Functional Neuroimaging of
Speech Perception in Infants
Ghislaine Dehaene-Lambertz,1* Stanislas Dehaene,2
Human infants begin to acquire their native language in the first months of life.
To determine which brain regions support language processing at this young
age, we measured with functional magnetic resonance imaging the brain ac-
infants. Left-lateralized brain regions similar to those of adults, including the
superior temporal and angular gyri, were already active in infants. Additional
activation in right prefrontal cortex was seen only in awake infants processing
normal speech. Thus, precursors of adult cortical language areas are already
active in infants, well before the onset of speech production.
The adult human brain exhibits anatomical
and functional specialization for speech pro-
cessing (1–5). To understand this adult orga-
nization, one must ultimately clarify how it
emerges in the course of development
through a combination of brain maturation
constraints and environmental influences.
Behavioral studies in infants indicate that
considerable language learning is already tak-
ing place in the first year of life in the do-
mains of phonology, prosody, and word seg-
mentation [reviewed in (6)]. However, little
is known about the brain mechanisms under-
lying those abilities. At present, the only
evidence comes from recordings from event-
related potentials (ERPs). They indicate that
the temporal lobes contain neural circuits for
phoneme discrimination, which become at-
tuned to the mother language in the first year
of life (7–9). ERPs, however, do not provide
spatially accurate information on the active
brain areas. Here, we show that functional
magnetic resonance imaging (fMRI) can be
used to study the functional organization of
the infant brain. Provided that precautions are
taken to avoid introducing metallic objects in
the magnetic field, magnetic resonance is a
safe method that has been used in neurope-
diatric practice and research for the last 20
years with a variety of age ranges, including
healthy infants (10–13) and even fetuses
We collected fMRI images from 20
healthy nonsedated infants (2 to 3 months
old) while they listened to 20 s of speech
stimuli alternating with 20 s of silence (15).
On alternate blocks, the same excerpts from a
highly intonated female voice reading a chil-
dren’s book were presented in the infant’s
native language (French), with the recording
playing either forward or backward. Back-
ward speech violates several segmental and
suprasegmental phonological properties that
are universally observed in human speech (4,
16). Behavioral studies indicate that infants
are sensitive to these properties. For instance,
4-day-old neonates and 2-month-old infants
discriminate sentences in their native lan-
guage from sentences in a foreign language,
but this performance vanishes when the stim-
uli are played backward (17–19). We there-
fore expected that forward speech would elic-
it stronger activation than backward speech in
brain areas engaged in the recognition of
segmental and suprasegmental properties of
the native language. Nonetheless, forward
and backward speech both contain fast tem-
poral auditory transitions and phonetic infor-
mation conveyed by temporally symmetrical
phonemes. Thus, brain areas sensitive to
those properties were expected to be jointly
activated by those two conditions.
To obtain reliable fMRI images in nonse-
dated infants, we took several precautions to
1Laboratoire de Sciences Cognitives et Psycholinguis-
tique, CNRS & Ecole des Hautes Etudes en Sciences
Sociales, 54 Boulevard Raspail, 75270 Paris Cedex 06,
France.2Unite ´ INSERM 562,3Unite ´ de Neuroanato-
mie Fonctionnelle, Service Hospitalier Fre ´de ´ric Joliot,
Commissariat a ` l’Energie Atomique, 91401 Orsay Ce-
dex, France.4Service de Radiologie Pe ´diatrique, Ho ˆpi-
tal Necker–Enfants Malades, Assistance Publique–Ho ˆ-
pitaux de Paris, 75015 Paris, France.
*To whom correspondence should be addressed. E-
Fig. 1. Characteristics of the hemodynamic response to sound (forward and backward speech) in
2- to 3-month-old infants. (A) Sample nonaveraged data recorded from a left temporal voxel
during a single run, with forward (Fw) and backward (Bw) speech periods alternating with silent
periods. fMRI measurements (blue) were modeled by a sinusoidal function with additional
regressors accounting for head motion (red curve). (B) Distribution of phase lags for all voxels with
a significant sinusoidal activation (P ? 0.001), cumulated across all 20 infants. The most frequent
response occurred with a delay of 3 to 7 s between activation onset and sound onset, similar to
the adult HRF. Another smaller peak approximately a half-period away indicated the presence of
occasional inverse BOLD responses, which could be due either to deactivation during sound
presentation or to a normal activation by sound offset. However, deactivation responses were
never significant in the random-effect group analyses. (C) Activation induced by sound, observed
in an individual infant (P ? 0.001) when fitting the data using an idealized adult HRF. The figure
shows a transparent brain view, an axial slice at the level of the superior temporal cortices, and the
response curve at the indicated location, averaged across all sound presentations (mean ? SE). The
color scale indicates the value of the t test assessing the significance of the correlation of the
observed data with the model (d.f., degrees of freedom). The infant was asleep and activation was
largely confined to bilateral superior temporal cortices.
R E P O R T S
www.sciencemag.orgSCIENCEVOL 2986 DECEMBER 2002
ensure their comfort, to minimize head mo-
tion and noise exposure, and to permit con-
stant experimenter monitoring (15). Data pro-
cessing included rejection of images with
severe motion artifacts, realignment of the
remaining images, and incorporation of the
movement parameters as regressors in a lin-
ear model of the blood oxygen level–depen-
dent (BOLD) response appropriate for tem-
poral sequences with occasional missing
data. An important issue was the determina-
tion of the hemodynamic response function
(HRF) in infants, which may differ in the
immature brain. The small number of previ-
ous fMRI studies on this topic, all performed
with sedated or sleeping infants and using
various stimuli, are contradictory. Some re-
port a normal adult response (13, 14); others
suggest that the response can reverse at an
age that varies from region to region (10–12).
To characterize the latency and sign of the
HRF in our data set, we devised a sinusoidal
model that allowed us to detect any periodic
response at the frequency imposed by the
periodic alternation of sound and silence in
the stimuli, and to measure its activation de-
lay relative to sound onset (15). Within the
temporal resolution of the present block de-
sign, most activated voxels showed a delay of
about 5 s, a time course compatible with the
normal adult HRF (Fig. 1). Our use of eco-
logically natural stimuli in nonsedated infants
may have contributed to the observation of
We then identified active brain areas by
convolving the standard adult HRF with the
time course of the forward or backward speech
signal. A random-effect analysis of the 20 in-
fants revealed stimulus-induced activation in a
large extent of the left temporal lobe (Fig. 2A).
Activation ranged from the superior temporal
gyrus, encompassing Heschl’s gyrus, to sur-
rounding areas of the superior temporal sulcus
and the temporal pole. Symmetrical areas of the
right temporal lobe also showed a small activa-
tion, which did not remain significant after
correction for multiple comparisons (see table
S1). Activation was significantly greater in the
the planum temporale (Fig. 2B).
We then studied the differences between
forward and backward speech. The left angu-
lar gyrus and the left mesial parietal lobe
(precuneus) were significantly more activated
by forward speech than by backward speech
(Fig. 2C). Conversely, no region showed
greater activation by backward speech than
by forward speech. To investigate the effect
of wakefulness on these responses, we com-
pared the activation patterns of six infants
who stayed awake during the entire session
with those of five infants who were deeply
asleep (in the remaining infants, the state of
wakefulness was too variable to permit un-
ambiguous classification). Although the main
effect of wakefulness on sound-induced acti-
vation was not significant, a significant inter-
action between the nature of the stimulus and
the infants’ wakefulness was observed in the
right dorsolateral prefrontal cortex (DLPFC).
This region showed a greater activation by
forward speech than by backward speech in
awake infants but not in sleeping infants (Fig.
3). The right lateralization of this activation
was significant. The converse interaction re-
vealed a greater activation by backward
speech than by forward speech bilaterally in
the posterior part of the superior temporal
sulci, again only in awake infants.
Our results show that the infant cortex is
already structured into several functional re-
gions. As in adults (1–4), listening to speech
activates a large subset of the temporal lobe,
with a significant left-hemispheric domi-
nance. This is consistent with the left lateral-
ization observed in ERPs (7, 8) and in dichot-
ic listening during syllable discrimination
tasks in infants (20). An anatomical asymme-
try is detectable in the planum temporale as
early as 31 weeks of gestation (21). Our fMRI
results indicate that this anatomical differ-
ence supports an early functional asymmetry
in the processing capacities of the two hemi-
spheres. It is not yet known, however, wheth-
er this asymmetry reflects an early special-
ization for speech perception or a greater
responsivity of the left temporal cortex to any
auditory stimulus (or perhaps to any stimulus
with fast temporal changes).
In adults, listening to the native language
(versus listening to backward speech or a
foreign language) induces greater activation
all along the left superior temporal sulcus,
extending posteriorly into the left angular
gyrus (1–3). We did not observe any differ-
ence between forward and backward speech
in the infant temporal lobe. This suggests that
Fig. 2. Localization of activation in random-effect group analyses (voxel P ? 0.01, cluster P ? 0.05,
corrected). (A) Activation evoked by sound presentation (forward and backward speech), relative to
silent periods, at three different axial levels in the left temporal lobe (see table S1 for coordinates
of activation peaks). (B) Statistical map of asymmetry of activation by sound, showing that the
planum temporale was significantly more activated in the left hemisphere than in the right. (C)
Statistical map of the comparison between forward and backward speech, showing greater
activation by forward speech in the left angular gyrus.
R E P O R T S
6 DECEMBER 2002VOL 298SCIENCEwww.sciencemag.org
this area is undergoing changes during infan-
cy and has not yet acquired its full compe-
tence for the native language by 3 months.
However, we found a significant advantage
for the native language in the left angular
gyrus, the left precuneus, and, in awake ba-
bies only, the right prefrontal cortex. In
adults, the left angular gyrus shows greater
activation when subjects hear words than
when they hear nonwords (4, 5) or when they
hear sentences in a known language relative
to hearing sentences in an unknown language
or backward speech (1–3). Moreover, the pre-
cuneus and DLPFC are activated, often with
a right lateralization, when adults retrieve
verbal information from memory (22–24).
Activation of both regions in 3-month-olds
may indicate the early engagement of active
memory retrieval mechanisms. This would fit
with behavioral evidence that infants of that
age have already memorized the prosodic
contours of their native language (17, 19),
although they may not remember single
words until the age of 7 months (6).
Two mechanisms of language acquisition
are classically opposed. According to one view,
the human brain is equipped with genetically
that endow the infant with an early linguistic
competence without which language acquisi-
tion would be impossible (25). For others, the
infant brain is initially immature and plastic,
and exposure to speech inputs progressively
shapes its organization through domain-general
mechanisms of learning and plasticity (26).
Without resolving this debate, our results pro-
vide evidence that should be accommodated by
any model of language development. First, the
areas activated by the native language are not
confined to primary auditory cortices, even dur-
ing the first few months of life. Second, neither
do they extend widely into other cortical terri-
tories such as the visual areas; rather, they
remain confined to regions that are similar to
those observed in adults in both their localiza-
tion and their lateralization. Third, frontal cor-
tex already exhibits functional specificity in
infants and thus can no longer be assumed to be
silent in the first months of life. The delayed
synaptogenesis and myelination of this area,
and its immature level of metabolic activity,
need not imply that it does not contribute to
early cognitive processes (27). Overall, the pre-
frontal activation, in coordination with a left-
lateralized temporoparietal activation partially
similar to that found in adults, favors a descrip-
tion of language acquisition as a progressive
differentiation of a preconstrained network of
left-hemispheric regions under the influence of
active mechanisms of attention and effort (28).
References and Notes
1. D. Perani et al., Neuroreport 7, 2439 (1996).
2. B. M. Mazoyer et al., J. Cogn. Neurosci. 5, 467 (1993).
3. S. Dehaene et al., Neuroreport 8, 3809 (1997).
4. J. R. Binder et al., Cereb. Cortex 10, 512 (2000).
5. J.-F. De ´monet et al., Brain 115, 1753 (1992).
6. P. W. Jusczyk, The Discovery of Spoken Language (MIT
Press, Cambridge, MA, 1997).
7. G. Dehaene-Lambertz, S. Baillet, Neuroreport 9, 1885
8. G. Dehaene-Lambertz, S. Dehaene, Nature 370, 292
9. M. Cheour et al., Nature Neurosci. 1, 351 (1998).
10. T. Morita et al., Neurosci. Res. 38, 63 (2000).
11. A. P. Born et al., Neuropediatrics 31, 24 (2000).
12. A. W. Anderson et al., Magn. Reson. Imaging 19, 1
13. N. R. Altman, B. Bernal, Radiology 221, 56 (2001).
14. R. J. Moore et al., Hum. Brain Mapp. 12, 94 (2001).
15. See supporting data on Science Online.
16. J. Vaissie `re, in Prosody: Models and Measurements, A.
Cutler, D. R. Ladd, Eds. (Springer-Verlag, New York,
1983), pp. 53–66.
17. J. Mehler et al., Cognition 29, 143 (1988).
18. F. Ramus, M. D. Hauser, C. Miller, D. Morris, J. Mehler,
Science 288, 349 (2000).
19. G. Dehaene-Lambertz, D. Houston, Lang. Speech 41,
20. J. Bertoncini et al., Brain Lang. 37, 591 (1989).
21. J. A. Wada, R. Clarke, A. Hamm, Arch. Neurol. 32, 239
22. T. Shallice et al., Nature 368, 633 (1994).
23. B. J. Krause et al., Brain 122, 255 (1999).
24. P. C. Fletcher, R. N. Henson, Brain 124, 849 (2001).
25. S. Pinker, The Language Instinct: How the Mind Cre-
ates Language (Penguin, London, 1994).
26. J. L. Elman et al., Rethinking Innateness: A Connec-
tionist Perspective on Development (MIT Press, Cam-
bridge, MA, 1996).
27. P. R. Huttenlocher, A. S. Dabholkar, J. Comp. Neurol.
387, 167 (1997).
28. M. H. Johnson, Nature Rev. Neurosci. 2, 475 (2001).
29. We thank F. Brunelle, M. Dutat, E. Giacomini, T.
Gliga, F. Hennel, A. Jobert, D. Le Bihan, J. F. Mangin,
S. Margules, C. Pallier, J. B. Poline, and D. Rivie `re for
support and advice. Supported by CNRS, INSERM,
Institut Fe ´de ´ratif de Recherches 49, Ho ˆpital Neck-
er–Enfants Malades, and the McDonnell Founda-
Supporting Online Material
Materials and Methods
6 August 2002; accepted 21 October 2002
Fig. 3. Interaction between wakefulness and the linguistic nature of the stimuli (voxel P ? 0.01,
cluster P ? 0.05, corrected). This comparison isolated a right dorsolateral prefrontal region that
showed greater activation by forward speech than by backward speech in awake infants, but not
in sleeping infants.
R E P O R T S
www.sciencemag.orgSCIENCE VOL 2986 DECEMBER 2002