Infants’ perception of repetition-based regularities in speech: a look from the perspective of the same/different distinction

Abstract

We review the existing evidence, behavioral and neural, of infants’ ability to encode repetition- (‘same’) and diversity- (‘different’) based regularities in speech. These studies show that, from birth, infants exhibit a robust capacity for learning repetition-based rules from speech (e.g. AAB or ABA, in which A = A). Further, the ability to generalize such repetition-based structures is not strictly language-specific, as infants’ extract repetition-based structures from musical tones, animal pictures, abstract geometrical shapes, or faces under some conditions. However, this capacity is strongest when presented with speech or other communicative/meaningful stimuli. Additionally and importantly, recent brain-imaging studies suggest that by six months of age, infants also distinctly encode the notion of difference in speech stimuli. This is the youngest age at which this ability has been shown.

Full text

Infants’
perception
of
repetition-based
regularities
in
speech:
a
look
from
the
perspective
of
the
same/different
distinction
Irene
de
la
Cruz-Pavı
´a
1,2,3,4
and
Judit
Gervain
1,2
We
review
the
existing
evidence,
behavioral
and
neural,
of
infants’
ability
to
encode
repetition-
(‘same’)
and
diversity-
(‘different’)
based
regularities
in
speech.
These
studies
show
that,
from
birth,
infants
exhibit
a
robust
capacity
for
learning
repetition-based
rules
from
speech
(e.g.
AAB
or
ABA,
in
which
A
=
A).
Further,
the
ability
to
generalize
such
repetition-based
structures
is
not
strictly
language-specific,
as
infants’
extract
repetition-based
structures
from
musical
tones,
animal
pictures,
abstract
geometrical
shapes,
or
faces
under
some
conditions.
However,
this
capacity
is
strongest
when
presented
with
speech
or
other
communicative/meaningful
stimuli.
Additionally
and
importantly,
recent
brain-imaging
studies
suggest
that
by
six
months
of
age,
infants
also
distinctly
encode
the
notion
of
difference
in
speech
stimuli.
This
is
the
youngest
age
at
which
this
ability
has
been
shown.
Addresses
1
Department
of
Developmental
and
Social
Psychology,
University
of
Padua,
Padua,
Italy
2
Integrative
Neuroscience
and
Cognition
Center,
CNRS
&
University
of
Paris,
Paris,
France
3
Department
of
Linguistics
and
Basque
Studies,
University
of
the
Basque
Country
(UPV/EHU),
Vitoria-Gasteiz,
Spain
4
IKERBASQUE,
Basque
Foundation
for
Science,
Bilbao,
Spain
Corresponding
author:
Gervain,
Judit
(judit.gervain@unipd.it)
Current
Opinion
in
Behavioral
Sciences
2020,
37:125–132
This
review
comes
from
a
themed
issue
on
Same-different
conceptualization
Edited
by
Edward
A
Wasserman,
Jean-Re
´my
Hochmann
and
Susan
Carey
https://doi.org/10.1016/j.cobeha.2020.11.014
2352-1546/ã
2020
The
Authors.
Published
by
Elsevier
Ltd.
This
is
an
open
access
article
under
the
CC
BY-NC-ND
license
(http://creative-
commons.org/licenses/by-nc-nd/4.0/).
Introduction
Humans’
ability
to
learn
linguistic
rules
is
central
to
debates
about
the
nature
of
the
language
faculty.
A
pivotal
question
in
the
nativist
versus
empiricist
debate
has
been
whether
humans
are
biologically
endowed
with
the
capacity
to
learn
abstract
rules
containing
symbolic
variables.
Such
rules
would
apply
to
any
item
in
a
set
or
category
that
the
variable
is
a
placeholder
for,
thus
guaranteeing
a
combinatorially
productive
system.
Nativist
accounts
[1–3]
argue
that
this
ability
is
a
spe-
cies-specific,
innate
trait,
at
the
core
of
human
language’s
infinite
computational
power.
Empiricist
accounts
[4,5],
by
contrast,
claim
that
while
the
ability
to
learn
abstract,
symbolic
rules
may
be
present
in
adults,
it
is
by
no
means
innate
and
is
learned
from
the
input
through
experience.
Critical to this
debate
then
is infants’ ability
to learn abstract
rules
(see,
for
example,
[6

]
for
a
review).
If
infants,
who
have
little
experience
with
language,
can
nevertheless
learn
rules
from
language,
then
this
ability
must
be
innate.
Probing
infants’
rule
learning
abilities,
Marcus
et
al.
[7

]
showed
that
seven-month-olds
extract
the
identity
rela-
tionship
(A
=
A),
the
simplest
abstract
regularity,
from
artificial
grammars
generating
three-word
sequences
in
which
two
words
are
identical
(e.g.
ABB,
AAB,
ABA).
A
large
body
of
literature
followed,
exploring
the
nature
and
scope
of
infants’
rule-learning
abilities.
Given
the
theoretical
relevance
of
these
questions,
infants
have
been
tested
from
the
youngest
age
possible,
namely
birth.
The
tasks,
stimuli
and
structures
tested
in
this
field
are
often
highly
similar
to
the
ones
used
to
examine
infants’
encoding
of
sameness
and
difference
in
the
conceptual
domain.
Infants’
ability
to
learn
repeti-
tion-based
linguistic
regularities
could,
therefore,
be
rel-
evant
for
our
understanding
of
how
they
encode
the
notions
of
same
and
different.
The
current
paper
reviews
this
empirical
evidence,
behavioral
and
neural,
and
argues
that
by
six
months
of
age,
infants
can
represent
not
only
sameness,
but
also
distinctly
encode
the
notion
of
difference
in
speech
stimuli.
This
is
the
youngest
age
at
which
this
ability
has
been
shown
[8,9].
Behavioral
evidence
In
their
original
study
using
the
head-turn
preference
procedure,
Marcus
et
al.
[7

]
showed
that
seven-month-
old
infants
exposed
to
speech
sequences
rapidly
extract
and
discriminate
structures
containing
adjacent
(ABB:
ga
ti
ti)
and
non-adjacent
repetitions
(ABA:
ga
ti
ga;
but
fail
at
five
months
of
age
[10]),
as
well
as
structures
containing
adja-
cent
repetitions
at
different
positions
(i.e.
ABB:
ga
ti
ti
versus
AAB:
ga
ga
ti).
Importantly,
rather
than
memorizing
item-based
information,
infants
generalize
the
underlying
structure
of
the
sequences
heard
during
familiarization
to
novel
exemplars
presented
in
test
(Figure
1).
Available
online
at
www.sciencedirect.com
ScienceDirect
www.sciencedirect.com
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132

This
robust
finding,
replicated
in
[11

,12

,13]
and
a
wealth
of
subsequent
work
have
revealed
that
learning
repetition-based
structures
is
not
a
language-specific
capacity,
as
infants
extract
such
patterns
from
a
range
of
stimuli,
but
it
is
strongest
when
presented
with
speech
or
other
communicative
or
meaningful
stimuli
(see
Ref.
[14

]
for
a
recent
meta-analysis).
Infants
learn
ABB
and
ABA
generalizations
in
musical
chords
and
tones
at
four
months
of
age,
but
lose
this
ability
by
seven
months
[15],
presumably
having
learned
that
speech,
but
not
music,
is
a
suitable
input
for
this
type
of
rule.
Interestingly,
while
seven-month-old
infants
fail
to
extract
these
rules
directly
from
chords,
tones,
instrument
timbres,
or
animal
sounds,
they
succeed
if
they
are
first
familiarized
with
repetition-
based
structures
implemented
in
speech,
transferring
the
rules
across
stimuli
[12

].
Similarly,
7.5-month-old
infants
learn
repetition-based
rules
from
sine-wave
tones
only
if
they
are
first
primed
to
consider
them
as
communicative
signals,
and
can
even
transfer
the
rules
to
speech
sequences
[16].
Infants’
ability
to
generalize
repetition-based
structures
is
not
limited
to
the
auditory
domain.
At
3–4
months,
infants
discriminate
ABB
and
ABA
patterns
implemented
over
pictures
of
dogs
[17

]
and
at
seven
months,
they
succeed
with
pictures
of
dogs,
cats
[18],
and
upright,
but
not
inverted
faces
[19].
Studies
using
linguistic
visual
stimuli,
that
is,
sign
language,
reveal
an
unclear
picture.
One
study
[20]
finds
that
7.5-month-old
infants
generalize
AAB
but
not
ABB
patterns
in
sign,
while
another
[14

]
reports
that
infants
at
this
age
learn
ABB
and
ABA
patterns
from
sign
only
if
previously
primed
to
treat
them
as
communicative.
This
conclusion
needs
to
be
taken
with
caution,
though,
as
it
is
supported
by
planned
comparisons
but
not
the
study’s
omnibus
statistical
test.
Studies
examining
infants’
ability
to
extract
rules
from
abstract
geometric
shapes,
i.e.
non-meaningful
and
non-
communicative
visual
stimuli,
report
inconsistent
evi-
dence
for
learning
[10,21],
supporting
the
hypothesis
that
meaningless
stimuli
hinder
infants’
rule
learning
abilities
[14

].
However,
two
recent
studies
suggest
that
these
variable
findings
may
instead
result
from
differences
in
stimuli
presentation
[17

,22].
Indeed,
from
three
months
of
age,
infants
generalize
ABB
and
ABA
rules
over
geo-
metric
shapes
if
the
sequences
are
spatially
structured,
that
is,
presented
left-to-right
[17

,22].
Redundancy
within
and
across
modalities
also
enhances
infants’
ability
to
learn
repetition-based
rules.
Thus,
five-
month-old
infants
extract
ABB
and
ABA
rules
from
sequences
of
syllables
paired
with
spatially
unstructured
geometric
shapes,
but
fail
when
presented
with
either
modality
separately
[20,25],
and
seven-month-old
infants’
ability
to
extract
these
rules
from
speech
is
hindered
when
the
input
is
ambiguous
(i.e.
when
vowels
carry
the
ABA/ABB
rule,
while
consonants
carry
a
contradictory
AAA
rule:
ba
bi
ba
versus
ba
bi
bi;
[13].
Are
all
repetition
patterns
equally
easy
to
learn?
In
challenging
contexts,
adjacent
repetitions
are
seemingly
more
reliably
extracted
than
non-adjacent
repetitions.
Thus,
7-month-old
and
12-month-old
monolingual
infants
learn
only
the
adjacent
rule
(AAB)
when
pre-
sented
simultaneously
with
a
non-adjacent
one
[23

]
(while
12-month-old
bilinguals
learn
both
[24])
in
an
anticipatory
eye-tracking
task,
and
so
do
78-month-olds
presented
with
spatially
unstructured
shapes
(ABB)
[21],
or
with
syllables
in
which
the
repetition
(ABB)
is
instan-
tiated
only
in
the
vowels
[13].
However,
a
recent
meta-
analysis
[14

]
does
not
find
evidence
of
adjacent
repeti-
tion
patterns
being
easier
than
patterns
containing
non-
adjacent
repetitions.
This
conclusion
is
also
confirmed
empirically
in
a
study
[25]
that
showed
that
seven-month-
old
infants
discriminate
both
adjacent
and
non-adjacent
repetitions
from
the
diversity-based
structure
ABC.
Fur-
thermore,
in
the
absence
of
familiarization,
infants
show
no
spontaneous
preference
for
repetition-based
structures
over
the
diversity-based
ones
[25].
Other
studies
suggest
that
sequence-final
repetitions
may
be
easier
to
learn
than
sequence-initial
ones
[21].
This
is
confirmed
by
a
recent
meta-analysis,
which
reports
a
larger
effect
for
ABB
as
compared
with
AAB
patterns
[14

].
In
sum,
very
young
infants
exhibit
a
robust
capacity
for
learning
rules
based
on
the
identity
relation,
that
is,
the
126
Same-different
conceptualization
Figure
1
8
7
6
5
4
3
2
1
0
consistent inconsistent
looking time (sec)
Current Opinion in Behavioral Sciences
The
results
of
Marcus
et
al.
[7

]:
looking
times
in
the
test
phase
to
the
test
items
that
were
consistent
(green)
or
inconsistent
(violet)
with
the
grammar
heard
during
familiarization.
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132
www.sciencedirect.com

relation
of
‘same’
from
speech
and
other
stimuli.
Do
infants
similarly
detect
rules
based
on
diversity,
that
is,
the
relation
‘different’?
The
behavioral
work
available
doesn’t
provide
evidence
for
this
ability
during
the
first
year
of
life.
When
familiarized
simultaneously
with
an
identity-based
(AA:
va
va,
or
ABA:
du
ba
du)
and
a
diversity-based
rule
(AB:
va
lu,
or
ABC:
du
ba
lo)
in
speech,
in
an
eye-tracking
study
measuring
anticipation
in
response
to
the
rules,
7-month-olds
and
12-month-olds
generalize
only
the
identity-based
pattern
to
novel
tokens
in
test
[23

].
A
similar
pattern
emerges
with
non-speech
stimuli.
When
presented
with
geometrical
shapes,
7-
month-old
and
12-month-old
infants
generalize
a
rule
based
on
identity
(AA)
but
fail
to
generalize
a
rule
based
on
diversity
(AB)
[9].
Neural
evidence
The
advent
of
brain
imaging,
such
as
near-infrared
spec-
troscopy
(NIRS),
has
made
it
possible
to
assess
newborns’
brain
responses
to
sequence-final
adjacent
repetitions
within
trisyllabic
sequences
(ABB:
mubaba)
and
to
other-
wise
matched
random
controls,
that
is,
diversity-based
structures
(ABC:
mubage)
[26

].
This
paradigm
does
not
include
familiarization
and
test
phases,
measuring
instead
infants’
responses
to
stimulus
blocks
of
ABB
or
ABC
sequences.
To
capture
responses
to
the
structure
of
the
sequences
rather
than
to
individual
items
or
perceptual
properties,
a
high
number
of
strongly
variable
items
(140
per
condition)
are
presented,
with
no
repetitions,
exceeding
newborns’
memory
capacity
for
rote
learning.
Under
these
conditions,
newborns
showed
a
greater
response
to
ABB
than
to
ABC
patterns
(Figure
2a)
in
the
bilateral
temporal
and
left
inferior
frontal
(involving
Broca’s
area)
regions,
suggesting
that
newborns
already
process
the
repetition-based
pattern
differently
from
the
diversity-based
control
in
the
language
network.
Interestingly,
the
advantage
for
the
repetition
sequence
was
present
from
the
beginning
of
the
experiment
and
increased
over
its
time
course
(Figure
2b).
This
suggests
Encoding
repetition-based
structures
from
speech
de
la
Cruz-Pavı
´a and
Gervain
127
Figure
2
(a)
(b)
oxyHb concentration (mmol*mm)
ABC ABB
Blocks 1-4
Blocks 11-14
Current Opinion in Behavioral Sciences
Newborns’
hemodynamic
responses
to
sequence-final
adjacent
repetition-based
and
diversity-based
structures,
adapted
from
Ref.
[26

].
(a)
The
hemodynamic
response
in
the
left
temporal
area.
(b)
The
time
course
of
the
hemodynamic
response
in
the
left
frontal
area:
an
immediate
advantage
for
ABB
over
ABC
(Blocks
1-4),
which
further
increases
over
time
(Blocks
11-14).
www.sciencedirect.com
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132

two
mechanisms
involved
in
processing
the
repetition
structure:
one
immediately
detecting
the
repetition
and
one
building
up
knowledge
about
it
over
time
[26

].
These
two
neural
signatures
have
also
been
observed
when
stimuli
were
less
complex,
only
12
items
per
condition
instead
of
140,
with
several
repetitions
of
each
item
[27].
However,
under
these
less
complex,
more
redundant
conditions,
the
early
increased
responses
to
repetitions
remained
stable
over
time,
while
the
response
to
the
diversity-based
structure
increased,
likely
due
to
the
rote
learning
of
(at
least
some)
individual
items.
With
NIRS,
it
is
possible
to
not
only
compare
the
two
conditions,
but
also
to
establish
whether
a
given
condition
evoked
activation,
that
is,
was
processed
or
responded
to,
at
all.
Compared
to
a
silent
baseline,
in
this
study,
activation
was
significant
to
the
ABB
structure
in
the
bilateral
temporal
and
left
frontal
areas,
whereas
the
ABC
pattern
elicited
only
weak
responses
in
a
single
right
temporal
channel.
These
results
suggest
that
while
new-
borns
robustly
encode
the
sequence-final
adjacent
repe-
tition
pattern,
their
ability
to
represent
diversity-based
patterns
is
restricted.
The
newborn
brain
also
shows
an
advantage
for
sequence-initial
adjacent
repetitions
(AAB:
babamu)
as
compared
to
diversity-based
ABC
controls,
with
a
pattern
of
activation
similar
to
the
one
elicited
by
sequence-final
repetitions
[28

].
When
ABB
and
AAB
structures
are
compared
directly,
that
is,
when
infants
are
presented
with
blocks
containing
only
ABB
or
only
AAB
sequences,
both
structures
elicit
the
same
increased
activation
[28

].
But
when
their
dis-
crimination
is
tested
indirectly
in
an
alternating/non-
alternating
paradigm
(Figure
3),
newborns
respond
dif-
ferentially
in
the
left
inferior
frontal
areas
to
alternating
and
non-alternating
blocks
(i.e.
blocks
containing
tokens
of
the
two
structures
in
strict
alternation:
pepena
talulu
kokofe
bisoso
.
.
.
versus
blocks
containing
tokens
of
only
one
structure:
e.g.
pepena
kokofe
duduzi
.
.
.
)
[28

].
New-
borns,
as
seven-month-olds
[7

],
successfully
discriminate
then
between
sequence-initial
(e.g.
pepena)
and
–final
(e.g.
talulu)
adjacent
repetitions,
which
entails
encoding
not
only
the
repetition
itself
(
pepe,
lulu),
but
also
its
serial
position
(initial:
pepe–
versus
final:
–lulu)
in
the
sequence.
By
contrast,
newborns’
neural
responses
to
non-adjacent
repetitions
(ABA:
bamuba)
and
the
diversity-based
ABC
controls
do
not
differ
[26

],
suggesting
that
non-adjacent
repetitions
are
more
challenging
for
the
newborn
brain
than
adjacent
ones.
These
results
establish
the
earliest
possible
developmen-
tal
onset
for
the
sensitivity
to
repetition,
that
is,
identity-
based
linguistic
rules.
The
asymmetry
between
adjacent
and
non-adjacent
repetitions
converges
with
the
one
observed
behaviorally
in
older
infants
in
some
[20,21],
although
not
all
studies
[14

].
Interestingly,
newborns
process
initial
and
final
repetitions
equally
well,
not
demonstrating
the
advantage
some
behavioral
studies
found
for
sequence-final
repetitions
[14

].
It
remains
to
be
investigated
whether
these
divergences
represent
developmental
changes
or
are
rather
attributable
to
meth-
odological
differences.
The
handful
of
NIRS
studies
investigating
older
infants
point
to
an
interesting
developmental
change
in
encoding
diversity-based
structures.
In
simple
block
presentation
paradigms
similar
to
those
used
with
newborns,
six-
month-olds
show
an
increased
response,
that
is,
higher
than
a
silent
baseline,
to
simple
blocks
of
both
the
repeti-
tion
and
the
random
control,
whether
trisyllabic
[29]
or
bisyllabic
[30

],
and
as
a
result,
no
difference
between
the
two
conditions.
Nevertheless,
when
the
two
structures
are
presented
in
an
alternating/non-alternating
paradigm,
six-
month-olds
can
discriminate
them
[31].
Each
of
these
two
results,
taken
separately,
could
be
explained
without
positing
developmental
changes
in
infants’
encoding
of
diversity-based
structures.
The
simi-
lar
activation
found
in
the
repeated
and
random
condi-
tions,
when
presented
in
simple
blocks,
could
result
from
the
low-level
auditory
processing
of
the
sequences,
with-
out
the
processing
of
their
structure.
Independently,
infants
could
discriminate
the
two
structures
in
the
alter-
nating/non-alternating
paradigm
simply
by
encoding
the
repetition-based
structures
alone,
without
encoding
the
fact
that
the
random
sequences
contain
different
syllables.
However,
when
the
two
findings
are
taken
together,
these
explanations
are
not
sufficient,
and
the
ability
to
represent
difference
in
syllables
needs
to
be
posited.
Indeed,
if
the
response
to
repetitions
and
random
sequences
is
auditory
only,
excluding
the
processing
of
their
structure,
then
we
would
predict
no
discrimi-
nation
in
the
alternating/non-alternating
paradigm
either.
But
discrimination
is
actually
observed
there.
Conversely,
if
we
assume,
on
the
basis
of
the
alternat-
ing/non-alternating
paradigm,
that
only
repetitions,
that
is,
sameness,
is
encoded,
but
not
difference,
then
we
would
expect
an
advantage
for
it
in
the
simple
block
comparison,
just
like
in
the
case
of
newborns.
But
equally
high
responses
are
found
for
both
repetitions
and
random
sequences.
These
findings
can
only
be
reconciled,
if
we
assume
that
six-month-old
infants
encode
diversity-based
patterns
as
well
as
identity-
based
ones.
This
result
contrasts
with
infants’
failure
to
detect
diversity-based
rules
even
at
12
months
of
age
in
behavioral
studies
(shapes:
[9],
syllables:
[23

]).
How-
ever,
in
these
studies
infants
needed
to
learn
a
regular-
ity
and
associate
it
with
a
position
on
the
screen.
Infants’
ability
to
encode
difference
might
be
too
weak
to
support
such
complex
tasks.
128
Same-different
conceptualization
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132
www.sciencedirect.com

Encoding
repetition-based
structures
from
speech
de
la
Cruz-Pavı
´a and
Gervain
129
Figure
3
(a)
(b)
(c)
Current Opinion in Behavioral Sciences
Newborns
discriminate
sequence-initial
and
sequence-final
adjacent
repetitions.
(a)
The
simple
block
and
(b)
the
alternating/non-alternating
designs
used
in
[28

].
(c)
The
differential
response
to
non-alternating
blocks
in
the
left
inferior
frontal
region.
www.sciencedirect.com
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132

Potential
support
for
infants’
emerging
ability
to
repre-
sent
difference,
at
least
in
linguistic
sequences,
comes
from
the
finding
that
by
nine
months
infants
show
stron-
ger
activation
to
diversity-based
sequences
than
to
repe-
tition-based
ones
[32].
This
pattern
of
results
could
suggest
a
complete
developmental
shift.
However,
this
result
has
to
be
taken
with
great
caution,
due
to
its
small
sample
size,
and
needs
to
be
replicated,
especially
because
it
contrasts
with
behavioral
findings
(shapes:
[9],
syllables:
[23

]),
which
show
a
repetition-advantage
at
that
age.
To
what
extent
methodological
differences
(imaging
versus
behavioral
methods)
plays
a
role
also
requires
further
investigation.
Older
infants’
neural
ability
to
detect
repetition-based
regularities
in
visual
input
has
been
investigated
only
for
linguistic
signs
and
their
non-linguistic
visual
controls
[33].
At
6
months,
hearing
infants
never
exposed
to
sign
language
show
an
advantage
for
repetition-based
sequences
of
two
nonsense
signs
(AA)
over
random
sequences
(AB)
in
bilateral
fronto-temporal
brain
areas
overlapping
with
what
is
identified
in
adults
as
the
language
network,
unlike
their
age-matched
peers’
equal
response
to
repetition-based
and
diversity-based
regular-
ities
in
speech,
but
similarly
to
newborns’
repetition
advantage
response.
This
suggests
that
in
language,
experience
modulates
how
regularities
are
processed.
Interestingly,
non-linguistic
visual
controls,
matched
in
spatio-temporal
dynamics
and
shape
to
the
signs
but
represented
as
a
cartoon
tree,
triggered
greater
responses
to
the
diversity-based
than
to
the
repetition-based
struc-
ture,
pointing
to
mechanisms
that
process
repetitions
differently
as
a
function
of
their
relevance
in
a
given
cognitive
domain.
Discussion
and
conclusion
Taken
together,
the
behavioral
and
neuroimaging
studies
provide
firm
evidence
for
young
infants’
ability
to
encode
regularities
predicated
over
identical
elements
in
speech
from
birth,
and
stimulus-dependent,
more
variable
abili-
ties
for
other
stimuli
from
3
to
5
months.
Is
this
ability
an
abstract,
symbolic
rule-learning
mecha-
nism,
as
initially
conceptualized
[7

],
or
could
repetition-
based
structures
(ABB,
AAB,
ABA
etc.)
be
learned
relying
on
lower
level
perceptual
and
memory
mechanisms
[34]?
Adjacent
repetitions
have
been
argued
to
be
Gestalts
automatically
detected
by
the
perceptual
system,
without
relying
on
abstract
symbols,
since
different
animals
are
also
sensitive
to
immediate
repetition
[35].
Furthermore,
while
from
a
symbolic
perspective
identity
is
a
two-place
predicate
similar
to
the
‘greater/smaller
than’
ordinal
relationship,
adults
are
better
at
learning
identity-based
than
ordinal
relations
[36].
Also,
in
the
tested
structures,
repetitions
always
appear
at
sequence
edges,
a
position
known
from
the
memory
literature
to
be
special
(primacy/
recency
effects).
Indeed,
adults
are
better
at
detecting
repetitions
at
the
edges
of
5-syllable-long
or
7-syllable-
long
sequences
than
in
medial
positions
[37,38].
How-
ever,
to
distinguish
sequence-initial
and
–final
repetitions
[7

,28

],
even
if
the
repetition
is
detected
by
a
perceptual
Gestalt
and
the
edges
as
memory
primitives,
the
two
need
to
be
combined
into
a
joint
representation.
This
repre-
sentation
is,
therefore,
at
least
one
level
more
abstract
than
the
two
low-level
mechanisms
that
feed
into
it.
It
is,
therefore,
reasonable
to
conclude
that
young
infants
have
the
ability
to
encode
repetitions,
that
is,
identity
or
sameness,
from
birth,
at
least
in
language.
The
evidence
is
less
conclusive
about
the
representation
of
difference.
Many
results
can
be
explained
by
assuming
that
infants
can
represent
sameness
only
and
develop
strategies
to
choose
or
avoid
it
(for
a
detailed
discussion,
see
Ref.
[9]).
This
is
because
many
studies
use
methods
that
cannot
establish
an
absolute
measure
of
infants’
preference,
but
rather
compare
measures
between
conditions.
In
illustra-
tion,
looking
time
for
a
stimulus
type
is
not
in
itself
interpretable.
Rule
learning
is
inferred
from
looking
time
data,
if
there
is
a
difference
in
looking
times
between
two
relevant
conditions,
for
example,
repetition
versus
non-
repetition.
A
few
studies,
however,
used
methods
that
establish
a
measure
of
infants’
processing
of
a
single
condition
compared
to
some
baseline.
Anticipatory
looking
assessed
by
eye-tracking
is
one
such
paradigm.
Using
this
design,
7-month-olds
and
12-month-olds
have
been
found
to
correctly
learn
to
anticipate
when
hearing
a
repetition-
but
not
a
diversity-based
regularity
[23

].
This
result
suggests
that
the
representation
of
diversity
may
not
be
in
place
in
the
first
year
of
life.
This
task,
however,
is
complex
–
infants
need
to
learn
the
regu-
larity
and
its
association
with
a
position
on
the
screen
to
anticipate
correctly.
The
ability
to
encode
diversity
could
be
present,
but
too
weak
to
support
this
task.
The
NIRS
data
indeed
suggests
this
to
be
the
case.
While
the
response
to
ABC
patterns
is
not
distinguish-
able
from
baseline
at
birth
[26

],
by
six
months,
it
becomes
significant,
reaching
amplitudes
similar
to
the
responses
to
repetitions
[29,30

].
This
is
the
earliest
evidence
available
to
suggest
that
infants
can
represent
not
only
sameness,
but
also
differ-
ence
–
at
least
for
speech
stimuli.
Given
that
at
the
same
age,
infants
show
a
repetition-advantage
for
sign,
their
emerging
ability
to
encode
diversity
in
speech
is
likely
linked
to
their
experience
and
the
developmental
trajec-
tory
of
language
learning.
It
remains
an
open
question
whether,
tested
using
brain
imaging,
infants
show
evi-
dence
of
encoding
difference
in
other
domains,
and
if
yes,
whether
and
how
this
may
be
linked
to
developmental
changes
and
learning
constraints
characteristic
of
these
domains,
for
instance
how
spatial
structure
versus
130
Same-different
conceptualization
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132
www.sciencedirect.com

meaningfulness
play
a
role
in
extracting
regularities
from
visual
stimuli.
Conflict
of
interest
statement
Nothing
declared.
Acknowledgements
This
work
was
supported
by
the
ERC
Consolidator
Grant
“BabyRhythm”
nr.
773202
to
JG,
and
the
Basque
Foundation
for
Science
Ikerbasque
and
the
Spanish
Ministry
of
Science
and
Innovation
[Grant
nr.
PID2019-
105100RJ-I00]
to
IdlCP.
References
and
recommended
reading
Papers
of
particular
interest,
published
within
the
period
of
review,
have
been
highlighted
as:

of
special
interest
1.
Chomsky
N:
A
review
of
B.
F.
Skinner’s
verbal
behavior.
Language
1959,
35:26-58.
2.
Guasti
MT:
Language
Acquisition
:
The
Growth
of
Grammar.
MIT
Press;
2002.
3.
Pinker
S:
Language
Learnability
and
Language
Development.
Harvard
University
Press;
1984.
4.
Elman
J,
Bates
E,
Johnson
M,
Karmiloff-Smith
A,
Parisi
D,
Plunkett
K:
Rethinking
Innateness:
A
Connectionist
Perspective
on
Development
(Neural
Networks
and
Connectionist
Modeling).
The
MIT
Press;
1997.
5.
Tomasello
M:
Do
young
children
have
adult
syntactic
competence?
Cognition
2000,
74:209-253.
6.

Gervain
J,
de
la
Cruz
Pavia
I,
Gerken
L:
Behavioural
and
imaging
studies
of
infant
artificial
grammar
learning.
Top
Cogn
Sci
2020,
12:815-827
This
review
discusses
the
literature
testing
infants’
ability
to
learn
gram-
matical
properties
of
language,
focusing
on
studies
that
use
the
Artificial
Grammar
Learning
paradigm.
7.

Marcus
GF,
Vijayan
S,
Rao
SB,
Vishton
PM:
Rule
learning
by
seven-month-old
infants.
Science
1999,
283:77-80
This
pioneering
study
uncovered
young
infants’
abilities
to
encode
and
discriminate
structures
containing
adjacent
and
non-adjacent
repetition-
based
regularities.
8.
Hochmann
J-R,
Mody
S,
Carey
S:
Infants’
representations
of
same
and
different
in
match-
and
non-
match-to
sample.
Cogn
Psychol
2016,
86:87-111.
9.
Hochmann
J-R,
Carey
S,
Mehler
J:
Infants
learn
a
rule
predicated
on
the
relation
same
but
fail
to
simultaneously
learn
a
rule
predicated
on
the
relation
different.
Cognition
2018,
177:49-57.
10.
Frank
MC,
Slemmer
JA,
Marcus
GF,
Johnson
SP:
Information
from
multiple
modalities
helps
5-month-olds
learn
abstract
rules.
Dev
Sci
2009,
12:504-509.
11.

Gerken
L:
Decisions,
decisions:
Infant
language
learning
when
multiple
generalizations
are
possible.
Cognition
2006,
98:B67-
B74
The
author
investigates
the
information
in
which
infants
focus
to
general-
ize
a
repetition-based
rule
to
new
stimuli.
She
presents
infants
with
sequences
from
an
artificial
language
containing
two
logically
possible
generalizations.
The
results
of
two
studies
show
that
infants
made
the
generalization
most
statistically
consistent
with
the
input
they
had
received,
rather
than
making
the
more
abstract
generalization.
12.

Marcus
GF,
Fernandes
KJ,
Johnson
SP:
Infant
rule
learning
facilitated
by
speech.
Psychol
Sci
2007,
18:387-391
The
authors
show
that
young
infants
fail
to
encode
repetition-based
regularities
directly
from
a
number
of
non-linguistic
auditory
stimuli
such
as
musical
tones
or
animal
sounds.
Strikingly,
infants
succeed
if
they
are
first
familiarized
with
these
regularities
in
speech
sequences.
13.
Thiessen
ED:
Effects
of
inter-and
intra-modal
redundancy
on
infants’
rule
learning.
Lang
Learn
Dev
2012,
8:197-214.
14.

Rabagliati
H,
Ferguson
B,
Lew-Williams
C:
The
profile
of
abstract
rule
learning
in
infancy:
Meta-analytic
and
experimental
evidence.
Dev
Sci
2019,
22
e12704
This
recent
meta-analysis
examines
20
previous
reports
on
infants’
learning
of
repetition-based
regularities.
It
additionally
contains
an
experi-
ment
testing
young
infants’
ability
to
extract
rules
from
sign
in
different
communicative
contexts.
The
authors
argue
that
the
ability
to
encode
repetition-based
structures
is
facilitated
when
the
signal
is
‘meaningful’
to
the
infant.
15.
Dawson
C,
Gerken
L:
From
domain-generality
to
domain-
sensitivity:
4-month-olds
learn
an
abstract
repetition
rule
in
music
that
7-month-olds
do
not.
Cognition
2009,
111:378-382.
16.
Ferguson
B,
Lew-Williams
C:
Communicative
signals
support
abstract
rule
learning
by
7-month-old
infants.
Sci
Rep
2016,
6:1-7.
17.

Ferguson
B,
Franconeri
SL,
Waxman
SR:
Very
young
infants
learn
abstract
rules
in
the
visual
modality.
PLoS
One
2018,
13
e0190185
This
study
shows
that
infants
as
young
as
3–4
months
of
age
learn
repetition-based
rules
implemented
over
abstract
geometrical
shapes
when
the
visual
sequences
are
presented
spatially
structured,
namely,
left-to-right
and
showing
all
elements
in
a
sequence
simultaneously.
This
result
evidences
the
different
requirements
of
the
auditory
and
visual
systems
in
rule
learning.
18.
Saffran
JR,
Pollak
SD,
Seibel
RL,
Shkolnik
A:
Dog
is
a
dog
is
a
dog:
Infant
rule
learning
is
not
specific
to
language.
Cognition
2007,
105:669-680.
19.
Bulf
H,
Brenna
V,
Valenza
E,
Johnson
SP,
Turati
C:
Many
faces,
one
rule:
the
role
of
perceptual
expertise
in
infants’
sequential
rule
learning.
Front
Psychol
2015,
6:1595.
20.
Rabagliati
H,
Senghas
A,
Johnson
S,
Marcus
GF:
Infant
rule
learning:
advantage
language,
or
advantage
speech?
PLoS
One
2012,
7:e40517.
21.
Johnson
SP,
Fernandes
KJ,
Frank
MC,
Kirkham
N,
Marcus
G,
Rabagliati
H,
Slemmer
JA:
Abstract
rule
learning
for
visual
sequences
in
8-and
11-month-olds.
Infancy
2009,
14:2-18.
22.
Bulf
H,
de
Hevia
MD,
Gariboldi
V,
Cassia
VM:
Infants
learn
better
from
left
to
right:
A
directional
bias
in
infants’
sequence
learning.
Sci
Rep
2017,
7
2437.
23.

Kova
´cs
A
´M:
Extracting
regularities
from
noise:
do
infants
encode
patterns
based
on
same
and
different
relations?
Lang
Learn
2014,
64:65-85
Using
eye-tracking,
the
author
investigates
whether
young
infants
encode
repetition-based
and
diversity-based
structures,
and
use
them
to
antici-
pate
the
occurrence
of
a
visual
stimulus
in
a
given
position
on
a
screen.
This
research
shows
that
infants
generalized
repetition-based
but
not
diversity-based
regularities.
24.
Kova
´cs
A
´M,
Mehler
J:
Flexible
learning
of
multiple
speech
structures
in
bilingual
infants.
Science
2009,
325:611-612.
25.
Gervain
J,
Werker
JF:
Learning
non-adjacent
regularities
at
age
0;
7.
J
Child
Lang
2013,
40:860-872.
26.

Gervain
J,
Macagno
F,
Cogoi
S,
Pena
M,
Mehler
J:
The
neonate
brain
detects
speech
structure.
Proc
Natl
Acad
Sci
U
S
A
2008,
105:14222-14227
Using
near-infrared
spectroscopy
(NIRS),
the
authors
show
that
newborn
infants
encode
adjacent
repetitions
from
speech
input
(ABB),
and
dis-
criminate
them
from
random
sequences
(ABC).
In
contrast,
newborns
do
not
discriminate
random
sequences
from
sequences
containing
non-
adjacent
repetitions
(ABA).
27.
Bouchon
C,
Nazzi
T,
Gervain
J:
Hemispheric
asymmetries
in
repetition
enhancement
and
suppression
effects
in
the
newborn
brain.
PLoS
One
2015,
10:e0140160.
28.

Gervain
J,
Berent
I,
Werker
JF:
Binding
at
birth:
the
newborn
brain
detects
identity
relations
and
sequential
position
in
speech.
J
Cogn
Neurosci
2012,
24:564-574
Using
near-infrared
spectroscopy
(NIRS),
the
authors
show
that
newborn
infants
discriminate
speech
sequences
containing
adjacent
repetitions
at
the
initial
and
final
edges
(i.e.
AAB
versus
ABB).
This
result
evidences
a
certain
level
of
abstraction,
as
infants
need
to
detect
the
repetition
as
well
as
its
position
within
the
sequence.
Encoding
repetition-based
structures
from
speech
de
la
Cruz-Pavı
´a and
Gervain
131
www.sciencedirect.com
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132

29.
Radulescu
S,
Wijnen
F,
Avrutin
S,
Gervain
J:
Same
processing
costs
for
encoding
sameness
and
difference
in
the
developing
brain:
An
fNIRS
study
with
6-7-month-olds.
.
In
preparation
2020.
30.

de
la
Cruz-Pavı
´a
I,
Gervain
J,
Berent
I:
The
detection
of
repetition-
based
regularities
from
visual
input
at
6
months
of
age.
2018.
Poster
presented
at
fNIRS
2018,
5-8
Oct
2018,
Tokyo,
Japan
This
work
shows
that
young
infants
never
exposed
to
sign
language
nonetheless
extract
repetition-based
structures
(AA)
from
dynamic
signs,
as
well
as
from
visual
analogs.
Interestingly,
infants’
brain
activation
measured
using
NIRS
is
greater
for
repetitions
in
sign
language,
but
for
random
sequences
(AB)
in
visual
analogs.
When
presented
with
speech
stimuli,
infants
exhibit
a
similar
degree
of
activation
to
repeated
and
random
sequences.
31.
de
la
Cruz-Pavı
´a I,
Gervain
J:
Representing
difference:
determining
6-month-old
infants’
abilities
to
learn
random
sequences
of
syllables.
2020.
Poster
presented
at
ICIS
Biennial
Congress,
July
6
–
9,
2020,
Glasgow,
Scotland.
32.
Wagner
JB,
Fox
SE,
Tager-Flusberg
H,
Nelson
CA:
Neural
processing
of
repetition
and
non-repetition
grammars
in
7-
and
9-month-old
infants.
Front
Psychol
2011,
2:168.
33.
Berent
I,
de
la
Cruz-Pavı
´a
I,
Gervain
J:
Infants
preferentially
extract
rules
from
(sign)
language.
.
Under
review
2020.
34.
Endress
AD,
Nespor
M,
Mehler
J:
Perceptual
and
memory
constraints
on
language
acquisition.
Trends
Cogn
Sci
2009,
13:348-353.
35.
Giurfa
M,
Zhang
S,
Jenett
A,
Menzel
R,
Srinivasan
MV:
The
concepts
of
‘sameness’
and
‘difference’
in
an
insect.
Nature
2001,
410:930-933.
36.
Endress
AD,
Dehaene-Lambertz
G,
Mehler
J:
Perceptual
constraints
and
the
learnability
of
simple
grammars.
Cognition
2007,
105:577-614.
37.
Endress
AD,
Scholl
BJ,
Mehler
J:
The
role
of
salience
in
the
extraction
of
algebraic
rules.
J
Exp
Psychol
2005,
134:406-419.
38.
Gervain
J,
Endress
AD:
Learning
multiple
rules
simultaneously:
affixes
are
more
salient
than
reduplications.
Mem
Cogn
2017,
45:508-527.
132
Same-different
conceptualization
Current
Opinion
in
Behavioral
Sciences
2021,
37:125–132
www.sciencedirect.com