A hierarchy of cortical responses to sequence violations in three-month-old infants

Full text

A hierarchy of cortical responses to sequence violations
in three-month-old infants
Anahita Basirat
a,b,c,
⇑
, Stanislas Dehaene
a,b,c,d
, Ghislaine Dehaene-Lambertz
a,b,c
a
INSERM, U992, Cognitive Neuroimaging Unit, F-91191 Gif/Yvette, France
b
CEA, DSV/I2BM, NeuroSpin Center, F-91191 Gif/Yvette, France
c
University Paris-Sud, Cognitive Neuroimaging Unit, F-91191 Gif/Yvette, France
d
Collège de France, F-75005 Paris, France
article info
Article history:
Received 3 August 2012
Revised 30 January 2014
Accepted 28 March 2014
Available online 5 May 2014
Keywords:
Mismatch response
Prediction
Brain development
Sequence learning
Rule extraction
abstract
The adult human brain quickly adapts to regular temporal sequences, and emits a sequence
of novelty responses when these regularities are violated. These novelty responses have
been interpreted as error signals that reflect the difference between the incoming signal
and predictions generated at multiple cortical levels. Do infants already possess such a
hierarchy of violation-detection mechanisms? Using high-density recordings of event-
related potentials during an auditory local–global violation paradigm, we show that
three-month-old infants process novelty in temporal sequences at two distinct levels. Vio-
lations of local expectancies, such as perceiving a deviant vowel ‘‘a’’ after repeated presen-
tation of another vowel i-i-i, elicited an early auditory mismatch response. Conversely,
violations of global expectancies, such as hearing the rare sequence a-a-a-a instead of
the frequent sequence a-a-a-i, modulated this early mismatch response and led to a late
frontal negative slow wave, whose cortical sources included the left inferior frontal region.
These results suggest that the infant brain already possesses two dissociable systems for
temporal sequence learning.
Ó2014 Elsevier B.V. All rights reserved.
1. Introduction
The classical constructivist perspective postulates that
learning starts at an early sensory level and very slowly
progresses towards increasingly abstract and logical levels
(Piaget, 1954; Quartz & Sejnowski, 1997). The first func-
tional MRI and NIRS studies in infants were thus surprising,
revealing the involvement of high-level brain areas such as
dorsolateral prefrontal cortex and Broca’s area at an early
age (Dehaene-Lambertz, Dehaene, & Hertz-Pannier, 2002;
Dehaene-Lambertz et al., 2010; Mahmoudzadeh et al.,
2013; Perani et al., 2010), supported by an efficiency
long-range connectivity (Leroy et al., 2011). Learning in
infants might thus not be limited to low-level processes,
but might occur at all levels along the processing hierarchy,
as proposed by recent Bayesian models of child develop-
ment (Tenenbaum, Kemp, Griffiths, & Goodman, 2011;
Téglás et al., 2011), with high-level regions generating
top-down predictions modulating the down-stream com-
putations (Friston, 2005; Rao & Ballard, 1999). Here, we test
the hypothesis that the infant brain, at three months of age,
already processes information about auditory sequences at
two hierarchical levels. Using an auditory violation para-
digm, we demonstrate that the infant brain contains a hier-
archy of error signals that respond, respectively, to
violations of local and global auditory sequences. We argue
that the presence of these signals suggest that the infant
http://dx.doi.org/10.1016/j.cognition.2014.03.013
0010-0277/Ó2014 Elsevier B.V. All rights reserved.
⇑
Corresponding author at: INSERM, U992, Cognitive Neuroimaging
Unit, F-91191 Gif/Yvette, France.
E-mail address: anahita@basirat.fr (A. Basirat).
Cognition 132 (2014) 137–150
Contents lists available at ScienceDirect
Cognition
journal homepage: www.elsevier.com/locate/COGNIT

brain, at three months, already generates top-down predic-
tions about future incoming stimuli.
A simple and widely used paradigm to study infant
auditory perception is the auditory oddball paradigm, in
which a novel sound is introduced after a series of repeated
sounds. This abrupt change generally elicits an early
mismatch response (MMR), often consisting of a frontal
positivity synchronous of a posterior negativity, around
200–400 ms after the deviant stimulus, generally followed,
around 700 ms, by a late frontal Negative Slow Wave
(NSW; Dehaene-Lambertz & Dehaene, 1994; Friederici,
Friedrich, & Weber, 2002).
This two-stage response in infants is reminiscent of the
MMN/P300 complex reported in adults, even if the laten-
cies and topographies of these responses are different
due to the immaturity of the infant’s brain. In adults, an
early and automatic mismatch response is recorded
around 100 ms and consists of a frontal negativity with a
polarity reversal above temporal regions (mismatch nega-
tivity or MMN, Näätänen, Gaillard, & Mäntysalo, 1978), and
a late central positivity is recorded around 300 ms (P300 or
P3b, Squires, Squires, & Hillyard, 1975). In adults, these
two components differ in their functional properties: The
MMN is present even if the subject does not pay attention
to the stimuli, is asleep (Atienza, Cantero, & Gomez, 1997)
or in coma (Fischer et al., 1999), but it disappears when the
inter-stimulus-interval (ISI) is increased beyond a few sec-
onds (Mäntysalo & Näätänen, 1987; Pegado et al., 2010). By
contrast, the P300 is only present if the subject is conscious
and attentive (Bekinschtein et al., 2009), and is not affected
by long ISI (Wetter, Polich, & Murphy, 2004). The MMN has
been associated with unconscious processing of auditory
transition probabilities (Näätänen, Paavilainen, Rinne, &
Alho, 2007; Wacongne, Changeux, & Dehaene, 2012;
Winkler, 2007), and the P300 with conscious detection of
novelty and ‘‘context updating’’ (Dehaene & Changeux,
2011; Donchin & Coles, 1988; Sergent, Baillet, & Dehaene,
2005).
Partially similar observations have been made in
infants. The early MMR can be elicited in non-attentive
or sleeping infants (Dehaene-Lambertz & Peña, 2001),
and is reduced by long ISI (Cheour et al., 2002). Its brain
sources are mainly located in the superior temporal
regions (Bristow et al., 2009; Dehaene-Lambertz &
Dehaene, 1994), congruent with the adults’ description of
the MMN sources (Celsis et al., 1999; Halgren, Sherfey,
Irimia, Dale, & Marinkovic, 2011). The late response
(NSW) is less often reported than the MMR, but this could
simply be because it occurs too late to fit within the length
of the studied ERP epoch. The NSW belongs to a set of late
components observed in infants which have been linked to
attention and novelty detection (Csibra, Kushnerenko, &
Grossmann, 2008), and more recently to conscious percep-
tion (Kouider et al., 2013). Indeed, when comparing awake
and asleep infants, Friederici et al. (2002) observed the
NSW after a deviant sound only in awake infants.
The functional similarities of the MMR/NSW with the
adult MMN/P300 components suggest a putative parallel
with the adult functional architecture (i.e. an early auto-
matic local response vs. a late context-dependent
response). Yet at present, no study has tried to disentangle
whether the MMR and NSW are sensitive to different types
of violations. A recently introduced hierarchical ‘‘local–glo-
bal’’ paradigm epitomizes the two distinct processing
stages behind the generation of a MMN and a P300 in
adults (Bekinschtein et al., 2009). This paradigm measures
brain responses to auditory novelty at two hierarchical lev-
els. At the first level, a novel sound is introduced after a
series of repeated sounds (e.g. xxxY, where x denotes the
repeated sound and Y the novel sound), generating a
‘‘local’’ deviancy. At the second level, a series of sounds is
selected as the frequent global sequence for a block of tri-
als (e.g. xxxY), and then this sequence is violated on a rare
subset of trials (e.g. by occasionally presenting the
sequence xxxx). With this paradigm, Bekinschtein et al.
(2009) disentangled two properties of the adult MMN
and P300 responses. First, local deviants (the last sound Y
in sequence xxxY) systematically elicit a MMN, even when
the sequence itself is frequent and predictable; this
response is automatic and remains present in inattentive
or comatose subjects. It corresponds to an automatic
error-signal generated when the incoming sound differed
from what was expected given the previous sounds
(Garrido, Kilner, Kiebel, & Friston, 2007; Garrido et al.,
2008; Näätänen et al., 1978; Wacongne et al., 2011,
2012; Winkler, 2007). Second, global deviants (rare
sequences) systematically elicit a P300 response, even
when the rare deviating sequence is a monotonous
sequence of repeated sounds (xxxx).
The latter finding is particularly diagnostic of a second-
order computation. While the first stage (MMN) simply
weights the incoming sound against predictions based on
past events, the second stage (P300) seems sensitive to
the global rule governing the entire sequence. Especially,
generating an error signal to a perfectly monotonic ‘‘xxxx’’
sequence (in a block where most trials are xxxY) can only
be performed by a system that actively generates an expec-
tation that the sequence should end with a different sound
(xxxY). This second stage, in adults, requires attention to
the sequence (Bekinschtein et al., 2009).
In the present study, we probed the existence of
hierarchical novelty detection and predictive processes in
three-month-old infants using high-density recordings of
event-related potentials during a variant of Bekinschtein
et al. (2009) auditory local–global paradigm. To maximize
attention, although we were obviously unable to give
instructions to our preverbal participants, we used audio-
visual speech stimuli which infants spontaneously find
strongly attractive (Fig. 1). Stimuli were presented in short
series of four vowels, following an xxxY or xxxx pattern in
distinct blocks (see Fig. 2 for experimental design). After a
short training phase, which let infants learn the global
sequence governing the present block, sequences violating
this global pattern were randomly presented (i.e. xxxx tri-
als in blocks with rule xxxY, and xxxY trials in blocks with
rule xxxx).
Our predictions were simple. Based on the infant litera-
ture, we expected to record a mismatch response (MMR)
around 200–400 ms after a deviant sound, followed by a
late frontal negativity if infants direct their attention
toward this novel event (Dehaene-Lambertz & Dehaene,
1994; Friederici et al., 2002). If the infant’s MMR is
138 A. Basirat et al. / Cognition 132 (2014) 137–150

equivalent to the adults’ MMN, it would occur after any
local change (i.e. any xxxY sequence). Because the size of
the mismatch depends on the frequency of the standard
and deviant sounds in the preceding seconds (King et al.,
2013; Sato et al., 2000; Winkler, Cowan, Csépe, Czigler, &
Näätänen, 1996), the MMR might be modulated by block
structure and be smaller on xxxY blocks (where local viola-
tions are frequent) than on xxxx blocks (where they are
rare). Crucially however, if infants and adults share a similar
hierarchical functional architecture, the late negative slow
wave (NSW) should depend on the presence of a global rule
violation. Thus, it should be elicited by any rare sequence,
even when it merely consists in the repetition of the same
sound (xxxx). If such a late response was observed, it would
suggest that three-month-old infants are able to represent
local and global auditory regularities and detect their viola-
tions at two hierarchical levels, as in adults.
2. Methods
2.1. Subjects
Twenty-nine healthy full-term infants (twelve females)
were tested between 11 and 15 weeks after birth (mean
age = 13.6 weeks, SD = 1.1 weeks). Twenty additional
infants were tested but rejected for fussiness, excessive
movement, bad recording or lack of enough data (<150 tri-
als in total or <20 artifact-free trials in the test phase of
each rule-type condition [xxxx and xxxY]). The study was
approved by the regional ethical committee for biomedical
research, and parents gave their written informed consent.
2.2. Stimuli
To maximize infants’ attention, we used audio–visual
speech stimuli. Two males and two females were filmed
articulating /a/ and /i/. Four frames were extracted from
each clip: mouth fully opened (frame 1), mouth gradually
closed (frames 2 and 3), mouth closed (frame 4). Auditory
stimuli matched in duration (200 ms) and subjective inten-
sity and produced by the same speakers were extracted
from the audio track.
2.3. Procedure
Infants were seated on their parent’s lap in front of a
computer screen inside a shielded room (Faraday cage),
with the loudspeakers located in the roof of the room. Each
Fig. 1. Temporal structure of a trial. On each trial, infants were presented with a series of four successive vowels (/a/ or /i/). The first three sounds were
always identical, and while the last sound could be identical or different. To maximize the infants’ attention and minimize fussiness, stimuli were
accompanied by a congruent view of a face speaking the same vowel. At the beginning of each trial, a face (and the corresponding voice) was randomly
chosen among four different speakers. The mouth was initially closed, and opened suddenly in synchrony with sound onset. At the end of each sound, the
mouth closed (two intermediate frames).
Fig. 2. Experimental design. Two types of trials were presented. Local standards, denoted xxxx, comprised four identical vowels (/aaaa/ or /iiii/). Local
deviants, denoted xxxY, comprised three identical vowels followed by a different one (/aaai/ or /iiia/). In a given block of trials, a rule (xxxx or xxxY) was
selected. During an initial learning phase (15 trials), 100% of trials followed this rule. Then during the test phase, 75% of the trials followed the rule (global
standard) while 25% violated it (global deviant). The rule was alternated across blocks in order to expose each infant to both rules. The order was
counterbalanced across infants.
A. Basirat et al. / Cognition 132 (2014) 137–150 139

trial consisted of the audio–visual presentation of four
vowels with a 600 ms stimulus onset asynchrony (SOA).
The onset of the auditory vowel always coincided with a
visual presentation of the mouth fully opened as if pro-
nouncing the corresponding vowel. At the end of the sound
(after 200 ms), the mouth was gradually closed (two
frames of 60 ms duration each), then the face with a closed
mouth was presented for 280 ms until the onset of the next
vowel (Fig. 1). This design ensured a sufficiently natural
mouth movement while perfectly controlling the onset of
the audio–visual stimulation. Note that, with this design,
the novelty of the fourth event was simultaneously con-
veyed by audition (vowel) and by vision (mouth shape), a
feature which should only facilitate novelty detection and
global rule extraction. 1400 ms after the onset of the last
syllable of the trial, a new face was presented, randomly
chosen amongst the three remaining faces. 500 ms later,
the next audio–visual presentation began. Thus, the inter-
trial interval was 1900 ms long. Two types of trials were
presented: local standard trials, denoted xxxx, in which
the same vowel was repeated (/aaaa/ or /iiii/), and local
deviant trials, denoted xxxY, in which the last syllable
was changed (/aaai/ or /iiia/).
The experiment was separated into four blocks of 75 tri-
als, each lasting 5 min, following two different types of
rules (Fig. 2). In xxxx blocks (aaaa or iiii), infants were
expected to learn that all trials were of the xxxx form,
whereas in xxxY blocks (aaai or iiia), they were expected
to learn that all trials followed the xxxY form. In order to
establish the global sequence, during an initial learning
phase, the first 15 trials always respected it. This phase
was followed by a test phase of 60 trials, in which 75% of
the trials were ‘‘global standards’’ which respected the
sequence chosen for that block (45 trials) but 25% were
‘‘global deviants’’ which violated it (15 trials). These devi-
ant trials were always followed by at least one standard
trial, which was not included in the analyses. The repeated
vowel was kept constant during the block, but its gender
could change from trial to trial in congruence with the
visual stimuli. The rule order was counterbalanced across
infants and alternated from one block to the next (e.g. /
aaaa/ /iiia/ /iiii/ /aaai/). Stimuli were presented using E-
prime v1.2 (Psychology Software Tools, Inc.). The experi-
ment was stopped if the infant became fussy or after the
four blocks. All infants were exposed to both rules (see
exclusion criteria).
2.4. EEG recordings
Scalp voltages were collected using a 128-channel Geo-
desic Sensor Net (EGI) referenced to the vertex. They were
amplified, sampled at 250 Hz, and filtered between 0.5 and
20 Hz. Trials were segmented relative to the onset of the
fourth (or test) syllable (2100 to 1800 ms). Channels con-
taminated by eye or motion artifacts were automatically
rejected and trials with more than 50% bad channels were
excluded. Due to an initial problem in the stimulus presen-
tation program, six infants did not receive the learning
phase for the first block. The entire first block was rejected
for these infants. Artifact-free trials were averaged per
infant and per condition. Averages were baseline corrected
(using 300 ms before the onset of the first sound of the
trial) and average referenced. An average number of 97.1
trials were kept in each infant, corresponding to 34.5 xxxx
and 17.2 xxxY trials on xxxx blocks, and 30.1 xxxY and 15.3
xxxx trials on xxxY blocks.
2.5. Data analysis
Our 2 2 experimental design combined two factors:
local deviance (whether or not the last item of the
sequence changed, i.e. xxxx vs. xxxY) and global deviance
(whether or not the trial sequence violated the block rule).
We analyzed the local and global effect by merging the 4
types of trials accordingly: for the local effect (xxxY vs.
xxxx), frequent xxxY in the xxxY blocks and rare xxxY in
the xxxx blocks were compared to frequent xxxx in the
xxxx blocks and rare xxxx in the xxxY blocks. For the global
effect, rare xxxY and rare xxxx were compared to frequent
xxxx and frequent xxxY.
Given that we used a 128-channels recording system
and a time resolution of 4 ms (250 Hz), performing a sys-
tematic comparison of all samples would run into the
severe risk of false positives. We thus used two orthogonal
approaches. We first tested whether, and at which time,
the expected two stages of novelty processing were found.
We did this computing at each time point the mean across
the 129 electrodes of the absolute value of the voltage dif-
ference related to the studied effect (local or global devi-
ance). This analysis therefore discarded topographic
information and reduced the data to one point per time
sample (Fig. 3). We compared the grand-average computed
across the infants, to surrogate data obtained by permuting
the labels of the different conditions within-subjects. 1000
permutations were done in each infant and 1000
grand-averages were thus computed to which the real
grand-average was compared. This evaluated the chance
probability of obtaining a difference at least as large as
the observed one at each time-point.
Second, to evaluate the topography of the effects while
avoiding a combinatorial explosion in the number of statis-
tical tests, we based our statistical analyses on the infant
literature and averaged the voltage across pre-selected
clusters of electrodes and time-windows, which were
determined a priori using previous studies of the early mis-
match response and the late slow wave in infants (Bristow
et al., 2009; Dehaene-Lambertz & Gliga, 2004; Friederici
et al., 2002). Our goal was to understand how these known
components were affected by the two types of deviance.
For the MMR, Bristow et al. (2009) performed an audio–
visual speech perception experiment in two-month-olds
using the same stimuli as here. They reported a dipole con-
figuration with a right positive anterior (around C4, Fz, F4,
and F8) and a left posterior negative pole (around O1, P3,
T5, mastoid) at about 300 ms after the onset of the deviant
sound (see also Dehaene-Lambertz & Dehaene, 1994).
Thus, in our analysis, we selected a time window centered
on the peak response (270–370 ms) congruent with this
study and using clusters of electrodes placed over the same
regions (15 anterior channels and 12 posterior electrodes
on each side, Fig. 4). We also selected their homologues
on the contralateral hemisphere to enter hemisphere as a
140 A. Basirat et al. / Cognition 132 (2014) 137–150

factor in our analyses and test for hemispheric asymme-
tries in novelty detection. Indeed, the human brain dis-
plays structural and functional asymmetries from the
preterm period on. Markers of maturational development
show a differential calendar in the left and right hemi-
spheres (Chiron et al., 1997; Dubois et al., 2008; Leroy
et al., 2011; Lin et al., 2012) and several studies using near
infra-red spectroscopy (Mahmoudzadeh et al., 2013;
Telkemeyer et al., 2011), fMRI (Dehaene-Lambertz et al.,
2010; Perani et al., 2010), and ERPs (Bristow et al., 2009)
describe functional asymmetries which suggest a func-
tional counterpart of the structural asymmetries.
As an additional control for the possibility that the
observed differences in response to the last sound were
due to chance, we introduced in our analyses the factor
‘‘stimulus number’’, and tested whether the difference
between conditions was significantly larger for the last
vowel than during the first three vowels. We thus per-
formed a repeated-measure analysis of variance (ANOVA)
on the voltage averaged over the selected time window
and cluster of electrodes with stimulus number (mean of
first three vowels vs. fourth vowel), local deviance (local
deviant vs. local standard), global deviance (global deviant
vs. global standard), electrode location (anterior positive
vs. posterior negative pole) and hemisphere (left vs. right)
as within-subject factors. In this analysis, the only perti-
nent comparisons were the interactions of local/global
deviance with stimulus number and electrode location (±
hemisphere). We considered only interactions with elec-
trode location (and not the main effect over all locations)
because the electrode clusters were located on the positive
and negative poles of the dipolar fields evoked by the devi-
ant stimuli, and thus deviancy was expected to yield more
positives voltages at one location and more negative volt-
ages at the other, i.e. a deviancy location interaction.
The late slow wave belongs to a set of late responses
observed after 600 ms in infants. In previous auditory stud-
ies a bilateral frontal negative wave, without a clear dipo-
lar configuration, was reported from 700 to more than
1000 ms after the deviant stimulus (e.g., Dehaene-
Lambertz & Dehaene, 1994; Friederici et al., 2002). We thus
selected a cluster of 10 frontal electrodes for this analysis
(Fig. 5C) on the 900–1200 ms time window, where the
maximum of the response was recorded. An ANOVA was
performed on the mean ERPs across this time window with
local deviance (local deviant vs. local standard) and global
deviance (global deviant vs. global standard) as within-
subject factors. Note that, since the duration between
sounds (SOA) within a trial sequence was 600 ms, the stim-
ulus number could not be included as a factor in this
ANOVA. We thus examined for this analysis only the main
effects of local and global deviance.
2.6. Cortical sources modeling
Although ERPs have a coarse spatial resolution, it is pos-
sible to approximately infer cortical sources from the scalp
voltage even in infants (Dehaene-Lambertz & Dehaene,
1994; Richards, 2005). Using a two-month-old infant tem-
plate (Kabdebon et al., submitted for publication), we mod-
eled the sources of the grand-averages of the local and
global effects using Brainstorm, a matlab software package
(Tadel, Baillet, Mosher, Pantazis, & Leahy, 2011). We used a
distributed model of 14,000 current dipoles, whose loca-
tions were constrained to the cortex of the infant template.
The EEG forward model was computed using overlapping
spheres whose radiuses were adapted to the infant head,
skull and brain size. Conductivities were also modified to
take into account infants’ specific tissue properties. Corti-
cal current maps were computed from the EEG time series
using a linear inverse estimator (weighted minimum-norm
current estimate).
3. Results
3.1. A two-stage response to auditory deviance
Our first analysis of the local and global deviance, based
on permutations of these labels within-subjects, isolated
two significant time periods for each effect (Fig. 3). For
the local deviance (xxxY vs. xxxx), a large significant differ-
ence was seen at the classical latency of the infants’ MMR
(p< .05 from 180 to 500 ms), followed by a later difference
(p< .05 from 992 to 1140 ms). The topography of the local
difference during the early time-window was character-
ized by a frontal positivity synchronous of a posterior neg-
ativity whereas the late response corresponded to a frontal
negativity with two posterior positive poles. These
responses are congruent with the MMR/late frontal nega-
tive response described in attentive infants when a novel
Fig. 3. Auditory deviance affected two distinct time windows
(MMR = mismatch response, and NSW = negative slow wave). Grand-
averages of the local and the global effects are compared to surrogate data
obtained through 1000 permutations of the condition labels within each
subject. P-values are presented at the bottom of the plot (scale on the
right yaxis). The p-value reached .05 from 252 to 452 ms and from 884 to
1140 ms for the global effect (green lines) and from 180 to 500 ms and
from 992 to 1140 ms for the local effect (red lines). Topographies of the
voltage averaged across each significant time-window are presented
above the plot (green circles for the global effect and red circles for the
local effect). (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
A. Basirat et al. / Cognition 132 (2014) 137–150 141

auditory stimulus is introduced after repeated stimuli
(Dehaene-Lambertz & Dehaene, 1994; Friederici et al.,
2002). For the global deviance (violation of the block struc-
ture), a two-stage response was also observed with a
shorter early mismatch response (significant from 252 to
452 ms) and a longer late response (884–1140 ms) relative
to the local effect. The topographies appeared more left lat-
eralized than for the local deviance.
To analyze these effects topographically while avoiding
double dipping problems (Vul & Pashler, 2012), we based
our following analyses on the time-windows and sets of
electrodes previously used in the infants’ literature.
3.2. Early time-window: mismatch response
A bilateral mismatch response was observed when the
trial was locally deviant (xxxY). Four clusters of elec-
trodes selected from the literature captured quite well
the topography of the response. Fig. 4 shows the time
course of ERPs for each of the four types of trials over
these clusters of electrodes, while Fig. 5A shows their
match to the full topography of the local effect. The
ANOVA computed across the time window 270–370 ms
reported a significant local effect (local deviance 
stimulus number electrode location: F(1,28) = 21.13,
p< 0.001) which did not interact with hemisphere (local
deviance hemisphere stimulus number electrode
location: F(1,28) = 1.02). This response was considerably
larger when the xxxY trials were rare (in xxxx blocks)
than when they were frequent (in xxxY blocks) as shown
by a significant interaction of local and global deviance
(Figs. 5 and 6, global deviance local deviance stimulus
number electrode location: F(1,28) = 4.47, p< 0.05).
We also observed a global effect in this early time-
window (global deviance stimulus number electrode
location: F(1,28) = 8.81, p< 0.01, Fig. 5B). This response
was significantly different over the left and right
hemisphere (4-way interaction of global deviance hemi-
sphere stimulus number electrode location, F(1, 28) =
4.36, p< 0.05). Post hoc analyses restricted to anterior
electrodes revealed a significant hemispheric difference
for the global effect (global deviance hemisphere
Fig. 4. Grand-average event-related potentials averaged across the electrodes displayed in each panel. Time 0 marks the onset of the fourth sound. Dashed
vs. solid lines indicate ERPs to local deviants vs. local standards, while purple vs. green lines indicate ERPs to global deviants (rare rule violations) vs. global
standards (frequent rule-governed stimuli). Within each condition, a baseline corresponding to the mean ERP during the presentation of the first three
sounds (i.e. time window from 1800 to 0 ms) was subtracted from the corresponding ERPs. A mismatch response is visible on all selected clusters around
300 ms after the onset of the local deviant sound (dashed vs. solid lines), whether it is rare (in xxxx rule blocks) or frequent (in xxxY rule blocks). However,
the amplitude of this response is reduced in xxxY rule blocks (green vs. purple dashed lines). In the same time window, a response to rare violations is
mainly seen over the left anterior electrodes (purple vs. green lines in the top left panel). (For interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)
142 A. Basirat et al. / Cognition 132 (2014) 137–150

stimulus number, F(1,28) = 6.57, p< 0.05). While the
local effect was significant over both left and right ante-
rior electrodes (local deviance stimulus number:
respectively, F(1,28) = 6.79, p< 0.05 and F(1, 28) = 8.45,
p< 0.01), the global effect resulted in a significant positiv-
ity only over the anterior left hemisphere (global devi-
ance stimulus number: left F(1,28) = 18.03, p< 0.0001,
right F(1,28) < 1).
3.3. Late time-window: frontal negative slow wave
We then examined the presence of a late response to
auditory deviance over frontal electrodes and the 900–
1200 ms time window (Fig. 5C, Fig. 6B and C). The two-
way ANOVA showed a significant global effect
(F(1,28) = 5.75, p< 0.05). A significant interaction of global
deviance local deviance was observed (F(1,28) = 4.29,
p< 0.05). At this moment and on this cluster, the local
effect itself was not significant (F(1,28) < 1).
3.4. A global novelty response to xxxx trials
A notable feature of the response to global deviance is
that the brain should generate a novelty response to a
monotonous xxxx stimulus sequence when this sequence
is presented within xxxY blocks and is therefore rare and
unexpected. In order to test whether this effect was pres-
ent in infants, we compared the response to the same xxxx
trials in the blocks following the xxxx rule and those fol-
lowing the xxxY rule. A global deviance effect was indeed
Fig. 5. Topographies of the local and global effects. Each panel shows the projection, on a 2-D representation of the head surface, of the average voltage
measured in a time window of interest in the two compared experimental conditions, as well as the corresponding z-scores. Electrodes selected for further
statistical analysis are shown at right. (A) Local effect. Averaged ERPs across the time window 270–320 ms after the onset of the fourth sound for local
standard (xxxx) and local deviant (xxxY) sequences, and the z-score of the mismatch response (xxxY–xxxx). A bilateral anterior positivity and posterior
negativity are evoked by local deviant sounds. (B and C) Global effect. Averaged ERPs across the time window 270–320 ms (in B) and 900–1200 ms (in C)
after the onset of the fourth sound in global standard (frequent) and global deviant (rare) sequences, and the z-score of the difference (rare–frequent). A left-
lateralized positivity over anterior electrodes, followed by a late negativity over frontal electrodes, are seen for rare violations relative to the rule-governed
sequences.
A. Basirat et al. / Cognition 132 (2014) 137–150 143

observed at both time-windows: at 270–370 ms on the left
anterior electrodes (global deviance stimulus number:
F(1,28) = 5.58, p< 0.05) and at 900–1200 ms on frontal
electrodes (mean of difference = 6.67
l
v, t=3.08,
p< 0.01). This is particularly visible in Fig. 6B for the NSW.
The preceding analysis compared xxxx trials coming
from different experimental blocks. As a control, we
checked whether these blocks already differed prior to
the fourth sound. We compared brain responses to the first
three sounds in xxxx and xxxY blocks and in the learning
and test phases during the early time-window (i.e. 270–
370 ms after the onset of the sounds) on the left anterior
electrodes which showed a global effect. We thus per-
formed a repeated-measure ANOVA with stimulus number
(first, second or third sound), phase (learning vs. test) and
rule (xxxx vs. xxxY) as within-subject factors. The ANOVA
showed a significant effect of stimulus number
(F(1,28) = 31.14, p< 0.0001) but no effect of experimental
phase (F(1,28) = 2.23) or rule type (F(1, 28) = 2.96) was
observed, nor any interaction of these factors. Post hoc
analyses restricted to the first and second sounds and to
the second and third sounds revealed a significantly larger
response to the first than to the second sound
(F(1,28) = 34.34, p< 0.0001) while the response to the sec-
ond and the third sounds was not significantly different
(F(1,28) < 1), as already described in Dehaene-Lambertz
and Dehaene (1994).
3.5. Source localization
Although our main goal was to determine the temporal
dynamics and functional properties of auditory novelty
responses in infants, our high-density recordings allow
some tentative inferences about the underlying cortical
generators. Fig. 7 displays the difference in source currents
associated with the local and global effects in the early and
late time-windows studied above. Bilateral temporal, infe-
rior parietal and ventral and dorsal prefrontal cortices are
involved in the early local and global effect (i.e. 270–
370 ms after the onset of the 4th sound). During the late
time window of the global effect (900–1200 ms after the
onset of the 4th sound), activity remained in a large net-
work of distributed areas mainly in the left hemisphere
involving notably the left inferior frontal region. For illus-
tration, Fig. 7 shows the reconstructed temporal profile of
activations in two exemplary regions, the left superior
temporal (primarily showing repetition suppression and
the local effect) and inferior frontal regions (primarily
showing the global effect).
4. Discussion
In agreement with earlier studies using an auditory
oddball paradigm (Dehaene-Lambertz & Dehaene, 1994;
Friederici et al., 2002), we observed that the detection of
a change in a series of repeated sounds induces two succes-
sive brain responses in awake three-month-old infants: a
mismatch response followed by a late frontal negativity.
By using the local–global paradigm (Bekinschtein et al.,
2009), we now demonstrate that these responses corre-
spond to partially dissociable processing stages, related
respectively to the detection of local probability changes
vs. violations of a global sequence.
4.1. An early mismatch response to local violations
We recorded a classical early mismatch response
(around 300 ms) when a novel vowel was introduced
Fig. 6. Grand average of the voltage across the time-windows and clusters of electrodes selected for the MMR (A) and the NSW (B). For the MMR (A), the
response was larger for xxxY trials relative to xxxx trials (local effect), both for frequent sequences (global standards) and for rare sequences (global
deviants). In addition, over the left anterior cluster, global deviant trials had a higher amplitude than global standard trials (p< .0001). This was not the case
over the right anterior cluster, leading to a significant interaction of global deviance with hemisphere (p< .05). For the NSW (B), a larger negativity was
recorded for rare sequences relative to frequent sequences (global effect). The topographies of this global effect (rare–frequent) are presented below,
separately for xxxx and xxxY sequences.
144 A. Basirat et al. / Cognition 132 (2014) 137–150

following a series of repeated vowels (xxxY trials). As is
well known from earlier research (e.g. Dehaene-Lambertz
& Dehaene, 1994; Friederici et al., 2002; Stefanics et al.,
2007), this effect, analogous to the adult MMN, indicates
that the infant brain is sensitive to elementary statistical
regularities, such as the fact that one sound is presented
more frequently than the other.
There is a debate as to whether the adult MMN can be
explained solely by a passive process of adaptation to rep-
etition (May & Tiitinen, 2010) or implies an active mecha-
nism of prediction and/or comparison of the past with the
present (Näätänen et al., 2007; Wacongne et al., 2012;
Winkler, 2007). According to predictive coding models,
the amplitude of the auditory response is determined by
the difference between the current incoming auditory
input and its prediction based on the temporal contingen-
cies experienced in the recent past (Bendixen, SanMiguel,
& Schröger, 2012; Bendixen, Schröger, & Winkler, 2009;
Friston, 2005; Wacongne et al., 2012; Winkler & Czigler,
2012). Wacongne et al. (2012) proposed a neuronal imple-
mentation of such a predictive model where the mismatch
response is generated by ‘‘prediction errors’’ neurons
which sum excitatory input from the thalamus and inhib-
itory input from a population of predictive neurons in
supra-granular layers.
In adults, the predictive coding hypothesis is supported
by the observation of an MMN to omitted sounds in a pat-
terned sequence (Bendixen et al., 2009; Wacongne et al.,
2012), or to rare repeated sounds (AA) amidst frequent dif-
ferent ones (AB) (Horváth & Winkler, 2004; Wacongne
et al., 2011). Both of these findings are difficult to explain
by a purely passive adaptation process. In infants, although
fewer data is available, He, Hotson, and Trainor (2009) also
showed that 4-month-old infants display a mismatch
response to rare repeated pairs (AA) amidst frequent dif-
ferent ones (AB), suggesting that a similar activation pre-
dictive mechanism may already be present at four month
of age (evidence was ambiguous in younger infants).
4.2. The early response is modulated by global context
Our three-month-olds’ local mismatch response was
larger on xxxx blocks, where xxxY deviants were rare, than
on xxxY blocks, when the deviants were frequent and lar-
gely predictable. Similarly, the response to xxxx trials was
larger when these trials were rare (in xxxY blocks), than
when they were frequent (in xxxx blocks). As a result,
the local effect (difference between xxxY and xxxx trials)
was larger when these sequences were rare than when
they were frequent (Fig. 6). These effects also induced a
significant difference between rare and frequent segments
(early global effect), mostly visible on the left anterior
channels.
Although these findings adds complexity to the simple
two-stage picture of the local–global design, with an early
MMR to local violations and a late SNW to global violations
(Bekinschtein et al., 2009), a modulation of the early MMR
according to context is not surprising given that the struc-
ture of the blocks affects sound transition probabilities. In
adults too, although the first publication of the local–global
design did not detect any interaction of the two effects
(Bekinschtein et al., 2009), a more recent multivariate
Fig. 7. Approximate reconstruction of cortical sources for the local and global effects, as computed from the grand average. (A) In each panel, the colors,
projected on an infant’ left- and right-hemispheric cortical surfaces, code for the difference in source currents associated with the experimental
comparisons depicted in Fig. 5. Bilateral temporal, inferior parietal and ventral and dorsal prefrontal cortices are involved in the early local and global effect
(i.e. 270–370 ms after the onset of the 4th sound). Left inferior frontal areas are involved in the late global effect (900–1200 ms after the onset of the 4th
sound). (B) Time courses represent the mean activity in two representative regions of interest: left inferior frontal cortex and left superior temporal cortex.
Two temporally distinct responses to violations are clearly visible in the time windows where a mismatch response (MMR) and a Negative Slow Wave
(NSW) are seen on the scalp (see Fig. 5). The left superior temporal cortex shows primarily an early MMR to local deviance (xxxY relative to xxxx trials). The
left frontal region is involved in the detection of global deviants, with an earlier response when the rare sequence is xxxY than when it is xxxx.
A. Basirat et al. / Cognition 132 (2014) 137–150 145

pattern analysis of adult data demonstrated such an effect
as early as 150 ms in attentive subjects (King et al., 2013).
There are several possible interpretations of this early
global difference. One interpretation that we can exclude
is a stronger attention to one type of block than to the
other. All audio–visual blocks comprised face movements
and speech stimuli attractive for infants of this age and
an analysis performed on the first three sounds of trials
showed no difference between blocks. Furthermore,
because our rare vs. frequent comparison is orthogonal to
the block factor, a difference in attention between blocks
would not be sufficient to explain this early global
difference.
A more likely interpretation is that the amplitude of the
auditory response is modulated by the frequency of occur-
rence of sounds and their transitions, not only on a given
trial, but accumulated from one trial to the next as imple-
mented in a recent model of the adult MMN (Wacongne
et al., 2012). It has already been shown that the mismatch
response is modulated by deviant frequency in adults: its
amplitude is increasingly larger as the event frequency is
low (Matuoka et al., 2006; Sato et al., 2000). It is also elic-
ited after a smaller number of repetitions when the
repeated sound is kept constant across trials than when
it changes from one trial to the next (Winkler et al.,
1996). Here by necessity, the proportion of Y vowels was
higher in xxxY blocks than in xxxx blocks (18.75% vs.
6.25% during the test phase). Thus, as in adults, the infant
mismatch response to a local change appears to be modu-
lated by the block structure, being larger on xxxx blocks
where the Y sound is less frequent.
What about xxxx trials? Even there, a similar explana-
tion may hold. During the test phase, the occurrence of x
stimuli was 93.75% in the blocks following the xxxx rule,
vs. only 81.25% in the xxxY blocks. Furthermore, in terms
of transition probabilities, the x–x transition occurred on
91.7% of test trials in xxxx blocks, vs. 75% in xxxY blocks.
Here again, a modulation of early auditory responses by
the frequency of occurrence of sounds and/or their transi-
tions, and thus by local sound predictability, can explain
the difference between frequent and rare xxxx segments.
This interpretation suggests that a record of event frequen-
cies, or conditional transition probabilities, is accumulated
from one trial to the next, being only limited by the decay
of sound representations in auditory memory.
We cannot exclude, however, another interpretation,
according to which even global rules impact on early
responses. The two rules that we tested, a strict repetition
in xxxx blocks and a final deviant sound in xxxY blocks, are
simple enough that they might be partially learned at an
early auditory level (Bendixen et al., 2012), attenuating
the auditory response by a higher-order ‘‘global’’ expecta-
tion of the overall sequence pattern. Two experiments in
infants suggest that some simple regular sequences may
indeed modulate responses in the auditory cortex.
Stefanics et al. (2007) recorded a weaker negativity in neo-
nates between 150 and 250 ms in a regular sequence
(xxxxY) relative to a random sequence of the same x and
Y tones, suggesting an amplitude modulation of the early
response by sound predictability (see also He et al.,
2009). Carral et al. (2005) observed a mismatch response
in neonates to changes in the direction of pitch change
between two tones in a pair (e.g. a deviant descending-
pitch pair within a series of ascending-pitch pairs) inde-
pendently of their absolute frequency. These experiments
emphasize that even in infants, prediction of the following
sound is an early-functioning learning mechanism.
In adults, the early global effect was only observed in
attentive adults (King et al., 2013). Here, our infants were
awake and attentive, as they were strongly attracted to
the speaking faces, as in He et al.’s study (2009), but they
were asleep in Stefanics et al.’s (2007), and in Carral
et al.’s (2005) studies. Further studies are needed to specify
what type of auditory patterns may or may not be learned
in inattentive or even sleeping infants. Ascending and
descending pairs (Carral et al., 2005) and regular vs. ran-
dom presentation (Stefanics et al., 2007) might be consid-
ered as gestalts immediately identified at the auditory level
(Endress, Dehaene-Lambertz, & Mehler, 2007), whereas
more complex patterns, such as the repetition of xxxY seg-
ments, might need an attentive participant before being
internalized at the auditory level.
4.3. Left–right hemispheric differences
Spatial differences in electrophysiological recordings
must be treated cautiously, as event-related potentials
are limited in their spatial resolution and different cortical
sources may mix on the scalp, making it difficult to infer
cortical sources from surface recordings. Reconstruction
of cortical sources should also be interpreted with caution,
given the anatomical asymmetries present in the human
brain from the fetal age on (Glasel et al., 2011), the lack
of individual MRI in our data-set, and the inevitable reduc-
tion in accuracy incurred when modeling the group aver-
age rather than individual subjects. However, the distinct
difference in amplitude modulation observed over the left
and right anterior channels for the early local and global
effect might suggest that the two hemispheres do not react
to exactly the same information. Numerous studies have
shown that the left and right hemisphere follow a different
developmental time-course from the preterm period on,
with numerous regions of the left hemisphere slightly
delayed relative to the right. Sulci appear generally earlier
in the right than in the left hemisphere (Dubois et al.,
2008), cerebral blood flow is larger in the right than in
the left hemisphere during infancy (Chiron et al., 1997;
Lin et al., 2012) and the superior temporal region show
indices of faster maturation during the first months of life
(Leroy et al., 2011). Furthermore, processing differences
are also observed early on: processing of fast temporal
transitions is preferentially performed in the left hemi-
sphere whereas spectral information is preferentially
channeled to the right hemisphere already during infancy
(Bristow et al., 2009; Dehaene-Lambertz et al., 2010;
Mahmoudzadeh et al., 2013; Perani et al., 2010;
Telkemeyer et al., 2011). The present experiment provides
yet another observation of the different functional proper-
ties displayed very early on by the left and right hemi-
spheres in human infants. It might suggest that left and
right auditory buffers might be of different size, allowing
the left hemisphere to keep track of a sound during a
146 A. Basirat et al. / Cognition 132 (2014) 137–150

longer time-window, or that top-down modulation from
higher-levels regions are stronger on the left than on the
right hemisphere: Whereas maturation indices are delayed
in the left temporal region relative to the right, the reverse
pattern is observed in Broca’s area (Leroy et al., 2011) and
in the arcuate fasciculus (Dubois et al., 2009), suggesting a
stronger short-term auditory loop in the left hemisphere
than in the right (Dehaene-Lambertz et al., 2006; Leroy
et al., 2011) from the first months of life on.
4.4. A Negative Slow Wave (NSW) response to global
violations
During the late time-window (900–1200 ms), we
observed a specific response to violations of the global
sequence pattern. A bilateral frontal negativity was
observed only for the global, but not local, deviance. A sim-
ilar late response has been previously reported by
Dehaene-Lambertz and Dehaene (1994) and Friederici
et al. (2002) in awake infants after a deviant sound. This
response was similar in topography and latency to the neg-
ative slow wave evoked by deviant visual events in visual
oddball experiments, and which is interpreted either as
an orientation response to novelty, or as a generalized
arousal effect (see Nelson, 1994, and for review, Csibra
et al., 2008). Taken together, these results therefore sug-
gest that, similarly to the adult P300 response, the infant’s
negative slow wave indexes a multimodal inferential sys-
tem capable of detecting cross-modal regularities (Ritter,
Sussman, Deacon, Cowan, & Vaughan, 1999) and that
infants not only encode the present incoming stimulus,
but are also able to place it in a more global and abstract
context.
Our auditory–visual paradigm was designed to maxi-
mally engage the infant. At this age when infants learn
the vowel repertoire of their native language and to associ-
ate mouth movement and speech sounds (Kuhl & Meltzoff,
1982; Patterson & Werker, 2003), talking faces are strong
attractors for infants. Thus even in the case of repeated
xxxY, the change of vowel might still have grabed infants’
attention, but rare xxxY among frequent xxxx were obvi-
ously noticeable given the large auditory and visual differ-
ence between /a/ and /i/. This huge effect explains that our
first analysis also uncovered a late local effect during the
time window of the NSW. Most important, however, is
the observation that the mere repetition of the same
vowel, when it was unexpected (rare xxxx among frequent
xxxY) elicited a long-lasting NSW (around a second). That
NSW can only be explained by the novelty of the overall
sequence, signaling an orientation of infants’ attention
due to the violation of their expectation.
Detecting that a monotonic sequence xxxx is deviant
suggests a hierarchical novelty-detection process: as
shown by the early mismatch response, the infant brain
reacts to deviant final sounds (xxxY), but it is also capable
of detecting the absence of such a local deviant in the xxxx
sequence, which, in turn, signals that this sequence is in
fact novel. The existence of a brain response to rare xxxx
trials has therefore been interpreted as fitting with the
hierarchical Bayesian framework (Wacongne et al., 2011),
according to which the brain generates predictive signals
at multiple hierarchical levels, and event-related response
reflect a sequence of error responses that index the differ-
ence between the predicted and actual signals (Friston,
2005; Rao & Ballard, 1999). Like adults’, infants’ brain
responses to auditory violations would be separated into
two stages: the first one mainly driven by the local proba-
bility of occurrence of a sound, and the second sensitive to
the overall regularity of the global sequence of sounds rel-
ative to past sequences. Hierarchical Bayesian inference
would thus already be available to infants, as also sug-
gested by other predictive and surprise paradigms (e.g.
Gopnik & Schulz, 2004; Gweon, Tenenbaum, & Schulz,
2010; Tenenbaum et al., 2011; Téglás et al., 2011).
Although source localization is only indicative in
infants, it ascribed the late global effect to a prolonged
activity in a left inferior frontal source over Broca’s area
whereas the sources of the local and global mismatch
response were predominantly assigned to a early phasic
activation in the superior temporal cortex (Fig. 7).
Although speculative at this time, it is interesting to under-
line potential similarities between our infant sources and
the adult sources reported using MEG by several teams
using similar auditory paradigms (Halgren et al., 2011;
Wacongne et al., 2011). These authors reported a sequence
of activations, first confined to temporal areas, then involv-
ing a broader network expanding notably into frontal areas
around 250 ms. Once again, infants appear to show a sim-
ilar sequence, only at a much slower pace, compatible with
their reduced synaptic connectivity and weak myelination
(see also Kouider et al., 2013 for another delayed response
signaling higher-level processing). The involvement of a
similar cortical network in infants and adults should obvi-
ously be addressed in future studies using more precise
localization methods such as fMRI.
5. Limits, implications and perspectives
The present experimental design leaves open the degree
of abstractness of the global sequence pattern that infants
detected, a point that will require further experimentation.
Infants may be memorizing the recurrence of a particular
sequence of vowels in a given 5-min block (e.g. a-a-a-i).
Indeed, a memory for specific words (Benavides-Varela
et al., 2011; Shi, Werker, & Morgan, 1999) and sentences
(Dehaene-Lambertz et al., 2006) is known to be available
in the first few months of life. Alternatively, they may be
abstracting away from the individual phonetic items and
storing a more abstract rule such as xxxY (‘‘3 identical
items are followed by a different one’’) or yet more abstract
rules (e.g. ‘‘the last item is different’’).
The ability of older infants to quickly infer abstract reg-
ularities after a brief exposure to sequences of auditory or
visual stimuli has been vastly studied (review in Aslin &
Newport, 2012; Johnson et al., 2009; Kirkham, Slemmer,
Richardson, & Johnson, 2007; Marcus, Fernandes, &
Johnson, 2007; Marcus, Vijayan, Bandi Rao, & Vishton,
1999; Saffran, Pollak, Seibel, & Shkolnik, 2007; Saffran &
Thiessen, 2003). In a classical study, Marcus et al. (1999)
showed that 7-month-old infants which were familiarized
to rules like ABA (e.g. ga ti ga, po tu po) or ABB (e.g. ga ti ti,
A. Basirat et al. / Cognition 132 (2014) 137–150 147

po tu tu) were able to extract the abstract rule and gener-
alize it to new syllables, thus detecting sequences that vio-
lated the learned rule. Using the same type of paradigm,
Frank, Slemmer, Marcus, and Johnson (2009) observed that
5-month-old infants could learn ABA or AAB rules when
they were presented in an audio–visual modality: a sylla-
ble was accompanied by a colored looming shape (e.g.
‘‘ba-octagon de-square ba-octagon’’ for ABA rule). How-
ever, no significant effect of abstract rule extraction was
observed when the rule was presented in the auditory or
the visual modalities alone. Similarly, Gervain, Macagno,
Cogoi, Peña, and Mehler (2008), using near-infrared spec-
troscopy, observed that neonates were sensitive to the
repetitive structure ABB of auditory pseudowords relative
to a random structure ABC, but did not differentiate the
ABA structure from the random ABC, suggesting sensitivity
to adjacent repetitions only. To our knowledge, there is no
other report in the literature on the rule-learning abilities
of younger infants. In the future, the present paradigm
could be used to determine whether three-month-old
infants can already detect an abstract rule, such as ‘‘the last
item is different’’, for instance by varying the items during
the learning phase (e.g. mixing a-a-a-i and i-i-i-a) and test-
ing for brain responses to rule-respecting and rule-violat-
ing sequences with a variable degree of generalization to
novel items (see Fitch & Friederici, 2012).
Besides its relevance for temporal-sequence learning,
the local–global test was first introduced as a probe of con-
scious-level processing in adult brain-lesioned patients
with vegetative state and related disorders (Bekinschtein
et al., 2009). The rationale was that (1) the identification,
over multiple trials, of a recurring global sequence such
as xxxY requires a temporal scope which is beyond the typ-
ical exponential decay of unconscious processes, and
requires an explicit maintenance in conscious working
memory (Dehaene & Naccache, 2001); and (2) global devi-
ants elicit a late P3b electrophysiological response which,
in many paradigms contrasting conscious vs. non-con-
scious stimuli, has been repeatedly demonstrated to corre-
late with conscious-level processing in healthy subjects
(Donchin & Coles, 1988; Sergent et al., 2005; review in
Dehaene & Changeux, 2011). Subsequently, empirical stud-
ies using the local–global paradigm in adult brain-lesioned
patients with coma, vegetative state or minimal conscious-
ness indeed demonstrated that the presence of the global
P3b response almost invariably indexes a conscious subject
(Bekinschtein et al., 2009; Faugeras et al., 2011), while the
earlier local response (mismatch negativity) remains pres-
ent in many comatose subjects (Fischer, Luauté, Adeleine,
& Morlet, 2004; Fischer, Luauté, & Morlet, 2010).
Following this logic, the unambiguous finding that the
brain of three-month-old infants responds to global audi-
tory violations suggests the tentative conclusion that an
elementary form of conscious-level processing may
already be operative in three-month-old infants. This con-
clusion fits with our recent observation of a tight parallel
between infants’ and adults’ event-related responses to
visible and invisible masked faces, also suggesting that a
late frontal negative response characterizes conscious
access to visual stimuli in infants (Kouider et al., 2013).
Lagercrantz and Changeux (2009, 2010) review the
considerable, though still highly fragmentary, pediatric
and neuroimaging evidence suggesting that even new-
borns may already present integrative responses to pain,
olfactory, visual and auditory stimulation. Long-distance
bi-hemispheric cortical networks thought to support a
‘‘conscious global workspace’’ (Dehaene, Kerszberg, &
Changeux, 1998; Dehaene & Naccache, 2001) are already
anatomically present (Dubois et al., 2008; Takahashi,
Folkerth, Galaburda, & Grant, 2012) and may already be
functionally operative in the first few months of life
(Dehaene-Lambertz et al., 2002; Fransson et al., 2007;
Kouider et al., 2013).
At present, any conclusion on this topic must remain
speculative, as we cannot exclude that specific mecha-
nisms of rule extraction may be operative non-consciously
in three-month-old infants, but may require consciousness
at a later age. Extensions of the present paradigm, compar-
ing responses arising from various levels of cortical inte-
gration in infants and adults with variable degrees of
residual consciousness, could be useful in order to estab-
lish more unambiguous behavioral and electrophysiologi-
cal signatures of the level of consciousness during infancy.
6. Conclusion
The present study provides evidence that hierarchical
brain mechanisms underlying temporal sequence learning
are already present at three months of life. At least two dis-
tinct cortical systems, one sensitive to local probabilities
and the other to global regularities, are present in young
infants. As in adults, violations of local and global regular-
ities affect distinct time windows: only global, but not local
violations, elicit a late anterior negative slow wave over
bilateral frontal areas. In that respect, the present results
fit with previous brain imaging studies demonstrating that
frontal regions are already active in three-month-old
infants (Bristow et al., 2009; Dehaene-Lambertz et al.,
2002). To our knowledge, the present study provides the
earliest evidence so far that young infants can process mul-
tiple levels of statistical violations, and supports the
emerging Bayesian framework (Tenenbaum et al., 2011)
according to which hierarchical learning mechanisms
may be operative early on during development.
Acknowledgments
This research was supported by European Research
Council (NeuroConsc Program), CEA (Commissariat à
l’Energie Atomique et aux Energies Alternatives) and
INSERM (Institut National de la Santé et de la Recherche
Médicale). We are particularly grateful to Giovanna Sant-
oro and Gaëlle Mediouni for their help in baby recruitment
and testing.
References
Aslin, R. N., & Newport, E. L. (2012). Statistical learning: From acquiring
specific items to forming general rules. Current Directions in
Psychological Science, 21(3), 170–176.
Atienza, M., Cantero, J. L., & Gomez, C. M. (1997). The mismatch negativity
component reveals the sensory memory during REM sleep in humans.
Neuroscience Letters, 237(1), 21–24.
148 A. Basirat et al. / Cognition 132 (2014) 137–150

Bekinschtein, T. A., Dehaene, S., Rohaut, B., Tadel, F., Cohen, L., & Naccache,
L. (2009). Neural signature of the conscious processing of auditory
regularities. Proceedings of the National Academy of Sciences of the
United States of America, 106(5), 1672–1677.
Benavides-Varela, S., Gómez, D., Macagno, F., Bion, R., Peretz, I., & Mehler,
J. (2011). Memory in the neonate brain. PLoS One, 6(11), e27497.
Bendixen, A., SanMiguel, I., & Schröger, E. (2012). Early
electrophysiological indicators for predictive processing in audition:
A review. International Journal of Psychophysiology, 83(2), 120–131.
Bendixen, A., Schröger, E., & Winkler, I. (2009). I heard that coming:
Event-related potential evidence for stimulus-driven prediction in the
auditory system. The Journal of Neuroscience, 29(26), 8447–8451.
Bristow, D., Dehaene-Lambertz, G., Mattout, J., Soares, C., Gliga, T., Baillet,
S., et al. (2009). Hearing faces: How the infant brain matches the face
it sees with the speech it hears. Journal of Cognitive Neuroscience,
21(5), 905–921.
Carral, V., Huotilainen, M., Ruusuvirta, T., Fellman, V., Näätänen, R., &
Escera, C. (2005). A kind of auditory ‘‘primitive intelligence’’ already
present at birth. European Journal of Neuroscience, 21(11), 3201–3204.
Celsis, P., Boulanouar, K., Doyon, B., Ranjeva, J. P., Berry, I., & Chollet, F.
(1999). Differential fMRI responses in the left posterior superior
temporal gyrus and left supramargynal gyrus to habituation and
change detection in syllables and tones. NeuroImage, 9, 135–144.
Cheour, M., C
ˇe
˙poniené, R., Leppänen, P., Alho, K., Kujala, T., Renlund, M.,
et al. (2002). The auditory sensory memory trace decays rapidlyin
newborns. Scandinavian Journal of Psychology, 43(1), 33–39.
Chiron, C., Jambaque, I., Nabbout, R., Lounes, R., Syrota, A., & Dulac, O.
(1997). The right brain hemisphere is dominant in human infants.
Brain, 120, 1057–1065.
Csibra, G., Kushnerenko, E., & Grossmann, T. (2008). Electrophysiological
methods in studying infant cognitive development. In C. A. Nelson &
M. Luciana (Eds.), Handbook of developmental cognitive neuroscience
(2nd ed., pp. 247–262). Cambridge, Massachusetts: MIT Press.
Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical
approaches to conscious processing. Neuron, 70(2), 200–227.
Dehaene, S., Kerszberg, M., & Changeux, J.-P. (1998). A neuronal model of
a global workspace in effortful cognitive tasks. Proceedings of the
National Academy of Sciences of the United States of America, 95(24),
14529–14534.
Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of
consciousness: Basic evidence and a workspace framework. Cognition,
79(1–2), 1–37.
Dehaene-Lambertz, G., & Dehaene, S. (1994). Speed and cerebral
correlates of syllable discrimination in infants. Nature, 370(6487),
292–295.
Dehaene-Lambertz, G., Dehaene, S., & Hertz-Pannier, L. (2002). Functional
neuroimaging of speech perception in infants. Science, 298(5600),
2013–2015.
Dehaene-Lambertz, G., & Gliga, T. (2004). Common neural basis for
phoneme processing in infants and adults. Journal of Cognitive
Neuroscience, 16(8), 1375–1387.
Dehaene-Lambertz, G., Hertz-Pannier, L., Dubois, J., Mériaux, S., Roche, A.,
Sigman, M., et al. (2006). Functional organization of perisylvian
activation during presentation of sentences in preverbal infants.
Proceedings of the National Academy of Sciences of the United States of
America, 103(38), 14240–14245.
Dehaene-Lambertz, G., Montavont, A., Jobert, A., Allirol, L., Dubois, J., &
Hertz-Pannier, L. (2010). Language or music, mother or mozart?
Structural and environmental influences on infants’ language
networks. Brain and Language, 114(2), 53–65.
Dehaene-Lambertz, G., & Peña, M. (2001). Electrophysiological evidence
for automatic phonetic processing in neonates. Neuroreport, 12(14),
3155–3158.
Donchin, E., & Coles, M. (1988). Is the P300 component a manifestation of
context updating? Behavioral and Brain Sciences, 11(3), 357–374.
Dubois, J., Benders, M., Cachia, A., Lazeyras, F., Ha-Vinh Leuchter, R.,
Sizonenko, S. V., et al. (2008). Mapping the early cortical folding
process in the preterm newborn brain. Cerebral Cortex, 18(6),
1444–1454.
Dubois, J., Hertz-Pannier, L., Cachia, A., Mangin, J.-F., Le Bihan, D., &
Dehaene-Lambertz, G. (2009). Structural asymmetries in the infant
language and sensori-motor networks. Cerebral Cortex, 19(2),
414–423.
Endress, A. D., Dehaene-Lambertz, G., & Mehler, J. (2007). Perceptual
constraints and the learnability of simple grammars. Cognition,
105(3), 577–614.
Faugeras, F., Rohaut, B., Weiss, N., Bekinschtein, T. A., Galanaud, D.,
Puybasset, L., et al. (2011). Probing consciousness with event-related
potentials in the vegetative state. Neurology, 77(3), 264–268.
Fischer, C., Luauté, J., Adeleine, P., & Morlet, D. (2004). Predictive value of
sensory and cognitive evoked potentials for awakening from coma.
Neurology, 63(4), 669–673.
Fischer, C., Luauté, J., & Morlet, D. (2010). Event-related potentials (MMN
and novelty P3) in permanent vegetative or minimally conscious
states. Clinical Neurophysiology, 121(7), 1032–1042.
Fischer, C., Morlet, D., Bouchet, P., Luauté, J., Jourdan, C., & Salord, F.
(1999). Mismatch negativity and late auditory evoked potentials in
comatose patients. Clinical Neurophysiology, 110(9), 1601–1610.
Fitch, W., & Friederici, A. D. (2012). Artificial grammar learning meets
formal language theory: An overview. Philosophical Transactions of the
Royal Society B: Biological Sciences, 367(1598), 1933–1955.
Frank, M. C., Slemmer, J. A., Marcus, G. F., & Johnson, S. P. (2009).
Information from multiple modalities helps 5-month-olds learn
abstract rules. Developmental Science, 12(4), 504–509.
Fransson, P., Skiöld, B., Horsch, S., Nordell, A., Blennow, M.,
Lagercrantz, H., et al. (2007). Resting-state networks in the
infant brain. Proceedings of the National Academy of Sciences of
the United States of America, 104(39), 15531–15536.
Friederici, A. D., Friedrich, M., & Weber, C. (2002). Neural manifestation of
cognitive and precognitive mismatch detection in early infancy.
Neuroreport, 13(10), 1251–1254.
Friston, K. (2005). A theory of cortical responses. Philosophical Transactions
of the Royal Society B: Biological Sciences, 360(1456), 815–836.
Garrido, M. I., Friston, K. J., Kiebel, S. J., Stephan, K. E., Baldeweg, T., &
Kilner, J. M. (2008). The functional anatomy of the MMN: A DCM
study of the roving paradigm. NeuroImage, 42(2), 936–944.
Garrido, M. I., Kilner, J. M., Kiebel, S. J., & Friston, K. J. (2007). Evoked brain
responses are generated by feedback loops. Proceedings of the National
Academy of Sciences of the United States of America, 104(52),
20961–20966.
Gervain, J., Macagno, F., Cogoi, S., Peña, M., & Mehler, J. (2008). The
neonate brain detects speech structure. Proceedings of the National
Academy of Sciences of the United States of America, 105(37),
14222–14227.
Glasel, H., Leroy, F., Dubois, J., Hertz-Pannier, L., Mangin, J.-F., & Dehaene-
Lambertz, G. (2011). A robust cerebral asymmetry in the infant brain:
The rightward superior temporal sulcus. NeuroImage, 58(3), 716–723.
Gopnik, A., & Schulz, L. (2004). Mechanisms of theory formation in young
children. Trends in Cognitive Sciences, 8(8), 371–377.
Gweon, H., Tenenbaum, J. B., & Schulz, L. E. (2010). Infants consider
both the sample and the sampling process in inductive
generalization. Proceedings of the National Academy of Sciences
of the United States of America, 107(20), 9066–9071.
Halgren, E., Sherfey, J., Irimia, A., Dale, A., & Marinkovic, K. (2011).
Sequential temporo-fronto-temporal activation during monitoring of
the auditory environment for temporal patterns. Human Brain
Mapping, 32(8), 1260–1276.
He, C., Hotson, L., & Trainor, L. (2009). Development of infant mismatch
responses to auditory pattern changes between 2 and 4 months old.
European Journal of Neuroscience, 29(4), 861–867.
Horváth, J., & Winkler, I. (2004). How the human auditory system treats
repetition amongst change. Neuroscience Letters, 368(2), 157–161.
Johnson, S. P., Fernandas, K. J., Frank, M. C., Kirkham, N., Marcus, G.,
Rabagliati, H., et al. (2009). Abstract rule learning for visual sequences
in 8- and 11-month-olds. Infancy, 14(1), 2–18.
Kabdebon, C., Leroy, F., Simonnet, H., Perrot, M. Dubois, J., & Dehaene-
Lambertz, G. (submitted for publication). Anatomical correlations of
the international 10–20 sensor placement system in infants.
King, J. R., Faugeras, F., Gramfort, A., Schurger, A., El Karoui, I., Sitt, J., et al.
(2013). Single-trial decoding of auditory novelty responses facilitates
the detection of residual consciousness. NeuroImage, 83, 726–738.
Kirkham, N. Z., Slemmer, J. A., Richardson, D. C., & Johnson, S. P. (2007).
Location, location, location: Development of spatiotemporal sequence
learning in infancy. Child Development, 78(5), 1559–1571.
Kouider, S., Stahlhut, C., Gelskov, S. V., Barbosa, L. S., Dutat, M., de
Gardelle, V., et al. (2013). A neural marker of perceptual
consciousness in infants. Science, 340(6130), 376–380.
Kuhl, P. K., & Meltzoff, A. N. (1982). The bimodal perception of speech in
infancy. Science, 218(4577), 1138–1141.
Lagercrantz, H., & Changeux, J.-P. (2009). The emergence of human
consciousness: From fetal to neonatal life. Pediatric Research, 65(3),
255–260.
Lagercrantz, H., & Changeux, J.-P. (2010). Basic consciousness of the
newborn. Seminars in Perinatology, 34(3), 201–206.
Leroy, F., Glasel, H., Dubois, J., Hertz-Pannier, L., Thirion, B., Mangin, J.-F.,
et al. (2011). Early maturation of the linguistic dorsal pathway in
human infants. The Journal of Neuroscience, 31(4), 1500–1506.
A. Basirat et al. / Cognition 132 (2014) 137–150 149

Lin, P.-Y., Roche-Labarbe, N., Dehaes, M., Fenoglio, A., Grant, P. E., &
Franceschini, M. A. (2012). Regional and hemispheric asymmetries of
cerebral hemodynamic and oxygen metabolism in newborns. Cerebral
Cortex.http://dx.doi.org/10.1093/cercor/bhs023.
Mahmoudzadeh, M., Dehaene-Lambertz, G., Fournier, M., Kongolo, G.,
Goudjil, S., Dubois, J., et al. (2013). Syllabic discrimination in
premature human infants prior to complete formation of cortical
layers. Proceedings of the National Academy of Sciences of the United
States of America, 110(12), 4846–4851.
Mäntysalo, S., & Näätänen, R. (1987). The duration of a neuronal trace of
an auditory stimulus as indicated by event-related potentials.
Biological Psychology, 24(3), 183–195.
Marcus, G., Fernandes, K., & Johnson, S. (2007). Infant rule learning
facilitated by speech. Psychological Science, 18(5), 387–391.
Marcus, G. F., Vijayan, S., Bandi Rao, S., & Vishton, P. M. (1999). Rule
learning by seven-month-old infants. Science, 283(5398), 77–80.
Matuoka, T., Yabe, H., Shinozaki, N., Sato, Y., Hiruma, T., Ren, A., et al.
(2006). The development of memory trace depending on the number
of the standard stimuli. Clinical EEG and Neuroscience, 37(3), 223–229.
May, P., & Tiitinen, H. (2010). Mismatch negativity (MMN), the deviance-
elicited auditory deflection, explained. Psychophysiology, 47(1),
66–122.
Näätänen, R., Gaillard, A., & Mäntysalo, S. (1978). Early selective-attention
effect on evoked potential reinterpreted. Acta Psychologica, 42(4),
313–329.
Näätänen, R., Paavilainen, P., Rinne, T., & Alho, K. (2007). The mismatch
negativity (MMN) in basic research of central auditory processing: A
review. Clinical Neurophysiology, 118(12), 2544–2590.
Nelson, C. A. (1994). Neural correlates of recognition memory in the first
postnatal year of life. In G. Dawson & K. Fischer (Eds.), Human behavior
and the developing brain (pp. 269–313). New York: Guilford Press.
Patterson, M. L., & Werker, J. F. (2003). Two-month old infants match
phonetic information in lips and voice. Developmental Science, 6(2),
193–198.
Pegado, F., Bekinschtein, T. A., Chausson, N., Dehaene, S., Cohen, L., &
Naccache, L. (2010). Probing the lifetimes of auditory novelty
detection processes. Neuropsychologia, 48(10), 3145–3154.
Perani, D., Saccuman, M., Scifo, P., Spada, D., Andreolli, G., Rovelli, R., et al.
(2010). Functional specializations for music processing in the human
newborn brain. Proceedings of the National Academy of Sciences of the
United States of America, 107(10), 4758–4763.
Piaget, J. (1954). The construction of reality in the child. New York: Basic
Books.
Quartz, S. R., & Sejnowski, T. J. (1997). The neural basis of cognitive
development: A constructivist manifesto. Behavioral and Brain
Sciences, 20(4), 537–556. discussion 556–596.
Rao, R. P. N., & Ballard, D. H. (1999). Predictive coding in the visual cortex:
A functional interpretation of some extra-classical receptive-field
effects. Nature Neuroscience, 2(1), 79–87.
Richards, J. E. (2005). Localizing cortical sources of event-related
potentials in infants’ covert orienting. Developmental Science, 8(3),
255–278.
Ritter, W., Sussman, E., Deacon, D., Cowan, N., & Vaughan, H. G. (1999).
Two cognitive systems simultaneously prepared for opposite events.
Psychophysiology, 36(6), 835–838.
Saffran, J. R., Pollak, S. D., Seibel, R. L., & Shkolnik, A. (2007). Dog is a dog is
a dog: Infant rule learning is not specific to language. Cognition,
105(3), 669–680.
Saffran, J. R., & Thiessen, E. D. (2003). Pattern induction by infant language
learners. Developmental Psychology, 39(3), 484–494.
Sato, Y., Yabe, H., Hiruma, T., Sutoh, T., Shinozaki, N., Nashida, T., et al.
(2000). The effect of deviant stimulus probability on the human
mismatch process. NeuroReport, 11(17), 3703–3708.
Sergent, C., Baillet, S., & Dehaene, S. (2005). Timing of the brain events
underlying access to consciousness during the attentional blink.
Nature Neuroscience, 8(10), 1391–1400.
Shi, R., Werker, J. F., & Morgan, J. L. (1999). Newborn infants’ sensitivity to
perceptual cues to lexical and grammatical words. Cognition, 72(2),
B11–B21.
Squires, N., Squires, K., & Hillyard, S. (1975). Two varieties of long-latency
positive waves evoked by unpredictable auditory stimuli in man.
Electroencephalography and Clinical Neurophysiology, 38(4), 387–401.
Stefanics, G., Háden, G., Huotilainen, M., Balázs, L., Sziller, I., Beke, A., et al.
(2007). Auditory temporal grouping in newborn infants.
Psychophysiology, 44(5), 697–702.
Tadel, F., Baillet, S., Mosher, J., Pantazis, D., & Leahy, R. (2011). Brainstorm:
A user-friendly application for MEG/EEG analysis [electronic version].
Computational Intelligence and Neuroscience (13p.).
Takahashi, E., Folkerth, R., Galaburda, A., & Grant, P. (2012). Emerging
cerebral connectivity in the human fetal brain: An MR tractography
study. Cerebral Cortex, 22(2), 455–464.
Téglás, E., Vul, E., Girotto, V., Gonzalez, M., Tenenbaum, J., & Bonatti, L.
(2011). Pure reasoning in 12-month-old infants as probabilistic
inference. Science, 332(6033), 1054–1059.
Telkemeyer, S., Rossi, S., Nierhaus, T., Steinbrink, J., Obrig, H., &
Wartenburger, I. (2011). Acoustic processing of temporally
modulated sounds in infants: Evidence from a combined near-
infrared spectroscopy and EEG study. Frontiers in Psychology, 1,62.
Tenenbaum, J., Kemp, C., Griffiths, T., & Goodman, N. (2011). How to grow
a mind: Statistics, structure, and abstraction. Science, 331(6022),
1279–1285.
Vul, E., & Pashler, H. (2012). Voodoo and circularity errors. NeuroImage,
62(2), 945–948.
Wacongne, C., Changeux, J.-P., & Dehaene, S. (2012). A neuronal model of
predictive coding accounting for the mismatch negativity. The Journal
of Neuroscience, 32(11), 3665–3678.
Wacongne, C., Labyt, E., van Wassenhove, V., Bekinschtein, T. A., Naccache,
L., & Dehaene, S. (2011). Evidence for a hierarchy of predictions and
prediction errors in human cortex. Proceedings of the National
Academy of Sciences of the United States of America, 108(51),
20754–20759.
Wetter, S., Polich, J., & Murphy, C. (2004). Olfactory, auditory, and visual
ERPs from single trials: No evidence for habituation. International
Journal of Psychophysiology, 54(3), 263–272.
Winkler, I. (2007). Interpreting the mismatch negativity. Journal of
Psychophysiology, 21(3), 147–163.
Winkler, I., Cowan, N., Csépe, V., Czigler, I., & Näätänen, R. (1996).
Interactions between transient and long-term auditory memory as
reflected by the mismatch negativity. Journal of Cognitive
Neuroscience, 8(5), 403–415.
Winkler, I., & Czigler, I. (2012). Evidence from auditory and visual event-
related potential (ERP) studies of deviance detection (MMN and
vMMN) linking predictive coding theories and perceptual object
representations. International Journal of Psychophysiology, 83(2),
132–143.
150 A. Basirat et al. / Cognition 132 (2014) 137–150