoustic networks to create acousticrepresentations along these dimensions The joint rep-resentation of speech within both an acoustic represen-tation and a phonemic representation is exemplified bythe perception of sine wave stimuli that can be heardboth linguistically and nonlinguistically (LiebenthalBinder Piorkowski amp Remez 2003 Remez Pardo

Dehaene-Lambertz and Gliga 1377

Piorkowski amp Rubin 2001) Acoustic representations arealso used to discriminate within-category changes andforeign contrasts However these acoustic networks areless efficient and compute less stable memory traces thanthe phonemic network as demonstrated by our poorperformance in perceiving these changes It is also possi-ble that the phonemic network once activated by speechexerts an inhibitory influence over these auditory repre-sentations to prevent interference from nonlinguisticallypertinent differences (Liebenthal et al 2003 LibermanIsenberg amp Rakerd 1981 Dehaene-Lambertz PallierSerniclaes Sprenger-Charolle amp Dehaene submitted)

Where is the phonemic network located Dipolemodels of the phonemic MMN suggest that this re-sponse originates from the planum temporale and isasymmetric favoring the left hemisphere (Tervaniemiet al 1999 Naatanen et al 1997) This is congruent withneuropsychological observations that have establishedthat correct phoneme perception depends on the integ-rity of the posterior part of the left perisylvian regions(Dronkers Redfern amp Knight 2000 Caplan Gow ampMakris 1995) Brain imaging studies (Jacquemot PallierLeBihan Dehaene amp Dupoux 2003 VouloumanosKiehl Werker amp Liddle 2001 Binder et al 2000) anddiscrimination deficits elicited by electrical stimulation inpatients with implanted subdural electrodes arrays(Boatman Lesser amp Gordon 1995) have confirmedthe crucial role of the posterior and superior part ofthe left temporal lobe and of the left inferior parietalgyrus in phoneme perception in adults The asymmetryfavoring the left hemisphere for phoneme perceptionhas been related by some authors to the fact that the lefthemisphere is specialized to process stimuli with fasttemporal changes which are frequent in speech (Za-torre Belin amp Penhune 2002) Other authors suggest agenetic predisposition of the left hemisphere to processall aspects of language from phoneme perception tosyntax fMRI studies have shown that activation in someleft regions does not depend only on the physicalcharacteristics of the stimuli but rather on their linguisticvalue Significant differences are observed between sub-jects with different native languages in the left temporaland parietal lobe (Jacquemot et al 2003 Klein ZatorreMilner amp Zhao 2001) or in the left frontal region(Gandour et al 2002) Using fMRI with sine waveanalogues of ba da Dehaene-Lambertz et al (submit-ted) observed that brain activations are always signifi-cantly asymmetric favoring the left side even when sinewave stimuli were perceived as whistles confirming thatthe left hemisphere is better suited to process stimuliwith fast transitions However two brain areas in the lefthemisphere the posterior part of the superior temporalsulcus and the supramarginal gyrus were specificallyactivated when the sine waves were perceived as sylla-bles These results show that the acoustical properties ofthe signal are not sufficient to explain brain activationsto speech stimuli For equal stimulus characteristics

some networks in the left hemisphere are specificallycorrelated with the perception of stimuli as speech

To summarize ERP experiments in adults suggestthat several representations are computed in parallelgenerating different mismatch responses around 100ndash300 msec One of these corresponds to a phonemic rep-resentation that presents the same functional propertiesas those identified in behavioral studies of speechprocessing Neurospychology and brain imaging studieslocate this network in the left temporoparietal junction(Wernickersquos area) Can we isolate a network in infantsdemonstrating the same properties as the one presentin adults

Cerebral Bases of PhonemePerception in Infants

We have hypothesized that mismatch responses arerelated to a decrease in activity due to repetition Ininfants as in adults (Woods amp Elmasian 1986) repeti-tion of the same stimulus induces a decrease in ampli-tude of the event-related response (Dehaene-Lambertzamp Dehaene 1994) and this decrease in amplitude islarger when the interstimulus interval (ISI) is shorter(Leppanen Pihko Eklund amp Lyytinen 1999) The effectof repetition is similar across ages inducing a sharp andsignificant decrease between the first and second pre-sentation (Figure 1) When syllables belonging to thesame phonemic category are produced by differentspeakers the same decrease in amplitude is observedsuggesting that a normalization process is involved(Dehaene-Lambertz Pena Christophe amp Landrieu2004 Dehaene-Lambertz amp Pena 2001) and that aphonemic representation is accessed (Figure 1)

The introduction of a new stimulus induces a mis-match response that displays functional similarities withadultsrsquo MMN It is recorded even when attention is notdirected toward the stimuli either because infants arelooking at interesting visual stimuli to keep them quietduring recording (Dehaene-Lambertz amp Dehaene 1994)or because they are asleep (Dehaene-Lambertz amp Pena2001 Alho Saino Sajaniemi Reinikainen amp Naatanen1990) or even in a coma vigil (Dehaene-Lambertz et al2004) When babies are awake this response precedes alate frontal negativity beginning around 800 msec (Frie-derici Friedrich amp Weber 2002 Dehaene-Lambertz ampDehaene 1994 Kurtzberg Stone amp Vaughan 1986)Because this late negativity is recorded after unexpectedvisual and auditory stimuli (Nelson amp deRegnier 1992Kurtzberg et al 1986 Courchesne 1983) and is notpresent when babies are asleep (Friederici et al 2002)it is seen as reflecting general attention-orienting pro-cesses In adults the auditory mismatch response isfollowed by a P300 when subjects have to detect thechange of stimulus Although no study has recordedbehavioral responses and ERPs in the same session ininfants this two-step progression auditory mismatch


response followed by an amodal attentional componentis reminiscent of the two stages observed in adults aftera deviant attended stimulus (Figure 2)

As in adults there is not one but several mismatchresponses depending on the stimulus and on the featureof the stimulus that changes For example within thesame infants the response to a change in voice has adifferent topography than the response to a change inphoneme demonstrating that voice and phoneme cat-egory among other features of the stimulus have beencoded in parallel by different networks (Dehaene-Lam-bertz 2000) The mismatch response to syllables dis-plays properties essential for phoneme perception such

as normalization across speakers In neonates a similarmismatch response is found when the same physicalstimulus is repeated before the phonemic change andwhen each syllable is produced by a different speakerand thus is physically different (Dehaene-Lambertz ampPena 2001) This implies that the representation com-puted from the speech signal in this case is independentof the acoustic variations related to changes of intensityduration prosodic contour and timbre of voice and thatit is only affected by a change of phoneme

Categorical perception is demonstrated by the factthat when syllables varying along a place of articulation(ba to da) are used the mismatch response is larger

Figure 2 Grand average

recorded from a left frontal

electrode ( on the maps) in16 three-month-old infants

during an entire trial for the

standard (blue line) and

deviant condition (red line)A first mismatch response

originating from the temporal

lobes is recorded around

400 msec A second slow wavedevelops after 600 msec over

the frontal areas (adapted from

Dehaene-Lambertz ampDehaene 1994)

Figure 1 Repetition of the

same syllable induces a

decrease in ERPs in adults

(top) 3-month-old infants(middle) and neonates

(bottom) already significant

between the first and the

second syllable () Leftgrand-average across subjects

of recordings at the vertex For

neonates two waveforms arepresented showing similar

habituation when the syllable

is physically identical (blue

line) and when the syllable isproduced by different speakers

(red line) Right topographic

maps of the voltage at the

same time point (arrow on thewaveforms) for the first (S1)

and the third syllable (S3)

Although morphology andlatencies of the ERPS are very

different across ages the same

repetition priming effect is

observed (adapted fromDehaene-Lambertz 1997

Dehaene-Lambertz amp Baillet

1998 Dehaene-Lambertz amp

Pena 2001)


for a change that crosses the phonemic boundary (AC)than for a change of similar amplitude within thephonemic category (WC) (Dehaene-Lambertz amp Baillet1998) As in adults the mismatch response to the within-category change does not precede the mismatch re-sponse to across-category change but occurs at thesame latency Furthermore their topography differssuggesting that different brain areas are involved(Figure 3) The mismatch response to the within-cate-gory change is more central than the response to theacross-category change which is present above the rightfrontal and left occipital areas When a dipole model ofthese responses is computed the proposed sources arelocated more posterior and dorsal for the linguisticresponse (AC) than for the acoustical response (WC)(Dehaene-Lambertz amp Baillet 1998) In similar mismatchparadigms using fMRI adults display greater activationsduring phonemic discrimination than during acousticdiscrimination in the posterior part of the superiortemporal sulcus and in the adjacent parietal region(supramarginal and angular gyrus) ( Jacquemot et al2003 Dehaene-Lambertz et al submitted) The moreimportant contribution of posterior perisylvian areas inphonemic coding is thus compatible with the shifting ofthe source localization found in infants (Figure 3) Thisresult suggests that as in adults distinct acoustic andphonemic representations are computed in parallelfrom the same syllable

Dipole modeling of the mismatch response to sylla-bles locates the brain areas activated in infants in theposterior part of the temporal lobe (Dehaene-Lambertzamp Baillet 1998 Dehaene-Lambertz amp Dehaene 1994) Inadults this response is asymmetric originating mainlyfrom the left hemisphere In infants both ERPs to CVsyllables and mismatch responses are asymmetric favor-ing the left hemisphere but a similar asymmetry isobserved for acoustic processing For example thedifference in the timbre of two tones elicits an asym-metric mismatch response which is larger above the lefthemisphere (Dehaene-Lambertz 2000 Duclaux Challa-mel Collet Roullet-Solignac amp Revol 1991) In a dipolemodel the vector is longer on the left side than on theright side both for a within-category and an acrosscategory change (Figure 3) Thus the left auditory cor-tex seems to present a higher responsiveness than theright for auditory stimuli in general and not specificallyfor linguistic stimuli A similar pattern is observed withfMRI in 3-month-old infants (Dehaene-Lambertz et al2002) Forward but also backward speech activates theleft temporal areas more than the right and no interac-tion was observed between the linguistic pertinence ofthe stimuli and the hemisphere activated Howeverboth stimuli contain high-frequency transitions Usingoptical topography Sato et al (2003) compared thelateralization of the bold response in auditory areas toa pitch change (itta vs itta) and to a phonemechange (itta vs itte) during the first year of life A

difference in laterality between the two conditions wasnot found until the age of 11ndash12 months on Morestudies with high spatial resolution such as fMRI oroptical topography are needed to confirm whether theleft hemisphere is more responsive than the right to anysound or only to sounds with fast transitions Thehigher responsiveness of the left hemisphere to soundsduring the first months of life while babies are inten-sively learning their native language might be onedevelopmental bias that pressures language processingtoward the left hemisphere This leftward asymmetrydescribed in normal infants appears to be more a biasthan to be related to incapacity of the right hemisphereto process syllables Patient LG suffers from a left sylvianinfarct that occurred at birth Tested 3 weeks after theacute lesion her right hemisphere was able to discrim-inate a change of phoneme even when several speakersproduced the stimuli (Dehaene-Lambertz et al 2004)On the other hand patient SD with an analogous righthemisphere lesion was easily able to discriminate achange in timbre with his left hemisphere In adultssuch a lesion would have induced a severe deficit in thistimbre discrimination task based on spectral computa-tion (Figure 4) These results suggest a different lateral-ization pattern in infants and adults

The influence of native language on phoneme per-ception is not noticeable in behavior before 5ndash6 monthsof age A phonemic mismatch response should beaffected in a similar way Only one published experimenthas explored the perception of foreign phonemic con-trasts with ERPs Cheour Ceponiene et al (1998) re-corded the response to a change of vowel from e to oand o in Estonian and Finnish infants o is not presentin Finnish Finnish adults contrary to Estonians show asmaller MMN for o than for o although the acousticdistance is wider for the change endasho than for endashoAt 1 year of age the mismatch response recorded for theforeign vowel o is smaller in Finns than in Estonianswhereas at 6 months the response is similar in bothpopulations In adults the duration of the mismatchresponse is strongly correlated with discrimination per-formance It is thus surprising to record a similar mis-match response at 6 months in Finns and Estoniansespecially considering that numerous behavioral experi-ments have shown that vowel perception is alreadyaffected by native language at this age (Polka amp Werker1994 Kuhl et al 1992) However we do not havebehavioral data on the precise contrast used in Cheouret alrsquos electrophysiological experiment The typical de-velopment often presented in the literature (effects ofnative language at 6 months for vowel perception and at10 months for consonant perception) may be oversim-plified Some contrasts may be obscured at differentages depending on the adjacent phonemes in thephonemic space of the native language or on therobustness of the phonemic boundary (Maye Werkeramp Gerken 2002 Best McRoberts amp Sithole 1988)


Figure 4 Mismatch re-

sponses for a phoneme discri-

mination task and a timbrediscrimination task in normal

2-month-old infants and in two

patients SD and LG These two

babies had suffered from animportant hemispheric lesion

due to a sylvian infarct at birth

on the left side for LG and on

the right side for SD Despitetheir lesions they show a

mismatch response in the

contralateral healthyhemisphere (adapted from

Dehaene-Lambertz et al 2004

Dehaene-Lambertz amp Pena

2001 for the normal infantsDehaene-Lambertz 1997 for

LG and unpublished data

for SD)

Figure 3 Categorical

perception in 3-month-old

infants along a synthetic

continuum (ba to da)Control (ba1 ba1 ba1

ba1) within-category change

(ba2 ba2 ba2 ba1) and

across-category change (dada da ba1) trials were

randomly presented The

physical distance was similaralong the continuum for the

within-category change and

the across-category change

(top) Voltage cartographies ofthe same test syllable (ba1) at

the maximum of the mismatch

response (middle) and dipole

models of the active brainregions (bottom) for the three

conditions The dipole for the

across-category change is moreposterior and dorsal than

for the within-category

change (adapted from

Dehaene-Lambertz amp Baillet1998)


Second it is possible that the mismatch response was infact already differentiable in both populations at6 months but that this was not observed because thespatial sampling of the voltage on the scalp was scarce(only Cz and Fz were recorded) and because the acous-tical distance of the change in this experiment wasconfounded with its linguistic pertinence Further ex-periments testing other phonemic contrasts and com-bining behavioral and ERP results should be carried outto clarify these points However this result shows atleast that the difference in behavior observed acrosssubjects from different linguistic backgrounds fromthe end of the first year of life on is based on a differ-ence in the early representations computed from thespeech signal

Differences Between Infantsrsquo and AdultsrsquoMismatch Responses

Using the same ERP paradigm in infants as in adults wehave observed a mismatch response to a change ofsyllable that displays similar functional properties inboth populations categorical perception and normali-zation It is elicited automatically even when infants areasleep and this response precedes an amodal frontalresponse when infants are awake reminiscent of what isobserved in adults processing an attended or unattend-ed stimulus Dipole modeling suggests a temporal originas in adults Although the functional properties ofmismatch responses and their corresponding activebrain areas appear to be similar in infants and adultssuggesting similar phonemic representations it shouldbe noted that the latency of the first mismatch effect isdelayed in infants relative to adults For instance themaximum of the mismatch response for the samestimuli ba da is at 280 msec in adults versus 454 msecin 3-month-old babies (Dehaene-Lambertz 1997 De-haene-Lambertz amp Baillet 1998) Secondly in most ofthe experiments cited above the polarity of the mis-match response is reversed Whereas in adults themismatch response is described as a frontal negativity(from which the name mismatch negativity arises) witha positivity over the temporal regions in our experimentwith infants the mismatch response is positive overfrontal areas and negative over temporo-occipital areasalong a similar right frontalndashleft posterior axis as inadults (Dehaene-Lambertz 1997 Dehaene-Lambertz ampBaillet 1998) Several other authors have also recorded asimilar mismatch frontal positivity in infants (Winkleret al 2003 Friederici et al 2002 Morr Shafer Kreuzeramp Kurtzberg 2002 Pihko et al 1999 Duclaux et al1991 Alho Sajaniemi Niittyvuopio Saino amp Naatanen1990) and in toddlers (Maurer Bucher Brem amp Bran-deis 2003) We have already discussed the hypothesisthat activity of the same networks might result indifferent voltage topographies on the scalp across agesbecause of the structural changes occurring in the brain

during development However some authors have ob-served frontal negativities (Cheour Alho et al 1998Cheour Ceponiene et al 1998 Cheour et al 2002) andhave suggested a strict similarity of the mismatch re-sponse across ages (Cheour Alho et al 1998) Thesenegativities however are usually very long severalhundred milliseconds and do not show the same sharptemporal profile as in adults (eg compare the 100-msecduration of the MMN for a vowel contrast in adultsand the 400-msec duration for the same contrast in 12-month-old infants (Cheour Ceponiene et al 1998Naatanen et al 1997) Moreover shortcomings that donot facilitate the interpretation of the results across agesare sometimes present in infant studies First the refer-ence electrode affects the visualization of ERPs and itshould be chosen to be as neutral as possible either byusing a distant electrode that is not affected by theactivity of the voltage generators or by computing anaverage reference if a large number of electrodes is usedIn the latter case subtracting the average voltage fromthe voltage recorded at each electrode creates refer-ence-free voltages (Bertrand Perrin amp Pernier 1985)However EEG recordings in infants have often useda mastoid reference As can be seen in Figure 1 theauditory average reference ERP is negative above themastoid in infants and the mastoid should not beconsidered as neutral Thus a comparison across agesshould be made cautiously when the reference is locatedon an active site Second in order to measure a mis-match response in classical oddball designs the re-sponse to a deviant presented 20 of the time iscompared with the same stimulus presented in anotherblock 100 of the time The block effect is thus con-founded with time Time could induce habituation of low-level processing or changes in listening strategy evenduring passive tasks This could be worse in childrenwho grow increasingly restless as time passes with pos-sible increases in movement artifacts changes in atten-tion and others Often this second block is not recordedin infants and the comparison is thus done betweendifferent standard and deviant stimuli within the firstblock In this case if a difference is recorded it couldbe related either to a genuine discrimination process or todifferences in ERP morphologies and latencies due to thephysical differences between the two stimuli For exam-ple the latency and the morphology of N1 are affected byVOT A comparison between two stimuli of a VOT con-tinuum would show differences but these are not sys-tematically associated with the perception of a difference(Sharma et al 2000) The morphology of infantsrsquo ERPs isalso affected by the physical characteristics of the stimu-lus such as its duration (Kushnerenko Ceponiene BalanFellman amp Naatanen 2002) its intensity envelope (De-haene-Lambertz et al 2004) the tone onset time (Simosamp Molfese 1997) or the VOT (Kurtzberg Stapells ampWallace 1988) These should not be neglected therecording of a difference between physically different


standard and deviant stimuli is not sufficient to establishthe existence of a mismatch response3

Vigilance attention toward the stimuli number ofrepetitions of the standard before a deviant size of thedeviance random or predictable presentation of thedeviant and the maturation stage of the network itselftargeted by the experimental paradigm are other factorsthat differ across infant experiments For instance asalient or predictable deviant might induce more inter-actions between frontal selective attention mechanismsand auditory representations Networks computing com-plex cues in which numerous assemblies of neurons aresynchronized might imply larger cortico-cortical infor-mation flow within the network than within networkscomputing simple cues especially during learning peri-ods Polarity of ERPs depending on the weight of activityin the different cortical layers and increase of activity incortico-cortical connections might have an effect on thepolarity of the response In any case mismatch negativ-ity should not be considered as a flag that the brainexhibits when an auditory change is perceived and thatinfants possess or not but mismatch responses shouldbe understood as the result of complex interactionsdetermined by the experimental conditions betweeninput and output information flow within assemblies ofneurons in a developing and learning brain Furtherstudies are needed in which all these factors should besystematically manipulated to understand the importantcontributing factors to record one or the other polarityover the frontal areas

In summary infants and adults exhibit similar behav-ioral performance in phoneme perception ERPs revealthat this performance is subserved by similar computa-tions of temporal neurons in both populations There-fore we hypothesize that in the case of phonemeprocessing there is continuity between neonates andadults and that from birth on infants are able to spon-taneously compute phonemic representations At thebeginning of the life span these phonemic representa-tions are universal4 This phonemic network effectivefrom the first days of life is adequately configured toprocess the relevant properties of the speech environ-ment and to detect any inherent regularities present ininput Exposed to a specific linguistic environment theweights in the network are then modified to give rise tothe adult phonemic representations without explicitreinforcement A plausible explanation of this environ-mental effect is that infants exploit the statistical prop-erties of language input (Maye et al 2002 Kuhl 2000Holt Lotto amp Kluender 1998) Each language uses onlya reduced repertoire out of the set of all possiblephonemes and productions of native speakers clusteraround prototypes For example Lisker and Abramson(1964) reported that although variability exists withinand between speakers VOT values for the production ofa particular phoneme tend to cluster around a meanvalue If a phonemic feature is contrastive tokens

present a bimodal distribution along this specific dimen-sion whereas if it is not contrastive the distribution isunimodal (Maye et al 2002) This particular distributionmay distort the initial perceptive space increasing theresponse to a change of category and decreasing theresponse to changes around the mean value The resultwould be a magnet effect as suggested by Kuhl (2000)which would lead to a reduction of mismatch responsesto foreign contrast as observed in adults and in infants asearly as 1 year of age In a recent article Kuhl et al(2003) showed that mere exposure to foreign speech isnot in itself sufficient to modify the perceptual spaceSocial interactions are also necessary either becausesocial interactions direct infantsrsquo attention to the visualcues of speech production and thus consolidate phone-mic representations through cross modal integration orbecause a situation of social interaction may by itselfactivate linguistic networks given that speech is themain communication medium in our species

An argument often presented against a phonemicnetwork is that animals show perceptual discontinuitiessimilar to those observed in humans For example amismatch response is recorded to a change in VOT fromthe brainstem of guinea pigs (King McGee Rubel Nicolamp Kraus 1995) and a common boundary along the VOTdimension is found in humans macaques (Kuhl ampPadden 1983) and chinchillas (Kuhl amp Miller 1975)There is no reason to believe that if a strong acousticdiscontinuity exists in perception a linguistic networkshould ignore it However phonemic perception is notan exact mirror of acoustic perception Using sine waveanalogues of syllables Serniclaes Sprenger-CharollesCarre and Demonet (2001) explored discriminationfunctions along a bandashda continuum in children whenthe stimuli were perceived as electronic glissando andwhen they were perceived as speech The location of thecategorical boundary differed by one step between thespeech mode and the nonspeech mode of perceptionAlong an rndashl continuum humans and monkeys havedifferent discrimination performance (Sinnott amp Brown1997) whereas 2-month-old infants and English adultsshow the same categorical boundary (Eimas 1975) Onthe other hand because hearing is demonstrated duringthe last trimester of pregnancy it is possible that theproperties demonstrated in infants are the consequen-ces of exposure to speech However the characteristicsof the fetus environment would predict rather differentproperties for a learning network The predominance ofthe mother voice inside the womb would favor preciserepresentations of her productions and difficulties tonormalize across different speakers The perception ofsome phonemic contrasts such as place of articulationis very sensitive to noise Recording within the uterus ofa pregnant sheep Griffiths Brown Gerhardt Abramsand Morris (1994) show that intelligibility was not goodalong this dimension However neonates have no prob-lem discriminating phonemes such as pa and ta


(Dehaene-Lambertz amp Pena 2001) Infants are also ableto discriminate foreign contrasts not present in theirenvironment (Best et al 1988) These facts suggest thatit is not exposure to speech that creates the capabilitiesdescribed in infants even if the environment is certainlyalready shaping the phonemic representations

In this article we have emphasized the similaritiesbetween infants and adults suggesting that phonemeperception relies on the same neural bases in infants andadults However we are far from understanding theprecise mechanisms that govern the effect of the envi-ronment on this network We do not know for thepresent how much exposure to a specific contrast isnecessary how long these environmental effects lastwithout requiring new exposure to speech whethervisual cues are important whether social interactionshave the same impact in the first months of life and lateron or which exact statistical properties infants aresensitive to Beyond phoneme perception which wehave discussed here the recent progress in brain imag-ing techniques gives us access to the organization of thehuman brain from very early on and will permit us tobetter understand the ontogenesis of higher cognitivefunctions in the human brain

Acknowledgments

This study was supported by grants from the Ministere Francaisde la Sante et de la Recherche PHRC 1995 No AOM95011Groupement drsquoInteret Scientifique Sciences de la CognitionNo PO 9004 ACI Blanche 1999 lsquolsquoPlasticiteneuronale etacquisition du langagersquorsquo and a McDonnell Foundation grant

Reprint requests should be sent to Ghislaine Dehaene-Lambertz Unite INSERM 562 Service Hospitalier FredericJoliot CEADRMDSV 4 place du general Leclerc 91401 Orsaycedex France or via e-mail ghislscpehessfr

Notes

1 Linguists distinguish phones (which are physical catego-ries) and phonemes (which are functional classes that distin-guish words) Here we use phoneme in a psychological sensethat is as a perceptive category2 The MacGurk effect is related to the visual integration ofincongruent articulatory movements during phoneme percep-tion that induces a change of auditory perception although theauditory stimulus itself does not change The ventriloquy effectis the attribution of the source of a sound to a person althoughthe sound originates from a different location in space Duplexperception is the integration of two stimuli into one singlepercept although they are presented separately in each ear3 To avoid these problems we have adapted the classicaloddball paradigm Stimuli are presented in trials of four stimulieach Two types of trials standard and deviant are randomlypresented In standard trials the four stimuli are identical (A AA A) In deviant trials the last stimulus of the trial is differentfrom the first three (B B B A) Across trials A and B arerandomly swapped both stimuli being alternatively standardand deviant Because the waveforms from the same syllable arecompared the observed difference can only be related to thecontext in which the last syllable was presented and to the

distance between this syllable and the context This design hasseveral important advantages over the classical oddball designThe introduction of the deviant stimulus is better controlled (itis always preceded by three standard stimuli) The response tothe same stimulus at the same position in the trial is comparedin standard and deviant trials and the same number of trials isaveraged in each condition Because from trial to trial thecontext and test stimulus are exchanged we can compare ERPsto the same stimulus when it is presented as standard and asdeviant and at a short time interval4 Universal in the sense that the same performance is shownby every human neonate and not in the sense that everyphonemic boundary present across all languages of the worldis already present at birth

REFERENCES

Alho K Saino K Sajaniemi N Reinikainen K amp NaatanenR (1990) Event-related potential of human newbornsto pitch change of an acoustic stimulusElectroencephalography and Clinical Neurology77 151ndash155

Alho K Sajaniemi N Niittyvuopio T Saino K amp NaatanenR (1990) ERPs to an auditory stimulus change in pre-termand full-terms infants Psychophysiological Brain Research2 139ndash142

Altman N R amp Bernal B (2001) Brain activation in sedatedchildren Auditory and visual functional MR imagingRadiology 221 56ndash63

Anderson A W Marois R Colson E R Peterson B SDuncan C C amp Ehrenkranz R A (2001) Neonatal auditoryactivation detected by functional magnetic resonanceimaging Magnetic Resonance Imaging 19 1ndash5

Barnet A B Ohlrich E S Weiss I P amp Shanks B (1975)Auditory evoked potentials during sleep in normal childrenfrom ten days to three years of age Electroencephalographyand Clinical Neurophysiology 39 29ndash41

Bertoncini J Bijeljac-Babic R Jusczyk P W Kennedy Lamp Mehler J (1988) An investigation of young infantsrsquoperceptual representations of speech sounds Journal ofExperimental Psychology General 117 21ndash33

Bertrand O Perrin F amp Pernier J (1985) A theoreticaljustification of the average reference in topographic evokedpotential studies Electroencephalography and ClinicalNeurophysiology 62 462ndash464

Best C T McRoberts G W amp Sithole N M (1988)Examination of the perceptual reorganization for speechcontrasts Zulu click discrimination by English-speakingadults and infants Journal of Experimental PsychologyHuman Perception and Performance 3 345ndash360

Binder J R Frost J A Hammeke T A Bellgowan P SSpringer J A amp Kaufman J N (2000) Human temporallobe activation by speech and nonspeech sounds CerebralCortex 10 512ndash528

Boatman D Lesser R P amp Gordon B (1995) Auditoryspeech processing in the left temporal lobe An electricalinterference study Brain and Language 51 269ndash290

Bushnell I W R (1982) Discrimination of faces by younginfants Journal of Experimental Child Psychology33 298ndash308

Caplan D Gow D amp Makris N (1995) Analysis of lesions byMRI in stroke patients with acousticndashphonetic processingdeficits Neurology 45 293ndash298

Cheour M Alho K Ceponiene R Reinikainen K Sainio KPohjavuori M Aaltonen O amp Naatanen R (1998)Maturation of mismatch negativity in infants InternationalJournal of Psychophysiology 29 217ndash226


Cheour M Ceponiene R Lehtokoski A Luuk A Allik JAlho K amp Naatanen R (1998) Development oflanguage-specific phoneme representations in the infantbrain Nature Neuroscience 1 351ndash353

Cheour M Martynova O Naatanen R Erkkola R SillanpaaM Kero P Raz A Kaipio M L Hiltunen J Aaltonen OSavela J amp Hamalainen H (2002) Speech sounds learnedby sleeping newborns Nature 415 599ndash600

Colin C Radeau M Soquet A Demolin D Colin F ampDeltenre P (2002) Mismatch negativity evoked by theMcGurkndashMacDonald effect A phonetic representationwithin short-term memory Clinical Neurophysiology 113495ndash506

Courchesne E (1983) Cognitive components of theevent-related brain potential Changes associated withdevelopment In A W K Gaillard amp W Ritter (Eds)Tutorials in event-related potential research Endogenouscomponents (pp 329ndash344) North-Holland Amsterdam

Dehaene-Lambertz G (1997) Electrophysiological correlatesof categorical phoneme perception in adults NeuroReport8 919ndash924

Dehaene-Lambertz G (2000) Cerebral specialization forspeech and non-speech stimuli in infants Journal ofCognitive Neuroscience 12 449ndash460

Dehaene-Lambertz G amp Baillet S (1998) A phonologicalrepresentation in the infant brain NeuroReport 91885ndash1888

Dehaene-Lambertz G amp Dehaene S (1994) Speed andcerebral correlates of syllable discrimination in infantsNature 370 292ndash295

Dehaene-Lambertz G Dehaene S amp Hertz-Pannier L(2002) Functional neuroimaging of speech perception ininfants Science 298 2013ndash2015

Dehaene-Lambertz G Dupoux E amp Gout A (2000)Electrophysiological correlates of phonological processingA cross-linguistic study Journal of Cognitive Neuroscience12 635ndash647

Dehaene-Lambertz G Pallier C Serniclaes WSprenger-Charolle L amp Dehaene S Neural correlates ofswitching from auditory to speech perception Manuscriptsubmitted for publication

Dehaene-Lambertz G amp Pena M (2001) Electrophysiologicalevidence for automatic phonetic processing in neonatesNeuroReport 12 3155ndash3158

Dehaene-Lambertz G Pena M Christophe A amp Landrieu P(2004) Phonetic processing in a neonate with a left sylvianinfarct Brain and Language 88 26ndash38

Dronkers N F Redfern B B amp Knight R T (2000) Theneural architecture of language disorders In M S Gazzaniga(Ed) The new cognitive neurosciences (pp 949ndash958)Cambridge MIT Press

Duclaux R Challamel M J Collet L Roullet-Solignac I ampRevol M (1991) Hemispheric asymmetry of late auditoryevoked response induced by pitch changes in infantsInfluence of sleep stages Brain Research 566152ndash158

Dupoux E Kakehi K Hirose Y Pallier C amp Mehler J(1999) Epenthetic vowels in Japanese A perceptual illusionJournal of Experimental Psychology Human Perceptionand Performance 25 1568ndash1578

Eggermont J J amp Ponton C W (2002) The neurophysiologyof auditory perception From single units to evokedpotentials Audiology amp Neuro-otology 7 71ndash99

Eimas P D (1975) Auditory and phonetic coding of the cuesfor speech Discrimination of the rndashl distinction by younginfants Perception and Psychophysics 18341ndash347

Eimas P D amp Miller J L (1992) Organization in the

perception of speech by young infants PsychologicalScience 3 340ndash345

Eimas P D Siqueland E R Jusczyk P W amp Vigorito J(1971) Speech perception in infants Science 171303ndash306

Feigenson L Carey S amp Spelke E (2002) Infantsrsquodiscrimination of number vs continuous extentCognitive Psychology 44 33ndash66

Fowler C A amp Rosenblum L D (1990) Duplex perception Acomparison of monosyllables and slamming doors Journalof Experimental Psychology Human Perception andPerformance 16 742ndash754

Friederici A D Friedrich M amp Weber C (2002) Neuralmanifestation of cognitive and precognitive mismatchdetection in early infancy NeuroReport 13 1251ndash1254

Gandour J Wong D Lowe M Dzemidzic MSatthamnuwong N amp Tong Y (2002) A cross-linguisticfMRI study of spectral and temporal cues underlyingphonological processing Journal of CognitiveNeuroscience 14 1076ndash1087

Giard M H Lavikainen J Reinikainen R Perrin FBertrand O amp Pernier J (1995) Separate representationsof stimulus frequency intensity and duration in auditorysensory memory An event-related potential anddipole-model analysis Journal of Cognitive Neuroscience7 133ndash143

Goto H (1971) Auditory perception by normal Japaneseadults of the sounds lsquolsquoLrsquorsquo and lsquolsquoRrsquorsquo Neuropsychologia 9317ndash323

Griffiths S K Brown W S Gerhardt K J Abrams R Mamp Morris R J (1994) The perception of speech soundsrecorded within the uterus of a pregnant sheep Journalof the Acoustical Society of America 96

Hauser M D Dehaene S Dehaene-Lambertz G amp PatalanoA (2002) Spontaneous number discrimination ofmulti-format auditory stimuli in cotton-top tamarins(Saguinus oedipus) Cognition 86 B23ndashB32

Holt L L Lotto A J amp Kluender K R (1998) Incorporatingprinciples of general learning in theories of languageacquisition In M Gruber C D Higgins K S Olson ampT Wysocki (Eds) Chicago linguistic society (pp 253ndash268)Chicago Chicago Linguistic Society

Horvath J Czigler I Sussman E amp Winkler I (2001)Simultaneously active pre-attentive representations oflocal and global rules for sound sequences in the humanbrain Brain Research Cognitive Brain Research 12131ndash144

Jacquemot C Pallier C LeBihan D Dehaene S amp DupouxE (2003) Phonological grammar shapes the auditory cortexA functional magnetic resonance imaging study Journal ofNeuroscience 23 9541ndash9546

Jusczyk P W Luce P A amp Charles-Luce J (1994) Infantsrsquosensitivity to phonotactic patterns in the native languageJournal of Memory and Language 33 630ndash645

Jusczyk P W Pisoni D B amp Mullennix J (1992) Someconsequences of stimulus variability on speech processingby 2-month-old infants Cognition 43 253ndash291

King C McGee T Rubel E W Nicol T amp Kraus N (1995)Acoustic features and acoustic change are represented bydifferent central pathways Hearing Research 8545ndash52

Klein D Zatorre R J Milner B amp Zhao V (2001) Across-linguistic PET study of tone perception in MandarinChinese and English speakers Neuroimage 13646ndash653

Kuhl P K (1983) Perception of auditory equivalenceclasses for speech in early infancy Infant Behavior andDevelopment 6 263ndash285


Kuhl P K (2000) A new view of language acquisitionProceedings of the National Academy of Sciences USA 9711850ndash11857

Kuhl P K amp Meltzoff A N (1984) The intermodalrepresentation of speech in infants Infant Behavior andDevelopment 7 361ndash381

Kuhl P K amp Miller J D (1975) Speech perception by thechinchilla Voicedndashvoiceless distinction in alveolar plosiveconsonants Science 190 69ndash72

Kuhl P K amp Padden D M (1983) Enhanced discriminabilityat the phonetic boundaries for the place feature inmacaques Journal of the Acoustical Society of America73 1003ndash1010

Kuhl P K Tsao F M amp Liu H M (2003) Foreign-languageexperience in infancy Effects of short-term exposure andsocial interaction on phonetic learning Proceedings of theNational Academy of Sciences USA 100 9096ndash9106

Kuhl P K Williams K A Lacerda F Stevens K N ampLindblom B (1992) Linguistic experiences alter phoneticperception in infants by 6 months of age Science 255606ndash608

Kurtzberg D Stapells D R amp Wallace I F (1988)Event-related potential assessment of auditory systemintegrity Implications for language development In P MVietze amp H G Vaughan (Eds) Early identification ofinfants with developmental disabilities (pp 160ndash179)Boston Allyn and Bacon

Kurtzberg D Stone C L amp Vaughan H G (1986) Corticalresponses to speech sounds in the infant In R Craccoamp I Bodis-Wollner (Eds) Frontiers of ClinicalNeuroscience Vol 3 Evoked potentials (pp 513ndash520)New York Alan R Liss

Kushnerenko E Ceponiene R Balan P Fellman VHuotilaine M amp Naatanen R (2002) Maturation of theauditory event-related potentials during the first year oflife NeuroReport 13 47ndash51

Kushnerenko E Ceponiene R Balan P Fellman V ampNaatanen R (2002) Maturation of the auditory changedetection response in infants A longitudinal ERP studyNeuroReport 13 1843ndash1848

Leppanen P H Pihko E Eklund K M amp Lyytinen H(1999) Cortical responses of infants with and without agenetic risk for dyslexia II Group effects NeuroReport10 969ndash973

Liberman A M Isenberg D amp Rakerd B (1981) Duplexperception of cues for stop consonants Evidence for aphonetic mode Perception amp Psychophysics 30 133ndash143

Liebenthal E Binder J R Piorkowski R L amp Remez R E(2003) Short-term reorganization of auditory analysisinduced by phonetic experience Journal of CognitiveNeuroscience 15 549ndash558

Lisker L amp Abramson A S (1964) A cross-language study ofvoicing in initial stops Acoustical measurements Word 20384ndash482

Mattys S L amp Jusczyk P W (2001) Phonotactic cues forsegmentation of fluent speech by infants Cognition 7891ndash121

Maurer U Bucher K Brem S amp Brandeis D (2003)Development of the automatic mismatch response Fromfrontal positivity in kindergarten children to the mismatchnegativity Clinical Neurophysiology 114 808ndash817

Maye J Werker J F amp Gerken L (2002) Infant sensitivityto distributional information can affect phoneticdiscrimination Cognition 82 B101ndashB111

Mehler J Jusczyk P Lambertz G Halsted N Bertoncini Jamp Amiel-Tison C (1988) A precursor of languageacquisition in young infants Cognition 29 143ndash178

Miller E K Li L amp Desimone R (1991) A neural mechanism

for working and recognition memory in inferior temporalcortex Science 254 1377ndash1379

Morr M L Shafer V L Kreuzer J A amp Kurtzberg D (2002)Maturation of mismatch negativity in typically developinginfants and preschool children Ear and Hearing 23118ndash136

Naatanen R (2001) The perception of speech sounds by thehuman brain as reflected by the mismatch negativity (MMN)and its magnetic equivalent (MMNm) Psychophysiology 381ndash21

Naatanen R Lehtokovski A Lennes M Cheour MHuotilainen M amp Iivonen A (1997) Language-specificphoneme representations revealed by electric and magneticbrain responses Nature 385 432ndash434

Naccache L amp Dehaene S (2001) The priming methodImaging unconscious repetition priming reveals an abstractrepresentation of number in the parietal lobes CerebralCortex 11 966ndash974

Nelson C A amp deRegnier R (1992) Neural correlates ofattention and memory in the first year of life DevelopmentalNeuropsychology 8 119ndash134

Novak G P Kurtzberg D Kreuzer J A amp Vaughan J H G(1989) Cortical responses to speech sounds ant theirformants in normal infants Maturational sequences andspatiotemporal analysis Electroencephalography andClinical Neurophysiology 73 295ndash305

Pallier C Bosch L amp Sebastian N (1997) A limit onbehavioral plasticity in speech perception Cognition64 B9ndashB17

Parr L A Winslow J T Hopkins W D amp de Waal F B(2000) Recognizing facial cues Individual discriminationby chimpanzees (Pan troglodytes) and rhesus monkeys(Macaca mulatta) Journal of Comparative Psychology114 47ndash60

Patterson M L amp Werker J F (2003) Two-month oldinfants match phonetic information in lips and voiceDevelopmental Science 6 193ndash198

Phillips C Pellathy T Marantz A Yellin E Wexler K ampPoeppel D (2000) Auditory cortex accesses phonologicalcategories An MEG mismatch study Journal of CognitiveNeuroscience 12 1038ndash1055

Pihko E Leppanen P H Eklund K M Cheour M GuttormT K amp Lyytinen H (1999) Cortical responses of infantswith and without a genetic risk for dyslexia I Age effectsNeuroReport 10 901ndash905

Pisoni D B (1977) Identification and discrimination of therelative onset time of two components tones Implicationsfor voicing perception in stops Journal of the AcousticalSociety of America 61 1352ndash1361

Polka L amp Werker J F (1994) Developmental changes inperception of nonnative vowel contrasts Journal ofExperimental Psychology Human Perception andPerformance 20 421ndash435

Pulvermuller F Kujala T Shtyrov Y Simola J Tiitinen Hamp Alku P (2001) Memory traces for words as revealed bythe mismatch negativity Neuroimage 14 607ndash616

Ramus F Hauser M D Miller C Morris D amp Mehler J(2000) Language discrimination by human newborns andby cotton-top tamarin monkeys Science 288349ndash351

Remez R E Pardo J S Piorkowski R L amp Rubin P E(2001) On the bistability of sine wave analogues of speechPsychological Science 12 24ndash29

Rivera-Gaxiola M Csibra G Johnson M H ampKarmiloff-Smith A (2000) Electrophysiological correlatesof cross-linguistic speech perception in native Englishspeakers Behavioural Brain Research 111 13ndash23

Rosenblum L D Schmuckler M A amp Johnson J A (1997)


The McGurk effect in infants Perception and Psychophysics59 347ndash357

Sams M Aulanko R Halalainen M Hari M LounasmaaO V amp Lu S T (1991) Seeing speech Visual informationfrom lip movement modified activity in the human auditorycortex Neuroscience Letters 127 141ndash145

Sato Y Mori K Furuya I Hayashi R Minagawa-Kawai Yamp Koizumi T (2003) Developmental changes in cerebrallateralization to spoken language in infants Measured bynear-infrared spectroscopy The Japan Journal ofLogopedics and Phoniatrics 44 165ndash171

Serniclaes W Sprenger-Charolles L Carre R amp DemonetJ F (2001) Perceptual discrimination of speech sounds indevelopmental dyslexia Journal of Speech Language andHearing Research 44 384ndash399

Sharma A amp Dorman M F (1999) Cortical auditory evokedpotential correlates of categorical perception of voice-onsettime Journal of the Acoustical Society of America 1061078ndash1083

Sharma A amp Dorman M F (2000) Neurophysiologiccorrelates of cross-language phonetic perception Journal ofthe Acoustical Society of America 107(5 Pt 1) 2697ndash2703

Sharma A Marsh C M amp Dorman M F (2000) Relationshipbetween N1 evoked potential morphology and theperception of voicing Journal of the Acoustical Society ofAmerica 108 3030ndash3035

Shtyrov Y amp Pulvermuller F (2002) Memory traces forinflectional affixes as shown by mismatch negativityEuropean Journal of Neuroscience 15 1085ndash1091

Simos P G amp Molfese D L (1997) Electrophysiologicalresponses from a temporal order continuum in the newborninfant Neuropsychologia 35 89ndash98

Sinnott J M amp Brown C H (1997) Perception of theAmerican English liquid randashla contrast by humans andmonkeys Journal of the Acoustical Society of America 102588ndash602

Takegata R Paavilainen P Naatanen R amp Winkler I (1999)Independent processing of changes in auditory singlefeatures and feature conjunctions in humans as indexed bythe mismatch negativity Neuroscience Letters 266109ndash112

Tervaniemi M Kujala A Alho K Virtanen J IlmoniemiR J amp Naatanen R (1999) Functional specialization of thehuman auditory cortex in processing phonetic and musicalsounds A magnetoencephalographic (MEG) studyNeuroimage 9 330ndash336

Ulanovsky N Las L amp Nelken I (2003) Processing oflow-probability sounds by cortical neurons NatureNeuroscience 6 391ndash398

Vouloumanos A Kiehl K A Werker J F amp Liddle P F(2001) Detection of sounds in the auditory streamEvent-related fMRI evidence for differential activation tospeech and nonspeech Journal of Cognitive Neuroscience13 994ndash1005

Werker J F amp Tees R C (1984a) Cross-language speechperception Evidence for perceptual reorganisation duringthe first year of life Infant Behavior and Development 749ndash63

Werker J F amp Tees R C (1984b) Phonemic and phoneticfactors in adult cross-language speech perception Journalof the Acoustical Society of America 75 1866ndash1878

Werker J F Gilbert J H V Humphrey K amp Tees R C(1981) Developmental aspects of cross-language speechperception Child Development 52 349ndash355

Whalen D H amp Liberman A M (1987) Speech perceptiontakes precedence over nonspeech perception Science 237169ndash171

Winkler I Kujala T Tiitinen H Sivonen P Alku PLehtokoski A (1999) Brain responses reveal the learning offoreign language phonemes Psychophysiology 36638ndash642

Winkler I Kushnerenko E Horvath J Ceponiene RFellman V amp Huotilainen M (2003) Newborn infants canorganize the auditory world Proceedings of the NationalAcademy of Sciences USA 100 11812ndash11815

Woods D L amp Elmasian R (1986) The habituation ofevent-related potentials to speech sounds and tonesElectroencephalography and Clinical Neurophysiology66 447ndash459

Zatorre R J Belin P amp Penhune V B (2002) Structure andfunction of auditory cortex Music and speech Trends inCognitive Sciences 6 37ndash46


nating foreign phonemic contrasts that are not presentin their native language such as rndashl for Japanesespeakers (Goto 1971) endashgt for Spanish speakers (Pal-lier Bosch amp Sebastian 1997) retroflex versus dental dfor English speakers (Werker amp Tees 1984b) amongmany examples The influence of native language onspeech perception is not limited to the phoneme reper-toire of the language but concerns all aspects of phonol-ogy Acoustic cues coding suprasegmental informationsuch as tones or stress and phonotactics the set of rulesthat govern how phonemes are combined within wordsare processed differently by subjects from different lan-guages For example Japanese adults have difficulties indiscriminating two pseudowords like ebzo and ebuzoThey perceive an illegal word like ebzo as ebuzoinserting an illusory vowel between consonants becauseJapanese does not allow for consonant clusters (DupouxKakehi Hirose Pallier amp Mehler 1999)

Phoneme Perception in Infants

Behavioral studies have demonstrated striking similari-ties between the phoneme perception capacities ofinfants and adults Like adults infants perceive pho-nemes categorically along acoustic dimensions such asvoice onset time (VOT) (Eimas Siqueland Jusczyk ampVigorito 1971) place of articulation (Werker GilbertHumphrey amp Tees 1981) and others They are able tonormalize across irrelevant acoustic variations such asthose related to different voices or pitches (JusczykPisoni amp Mullennix 1992 Kuhl 1983) or coarticulationcontext (Bertoncini Bijeljac-Babic Jusczyk Kennedy ampMehler 1988) They display duplex perception (Eimas ampMiller 1992) They are able to detect a match betweena vowel sound and the correct corresponding face move-ments (Patterson amp Werker 2003 Kuhl amp Meltzoff 1984)and are sensitive to the MacGurk effect (RosenblumSchmuckler amp Johnson 1997) The linguistic environ-ment rapidly modifies these initial capacities Around6 months infants show greater sensitivity to vowel cate-gories of their native language (Polka amp Werker 1994Kuhl Williams Lacerda Stevens amp Lindblom 1992) andaround 10 months to consonants (Kuhl Tsao amp Liu2003 Werker amp Tees 1984a) Although no experimenthas directly demonstrated a decrease of discriminationcapacities for other aspects of phonology than thoserelated to the repertoire of phonemes several experi-ments have shown that around 9ndash10 months of age in-fants use their knowledge of phonotactics to segmentwords (Mattys amp Jusczyk 2001) or to distinguish list ofwords differing on the probability of sequences of pho-nemes that constitute them (Jusczyk Luce amp Charles-Luce 1994) Thus at the end of the first year of life infantscertainly have access to phonological representationssimilar to those present in adults Are the same behav-ioral capacities in infants and adults sustained by thesame brain organization What is the initial state of or-

ganization of the human brain and how can it facilitatelanguage acquisition

METHODOLOGICAL ISSUES MISMATCHRESPONSES AS A TOOL TO EXPLOREAUDITORY REPRESENTATIONS

Event-related potentials (ERPs) are easily recorded evenin young children and elicitation of mismatch responsesin oddball paradigms is a powerful tool to access theproperties of the representations computed by thecortex from a sound It has been observed in single-cellrecordings in the visual (Miller Li amp Desimone 1991)and auditory cortex (Ulanovsky Las amp Nelken 2003)that the response of neurons decreases with repetitiveexposure to the same stimulus When a new or deviantstimulus is presented it elicits a response of a new set ofneurons Using this property called repetition suppres-sion it is possible to isolate cells that code differentproperties of the stimulus such as object identityindependently of its size or location (Miller et al1991) Of course the spatial resolution of brain imagingtechniques is too low to distinguish activity at the scaleof single neurons or columns of neurons that are thecoding elements of the nervous system However it ispossible to access this code at a macroscopic level bycontrasting the decreased activity recorded when thestimulus is repeated with the renewed activity due to theresponse of even a close set of neurons in the samebrain volume (see Naccache amp Dehaene 2001 for acomplete discussion of the priming method in brainimagery) This difference in activity is certainly thesource of mismatch responses recorded in ERP oddballparadigms By manipulating what elicits a mismatchresponse that is what counts or not as a repetitionfor a given neural network it is possible to infer theformat of representation that is computed by the net-work For example a network computing a phonemicrepresentation should habituate to the repetition of thesame phoneme independently of the speaker and re-cover when the phoneme is changed whereas a net-work coding for speaker should be affected by a voicechange but not by a phoneme change Although mis-match negativities (MMNs) observed in auditory oddballparadigms have been interpreted as a specific auditoryphenomenon related to a trace of the repeated stimulusin echoic memory (Naatanen 2001) these responses arebetter understood in this more general context ofrepetition suppression The experimental conditionsby selecting the level of representation that is targetedby the repetition context generate different MMNsrecorded with different topographies and latencies be-cause of the activation of different networks within thetemporal lobes Some MMNs are thus related to thecomputation of simple acoustic cues (eg durationintensity of the stimulus Giard et al 1995) but othersimply that multiple cues such as conjunction of features






















































Pena 2001)






















for SD)



































Acknowledgments



Notes



REFERENCES







































































































































































Pena 2001)






















for SD)



































Acknowledgments



Notes



REFERENCES
































































































































































Pena 2001)






















for SD)



































Acknowledgments



Notes



REFERENCES
























































































































































Pena 2001)






















for SD)



































Acknowledgments



Notes



REFERENCES







































































































































for SD)



































Acknowledgments



Notes



REFERENCES


































































































































for SD)



































Acknowledgments



Notes



REFERENCES
































































































































Acknowledgments



Notes



REFERENCES



























































































































Acknowledgments



Notes



REFERENCES





















































































































Acknowledgments



Notes



REFERENCES












































































































































































































































































































Documents

Common Neural Basis for Phoneme Processing in Infants ......Common Neural Basis for Phoneme Processing in Infants and Adults G. Dehaene-Lambertz 1,2 and T. Gliga 1 Abstract & Investigating